Split (2)

train · 31 rows

PMCID stringlengths 10 10	Title stringlengths 46 162	FullText stringlengths 17.4k 118k
PMC4802042	A conserved motif in JNK/p38-specific MAPK phosphatases as a determinant for JNK1 recognition and inactivation	Mitogen-activated protein kinases (MAPKs), important in a large array of signalling pathways, are tightly controlled by a cascade of protein kinases and by MAPK phosphatases (MKPs). MAPK signalling efficiency and specificity is modulated by protein–protein interactions between individual MAPKs and the docking motifs in cognate binding partners. Two types of docking interactions have been identified: D-motif-mediated interaction and FXF-docking interaction. Here we report the crystal structure of JNK1 bound to the catalytic domain of MKP7 at 2.4-Å resolution, providing high-resolution structural insight into the FXF-docking interaction. The FNFL segment in MKP7 directly binds to a hydrophobic site on JNK1 that is near the MAPK insertion and helix αG. Biochemical studies further reveal that this highly conserved structural motif is present in all members of the MKP family, and the interaction mode is universal and critical for the MKP-MAPK recognition and biological function.The mitogen-activated protein kinases (MAPKs) are central components of the signal-transduction pathways, which mediate the cellular response to a variety of extracellular stimuli, ranging from growth factors to environmental stresses123. The MAPK signalling pathways are evolutionally highly conserved. The basic assembly of MAPK pathways is a three-tier kinase module that establishes a sequential activation cascade: a MAPK kinase kinase activates a MAPK kinase, which in turn activates a MAPK. The three best-characterized MAPK signalling pathways are mediated by the kinases extracellular signal-regulated kinase (ERK), c-Jun N-terminal kinase (JNK) and p38. The ERK pathway is activated by various mitogens and phorbol esters, whereas the JNK and p38 pathways are stimulated mainly by environmental stress and inflammatory cytokines456. The MAPKs are activated by MAPK kinases that phosphorylate the MAPKs at conserved threonine and tyrosine residues within their activation loop. After activation, each MAPK phosphorylates a distinct set of protein substrates, which act as the critical effectors that enable cells to mount the appropriate responses to varied stimuli. MAPKs lie at the bottom of conserved three-component phosphorylation cascades and utilize docking interactions to link module components and bind substrates78. Two types of docking motifs have been identified in MAPK substrates and cognate proteins: kinase-interacting motif (D-motif) and FXF-motif (also called DEF motif, docking site for ERK FXF). The best-studied docking interactions are those between MAP kinases and ‘D-motifs', which consists of two or more basic residues followed by a short linker and a cluster of hydrophobic residues. The D-motif-docking site (D-site) in MAPKs is situated in a noncatalytic region opposite of the kinase catalytic pocket and is comprised of a highly acidic patch and a hydrophobic groove. D-motifs are found in many MAPK-interacting proteins, including substrates, activating kinases and inactivating phosphatases, as well as scaffolding proteins. A second docking motif for MAPKs consists of two Phe residues separated by one residue (FXF-motif). This motif has been observed in several MAPK substrates910111213. The FXF-motif-binding site of ERK2 has been mapped to a hydrophobic pocket formed between the P+1 site, αG helix and the MAPK insert14. However, the generality and mechanism of the FXF-mediated interaction is unclear. The physiological outcome of MAPK signalling depends on both the magnitude and the duration of kinase activation15. Downregulation of MAPK activity can be achieved through direct dephosphorylation of the phospho-threonine and/or tyrosine residues by various serine/threonine phosphatases, tyrosine phosphatases and dual-specificity phosphatases (DUSPs) termed MKPs. MKPs constitute a group of DUSPs that are characterized by their ability to dephosphorylate both phosphotyrosine and phosphoserine/phospho-threonine residues within a substrate1617. Dysregulated expression of MKPs has been associated with pathogenesis of various diseases, and understanding their precise recognition mechanism presents an important challenge and opportunity for drug development1819. Here, we present the crystal structure of JNK1 in complex with the catalytic domain of MKP7. This structure reveals the molecular mechanism underlying the docking interaction between MKP7 and JNK1. In the JNK1–MKP7 complex, a hydrophobic motif (FNFL) that initiates the helix α5 in the MKP7 catalytic domain directly binds to the FXF-motif-binding site on JNK1, providing the structural insight into the classic FXF-type docking interaction. Biochemical and modelling studies further demonstrate that the molecular interactions mediate this key element for substrate recognition are highly conserved among all MKP-family members. Thus, our study reveals a hitherto unrecognized interaction mode for encoding complex target specificity among MAPK isoforms. DUSPs belong to the protein-tyrosine phosphatases (PTPase) superfamily, which is defined by the PTPase-signature motif CXXGXXR20. MKPs represent a distinct subfamily within a larger group of DUSPs. In mammalian cells, the MKP subfamily includes 10 distinct catalytically active MKPs. All MKPs contain a highly conserved C-terminal catalytic domain (CD) and an N-terminal kinase-binding domain (KBD)1521. The KBD is homologous to the rhodanese family and contains an intervening cluster of basic amino acids, which has been suggested to be important for interacting with the target MAPKs. On the basis of sequence similarity, substrate specificity and predominant subcellular localization, the MKP family can be further divided into three groups (Fig. 1). Biochemical and structural studies have revealed that the KBD of MKPs is critical for MKP3 docking to ERK2, and MKP5 binding to p38α, although their binding mechanisms are completely different2223. However, it is unknown if other MAPKs can interact with the KBD of their cognate phosphatases in the same manner as observed for recognition of ERK2 and p38α by their MKPs, or whether they recognize distinct docking motifs of MKPs. MKP7, the biggest molecule in the MKP family, selectively inactivates JNK and p38 following stress activation24. In addition to the CD and KBD, MKP7 has a long C-terminal region that contains both nuclear localization and export sequences by which MKP7 shuttles between the nucleus and the cytoplasm (Fig. 2a). To quantitatively assess the contribution of the N-terminal domain to the MKP7-catalysed JNK1 dephosphorylation, we first measured the kinetic parameters of the C-terminal truncation of MKP7 (MKP7ΔC304, residues 5–303) and MKP7-CD (residues 156–301) towards phosphorylated JNK1 (pJNK1). Figure 2b shows the variation of initial rates of the MKP7ΔC304 and MKP7-CD-catalysed reaction with the concentration of phospho-JNK1. Because the concentrations of MKP7 and pJNK1 were comparable in the reaction, the assumption that the free-substrate concentration is equal to the total substrate concentration is not valid. Thus, the kinetic data were analysed using the general initial velocity equation, taking substrate depletion into account: The kcat and Km of the MKP7-CD (0.028 s and 0.26 μM) so determined were nearly identical to those of MKP7ΔC304 (0.029 s and 0.27 μM), indicating that the MKP7-KBD has no effect on enzyme catalysis. We next examined the interaction of JNK1 with the CD and KBD of MKP7 by gel filtration analysis. When 3 molar equivalents of CD were mixed with 1 molar equivalent of JNK1, a significant amount of CD co-migrated with JNK1 to earlier fractions, and the excess amount of CD was eluted from the size exclusion column as a monomer, indicating stable complex formation (Fig. 2c). In contrast, no KBD–JNK1 complex was detected when 3 molar equivalents of KBD were mixed with 1 molar equivalent of JNK1. To further confirm the JNK1–MKP7-CD interaction, we performed a pull-down assay using the purified proteins. As shown in Fig. 2d, the CD of MKP7 can be pulled down by JNK1, while the KBD failed to bind to the counterpart protein. Taken together, our data indicate that the CD of MKP7, but not the KBD domain, is responsible for JNK substrate-binding and enzymatic specificity. To understand the molecular basis of JNK1 recognition by MKP7, we determined the crystal structure of unphosphorylated JNK1 in complex with the MKP7-CD (Fig. 3a, Supplementary Fig. 1a and Table 1). In the complex, JNK1 has its characteristic bilobal structure comprising an N-terminal lobe rich in β-sheet and a C-terminal lobe that is mostly α-helical. The overall folding of MKP7-CD is typical of DUSPs, with a central twisted five-stranded β-sheet surrounded by six α-helices. One side of the β-sheet is covered with two α-helices and the other is covered with four α-helices (Fig. 3b). The catalytic domain of MKP7 interacts with JNK1 through a contiguous surface area that is remote from the active site. MKP7-CD is positioned onto the JNK1 molecule so that the active site of the phosphatase faces towards the activation segment. In an alignment of the structure of MKP7-CD with that of VHR25, an atypical ‘MKP' consisting of only a catalytic domain, 119 of 147 MKP7-CD residues could be superimposed with a r.m.s.d. (root mean squared deviation) of 1.05 Å (Fig. 3c). The most striking difference is that helix α0 and loop α0–β1 of VHR are absent in MKP7-CD. Another region that cannot be aligned with VHR is found in loop β3–β4. This loop is shortened by nine residues in MKP7-CD compared with that in VHR. Since helix α0 and the following loop α0–β1 are known for a substrate-recognition motif of VHR and other phosphatases, the absence of these moieties implicates a different substrate-binding mode of MKP7. The active site of MKP7 consists of the phosphate-binding loop (P-loop, Cys244-Leu245-Ala246-Gly247-Ile248-Ser249-Arg250), and Asp213 in the general acid loop (Fig. 3b and Supplementary Fig. 1b). The MKP7-CD structure near the active site exhibits a typical active conformation as found in VHR and other PTPs25. The catalytic residue, Cys244, is located just after strand β5 and optimally positioned for nucleophilic attack262728. Asp213 in MKP7 also adopts a position similar to that of Asp92 in VHR (Supplementary Fig. 1c), indicating that Asp213 is likely to function as the general acid in MKP7. We also observed the binding of a chloride ion in the active site of MKP7-CD. It is located 3.36 Å from the Cys244 side chain and makes electrostatic interactions with the dipole moment of helix α3 and with several main-chain amide groups. The side chain of strictly conserved Arg250 is oriented towards the negatively charged chloride, similar to the canonical phosphate-coordinating conformation. Thus this chloride ion is a mimic for the phosphate group of the substrate, as revealed by a comparison with the structure of PTP1B in complex with phosphotyrosine29 (Supplementary Fig. 1d). Although the catalytically important residues in MKP7-CD are well aligned with those in VHR, the residues in the P-loop of MKP7 are smaller and have a more hydrophobic character than those of VHR (Cys124-Arg125-Glu126-Gly127-Tyr128-Gly129-Arg130; Fig. 3b,c). The difference in the polarity/hydrophobicity of the surface may also point to the origin of the differences in the substrate-recognition mechanism for these two phosphatases (Supplementary Fig. 1e,f). In the complex, MKP7-CD and JNK1 form extensive protein–protein interactions involving the C-terminal helices of MKP7-CD and C-lobe of JNK1 (Fig. 3d,e). As a result, the buried solvent-accessible surface area is ∼1,315 Å. In the C-terminal domain, JNK1 has an insertion after the helix αG. This insertion consists of two helices (α1L14 and α2L14) that are common to all members of the MAPK family. The interactive surface in JNK1, formed by the helices αG and α2L14, displays a hydrophobic region, centred at Trp234 (Fig. 3d). The MKP7-docking region includes two helices, α4 and α5, and the general acid loop. The aromatic ring of Phe285 on MKP7 α5-helix is nestled in a hydrophobic pocket on JNK1, formed by side chains of Ile197, Leu198, Ile231, Trp234, Val256, Tyr259, Val260 and the aliphatic portion of His230 (Fig. 3d,f and Supplementary Fig. 1g). In addition, there are hydrogen bonds between Ser282 and Asn286 of MKP7 and His230 and Thr255 of JNK1, and the main chain of Phe215 in the general acid loop of MKP7 is hydrogen-bonded to the side chain of Gln253 in JNK1. The second interactive area involves the α4 helix of MKP7 and charged/polar residues of JNK1 (Fig. 3e). The carboxylate of Asp268 in MKP7 forms a salt bridge with side chain of Arg263 in JNK1, and Lys275 of MKP7 forms a hydrogen bond and a salt bridge with Thr228 and Asp229 of JNK1, respectively. To assess the importance of the aforementioned interactions, we generated a series of point mutations on the MKP7-CD and examined their effect on the MKP7-catalysed JNK1 dephosphorylation (Fig. 4a). When the hydrophobic residues Phe285 and Phe287 on the α5 of MKP7-CD were replaced by Asp or Ala, their phosphatase activities for JNK1 dephosphorylation decreased ∼10-fold. In comparison, replacement of the other residues (Phe215, Asp268, Lys275, Ser282, Asn286 and Leu292) with an Ala or Asp individually led to a modest decrease in catalytic efficiencies, suggesting that this position may only affect some selectivity of MKP. Mutation of Leu288 markedly reduced its solubility when expressed in Escherichia coli, resulting in the insoluble aggregation of the mutant protein. Gel filtration analysis further confirmed the key role of Phe285 in the MKP7–JNK1 interaction: no F285D–JNK1 complex was detected when 3 molar equivalents of MKP7-CD (F285D) were mixed with 1 molar equivalent of JNK1 (Fig. 4b). Interestingly, mutation of Phe287 results in a considerable loss of activity against pJNK1 without altering the affinity of MKP7-CD for JNK1 (Supplementary Fig. 2a). We also generated a series of point mutations in the JNK1 and assessed the effect on JNK1 binding using the GST pull-down assay (Fig. 4c). Substitution at Asp229, Trp234, Thr255, Val256, Tyr259 and Val260 significantly reduced the binding affinity of MKP7-CD for JNK. To determine whether the deficiencies in their abilities to bind partner proteins or carry out catalytic function are owing to misfolding of the purified mutant proteins, we also examined the folding properties of the JNK1 and MKP7 mutants with circular dichroism. The spectra of these mutants are similar to the wild-type proteins, indicating that these mutants fold as well as the wild-type proteins (Fig. 4d,e). Taken together, these results are consistent with the present crystallographic model, which reveal the hydrophobic contacts between the MKP7 catalytic domain and JNK1 have a predominant role in the enzyme–substrate interaction, and hydrophobic residue Phe285 in the MKP7-CD is a key residue for its high-affinity binding to JNK1. It has previously been reported that several cytosolic and inducible nuclear MKPs undergo catalytic activation upon interaction with the MAPK substrates15. This allosteric activation of MKP3 has been well-documented in vitro using pNPP, a small-molecule phosphotyrosine analogue of its normal substrate3031. We then assayed pNPPase activities of MKP7ΔC304 and MKP7-CD in the presence of JNK1. Incubation of MKP7 with JNK1 did not markedly stimulate the phosphatase activity, which is consistent with previous results that MKP7 solely possesses the intrinsic activity (Supplementary Fig. 2b). The small pNPP molecule binds directly at the enzyme active site and can be used to probe the reaction mechanism of protein phosphatases. We therefore examined the effects of the MKP7-CD mutants on their pNPPase activities. As shown in Fig. 4f, all the mutants, except F287D/A, showed little or no activity change compared with the wild-type MKP7-CD. In the JNK1/MKP7-CD complex structure, Phe287 of MKP7 does not make contacts with JNK1 substrate. It penetrates into a pocket formed by residues from the P-loop and general acid loop and forms hydrophobic contacts with the aliphatic portions of side chains of Arg250, Glu217 and Ile219, suggesting that Phe287 in MKP7 would play a similar role to that of its structural counterpart in the PTPs (Gln266 in PTP1B) and VHR (Phe166 in VHR) in the precise alignment of active-site residues in MKP7 with respect to the substrate for efficient catalysis32333435 (Supplementary Fig. 2c). Kinase-associated phosphatase (KAP), a member of the DUSP family, plays a crucial role in cell cycle regulation by dephosphorylating the pThr160 residue of CDK2 (cyclin-dependent kinase 2). The crystal structure of the CDK2/KAP complex has been determined at 3.0 Å (Fig. 5a)36. The interface between these two proteins consists of three discontinuous contact regions. Biochemical results suggested that the affinity and specificity between KAP and CDK2 results from the recognition site comprising CDK2 residues from the αG helix and L14 loop and the N-terminal helical region of KAP (Fig. 5b). There is a hydrogen bond between the main-chain nitrogen of Ile183 (KAP) and side chain oxygen of Glu208 (CDK2), and salt bridges between Lys184 of KAP and Asp235 of CDK2. Structural analysis and sequence alignment reveal that one of the few differences between MKP7-CD and KAP in the substrate-binding region is the presence of the motif FNFL in MKP7-CD, which corresponds to IKQY in KAP (Fig. 5c). The substitution of the two hydrophobic residues with charged/polar residues (F285I/N286K) seriously disrupts the hydrophobic interaction required for MKP7 binding on JNK1 (Fig. 4a). In addition, His230 and Val256 in JNK1 are replaced by the negatively charged residues Glu208 and Asp235 in CDK2 (Fig. 5d), and the charge distribution on the CDK2 interactive surface is quite different from that of JNK. These data indicated that a unique hydrophobic pocket formed between the MAPK insert and αG helix plays a major role in the substrate recognition by MKPs. JNK is activated following cellular exposure to a number of acute stimuli such as anisomycin, H2O2, ultraviolet light, sorbitol, DNA-damaging agents and several strong apoptosis inducers (etoposide, cisplatin and taxol)373839. To assess the effects of MKP7 and its mutants on the activation of endogenous JNK in vivo, HEK293T cells were transfected with blank vector or with HA-tagged constructs for full-length MKP7, MKP7ΔC304 and MKP7-CD or MKP7 mutants, and stimulated with ultraviolet or etoposide treatment. As shown in Fig. 6a–c, immunobloting showed similar expression levels for the different MKP7 constructs in all the cells. Overexpressed full-length MKP7, MKP7ΔC304 and MKP7-CD significantly reduced the endogenous level of phosphorylated JNK compared with vector-transfected cells. Parallel experiments showed clearly that the D-motif mutants (R56A/R57A and V63A/I65A) dephosphorylated JNK as did the wild type under the same conditions, further confirming that the MKP7-KBD is not required for the JNK inactivation in vivo. Consistent with the in vitro data, the level of phosphorylated JNK was not or little altered in MKP7 FXF-motif mutants (F285D, F287D and L288D)-transfected cells, and the MKP7 D268A and N286A mutants retained the ability to reduce the phosphorylation levels of JNK. We next tested in vivo interactions between JNK1 mutants and full-length MKP7 by coimmunoprecipitation experiments under unstimulated conditions. When co-expressed in HEK293T cells, wild-type (HA)-JNK1 was readily precipitated with (Myc)-MKP7 (Fig. 6d), indicating that MKP7 binds dephosphorylated JNK1 protein in vivo. In agreement with the in vitro pull-down results, the mutants D229A, W234D and Y259D were not co-precipitated with MKP7, and the I231D mutant had only little effect on the JNK1–MKP7 interaction (Fig. 6d and Supplementary Fig. 3a). Activation of the JNK signalling pathway is frequently associated with apoptotic cell death, and inhibition of JNK can prevent apoptotic death of multiple cells640414243. To examine whether the inhibition of JNK activity by MKP7 would provide protections against the apoptosis, we analysed the rate of apoptosis in ultraviolet-irradiated cells transfected with MKP7 (wild type or mutants) by flow cytometry. The results showed similar apoptotic rates between cells transfected with blank vector or with MKP7 (wild type or mutants) under unstimulated conditions (Supplementary Fig. 3b), while ultraviolet-irradiation significantly increased apoptotic rate in cells transfected with blank vector (Fig. 6e). Expressions of wild-type MKP7, MKP7ΔC304 and MKP7-CD significantly decreased the proportion of apoptotic cells after ultraviolet treatment. Moreover, treatment of cells expressing MKP7-KBD mutants (R56A/R57A and V63A/I65A) decreased the apoptosis rates to a similar extent as MKP7 wild type did. In contrast, cells transfected with the MKP7 FXF-motif mutants (F285D, F287D and L288D) showed little protective effect after ultraviolet treatment and similar levels of apoptosis rates were detected to cells transfected with control vectors (Fig. 6e,f). Taken together, our results suggested that FXF-motif-mediated, rather than KBD-mediated, interaction is essential for MKP7 to block ultraviolet-induced apoptosis. MKP5 belongs to the same subfamily as MKP7. MKP5 is unique among the MKPs in possessing an additional domain of unknown function at the N-terminus44 (Fig. 7a). The KBD of MKP5 interacts with the D-site of p38α to mediate the enzyme–substrate interaction. Deletion of the KBD in MKP5 leads to a 280-fold increase in Km for p38α substrate23. In contrast to p38α substrate, deletion of the MKP5-KBD had little effects on the kinetic parameters for the JNK1 dephosphorylation, indicating that the KBD of MKP5 is not required for the JNK1 dephosphorylation (Fig. 7b). The substrate specificity constant kcat /Km value for MKP5-CD was calculated as 1.0 × 10 M s, which is very close to that of MKP7-CD (1.07 × 10 M s). The crystal structure of human MKP5-CD has been determined45. Comparisons between catalytic domains structures of MKP5 and MKP7 reveal that the overall folds of the two proteins are highly similar, with only a few regions exhibiting small deviations (r.m.s.d. of 0.79 Å; Fig. 7c). Given the distinct interaction mode revealed by the crystal structure of JNK1–MKP7-CD, one obvious question is whether this is a general mechanism used by all members of the JNK-specific MKPs. To address this issue, we first examined the docking ability of JNK1 to the KBD and CD of MKP5 using gel filtration analysis and pull-down assays. It can be seen from gel filtration experiments that JNK1 can forms a stable heterodimer with MKP5-CD in solution, but no detectable interaction was found with the KBD domain (Fig. 7d). Pull-down assays also confirmed the protein–protein interactions observed above. The catalytic domain of MKP5, but not its KBD, was able to pull-down a detectable amount of JNK1 (Fig. 7e), implicating a different substrate-recognition mechanisms for p38 and JNK MAPKs. To further test our hypothesis, we generated forms of MKP5-CD bearing mutations corresponding to the changes we made on MKP7-CD on the basis of sequence and structural alignment and examined their effects on the phosphatase activity. As shown in Fig. 7f, the T432A and L449F MKP5 mutant showed little or no difference in phosphatase activity, whereas the other mutants showed reduced specific activities of MKP5. As in the case of MKP7, all the mutants, except F451D/A, showed no pNPPase activity changes compared with the wild-type MKP5-CD (Fig. 7g), and the point mutations in JNK1 also reduced the binding affinity of MKP5-CD for JNK1 (Fig. 7h). In addition, there were no significant differences in the CD spectra between wild-type and mutant proteins, indicating that the overall structures of these mutants did not change significantly from that of wild-type MKP5 protein (Supplementary Fig. 4a). Taken together, our results suggest that MKP5 binds JNK1 in a docking mode similar to that in the JNK1–MKP7 complex, and the detailed interaction model can be generated using molecular dynamics simulation based on the structure of JNK1–MKP7-CD complex (Supplementary Fig. 4b,c). In this model, the MKP5-CD adopts a conformation nearly identical to that in its unbound form, suggesting that the conformation of the catalytic domain undergoes little change, if any at all, upon JNK1 binding. In particular, Leu449 of MKP5, which is equivalent to the key residue Phe285 of MKP7, buried deeply within the hydrophobic pocket of JNK1 in the same way as Phe285 in the JNK1–MKP7-CD complex (Supplementary Fig. 4d). Despite the strong similarities between JNK1–MKP5-CD and JNK1–MKP7-CD, however, there are differences. The JNK1–MKP7-CD interaction is better and more extensive. Asp268 of MKP7-CD forms salt bridge with JNK1 Arg263, whereas the corresponding residue Thr432 in MKP5-CD may not interact with JNK1. In addition, the key interacting residues of MKP7-CD, Phe215, Leu267 and Leu288, are replaced by less hydrophobic residues, Asn379, Met431 and Met452 in MKP5-CD (Fig. 5c), respectively, which may result in weaker hydrophobic interactions between MKP5-CD and JNK1. This is consistent with the experimental observation showing that JNK1 binds to MKP7-CD much more tightly than MKP5-CD (Km value of MKP5-CD for pJNK1 substrate is ∼20-fold higher than that of MKP7-CD). The MAPKs p38, ERK and JNK, are central to evolutionarily conserved signalling pathways that are present in all eukaryotic cells. Each MAPK cascade is activated in response to a diverse array of extracellular signals and culminates in the dual-phosphorylation of a threonine and a tyrosine residue in the MAPK-activation loop2. The propagation of MAPK signals is attenuated through the actions of the MKPs. Most studies have focused on the dephosphorylation of MAPKs by phosphatases containing the ‘kinase-interaction motif ' (D-motif), including a group of DUSPs (MKPs) and a distinct subfamily of tyrosine phosphatases (HePTP, STEP and PTP-SL)4647. Crystal structures of ERK2 bound with the D-motif sequences derived from MKP3 and HePTP have been reported2248. These structures revealed that linear docking motifs in interacting proteins bind to a common docking site on MAPKs outside the kinase active site. The particular amino acids and their spacing within D-motif sequences and amino acid composition of the docking sites on MAPKs appear to determine the specificity of D-motifs for individual MAPKs. Recently, the crystal structure of a complex between the KBD of MKP5 and p38α has been obtained23. This complex has revealed a distinct interaction mode for MKP5. The KBD of MKP5 binds to p38α in the opposite polypeptide direction compared with how the D-motif of MKP3 binds to ERK2. In contrast to the canonical D-motif-binding mode, separate helices, α2 and α3′, in the KBD of MKP5 engage the p38α-docking site. Further structural and biochemical studies indicate that KBD of MKP7 may interact with p38α in a similar manner to that of MKP5. In contrast to MKP5, removal of the KBD domain from MKP7 does not drastically affect enzyme catalysis, and the kinetic parameters of MKP7-CD for p38α substrate are very similar to those for JNK1 substrate23. Taken together, these results suggest that MKP7 utilizes a bipartite recognition mechanism to achieve the efficiency and fidelity of p38α signalling. The MKP7-KBD docks to the D-site located on the back side of the p38α catalytic pocket for high-affinity association, whereas the interaction of the MKP7-CD with another p38α structural region, which is close to the activation loop, may not only stabilize binding but also provide contacts crucial for organizing the MKP7 active site with respect to the phosphoreceptor in the p38α substrate for efficient dephosphorylation. In addition to the canonical D-site, the MAPK ERK2 contains a second binding site utilized by transcription factor substrates and phosphatases, the FXF-motif-binding site (also called F-site), that is exposed in active ERK2 and the D-motif peptide-induced conformation of MAPKs91049. This hydrophobic site was first identified by changes in deuterium exchange profiles, and is near the MAPK insertion and helix αG. Interestingly, many of the equivalent residues in JNK1, important for MKP7-CD recognition, are also used for substrate binding by ERK2 (ref. 14), indicating that this site is overlapped with the DEF-site previously identified in ERK2 (Fig. 5d). MKP3 is highly specific in dephosphorylating and inactivating ERK2, and the phosphatase activity of the MKP3-catalysed pNPP reaction can be markedly increased in the presence of ERK2 (refs 30, 31). Sequence alignment of all MKPs reveals a high degree of conservation of residues surrounding the interacting region observed in JNK1–MKP7-CD complex (Supplementary Fig. 5). Therefore, it is tempting to speculate that the catalytic domain of MKP3 may bind to ERK2 in a manner analogous to the way by which MKP7-CD binds to JNK1. A comprehensive examination of the molecular basis of the specific ERK2 recognition by MKP3 is underway. The ongoing work demonstrates that although the overall interaction modes are similar between the JNK1–MKP7-CD and ERK2–MKP3-CD complexes, the ERK2–MKP3-CD interaction is less extensive and helix α4 from MKP3-CD does not interact directly with ERK2. The FXF-motif-mediated interaction is critical for both pERK2 inactivation and ERK2-induced MKP3 activation (manuscript in preparation). In summary, we have resolved the structure of JNK1 in complex with the catalytic domain of MKP7. This structure reveals an FXF-docking interaction mode between MAPK and MKP. Results from biochemical characterization of the Phe285 and Phe287 MKP7 mutants combined with structural information support the conclusion that the two Phe residues serve different roles in the catalytic reaction. Phe285 is essential for JNK1 substrate binding, whereas Phe287 plays a role for the precise alignment of active-site residues, which are important for transition-state stabilization32. This newly identified FXF-type motif is present in all MKPs, except that the residue at the first position in MKP5 is an equivalent hydrophobic leucine residue (see also Fig. 7f,g), suggesting that these two Phe residues would play a similar role in facilitating substrate recognition and catalysis, respectively. An important feature of MKP–JNK1 interactions is that MKP7 or MKP5 only interact with the F-site of JNK1. One possible explanation is that JNK1 needs to use the D-site to interact with JIP-1, a scaffold protein for JNK signalling5051. The N-terminal JNK-binding domain of JIP-1 interacts with the D-site on JNK and this interaction is required for JIP-1-mediated enhancement of JNK activation5253. In addition, JIP-1 can also associate with MKP7 via the C-terminal region of MKP7 (ref. 54). When MKP7 is bound to JIP-1, it reduces JNK activation, leading to reduced phosphorylation of the JNK target c-Jun. Thus, our biochemical and structural data allow us to present a model for the JNK1–JIP-1–MKP7 ternary complex and provide an important insight into the assembly and function of JNK signalling modules (Supplementary Fig. 6). The cDNAs of human MKP7 and MKP5 were kindly provided by Dr Mathijs Baens (University of Leuven) and Dr Eisuke Nishida (Kyoto University), respectively. The cDNAs of human ASK1, MKK4, MKK7 and JNK1 were kindly provided by Dr Zhenguo Wu (Hong Kong University of Science and Technology). The catalytic domains of MKP7 (MKP7-CD, residues 156–301) and MKP5 (MKP5-CD, 320–467) and the full-length MKP5 were cloned into the pET15b vector, resulting in the N-terminal His-fusion proteins. The KBD domains of MKP7 (MKP7-KBD, 5–138) and MKP5 (MKP5-KBD, 139–287), and the C-terminal truncation of MKP7 (MKP7ΔC304, 5–303) were cloned into pET21b vector for generation of C-terminal His-tagged proteins. The human full-length JNK1, MKK4, MKK7 and the kinase domain of ASK1 (659–951) were cloned into pGEX4T-2, pET15b and/or pET21b vectors to produce a GST- or His-tagged protein. Mutations of MKP7-CD, MKP5-CD and JNK1 were generated by overlap PCR procedure. All constructs were verified by DNA sequencing. All proteins, overexpressed in BL21(DE3) cells at 20 °C, were first purified over Ni-NTA (Qiagen) or GS4B (GE Healthcare) columns, and then by ion exchange and gel filtration chromatography (Source-15Q/15S and Superdex-200, GE Healthcare) at 4 °C. The double-phosphorylated JNK1 (phospho-JNK1) was generated by mixing JNK1 with upstream kinases MKK4, MKK7 and ASK1 in buffer containing 10 mM MgCl2 and 2 mM ATP, and further purified by gel filtration chromatography (Superdex-200, GE Healthcare) at 4 °C (ref. 55). Proteins were stored at −80 °C, and stocks for phosphatase assays were supplemented with glycerol to a final concentration of 20% (v/v). Protein concentrations were determined spectrophotometrically using theoretical molar extinction coefficients at 280 nm (ref. 56). The mixture of unphosphorylated JNK1 and MKP7-CD at 1:1 molar ratio was subjected to crystallization trials. Crystals were grown by the vapor-diffusion technique in hanging drops, and the drops were prepared by mixing equal volumes of protein with reservoir solution containing 0.1 M HEPES, pH 7.0, 14% PEG3350, 0.2 M MgCl2, 6% 1,6-Hexanediol and 0.005 M EDTA at 21 °C. Crystals were cryo-protected in reservoir solutions supplemented with 10% glycerol and then flash frozen in liquid nitrogen. The diffraction data sets were collected at beamline 17U at Shanghai Synchrotron Radiation Facility and processed with the HKL2000 package57. The crystals belong to space group P1 and comprise eight molecules per asymmetric unit (four complexes). Structure was solved by molecular replacement using Phaser58 with JNK1 (PDB 1UKH) and MKP5-CD (PDB 1ZZW) as the search models. Standard refinement was performed with programs PHENIX59 and Coot60. The crystal structure of unphosphorylated JNK1 in complex with the catalytic domain of MKP7 was refined to 2.4 Å resolution. Initial structural refinement was performed with NCS restraints, and after several rounds the restraints were removed from the calculations. The final Rwork and Rfree were 21.7 and 23.9%, respectively. The crystallographic asymmetric unit contains four JNK1–MKP7-CD complexes. The four complexes are nearly identical with an r.m.s.d.<1 Å for any complex pair in the asymmetric unit. Ramachandran analysis was carried out using PROCHECK61. Additional density at the active site of MKP7-CD was attributed to a chloride ion incorporated as a crystallizing agent, similar to those observed in the structures of MKP3-CD and MKP5-CD (refs 62, 63). The data collection and refinement statistics are summarized in Table 1. All structural representations in this paper were prepared with PYMOL (http://www.pymol.org). The activities of MKP7 and MKP5 was assayed using phospho-JNK1 as substrate in the coupled enzyme system containing 50 mM MOPS, pH 7.0, 100 mM NaCl, 0.1 mM EDTA, 50 μM MESG and 0.1 mg ml PNPase. This coupled assay uses PNPase and its chromogenic substrate MESG to monitor the production of inorganic phosphate64. The reactions were initiated by addition of 0.4 μM MKP7-CD and MKP7ΔC304 (or 0.1 μM MKP5 full-length and 0.15 μM MKP5-CD) for substrate phospho-JNK1. All experiments were carried out at 25 °C in 1.8 ml reaction mixtures, and the continuous absorbance changes were recorded with a PerkinElmer LAMBDA 45 spectrophotometer equipped with a magnetic stirrer in the cuvette holder. The quantification of inorganic phosphate produced was monitored at 360 nm with the extinction coefficient of 11,200 M cm (ref. 65). The initial rates were determined from the linear slope of the progress curves obtained. The activity of MKP7-CD or MKP5-CD mutants were assayed using pNPP or phospho-JNK1 as substrate. The phospho-JNK1 assay was performed as the same procedure mentioned above, and in the presence of wild type (as a control) or indicated mutants, and equal concentrations of phospho-JNK1. The pNPP assay was performed in the reaction mixture containg 50 mM MOPS, pH 7.0, 100 mM NaCl, 0.1 mM EDTA and 20 mM pNPP. The amount of the product p-nitrophenol was determined from the absorbance at 405 nm using a molar extinction coefficient of 18,000 M cm (ref. 31). The interactions of JNK1 with the CD and KBD domains of MKP7 and MKP5 were examined by gel filtration analyses using a Superdex-200 10/300 column on an ÄKTA FPLC (GE Healthcare). The column was equilibrated with a buffer containing 10 mM HEPES, pH 7.5, 150 mM NaCl and 2 mM dithiothreitol, and calibrated with molecular mass standards. Samples of individual proteins and indicated mixtures (500 μl each) were loaded to the Superdex-200 column and then eluted at a flow rate of 0.5 ml min. Fractions of 0.5 ml each were collected, and aliquots of relevant fractions were subjected to SDS-polyacrylamide gel electrophoresis (PAGE) followed Coomassie Blue staining. The interactions between various JNK1 mutants and MKP7-CD or MKP5-CD were assessed by GST-mediated pull-down assays at 4 °C. First, 0.5 ml GST-JNK1 proteins (6 μM) were loaded to 0.2 ml GS4B resin. The excess unbound JNK1 or other contaminants were removed by washing the column 5 times, each with 1.0 ml buffer containing 25 mM Tris-HCl, pH 8.0, 150 mM NaCl and 2 mM dithiothreitol. Then, 0.5 ml MKP7-CD or MKP5-CD (20 μM) was allowed to flow through the JNK1-bound column. After extensive washing, the bound proteins were eluted with 0.5 ml reduced glutathione (10 mM). The interactions of JNK1 with the CD and KBD domains of MKP7 and MKP5 were also examined by GST-mediated pull-down assays. The GST protein alone was used as a control. Aliquots of all eluates were subjected to SDS-PAGE, and proteins were visualized by Coomassie Blue staining. The uncropped gels are shown in Supplementary Fig. 7. The model of catalytic domain of MKP5 bound to JNK was constructed by superimposition of previous deposited structures of MKP5-CD (PDB 1ZZW) to the corresponding domains in the crystal structure of JNK1–MKP7-CD. The fractured loops in deposited structures were computationally generated using Modeller66. The program CHARMM22 (ref. 67) was then used to add hydrogen atoms, N- and C-terminal patches to the model. The model was then subjected to restrained energy minimization to optimize bonds and remove any nonbonded steric clashes. Refinement of the modelled complex was performed using NAMD2.9 package68 at 1 atm pressure and 300 K. The generated complex structure was solvated and neutralized in a box with TIP3P water at a minimum of 13 Å between the model and the wall of the box. The simulation was first set up with 1 fs time step under periodic boundary conditions. The particle mesh Ewald method was applied to model the electrostatics and the van der Waals interactions cutoff was set at 12 Å. The system was restrained for 5 ps minimization and 5 ps simulation, and followed by removing all the restraints and performing a minimization of 10 ps and an equilibration of 10 ns. Simulations were viewed using VMD69. The experiments were performed on a Chirascan-plus circular dichroism Spectrometer (Applied Photophysics, Surrey, UK) using 0.1 mm quartz cuvette. The protein sample were analysed at a concentration of 0.5 mg ml. Data were collected over a wavelength range from 260 to 190 nm with 1 nm intervals at room temperature, three scans were averaged, and the baseline spetrum of solution buffer containing 10 mM HEPES (pH 7.5) and 150 mM NaCl was subtracted. pcDNA3.3-Myc-MKP7, pCMV5-3HA-JNK1 and pBOBI-HA-MKP7 were generated with standard molecular techniques. Mutants with amino acid substitution and truncation constructs were generated through PCR-based site-directed mutagenesis method using Pfu polymerase (Stratagene). The authenticities of all constructs were confirmed by sequencing (Invitrogen, China). HEK293T and HeLa cells (ATCC) were maintained in DMEM supplemented with 10% fetal bovine serum, 100 IU penicillin, 100 mg ml streptomycin at 37 °C in a humidified incubator containing 5% CO2. Polyethylenimine (Polysciences, #23966) at a final concentration of 10 μM was used to transfect HEK293T cells. Total DNA for each plate was adjusted to the same amount by adding relevant blank vector. Lentiviruses for infection were packaged in HEK293T cells after transfection using Lipofectamine 2000 (Invitrogen, 11668-027). At 30 h post transfection, medium was collected for further infection. Cells were lysed in a lysis buffer containing 20 mM Tris-HCl (pH 7.4), 150 mM NaCl, 0.5% NP-40, 1 mM EDTA, 2 mM Na3VO4, 25 mM NaF, 1 mM phenylmethanesulfonyl fluoride, 1 μg ml leupeptin and 1 μg ml aprotinin. Cell lysates were incubated with respective antibodies overnight at 4 °C. Protein aggregates resulting from the overnight incubation were removed by centrifugation, and protein A/G beads (Santa-Cruz Biotechnology, Dallas, TX, USA) were then added into the lysates and incubated for another 3 h. After spinning and washing for three times with the lysis buffer, the beads were mixed with 2 × SDS sample buffer, boiled and subjected to 15% SDS/PAGE. The samples were transferred to PVDF membranes (Millipore), and immunoblotted with indicated antibodies. Levels of total proteins and the levels of phosphorylation of proteins were analysed on separate gels. The uncropped blots are shown in Supplementary Fig. 7. Antibodies used in this study: mouse anti-HA (1:100 for immunoprecipitation; F-7) and anti-JNK1 (1:1,000 for immunoblotting; F-3) antibodies were purchased from Santa-Cruz Biotechnology. Anti-c-Myc Agarose Affinity Gel antibody produced in rabbit (1:200 for immunoprecipitation; A7470) was purchased from Sigma. Rabbit anti-HA-tag (1:1,000 for immunoblotting; #3724), anti-phospho-JNK-T183/Y185 (1:1,000 for immunoblotting; #4668) and mouse anti-Myc-tag (1:1,000 for immunoblotting; #2276) antibodies were purchased from Cell Signaling Technology. Etoposide (E1383) was purchased from Sigma. HeLa cells were infected with lentiviruses expressing MKP7 or its mutants. At 36 h post infection, cells were irradiated with 25 J m ultraviolet light and collected at 6 h after irradiation. Cells were then stained with the Annexin-V-APC/PI double-staining solution (BD Biosciences) and analysed with a flow cytometer (BD LSRFortessa). The percentages of apoptotic cells were quantified with FlowJo 7.6.1 software. Accession codes: The coordinates and structure factors have been deposited in the Protein Data Bank with accession codes 4YR8 for the JNK1–MKP7-CD structure. How to cite this article: Liu, X. et al. A conserved motif in JNK/p38-specific MAPK phosphatases as a determinant for JNK1 recognition and inactivation. Nat. Commun. 7:10879 doi: 10.1038/ncomms10879 (2016).
PMC4968113	Structural diversity in a human antibody germline library	To support antibody therapeutic development, the crystal structures of a set of 16 germline variants composed of 4 different kappa light chains paired with 4 different heavy chains have been determined. All four heavy chains of the antigen-binding fragments (Fabs) have the same complementarity-determining region (CDR) H3 that was reported in an earlier Fab structure. The structure analyses include comparisons of the overall structures, canonical structures of the CDRs and the VH:VL packing interactions. The CDR conformations for the most part are tightly clustered, especially for the ones with shorter lengths. The longer CDRs with tandem glycines or serines have more conformational diversity than the others. CDR H3, despite having the same amino acid sequence, exhibits the largest conformational diversity. About half of the structures have CDR H3 conformations similar to that of the parent; the others diverge significantly. One conclusion is that the CDR H3 conformations are influenced by both their amino acid sequence and their structural environment determined by the heavy and light chain pairing. The stem regions of 14 of the variant pairs are in the ‘kinked’ conformation, and only 2 are in the extended conformation. The packing of the VH and VL domains is consistent with our knowledge of antibody structure, and the tilt angles between these domains cover a range of 11 degrees. Two of 16 structures showed particularly large variations in the tilt angles when compared with the other pairings. The structures and their analyses provide a rich foundation for future antibody modeling and engineering efforts.At present, therapeutic antibodies are the largest class of biotherapeutic proteins that are in clinical trials. The use of monoclonal antibodies as therapeutics began in the early 1980s, and their composition has transitioned from murine antibodies to generally less immunogenic humanized and human antibodies. The technologies currently used to obtain human antibodies include transgenic mice containing human antibody repertoires, cloning directly from human B cells, and in vitro selection from antibody libraries using various display technologies. Once a candidate antibody is identified, protein engineering is usually required to produce a molecule with the right biophysical and functional properties. All engineering efforts are guided by our understanding of the atomic structures of antibodies. In such efforts, the crystal structure of the specific antibody may not be available, but modeling can be used to guide the engineering efforts. Today's antibody modeling approaches, which normally focus on the variable region, are being developed by the application of structural principles and insights that are evolving as our knowledge of antibody structures continues to expand. Our current structural knowledge of antibodies is based on a multitude of studies that used many techniques to gain insight into the functional and structural properties of this class of macromolecule. Five different antibody isotypes occur, IgG, IgD, IgE, IgA and IgM, and each isotype has a unique role in the adaptive immune system. IgG, IgD and IgE isotypes are composed of 2 heavy chains (HCs) and 2 light chains (LCs) linked through disulfide bonds, while IgA and IgM are double and quintuple versions of antibodies, respectively. Isotypes IgG, IgD and IgA each have 4 domains, one variable (V) and 3 constant (C) domains, while IgE and IgM each have the same 4 domains along with an additional C domain. These multimeric forms are linked with an additional J chain. The LCs that associate with the HCs are divided into 2 functionally indistinguishable classes, κ and λ. Both κ and λ polypeptide chains are composed of a single V domain and a single C domain. The heavy and light chains are composed of structural domains that have ∼110 amino acid residues. These domains have a common folding pattern often referred to as the “immunoglobulin fold,” formed by the packing together of 2 anti-parallel β-sheets. All immunoglobulin chains have an N-terminal V domain followed by 1 to 4 C domains, depending upon the chain type. In antibodies, the heavy and light chain V domains pack together forming the antigen combining site. This site, which interacts with the antigen (or target), is the focus of current antibody modeling efforts. This interaction site is composed of 6 complementarity-determining regions (CDRs) that were identified in early antibody amino acid sequence analyses to be hypervariable in nature, and thus are responsible for the sequence and structural diversity of our antibody repertoire. The sequence diversity of the CDR regions presents a substantial challenge to antibody modeling. However, an initial structural analysis of the combining sites of the small set of structures of immunoglobulin fragments available in the 1980s found that 5 of the 6 hypervariable loops or CDRs had canonical structures (a limited set of main-chain conformations). A CDR canonical structure is defined by its length and conserved residues located in the hypervariable loop and framework residues (V-region residues that are not part of the CDRs). Furthermore, studies of antibody sequences revealed that the total number of canonical structures are limited for each CDR, indicating possibly that antigen recognition may be affected by structural restrictions at the antigen-binding site. Later studies found that the CDR loop length is the primary determining factor of antigen-binding site topography because it is the primary factor for determining a canonical structure. Additional efforts have led to our current understanding that the LC CDRs L1, L2, and L3 have preferred sets of canonical structures based on length and amino acid sequence composition. This was also found to be the case for the H1 and H2 CDRs. Classification schemes for the canonical structures of these 5 CDRs have emerged and evolved as the number of depositions in the Protein Data Bank of Fab fragments of antibodies grow. Recently, a comprehensive CDR classification scheme was reported identifying 72 clusters of conformations observed in antibody structures. The knowledge and predictability of these CDR canonical structures have greatly advanced antibody modeling efforts. In contrast to CDRs L1, L2, L3, H1 and H2, no canonical structures have been observed for CDR H3, which is the most variable in length and amino acid sequence. Some clustering of conformations was observed for the shortest lengths; however, for the longer loops, only the portions nearest the framework (torso, stem or anchor region) were found to have defined conformations. In the torso region, 2 primary groups could be identified, which led to sequence-based rules that can predict with some degree of reliability the conformation of the stem region. The “kinked” or “bulged” conformation is the most prevalent, but an “extended” or “non-bulged” conformation is also, but less frequently, observed. The cataloging and development of the rules for predicting the conformation of the anchor region of CDR H3 continue to be refined, producing new insight into the CDR H3 conformations and new tools for antibody engineering. Current antibody modeling approaches take advantage of the most recent advances in homology modeling, the evolving understanding of the CDR canonical structures, the emerging rules for CDR H3 modeling and the growing body of antibody structural data available from the PDB. Recent antibody modeling assessments show continued improvement in the quality of the models being generated by a variety of modeling methods. Although antibody modeling is improving, the latest assessment revealed a number of challenges that need to be overcome to provide accurate 3-dimensional models of antibody V regions, including accuracies in the modeling of CDR H3. The need for improvement in this area was also highlighted in a recent study reporting an approach and results that may influence future antibody modeling efforts. One important finding of the antibody modeling assessments was that errors in the structural templates that are used as the basis for homology models can propagate into the final models, producing inaccuracies that may negatively influence the predictive nature of the V region model. To support antibody engineering and therapeutic development efforts, a phage library was designed and constructed based on a limited number of scaffolds built with frequently used human germ-line IGV and IGJ gene segments that encode antigen combining sites suitable for recognition of peptides and proteins. This Fab library is composed of 3 HC germlines, IGHV1-69 (H1-69), IGHV3-23 (H3-23) and IGHV5-51(H5-51), and 4 LC germlines (all κ), IGKV1-39 (L1-39), IGKV3-11 (L3-11), IGKV3-20 (L3-20) and IGKV4-1 (L4-1). Selection of these genes was based on the high frequency of their use and their cognate canonical structures that were found binding to peptides and proteins, as well as their ability to be expressed in bacteria and displayed on filamentous phage. The implementation of the library involves the diversification of the human germline genes to mimic that found in natural human libraries. The crystal structure determinations and structural analyses of all germline Fabs in the library described above along with the structures of a fourth HC germline, IGHV3-53 (H3-53), paired with the 4 LCs of the library have been carried out to support antibody therapeutic development. All 16 HCs of the Fabs have the same CDR H3 that was reported in an earlier Fab structure. This is the first systematic study of the same VH and VL structures in the context of different pairings. The structure analyses include comparisons of the overall structures, canonical structures of the L1, L2, L3, H1 and H2 CDRs, the structures of all CDR H3s, and the VH:VL packing interactions. The structures and their analyses provide a foundation for future antibody engineering and structure determination efforts. The crystal structures of a germline library composed of 16 Fabs generated by combining 4 HCs (H1-69, H3-23, H3-53 and H5-51) and 4 LCs (L1-39, L3-11, L3-20 and L4-1) have been determined. The Fab heavy and light chain sequences for the variants numbered according to Chothia are shown in Fig. S1. The four different HCs all have the same CDR H3 sequence, ARYDGIYGELDF. Crystallization of the 16 Fabs was previously reported. Three sets of the crystals were isomorphous with nearly identical unit cells (Table 1). These include (1) H3-23:L3-11 and H3-23:L4-1 in P212121, (2) H3-53:L1-39, H3-53:L3-11 and H3-53:L3-20 in P6522, and (3) H5-51:L1-39, H5-51:L3-11 and H5-51:L3-20 in P212121. Crystallization conditions for the 3 groups are also similar, but not identical (Table 1). Variations occur in the pH (buffer) and the additives, and, in group 3, PEG 3350 is the precipitant for one variants while ammonium sulfate is the precipitant for the other two. The similarity in the crystal forms is attributed in part to cross-seeding using the microseed matrix screening for groups 2 and 3.Table 1.Crystal data, X-ray data, and refinement statistics.FabH1-69:L1-39H1-69:L3-11H1-69:L3-20H1-69:L4-1PDB identifier5I155I165I175I18Crystal Data Crystallization Solution Buffer, pH0.1 M MES- pH 6.50.1 M MES pH 6.50.1 M MES, pH 6.50.1 M HEPES, pH 7.5 Precipitant5 M Na Formate25% PEG 33502.0 M Amm Sulfate10% PEG 8000 Additive 0.2 M Na Formate5% MPD8% EG Space GroupP3121C2P422P4212 Molecules/AU1221 Unit Cell a(Å)129.2212.0152.5120.0 b(Å)129.255.1152.5120.0 c(Å)91.880.3123.464.2 β(°)90.097.890.090.0 γ(°)120.090.090.090.0 Vm (Å/Da)4.672.443.772.39 Solvent Content (%)74506748X-Ray Data Resolution (Å)30-2.6 (2.7-2.6)30.0-1.9 (1.95-1.9)30.0-3.3 (3.4-3.3)30-1.9 (2.0-1.9) Measured Reflections136,745 (8,650)241,145 (16,580)237,504 (15,007)801,080 (19,309) Unique Reflections27,349 (1,730)71,932 (5,198)22,379 (1,590)35,965 (2,194) Completeness (%)99.3 (98.7)99.0 (97.3)99.5 (96.8)98.5 (82.8) Redundancy5.0 (5.0)3.4 (3.2)10.6 (9.4)22.3 (8.8) Rmerge0.048 (0.522)0.044 (0.245)0.086 (0.536)0.093 (0.231) < I/σ >21.2 (3.9)17.8 (4.7)25.5 (4.5)29.2 (8.1) B-factor (Å)60.533.261.019.6Refinement Resolution (Å)15-2.615-1.915-3.315-1.9 Number of Reflections26,23870,34621,19734,850 Number of All Atoms3,2246,9756,3983,695 Number of Waters24720399 R-factor (%)20.519.220.216.7 R-free (%)24.122.224.721.3RMSD Bond Lengths (Å)0.0060.0050.0050.008 Bond Angles (°)1.21.11.01.1 Mean B-factor (Å)65.334.480.120.0Ramachandran Plot (%) Outliers0.00.00.90.0 Favored92.396.993.196.91Abbreviations: Amm, ammonium;EG, ethylene glycol; PEG, polyethylene glycol.2Values for high-resolution shell are in parentheses.Table 1.(Continued) Crystal data, X-ray data, and refinement statistics.FabH3-23:L1-39H3-23:L3-11H3-23:L3-20H3-23:L4-1 PDB identifier5I195I1A5I1C5I1DCrystal Data Crystallization Solution Buffer, pHNo Buffer0.1 M Na Acetate, pH 4.50.1 M MES, pH 6.50.1 M HEPES, pH 7.5 Precipitant20% PEG 33502.0 M Amm Sulfate16% PEG 33502.0 M Amm Sulfate Additive0.2 M Li Citrate5% PEG 4000.2 M Amm Acetate2% PEG 400 Space GroupP41212P212121P6222P212121 Molecules/AU1212Unit Cell a(Å)96.660.9121.562.7 b(Å)96.6110.6121.5111.0 c(Å)105.4158.9160.4160.0 β(°)90909090 Vm (Å/Da)2.602.823.602.90 Solvent Content (%)53566657X-Ray Data Resolution (Å)30-2.8 (2.9-2.8)30-2.0 (2.1-2.0)30-2.25 (2.3-2.25)30-2.0 (2.1-2.0) Measured Reflections177,681 (12,072)351,312 (8,634)887,349 (59,919)873,523 (49,118) Unique Reflections12,678 (899)58,989 (2,870)32,572 (2,300)75,540 (5,343) Completeness (%)99.5 (97.4)80.9 (54.2)96.9 (94.8)99.7 (96.9) Redundancy14.0 (13.4)6.0 (3.0)27.2 (26.1)11.6 (9.2) Rmerge0.091 (0.594)0.066 (0.204)0.086 (0.478)0.094 (0.488) < I/σ >31.2 (5.1)20.4 (4.6)37.0 (10.4)21.6 (5.0) B-factor (Å)42.827.133.729.4Refinement Resolution (Å)15-2.815-2.015-2.2515-2.0 Number of Reflections11,97257,59931,41174,238 Number of All Atoms3,2346,9483,4727,210 Number of Waters0416222635 R-factor (%)23.920.522.021.6 R-free (%)31.525.526.625.1RMSD Bond Lengths (Å)0.0090.0100.0050.008 Bond Angles (°)1.31.31.01.1 Mean B-factor (Å)48.436.747.746.4Ramachandran Plot (%) Outliers0.00.00.00.0 Favored92.396.897.597.61Abbreviations: Amm, ammonium; PEG, polyethylene glycol.2Values for high-resolution shell are in parentheses.Table 1.(Continued) Crystal data, X-ray data, and refinement statistics.FabH3-53:L1-39H3-53:L3-11H3-53:L3-20H3-53:L4-1PDB indentifier5I1E5I1G5I1H5I1ICrystal Data Crystallization Solution Buffer, pHNo buffer0.1 M Na Acetate pH 4.50.1 M Na Acetate pH 4.50.1M MES, pH 6.5 Precipitant16% PEG 335025% PEG 335019% PEG 400017% PEG 3350 Additive0.2 M Amm Sulfate 5% Dioxane0.2 M Li2SO40.2 M Amm Sulfate0.2 M Na Formate, 5% MPD Space GroupP6522P6522P6522P31 Molecules/AU1111Unit Cell a(Å)89.488.189.468.1 b(Å)89.488.189.468.1 c(Å)212.4219.6211.795.6 β(°)90909090 γ(°)120120120120 Vm (Å/Da)2.572.642.572.64 Solvent Content (%)52535253X-Ray Data Resolution (Å)30-2.7 (2.8-2.7)30-2.3 (2.4-2.3)30-2.2 (2.3-2.0)30-2.5 (2.6-2.5) Measured Reflections297,367 (19,369)333,739 (8,008)381,125 (1,591)137,992 (9,883) Unique Reflections14,402 (1,003)21,683 (1,135)24,323 (964)16,727 (1,227) Completeness (%)99.6 (96.8)93.8 (68.4)95.3 (52.0)98.6 (98.1) Redundancy20.6 (19.3)15.4 (7.1)15.7 (1.7)8.2 (8.1) Rmerge0.095 (0.451)0.057 (0.324)0.062 (0.406)0.047 (0.445) < I/σ >38.3 (8.1)36.7 (5.5)36.2 (1.6)31.6 (5.6) B-factor (Å)33.237.333.754.8Refinement Resolution (Å)15-2.715-2.315-2.215-2.5 Number of Reflections13,58320,25524,96215,811 Number of All Atoms3,3353,2713,2983,239 Number of Waters88707121 R-factor (%)19.129.822.825.0 R-free (%)26.438.326.633.7RMSD Bond Lengths (Å)0.0080.0050.0050.006 Bond Angles (°)1.21.01.01.1 Mean B-factor (Å)49.146.351.788.9Ramachandran Plot (%) Outliers0.20.20.21.2 Favored96.797.196.590.91Abbreviations: Amm, ammonium; PEG, polyethylene glycol.2Values for high-resolution shell are in parentheses.Table 1.(Continued) Crystal data, X-ray data, and refinement statistics.FabH5-51:L1-39H5-51:L3-11H5-51:L3-20H5-51:L4-1 PDB identifier4KMT5I1J5I1K5I1LCrystal Data Crystallization Solution Buffer, pH0.1 M CHES, pH 9.50.1 M Tris, pH 8.50.1 M CHES, pH 9.50.1 M Tris, pH 8.5 Precipitant1.8 M Amm Sulfate25% PEG 33501.0 M Amm Sulfate24% PEG 3350 Additive5% dioxane0.2 M MgCl2 0.2 M Amm Sulfate Space GroupP212121P212121P212121P21 Molecules/AU1112Unit Cell a(Å)63.764.163.8106.0 b(Å)73.873.874.138.0 c(Å)103.1103.0103.0112.3 β(°)909090100.4 Vm (Å/Da)2.532.562.542.28 Solvent Content (%)51525146X-Ray Data Resolution (Å)30-2.1 (2.2-2.1)30-2.5 (2.6-2.5)30-1.65 (1.7-1.65)30-1.95 (2.0-1.95) Measured Reflections131,839 (6,655)120,521 (7,988)246,750 (4,142)320,324 (12,119) Unique Reflections27,026 (1,885)17,286 (1,236)53,058 (2,141)61,554 (3,243) Completeness (%)93.6 (89.8)99.7 (97.3)89.8 (49.8)94.4 (67.1) Redundancy4.9 (3.5)7.0 (6.5)4.7 (1.9)5.2 (3.7) Rmerge0.079 (0.278)0.080 (0.281)0.034 (0.131)0.060 (0.395) < I/σ >16.8 (5.7)21.1(6.9)27.5 (5.8)19.7 (3.1) B-factor (Å)26.027.021.631.4Refinement Resolution (Å)15-2.115-2.515-1.6515-1.95 Number of Reflections25,85716,32851,88260,181 Number of All Atoms3,6763,4543,8147,175 Number of Waters302196527445 R-factor (%)17.117.717.219.4 R-free (%)22.025.819.725.8RMSD Bond Lengths (Å)0.0060.0090.0050.009 Bond Angles (°)1.01.31.31.3 Mean B-factor (Å)25.238.220.019.5Ramachandran Plot (%) Outliers0.00.00.00.0 Favored98.497.998.198.01Abbreviations: Amm, ammonium; PEG, polyethylene glycol.2Values for high-resolution shell are in parentheses. Crystal data, X-ray data, and refinement statistics. Abbreviations: Amm, ammonium;EG, ethylene glycol; PEG, polyethylene glycol. Values for high-resolution shell are in parentheses. (Continued) Crystal data, X-ray data, and refinement statistics. Abbreviations: Amm, ammonium; PEG, polyethylene glycol. Values for high-resolution shell are in parentheses. (Continued) Crystal data, X-ray data, and refinement statistics. Abbreviations: Amm, ammonium; PEG, polyethylene glycol. Values for high-resolution shell are in parentheses. (Continued) Crystal data, X-ray data, and refinement statistics. Abbreviations: Amm, ammonium; PEG, polyethylene glycol. Values for high-resolution shell are in parentheses. The crystal structures of the 16 Fabs have been determined at resolutions ranging from 3.3 Å to 1.65 Å (Table 1). The number of Fab molecules in the crystallographic asymmetric unit varies from 1 (for 12 Fabs) to 2 (for 4 Fabs). Overall the structures are fairly complete, and, as can be expected, the models for the higher resolution structures are more complete than those for the lower resolution structures (Table S1). Invariably, the HCs have more disorder than the LCs. For the LC, the disorder is observed at 2 of the C-terminal residues with few exceptions. Apart from the C-terminus, only a few surface residues in LC are disordered. The HCs feature the largest number of disordered residues, with the lower resolution structures having the most. The C-terminal residues including the 6xHis tags are disordered in all 16 structures. In addition to these, 2 primary disordered stretches of residues are observed in a number of structures (Table S1). One involves the loop connecting the first 2 β-strands of the constant domain (in all Fabs except H3-23:L1-39, H3-23:L3-11 and H3-53:L1-39). The other is located in CDR H3 (in H5-51:L3-11, H5-51:L3-20 and in one of 2 copies of H3-23:L4-1). CDR H1 and CDR H2 also show some degree of disorder, but to a lesser extent. Several CDR definitions have evolved over decades of antibody research. Depending on the focus of the study, the CDR boundaries differ slightly between various definitions. In this work, we use the CDR definition of North et al., which is similar to that of Martin with the following exceptions: 1) CDRs H1 and H3 begin immediately after the Cys; and 2) CDR L2 includes an additional residue at the N-terminal side, typically Tyr. The four HCs feature CDR H1 of the same length, and their sequences are highly similar (Table 2). The CDR H1 backbone conformations for all variants for each of the HCs are shown in Fig. 1. Three of the HCs, H3-23, H3-53 and H5-51, have the same canonical structure, H1-13-1, and the backbone conformations are tightly clustered for each set of Fab structures as reflected in the rmsd values (Fig. 1B-D). Some deviation is observed for H3-53, mostly due to H3-53:L4-1, which exhibits a significant degree of disorder in CDR H1. The electron density for the backbone is weak and discontinuous, and completely missing for several side chains. Figure 1.The superposition of CDR H1 backbones for all HC:LC pairs with heavy chains: (A) H1-69, (B) H3-23, (C) H3-53 and (D) H5-51. Table 2.Canonical structures.PairsPDBCDR H1CDR H2CDR H3H1-69 KASGGTFSSYAISGIIPIFGTANARYDGIYGELDFH1-69:L1-395I15H1-13-4H2-10-1H3-12-1H1-69:L3-115I16H1-13-1/H1-13-1H2-10-1/H2-10-1H3-12-1/H3-12-1H1-69:L3-205I17H1-13-3/H1-13-6H2-10-1/NAH3-12-1/H3-12-1H1-69:L4-15I18H1-13-10H2-10-1H3-12-1H3-23 AASGFTFSSYAMSAISGSGGSTYAKYDGIYDGIYGELDFH3-23:L1-395I19H1-13-1H2-10-2H3-12-1H3-23:L3-115I1AH1-13-1/H1-13-1H2-10-2/H2-10-2H3-12-1/H3-12-1H3-23:L3-205I1CH1-13-1H2-10-2H3-12-1H3-23:L4-15I1DH1-13-1/H1-13-1H2-10-2/H2-10-2H3-12-1/NAH3-53 AASGFTVSSNYMSVIYSGGSTYARYDGIYGELDFH3-53:L1-395I1EH1-13-1H2-9-3H3-12-1H3-53:L3-115I1GH1-13-1H2-9-3H3-12-1H3-53:L3-205I1HH1-13-1H2-9-3H3-12-1H3-53:L4-15I1IH1-13-1H2-9-3NAH5-51 KGSGYSFTSYWIGIIYPGDSDTRARYDGIYGELDFH5-51:L1-394KMTH1-13-1H2-10-1H3-12-1H5-51:L3-115I1JH1-13-1H2-10-1NAH5-51:L3-205I1KH1-13-1H2-10-1NAH5-51:L4-15I1LH1-13-1/H1-13-1H2-10-1/H2-10-1H3-12-1/H3-12-1 CDR L1CDR L2CDR L3L1-39 RASQSISSYLNYAASSLQSQQSYSTPLTH1-69:L1-395I15L1-11-1L2-8-1L3-9-cis7-1H3-23:L1-395I19L1-11-1L2-8-1L3-9-cis7-1H3-53:L1-395I1EL1-11-1L2-8-1L3-9-cis7-1H5-51:L1-394KMTL1-11-1L2-8-1L3-9-cis7-1L3-11 RASQSVSSYLAYDASNRATQQRSNWPLTH1-69:L3-115I16L1-11-1/L1-11-1L2-8-1/L2-8-1L3-9-cis7-1/L3-9-cis7-1H3-23:L3-115I1AL1-11-1/L1-11-1L2-8-1/L2-8-1L3-9-cis7-1/L3-9-cis7-1H3-53:L3-115I1GL1-11-1L2-8-1L3-9-cis7-1H5-51:L3-115I1JL1-11-1L2-8-1L3-9-cis7-1L3-20 RASQSVSSSYLAYGASSRATQQYGSSPLTH1-69:L3-205I17L1-12-2/L1-12-1L2-8-1/L2-8-1L3-9-cis7-1/L3-9-cis7-1H3-23:L3-205I1CL1-12-2L2-8-1L3-9-cis7-1H3-53:L3-205I1HL1-12-1L2-8-1L3-9-cis7-1H5-51:L3-205I1KL1-12-1L2-8-1L3-9-cis7-1L4-1 KSSQSVLYSSNNKNYLAYWASTRESQQYYSTPLTH1-69:L4-15I18L1-17-1L2-8-1L3-9-cis7-1H3-23:L4-15I1DL1-17-1/L1-17-1L2-8-1/L2-8-1L3-9-cis7-1/L3-9-cis7-1H3-53:L4-15I1IL1-17-1L2-8-1L3-9-cis7-1H5-51:L4-15I1LL1-17-1/L1-17-1L2-8-1/L2-8-1L3-9-cis7-1/L3-9-cis7-11CDRs are defined using the Dunbrack convention . Assignments for 2 copies of the Fab in the asymmetric unit are given for 5 structures. No assignment (NA) for CDRs with missing residues. The superposition of CDR H1 backbones for all HC:LC pairs with heavy chains: (A) H1-69, (B) H3-23, (C) H3-53 and (D) H5-51. Canonical structures. CDRs are defined using the Dunbrack convention . Assignments for 2 copies of the Fab in the asymmetric unit are given for 5 structures. No assignment (NA) for CDRs with missing residues. The CDR H1 structures with H1-69 shown in Fig. 1A are quite variable, both for the structures with different LCs and for the copies of the same Fab in the asymmetric unit, H1-69:L3-11 and H1-69:L3-20. In total, 6 independent Fab structures produce 5 different canonical structures, namely H1-13-1, H1-13-3, H1-13-4, H1-13-6 and H1-13-10. A major difference of H1-69 from the other germlines in the experimental data set is the presence of Gly instead of Phe or Tyr at position 27 (residue 5 of 13 in CDR H1). Glycine introduces the possibility of a higher degree of conformational flexibility that undoubtedly translates to the differences observed, and contributes to the elevated thermal parameters for the atoms in the amino acid residues in this region. The canonical structures of CDR H2 have fairly consistent conformations (Table 2, Fig. 2). Each of the 4 HCs adopts only one canonical structure regardless of the pairing LC. Germlines H1-69 and H5-51 have the same canonical structure assignment H2-10-1, H3-23 has H2-10-2, and H3-53 has H2-9-3. The conformations for all of these CDR H2s are tightly clustered (Fig. 2). In one case, in the second Fab of H1-69:L3-20, CDR H2 is partially disordered (Δ55-60). Figure 2.The superposition of CDR H2 backbones for all HC:LC pairs with heavy chains: (A) H1-69, (B) H3-23, (C) H3-53 and (D) H5-51. The superposition of CDR H2 backbones for all HC:LC pairs with heavy chains: (A) H1-69, (B) H3-23, (C) H3-53 and (D) H5-51. Although three of the germlines have CDR H2 of the same length, 10 residues, they adopt 2 distinctively different conformations depending mostly on the residue at position 71 from the so-called CDR H4. Arg71 in H3-23 fills the space between CDRs H2 and H4, and defines the conformation of the tip of CDR H2 so that residue 54 points away from the antigen binding site. Germlines H1-69 and H5-51 are unique in the human repertoire in having an Ala at position 71 that leaves enough space for H-Pro52a to pack deeper against CDR H4 so that the following residues 53 and 54 point toward the putative antigen. Conformations of CDR H2 in H1-69 and H5-51, both of which have canonical structure H2-10-1, show little deviation within each set of 4 structures. However, there is a significant shift of the CDR as a rigid body when the 2 sets are superimposed. Most likely this is the result of interaction of CDR H2 with CDR H1, namely with the residue at position 33 (residue 11 of 13 in CDR H1). Germline H1-69 has Ala at position 33 whereas in H5-51 position 33 is occupied by a bulky Trp, which stacks against H-Tyr52 and drives CDR H2 away from the center. The four LC CDRs L1 feature 3 different lengths (11, 12 and 17 residues) having a total of 4 different canonical structure assignments. Of these LCs, L1-39 and L3-11 have the same canonical structure, L1-11-1, and superimpose very well (Fig. 3A, B). For the remaining 2, L3-20 has 2 different assignments, L1-12-1 and L1-12-2, while L4-1 has a single assignment, L1-17-1. Figure 3.The superposition of CDR L1 backbones for all HC:LC pairs with light chains: (A) L1-39, (B) L3-11, (C) L3-20 and (D) L4-1. The superposition of CDR L1 backbones for all HC:LC pairs with light chains: (A) L1-39, (B) L3-11, (C) L3-20 and (D) L4-1. L4-1 has the longest CDR L1, composed of 17 amino acid residues (Fig. 3D). Despite this, the conformations are tightly clustered (rmsd is 0.20 Å). The backbone conformations of the stem regions superimpose well. Some changes in conformation occur between residues 30a and 30f (residues 8 and 13 of 17 in CDR L1). This is the tip of the loop region, which appears to have similar conformations that fan out the structures because of the slight differences in torsion angles in the backbone near Tyr30a and Lys30f. L3-20 is the most variable in CDR L1 among the 4 germlines as indicated by an rmsd of 0.54 Å (Fig. 3C). Two structures, H3-53:L3-20 and H5-51:L3-20 are assigned to canonical structure L1-12-1 with virtually identical backbone conformations. The third structure, H3-23:L3-20, has CDR L1 as L1-12-2, which deviates from L1-12-1 at residues 29-32, i.e., at the site of insertion with respect to the 11-residue CDR. The fourth member of the set, H1-69:L3-20, was crystallized with 2 Fabs in the asymmetric unit. The conformation of CDR L1 in these 2 Fabs is slightly different, and both conformations fall somewhere between L1-12-1 and L1-12-2. This reflects the lack of accuracy in the structure due to low resolution of the X-ray data (3.3 Å). All four LCs have CDR L2 of the same length and canonical structure, L2-8-1 (Table 2). The CDR L2 conformations for each of the LCs paired with the 4 HCs are clustered more tightly than any of the other CDRs (rmsd values are in the range 0.09-0.16 Å), and all 4 sets have virtually the same conformation despite the sequence diversity of the loop. No significant conformation outliers are observed (Fig. 4). Figure 4.The superposition of CDR L2 backbones for all HC:LC pairs with light chains: (A) L1-39, (B) L3-11, (C) L3-20 and (D) L4-1. The superposition of CDR L2 backbones for all HC:LC pairs with light chains: (A) L1-39, (B) L3-11, (C) L3-20 and (D) L4-1. As with CDR L2, all 4 LCs have CDR L3 of the same length and canonical structure, L3-9-cis7-1 (Table 2). The conformations of CDR L3 for L1-39, L3-11, and particularly for L320, are not as tightly clustered as those of L4-1 (Fig. 5). The slight conformational variability occurs in the region of amino acid residues 90-92, which is in contact with CDR H3. Figure 5.The superposition of CDR L3 backbones for all HC:LC pairs with light chains: (A) L1-39, (B) L3-11, (C) L3-20 and (D) L4-1. The superposition of CDR L3 backbones for all HC:LC pairs with light chains: (A) L1-39, (B) L3-11, (C) L3-20 and (D) L4-1. As mentioned earlier, all 16 Fabs have the same CDR H3, for which the amino acid sequence is derived from the anti-CCL2 antibody CNTO 888. The loop and the 2 β-strands of the CDR H3 in this ‘parent’ structure are stabilized by H-bonds between the carbonyl oxygen and peptide nitrogen atoms in the 2 strands. An interesting feature of these CDR H3 structures is the presence of a water molecule that interacts with the peptide nitrogens and carbonyl oxygens near the bridging loop connecting the 2 β-strands. This water is present in both the bound (4DN4) and unbound (4DN3) forms of CNTO 888. The stem region of CDR H3 in the parental Fab is in a ‘kinked’ conformation, in which the indole nitrogen of Trp103 forms a hydrogen bond with the carbonyl oxygen of Leu100b. The carboxyl group of Asp101 forms a salt bridge with Arg94. These interactions are illustrated in Fig. S2. Despite having the same amino acid sequence in all variants, CDR H3 has the highest degree of structural diversity and disorder of all of the CDRs in the experimental set. Three of the 21 Fab structures (including multiple copies in the asymmetric unit), H5-51:L3-11, H551:L3-20 and H3-23:L4-1 (one of the 2 Fabs), have missing (disordered) residues at the apex of the CDR loop. Another four of the Fabs, H3-23:L1-39, H3-53:L1-39, H3-53:L3-11 and H3-53:L4-1 have missing side-chain atoms. The variations in CDR H3 conformation are illustrated in Fig. 6 for the 18 Fab structures that have ordered backbone atoms. Figure 6.Ribbon representations of (A) the superposition of all CDR H3s of the structures with complete backbone traces. (B) The CDR H3s rotated 90° about the y axis of the page. The structure of each CDR H3 is represented with a different color. Ribbon representations of (A) the superposition of all CDR H3s of the structures with complete backbone traces. (B) The CDR H3s rotated 90° about the y axis of the page. The structure of each CDR H3 is represented with a different color. In 10 of the 18 Fab structures, H1-69:L1-39, H1-69:L3-11 (2 Fabs), H1-69:L4-1, H3-23:L3-11 (2 Fabs), H3-23:L3-20, H3-53:L3-11, H3-53:L3-20 and H5-51:L1-39, the CDRs have similar conformations to that found in 4DN3. The bases of these structures have the ‘kinked’ conformation with the H-bond between Trp103 and Leu100b. A representative CDR H3 structure for H1-69:L1-39 illustrating this is shown in Fig. 7A. The largest backbone conformational deviation for the set is at Tyr99, where the C=O is rotated by 90° relative to that observed in 4DN3. Also, it is worth noting that only one of these structures, H1-69:L4-1, has the conserved water molecule in CDR H3 observed in the 4DN3 and 4DN4 structures. In fact, it is the only Fab in the set that has a water molecule present at this site. The CDR H3 for this structure is shown in Fig. S3. Figure 7.A comparison of representatives of the “kinked” and “extended” structures. (A) The “kinked” CDR H3 of H1-69:L3-11 with purple carbon atoms and yellow dashed lines connecting the H-bond pairs for Leu100b O and Trp103 NE1, Arg94 NE and Asp101 OD1, and Arg94 NH2 and Asp101 OD2. (B) The “extended” CDR H3 of H1-69:L3-20 with green carbon atoms and yellow dashed lines connecting the H-bond pairs for Asp101 OD1 and OD2 and Trp103 NE1. A comparison of representatives of the “kinked” and “extended” structures. (A) The “kinked” CDR H3 of H1-69:L3-11 with purple carbon atoms and yellow dashed lines connecting the H-bond pairs for Leu100b O and Trp103 NE1, Arg94 NE and Asp101 OD1, and Arg94 NH2 and Asp101 OD2. (B) The “extended” CDR H3 of H1-69:L3-20 with green carbon atoms and yellow dashed lines connecting the H-bond pairs for Asp101 OD1 and OD2 and Trp103 NE1. The remaining 8 Fabs can be grouped into 5 different conformational classes. Three of the Fabs, H3-23:L1-39, H3-23:L4-1 and H3-53:L1-39, have distinctive conformations. The stem regions in these 3 cases are in the ‘kinked’ conformation consistent with that observed for 4DN3. The five remaining Fabs, H5-51:L4-1 (2 copies), H1-69:L3-20 (2 copies) and H3-53:L4-1, have 3 different CDR H3 conformations (Fig. S4). The stem regions of CDR H3 for the H5-51:L4-1 Fabs are in the ‘kinked’ conformation while, surprisingly, those of the H1-69:L3-20 pair and H3-53:L4-1 are in the ‘extended’ conformation (Fig. 7B). The VH and VL domains have a β-sandwich structure (also often referred as a Greek key motif) and each is composed of a 4-stranded and a 5-stranded antiparallel β-sheets. The two domains pack together such that the 5-stranded β-sheets, which have hydrophobic surfaces, interact with each other bringing the CDRs from both the VH and VL domains into close proximity. The domain packing of the variants was assessed by computing the domain interface interactions, the VH:VL tilt angles, the buried surface area and surface complementarity. The results of these analyses are shown in Tables 3, 4 and S2. The VH:VL interface is pseudosymmetric, and involves 2 stretches of the polypeptide chain from each domain, namely CDR3 and the framework region between CDRs 1 and 2. These stretches form antiparallel β-hairpins within the internal 5-stranded β-sheet. There are a few principal inter-domain interactions that are conserved not only in the experimental set of 16 Fabs, but in all human antibodies. They include: 1) a bidentate hydrogen bond between L-Gln38 and H-Gln39; 2) H-Leu45 in a hydrophobic pocket between L-Phe98, L-Tyr87 and L-Pro44; 3) L-Pro44 stacked against H-Trp103; and 4) L-Ala43 opposite the face of H-Tyr91 (Fig. 8). With the exception of L-Ala43, all other residues are conserved in human germlines. Position 43 may be alternatively occupied by Ser, Val or Pro (as in L4-1), but the hydrophobic interaction with H-Tyr91 is preserved. These core interactions provide enough stability to the VH:VL dimer so that additional VH-VL contacts can tolerate amino acid sequence variations in CDRs H3 and L3 that form part of the VH:VL interface. Figure 8.The conserved VH:VL interactions as viewed along the VH/VL axis. The VH residues are in blue, the VL residues are in orange. The conserved VH:VL interactions as viewed along the VH/VL axis. The VH residues are in blue, the VL residues are in orange. In total, about 20 residues are involved in the VH:VL interactions on each side (Fig. S5). Half of them are in the framework regions and those residues (except residue 61 in HC, which is actually in CDR2 in Kabat's definition) are conserved in the set of 16 Fabs. The side chain conformations of these conserved residues are also highly similar. One notable exception is H-Trp47, which exhibits 2 conformations of the indole ring. In most of the structures, it has the χ2 angle of ∼80°, while the ring is flipped over (χ2 = −100°) in H5-51:L3:11 and H5-51:L3-20. Interestingly, these are the only 2 structures with residues missing in CDR H3 because of disorder, although both structures are determined at high resolution and the rest of the structure is well defined. Apparently, residues flanking CDR H3 in the 2 VH:VL pairings are inconsistent with any stable conformation of CDR H3, which translates into a less restricted conformational space for some of them, including H-Trp47. The relative orientation of VH and VL has been measured in a number of different ways. Presented here are the results of 2 different approaches for determining the orientation of one domain relative to the other. The first approach uses ABangles, the results of which are shown in Table S2. The four LCs all are classified as Type A because they have a proline at position 44, and the results for each orientation parameter are within the range of values of this type reported by Dunbar and co-workers. In fact, the parameter values for the set of 16 Fabs are in the middle of the distribution observed for 351 non-redundant antibody structures determined at 3.0 Å resolution or better. The only exception is HC1, which is shifted toward smaller angles with the mean value of 70.8° as compared to the distribution centered at 72° for the entire PDB. This probably reflects the invariance of CDR H3 in the current set as opposed to the CDR H3 diversity in the PDB. The second approach used for comparing tilt angles involved computing the difference in the tilt angles between all pairs of structures. For structures with 2 copies of the Fab in the asymmetric unit, only one structure was used. The differences between independent Fabs in the same structure are 4.9° for H1-69:L3-20, 1.6° for H1-69:L3-11, 1.4° for H3-23:L4-1, 3.3° for H3-23:L3-11, and 2.5° for H5-51:L4-1. With the exception of H1-69:L3-20, the angles are within the range of 2-3° as are observed in the identical structures in the PDB. In H1-69:L3-20, one of the Fabs is substantially disordered so that part of CDR H2 (the outer β-strand, residues 55-60) is completely missing. This kind of disorder may compromise the integrity of the VH domain and its interaction with the VL. Indeed, this Fab has the largest twist angle HC2 within the experimental set that exceeds the mean value by 2.5 standard deviations (Table S2). The differences in the tilt angle are shown for all pairs of V regions in Table 3. They range from 0.6° to 11.0°. The smallest differences in the tilt angle are between the Fabs in isomorphous crystal forms. The largest deviations in the tilt angle, up to 11.0°, are found for 2 structures, H1-69:L3-20 and H3-23:L3-20, that stand out from the other Fabs. One of the 2 structures, H1-69:L3-20, has its CDR H3 in the ‘extended’ conformation; the other structure has it in the ‘kinked’ conformation. Two examples illustrating large (10.5°) and small (1.6°) differences in the tilt angles are shown in Fig. 9. Figure 9.An illustration of the difference in tilt angle for 2 pairs of variants by the superposition of the VH domains of (A) H1-69:L3-20 on that of H5-51:L1-39 (the VL domain is off by a rigid-body roatation of 10.5°) and (B) H1-69:L4-1 on that of H5-51:L1-39 (the VL domain is off by a rigid-body roatation of 1.6°). Table 3.Differences in VH:VL tilt angles. H1-69:L1-39H1-69:L3-11H1-69:L3-20H1-69:L4-1H3-23:L1-39H3-23:L3-11H3-23:L3-20H3-23:L4-1H3-53:L1-39H3-53:L3-11H3-53:L3-20H3-53:L4-1H5-51:L1-39H5-51:L3-11H5-51:L3-20H5-51:L4-1H1-69:L1-3902.18.91.14.23.09.51.53.33.63.11.61.82.92.45.2H1-69:L3-11 07.32.92.52.08.41.32.62.93.21.83.94.64.45.0H1-69:L3-20 09.25.08.77.47.68.98.69.47.910.510.111.09.7H1-69:L4-1 04.63.910.11.84.44.74.12.31.62.52.36.2H3-23:L1-39 04.08.02.84.84.85.43.66.06.26.66.6H3-23:L3-11 09.33.02.12.93.33.34.65.85.05.2H3-23:L3-20 08.97.97.07.67.910.59.710.76.2H3-23:L4-1 03.63.83.71.53.23.73.85.6H3-53:L1-39 01.01.62.94.65.34.83.1H3-53:L3-11 01.32.94.85.25.02.3H3-53:L3-20 02.53.84.23.92.2H3-53:L4-1 02.93.03.34.2H5-51:L1-39 01.90.65.8H5-51:L3-11 01.95.7H5-51:L3-20 05.8H5-51:L4-1 0 An illustration of the difference in tilt angle for 2 pairs of variants by the superposition of the VH domains of (A) H1-69:L3-20 on that of H5-51:L1-39 (the VL domain is off by a rigid-body roatation of 10.5°) and (B) H1-69:L4-1 on that of H5-51:L1-39 (the VL domain is off by a rigid-body roatation of 1.6°). Differences in VH:VL tilt angles. The results of the PISA contact surface calculation and surface complementarity calculation are shown in Table 4. The interface areas are calculated as the average of the VH and VL contact surfaces. Six of the 16 structures have CDR H3 side chains or complete residues missing, and therefore their interfaces are much smaller than in the other 10 structures with complete CDRs (the results are provided for all Fabs for completeness). Among the complete structures, the interface areas range from 684 to 836 Å. Interestingly, the 2 structures that have the largest tilt angle differences with the other variants, H3-23:L3-20 and H1-69:L3-20, have the smallest VH:VL interfaces, 684 and 725 Å, respectively. H3-23:L3-20 is also unique in that it has the lowest value (0.676) of surface complementarity. Table 4.VH:VL surface areas and surface complementarity.Chain PairsPDBContact surfaceVH (Å)Contact surfaceVL (Å)Interface(Å)Surface complementarityH1-69:L1-395I157277717490.743H1-69:L3-115I168028708360.762H1-69:L3-205I177137367250.723H1-69:L4-15I187297367330.734H3-23:L1-395I197958178060.722H3-23:L3-115I1A8228348280.725H3-23:L3-205I1C6706986840.676H3-23:L4-15I1D7437707570.708H3-53:L1-395I1E6987197090.712H3-53:L3-115I1G7477587530.690H3-53:L3-205I1H7437357390.687H3-53:L4-15I1I6896936910.711H5-51:L1-394KMT7618087850.728H5-51:L3-115I1J6487146810.717H5-51:L3-205I1K6226436330.740H5-51:L4-15I1L7907927910.7041Some side chain atoms in CDR H3 are missing.2Residues in CDR H3 are missing: YGE in H5-51:L3-11, GIY in H5-51:L3-20. VH:VL surface areas and surface complementarity. Some side chain atoms in CDR H3 are missing. Residues in CDR H3 are missing: YGE in H5-51:L3-11, GIY in H5-51:L3-20. Melting temperatures (Tm) were measured for all Fabs using differential scanning calorimetry (Table 5). It appears that for each given LC, the Fabs with germlines H1-69 and H3-23 are substantially more stable than those with germlines H3-53 and H5-51. In addition, L1-39 provides a much higher degree of stabilization than the other 3 LC germlines when combined with any of the HCs. As a result, the Tm for pairs H1-69:L1-39 and H3-23:L1-39 is 12-13° higher than for pairs H3-53:L3-20, H3-53:L4-1, H5-51:L3-20 and H5-51:L4-1. Table 5.Melting temperatures for the 16 Fabs. L3-20L4-1L3-11L1-39HC averageH1-6973.674.875.680.376.1H3-2374.875.24.881.576.6H3-5368.468.071.573.970.5H5-5168.468.471.977.071.4LC average71.371.673.578.2 1Colors: blue (Tm < 70°C), green (70°C < Tm < 73°C), yellow (73°C < Tm < 78°C), orange (Tm > 78°C). Melting temperatures for the 16 Fabs. Colors: blue (Tm < 70°C), green (70°C < Tm < 73°C), yellow (73°C < Tm < 78°C), orange (Tm > 78°C). These findings correlate well with the degree of conformational disorder observed in the crystal structures. Parts of CDR H3 main chain are completely disordered, and were not modeled in Fabs H5-51:L3-20 and H5-51:L3-11 that have the lowest Tms in the set. No electron density is observed for a number of side chains in CDRs H3 and L3 in all Fabs with germline H3-53, which indicates loose packing of the variable domains. All those molecules are relatively unstable, as is reflected in their low Tms. This is the first report of a systematic structural investigation of a phage germline library. The 16 Fab structures offer a unique look at all pairings of 4 different HCs (H1-69, H3-23, H3-53, and H5-51) and 4 different LCs (L1-39, L3-11, L3-20 and L4-1), all with the same CDR H3. The structural data set taken as a whole provides insight into how the backbone conformations of the CDRs of a specific heavy or light chain vary when it is paired with 4 different light or heavy chains, respectively. A large variability in the CDR conformations for the sets of HCs and LCs is observed. In some cases the CDR conformations for all members of a set are virtually identical, for others subtle changes occur in a few members of a set, and in some cases larger deviations are observed within a set. The five variants that crystallized with 2 copies of the Fab in the asymmetric unit serve somewhat as controls for the influence of crystal packing on the conformations of the CDRs. In four of the 5 structures the CDR conformations are consistent. In only one case, that of H1-69:L3-20 (the lowest resolution structure), do we see differences in the conformations of the 2 copies of CDRs H1 and L1. This variability is likely a result of 2 factors, crystal packing interactions and internal instability of the variable domain. For the CDRs with canonical structures, the largest changes in conformation occur for CDR H1 of H1-69 and H3-53. The other 2 HCs, H3-23 and H5-51, have canonical structures that are remarkably well conserved (Fig. 1). Of the 4 HCs, H1-69 has the greatest number of canonical structure assignments (Table 2). H1-69 is unique in having a pair of glycine residues at positions 26 and 27, which provide more conformational freedom in CDR H1. Besides IGHV1-69, only the germlines of the VH4 family possess double glycines in CDR H1, and it will be interesting to see if they are also conformationally unstable. Having all 16 VH:VL pairs with the same CDR H3 provides some insights into why molecular modeling efforts of CDR H3 have proven so difficult. As mentioned in the Results section, this data set is composed of 21 Fabs, since 5 of the 16 variants have 2 Fab copies in the asymmetric unit. For the 18 Fabs with complete backbone atoms for CDR H3, 10 have conformations similar to that of the parent, while the others have significantly different conformations (Fig. 6). Thus, it is likely that the CDR H3 conformation is dependent upon 2 dominating factors: 1) amino acid sequence; and 2) VH and VL context. More than half of the variants retain the conformation of the parent despite having differences in the VH:VL pairing. This subset includes 2 structures with 2 copies of the Fab in the asymmetric unit, all of which are nearly identical in conformation. This provides an internal control showing a consistency in the conformations. The remaining 8 structures exhibit “non-parental” conformations, indicating that the VH and VL context can also be a dominating factor influencing CDR H3. Importantly, there are 5 distinctive conformations in this subset. This subset also has 2 structures with 2 Fab copies in the asymmetric unit. Each pair has nearly identical conformations providing an internal check on the consistency of the conformations. Interestingly, as described earlier, these 2 pairs differ in the stem regions with the H1-69:L3-20 pair in the ‘extended’ conformation and H5-51:L4-1 pair in the ‘kinked’ conformation. The conformations are different from each other, as well as from the parent. The CDR H3 conformational analysis shows that, for each set of variants of one HC paired with the 4 different LCs, both “parental” and “non-parental” conformations are observed. The same variability is observed for the sets of variants composed of one LC paired with each of the 4 HCs. Thus, no patterns of conformational preference for a particular HC or LC emerge to shed any direct light on what drives the conformational differences. This finding supports the hypothesis of Weitzner et al. that the H3 conformation is controlled both by its sequence and its environment. In looking at a possible correlation between the tilt angle and the conformation of CDR H3, no clear trends are observed. Two variants, H1-69:L3-20 and H3-23:L3-20, have the largest differences in the tilt angles compared to other variants as seen in Table 3. The absolute VH:VL orientation parameters for the 2 Fabs (Table S2) show significant deviation in HL, LC1 and HC2 values (2-3 standard deviations from the mean). One of the variants, H3-23:L3-20, has the CDR H3 conformation similar to the parent, but the other, H1-69:L3-20, is different. As noted in the Results section, the 2 variants, H1-69:L3-20 and H3-23:L3-20, are outliers in terms of the tilt angle; at the same time, both have the smallest VH:VL interface. These smaller interfaces may perhaps translate to a significant deviation in how VH is oriented relative to VL than the other variants. These deviations from the other variants can also be seen to some extent in VH:VL orientation parameters in Table S2, as well as in the smaller number of residues involved in the VH:VL interfaces of these 2 variants (Fig. S5). These differences undoubtedly influence the conformation of the CDRs, in particular CDR H1 (Fig. 1A) and CDR L1 (Fig. 3C), especially with the tandem glycines and multiple serines present, respectively. Pairing of different germlines yields antibodies with various degrees of stability. As indicated by the melting temperatures, germlines H1-69 and H3-23 for HC and germline L1-39 for LC produce more stable Fabs compared to the other germlines in the experimental set. Structural determinants of the differential stability are not always easy to decipher. One possible explanation of the clear preference of LC germline L1-39 is that CDR L3 has smaller residues at positions 91 and 94, allowing for more room to accommodate CDR H3. Other germlines have bulky residues, Tyr, Arg and Trp, at these positions, whereas L1-39 has Ser and Thr. Various combinations of germline sequences for VL and VH impose certain constraints on CDR H3, which has to adapt to the environment. A more compact CDR L3 may be beneficial in this situation. At the other end of the stability range is LC germline L3-20, which yields antibodies with the lowest Tms. While pairings with H3-53 and H5-51 may be safely called a mismatch, those with H1-69 and H3-23 have Tms about 5-6° higher. Curiously, the 2 Fabs, H1-69:L3-20 and H3-23:L3-20, deviate markedly in their tilt angles from the rest of the panel. It is possible that by adopting extreme tilt angles the structure modulates CDR H3 and its environment, which apparently cannot be achieved solely by conformational rearrangement of the CDR. Note that most of the VH:VL interface residues are invariant; therefore, significant change of the tilt angle must come with a penalty in free energy. Yet, for the 2 antibodies, the total gain in stability merits the domain repacking. Overall, the stability of the Fab, as measured by Tm, is a result of the mutual adjustment of the HC and LC variable domains and adjustment of CDR H3 to the VH:VL cleft. The final conformation represents an energetic minimum; however, in most cases it is very shallow, so that a single mutation can cause a dramatic rearrangement of the structure. In summary, the analysis of this structural library of germline variants composed of all pairs of 4 HCs and 4LCs, all with the same CDR H3, offers some unique insights into antibody structure and how pairing and sequence may influence, or not, the canonical structures of the L1, L2, L3, H1 and H2 CDRs. Comparison of the CDR H3s reveals a large set of variants with conformations similar to the parent, while a second set has significant conformational variability, indicating that both the sequence and the structural context define the CDR H3 conformation. Quite unexpectedly, 2 of the variants, H1-69:L3-20 and H3-53:L4-1, have the ‘extended’ stem region differing from the other 14 that have a ‘kinked’ stem region. Why this is the case is unclear at present. These data reveal the difficulty of modeling CDR H3 accurately, as shown again in Antibody Modeling Assessment II. Furthermore, antibody CDRs, H3 in particular, may go through conformational changes upon binding their targets, making structural prediction for docking purposes an even more difficult task. Fortunately, for most applications of antibody modeling, such as engineering affinity and biophysical properties, an accurate CDR H3 structure is not always necessary. For those applications where accurate CDR structures are essential, such as docking, the results in this work demonstrate the importance of experimental structures. With the recent advances in expression and crystallization methods, Fab structures can be obtained rapidly. The set of 16 germline Fab structures offers a unique dataset to facilitate software development for antibody modeling. The results essentially support the underlying idea of canonical structures, indicating that most CDRs with germline sequences tend to adopt predefined conformations. From this point of view, a novel approach to design combinatorial antibody libraries would be to cover the range of CDR conformations that may not necessarily coincide with the germline usage in the human repertoire. This would insure more structural diversity, leading to a more diverse panel of antibodies that would bind to a broad spectrum of targets. The production, purification and crystallization of the Fabs reported in this article were described previously. Briefly, the 16 Fabs were produced by combining 4 different HC and 4 different LC germline constructs. The human HC germlines were IGHV1-69 (H1-69), IGHV3-23 (H3-23), IGHV3-53 (H3-53) and IGHV5-51 (H5-51) in the IMGT nomenclature. The human LC germlines were IGKV1-39 (L1-39), IGKV3-11 (L3-11), IGKV3-20 (L3-20) and IGKV4-1 (L4-1) corresponding to O12, L6, A27 and B3 in the V-BASE nomenclature. CDR H3 of the anti-CCL2 antibody CNTO 888 with the amino acid sequence ARYDGIYGELDF was used in all Fab constructs. The J region genes were IGHJ1 for the HC and IGKJ1 for the LC for all Fabs. Human IgG1 and κ constant regions were used in all Fab constructs. A 6xHis tag was added to the C-terminus of the HC to facilitate purification. The Fabs were expressed in HEK 293E cells and purified by affinity and size-exclusion chromatography. For crystallization, the Fabs were dialyzed into 20 mM Tris buffer, pH 7.4, with 50 mM NaCl and concentrated to 12-18 mg/mL. Automated crystallization screening was carried out using the vapor diffusion method at 20°C with an Oryx4 (Douglas Instruments) or a Mosquito (TTP Labtech) crystallization robot in a sitting drop format using Corning 3550 plates. Initial screening was carried out with an in-house 192-well screen optimized for Fab crystallization and the Hampton 96-well Crystal Screen HT (Hampton Research). For the majority of the Fabs, the crystallization protocol employed microseed matrix screening using self-seeding or cross-seeding approaches. A summary of the final crystallization conditions for each of the Fabs is presented in Table 1. For 13 of the Fab crystals, X-ray data collection was carried out at Janssen Research and Development, LLC using a Rigaku MicroMax™-007HF microfocus X-ray generator equipped with a Saturn 944 CCD detector and an X-stream™ 2000 cryocooling system (Rigaku), and for the remaining 3, X-ray data collection was carried out at the Advanced Photon Source (APS) synchrotron at Argonne National Laboratory using the IMCA 17-ID beamline with a Pilatus 6M detector. For X-ray data collection, the Fab crystals were soaked for a few seconds in a cryo-protectant solution containing the corresponding mother liquor supplemented with 17-25% glycerol (Table S1). The crystals for which data were collected in-house were flash cooled in the stream of nitrogen at 100 K. Crystals sent to the APS were flash cooled in liquid nitrogen prior to shipping them to the synchrotron. Diffraction data for all variants were processed with the program XDS. X-ray data statistics are given in Table 1. A summary of the methods used in the structure solution and refinement of the 16 Fabs is presented in Table S1. Twelve of the structures were solved by molecular replacement with Phaser using different combinations of search models for the VH, VL and constant domains. Four of the structures, H3-53:L1-39, H3-53:L3-11, H5-51:L1-39 and H5-51:L3-11, were solved by direct replacement followed by rigid body refinement with REFMAC. All structures were refined using REFMAC. Model adjustments were carried out using the program Coot. The refinement statistics are given in Table 1. Other crystallographic calculations were performed with the CCP4 suite of programs. The structural figures were prepared using the PyMOL Molecular Graphics System, Version 1.0 (Schrödinger, LLC). The canonical structure assignments (Table 2) were made using PyIgClassify, an online canonical structure classification tool (http://dunbrack2.fccc.edu/pyigclassify/) that uses the rules set forth by Dunbrack and coworkers. The conformational variability within the CDRs was assessed by calculating the root-mean-square deviation (rmsd) from the average structure that was generated after superposition of all structures of the set using the main-chain atoms of the CDR in question. The rmsd was calculated for all main-chain atoms (N, CA, C, O) of the CDR. The contact surface areas of the VH and VL domains at the VH:VL inteface were computed with the CCP4 program PISA. The surface complementarity of the VH and VL domains was computed using the CCP4 program SC. The orientation of the VH domain with respect to the VL domain was assessed using 2 different approaches. The first approach calculates the 6 VH:VL orientation parameters that describe the VH:VL relationship according to Dunbar and co-workers using a script downloaded from the website (http://opig.stats.ox.ac.uk/webapps/abangle). The six parameters include 5 angles, HL, H1, H2, L1 and L2, and a distance, dc. These parameters are derived by first defining 2 planes, one for each domain, based on core residues in the domains. The distance between the planes, dc, is determined along a vector between the planes that is used to establish a consistent coordinate system. The torsion angle between the domains, HL, is much like the VH:VL packing angle defined by Abhinandan and Martin. The tilt of one domain relative to the other is defined by the HC1 and LC1 angles, and the twist of one domain relative to the other is defined by the HC2 and LC2 angles. The second approach calculates the difference in the tilt angle between pairs of Fvs, which reflects the relative orientation between the VH and VL domains. The difference with respect to the reference structure is calculated by sequential root-mean-square superposition of the VL and VH domains using β-sheet core Cα positions (Chothia numbering scheme): 3–13, 18–25, 33–38, 43–49, 61–67, 70–76, 85–90, 97–103 for VL; 3–7, 18–24, 34–40, 44–51, 56–59, 67–72, 77–82a, 87–94, 102–110 for VH. The κ angle in the spherical polar angular system (ω, ϕ, κ) of the latter transformation is the difference in the tilt angle. DSC experiments were performed on a VP-capillary DSC system (MicroCal Inc., Northampton, MA) in which temperature differences between the reference and sample cell are continuously measured and calibrated to power units. Samples were heated from 10°C to 95°C at a heating rate of 60°C/hour. The pre-scan time was 15 minutes and the filtering period was 10 seconds. The concentration used in the DSC experiments was about 0.4 mg/mL in phosphate-buffered saline. Analysis of the resulting thermograms was performed using MicroCal Origin 7 software. Melting temperature of proteins was determined by deconvolution of the DSC scans using non-2 state model in the MicroCal Origin 7 software. Scans were deconvoluted using a non-2 state model with either 1-step transition or 2-step transition depending on the number of resolved peaks observed in a scan. Atomic coordinates and structure factors have been deposited in the Protein Data Bank with accession numbers 4KMT, 5I15, 5I16, 5I17, 5I18, 5I19, 5I1A, 5I1C, 5I1D, 5I1E, 5I1G, 5I1H, 5I1I, 5I1J, 5I1K and 5I1L.
PMC4854314	RNA protects a nucleoprotein complex against radiation damage	Systematic analysis of radiation damage within a protein–RNA complex over a large dose range (1.3–25 MGy) reveals significant differential susceptibility of RNA and protein. A new method of difference electron-density quantification is presented.With the wide use of high-flux third-generation synchrotron sources, radiation damage (RD) has once again become a dominant reason for the failure of structure determination using macromolecular crystallography (MX) in experiments conducted both at room temperature and under cryocooled conditions (100 K). Significant progress has been made in recent years in understanding the inevitable manifestations of X-ray-induced RD within protein crystals, and there is now a body of literature on possible strategies to mitigate the effects of RD (e.g. Zeldin, Brockhauser et al., 2013 ▸; Bourenkov & Popov, 2010 ▸). However, there is still no general consensus within the field on how to minimize RD during MX data collection, and debates on the dependence of RD progression on incident X-ray energy (Shimizu et al., 2007 ▸; Liebschner et al., 2015 ▸) and the efficacy of radical scavengers (Allan et al., 2013 ▸) have yet to be resolved. RD manifests in two forms. Global radiation damage is observed within reciprocal space as the overall decay of the summed intensity of reflections detected within the diffraction pattern as dose increases (Garman, 2010 ▸; Murray & Garman, 2002 ▸). Dose is defined as the absorbed energy per unit mass of crystal in grays (Gy; 1 Gy = 1 J kg), and is the metric against which damage progression should be monitored during MX data collection, as opposed to time. At 100 K, an experimental dose limit of 30 MGy has been recommended as an upper limit beyond which the biological information derived from any macromolecular crystal may be compromised (Owen et al., 2006 ▸). Specific radiation damage (SRD) is observed in the real-space electron density, and has been detected at much lower doses than any observable decay in the intensity of reflections. Indeed, the C—Se bond in selenomethionine, the stability of which is key for the success of experimental phasing methods, can be cleaved at a dose as low as 2 MGy for a crystal maintained at 100 K (Holton, 2007 ▸). SRD has been well characterized in a large range of proteins, and is seen to follow a reproducible order: metallo-centre reduction, disulfide-bond cleavage, acidic residue decarboxylation and methionine methylthio cleavage (Ravelli & McSweeney, 2000 ▸; Burmeister, 2000 ▸; Weik et al., 2000 ▸; Yano et al., 2005 ▸). Furthermore, damage susceptibility within each residue type follows a preferential ordering influenced by a combination of local environment factors (solvent accessibility, conformational strain, proximity to active sites/high X-ray cross-section atoms; Holton, 2009 ▸). Deconvoluting the individual roles of these parameters has been surprisingly challenging, with factors such as solvent accessibility currently under active investigation (Weik et al., 2000 ▸; Fioravanti et al., 2007 ▸; Gerstel et al., 2015 ▸). There are a number of cases where SRD manifestations have compromised the biological information extracted from MX-determined structures at much lower doses than the recommended 30 MGy limit, leading to false structural interpretations of protein mechanisms. Active-site residues appear to be particularly susceptible, particularly for photosensitive proteins and in instances where chemical strain is an intrinsic feature of the reaction mechanism. For instance, structure determination of the purple membrane protein bacteriorhodopsin required careful corrections for radiation-induced structural changes before the correct photosensitive intermediate states could be isolated (Matsui et al., 2002 ▸). The significant chemical strain required for catalysis within the active site of phosphoserine aminotransferase has been observed to diminish during X-ray exposure (Dubnovitsky et al., 2005 ▸). Since the majority of SRD studies to date have focused on proteins, much less is known about the effects of X-ray irradiation on the wider class of crystalline nucleoprotein complexes or how to correct for such radiation-induced structural changes. Understanding RD to such complexes is crucial, since DNA is rarely naked within a cell, instead dynamically interacting with proteins, facilitating replication, transcription, modification and DNA repair. As of early 2016, >5400 nucleoprotein complex structures have been deposited within the PDB, with 91% solved by MX. It is essential to understand how these increasingly complex macromolecular structures are affected by the radiation used to solve them. Nucleoproteins also represent one of the main targets of radiotherapy, and an insight into the damage mechanisms induced by X-ray irradiation could inform innovative treatments. When a typical macromolecular crystal is irradiated with ionizing X-rays, each photoelectron produced via interactions with both the macromolecule (direct damage) and solvent (indirect damage) can induce cascades of up to 500 secondary low-energy electrons (LEEs) that are capable of inducing further ionizations. Investigations on sub-ionization-level LEEs (0–15 eV) interacting with both dried and aqueous oligonucleotides (Alizadeh & Sanche, 2014 ▸; Simons, 2006 ▸) concluded that resonant electron attachment to DNA bases and the sugar-phosphate backbone could lead to the preferential cleavage of strong (∼4 eV, 385 kJ mol) sugar-phosphate C—O covalent bonds within the DNA backbone and then base-sugar N1—C bonds, eventually leading to single-strand breakages (SSBs; Ptasińska & Sanche, 2007 ▸). Electrons have been shown to be mobile at 77 K by electron spin resonance spectroscopy studies (Symons, 1997 ▸; Jones et al., 1987 ▸), with rapid electron quantum tunnelling and positive hole migration along the protein backbone and through stacked DNA bases indicated as a dominant mechanism by which oxidative and reductive damage localizes at distances from initial ionization sites at 100 K (O’Neill et al., 2002 ▸). The investigation of naturally forming nucleoprotein complexes circumvents the inherent challenges in making controlled comparisons of damage mechanisms between protein and nucleic acids crystallized separately. Recently, for a well characterized bacterial protein–DNA complex (C.Esp1396I; PDB entry 3clc; resolution 2.8 Å; McGeehan et al., 2008 ▸) it was concluded that over a wide dose range (2.1–44.6 MGy) the protein was far more susceptible to SRD than the DNA within the crystal (Bury et al., 2015 ▸). Only at doses above 20 MGy were precursors of phosphodiester-bond cleavage observed within AT-rich regions of the 35-mer DNA. For crystalline complexes such as C.Esp1396I, whether the protein is intrinsically more susceptible to X-ray-induced damage or whether the protein scavenges electrons to protect the DNA remains unclear in the absence of a non-nucleic acid-bound protein control obtained under exactly the same crystallization and data-collection conditions. To monitor the effects of nucleic acid binding on protein damage susceptibility, a crystal containing two protein molecules per asymmetric unit, only one of which was bound to RNA, is reported here (Fig. 1 ▸). Using newly developed methodology, we present a controlled SRD investigation at 1.98 Å resolution using a large (∼91 kDa) crystalline protein–RNA complex: trp RNA-binding attenuation protein (TRAP) bound to a 53 bp RNA sequence (GAGUU)10GAG (PDB entry 1gtf; Hopcroft et al., 2002 ▸). TRAP consists of 11 identical subunits assembled into a ring with 11-fold rotational symmetry. It binds with high affinity (K d ≃ 1.0 nM) to RNA segments containing 11 GAG/UAG triplets separated by two or three spacer nucleotides (Elliott et al., 2001 ▸) to regulate the transcription of tryptophan biosynthetic genes in Bacillus subtilis (Antson et al., 1999 ▸). In this structure, the bases of the G1-A2-G3 nucleotides form direct hydrogen bonds to the protein, unlike the U4-U5 nucleotides, which appear to be more flexible. Ten successive 1.98 Å resolution MX data sets were collected from the same TRAP–RNA crystal to analyse X-ray-induced structural changes over a large dose range (d 1 = 1.3 MGy to d 10 = 25.0 MGy). To avoid the previous necessity for visual inspection of electron-density maps to detect SRD sites, a computational approach was designed to quantify the electron-density change for each refined atom with increasing dose, thus providing a rapid systematic method for SRD study on such large multimeric complexes. By employing the high 11-fold structural symmetry within each TRAP macromolecule, this approach permitted a thorough statistical quantification of the RD effects of RNA binding to TRAP. As previously described (Hopcroft et al., 2002 ▸), the 53-base RNA (GAGUU)10GAG was synthesized by in vitro transcription with T7 RNA polymerase and gel-purified. TRAP from B. stearothermophilus was overexpressed in Escherichia coli and purified. TRAP–RNA crystals were prepared using a previously established hanging-drop crystallization protocol (Antson et al., 1999 ▸). By using a 2:1 molar ratio of TRAP to RNA, crystals successfully formed from the protein–RNA complex (∼15 mg ml) in a solution containing 70 mM potassium phosphate pH 7.8 and 10 mM l-tryptophan. The reservoir consisted of 0.2 M potassium glutamate, 50 mM triethanolamine pH 8.0, 10 mM MgCl2, 8–11% monomethyl ether PEG 2000. In order to accelerate crystallization, a further gradient was induced by adding 0.4 M KCl to the reservoir after 1.5 µl protein solution had been mixed with an equal volume of the reservoir solution. Wedge-shaped crystals of approximate length 70 µm (longest dimension) grew within 3 d and were vitrified and stored in liquid nitrogen immediately after growth. The cryosolution consisted of 12% monomethyl ether PEG 2000, 30 mM triethanolamine pH 8.0, 6 mM l-tryptophan, 0.1 M potassium glutamate, 35 mM potassium phosphate pH 7.8, 5 mM MgCl2 with 25% 2-methyl-2,4-pentanediol (MPD) included as a cryoprotectant. Data were collected at 100 K from a wedge-shaped TRAP–RNA crystal of approximate dimensions 70 × 20 × 40 µm (see Supplementary Fig. S2) on beamline ID14-4 at the ESRF using an incident wavelength of 0.940 Å (13.2 keV) and an ADSC Q315R mosaic CCD detector at 304.5 mm from the crystal throughout the data collection. The beam size was slitted to 0.100 mm (vertical) × 0.160 mm (horizontal), with a uniformly distributed profile, such that the crystal was completely bathed within the beam throughout data collection. Ten successive (1.98 Å resolution) 180° data sets (with Δφ = 1°) were collected over the same angular range from a TRAP–RNA crystal at 28.9% beam transmission. The TRAP–RNA macromolecule crystallized in space group C2, with unit-cell parameters a = 140.9, b = 110.9, c = 137.8 Å, α = γ = 90, β = 137.8° (the values quoted are for the first data set; see Supplementary Table S1 for subsequent values). For the first nine data sets the attenuated flux was recorded to be ∼5 × 10 photons s. A beam refill took place immediately before data set 10, requiring a flux-scale factor increase of 1.42 to be applied, based on the ratio of observed relative intensity I D/I 1 at data set 10 to that extrapolated from data set 9. RADDOSE-3D (Zeldin, Gerstel et al., 2013 ▸) was used to calculate the absorbed dose distribution during each data set (see input file; Supplementary Figs. S1 and S2). The crystal composition was calculated from the deposited TRAP–RNA structure (PDB entry 1gtf; Hopcroft et al., 2002 ▸). Crystal absorption coefficients were calculated in RADDOSE-3D using the concentration (mmol l) of solvent heavy elements from the crystallization conditions. The beam-intensity profile was modelled as a uniform (‘top-hat’) distribution. The diffraction-weighted dose (DWD) values (Zeldin, Brockhauser et al., 2013 ▸) are given in Supplementary Table S1. Each data set was integrated using iMosflm (Leslie & Powell, 2007 ▸) and was scaled using AIMLESS (Evans & Murshudov, 2013 ▸; Winn et al., 2011 ▸) using the same 5% R free set of test reflections for each data set. To phase the structure obtained from the first data set, molecular replacement was carried out with Phaser (McCoy et al., 2007 ▸), using an identical TRAP–RNA structure (PDB entry 1gtf; resolution 1.75 Å; Hopcroft et al., 2002 ▸) as a search model. The resulting TRAP–RNA structure (TR1) was refined using REFMAC5 (Murshudov et al., 2011 ▸), initially using rigid-body refinement, followed by repeated cycles of restrained, TLS and isotropic B-factor refinement, coupled with visual inspection in Coot (Emsley et al., 2010 ▸). TR1 was refined to 1.98 Å resolution, with a dimeric assembly of non-RNA-bound and RNA-bound TRAP rings within the asymmetric unit. Consistent with previous structures of the TRAP–RNA complex, the RNA sequence termini were not observed within the 2F o − F c map; the first spacer (U4) was then modelled at all 11 repeats around the TRAP ring and the second spacer (U5) was omitted from the final refined structure. For the later data sets, the observed structure-factor amplitudes from each separately scaled data set (output from AIMLESS) were combined with the phases of TR1 and the resulting higher-dose model was refined with phenix.refine (Adams et al., 2010 ▸) using only rigid-body and isotropic B-factor refinement. During this refinement, the TRAP–RNA complex and nonbound TRAP ring were treated as two separate rigid bodies within the asymmetric unit. Supplementary Table S1 shows the relevant summary statistics. The CCP4 program CAD was used to create a series of nine merged .mtz files combining observed structure-factor amplitudes for the first data set F obs(d 1) with each later data set F obs(d n) (for n = 2, …, 10). All later data sets were scaled against the initial low-dose data set in SCALEIT. For each data set an atom-tagged .map file was generated using the ATMMAP mode in SFALL (Winn et al., 2011 ▸). A full set of nine Fourier difference maps F obs(d n) − F obs(d 1) were calculated using FFT (Ten Eyck, 1973 ▸) over the full TRAP–RNA unit-cell dimensions, with the same grid-sampling dimensions as the atom-tagged .map file. All maps were cropped to the TRAP asymmetric unit in MAPMASK. Comparing the atom-tagged .map file and F obs(d n) − F obs(d 1) difference map at each dose, each refined atom was assigned a set of density-change values X. The maximum density-loss metric, D loss (units of e Å), was calculated to quantify the per-atom electron-density decay at each dose, assigned as the absolute magnitude of the most negative Fourier difference map voxel value in a local volume around each atom as defined by the set X. Model calculations were run for the simple amino acids glutamate and aspartate. In order to avoid decarboxylation at the C-terminus instead of the side chain on the C atom, the C-terminus of each amino acid was methylated. While the structures of the closed shell acids are well known, the same is not true of those in the oxidized state. The quantum-chemical calculations employed were chosen to provide a satisfactory description of the structure of such radical species and also provide a reliable estimation of the relative C—C(O2) bond strengths, which are otherwise not available. Structures of methyl-terminated (at the N- and C-termini) carboxylates were determined using analytic energy gradients with density functional theory (B3LYP functional; Becke, 1993 ▸) and a flexible basis set of polarized valence triple-zeta size with diffuse functions on the non-H atoms [6-311+G(d,p)] in the Gaussian 09 computational chemistry package (Frisch et al., 2009 ▸). The stationary points obtained were characterized as at least local minima by examination of the associated analytic Hessian. Effects of the medium were modelled using a dielectric cavity approach (Tomasi et al., 1999 ▸) parameterized for water. To quantify the exact effects of nucleic acid binding to a protein on SRD susceptibility, a high-throughput and automated pipeline was created to systematically calculate the electron-density change for every refined atom within the TRAP–RNA structure as a function of dose. This provides an atom-specific quantification of density–dose dynamics, which was previously lacking within the field. Previous studies have characterized SRD sites by reporting magnitudes of F obs(d n) − F obs(d 1) Fourier difference map peaks in terms of the sigma (σ) contour level (the number of standard deviations from the mean map electron-density value) at which peaks become visible. However, these σ levels depend on the standard deviation values of the map, which can deviate between data sets, and are thus unsuitable for quantitative comparison of density between different dose data sets. Instead, we use here a maximum density-loss metric (D loss), which is the per-atom equivalent of the magnitude of these negative Fourier difference map peaks in units of e Å. Large positive D loss values indicate radiation-induced atomic disordering reproducibly throughout the unit cells with respect to the initial low-dose data set. For each TRAP–RNA data set, the D loss metric successfully identified the recognized forms of protein SRD (Fig. 2 ▸ a), with clear Glu and Asp side-chain decarboxylation even in the first difference map calculated (3.9 MGy; Fig. 3 ▸ a). The main sequence of TRAP does not contain any Trp and Cys residues (and thus contains no disulfide bonds). The substrate Trp amino-acid ligands also exhibited disordering of the free terminal carboxyl groups at higher doses (Fig. 2 ▸ a); however, no clear Fourier difference peaks could be observed visually. Even for radiation-insensitive residues (e.g. Gly) the average D loss increases with dose: this is the effect of global radiation damage, since as dose increases the electron density associated with each refined atom becomes weaker as the atomic occupancy decreases (Fig. 2 ▸ b). Only Glu and Asp residues exhibit a rate of D loss increase that consistently exceeds the average decay (Fig. 2 ▸ b, dashed line) at each dose. Additionally, the density surrounding ordered solvent molecules was determined to significantly diminish with increasing dose (Fig. 2 ▸ b). The rate of D loss (attributed to side-chain decarboxylation) was consistently larger for Glu compared with Asp residues over the large dose range (Fig. 2 ▸ b and Supplementary Fig. S3); this observation is consistent with our calculations on model systems (see above) that suggest that, without considering differential hydrogen-bonding environments, CO2 loss is more exothermic by around 8 kJ mol from oxidized Glu residues than from their Asp counterparts. Visual inspection of Fourier difference maps illustrated the clear lack of RNA electron-density degradation with increasing dose compared with the obvious protein damage manifestations (Figs. 3 ▸ b and 3 ▸ c). Only at the highest doses investigated (>20 MGy) was density loss observed at the RNA phosphate and C—O bonds of the phosphodiester backbone. However, the median D loss was lower by a factor of >2 for RNA P atoms than for Glu and Asp side-chain groups at 25.0 MGy (Supplementary Fig. S4), and furthermore could not be numerically distinguished from Gly C atoms within TRAP, which are not radiation-sensitive at the doses tested here (Supplementary Fig. S3). For the large number of acidic residues per TRAP ring (four Asp and six Glu residues per protein monomer), a strong dependence of decarboxylation susceptibility on local environment was observed (Fig. 4 ▸). For each Glu C or Asp C atom, D loss provided a direct measure of the rate of side-chain carboxyl-group disordering and subsequent decarboxylation. For acidic residues with no differing interactions between nonbound and bound TRAP (Fig. 4 ▸ a), similar damage was apparent between the two rings within the asymmetric unit, as expected. However, TRAP residues directly on the RNA-binding interfaces exhibited greater damage accumulation in nonbound TRAP (Fig. 4 ▸ b), and for residues at the ring–ring interfaces (where crystal contacts were detected) bound TRAP exhibited enhanced SRD accumulation (Fig. 4 ▸ c). Three acidic residues (Glu36, Asp39 and Glu42) are involved in RNA interactions within each of the 11 TRAP ring subunits, and Fig. 5 ▸ shows their density changes with increasing dose. Hotelling’s T-squared test (the multivariate counterpart of Student’s t-test) was used to reject the null hypothesis that the means of the D loss metric were equal for the bound and nonbound groups in Fig. 5 ▸. A significant reduction in D loss is seen for Glu36 in RNA-bound compared with nonbound TRAP, indicative of a lower rate of side-chain decarboxylation (Fig. 5 ▸ a; p = 6.06 × 10). For each TRAP ring subunit, the Glu36 side-chain carboxyl group accepts a pair of hydrogen bonds from the two N atoms of the G3 RNA base. In our analysis, Asp39 in the TRAP–(GAGUU)10GAG structure appears to exhibit two distinct hydrogen bonds to the G1 base within each of the 11 TRAP–RNA interfaces, as does Glu36 to G3; however, the reduction in density disordering upon RNA binding is far less significant for Asp39 than for Glu36 (Fig. 5 ▸ b, p = 0.0925). One oxygen (O) of Glu42 appears to form a hydrogen bond to a nearby water within each TRAP RNA-binding pocket, with the other (O) being involved in a salt-bridge interaction with Arg58 (Hopcroft et al., 2002 ▸; Antson et al., 1999 ▸). Salt-bridge interactions have previously been suggested to reduce the glutamate decarboxylation rate within the large (∼62.4 kDa) myrosinase protein structure (Burmeister, 2000 ▸). A significant difference was observed between the D loss dynamics for the nonbound/bound Glu42 O atoms (Fig. 5 ▸ c; p = 0.007) but not for the Glu42 O atoms (Fig. 5 ▸ d; p = 0.239), indicating that the stabilizing strength of this salt-bridge interaction was conserved upon RNA binding and that the water-mediated hydrogen bond had a greater relative susceptibility to atomic disordering in the absence of RNA. The density-change dynamics were statistically indistinguishable between bound and nonbound TRAP for each Glu42 carboxyl group C atom (p = 0.435), indicating that upon RNA binding the conserved salt-bridge interaction ultimately dictated the overall Glu42 decarboxylation rate. The RNA-stabilizing effect was not restricted to radiation-sensitive acidic residues. The side chain of Phe32 stacks against the G3 base within the 11 TRAP RNA-binding interfaces (Antson et al., 1999 ▸). With increasing dose, the D loss associated with the Phe32 side chain was significantly reduced upon RNA binding (Fig. 5 ▸ e; Phe32 C; p = 0.0014), an indication that radiation-induced conformation disordering of Phe32 had been reduced. The extended aliphatic Lys37 side chain stacks against the nearby G1 base, making a series of nonpolar contacts within each RNA-binding interface. The D loss for Lys37 side-chain atoms was also reduced when stacked against the G1 base (Fig. 5 ▸ f; p = 0.0243 for Lys37 C atoms). Representative Phe32 and Lys37 atoms were selected to illustrate these trends. Here, MX radiation-induced specific structural changes within the large TRAP–RNA assembly over a large dose range (1.3–25.0 MGy) have been analysed using a high-throughput quantitative approach, providing a measure of the electron-density distribution for each refined atom with increasing dose, D loss. Compared with previous studies, the results provide a further step in the detailed characterization of SRD effects in MX. Our methodology, which eliminated tedious and error-prone visual inspection, permitted the determination on a per-atom basis of the most damaged sites, as characterized by F obs(d n) − F obs(d 1) Fourier difference map peaks between successive data sets collected from the same crystal. Here, it provided the precision required to quantify the role of RNA in the damage susceptibilities of equivalent atoms between RNA-bound and nonbound TRAP, but it is applicable to any MX SRD study. The RNA was found to be substantially more radiation-resistant than the protein, even at the highest doses investigated (∼25.0 MGy), which is in strong concurrence with our previous SRD investigation of the C.Esp1396I protein–DNA complex (Bury et al., 2015 ▸). Consistent with that study, at high doses of above ∼20 MGy, F obs(d n) − F obs(d 1) map density was detected around P, O3′ and O5′ atoms of the RNA backbone, with no significant difference density localized to RNA ribose and basic subunits. RNA backbone disordering thus appears to be the main radiation-induced effect in RNA, with the protein–base interactions maintained even at high doses (>20 MGy). The U4 phosphate exhibited marginally larger D loss values above 20 MGy than G1, A2 and G3 (Supplementary Fig. S4). Since U4 is the only refined nucleotide not to exhibit significant base–protein interactions around TRAP (with a water-mediated hydrogen bond detected in only three of the 11 subunits and a single Arg58 hydrogen bond suggested in a further four subunits), this increased U4 D loss can be explained owing to its greater flexibility. At 25.0 MGy, the magnitude of the RNA backbone D loss was of the same order as for the radiation-insensitive Gly C atoms and on average less than half that of the acidic residues of the protein (Supplementary Fig. S3). Consequently, no clear single-strand breaks could be located, and since RNA-binding within the current TRAP–(GAGUU)10GAG complex is mediated predominantly through base–protein interactions, the biological integrity of the RNA complex was dictated by the rate at which protein decarboxylation occurred. RNA interacting with TRAP was shown to offer significant protection against radiation-induced structural changes. Both Glu36 and Asp39 bind directly to RNA, each through two hydrogen bonds to guanine bases (G3 and G1, respectively). However, compared with Asp39, Glu36 is strikingly less decarboxylated when bound to RNA (Fig. 4 ▸). This is in good agreement with previous mutagenesis and nucleoside analogue studies (Elliott et al., 2001 ▸), which indicated that the G1 nucleotide does not bind to TRAP as strongly as do A2 and G3, and plays little role in the high RNA-binding affinity of TRAP (K d ≃ 1.1 ± 0.4 nM). For Glu36 and Asp39, no direct quantitative correlation could be established between hydrogen-bond length and D loss (linear R of <0.23 for all doses; Supplementary Fig. S5). Thus, another factor must be responsible for this clear reduction in Glu36 CO2 decarboxylation in RNA-bound TRAP. The Glu36 carboxyl side chain also potentially forms hydrogen bonds to His34 and Lys56, but since these interactions are conserved irrespective of G3 nucleotide binding, this cannot directly account for the stabilization effect on Glu36 in RNA-bound TRAP. Radiation-induced decarboxylation has been proposed to be mediated by preferential positive-hole migration to the side-chain carboxyl group, with rapid proton transfer trapping the hole at the carboxyl group (Burmeister, 2000 ▸; Symons, 1997 ▸):where the forward rate is K 1 and the backward rate is K −1, where the forward rate is K 2. When bound to RNA, the average solvent-accessible area of the Glu36 side-chain O atoms is reduced from ∼15 to 0 Å. We propose that with no solvent accessibility Glu36 decarboxylation is inhibited, since the CO2-formation rate K 2 is greatly reduced, and suggest that steric hindrance prevents each radicalized Glu36 CO2 group from achieving the planar conformation required for complete dissociation from TRAP. The electron-recombination rate K −1 remains high, however, owing to rapid electron migration through the protein–RNA complex to refill the Glu36 positive hole (the precursor for Glu decarboxylation). Upon RNA binding, the Asp39 side-chain carboxyl group solvent-accessible area changes from ∼75 to 35 Å, still allowing a high CO2-formation rate K 2. Previous studies have reported inconsistent results concerning the dependence of the acidic residue decarboxylation rate on solvent accessibility (Weik et al., 2000 ▸; Fioravanti et al., 2007 ▸; Gerstel et al., 2015 ▸). The prevalence of radical attack from solvent channels surrounding the protein in the crystal is a questionable cause, considering previous observations indicating that the strongly oxidizing hydroxyl radical is immobile at 100 K (Allan et al., 2013 ▸; Owen et al., 2012 ▸). Furthermore, the suggested electron hole-trapping mechanism which induces decarboxylation within proteins at 100 K has no clear mechanistic dependence on the solvent-accessible area of each carboxyl group. By comparing equivalent acidic residues with and without RNA, we have now deconvoluted the role of solvent accessibility from other local protein environment factors, and thus propose a suitable mechanism by which exceptionally low solvent accessibility can reduce the rate of decarboxylation. Overall, no direct correlation between solvent accessibility and decarboxylation susceptibility was observed, but it is very clear that inaccessible residues are protected. Apart from these RNA-binding interfaces, RNA binding was seen to enhance decarboxylation for residues Glu50, Glu71 and Glu73, all of which are involved in crystal contacts between TRAP rings (Fig. 4 ▸ c). However, for each of these residues the exact crystal contacts are not preserved between bound and nonbound TRAP or even between monomers within one TRAP ring. For example, in bound TRAP, Glu73 hydrogen-bonds to a nearby lysine on each of the 11 subunits, whereas in nonbound TRAP no such interaction exists and Glu73 interacts with a variable number of refined waters in each subunit. Thus, the dependence of decarboxylation rates on these interactions could not be established. Radiation-induced side-chain conformational changes have been poorly characterized in previous SRD investigations owing to their strong dependence on packing density and geometric strain. Such structural changes are known to have significant roles within enzymatic pathways, and experimenters must be aware of these possible confounding factors when assigning true functional mechanisms using MX. Our results show that RNA binding to TRAP physically stabilizes non-acidic residues within the TRAP macromolecule, most notably Lys37 and Phe32, which stack against the G1 and G3 bases, respectively. It has been suggested (Burmeister, 2000 ▸) that Tyr residues can lose their aromatic –OH group owing to radiation-induced effects; however, no energetically favourable pathway for –OH cleavage exists and this has not been detected in aqueous radiation-chemistry studies. In TRAP, D loss increased at a similar rate for both the Tyr O atoms and aromatic ring atoms, suggesting that full ring conformational disordering is more likely. Indeed, no convincing reproducible Fourier difference peaks above the background map noise were observed around any Tyr terminal –OH groups. The RNA-stabilization effects on protein are observed at short ranges and are restricted to within the RNA-binding interfaces around the TRAP ring. For example, Asp17 is located ∼6.8 Å from the G1 base, outside the RNA-binding interfaces, and has indistinguishable C atom D loss dose-dynamics between RNA-bound and nonbound TRAP (p > 0.9). An increase in the dose at which functionally important residues remain intact has biological ramifications for understanding the mechanisms at which ionizing radiation damage is mitigated within naturally forming DNA–protein and RNA–protein complexes. Observations of lower protein radiation-sensitivity in DNA-bound forms have been recorded in solution at RT at much lower doses (∼1 kGy) than those used for typical MX experiments [e.g. an oestrogen response element–receptor complex (Stísová et al., 2006 ▸) and a DNA glycosylase and its abasic DNA target site (Gillard et al., 2004 ▸)]. In these studies, the main damaging species is predicted to be the oxidizing hydroxyl radical produced through solvent irradiation, which is known to add to double covalent bonds within both DNA and RNA bases to induce strand breaks and base modification (Spotheim-Maurizot & Davídková, 2011 ▸; Chance et al., 1997 ▸). It was suggested that physical screening of DNA by protein shielded the DNA–protein interaction sites from radical damage, yielding an extended life-dose for the nucleoprotein complex compared with separate protein and DNA constituents at RT. However, in the current MX study at 100 K, the main damaging species are believed to be migrating LEEs and holes produced directly within the protein–RNA components or in closely associated solvent. The results presented here suggest that biologically relevant nucleoprotein complexes also exhibit prolonged life-doses under the effect of LEE-induced structural changes, involving direct physical protection of key RNA-binding residues. Such reduced radiation-sensitivity in this case ensures that the interacting protein remains bound long enough to the RNA to complete its function, even whilst exposed to ionizing radiation. Within the nonbound TRAP macromolecule, the acidic residues within the unoccupied RNA-binding interfaces (Asp39, Glu36, Glu42) are notably amongst the most susceptible residues within the asymmetric unit (Fig. 4 ▸). When exposed to X-rays, these residues will be preferentially damaged by X-rays and subsequently reduce the affinity with which TRAP binds to RNA. Within the cellular environment, this mechanism could reduce the risk that radiation-damaged proteins might bind to RNA, thus avoiding the detrimental introduction of incorrect DNA-repair, transcriptional and base-modification pathways. The Python scripts written to calculate the per atom D loss metric are available from the authors on request. The following references are cited in the Supporting Information for this article: Chen et al. (2010 ▸).
PMC4792962	A unified mechanism for proteolysis and autocatalytic activation in the 20S proteasome	Biogenesis of the 20S proteasome is tightly regulated. The N-terminal propeptides protecting the active-site threonines are autocatalytically released only on completion of assembly. However, the trigger for the self-activation and the reason for the strict conservation of threonine as the active site nucleophile remain enigmatic. Here we use mutagenesis, X-ray crystallography and biochemical assays to suggest that Lys33 initiates nucleophilic attack of the propeptide by deprotonating the Thr1 hydroxyl group and that both residues together with Asp17 are part of a catalytic triad. Substitution of Thr1 by Cys disrupts the interaction with Lys33 and inactivates the proteasome. Although a Thr1Ser mutant is active, it is less efficient compared with wild type because of the unfavourable orientation of Ser1 towards incoming substrates. This work provides insights into the basic mechanism of proteolysis and propeptide autolysis, as well as the evolutionary pressures that drove the proteasome to become a threonine protease.The 20S proteasome core particle (CP) is the key non-lysosomal protease of eukaryotic cells. Its seven different α and seven different β subunits assemble into four heptameric rings that are stacked on each other to form a hollow cylinder. While the inactive α subunits build the two outer rings, the β subunits form the inner rings. Only three out of the seven different β subunits, namely β1, β2 and β5, bear N-terminal proteolytic active centres, and before CP maturation these are protected by propeptides123. In the last stage of CP biogenesis, the prosegments are autocatalytically removed through nucleophilic attack by the active site residue Thr1 on the preceding peptide bond involving Gly(-1)45. Release of the propeptides creates a functionally active CP that cleaves proteins into short peptides. Although the chemical nature of the substrate-binding channel and hence substrate preferences are unique to each of the distinct active β subunits67, all active sites employ an identical reaction mechanism to hydrolyse peptide bonds2. Nucleophilic attack of Thr1O on the carbonyl carbon atom of the scissile peptide bond creates a first cleavage product and a covalent acyl-enzyme intermediate. Hydrolysis of this complex by the addition of a nucleophilic water molecule regenerates the enzyme and releases the second peptide fragment89. The proteasome belongs to the family of N-terminal nucleophilic (Ntn) hydrolases10, and the free N-terminal amine group of Thr1 was proposed to deprotonate the Thr1 hydroxyl group to generate a nucleophilic Thr1O for peptide-bond cleavage2911. This mechanism, however, cannot explain autocatalytic precursor processing because in the immature active sites, Thr1N is part of the peptide bond with Gly(-1), the bond that needs to be hydrolysed. An alternative candidate for deprotonating the Thr1 hydroxyl group is the side chain of Lys33 as it is within hydrogen-bonding distance to Thr1OH (2.7 Å). In principle it could function as the general base during both autocatalytic removal of the propeptide and protein substrate cleavage. Here we provide experimental evidences for this distinct view of the proteasome active-site mechanism. Data from biochemical and structural analyses of proteasome variants with mutations in the β5 propeptide and the active site strongly support the model and deliver novel insights into the structural constraints required for the autocatalytic activation of the proteasome. Furthermore, we determine the advantages of Thr over Cys or Ser as the active-site nucleophile using X-ray crystallography together with activity and inhibition assays. Proteasome-mediated degradation of cell-cycle regulators and potentially toxic misfolded proteins is required for the viability of eukaryotic cells8. Inactivation of the active site Thr1 by mutation to Ala has been used to study substrate specificity and the hierarchy of the proteasome active sites1412131415. Yeast strains carrying the single mutations β1-T1A or β2-T1A, or both, are viable, even though one or two of the three distinct catalytic β subunits are disabled and carry remnants of their N-terminal propeptides4 (Table 1). These results indicate that the β1 and β2 proteolytic activities are not essential for cell survival. By contrast, the T1A mutation in subunit β5 has been reported to be lethal or nearly so113. Viability is restored if the β5-T1A subunit has its propeptide (pp) deleted but expressed separately in trans (β5-T1A pp trans), although substantial phenotypic impairment remains11516 (Table 1). Our present crystallographic analysis of the β5-T1A pp trans mutant demonstrates that the mutation per se does not structurally alter the catalytic active site and that the trans-expressed β5 propeptide is not bound in the β5 substrate-binding channel (Supplementary Fig. 1a). The extremely weak growth of the β5-T1A mutant pp cis described by Chen and Hochstrasser1 compared with the inviability reported by Heinemeyer et al.13 prompted us to analyse this discrepancy. Sequencing of the plasmids, testing them in both published yeast strain backgrounds and site-directed mutagenesis revealed that the β5-T1A mutant pp cis is viable, but suffers from a marked growth defect that requires extended incubation of 4–5 days for initial colony formation (Table 1 and Supplementary Methods). We also identified an additional point mutation K81R in subunit β5 that was present in the allele used in ref. 1. This single amino-acid exchange is located at the interface of the subunits α4, β4 and β5 (Supplementary Fig. 1b) and might weakly promote CP assembly by enhancing inter-subunit contacts. The slightly better growth of the β5-T1A-K81R mutant allowed us to solve the crystal structure of a yeast proteasome (yCP) with the β5-T1A mutation, which is discussed in the following section (for details see Supplementary Note 1). In the final steps of proteasome biogenesis, the propeptides are autocatalytically cleaved from the mature β-subunit domains1. For subunit β1, this process was previously inferred to require that the propeptide residue at position (-2) of the subunit precursor occupies the S1 specificity pocket of the substrate-binding channel formed by amino acid 45 (for details see Supplementary Note 2)5. Furthermore, it was observed that the prosegment forms an antiparallel β-sheet in the active site, and that Gly(-1) adopts a γ-turn conformation, which by definition is characterized by a hydrogen bond between Leu(-2)O and Thr1NH (ref. 5). Here we again analysed the β1-T1A mutant crystallographically but in addition determined the structures of the β2-T1A single and β1-T1A-β2-T1A double mutants (Protein Data Bank (PDB) entry codes are provided in Supplementary Table 1). In subunit β1, we found that Gly(-1) indeed forms a sharp turn, which relaxes on prosegment cleavage (Fig. 1a and Supplementary Fig. 2a). However, the γ-turn conformation and the associated hydrogen bond initially proposed is for geometric and chemical reasons inappropriate and would not perfectly position the carbonyl carbon atom of Gly(-1) for nucleophilic attack by Thr1. Regarding the β2 propeptide, Thr(-2) occupies the S1 pocket but is less deeply anchored compared with Leu(-2) in β1, which might be due to the rather large β2-S1 pocket created by Gly45. Thr(-2) positions Gly(-1)O via hydrogen bonding (∼2.8 Å) in a perfect trajectory for the nucleophilic attack by Thr1O (Fig. 1b and Supplementary Fig. 2b). Next, we examined the position of the β5 propeptide in the β5-T1A-K81R mutant. Surprisingly, Gly(-1) is completely extended and forces the histidine side chain at position (-2) to occupy the S2 instead of the S1 pocket, thereby disrupting the antiparallel β-sheet. Nonetheless, the carbonyl carbon of Gly(-1) would be ideally placed for nucleophilic attack by Thr1O (Fig. 1c and Supplementary Fig. 2c,d). As the K81R mutation is located far from the active site (Thr1C–Arg81C: 24 Å), any influence on propeptide conformation can be excluded. Instead, the plasticity of the β5 S1 pocket caused by the rotational flexibility of Met45 might prevent stable accommodation of His(-2) in the S1 site and thus also promote its immediate release after autolysis. Processing of β-subunit precursors requires deprotonation of Thr1OH; however, the general base initiating autolysis is unknown. Remarkably, eukaryotic proteasomal β5 subunits bear a His residue in position (-2) of the propeptide (Supplementary Fig. 3a). As histidine commonly functions as a proton shuttle in the catalytic triads of serine proteases17, we investigated the role of His(-2) in processing of the β5 propeptide by exchanging it for Asn, Lys, Phe and Ala. All yeast mutants were viable at 30 °C, but suffered from growth defects at 37 °C with the H(-2)N and H(-2)F mutants being most affected (Supplementary Fig. 3b and Table 1). In agreement, the chymotrypsin-like (ChT-L) activity of H(-2)N and H(-2)F mutant yCPs was impaired in situ and in vitro (Supplementary Fig. 3c). Structural analyses revealed that the propeptides of all mutant yCPs shared residual 2FO–FC electron densities. Gly(-1) and Phe/Lys(-2) were visualized at low occupancy, while Ala/Asn(-2) could not be assigned. This observation indicates a mixture of processed and unprocessed β5 subunits and partially impaired autolysis18, thereby excluding any essential role of residue (-2) as the general base. Next, we examined the effect of residue (-2) on the orientation of the propeptide by creating mutants that combine the T1A (K81R) mutation(s) with H(-2)L, H(-2)T or H(-2)A substitutions. Leu(-2) is encoded in the yeast β1 subunit precursor (Supplementary Fig. 3a); Thr(-2) is generally part of β2-propeptides (Supplementary Fig. 3a); and Ala(-2) was expected to fit the β5-S1 pocket without inducing conformational changes of Met45, allowing it to accommodate ‘β1-like' propeptide positioning. As expected from β5-T1A mutants, the yeasts show severe growth phenotypes, with minor variations (Supplementary Fig. 4a and Table 1). We determined crystal structures of the β5-H(-2)L-T1A, β5-H(-2)T-T1A and the β5-H(-2)A-T1A-K81R mutants (Supplementary Table 1). For the β5-H(-2)A-T1A-K81R variant, only the residues Gly(-1) and Ala(-2) could be visualized, indicating that Ala(-2) leads to insufficient stabilization of the propeptide in the substrate-binding channel (Supplementary Fig. 4d). By contrast, the prosegments of the β5-H(-2)L-T1A and the β5-H(-2)T-T1A mutants were significantly better resolved in the 2FO–FC electron-density maps yet not at full occupancy (Supplementary Fig. 4b,c and Supplementary Table 1), suggesting that the natural propeptide bearing His(-2) is most favourable. Nevertheless, both Leu(-2) and Thr(-2) were found to occupy the S1 specificity pocket formed by Met45 (Fig. 2a,b and Supplementary Fig. 4f–h). This result proves that the naturally occurring His(-2) of the β5 propeptide does not stably fit into the S1 site. Since Gly(-1) adopts the same position in both wild-type (WT) and mutant β5 propeptides, and since in all cases its carbonyl carbon is perfectly placed for nucleophilic attack by Thr1O (Fig. 2b), we propose that neither binding of residue (-2) to the S1 pocket nor formation of the antiparallel β-sheet is essential for autolysis of the propeptide. Next, we determined the crystal structure of a chimeric yCP having the yeast β1-propeptide replaced by its β5 counterpart18. Although we observed fragments of 2FO–FC electron density in the β1 active site, the data were not interpretable. Bearing in mind that in contrast to Thr(-2) in β2, Leu(-2) in subunit β1 is not conserved among species (Supplementary Fig. 3a), we created a β2-T(-2)V proteasome mutant. As proven by the β2-T1A crystal structures, Thr(-2) hydrogen bonds to Gly(-1)O. Although this interaction was not observed for the β5-H(-2)T-T1A mutant (Fig. 2c and Supplementary Fig. 4c,i), exchange of Thr(-2) by Val in β2, a conservative mutation regarding size but drastic with respect to polarity, was found to inhibit maturation of this subunit (Fig. 2d and Supplementary Fig. 4e,j). Notably, the 2FO–FC electron-density map displays a different orientation for the β2 propeptide than has been observed for the β2-T1A proteasome. In particular, Val(-2) is displaced from the S1 site and Gly(-1) is severely shifted (movement of the carbonyl oxygen atom of 3.8 Å), thereby preventing nucleophilic attack of Thr1 (Fig. 2d and Supplementary Fig. 4j,k). These results further confirm that correct positioning of the active-site residues and Gly(-1) is decisive for the maturation of the proteasome. Proton shuttling from the proteasomal active site Thr1OH to Thr1NH2 via a nucleophilic water molecule was suggested to initiate peptide-bond hydrolysis2910. However, in the immature particle Thr1NH2 is blocked by the propeptide and cannot activate Thr1O. Instead, Lys33NH2, which is in hydrogen-bonding distance to Thr1O (2.7 Å) in all catalytically active β subunits (Fig. 3a,b)29, was proposed to serve as the proton acceptor19. Owing to its likely protonation at neutral pH, however, it was assumed not to act as the general base259. A proposed catalytic tetrad model involving Thr1OH, Thr1NH2, Lys33NH2 and Asp17O, as well as a nucleophilic water molecule as the proton shuttle appeared to accommodate all possible views of the proteasomal active site8920. Twenty years later, with a plethora of yCP X-ray structures in hand, we decided to re-analyse the active site of the proteasome and to resolve the uncertainty regarding the nature of the general base. Mutation of β5-Lys33 to Ala causes a strongly deleterious phenotype, and previous structural and biochemical analyses confirmed that this is caused by failure of propeptide cleavage, and consequently, lack of ChT-L activity1413 (Fig. 4a, Supplementary Fig. 3b and Table 1; for details see Supplementary Note 1). The phenotype of the β5-K33A mutant was however less pronounced than for the β5-T1A-K81R yeast (Fig. 4a). This discrepancy in growth was traced to an additional point mutation L(-49)S in the β5-propeptide of the β5-K33A mutant (see also Supplementary Note 1). Structural comparison of the β5-L(-49)S-K33A and β5-T1A-K81R active sites revealed that mutation of Lys33 to Ala creates a cavity that is filled with Thr1 and the remnant propeptide. This structural alteration destroys active-site integrity and abolishes catalytic activity of the β5 active site4 (Supplementary Fig. 5a). Additional proof for the key function of Lys33 was obtained from the β5-K33A mutant, with the propeptide expressed separately from the main subunit (pp trans)15. The Thr1 N terminus of this mutant is not blocked by the propeptide, yet its catalytic activity is reduced by ∼83% (Supplementary Fig. 6b). Consistent with this, the crystal structure of the β5-K33A pp trans mutant in complex with carfilzomib only showed partial occupancy of the ligand at the β5 active sites (Supplementary Fig. 5b and Supplementary Table 1). Since no acetylation of the Thr1 N terminus was observed for the β5-K33A pp trans apo crystal structure416, the reduced reactivity towards substrates and inhibitors indicates that Lys33NH2, rather than Thr1NH2, deprotonates and activates Thr1OH. Furthermore, the crystal structure of the β5-K33A pp trans mutant without inhibitor revealed that Thr1O strongly coordinates a well-defined water molecule (∼2 Å; Fig. 3c and Supplementary Fig. 5c,d). This water hydrogen bonds also to Arg19O (∼3.0 Å) and Asp17O (∼3.0 Å), and thereby presumably enables residual activity of the mutant. Remarkably, the solvent molecule occupies the position normally taken by Lys33NH2 in the WT proteasome structure (Fig. 3c), further corroborating the essential role of Lys33 as the general base for autolysis and proteolysis. Conservative substitution of Lys33 by Arg delays autolysis of the β5 precursor and impairs yeast growth (for details see Supplementary Note 1). While Thr1 occupies the same position as in WT yCPs, Arg33 is unable to hydrogen bond to Asp17, thereby inactivating the β5 active site24 (Supplementary Fig. 5e). The conservative mutation of Asp17 to Asn in subunit β5 of the yCP also provokes a severe growth defect (Supplementary Note 1, Supplementary Fig. 6a and Table 1). Notably, only with the additional point mutation L(-49)S present in the β5 propeptide could we purify a small amount of the β5-D17N mutant yCP. As determined by crystallographic analysis, this mutant β5 subunit was partially processed (Table 1) but displayed impaired reactivity towards the proteasome inhibitor carfilzomib compared with the subunits β1 and β2, and with WT β5 (Supplementary Fig. 7a). In contrast to the cis-construct, expression of the β5 propeptide in trans allowed straightforward isolation and crystallization of the D17N mutant proteasome. The ChT-L activity of the β5-D17N pp in trans CP towards the canonical β5 model substrates N-succinyl-Leu-Leu-Val-Tyr-7-amino-4-methylcoumarin (Suc-LLVY-AMC) and carboxybenzyl-Gly-Gly-Leu-para-nitroanilide (Z-GGL-pNA) was severely reduced (Supplementary Fig. 6b), confirming that Asp17 is of fundamental importance for the catalytic activity of the mature proteasome. Even though the β5-D17N pp trans yCP crystal structure appeared identical to the WT yCP (Supplementary Fig. 7b), the co-crystal structure with the α′, β′ epoxyketone inhibitor carfilzomib visualized only partial occupancy of the ligand in the β5 active site (Supplementary Fig. 7a). This observation is consistent with a strongly reduced reactivity of β5-Thr1 and the crystal structure of the β5-D17N pp cis mutant in complex with carfilzomib. Autolysis and residual catalytic activity of the β5-D17N mutants may originate from the carbonyl group of Asn17, which albeit to a lower degree still can polarize Lys33 for the activation of Thr1. In agreement, an E17A mutant in the proteasomal β-subunit of the archaeon Thermoplasma acidophilum prevents autolysis and catalysis21. Strikingly, although the X-ray data on the β5-D17N mutant with the propeptide expressed in cis and in trans looked similar, there was a pronounced difference in their growth phenotypes observed (Supplementary Fig. 6a and Supplementary Fig. 7b). On the basis of these results, we propose that CPs from all domains of life use a catalytic triad consisting of Thr1, Lys33 and Asp/Glu17 for both autocatalytic precursor processing and proteolysis (Fig. 3d). This model is also consistent with the fact that no defined water molecule is observed in the mature WT proteasomal active site that could shuttle the proton from Thr1O to Thr1NH2. To explore this active-site model further, we exchanged the conserved Asp166 residue for Asn in the yeast β5 subunit. Asp166O is hydrogen-bonded to Thr1NH2 via Ser129OH and Ser169OH, and therefore was proposed to be involved in catalysis2. The β5-D166N pp cis yeast mutant is significantly impaired in growth and its ChT-L activity is drastically reduced (Supplementary Fig. 6a,b and Table 1). X-ray data on the β5-D166N mutant indicate that the β5 propeptide is hydrolysed, but due to reorientation of Ser129OH, the interaction with Asn166O is disrupted (Supplementary Fig. 8a). Instead, a water molecule is bound to Ser129OH and Thr1NH2 (Supplementary Fig. 8b), which may enable precursor processing. The hydrogen bonds involving Ser169OH are intact and may account for residual substrate turnover. Soaking the β5-D166N crystals with carfilzomib and MG132 resulted in covalent modification of Thr1 at high occupancy (Supplementary Fig. 8c). In the carfilzomib complex structure, Thr1O and Thr1N incorporate into a morpholine ring structure and Ser129 adopts its WT-like orientation. In the MG132-bound state, Thr1N is unmodified, and we again observe that Ser129 is hydrogen-bonded to a water molecule instead of Asn166. Whereas Asn can to some degree replace Asp166 due to its carbonyl group in the side chain, Ala at this position was found to prevent both autolysis and catalysis21. These results suggest that Asp166 and Ser129 function as a proton shuttle and affect the protonation state of Thr1N during autolysis and catalysis. Mutation of Thr1 to Cys inactivates the 20S proteasome from the archaeon T. acidophilum21. In yeast, this mutation causes a strong growth defect (Fig. 4a and Table 1), although the propeptide is hydrolysed, as shown here by its X-ray structure. In one of the two β5 subunits, however, we found the cleaved propeptide still bound in the substrate-binding channel (Fig. 4c). His(-2) occupies the S2 pocket like observed for the β5-T1A-K81R mutant, but in contrast to the latter, the propeptide in the T1C mutant adopts an antiparallel β-sheet conformation as known from inhibitors like MG132 (Fig. 4c–e and Supplementary Fig. 9b). On the basis of the phenotype of the T1C mutant and the propeptide remnant identified in its active site, we suppose that autolysis is retarded and may not have been completed before crystallization. Owing to the unequal positions of the two β5 subunits within the CP in the crystal lattice, maturation and propeptide displacement may occur at different timescales in the two subunits. Despite propeptide hydrolysis, the β5-T1C active site is catalytically inactive (Fig. 4b and Supplementary Fig. 9a). In agreement, soaking crystals with the CP inhibitors bortezomib or carfilzomib modifies only the β1 and β2 active sites, while leaving the β5-T1C proteolytic centres unmodified even though they are only partially occupied by the cleaved propeptide remnant. Moreover, the structural data reveal that the thiol group of Cys1 is rotated by 74° with respect to the hydroxyl side chain of Thr1 (Fig. 4f and Supplementary Fig. 9b). This presumably results from the larger radius of the sulfur atom compared with oxygen. Consequently, the hydrogen bond bridging the active-site nucleophile and Lys33 in WT CPs is broken with Cys1. Notably, the 2FO–FC electron-density map of the T1C mutant also indicates that Lys33NH2 is disordered. Together, these observations suggest that efficient peptide-bond hydrolysis requires that Lys33NH2 hydrogen bonds to the active site nucleophile. All proteasomes strictly employ threonine as the active-site residue instead of serine. To investigate the reason for this singularity, we analysed a β5-T1S mutant, which is viable but suffers from growth defects (Fig. 4a and Table 1). Activity assays with the β5-specific substrate Suc-LLVY-AMC demonstrated that the ChT-L activity of the T1S mutant is reduced by 40–45% compared with WT proteasomes depending on the incubation temperature (Fig. 4b and Supplementary Fig. 9c). By contrast, turnover of the substrate Z-GGL-pNA, used to monitor ChT-L activity in situ but in a less quantitative fashion, is not detectably impaired (Supplementary Fig. 9a). Crystal structure analysis of the β5-T1S mutant confirmed precursor processing (Fig. 4g), and ligand-complex structures with bortezomib and carfilzomib unambiguously corroborated the reactivity of Ser1 (Fig. 5). However, the apo crystal structure revealed that Ser1O is turned away from the substrate-binding channel (Fig. 4g). Compared with Thr1O in WT CP structures, Ser1O is rotated by 60°. This renders it unavailable for direct nucleophilic attack onto incoming substrates and first requires its reorientation, which is expected to delay substrate turnover. Because both conformations of Ser1O are hydrogen-bonded to Lys33NH2 (Fig. 4h), the relay system is capable of hydrolysing peptide substrates, albeit at lower rates compared with Thr1. The active-site residue Thr1 is fixed in its position, as its methyl group is engaged in hydrophobic interactions with Thr3 and Ala46 (Fig. 4h). Consequently, the hydroxyl group of Thr1 requires no reorientation before substrate cleavage and is thus more catalytically efficient than Ser1. In agreement, at an elevated growing temperature of 37 °C the T1S mutant is unable to grow (Fig. 4a). In vitro, the mutant proteasome is less susceptible to proteasome inhibition by bortezomib (3.7-fold) and carfilzomib (1.8-fold; Fig. 5). Nevertheless, inhibitor complex structures indicate identical binding modes compared with the WT yCP structures, with the same inhibitors2223. Notably, the affinity of the tetrapeptide carfilzomib is less impaired, as it is better stabilized in the substrate-binding channel than the dipeptide bortezomib, which lacks a defined P3 site and has only a few interactions with the surrounding protein. Hence, the mean residence time of carfilzomib at the active site is prolonged and the probability to covalently react with Ser1 is increased. Considered together, these results provide a plausible explanation for the invariance of threonine as the active-site nucleophile in proteasomes in all three domains of life, as well as in proteasome-like proteases such as HslV (ref. 24). The 20S proteasome CP is the major non-lysosomal protease in eukaryotic cells, and its assembly is highly organized. The β-subunit propeptides, particularly that of β5, are key factors that help drive proper assembly of the CP complex1. In addition, they prevent irreversible inactivation of the Thr1 N terminus by N-acetylation41516. By contrast, the prosegments of β subunits are dispensable for archaeal proteasome assembly, at least when heterologously expressed in Escherichia coli25. In eukaryotes, deletion of or failure to cleave the β1 and β2 propeptides is well tolerated513141516. However, removal of the β5 prosegment or any interference with its cleavage causes severe phenotypic defects113. These observations highlight the unique function and importance of the β5 propeptide as well as the β5 active site for maturation and function of the eukaryotic CP. Here we have described the atomic structures of various β5-T1A mutants, which allowed for the first time visualization of the residual β5 propeptide. Depending on the (-2) residue we observed various propeptide conformations, but Gly(-1) is in all structures perfectly located for the nucleophilic attack by Thr1O, although it does not adopt the tight turn observed for the prosegment of subunit β1. From these data we conclude that only the positioning of Gly(-1) and Thr1 as well as the integrity of the proteasomal active site are required for autolysis. In this regard, inappropriate N-acetylation of the Thr1 N terminus cannot be removed by Thr1O due to the rotational freedom and flexibility of the acetyl group. The propeptide needs some anchoring in the substrate-binding channel to properly position Gly(-1), but this seems to be independent of the orientation of residue (-2). Autolytic activation of the CP constitutes one of the final steps of proteasome biogenesis26, but the trigger for propeptide cleavage had remained enigmatic. On the basis of the numerous CP:ligand complexes solved during the past 18 years and in the current study, we provide a revised interpretation of proteasome active-site architecture. We propose a catalytic triad for the active site of the CP consisting of residues Thr1, Lys33 and Asp/Glu17, which are conserved among all proteolytically active eukaryotic, bacterial and archaeal proteasome subunits. Lys33NH2 is expected to act as the proton acceptor during autocatalytic removal of the propeptides19, as well as during substrate proteolysis, while Asp17O orients Lys33NH2 and makes it more prone to protonation by raising its pKa (hydrogen bond distance: Lys33NH3–Asp17O: 2.9 Å). Analogously to the proteasome, a Thr–Lys–Asp triad is also found in L-asparaginase27. Thus, specific protein surroundings can significantly alter the chemical properties of amino acids such as Lys to function as an acid–base catalyst28. In this new view of the proteasomal active site, the positively charged Thr1NH3-terminus hydrogen bonds to the amide nitrogen of incoming peptide substrates and stabilizes as well as activates them for the endoproteolytic cleavage by Thr1O (Fig. 3d). Consistent with this model, the positively charged Thr1 N terminus is engaged in hydrogen bonds with inhibitory compounds like fellutamide B (ref. 29), α-ketoamides30, homobelactosin C (ref. 31) and salinosporamide A (ref. 32). Furthermore, opening of the β-lactone compound omuralide2 by Thr1 creates a C3-hydroxyl group, whose proton originates from Thr1NH3. The resulting uncharged Thr1NH2 is hydrogen-bridged to the C3-OH group. In agreement, acetylation of the Thr1 N terminus irreversibly blocks hydrolytic activity1516, and binding of substrates is prevented for steric reasons. By acting as a proton donor during catalysis, the Thr1 N terminus may also favour cleavage of substrate peptide bonds (Fig. 3d). In all proteases, collapse of the tetrahedral transition state results in selective breakage of the substrate amide bond, while the covalent interaction between the substrate and the enzyme persists. Cleavage of the scissile peptide bond requires protonation of the emerging free amine, and in the proteasome, the Thr1 amine group is likely to assume this function. Analogously, Thr1NH3 might promote the bivalent reaction mode of epoxyketone inhibitors by protonating the epoxide moiety to create a positively charged trivalent oxygen atom that is subsequently nucleophilically attacked by Thr1NH2. During autolysis the Thr1 N terminus is engaged in a hydroxyoxazolidine ring intermediate (Fig. 3d), which is unstable and short-lived. Breakdown of this tetrahedral transition state releases the Thr1 N terminus that is protonated by aspartic acid 166 via Ser129OH to yield Thr1NH3. The residues Ser129 and Asp166 are expected to increase the pKa value of Thr1N, thereby favouring its charged state. Consistent with playing an essential role in proton shuttling, the mutation D166A prevents autolysis of the archaeal CP21 and the exchange D166N impairs catalytic activity of the yeast CP about 60%. The mutation D166N lowers the pKa of Thr1N, which is thus more likely to exist in the uncharged deprotonated state (Thr1NH2). This renders the N terminus less suitable to stabilize substrates and to protonate the first cleavage product during catalysis, although it favours its ability to act as a nucleophile. This interpretation agrees with the strongly reduced catalytic activity of the β5-D166N mutant on the one hand, and the ability to react readily with carfilzomib on the other. Hence, the proteasome can be viewed as having a second triad that is essential for efficient proteolysis. While Lys33NH2 and Asp17O are required to deprotonate the Thr1 hydroxyl side chain, Ser129OH and Asp166OH serve to protonate the N-terminal amine group of Thr1. In accord with the proposed Thr1–Lys33–Asp17 catalytic triad, crystallographic data on the proteolytically inactive β5-T1C mutant demonstrate that the interaction of Lys33NH2 and Cys1 is broken. Consequently, efficient substrate turnover or covalent modification by ligands is prevented. However, owing to Cys being a strong nucleophile, the propeptide can still be cleaved off over time. While only one single turnover is necessary for autolysis, continuous enzymatic activity is required for significant and detectable substrate hydrolysis. Notably, in the Ntn hydrolase penicillin acylase, substitution of the catalytic N-terminal Ser residue by Cys also inactivates the enzyme but still enables precursor processing33. To investigate why the CP specifically employs threonine as its active-site residue, we used a β5-T1S mutant of the yCP and characterized it biochemically and structurally. Activity assays with the β5-T1S mutant revealed reduced turnover of Suc-LLVY-AMC. We also observed slightly lower affinity of the β5-T1S mutant yCP for the Food and Drug Administration-approved proteasome inhibitors bortezomib and carfilzomib. Structural analyses support these findings with the T1S mutant and provide an explanation for the strict use of Thr residues in proteasomes. Thr1 is well anchored in the active site by hydrophobic interactions of its C methyl group with Ala46 (C), Lys33 (carbon side chain) and Thr3 (C). Notably, proteolytically active proteasome subunits from archaea, yeast and mammals, including constitutive, immuno- and thymoproteasome subunits, either encode Thr or Ile at position 3, indicating the importance of the C for fixing the position of the nucleophilic Thr1. In contrast to Thr1, the hydroxyl group of Ser1 occupies the position of the Thr1 methyl side chain in the WT enzyme, which requires its reorientation relative to the substrate to allow cleavage (Fig. 4g,h). Notably, in the threonine aspartase Taspase1, mutation of the active-site Thr234 to Ser also places the side chain in the position of the methyl group of Thr234 in the WT, thereby reducing catalytic activity34. Similarly, although the serine mutant is active, threonine is more efficient in the context of the proteasome active site. The greater suitability of threonine for the proteasome active site, which has been noted in biochemical as well as in kinetic studies35, constitutes a likely reason for the conservation of the Thr1 residue in all proteasomes from bacteria to eukaryotes. Site-directed mutagenesis was performed by standard techniques using oligonucleotides listed in Supplementary Table 2. The pre2/doa3 (β5) mutant alleles in the centromeric, TRP1- or LEU2-marked shuttle vectors YCplac22 and pRS315, respectively, were verified by sequencing and subsequently introduced into the yeast strains MHY784 (ref. 1) or YWH20 (ref. 13), which express WT PRE2 from a URA3-marked plasmid. Counter-selection against the URA3 marker with 5-fluoroorotic acid yielded strains expressing only the mutant forms of β5. The strain producing a processed β5-T1A variant and the β5 propeptide in trans is a derivative of YWH212 (ref. 15). It carries an additional deletion of the NAT1 gene to avoid N-acetylation of Ala1; this strain exhibits extremely slow growth rates and served for crystallographic analysis only. All strains used in this study are listed in Supplementary Table 3. Yeast strains were grown in 18-l cultures at 30 °C in YPD into early stationary phase, and the yCPs were purified according to published procedures36. In brief, 120 g yeast cells were solubilized in 150 ml of 50 mM KH2PO4/K2HPO4 buffer (pH 7.5) and disrupted with a French press. Cell debris were removed by centrifugation for 30 min at 21,000 r.p.m. (4 °C). The resulting supernatant was filtered and ammonium sulfate (saturated solution) was added to a final concentration of 30% (v/v). This solution was loaded on a Phenyl Sepharose 6 Fast Flow column (GE Healthcare) pre-equilibrated with 1 M ammonium sulfate in 20 mM KH2PO4/K2HPO4 (pH 7.5). CPs were eluted by applying a linear gradient from 1 to 0 M ammonium sulfate. Proteasome-containing fractions were pooled and loaded onto a hydroxyapatite column (Bio-Rad) equilibrated with 20 mM KH2PO4/K2HPO4 (pH 7.5). Elution of the CPs was achieved by increasing the phosphate buffer concentration from 20 to 500 mM. Anion-exchange chromatogaphy (Resource Q column (GE Healthcare), elution gradient from 0 to 500 mM sodium chloride in 20 mM Tris-HCl (pH 7.5)) and subsequent size-exclusion chromatography (Superose 6 10/300 GL (GE Healthcare), 20 mM Tris-HCl (pH 7.5) and 150 mM NaCl) resulted in pure CPs for crystallization and activity assays. ChT-L (β5) activity of CPs was monitored by fluorescence spectroscopy using the model substrate Suc-LLVY-AMC. Purified yCPs (66 nM in 100 mM Tris-HCl, pH 7.5) were incubated with 300 μM substrate for 1 h at room temperature or 37 °C. The reactions were stopped by diluting samples 1:10 in 20 mM Tris-HCl, pH 7.5. AMC fluorophores released by proteasomal activity were measured in triplicate with a Varian Cary Eclipse Fluorescence Spectrophotometer (Agilent Technologies) at λexc=360 nm and λem=460 nm. Purified yCPs were mixed with dimethylsulfoxide as a control or serial dilutions of inhibitor and incubated for 45 min at room temperature. A final concentration of yCP of 66 nM was used. After addition of the peptide substrate Suc-LLVY-AMC to a final concentration of 200 μM and incubation for 1 h at room temperature, the reaction was stopped by diluting the samples 1:10 in 20 mM Tris-HCl, pH 7.5. AMC fluorophores released by residual proteasomal activity were measured in triplicate at λexc=360 nm and λem=460 nm. Relative fluorescence units were normalized to the dimethylsulfoxide-treated control. The calculated residual activities were plotted against the logarithm of the applied inhibitor concentration and fitted with GraphPad Prism 5. The IC50 value, the ligand concentration that leads to 50% inhibition of the enzymatic activity, was deduced from the fitted data. Mutant yCPs were crystallized as previously described for WT 20S proteasomes3637. Crystals were grown at 20 °C using the hanging drop vapour diffusion method. Drops contained a 1:1 mixture of protein (40 mg ml) and reservoir solution (25 mM magnesium acetate, 100 mM 2-(N-morpholino)ethanesulfonic acid (MES; pH 6.8) and 9–13% (v/v) 2-methyl-2,4-pentanediol (MPD)). Crystals were cryoprotected by addition of 5 μl cryobuffer (20 mM magnesium acetate, 100 mM MES, pH 6.8, and 30% (v/v) MPD). Inhibitor complex structures were obtained by incubating crystals in 5 μl cryobuffer supplemented with bortezomib or carfilzomib at a final concentration of 1.5 mM for at least 8 h. Diffraction data were collected at the beamline X06SA at the Paul Scherrer Institute, SLS, Villigen, Switzerland (λ=1.0 Å). Evaluation of reflection intensities and data reduction were performed with the programme package XDS38. Molecular replacement was carried out with the coordinates of the yeast 20S proteasome (PDB entry code: 5CZ4) by rigid body refinements (REFMAC5; ref. 39). MAIN40 and COOT41 were used to build models. TLS (Translation/Libration/Screw) refinements finally yielded excellent Rwork and Rfree, as well as root mean squared deviation bond and angle values. The coordinates, proven to have good stereochemistry from the Ramachandran plots, were deposited in the RCSB Protein Data Bank (Supplementary Table 1). The coordinates for the yeast 20S proteasome deposited under the entry code 1RYP do not represent the WT yCP but the double-mutant β5-K33R β1-T1A. At the time of deposition (in 1997), these data were the best available on the yCP. As yCP structure determination has become routine today, and structure refinement procedures have significantly improved, we here provide coordinates for the WT yCP at 2.3 Å resolution (PDB entry code: 5CZ4). Furthermore, the structures of most mutant yCPs described in this work were determined in their apo and ligand-bound states. For mutants with proteolytically inactive β5 subunits, the best crystallographic data obtained are given. For ligands or propeptide segments that were only partially defined in the 2FO–FC electron-density map the occupancy was reduced (for details see Supplementary Table 1). Accession codes: Coordinates and structure factors have been deposited in the Protein Data Bank, www.pdb.org (for PDB entry codes see Supplementary Table 1). How to cite this article: Huber, E. M. et al. A unified mechanism for proteolysis and autocatalytic activation in the 20S proteasome. Nat. Commun. 7:10900 doi: 10.1038/ncomms10900 (2016).
PMC4937829	Visualizing chaperone-assisted protein folding	Challenges in determining the structures of heterogeneous and dynamic protein complexes have greatly hampered past efforts to obtain a mechanistic understanding of many important biological processes. One such process is chaperone-assisted protein folding, where obtaining structural ensembles of chaperone:substrate complexes would ultimately reveal how chaperones help proteins fold into their native state. To address this problem, we devised a novel structural biology approach based on X-ray crystallography, termed Residual Electron and Anomalous Density (READ). READ enabled us to visualize even sparsely populated conformations of the substrate protein immunity protein 7 (Im7) in complex with the E. coli chaperone Spy. This study resulted in a series of snapshots depicting the various folding states of Im7 while bound to Spy. The ensemble shows that Spy-associated Im7 samples conformations ranging from unfolded to partially folded and native-like states, and reveals how a substrate can explore its folding landscape while bound to a chaperone.High-resolution structural models of protein-protein interactions are critical for obtaining mechanistic insights into biological processes. However, many protein-protein interactions are highly dynamic, making it difficult to obtain high-resolution data. Particularly challenging are interactions of intrinsically or conditionally disordered sections of proteins with their partner proteins. Recent advances in X-ray crystallography and NMR spectroscopy continue to improve our ability to analyze biomolecules that exist in multiple conformations. X-ray crystallography has historically provided valuable information on small-scale conformational changes, but observing large-amplitude heterogeneous conformational changes often falls beyond the reach of current crystallographic techniques. NMR can theoretically be used to determine heterogeneous ensembles, but in practice, this proves to be very challenging. Despite the importance of understanding how proteins fold into their native state within the cell, our knowledge about this critical process remains limited. It is clear that molecular chaperones aid in protein folding. However, exactly how they facilitate the folding process is still being debated. Structural characterization of chaperone-assisted protein folding likely would help bring clarity to this question. Structural models of chaperone-substrate complexes have recently begun to provide information as to how a chaperone can recognize its substrate. However, the impact that chaperones have on their substrates, and how these interactions affect the folding process remain largely unknown. For most chaperones, it is still unclear whether the chaperone actively participates in and affects the folding of the substrate proteins, or merely provides a suitable microenvironment enabling the substrate to fold on its own. This is a truly fundamental question in the chaperone field, and one that has eluded the community largely because of the highly dynamic nature of the chaperone-substrate complexes. To address this question, we investigated the ATP-independent Escherichia coli periplasmic chaperone Spy. Spy prevents protein aggregation and aids in protein folding under various stress conditions, including treatment with tannin and butanol. We originally discovered Spy by its ability to stabilize the protein-folding model Im7 in vivo and recently demonstrated that Im7 folds while associated with Spy. The crystal structure of Spy revealed that it forms a thin α-helical homodimeric cradle. Crosslinking and genetic experiments suggested that Spy interacts with substrates somewhere on its concave side. By using a novel X-ray crystallography-based approach to model disorder in crystal structures, we have now determined the high-resolution ensemble of the dynamic Spy:Im7 complex. This work provides a detailed view of chaperone-mediated protein folding and shows how substrates like Im7 find their native fold while bound to their chaperones. We reasoned that to obtain crystals of complexes between Spy (domain boundaries in Supplementary Fig. 1) and its substrate proteins, our best approach was to identify crystallization conditions that yielded Spy crystals in the presence of protein substrates but not in their absence. We therefore screened crystallization conditions for Spy with four different substrate proteins: a fragment of the largely unfolded bovine α-casein protein, wild-type (WT) E. coli Im7, an unfolded variant of Im7 (L18A L19A L37A), and the N-terminal half of Im7 (Im76-45), which encompasses the entire Spy-binding portion of Im7. We found conditions in which all four substrates co-crystallized with Spy, but in which Spy alone did not yield crystals. Subsequent crystal washing and dissolution experiments confirmed the presence of the substrates in the co-crystals (Supplementary Fig. 2). The crystals diffracted to ~1.8 Å resolution. We used Spy:Im76-45 selenomethionine crystals for phasing with single-wavelength anomalous diffraction (SAD) experiments, and used this solution to build the well-ordered Spy portions of all four complexes. However, modeling of the substrate in the complex proved to be a substantial challenge, as the electron density of the substrate was discontinuous and fragmented. Even the minimal binding portion of Im7 (Im76-45) showed highly dispersed electron density (Fig. 1a). We hypothesized that the fragmented density was due to multiple, partially occupied conformations of the substrate bound within the crystal. Such residual density is typically not considered usable by traditional X-ray crystallography methods. Thus, we developed a new approach to interpret the chaperone-bound substrate in multiple conformations. To determine the structure of the substrate portion of these Spy:substrate complexes, we conceived of an approach that we term READ, for Residual Electron and Anomalous Density. We split this approach into five steps: (1) By using a well-diffracting Spy:substrate co-crystal, we first determined the structure of the folded domain of Spy and obtained high quality residual electron density within the dynamic regions of the substrate. (2) We then labeled individual residues in the flexible regions of the substrate with the strong anomalous scatterer iodine, which serves to locate these residues in three-dimensional space using their anomalous density. (3) We performed molecular dynamics (MD) simulations to generate a pool of energetically reasonable conformations of the dynamic complex and (4) applied a sample-and-select algorithm to determine the minimal set of substrate conformations that fit both the residual and anomalous density. (5) Finally, we validated the ensemble using multiple statistical tests. Importantly, even though we only labeled a subset of the residues in the flexible regions of the substrate with iodine, the residual electron density can provide spatial information on many of the other flexible residues. These two forms of data are therefore complementary: by labeling individual residues, one can locate them to specific points in space. The electron density then allowed us to connect the labeled residues of the substrate by confining the protein chain within regions of detectable density. In this way, the two forms of data together were able to describe multiple conformations of the substrate within the crystal. As described in detail below, we developed the READ method to uncover the ensemble of conformations that the Spy-binding domain of Im7 (i.e., Im76-45) adopts while bound to Spy. However, we believe that READ will prove generally applicable to visualizing heterogeneous and dynamic complexes that have previously escaped detailed structural analysis. To apply the READ technique to the folding mechanism employed by the chaperone Spy, we selected Im76-45 for further investigation because NMR data suggested that Im76-45 could recapitulate unfolded, partially folded, and native-like states of Im7 (Supplementary Fig. 3). Moreover, binding experiments indicated that Im76-45 comprises the entire Spy-binding region. To introduce the anomalous scatterer iodine, we replaced eight Im76-45 residues with the non-canonical amino acid 4-iodophenylalanine (pI-Phe). Its strong anomalous scattering allowed us to track the positions of these individual Im76-45 residues one at a time, potentially even if the residue was found in several locations in the same crystal. We then co-crystallized Spy and the eight Im76-45 peptides, each of which harbored an individual pI-Phe substitution at one distinct position, and collected anomalous data for all eight Spy:Im76-45 complexes (Fig. 1B, Supplementary Table 1 Supplementary Dataset 1, and Supplementary Table 2). Consistent with our electron density map, we found that the majority of anomalous signals emerged in the cradle of Spy, implying that this is the likely Im7 substrate binding site. Consistent with the fragmented density, however, we observed multiple iodine positions for seven of the eight substituted residues. Together, these results indicated that the Im7 substrate binds Spy in multiple conformations. To determine the structural ensemble that Im76-45 adopts while bound to Spy, we combined the residual electron density and the anomalous signals from our pI-Phe substituted Spy:Im76-45 complexes. To generate an accurate depiction of the chaperone-substrate interactions, we devised a selection protocol based on a sample-and-select procedure employed in NMR spectroscopy. This procedure iteratively constructs structural ensembles and then compares them to the experimental data. During each round of the selection, a genetic algorithm alters the ensemble and its agreement to the experimental data is re-evaluated. If successful, the selection identifies the smallest group of specific conformations that best fits the residual electron density and anomalous signals. The READ sample-and-select algorithm is diagrammed in Fig. 2. Prior to performing the selection, we generated a large and diverse pool of chaperone-substrate complexes using coarse-grained MD simulations in a pseudo-crystal environment (Fig. 2 and Supplementary Fig. 4). The coarse-grained simulations are based on a single-residue resolution model for protein folding and were extended here to describe Spy-Im76-45 binding events (Online Methods). The initial conditions of the binding simulations are not biased toward a particular conformation of the substrate or any specific chaperone-substrate interaction (Online Methods). Im76-45 binds and unbinds to Spy throughout the simulations. This strategy allows a wide range of substrate conformations to interact with the chaperone. From the MD simulations, we extracted ~10,000 diverse Spy:Im76-45 complexes to be used by the ensuing selection. Each complex within this pool comprises one Spy dimer bound to a single Im76-45 substrate. This pool was then used by the selection algorithm to identify the minimal ensemble that best satisfies both the residual electron and anomalous crystallographic data. The anomalous scattering portion of the selection uses our basic knowledge of pI-Phe geometry: the iodine is separated from its respective Cα atom in each coarse-grained conformer by 6.5 Å. The selection then picks ensembles that best reproduce the collection of iodine anomalous signals. Simultaneously, it uses the residual electron density to help choose ensembles. To make the electron density selection practical, we needed to develop a method to rapidly evaluate the agreement between the selected sub-ensembles and the experimental electron density on-the-fly during the selection procedure. To accomplish this task, we generated a compressed version of the experimental 2mFo−DFc electron density map for use in the selection. This process provided us with a target map that the ensuing selection tried to recapitulate. To reduce the extent of 3D space to be explored, this compressed map was created by only using density from regions of space significantly sampled by Im76-45 in the Spy:Im76-45 MD simulations. For each of the ~10,000 complexes in the coarse-grained MD pool, the electron density at the Cα positions of Im76-45 was extracted and used to construct an electron density map (Online Methods). These individual electron density maps from the separate conformers could then be combined (Fig. 2) and compared to the averaged experimental electron density map as part of the selection algorithm. This approach allowed us to simultaneously use both the iodine anomalous signals and the residual electron density in the selection procedure. The selection resulted in small ensembles from the MD pool that best fit the READ data (Fig. 1c,d). Before analyzing the details of the Spy:Im76-45 complex, we first engaged in a series of validation tests to verify the ensemble and selection procedure (Supplementary Note 1, Figures 1c,d, Supplemental Figures 5-7). Combined, these validation tests confirmed that the selection procedure and selected six-member ensemble recapitulate the experimental data. Of note, the final six-membered ensemble was the largest ensemble that could simultaneously decrease the RFree and pass the 10-fold cross-validation test. This ensemble depicts the conformations that the substrate Im76-45 adopts while bound to the chaperone Spy (Fig. 3 Supplementary Movie 1, and Table 1). Our results showed that by using this novel READ approach, we were able to obtain structural information about the dynamic interaction of a chaperone with its substrate protein. We were particularly interested in finding answers to one of the most fundamental questions in chaperone biology—how does chaperone binding affect substrate structure and vice versa. By analyzing the individual structures of the six-member ensemble of Im76-45 bound to Spy, we observed that Im76-45 takes on several different conformations while bound. We found these conformations to be highly heterogeneous and to include unfolded, partially folded, and native-like states (Fig. 3). The ensemble primarily encompasses Im76-45 laying diagonally within the Spy cradle in several different orientations, but some conformations traverse as far as the tips or even extend over the side of the cradle (Figs. 3,4a). We constructed a contact map of the complex, which shows the frequency of interactions for chaperone-substrate residue pairs (Fig. 4). We found that the primary interaction sites on Spy reside at the N and C termini (Arg122, Thr124, and Phe29) as well as on the concave face of the chaperone (Arg61, Arg43, Lys47, His96, and Met46). The Spy-contacting residues comprise a mixture of charged, polar, and hydrophobic residues. Surprisingly, we noted that in the ensemble, Im76-45 interacts with only 38% of the hydrophobic residues in the Spy cradle, but interacts with 61% of the hydrophilic residues in the cradle. This mixture suggests the importance of both electrostatic and hydrophobic components in binding the Im76-45 ensemble. With respect to the substrate, we observed that nearly every residue in Im76-45 is in contact with Spy (Fig. 4a). However, we did notice that despite this uniformity, regions of Im76-45 preferentially interact with different regions in Spy (Fig. 4b). For example, the N-terminal half of Im76-45 binds more consistently in the Spy cradle, whereas the C-terminal half predominantly binds to the outer edges of Spy’s concave surface. Not unexpectedly, we found that as Im76-45 progresses from the unfolded to the native state, its interactions with Spy shift accordingly. Whereas the least-folded Im76-45 pose in the ensemble forms the most hydrophobic contacts with Spy (Fig. 3), the two most-folded conformations form the fewest hydrophobic contacts (Fig. 3). This shift in contacts is likely due to hydrophobic residues of Im76-45 preferentially forming intra-molecular contacts upon folding (i.e., hydrophobic collapse), effectively removing themselves from the interaction sites. The diversity of conformations and binding sites observed here emphasizes the dynamic and heterogeneous nature of the chaperone-substrate ensemble. Although we do not yet have time resolution data of these various snapshots of Im76-45, this ensemble illustrates how a substrate samples its folding landscape while bound to a chaperone. Comparing the structure of Spy in its substrate-bound and apo states revealed that the Spy dimer also undergoes significant conformational changes upon substrate binding (Fig. 5a and Supplementary Movie 2). Upon substrate binding, the Spy dimer twists 9° about its center relative to its apo form. This twist yields asymmetry and results in substantially different interaction patterns in the two Spy monomers (Fig. 4b). It is possible that this twist serves to increase heterogeneity in Spy by providing more binding poses. Additionally, we observed that the linker region (residues 47–57) of Spy, which participates in substrate interaction, becomes mostly disordered upon binding the substrate. This increased disorder might explain how Spy is able to recognize and bind different substrates and/or differing conformations of the same substrate. Importantly, we observed the same structural changes in Spy regardless of which of the four substrates was bound (Fig. 5b, Table 1). The RMSD between the well-folded sections of Spy in the four chaperone-substrate complexes was very small, less than 0.3 Å. Combined with competition experiments showing that the substrates compete in solution for Spy binding (Fig. 5c and Supplementary Fig. 8), we conclude that all the tested substrates share the same overall Spy binding site. To shed light on how chaperones interact with their substrates, we developed a novel structural biology method (READ) and applied it to determine a conformational ensemble of the chaperone Spy bound to substrate. As a substrate, we used Im76-45, the chaperone-interacting portion of the protein-folding model protein Im7. In the chaperone-bound ensemble, Im76-45 samples unfolded, partially folded, and native-like states. The ensemble provides an unprecedented description of the conformations that a substrate assumes while exploring its chaperone-associated folding landscape. This substrate-chaperone ensemble helps accomplish the longstanding goal of obtaining a detailed view of how a chaperone aids protein folding. We recently showed that Im7 can fold while remaining continuously bound to Spy. The high-resolution ensemble obtained here now provides insight into exactly how this occurs. The structures of our ensemble agree well with lower-resolution crosslinking data, which indicate that chaperone-substrate interactions primarily occur on the concave surface of Spy. The ensemble suggests a model in which Spy provides an amphipathic surface that allows substrate proteins to assume different conformations while bound to the chaperone. This model is consistent with previous studies postulating that the flexible binding of chaperones allows for substrate protein folding. The amphipathic concave surface of Spy likely facilitates this flexible binding and may be a crucial feature for Spy and potentially other chaperones, allowing them to bind multiple conformations of many different substrates. In contrast to Spy’s binding hotspots, Im76-45 displays substantially less specificity in its binding sites. Nearly all Im76-45 residues come in contact with Spy. Unfolded substrate conformers interact with Spy through both hydrophobic and hydrophilic interactions, whereas the binding of native-like states is mainly hydrophilic. This trend suggests that complex formation between an ATP-independent chaperone and its unfolded substrate may initially involve hydrophobic interactions, effectively shielding the exposed aggregation-sensitive hydrophobic regions in the substrate. Once the substrate begins to fold within this protected environment, it progressively buries its own hydrophobic residues, and its interactions with the chaperone shift towards becoming more electrostatic. Notably, the most frequent contacts between Spy and Im76-45 are charge-charge interactions. The negatively charged Im7 residues Glu21, Asp32, and Asp35 reside on the surface of Im7 and form interactions with Spy’s positively charged cradle in both the unfolded and native-like states. Residues Asp32 and Asp35 are close to each other in the folded state of Im7. This proximity likely causes electrostatic repulsion that destabilizes Im7’s native state. Interaction with Spy’s positively-charged residues likely relieves the charge repulsion between Asp32 and Asp35, promoting their compaction into a helical conformation. As inter-molecular hydrophobic interactions between Spy and the substrate become progressively replaced by intra-molecular interactions within the substrate, the affinity between chaperone and substrates could decrease, eventually leading to release of the folded client protein. Recently, we employed a genetic selection system to improve the chaperone activity of Spy. This selection resulted in “Super Spy” variants that were more effective at both preventing aggregation and promoting protein folding. In conjunction with our bound Im76-45 ensemble, these mutants now allowed us to investigate structural features important to chaperone function. Previous analysis revealed that the Super Spy variants either bound Im7 tighter than WT Spy, increased chaperone flexibility as measured via H/D exchange, or both. Our ensemble revealed that two of the Super Spy mutations (H96L and Q100L) form part of the chaperone contact surface that binds to Im76-45 (Fig. 4a). Moreover, our co-structure suggests that the L32P substitution, which increases Spy’s flexibility, could operate by unhinging the N-terminal helix and effectively expanding the size of the disordered linker. This possibility is supported by the Spy:substrate structures, in which the linker region becomes more flexible compared to the apo state (Fig. 6a). This expansion would increase the structural plasticity for substrate binding. By sampling multiple conformations, this linker region may allow diverse substrate conformations to be accommodated. Other Super Spy mutations (F115I and F115L) caused increased flexibility but not tighter substrate binding. This residue does not directly contact Im76-45 in our READ-derived ensemble. Instead, when Spy is bound to substrate, F115 engages in close CH⋯π hydrogen bonds with Tyr104 (Fig. 6b). This interaction presumably reduces the mobility of the C-terminal helix. The F115I/L substitutions would replace these hydrogen bonds with hydrophobic interactions that have little angular dependence. As a result, the C-terminus, and possibly also the flexible linker, is likely to become more flexible and thus more accommodating of different conformations of substrates. Overall, comparison of our ensemble to the Super Spy variants provides specific examples to corroborate the importance of conformational flexibility in chaperone-substrate interactions. Despite extensive studies, exactly how complex chaperone machines help proteins fold remains controversial. Our study indicates that the chaperone Spy employs a simple surface binding approach that allows the substrate to explore various conformations and form transiently favorable interactions while being protected from aggregation. We speculate that many other chaperones could utilize a similar strategy. ATP and co-chaperone dependencies may have emerged later through evolution to better modulate and control chaperone action. In addition to insights into chaperone function, this work presents a new method for determining heterogeneous structural ensembles via a hybrid methodology of X-ray crystallography and computational modeling. Heterogeneous dynamic complexes or disordered regions of single proteins, once considered solely approachable by NMR spectroscopy, can now be visualized through X-ray crystallography. Consequently, this technique could enable structural characterization of many important dynamic and heterogeneous biomolecular systems. For computational methods, including simulations of Spy-substrate interactions, binning the residual Im7 electron density, ensemble selection, validation tests, and contact map generation, please see Supplementary Note 1. To facilitate crystallization, we used Spy 29-124, a truncated Spy version that removes the unstructured N- and C-terminal tails (full length Spy is 138 amino acids). To determine if these alterations impact Spy’s chaperone activity in vitro, we performed in vitro chaperone activity assays and found that they had no significant effect; these deletions also had only a minor effect on Spy’s ability to stabilize Im7 in vivo (Supplementary Fig. 1). The in vitro activity of Spy 29-124 was assessed using the aldolase refolding assay as previously described. Briefly, in the denaturing step, 100 μM aldolase was denatured in buffer containing 6.6 M GdmCl, 40 mM HEPES pH 7.5, and 50 mM NaCl overnight at 22 °C (room temperature). In the refolding step, denatured aldolase was diluted to 3 μM in refolding buffer (40 mM HEPES, 150 mM NaCl, 5 mM DTT pH 7.5) in the presence of 6 μM WT Spy or Spy 29-124 (Spy:aldolase = 2:1). As a control, an identical experiment without Spy added was also performed. The refolding temperature was 37 °C with continuous shaking. The refolding status was monitored at different time points (1 min, 4 min, 10 min, and 20 min) and tested by diluting the refolding sample by 15-fold into the reaction buffer (0.15 mM NADH, 2 mM F1,6-DP, 1.8 U/ml GDH/TPI, 40 mM HEPES, and 150 mM NaCl pH 7.5) at 28 °C. The absorbance was monitored for 1.5 min at 340 nm. The percentage refolding was calculated and averaged over three repeats. To determine the in vivo activity of the Spy mutants, the quantity of the unstable Im7 variant L53A I54A expressed in the periplasm was compared during Spy variant co-expression as previously described. Plasmid Spy (pTrc-spy) was used as the template for the construction of the variant plasmids of Spy for in vivo chaperone activity measurement (Supplementary Table 3). To use the native signal sequence of spy for the periplasmic export of the Spy variants, an NheI site was first introduced between the signal sequence and the mature protein coding region of Spy. The vector was then digested with NheI and BamHI, purified, and ligated with the linear fragments corresponding to truncated sequences (21–130, 24–130, 27–130, 30–130, and 33–130) of Spy. Cells containing a strain that expressed the unstable Im7 mutant IL53A I54A (pCDFTrc-ssIm7L53A I54A) were transformed with plasmids that expressed either WT or one of the five truncated Spy mutants and grown to mid-log phase in LB medium at 37 °C. Im7 L53A I54A and Spy expression were induced with various concentrations of IPTG for 2 h to compare the in vivo chaperone activity of WT Spy and the truncated Spy mutants at similar expression levels. Periplasmic fractions were prepared as previously described and were separated on 16% Tricine gel (Life Technologies Inc.). The bands corresponding to Spy and the C-terminal His-tagged Im7 were either directly visualized on Coomassie stained gels or determined by western blot using anti-His antibody (Abcam ab1187; validation provided on manufacturer’s website). The gene for spy 29-124 was amplified from plasmid pET28sumo-spy with primer 1 (5′-CGC GGG ATC CTT CAA AGA CCT GAA CCT GAC CG-3′) and primer 2 (5′-CGC GCT CGA GTT ATG TCA GAC GCT TCT CAA AAT TAG C-3′), and was cloned into pET28sumo via BamHI and XhoI sites. The H96L variant was made by Phusion site-directed mutagenesis (New England Biolabs). WT and H96L Spy 29-124 were expressed and purified as described previously with the exception that Ni-HisTrap columns (GE Healthcare) were utilized instead of the Ni-NTA beads and mini-chromatography column. ULP1 cleavage occurred following elution from the Ni-HisTrap column overnight at 4 °C while dialyzing to 40 mM Tris, 300 mM NaCl, pH 8.0. After dialysis, Spy was passed over the HisTrap column to remove the cleaved SUMO tag (20 mM imidazole was left over from the dialysis). Cleavage of the SUMO tag leaves a single serine in position 28 of Spy. The flow-through was then concentrated and diluted 5 times with 20 mM Tris, pH 8 for further purification on a HiTrap Q column. Spy has an isoelectric point of 9.5 and therefore was collected in the flow-through. The flow-through containing Spy was concentrated and diluted 5-fold with 50 mM sodium phosphate at pH 6.5 and passed over a HiTrap SP column. Spy was then eluted with a gradient from 0 M to 1 M NaCl. Re-buffering to the final reaction buffer was accomplished by gel filtration, passing the pooled and concentrated fractions containing Spy over a HiLoad 75 column in 40 mM HEPES, 100 mM NaCl, pH 7.5. Fractions containing Spy were then concentrated, frozen in liquid nitrogen, and stored at −80 °C. WT Im7, Im7 L18A L19A L37A H40W, and Im7 L18A L19A L37A were purified by the same protocol as Spy, but without the SP column step. In addition to WT Im7 and these various Im7 mutants, co-crystallization experiments extensively utilized Im76-45, a minimal Spy-binding segment that encompasses the first two helices of Im7 and contains 46% of the total Im7 sequence. It displays partial helicity when free in solution (Supplementary Fig. 3). The 6-45 portion of Im7 (H2N-SISDYTEAEFVQLLKEIEKENVAATDDVLD VLLEHFVKIT-OH), 4-iodophenlyalanine variants, and a peptide corresponding to a portion of bovine alpha casein S1 148-177 (Ac-ELFRQFYQLDAYPSGAWYYVPLGTQYTDAP-amide) were obtained from New England Peptide at ≥ 95% purity. Anomalous signals for residues E12, E14, L19, and E21 substitutions were determined using a peptide containing Im7 6–26, which was also obtained from New England Peptide at ≥ 95% purity. Co-crystals of WT Spy 29-124 and Spy H96L 29-124 in complex with Im7 variants and casein were grown by vapor diffusion. 25–130 mg/ml dimer Spy was incubated with various Im7 or casein substrates at concentrations ranging from equimolar to three-fold excess substrate in 22%–33% PEG 3000, 0.88–1.0 M imidazole pH 8.0, and 40–310 mM zinc acetate at 20 °C. Crystals were flash frozen in liquid nitrogen using 35% PEG 3000 as a cryo-protectant. It is worthwhile to note that the flash freezing could somewhat bias the conformations observed in the crystal structure. However, we chose to freeze the crystal to provide us with the maximum capability to identify and interpret the iodine anomalous signals. Crystals were washed by sequential transfer between three to six 2 μl drops of mother liquor, incubating in each wash solution for 2–10 s in an effort to remove all surface bound and precipitated substrate protein before being dissolved for visualization by SDS-PAGE. Before loading, samples were boiled for 10 min in reducing loading buffer, and then loaded onto 16% Tricine gels. Wash samples and dissolved crystal samples were analyzed by Lumetein staining (Biotium) and Flamingo staining (Bio-Rad) per manufacturer’s instructions, and imaged using a FluorChem M Imager (ProteinSimple). Data were collected at the LS-CAT beamlines at the Advanced Photon Source at 100 K. SeMet and native Spy:Im76-45 crystals were collected at 12.7 keV and 9.7 keV, respectively. Spy:Casein 148-177, and Spy H96L:WT Im7 crystals were collected also collected at 12.7 keV. Data integration and scaling were performed with iMosflm and AIMLESS, respectively. As molecular replacement attempts using the previously published apo Spy structures (PDB IDs: 3O39 and 3OEO) were unsuccessful, the Spy:Im76-45 complex was solved using Se-SAD phasing with SeMet-Spy, followed by density modification and initial model building by AutoSol in Phenix. The initial model was completed and refined using the native Spy:Im76-45 complex data. The rest of the structures were built using the native Spy:Im76-45 structure as a molecular replacement search model. Refinements, including TLS refinement, were performed using COOT and Phenix. All refined structures were validated using the Molprobity server, with Clashscores ranking better than the 90 percentile for all structures. Structural figures were rendered using PyMOL and UCSF Chimera, and movies generated using UCSF Chimera. Several partially occupied zinc atoms were observed in the crystal structure. Although some of these zinc atoms could also potentially modelled as water molecules, doing so resulted in an increase in the RFree. Additionally, a section of density near His A96 that is potentially partially occupied by a combination of water, Spy linker region, and possibly zinc, was modelled as containing water molecules. Spy H96L:Im76-45 was employed for iodine anomalous scattering experiments due to increased robustness and reproducibility of the crystals. The expected anomalous scatterers in the structures were S in methionine residues of Spy, Zn from the crystallization buffer, and I in the single pI-Phe residue of each synthetic Im76-45 peptide. Each I site is expected to be partially occupied as Im76-45 had diffuse density corresponding to multiple, partially occupied conformers; the Zn sites also may be partially occupied. To identify I, S, and Zn atomic positions using anomalous scattering, datasets were collected at 6.5 keV and 14.0 keV at 100 K using the ID-D beamline at LS-CAT. Anomalous difference maps for initial anomalous signal screening were calculated with phases from a molecular replacement search using the native Spy:Im76-45 (with no Im76-45 built in) complex as the search model. Anomalous difference maps calculated with the 14.0 keV data were used as controls to distinguish iodine from zinc atoms, as the iodine and zinc anomalous scattering factors are comparable at 14.0 keV, whereas at 6.5 keV, f″ is ~9-fold greater for iodine than for zinc. Anomalous differences were also collected and analyzed for a crystal of WT Spy 29-124:Im76-45 containing no iodine. The resulting anomalous difference map was inspected for peaks corresponding to sulfur, which were then excluded when selecting iodine peaks. Also, peaks that overlapped with Spy in the crystal lattice were excluded from analysis. As an initial screen for placing iodine atoms in the 6.5 keV anomalous difference maps, the median methionine sulfur signal was used as a cutoff for each individual map to control for varying data quality between crystals. Then, all anomalous atoms were refined in Phenix using anomalous group refinement. Refined B-factor of placed iodine ions was then used to estimate the positional fluctuation of the anomalous signals. This positional fluctuation was used as estimated error in the ensuing selection. A summary of all the anomalous signal heights (Supplementary Table 1) and anomalous difference maps (Supplemental Dataset 1) are displayed at varying contour levels for maximum clarity of iodine and methionine peak heights. The dissociation constant of Im76-45 was determined via a fluorescence-based competition experiment with Im7 L18A L19A L37A H40W, and its ability to compete with casein 148-177 for Spy binding was tested. Im7 L18A L19A L37A H40W was chosen for competition experiments due to its tight binding (Supplementary Fig. 8) and substantial fluorescence change upon binding. This mutant binds to Spy tighter than Im7 L18A L19A L37A. 10 μM Spy 29-124 dimer was mixed with 10 μM Im7 L18A L19A L37A H40W or casein 148-177 to form a 1:1 complex in a buffer containing 40 mM HEPES pH 7.5 and 100 mM NaCl at 22 °C. Complex formation was monitored with a QuantaMaster 400 (Photon Technology International) using the tryptophan fluorescence of Im7 L18A L19A L37A H40W. Naturally tryptophan-free Im76-45 was then titrated into the complex to compete with Im7 L18A L19A L37A H40W for Spy binding. The observed fluorescence intensity at 350 nm was plotted as a function of the logarithm of the Im76-45 or casein 148-177 concentration. The data were fit for a one-site-binding competition model (OriginLab 9.1): y=A2+A1-A21+10x-logx0 where A1 and A2 are the maximum and minimum asymptotes, respectively, and x is the concentration of Im76-45. x0 is the apparent KD for Im76-45 based on its ability to compete with Im7 L18A L19A L37A H40W. Using the KD of Im7 L18A L19A L37A H40W binding to Spy 29-124, we then calculated the KD for Im76-45 binding to Spy 29-124 using the Cheng-Prusoff equation: Ki=x01+L/KD where L is the concentration of Im7 L18A L19A L37A H40W and KD is the dissociation constant for Im7 L18A L19A L37A H40W binding to Spy. Due to interaction between higher oligomer states of Im76-45 and casein 148-177 (Supplementary Fig. 8), the competition curve was unable to be fit for casein 148-177 competing with Im76-45. The stoichiometry of binding of casein 148-177 and Spy was determined by tryptophan fluorescence of the casein upon Spy 29-124 addition. Increasing concentrations of Spy 29-124 were titrated to 20 μM of casein 148-177 in 40 mM HEPES (pH 7.5), 100 mM NaCl, at 22 °C. Complex formation was monitored with a QuantaMaster 400 (Photon Technology International) using the tryptophan fluorescence of casein 148-177. The observed fluorescence intensity at 339 nm was plotted as a function of the Spy 29-124 dimer concentration and fit with a quadratic equation using Origin 9.1 (OriginLab). To determine the dissociation constant, increasing concentrations of Spy 29-124 were titrated to 0.25 μM of casein in 40 mM HEPES (pH 7.5), 100 mM NaCl, at 22 °C. Complex formation was monitored with a QuantaMaster 400 (Photon Technology International) using the tryptophan fluorescence of casein 148-177. The observed fluorescence intensity at 339 nm was corrected for dilution due to the titration and then plotted as a function of the Spy 29-124 dimer concentration. The data were fit using a square hyperbola function in Origin 9.1 (OriginLab): F=Fmax×LKD+L+C where F is the recorded fluorescence signal, Fmax is the maximum fluorescence reached upon saturation of the complex, L is the concentration of free Spy in solution, KD is the dissociation constant, and C is a parameter for the offset. The calculated KD is an average of three independent repetitions. The measured dissociation constants for the different substrates ranged from 0.1 to 1 μM. Spy 29-124 and Im7-L18A L19A L37A H40W were dialyzed overnight against 40 mM HEPES, 100 mM NaCl, pH 7.5. 165 μM Spy dimer was loaded into a syringe and titrated into a cell containing 15 μM Im7 L18A L19 AL37A H40W at 25 °C in an iTC200 (Malvern Instruments) with an injection interval of 120 s and an initial delay time of 60 s. The solution was stirred at 1000 rpm, and the reference power was set to 6 μcal s in high feedback mode. Data analysis was conducted using a plugin for Origin 7 (OriginLab), the software provided by the manufacturer. Sedimentation velocity experiments for the Im76-45 and the bovine α-S1-casein peptide were performed using a Beckman Proteome Lab XL-I analytical ultracentrifuge (Beckman Coulter). Both peptides were first dialyzed against 40 mM HEPES, 100 mM NaCl, pH 7.5, then diluted to a concentration of 10 μM using the dialysis buffer. Samples were loaded into cells containing standard sector shaped 2-channel Epon centerpieces with 1.2 cm path-length (Beckman Coulter) and equilibrated to 22 °C for at least 1 h prior to sedimentation. All samples were spun at 48,000 rpm in a Beckman AN-50 Ti rotor, and the sedimentation of the protein was monitored continuously using interference optics, since the Im76-45 does not absorb strongly at 280 nm. Data analysis was conducted with SEDFIT (version 14.1), using the continuous c(s) distribution model. The confidence level for the maximum entropy (ME) regularization was set to 0.95. Buffer density and viscosity were calculated using SEDNTERP (http://sednterp.unh.edu/).
PMC4848761	Predictive features of ligand‐specific signaling through the estrogen receptor	Some estrogen receptor‐α (ERα)‐targeted breast cancer therapies such as tamoxifen have tissue‐selective or cell‐specific activities, while others have similar activities in different cell types. To identify biophysical determinants of cell‐specific signaling and breast cancer cell proliferation, we synthesized 241 ERα ligands based on 19 chemical scaffolds, and compared ligand response using quantitative bioassays for canonical ERα activities and X‐ray crystallography. Ligands that regulate the dynamics and stability of the coactivator‐binding site in the C‐terminal ligand‐binding domain, called activation function‐2 (AF‐2), showed similar activity profiles in different cell types. Such ligands induced breast cancer cell proliferation in a manner that was predicted by the canonical recruitment of the coactivators NCOA1/2/3 and induction of the GREB1 proliferative gene. For some ligand series, a single inter‐atomic distance in the ligand‐binding domain predicted their proliferative effects. In contrast, the N‐terminal coactivator‐binding site, activation function‐1 (AF‐1), determined cell‐specific signaling induced by ligands that used alternate mechanisms to control cell proliferation. Thus, incorporating systems structural analyses with quantitative chemical biology reveals how ligands can achieve distinct allosteric signaling outcomes through ERα.Many drugs are small‐molecule ligands of allosteric signaling proteins, including G protein‐coupled receptors (GPCRs) and nuclear receptors such as ERα. These receptors regulate distinct phenotypic outcomes (i.e., observable characteristics of cells and tissues, such as cell proliferation or the inflammatory response) in a ligand‐dependent manner. Small‐molecule ligands control receptor activity by modulating recruitment of effector enzymes to distal regions of the receptor, relative to the ligand‐binding site. Some of these ligands achieve selectivity for a subset of tissue‐ or pathway‐specific signaling outcomes, which is called selective modulation, functional selectivity, or biased signaling, through structural mechanisms that are poorly understood (Frolik et al, 1996; Nettles & Greene, 2005; Overington et al, 2006; Katritch et al, 2012; Wisler et al, 2014). For example, selective estrogen receptor modulators (SERMs) such as tamoxifen (Nolvadex; AstraZeneca) or raloxifene (Evista; Eli Lilly) (Fig 1A) block the ERα‐mediated proliferative effects of the native estrogen, 17β‐estradiol (E2), on breast cancer cells, but promote beneficial estrogenic effects on bone mineral density and adverse estrogenic effects such as uterine proliferation, fatty liver, or stroke (Frolik et al, 1996; Fisher et al, 1998; McDonnell et al, 2002; Jordan, 2003). Chemical structures of some common ERα ligands. BSC, basic side chain. E2‐rings are numbered A‐D. The E‐ring is the common site of attachment for BSC found in many SERMS.ERα domain organization lettered, A‐F. DBD, DNA‐binding domain; LBD, ligand‐binding domain; AF, activation functionSchematic illustration of the canonical ERα signaling pathway.Linear causality model for ERα‐mediated cell proliferation.Branched causality model for ERα‐mediated cell proliferation. Chemical structures of some common ERα ligands. BSC, basic side chain. E2‐rings are numbered A‐D. The E‐ring is the common site of attachment for BSC found in many SERMS. ERα domain organization lettered, A‐F. DBD, DNA‐binding domain; LBD, ligand‐binding domain; AF, activation function Schematic illustration of the canonical ERα signaling pathway. Linear causality model for ERα‐mediated cell proliferation. Branched causality model for ERα‐mediated cell proliferation. ERα contains structurally conserved globular domains of the nuclear receptor superfamily, including a DNA‐binding domain (DBD) that is connected by a flexible hinge region to the ligand‐binding domain (LBD), as well as unstructured AB and F domains at its amino and carboxyl termini, respectively (Fig 1B). The LBD contains a ligand‐dependent coactivator‐binding site called activation function‐2 (AF‐2). However, the agonist activity of SERMs derives from activation function‐1 (AF‐1)—a coactivator recruitment site located in the AB domain (Berry et al, 1990; Shang & Brown, 2002; Abot et al, 2013). AF‐1 and AF‐2 bind distinct but overlapping sets of coregulators (Webb et al, 1998; Endoh et al, 1999; Delage‐Mourroux et al, 2000; Yi et al, 2015). AF‐2 binds the signature LxxLL motif peptides of coactivators such as NCOA1/2/3 (also known as SRC‐1/2/3). AF‐1 binds a separate surface on these coactivators (Webb et al, 1998; Yi et al, 2015). Yet, it is unknown how different ERα ligands control AF‐1 through the LBD, and whether this inter‐domain communication is required for cell‐specific signaling or anti‐proliferative responses. In the canonical model of the ERα signaling pathway (Fig 1C), E2‐bound ERα forms a homodimer that binds DNA at estrogen‐response elements (EREs), recruits NCOA1/2/3 (Metivier et al, 2003; Johnson & O'Malley, 2012), and activates the GREB1 gene, which is required for proliferation of ERα‐positive breast cancer cells (Ghosh et al, 2000; Rae et al, 2005; Deschenes et al, 2007; Liu et al, 2012; Srinivasan et al, 2013). However, ERα‐mediated proliferative responses vary in a ligand‐dependent manner (Srinivasan et al, 2013); thus, it is not known whether this canonical model is widely applicable across diverse ERα ligands. Our long‐term goal is to be able to predict proliferative or anti‐proliferative activity of a ligand in different tissues from its crystal structure by identifying different structural perturbations that lead to specific signaling outcomes. The simplest response model for ligand‐specific proliferative effects is a linear causality model, where the degree of NCOA1/2/3 recruitment determines GREB1 expression, which in turn drives ligand‐specific cell proliferation (Fig 1D). Alternatively, a more complicated branched causality model could explain ligand‐specific proliferative responses (Fig 1E). In this signaling model, multiple coregulator binding events and target genes (Won Jeong et al, 2012; Nwachukwu et al, 2014), LBD conformation, nucleocytoplasmic shuttling, the occupancy and dynamics of DNA binding, and other biophysical features could contribute independently to cell proliferation (Lickwar et al, 2012). To test these signaling models, we profiled a diverse library of ERα ligands using systems biology approaches to X‐ray crystallography and chemical biology (Srinivasan et al, 2013), including a series of quantitative bioassays for ERα function that were statistically robust and reproducible, based on the Z’‐statistic (Fig EV1A and B; see Materials and Methods). We also determined the structures of 76 distinct ERα LBD complexes bound to different ligand types, which allowed us to understand how diverse ligand scaffolds distort the active conformation of the ERα LBD. Our findings here indicate that specific structural perturbations can be tied to ligand‐selective domain usage and signaling patterns, thus providing a framework for structure‐based design of improved breast cancer therapeutics, and understanding the different phenotypic effects of environmental estrogens. Summary of ligand screening assays used to measure ER‐mediated activities. ERE, estrogen‐response element; Luc, luciferase reporter gene; M2H, mammalian 2‐hybrid; UAS, upstream‐activating sequence.Controls for screening assays described in panel (A), above. Error bars indicate mean ± SEM, n = 3. Summary of ligand screening assays used to measure ER‐mediated activities. ERE, estrogen‐response element; Luc, luciferase reporter gene; M2H, mammalian 2‐hybrid; UAS, upstream‐activating sequence. Controls for screening assays described in panel (A), above. Error bars indicate mean ± SEM, n = 3. To compare ERα signaling induced by diverse ligand types, we synthesized and assayed a library of 241 ERα ligands containing 19 distinct molecular scaffolds. These include 15 indirect modulator series, which lack a SERM‐like side chain and modulate coactivator binding indirectly from the ligand‐binding pocket (Fig 2A–E; Dataset EV1) (Zheng et al, 2012) (Zhu et al, 2012) (Muthyala et al, 2003; Seo et al, 2006) (Srinivasan et al, 2013) (Wang et al, 2012) (Liao et al, 2014) (Min et al, 2013). We also generated four direct modulator series with side chains designed to directly dislocate h12 and thereby completely occlude the AF‐2 surface (Fig 2C and E; Dataset EV1) (Kieser et al, 2010). Ligand profiling using our quantitative bioassays revealed a wide range of ligand‐induced GREB1 expression, reporter gene activities, ERα‐coactivator interactions, and proliferative effects on MCF‐7 breast cancer cells (Figs EV1 and EV2A–J). This wide variance enabled us to probe specific features of ERα signaling using ligand class analyses, and identify signaling patterns shared by specific ligand series or scaffolds. AStructure of the E2‐bound ERα LBD in complex with an NCOA2 peptide of (PDB 1GWR).B–DStructural details of the ERα LBD bound to the indicated ligands. Unlike E2 (PDB 1GWR), TAM is a direct modulator with a BSC that dislocates h12 to block the NCOA2‐binding site (PDB 3ERT). OBHS is an indirect modulator that dislocates the h11 C‐terminus to destabilize the h11–h12 interface (PDB 4ZN9).EThe ERα ligand library contains 241 ligands representing 15 indirect modulator scaffolds, plus 4 direct modulator scaffolds. The number of compounds per scaffold is shown in parentheses (see Dataset EV1 for individual compound information and Appendix Supplementary Methods for synthetic protocols). Structure of the E2‐bound ERα LBD in complex with an NCOA2 peptide of (PDB 1GWR). Structural details of the ERα LBD bound to the indicated ligands. Unlike E2 (PDB 1GWR), TAM is a direct modulator with a BSC that dislocates h12 to block the NCOA2‐binding site (PDB 3ERT). OBHS is an indirect modulator that dislocates the h11 C‐terminus to destabilize the h11–h12 interface (PDB 4ZN9). The ERα ligand library contains 241 ligands representing 15 indirect modulator scaffolds, plus 4 direct modulator scaffolds. The number of compounds per scaffold is shown in parentheses (see Dataset EV1 for individual compound information and Appendix Supplementary Methods for synthetic protocols). A–JScreening data from individual ligands are shown, grouped by scaffold. Each data point represents the activity of a distinct compound. Error bars indicate the class average (mean) ± range. Direct modulator. Screening data from individual ligands are shown, grouped by scaffold. Each data point represents the activity of a distinct compound. Error bars indicate the class average (mean) ± range. Direct modulator. Source data are available online for this figure. We first asked whether direct modulation of the receptor with an extended side chain is required for cell‐specific signaling. To this end, we compared the average ligand‐induced GREB1 mRNA levels in MCF‐7 cells and 3×ERE‐Luc reporter gene activity in Ishikawa endometrial cancer cells (E‐Luc) or in HepG2 cells transfected with wild‐type ERα (L‐Luc ERα‐WT) (Figs 3A and EV2A–C). Direct modulators showed significant differences in average activity between cell types except OBHS‐ASC analogs, which had similar low agonist activities in the three cell types. The other direct modulators had low agonist activity in Ishikawa cells, no or inverse agonist activity in MCF‐7 cells, and more variable activity in HepG2 liver cells. While it was known that direct modulators such as tamoxifen drive cell‐specific signaling, these experiments reveal that indirect modulators also drive cell‐specific signaling, since eight of fourteen classes showed significant differences in average activity (Figs 3A and EV2A–C). A, B(A) Ligand‐specific ERα activities in HepG2, Ishikawa and MCF‐7 cells. The ligand‐induced L‐Luc ERα‐WT and E‐Luc activities and GREB1 mRNA levels are shown by scaffold (mean + SD). (B) Ligand class analysis of the L‐Luc ERα‐WT and ERα‐ΔAB activities in HepG2 cells. Significant sensitivity to AB domain deletion was determined by Student's t‐test (n = number of ligands per scaffold in Fig 2). The average activities of ligands classes are shown (mean + SEM).C–FCorrelation and regression analyses in a large test set. The r values are plotted as a heat map. In cluster 1, the first three comparisons (rows) showed significant positive correlations (F‐test for nonzero slope, P ≤ 0.05). In cluster 2, only one of these comparisons revealed a significant positive correlation, while none was significant in cluster 3. +, statistically significant correlations gained by deletion of the AB or F domains. −, significant correlations lost upon deletion of AB or F domains. (A) Ligand‐specific ERα activities in HepG2, Ishikawa and MCF‐7 cells. The ligand‐induced L‐Luc ERα‐WT and E‐Luc activities and GREB1 mRNA levels are shown by scaffold (mean + SD). (B) Ligand class analysis of the L‐Luc ERα‐WT and ERα‐ΔAB activities in HepG2 cells. Significant sensitivity to AB domain deletion was determined by Student's t‐test (n = number of ligands per scaffold in Fig 2). The average activities of ligands classes are shown (mean + SEM). Correlation and regression analyses in a large test set. The r values are plotted as a heat map. In cluster 1, the first three comparisons (rows) showed significant positive correlations (F‐test for nonzero slope, P ≤ 0.05). In cluster 2, only one of these comparisons revealed a significant positive correlation, while none was significant in cluster 3. +, statistically significant correlations gained by deletion of the AB or F domains. −, significant correlations lost upon deletion of AB or F domains. Source data are available online for this figure. Tamoxifen depends on AF‐1 for its cell‐specific activity (Sakamoto et al, 2002); therefore, we asked whether cell‐specific signaling observed here is due to a similar dependence on AF‐1 for activity (Fig EV1). To test this idea, we compared the average L‐Luc activities of each scaffold in HepG2 cells co‐transfected with wild‐type ERα or with ERα lacking the AB domain (Figs 1B and EV1). While E2 showed similar L‐Luc ERα‐WT and ERα‐ΔAB activities, tamoxifen showed complete loss of activity without the AB domain (Fig EV1B). Deletion of the AB domain significantly reduced the average L‐Luc activities of 14 scaffolds (Student's t‐test, P ≤ 0.05) (Fig 3B). These “AF‐1‐sensitive” activities were exhibited by both direct and indirect modulators, and were not limited to scaffolds that showed cell‐specific signaling (Fig 3A and B). Thus, the strength of AF‐1 signaling does not determine cell‐specific signaling. As another approach to identifying cell‐specific signaling, we determined the degree of correlation between ligand‐induced activities in the different cell types. Here, we compared ligands within each class (Fig 3C), instead of comparing average activities (Fig 3A and B). For each ligand class or scaffold, we calculated the Pearson's correlation coefficient, r, for pairwise comparison of activity profiles in breast (GREB1), liver (L‐Luc), and endometrial cells (E‐Luc). The value of r ranges from −1 to 1, and it defines the extent to which the data fit a straight line when compounds show similar agonist/antagonist activity profiles between cell types (Fig EV3A). We also calculated the coefficient of determination, r , which describes the percentage of variance in a dependent variable such as proliferation that can be predicted by an independent variable such as GREB1 expression. We present both calculations as r to readily compare signaling specificities using a heat map on which the red–yellow palette indicates significant positive correlations (P ≤ 0.05, F‐test for nonzero slope), while the blue palette denotes negative correlations (Fig 3C–F). A–CCorrelation analysis of OBHS versus OBHS‐BSC activity across cell types.D, ECorrelation analysis of L‐Luc ERα‐ΔAB activity versus endogenous ERα activity of OBHS analogs. In panel (D), L‐Luc ERα‐WT activity from panel (B) is shown for comparison.FCorrelation analysis of L‐Luc ERα‐ΔF activity versus endogenous ERα activities of OBHS analogs.G, HCorrelation analysis of MCF‐7 cell proliferation versus NCOA2/3 recruitment or GREB1 levels observed in response to (G) OBHS‐N and (H) OBHS‐BSC analogs.Data information: In each panel, a data point indicates the activity of a distinct compound.Source data are available online for this figure. Correlation analysis of OBHS versus OBHS‐BSC activity across cell types. Correlation analysis of L‐Luc ERα‐ΔAB activity versus endogenous ERα activity of OBHS analogs. In panel (D), L‐Luc ERα‐WT activity from panel (B) is shown for comparison. Correlation analysis of L‐Luc ERα‐ΔF activity versus endogenous ERα activities of OBHS analogs. Correlation analysis of MCF‐7 cell proliferation versus NCOA2/3 recruitment or GREB1 levels observed in response to (G) OBHS‐N and (H) OBHS‐BSC analogs. This analysis revealed diverse signaling specificities that we grouped into three clusters. Scaffolds in cluster 1 exhibited strongly correlated GREB1 levels, E‐Luc and L‐Luc activity profiles across the three cell types (Fig 3C lanes 1–4), suggesting these ligands use similar ERα signaling pathways in the breast, endometrial, and liver cell types. This cluster includes WAY‐C, OBHS, OBHS‐N, and triaryl‐ethylene analogs, all of which are indirect modulators. Cluster 2 contains scaffolds with activities that were positively correlated in only two of the three cell types, indicating cell‐specific signaling (Fig 3C lanes 5–12). This cluster includes two classes of direct modulators (cyclofenil‐ASC and WAY dimer), and six classes of indirect modulators (2,5‐DTP, 3,4‐DTP, S‐OBHS‐2 and S‐OBHS‐3, furan, and WAY‐D). In this cluster, the correlated activities varied by scaffold. For example, 3,4‐DTP, furan, and S‐OBHS‐2 drove positively correlated GREB1 levels and E‐Luc but not L‐Luc ERα‐WT activity (Fig 3C lanes 5–7). In contrast, WAY dimer and WAY‐D analogs drove positively correlated GREB1 levels and L‐Luc ERα‐WT but not E‐Luc activity (Fig 3C lanes 8 and 9). The last set of scaffolds, cluster 3, displayed cell‐specific activities that were not correlated in any of the three cell types (Fig 3C lanes 13–19). This cluster includes two direct modulator scaffolds (OBHS‐ASC and OBHS‐BSC), and five indirect modulator scaffolds (A‐CD, cyclofenil, 3,4‐DTPD, imine, and imidazopyridine). These results suggest that addition of an extended side chain to an ERα ligand scaffold is sufficient to induce cell‐specific signaling, where the relative activity profiles of the individual ligands change between cell types. This is demonstrated by directly comparing the signaling specificities of matched OBHS (indirect modulator, cluster 1) and OBHS‐BSC analogs (direct modulator, cluster 3), which differ only in the basic side chain (Fig 2E). The activities of OBHS analogs were positively correlated across the three cell types, but the side chain of OBHS‐BSC analogs was sufficient to abolish these correlations (Figs 3C lanes 1 and 19, and EV3A–C). The indirect modulator scaffolds in clusters 2 and 3 showed cell‐specific signaling patterns without the extended side chain typically viewed as the primary chemical and structural mechanism driving cell‐specific activity. Many of these scaffolds drove similar average activities of the ligand class in the different cell types (Fig 3A), but the individual ligands in each class had different cell‐specific activities (Fig EV2A–C). Thus, examining the correlated patterns of ERα activity within each scaffold demonstrates that an extended side chain is not required for cell‐specific signaling. To evaluate the role of AF‐1 and the F domain in ERα signaling specificity, we compared activity of truncated ERα constructs in HepG2 liver cells with endogenous ERα activity in the other cell types. The positive correlation between the L‐Luc and E‐Luc activities or GREB1 levels induced by scaffolds in cluster 1 was generally retained without the AB domain, or the F domain (Fig 3D lanes 1–4). This demonstrates that the signaling specificities underlying these positive correlations are not modified by AF‐1. OBHS analogs showed an average L‐Luc ERα‐ΔAB activity of 3.2% ± 3 (mean + SEM) relative to E2. Despite this nearly complete lack of activity, the pattern of L‐Luc ERα‐ΔAB activity was still highly correlated with the E‐Luc activity and GREB1 expression (Fig EV3D and E), demonstrating that very small AF‐2 activities can be amplified by AF‐1 to produce robust signals. Similarly, deletion of the F domain did not abolish correlations between the L‐Luc and E‐Luc or GREB1 levels induced by OBHS analogs (Fig EV3F). These similar patterns of ligand activity in the wild‐type and deletion mutants suggest that AF‐1 and the F domain purely amplify the AF‐2 activities of ligands in cluster 1. In contrast, AF‐1 was a determinant of signaling specificity for scaffolds in cluster 2. Deletion of the AB or F domain altered correlations for six of the eight scaffolds in this cluster (2,5‐DTP, 3,4‐DTP, S‐OBHS‐3, WAY‐D, WAY dimer, and cyclofenil‐ASC) (Fig 3D lanes 5–12). Comparing Fig 3C and D, the + and − signs indicate where the deletion mutant assays led to a gain or loss of statically significant correlation, respectively. Thus, in cluster 2, AF‐1 substantially modulated the specificity of ligands with cell‐specific activity (Fig 3D lanes 5–12). For ligands in cluster 3, we could not eliminate a role for AF‐1 in determining signaling specificity, since this cluster lacked positively correlated activity profiles (Fig 3C), and deletion of the AB or F domain rarely induced such correlations (Fig 3D), except for A‐CD and OBHS‐ASC analogs, where deletion of the AB domain or F domain led to positive correlations with E‐Luc activity and/or GREB1 levels (Fig 3D lanes 13 and 18). Thus, ligands in cluster 2 rely on AF‐1 for both activity (Fig 3B) and signaling specificity (Fig 3D). As discussed below, this cell specificity derives from alternate coactivator preferences. To determine whether ligand classes control expression of native ERα target genes through the canonical linear signaling pathway, we performed pairwise linear regression analyses using ERα–NCOA1/2/3 interactions in M2H assay as independent predictors of GREB1 expression (the dependent variable) (Figs EV1 and EV2A, F–H). In cluster 1, the recruitment of NCOA1 and NCOA2 was highest for WAY‐C, followed by triaryl‐ethylene, OBHS‐N, and OBHS series, while for NCOA3, OBHS‐N compounds induced the most recruitment and OBHS ligands were inverse agonists (Fig EV2F–H). The average induction of GREB1 by cluster 1 ligands showed greater variance, with a range between ~25 and ~75% for OBHS and a range from full agonist to inverse agonist for the others in cluster 1 (Fig EV2A). GREB1 levels induced by OBHS analogs were determined by recruitment of NCOA1 but not NCOA2/3 (Fig 3E lane 1), suggesting that there may be alternate or preferential use of these coactivators by different classes. However, in cluster 1, NCOA1/2/3 recruitment generally predicted GREB1 levels (Fig 3E lanes 1–4), consistent with the canonical signaling model (Fig 1D). For clusters 2 and 3, GREB1 activity was generally not predicted by NCOA1/2/3 recruitment. Direct modulators showed low NCOA1/2/3 recruitment (Fig EV2F–H), but only OBHS‐ASC analogs had NCOA2 recruitment profiles that predicted a full range of effects on GREB1 levels (Figs 3E lanes 9, 11, 18–19, and EV2A). The indirect modulators in clusters 2 and 3 stimulated NCOA1/2/3 recruitment and GREB1 expression with substantial variance (Figs 3A and EV2F–H). However, ligand‐induced GREB1 levels were generally not determined by NCOA1/2/3 recruitment (Fig 3E lanes 5–19), consistent with an alternate causality model (Fig 1E). Out of 11 indirect modulator series in cluster 2 or 3, only the S‐OBHS‐3 class had NCOA1/2/3 recruitment profiles that predicted GREB1 levels (Fig 3E lane 12). These results suggest that compounds that show cell‐specific signaling do not activate GREB1, or use coactivators other than NCOA1/2/3 to control GREB1 expression (Fig 1E). To determine mechanisms for ligand‐dependent control of breast cancer cell proliferation, we performed linear regression analyses across the 19 scaffolds using MCF‐7 cell proliferation as the dependent variable, and the other activities as independent variables (Fig 3F). In cluster 1, E‐Luc and L‐Luc activities, NCOA1/2/3 recruitment, and GREB1 levels generally predicted the proliferative response (Fig 3F lanes 2–4). With the OBHS‐N compounds, NCOA3 and GREB1 showed near perfect prediction of proliferation (Fig EV3G), with unexplained variance similar to the noise in the assays. The lack of significant predictors for OBHS analogs (Fig 3F lane 1) reflects their small range of proliferative effects on MCF‐7 cells (Fig EV2I). The significant correlations with GREB1 expression and NCOA1/2/3 recruitment observed in this cluster are consistent with the canonical signaling model (Fig 1D), where NCOA1/2/3 recruitment determines GREB1 expression, which then drives proliferation. Ligands in cluster 2 and cluster 3 showed a wide range of proliferative effects on MCF‐7 cells (Fig EV2I). Despite this phenotypic variance, proliferation was not generally predicted by correlated NCOA1/2/3 recruitment and GREB1 induction (Figs 3F lanes 5–19, and EV3H). Out of 15 ligand series in these clusters, only 2,5‐DTP analogs induced a proliferative response that was predicted by GREB1 levels, which were not determined by NCOA1/2/3 recruitment (Fig 3E and F lane 10). 3,4‐DTP, cyclofenil, 3,4‐DTPD, and imidazopyridine analogs had NCOA1/3 recruitment profiles that predicted their proliferative effects, without determining GREB1 levels (Fig 3E and F, lanes 5 and 14–16). Similarly, S‐OBHS‐3, cyclofenil‐ASC, and OBHS‐ASC had positively correlated NCOA1/2/3 recruitment and GREB1 levels, but none of these activities determined their proliferative effects (Fig 3E and F lanes 11–12 and 18). For ligands that show cell‐specific signaling, ERα‐mediated recruitment of other coregulators and activation of other target genes likely determine their proliferative effects on MCF‐7 cells. We also questioned whether promoter occupancy by coactivators is statistically robust and reproducible for ligand class analysis using a chromatin immunoprecipitation (ChIP)‐based quantitative assay, and whether it has a better predictive power than the M2H assay. ERα and NCOA3 cycle on and off the GREB1 promoter (Nwachukwu et al, 2014). Therefore, we first performed a time‐course study, and found that E2 and the WAY‐C analog, AAPII‐151‐4, induced recruitment of NCOA3 to the GREB1 promoter in a temporal cycle that peaked after 45 min in MCF‐7 cells (Fig 4A). At this time point, other WAY‐C analogs also induced recruitment of NCOA3 at this site to varying degrees (Fig 4B). The Z’ for this assay was 0.6, showing statistical robustness (see Materials and Methods). We prepared biological replicates with different cell passage numbers and separately prepared samples, which showed r of 0.81, demonstrating high reproducibility (Fig 4C). AKinetic ChIP assay examining recruitment of NCOA3 to the GREB1 gene in MCF‐7 cells stimulated with E2 or the indicated WAY‐C analog. The average of duplicate experiments (mean ± SEM) is shown.B, CNCOA3 occupancy at GREB1 was compared by ChIP assay 45 min after stimulation with vehicle, E2, or the WAY‐C analogs. In panel (B), the average recruitment of two biological replicates are shown as mean + SEM, and the Z‐score is indicated. In panel (C), correlation analysis was performed for two biological replicates.DLinear regression analyses comparing the ability of NCOA3 recruitment, measured by ChIP or M2H, to predict other agonist activities of WAY‐C analogs. Significant positive correlation (F‐test for nonzero slope, P‐value). Kinetic ChIP assay examining recruitment of NCOA3 to the GREB1 gene in MCF‐7 cells stimulated with E2 or the indicated WAY‐C analog. The average of duplicate experiments (mean ± SEM) is shown. NCOA3 occupancy at GREB1 was compared by ChIP assay 45 min after stimulation with vehicle, E2, or the WAY‐C analogs. In panel (B), the average recruitment of two biological replicates are shown as mean + SEM, and the Z‐score is indicated. In panel (C), correlation analysis was performed for two biological replicates. Linear regression analyses comparing the ability of NCOA3 recruitment, measured by ChIP or M2H, to predict other agonist activities of WAY‐C analogs. Significant positive correlation (F‐test for nonzero slope, P‐value). Source data are available online for this figure. The M2H assay for NCOA3 recruitment broadly correlated with the other assays, and was predictive for GREB1 expression and cell proliferation (Fig 3E). However, the ChIP assays for WAY‐C‐induced recruitment of NCOA3 to the GREB1 promoter did not correlate with any of the other WAY‐C activity profiles (Fig 4D), although the positive correlation between ChIP assays and NCOA3 recruitment via M2H assay showed a trend toward significance with r = 0.36 and P = 0.09 (F‐test for nonzero slope). Thus, the simplified coactivator‐binding assay showed much greater predictive power than the ChIP assay for ligand‐specific effects on GREB1 expression and cell proliferation. One difference between MCF‐7 breast cancer cells and Ishikawa endometrial cancer cells is the contribution of ERβ to estrogenic response, as Ishikawa cells may express ERβ (Bhat & Pezzuto, 2001). When overexpressed in MCF‐7 cells, ERβ alters E2‐induced expression of only a subset of ERα‐target genes (Wu et al, 2011), raising the possibility that ligand‐induced ERβ activity may contribute to E‐Luc activities, and thus underlie the lack of correlation between the E‐Luc and L‐Luc ERα‐WT activities or GREB1 levels induced by cell‐specific modulators in cluster 2 and cluster 3 (Fig 3C). To test this idea, we determined the L‐Luc ERβ activity profiles of the ligands (Fig EV1). All direct modulator and two indirect modulator scaffolds (OBHS and S‐OBHS‐3) lacked ERβ agonist activity. However, the other ligands showed a range of ERβ activities (Fig EV2J). For most scaffolds, L‐Luc ERβ and E‐Luc activities were not correlated, except for 2,5‐DTP and cyclofenil analogs, which showed moderate but significant correlations (Fig EV4A). Nevertheless, the E‐Luc activities of both 2,5‐DTP and cyclofenil analogs were better predicted by their L‐Luc ERα‐WT than L‐Luc ERβ activities (Fig EV4A and B). Thus, ERβ activity was not an independent determinant of the observed activity profiles. ERβ activity in HepG2 cells rarely correlates with E‐Luc activity.ERα activity of 2,5‐DTP and cyclofenil analogs correlates with E‐Luc activity.Data information: The r and P values for the indicated correlations are shown in both panels. Significant positive correlation (F‐test for nonzero slope, P‐value) ERβ activity in HepG2 cells rarely correlates with E‐Luc activity. ERα activity of 2,5‐DTP and cyclofenil analogs correlates with E‐Luc activity. To overcome barriers to crystallization of ERα LBD complexes, we developed a conformation‐trapping X‐ray crystallography approach using the ERα‐Y537S mutation (Nettles et al, 2008; Bruning et al, 2010; Srinivasan et al, 2013). To further validate this approach, we solved the structure of the ERα‐Y537S LBD in complex with diethylstilbestrol (DES), which bound identically in the wild‐type and ERα‐Y537S LBDs, demonstrating again that this surface mutation stabilizes h12 dynamics to facilitate crystallization without changing ligand binding (Appendix Fig S1A and B) (Nettles et al, 2008; Bruning et al, 2010; Delfosse et al, 2012). Using this approach, we solved 76 ERα LBD structures in the active conformation and bound to ligands studied here (Appendix Fig S1C). Eleven of these structures have been published, while 65 are new, including the DES‐bound ERα‐Y537S LBD. We present 57 of these new structures here (Dataset EV2), while the remaining eight new structures bound to OBHS‐N analogs will be published elsewhere (S. Srinivasan et al, in preparation). Examining many closely related structures allows us to visualize subtle structural differences, in effect using X‐ray crystallography as a systems biology tool. The indirect modulator scaffolds in cluster 1 did not show cell‐specific signaling (Fig 3C), but shared common structural perturbations that we designed to modulate h12 dynamics. Based on our original OBHS structure, the OBHS, OBHS‐N, and triaryl‐ethylene compounds were modified with h11‐directed pendant groups (Zheng et al, 2012; Zhu et al, 2012; Liao et al, 2014). Superposing the LBDs based on the class of bound ligands provides an ensemble view of the structural variance and clarifies what part of the ligand‐binding pocket is differentially perturbed or targeted. The 24 structures containing OBHS, OBHS‐N, or triaryl‐ethylene analogs showed structural diversity in the same part of the scaffolds (Figs 5A and EV5A), and the same region of the LBD—the C‐terminal end of h11 (Figs 5B and C, and EV5B), which in turn nudges h12 (Fig 5C and D). We observed that the OBHS‐N analogs displaced h11 along a vector away from Leu354 in a region of h3 that is unaffected by the ligands, and toward the dimer interface. For the triaryl‐ethylene analogs, the displacement of h11 was in a perpendicular direction, away from Ile424 in h8 and toward h12. Remarkably, these individual inter‐atomic distances showed a ligand class‐specific ability to significantly predict proliferative effects (Fig 5E and F), demonstrating the feasibility of developing a minimal set of activity predictors from crystal structures. AStructure‐class analysis of triaryl‐ethylene analogs. Triaryl‐ethylene analogs bound to the superposed crystal structures of the ERα LBD are shown. Arrows indicate chemical variance in the orientation of the different h11‐directed ligand side groups (PDB 5DK9, 5DKB, 5DKE, 5DKG, 5DKS, 5DL4, 5DLR, 5DMC, 5DMF and 5DP0).B, CTriaryl‐ethylene analogs induce variance of ERα conformations at the C‐terminal region of h11. Panel (B) shows the crystal structure of a triaryl‐ethylene analog‐bound ERα LBD (PDB 5DLR). The h11–h12 interface (circled) includes the C‐terminal part of h11. This region was expanded in panel (C), where the 10 triaryl‐ethylene analog‐bound ERα LBD structures (see Datasets EV1 and EV2) were superposed to show variations in the h11 C‐terminus (PDB 5DK9, 5DKB, 5DKE, 5DKG, 5DKS, 5DL4, 5DLR, 5DMC, 5DMF, and 5DP0).DERα LBDs in complex with diethylstilbestrol (DES) or a triaryl‐ethylene analog were superposed to show that the ligand‐induced difference in h11 conformation is transmitted to the C‐terminus of h12 (PDB 4ZN7, 5DMC).E, FInter‐atomic distances predict the proliferative effects of specific ligand series. Ile424–His524 distance measured in the crystal structures correlates with the proliferative effect of triaryl‐ethylene analogs in MCF‐7 cells. In contrast, the Leu354–Leu525 distance correlates with the proliferative effects of OBHS‐N analogs in MCF‐7 cells.G, HStructure‐class analysis of WAY‐C analogs. WAY‐C side groups subtly nudge h12 Leu540. ERα LBD structures bound to 4 distinct WAY‐C analogs were superposed (PDB 4 IU7, 4IV4, 4IVW, 4IW6) (see Datasets EV1 and EV2). Structure‐class analysis of triaryl‐ethylene analogs. Triaryl‐ethylene analogs bound to the superposed crystal structures of the ERα LBD are shown. Arrows indicate chemical variance in the orientation of the different h11‐directed ligand side groups (PDB 5DK9, 5DKB, 5DKE, 5DKG, 5DKS, 5DL4, 5DLR, 5DMC, 5DMF and 5DP0). Triaryl‐ethylene analogs induce variance of ERα conformations at the C‐terminal region of h11. Panel (B) shows the crystal structure of a triaryl‐ethylene analog‐bound ERα LBD (PDB 5DLR). The h11–h12 interface (circled) includes the C‐terminal part of h11. This region was expanded in panel (C), where the 10 triaryl‐ethylene analog‐bound ERα LBD structures (see Datasets EV1 and EV2) were superposed to show variations in the h11 C‐terminus (PDB 5DK9, 5DKB, 5DKE, 5DKG, 5DKS, 5DL4, 5DLR, 5DMC, 5DMF, and 5DP0). ERα LBDs in complex with diethylstilbestrol (DES) or a triaryl‐ethylene analog were superposed to show that the ligand‐induced difference in h11 conformation is transmitted to the C‐terminus of h12 (PDB 4ZN7, 5DMC). Inter‐atomic distances predict the proliferative effects of specific ligand series. Ile424–His524 distance measured in the crystal structures correlates with the proliferative effect of triaryl‐ethylene analogs in MCF‐7 cells. In contrast, the Leu354–Leu525 distance correlates with the proliferative effects of OBHS‐N analogs in MCF‐7 cells. Structure‐class analysis of WAY‐C analogs. WAY‐C side groups subtly nudge h12 Leu540. ERα LBD structures bound to 4 distinct WAY‐C analogs were superposed (PDB 4 IU7, 4IV4, 4IVW, 4IW6) (see Datasets EV1 and EV2). Source data are available online for this figure. A, BStructure‐class analysis of indirect modulators in cluster 1. Crystal structures of the ERα LBD bound to OBHS and OBHS‐N analogs were superposed. The bound ligands are shown in panel (A). Arrows indicate chemical variance in the orientation of the different h11‐directed ligand side groups. Panel (B) shows the ligand‐induced conformational variation at the C‐terminal region of h11 (OBHS: PDB 4ZN9, 4ZNH, 4ZNS, 4ZNT, 4ZNU, 4ZNV, and 4ZNW; OBHS‐N: PDB 4ZUB, 4ZUC, 4ZWH, 4ZWK, 5BNU, 5BP6, 5BPR, and 5BQ4).C–GStructure‐class analysis of indirect modulators in clusters 2 and 3. Crystal structures of the ERα LBD bound to ligands with cell‐specific activities were superposed. The bound ligands are shown, and arrows indicate considerable variation in the orientation of the different h3‐, h8‐, h11‐, or h12‐directed ligand side groups. Structure‐class analysis of indirect modulators in cluster 1. Crystal structures of the ERα LBD bound to OBHS and OBHS‐N analogs were superposed. The bound ligands are shown in panel (A). Arrows indicate chemical variance in the orientation of the different h11‐directed ligand side groups. Panel (B) shows the ligand‐induced conformational variation at the C‐terminal region of h11 (OBHS: PDB 4ZN9, 4ZNH, 4ZNS, 4ZNT, 4ZNU, 4ZNV, and 4ZNW; OBHS‐N: PDB 4ZUB, 4ZUC, 4ZWH, 4ZWK, 5BNU, 5BP6, 5BPR, and 5BQ4). Structure‐class analysis of indirect modulators in clusters 2 and 3. Crystal structures of the ERα LBD bound to ligands with cell‐specific activities were superposed. The bound ligands are shown, and arrows indicate considerable variation in the orientation of the different h3‐, h8‐, h11‐, or h12‐directed ligand side groups. As visualized in four LBD structures (Srinivasan et al, 2013), WAY‐C analogs were designed with small substitutions that slightly nudge h12 Leu540, without exiting the ligand‐binding pocket (Fig 5G and H). Therefore, changing h12 dynamics maintains the canonical signaling pathway defined by E2 (Fig 1D) to support AF‐2‐driven signaling and recruit NCOA1/2/3 for GREB1‐stimulated proliferation. Direct modulators like tamoxifen drive AF‐1‐dependent cell‐specific activity by completely occluding AF‐2, but it is not known how indirect modulators produce cell‐specific ERα activity. Therefore, we examined another 50 LBD structures containing ligands in clusters 2 and 3. These structures demonstrated that cell‐specific activity derived from altering the shape of the AF‐2 surface without an extended side chain. Ligands in cluster 2 and cluster 3 showed conformational heterogeneity in parts of the scaffold that were directed toward multiple regions of the receptor including h3, h8, h11, h12, and/or the β‐sheets (Fig EV5C–G). For instance, S‐OBHS‐2 and S‐OBHS‐3 analogs (Fig 2) had similar ERα activity profiles in the different cell types (Fig EV2A–C), but the 2‐ versus 3‐methyl substituted phenol rings altered the correlated signaling patterns in different cell types (Fig 3B lanes 7 and 12). Structurally, the 2‐ versus 3‐methyl substitutions changed the binding position of the A‐ and E‐ring phenols by 1.0 Å and 2.2 Å, respectively (Fig EV5C). This difference in ligand positioning altered the AF‐2 surface via a shift in the N‐terminus of h12, which directly contacts the coactivator. This effect is evident in a single structure due to its 1 Å magnitude (Fig 6A and B). The shifts in h12 residues Asp538 and Leu539 led to rotation of the coactivator peptide (Fig 6C). Thus, cell‐specific activity can stem from perturbation of the AF‐2 surface without an extended side chain, which presumably alters the receptor–coregulator interaction profile. A–CS‐OBHS‐2/3 analogs subtly distort the AF‐2 surface. Panel (A) shows the crystal structure of an S‐OBHS‐3‐bound ERα LBD (PDB 5DUH). The h3–h12 interface (circled) at AF‐2 (pink) was expanded in panels (B, C). The S‐OBHS‐2/3‐bound ERα LBDs were superposed to show shifts in h3 (panel B) and the NCOA2 peptide docked at the AF‐2 surface (panel C).DCrystal structures show that 2,5‐DTP analogs shift h3 and h11 further apart compared to an A‐CD‐ring estrogen (PDB 4PPS, 5DRM, 5DRJ). The 2F o‐F c electron density map and F o‐F c difference map of a 2,5‐DTP‐bound structure (PDB 5DRJ) were contoured at 1.0 sigma and ± 3.0 sigma, respectively.EAverage (mean + SEM) α‐carbon distance measured from h3 Thr347 to h11 Leu525 of A‐CD‐, 2,5‐DTP‐, and 3,4‐DTPD‐bound ERα LBDs. Two‐tailed Student's t‐test, P = 0.002 (PDB A‐CD: 5DI7, 5DID, 5DIE, 5DIG, and 4PPS; 2,5‐DTP: 4IWC, 5DRM, and 5DRJ; 3,4‐DTPD: 5DTV and 5DU5).F, GCrystal structures show that a 3,4‐DTPD analog shifts h3 (F) and the NCOA2 (G) peptide compared to an A‐CD‐ring estrogen (PDB 4PPS, 5DTV).HHierarchical clustering of ligand‐specific binding of 154 interacting peptides to the ERα LBD was performed in triplicate by MARCoNI analysis. S‐OBHS‐2/3 analogs subtly distort the AF‐2 surface. Panel (A) shows the crystal structure of an S‐OBHS‐3‐bound ERα LBD (PDB 5DUH). The h3–h12 interface (circled) at AF‐2 (pink) was expanded in panels (B, C). The S‐OBHS‐2/3‐bound ERα LBDs were superposed to show shifts in h3 (panel B) and the NCOA2 peptide docked at the AF‐2 surface (panel C). Crystal structures show that 2,5‐DTP analogs shift h3 and h11 further apart compared to an A‐CD‐ring estrogen (PDB 4PPS, 5DRM, 5DRJ). The 2F o‐F c electron density map and F o‐F c difference map of a 2,5‐DTP‐bound structure (PDB 5DRJ) were contoured at 1.0 sigma and ± 3.0 sigma, respectively. Average (mean + SEM) α‐carbon distance measured from h3 Thr347 to h11 Leu525 of A‐CD‐, 2,5‐DTP‐, and 3,4‐DTPD‐bound ERα LBDs. *Two‐tailed Student's t‐test, P = 0.002 (PDB A‐CD: 5DI7, 5DID, 5DIE, 5DIG, and 4PPS; 2,5‐DTP: 4IWC, 5DRM, and 5DRJ; 3,4‐DTPD: 5DTV and 5DU5). Crystal structures show that a 3,4‐DTPD analog shifts h3 (F) and the NCOA2 (G) peptide compared to an A‐CD‐ring estrogen (PDB 4PPS, 5DTV). Hierarchical clustering of ligand‐specific binding of 154 interacting peptides to the ERα LBD was performed in triplicate by MARCoNI analysis. Source data are available online for this figure. The 2,5‐DTP analogs showed perturbation of h11, as well as h3, which forms part of the AF‐2 surface. These compounds bind the LBD in an unusual fashion because they have a phenol‐to‐phenol length of ~12 Å, which is longer than steroids and other prototypical ERα agonists that are ~10 Å in length. One phenol pushed further toward h3 (Fig 6D), while the other phenol pushed toward the C‐terminus of h11 to a greater extent than A‐CD‐ring estrogens (Nwachukwu et al, 2014), which are close structural analogs of E2 that lack a B‐ring (Fig 2). To quantify this difference, we compared the distance between α‐carbons at h3 Thr347 and h11 Leu525 in the set of structures containing 2,5‐DTP analogs (n = 3) or A‐CD‐ring analogs (n = 5) (Fig 6E). We observed a difference of 0.4 Å that was significant (two‐tailed Student's t‐test, P = 0.002) due to the very tight clustering of the 2,5‐DTP‐induced LBD conformation. The shifts in h3 suggest these compounds are positioned to alter coregulator preferences. The 2,5‐DTP and 3,4‐DTP scaffolds are isomeric, but with aryl groups at obtuse and acute angles, respectively (Fig 2). The crystal structure of ERα in complex with a 3,4‐DTP is unknown; however, we solved two crystal structures of ERα bound to 3,4‐DTPD analogs and one structure containing a furan ligand—all of which have a 3,4‐diaryl configuration (Fig 2; Datasets EV1 and EV2). In these structures, the A‐ring mimetic of the 3,4‐DTPD scaffold bound h3 Glu353 as expected, but the other phenol wrapped around h3 to form a hydrogen bond with Thr347, indicating a change in binding epitopes in the ERα ligand‐binding pocket (Fig 6F). The 3,4‐DTPD analogs also induced a shift in h3 positioning, which translated again into a shift in the bound coactivator peptide (Fig 6F). Therefore, these indirect modulators, including S‐OBHS‐2, S‐OBHS‐3, 2,5‐DTP, and 3,4‐DTPD analogs—all of which show cell‐specific activity profiles—induced shifts in h3 and h12 that were transmitted to the coactivator peptide via an altered AF‐2 surface. To test whether the AF‐2 surface shows changes in shape in solution, we used the microarray assay for real‐time coregulator–nuclear receptor interaction (MARCoNI) analysis (Aarts et al, 2013). Here, the ligand‐dependent interactions of the ERα LBD with over 150 distinct LxxLL motif peptides were assayed to define structural fingerprints for the AF‐2 surface, in a manner similar to the use of phage display peptides as structural probes (Connor et al, 2001). Despite the similar average activities of these ligand classes (Fig 3A and B), 2,5‐DTP and 3,4‐DTP analogs displayed remarkably different peptide recruitment patterns (Fig 6H), consistent with the structural analyses. Hierarchical clustering revealed that many of the 2,5‐DTP analogs recapitulated most of the peptide recruitment and dismissal patterns observed with E2 (Fig 6H). However, there was a unique cluster of peptides that were recruited by E2 but not the 2,5‐DTP analogs. In contrast, 3,4‐DTP analogs dismissed most of the peptides from the AF‐2 surface (Fig 6H). Thus, the isomeric attachment of diaryl groups to the thiophene core changed the AF‐2 surface from inside the ligand‐binding pocket, as predicted by the crystal structures. Together, these findings suggest that without an extended side chain, cell‐specific activity stems from different coregulator recruitment profiles, due to unique ligand‐induced conformations of the AF‐2 surface, in addition to differential usage of AF‐1. Indirect modulators in cluster 1 avoid this by perturbing the h11–h12 interface, and modulating the dynamics of h12 without changing the shape of AF‐2 when stabilized. Our goal was to identify a minimal set of predictors that would link specific structural perturbations to ERα signaling pathways that control cell‐specific signaling and proliferation. We found a very strong set of predictors, where ligands in cluster 1, defined by similar signaling across cell types, showed indirect modulation of h12 dynamics via the h11–12 interface or slight contact with h12. This perturbation determined proliferation that correlated strongly with AF‐2 activity, recruitment of NCOA1/2/3 family members, and induction of the GREB1 gene, consistent with the canonical ERα signaling pathway (Fig 1D). For ligands in cluster 1, deletion of AF‐1 reduced activity to varying degrees, but did not change the underlying signaling patterns established through AF‐2. In contrast, an extended side chain designed to directly reposition h12 and completely disrupt the AF‐2 surface results in cell‐specific signaling. This was demonstrated with direct modulators in clusters 2 and 3. Cluster 2 was defined by ligand classes that showed correlated activities in two of the three cell types tested, while ligand classes in cluster 3 did not show correlated activities among any of the three cell types. Compared to cluster 1, the structural rules are less clear in clusters 2 and 3, but a number of indirect modulator classes perturbed the LBD conformation at the intersection of h3, the h12 N‐terminus, and the AF‐2 surface. Ligands in these classes altered the shape of AF‐2 to affect coregulator preferences. For direct and indirect modulators in cluster 2 or 3, the canonical ERα signaling pathway involving recruitment of NCOA1/2/3 and induction of GREB1 did not generally predict their proliferative effects, indicating an alternate causal model (Fig 1E). These principles outlined above provide a structural basis for how the ligand–receptor interface leads to different signaling specificities through AF‐1 and AF‐2. It is noteworthy that regulation of h12 dynamics indirectly through h11 can virtually abolish AF‐2 activity, and yet still drive robust transcriptional activity through AF‐1, as demonstrated with the OBHS series. This finding can be explained by the fact that NCOA1/2/3 contain distinct binding sites for interaction with AF‐1 and AF‐2 (McInerney et al, 1996; Webb et al, 1998), which allows ligands to nucleate ERα–NCOA1/2/3 interaction through AF‐2, and reinforce this interaction with additional binding to AF‐1. Completely blocking AF‐2 with an extended side chain or altering the shape of AF‐2 changes the preference away from NCOA1/2/3 for determining GREB1 levels and proliferation of breast cancer cells. AF‐2 blockade also allows AF‐1 to function independently, which is important since AF‐1 drives tissue‐selective effects in vivo. This was demonstrated with AF‐1 knockout mice that show E2‐dependent vascular protection, but not uterine proliferation, thus highlighting the role of AF‐1 in tissue‐selective or cell‐specific signaling (Billon‐Gales et al, 2009; Abot et al, 2013). One current limitation to our approach is the identification of statistical variables that predict ligand‐specific activity. Here, we examined many LBD structures and tested several variables that were not predictive, including ERβ activity, the strength of AF‐1 signaling, and NCOA3 occupancy at the GREB1 gene. Similarly, we visualized structures to identify patterns. There are many systems biology approaches that could contribute to the unbiased identification of predictive variables for statistical modeling. For example, phage display was used to identify the androgen receptor interactome, which was cloned into an M2H library and used to identify clusters of ligand‐selective interactions (Norris et al, 2009). Also, we have used siRNA screening to identify a number of coregulators required for ERα‐mediated repression of the IL‐6 gene (Nwachukwu et al, 2014). However, the use of larger datasets to identify such predictor variables has its own limitations, one of the major ones being the probability of false positives from multiple hypothesis testing. If we calculated inter‐atomic distance matrices containing 4,000 atoms per structure × 76 ligand–receptor complexes, we would have 3 × 10 predictions. One way to address this issue is to use the cross‐validation concept, where hypotheses are generated on training sets of ligands and tested with another set of ligands. Based on this work, we propose several testable hypotheses for drug discovery. We have identified atomic vectors for the OBHS‐N and triaryl‐ethylene classes that predict ligand response (Fig 5E and F). These ligands in cluster 1 drive consistent, canonical signaling across cell types, which is desirable for generating full antagonists. Indeed, the most anti‐proliferative compound in the OBHS‐N series had a fulvestrant‐like profile across a battery of assays (S. Srinivasan et al, in preparation). Secondly, our finding that WAY‐C compounds do not rely of AF‐1 for signaling efficacy may derive from the slight contacts with h12 observed in crystal structures (Figs 3B and 5H), unlike other compounds in cluster 1 that dislocate h11 and rely on AF‐1 for signaling efficacy (Figs 3B and 5C, and EV5B). Thirdly, we found ligands that achieved cell‐specific activity without a prototypical extended side chain. Some of these ligands altered the shape of the AF‐2 surface by perturbing the h3–h12 interface, thus providing a route to new SERM‐like activity profiles by combining indirect and direct modulation of receptor structure. Incorporation of statistical approaches to understand relationships between structure and signaling variables moves us toward predictive models for complex ERα‐mediated responses such as in vivo uterine proliferation or tumor growth, and more generally toward structure‐based design for other allosteric drug targets including GPCRs and other nuclear receptors. Correlation and linear regression analyses were performed using GraphPad Prism software. For correlation analysis, the degree to which two datasets vary together was calculated with the Pearson correlation coefficient (r). However, we reported r rather than r, to facilitate comparison with the linear regression results for which we calculated and reported r (Fig 3C–F). Significance for r was determined using the F‐test for nonzero slope. High‐throughput assays were considered statistically robust if they show Z’ > 0.5, where Z’ = 1 − (3(σp+σn)/\|μp−μn\|), for the mean (σ) and standard deviations (μ) of the positive and negative controls (Fig EV1A and B). The library of compounds examined includes both previously reported (Srinivasan et al, 2013) and newly synthesized compound series (see Dataset EV1 for individual compound information, and Appendix Supplementary Methods for synthetic protocols). Cells were transfected with FugeneHD reagent (Roche Applied Sciences, Indianapolis, IN) in 384‐well plates. After 24 h, cells were stimulated with 10 μM compounds dispensed using a 100‐nl pintool Biomeck NXP workstation (Beckman Coulter Inc.). Luciferase activity was measured 24 h later (see Appendix Supplementary Methods for more details). HEK293T cells were transfected with 5× UAS‐luciferase reporter, and wild‐type ERα‐VP16 activation domain plus full‐length NCOA1/2/3‐GAL4 DBD fusion protein expression plasmids, using the TransIT‐LT1 transfection reagent (Mirus Bio LLC, Madison, WI). The next day, cells were stimulated with 10 μM compounds using a 100‐nl pintool Biomeck NXP workstation (Beckman Coulter Inc.). Luciferase activity was measured after 24 h (see Appendix Supplementary Methods for more details). MCF‐7 cells were plated on 384‐well plates in phenol red‐free media plus 10% FBS and stimulated with 10 μM compounds using 100‐nl pintool Biomeck NXP workstation (Beckman Coulter Inc.). Cell numbers determined 1 week later (see Appendix Supplementary Methods for more details). MCF‐7 cells were steroid‐deprived and stimulated with compounds for 24 h. Total RNA was extracted and reverse‐transcribed. The cDNA was analyzed using TaqMan Gene Expression Master Mix (Life Technologies, Grand Island, NY), GREB1 and GAPDH (control) primers, and hybridization probes (see Appendix Supplementary Methods for more details). This assay was performed as previously described with the ERα LBD, 10 μM compounds, and a PamChiP peptide microarray (PamGene International) containing 154 unique coregulator peptides (Aarts et al, 2013) (see Appendix Supplementary Methods for more details). ERα protein was produced as previously described (Bruning et al, 2010). New ERα LBD structures (see Dataset EV2 for data collection and refinement statistics) were solved by molecular replacement using PHENIX (Adams et al, 2010), refined using ExCoR as previously described (Nwachukwu et al, 2013), and COOT (Emsley & Cowtan, 2004) for ligand‐docking and rebuilding. Crystal structures analyzed in this study include the following: 1GWR (Warnmark et al, 2002), 3ERD and 3ERT (Shiau et al, 1998), 4ZN9 (Zheng et al, 2012), 4IWC, 4 IU7, 4IV4, 4IVW, 4IW6, 4IUI, 4IV2, 4IVY and 4IW8 (Srinivasan et al, 2013), and 4PPS (Nwachukwu et al, 2014). New crystal structures analyzed in this study were deposited in the RCSB protein data bank (http://www.pdb.org): 4ZN7, 4ZNH, 4ZNS, 4ZNT, 4ZNU, 4ZNV, 4ZNW, 5DI7, 5DID, 5DIE, 5DIG, 5DK9, 5DKB, 5DKE, 5DKG, 5DKS, 5DL4, 5DLR, 5DMC, 5DMF, 5DP0, 5DRM, 5DRJ, 5DTV, 5DU5, 5DUE, 5DUG, 5DUH, 5DXK, 5DXM, 5DXP, 5DXQ, 5DXR, 5EHJ, 5DY8, 5DYB, 5DYD, 5DZ0, 5DZ1, 5DZ3, 5DZH, 5DZI, 5E0W, 5E0X, 5E14, 5E15, 5E19, 5E1C, 5DVS, 5DVV, 5DWE, 5DWG, 5DWI, 5DWJ, 5EGV, 5EI1, 5EIT.
PMC4869123	Inhibiting complex IL-17A and IL-17RA interactions with a linear peptide	IL-17A is a pro-inflammatory cytokine that has been implicated in autoimmune and inflammatory diseases. Monoclonal antibodies inhibiting IL-17A signaling have demonstrated remarkable efficacy, but an oral therapy is still lacking. A high affinity IL-17A peptide antagonist (HAP) of 15 residues was identified through phage-display screening followed by saturation mutagenesis optimization and amino acid substitutions. HAP binds specifically to IL-17A and inhibits the interaction of the cytokine with its receptor, IL-17RA. Tested in primary human cells, HAP blocked the production of multiple inflammatory cytokines. Crystal structure studies revealed that two HAP molecules bind to one IL-17A dimer symmetrically. The N-terminal portions of HAP form a β-strand that inserts between two IL-17A monomers while the C-terminal section forms an α helix that directly blocks IL-17RA from binding to the same region of IL-17A. This mode of inhibition suggests opportunities for developing peptide antagonists against this challenging target.The family of IL-17 cytokines and receptors consists of six polypeptides, IL-17A-F, and five receptors, IL-17RA-E1. IL-17A is secreted from activated Th17 cells, and several innate immune T cell types including macrophages, neutrophils, natural killer cells, and dendritic cells2. IL-17A signals through a specific cell surface receptor complex which consists of IL-17RA and IL-17RC3. IL-17A’s downstream signaling leads to increased production of inflammatory cytokines such as IL-6, IL-8, CCL-20 and CXCL1 by various mechanisms including stimulation of transcription and stabilization of mRNA4567. Although various cell types have been reported to express IL-17RA, the highest responses to IL-17A come from epithelial cells, endothelial cells, keratinocytes and fibroblasts4. IL-17A and its signaling is important in host defense against certain fungal and bacterial infections as demonstrated by patients with autoantibodies against IL-17A and IL-17F, or with inborn errors of IL-17 immunity89. In addition to its physiological role, IL-17A is a key pathogenic factor in inflammatory and autoimmune diseases. In phase II and III clinical trials, neutralizing monoclonal antibodies against IL-17A (secukinumab and ixekizumab)101112 or its receptor IL-17RA (brodalumab)13 are highly efficacious in treating moderate to severe plaque psoriasis and psoriatic arthritis. Secukinumab has been approved recently as a new psoriasis drug by the US Food and Drug Administration (Cosentyx™)14. In addition to psoriasis and psoriatic arthritis, IL-17A blockade has also shown preclinical and clinical efficacies in ankylosing spondylitis and rheumatoid arthritis151617181920. Among IL-17 cytokines, IL-17A and IL-17F share the highest homology. These polypeptides form covalent homodimers, and IL-17A and IL-17F also form an IL-17A/IL-17F hetereodimer21. Structures are known for apo IL-17F22 and its complex with IL-17RA23, for apo IL-17A24, its complex with an antibody Fab25, and its complex with IL-17RA24. In these structures, both IL-17A and IL-17F adopt a cysteine-knot fold with two intramolecular disulfides and two interchain disulfide bonds that covalently link two monomers. There has been active research in identifying orally available chemical entities that would functionally antagonize IL-17A-mediated signaling. Developing small molecules targeting protein-protein interactions is difficult with particular challenges associated with the large, shallow IL-17A/IL-17RA interfaces. Since IL-17RA is a shared receptor for at least IL-17A, IL-17F, IL-17A/IL-17F and IL-17E212226, we chose to seek IL-17A-specific inhibitors that may have more defined pharmacological responses than IL-17RA inhibitors. Our initial approach was to identify peptide inhibitors which could serve as leads for the development of anti-inflammatory therapeutics that could be used alone or in combination with other agents. Our efforts resulted in discovery of a high affinity IL-17A peptide antagonist (HAP), which we attempted to increase the functional production and pharmacokinetics after fusing HAP to antibodies for evaluation as a bispecific therapeutic in animal studies2728. Unfortunately, this past work revealed stability issues of the uncapped HAP in cell culture Here, we provide the details of the discovery and optimization that led to HAP and report the complex structure of IL-17A with HAP, which provides structure based rationalization of peptide optimization and structure activity relationship (SAR). Peptides specifically binding to human IL-17A were identified from phage panning using cyclic and linear peptide libraries (Supplementary Figure S1). Positive phage pools were then sub-cloned into a maltose-binding protein (MBP) fusion system. Single clones were isolated and sub-cultured in growth medium, and culture supernatants were used in an enzyme-linked immunosorbent assay (ELISA) to identify specific IL-17A-binding clones. The positive binding supernatants were tested for the ability to block biotinylated IL-17A signaling through IL-17RA in an IL-17A/IL-17RA competition ELISA assay where unlabeled IL-17A was used as positive control to inhibit biotinylated IL-17A binding. Approximately 10% of the clones that specifically bound to IL-17A also prevented the cytokine from binding to IL-17RA. Sequences identified from phage clones were chemically synthesized (Supplementary Table 1) and tested for inhibition of IL-17A binding to IL-17RA (Table 1). A 15-mer linear peptide 1 was shown to block IL-17A/IL-17RA binding with an IC50 of 80 nM in the competition ELISA assay (Table 1). This peptide was then tested in a cell-based functional assay wherein production of GRO-α in BJ human fibroblast cells was measured as a function of IL-17A stimulation using 1 ng/ml IL-17A. Peptide 1 was found to be active in this functional assay with an IC50 of 370 nM. A SAR campaign was undertaken to improve the potency of peptide 1. An alanine scan of peptide 2, an analogue of 1 with a lysine to arginine substitution at position 14, was initiated. When alanine was already present (positions 7 and 15), substitution was made with lysine (Table 1, peptides 3–17). Positions 1, 2, 4, 5, 7, 14 and 15 were shown to be amenable to substitution without significant loss (less than 3-fold) of binding affinity as measured by the IL-17A/IL-17RA competition ELISA. In particular, at position 5 (13), substitution of methionine with alanine resulted in a seven fold improvement in potency (80 nM versus 11 nM respectively). In order to rapidly evaluate the effects of substitution of natural amino acids at tolerant positions identified by the alanine scan, the lead sequence was subjected to site-specific saturation mutagenesis using MBP29. Each of the seven positions identified by the alanine scan was individually modified while keeping the rest of the sequence constant. Modifications at positions 2 and 14 were shown to display improvement in binding affinity (data not shown). Peptides with beneficial point mutations at positions 2, 5, and 14 were synthesized and evaluated in the competition ELISA (Table 1). Two of the changes, V2H (18) or V2T (21) displayed improved binding in the competition ELISA. Since the replacement of methionine at position 5 with alanine was beneficial, the additional hydrophobic amino acids isoleucine (24), leucine (25) and valine (26) were evaluated and an additional two-three fold improvement in binding was observed for the valine and isoleucine replacements in comparison with alanine. Introduction of a methionine (27) or a carboxamide (28 and 29) at position 14 was shown to improve the binding affinity of the lead peptide. In general, there was good agreement between the respective binding affinities of the synthesized peptides and their MBP fusion counterparts, except for substitution of valine at position 2 to a tryptophan (22), which resulted in a fivefold loss of affinity, for the free peptide when compared with the MBP fusion. Combining the key amino-acid residues identified by SAR into a single peptide sequence resulted in peptide 30, named high affinity peptide (HAP), that was found to inhibit IL-17A signaling in a BJ human fibroblast cell assay with an IC50 of 17 nM, a more than 20-fold improvement over the phage peptide 1 (Table 2 and Supplementary Figure S2). We also examined the effect of removing the acetyl group at the N-terminus of HAP (which is present in all the peptides made, see Supplementary Material). The un-capped peptide (31) had an IC50 of 420 nM in the cell-based assay. The loss of cellular activity of 31 was most likely due to the degradation of the N-terminus of 31, since peptide 31 was shown to be able to bind to IL-17A with similar affinity as HAP itself27. Furthermore, our previous work had reported that in antibody fusions the uncapped peptide was degraded under cell assay conditions with removal of the first 1-3 residues to inactive products with the same N-terminal sequences as peptides 32–3427. In this work, 32–34 are capped by protective acetyl group and reflect the same inactivity as reported. C-terminal truncations showed a more gradual reduction in activity (35–37; Table 2). After deletion of three amino acids from the C-terminal end (37), the peptide is no longer active. We reasoned that since the IL-17A protein is almost exclusively present in a dimeric form3031, dimerizing the IL-17A binding peptides could result in an improvement in binding affinity and inhibitory activity. Homodimers of HAP were made through attachment of polyethylene glycol (PEG) spacers of different lengths at amino acids 4, 7 and 14, as these positions were identified in the alanine scan analysis as not contributing significantly to the activity, and at each N-terminus (Supplementary Table S2). Due to the high reactivity of the pentafluoroester (PFP) group used as the activating group in the PEG, the histidine at position 2 and the lysine at position 15 were replaced with threonine and dimethyllysine respectively to prevent formation of side products, which resulted in peptide 38 that was comparable in activity with HAP. This exercise revealed that several dimeric peptides with the longer PEG21 spacer were significantly more potent than the monomer peptide in the cell-based assay (Supplementary Table S2). Peptide 45, dimerized via attachment of a PEG21 spacer at position 14 (Supplementary Scheme S1 and Figure S3), was the most potent with cellular IC50 of 0.1 nM. This significant improvement in antagonism was not seen in the peptide monomer functionalized with a PEG21 group at position 14 as peptide 48 had an IC50 of 21 nM (Supplementary Scheme S2). The species cross-reactivity of the dimeric peptide 45 and HAP were assessed in a murine functional cell assay using 15 ng/ml murine IL-17A. Peptide 45 blocked the receptor binding of murine IL-17A although with potency two orders of magnitude weaker than that observed against human IL-17A (IC50 = 41 nM vs IC50 = 0.1 nM, respectively). The monomer HAP was much weaker (IC50 >1 μM) in inhibiting murine IL-17A signaling (Supplementary Figure S4). Although the dimeric peptide 45 is much more potent than HAP in the cell-based assay, in subsequent studies we decided to focus our efforts solely on characterizations of the monomeric peptide HAP in hopes to identify smaller peptide inhibitors containing the best minimal functional group. To further characterize the interaction of HAP with IL-17A, we set out to determine its in vitro binding affinity, specificity and kinetic profile using Surface Plasmon Resonance (SPR) methods (Fig. 1A). HAP binds to immobilized human IL-17A homodimer tightly (Table 3). It has slightly weaker affinity for human IL-17A/F heterodimer and >10 fold weaker affinity for mouse IL-17A (Table 3). HAP does not show significant binding to immobilized human IL-17F homodimer or IL-17RA at concentrations up to 100 nM. Additionally, we investigated the antagonism of the human IL-17A/IL-17RA interaction by HAP using orthogonal methods including SPR and Förster resonance energy transfer (FRET) competition assays (Fig. 1B,C). In both assays, incubation of IL-17A with HAP effectively blocks the binding of IL-17A to immobilized IL-17RA with similar sub-nM IC50 (Table 3). While either IL-17A or TNF-α alone can stimulate the release of multiple inflammatory cytokines, when acting together they can synergistically enhance each other’s effects (Supplementary Figure S5). These integrative responses to IL-17A and TNF-α in human keratinocytes have been reported to account for key inflammatory pathogenic circuits in psoriasis32. Thus, we chose to study HAP’s efficacy in blocking the production of IL-8, IL-6 and CCL-20 by primary human keratinocytes stimulated by IL-17A in the presence of TNF-α, an assay which may be more disease-relevant. HAP inhibits the production of all three cytokines in a dose-dependent fashion (Fig. 1D). Significantly, the baseline levels of IL-8, IL-6 and CCL-20 stimulated by TNF-α alone are not inhibited by HAP, further indicating the selectivity of HAP (Fig. 1D). Such pharmacological selectivity may be important to suppress inflammatory pathogenic circuits in psoriasis, while sparing the anti-infectious immune responses produced by TNF-α. The relatively high IC50 values in this assay (Table 3) are probably due to the high IL-17A concentration (100 ng/ml) needed for detection of IL-6. As a reference, a commercial anti-IL-17A antibody (R&D Systems) inhibits the production of IL-8 with an IC50 of 13(±6) nM (N = 3). Indeed, the IC50 was 14(±9) nM (N = 12) for HAP inhibition of IL-8 production when only 5 ng/ml IL-17A was used in this assay. In patients, the concentration of IL-17A in psoriatic lesions is reported to be 0.01 ng/ml, well below the EC50 (5–10ng/ml) of IL-17A induced IL-8 production in vitro333. Similar to keratinocytes assay results, while HAP inhibits IL-17A stimulated IL-6 production by BJ human fibroblast potently (IC50 of 17 nM), it does not inhibit TNF-α stimulated IL-6 production at concentrations up to 10 μM (Supplementary Figure S2). Extensive crystallization trials, either by co-crystallization or by soaking HAP into preformed apo IL-17A crystals24, failed to lead to an IL-17A/HAP complex crystals. We theorized that HAP binding induced large conformational changes in IL-17A that led to the difficulty of getting an IL-17A/HAP binary complex crystal. It is known that an antibody antigen-binding fragment (Fab) can be used as crystallization chaperones in crystallizing difficult targets34. We hypothesized that HAP may target the N-terminal of IL-17A which is known to be more flexible than its C-terminal2425 and conformational changes needed for HAP binding may be more likely there. We designed an antibody Fab known to target the C-terminal half of IL-17A based on a published IL-17A/Fab complex crystal structure25, and produced it in HEK293 cells. In an SPR assay HAP and this Fab were able to co-bind IL-17A without large changes in their binding affinities and kinetics, confirming our hypothesis (Supplementary Figure S6). Furthermore, since it binds to an area far away from that of HAP (see below), this Fab should have minimum effects on HAP binding conformation. Crystals of Fab/IL-17A/HAP ternary complex were obtained readily in crystallization screens. Crystallization of IL-17A and its binding partners was accomplished using two forms of IL-17A. These were, respectively, a presumably more homogeneous form of IL-17A that lacked the disordered N-terminal peptide and a full-length form of the cytokine with a full complement of disulfide bonds. (see Method). Crystals of the Fab/truncated IL-17A/HAP complex diffracted to 2.2 Å, and the Fab/full length IL-17A/HAP complex diffracted to 3.0 Å (Supplementary Table S3). Both structures were solved by molecular replacement. Both complexes crystallized in the space group of P321, with half the complex (1 Fab/1 IL-17A monomer/1 HAP) in the asymmetric unit. The intact complex can be generated by applying crystallographic 2-fold symmetry. Electron densities for HAP residues Ile1-Asn14 were readily interpretable with the exception of Lys15, which is disordered. When considering the protein, the complex structure containing the full length IL-17A is identical to that of the truncated IL-17A, with the exception of Cys106 (Ser106 in the truncated IL-17A), which is disordered. Cys106 is covalently linked to Cys10 that resides in the disordered N-terminal peptide in the full length IL-17A. In a similar manner to the published structure of Fab/IL-17A complex25, two Fab molecules bind symmetrically to the C-terminal of the cytokine dimer, interacting with epitopes from both monomers (Fig. 2A). Two copies of HAP bind to the N-terminal of the cytokine dimer, also symmetrically, and each HAP molecule also interacts with both IL-17A monomers (Fig. 2). Based on disclosed epitopes of Secukinumab and Ixekizumab3536, HAP binds to IL-17A at an area that is also different from those of those two antibodies. The N-terminal 5 residues of HAP, IHVTI, form an amphipathic β-strand that inserts between β-strand 4 of one IL-17A monomer and β-strand 0 (the first ordered peptide of IL-17A) of the second monomer. This β-strand is parallel to both strands 0 and 4 (Fig. 3B). Strands 0 of two IL-17A monomer are antiparallel, as appeared in other IL-17A structures24. The C-terminal 8 residues of the HAP that are ordered in the structure, ADLWDWIN, form an amphipathic α-helix interacting with the second IL-17A monomer. Pro6 of HAP makes a transition between the N-terminal β-strand and the C-terminal α-helix of HAP. As a comparison, an IL-17A/IL-17RA complex structure (PDB code 4HSA) is also shown with IL-17A in the same orientation (Fig. 2C). IL-17RA binds IL-17A at three regions on the IL-17A homodimer24. HAP binds IL-17A at region I. Region I is formed by residues at the ends of β strands 0 and 4, and from loops 1–2 and 3–4 of IL-17A (Fig. 2). Conformational changes in region I induced by HAP binding alone may allosterically affect IL-17RA binding, but more importantly, the α-helix of HAP directly competes with IL-17RA for binding to IL-17A (Fig. 3). The most significant interactions between the α helix of HAP and IL-17A involve Trp12 of HAP, which binds in a hydrophobic pocket in IL-17A formed by the side chains of Phe110, Tyr62, Pro59 and the hydrophobic portion of the Arg101 side chain (Fig. 3A). The Trp12 side chain of HAP donates a hydrogen bond to the main chain oxygen of Pro69 of IL-17A. The positively charged Arg101 side chain of the IL-17A engages in a charge-helix dipole interaction with the main chain oxygen of Trp12. Additionally, Leu9 and Ile13 of the HAP have hydrophobic interactions with IL-17A, and the Asp8 side chain has hydrogen bond and ion pair interactions with Tyr62 and Lys114 of IL-17A, respectively. In region I, an IL-17RA peptide interacts with IL-17A in a very similar fashion to the α-helix of HAP. The IL-17RA peptide has sequences of LDDSWI, and part of the peptide is also α-helical (Fig. 3B). Leu7, Trp31 and Ile32 of IL-17RA interact very similarly with the same residues of IL-17A as Leu9, Trp12 and Ile13 of HAP (Fig. 3B). In this sense, the α-helix of HAP with a sequence of LWDWI is a good mimetic of the LDDSWI peptide of IL-17RA. The β-strand of HAP has no equivalent in IL-17RA. However, it mimics the β-strand 0 of IL-17A. The amphipathic β-strand of HAP orients the hydrophilic side chains of His2 and Thr4 outwards, and the hydrophobic side chains of Ile1, Val3 and Ile5 inward (Fig. 3A). β-strand 0 in IL-17A is also amphipathic with the sequence of TVMVNLNI. In all IL-17A structures obtained to date, β-strand 0 orients the hydrophilic side chains of Thr21, Asn25 and Asn27 outward, and the hydrophobic side chains of Val22, Val24, Leu26 and Ile28 inward. The binding pocket occupied by either Trp12 of HAP or Trp31 of IL-17RA is not formed in the apo IL-17A structure (Fig. 3C). Conformational changes of IL-17A are needed for both HAP and IL-17RA to bind to that region. Particularly for HAP, β-strands 0 have to shift out of the hydrophobic cleft formed by the main body of the IL-17A by as much as 10 Å between Cα atoms (Fig. 3C). Disruptions of the apo IL-17A structure by HAP binding are apparently compensated for by formation of the new interactions that involve almost the entire HAP molecule (Fig. 3B). The IL-17A/HAP complex structure obtained is very consistent with the observed SAR of our identified peptide inhibitors, explaining well how the evolution of the initial phage peptide 1 to HAP and 45 improved its potency (Supplementary Figure S7). The important interactions involving Trp12 of HAP explain the >90 times drop in potency of the W12A variant (6 vs 1, Table 1). The amphipathic nature of the HAP β-strand explains the preference of the hydrophilic residues at the 2 and 4 positions of peptides (14, 18, 19, 21 and 23 vs 1 and 22, Table 1). All N-terminal residues of HAP are part of the β-sheet with β-stands 0 and 4 of IL-17A, which explains why removal of the first 1–3 residues completely abolishes the ability of HAP to block IL-17A cell signaling (31,32 and 33, Table 2). The C-terminal Asn14 and Lys15 of HAP are not directly involved in interactions with IL-17A, and this is reflected in the gradual reduction in activity caused by C-terminal truncations (35 and 36, Table 2). Each peptide monomer in 45 may not necessarily be more potent than HAP, but two monomer peptides within the same molecule that can simultaneously bind to IL-17A can greatly improve its potency due to avidity effects. HAP targets region I of IL-17A, an area that has the least sequence conservation in IL-17 cytokines2224. This lack of sequence conservation in the HAP binding site explains the observed specificity of HAP binding to human IL-17A. For example, inspection of the published IL-17F crystal structure (PDB code 1JPY) revealed a pocket of IL-17F similar to that of IL-17A for W12 of HAP binding, but it is occupied by a Phe-Phe motif at the N-terminal peptide of IL-17F. This Phe-Phe motif is missing in IL-17A. Sequence alignments between human and mouse IL-17A24 indicated that among IL-17A residues that interacting with HAP, majority differences occur in strand 0 of IL-17A which interacts with the N-terminal β-strand of HAP. In human IL-17A the sequences are TVMVNLNI, and in mouse they are NVKVNLKV. Using a combination of phage display and SAR we have discovered novel peptides that are IL-17A antagonists. One of those peptides, HAP, also shows activity in inhibiting the production of multiple inflammatory cytokines by primary human keratinocytes stimulated by IL-17A and TNF-α, a disease relevant-model. We have also determined the complex structure of IL-17A/HAP, which provides the structural basis for HAP’s antagonism to IL-17A signaling. During IL-17A signaling, IL-17A binds to one copy of IL-17RA and one copy of IL-17RC212324. Since apo IL-17A is a homodimer with 2 fold symmetry, IL-17RA potentially can bind to either face of the IL-17A dimer. With two HAP molecules covering both faces of the IL-17A dimer, HAP can block IL-17RA approaching from either face. To form the 1:2 complex observed in crystal structure, it is important that there is no strong negative cooperativity in the binding of two HAP molecules. In fact, in native electrospray ionization mass spectrometry analysis only 1:2 IL-17A/HAP complex was observed even when IL-17A was in excess (Supplementary Figure S8), indicating a positive binding cooperativity that favors inhibition of IL-17RA binding by HAP. HAP, with only 15 residues, can achieve almost the same binding affinity as the much larger IL-17RA molecule, indicating a more efficient way of binding to IL-17A. The interaction of IL-17A with IL-17RA has an extensive interface, covering ~2,200 Å surface area of IL-17A24. Due to the discontinuous nature of the IL-17A/IL-17RA binding interface, it is classified as having tertiary structural epitopes on both binding partners, and is therefore hard to target using small molecules37. Our studies of HAP demonstrated an uncommon mode of action for a peptide in inhibiting such a difficult protein-protein interaction target, and suggest further possible improvements in its binding potency. One way of further improving HAP’s potency is by dimerization. Homo-dimerization of HAP (45) achieved sub-nanomolar potency against human IL-17A in cell assay. In the crystal structure, the distance between the carbonyl of Asn14 of one HAP molecule and the N-terminus of the second is only 15.7 Å, suggesting the potential for more potent dimeric peptides to be designed by using linkers of different lengths at different positions. Another direction of improving HAP is by reducing its size. As demonstrated by the crystal structure, binding of the α-helix of HAP should be sufficient for preventing IL-17RA binding to IL-17A. Theoretically, it is possible to design chemicals such as stapled α-helical peptides to block α-helix-mediated IL-17A/IL-17RA interactions. Such peptides may have smaller sizes with more favorable physical properties (proteolytic resistance, serum half-life, permeability, etc)383940. In summary, these peptide-based anti-IL-17A modalities could be further developed as alternative therapeutic options to the reported monoclonal antibodies. We are also very interested in finding non-peptidic small molecule IL-17A antagonists, and HAP can be used as an excellent tool peptide. The strategy utilized in generating the complex structures of HAP may also be useful for enabling structure based design of some known small molecule IL-17A antagonists41. Peptides were synthesized either on a PTI Symphony or Biotage Syro II synthesizer employing standard Fmoc chemistry on Rink resin with N-terminal amine capped with acetyl group unless noted otherwise. Resin, amino acids and solvents were purchased from EMD Chemicals. Peptides were then removed from polymer support using the cleavage cocktail (88:5:5:2, TFA:PhOH:H2O:TIPS) and precipitated with cold diethyl ether. For the dimers, the native peptides incorporating a primary amine (N-terminal amine or lysine at specific positions in the peptide sequence) were mixed with the bis-pentafluoroesters of the desired PEG (purchased from Quanta BioDesign) in the presence of DIEA in DMF. Peptides were purified by reversed phase HPLC on Phenomenex preparative Luna C18 columns using water:acetonitrile gradients and lyophilized. All peptides were assessed for purity by analytical C18 RP-HP-LCMS prior to use in biological assays. Buffers used were 0.1% trifluoroacetic acid in water (A) and 0.1% trifluoroacetic acid in acetonitrile (B). The standard method (1) consisted of a linear gradient of 5% to 95% B over 10 minutes on Agilent 1100 series HPLC-MSD and method (2) consisted of a linear gradient of 5% to 95% B over 8 minutes on Waters 2767 series HPLC-MSD. The C18 column (Phenomenex, Luna C18, 4.6 × 150mm) effluent was immediately mass analyzed in electrospray positive mode. Accurate mass measurements of final peptides were performed using C18 reversed-phase chromatography mass spectrometry (RPHPLCMS) and mass detected on a Waters Synapt G2 Q-Tof mass spectrometer tuned to a resolution (FWHM) of 25,000. Exact intact masses were calculated based on the monoisotopic m/z value of the base peak charge state. All peptides were analyzed using these methods. Supporting data of peptide 45 is shown (Supplementary Figure S2) as representative data sets of the all the molecules investigated (characterization of all peptides presented on Table 1S). rhIL-17 (#317-ILB), IL-17R (#2269-IL) were from R&D Systems. ELISA wash buffer (50-63-01) and TMB SureBlue Microwell Peroxidase substrate (#50-63-01 or #52-00-00) were from KPL. Half-well high-binding ELISA plates (#3690) and full-well plates (#9018) were from Costar. Superblock (#2011 or AAA500) was from ScyTek. No-Weigh NSH-PEG4-Biotin (21329) and Zeba Desalt Spin Columns (89882) were from Pierce. ELISAs were as follows in duplicate unless otherwise specified. Plates were coated in PBS overnight at 4 °C and blocked using Superblock for one hour at RT. Subsequent incubations were for 1 hr at room temperature with dilutions in Superblock. Plates were washed 3X between steps using a Biotek ELx405 plate washer and developed using TMB substrate. Reactions were stopped in 2 M H2SO4 and OD450 values were read on a Molecular Devices SpectraMax Plus plate reader. IL-17 RA competition: half-well plates were coated with 0.5 μg/mL IL-17R-Fc and blocked. Peptides were titrated in Superblock containing 0.4 μg/mL Biotinylated IL-17A. Biotinylated IL-17A was detected by adding tetramethylbenzidine (TMB). Reactions were stopped in 2 M H2SO4 and OD450 values were read on a Molecular Devices SpectraMax Plus plate reader. For inhibitor screen, human BJ fibroblast cells (ATCC CRL2522) (American Type Culture Collection, VA) were used. For mouse cell based assay MLE-12 mouse epithelial cells (ATCC CRL2110) were used. Both cell lines were maintained in ATCC recommended media. Cells were seeded at 5 × 10 cells/well into 96-well flat-bottom microtiter plates in which peptides that had been pre-diluted with cytokines (1 ng/mL for human IL-17A or 15 ng/mL for mouse IL-17A) in culture medium. Cells were incubated at 37 °C for 16–24 hrs, and supernatants were collected and analyzed by ELISA for either human CXCL1/GRO-α (R&D Systems DY275) or mouse CXCL1/KC (R&D Systems DY453). Primary human keratinocytes were cultured in Epilife medium with EDGS (Life Technology, Cascade Biologics) following product instructions. 5 days after establishing the culture from frozen vials, cells were plated at 10,000/well (80 μl) in culture media in 384 well plate. 4 hours after plating the cells, 10 μl of 10X peptide stocks were added. Final DMSO concentration was 1%. Immediately after peptide addition, 10 μl of a mixture of recombinant human IL-17A (endotoxin Level <0.10 EU per 1 μg of the protein, E.coli expression, >97% pure judged by SDS/page, R&D System, Minneapolis, MN) 100 ng /ml (final) and TNF-α (Sigma-Aldrich, St Louis, MO) 10 ng/ ml (final), or TNF-α 10 ng/ml only were added to the cells. Cell assay plates were incubated for 48 hours at 37 °C in a tissue culture incubator before harvesting culture supernatants for analysis of IL-6, IL-8 and CCL-20 production using kits K211AKB-2, K211ANB-2, and K211BEB-2, respectively (Meso Scale Discovery, Rockville, MD). Details of non-commercial protein constructs used in this study and their purifications are in the Supplementary Material. The IL-17 SPR binding assay was run on a Biacore 3000 SPR instrument (GE Healthcare). Biotinylated human IL-17A, or IL-17A/F heteromer, or IL-17F, or mouse IL-17A (Cell Signaling) was captured on a Biacore Streptavidin chip to achieve protein density of about 2500 to 3500 RUs on the surface. The SPR running buffer was 10 mM HEPES, pH 7.4, 150 mM NaCl, 0.01% P20 with 3% DMSO. Peptide samples were injected at a flow rate of 50 μl/min for 180 seconds association and at least 600 seconds dissociation using a 2–3 fold dilution series. Multiple blank injections were run before and after each peptide series for references. The data were processed and analyzed with Scrubber 2.0 (BioLogical Software) and Biaeval software (GE Healthcare) to calculate binding constants and on and off rates. The IL-17A SPR competition assay was run on a BioRAD ProteON instrument (BioRAD Laboratories). IL-17RA-Fc fusion protein (R&D Systems) was captured on a GLM chip using the standard amine coupling reaction. The chip surface activated using a mixture of sulfo-NHS and EDC, was exposed to the receptor dissolved in acetate buffer pH 5.0 at 0.01 mg/ml concentration. The surface was deactivated using 1 M ethanolamine HCl. An adjacent flow cell on the chip was treated identically without the protein and was used as a reference surface for subsequent data analysis. Three-fold dilution series of peptides mixed with 5 nM IL-17A were injected at a flow rate of 50 μl/min for 120 seconds, followed by 300 seconds of dissociation. The receptor on the surface was regenerated using a 30 second injection of 3 M MgCl2. Observed signal between 180 and 200 seconds was averaged for each sample. Average response observed for multiple 5 nM IL-17A samples without any peptide was used as 100% signal for calculating % inhibition for each concentration of compound. The data were fit with Microcal Origin software (OriginLab, MA) to calculate IC50 for peptides. The FRET signal of a Eu-labeled IL-17A donor and an Alexa Fluor 647 labeled IL-17RA acceptor was measured to monitor the interaction of IL-17A and IL-17RA. Maximal FRET was observed when IL-17A was bound to IL-17RA and diminished FRET was observed when IL-17A was separated from IL-17RA. The excitation of the donor at a wavelength of 320 nm triggers fluorescence at 615 nM and this in turn serves to excite the acceptor, which then fluoresces at a wavelength of 665 nm. The fluorescence at both 615 nm and 665 nm were measured and the ratio of 665 nm/615 nm was used to monitor the IL-17A/IL-17RA binding. Final assay concentrations were 1 nM biotinylated IL-17A labeled with 0.67 nM Europium-Streptavidin (Invitrogen), 6 nM IL-17RA-Fc fusion protein (R&D Systems) labeled with 1 nM Alexa Fluor 647 antibody (BioLegand) in a buffer containing 10 mM HEPES pH 7.4, 150 mM NaCl, 0.02% BSA, and 0.01% Tween 20. Peptides were tested using a half log dilution series of 11 concentrations. The IL-17A was incubated with the Europium-Streptavidin to 1.5X final assay concentration for one hour at room temperature. Peptides were prepared at 50X final concentration in 100% DMSO and 300 nl were added to a 384-well white assay plate (Greiner). 10 μl of the Eu-labeled IL-17A was added to the peptides and incubated at room temperature for one hour. During this pre-incubation of peptide and IL-17A, IL-17RA was incubated with Alexa Fluor 647 antibody to 3X final assay concentration at room temperature for one hour and then 5 μl of the 3X Alexa Fluor 647 labeled IL-17RA was added to the assay for a total volume of 15.3 μl and a final DMSO concentration of 2%. The plates were covered and incubated at room temperature for 3 hours. The FRET signal of the IL-17A/IL-17RA interaction was measured using an EnVision Multilabel plate reader (PerkinElmer). The peptide data was converted into % inhibition, using 0% (no HAP) and 100% inhibition (100 nM HAP) as controls. A four parameter logistic nonlinear regression model using the percent inhibition at each concentration was used to calculate an IC50 for each peptide. To crystallize Fab/IL-17A/HAP complex, 30 mM HAP in DMSO stock was added to the Fab/IL-17A complex to a final concentration of 1 mM. The complex was then screened for crystals using commercial screen kits using a sitting drop vapor diffusion format. Crystals of Fab/truncated IL-17A/HAP complex were obtained under condition of 0.02 mM CdCl2, 0.02 M MgCl2, 0.02 M NiCl2, 0.1 M NaOAc pH = 4.2–4.9, and 24–28% PEG MME 2000. Crystals of Fab/full length IL-17A covalent dimer/HAP complex were obtained under conditions of 0.1 M NaOAc, pH = 4.5 and 30% PEG MME 5000. Crystals were soaked briefly in cryo solutions of the mother liquor supplemented with 25% glycerol before flash cooled in liquid nitrogen. Crystal data sets were collected at APS IMCA 17ID beamline (Chicago, IL), processed with autoPROC42. Data collection statistics are listed in Supplementary Table 2. Fab/IL-17A/HAP complex structures were solved with molecular replacement method using the published FAN/IL-17A crystal structure (pdb code 2VXS), using program Phaser43. Structure refinements was carried out using program Buster44 and manual model building using program COOT45. Final refinement statistics are listed in Supplementary Table 2. How to cite this article: Liu, S. et al. Inhibiting complex IL-17A and IL-17RA interactions with a linear peptide. Sci. Rep. 6, 26071; doi: 10.1038/srep26071 (2016).
PMC4841544	Molecular Basis of Ligand-Dependent Regulation of NadR, the Transcriptional Repressor of Meningococcal Virulence Factor NadA	Neisseria adhesin A (NadA) is present on the meningococcal surface and contributes to adhesion to and invasion of human cells. NadA is also one of three recombinant antigens in the recently-approved Bexsero vaccine, which protects against serogroup B meningococcus. The amount of NadA on the bacterial surface is of direct relevance in the constant battle of host-pathogen interactions: it influences the ability of the pathogen to engage human cell surface-exposed receptors and, conversely, the bacterial susceptibility to the antibody-mediated immune response. It is therefore important to understand the mechanisms which regulate nadA expression levels, which are predominantly controlled by the transcriptional regulator NadR (Neisseria adhesin A Regulator) both in vitro and in vivo. NadR binds the nadA promoter and represses gene transcription. In the presence of 4-hydroxyphenylacetate (4-HPA), a catabolite present in human saliva both under physiological conditions and during bacterial infection, the binding of NadR to the nadA promoter is attenuated and nadA expression is induced. NadR also mediates ligand-dependent regulation of many other meningococcal genes, for example the highly-conserved multiple adhesin family (maf) genes, which encode proteins emerging with important roles in host-pathogen interactions, immune evasion and niche adaptation. To gain insights into the regulation of NadR mediated by 4-HPA, we combined structural, biochemical, and mutagenesis studies. In particular, two new crystal structures of ligand-free and ligand-bound NadR revealed (i) the molecular basis of ‘conformational selection’ by which a single molecule of 4-HPA binds and stabilizes dimeric NadR in a conformation unsuitable for DNA-binding, (ii) molecular explanations for the binding specificities of different hydroxyphenylacetate ligands, including 3Cl,4-HPA which is produced during inflammation, (iii) the presence of a leucine residue essential for dimerization and conserved in many MarR family proteins, and (iv) four residues (His7, Ser9, Asn11 and Phe25), which are involved in binding 4-HPA, and were confirmed in vitro to have key roles in the regulatory mechanism in bacteria. Overall, this study deepens our molecular understanding of the sophisticated regulatory mechanisms of the expression of nadA and other genes governed by NadR, dependent on interactions with niche-specific signal molecules that may play important roles during meningococcal pathogenesis.The ‘Reverse Vaccinology’ approach was pioneered to identify antigens for a protein-based vaccine against serogroup B Neisseria meningitidis (MenB), a human pathogen causing potentially-fatal sepsis and invasive meningococcal disease . Indeed, Reverse Vaccinology identified Neisseria adhesin A (NadA), a surface-exposed protein involved in epithelial cell invasion and found in ~30% of clinical isolates [2–4]. Recently, we reported the crystal structure of NadA, providing insights into its biological and immunological functions . Recombinant NadA elicits a strong bactericidal immune response and is therefore included in the Bexsero vaccine that protects against MenB and which was recently approved in over 35 countries worldwide . Previous studies revealed that nadA expression levels are mainly regulated by the Neisseria adhesin A Regulator (NadR) . Although additional factors influence nadA expression, we focused on its regulation by NadR, the major mediator of NadA phase variable expression [8, 9]. Studies of NadR also have broader implications, since a genome-wide analysis of MenB wild-type and nadR knock-out strains revealed that NadR influences the regulation of > 30 genes, including maf genes, from the multiple adhesin family . These genes encode a wide variety of proteins connected to many biological processes contributing to bacterial survival, adaptation in the host niche, colonization and invasion [11, 12]. NadR belongs to the MarR (Multiple Antibiotic Resistance Regulator) family, a group of ligand-responsive transcriptional regulators ubiquitous in bacteria and archaea. MarR family proteins can promote bacterial survival in the presence of antibiotics, toxic chemicals, organic solvents or reactive oxygen species [13, 14] and can regulate virulence factor expression . MarR homologues can act either as transcriptional repressors or as activators . Although > 50 MarR family structures are known, a molecular understanding of their ligand-dependent regulatory mechanisms is still limited, often hampered by lack of identification of their ligands and/or DNA targets. A potentially interesting exception comes from the ligand-free and salicylate-bound forms of the Methanobacterium thermoautotrophicum protein MTH313 which revealed that two salicylate molecules bind to one MTH313 dimer and induce large conformational changes, apparently sufficient to prevent DNA binding . However, the homologous archeal Sulfolobus tokodaii protein ST1710 presented essentially the same structure in ligand-free and salicylate-bound forms, apparently contrasting the mechanism proposed for MTH313 . Despite these apparent differences, MTH313 and ST1710 bind salicylate in approximately the same site, between their dimerization and DNA-binding domains. However, it is unknown whether salicylate is a relevant in vivo ligand of either of these two proteins, which share ~20% sequence identity with NadR, rendering unclear the interpretation of these findings in relation to the regulatory mechanisms of NadR or other MarR family proteins . NadR binds the nadA promoter and represses gene transcription . NadR binds nadA on three different operators (OpI, OpII and OpIII) . The DNA-binding activity of NadR is attenuated in vitro upon addition of various hydroxyphenylacetate (HPA) derivatives, including 4-HPA. 4-HPA is a small molecule derived from mammalian aromatic amino acid catabolism and is released in human saliva, where it has been detected at micromolar concentration . In the presence of 4-HPA, NadR is unable to bind the nadA promoter and nadA gene expression is induced [9, 10]. In vivo, the presence of 4-HPA in the host niche of N. meningitidis serves as an inducer of NadA production, thereby promoting bacterial adhesion to host cells . Further, we recently reported that 3Cl,4-HPA, produced during inflammation, is another inducer of nadA expression . Extending our previous studies based on hydrogen-deuterium exchange mass spectrometry (HDX-MS) , here we sought to reveal the molecular mechanisms and effects of NadR/HPA interactions via X-ray crystallography, NMR spectroscopy and complementary biochemical and in vivo mutagenesis studies. We obtained detailed new insights into ligand specificity, how the ligand allosterically influences the DNA-binding ability of NadR, and the regulation of nadA expression, thus also providing a deeper structural understanding of the ligand-responsive MarR super-family. Moreover, these findings are important because the activity of NadR impacts the potential coverage provided by anti-NadA antibodies elicited by the Bexsero vaccine and influences host-bacteria interactions that contribute to meningococcal pathogenesis . Recombinant NadR was produced in E. coli using an expression construct prepared from N. meningitidis serogroup B strain MC58. Standard chromatographic techniques were used to obtain a highly purified sample of NadR (see Materials and Methods). In analytical size-exclusion high-performance liquid chromatography (SE-HPLC) experiments coupled with multi-angle laser light scattering (MALLS), NadR presented a single species with an absolute molecular mass of 35 kDa (S1 Fig). These data showed that NadR was dimeric in solution, since the theoretical molecular mass of the NadR dimer is 33.73 kDa; and, there was no change in oligomeric state on addition of 4-HPA. The thermal stability of NadR was examined using differential scanning calorimetry (DSC). Since ligand-binding often increases protein stability, we also investigated the effect of various HPAs (Fig 1A) on the melting temperature (Tm) of NadR. As a control of specificity, we also tested salicylate, a known ligand of some MarR proteins previously reported to increase the Tm of ST1710 and MTH313 . The Tm of NadR was 67.4 ± 0.1°C in the absence of ligand, and was unaffected by salicylate. However, an increased thermal stability was induced by 4-HPA and, to a lesser extent, by 3-HPA. Interestingly, NadR displayed the greatest Tm increase upon addition of 3Cl,4-HPA (Table 1 and Fig 1B). (A) Molecular structures of 3-HPA (MW 152.2), 4-HPA (MW 152.2), 3Cl,4-HPA (MW 186.6) and salicylic acid (MW 160.1). (B) DSC profiles, colored as follows: apo-NadR (violet), NadR+salicylate (red), NadR+3-HPA (green), NadR+4-HPA (blue), NadR+3Cl,4-HPA (pink). All DSC profiles are representative of triplicate experiments. Dissociation constants (KD) of the NadR/ligand interactions from SPR steady-state binding experiments. n.a.: not applicable; n.d.: not determinable To further investigate the binding of HPAs to NadR, we used surface plasmon resonance (SPR). The SPR sensorgrams revealed very fast association and dissociation events, typical of small molecule ligands, thus prohibiting a detailed study of binding kinetics. However, steady-state SPR analyses of the NadR-HPA interactions allowed determination of the equilibrium dissociation constants (KD) (Table 1 and S2 Fig). The interactions of 4-HPA and 3Cl,4-HPA with NadR exhibited KD values of 1.5 mM and 1.1 mM, respectively. 3-HPA showed a weaker interaction, with a KD of 2.7 mM, while salicylate showed only a very weak response that did not reach saturation, indicating a non-specific interaction with NadR. A ranking of these KD values showed that 3Cl,4-HPA was the tightest binder, and thus matched the ranking of ligand-induced Tm increases observed in the DSC experiments. Although these KD values indicate rather weak interactions, they are similar to the values reported previously for the MarR/salicylate interaction (KD ~1 mM) and the MTH313/salicylate interaction (KD 2–3 mM) , and approximately 20-fold tighter than the ST1710/salicylate interaction (KD ~20 mM) . To fully characterize the NadR/HPA interactions, we sought to determine crystal structures of NadR in ligand-bound (holo) and ligand-free (apo) forms. First, we crystallized NadR (a selenomethionine-labelled derivative) in the presence of a 200-fold molar excess of 4-HPA. The structure of the NadR/4-HPA complex was determined at 2.3 Å resolution using a combination of the single-wavelength anomalous dispersion (SAD) and molecular replacement (MR) methods, and was refined to R work/R free values of 20.9/26.0% (Table 2). Despite numerous attempts, we were unable to obtain high-quality crystals of NadR complexed with 3Cl,4-HPA, 3,4-HPA, 3-HPA or DNA targets. However, it was eventually possible to crystallize apo-NadR, and the structure was determined at 2.7 Å resolution by MR methods using the NadR/4-HPA complex as the search model. The apo-NadR structure was refined to R work/R free values of 19.1/26.8% (Table 2). Statistics for the highest-resolution shell are shown in parentheses. R sym = Σhkl Σi \|Ii(hkl)—<I(hkl)>\| / Σhkl Σi Ii(hkl) * R meas = redundancy-independent (multiplicity-weighted) R merge as reported from AIMLESS . R work = Σ\|\|F(obs)\|- \|F(calc)\|\|/Σ\|F(obs)\| R free = as for R work, calculated for 5.0% of the total reflections, chosen at random, and omitted from refinement. Values obtained using Molprobity . The asymmetric unit of the NadR/4-HPA crystals (holo-NadR) contained one NadR homodimer, while the apo-NadR crystals contained two homodimers. In the apo-NadR crystals, the two homodimers were related by a rotation of ~90°; the observed association of the two dimers was presumably merely an effect of crystal packing, since the interface between the two homodimers is small (< 550 Å of buried surface area), and is not predicted to be physiologically relevant by the PISA software . Moreover, our SE-HPLC/MALLS analyses (see above) revealed that in solution NadR is dimeric, and previous studies using native mass spectrometry (MS) revealed dimers, not tetramers . The NadR homodimer bound to 4-HPA has a dimerization interface mostly involving the top of its ‘triangular’ form, while the two DNA-binding domains are located at the base (Fig 2A). High-quality electron density maps allowed clear identification of the bound ligand, 4-HPA (Fig 2B). The overall structure of NadR shows dimensions of ~50 × 65 × 50 Å and a large homodimer interface that buries a total surface area of ~ 4800 Å. Each NadR monomer consists of six α-helices and two short β-strands, with helices α1, α5, and α6 forming the dimer interface. Helices α3 and α4 form a helix-turn-helix motif, followed by the “wing motif” comprised of two short antiparallel β-strands (β1-β2) linked by a relatively long and flexible loop. Interestingly, in the α4-β2 region, the stretch of residues from R64-R91 presents seven positively-charged side chains, all available for potential interactions with DNA. Together, these structural elements constitute the winged helix-turn-helix (wHTH) DNA-binding domain and, together with the dimeric organization, are the hallmarks of MarR family structures . (A) The holo-NadR homodimer is depicted in green and blue for chains A and B respectively, while yellow sticks depict the 4-HPA ligand (labelled). For simplicity, secondary structure elements are labelled for chain B only. Red dashes show hypothetical positions of chain B residues 88–90 that were not modeled due to lack of electron density. (B) A zoom into the pocket occupied by 4-HPA shows that the ligand contacts both chains A and B; blue mesh shows electron density around 4-HPA calculated from a composite omit map (omitting 4-HPA), using phenix . The map is contoured at 1σ and the figure was prepared with a density mesh carve factor of 1.7, using Pymol (www.pymol.org). The NadR dimer interface is formed by at least 32 residues, which establish numerous inter-chain salt bridges or hydrogen bonds, and many hydrophobic packing interactions (Fig 3A and 3B). To determine which residues were most important for dimerization, we studied the interface in silico and identified several residues as potential mediators of key stabilizing interactions. Using site-directed mutagenesis, a panel of eight mutant NadR proteins was prepared (including mutations H7A, S9A, N11A, D112A, R114A, Y115A, K126A, L130K and L133K), sufficient to explore the entire dimer interface. Each mutant NadR protein was purified, and then its oligomeric state was examined by analytical SE-HPLC. Almost all the mutants showed the same elution profile as the wild-type (WT) NadR protein. Only the L130K mutation induced a notable change in the oligomeric state of NadR (Fig 3C). Further, in SE-MALLS analyses, the L130K mutant displayed two distinct species in solution, approximately 80% being monomeric (a 19 kDa species), and only 20% retaining the typical native dimeric state (a 35 kDa species) (Fig 3D), demonstrating that Leu130 is crucial for stable dimerization. It is notable that L130 is usually present as Leu, or an alternative bulky hydrophobic amino acid (e.g. Phe, Val), in many MarR family proteins, suggesting a conserved role in stabilizing the dimer interface. In contrast, most of the other residues identified in the NadR dimer interface were poorly conserved in the MarR family. (A) Both orientations show chain A, green backbone ribbon, colored red to highlight all locations involved in dimerization; namely, inter-chain salt bridges or hydrogen bonds involving Q4, S5, K6, H7, S9, I10, N11, I15, Q16, R18, D36, R43, A46, Q59, C61, Y104, D112, R114, Y115, D116, E119, K126, E136, E141, N145, and the hydrophobic packing interactions involving I10, I12, L14, I15, R18, Y115, I118, L130, L133, L134 and L137. Chain B, grey surface, is marked blue to highlight residues probed by site-directed mutagenesis (E136 only makes a salt bridge with K126, therefore it was sufficient to make the K126A mutation to assess the importance of this ionic interaction; the H7 position is labelled for monomer A, since electron density was lacking for monomer B). (B) A zoom into the environment of helix α6 to show how residue L130 chain B (blue side chain) is a focus of hydrophobic packing interactions with L130, L133, L134 and L137 of chain A (red side chains). (C) SE-HPLC analyses of all mutant forms of NadR are compared with the wild-type (WT) protein. The WT and most of the mutants show a single elution peak with an absorbance maximum at 17.5 min. Only the mutation L130K has a noteworthy effect on the oligomeric state, inducing a second peak with a longer retention time and a second peak maximum at 18.6 min. To a much lesser extent, the L133K mutation also appears to induce a ‘shoulder’ to the main peak, suggesting very weak ability to disrupt the dimer. (D) SE-HPLC/MALLS analyses of the L130K mutant, shows 20% dimer and 80% monomer. The curves plotted correspond to Absorbance Units (mAU) at 280nm wavelength (green), light scattering (red), and refractive index (blue). The NadR/4-HPA structure revealed the ligand-binding site nestled between the dimerization and DNA-binding domains (Fig 2). The ligand showed a different position and orientation compared to salicylate complexed with MTH313 and ST1710 [17, 18] (see Discussion). The binding pocket was almost entirely filled by 4-HPA and one water molecule, although there also remained a small tunnel 2-4Å in diameter and 5-6Å long leading from the pocket (proximal to the 4-hydroxyl position) to the protein surface. The tunnel was lined with rather hydrophobic amino acids, and did not contain water molecules. Unexpectedly, only one monomer of the holo-NadR homodimer contained 4-HPA in the binding pocket, whereas the corresponding pocket of the other monomer was unoccupied by ligand, despite the large excess of 4-HPA used in the crystallization conditions. Inspection of the protein-ligand interaction network revealed no bonds from NadR backbone groups to the ligand, but several key side chain mediated hydrogen (H)-bonds and ionic interactions, most notably between the carboxylate group of 4-HPA and Ser9 of chain A (SerA9), and chain B residues TrpB39, ArgB43 and TyrB115 (Fig 4A). At the other ‘end’ of the ligand, the 4-hydroxyl group was proximal to AspB36, with which it may establish an H-bond (see bond distances in Table 3). The water molecule observed in the pocket was bound by the carboxylate group and the side chains of SerA9 and AsnA11. A) A stereo-view zoom into the binding pocket showing side chain sticks for all interactions between NadR and 4-HPA. Green and blue ribbons depict NadR chains A and B, respectively. 4-HPA is shown in yellow sticks, with oxygen atoms in red. A water molecule is shown by the red sphere. H-bonds up to 3.6Å are shown as dashed lines. The entire set of residues making H-bonds or non-bonded contacts with 4-HPA is as follows: SerA9, AsnA11, LeuB21, MetB22, PheB25, LeuB29, AspB36, TrpB39, ArgB43, ValB111 and TyrB115 (automated analysis performed using PDBsum and verified manually). Residues AsnA11 and ArgB18 likely make indirect yet local contributions to ligand binding, mainly by stabilizing the position of AspB36. Bond distances for interacting polar atoms are provided in Table 3. Side chains mediating hydrophobic interactions are shown in orange. (B) A model was prepared to visualize putative interactions of 3Cl,4-HPA (pink) with NadR, revealing the potential for additional contacts (dashed lines) of the chloro moiety (green stick) with LeuB29 and AspB36. * Bond distance between the ligand carboxylate group and the water molecule, which in turn makes H-bond to the SerA9 and AsnA11 side chains. In addition to the H-bonds involving the carboxylate and hydroxyl groups of 4-HPA, binding of the phenyl moiety appeared to be stabilized by several van der Waals’ contacts, particularly those involving the hydrophobic side chain atoms of LeuB21, MetB22, PheB25, LeuB29 and ValB111 (Fig 4A). Notably, the phenyl ring of PheB25 was positioned parallel to the phenyl ring of 4-HPA, potentially forming π-π parallel-displaced stacking interactions. Consequently, residues in the 4-HPA binding pocket are mostly contributed by NadR chain B, and effectively created a polar ‘floor’ and a hydrophobic ‘ceiling’, which house the ligand. Collectively, this mixed network of polar and hydrophobic interactions endows NadR with a strong recognition pattern for HPAs, with additional medium-range interactions potentially established with the hydroxyl group at the 4-position. We modelled the binding of other HPAs by in silico superposition onto 4-HPA in the holo-NadR structure, and thereby obtained molecular explanations for the binding specificities of diverse ligands. For example, similar to 4-HPA, the binding of 3Cl,4-HPA could involve multiple bonds towards the carboxylate group of the ligand and some to the 4-hydroxyl group. Additionally, the side chains of LeuB29 and AspB36 would be only 2.6–3.5 Å from the chlorine atom, thus providing van der Waals’ interactions or H-bonds to generate the additional binding affinity observed for 3Cl,4-HPA (Fig 4B). The presence of a single hydroxyl group at position 2, as in 2-HPA, rather than at position 4, would eliminate the possibility of favorable interactions with AspB36, resulting in the lack of NadR regulation by 2-HPA described previously . Finally, salicylate is presumably unable to specifically bind NadR due to the 2-hydroxyl substitution and the shorter aliphatic chain connecting its carboxylate group (Fig 1A): the compound simply seems too small to simultaneously establish the network of beneficial bonds observed in the NadR/HPA interactions. We attempted to investigate further the binding stoichiometry using solution-based techniques. However, studies based on tryptophan fluorescence were confounded by the fluorescence of the HPA ligands, and isothermal titration calorimetry (ITC) was unfeasible due to the need for very high concentrations of NadR in the ITC chamber (due to the relatively low affinity), which exceeded the solubility limits of the protein. However, it was possible to calculate the binding stoichiometry of the NadR-HPA interactions using an SPR-based approach. In SPR, the signal measured is proportional to the total molecular mass proximal to the sensor surface; consequently, if the molecular weights of the interactors are known, then the stoichiometry of the resulting complex can be determined . This approach relies on the assumption that the captured protein (‘the ligand’, according to SPR conventions) is 100% active and freely-accessible to potential interactors (‘the analytes’). This assumption is likely valid for this pair of interactors, for two main reasons. Firstly, NadR is expected to be covalently immobilized on the sensor chip as a dimer in random orientations, since it is a stable dimer in solution and has sixteen lysines well-distributed around its surface, all able to act as potential sites for amine coupling to the chip, and none of which are close to the ligand-binding pocket. Secondly, the HPA analytes are all very small (MW 150–170, Fig 1A) and therefore are expected to be able to diffuse readily into all potential binding sites, irrespective of the random orientations of the immobilized NadR dimers on the chip.The stoichiometry of the NadR-HPA interactions was determined using Eq 1 (see Materials and Methods), and revealed stoichiometries of 1.13 for 4-HPA, 1.02 for 3-HPA, and 1.21 for 3Cl,4-HPA, strongly suggesting that one NadR dimer bound to 1 HPA analyte molecule. The crystallographic data, supported by the SPR studies of binding stoichiometry, revealed the lack of a second 4-HPA molecule in the homodimer, suggesting negative co-operativity, a phenomenon previously described for the MTH313/salicylate interaction and for other MarR family proteins . To explore the molecular basis of asymmetry in holo-NadR, we superposed its ligand-free monomer (chain A) onto the ligand-occupied monomer (chain B). Overall, the superposition revealed a high degree of structural similarity (Cα root mean square deviation (rmsd) of 1.5Å), though on closer inspection a rotational difference of ~9 degrees along the long axis of helix α6 was observed, suggesting that 4-HPA induced a slight conformational change (Fig 5A). However, since residues of helix α6 were not directly involved in ligand binding, an explanation for the lack of 4-HPA in monomer A did not emerge by analyzing only these backbone atom positions, suggesting that a more complex series of allosteric events may occur. Indeed, we noted interesting differences in the side chains of Met22, Phe25 and Arg43, which in monomer B are used to contact the ligand while in monomer A they partially occupied the pocket and collectively reduced its volume significantly. Specifically, upon analysis with the CASTp software , the pocket in chain B containing the 4-HPA exhibited a total volume of approximately 370 Å, while the pocket in chain A was occupied by these three side chains that adopted ‘inward’ positions and thereby divided the space into a few much smaller pockets, each with volume < 50 Å, evidently rendering chain A unfavorable for ligand binding. Most notably, atomic clashes between the ligand and the side chains of MetA22, PheA25 and ArgA43 would occur if 4-HPA were present in the monomer A pocket (Fig 5B). Subsequently, analyses of the pockets in apo-NadR revealed that in the absence of ligand the long Arg43 side chain was always in the open ‘outward’ position compatible with binding to the 4-HPA carboxylate group. In contrast, the apo-form Met22 and Phe25 residues were still encroaching the spaces of the 4-hydroxyl group and the phenyl ring of the ligand, respectively (Fig 5C). The ‘outward’ position of Arg43 generated an open apo-form pocket with volume approximately 380Å. Taken together, these observations suggest that Arg43 is a major determinant of ligand binding, and that its ‘inward’ position inhibits the binding of 4-HPA to the empty pocket of holo-NadR. (A) Aligned monomers of holo-NadR (chain A: green; chain B: blue), reveal major overall differences by the shift of helix α6. (B) Comparison of the two binding pockets in holo-NadR shows that in the ligand-free monomer A (green) residues Met22, Phe25 and Arg43 adopt ‘inward’ positions (highlighted by arrows) compared to the ligand-occupied pocket (blue residues); these ‘inward’ conformations appear unfavorable for binding of 4-HPA due to clashes with the 4-hydroxyl group, the phenyl ring and the carboxylate group, respectively. In these crystals, the ArgA43 side chain showed two alternate conformations, modelled with 50% occupancy in each state, as indicated by the two ‘mirrored’ arrows. The inner conformer is the one that would display major clashes if 4-HPA were present. (C) Comparison of the empty pocket from holo-NadR (green residues) with the four empty pockets of apo-NadR (grey residues), shows that in the absence of 4-HPA the Arg43 side chain is always observed in the ‘outward’ conformation. Finally, we applied N heteronuclear solution NMR spectroscopy to examine the interaction of 4-HPA with apo NadR. We collected NMR spectra on NadR in the presence and absence of 4-HPA (see Materials and Methods). The H-N TROSY-HSQC spectrum of apo-NadR, acquired at 25°C, displayed approximately 140 distinct peaks (Fig 6A), most of which correspond to backbone amide N-H groups. The broad spectral dispersion and the number of peaks observed, which is close to the number of expected backbone amide N-H groups for this polypeptide, confirmed that apo-NadR is well-folded under these conditions and exhibits one conformation appreciable on the NMR timescale, i.e. in the NMR experiments at 25°C, two or more distinct conformations of apo-NadR monomers were not readily apparent. Upon the addition of 4-HPA, over 45 peaks showed chemical shift perturbations, i.e. changed position in the spectrum or disappeared, while the remaining peaks remained unchanged. This observation showed that 4-HPA was able to bind NadR and induce notable changes in specific regions of the protein. (A) Superposition of two H-N TROSY-HSQC spectra recorded at 25°C on apo-NadR (cyan) and on NadR in the presence of 4-HPA (red). (B,C) Overlay of selected regions of the H-N TROSY-HSQC spectra acquired at 25°C of apo-NadR (cyan) and NadR/4-HPA (red) superimposed with the spectra acquired at 10°C of apo-NadR (blue) and NadR/4-HPA (green). The spectra acquired at 10°C are excluded from panel A for simplicity. However, in the presence of 4-HPA, the H-N TROSY-HSQC spectrum of NadR displayed approximately 140 peaks, as for apo-NadR, i.e. two distinct stable conformations (that might have potentially revealed the molecular asymmetry observed crystallographically) were not notable. Considering the small size, fast diffusion and relatively low binding affinity of 4-HPA, it would not be surprising if the ligand associates and dissociates rapidly on the NMR time scale, resulting in only one set of peaks whose chemical shifts represent the average environment of the bound and unbound states. Interestingly, by cooling the samples to 10°C, we observed that a number of those peaks strongly affected by 4-HPA (and therefore likely to be in the ligand-binding site) demonstrated evidence of peak splitting, i.e. a tendency to become two distinct peaks rather than one single peak (Fig 6B and 6C). These doubled peaks may therefore reveal that the cooler temperature partially trapped the existence in solution of two distinct states, in presence or absence of 4-HPA, with minor conformational differences occurring at least in proximity to the binding pocket. Although more comprehensive NMR experiments and full chemical shift assignment of the spectra would be required to precisely define this multi-state behavior, the NMR data clearly demonstrate that NadR exhibits conformational flexibility which is modulated by 4-HPA in solution. The apo-NadR crystal structure contained two homodimers in the asymmetric unit (chains A+B and chains C+D). Upon overall structural superposition, these dimers revealed a few minor differences in the α6 helix (a major component of the dimer interface) and the helices α4-α5 (the DNA binding region), and an rmsd of 1.55Å (Fig 7A). Similarly, the entire holo-homodimer could be closely superposed onto each of the apo-homodimers, showing rmsd values of 1.29Å and 1.31Å, and with more notable differences in the α6 helix positions (Fig 7B). The slightly larger rmsd between the two apo-homodimers, rather than between apo- and holo-homodimers, further indicate that apo-NadR possesses a notable degree of intrinsic conformational flexibility. (A) Pairwise alignment of the two distinct apo-NadR homodimers (AB and CD) present in the apo-NadR crystals. (B) Alignment of the holo-NadR homodimer (green and blue chains) onto the apo-NadR homodimers. Here, larger differences are observed in the α6 helices (top). To further investigate the conformational rearrangements of NadR, we performed local structural alignments using only a subset of residues in the DNA-binding helix (α4). By selecting and aligning residues Arg64-Ala77 of one α4 helix per dimer, superposition of the holo-homodimer onto the two apo-homodimers revealed differences in the monomer conformations of each structure. While one monomer from each structure was closely superimposable (Fig 8A, left side), the second monomer displayed quite large differences (Fig 8A, right side). Most notably, the position of the DNA-binding helix α4 shifted by as much as 6 Å (Fig 8B). Accordingly, helix α4 was also found to be one of the most dynamic regions in previous HDX-MS analyses of apo-NadR in solution . (A) The holo-homodimer structure is shown as green and blue cartoons, for chain A and B, respectively, while the two homodimers of apo-NadR are both cyan and pale blue for chains A/C and B/D, respectively. The three homodimers (chains AB holo, AB apo, and CD apo) were overlaid by structural alignment exclusively of all heavy atoms in residues R64-A77 (shown in red, with side chain sticks) of chains A holo, A apo, and C apo, belonging to helix α4 (left). The α4 helices aligned closely, Cα rmsd 0.2Å for 14 residues. (B) The relative positions of the α4 helices of the 4-HPA-bound holo homodimer chain B (blue), and of apo homodimers AB and CD (showing chains B and D) in pale blue. Dashes indicate the Ala77 Cα atoms, in the most highly shifted region of the ‘non-fixed’ α4 helix. (C) The double-stranded DNA molecule (grey cartoon) from the OhrR-ohrA complex is shown after superposition with NadR, to highlight the expected positions of the NadR α4 helices in the DNA major grooves. The proteins share ~30% amino acid sequence identity. For clarity, only the α4 helices are shown in panels (B) and (C). (D) Upon comparison with the experimentally-determined OhrR:ohrA structure (grey), the α4 helix of holo-NadR (blue) is shifted ~8Å out of the major groove. However, structural comparisons revealed that the shift of holo-NadR helix α4 induced by the presence of 4-HPA was also accompanied by several changes at the holo dimer interface, while such extensive structural differences were not observed in the apo dimer interfaces, particularly notable when comparing the α6 helices (S3 Fig). In summary, compared to ligand-stabilized holo-NadR, apo-NadR displayed an intrinsic flexibility focused in the DNA-binding region. This was also evident in the greater disorder (i.e. less well-defined electron density) in the β1-β2 loops of the apo dimers (density for 16 residues per dimer was missing) compared to the holo dimer (density for only 3 residues was missing). In holo-NadR, the distance separating the two DNA-binding α4 helices was 32 Å, while in apo-NadR it was 29 Å for homodimer AB, and 34 Å for homodimer CD (Fig 8C). Thus, the apo-homodimer AB presented the DNA-binding helices in a conformation similar to that observed in the protein:DNA complex of OhrR:ohrA from Bacillus subtilis (Fig 8C). Interestingly, OhrR contacts ohrA across 22 base pairs (bp), and similarly the main NadR target sites identified in the nadA promoter (the operators Op I and Op II) both span 22 bp [9, 10]. Pairwise superpositions showed that the NadR apo-homodimer AB was the most similar to OhrR (rmsd 2.6 Å), while the holo-homodimer was the most divergent (rmsd 3.3 Å) (Fig 8C). Assuming the same DNA-binding mechanism is used by OhrR and NadR, the apo-homodimer AB seems ideally pre-configured for DNA binding, while 4-HPA appeared to stabilize holo-NadR in a conformation poorly suited for DNA binding. Specifically, in addition to the different inter-helical translational distances, the α4 helices in the holo-NadR homodimer were also reoriented, resulting in movement of α4 out of the major groove, by up to 8Å, and presumably preventing efficient DNA binding in the presence of 4-HPA (Fig 8D). When aligned with OhrR, the apo-homodimer CD presented yet another different intermediate conformation (rmsd 2.9Å), apparently not ideally pre-configured for DNA binding, but which in solution can presumably readily adopt the AB conformation due to the intrinsic flexibility described above. While previous studies had correctly suggested the involvement of several NadR residues in ligand binding , the crystal structures presented here revealed additional residues with previously unknown roles in dimerization and/or binding to 4-HPA. To explore the functional involvement of these residues, we characterized the behavior of four new NadR mutants (H7A, S9A, N11A and F25A) in an in vivo assay using the previously described MC58-Δ1843 nadR-null mutant strain , which was complemented either by wild-type nadR or by the nadR mutants. NadA protein abundance levels were assessed by Western blotting to evaluate the ability of the NadR mutants to repress the nadA promoter, in the presence or absence of 4-HPA. The nadR H7A, S9A and F25A complemented strains showed hyper-repression of nadA expression in vivo, i.e. these mutants repressed nadA more efficiently than the NadR WT protein, either in the presence or absence of 4-HPA, while complementation with wild-type nadR resulted in high production of NadA only in the presence of 4-HPA (Fig 9). Interestingly, and on the contrary, the nadR N11A complemented strain showed hypo-repression (i.e. exhibited high expression of nadA both in absence and presence of 4-HPA). This mutagenesis data revealed that NadR residues His7, Ser9, Asn11 and Phe25 play key roles in the ligand-mediated regulation of NadR; they are each involved in the controlled de-repression of the nadA promoter and synthesis of NadA in response to 4-HPA in vivo. Western blot analyses of wild-type (WT) strain (lanes 1–2) or isogenic nadR knockout strains (ΔNadR) complemented to express the indicated NadR WT or mutant proteins (lanes 3–12) or not complemented (lanes 13–14), grown in the presence (even lanes) or absence (odd lanes) of 5mM 4-HPA, showing NadA and NadR expression. Complementation of ΔNadR with WT NadR enables induction of nadA expression by 4-HPA. The H7A, S9A and F25A mutants efficiently repress nadA expression but are less ligand-responsive than WT NadR. The N11A mutant does not efficiently repress nadA expression either in presence or absence of 4-HPA. (The protein abundance levels of the meningococcal factor H binding protein (fHbp) were used as a gel loading control). NadA is a surface-exposed meningococcal protein contributing to pathogenesis, and is one of three main antigens present in the vaccine Bexsero . A detailed understanding of the in vitro repression of nadA expression by the transcriptional regulator NadR is important, both because it is a relevant disease-related model of how small-molecule ligands can regulate MarR family proteins and thereby impact bacterial virulence, and because nadA expression levels are linked to the prediction of vaccine coverage . The repressive activity of NadR can be relieved by hydroxyphenylacetate (HPA) ligands , and HDX-MS studies previously indicated that 4-HPA stabilizes dimeric NadR in a configuration incompatible with DNA binding . Despite these and other studies , the molecular mechanisms by which ligands regulate MarR family proteins are relatively poorly understood and likely differ depending on the specific ligand. Given the importance of NadR-mediated regulation of NadA levels in the contexts of meningococcal pathogenesis, we sought to characterize NadR, and its interaction with ligands, at atomic resolution. Firstly, we confirmed that NadR is dimeric in solution and demonstrated that it retains its dimeric state in the presence of 4-HPA, indicating that induction of a monomeric status is not the manner by which 4-HPA regulates NadR. These observations were in agreement with (i) a previous study of NadR performed using SEC and mass spectrometry , and (ii) crystallographic studies showing that several MarR homologues are dimeric . We also used structure-guided site-directed mutagenesis to identify an important conserved residue, Leu130, which stabilizes the NadR dimer interface, knowledge of which may also inform future studies to explore the regulatory mechanisms of other MarR family proteins. Secondly, we assessed the thermal stability and unfolding of NadR in the presence or absence of ligands. All DSC profiles showed a single peak, suggesting that a single unfolding event simultaneously disrupted the dimer and the monomer. HPA ligands specifically increased the stability of NadR. The largest effects were induced by the naturally-occurring compounds 4-HPA and 3Cl,4-HPA, which, in SPR assays, were found to bind NadR with KD values of 1.5 mM and 1.1 mM, respectively. Although these NadR/HPA interactions appeared rather weak, their distinct affinities and specificities matched their in vitro effects [9, 20] and their biological relevance appears similar to previous proposals that certain small molecules, including some antibiotics, in the millimolar concentration range may be broad inhibitors of MarR family proteins [13, 17]. Indeed, 4-HPA is found in human saliva and 3Cl,4-HPA is produced during inflammatory processes , suggesting that these natural ligands are encountered by N. meningitidis in the mucosa of the oropharynx during infections. It is also possible that NadR responds to currently unidentified HPA analogues. Indeed, in the NadR/4-HPA complex there was a water molecule close to the carboxylate group and also a small unfilled tunnel ~5Å long, both factors suggesting that alternative larger ligands could occupy the pocket. It is conceivable that such putative ligands may establish different bonding networks, potentially binding in a 2:2 ratio, rather than the 1:2 ratio observed herein. The ability to respond to various ligands might enable NadR in vivo to orchestrate multiple response mechanisms and modulate expression of genes other than nadA. Ultimately, confirmation of the relevance of each ligand will require a deeper understanding of the available concentration in vivo in the host niche during bacterial colonization and inflammation. Here, we determined the first crystal structures of apo-NadR and holo-NadR. These experimentally-determined structures enabled a new detailed characterization of the ligand-binding pocket. In holo-NadR, 4-HPA interacted directly with at least 11 polar and hydrophobic residues. Several, but not all, of these interactions were predicted previously by homology modelling combined with ligand docking in silico . Subsequently, we established the functional importance of His7, Ser9, Asn11 and Phe25 in the in vitro response of meningococcus to 4-HPA, via site-directed mutagenesis. More unexpectedly, the crystal structure revealed that only one molecule of 4-HPA was bound per NadR dimer. We confirmed this stoichiometry in solution using SPR methods. We also used heteronuclear NMR spectroscopy to detect substantial conformational changes of NadR occurring in solution upon addition of 4-HPA. Moreover, NMR spectra at 10°C suggested the existence of two distinct conformations of NadR in the vicinity of the ligand-binding pocket. More powerfully, our unique crystallographic observation of this ‘occupied vs unoccupied site’ asymmetry in the NadR/4-HPA interaction is, to our knowledge, the first example reported for a MarR family protein. Structural analyses suggested that ‘inward’ side chain positions of Met22, Phe25 and especially Arg43 precluded binding of a second ligand molecule. Such a mechanism indicates negative cooperativity, which may enhance the ligand-responsiveness of NadR. Comparisons of the NadR/4-HPA complex with available MarR family/salicylate complexes revealed that 4-HPA has a previously unobserved binding mode. Briefly, in the M. thermoautotrophicum MTH313 dimer, one molecule of salicylate binds in the pocket of each monomer, though with two rather different positions and orientations, only one of which (site-1) is thought to be biologically relevant (Fig 10A). In the S. tokodaii protein ST1710, salicylate binds to the same position in each monomer of the dimer, in a site equivalent to the putative biologically relevant site of MTH313 (Fig 10B). Unlike other MarR family proteins which revealed multiple ligand binding interactions, we observed only 1 molecule of 4-HPA bound to NadR, suggesting a more specific and less promiscuous interaction. In NadR, the single molecule of 4-HPA binds in a position distinctly different from the salicylate binding site: translated by > 10 Å and with a 180° inverted orientation (Fig 10C). (A) A structural alignment of MTH313 chains A and B shows that salicylate is bound in distinct locations in each monomer; site-1 (thought to be the biologically relevant site) and site-2 differ by ~7Å (indicated by black dotted line) and also by ligand orientation. (B) A structural alignment of MTH313 chain A and ST1710 (pink) (Cα rmsd 2.3Å), shows that they bind salicylate in equivalent sites (differing by only ~3Å) and with the same orientation. (C) Addition of holo-NadR (chain B, blue) to the alignment reveals that bound 4-HPA differs in position by > 10 Å compared to salicylate, and adopts a novel orientation. Interestingly, a crystal structure was previously reported for a functionally-uncharacterized meningococcal homologue of NadR, termed NMB1585, which shares 16% sequence identity with NadR . The two structures can be closely aligned (rmsd 2.3 Å), but NMB1585 appears unsuited for binding HPAs, since its corresponding ‘pocket’ region is occupied by several bulky hydrophobic side chains. It can be speculated that MarR family members have evolved separately to engage distinct signaling molecules, thus enabling bacteria to use the overall conserved MarR scaffold to adapt and respond to diverse changing environmental stimuli experienced in their natural niches. Alternatively, it is possible that other MarR homologues (e.g. NMB1585) may have no extant functional binding pocket and thus may have lost the ability to respond to a ligand, acting instead as constitutive DNA-binding regulatory proteins. The apo-NadR crystal structures revealed two dimers with slightly different conformations, most divergent in the DNA-binding domain. It is not unusual for a crystal structure to reveal multiple copies of the same protein in very slightly different conformations, which are likely representative of the lowest-energy conformations sampled by the dynamic ensemble of molecular states occurring in solution, and which likely have only small energetic differences, as described previously for MexR (a MarR protein) or more recently for the solute-binding protein FhuD2 [31, 32]. Further, the holo-NadR structure was overall more different from the two apo-NadR structures (rmsd values ~1.3Å), suggesting that the ligand selected and stabilized yet another conformation of NadR. These observations suggest that 4-HPA, and potentially other similar ligands, can shift the molecular equilibrium, changing the energy barriers that separate active and inactive states, and stabilizing the specific conformation of NadR poorly suited to bind DNA. Comparisons of the apo- and holo-NadR structures revealed that the largest differences occurred in the DNA-binding helix α4. The shift of helix α4 in holo-NadR was also accompanied by rearrangements at the dimer interface, involving helices α1, α5, and α6, and this holo-form appeared poorly suited for DNA-binding when compared with the known OhrR:ohrA complex . While some flexibility of helix α4 was also observed in the two apo-structures, concomitant changes in the dimer interfaces were not observed, possibly due to the absence of ligand. One of the two conformations of apo-NadR appeared ideally suited for DNA-binding. Overall, these analyses suggest that the apo-NadR dimer has a pre-existing equilibrium that samples a variety of conformations, some of which are compatible with DNA binding. This intrinsically dynamic nature underlies the possibility for different conformations to inter-convert or to be preferentially selected by a regulatory ligand, as generally described in the ‘conformational selection’ model for protein-ligand interactions (the Monod-Wyman-Changeux model), rather than an ‘induced fit’ model (Koshland-Nemethy-Filmer) . The noted flexibility may also explain how NadR can adapt to bind various DNA target sequences with slightly different structural features. Subsequently, upon ligand binding, holo-NadR adopts a structure less suited for DNA-binding and this conformation is selected and stabilized by a network of protein-ligand interactions and concomitant rearrangements at the NadR holo dimer interface. In an alternative and less extensive manner, the binding of two salicylate molecules to the M. thermoautotrophicum protein MTH313 appeared to induce large changes in the wHTH domain, which was associated with reduced DNA-binding activity . Here we have presented two new crystal structures of the transcription factor, NadR, which regulates expression of the meningococcal surface protein, virulence factor and vaccine antigen NadA. Detailed structural analyses provided a molecular explanation for the ligand-responsive regulation by NadR on the majority of the promoters of meningococcal genes regulated by NadR, including nadA . Intriguingly, NadR exhibits a reversed regulatory mechanism on a second class of promoters, including mafA of the multiple adhesin family–i.e. NadR represses these genes in the presence but not absence of 4-HPA. The latter may influence the surface abundance or secretion of maf proteins, an emerging class of highly conserved meningococcal putative adhesins and toxins with many important roles [11, 12]. Further work is required to investigate how the two different promoter types influence the ligand-responsiveness of NadR during bacterial infection and may provide insights into the regulatory mechanisms occurring during these host-pathogen interactions. Ultimately, knowledge of the ligand-dependent activity of NadR will continue to deepen our understanding of nadA expression levels, which influence meningococcal pathogenesis. In this study we used N. meningitidis MC58 wild type strain and related mutant derivatives. The MC58 isolate was kindly provided to us by Professor E. Richard Moxon, University of Oxford, UK, and was previously submitted to the Meningococcal Reference Laboratory, Manchester, UK . Strains were routinely cultured, stocked, and transformed as described previously . To generate N. meningitidis MC58 mutant strains expressing only the amino acid substituted forms of NadR, plasmids containing the sequence of nadR mutated to insert alanine codons to replace His7, Ser9, Asn11 or Phe25 were constructed using the QuikChange II XL Site-Directed Mutagenesis Kit (Stratagene). The nadR gene (also termed NMB1843) was mutated in the pComEry-1843 plasmid using couples of mutagenic primers (forward and reverse). The resulting plasmids were named pComEry-1843H7A, -1843S9A, -1843N11A or -1843F25A, and contained a site-directed mutant allele of the nadR gene, in which the selected codons were respectively substituted by a GCG alanine codon, and were used for transformation of the MC-Δ1843 strain. Total lysates from single colonies of all transformants were used as a template for PCR analysis to confirm the correct insertion by double homologous recombinant event. When indicated, bacterial strains were grown in presence of 5 mM 4-HPA (MW 152, Sigma-Aldrich). The preparation of the expression construct enabling production of soluble NadR with an N-terminal His-tag followed by a thrombin cleavage site (MGSSHHHHHHSSGLVPR↓GSH-) (where the arrow indicates the cleavage site) and then NadR residues M1-S146 (Uniprot code Q7DD70), and methods to generate site-directed mutants, were described previously . Briefly, site-directed mutagenesis was performed using two overlapping primers containing the desired mutation to amplify pET15b containing several NadR variants. (Full oligonucleotide sequences of primers are available upon request). 1–10 ng of plasmid DNA template were amplified using Kapa HiFi DNA polymerase (Kapa Biosystems) and with the following cycling conditions: 98°C for 5 min, 15 cycles of (98°C for 30 s, 60°C for 30 s, 72°C for 6 min) followed by a final extension of 10 min at 72°C. Residual template DNA was digested by 30 min incubation with FastDigest DpnI (Thermo Scientific) at 37°C and 1 μl of this reaction was used to transform E. coli DH5α competent cells. The full recombinant tagged NadR protein generated contained 166 residues, with a theoretical MW of 18746, while after thrombin-cleavage the untagged protein contained 149 residues, with a theoretical MW of 16864. The NadR expression constructs (wild-type or mutant clones) were transformed into E. coli BL21 (DE3) cells and were grown at 37°C in Luria-Bertani (LB) medium supplemented with 100 μg/mL ampicillin, until an OD600 of 0.5 was reached. Target protein production was induced by the addition of 1 mM IPTG followed by incubation with shaking overnight at 21°C. For production of the selenomethionine (SeMet) derivative form of NadR for crystallization studies, essentially the same procedure was followed, but using the E. coli B834 strain grown in a modified M9 minimal medium supplemented with 40 mg/L L-SeMet. For production of N-labeled NadR for NMR analyses, the EnPresso B Defined Nitrogen-free medium (Sigma-Aldrich) was used; in brief, BL21 (DE3) cells were grown in BioSilta medium at 30°C for 30 h, and production of the N-labeled NadR was enabled by the addition of 2.5 g/L NH4Cl and further incubation for 2 days. Cells were harvested by centrifugation (6400 g, 30 min, 4°C), resuspended in 20 mM HEPES pH 8.0, 300 mM NaCl, 20 mM imidazole, and were lysed by sonication (Qsonica Q700). Cell lysates were clarified by centrifugation at 2800 g for 30 min, and the supernatant was filtered using a 0.22 μm membrane (Corning filter system) prior to protein purification. NadR was purified by affinity chromatography using an AKTA purifier (GE Healthcare). All steps were performed at room temperature (18–26°C), unless stated otherwise. The filtered supernatant was loaded onto an Ni-NTA resin (5 mL column, GE Healthcare), and NadR was eluted using 4 steps of imidazole at 20, 30, 50 and 250 mM concentration, at a flow rate of 5 mL/min. Eluted fractions were examined by reducing and denaturing SDS-PAGE analysis. Fractions containing NadR were identified by a band migrating at ~17 kDa, and were pooled. The N-terminal 6-His tag was removed enzymatically using the Thrombin CleanCleave Kit (Sigma-Aldrich). Subsequently, the sample was reloaded on the Ni-NTA resin to capture the free His tag (or unprocessed tagged protein), thus allowing elution in the column flow-through of tagless NadR protein, which was used in all subsequent studies. The NadR sample was concentrated and loaded onto a HiLoad Superdex 75 (16/60) preparative size-exclusion chromatography (SEC) column equilibrated in buffer containing 20 mM HEPES pH 8.0, 150 mM NaCl, at a flow-rate of 1 mL/min. NadR protein was collected and the final yield of purified protein obtained from 0.5 L LB growth medium was approximately 8 mg (~2 mg protein per g wet biomass). Samples were used immediately for crystallization or analytical experiments, or were frozen for storage at -20°C. MALLS analyses were performed online with SE-HPLC using a Dawn TREOS MALLS detector (Wyatt Corp., Santa Barbara, CA, USA) and an incident laser wavelength of 658 nm. The intensity of the scattered light was measured at 3 angles simultaneously. Data analysis was performed using the Astra V software (Wyatt) to determine the weighted-average absolute molecular mass (MW), the polydispersity index (MW/Mn) and homogeneity (Mz/Mn) for each oligomer present in solution. Normalization of the MALLS detectors was performed in each analytical session by use of bovine serum albumin. The thermal stability of NadR proteins was assessed by differential scanning calorimetry (DSC) using a MicroCal VP-Capillary DSC instrument (GE Healthcare). NadR samples were prepared at a protein concentration of 0.5 mg/mL (~30 μM) in buffer containing 20 mM HEPES, 300 mM NaCl, pH 7.4, with or without 6 mM HPA or salicylate. The DSC temperature scan ranged from 10°C to 110°C, with a thermal ramping rate of 200°C per hour and a 4 second filter period. Data were analyzed by subtraction of the reference data for a sample containing buffer only, using the Origin 7 software. All experiments were performed in triplicate, and mean values of the melting temperature (Tm) were determined. Surface plasmon resonance binding analyses were performed using a Biacore T200 instrument (GE Healthcare) equilibrated at 25°C. The ligand (NadR) was covalently immobilized by amine-coupling on a CM-5 sensor chip (GE Healthcare), using 20 μg/mL purified protein in 10 mM sodium acetate buffer pH 5, injected at 10 μl/min for 120 s until ~9000 response units (RU) were captured. A high level of ligand immobilization was required due to the small size of the analytes. An unmodified surface was used as the reference channel. Titrations with analytes (HPAs or salicylate) were performed with a flow-rate of 30 μl/min, injecting the compounds in a concentration range of 10 μM to 20 mM, using filtered running buffer containing Phosphate Buffered Saline (PBS) with 0.05% Tween-20, pH 7.4. Following each injection, sensor chip surfaces were regenerated with a 30-second injection of 10 mM Glycine pH 2.5. Each titration series contained 20 analyte injections and was performed in triplicate. Titration experiments with long injection phases (> 15 mins) were used to enable steady-state analyses. Data were analyzed using the BIAcore T200 evaluation software and the steady-state affinity model. A buffer injection was subtracted from each curve, and reference sensorgrams were subtracted from experimental sensorgrams to yield curves representing specific binding. The equilibrium dissociation constant, KD, was determined from the plot of RUeq against analyte concentration (S2 Fig), as described previously . Determination of binding stoichiometry: From each plot of RUeq against analyte concentration, obtained from triplicate experiments, the Rmax value (maximum analyte binding capacity of the surface) was extrapolated from the experimental data (S2 Fig). Stoichiometry was calculated using the molecular weight of dimeric NadR as ligand molecule (MWligand) and the molecular weights of the HPA analyte molecules (MWanalyte), and the following equation: Stoichiometry=Rmax×MWligandMWanalyte×Rligand(1) where Rligand is recorded directly from the sensorgram during ligand immobilization prior to the titration series, as described previously . The stoichiometry derived therefore represented the number of HPA molecules bound to one dimeric NadR protein. Purified NadR was concentrated to 2.7 mg/mL (~160 μM) using a centrifugal concentration device (Amicon Ultra-15 Centrifugal Filter Unit with Ultracel-10 membrane with cut-off size 10 kDa; Millipore) running at 600 g in a bench top centrifuge (Thermo Scientific IEC CL40R) refrigerated at 2–8°C. To prepare holo-NadR samples, HPA ligands were added at a 200-fold molar excess prior to the centrifugal concentration step. The concentrated holo- or apo-NadR was subjected to crystallization trials performed in 96-well low-profile Intelli-Plates (Art Robbins) or 96-well low-profile Greiner crystallization plates, using a nanodroplet sitting-drop vapour-diffusion format and mixing equal volumes (200 nL) of protein samples and crystallization buffers using a Gryphon robot (Art Robbins). Crystallization trays were incubated at 20°C. Crystals of apo-NadR were obtained in 50% PEG 3350 and 0.13 M di-Ammonium hydrogen citrate, whereas crystals of SeMet–NadR in complex with 4-HPA grew in condition H4 of the Morpheus screen (Molecular Dimensions), which contains 37.5% of the pre-mixed precipitant stock MPD_P1K_PEG 3350, buffer system 1 and 0.1 M amino acids, at pH 6.5. All crystals were mounted in cryo-loops using 10% ethylene glycol or 10% glycerol as cryo-protectant before cooling to 100 K for data collection. X-ray diffraction data from crystals of apo-NadR and SeMet–NadR/4-HPA were collected on beamline PXII-X10SA of the Swiss Light Source (SLS) at the Paul Scherrer Institut (PSI), Villigen, Switzerland. All diffraction data were processed with XDS and programs from the CCP4 suite . Crystals of apo-NadR and 4-HPA-bound SeMet-NadR belonged to space group P43 21 2 (see Table 2). Apo-NadR crystals contained four molecules (two dimers) in the asymmetric unit (Matthews coefficient 2.25 Å Da, for a solvent content of 45%), while crystals of SeMet–NadR/4-HPA contained two molecules (one dimer) in the asymmetric unit (Matthews coefficient 1.98 Å Da, for a solvent content of 38%). In solving the holo-NadR structure, an initial and marginal molecular replacement (MR) solution was obtained using as template search model the crystal structure of the transcriptional regulator PA4135 (PBD entry 2FBI), with which NadR shares ~54% sequence identity. This solution was combined with SAD data to aid identification of two selenium sites in NadR, using autosol in phenix and this allowed generation of high-quality electron density maps that were used to build and refine the structure of the complex. Electron densities were clearly observed for almost the entire dimeric holo-NadR protein, except for a new N-terminal residues and residues 88–90 of chain B. The crystal structure of apo-NadR was subsequently solved by MR in Phaser at 2.7 Å, using the final refined model of SeMet-NadR/4-HPA as the search model. For apo-NadR, electron densities were clearly observed for almost the entire protein, although residues 84–91 of chains A, C, and D, and residues 84–90 of chain B lacked densities suggesting local disorder. Both structures were refined and rebuilt using phenix and Coot , and structural validation was performed using Molprobity . Data collection and refinement statistics are reported in Table 2. Atomic coordinates of the two NadR structures have been deposited in the Protein Data Bank, with entry codes 5aip (NadR bound to 4-HPA) and 5aiq (apo-NadR). All crystallographic software was compiled, installed and maintained by SBGrid . For heteronuclear NMR experiments, the NadR protein concentration used was 85 μM (~ 1.4 mg/mL) in a solution containing 100 mM sodium phosphate buffer (90% H2O and 10% D2O) and 200 mM NaCl, prepared in the apo-form or in the presence of a 200-fold molar excess of 4-HPA, at pH 6.5. The stability, integrity and dimeric state of the protein in the NMR buffer was confirmed by analytical SEC (Superdex 75, 10/300 column) prior to NMR studies. H-N transverse relaxation-optimized spectroscopy (TROSY)-heteronuclear single quantum coherence (HSQC) experiments on apo-NadR and NadR in the presence of 4-HPA were acquired using an Avance 950 Bruker spectrometer, operating at a proton frequency of 949.2 MHz and equipped with triple resonance cryogenically-cooled probe at two different temperatures (298 K and 283 K). H-N TROSY-HSQC experiments were recorded for 12 h, with a data size of 1024 x 232 points. Spectra were processed using the Bruker TopSpin 2.1 and 3.1 software packages. Western blot analysis was performed as described previously .
PMC4822561	Structure of a quinolone-stabilized cleavage complex of topoisomerase IV from Klebsiella pneumoniae and comparison with a related Streptococcus pneumoniae complex	Crystal structures of the cleavage complexes of topoisomerase IV from Gram-negative (K. pneumoniae) and Gram-positive (S. pneumoniae) bacterial pathogens stabilized by the clinically important antibacterial drug levofloxacin are presented, analysed and compared. For K. pneumoniae, this is the first high-resolution cleavage complex structure to be reported.Klebsiella is a genus belonging to the Enterobacteriaceae family of Gram-negative bacilli, which is divided into seven species with demonstrated similarities in DNA homology: K. pneumoniae, K. ozaenae, K. rhinoscleromatis, K. oxytoca, K. planticola, K. terrigena and K. ornithinolytica. K. pneumoniae is the most medically important species of the genus owing to its high resistance to antibiotics. Significant morbidity and mortality has been associated with an emerging, highly drug-resistant strain of K. pneumoniae characterized as producing the carbapenemase enzyme (KPC-producing bacteria; Nordmann et al., 2009 ▸). The best therapeutic approach to KPC-producing organisms has yet to be defined. However, common treatments (based on in vitro susceptibility testing) are the polymyxins, tigecycline and, less frequently, aminoglycoside antibiotics (Arnold et al., 2011 ▸). Another effective strategy involves the limited use of certain antimicrobials, specifically fluoroquinolones and cephalosporins (Gasink et al., 2009 ▸). Several new antibiotics are under development for KPC producers. These include combinations of existing β-lactam antibiotics with new β-lactamase inhibitors able to circumvent KPC resistance. Neoglycosides are novel aminoglycosides that have activity against KPC-producing bacteria that are also being developed (Chen et al., 2012 ▸). Type II topoisomerase enzymes play important roles in prokaryotic and eukaryotic DNA replication, recombination and transcription (Drlica et al., 2008 ▸; Laponogov et al., 2013 ▸; Lee et al., 2013 ▸; Nitiss, 2009a ▸,b ▸; Schoeffler & Berger, 2008 ▸; Sissi & Palumbo, 2009 ▸; Vos et al., 2011 ▸; Wendorff et al., 2012 ▸; Wu et al., 2011 ▸, 2013 ▸). In bacteria, topoisomerase IV, a tetramer of two ParC and two ParE subunits, unlinks daughter chromosomes prior to cell division, whereas the related enzyme gyrase, a GyrA2GyrB2 tetramer, supercoils DNA and helps unwind DNA at replication forks. Both enzymes act via a double-strand DNA break involving a cleavage complex and are targets for quinolone antimicrobials that act by trapping these enzymes at the DNA-cleavage stage and preventing strand re-joining (Drlica et al., 2008 ▸). Levofloxacin is a broad-spectrum third-generation fluoroquinolone antibiotic. It is active against Gram-positive and Gram-negative bacteria and functions by inhibiting gyrase and topoisomerase IV (Drlica & Zhao, 1997 ▸; Laponogov et al., 2010 ▸). Acquiring a deep structural and functional understanding of the mode of action of fluoroquinolones (Tomašić & Mašič, 2014 ▸) and the development of new drugs targeted against topoisomerase IV and gyrase from a wide range of Gram-positive and Gram-negative pathogenic bacteria are highly active areas of current research directed at overcoming the vexed problem of drug resistance (Bax et al., 2010 ▸; Chan et al., 2015 ▸; Drlica et al., 2014 ▸; Mutsaev et al., 2014 ▸; Pommier, 2013 ▸; Srikannathasan et al., 2015 ▸). Here, we report the first three-dimensional X-ray structure of a K. pneumoniae topoisomerase IV ParC/ParE cleavage complex with DNA stabilized by levofloxacin. The crystal structure provides structural information on topoisomerase IV from K. pneumoniae, a pathogen for which drug resistance is a serious concern. The structure of the ParC/ParE–DNA–levofloxacin binding site highlights the details of the cleavage-complex assembly that are essential for the rational design of Klebsiella topoisomerase inhibitors. Cloning, expression and purification protocols are described in detail in the Supporting Information. Table 1 ▸ contains the sequence information for all of the components of the complexes. Fig. 1 ▸ provides information about the protein and DNA constructs used in the experimental work. For the K. pneumoniae cleavage complex, two DNA oligomers (5′-CGTATTACGTTGTAT-3′ and 5′-GATCATACAACGTAATACG-3′) were synthesized by solid-phase phosphoramidite chemistry and doubly HPLC purified by Metabion, Munich, Germany. The DNA sequence was designed to make a complementary DNA 34-mer that contained the ‘pre-cut’ binding-site fragment: 5′-CGTATTACGTTGTAT↓GATCATACAACGTAATACG-3′ and 3′-GCATAATGCAACATACTAG↓TATGTTGCATTATGC-5′ (the cuts are shown by arrows; see Fig. 1 ▸ b). For the S. pneumoniae cleavage complex, two DNA oligomers (5′-CATGAATGACTATGCACG-3′ and 5′-CGTGCATAGTCATTCATG-3′) were synthesized by solid-phase phosphoramidite chemistry and doubly HPLC purified by Metabion, Munich, Germany. The DNA sequence corresponds to the E-site 18-mer, which was found to be a better DNA length for crystallization of the S. pneumoniae topoisomerase IV cleavage complexes in order to give stable reproducible crystals (see Fig. 1 ▸ b). DNA stock solutions were made by mixing the required oligomers (at 1 mM in 20 mM Tris pH 7.5, 200 mM NaCl, 1 mM β-mercaptothanol, 0.05% NaN3) in equal volumes. For DNA annealing, the mixtures of complementary oligomers were heated to 98°C and then slowly cooled to 4°C over a 48 h period. Crystallization information for both the S. pneumoniae and the K. pneumoniae topoisomerase IV cleavage complexes is summarized in Table 2 ▸. Data-collection statistics and details are provided in Table 3 ▸. Structure-solution and refinement details are provided in Table 4 ▸. Protein was mixed with DNA in a 1:1:1.2 molar ratio (ParC55:ParE30:18-mer E-site DNA) with an overall concentration of 4 mg ml. Levofloxacin and magnesium chloride were added to final concentrations of 2 and 10 mM, respectively. The mixture was pre-incubated at room temperature overnight. Initial crystallization screening was performed by sitting-drop vapour diffusion in a 96-well MRC crystallization plate (600 nl protein mixture + 400 nl reservoir solution) using a Mosquito robot (TTP Labtech; http://www.ttplabtech.com). The best crystals were obtained using capillary counter-diffusion against 50 mM sodium cacodylate pH 6.5, 2.5% Tacsimate (Hampton Research; McPherson & Cudney, 2006 ▸), 7% 2-propanol, 62.5 mM KCl, 7.5 mM MgCl2 at 304 K. The crystals were flash-cooled at 100 K in cryoprotectant buffer C [50 mM sodium cacodylate pH 6.5, 2.5% Tacsimate, 62.5 mM KCl, 7.5 mM MgCl2, 1 mM β-mercaptoethanol, 30%(v/v) MPD]. The best data set was collected on beamline I03 at Diamond Light Source at a wavelength of 0.9763 Å using an ADSC Quantum 315 detector. The data extended to 2.6 Å resolution anisotropically and were used in refinement with a maximum-likelihood target in the initial refinement cycles; they were deposited in the PDB without introducing a resolution cutoff. However, owing to the high R merge values in the outer shells, the final resolution is given as 2.9 Å and the statistics are reported according to this ‘trimmed’ resolution. The resolution cutoff was based on the rejection criteria R merge < 50% and I/σ(I) > 1.5 in the highest resolution shell. The data were integrated using HKL-2000 (Otwinowski & Minor, 1997 ▸). The space group was determined to be P3121, with unit-cell parameters a = b = 157.83, c = 211.15 Å. The structure was solved by molecular replacement using Phaser (McCoy et al., 2007 ▸) as implemented within the CCP4 suite (Winn et al., 2011 ▸) and our previously published topoisomerase IV–levofloxacin structure (PDB entry 3k9f; Laponogov et al., 2010 ▸). Refinement was performed in PHENIX (Adams et al., 2002 ▸, 2010 ▸) with manual inspection and corrections performed in WinCoot (Emsley & Cowtan, 2004 ▸; Emsley et al., 2010 ▸).The structure was verified using WinCoot and PROCHECK (Laskowski et al., 1993 ▸). ParC55/ParE30 protein stock in incubation buffer (at 4.5 mg ml) was mixed with the ‘pre-cut’ 34-mer DNA stock in a 1:1.2 protein:DNA molar ratio. High-concentration stocks of levofloxacin and MgCl2 were added to give final concentrations of 2 and 10 mM, respectively. The mixture was incubated overnight at room temperature. Initial crystallization screening was performed by sitting-drop vapour diffusion in 96-well MRC crystallization plates (600 nl protein mixture + 300 nl reservoir solution) using a Mosquito robot. When the optimal crystallization conditions had been established, conventional hanging-drop vapour diffusion in 24-well Linbro plates (4 µl protein mixture + 2 µl reservoir solution) was used to increase the crystal size. Crystals formed after ∼7–10 d at room temperature. The crystallization conditions varied slightly from batch to batch in the range 0.1 M Tris pH 7.5–8.0, 0–50 mM NaCl, 4–8% PEG 4000, 12–15% glycerol. It should be mentioned that several other DNA oligomers with the same binding-site sequence were tried for crystallization (i.e. 20-mer, ‘pre-cut’ 20-mer and 34-mer DNA sequences). However, these protein–DNA–drug complexes did not produce good-quality crystals for data collection. Crystals were tested in-house for diffraction quality using an Oxford Xcalibur Nova CCD diffractometer and were then transported for high-resolution data collection at Diamond Light Source (Harwell Science and Innovation Campus, Oxfordshire, England). The data were collected on beamline I03 (wavelength 0.9762 Å) using a Pilatus 6M-F detector (0.2° oscillation per image, 100 K nitrogen stream). The best crystals diffracted to ∼3.2 Å resolution. All data sets were integrated with MOSFLM (Leslie & Powell, 2007 ▸) and merged with SCALA (Evans, 2006 ▸) as implemented in CCP4 (Winn et al., 2011 ▸). The ParC55/ParE30–DNA–levofloxacin crystals belonged to space group P21, with unit-cell parameters a = 102.07, b = 161.53, c = 138.60 Å, α = 90.00, β = 94.22, γ = 90.00°. They contained two ParC/ParE–DNA heterodimers in the asymmetric unit. Several data sets were collected, some of which contained visible diffraction to 3.2 Å resolution, but owing to potential internal twinning and space-group ambiguity (most data sets could be integrated in space groups P21 and P212121) and the fact that the structure solution could be obtained in both space groups, careful selection of the integration ranges as well as appropriate data truncation were necessary. The best region of data was integrated to 3.35 Å (see Table 3 ▸ for statistics). The resolution cutoff was based on the rejection criteria R merge < 50% and I/σ(I) > 1.5 in the highest resolution shell. The structure was solved by the molecular-replacement method in Phaser (McCoy et al., 2007 ▸) using the levofloxacin–DNA cleavage complex of topoisomerase IV from S. pneumoniae as a search model (PDB entry 3rae; ∼41.8% sequence identity). Refinement was performed in PHENIX (Adams et al., 2002 ▸, 2010 ▸) using secondary-structure restraints derived by superposition of the K. pneumoniae ParC/ParE model with the previously solved complex of S. pneumoniae ParC/ParE. Rigid-body, positional and TLS refinements were performed. Levofloxacin molecules and magnesium ions were placed during the final stages of refinement based on missing electron density in the σA-weighted 2F obs − F calc and F obs − F calc maps. WinCoot (Emsley & Cowtan, 2004 ▸) was used for interactive model fitting. The structure was verified using WinCoot and PROCHECK (Laskowski et al., 1993 ▸). The resulting model had good geometry, with 87.8, 9.9 and 1.3% of residues in the favoured, allowed and generously allowed regions of the Ramachandran plot, respectively, and no more than 1% of residues in disallowed regions. The data-collection and final refinement statistics are given in Tables 3 ▸ and 4 ▸. Sequence alignment was performed in ClustalW (Larkin et al., 2007 ▸, McWilliam et al., 2013 ▸). Figures were prepared using PyMOL (DeLano, 2008 ▸), CHEMDRAW (Evans, 2014 ▸) and CorelDRAW (http://www.coreldraw.com). We have co-crystallized the K. pneumoniae topoisomerase IV ParC/ParE breakage-reunion domain (ParC55; residues 1–490) and ParE TOPRIM domain (ParE30; residues 390–631) with a precut 34 bp DNA duplex (the E-site), stabilized by levofloxacin. The X-ray crystal structure of the complex was determined to 3.35 Å resolution, revealing a closed ParC55 dimer flanked by two ParE30 monomers (Figs. 1 ▸, 2 ▸ and 3 ▸). The overall architecture of this complex is similar to that found for S. pneumoniae topoisomerase–DNA–drug complexes (Laponogov et al., 2009 ▸, 2010 ▸). Residues 6–30 of the N-terminal α-helix α1 of the ParC subunit again embrace the ParE subunit, ‘hugging’ the ParE subunits close to either side of the ParC dimer (Laponogov et al., 2010 ▸). Deletion of this ‘arm’ α1 results in loss of DNA-cleavage activity (Laponogov et al., 2007 ▸) and is clearly very important in complex stability (Fig. 3 ▸). This structural feature was absent in our original ParC55 structure (Laponogov et al., 2007 ▸; Sohi et al., 2008 ▸). The upper region of the topoisomerase complex consists of the E-subunit TOPRIM metal-binding domain formed of four parallel β-sheets and the surrounding α-helices. The C-subunit provides the WHD (winged-helix domain; a CAP-like structure; McKay & Steitz, 1981 ▸) and the ‘tower’ which form the U groove-shaped protein region into which the G-gate DNA binds with an induced U-shaped bend. The lower C-gate region (Fig. 3 ▸) consists of the same disposition of pairs of two long α-helices terminated by a spanning short α-helix forming a 30 Å wide DNA-accommodating cavity through which the T-gate DNA passes as found in the S. pneumoniae complex. Owing to the structural similarity, it appears that the topoisomerases IV from K. pneumoniae and S. pneumoniae are likely to follow a similar overall topoisomerase catalytic cycle as shown in Fig. 4 ▸; we have confirmation of one intermediate from our recent structure of the full complex (the holoenzyme less the CTD β-pinwheel domain) with the ATPase domain in the open conformation (Laponogov et al., 2013 ▸). The G-gate DNA for the S. pneumoniae complex consists of an 18-base-pair E-site sequence (our designation for a DNA site which we first found from DNA-mapping studies; Leo et al., 2005 ▸; Arnoldi et al., 2013 ▸; Fig. 1 ▸). The crystallized complex was formed by turning over the topoisomerase tetramer in the presence of DNA and levofloxacin and crystallizing the product. In contrast, the K. pneumoniae complex was formed by co-crystallizing the topoisomerase tetramer complex in the presence of a 34-base-pair pre-cleaved DNA in the presence of levofloxacin. In both cases the DNA is bent into a U-form and bound snugly against the protein of the G-gate. We have been able to unambiguously read off the DNA sequences in the electron-density maps. There is 41.6% sequence identity and 54.4% sequence homology between the ParE subunit of K. pneumoniae and that of S. pneumoniae. For the ParC subunits, the figures are 40.8 identity and 55.6% homology between the two organisms. The sequence alignment is given in Supplementary Fig. S1, with the key metal-binding residues and those which give rise to quinolone resistance highlighted. The binding of levofloxacin in the K. pneumoniae complex is shown in Figs. 2 ▸, 3 ▸ and 5 ▸ and is hemi-intercalated into the DNA and stacked against the DNA bases at the cleavage site (positions −1 and +1 of the four-base-pair staggered cut in the 34-mer DNA) which is similar to that found for the S. pneumoniae complex. Fig. 5 ▸ presents side-by-side views of the K. pneumoniae and S. pneumoniae active sites which shows that levofloxacin binds in a very similar manner in these two complexes with extensive π–π stacking interaction between the bases and the drug. The methylpiperazine at C7 (using the conventional quinolone numbering; C9 in the IUPAC numbering) on the drug extends towards residues Glu474 and Glu475 for S. pneumoniae and towards Gln460 and Glu461 for K. pneumoniae, where the glutamate at 474 is substituted by a glutamine at 460 in the Klebsiella strain. Interestingly, for S. pneumoniae we observe only one possible orientation of the C7 groups in both subunits, while for K. pneumoniae we can see two: one with the same orientation as in S. pneumoniae and other rotated 180° away. They both exist within the same crystal in the two different dimers in the asymmetric unit. The side chains surrounding them in ParE are quite disordered and are more defined in K. pneumoniae (even though this complex is at lower resolution) than in S. pneumoniae. There are no direct hydrogen bonds from the drug to these residues (although it is possible that some are formed through water, which cannot be observed at this resolution). Obviously, the drug–ParE interaction in this region is less strong compared with PD 0305970 binding to the S. pneumoniae DNA complex, where PD 0305970 forms a hydrogen bond to ParE residue Asp475 and can form one to Asp474 if the bond rotates (Laponogov et al., 2010 ▸). This may explain why drug-resistance mutations for levofloxacin are more likely to form in the ParC subunits rather than in the ParE subunits. For both complexes there is a Mg ion bound to levofloxacin between the carbonyl group at position 4 of the quinolone and the carboxyl at position 6 (Figs. 2 ▸ and 5 ▸ and Supplementary Fig. 2 ▸). For S. pneumoniae topoisomerase IV, one of the O atoms of the carboxyl of Asp83 points towards the Mg ion and is within hydrogen-bonding distance (5.04 Å) through an Mg-coordinated water. For K. pneumoniae both of the carboxyl O atoms are pointing towards the Mg ion at distances of 4.86 and 4.23 Å. These residues are ordered in only one of the two dimers in the K. pneumoniae crystal (the one in which the C7 group is pointing towards the DNA away from ParE, although the conformations of these two groups on the drug are probably not correlated). The topoisomerase IV ParE27-ParC55 fusion protein from K. pneumoniae was fully active in promoting levofloxacin-mediated cleavage of DNA (Fig. 6 ▸). In the absence of the drug and ATP, the protein converted supercoiled pBR322 into a ladder of bands corresponding to relaxed DNA. The inclusion of levofloxacin produced linear DNA in a dose-dependent and ATP-independent fashion. Similar behaviour was observed for the S. pneumoniae topoisomerase IV ParE30-ParC55 fusion protein. The CC25 (the drug concentration that converted 25% of the supercoiled DNA substrate to a linear form) was 0.5 µM for the Klebsiella enzyme and 1 µM for the pneumococcal enzyme. Interestingly, K. pneumoniae strains are much more susceptible to levofloxacin than S. pneumoniae, with typical MIC values of 0.016 and 1 mg l, respectively (Odenholt & Cars, 2006 ▸), reflecting differences in multiple factors (in addition to binding affinity) that influence drug responses, including membrane, peptidoglycan structure, drug-uptake and efflux mechanisms. Moreover, although topoisomerase IV is primarily the target of levofloxacin in S. pneumoniae, it is likely to be gyrase in the Gram-negative K. pneumoniae. In summary, we have determined the first structure of a quinolone–DNA cleavage complex involving a type II topoisomerase from K. pneumoniae. Given the current concerns about drug-resistant strains of Klebsiella, the structure reported here provides key information in understanding the action of currently used quinolones and should aid in the development of other topoisomerase-targeting therapeutics active against this major human pathogen.
PMC4850273	Molecular Dissection of Xyloglucan Recognition in a Prominent Human Gut Symbiont	Polysaccharide utilization loci (PUL) within the genomes of resident human gut Bacteroidetes are central to the metabolism of the otherwise indigestible complex carbohydrates known as “dietary fiber.” However, functional characterization of PUL lags significantly behind sequencing efforts, which limits physiological understanding of the human-bacterial symbiosis. In particular, the molecular basis of complex polysaccharide recognition, an essential prerequisite to hydrolysis by cell surface glycosidases and subsequent metabolism, is generally poorly understood. Here, we present the biochemical, structural, and reverse genetic characterization of two unique cell surface glycan-binding proteins (SGBPs) encoded by a xyloglucan utilization locus (XyGUL) from Bacteroides ovatus, which are integral to growth on this key dietary vegetable polysaccharide. Biochemical analysis reveals that these outer membrane-anchored proteins are in fact exquisitely specific for the highly branched xyloglucan (XyG) polysaccharide. The crystal structure of SGBP-A, a SusD homolog, with a bound XyG tetradecasaccharide reveals an extended carbohydrate-binding platform that primarily relies on recognition of the β-glucan backbone. The unique, tetra-modular structure of SGBP-B is comprised of tandem Ig-like folds, with XyG binding mediated at the distal C-terminal domain. Despite displaying similar affinities for XyG, reverse-genetic analysis reveals that SGBP-B is only required for the efficient capture of smaller oligosaccharides, whereas the presence of SGBP-A is more critical than its carbohydrate-binding ability for growth on XyG. Together, these data demonstrate that SGBP-A and SGBP-B play complementary, specialized roles in carbohydrate capture by B. ovatus and elaborate a model of how vegetable xyloglucans are accessed by the Bacteroidetes.The human gut microbiota influences the course of human development and health, playing key roles in immune stimulation (1), intestinal cell proliferation (2), and metabolic balance (3, 4). This microbial community is largely bacterial, with the Bacteroidetes, Firmicutes, and Actinobacteria comprising the dominant phyla (5, 6). The ability to acquire energy from carbohydrates of dietary or host origin is central to the adaptation of human gut bacterial species to their niche. More importantly, this makes diet a tractable way to manipulate the abundance and metabolic output of the microbiota toward improved human health. However, there is a paucity of data regarding how the vast array of complex carbohydrate structures are selectively recognized and imported by members of the microbiota, a critical process that enables these organisms to thrive in the competitive gut environment. The human gut bacteria Bacteroidetes share a profound capacity for dietary glycan degradation, with many species containing >250 predicted carbohydrate-active enzymes (CAZymes), compared to 50 to 100 within many Firmicutes and only 17 in the human genome devoted toward carbohydrate utilization (7). A remarkable feature of the Bacteroidetes is the packaging of genes for carbohydrate catabolism into discrete polysaccharide utilization loci (PUL) (8), which are transcriptionally regulated by specific substrate signatures (9–11). The archetypal PUL-encoded system is the starch utilization system (Sus) (Fig. 1B) of Bacteroides thetaiotaomicron (12, 13). The Sus includes a lipid-anchored, outer membrane endo-amylase, SusG (14, 15); a TonB-dependent transporter (TBDT), SusC, which imports oligosaccharides with the help of an associated starch-binding protein, SusD (13, 16, 17); two additional carbohydrate-binding lipoproteins, SusE and SusF (18); and two periplasmic exo-glucosidases, SusA and SusB, which generate glucose for transport into the cytoplasm (19). The importance of PUL as a successful evolutionary strategy is underscored by the observation that Bacteroidetes such as B. thetaiotaomicron and Bacteroides ovatus devote ~18% of their genomes to these systems (11). Moving beyond seminal genomic and transcriptomic analyses, the current state-of-the-art PUL characterization involves combined reverse-genetic, biochemical, and structural studies to illuminate the molecular details of PUL function (20–23). Xyloglucan and the Bacteroides ovatus xyloglucan utilization locus (XyGUL). (A) Representative structures of common xyloglucans (60) using the Consortium for Functional Glycomics Symbol Nomenclature (http://www.functionalglycomics.org/static/consortium/Nomenclature.shtml). Cleavage sites for BoXyGUL glycosidases (GHs) are indicated for solanaceous xyloglucan. (B) BtSus and BoXyGUL. (C) Localization of BoXyGUL-encoded proteins in cellular membranes and concerted modes of action in the degradation of xyloglucans to monosaccharides. The location of SGBP-A/B is presented in this work; the location of GH5 has been empirically determined, and the enzymes have been placed based upon their predicted cellular location (23). We recently reported the detailed molecular characterization of a PUL that confers the ability of the human gut commensal B. ovatus ATCC 8483 to grow on a prominent family of plant cell wall glycans, the xyloglucans (XyG) (23). XyG variants (Fig. 1A) constitute up to 25% of the dry weight of common vegetables (24, 25). Analogous to the Sus locus, the xyloglucan utilization locus (XyGUL) encodes a cohort of carbohydrate-binding, -hydrolyzing, and -importing proteins (Fig. 1B and C). The number of glycoside hydrolases (GHs) encoded by the XyGUL is, however, more expansive than that by the Sus locus (Fig. 1B), which reflects the greater complexity of glycosidic linkages found in XyG vis-à-vis starch (10). Whereas our previous study focused on the characterization of the linkage specificity of these GHs (23), a key outstanding question regarding this locus is how XyG recognition is mediated at the cell surface. In the archetypal starch utilization system of B. thetaiotaomicron, starch binding to the cell surface is mediated at eight distinct starch-binding sites distributed among four surface glycan-binding proteins (SGBPs): two within the amylase SusG, one within SusD, two within SusE, and three within SusF (15, 16, 18). The functional redundancy of many of these sites is high: whereas SusD is essential for growth on starch, combined mutations of the SusE, SusF, and SusG binding sites are required to impair growth on the polysaccharide (16, 26). Bacteroidetes PUL ubiquitously encode homologs of SusC and SusD, as well as proteins whose genes are immediately downstream of susD, akin to susE/F, and these are typically annotated as “putative lipoproteins” (8, 27). The genes coding for these proteins, sometimes referred to as “susE/F positioned,” display products with a wide variation in amino acid sequence and which have little or no homology to other PUL-encoded proteins or known carbohydrate-binding proteins (28, 29). As the Sus SGBPs remain the only structurally characterized cohort to date, we therefore wondered whether such glycan binding and function are extended to other PUL that target more complex and heterogeneous polysaccharides, such as XyG. We describe here the detailed functional and structural characterization of the noncatalytic SGBPs encoded by Bacova_02651 and Bacova_02650 of the XyGUL, here referred to as SGBP-A and SGBP-B, to elucidate their molecular roles in carbohydrate acquisition in vivo. Combined biochemical, structural, and reverse-genetic approaches clearly illuminate the distinct, yet complementary, functions that these two proteins play in XyG recognition as it impacts the physiology of B. ovatus. These data extend our current understanding of the Sus-like glycan uptake paradigm within the Bacteroidetes and reveals how the complex dietary polysaccharide xyloglucan is recognized at the cell surface. SGBP-A, encoded by the XyGUL locus tag Bacova_02651 (Fig. 1B), shares 26% amino acid sequence identity (40% similarity) with its homolog, B. thetaiotaomicron SusD (16), and similar homology with the SusD-like proteins encoded within syntenic XyGUL identified in our earlier work (23). In contrast, SGBP-B, encoded by locus tag Bacova_02650, displays little sequence similarity to the products of similarly positioned genes in syntenic XyGUL nor to any other gene product among the diversity of Bacteroidetes PUL (18, 27). Whereas sequence similarity among SusC/SusD homolog pairs often serves as a hallmark for PUL identification, the sequence similarities of downstream genes encoding SGBPs are generally too low to allow reliable bioinformatic classification of their products into protein families, let alone prediction of function (30). Hence, there is a critical need for the elucidation of detailed structure-function relationships among PUL SGBPs, in light of the manifold glycan structures in nature. Immunofluorescence of formaldehyde-fixed, nonpermeabilized cells grown in minimal medium with XyG as the sole carbon source to induce XyGUL expression (23), reveals that both SGBP-A and SGBP-B are presented on the cell surface by N-terminal lipidation, as predicted by signal peptide analysis with SignalP (Fig. 2). Here, the SGBPs very likely work in concert with the cell-surface-localized endo-xyloglucanase B. ovatus GH5 (BoGH5) (23) to recruit and cleave XyG for subsequent periplasmic import via the SusC-like TBDT of the XyGUL (Fig. 1B and C). SGBP-A and SGBP-B visualized by immunofluorescence. Formalin-fixed, nonpermeabilized B. ovatus cells were grown in minimal medium plus XyG, probed with custom rabbit antibodies to SGBP-A or SGBP-B, and then stained with Alexa Fluor 488 goat anti-rabbit IgG. (A) Overlay of bright-field and FITC images of B. ovatus cells labeled with anti-SGBP-A. (B) Overlay of bright-field and FITC images of B. ovatus cells labeled with anti-SGBP-B. (C) Bright-field image of ΔSGBP-B cells labeled with anti-SGBP-B antibodies. (D) FITC images of ΔSGBP-B cells labeled with anti-SGBP-B antibodies. Cells lacking SGBP-A (ΔSGBP-A) do not grow on XyG and therefore could not be tested in parallel. In our initial study focused on the functional characterization of the glycoside hydrolases of the XyGUL, we reported preliminary affinity PAGE and isothermal titration calorimetry (ITC) data indicating that both SGBP-A and SGBP-B are competent xyloglucan-binding proteins (affinity constant [Ka] values of 3.74 × 10 M and 4.98 × 10 M, respectively ). Additional affinity PAGE analysis (Fig. 3) demonstrates that SGBP-A also has moderate affinity for the artificial soluble cellulose derivative hydroxyethyl cellulose [HEC; a β(1 → 4)-glucan] and limited affinity for mixed-linkage β(1→3)/β(1→4)-glucan (MLG) and glucomannan (GM; mixed glucosyl and mannosyl backbone), which together indicate general binding to polysaccharide backbone residues and major contributions from side-chain recognition. In contrast, SGBP-B bound to HEC more weakly than SGBP-A and did not bind to MLG or GM. Neither SGBP recognized galactomannan (GGM), starch, carboxymethylcellulose, or mucin (see Fig. S1 in the supplemental material). Together, these results highlight the high specificities of SGBP-A and SGBP-B for XyG, which is concordant with their association with XyG-specific GHs in the XyGUL, as well as transcriptomic analysis indicating that B. ovatus has discrete PUL for MLG, GM, and GGM (11). Notably, the absence of carbohydrate-binding modules in the GHs encoded by the XyGUL (23) implies that noncatalytic recognition of xyloglucan is mediated entirely by SGBP-A and -B. SGBP-A and SGBP-B preferentially bind xyloglucan. Affinity electrophoresis (10% acrylamide) of SGBP-A and SGBP-B with BSA as a control protein. All samples were loaded on the same gel next to the BSA controls; thin black lines indicate where intervening lanes were removed from the final image for both space and clarity. The percentage of polysaccharide incorporated into each native gel is displayed. The vanguard endo-xyloglucanase of the XyGUL, BoGH5, preferentially cleaves the polysaccharide at unbranched glucosyl residues to generate xylogluco-oligosaccharides (XyGOs) comprising a Glc4 backbone with variable side-chain galactosylation (XyGO1) (Fig. 1A; n = 1) as the limit of digestion products in vitro (23); controlled digestion and fractionation by size exclusion chromatography allow the production of higher-order oligosaccharides (e.g., XyGO2) (Fig. 1A; n = 2). ITC demonstrates that SGBP-A binds to XyG polysaccharide and XyGO2 (based on a Glc8 backbone) with essentially equal affinities, while no binding of XyGO1 (Glc4 backbone) was detectable (Table 1; see Fig. S2 and S3 in the supplemental material). Similarly, SGBP-B also bound to XyG and XyGO2 with approximately equal affinities, although in both cases, Ka values were nearly 10-fold lower than those for SGBP-A. Also in contrast to SGBP-A, SGBP-B also bound to XyGO1, yet the affinity for this minimal repeating unit was poor, with a Ka value of ca. 1 order of magnitude lower than for XyG and XyGO2. Together, these data clearly suggest that polysaccharide binding of both SGBPs is fulfilled by a dimer of the minimal repeat, corresponding to XyGO2 (cf. Fig. 1A). The observation by affinity PAGE that these proteins specifically recognize XyG is further substantiated by their lack of binding for the undecorated oligosaccharide cellotetraose (Table 1; see Fig. S3). Furthermore, SGBP-A binds cellohexaose with ~770-fold weaker affinity than XyG, while SGBP-B displays no detectable binding to this linear hexasaccharide. To provide molecular-level insight into how the XyGUL SGBPs equip B. ovatus to specifically harvest XyG from the gut environment, we performed X-ray crystallography analysis of both SGBP-A and SGPB-B in oligosaccharide-complex forms. Summary of thermodynamic parameters for wild-type SGBP-A and SGBP-B obtained by isothermal titration calorimetry at 25°C Shown are average values ± standard errors from two independent titrations, unless otherwise indicated. Binding thermodynamics for XyG based on the concentration of the binding unit, XyGO2. Values from a single titration. NB, no binding observed. As anticipated by sequence similarity, the high-resolution tertiary structure of apo-SGBP-A (1.36 Å, Rwork = 14.7%, Rfree = 17.4%, residues 28 to 546) (Table 2) displays the canonical “SusD-like” protein fold dominated by four tetratrico-peptide repeat (TPR) motifs that cradle the rest of the structure (Fig. 4A) (31). Specifically, SGBP-A overlays B. thetaiotaomicron SusD (BtSusD) with a root mean square deviation (RMSD) value of 2.2 Å for 363 Cα pairs, which is notable given the 26% amino acid identity (40% similarity) between these homologs (Fig. 4C). Cocrystallization of SGBP-A with XyGO2 generated a substrate complex structure (2.3 Å, Rwork = 21.8%, Rfree = 24.8%, residues 36 to 546) (Fig. 4A and B; Table 2) that revealed the distinct binding-site architecture of the XyG binding protein. The SGBP-A:XyGO2 complex superimposes closely with the apo structure (RMSD of 0.6 Å) and demonstrates that no major conformational change occurs upon substrate binding; small deviations in the orientation of several surface loops are likely the result of differential crystal packing. It is particularly notable that although the location of the ligand-binding site is conserved between SGBP-A and SusD, that of SGBP-A displays an ~29-Å-long aromatic platform to accommodate the extended, linear XyG chain (see reference 32 for a review of XyG secondary structure), versus the shorter, ~18-Å-long, site within SusD that complements the helical conformation of amylose (16) (Fig. 4C and D). Molecular structure of SGBP-A (Bacova_02651). (A) Overlay of SGBP-A from the apo (rainbow) and XyGO2 (gray) structures. The apo structure is color ramped from blue to red. An omit map (2σ) for XyGO2 (orange and red sticks) is displayed. (B) Close-up view of the omit map as in panel A, rotated 90° clockwise. (C) Overlay of the Cα backbones of SGBP-A (black) with XyGO2 (orange and red spheres) and BtSusD (blue) with maltoheptaose (pink and red spheres), highlighting the conservation of the glycan-binding site location. (D) Close-up of the SGBP-A (black and orange) and SusD (blue and pink) glycan-binding sites. The approximate length of each glycan-binding site is displayed, colored to match the protein structures. (E) Stereo view of the xyloglucan-binding site of SGBP-A, displaying all residues within 4 Å of the ligand. The backbone glucose residues are numbered from the nonreducing end; xylose residues are labeled X1 and X2. Potential hydrogen-bonding interactions are shown as dashed lines, and the distance is shown in angstroms. X-ray data collection and refinement statistics Numbers in parentheses are for the highest-resolution shell. CC1/2, Pearson correlation coefficient between the average intensities of each subset. CC, Pearson correlation coefficient for correlation between the observed data set and true signal. Seven of the eight backbone glucosyl residues of XyGO2 could be convincingly modeled in the ligand electron density, and only two α(1→6)-linked xylosyl residues were observed (Fig. 4B; cf. Fig. 1). Indeed, the electron density for the ligand suggests some disorder, which may arise from multiple oligosaccharide orientations along the binding site. Three aromatic residues—W82, W283, W306—comprise the flat platform that stacks along the naturally twisted β-glucan backbone (Fig. 4E). The functional importance of this platform is underscored by the observation that the W82A W283A W306A mutant of SGBP-A, designated SGBP-A, is completely devoid of XyG affinity (Table 3; see Fig. S4 in the supplemental material). Dissection of the individual contribution of these residues reveals that the W82A mutant displays a significant 4.9-fold decrease in the Ka value for XyG, while the W306A substitution completely abolishes XyG binding. Contrasting with the clear importance of these hydrophobic interactions, there are remarkably few hydrogen-bonding interactions with the ligand, which are provided by R65, N83, and S308, which are proximal to Glc5 and Glc3. Most surprising in light of the saccharide-binding data, however, was a lack of extensive recognition of the XyG side chains; only Y84 appeared to provide a hydrophobic interface for a xylosyl residue (Xyl1). Summary of thermodynamic parameters for site-directed mutants of SGBP-A and SGBP-B obtained by ITC with XyG at 25°C Shown are average values ± standard deviations from two independent titrations, unless otherwise indicated. Binding thermodynamics are based on the concentration of the binding unit, XyGO2. Weak binding represents a Ka of <500 M. ND, not determined; NB, no binding. Ka fold change = Ka of wild-type protein/Ka of mutant protein for xyloglucan binding. Values from a single titration. The crystal structure of full-length SGBP-B in complex with XyGO2 (2.37 Å, Rwork = 19.9%, Rfree = 23.9%, residues 34 to 489) (Table 2) revealed an extended structure composed of three tandem immunoglobulin (Ig)-like domains (domains A, B, and C) followed at the C terminus by a novel xyloglucan-binding domain (domain D) (Fig. 5A). Domains A, B, and C display similar β-sandwich folds; domains B (residues 134 to 230) and C (residues 231 to 313) can be superimposed onto domain A (residues 34 to 133) with RMSDs of 1.1 and 1.2 Å, respectively, for 47 atom pairs (23% and 16% sequence identity, respectively). These domains also display similarity to the C-terminal β-sandwich domains of many GH13 enzymes, including the cyclodextrin glucanotransferase of Geobacillus stearothermophilus (Fig. 5B). Such domains are not typically involved in carbohydrate binding. Indeed, visual inspection of the SGBP-B structure, as well as individual production of the A and B domains and affinity PAGE analysis (see Fig. S5 in the supplemental material), indicates that these domains do not contribute to XyG capture. On the other hand, production of the fused domains C and D in tandem (SGBP-B residues 230 to 489) retains complete binding of xyloglucan in vitro, with the observed slight increase in affinity likely arising from a reduced potential for steric hindrance of the smaller protein construct during polysaccharide interactions (Table 3). While neither the full-length protein nor domain D displays structural homology to known XyG-binding proteins, the topology of SGBP-B resembles the xylan-binding protein Bacova_04391 (PDB 3ORJ) encoded within a xylan-targeting PUL of B. ovatus (22) (Fig. 5C). The structure-based alignment of these proteins reveals 17% sequence identity, with a core RMSD of 3.6 Å for 253 aligned residues. While there is no substrate-complexed structure of Bacova_04391 available, the binding site is predicted to include W241 and Y404 (22), which are proximal to the XyGO binding site in SGBP-B. However, the opposing, clamp-like arrangement of these residues in Bacova_04391 is clearly distinct from the planar surface arrangement of the residues that interact with XyG in SGBP-B (described below). Multimodular structure of SGBP-B (Bacova_02650). (A) Full-length structure of SGBP-B, color coded by domain as indicated. Prolines between domains are indicated as spheres. An omit map (2σ) for XyGO2 is displayed to highlight the location of the glycan-binding site. (B) Overlay of SGBP-B domains A, B, and C (colored as in panel A), with a C-terminal Ig-like domain of the G. stearothermophilus cyclodextrin glucanotransferase (PDB 1CYG [residues 375 to 493]) in green. (C) Cα overlay of SGBP-B (gray) and Bacova_04391 (PDB 3ORJ) (pink). (D) Close-up omit map for the XyGO2 ligand, contoured at 2σ. (E) Stereo view of the xyloglucan-binding site of SGBP-B, displaying all residues within 4 Å of the ligand. The backbone glucose residues are numbered from the nonreducing end, xylose residues are shown as X1, X2, and X3, potential hydrogen-bonding interactions are shown as dashed lines, and the distance is shown in angstroms. Inspection of the tertiary structure indicates that domains C and D are effectively inseparable, with a contact interface of 396 Å. Domains A, B, and C do not pack against each other. Moreover, the five-residue linkers between these first three domains all feature a proline as the middle residue, suggesting significant conformational rigidity (Fig. 5A). Despite the lack of sequence and structural conservation, a similarly positioned proline joins the Ig-like domains of the xylan-binding Bacova_04391 and the starch-binding proteins SusE and SusF. We speculate that this is a biologically important adaptation that serves to project the glycan binding site of these proteins far from the membrane surface. Any mobility of SGBP-B on the surface of the cell (beyond lateral diffusion within the membrane) is likely imparted by the eight-residue linker that spans the predicted lipidated Cys (C28) and the first β-strand of domain A. Other outer membrane proteins from various Sus-like systems possess a similar 10- to 20-amino-acid flexible linker between the lipidated Cys that tethers the protein to the outside the cell and the first secondary structure element (15, 16, 33). Analogously, the outer membrane-anchored endo-xyloglucanase BoGH5 of the XyGUL contains a 100-amino-acid, all-β-strand, N-terminal module and flexible linker that imparts conformational flexibility and distances the catalytic module from the cell surface (23). XyG binds to domain D of SGBP-B at the concave interface of the top β-sheet, with binding mediated by loops connecting the β-strands. Six glucosyl residues, comprising the main chain, and three branching xylosyl residues of XyGO2 can be modeled in the density (Fig. 5D; cf. Fig. 1A). The backbone is flat, with less of the “twisted-ribbon” geometry observed in some cello- and xylogluco-oligosaccharides (34–36). The aromatic platform created by W330, W364, and Y363 spans four glucosyl residues, compared to the longer platform of SGBP-A, which supports six glucosyl residues (Fig. 5E). The Y363A site-directed mutant of SGBP-B displays a 20-fold decrease in the Ka for XyG, while the W364A mutant lacks XyG binding (Table 3; see Fig. S6 in the supplemental material). There are no additional contacts between the protein and the β-glucan backbone and surprisingly few interactions with the side-chain xylosyl residues, despite that fact that ITC data demonstrate that SGBP-B does not measurably bind the cellohexaose (Table 1). F414 stacks with the xylosyl residue of Glc3, while Q407 is positioned for hydrogen bonding with the O4 of xylosyl residue Xyl1. Surprisingly, an F414A mutant of SGBP-B displays only a mild 3-fold decrease in the Ka value for XyG, again suggesting that glycan recognition is primarily mediated via contact with the β-glucan backbone (Table 3; see Fig. S6). Additional residues surrounding the binding site, including Y369 and E412, may contribute to the recognition of more highly decorated XyG, but precisely how this is mediated is presently unclear. Hoping to achieve a higher-resolution view of the SGBP-B–xyloglucan interaction, we solved the crystal structure of the fused CD domains in complex with XyGO2 (1.57 Å, Rwork = 15.6%, Rfree = 17.1%, residues 230 to 489) (Table 2). The CD domains of the truncated and full-length proteins superimpose with a 0.4-Å RMSD of the Cα backbone, with no differences in the position of any of the glycan-binding residues (see Fig. S7A in the supplemental material). While density is observed for XyGO2, the ligand could not be unambiguously modeled into this density to achieve a reasonable fit between the X-ray data and the known stereochemistry of the sugar (see Fig. S7B and C). While this may occur for a number of reasons in crystal structures, it is likely that the poor ligand density even at higher resolution is due to movement or multiple orientations of the sugar averaged throughout the lattice. The similarity of the glycan specificity of SGBP-A and SGBP-B presents an intriguing conundrum regarding their individual roles in XyG utilization by B. ovatus. To disentangle the functions of SGBP-A and SGBP-B in XyG recognition and uptake, we created individual in-frame deletion and complementation mutant strains of B. ovatus. In these growth experiments, overnight cultures of strains grown on minimal medium plus glucose were back-diluted 1:100-fold into minimal medium containing 5 mg/ml of the reported carbohydrate. Growth on glucose displayed the shortest lag time for each strain, and so lag times were normalized for each carbohydrate by subtracting the lag time of that strain in glucose (Fig. 6; see Fig. S8 in the supplemental material). A strain in which the entire XyGUL is deleted displays a lag of 24.5 h during growth on glucose compared to the isogenic parental wild-type (WT) Δtdk strain, for which exponential growth lags for 19.8 h (see Fig. S8D). It is unknown whether this is because cultures were not normalized by the starting optical density (OD) or viable cells or reflects a minor defect for glucose utilization. The former seems more likely as the growth rates are nearly identical for these strains on glucose and xylose. The ΔXyGUL and WT Δtdk strains display normalized lag times on xylose within experimental error, and curiously some of the mutant and complemented strains display a nominally shorter lag time on xylose than the WT Δtdk strain. Complementation of the ΔSGBP-A strain (ΔSGBP-A::SGBP-A) restores growth to wild-type rates on xyloglucan and XyGO1, yet the calculated rate of the complemented strain is ~72% that of the WT Δtdk strain on XyGO2; similar results were obtained for the SGBP-B complemented strain despite the fact that the growth curves do not appear much different (see Fig. S8C and F). The reason for this observation on XyGO2 is unclear, as the ΔSGBP-B mutant does not have a significantly different growth rate from the WT on XyGO2. Therefore, we limit our discussion to those mutants that displayed the most obvious defects in growth on particular substrates. Growth of select XyGUL mutants on xyloglucan and oligosaccharides. B. ovatus mutants were created in a thymidine kinase deletion (Δtdk) mutant as described previously (23). SGBP-A* denotes the Bacova_02651 (W82A W283A W306A) allele, and the GH9 gene is Bacova_02649. Growth was measured over time in minimal medium containing (A) XyG, (B) XyGO2, (C) XyGO1, (D) glucose, and (E) xylose. In panel F, the growth rate of each strain on the five carbon sources is displayed, and in panel G, the normalized lag time of each culture, relative to its growth on glucose, is displayed. Solid bars indicate conditions that are not statistically significant from the WT Δtdk cultures grown on the indicated carbohydrate, while open bars indicate a P value of <0.005 compared to the WT Δtdk strain. Conditions denoted by the same letter (b, c, or d) are not statistically significant from each other but are significantly different from the condition labeled “a.” Complementation of ΔSGBP-A and ΔSBGP-B was performed by allelic exchange of the wild-type genes back into the genome for expression via the native promoter: these growth curves, quantified rates and lag times are displayed in Fig. S8 in the supplemental material. The ΔSGBP-A (ΔBacova_02651) strain (cf. Fig. 1B) was completely incapable of growth on XyG, XyGO1, and XyGO2, indicating that SGBP-A is essential for XyG utilization (Fig. 6). This result mirrors our previous data for the canonical Sus of B. thetaiotaomicron, which revealed that a homologous ΔsusD mutant is unable to grow on starch or malto-oligosaccharides, despite normal cell surface expression of all other PUL-encoded proteins (16, 26). More recently, we demonstrated that this phenotype is due to the loss of the physical presence of SusD; complementation of ΔsusD with SusD, a triple site-directed mutant (W96A W320A Y296A) that ablates glycan binding, restores B. thetaiotaomicron growth on malto-oligosaccharides and starch when sus transcription is induced by maltose addition (26). Similarly, the function of SGBP-A extends beyond glycan binding. Complementation of ΔSGBP-A with the SGBP-A (W82A W283A W306A) variant, which does not bind XyG, supports growth on XyG and XyGOs (Fig. 6; ΔSGBP-A::SGBP-A), with growth rates that are ~70% that of the WT. In previous studies, we observed that carbohydrate binding by SusD enhanced the sensitivity of the cells to limiting concentrations of malto-oligosaccharides by several orders of magnitude, such that the addition of 0.5 g/liter maltose was required to restore growth of the ΔsusD::SusD strain on starch, which nonetheless occurred following an extended lag phase (26). In contrast, the ΔSGBP-A::SGBP-A* strain does not display an extended lag time on any of the xyloglucan substrates compared to the WT (Fig. 6). The specific glycan signal that upregulates BoXyGUL is currently unknown. From our present data, we cannot eliminate the possibility that the glycan binding by SGBP-A enhances transcriptional activation of the XyGUL. However, the modest rate defect displayed by the SGBP-A::SGBP-A* strain suggests that recognition of XyG and product import is somewhat less efficient in these cells. Intriguingly, the ΔSGBP-B strain (ΔBacova_02650) (cf. Fig. 1B) exhibited a minor growth defect on both XyG and XyGO2, with rates 84.6% and 93.9% that of the WT Δtdk strain. However, growth of the ΔSGBP-B strain on XyGO1 was 54.2% the rate of the parental strain, despite the fact that SGBP-B binds this substrate ca. 10-fold more weakly than XyGO2 and XyG (Fig. 6; Table 1). As such, the data suggest that SGBP-A can compensate for the loss of function of SGBP-B on longer oligo- and polysaccharides, while SGBP-B may adapt the cell to recognize smaller oligosaccharides efficiently. Indeed, a double mutant, consisting of a crippled SGBP-A and a deletion of SGBP-B (ΔSGBP-A::SGBP-A/ΔSGBP-B), exhibits an extended lag time on both XyG and XyGO2, as well as XyGO1. Taken together, the data indicate that SGBP-A and SGBP-B functionally complement each other in the capture of XyG polysaccharide, while SGBP-B may allow B. ovatus to scavenge smaller XyGOs liberated by other gut commensals. This additional role of SGBP-B is especially notable in the context of studies on BtSusE and BtSusF (positioned similarly in the archetypal Sus locus) (Fig. 1B), for which growth defects on starch or malto-oligosaccharides have never been observed. Beyond SGBP-A and SGBP-B, we speculated that the catalytically feeble endo-xyloglucanase GH9, which is expendable for growth in the presence of GH5, might also play a role in glycan binding to the cell surface (23). However, combined deletion of the genes encoding GH9 (encoded by Bacova_02649) and SGBP-B does not exacerbate the growth defect on XyGO1 (Fig. 6; ΔSGBP-B/ΔGH9). The necessity of SGBP-B is elevated in the SGBP-A strain, as the ΔSGBP-A::SGBP-A/ ΔSGBP-B mutant displays an extended lag during growth on XyG and xylogluco-oligosaccharides, while growth rate differences are more subtle. The precise reason for this lag is unclear, but recapitulating our findings on the role of SusD in malto-oligosaccharide sensing in B. thetaiotaomicron, this extended lag may be due to inefficient import and thus sensing of xyloglucan in the environment in the absence of glycan binding by essential SGBPs. Our previous work demonstrates that B. ovatus cells grown in minimal medium plus glucose express low levels of the XyGUL transcript (23). Thus, in our experiments, we presume that each strain, initially grown in glucose, expresses low levels of the XyGUL transcript and thus low levels of the XyGUL-encoded surface proteins, including the vanguard GH5. Presumably without glycan binding by the SGBPs, the GH5 protein cannot efficiently process xyloglucan, and/or the lack of SGBP function prevents efficient capture and import of the processed oligosaccharides. It may then be that only after a sufficient amount of glycan is processed and imported by the cell is XyGUL upregulated and exponential growth on the glycan can begin. We hypothesize that during exponential growth the essential role of SGBP-A extends beyond glycan recognition, perhaps due to a critical interaction with the TBDT. In the BtSus, SusD and the TBDT SusC interact (37), and we speculate that this interaction is necessary for glycan uptake, as suggested by the fact that a ΔsusD mutant cannot grow on starch (16, 26), but a ΔsusD::SusD strain regains this ability if a transcriptional activator of the sus operon is supplied (26). Likewise, such cognate interactions between homologous protein pairs such as SGBP-A and its TBDT may underlie our observation that a ΔSGBP-A mutant cannot grow on xyloglucan. However, unlike the Sus, in which elimination of SusE and SusF does not affect growth on starch, SGBP-B appears to have a dedicated role in growth on small xylogluco-oligosaccharides. The ability of gut-adapted microorganisms to thrive in the gastrointestinal tract is critically dependent upon their ability to efficiently recognize, cleave, and import glycans. The human gut, in particular, is a densely packed ecosystem with hundreds of species, in which there is potential for both competition and synergy in the utilization of different substrates. Recent work has elucidated that Bacteroidetes cross-feed during growth on many glycans; the glycoside hydrolases expressed by one species liberate oligosaccharides for consumption by other members of the community (38, 39). Thus, understanding glycan capture at the cell surface is fundamental to explaining, and eventually predicting, how the carbohydrate content of the diet shapes the gut community structure as well as its causative health effects. Here, we demonstrate that the surface glycan binding proteins encoded within the BoXyGUL play unique and essential roles in the acquisition of the ubiquitous and abundant vegetable polysaccharide xyloglucan. Yet, a number of questions remain regarding the molecular interplay of SGBPs with their cotranscribed cohort of glycoside hydrolases and TonB-dependent transporters. A particularly understudied aspect of glycan utilization is the mechanism of import via TBDTs (SusC homologs) (Fig. 1), which are ubiquitous and defining components of all PUL (28, 40). PUL-encoded TBDTs in Bacteroidetes are larger than the well-characterized iron-targeting TBDTs from many Proteobacteria and are further distinguished as the only known glycan-importing TBDTs coexpressed with an SGBP (41). A direct interaction between the BtSusC TBDT and the SusD SGBP has been previously demonstrated (13, 37), as has an interaction between the homologous components encoded by an N-glycan-scavenging PUL of Capnocytophaga canimorsus (42). Our observation here that the physical presence of the SusD homolog SGBP-A, independent of XyG-binding ability, is both necessary and sufficient for XyG utilization further supports a model of glycan import whereby the SusC-like TBDTs and the SusD-like SGBPs must be intimately associated to support glycan uptake (Fig. 1C). It is yet presently unclear whether this interaction is static or dynamic and to what extent the association of cognate TBDT/SGBPs is dependent upon the structure of the carbohydrate to be imported. On the other hand, there is clear evidence for independent TBDTs in Bacteroidetes that do not require SGBP association for activity. For example, it was recently demonstrated that expression of nanO, which encodes a SusC-like TBDT as part of a sialic-acid-targeting PUL from B. fragilis, restored growth on this monosaccharide in a mutant strain of E. coli (43). In this instance, coexpression of the susD-like gene nanU was not required, nor did the expression of the nanU gene enhance growth kinetics. Similarly, the deletion of BT1762 encoding a fructan-targeting SusD-like protein in B. thetaiotaomicron did not result in a dramatic loss of growth on fructans (44). Thus, the strict dependence on a SusD-like SGBP for glycan uptake in the Bacteroidetes may be variable and substrate dependent. Furthermore, considering the broader distribution of TBDTs in PUL lacking SGBPs (sometimes known as carbohydrate utilization containing TBDT [CUT] loci; see reference 45 and reviewed in reference 40) across bacterial phyla, it appears that the intimate biophysical association of these substrate-transport and -binding proteins is the result of specific evolution within the Bacteroidetes. Equally intriguing is the observation that while SusD-like proteins such as SGBP-A share moderate primary and high tertiary structural conservation, the genes for the SGBPs encoded immediately downstream (Fig. 1B [sometimes referred to as “susE positioned”]) encode glycan-binding lipoproteins with little or no sequence or structural conservation, even among syntenic PUL that target the same polysaccharide. Such is the case for XyGUL from related Bacteroides species, which may encode either one or two of these predicted SGBPs, and these proteins vary considerably in length (23). The extremely low similarity of these SGBPs is striking in light of the moderate sequence conservation observed among homologous GHs in syntenic PUL. This, together with the observation that these SGBPs, as exemplified by BtSusE and BtSusF (26) and the XyGUL SGBP-B of the present study, are expendable for polysaccharide growth, implies a high degree of evolutionary flexibility to enhance glycan capture at the cell surface. Because the intestinal ecosystem is a dense consortium of bacteria that must compete for their nutrients, these multimodular SGBPs may reflect ongoing evolutionary experiments to enhance glycan uptake efficiency. Whether organisms that express longer SGBPs, extending further above the cell surface toward the extracellular environment, are better equipped to compete for available carbohydrates is presently unknown. However, the natural diversity of these proteins represents a rich source for the discovery of unique carbohydrate-binding motifs to both inform gut microbiology and generate new, specific carbohydrate analytical reagents (46). In conclusion, the present study further illuminates the essential role that surface-glycan binding proteins play in facilitating the catabolism of complex dietary carbohydrates by Bacteroidetes. The ability of our resident gut bacteria to recognize polysaccharides is the first committed step of glycan consumption by these organisms, a critical process that influences the community structure and thus the metabolic output (i.e., short-chain fatty acid and metabolite profile) of these organisms. A molecular understanding of glycan uptake by human gut bacteria is therefore central to the development of strategies to improve human health through manipulation of the microbiota (47, 48). The gene fragments corresponding to Bacova_02650 (encoding SGBP-B residues 34 to 489) and Bacova_02651 (encoding SGBP-A residues 28 to 546) were amplified from Bacteroidetes ovatus ATCC 8483 genomic DNA by PCR using forward primers, including NdeI restriction sites, and reverse primers, including XhoI. The gene products were ligated into a modified version of pET-28a (EMD Biosciences) containing a recombinant tobacco etch virus (rTEV) protease recognition site (pET-28aTEV) preceding an N-terminal 6-His tag for affinity purification. The expression vector (pET-28TEV) containing SGBP-B was used for subsequent cloning of the domains A (residues 34 to 133), B (residues 134 to 229), and CD (residues 230 to 489). The pET-28TEV vector expressing residues 28 to 546 of SGBP-A was utilized for carbohydrate-binding experiments and crystallization of the apo structure. To obtain crystals of SGBP-A with XyGO2, the DNA sequence coding for residues 36 to 546 was PCR amplified from genomic DNA for ligation-independent cloning into the pETite N-His vector (Lucigen, Madison, WI) according to the manufacturer’s instructions. The N-terminal primer for pETite N-His insertion contained a TEV cleavage site immediately downstream of the complementary 18-bp overlap (encoding the His tag) to create a TEV-cleavable His-tagged protein. The site-directed mutants of SGBP-A and SGBP-B in pET-28TEV were created using the QuikChange II site-directed mutagenesis kit (Stratagene) according to the manufacturer’s instructions. The sequences of all primers to generate these constructs are displayed in Table S1 in the supplemental material. The plasmids containing the SGBP-A and SGBP-B genes were transformed into Escherichia coli BL21(DE3) or Rosetta(DE3) cells. For native protein expression, cells were cultured in Terrific Broth containing kanamycin (50 µg/ml) and chloramphenicol (20 µg/ml) at 37°C to the mid-exponential phase (A600 of ≈0.6), induced by the addition of 0.5 mM isopropyl β-d-1-thiogalactopyranoside (IPTG), and then incubated for 2 days at 16°C or 1 day at 20°C. Cells were harvested by centrifugation and frozen at −80°C prior to protein purification. For selenomethionine-substituted SGBP-B, the pET-28TEV-SGBP-B plasmid was transformed into E. coli Rosetta(DE3)/pLysS and plated onto LB supplemented with kanamycin (50 µg/ml) and chloramphenicol (20 µg/ml). After 16 h of growth at 37°C, colonies were harvested from the plates, used to inoculate 100 ml of M9 minimal medium supplemented with kanamycin (30 µg/ml) and chloramphenicol (20 µg/ml), and then grown at 37°C for 16 h. This overnight culture was used to inoculate a 2-liter baffled flask containing 1 liter of Molecular Dimensions SelenoMet premade medium supplemented with 50 ml of the recommended sterile nutrient mix, chloramphenicol, and kanamycin. Cultures were grown at 37°C to an A600 of ≈0.45 before adjusting the temperature to 20°C and supplementing each flask with 100 mg each of l-lysine, l-threonine, and l-phenylalanine and 50 mg each of l-leucine, l-isoleucine, l-valine, and l-selenomethionine (49). After 20 additional minutes of growth, the cells were induced with 0.5 mM IPTG, and cultures were grown for an additional 48 h. For the purification of native and selenomethionine-substituted protein, cells were thawed and lysed via sonication in His buffer (25 mM NaH2PO4, 500 mM NaCl, 20 mM imidazole, pH 7.5) and purified via immobilized nickel affinity chromatography (His-Trap; GE Healthcare) using a gradient of 20 to 300 mM imidazole, according to the manufacturer’s instructions. The His tag was removed by incubation with TEV protease (1:100 molar ratio relative to protein) at room temperature for 2 h and then overnight at 4°C while being dialyzed against His buffer. The cleaved protein was then repurified via nickel affinity chromatography to remove undigested target protein, the cleaved His tag, and His-tagged TEV protease. Purified proteins were dialyzed against 20 mM HEPES–100 mM NaCl (pH 7.0) prior to crystallization and concentrated using Vivaspin 15 (10,000-molecular-weight-cutoff [MWCO]) centrifugal concentrators (Vivaproducts, Inc.). Xyloglucan from tamarind seed, β-glucan from barley, and konjac glucomannan were purchased from Megazyme. Starch, guar, and mucin were purchased from Sigma. Hydroxyethyl cellulose was purchased from AMRESCO. Carboxymethyl cellulose was purchased from Acros Organics. Xylogluco-oligosaccharides XyGO1 and XyGO2 for biophysical analyses were prepared from tamarind seed XyG according to the method of Martinez-Fleites et al. (61) with minor modifications. XyGO2 for cocrystallization was purchased from Megazyme (O-XGHDP). Affinity PAGE was performed as described previously (50), with minor modification. Various polysaccharides were used at a concentration of 0.05 to 0.1% (wt/vol), and electrophoresis was carried out for 90 min at room temperature in native 10% (wt/vol) polyacrylamide gels. BSA was used as noninteracting negative-control protein. Isothermal titration calorimetry (ITC) of glycan binding by the SGPB-A mutants was performed using the TA Nano isothermal titration calorimeter calibrated to 25°C. Proteins were dialyzed against 20 mM HEPES–100 mM NaCl (pH 7.0), and sugars were prepared using the dialysis buffer. The protein (45 to 50 µM) was placed in the sample cell, and the syringe was loaded with 2.5 to 4 mg/ml XyG polysaccharide. Following an initial injection of 0.5 µl, 26 subsequent injections of 2 µl were performed with stirring at 350 rpm, and the resulting heat of reaction was recorded. Data were analyzed using the Nano Analyze software. All other ITC experiments were performed using a MicroCal VP-ITC titration calorimeter calibrated to 25°C. Proteins were dialyzed into 20 mM HEPES–100 mM NaCl (pH 7.0), and polysaccharides were prepared using the dialysis buffer. Proteins (micromolar concentrations) were placed in the sample cell, and a first injection of 2 µl was performed followed by 24 subsequent injections of 10 µl of 2 to 20 mM oligosaccharide (cellotetraose, cellohexaose, XyGO1, or XyGO2) or 1 to 2.5 mg/ml XyG polysaccharide. The solution was stirred at 280 rpm, and the resulting heat of reaction was recorded. Data were analyzed using the Origin software program. Structural integrity of the SGBP-B mutants was verified by differential scanning calorimetry (DSC). DSC studies were performed on a MicroCal VP-DSC (Malvern Instruments). Experiments were carried out in 50 mM HEPES (pH 7.0) at a scan rate of 60°C/h. All samples (40 µM protein) were degassed for 7 min with gentle stirring under vacuum prior to being loaded into the calorimeter. Background excess thermal power scans were obtained with buffer in both the sample and reference cells and subtracted from the scans for each sample solution to generate excess heat capacity versus temperature thermograms. The melting temperature decreased from 57.8 ± 0.9°C for the wild-type SGBP-B protein to 54.6 ± 0.1°C for the Y363A mutant, 54.2 ± 0.1°C for the W364A mutant, and 52 ± 1°C for the F414A mutant. All proteins were therefore in their stable folded state for the ITC measurements (see Fig. S9 in the supplemental material). Gene deletions and complementations were performed via allelic exchange in a Bacteroides ovatus thymidine kinase gene (Bacova_03071) deletion (Δtdk) derivative strain of ATCC 8483 using the vector pExchange-tdk, as previously described (23). Primers for the construction of B. ovatus mutants are listed in Table S1. The B. ovatus Δtdk strain and the B. ovatus ΔXyGUL mutant were a generous gift from Eric Martens, University of Michigan Medical School. All Bacteroides ovatus culturing was performed in a Coy anaerobic chamber (85% N2, 10% H2, 5% CO2) at 37°C. Prior to growth on minimal medium plus the carbohydrates indicated (Fig. 6; see Fig. S8 in the supplemental material), each strain was grown for 16 h from a freezer stock in tryptone-yeast extract-glucose (TYG) medium (51) and then back diluted 1:100 into Bacteroides minimal medium supplemented with 5 mg/ml glucose, as previously described (52). After growth for 24 h, cultures were back-diluted 1:100 into Bacteroides minimal medium supplemented with 5 mg/ml of glucose, xylose, XyG, XyGO1, or XyGO2. Growth experiments were performed in replicates of 12 (glucose, xylose, and xyloglucan) or 5 (XyGO1 and XyGO2) as 200-µl cultures in 96-well plates. Plates were loaded into a Biostack automated plate handling device coupled to a Powerwave HT absorbance reader (both devices from Biotek Instruments, Winooski, VT). Absorbance at 600 nm (A600; i.e., optical density at 600 nm [OD600]) was measured for each well at 20-min intervals. Data were processed using Gen5 software (BioTek) and Microsoft Excel. Growth was quantified in each assay by first identifying a minimum time point (Amin) at which A600 had increased by 15% over a baseline reading taken during the first 500 min of incubation. Next, we identified the time point at which A600 reached its maximum (Amax) immediately after exponential growth. The growth rate for each well was defined by (Amax − Amin)/(Tmax − Tmin), where Tmax and Tmin are the corresponding time values for each absorbance. To account for variations in inoculum density, for each strain, the lag time (Tmin) on glucose was subtracted from the lag time for the substrate of interest; in all cases, cultures had shorter lag times on glucose than other glycans. Custom rabbit antibodies to recombinant SGBP-A and SGBP-B were generated by Cocalico Biologicals, Inc. (Reamstown, PA). The B. ovatus ATCC 8483 Δtdk and ΔSGBP-B strains were grown in 1 ml minimal Bacteroides medium (11) supplemented with 5 mg/ml tamarind xyloglucan to an A600 of ≈0.6 and then harvested via centrifugation (7,000 × g for 3 min) and washed twice with phosphate-buffered saline (PBS). Cells were resuspended in 0.25 ml PBS, and 0.75 ml of 6% formalin in PBS was added. Cells were incubated with rocking at 20°C for 1.5 h and then washed twice with PBS. Cells were resuspended in 0.5 to 1 ml blocking solution (2% goat serum, 0.02% NaN3 in PBS) and incubated for 16 h at 4°C. Cells were centrifuged and resuspended in 0.5 ml of a 1/100 dilution of custom rabbit antibody sera in blocking solution and incubated by rocking for 2 h at 20°C. Cells were washed with PBS and then resuspended in 0.4 ml of a 1/500 dilution of Alexa Fluor 488 goat anti-rabbit IgG (Life Technologies) in blocking solution and incubated with rocking for 1 h at 20°C. Cells were washed three times with an excess of PBS and then resuspended in 20 µl of PBS plus 1 µl of ProLong Gold antifade (Life Technologies). Cells were spotted on 0.8% agarose pads and imaged at the Center for Live Cell Imaging at the University of Michigan Medical School, using an Olympus IX70 inverted confocal microscope. Images were processed with Metamorph Software. All X-ray diffraction data for both native and selenomethionine-substituted protein crystals were collected at the Life Science Consortium (LS-CAT) at the Advance Photon Source at Argonne National Laboratory in Argonne, IL. The native protein SGBP-B (residues 34 to 489) was concentrated to an A280 of 12.25 prior to crystallization and mixed with 10 mM XyGO2 (Megazyme, O-XGHDP). Hanging drop vapor diffusion was performed against mother liquor consisting of 1.1 to 1.5 M ammonium sulfate and 30 to 70 mM sodium cacodylate (pH 6.5). To decrease crystal nucleation, 0.3 ml of paraffin oil was overlaid on top of 0.5 ml of mother liquor yielding diffraction-quality crystals within 2 weeks. Selenomethionine-substituted crystals of SGBP-B were generated using the same conditions as the native crystals. Crystals of the truncated SGBP-B (domains CD, residues 230 to 489) were obtained by mixing concentrated protein (A600 of 20.6) with 10 mM XyGO2 for hanging drop vapor diffusion against a solution containing 2 M sodium formate and 0.1 M sodium acetate (pH 4.6). All SGBP-B crystals were flash-frozen prior to data collection by briefly soaking in a solution of 80% mother liquor–20% glycerol plus 10 mM xylogluco-oligosaccharide. Data were processed and scaled using HKL2000 and Scalepack (53). SAD phasing from a selenomethionine-substituted protein crystals was used to determine the structure of SGBP-B. The AutoSol (54) and Autobuild (55) algorithms within the Phenix (56) suite of programs were used to locate and refine the selenium positions and automatically build an initial model of the protein structure, respectively. Successive rounds of manual model building and refinement in Coot (57) and Phenix, respectively, were utilized to build a 2.7-Å model of the selenomethionine-substituted protein, which then was placed in the unit cell of the native data set. Additional rounds of manual model building and refinement were performed to complete the 2.37-Å structure of SGBP-B with XyGO2. The structure of the truncated protein (CD domains, residues 230 to 489) was solved via molecular replacement with Phaser (58) using the CD domains of the full-length protein as a model. The native protein SGBP-A (residues 28 to 546) was concentrated to an A280 of 28.6 and crystallized via hanging drop vapor diffusion from the Morpheus crystal screen (Molecular Dimensions). Crystals formed in well A1 (30 mM MgCl2, 30 mM CaCl2, 20% polyethylene glycol [PEG 500], 10% PEG 20K, 0.1 M imidazole-MES [morpholinethanesulfonic acid], pH 6.5), and were flash-frozen in liquid nitrogen without additional cryoprotectant. The truncated SGBP-A (residues 36 to 546) concentrated to an A280 of 38.2 yielded crystals with 10 mM XyGO2 via hanging drop vapor diffusion against 1.2 to 1.8 M sodium citrate (pH 6.15 to 6.25), and were flash-frozen in a cryoprotectant solution of 80% mother liquor–20% ethylene glycol with the glycan. Data were processed and scaled using HKL2000 and Scalepack (53). The structure of the apo protein was solved via molecular replacement with BALBES (59) using the homologous structure PDB 3JYS, followed by successive rounds of automatic and manual model building with Autobuild and Coot. The structure of SGBP-A with XyGO2 was solved via molecular replacement with Phaser (58) and refined with Phenix (56). X-data collection and refinement statistics are presented in Table 2.
PMC4896748	Ensemble cryo-EM uncovers inchworm-like translocation of a viral IRES through the ribosome	Internal ribosome entry sites (IRESs) mediate cap-independent translation of viral mRNAs. Using electron cryo-microscopy of a single specimen, we present five ribosome structures formed with the Taura syndrome virus IRES and translocase eEF2•GTP bound with sordarin. The structures suggest a trajectory of IRES translocation, required for translation initiation, and provide an unprecedented view of eEF2 dynamics. The IRES rearranges from extended to bent to extended conformations. This inchworm-like movement is coupled with ribosomal inter-subunit rotation and 40S head swivel. eEF2, attached to the 60S subunit, slides along the rotating 40S subunit to enter the A site. Its diphthamide-bearing tip at domain IV separates the tRNA-mRNA-like pseudoknot I (PKI) of the IRES from the decoding center. This unlocks 40S domains, facilitating head swivel and biasing IRES translocation via hitherto-elusive intermediates with PKI captured between the A and P sites. The structures suggest missing links in our understanding of tRNA translocation. DOI: http://dx.doi.org/10.7554/eLife.14874.001Virus propagation relies on the host translational apparatus. To efficiently compete with host mRNAs and engage in translation under stress, some viral mRNAs undergo cap-independent translation. To this end, internal ribosome entry site (IRES) RNAs are employed (reviewed in Deforges et al. (2015), Jackson et al. (2010), Lozano and Martinez-Salas (2015). An IRES is located at the 5’ untranslated region of the viral mRNA, preceding an open reading frame (ORF). To initiate translation, a structured IRES RNA interacts with the 40S subunit or the 80S ribosome, resulting in precise positioning of the downstream start codon in the small 40S subunit. The canonical scenario of cap-dependent and IRES-dependent initiation involves positioning of the AUG start codon and the initiator tRNA in the ribosomal peptidyl-tRNA (P) site, facilitated by interaction with initiation factors (Jackson et al., 2010). Subsequent binding of an elongator aminoacyl-tRNA to the ribosomal A site transitions the initiation complex into the elongation cycle of translation. Upon peptide bond formation, the two tRNAs and their respective mRNA codons translocate from the A and P to P and E (exit) sites, freeing the A site for the next elongator tRNA. An unusual strategy of initiation is used by intergenic-region (IGR) IRESs found in Dicistroviridae arthropod-infecting viruses. These include shrimp-infecting Taura syndrome virus (TSV; Cevallos and Sarnow, 2005; Hatakeyama et al., 2004), and insect viruses Plautia stali intestine virus (PSIV; Nishiyama et al. (2003), Sasaki and Nakashima (1999)) and Cricket paralysis virus (CrPV; Jan et al. (2001), Wilson et al. (2000)). The IGR IRES mRNAs do not contain an AUG start codon. The IGR-IRES-driven initiation does not involve initiator tRNA and initiation factors (Jan et al., 2001; Sasaki and Nakashima, 1999; Wilson et al., 2000). As such, this group of IRESs represents the most streamlined mechanism of eukaryotic translation initiation. A recent demonstration of bacterial translation initiation by an IGR IRES (Colussi et al., 2015) indicates that the IRESs take advantage of conserved structural and dynamic properties of the ribosome. Early electron cryo-microscopy (cryo-EM) studies have found that the CrPV IRES packs in the ribosome intersubunit space (Schuler et al., 2006; Spahn et al., 2004b). Recent cryo-EM structures of ribosome-bound TSV IRES (Koh et al., 2014) and CrPV IRES (Fernandez et al., 2014) revealed that IGR IRESs position the ORF by mimicking a translating ribosome bound with tRNA and mRNA. The ~200-nt IRES RNAs span from the A site beyond the E site. A conserved tRNA-mRNA–like structural element of pseudoknot I (PKI; Costantino et al., 2008) interacts with the decoding center in the A site of the 40S subunit (Fernandez et al., 2014; Koh et al., 2014). The codon-anticodon-like helix of PKI is stabilized by interactions with the universally conserved decoding-center nucleotides G577, A1755 and A1756 (G530, A1492 and A1493 in E. coli 16S ribosomal RNA, or rRNA). The downstream initiation codon—coding for alanine—is placed in the mRNA tunnel, preceding the decoding center. PKI of IGR IRESs therefore mimics an A-site elongator tRNA interacting with an mRNA sense codon, but not a P-site initiator tRNA and the AUG start codon. How this non-canonical initiation complex transitions to the elongation step is not fully understood. For a cognate aminoacyl-tRNA to bind the first viral mRNA codon, PKI has to be translocated from the A site, so that the first codon can be presented in the A site. A cryo-EM structure of the ribosome bound with a CrPV IRES and release factor eRF1 occupying the A site provided insight into the post-translocation state (Muhs et al., 2015). In this structure, PKI is positioned in the P site and the first mRNA codon is located in the A site. How the large IRES RNA translocates within the ribosome, allowing PKI translocation from the A to P site is not known. The structural similarity of PKI and the tRNA anticodon stem loop (ASL) bound to a codon suggests that their mechanisms of translocation are similar to some extent. Translocation of the IRES or tRNA-mRNA requires eukaryotic elongation factor 2 (eEF2) (Jan et al., 2003; Pestova and Hellen, 2003), a structural and functional homolog of the well-studied bacterial EF-G (Czworkowski et al., 1994; Evarsson et al., 1994). Pre-translocation tRNA-bound ribosomes contain a peptidyl- and deacyl-tRNA, both base-paired to mRNA codons in the A and P sites (termed 2tRNA•mRNA complex). Translocation of 2tRNA•mRNA involves two major large-scale ribosome rearrangements (Figure 1—figure supplement 1) (reviewed in Ling and Ermolenko, (2016)). First, studies of bacterial ribosomes showed that a ~10° rotation of the small subunit relative to the large subunit, known as intersubunit rotation, or ratcheting (Frank and Agrawal, 2000), is required for translocation (Horan and Noller, 2007). Intersubunit rotation occurs spontaneously upon peptidyl transfer, and is coupled with formation of hybrid tRNA states (Agirrezabala et al., 2008; Blanchard et al., 2004; Cornish et al., 2008; Ermolenko et al., 2007; Julián et al., 2008; Moazed and Noller, 1989). In the rotated pre-translocation ribosome, the peptidyl-tRNA binds the A site of the small subunit with its ASL and the P site of the large subunit with the CCA 3’ end (A/P hybrid state). Concurrently, the deacyl-tRNA interacts with the P site of the small subunit and the E site of the large subunit (P/E hybrid state). The ribosome can undergo spontaneous, thermally-driven forward-reverse rotation (Cornish et al., 2008) that shifts the two tRNAs between the hybrid and 'classical' states while the anticodon stem loops remain non-translocated. Binding of EF-G next to the A site and reverse rotation of the small subunit results in translocation of both ASLs on the small subunit (Ermolenko and Noller, 2011). EF-G is thought to 'unlock' the pre-translocation ribosome (Savelsbergh et al., 2003; Spirin, 1969), allowing movement of the 2tRNA•mRNA complex, however the structural details of this unlocking are not known. The second large-scale rearrangement involves rotation, or swiveling, of the head of the small subunit relative to the body. The head can rotate by up to ~20° around the axis nearly orthogonal to that of intersubunit rotation, in the absence of tRNA (Schuwirth et al., 2005) or in the presence of a single P/E tRNA and eEF2 (Taylor et al., 2007) or EF-G (Ratje et al., 2010). Förster resonance energy transfer (FRET) data suggest that head swivel of the rotated small subunit facilitates EF-G-mediated movement of 2tRNA•mRNA (Guo and Noller, 2012). Structures of the 70S•EF-G complex bound with two nearly translocated tRNAs (Ramrath et al., 2013; Zhou et al., 2014), exhibit a large 18° to 21° head swivel in a mid-rotated subunit, whereas no head swivel is observed in the fully rotated pre-translocation or in the non-rotated post-translocation 70S•2tRNA•EF-G structures (Brilot et al., 2013; Gao et al., 2009). The structural role of head swivel is not fully understood. The head swivel was proposed to facilitate transition of the tRNA from the P to E site by widening a constriction between these sites on the 30S subunit (Schuwirth et al., 2005). This widening allows the ASL to sample positions between the P and E sites (Ratje et al., 2010). Whether and how the head swivel mediates tRNA transition from the A to P site remains unknown. We sought to address the following questions by structural visualization of 80S•IRES•eEF2 translocation complexes: (1) How does a large IRES RNA move through the restricted intersubunit space, bringing PKI from the A to P site of the small subunit? (2) How does eEF2 mediate IRES translocation? (3) Does IRES translocation involve large rearrangements in the ribosome, similar to tRNA translocation? (4) What, if any, is the mechanistic role of 40S head rotation in IRES translocation? We used cryo-EM to visualize 80S•TSV IRES complexes formed in the presence of eEF2•GTP and the translation inhibitor sordarin, which stabilizes eEF2 on the ribosome. Although the mechanism of sordarin action is not fully understood, the inhibitor does not affect the conformation of eEF2•GDPNP on the ribosome (Taylor et al., 2007), rendering it an excellent tool in translocation studies. Maximum-likelihood classification using FREALIGN (Lyumkis et al., 2013) identified five IRES-eEF2-bound ribosome structures within a single sample (Figures 1 and 2). The structures differ in the positions and conformations of ribosomal subunits (Figures 1b and 2), IRES RNA (Figures 3 and 4) and eEF2 (Figures 5 and 6). This ensemble of structures allowed us to reconstruct a sequence of steps in IRES translocation induced by eEF2.10.7554/eLife.14874.002Figure 1.Cryo-EM structures of the 80S•TSV IRES bound with eEF2•GDP•sordarin.(a) Structures I through V. In all panels, the large ribosomal subunit is shown in cyan; the small subunit in light yellow (head) and wheat-yellow (body); the TSV IRES in red, eEF2 in green. Nucleotides C1274, U1191 of the 40S head and G904 of the platform (C1054, G966 and G693 in E. coli 16S rRNA) are shown in black to denote the A, P and E sites, respectively. Unresolved regions of the IRES in densities for Structures III and V are shown in gray. (b) Schematic representation of the structures shown in panel a, denoting the conformations of the small subunit relative to the large subunit. A, P and E sites are shown as rectangles. All measurements are relative to the non-rotated 80S•2tRNA•mRNA structure (Svidritskiy et al., 2014). The colors are as in panel a.DOI: http://dx.doi.org/10.7554/eLife.14874.00210.7554/eLife.14874.004Figure 1—figure supplement 1.Comparison of 70S•2tRNA•mRNA and 80S•IRES translocation complexes.(a) Structures of bacterial 70S•2tRNA•mRNA translocation complexes, ordered according to the position of the translocating A->P tRNA (orange). The large ribosomal subunit is shown in cyan; the small subunit in light yellow (head) and wheat-yellow (body), elongation factor G (EF-G) is shown in green. Nucleotides C1054, G966 and G693 of 16S rRNA are shown in black to denote the A, P and E sites, respectively. The extents of the 30S subunit rotation and head swivel relative to their positions in the post-translocation structure (Gao et al., 2009) are shown with arrows. References and PDB codes of the structures are shown. (b) Structures of the 80S•IRES complexes in the absence (Koh et al., 2014) and presence of eEF2 (this work). The large ribosomal subunit is shown in cyan; the small subunit in light yellow (head) and wheat-yellow (body); the TSV IRES in red, eEF2 in green. Nucleotides C1274, U1191 of the 40S head and G904 of the platform (corresponding to C1054, G966 and G693 in E. coli 16S rRNA) are shown in black to denote the A, P and E sites, respectively. Unresolved regions of the IRES in densities for Structures III and V are shown in gray. The extents of the 40S subunit rotation and head swivel relative to their positions in the post-translocation structure (Svidritskiy et al., 2014) are shown with arrows.DOI: http://dx.doi.org/10.7554/eLife.14874.00410.7554/eLife.14874.005Figure 1—figure supplement 2.Schematic of cryo-EM refinement and classification procedures.All particles were initially aligned to a single model. 3D classification using a 3D mask around the 40S head, TSV IRES and eEF2, of the 4x binned stack was used to identify particles containing both the IRES and eEF2. Subsequent 3D classification using a 2D mask comprising PKI and domain IV of eEF2 yielded 5 'purified' classes representing Structures I through V. Sub-classification of each class did not yield additional classes, but helped improve density in the PKI region of class III (estimated resolution and percentage of particles in the sub-classified reconstruction are shown in parentheses).DOI: http://dx.doi.org/10.7554/eLife.14874.00510.7554/eLife.14874.006Figure 1—figure supplement 3.Cryo-EM density of Structures I-V.In panels (a-e), the maps are segmented and colored as in Figure 1. The maps in all panels were B-softened by applying a B-factor of 30 Å. (a-e) Cryo-EM map of Structures I, II, III, IV and V. (f-j) Local resolution of unfiltered and unmasked cryo-EM reconstructions, assessed using Blocres from the BSoft package (Cardone et al., 2013), for Structures I, II, III, IV and V. (k-o) Cryo-EM density for the TSV IRES (red model) and eEF2 (green model) in Structures I, II, III, IV and V. (p) Fourier shell correlation (FSC) curves for Structures I-V. The horizontal axis is labeled with spatial frequency Å and with Å. The resolutions stated in the text correspond to an FSC threshold value of 0.143, shown as a dotted line, for the FREALIGN-derived FSC ('Part_FSC').DOI: http://dx.doi.org/10.7554/eLife.14874.006 (a) Structures I through V. In all panels, the large ribosomal subunit is shown in cyan; the small subunit in light yellow (head) and wheat-yellow (body); the TSV IRES in red, eEF2 in green. Nucleotides C1274, U1191 of the 40S head and G904 of the platform (C1054, G966 and G693 in E. coli 16S rRNA) are shown in black to denote the A, P and E sites, respectively. Unresolved regions of the IRES in densities for Structures III and V are shown in gray. (b) Schematic representation of the structures shown in panel a, denoting the conformations of the small subunit relative to the large subunit. A, P and E sites are shown as rectangles. All measurements are relative to the non-rotated 80S•2tRNA•mRNA structure (Svidritskiy et al., 2014). The colors are as in panel a. DOI: http://dx.doi.org/10.7554/eLife.14874.002 10.7554/eLife.14874.004Figure 1—figure supplement 1.Comparison of 70S•2tRNA•mRNA and 80S•IRES translocation complexes.(a) Structures of bacterial 70S•2tRNA•mRNA translocation complexes, ordered according to the position of the translocating A->P tRNA (orange). The large ribosomal subunit is shown in cyan; the small subunit in light yellow (head) and wheat-yellow (body), elongation factor G (EF-G) is shown in green. Nucleotides C1054, G966 and G693 of 16S rRNA are shown in black to denote the A, P and E sites, respectively. The extents of the 30S subunit rotation and head swivel relative to their positions in the post-translocation structure (Gao et al., 2009) are shown with arrows. References and PDB codes of the structures are shown. (b) Structures of the 80S•IRES complexes in the absence (Koh et al., 2014) and presence of eEF2 (this work). The large ribosomal subunit is shown in cyan; the small subunit in light yellow (head) and wheat-yellow (body); the TSV IRES in red, eEF2 in green. Nucleotides C1274, U1191 of the 40S head and G904 of the platform (corresponding to C1054, G966 and G693 in E. coli 16S rRNA) are shown in black to denote the A, P and E sites, respectively. Unresolved regions of the IRES in densities for Structures III and V are shown in gray. The extents of the 40S subunit rotation and head swivel relative to their positions in the post-translocation structure (Svidritskiy et al., 2014) are shown with arrows.DOI: http://dx.doi.org/10.7554/eLife.14874.004 (a) Structures of bacterial 70S•2tRNA•mRNA translocation complexes, ordered according to the position of the translocating A->P tRNA (orange). The large ribosomal subunit is shown in cyan; the small subunit in light yellow (head) and wheat-yellow (body), elongation factor G (EF-G) is shown in green. Nucleotides C1054, G966 and G693 of 16S rRNA are shown in black to denote the A, P and E sites, respectively. The extents of the 30S subunit rotation and head swivel relative to their positions in the post-translocation structure (Gao et al., 2009) are shown with arrows. References and PDB codes of the structures are shown. (b) Structures of the 80S•IRES complexes in the absence (Koh et al., 2014) and presence of eEF2 (this work). The large ribosomal subunit is shown in cyan; the small subunit in light yellow (head) and wheat-yellow (body); the TSV IRES in red, eEF2 in green. Nucleotides C1274, U1191 of the 40S head and G904 of the platform (corresponding to C1054, G966 and G693 in E. coli 16S rRNA) are shown in black to denote the A, P and E sites, respectively. Unresolved regions of the IRES in densities for Structures III and V are shown in gray. The extents of the 40S subunit rotation and head swivel relative to their positions in the post-translocation structure (Svidritskiy et al., 2014) are shown with arrows. DOI: http://dx.doi.org/10.7554/eLife.14874.004 10.7554/eLife.14874.005Figure 1—figure supplement 2.Schematic of cryo-EM refinement and classification procedures.All particles were initially aligned to a single model. 3D classification using a 3D mask around the 40S head, TSV IRES and eEF2, of the 4x binned stack was used to identify particles containing both the IRES and eEF2. Subsequent 3D classification using a 2D mask comprising PKI and domain IV of eEF2 yielded 5 'purified' classes representing Structures I through V. Sub-classification of each class did not yield additional classes, but helped improve density in the PKI region of class III (estimated resolution and percentage of particles in the sub-classified reconstruction are shown in parentheses).DOI: http://dx.doi.org/10.7554/eLife.14874.005 All particles were initially aligned to a single model. 3D classification using a 3D mask around the 40S head, TSV IRES and eEF2, of the 4x binned stack was used to identify particles containing both the IRES and eEF2. Subsequent 3D classification using a 2D mask comprising PKI and domain IV of eEF2 yielded 5 'purified' classes representing Structures I through V. Sub-classification of each class did not yield additional classes, but helped improve density in the PKI region of class III (estimated resolution and percentage of particles in the sub-classified reconstruction are shown in parentheses). DOI: http://dx.doi.org/10.7554/eLife.14874.005 10.7554/eLife.14874.006Figure 1—figure supplement 3.Cryo-EM density of Structures I-V.In panels (a-e), the maps are segmented and colored as in Figure 1. The maps in all panels were B-softened by applying a B-factor of 30 Å. (a-e) Cryo-EM map of Structures I, II, III, IV and V. (f-j) Local resolution of unfiltered and unmasked cryo-EM reconstructions, assessed using Blocres from the BSoft package (Cardone et al., 2013), for Structures I, II, III, IV and V. (k-o) Cryo-EM density for the TSV IRES (red model) and eEF2 (green model) in Structures I, II, III, IV and V. (p) Fourier shell correlation (FSC) curves for Structures I-V. The horizontal axis is labeled with spatial frequency Å and with Å. The resolutions stated in the text correspond to an FSC threshold value of 0.143, shown as a dotted line, for the FREALIGN-derived FSC ('Part_FSC').DOI: http://dx.doi.org/10.7554/eLife.14874.006 In panels (a-e), the maps are segmented and colored as in Figure 1. The maps in all panels were B-softened by applying a B-factor of 30 Å. (a-e) Cryo-EM map of Structures I, II, III, IV and V. (f-j) Local resolution of unfiltered and unmasked cryo-EM reconstructions, assessed using Blocres from the BSoft package (Cardone et al., 2013), for Structures I, II, III, IV and V. (k-o) Cryo-EM density for the TSV IRES (red model) and eEF2 (green model) in Structures I, II, III, IV and V. (p) Fourier shell correlation (FSC) curves for Structures I-V. The horizontal axis is labeled with spatial frequency Å and with Å. The resolutions stated in the text correspond to an FSC threshold value of 0.143, shown as a dotted line, for the FREALIGN-derived FSC ('Part_FSC'). DOI: http://dx.doi.org/10.7554/eLife.14874.006 We used single-particle cryo-EM and maximum-likelihood image classification in FREALIGN to obtain three-dimensional density maps from a single specimen. The translocation complex was formed using S. cerevisiae 80S ribosomes, Taura syndrome virus IRES, and S. cerevisiae eEF2 in the presence of GTP and the eEF2-binding translation inhibitor sordarin. Unsupervised cryo-EM data classification was combined with the use of three-dimensional and two-dimensional masking around the ribosomal A site (Figure 1—figure supplement 2). This approach revealed five 80S•IRES•eEF2•GDP structures at average resolutions of 3.5 to 4.2 Å, sufficient to locate IRES domains and to resolve individual residues in the core regions of the ribosome and eEF2 (Figures 3c,d, and 5f,h; see also Figure 1—figure supplement 2 and Figure 5—figure supplement 2), including the post-translational modification diphthamide 699 (Figure 3c). Our structures represent hitherto uncharacterized translocation complexes of the TSV IRES captured within globally distinct 80S conformations (Figures 1b and 2). We numbered the structures from I to V, according to the position of the tRNA-mRNA-like PKI on the 40S subunit (Figure 2—source data 1). Specifically, PKI is partially withdrawn from the A site in Structure I, and fully translocated to the P site in Structure V (Figure 4; see also Figure 3—figure supplement 1). Thus Structures I to IV represent different positions of PKI between the A and P sites (Figure 2—source data 1), suggesting that these structures describe intermediate states of translocation. Structure V corresponds to the post-translocation state.10.7554/eLife.14874.007Figure 2.Large-scale rearrangements in Structures I through V, coupled with the movement of PKI from the A to P site and eEF2 entry into the A site.(a) Comparison of the 40S-subunit rotational states in Structures I through V, sampling a ~10° range between Structure I (fully rotated) and Structure V (non-rotated). 18S ribosomal RNA is shown and ribosomal proteins are omitted for clarity. The superpositions of Structures I-V were performed by structural alignments of the 25S ribosomal RNAs. (b) Bar graph of the angles characterizing the 40S rotational and 40S head swiveling states in Structures I through V. Measurements for the two 80S•IRES (INIT) structures (Koh et al., 2014) are included for comparison. All measurements are relative to the non-rotated 80S•2tRNA•mRNA structure (Svidritskiy et al., 2014). (c) Comparison of the 40S conformations in Structures I through V shows distinct positions of the head relative to the body of the 40S subunit (head swivel). Conformation of the non-swiveled 40S subunit in the S. cerevisiae 80S ribosome bound with two tRNAs (Svidritskiy et al., 2014) is shown for reference (blue). (d) Comparison of conformations of the L1 and P stalks of the large subunit in Structures I through V with those in the 80S•IRES (Koh et al., 2014) and tRNA-bound 80S (Svidritskiy et al., 2014) structures. Superpositions were performed by structural alignments of 25S ribosomal RNAs. The central protuberance (CP) is labeled. (e) Bar graph of the positions of PKI and domain IV of eEF2 relative to the P site residues of the head (U1191) and body (C1637) in Structures I through V. (f and g) Close-up view of rearrangements in the A and P sites from the initiation state (INIT: PDB ID 3J6Y) to the post-translocation Structure V. The fragment shown within a rectangle in panel f is magnified in panel g. Nucleotides of the 40S body are shown in orange, 40S head in yellow. The superpositions of structures were performed by structural alignments of the 18S ribosomal RNAs excluding the head region (nt 1150–1620).DOI: http://dx.doi.org/10.7554/eLife.14874.00710.7554/eLife.14874.009Figure 2—figure supplement 1.Large-scale rearrangements in Structures I through V, coupled with the movement of PKI from the A to P site and eEF2 entry into the A site.(a) Rotational states of the 40S subunit in the 80S•IRES structure (INIT; PDB 3J6Y; Koh et al., 2014) and in 80S•IRES•eEF2 Structures I, II, III, IV and V (this work). For each structure, the triangle outlines the contours of the 40S body; the lower angle illustrates the extent of intersubunit (body) rotation. The sizes of the arrows correspond to the extent of the head swivel (yellow) and subunit rotation (black). The views were obtained by structural alignment of the 25S rRNAs; the sarcin-ricin loop (SRL) of 25S rRNA is shown in gray for reference. (b) Solvent view (opposite from that shown in (a)) of the 40S subunit in the 80S•IRES structure (INIT; PDB 3J6Y; Koh et al., 2014) and in 80S•IRES•eEF2 Structures I, II, III, IV and V (this work). The structures are colored as in Figure 1.DOI: http://dx.doi.org/10.7554/eLife.14874.009 (a) Comparison of the 40S-subunit rotational states in Structures I through V, sampling a ~10° range between Structure I (fully rotated) and Structure V (non-rotated). 18S ribosomal RNA is shown and ribosomal proteins are omitted for clarity. The superpositions of Structures I-V were performed by structural alignments of the 25S ribosomal RNAs. (b) Bar graph of the angles characterizing the 40S rotational and 40S head swiveling states in Structures I through V. Measurements for the two 80S•IRES (INIT) structures (Koh et al., 2014) are included for comparison. All measurements are relative to the non-rotated 80S•2tRNA•mRNA structure (Svidritskiy et al., 2014). (c) Comparison of the 40S conformations in Structures I through V shows distinct positions of the head relative to the body of the 40S subunit (head swivel). Conformation of the non-swiveled 40S subunit in the S. cerevisiae 80S ribosome bound with two tRNAs (Svidritskiy et al., 2014) is shown for reference (blue). (d) Comparison of conformations of the L1 and P stalks of the large subunit in Structures I through V with those in the 80S•IRES (Koh et al., 2014) and tRNA-bound 80S (Svidritskiy et al., 2014) structures. Superpositions were performed by structural alignments of 25S ribosomal RNAs. The central protuberance (CP) is labeled. (e) Bar graph of the positions of PKI and domain IV of eEF2 relative to the P site residues of the head (U1191) and body (C1637) in Structures I through V. (f and g) Close-up view of rearrangements in the A and P sites from the initiation state (INIT: PDB ID 3J6Y) to the post-translocation Structure V. The fragment shown within a rectangle in panel f is magnified in panel g. Nucleotides of the 40S body are shown in orange, 40S head in yellow. The superpositions of structures were performed by structural alignments of the 18S ribosomal RNAs excluding the head region (nt 1150–1620). DOI: http://dx.doi.org/10.7554/eLife.14874.007 10.7554/eLife.14874.009Figure 2—figure supplement 1.Large-scale rearrangements in Structures I through V, coupled with the movement of PKI from the A to P site and eEF2 entry into the A site.(a) Rotational states of the 40S subunit in the 80S•IRES structure (INIT; PDB 3J6Y; Koh et al., 2014) and in 80S•IRES•eEF2 Structures I, II, III, IV and V (this work). For each structure, the triangle outlines the contours of the 40S body; the lower angle illustrates the extent of intersubunit (body) rotation. The sizes of the arrows correspond to the extent of the head swivel (yellow) and subunit rotation (black). The views were obtained by structural alignment of the 25S rRNAs; the sarcin-ricin loop (SRL) of 25S rRNA is shown in gray for reference. (b) Solvent view (opposite from that shown in (a)) of the 40S subunit in the 80S•IRES structure (INIT; PDB 3J6Y; Koh et al., 2014) and in 80S•IRES•eEF2 Structures I, II, III, IV and V (this work). The structures are colored as in Figure 1.DOI: http://dx.doi.org/10.7554/eLife.14874.009 (a) Rotational states of the 40S subunit in the 80S•IRES structure (INIT; PDB 3J6Y; Koh et al., 2014) and in 80S•IRES•eEF2 Structures I, II, III, IV and V (this work). For each structure, the triangle outlines the contours of the 40S body; the lower angle illustrates the extent of intersubunit (body) rotation. The sizes of the arrows correspond to the extent of the head swivel (yellow) and subunit rotation (black). The views were obtained by structural alignment of the 25S rRNAs; the sarcin-ricin loop (SRL) of 25S rRNA is shown in gray for reference. (b) Solvent view (opposite from that shown in (a)) of the 40S subunit in the 80S•IRES structure (INIT; PDB 3J6Y; Koh et al., 2014) and in 80S•IRES•eEF2 Structures I, II, III, IV and V (this work). The structures are colored as in Figure 1. DOI: http://dx.doi.org/10.7554/eLife.14874.009 Using the post-translocation S. cerevisiae 80S ribosome bound with the P and E site tRNAs as a reference (80S•2tRNA•mRNA), in which both the subunit rotation and the head-body swivel are 0°(Svidritskiy et al., 2014), we found that the ribosome adopts four globally distinct conformations in Structures I through V (Figure 1b; see also Figure 1—figure supplement 1 and Figure 2—source data 1). Structure I comprises the most rotated ribosome conformation (~10°), characteristic of pre-translocation hybrid-tRNA states. From Structure I to V, the body of the small subunit undergoes backward (reverse) rotation (Figure 2b; see also Figure 1—figure supplement 2 and Figure 2—figure supplement 1). Structures II and III are in mid-rotation conformations (~5°). Structure IV adopts a slightly rotated conformation (~1°). Structure V is in a nearly non-rotated conformation (0.5°), very similar to that of post-translocation ribosome-tRNA complexes (Gao et al., 2009; Korostelev et al., 2006; Selmer et al., 2006; Svidritskiy et al., 2014). Thus, intersubunit rotation of ~9° from Structure I to V covers a nearly complete range of relative subunit positions, similar to what was reported for tRNA-bound yeast (Svidritskiy et al., 2014; Taylor et al., 2007), bacterial (Agirrezabala et al., 2008; Fischer et al., 2010; Julián et al., 2008) and mammalian (Budkevich et al., 2011) ribosomes. The pattern of 40S head swivel between the structures is different from that of intersubunit rotation (Figures 2c and d; see also Figure 2—source data 1). As with the intersubunit rotation, the small head swivel (~1°) in the non-rotated Structure V is closest to that in the 80S•2tRNA•mRNA post-translocation ribosome (Svidritskiy et al., 2014). However in the pre-translocation intermediates (from Structure I to IV), the beak of the head domain first turns toward the large subunit and then backs off (Figure 2—figure supplement 1). This movement reflects the forward and reverse swivel. The head samples a mid-swiveled position in Structure I (12°), then a highly-swiveled position in Structures II and III (17°) and a less swiveled position in Structure IV (14°). The maximum head swivel is observed in the mid-rotated complexes II and III, in which PKI transitions from the A to P site, while eEF2 occupies the A site partially. By comparison, the similarly mid-rotated (4°) 80S•TSV IRES initiation complex, in the absence of eEF2 (Koh et al., 2014), adopts a mid-swiveled position (~10°) (Figure 2c). These observations suggest that eEF2 is necessary for inducing or stabilizing the large head swivel of the 40S subunit characteristic for IRES translocation intermediates. In each structure, the TSV IRES adopts a distinct conformation in the intersubunit space of the ribosome (Figures 3 and 4). The IRES (nt 6758–6952) consists of two globular parts (Figure 3a): the 5’-region (domains I and II, nt 6758–6888) and the PKI domain (domain III, nt 6889–6952). We collectively term domains I and II the 5’ domain. The PKI domain comprises PKI and stem loop 3 (SL3), which stacks on top of the stem of PKI (Koh et al., 2014). The GCU triplet immediately following the PKI domain is the first codon of the open reading frame. In the eEF2-free 80S•IRES initiation complex (INIT) (Koh et al., 2014), the bulk of the 5’-domain (nt. 6758–6888) binds near the E site, contacting the ribosome mostly by means of three protruding structural elements: the L1.1 region and stem loops 4 and 5 (SL4 and SL5). In Structures I to IV, these contacts remain as in the initiation complex (Figure 1a). Specifically, the L1.1 region interacts with the L1 stalk of the large subunit, while SL4 and SL5 bind at the side of the 40S head and interact with proteins uS7, uS11 and eS25 (Figure 3—figure supplement 2 and Figure 3—figure supplement 3; ribosomal proteins are termed according to Ban et al., 2014). In Structures I-IV, the minor groove of SL4 (at nt 6840–6846) binds next to an α-helix of uS7, which is rich in positively charged residues (K212, K213, R219 and K222). The tip of SL4 binds in the vicinity of R157 in the β-hairpin of uS7 and of Y58 in uS11. The minor groove of SL5 (at nt 6862–6868) contacts the positively charged region of eS25 (R49, R58 and R68) (Figure 3—figure supplement 4). In Structure V, however, the density for SL5 is missing suggesting that SL5 is mobile, while weak SL4 density suggests that SL4 is shifted along the surface of uS7, ~20 Å away from its initial position (Figure 3—figure supplement 2c). The L1.1 region remains in contact with the L1 stalk (Figure 3—figure supplement 3).10.7554/eLife.14874.010Figure 3.Positions of the IRES relative to eEF2 and elements of the ribosome in Structures I through V.(a) Secondary structure of the TSV IRES. The TSV IRES comprises two domains: the 5' domain (blue) and the PKI domain (red). The open reading frame (gray) is immediately following pseudoknot I (PKI). (b) Three-dimensional structure of the TSV IRES (Structure II). Pseudoknots and stem loops are labeled and colored as in (a). (c) Positions of the IRES and eEF2 on the small subunit in Structures I to V. The initiation-state IRES is shown in gray. The insert shows density for interaction of diphthamide 699 (eEF2; green) with the codon-anticodon-like helix (PKI; red) in Structure V. (d and e) Density of the P site in Structure V shows that interactions of PKI with the 18S rRNA nucleotides (c) are nearly identical to those in the P site of the 2tRNA•mRNA-bound 70S ribosome (d; Svidritskiy et al., 2013).DOI: http://dx.doi.org/10.7554/eLife.14874.01010.7554/eLife.14874.011Figure 3—figure supplement 1.Comparison of the TSV IRES and eEF2 positions in Structures I through V.(a) Positions of the IRES and eEF2 in the initiation, pre-translocation (I) and post-translocation (V) states, relative to the body of the 40S subunit (not shown) (b) Positions of the IRES and eEF2 in the initiation state (INIT) and intermediate steps of translocation (II, III and IV), relative to the body of the 40S subunit (not shown). Superpositions were obtained by structural alignments of the 18S rRNAs excluding the head domains (nt 1150–1620).DOI: http://dx.doi.org/10.7554/eLife.14874.01110.7554/eLife.14874.012Figure 3—figure supplement 2.Positions of the IRES relative to proteins uS7, uS11 and eS25.(a) Intra-IRES rearrangements from the 80SIRES initiation structure (INIT; PDB 3J6Y, Koh et al., 2014) to Structures I through V. For each structure (shown in red), the conformation from a preceding structure is shown in light red for comparison. Superpositions were obtained by structural alignments of 18S rRNA. (b) Positions of the IRES and eEF2 relative to those of classical P- and E-site tRNAs in the 80S•tRNA complex (Svidritskiy et al, 2014). (c) Positions of the IRES relative to proteins uS11 (40S platform) and uS7 and eS25 (40S head), which interact with the 5′ domain of the IRES in the initiation state (left panel). In all panels, superpositions were obtained by structural alignments of the 18S rRNAs. Ribosomal proteins of the initiation state are shown in gray for comparison.DOI: http://dx.doi.org/10.7554/eLife.14874.01210.7554/eLife.14874.013Figure 3—figure supplement 3.Positions of the L1stalk, tRNA and TSV IRES relative to proteins uS7 and eS25, in 80S•tRNA structures (Svidritskiy et al., 2014) and 80S•IRES structures I and V (this work).The view shows the vicinity of the ribosomal E site. Loop 1.1 and stem loops 4 and 5 of the IRES are labeled.DOI: http://dx.doi.org/10.7554/eLife.14874.01310.7554/eLife.14874.014Figure 3—figure supplement 4.Interactions of the stem loops 4 and 5 of the TSV with proteins uS7 and eS25.DOI: http://dx.doi.org/10.7554/eLife.14874.01410.7554/eLife.14874.015Figure 3—figure supplement 5.Position and interactions of loop 3 (variable loop region) of the PKI domain in Structure V (this work) resembles those of the anticodon stem loop of the E-site tRNA (blue) in the 80S•2tRNA•mRNA complex (Svidritskiy et al., 2014).DOI: http://dx.doi.org/10.7554/eLife.14874.01510.7554/eLife.14874.016Figure 3—figure supplement 6.Positions of tRNAs and the TSV IRES relative to the A-site finger (blue, nt 1008–1043 of 25S rRNA) and the P site of the large subunit, comprising helix 84 of 25S rRNA (nt. 2668–2687) and protein uL5 (collectively labeled as central protuberance, CP, in the upper-row first figure, and individually labeled in the lower-row first figure).Structures of translocation complexes of the bacterial 70S ribosome bound with two tRNAs and yeast 80S complexes with tRNAs are shown in the upper row and labeled. Structures of 80S•IRES complexes in the absence of eEF2 (INIT; PDB 3J6Y, Koh et al., 2014) and in the presence of eEF2 (this work) are shown in the lower row and labeled.DOI: http://dx.doi.org/10.7554/eLife.14874.01610.7554/eLife.14874.017Figure 3—figure supplement 7.Interactions of the TSV IRES with uL5 and eL42.Structures of 80S•IRES complexes in the absence of eEF2 (INIT; PDB 3J6Y, Koh et al., 2014) and in the presence of eEF2 (this work) are shown in the upper row and labeled. Structures of the 80S complexes with tRNAs (Svidritskiy et al., 2014) are shown in the lower row in a view similar to that for the 80S•IRES complex.DOI: http://dx.doi.org/10.7554/eLife.14874.017 (a) Secondary structure of the TSV IRES. The TSV IRES comprises two domains: the 5' domain (blue) and the PKI domain (red). The open reading frame (gray) is immediately following pseudoknot I (PKI). (b) Three-dimensional structure of the TSV IRES (Structure II). Pseudoknots and stem loops are labeled and colored as in (a). (c) Positions of the IRES and eEF2 on the small subunit in Structures I to V. The initiation-state IRES is shown in gray. The insert shows density for interaction of diphthamide 699 (eEF2; green) with the codon-anticodon-like helix (PKI; red) in Structure V. (d and e) Density of the P site in Structure V shows that interactions of PKI with the 18S rRNA nucleotides (c) are nearly identical to those in the P site of the 2tRNA•mRNA-bound 70S ribosome (d; Svidritskiy et al., 2013). DOI: http://dx.doi.org/10.7554/eLife.14874.010 10.7554/eLife.14874.011Figure 3—figure supplement 1.Comparison of the TSV IRES and eEF2 positions in Structures I through V.(a) Positions of the IRES and eEF2 in the initiation, pre-translocation (I) and post-translocation (V) states, relative to the body of the 40S subunit (not shown) (b) Positions of the IRES and eEF2 in the initiation state (INIT) and intermediate steps of translocation (II, III and IV), relative to the body of the 40S subunit (not shown). Superpositions were obtained by structural alignments of the 18S rRNAs excluding the head domains (nt 1150–1620).DOI: http://dx.doi.org/10.7554/eLife.14874.011 (a) Positions of the IRES and eEF2 in the initiation, pre-translocation (I) and post-translocation (V) states, relative to the body of the 40S subunit (not shown) (b) Positions of the IRES and eEF2 in the initiation state (INIT) and intermediate steps of translocation (II, III and IV), relative to the body of the 40S subunit (not shown). Superpositions were obtained by structural alignments of the 18S rRNAs excluding the head domains (nt 1150–1620). DOI: http://dx.doi.org/10.7554/eLife.14874.011 10.7554/eLife.14874.012Figure 3—figure supplement 2.Positions of the IRES relative to proteins uS7, uS11 and eS25.(a) Intra-IRES rearrangements from the 80SIRES initiation structure (INIT; PDB 3J6Y, Koh et al., 2014) to Structures I through V. For each structure (shown in red), the conformation from a preceding structure is shown in light red for comparison. Superpositions were obtained by structural alignments of 18S rRNA. (b) Positions of the IRES and eEF2 relative to those of classical P- and E-site tRNAs in the 80S•tRNA complex (Svidritskiy et al, 2014). (c) Positions of the IRES relative to proteins uS11 (40S platform) and uS7 and eS25 (40S head), which interact with the 5′ domain of the IRES in the initiation state (left panel). In all panels, superpositions were obtained by structural alignments of the 18S rRNAs. Ribosomal proteins of the initiation state are shown in gray for comparison.DOI: http://dx.doi.org/10.7554/eLife.14874.012 (a) Intra-IRES rearrangements from the 80SIRES initiation structure (INIT; PDB 3J6Y, Koh et al., 2014) to Structures I through V. For each structure (shown in red), the conformation from a preceding structure is shown in light red for comparison. Superpositions were obtained by structural alignments of 18S rRNA. (b) Positions of the IRES and eEF2 relative to those of classical P- and E-site tRNAs in the 80S•tRNA complex (Svidritskiy et al, 2014). (c) Positions of the IRES relative to proteins uS11 (40S platform) and uS7 and eS25 (40S head), which interact with the 5′ domain of the IRES in the initiation state (left panel). In all panels, superpositions were obtained by structural alignments of the 18S rRNAs. Ribosomal proteins of the initiation state are shown in gray for comparison. DOI: http://dx.doi.org/10.7554/eLife.14874.012 10.7554/eLife.14874.013Figure 3—figure supplement 3.Positions of the L1stalk, tRNA and TSV IRES relative to proteins uS7 and eS25, in 80S•tRNA structures (Svidritskiy et al., 2014) and 80S•IRES structures I and V (this work).The view shows the vicinity of the ribosomal E site. Loop 1.1 and stem loops 4 and 5 of the IRES are labeled.DOI: http://dx.doi.org/10.7554/eLife.14874.013 The view shows the vicinity of the ribosomal E site. Loop 1.1 and stem loops 4 and 5 of the IRES are labeled. DOI: http://dx.doi.org/10.7554/eLife.14874.013 10.7554/eLife.14874.014Figure 3—figure supplement 4.Interactions of the stem loops 4 and 5 of the TSV with proteins uS7 and eS25.DOI: http://dx.doi.org/10.7554/eLife.14874.014 DOI: http://dx.doi.org/10.7554/eLife.14874.014 10.7554/eLife.14874.015Figure 3—figure supplement 5.Position and interactions of loop 3 (variable loop region) of the PKI domain in Structure V (this work) resembles those of the anticodon stem loop of the E-site tRNA (blue) in the 80S•2tRNA•mRNA complex (Svidritskiy et al., 2014).DOI: http://dx.doi.org/10.7554/eLife.14874.015 DOI: http://dx.doi.org/10.7554/eLife.14874.015 10.7554/eLife.14874.016Figure 3—figure supplement 6.Positions of tRNAs and the TSV IRES relative to the A-site finger (blue, nt 1008–1043 of 25S rRNA) and the P site of the large subunit, comprising helix 84 of 25S rRNA (nt. 2668–2687) and protein uL5 (collectively labeled as central protuberance, CP, in the upper-row first figure, and individually labeled in the lower-row first figure).Structures of translocation complexes of the bacterial 70S ribosome bound with two tRNAs and yeast 80S complexes with tRNAs are shown in the upper row and labeled. Structures of 80S•IRES complexes in the absence of eEF2 (INIT; PDB 3J6Y, Koh et al., 2014) and in the presence of eEF2 (this work) are shown in the lower row and labeled.DOI: http://dx.doi.org/10.7554/eLife.14874.016 Structures of translocation complexes of the bacterial 70S ribosome bound with two tRNAs and yeast 80S complexes with tRNAs are shown in the upper row and labeled. Structures of 80S•IRES complexes in the absence of eEF2 (INIT; PDB 3J6Y, Koh et al., 2014) and in the presence of eEF2 (this work) are shown in the lower row and labeled. DOI: http://dx.doi.org/10.7554/eLife.14874.016 10.7554/eLife.14874.017Figure 3—figure supplement 7.Interactions of the TSV IRES with uL5 and eL42.Structures of 80S•IRES complexes in the absence of eEF2 (INIT; PDB 3J6Y, Koh et al., 2014) and in the presence of eEF2 (this work) are shown in the upper row and labeled. Structures of the 80S complexes with tRNAs (Svidritskiy et al., 2014) are shown in the lower row in a view similar to that for the 80S•IRES complex.DOI: http://dx.doi.org/10.7554/eLife.14874.017 Structures of 80S•IRES complexes in the absence of eEF2 (INIT; PDB 3J6Y, Koh et al., 2014) and in the presence of eEF2 (this work) are shown in the upper row and labeled. Structures of the 80S complexes with tRNAs (Svidritskiy et al., 2014) are shown in the lower row in a view similar to that for the 80S•IRES complex. DOI: http://dx.doi.org/10.7554/eLife.14874.017 The shape of the IRES changes considerably from the initiation state to Structures I through V, from an extended to compact to extended conformation (Figure 4; see also Figure 3—figure supplement 2a). Because in Structures I to IV the PKI domain shifts toward the P site, while the 5’ remains unchanged near the E site, the distance between the domains shortens (Figure 4). In the 80S•IRES initiation state (Koh et al., 2014), the A-site-bound PKI is separated from SL4 by almost 50 Å (Figure 4). In Structures I and II, the PKI is partially retracted from the A site and the distance from SL4 shortens to ~35 Å. As PKI moves toward the P site in Structures III and IV, the PKI domain approaches to within ~25 Å of SL4. Because the 5’-domain in the following structure (V) moves by ~20 Å along the 40S head, the IRES returns to an extended conformation (~45 Å) that is similar to that in the 80S•IRES initiation complex.10.7554/eLife.14874.018Figure 4.Inchworm-like translocation of the TSV IRES.Conformations and positions of the IRES in the initiation state and in Structures I-V are shown relative to those of the A-, P- and E-site tRNAs. The view was obtained by structural alignment of the body domains of 18S rRNAs of the corresponding 80S structures. Distances between nucleotides 6848 and 6913 in SL4 and PKI, respectively, are shown (see also Figure 2—source data 1).DOI: http://dx.doi.org/10.7554/eLife.14874.018 Conformations and positions of the IRES in the initiation state and in Structures I-V are shown relative to those of the A-, P- and E-site tRNAs. The view was obtained by structural alignment of the body domains of 18S rRNAs of the corresponding 80S structures. Distances between nucleotides 6848 and 6913 in SL4 and PKI, respectively, are shown (see also Figure 2—source data 1). DOI: http://dx.doi.org/10.7554/eLife.14874.018 Rearrangements of the IRES involve restructuring of several interactions with the ribosome. In Structure I, SL3 of the PKI domain is positioned between the A-site finger (nt 1008–1043 of 25S rRNA) and the P site of the 60S subunit, comprising helix 84 of 25S rRNA (nt. 2668–2687) and protein uL5 (Figure 3—figure supplement 6). This position of SL3 is ~25 Å away from that in the 80S•IRES initiation state (Koh et al., 2014), in which PKI and SL3 closely mimic the ASL and elbow of the A-site tRNA, respectively (Koh et al., 2014). As such, the transition from the initiation state to Structure I involves repositioning of SL3 around the A-site finger, resembling the transition between the pre-translocation A/P and A/P tRNA (Brilot et al., 2013). The second set of major structural changes involves interaction of the P site region of the large subunit with the hinge point of the IRES bending between the 5´ domain and the PKI domain (nt. 6886–6890). In the highly bent Structures III and IV, the hinge region interacts with the universally conserved uL5 and the C-terminal tail of eL42 (Figure 3—figure supplement 7). However, in the extended conformations, these parts of the IRES and the 60S subunit are separated by more than 10 Å, suggesting that an interaction between them stabilizes the bent conformations but not the extended ones. Another local rearrangement concerns loop 3, also known as the variable loop region (Ruehle et al., 2015; Ren et al., 2014; Au and Jan, 2012), which connects the ASL- and mRNA-like parts of PKI. This loop is poorly resolved in Structures I through IV, suggesting conformational flexibility in agreement with structural studies of the isolated PKI (Costantino et al., 2008; Zhu et al., 2011) and biochemical studies of unbound IRESs (Jan and Sarnow, 2002; Pfingsten et al., 2010). In Structure V, loop 3 is bound in the 40S E site and the backbone of loop 3 near the codon-like part of PKI (at nt. 6945–6946) interacts with R148 and R157 in β-hairpin of uS7. The interaction of loop 3 backbone with uS7 resembles that of the anticodon-stem loop of E-site tRNA in the post-translocation 80S•2tRNA•mRNA structure (Figure 3—figure supplement 5) (Svidritskiy et al., 2014). Ordering of loop 3 suggests that this flexible region contributes to the stabilization of the PKI domain in the post-translocation state. This interpretation is consistent with the recent observation that alterations in loop 3 of the CrPV IRES result in decreased efficiency of translocation (Ruehle et al., 2015). Elongation factor eEF2 in all five structures is bound with GDP and sordarin (Figure 5). The elongation factor consists of three dynamic superdomains (Jorgensen et al., 2003): an N-terminal globular superdomain formed by the G (GTPase) domain (domain I) and domain II; a linker domain III; and a C-terminal superdomain comprising domains IV and V (Figure 5a). Domain IV extends from the main body and is critical for translocation catalyzed by eEF2 or EF-G. ADP-ribosylation of eEF2 at the tip of domain IV (Davydova and Ovchinnikov, 1990; Nygard and Nilsson, 1990) or deletion of domain IV from EF-G (Martemyanov and Gudkov, 1999; Rodnina et al., 1997) abrogate translocation. In post-translocation-like 80S•tRNA•eEF2 complexes, domain IV binds in the 40S A site, suggesting direct involvement of domain IV in translocation of tRNA from the A to P site (Spahn et al., 2004a; Taylor et al., 2007). GDP in our structures is bound in the GTPase center (Figures 5d, e and f) and sordarin is sandwiched between the β-platforms of domains III and V (Figures 5g and h), as in the structure of free eEF2•sordarin complex (Jorgensen et al., 2003).10.7554/eLife.14874.019Figure 5.Conformations and interactions of eEF2.(a) Conformations of eEF2 in Structures I-V and domain organization of eEF2 are shown. Roman numerals denote eEF2 domains. Superposition was obtained by structural alignment of domains I and II. (b) Elements of the 80S ribosome in Structures I and V that contact eEF2. eEF2 is shown in green, IRES RNA in red, 40S subunit elements in orange, 60S in cyan/teal. (c) Comparison of conformations of eEF2•sordarin in Structure I (light green) with those of free apo-eEF2 (magenta) and eEF2•sordarin (teal) (Jorgensen et al., 2003). (d) Interactions of the GTPase domains with the 40S and 60S subunits in Structure I (colored in green/blue, eEF2; orange, 40S; cyan/teal, 60S) and in Structure II (gray). Switch loop I (SWI) in Structure I is in blue; dashed line shows the putative location of unresolved switch loop I in Structure II. Superposition was obtained by structural alignment of the 25S rRNAs. (e) Comparison of the GTP-like conformation of eEF2•GDP in Structure I (light green) with those of 70S-bound elongation factors EF-Tu•GDPCP (teal; Voorhees et al. 2010) and EF-G•GDP•fusidic acid (magenta; fusidic acid not shown; Zhou et al., 2013). (f) Cryo-EM density showing guanosine diphosphate bound in the GTPase center (green) next to the sarcin-ricin loop of 25S rRNA (cyan) of Structure II. (g) Comparison of the sordarin-binding sites in the ribosome-bound (light green; Structure II) and isolated eEF2 (teal; Jorgensen et al., 2003). (h) Cryo-EM density showing the sordarin-binding pocket of eEF2 (Structure II). Sordarin is shown in pink with oxygen atoms in red.DOI: http://dx.doi.org/10.7554/eLife.14874.01910.7554/eLife.14874.020Figure 5—figure supplement 1.Elements of the 80S ribosome that contact eEF2 in Structures I through V.The view and colors are as in Figure 5b: eEF2 is shown in green, IRES RNA in red, 40S subunit elements in orange, 60S in cyan/teal.DOI: http://dx.doi.org/10.7554/eLife.14874.02010.7554/eLife.14874.021Figure 5—figure supplement 2.Cryo-EM density of the GTPase region in Structures I and II.The switch loop I in Structure I is shown in blue. The putative position of the switch loop I, unresolved in the density of Structure II, is shown with a dashed line. Colors for the ribosome and eEF2 are as in Figure 1.DOI: http://dx.doi.org/10.7554/eLife.14874.021 (a) Conformations of eEF2 in Structures I-V and domain organization of eEF2 are shown. Roman numerals denote eEF2 domains. Superposition was obtained by structural alignment of domains I and II. (b) Elements of the 80S ribosome in Structures I and V that contact eEF2. eEF2 is shown in green, IRES RNA in red, 40S subunit elements in orange, 60S in cyan/teal. (c) Comparison of conformations of eEF2•sordarin in Structure I (light green) with those of free apo-eEF2 (magenta) and eEF2•sordarin (teal) (Jorgensen et al., 2003). (d) Interactions of the GTPase domains with the 40S and 60S subunits in Structure I (colored in green/blue, eEF2; orange, 40S; cyan/teal, 60S) and in Structure II (gray). Switch loop I (SWI) in Structure I is in blue; dashed line shows the putative location of unresolved switch loop I in Structure II. Superposition was obtained by structural alignment of the 25S rRNAs. (e) Comparison of the GTP-like conformation of eEF2•GDP in Structure I (light green) with those of 70S-bound elongation factors EF-Tu•GDPCP (teal; Voorhees et al. 2010) and EF-G•GDP•fusidic acid (magenta; fusidic acid not shown; Zhou et al., 2013). (f) Cryo-EM density showing guanosine diphosphate bound in the GTPase center (green) next to the sarcin-ricin loop of 25S rRNA (cyan) of Structure II. (g) Comparison of the sordarin-binding sites in the ribosome-bound (light green; Structure II) and isolated eEF2 (teal; Jorgensen et al., 2003). (h) Cryo-EM density showing the sordarin-binding pocket of eEF2 (Structure II). Sordarin is shown in pink with oxygen atoms in red. DOI: http://dx.doi.org/10.7554/eLife.14874.019 10.7554/eLife.14874.020Figure 5—figure supplement 1.Elements of the 80S ribosome that contact eEF2 in Structures I through V.The view and colors are as in Figure 5b: eEF2 is shown in green, IRES RNA in red, 40S subunit elements in orange, 60S in cyan/teal.DOI: http://dx.doi.org/10.7554/eLife.14874.020 The view and colors are as in Figure 5b: eEF2 is shown in green, IRES RNA in red, 40S subunit elements in orange, 60S in cyan/teal. DOI: http://dx.doi.org/10.7554/eLife.14874.020 10.7554/eLife.14874.021Figure 5—figure supplement 2.Cryo-EM density of the GTPase region in Structures I and II.The switch loop I in Structure I is shown in blue. The putative position of the switch loop I, unresolved in the density of Structure II, is shown with a dashed line. Colors for the ribosome and eEF2 are as in Figure 1.DOI: http://dx.doi.org/10.7554/eLife.14874.021 The switch loop I in Structure I is shown in blue. The putative position of the switch loop I, unresolved in the density of Structure II, is shown with a dashed line. Colors for the ribosome and eEF2 are as in Figure 1. DOI: http://dx.doi.org/10.7554/eLife.14874.021 The global conformations of eEF2 (Figure 5a) are similar in these structures (all-atom RMSD ≤ 2 Å), but the positions of eEF2 relative to the 40S subunit differ substantially as a result of 40S subunit rotation (Figure 2—source data 1). From Structure I to V, eEF2 is rigidly attached to the GTPase-associated center of the 60S subunit. The GTPase-associated center comprises the P stalk (L11 and L7/L12 stalk in bacteria) and the sarcin-ricin loop (SRL, nt 3012–3042). The tips of 25S rRNA helices 43 and 44 of the P stalk (nucleotides G1242 and A1270, respectively) stack on V754 and Y744 of domain V. An αββ motif of the eukaryote-specific protein P0 (aa 126–154) packs in the crevice between the long α-helix D (aa 172–188) of the GTPase domain and the β-sheet region (aa 246–263) of the GTPase domain insert (or G’ insert) of eEF2 (secondary-structure nomenclatures for eEF2 and EF-G (Czworkowski et al., 1994) are the same). Although the P/L11 stalk is known to be dynamic (Korostelev et al., 2008; Taylor et al., 2012), its position remains unchanged from Structure I to V: all-atom root-mean-square differences for the 25S rRNA of the P stalk (nt 1223–1286) are within 2.5 Å. However, with respect to its position in the 80S•IRES complex in the absence of eEF2 and in the 80S•2tRNA•mRNA complex, the P stalk is shifted by ~13 Å toward the A site (Figure 2d). The sarcin-ricin loop interacts with the GTP-binding site of eEF2 (Figures 5d and f). While the overall mode of this interaction is similar to that seen in 70S•EF-G crystal structures (Chen et al., 2013b; Gao et al., 2009; Pulk and Cate, 2013; Tourigny et al., 2013; Zhou et al., 2013; 2014), there is an important local difference between Structure I and Structures II-V in switch loop I, as discussed below. There are two modest but noticeable domain rearrangements between Structures I and V. Unlike in free eEF2, which can sample large movements of at least 50 Å of the C-terminal superdomain relative to the N-terminal superdomain (Figure 5c) (Jorgensen et al., 2003), eEF2 undergoes moderate repositioning of domain IV (~3 Å; Figure 5a) and domain III (~5 Å; Figure 6d). This limited flexibility of the ribosome-bound eEF2 is likely the result of simultaneous fixation of eEF2 superdomains, via domains I and V, by the GTPase-associated center of the large subunit. Domain IV of eEF2 binds at the 40S A site in Structures I to V but the mode of interaction differs in each complex (Figure 6). Because eEF2 is rigidly attached to the 60S subunit and does not undergo large inter-subunit rearrangements, gradual entry of domain IV into the A site between Structures I and V is due to 40S subunit rotation and head swivel. eEF2 settles into the A site from Structure I to V, as the tip of domain IV shifts by ~10 Å relative to the body and by ~20 Å relative to the swiveling head. Modest intra-eEF2 shifts of domain IV between Structures I to V outline a stochastic trajectory (Figure 5a), consistent with local adjustments of the domain in the A site. At the central region of eEF2, domains II and III contact the 40S body (mainly at nucleotides 48–52 and 429–432 of 18S rRNA helix 5 and uS12). From Structure I to V, these central domains migrate by ~10 Å along the 40S surface (Figure 6c). Comparison of eEF2 conformations reveals that in Structure V, domain III is displaced as a result of interaction with uS12, as discussed below.10.7554/eLife.14874.022Figure 6.Interactions of eEF2 with the 40S subunit.(a) eEF2 (green) interacts only with the body in Structure I (eEF2 domains are labeled with roman numerals in white), and with both the head and body in Structures II through V. Colors are as in Figure 1, except for the 40S structural elements that contact eEF2, which are labeled and shown in purple. (b) Entry of eEF2 into the 40S A site, from Structure I through V. Distances to the A-site accommodated eEF2 (Structure V) are shown. The view was obtained by superpositions of the body domains of 18S rRNAs. (c) Rearrangements, from Structure I through V, of a positively charged cluster of eEF2 (K613, R617 and R631) positioned over the phosphate backbone of 18S helices 33 and 34, suggesting a role of electrostatic interactions in eEF2 diffusion over the 40S surface. (d) Shift of the tip of domain III of eEF2, interacting with uS12 upon reverse subunit rotation from Structure I to Structure V. Structure I colored as in Figure 1, except uS12, which is in purple; Structure V is in gray.DOI: http://dx.doi.org/10.7554/eLife.14874.02210.7554/eLife.14874.023Figure 6—figure supplement 1.Repositioning (sliding) of the positively-charged cluster of domain IV of eEF2 over the phosphate backbone (red) of the 18S helices 33 and 34.Structures I through V are shown. Electrostatic surface of eEF2 is shown; negatively and positively charged regions are shown in red and blue, respectively. The view was obtained by structural alignment of the 18S rRNAs.DOI: http://dx.doi.org/10.7554/eLife.14874.023 (a) eEF2 (green) interacts only with the body in Structure I (eEF2 domains are labeled with roman numerals in white), and with both the head and body in Structures II through V. Colors are as in Figure 1, except for the 40S structural elements that contact eEF2, which are labeled and shown in purple. (b) Entry of eEF2 into the 40S A site, from Structure I through V. Distances to the A-site accommodated eEF2 (Structure V) are shown. The view was obtained by superpositions of the body domains of 18S rRNAs. (c) Rearrangements, from Structure I through V, of a positively charged cluster of eEF2 (K613, R617 and R631) positioned over the phosphate backbone of 18S helices 33 and 34, suggesting a role of electrostatic interactions in eEF2 diffusion over the 40S surface. (d) Shift of the tip of domain III of eEF2, interacting with uS12 upon reverse subunit rotation from Structure I to Structure V. Structure I colored as in Figure 1, except uS12, which is in purple; Structure V is in gray. DOI: http://dx.doi.org/10.7554/eLife.14874.022 10.7554/eLife.14874.023Figure 6—figure supplement 1.Repositioning (sliding) of the positively-charged cluster of domain IV of eEF2 over the phosphate backbone (red) of the 18S helices 33 and 34.Structures I through V are shown. Electrostatic surface of eEF2 is shown; negatively and positively charged regions are shown in red and blue, respectively. The view was obtained by structural alignment of the 18S rRNAs.DOI: http://dx.doi.org/10.7554/eLife.14874.023 Structures I through V are shown. Electrostatic surface of eEF2 is shown; negatively and positively charged regions are shown in red and blue, respectively. The view was obtained by structural alignment of the 18S rRNAs. DOI: http://dx.doi.org/10.7554/eLife.14874.023 In summary, between Structures I and V, a step-wise translocation of PKI by ~15 Å from the A to P site - within the 40S subunit – occurs simultaneously with the ~11 Å side-way entry of domain IV into the A site coupled with ~3 to 5 Å inter-domain rearrangements in eEF2. These shifts occur during the reverse rotation of the 40S body coupled with the forward-then-reverse head swivel. To elucidate the detailed structural mechanism of IRES translocation and the roles of eEF2 and ribosome rearrangements, we describe in the following sections the interactions of PKI and eEF2 with the ribosomal A and P sites in Structures I through V (Figure 2g; see also Figure 1—figure supplement 1). In the fully rotated Structure I, PKI is shifted toward the P site by ~3 Å relative to its position in the initiation complex but maintains interactions with the partially swiveled head. At the head, C1274 of the 18S rRNA (C1054 in E. coli) base pairs with the first nucleotide of the ORF immediately downstream of PKI. The C1274:G6953 base pair provides a stacking platform for the codon-anticodon–like helix of PKI. We therefore define C1274 as the foundation of the 'head A site'. Accordingly, we use U1191 (G966 in E. coli) and C1637 (C1400 in E. coli) as the reference points of the 'head P site' and 'body P site' (Figure 2g), respectively, because these nucleotides form a stacking foundation for the fully translocated mRNA-tRNA helix in tRNA-bound structures (Korostelev et al., 2006; Selmer et al., 2006; Svidritskiy et al., 2014) and in our post-translocation Structure V discussed below. The interaction of PKI with the 40S body is substantially rearranged relative to that in the initiation state. In the latter, PKI is stabilized by interactions with the universally conserved decoding-center nucleotides G577, A1755 and A1756 ('body A site'), as in the A-site tRNA bound complexes (Koh et al., 2014). In Structure I, PKI does not contact these nucleotides (Figures 2g and 7).10.7554/eLife.14874.024Figure 7.Interactions of the residues at the eEF2 tip with the decoding center of the IRES-bound ribosome.Key elements of the decoding center of the 'locked' initiation structure (Koh et al., 2014), 'unlocked' Structure I, and post-translocation Structure V (this work) are shown. The histidine-diphthamide tip of eEF2 is shown in green. The codon-anticodon-like helix of PKI is shown in red, the downstream first codon of the ORF in magenta. Nucleotides of the 18S rRNA body are in orange and head in yellow; 25S rRNA nucleotide A2256 is blue. A and P sites are schematically demarcated by dotted lines.DOI: http://dx.doi.org/10.7554/eLife.14874.024 Key elements of the decoding center of the 'locked' initiation structure (Koh et al., 2014), 'unlocked' Structure I, and post-translocation Structure V (this work) are shown. The histidine-diphthamide tip of eEF2 is shown in green. The codon-anticodon-like helix of PKI is shown in red, the downstream first codon of the ORF in magenta. Nucleotides of the 18S rRNA body are in orange and head in yellow; 25S rRNA nucleotide A2256 is blue. A and P sites are schematically demarcated by dotted lines. DOI: http://dx.doi.org/10.7554/eLife.14874.024 The position of eEF2 on the 40S subunit of Structure I is markedly distinct from those in Structures II to V. The translocase interacts with the 40S body but does not contact the head (Figures 5b and 6a; Figure 5—figure supplement 1). Domain IV is partially engaged with the body A site. The tip of domain IV is wedged between PKI and decoding-center nucleotides A1755 and A1756, which are bulged out of h44. This tip contains the histidine-diphthamide triad (H583, H694 and Diph699), which interacts with the codon-anticodon-like helix of PKI and A1756 (Figure 7). Histidines 583 and 694 interact with the phosphate backbone of the anticodon-like strand (at G6907 and C6908). Diphthamide is a unique posttranslational modification conserved in archaeal and eukaryotic EF2 (at residue 699 in S. cerevisiae) and involves addition of a ~7-Å long 3-carboxyamido-3-(trimethylamino)-propyl moiety to the histidine imidazole ring at CE1. The trimethylamino end of Diph699 packs over A1756 (Figure 7). The opposite surface of the tail is oriented toward the minor-groove side of the second base pair of the codon-anticodon helix (G6906:C6951). Thus, in comparison with the initiation state, the histidine-diphthamide tip of eEF2 replaces the codon-anticodon–like helix of PKI. The splitting of the interaction of A1755-A1756 and PKI is achieved by providing the histidine-diphthamine tip as a binding partner for both A1756 and the minor groove of the codon-anticodon helix (Figure 7). Unlike in Structures II to V, the conformation of the eEF2 GTPase center in Structure I resembles that of a GTP-bound translocase (Figure 5e). In translational GTPases, switch loops I and II are involved in the GTPase activity (reviewed in Voorhees and Ramakrishnan, (2013)). Switch loop II (aa 105–110), which carries the catalytic H108 (H92 in E. coli EF-G; (Cunha et al., 2013; Holtkamp et al., 2014; Koripella et al., 2015; Salsi et al., 2014) is well resolved in all five structures. The histidine resides next to the backbone of G3028 of the sarcin-ricin loop and near the diphosphate of GDP (Figure 5e). By contrast, switch loop I (aa 50–70 in S. cerevisiae eEF2) is resolved only in Structure I (Figure 5—figure supplement 2). The N-terminal part of the loop (aa 50–60) is sandwiched between the tip of helix 14 (CAAA) of the 18S rRNA of the 40S subunit and helix A (aa 32–42) of eEF2 (Figure 5d). Bulged A416 interacts with the switch loop in the vicinity of D53. Next to GDP, the C-terminal part of the switch loop (aa 61–67) adopts a helical fold. As such, the conformations of SWI and the GTPase center in general are similar to those observed in ribosome-bound EF-Tu (Voorhees et al., 2010) and EF-G (Pulk and Cate, 2013; Tourigny et al., 2013; Zhou et al., 2013) in the presence of GTP analogs. In Structure II, relative to Structure I, PKI is further shifted along the 40S body, traversing ~4 Å toward the P site (Figures 2e, f, and g), while stacking on C1274 at the head A site. Thus, the intermediate position of PKI is possible due to a large swivel of the head relative to the body, which brings the head A site close to the body P site. Domain IV of eEF2 is further entrenched in the A site by ~3 Å relative to the body and ~8 Å relative to the head, preserving its interactions with PKI. The decoding center residues A1755 and A1756 are rearranged to pack inside helix 44, making room for eEF2. This conformation of decoding center residues is also observed in the absence of A-site ligands (Ogle et al., 2001). The head interface of domain IV interacts with the 40S head (Figure 6a). Here, a positively charged surface of eEF2, formed by K613, R617 and R631 contacts the phosphate backbone of helix 33 (Figures 6c; see also Figure 6—figure supplement 1). Consistent with the similar head swivels in Structure III and Structure II, relative positions of the 40S head A site and body P site remain as in Structure II. Among the five structures, the PKI domain is least ordered in Structure III and lacks density for SL3. The map allows placement of PKI at the body P site (Figure 1—figure supplement 3). Thus, in Structure III, PKI has translocated along the 40S body, but the head remains fully swiveled so that PKI is between the head A and P sites. Lower resolution of the map in this region suggests that PKI is somewhat destabilized in the vicinity of the body P site in the absence of stacking with the foundations of the head A site (C1274) or P site (U1191). The position of eEF2 is similar to that in Structure II. In Structure IV, the 40S subunit is almost non-rotated relative to the 60S subunit, and the 40S head is mid-swiveled. Unwinding of the head moves the head P-site residue U1191 and body P-site residue C1637 closer together, resulting in a partially restored 40S P site. Whereas C1637 forms a stacking platform for the last base pair of PKI, U1191 does not yet stack on PKI because the head remains partially swiveled. This renders PKI partially accommodated in the P site (Figure 2g). Unwinding of the 40S head also positions the head A site closer to the body A site. This results in rearrangements of eEF2 interactions with the head, allowing eEF2 to advance further into the A site. To this end, the head-interacting interface of domain IV slides along the surface of the head by 5 Å. Helix A of domain IV is positioned next to the backbone of h34, with positively charged residues K613, R617 and R631 rearranged from the backbone of h33 (Figure 6c; see also Figure 6—figure supplement 1). In the nearly non-rotated and non-swiveled ribosome conformation in Structure V closely resembling that of the post-translocation 80S•2tRNA•mRNA complex (Svidritskiy et al., 2014), PKI is fully accommodated in the P site. The codon-anticodon–like helix is stacked on P-site residues U1191 and C1637 (Figure 3d), analogous to stacking of the tRNA-mRNA helix (Figure 3e). A notable conformational change in eEF2 from that in the preceding Structures is visible in the position of domain III, which contacts uS12 (Figure 6d). In Structure V, protein uS12 is shifted along with the 40S body as a result of intersubunit rotation. In this position, uS12 forms extensive interactions with eEF2 domains II and III. Specifically, the C-terminal tail of uS12 packs against the β-barrel of domain II, while the β-barrel of uS12 packs against helix A of domain III. This shifts the tip of helix A of domain III (at aa 500) by ~5 Å (relative to that in Structure I) toward domain I. Although domain III remains in contact with domain V, the shift occurs in the direction that could eventually disconnect the β-platforms of these domains. Domain IV of eEF2 is fully accommodated in the A site. The first codon of the open reading frame is also positioned in the A site, with bases exposed toward eEF2 (Figure 7), resembling the conformations of the A-site codons in EF-G-bound 70S complexes. As in the preceding Structures, the histidine-diphthamide tip is bound in the minor groove of the P-site codon-anticodon helix. Diph699 slightly rearranges, relative to that in Structure I (Figure 7), and interacts with four out of six codon-anticodon nucleotides. The imidazole moiety stacks on G6907 (corresponding to nt 36 in the tRNA anticodon) and hydrogen bonds with O2’ of G6906 (nt 35 of tRNA). The amide at the diphthamide end interacts with N2 of G6906 and O2 and O2’ of C6951 (corresponding to nt 2 of the codon). The trimethylamino-group is positioned over the ribose of C6952 (codon nt 3). In this work we have captured the structures of the TSV IRES, whose PKI samples positions between the A and P sites (Structures I–IV), as well as in the P site (Structure V). We propose that together with the previously reported initiation state (Koh et al., 2014), these structures represent the trajectory of eEF2-induced IRES translocation (shown as an animation in http://labs.umassmed.edu/korostelevlab/msc/iresmovie.gif and Video 1). Our structures reveal previously unseen intermediate states of eEF2 or EF-G engagement with the A site, providing the structural basis for the mechanism of translocase action. Furthermore, they provide insight into the mechanism of eEF2•GTP association with the pre-translocation ribosome and eEF2•GDP dissociation from the post-translocation ribosome, also delineating the mechanism of translation inhibition by the antifungal drug sordarin. In summary, the reported ensemble of structures substantially enhances our understanding of the translocation mechanism, including that of tRNAs as discussed below.Video 1.Animation showing the transition from the initiation 80S•TSV IRES structures (Koh et al., 2014) to eEF2-bound Structures I through V (this work).Four views (scenes) are shown: (1) A view down the intersubunit space, with the head of the 40S subunit oriented toward a viewer, as in Figure 1a; (2) A view at the solvent side of the 40S subunit, with the 40S head shown at the top, as in Figure 2—figure supplement 1; (3) A view down at the subunit interface of the 40S subunit; (4) A close-up view of the decoding center (A site) and the P site, as in Figure 2g. Each scene is shown twice. Colors are as in Figure 1. In scenes 1, 2 and 3, nucleotides C1274, U1191 of the 40S head and G904 of the 40S platform are shown in black to denote the A, P and E sites, respectively. In scene 4, C1274 and U1191 are labeled and shown in yellow; G577, A1755 and A1756 of the 40S body A site and C1637 of the body P site are labeled and shown in orange.DOI: http://dx.doi.org/10.7554/eLife.14874.02510.7554/eLife.14874.025 Four views (scenes) are shown: (1) A view down the intersubunit space, with the head of the 40S subunit oriented toward a viewer, as in Figure 1a; (2) A view at the solvent side of the 40S subunit, with the 40S head shown at the top, as in Figure 2—figure supplement 1; (3) A view down at the subunit interface of the 40S subunit; (4) A close-up view of the decoding center (A site) and the P site, as in Figure 2g. Each scene is shown twice. Colors are as in Figure 1. In scenes 1, 2 and 3, nucleotides C1274, U1191 of the 40S head and G904 of the 40S platform are shown in black to denote the A, P and E sites, respectively. In scene 4, C1274 and U1191 are labeled and shown in yellow; G577, A1755 and A1756 of the 40S body A site and C1637 of the body P site are labeled and shown in orange. DOI: http://dx.doi.org/10.7554/eLife.14874.025 Translocation of the TSV IRES on the 40S subunit globally resembles a step of an inchworm (Figure 4; see also Figure 3—figure supplement 2). At the start (initiation state), the IRES adopts an extended conformation (extended inchworm). The front 'legs' (SL4 and SL5) of the 5’-domain (front end) are attached to the 40S head proteins uS7, uS11 and eS25 (Figure 3—figure supplement 2). PKI, representing the hind end, is bound in the A site. In the first sub-step (Structures I to IV), the hind end advances from the A to the P site and approaches the front end, which remains attached to the 40S surface. This shortens the distance between PKI and SL4 by up to 20 Å relative to the initiating IRES structure, resulting in a bent IRES conformation (bent inchworm). Finally (Structures IV to V), as the hind end is accommodated in the P site, the front 'legs' advance by departing from their initial binding sites. This converts the IRES into an extended conformation, rendering the inchworm prepared for the next translocation step. Notably, at all steps, the head of the IRES inchworm (L1.1 region) is supported by the mobile L1 stalk. In the post-translocation CrPV IRES structure (Muhs et al., 2015), the 5’-domain similarly protrudes between the subunits and interacts with the L1 stalk, as in the initiation state for this IRES (Fernandez et al., 2014). This underlines structural similarity for the TSV and CrPV IRES translocation mechanisms. Upon translocation, the GCU start codon is positioned in the A site (Structure V), ready for interaction with Ala-tRNA upon eEF2 departure. Recent studies have shown that in some cases a fraction of IGR IRES-driven translation results from an alternative reading frame, which is shifted by one nucleotide relative to the normal ORF (Au and Jan, 2012; Ren et al., 2012; 2014; Wang and Jan, 2014). One of the mechanistic scenarios (discussed in Ren et al., 2014) involves binding of the first aminoacyl-tRNA to the post-translocated IRES mRNA frame shifted by one nucleotide (predominantly a +1 frame shift). In our structures, the IRES presents to the decoding center a pre-translocated or fully translocated ORF, rather than a +1 (more translocated) ORF, suggesting that eEF2 does not induce a highly populated fraction of +1 shifted IRES mRNAs. It is likely that alternative frame setting occurs following eEF2 release and that this depends on transient displacement of the start codon in the decoding center, allowing binding of the corresponding amino acyl-tRNA to an off-frame codon. Further structural studies involving 80S•IRES•tRNA complexes are necessary to understand the mechanisms underlying alternative reading frame selection. The presence of several translocation complexes in a single sample suggests that the structures represent equilibrium states of forward and reverse translocation of the IRES, which interconvert among each other. This is consistent with the observations that the intergenic IRESs are prone to reverse translocation. Specifically, biochemical toe-printing studies in the presence of eEF2•GTP identified IRES in a non-translocated position unless eEF1a•aa-tRNA is also present (Jan et al., 2003; Pestova and Hellen, 2003; Yamamoto et al., 2007). These findings indicate that IRES translocation by eEF2 is futile: the IRES returns to the A site upon releasing eEF2•GDP unless an amino-acyl tRNA enters the A site and blocks IRES back-translocation. This contrasts with the post-translocated 2tRNA•mRNA complex, in which the classical P and E-site tRNAs are stabilized in the non-rotated ribosome after translocase release (Chen et al., 2013a; Ermolenko and Noller, 2011). Thus, the meta-stability of the post-translocation IRES is likely due to the absence of stabilizing structural features present in the 2tRNA•mRNA complex. In the initiation state, the IRES resembles a pre-translocation 2tRNA•mRNA complex (Brilot et al., 2013) reduced to the A/P-tRNA anticodon-stem loop and elbow in the A site and the P/E-tRNA elbow contacting the L1 stalk. Because the anticodon-stem loop of the A-tRNA is sufficient for translocation completion (Joseph and Noller, 1998; Studer et al., 2003), we ascribe the meta-stability of the post-translocation IRES to the absence of the P/E-tRNA elements, either the ASL or the acceptor arm, or both. Furthermore, interactions of SL4 and SL5 with the 40S subunit likely contribute to stabilization of pre-translocation structures. Our structures delineate the mechanistic functions for intersubunit rotation and head swivel in translocation. These functions are partitioned. Specifically, intersubunit rotation allows eEF2 entry into the A site, while the head swivel mediates PKI translocation. Various degrees of intersubunit rotation have been observed in cryo-EM studies of the 80S•IRES initiation complexes (Fernandez et al., 2014; Koh et al., 2014). This suggests that the subunits are capable of spontaneous rotation, as is the case for tRNA-bound pre-translocation complexes (Cornish et al., 2008). The pre-translocation Structure I with eEF2 least advanced into the A site adopts a fully rotated conformation. Reverse intersubunit rotation from Structure I to V shifts the translocation tunnel (the tunnel between the A, P and E sites) toward eEF2, which is rigidly attached to the 60S subunit. This allows eEF2 to move into the A site. As such, reverse intersubunit rotation facilitates full docking of eEF2 in the A site. Because the histidine-diphthamide tip of eEF2 (H583, H694 and Diph699) attaches to the codon-anticodon-like helix of PKI, eEF2 appears to directly force PKI out of the A site. The head swivel allows gradual translocation of PKI to the P site, first with respect to the body and then to the head. The fully swiveled conformations of Structures II and III represent the mid-point of translocation, in which PKI relocates between the head A site and body P site. We note that such mid-states have not been observed for 2tRNA•mRNA, but their formation can explain the formation of subsequent pe/E hybrid (Ratje et al., 2010) and ap/P chimeric structures (Ramrath et al., 2013; Zhou et al., 2014) (Figure 1—figure supplement 1). Reverse swivel from Structure III to V brings the head to the non-swiveled position, restoring the A and P sites on the small subunit. To our knowledge, our work provides the first high-resolution view of the dynamics of a ribosomal translocase that is inferred from an ensemble of structures sampled under uniform conditions. The structures, therefore, offer a unique opportunity to address the role of the elongation factors during translocation. Translocases are efficient enzymes. While the ribosome itself has the capacity to translocate in the absence of the translocase, spontaneous translocation is slow (Cukras et al., 2003; Ermolenko et al., 2013; Gavrilova et al., 1976; Gavrilova and Spirin, 1972; Pestka, 1968). EF-G enhances the translocation rate by several orders of magnitude, aided by an additional 2- to 50-fold boost from GTP hydrolysis (Ermolenko and Noller, 2011; Rodnina et al., 1997). Due to the lack of structures of translocation intermediates, the mechanistic role of eEF2/EF-G is not fully understood. The 80S•IRES•eEF2 structures reported here suggest two main roles for eEF2 in translocation. As discussed above, the first role is to directly shift PKI out of the A site upon spontaneous reverse intersubunit rotation. In our structures, the tip of domain IV docks next to PKI, with diphthamide 699 fit into the minor groove of the codon-anticodon-like helix of PKI (Figure 7). This arrangement rationalizes inactivation of eEF2 by diphtheria toxin, which catalyzes ADP-ribosylation of the diphthamide (reviewed in Collier, 2001). The enzyme ADP-ribosylates the NE2 atom of the imidazole ring, which in our structures interacts with the first two residues of the anticodon-like strand of PKI. The bulky ADP-ribosyl moiety at this position would disrupt the interaction, rendering eEF2 unable to bind to the A site (Nygard and Nilsson, 1990) and/or stalled on ribosomes in a non-productive conformation (Bermek, 1976; Davydova and Ovchinnikov, 1990). As eEF2 shifts PKI toward the P site in the course of reverse intersubunit rotation, the 60S-attached translocase migrates along the surface of the 40S subunit, guided by electrostatic interactions. Positively-charged patches of domains II and III (R391, K394, R433, R510) and IV (K613, R617, R609, R631, K651) slide over rRNA of the 40S body (h5) and head (h18 and h33/h34), respectively. The Structures reveal hopping of the positive clusters over rRNA helices. For example, between Structures II and V, the K613/R617/R631 cluster of domain IV hops by ~19 Å (for Cα of R617) from the phosphate backbone of h33 (at nt 1261–1264) to that of the neighboring h34 (at nt 1442–1445). Thus, sliding of eEF2 involves reorganization of electrostatic, perhaps isoenergetic interactions, echoing those implied in extraordinarily fast ribosome inactivation rates by the small-protein ribotoxins (Korennykh et al., 2006) and in fast protein association and diffusion along DNA (Givaty and Levy, 2009; Gorman et al., 2010; Halford, 2009). Comparison of our structures with the 80S•IRES initiation structure reveals the structural basis for the second key function of the translocase: 'unlocking' of intrasubunit rearrangements that are required for step-wise translocation of PKI on the small subunit. The unlocking model of the ribosome•2tRNA•mRNA pre-translocation complex has been proposed decades ago (Spirin, 1969) and functional requirement of the translocase in this process has been implicated (Savelsbergh et al., 2003). However, the structural and mechanistic definitions of the locked and unlocked states have remained unclear, ranging from the globally distinct ribosome conformations (Valle et al., 2003) to unknown local rearrangements, e.g. those in the decoding center (Taylor et al., 2007). FRET data indicate that translocation of 2tRNA•mRNA on the 70S ribosome requires a forward-and-reverse head swivel (Guo and Noller, 2012), which may be related to the unlocking phenomenon. Whereas intersubunit rotation of the pre-translocation complex occurs spontaneously, the head swivel is induced by the eEF2/EF-G translocase, consistent with requirement of eEF2 for unlocking. Structural studies revealed large head swivels in various 70S•tRNA•EF-G (Ramrath et al., 2013; Ratje et al., 2010; Zhou et al., 2013) and 80S•tRNA•eEF2 (Taylor et al., 2007) complexes, but not in 'locked' complexes with the A site occupied by the tRNA in the absence of the translocase (Agirrezabala et al., 2008; Demeshkina et al., 2012; Jenner et al., 2010; Selmer et al., 2006). Our structures suggest that eEF2 induces head swivel by 'unlocking' the head-body interactions (Figure 7). Binding of the ASL to the A site is known from structural studies of bacterial ribosomes to result in 'domain closure' of the small subunit, i.e. closer association of the head, shoulder and body domains (Ogle et al., 2001). The domain closure 'locks' cognate tRNA in the A site via stacking on the head A site (C1274 in S. cerevisiae or C1054 in E. coli) and interactions with the body A-site nucleotides A1755 and A1756 (A1492 and A1493 in E. coli). This 'locked' state is identical to that observed for PKI in the 80S•IRES initiation structures in the absence of eEF2 (Fernandez et al., 2014; Koh et al., 2014). Structure I demonstrates that at an early pre-translocation step, the histidine-diphthamide tip of eEF2 is wedged between A1755 and A1756 and PKI. This destabilization allows PKI to detach from the body A site upon spontaneous reverse 40S body rotation, while maintaining interactions with the head A site. Destabilization of the head-bound PKI at the body A site thus allows mobility of the head relative to the body. The histidine-diphthamide-induced disengagement of PKI from A1755 and A1756 therefore provides the structural definition for the 'unlocking' mode of eEF2 action. In summary, our structures are consistent with a model of eEF2-induced translocation in which both PKI and eEF2 passively migrate into the P and A site, respectively, during spontaneous 40S body rotation and head swivel, the latter being allowed by 'unlocking' of the A site by eEF2. Observation of different PKI conformations sampling a range of positions between the A and P sites in the presence of eEF2•GDP implies that thermal fluctuations of the 40S head domain are sufficient for translocation along the energetically flat trajectory. The conformational rearrangements in eEF2 from Structure I through Structure V provide insights into the mechanisms of eEF2 association with the pre-translocation ribosome and dissociation from the post-translocation ribosome. In all five structures, the GTPase domain is attached to the P stalk and the sarcin-ricin loop. In the fully-rotated pre-translocation-like Structure I, an additional interaction exists. Here, switch loop I interacts with helix 14 (CAAA) of the 18S rRNA. This stabilization renders the GTPase center to adopt a GTP-bound conformation, similar to those observed in other translational GTPases in the presence of GTP analogs (Pulk and Cate, 2013; Tourigny et al., 2013; Voorhees et al., 2010; Zhou et al., 2013) and in the 80S•eEF2 complex bound with a transition-state mimic GDP•AlF (Sengupta et al., 2008). The switch loop contacts the base of A416 (invariable A344 in E. coli and A463 in H. sapiens). Mutations of residues flanking A344 in E. coli 16S rRNA modestly inhibit translation but do not specifically affect EF-G-mediated translocation (Sahu et al., 2012). However, the effect of A344 mutation on translation was not addressed in that study, leaving the question open whether this residue is critical for eEF2/EF-G function. The interaction between h14 and switch loop I is not resolved in Structures II to V, in all of which the small subunit is partially rotated or non-rotated, so that helix 14 is placed at least 6 Å farther from eEF2 (Figure 5d). We conclude that unlike other conformations of the ribosome, the fully rotated 40S subunit of the pre-translocation ribosome provides an interaction surface, complementing the P stalk and SRL, for binding of the GTP-bound translocase. This structural basis rationalizes the observation of transient stabilization of the rotated 70S ribosome upon EF-G•GTP binding and prior to translocation (Chen et al., 2013a; Ermolenko and Noller, 2011; Fei et al., 2008; Pan et al., 2007; Spiegel et al., 2007). The least rotated conformation of the post-translocation Structure V suggests conformational changes that may trigger eEF2 release from the ribosome at the end of translocation. The most pronounced inter-domain rearrangement in eEF2 involves movement of domain III. In the rotated or mid-rotated Structures I through III, this domain remains rigidly associated with domain V and the N-terminal superdomain and does not undergo noticeable rearrangements. In Structure V, however, the tip of helix A of domain III is displaced toward domain I by ~5 Å relative to that in mid-rotated or fully rotated structures. This displacement is caused by the 8 Å movement of the 40S body protein uS12 upon reverse intersubunit rotation from Structure I to V (Figure 6d). We propose that the shift of domain III by uS12 initiates intra-domain rearrangements in eEF2, which unstack the β-platform of domain III from that of domain V. This would result in a conformation characteristic of free eEF2 and EF-G in which the β-platforms are nearly perpendicular (Czworkowski et al., 1994; Evarsson et al., 1994; Jorgensen et al., 2003). As we discuss below, Structure V captures a 'pre-unstacking' state due to stabilization of the interface between domains III and V by sordarin. Sordarin is a potent antifungal antibiotic that inhibits translation. Based on biochemical experiments, two alternative mechanisms of action were proposed: sordarin either prevents eEF2 departure by inhibiting GTP hydrolysis (Dominguez et al., 1999) or acts after GTP hydrolysis (Justice et al., 1998). Although our complex was assembled using eEF2•GTP, density maps clearly show GDP and Mg in each structure (Figure 5g). Our structures therefore indicate that sordarin stalls eEF2 on the ribosome in the GDP-bound form, i.e. following GTP hydrolysis and phosphate release. The mechanism of stalling is suggested by comparison of pre-translocation and post-translocation structures in our ensemble. In all five structures, sordarin is bound between domains III and V of eEF2, stabilized by hydrophobic interactions identical to those in the isolated eEF2•sordarin complex (Figures 5g and h). In the nearly non-rotated post-translocation Structure V, the tip of domain III is shifted, however the interface between domains III and V remains unchanged, suggesting strong stabilization of this interface by sordarin. We note that Structure V is slightly more rotated than the 80S•2tRNA•mRNA complex in the absence of eEF2•sordarin, implying that sordarin interferes with the final stages of reverse rotation of the post-translocation ribosome. We propose that sordarin acts to prevent full reverse rotation and release of eEF2•GDP by stabilizing the interdomain interface and thus blocking uS12-induced disengagement of domain III from domain V. Because translocation of tRNA must involve large-scale dynamics, this step has long been regarded as the most puzzling step of translation. Intersubunit rearrangements and tRNA hybrid states have been proposed to play key roles half a century ago (Bretscher, 1968; Spirin, 1969). Despite an impressive body of biochemical, fluorescence and structural data accumulated since then, translocation remains the least understood step of elongation (Joseph, 2003; Ling and Ermolenko, 2016; Voorhees and Ramakrishnan, 2013). The structural understanding of ribosome and tRNA dynamics has been greatly aided by a wealth of X-ray and cryo-EM structures (reviewed in Agirrezabala and Valle, 2015; Dunkle and Cate, 2010; Korostelev et al., 2008). However, visualization of the eEF2/EF-G-induced translocation is confined to very early pre-EF-G-entry states (Brilot et al., 2013; Lin et al., 2015) and late (almost translocated or fully translocated) states (Gao et al., 2009; Ramrath et al., 2013; Zhou et al., 2014), leaving most of the path from the A to the P site uncharacterized (Figure 1—figure supplement 1). Our study provides new insights into the structural understanding of tRNA translocation. First, we propose that tRNA and IRES translocations occur via the same general trajectory. This is evident from the fact that ribosome rearrangements in translocation are inherent to the ribosome (Agirrezabala et al., 2008; Cornish et al., 2008; Gavrilova et al., 1976; Julián et al., 2008) and likely occur in similar ways in both cases. Furthermore, the step-wise coupling of ribosome dynamics with IRES translocation is overall consistent with that observed for 2tRNA•mRNA translocation in solution. For example, fluorescence and biochemical studies revealed that the early pre-translocation EF-G-bound ribosomes are fully rotated (Chen et al., 2013a; Ermolenko and Noller, 2011; Spiegel et al., 2007) and translocation of the tRNA-mRNA complex occurs during reverse rotation of the small subunit (Ermolenko and Noller, 2011), coupled with head swivel (Guo and Noller, 2012). The sequence of ribosome rearrangements during IRES translocation also agrees with that inferred from 70S•EF-G structures, including those in which the A-to-P-site translocating tRNA was not present. Specifically, an earlier translocation intermediate ribosome (TIpre) was proposed to adopt a rotated (7–9°) body and a partly rotated head (5–7.5°) (Chen et al., 2013b; Ratje et al., 2010; Tourigny et al., 2013), in agreement with the conformation of our Structure I. The most swiveled head (18–21°) was observed in a mid-rotated ribosome (3–5°) of a later translocation intermediate TIpost (Ramrath et al., 2013; Ratje et al., 2010), similar to the conformation of our Structure III. Overall, these correlations suggest that the intermediate locations of the elusive A-to-P-site translocating tRNA are similar to those of PKI in our structures. Second, the structures clarify the structural basis of the often-used but structurally undefined terms 'locking' and 'unlocking' with respect to the pre-translocation complex (Figure 6f). We deem the pre-translocation complex locked, because the A-site bound ASL-mRNA is stabilized by interactions with the decoding center (Ogle et al., 2001). These interactions are maintained for the classical- and hybrid-state tRNAs in the spontaneously sampled non-rotated and rotated ribosomes, respectively (Ermolenko et al., 2007; Spiegel et al., 2007). Unlocking involves separation of the codon-anticodon helix from the decoding center residues by the protruding tip of eEF2/EF-G (Figure 7), occurring in the fully rotated ribosome at an early pre-translocation step. This unlatches the head, allowing creation of hitherto elusive intermediate tRNA positions during spontaneous reverse body rotation. Third, our findings uncover a new role of the head swivel. Previous studies showed that this movement widens the constriction ('gate') between the P and E sites, thus allowing the P-tRNA passage to the E site (Schuwirth et al., 2005; Spahn et al., 2004a; Taylor et al., 2007; Zhou et al., 2014). In addition to the 'gate-opening' role, we now show that the head swivel brings the head A site to the body P site, allowing a step-wise conveying of the codon-anticodon helix between the A and P sites. Finally, the similar populations of particles (within a 2X range) in our 80S•IRES•eEF2 reconstructions (Figure 1—figure supplement 2) suggest that the intermediate translocation states sample several energetically similar and interconverting conformations. This is consistent with the idea of a rather flat energy landscape of translocation, suggested by recent work that measured mechanical work produced by the ribosome during translocation (Liu et al., 2014). Our findings implicate, however, that the energy landscape is not completely flat and contains local minima for transient positions of the codon-anticodon helix between the A and P sites. The shift of the PKI with respect to the body occurs during forward head swivel in two major sub-steps of ~4 Å each (initiation complex to I, and I to II), after which PKI undergoes small shifts to settle in the body P site in Structures III, IV and V (Figure 2—source data 1). Movement of PKI relative to the head occurs during the subsequent reverse swivel in three 3–7 Å sub-steps (II to III to IV to V). It is possible that additional meta-stable but less populated states exist between the conformations we observe. We note that four of our near-atomic resolution maps comprised ~30,000 particles each, the minimum number required for a near-atomic-resolution reconstruction of the ribosome (Bai et al., 2013). A larger data set will therefore be necessary to reveal additional sub-states. Our work sheds light on the dynamic mechanism of cap-independent translation by IGR IRESs, tightly coupled with the universally conserved dynamic properties of the ribosome. The cryo-EM structures demonstrate that the TSV IRES structurally and dynamically represents a chimera of the 2tRNA•mRNA translocating complex (A/P-tRNA • P/E-tRNA • mRNA). Like in the 2tRNA•mRNA translocating complex in which the two tRNAs move independently of each other, the PKI domain moves relative to the 5´-domain, causing the IRES to undergo an inchworm-walk translocation. A large structural difference between the IRES and the 2tRNA•mRNA complex exists, however, in that the IRES lacks three out of six tRNA-like domains involved in tRNA translocation. This difference likely accounts for the inefficient translocation of the IRES, which is difficult to stabilize in the post-translocation state and therefore is prone to reverse translocation. Although structurally handicapped, the TSV IRES manages to translocate by employing ribosome dynamics that are remarkably similar to that in 2tRNA•mRNA translocation. The uniformity of ribosome dynamics underscores the idea that translocation is an inherent and structurally-optimized property of the ribosome, supported also by translocation activity in the absence of the elongation factor. This property is rendered by the relative mobility of the three major building blocks, the 60S subunit and the 40S head and body, assisted by ligand-interacting extensions including the L1 stalk and the P stalk. Intergenic IRESs, in turn, represent a striking example of convergent molecular evolution. Viral mRNAs have evolved to adopt an atypical structure to employ the inherent ribosome dynamics, to be able to hijack the host translational machinery in a simple fashion. Our current understanding of macromolecular machines, such as the ribosome, is often limited by a gap between biophysical/biochemical studies and structural studies. For example, Förster resonance energy transfer can provide insight into the macromolecular dynamics of an assembly at the single-molecule level but is limited to specifically labeled locations within the assembly. High-resolution crystal structures, on the other hand, can provide static images of an assembly, and the structural dynamics can only be inferred by comparing structures that are usually obtained in different experiments and under different, often non-native, conditions. Cryo-EM offers the possibility of obtaining integrated information of both structure and dynamics as demonstrated in lower-resolution studies of bacterial ribosome complexes (Agirrezabala et al., 2008; Fischer et al., 2010; Julián et al., 2008). Recent technical advances, including direct electron detectors and image processing software (Cheng et al., 2015), have significantly improved the resolution at which such studies can be performed. The increased resolution, need for larger datasets and more sophisticated algorithms have also led to a massive increase in the computational power required to process the data. The available computing infrastructure and computational efficiency have therefore become deciding factors in how many different structural states can be resolved. This is presumably one of the reasons why most recent studies of ribosome complexes have focused on a single high-resolution structure despite the non-uniform local resolution of the maps that likely reflects structural heterogeneity. The computational efficiency of FREALIGN (Lyumkis et al., 2013) has allowed us to classify a relatively large dataset (1.1 million particles) into 15 classes (Figure 1—figure supplement 2) and obtain eight near-atomic-resolution structures from it. The classification, which followed an initial alignment of all particles to a single reference, required about 130,000 CPU hours or about five to six full days on a 1000-CPU cluster. While it would clearly be impractical to perform this type of analysis on a desktop computer, one may extrapolate using Moore’s law (Moore, 1965) that such practice will become routine in less than ten years. Therefore, cryo-EM has the potential to become a standard tool for uncovering detailed dynamic pathways of complex macromolecular machines. A particularly exciting application will be to infer the high-resolution temporal trajectory of a pathway from an ensemble of equilibrium states in a single sample, as described in our work, together with an analysis of samples quenched at different time points of the reaction (Chen et al., 2015; Fischer et al., 2010; Shaikh et al., 2014). 80S ribosomes used in this study were prepared from Saccharomyces cerevisiae strain W303 as described previously (Ben-Shem et al., 2011; Koh et al., 2014). To obtain ribosomal subunits, purified 80S was incubated in dissociation buffer (20 mM HEPES·KOH (pH 7.5), 0.5 M KCl, 2.5 mM magnesium acetate, 2 mM dithiothreitol (DTT), and 0.5 U/μl RNasin) for 1 hr at 4°C. The dissociated subunits were then layered on sucrose gradients (10% to 30% sucrose) in the dissociation buffer and centrifuged for 15 hr at 22,000 rpm in an SW32 rotor. Fractions corresponding to 40S and 60S subunits were pooled and buffer-exchanged to subunit storage buffer containing 50 mM Tris (pH7.5), 20 mM MgCl2, 100 mM KCl, and 2 mM DTT. Purified subunits were flash-frozen in liquid nitrogen and stored in aliquots at –80°C. Plasmid pUC57 (Genscript) containing the synthetic DNA encoding for nucleotides 6741–6990 of the TSV mRNA sequence was used to amplify the 250-nucleotide fragment by PCR. This DNA fragment (TSV IRES RNA) served as a template for in vitro transcription. The transcription reaction was incubated for 4 hr at 37°C, and the resulting transcription product was treated with DNase I for 30 mins at 37°C. The RNA was then extracted with acidic phenol/chloroform, gel-purified, and then ethanol precipitated with 100% ethanol, followed by an 80% ethanol wash. The resulting RNA pellet was air-dried at room temperature and suspended in RNase-free water. The TSV IRES transcription product was folded in 1X IRES refolding buffer (20 mM Potassium acetate pH 7.5 and 5 mM MgCl2), incubated at 65°C for 10 min and cooled down gradually at room temperature, prior to complex preparation for cryo-EM study. C-terminally His6-tagged eEF2 was produced in yeast TKY675 cells obtained from Terri Goss Kinzy. Yeast cells were grown in 4 liters of YPD media at 27°C and 160 rpm, to A600=1.5. Yeast cell pellet (~5 g) was obtained by centrifugation and re-suspended in 20 ml of the lysis buffer (50 mM potassium phosphate pH 7.6, 1 M KCl, 1% Tween 20, 10% Glycerol, 10 mM imidazole, 0.2 mM PMSF, 1 mM DTT, and 1 tablet of Roche miniComplete protease inhibitor). The suspension was lysed with microfluidizer at 25,000 psi at 4°C, and then clarified by centrifugation twice at 30,000 × g for 20 min. The supernatant was subjected to Ni-NTA affinity chromatography using the AKTAexplorer 100 system (GE Healthcare). After lysate application onto the column, the column was washed with a five-column volume of wash buffer (50 mM potassium phosphate pH 7.6, 1 M KCl, 1% Tween 20, 10% Glycerol, 20 mM imidazole, 0.2 mM PMSF and 1 mM DTT). A gradient elution method was used to reach the final imidazole concentration of 250 mM. Eluted fractions were buffer-exchanged into buffer A (30 mM HEPES·KOH (pH 7.5), 5% glycerol, 65 mM ammonium chloride, 7 mM β–mercaptoethanol and 1 tablet of miniComplete protease inhibitor) for HiTrap SP Sepharose High Performance cation-exchange chromatography (GE Healthcare). A gradient elution method was used to reach the final salt concentration of 1 M KCl in buffer A over the 20-column volume (100 ml). The peak fraction was concentrated and buffer-exchanged into buffer A, which is also the buffer used for the subsequent size-exclusion chromatography employing Superdex 200 (GE Healthcare). Fractions corresponding to the eEF2 peak were concentrated and stored in aliquots at -20°C. The IRES-eEF2-ribosome complex was assembled in two steps. First, refolded TSV IRES RNA (8 μM - all concentrations are specified for the final complex) was incubated with the yeast 40S small subunit (0.8 μM) for 15 min at 30°C, in the buffer containing 45 mM HEPES·KOH (pH 7.5), 10 mM MgCl2, 100 mM KCl, 2.5 mM spermine and 2 mM β–mercaptoethanol. The 60S subunit (0.8 μM) was then added and incubated for 15 min at 30°C. Subsequently, eEF2 (5 μM), sordarin (800 μM) and GTP (1 mM) were added and incubated for 15 min at 30°C. The solution was then incubated on ice for 10 min and flash-frozen in liquid nitrogen. Quantifoil Cu 200 mesh grids (SPI Supplies, West Chester, PA) were coated with a thin layer of carbon and glow discharged for 45 s at 25 mA. 3 µL of sample with a concentration of ~0.1 µM was applied to the grid, incubated for 30 s and plunged into liquid ethane using an FEI Vitrobot Mark 2 (FEI Company, Hillsboro, OR) after blotting for 3 s at 4°C and ~85% relative humidity. Cryo-EM data were collected in movie mode on an FEI Krios microscope (FEI Company, Hillsboro, OR) operating at 300 kV and equipped with a K2 Summit direct detector (Gatan Inc., Pleasanton, CA) operating in super-resolution mode with pixel size of 0.82 Å per super-resolution pixel. Each movie consisted of 50 frames collected over 18.8 s with an exposure per frame of 1.4 e-/Å2 as shown by Digital Micrograph (Gatan Inc., Pleasanton, CA), giving a total exposure of 70 e-/Å2. The defocus ranged between ~0.7 to ~2.5 µm underfocus. The gain-corrected super-resolution movie frames were corrected for anisotropic magnification using bilinear interpolation (Grant and Grigorieff, 2015a). The frames were downsampled by Fourier cropping to a pixel size of 1.64 Å. The downsampled frames were then motion-corrected and exposure filtered using Unblur (Grant and Grigorieff, 2015b). The image defocus was determined using CTFFIND4 (Rohou and Grigorieff, 2015) on non-exposure-filtered images and images with excessive motion, low contrast, ice contamination or poor power spectra were removed based on visual inspection using TIGRIS (http://tigris.sourceforge.net/). 50 particles were picked manually using TIGRIS, summed and rotationally averaged to serve as a reference for correlation-based particle picking in IMAGIC (van Heel et al., 1996). 1,105,737 two-dimensional images of ribosomes (termed 'particles') were picked automatically, extracted into 256 x 256 boxes and converted to MRC/CCP4 format with a corresponding list of micrograph numbers and defocus values for input to FREALIGN v9 (Lyumkis et al., 2013). The summary of procedures resulting in 3D cryo-EM maps is presented on Figure 1—figure supplement 2. FREALIGN v9 was used for refinement, classification and 3D reconstruction of all ribosome structures. Initial particle alignments were obtained by performing an angular grid search (FREALIGN mode 3) with a density map calculated from the atomic model of the non-rotated 80S ribosome bound with 2 tRNAs (PDB: 3J78 Svidritskiy et al., 2014). For this search, the resolution was limited to 20 Å and the resolution of the resulting reconstruction was 3.6 Å, as determined by the FSC = 0.143 threshold criterion (Rosenthal and Henderson, 2003). Four additional rounds of mode 3 with the resolution limited to 7 Å improved the resolution of the reconstruction to 3.5 Å. Starting with cycle 6, particles were classified into six classes using 21 rounds of mode 1 refinement. Inspection of the six classes suggested that several represented mixed conformations. The alignment parameters of the class containing the largest number of particles (25%) were therefore used to initialize classification into 15 classes. For this classification, particle images were downsampled by Fourier cropping to a pixel size of 3.28 Å to accelerate processing. 99 rounds of refinement and classification were performed using mode 1 with a resolution limit of 7 Å. To help separate different conformations affecting small subunit, IRES and eEF2, we used a 3D mask that included density belonging to these parts of the structure. This mask was applied in every cycle to the 3D reference structures prior to refinement and classification in 42 additional cycles. The mask was then changed to include only the head of the small subunit, IRES and eEF2, and a final 18 cycles of refinement and classification were run. We selected six out of the 15 final classes based on clear density present for IRES and eEF2 and continued all further processing with this subset of the data (312,698 particles). The six classes were grouped into three groups based on the rotational state of the small subunit, and each group was further refined and classified using between six and 36 cycles of mode 1 and particles downsampled to 1.64 Å pixel size. For this classification, FREALIGN’s focused mask feature was used to select either the region of IRES PKI (for classes showing intermediate rotation of the small subunit) or a region containing both IRES PKI and eEF2 domain 4 (for classes showing no rotation of the small subunit). This refinement and sub-classification produced eight new classes with more distinct features in the regions selected by the focused masks. These eight classes were used as starting references for a final 33 cycles of refinement and classification using mode 1 and focused mask with the radius of 45 Å covering the vicinity of the ribosomal A site. The first 26 cycles were performed using particles downsampled to 3.28 Å pixel size, followed by two cycles at a pixel size of 1.64 Å, and five cycles at a pixel size of 0.82 Å. The resolution limit for the final cycles was set at 5 Å. The resulting eight reconstructions were used for further analyses, model building and structural refinements, as described below. In parallel, to enhance resolution of the IRES 5´ domain, we performed classification and refinement of the eight classes using a mask with the radius of 50 Å covering the vicinity of the E site and L1 stalk; these maps were used for model building and confirmation of the IRES 5´ domain structure, but not for structure refinements. Among the resulting eight reconstructions, four reconstructions contained well defined PKI and eEF2 densities (I, II, IV and V) (Figure 1—figure supplement 1). PKI was poorly resolved in reconstruction III. Reconstruction VI represents the previously reported 80S•TSV IRES initiation complex in the least rotated conformation (Koh et al., 2014). Reconstructions VII and VIII correspond to ribosomes adopting intermediate rotational states, similar to that of Structure III, with weak density in the region of the 5’ domain of the IRES and no density for the PKI domain. To resolve heterogeneity of PKI in reconstruction III, we performed additional sub-classification of all eight classes into two or three classes each. This sub-classification did not result into different structures for the four classes of interest (I, II, IV and V), suggesting a high degree of homogeneity in the masked regions of PKI and eEF2 domain IV. Sub-classification of reconstruction III helped improve the PKI density, resulting in a 4.2 Å reconstruction. All maps were subsequently B-factor-filtered by bfactor.exe (Lyumkis et al., 2013), using the B-factors of -50 to -120 Å, as suggested by bfactor.exe for individual maps, and used for real-space structure refinements. FSC curves (Figure 1—figure supplement 3) were calculated by FREALIGN for even and odd particles half-sets. Blocres (Cardone et al., 2013) was used to calculate the local resolution of unfiltered and unmasked volumes using a box size 60 pixel, step size of 3 pixels and FSC resolution criterion (threshold 0.143). The output volumes were then colored according to the local resolution of the final reconstructions (Figure 1—figure supplement 3) using the Surface Color tool of Chimera (Pettersen et al., 2004) The starting structural models were assembled using the high-resolution crystal structure of S. cerevisiae 80S ribosome (Ben-Shem et al., 2011), the cryo-EM structure of the 80S•TSV IRES complex (Koh et al., 2014) and the crystal structure of the isolated eEF2•sordarin complex (Jorgensen et al., 2003). The structure of the diphthamide residue of eEF2 (699) was adopted from PDB: 1ZM4 (Jorgensen et al., 2005). Initial domain fitting into the cryo-EM maps was performed using Chimera (Pettersen et al., 2004), followed by manual modeling of local regions into the density using Pymol (DeLano, 2002) and Coot (Emsley and Cowtan, 2004). Parts of several ribosomal proteins were modeled using I-TASSER (Yang et al., 2015) and Phyre2 (Kelley et al., 2015). The structural models were refined by real-space simulated-annealing refinement using atomic electron scattering factors (Gonen et al., 2005), employing RSRef (Chapman, 1995; Korostelev et al., 2002) as described (Svidritskiy et al., 2014). Secondary-structure restrains for ribosomal proteins and base-pairing restraints for RNA molecules were employed, as described (Laurberg et al., 2008). The refined structural models closely agree with the corresponding maps, as shown by low real-space R-factors of ~0.2 to 0.27, and they have good stereochemical parameters, characterized by low deviation from ideal bond lengths and angles (Figure 1—source data 1). The maps revealed regions, which are differently resolved in Structures I to V. The most prominent difference is in the platform subdomain of the 40S subunit, which is well resolved in Structures I, IV and V but poorly resolved in Structures II and III. The following components of the 40S platform in Structures II and III lacked resolution: proteins eS1, uS11, eS26 and eL41, 18S rRNA nt 892–900, 900–918 and the 3´ end beyond nt 1792. These and other regions of low density were modeled as protein or RNA backbone. For structural comparisons, the distances and angles were calculated in Pymol and Chimera, respectively. To calculate an angle of the 40S subunit rotation between two 80S structures, the 25S rRNAs were aligned using Pymol, and the angle between 18S rRNAs was measured in Chimera. To calculate an angle of the 40S-head rotation (swivel) between two 80S structures, the 18S rRNAs of the bulk of the 40S body (18S nucleotides excluding nt 1150–1620) were aligned using Pymol, and the angle between the 18S rRNA residues 1150–1620 was measured in Chimera. Figures were prepared in Pymol and Chimera.
PMC4832331	Structural insights into the Escherichia coli lysine decarboxylases and molecular determinants of interaction with the AAA+ ATPase RavA	The inducible lysine decarboxylase LdcI is an important enterobacterial acid stress response enzyme whereas LdcC is its close paralogue thought to play mainly a metabolic role. A unique macromolecular cage formed by two decamers of the Escherichia coli LdcI and five hexamers of the AAA+ ATPase RavA was shown to counteract acid stress under starvation. Previously, we proposed a pseudoatomic model of the LdcI-RavA cage based on its cryo-electron microscopy map and crystal structures of an inactive LdcI decamer and a RavA monomer. We now present cryo-electron microscopy 3D reconstructions of the E. coli LdcI and LdcC, and an improved map of the LdcI bound to the LARA domain of RavA, at pH optimal for their enzymatic activity. Comparison with each other and with available structures uncovers differences between LdcI and LdcC explaining why only the acid stress response enzyme is capable of binding RavA. We identify interdomain movements associated with the pH-dependent enzyme activation and with the RavA binding. Multiple sequence alignment coupled to a phylogenetic analysis reveals that certain enterobacteria exert evolutionary pressure on the lysine decarboxylase towards the cage-like assembly with RavA, implying that this complex may have an important function under particular stress conditions.Enterobacterial inducible decarboxylases of basic amino acids lysine, arginine and ornithine have a common evolutionary origin and belong to the α-family of pyridoxal-5′-phosphate (PLP)-dependent enzymes12. They counteract acid stress experienced by the bacterium in the host digestive and urinary tract, and in particular in the extremely acidic stomach34. Each decarboxylase is induced by an excess of the target amino acid and a specific range of extracellular pH, and works in conjunction with a cognate inner membrane antiporter. Decarboxylation of the amino acid into a polyamine is catalysed by a PLP cofactor in a multistep reaction12 that consumes a cytoplasmic proton and produces a CO2 molecule passively diffusing out of the cell, while the polyamine is excreted by the antiporter in exchange for a new amino acid substrate. Consequently, these enzymes buffer both the bacterial cytoplasm and the local extracellular environment5. These amino acid decarboxylases are therefore called acid stress inducible or biodegradative to distinguish them from their biosynthetic lysine and ornithine decarboxylase paralogs catalysing the same reaction but responsible for the polyamine production at neutral pH. Inducible enterobacterial amino acid decarboxylases have been intensively studied since the early 194067 because the ability of bacteria to withstand acid stress can be linked to their pathogenicity in humans. In particular, the inducible lysine decarboxylase LdcI (or CadA) attracts attention due to its broad pH range of activity and its capacity to promote survival and growth of pathogenic enterobacteria such as Salmonella enterica serovar Typhimurium, Vibrio cholerae and Vibrio vulnificus under acidic conditions589. Furthermore, both LdcI and the biosynthetic lysine decarboxylase LdcC of uropathogenic Escherichia coli (UPEC) appear to play an important role in increased resistance of this pathogen to nitrosative stress produced by nitric oxide and other damaging reactive nitrogen intermediates accumulating during the course of urinary tract infections (UTI)1011. This effect is attributed to cadaverine, the diamine produced by decarboxylation of lysine by LdcI and LdcC, that was shown to enhance UPEC colonisation of the bladder11. In addition, the biosynthetic E. coli lysine decarboxylase LdcC, long thought to be constitutively expressed in low amounts, was demonstrated to be strongly upregulated by fluoroquinolones via their induction of RpoS1213. A direct correlation between the level of cadaverine and the resistance of E. coli to these antibiotics commonly used as a first-line treatment of UTI could be established12. Both acid pH and cadaverine induce closure of outer membrane porins thereby contributing to bacterial protection from acid stress, but also from certain antibiotics, by reduction in membrane permeability141516. The crystal structure of the E. coli LdcI17 as well as its low resolution characterisation by electron microscopy171819 (EM) showed that it is a decamer made of two pentameric rings. Each monomer is composed of three domains – an N-terminal wing domain (residues 1–129), a PLP-binding core domain (residues 130–563), and a C-terminal domain (CTD, residues 564–715). Monomers tightly associate via their core domains into 2-fold symmetrical dimers with two complete active sites, and further build a toroidal D5-symmetrical structure held by the wing and core domain interactions around the central pore, with the CTDs at the periphery. Ten years ago19 we showed that the E. coli AAA+ ATPase RavA, involved in multiple stress response pathways19202122, tightly interacted with LdcI but was not capable of binding to LdcC. We described how two double pentameric rings of the LdcI17 tightly associate with five hexameric rings of RavA21 to form a unique cage-like architecture that enables the bacterium to withstand acid stress even under conditions of nutrient deprivation eliciting stringent response192123. Furthermore, we recently solved the structure of the E. coli LdcI-RavA complex by cryo-electron microscopy (cryoEM) and combined it with the crystal structures of the individual proteins23. This allowed us to make a pseudoatomic model of the whole assembly, underpinned by a cryoEM map of the LdcI-LARA complex (with LARA standing for LdcI associating domain of RavA), and to identify conformational rearrangements and specific elements essential for complex formation23. The main determinants of the LdcI-RavA cage assembly appeared to be the N-terminal loop of the LARA domain of RavA and the C-terminal β-sheet of LdcI23. In spite of this wealth of structural information, the fact that LdcC does not interact with RavA, although the two lysine decarboxylases are 69% identical and 84% similar1924, and the physiological significance of the absence of this interaction remained unexplored. To solve this discrepancy, in the present work we provided a three-dimensional (3D) cryoEM reconstruction of LdcC and compared it with the available LdcI and LdcI-RavA structures. Given that the LdcI crystal structures were obtained at high pH where the enzyme is inactive (LdcIi, pH 8.5), whereas the cryoEM reconstructions of LdcI-RavA and LdcI-LARA were done at acidic pH optimal for the enzymatic activity, for a meaningful comparison, we also produced a 3D reconstruction of the LdcI at active pH (LdcIa, pH 6.2). This comparison pinpointed differences between the biodegradative and the biosynthetic lysine decarboxylases and brought to light interdomain movements associated to pH-dependent enzyme activation and RavA binding, notably at the predicted RavA binding site at the level of the C-terminal β-sheet of LdcI. Consequently, we tested the capacity of cage formation by LdcI-LdcC chimeras where we interchanged the C-terminal β-sheets in question. Finally, we performed multiple sequence alignment of 22 lysine decarboxylases from Enterobacteriaceae containing the ravA-viaA operon in their genome. Remarkably, this analysis revealed that several specific residues in the above-mentioned β-sheet, independently of the rest of the protein sequence, are sufficient to define if a particular lysine decarboxylase should be classified as an “LdcC-like” or an “LdcI-like”. Moreover, this classification perfectly agrees with the genetic environment of the lysine decarboxylase genes. This fascinating parallelism between the propensity for RavA binding and the genetic environment of an enterobacterial lysine decarboxylase, as well as the high degree of conservation of this small structural motif, emphasize the functional importance of the interaction between biodegradative enterobacterial lysine decarboxylases and the AAA+ ATPase RavA. In the frame of this work, we produced two novel subnanometer resolution cryoEM reconstructions of the E. coli lysine decarboxylases at pH optimal for their enzymatic activity – a 5.5 Å resolution cryoEM map of the LdcC (pH 7.5) for which no 3D structural information has been previously available (Figs 1A,B and S1), and a 6.1 Å resolution cryoEM map of the LdcIa, (pH 6.2) (Figs 1C,D and S2). In addition, we improved our earlier cryoEM map of the LdcI-LARA complex from 7.5 Å to 6.2 Å resolution (Figs 1E,F and S3). Based on these reconstructions, reliable pseudoatomic models of the three assemblies were obtained by flexible fitting of either the crystal structure of LdcIi or a derived structural homology model of LdcC (Table S1). Significant differences between these pseudoatomic models can be interpreted as movements between specific biological states of the proteins as described below. As a first step of a comparative analysis, we superimposed the three cryoEM reconstructions (LdcIa, LdcI-LARA and LdcC) and the crystal structure of the LdcIi decamer (Fig. 2 and Movie S1). This superposition reveals that the densities lining the central hole of the toroid are roughly at the same location, while the rest of the structure exhibits noticeable changes. Specifically, at the center of the double-ring the wing domains of the subunits provide the conserved basis for the assembly with the lowest root mean square deviation (RMSD) (between 1.4 and 2 Å for the Cα atoms only), whereas the peripheral CTDs containing the RavA binding interface manifest the highest RMSD (up to 4.2 Å) (Table S2). In addition, the wing domains of all structures are very similar, with the RMSD after optimal rigid body alignment (RMSDmin) less than 1.1 Å. Thus, taking the limited resolution of the cryoEM maps into account, we consider that the wing domains of all the four structures are essentially identical and that in the present study the RMSD of less than 2 Å can serve as a baseline below which differences may be assumed as insignificant. This preservation of the central part of the double-ring assembly may help the enzymes to maintain their decameric state upon activation and incorporation into the LdcI-RavA cage. Both visual inspection (Fig. 2) and RMSD calculations (Table S2) show that globally the three structures at active pH (LdcIa, LdcI-LARA and LdcC) are more similar to each other than to the structure determined at high pH conditions (LdcIi). The decameric enzyme is built of five dimers associating into a 5-fold symmetrical double-ring17 (two monomers making a dimer are delineated in Fig. 1). As common for the α family of the PLP-dependent decarboxylases1725, dimerization is required for the enzymatic activity because the active site is buried in the dimer interface (Fig. 3A,B). This interface is formed essentially by the core domains with some contribution of the CTDs. The core domain is built by the PLP-binding subdomain (PLP-SD, residues 184–417) flanked by two smaller subdomains rich in partly disordered loops – the linker region (residues 130–183) and the subdomain 4 (residues 418–563). Zooming in the variations in the PLP-SD shows that most of the structural changes concern displacements in the active site (Fig. 3C–F). The most conspicuous differences between the PLP-SDs can be linked to the pH-dependent activation of the enzymes. The resolution of the cryoEM maps does not allow modeling the position of the PLP moiety and calls for caution in detailed mechanistic interpretations in terms of individual amino acids. Therefore we restrict our analysis to secondary structure elements. In particular, transition from LdcIi to LdcI-LARA involves ~3.5 Å and ~4.5 Å shifts away from the 5-fold axis in the active site α-helices spanning residues 218–232 and 246–254 respectively (Fig. 3C–E). Between these two extremes, the PLP-SDs of LdcIa and LdcC are similar both in the context of the decamer (Fig. 3F) and in terms of RMSDmin = 0.9 Å, which probably reflects the fact that, at the optimal pH, these lysine decarboxylases have a similar enzymatic activity26. In addition, our earlier biochemical observation that the enzymatic activity of LdcIa is unaffected by RavA binding19 is consistent with the relatively small changes undergone by the active site upon transition from LdcIa to LdcI-LARA. Worthy of note, our previous comparison of the crystal structure of LdcIi with that of the inducible arginine decarboxylase AdiA17 revealed high conservation of the PLP-coordinating residues and identified a patch of negatively charged residues lining the active site channel as a potential binding site for the target amino acid substrate17 (Figs S3 and S4 in ref. 17). An inhibitor of the LdcI and LdcC activity, the stringent response alarmone ppGpp, is known to bind at the interface between neighboring monomers within each ring (Fig. S4). The ppGpp binding pocket is made up by residues from all domains and is located approximately 30 Å away from the PLP moiety17. Whereas the crystal structure of the ppGpp-LdcIi was solved to 2 Å resolution, only a 4.1 Å resolution structure of the ppGpp-free LdcIi could be obtained17. At this resolution, the apo-LdcIi and ppGpp-LdcIi structures (both solved at pH 8.5) appeared indistinguishable except for the presence of ppGpp17 (Fig. S11 in ref. 17). Thus, we speculated that inhibition of LdcI by ppGpp would be accompanied by a transduction of subtle structural changes at the level of individual amino acid side chains between the ppGpp binding pocket and the active site of the enzyme17. All our current cryoEM reconstructions of the lysine decarboxylases were obtained in the absence of ppGpp in order to be closer to the active state of the enzymes under study. While differences in the ppGpp binding site could indeed be visualized (Fig. S4), the level of resolution warns against speculations about their significance. The fact that interaction with RavA reduces the ppGpp affinity for LdcI21 despite the long distance of ~30 Å between the LARA domain binding site and the closest ppGpp binding pocket (Fig. S5) seems to favor an allosteric regulation mechanism. Interestingly, although a number of ppGpp binding residues are strictly conserved between LdcI and AdiA that also forms decamers at low pH optimal for its arginine decarboxylase activity, no ppGpp regulation of AdiA could be demonstrated26. Inspection of the superimposed decameric structures (Figs 2 and S6) suggests a depiction of the wing domains as an anchor around which the peripheral CTDs swing. This swinging movement seems to be mediated by the core domains and is accompanied by a stretching of the whole LdcI subunits attracted by the RavA magnets. Indeed, all CTDs have very similar structures (RMSDmin <1 Å). Yet the superposition of the decamers lays bare a progressive movement of the CTD as a whole upon enzyme activation by pH and the binding of LARA. The LdcIi monomer is the most compact, whereas LdcIa and especially LdcI-LARA gradually extend their CTDs towards the LARA domain of RavA (Figs 2 and 4). These small but noticeable swinging and stretching (up to ~4 Å) may be related to the incorporation of the LdcI decamer into the LdcI-RavA cage. In our previous contribution, based on the fit of the LdcIi and the LARA crystal structures into the LdcI-LARA cryoEM density, we predicted that the LdcI-RavA interaction should involve the C-terminal two-stranded β-sheet of the LdcI23. Our present cryoEM maps and pseudoatomic models provide first structure-based insights into the differences between the inducible and the constitutive lysine decarboxylases. However, at the level of this structural element the two proteins are actually surpisingly similar. Therefore, we wanted to check the influence of the primary sequence of the two proteins in this region on their ability to interact with RavA. To this end, we swapped the relevant β-sheets of the two proteins and produced their chimeras, namely LdcIC (i.e. LdcI with the C-terminal β-sheet of LdcC) and LdcCI (i.e. LdcC with the C-terminal β-sheet of LdcI) (Fig. 5A–C). Both constructs could be purified and could form decamers visually indistinguishable from the wild-type proteins. As expected, binding of LdcI to RavA was completely abolished by this procedure and no LdcIC-RavA complex could be detected. On the contrary, introduction of the C-terminal β-sheet of LdcI into LdcC led to an assembly of the LdcCI-RavA complex. On the negative stain EM grid, the chimeric cages appeared less rigid than the native LdcI-RavA, which probably means that the environment of the β-sheet contributes to the efficiency of the interaction and the stability of the entire architecture (Fig. 5D–F). Alignment of the primary sequences of the E. coli LdcI and LdcC shows that some amino acid residues of the C-terminal β-sheet are the same in the two proteins, whereas others are notably different in chemical nature. Importantly, most of the amino acid differences between the two enzymes are located in this very region. Thus, to advance beyond our experimental confirmation of the C-terminal β-sheet as a major determinant of the capacity of a particular lysine decarboxylase to form a cage with RavA, we set out to investigate whether certain residues in this β-sheet are conserved in lysine decarboxylases of different enterobacteria that have the ravA-viaA operon in their genome. We inspected the genetic environment of lysine decarboxylases from 22 enterobacterial species referenced in the NCBI database, corrected the gene annotation where necessary (Tables S3 and S4), and performed multiple sequence alignment coupled to a phylogenetic analysis (see Methods). This procedure yielded several unexpected and exciting results. First of all, consensus sequence for the entire lysine decarboxylase family was derived. Second, the phylogenetic analysis clearly split the lysine decarboxylases into two groups (Fig. 6A). All lysine decarboxylases predicted to be “LdcI-like” or biodegradable based on their genetic environment, as for example their organization in an operon with a gene encoding the CadB antiporter (see Methods), were found in one group, whereas all enzymes predicted as “LdcC-like” or biosynthetic partitioned into another group. Thus, consensus sequences could also be determined for each of the two groups (Figs 6B,C and S7). Inspection of these consensus sequences revealed important differences between the groups regarding charge, size and hydrophobicity of several residues precisely at the level of the C-terminal β-sheet that is responsible for the interaction with RavA (Fig. 6B–D). For example, in our previous study23, site-directed mutations identified Y697 as critically required for the RavA binding. Our current analysis shows that Y697 is strictly conserved in the “LdcI-like” group whereas the “LdcC-like” enzymes always have a lysine in this position; it also uncovers several other residues potentially essential for the interaction with RavA which can now be addressed by site-directed mutagenesis. The third and most remarkable finding was that exactly the same separation into “LdcI-like” and “LdcC”-like groups can be obtained based on a comparison of the C-terminal β-sheets only, without taking the rest of the primary sequence into account. Therefore the C-terminal β-sheet emerges as being a highly conserved signature sequence, sufficient to unambiguously discriminate between the “LdcI-like” and “LdcC-like” enterobacterial lysine decarboxylases independently of any other information (Figs 6 and S7). Our structures show that this motif is not involved in the enzymatic activity or the oligomeric state of the proteins. Thus, enterobacteria identified here (Fig. 6, Table S4) appear to exert evolutionary pressure on the biodegradative lysine decarboxylase towards the RavA binding. One of the elucidated roles of the LdcI-RavA cage is to maintain LdcI activity under conditions of enterobacterial starvation by preventing LdcI inhibition by the stringent response alarmone ppGpp21. Furthermore, the recently documented interaction of both LdcI27 and RavA22 with specific subunits of the respiratory complex I, together with the unanticipated link between RavA and maturation of numerous iron-sulfur proteins, tend to suggest an additional intriguing function for this 3.5 MDa assembly. The conformational rearrangements of LdcI upon enzyme activation and RavA binding revealed in this work, and our amazing finding that the molecular determinant of the LdcI-RavA interaction is the one that straightforwardly determines if a particular enterobacterial lysine decarboxylase belongs to “LdcI-like” or “LdcC-like” proteins, should give a new impetus to functional studies of the unique LdcI-RavA cage. Besides, the structures and the pseudoatomic models of the active ppGpp-free states of both the biodegradative and the biosynthetic E. coli lysine decarboxylases offer an additional tool for analysis of their role in UPEC infectivity. Together with the apo-LdcI and ppGpp-LdcIi crystal structures, our cryoEM reconstructions provide a structural framework for future studies of structure-function relationships of lysine decarboxylases from other enterobacteria and even of their homologues outside Enterobacteriaceae. For example, the lysine decarboxylase of Eikenella corrodens is thought to play a major role in the periodontal disease and its inhibitors were shown to retard gingivitis development282930. Finally, cadaverine being an important platform chemical for the production of industrial polymers such as nylon, structural information is valuable for optimisation of bacterial lysine decarboxylases used for its production in biotechnology313233. LdcI and LdcC were expressed and purified as described192326 from an E. coli strain that cannot produce ppGpp (MG1655 ΔrelA ΔspoT strain). LdcI was stored in 20 mM Tris-HCl, 100 mM NaCl, 1 mM DTT, 0.1 mM PLP, pH 6.8 (buffer A) and LdcC in 20 mM Tris-HCl, 100 mM NaCl, 1 mM DTT, 0.1 mM PLP, pH 7.5 (buffer B). Chimeric LdcIC and LdcCI were constructed, expressed and purified as follows. The chimeras were designed by exchange, between LdcI and LdcC, of residues from 631 to 640 and from 697 to the C-terminus, corresponding to the regions around the two strands of the C-terminal β-sheet (Fig. 5B,C), while leaving the rest of the sequence unaltered. The synthetic ldcIC and ldcCI genes (2148 bp and 2154 bp respectively), provided within a pUC57 vector (GenScript) were subcloned into pET-TEV vector based on pET-28a (Invitrogen) containing an N-terminal TEV-cleavable 6x-His-Tag. Proteins were expressed in Rosetta 2 (DE3) cells (Novagen) in LB medium supplemented with kanamycin and chloramphenicol at 37 °C, upon induction with 0.5 mM IPTG at 18 °C. Cells were harvested by centrifugation, the pellet resuspended in a 50 mM Tris-HCl, 150 mM NaCl, pH 8 buffer supplemented with Complete EDTA free (Roche) and 0.1 mM PMSF (Sigma), and disrupted by sonication at 4 °C. After centrifugation at 75000 g at 4 °C for 20 min, the supernatant was loaded on a Ni-NTA column. The eluted protein-containing fractions were pooled and the His-Tag removed by incubation with the TEV protease at 1/100 ratio and an extensive dialysis in a 50 mM Tris-HCl, 150 mM NaCl, 1 mM DTT, 5 mM EDTA, pH 8 buffer. After a second dialysis in a 50 mM Tris-HCl, 150 mM NaCl, pH 8 buffer supplemented with 10 mM imidazole, the sample was loaded on a Ni-NTA column in the same buffer, which allowed to separate the TEV protease and LdcCI/LdcIC. Finally, the pure proteins were obtained by size exclusion chromatography on a Superdex-S200 column in buffer A. LdcI was prepared at 2 mg/ml in a buffer containing 25 mM MES, 100 mM NaCl, 0.2 mM PLP, 1 mM DTT, pH 6.2. 3 μl of sample were applied to glow-discharged quantifoil grids 300 mesh 2/1 (Quantifoil Micro Tools GmbH, Germany), excess solution was blotted during 2.5 s with a Vitrobot (FEI) and the grid frozen in liquid ethane34. Data collection was performed on a FEI Polara microscope operated at 300 kV under low dose conditions. Micrographs were recorded on Kodak SO-163 film at 59,000 magnification, with defocus ranging from 0.6 to 4.9 μm. Films were digitized on a Zeiss scanner (Photoscan) at a step size of 7 μm giving a pixel size of 1.186 Å. The contrast transfer function (CTF) for each micrograph was determined with CTFFIND335. Initially ~2500 particles of 256 × 256 pixels were extracted manually, binned 4 times and subjected to one round of multivariate statistical analysis and classification using IMAGIC36. Representative class averages corresponding to projections in different orientations were used as input for an ab-initio 3D reconstruction by RICOserver (rico.ibs.fr/)37. The resulting 3D reconstruction was refined using RELION38, which yielded an 18 Å resolution map. Using projections of this model as a template, particles of size 256 × 256 pixels were semi-automatically selected from all the micrographs using the Fast Projection Matching (FPM) algorithm39. The resulting dataset of ~46000 particles was processed in RELION with the previous map as an initial model and with a full CTF correction after the first peak. The final map comprised 44207 particles with a resolution of 7.4 Å as per the gold-standard FSC = 0.143 criterion40. It was sharpened with EMBfactor41 using calculated B-factor of −350 Å and masked with a soft mask to obtain a final map with a resolution of 6.1 Å (Fig. S3, Table S1). LdcC was prepared at 2 mg/ml in a buffer containing 25 mM HEPES, 100 mM NaCl, 0.2 mM PLP, 1 mM DTT, pH 7.2. Grids were prepared and sample imaged as LdcIa. Data were processed essentially as LdcIa described above. Briefly, an initial ~20 Å resolution model was generated by angular reconstitution after manual picking of circa 3000 particles from the first micrographs, filtered to 60 Å resolution, refined with RELION and used for a semi-automatic selection with FPM. The dataset was processed in RELION with a full CTF correction to yield a final reconstruction comprising 61000 particles. The map was sharpened with Relion postprocessing, using a soft mask and automated B-factor estimation (−690 Å), yielding a map at 5.5 Å resolution (Fig. S1, Table S1). In our first study23, the dataset was processed in SPIDER and the CTF correction involved a simple phase-flipping. Therefore, for consistency with the present work, we revisited the dataset and processed it in RELION with a full CTF correction after the first peak. It was sharpened with EMBfactor41 using calculated B-factor of −350 Å and masked with a soft mask to obtain a final map with a resolution of 6.2 Å (Fig. S2). This reconstruction is of a slightly better quality in terms of the continuity of the internal density. Therefore we used this improved map for fitting of the atomic model and further analysis (Fig. S2, Table S1). As a crosscheck, each data set was also refined either from other initial models: a “featureless donut” with approximate dimensions of the decamer, and low pass-filtered reconstructions from the two other data sets (i.e. the LdcC reconstruction was used as a model for the LdcIa and LdcI-LARA data sets, etc). All refinements converged to the same solutions independently of the starting model. Local resolution of all maps was determined with ResMap42. 0.4 mg/ml of RavA (in a 20 mM Tris-HCl, 500 mM NaCl, 10 mM MgCl2, 1 mM DTT, 5% glycerol, pH 6.8 buffer) was mixed with 0.3 mg/ml of either LdcI, LdcC, LdcCI or LdcIC in the presence of 2 mM ADP and 10 mM MgCl2 in a buffer containing 20 mM Hepes and 150 mM NaCl at pH 7.4. After 10 minutes incubation at room temperature, 3 μl of mixture were applied to the clear side of the carbon on a carbon-mica interface and negatively stained with 2% uranyl acetate. Images were recorded with a JEOL 1200 EX II microscope at 100 kV at a nominal magnification of 15000 on a CCD camera yielding a pixel size of 4.667 Å. No complexes between RavA and LdcC or LdcIC could be observed, whereas the LdcCI-RavA preparation manifested cage-like particles similar to the previously published LdcI-RavA19, but also unbound RavA and LdcCI, which implies that the affinity of RavA to the LdcCI chimera is lower than its affinity to the native LdcI. 1260 particles of 96 × 96 pixels were extracted interactively from several micrographs. 2D centering, multivariate statistical analysis and classification were performed using IMAGIC36. Class-averages similar to the cage-like LdcI-RavA complex were used as references for multi-reference alignment followed by multivariate statistical analysis and classification. A homology model of LdcC was obtained using the atomic coordinates of the LdcI monomer (3N75) as the template in SWISS-MODEL server43. The RMSD between the template and the resulting model was 0.26 Å. The LdcC model was then fitted as a rigid body into the LdcC cryoEM map using the fit-in-map module of UCSF Chimera44. This rigid fit indicated movements of several parts of the protein. Therefore, the density corresponding to one LdcC monomer was extracted and flexible fitting was performed using IMODFIT45 at 8 Å resolution. This monomeric model was then docked into the decameric cryoEM map with URO46 and its graphical version VEDA47 that use symmetry information for fitting in Fourier space. The Cα RMSDmin between the initial model of the LdcC monomer and the final IMODFIT LdcC model is 1.2 Å. In the case of LdcIa, the density corresponding to an individual monomer was extracted and the fit performed similarly to the one described above, with the final model of the decameric particle obtained with URO and VEDA. The Cα RMSDmin between the LdcIi monomer and the final IMODFIT model is 1.4 Å. For LdcI-LARA, the density accounting for one LdcI monomer bound to a LARA domain was extracted and further separated into individual densities corresponding to LdcI and to LARA. The fit of LdcI was performed as for LdcC and LdcIa, while the crystal structure of LARA was docked into the monomeric LdcI-LARA map as a rigid body using SITUS. The resulting pseudoatomic models were used to create the final model of the LdcI-LARA decamer with URO and VEDA. The Cα RMSDmin between the LdcIi monomer and the final IMODFIT model is 1.4 Å. A brief summary of relevant parameters is provided in Table S1. Out of a non-exhaustive list of 50 species of Enterobacteriaceae (Table S3), 22 were found to contain genes annotated as ldcI or ldcC and containing the ravA-viaA operon (Table S4). An analysis using MUSCLE48 with default parameters showed that these genes share more than 65% identity. To verify annotation of these genes, we compared their genetic environment with that of E. coli ldcI and ldcC. Indeed, in E. coli the ldcI gene is in operon with the cadB gene encoding a lysine-cadaverine antiporter, whereas the ldcC gene is present between the accA gene (encoding an acetyl-CoA carboxylase alpha subunit carboxyltransferase) and the yaeR gene (coding for an unknown protein belonging to the family of Glyoxalase/Dioxygenase/Bleomycin resistance proteins). Compared with this genetic environment, the annotation of several ldcI and ldcC genes in enterobacteria was found to be inconsistent (Table S4). For example, several strains contain genes annotated as ldcC in the genetic background of ldcI and vice versa, as in the case of Salmonella enterica and Trabulsiella guamensi. Furthermore, the gene with an “ldcC-like” environment was found to be annotated as cadA in particular strains of Citrobacter freundii, Cronobacter sakazakii, Enterobacter cloacae subsp. Cloaca, Erwinia amylovora, Pantoea agglomerans, Rahnella aquatilis, Shigella dysenteriae, and Yersinia enterocolitica subsp. enterocolitica, whereas in Hafnia alvei, Kluyvera ascorbata, and Serratia marcescens subsp. marcescens, the gene with an “ldcI-like” environment was found to be annotated as ldcC. In addition, as far as the genetic environment is concerned, Plesiomonas appears to have two ldc genes with the organization of the E. coli ldcI (operon cadA-cadB). Consequently, we corrected for gene annotation consistent with the genetic environment and made multiple sequence alignments using version 8.0.1 of the CLC Genomics Workbench software. A phylogenetic tree was generated based on Maximum Likelihood and following the Neighbor-Joining method with the WAG protein substitution model49. The reliability of the generated phylogenetic tree was assessed by bootstrap analysis. The presented unrooted phylogenetic tree shows the nodes that are reliable over 95% (Fig. 6A). Remarkably, the multiple sequence alignment and the resulting phylogenetic tree clearly grouped together all sequences annotated as ldcI on the one side, and all sequences annotated as ldcC on the other side. Thus, we conclude that all modifications in gene annotations that we introduced for the sake of consistency with the genetic environment are perfectly corroborated by the multiple sequence alignment and the phylogenetic analysis. Since the regulation of the lysine decarboxylase gene (i.e. inducible or constitutive) cannot be assessed by this analysis, we call the resulting groups “ldcI-like” and “ldcC-like” as referred to the E. coli enzymes, and make a parallel between the membership in a given group and the ability of the protein to form a cage complex with RavA. Accession codes: CryoEM maps and corresponding fitted atomic structures (main chain atoms) have been deposited in the Electron Microscopy Data Bank and Protein Data Bank, respectively, with accession codes EMD-3205 and 5FKZ for LdcC, EMD-3204 and 5FKX for LdcIa and EMD-3206 and 5FL2 for LdcI-LARA. How to cite this article: Kandiah, E. et al. Structural insights into the Escherichia coli lysine decarboxylases and molecular determinants of interaction with the AAA+ ATPase RavA. Sci. Rep. 6, 24601; doi: 10.1038/srep24601 (2016).
PMC4872110	Ribosome biogenesis factor Tsr3 is the aminocarboxypropyl transferase responsible for 18S rRNA hypermodification in yeast and humans	The chemically most complex modification in eukaryotic rRNA is the conserved hypermodified nucleotide N1-methyl-N3-aminocarboxypropyl-pseudouridine (macpΨ) located next to the P-site tRNA on the small subunit 18S rRNA. While S-adenosylmethionine was identified as the source of the aminocarboxypropyl (acp) group more than 40 years ago the enzyme catalyzing the acp transfer remained elusive. Here we identify the cytoplasmic ribosome biogenesis protein Tsr3 as the responsible enzyme in yeast and human cells. In functionally impaired Tsr3-mutants, a reduced level of acp modification directly correlates with increased 20S pre-rRNA accumulation. The crystal structure of archaeal Tsr3 homologs revealed the same fold as in SPOUT-class RNA-methyltransferases but a distinct SAM binding mode. This unique SAM binding mode explains why Tsr3 transfers the acp and not the methyl group of SAM to its substrate. Structurally, Tsr3 therefore represents a novel class of acp transferase enzymes.Eukaryotic ribosome biogenesis is highly complex and requires a large number of non-ribosomal proteins and small non-coding RNAs in addition to ribosomal RNAs (rRNAs) and proteins (1). An increasing number of diseases—so called ribosomopathies—are associated with disturbed ribosome biogenesis (2–4). During eukaryotic ribosome biogenesis several dozens of rRNA nucleotides become chemically modified (1). The most abundant rRNA modifications are methylations at the 2′-OH ribose moieties and isomerizations of uridine residues to pseudouridine, catalyzed by small nucleolar ribonucleoprotein particles (snoRNPs) (5,6). In addition, 18S and 25S (yeast)/ 28S (humans) rRNAs contain several base modifications catalyzed by site-specific and snoRNA-independent enzymes. In Saccharomyces cerevisiae 18S rRNA contains four base methylations, two acetylations and a single 3-amino-3-carboxypropyl (acp) modification, whereas six base methylations are present in the 25S rRNA (7). While in humans the 18S rRNA base modifications are highly conserved, only three of the yeast base modifications catalyzed by ScRrp8/HsNML (8), ScRcm1/HsNSUN5 and ScNop2/HsNSUN1 (9,10) are preserved in the corresponding human 28S rRNA. Ribosomal RNA modifications have been suggested to optimize ribosome function, although in most cases this remains to be clearly established. They might contribute to increased RNA stability by providing additional hydrogen bonds (pseudouridines), improved base stacking (pseudouridines and base methylations) or an increased resistance against hydrolysis (ribose methylations) (11–13). Most modified rRNA nucleotides cluster in the vicinity of the decoding or the peptidyl transferase center, suggesting an influence on ribosome functionality and stability (14–16). Defects of rRNA modification enzymes often lead to disturbed ribosome biogenesis or functionally impaired ribosomes, although the lack of individual rRNA modifications often has no or only a slight influence on the cell (16,17). Importantly, malfunctions of several base modifying enzymes are linked to developmental diseases (18,2,19), aging (10) or tumorigenesis (20). The chemically most complex modification is located in the loop capping helix 31 of 18S rRNA (Supplementary Figure S1B). There a uridine (U1191 in yeast) is modified to 1-methyl-3-(3-amino-3-carboxypropyl)-pseudouridine (macpΨ, Figure 1A). This base modification was first described in 1968 for hamster cells (21) and is conserved in eukaryotes. This hypermodified nucleotide, which is located at the P-site tRNA, is synthesized in three steps beginning with the snR35 H/ACA snoRNP guided conversion of uridine into pseudouridine (22). In a second step, the essential SPOUT-class methyltransferase Nep1/Emg1 modifies the pseudouridine to N1-methylpseudouridine (23–25). Methylation can only occur once pseudouridylation has taken place, as the latter reaction generates the substrate for the former. The final acp modification leading to N1-methyl-N3-aminocarboxypropyl-pseudouridine occurs late during 40S biogenesis in the cytoplasm (26,27), while the two former reactions are taking place in the nucleolus and nucleus, and is independent from pseudouridylation or methylation (25). Both the methyl and the acp group are derived from S-adenosylmethionine (SAM) (21), but the enzyme responsible for acp modification remained elusive for more than 40 years. Tsr3 is necessary for acp modification of 18S rRNA in yeast and human. (A) Hypermodified nucleotide macpΨ is synthesized in three steps: pseudouridylation catalyzed by snoRNP35, N1-methylation catalyzed by methyltransferase Nep1 and N3-acp modification catalyzed by Tsr3. The asterisk indicates the C1-atom labeled in the C-incorporation assay. (B) RP-HPLC elution profile of yeast 18S rRNA nucleosides. Hypermodified macpΨ elutes at 7.4 min (wild type, left profile) and is missing in Δtsr3 (middle profile) and Δnep1 Δnop6 mutants (right profile). (C) C-acp labeling of 18S rRNAs. Wild type (WT) and plasmid encoded 18S rRNA (U1191U) show the C-acp signal, whereas the C-acp signal is missing in the U1191A mutant plasmid encoded 18S rRNA (U1191A) and Δtsr3 mutants (Δtsr3). Upper lanes show the ethidium bromide staining of the 18S rRNAs for quantification. All samples were loaded on the gel with two different amounts of 5 and 10 μl. (D) Primer extension analysis of acp modification in yeast 18S rRNA (right gel) including a sequencing ladder (left gel). The primer extension stop at nucleotide 1191 is missing exclusively in Δtsr3 mutants and Δtsr3 Δsnr35 recombinants. (E) Primer extension analysis of human 18S rRNA after siRNA knockdown of HsNEP1/EMG1 (541, 542 and 543) and HsTSR3 (544 and 545) (right gel), including a sequencing ladder (left gel). The primer extension arrest is reduced in HTC116 cells transfected with siRNAs 544 and 545. The efficiency of siRNA mediated HsTSR3 repression correlates with the primer extension signals (see Supplementary Figure S2A). As a loading control, a structural stop is shown (asterisks). Only a few acp transferring enzymes have been characterized until now (28). During the biosynthesis of wybutosine, a tricyclic nucleoside present in eukaryotic and archaeal phenylalanine tRNA (29), Tyw2 (Trm12 in yeast) transfers an acp group from SAM to an acidic carbon atom. Archaeal Tyw2 has a structure very similar to Rossmann-fold (class I) RNA-methyltransferases (30), but its distinctive SAM-binding mode enables the transfer of the acp group instead of the methyl group of the cofactor. Another acp modification has been described in the diphtamide biosynthesis pathway (31), where an acp group is transferred from SAM to the carbon atom of a histidine residue of eukaryotic translation elongation factor 2 by use of a radical mechanism (32,33). In a recent bioinformatic study, the uncharacterized yeast gene YOR006c was predicted to be involved in ribosome biogenesis (34). It is highly conserved among eukaryotes and archaea (Supplementary Figure S1A) and its deletion leads to an accumulation of the 20S pre-rRNA precursor of 18S rRNA, suggesting an influence on D-site cleavage during the maturation of the small ribosomal subunit. On this basis, YOR006C was renamed ‘Twenty S rRNA accumulation 3′ (TSR3). However, its function remained unclear although recently a putative nuclease function during 18S rRNA maturation was predicted (35). Here, we identify Tsr3 as the long-sought acp transferase that catalyzes the last step in the biosynthesis of the hypermodified nucleotide macpΨ in yeast and human cells. Furthermore using catalytically defective mutants of yeast Tsr3 we demonstrated that the acp modification is required for 18S rRNA maturation. Surprisingly, the crystal structures of archaeal homologs revealed that Tsr3 is structurally similar to the SPOUT-class RNA methyltransferases. In contrast, the only other structurally characterized acp transferase enzyme Tyw2 belongs to the Rossmann-fold class of methyltransferase proteins. Interestingly, the two structurally very different enzymes use similar strategies in binding the SAM-cofactor in order to ensure that in contrast to methyltransferases the acp and not the methyl group of SAM is transferred to the substrate. Detailed descriptions are available in Supplementary Data. HCT116(+/+) cells (CCL-247; ATCC) were grown at 37°C in a humidified incubator under 5% CO2 in the McCoy's 5a modified (Sigma-Aldrich)/10% FBS media. All media were supplemented with 50 U/ml penicillin and 50 μg/ml streptomycin (Life Technologies). Reverse transfection of HCT116 cells, DsiRNA inactivation and RT-qPCR using total human RNA are described in Supplementary Data. Detailed descriptions for analytical or preparative separations of ribosomal subunits or polysome gradients are provided in Supplementary Data. 40S subunits from 200 ml yeast culture were isolated by sucrose gradient centrifugation in a SW28 rotor as described above, and precipitated with 2.5 vol of 100% ethanol (−20°C over night). Precipitated 40S subunits were dissolved in water and the 18S rRNA was purified via spin columns (Ambion PureLink RNA Mini Kit). RNA fragments were hydrolysed and dephosphorylated as described by Gehrke and Kuo (36). HPLC analysis of rRNA nucleoside composition was performed using a Supelcosil LC-18S column (Sigma; 250 × 4.6 mm, 5 μm) with a pre-column (4.6 × 20 mm) as previously described (25). To enhance C-labeling, mutants of interest were recombined with a Δmet13 deletion. Resulting strains were cultivated with l-[1-C]-methionine (Hartmann Analytic, 0.1 mCi/ml, 54 mCi/mmol) as described before (25). From isotope labeled cells total RNA was isolated with the PureLink RNA Mini Kit (Ambion) after enzymatic cell lysis with zymolyase. Ribosomal RNAs were separated on a 4% denaturing polyacrylamide gel. After ethidium bromide straining gels were dried and analyzed by autoradiography for 3–5 days using a storage phosphor screen. Signals were visualized with the Typhoon 9100 (GE Healthcare). 5 μg of total yeast RNAs extracted with phenol/chloroform were separated on 1.2% agarose gels in BPTE buffer for 16 h at 60V (37) and afterwards transferred to a Biodyne B membrane by vacuum blotting. Oligonucleotides D/A2 or +1-A0 were radiolabeled using γ-[P]-ATP and T4-polynucleotide kinase and hybridized to the membrane at 37°C. Signals were visualized by phosphoimaging with the Typhoon 9100 (GE Healthcare). RNA extraction from human cells, gel-electrophoresis and northern blotting were performed as described before (38). 20 pmol of oligonucleotide PE-1191 complementary to yeast 18S rRNA nucleotides 1247–1228 were labeled with 50 μCi γ-[P]-ATP using T4-polynucleotide kinase, purified via Sephadex G-25 and annealed to 500 ng of 18S rRNA. Primer annealing and reverse transcription were carried out as described by Sharma et al. (39). After precipitation with ethanol and 3 M NaAc pH 5.2 pellets were washed with 70% ethanol, dried and dissolved in 12 μl formamide loading dye. 2 μl of primer extension samples were separated on sequencing or mini gels which were dried after running and exposed on a storage phosphor screen. Signals were visualized with the Typhoon 9100 (GE Healthcare). Primer extension on human RNA was performed using 5 μg of total RNA with AMV Reverse Transcriptase (Promega) and oligonucleotide PE_1248. Following alkaline hydrolysis, cDNAs were precipitated with ethanol, resuspended in acrylamide loading buffer and separated on a 6% (v/v) denaturing acrylamide gel in 0.5× TBE at 80 W for 1.5 h. After migration, the gels were dried and exposed to Fuji Imaging plates (Fujifilm). The signal was acquired with a Phosphor imager (FLA-7000, Fujifilm). A description of the western blot detection of HA-fused Tsr3 in yeast crude extracts or sucrose gradients fractions is provided in Supplementary Data. For cellular localization Tsr3 was expressed as N-terminal fusion with yEGFP in a yeast strain encoding for ScNop56-mRFP (40). Protein localization in exponentially growing cells was visualized using a Leica TCS SP5. Purified SsTsr3 protein in 25 mM Tris–HCl pH 7.8 250 mM NaCl was mixed with S-[methyl-C]-adenosyl-l-methionine (PerkinElmer; 20 μCi/ml, 58 mCi/mmol) and 0–10 mM non-labeled SAM in a binding buffer (50 mM Tris–HCl pH 7.8, 250 mM NaCl) in a total volume of 50 μl and incubated at 30°C for 10 min. Samples were passed over HAWP02500 membrane filters (Millipore) and unbound C-SAM was removed by washing three times with 5 ml buffer using a vacuum filtering equipment. Filter bound C-SAM was measured by liquid scintillation spectrometry in a Wallac 1401 scintillation counter. Genes coding for archaeal Tsr3 homologs without any tags were obtained commercially (Genscript) in pET11a vectors and overexpressed in Escherichia coli BL21(DE3). Proteins were purified by a combination of heat shock and appropriate column chromatography steps as described in detail in the Supplementary Data. Initial hits for VdTsr3 and SsTsr3 were obtained using the Morpheus Screen (Molecular dimensions) and further refined as described in the Supplementary Data. Diffraction data were collected at the Swiss Light Source (Paul Scherer Institut). The structure of VdTsr3 was determined at 1.6 Å by SAD using a selenomethionine derivative. The structure of SsTsr3 was determined at 2.25 Å by molecular replacement using VdTsr3 as the search model. A detailed description of the data collection, processing, structure calculation and refinement procedures can be found in the Supplementary Data and in Supplementary Table S1. Structures were deposited in the Protein Data Bank as entries 5APG (VdTsr3) and 5AP8 (SsTsr3). For analytical gel filtration experiments a Sephadex S75 10/300 GL column (GE Healthcare) was used. 100 μl protein samples (25 mM Tris–HCl pH 7.8, 250 mM NaCl, 2 mM β-mercaptoethanol) with a protein concentration of 150 μM were used. The flow rate was 0.5 ml/min. The column was calibrated using the marker proteins of the LMW gel filtration calibration kit (GE Healthcare). Protein elution was followed by recording the adsorption at a wavelength of λ = 280 nm. Fluorescence quenching and fluorescence anisotropy measurements were carried out in triplicates at 25°C on a Fluorolog 3 spectrometer (Horiba Jobin Yvon) equipped with polarizers. For fluorescence quenching with SAM, SAH and 5′-methylthioadenosine experiments the tryptophan fluorescence of SsTsr3 (200 nM in 25 mM Tris–HCl pH 7.8, 250 mM NaCl, 2 mM β-mercaptoethanol) was excited at 295 nm and emission spectra were recorded from 250 to 450 nm for each titration step. The fluorescence intensity at 351 nm for each titration step was normalized with regard to the fluorescence of the free protein and was used for deriving binding curves. KD's were derived by nonlinear regression with Origin 8.0 (Origin Labs) using Equation (1): (1)[12pt] } = a }} + 1 (F is the normalized fluorescence intensity, a is the change in fluorescence intensity, c is the ligand concentration and KD is the dissociation constant). 5′-Fluoresceine labeled RNAs for fluorescence anisotropy measurements were obtained commercially (Dharmacon), deprotected according to the manufacturer's protocol and the RNA concentration adjusted to 50 nM in 25 mM Tris–HCl pH 7.8, 250 mM NaCl. Fluoresceine fluorescence was excited at 492 nm and emission was recorded at 516 nm. The data were fitted to Equation (1) (F is the normalized fluorescence anisotropy, a is the change in fluorescence anisotropy). The S. cerevisiae 18S rRNA acp transferase was identified in a systematic genetic screen where numerous deletion mutants from the EUROSCARF strain collection (www.euroscarf.de) were analyzed by HPLC for alterations in 18S rRNA base modifications. For the Δtsr3 deletion strain the HPLC elution profile of 18S rRNA nucleosides (Figure 1B) was very similar to that of the pseudouridine-N1 methyltransferase mutant Δnep1, where a shoulder at ∼ 7.4 min elution time was missing in the elution profile. As previously reported this shoulder was identified by ESI-MS as corresponding to macpΨ (25). In order to directly analyze the presence of the acp modification of nucleotide 1191 we used an in vivoC incorporation assay with 1-C-methionine (25). Whereas the acp labeling of 18S rRNA was clearly present in the wild type strain no radioactive labeling could be observed in a Δtsr3 strain (Figure 1C). No radioactive labeling was detected in the 18S U1191A mutant which served as a control for the specificity of the C-aminocarboxypropyl incorporation. As previously shown, only the acp but none of the other modifications at U1191 of yeast 18S rRNA blocks reverse transcriptase activity. Therefore the presence of the acp modification can be directly assessed by primer extension (16,27). Indeed, in wild-type yeast a strong primer extension stop signal occurred at position 1192. In contrast, in a Δtsr3 mutant no primer extension stop signal was present at this position. As expected, in a Δsnr35 deletion preventing pseudouridylation and N1-methylation (resulting in acpU) as well as in a Δnep1 deletion strain where pseudouridine is not methylated (resulting in acpΨ) (25) a primer extension stop signal of similar intensity as in the wild type was observed. In a Δtsr3 Δsnr35 double deletion strain the 18S rRNA contains an unmodified U and the primer extension stop signal was missing (Figure 1D). The Tsr3 protein is highly conserved in yeast and humans (50% identity). Human 18S rRNA has also been shown to contain macpΨ in the 18S rRNA at position 1248 (41). After siRNA-mediated depletion of Tsr3 in human colon carcinoma HCT116(+/+) cells the acp primer extension arrest was reduced in comparison to cells transfected with a non-targeting scramble siRNA control (Figure 1E, compare lanes 544 and scramble). The efficiency of siRNA-mediated depletion was established by RT-qPCR and found to be very high with siRNA 544 (Supplementary Figure S2A, remaining TSR3 mRNA level of 2%). By comparison, treating cells with siRNA 545, which only reduced the TSR3 mRNA to 20%, did not markedly reduced the acp signal. This suggests that low residual levels of HsTsr3 are sufficient to modify the RNA. As a control for loading, a structural stop is shown (asterisk, Figure 1E). Thus, HsTsr3 is also responsible for the acp modification of 18S rRNA nucleotide Ψ1248 in helix 31. Similar to yeast, siRNA-mediated depletion of the Ψ1248 N1-methyltransferase Nep1/Emg1 had no influence on the primer extension arrest (Figure 1E). Although the acp modification of 18S rRNA is highly conserved in eukaryotes, yeast Δtsr3 mutants showed only a minor growth defect. However, the Δtsr3 deletion was synthetic sick with a Δsnr35 deletion preventing pseudouridylation and Nep1-catalyzed methylation of nucleotide 1191 (Figure 2A). Interestingly, no increased growth defect could be observed for Δtsr3 Δnep1 recombinants containing the nep1 suppressor mutation Δnop6 (42) as well as for Δtsr3 Δsnr35 Δnep1 recombinants with unmodified U1191 (Supplementary Figure S2D and E). Phenotypic characterization of yeast TSR3 deletion (Δtrs3) and human TSR3 depletion (siRNAs 544 and 545) and cellular localization of yeast Tsr3. (A) Growth of yeast wild type, Δtsr3, Δsnr35 and Δtsr3 Δsnr35 segregants after meiosis and tetrad dissection of Δtsr3/TSR3 Δsnr35/SNR35 heterozygous diploids. The Δtsr3 deletion is synthetic sick with a Δsnr35 deletion preventing U1191 pseudouridylation. (B) In agar diffusion assays the yeast Δtsr3 deletion mutant shows a hypersensitivity against paromomycin and hygromycin B which is further increased by recombination with Δsnr35. (C) Northern blot analysis with an ITS1 hybridization probe after siRNA depletion of HsTSR3 (siRNAs 544 and 545) and a scrambled siRNA as control. The accumulation of 18SE and 47S and/or 45S pre-RNAs is enforced upon HsTSR3 depletion. Right gel: Ethidium bromide staining showing 18S and 28S rRNAs. (D) Cytoplasmic localization of yeast Tsr3 shown by fluorescence microscopy of GFP-fused Tsr3. From left to right: differential interference contrast (DIC), green fluorescence of GFP-Tsr3, red fluorescence of Nop56-mRFP as nucleolar marker, and merge of GFP-Tsr3/Nop56-mRFP with DIC. (E) Elution profile (A254) after sucrose gradient separation of yeast ribosomal subunits and polysomes (upper part) and western blot analysis of 3xHA tagged Tsr3 (Tsr3-3xHA) after SDS-PAGE separation of polysome profile fractions taken every 20 s (lower part). The TSR3 gene was genetically modified at its native locus, resulting in a C-terminal fusion of Tsr3 with a 3xHA epitope expressed by the native promotor in yeast strain CEN.BM258-5B. The influence of the acp modification of nucleotide 1191 on ribosome function was analyzed by treating Δtsr3 mutants with protein synthesis inhibitors. Similar to a temperature-sensitive nep1 mutant (43), the Δtsr3 deletion caused hypersensitivity to paromomycin and, to a lesser extent, to hygromycin B (Figure 2B), but not to G418 or cycloheximide (data not shown). In accordance with the synthetic sick growth phenotype the paromomycin and hygromycin B hypersensitivity further increased in a Δtsr3 Δsnr35 recombination strain (Figure 2B). In a yeast Δtsr3 strain as well as in the Δtsr3 Δsnr35 recombinant 20S pre-rRNA accumulated significantly and the level of mature 18S rRNA was reduced (Supplementary Figures S2C and S3D), as reported previously (34). A minor effect on 20S rRNA accumulation was also observed for Δsnr35, but - probably due to different strain backgrounds – to a weaker extent than described earlier (16). In human cells, the depletion of HsTsr3 in HCT116(+/+) cells caused an accumulation of the human 20S pre-rRNA equivalent 18S-E suggesting an evolutionary conserved role of Tsr3 in the late steps of 18S rRNA processing (Figure 2C and Supplementary Figure S2B). Surprisingly, early nucleolar processing reactions were also inhibited, and this was observed in both yeast Δtsr3 cells (see accumulation of 35S in Supplementary Figure S2C) and Tsr3 depleted human cells (see 47S/45S accumulation in Figure 2C and Northern blot quantification in Supplementary Figure S2B). Consistent with its role in late 18S rRNA processing, TSR3 deletion leads to a ribosomal subunit imbalance with a reduced 40S to 60S ratio of 0.81 (σ = 0.024) which was further increased in a Δtsr3 Δsnr35 recombinant to 0.73 (σ = 0.023) (Supplementary Figure S2F). In polysome profiles, a reduced level of 80S ribosomes and a strong signal for free 60S subunits was observed in line with the 40S subunit deficiency (Supplementary Figure S2G). Fluorescence microscopy of GFP-tagged Tsr3 localized the fusion protein in the cytoplasm of yeast cells and no co-localization with the nucleolar marker protein Nop56 could be observed (Figure 2D). This agrees with previous biochemical data suggesting that the acp modification of 18S rRNA occurs late during 40S subunit biogenesis in the cytoplasm (26,27), and makes an additional nuclear localization as reported in a previous large-scale analysis (40) unlikely. After polysome gradient separation C-terminally epitope-labeled Tsr3-3xHA was exclusively detectable in the low-density fraction (Figure 2E). Such distribution on a density gradient suggests that Tsr3 only interacts transiently with pre-40S subunits, which presumably explains why it was not characterized in pre-ribosome affinity purifications. Searches for sequence homologs of S. cerevisiae Tsr3 (ScTsr3) by us and others (44) revealed that the genomes of many archaea contain genes encoding Tsr3-like proteins. However, these archaeal homologs are significantly smaller than ScTsr3 (∼190 aa in archaea vs. 313 aa in yeast) due to shortened N- and C-termini (Supplementary Figure S1A). To locate the domains most important for Tsr3 activity, ScTsr3 fragments of different lengths containing the highly conserved central part were expressed in a Δtsr3 mutant (Figure 3A) and analyzed by primer extension (Figure 3B) and Northern blotting (Figure 3C). N-terminal truncations of up to 45 aa and C-terminal truncations of up to 76 aa mediated acp modification as efficiently as the full-length protein and no significant increased levels of 20S pre-RNA were detected. Even a Tsr3 fragment with a 90 aa C-terminal truncation showed a residual primer extension stop, whereas N-terminal truncations exceeding 46 aa almost completely abolished the primer extension arrest (Figure 3B). Domain characterization of yeast Tsr3 and correlation of acp modification with late 18S rRNA processing steps. (A) Scheme of the TSR3 gene with truncation positions in the open reading frame. TSR3 fragments of different length were expressed under the native promotor from multicopy plasmids in a Δtsr3 deletion strain. (B) Primer extension analysis of 18S rRNA acp modification in yeast cells expressing the indicated TSR3 fragments. N-terminal deletions of 36 or 45 amino acids and C-terminal deletions of 43 or 76 residues show a primer extension stop comparable to the wild type. Tsr3 fragments 37–223 or 46–223 cause a nearly complete loss of the arrest signal. The box highlights the shortest Tsr3 fragment (aa 46–270) with wild type activity (strong primer extension block). (C) Northern blot analysis of 20S pre-rRNA accumulation. A weak 20S rRNA signal, indicating normal processing, is observed for Tsr3 fragment 46–270 (highlighted in a box) showing its functionality. Strong 20S rRNA accumulation similar to that of the Δtsr3 deletion is observed for Tsr3 fragments 37–223 or 46–223. Thus, the archaeal homologs correspond to the functional core of Tsr3. In order to define the structural basis for Tsr3 function, homologs from thermophilic archaea were screened for crystallization. We focused on archaeal species containing a putative Nep1 homolog suggesting that these species are in principle capable of synthesizing N1-methyl-N3-acp-pseudouridine. Well diffracting crystals were obtained for Tsr3 homologs from the two crenarchaeal species Vulcanisaeta distributa (VdTsr3) and Sulfolobus solfataricus (SsTsr3) which share 36% (VdTsr3) and 38% (SsTsr3) identity with the ScTsr3 core region (ScTsr3 aa 46–223). While for S. solfataricus the existence of a modified nucleotide of unknown chemical composition in the loop capping helix 31 of its 16S rRNA has been demonstrated (45), no information regarding rRNA modifications is yet available for V. distributa. Crystals of VdTsr3 diffracted to a resolution of 1.6 Å whereas crystals of SsTsr3 diffracted to 2.25 Å. Serendipitously, VdTsr3 was purified and crystallized in complex with endogenous (E. coli) SAM (Supplementary Figure S4) while SsTsr3 crystals contained the protein in the apo state. The structure of VdTsr3 was solved ab initio, by single-wavelength anomalous diffraction phasing (Se-SAD) with Se containing derivatives (selenomethionine and seleno-substituted SAM). The structure of SsTsr3 was solved by molecular replacement using VdTsr3 as a search model (see Supplementary Table S1 for data collection and refinement statistics). The structure of VdTsr3 can be divided into two domains (Figure 4A). The N-terminal domain (aa 1–92) has a mixed α/β-structure centered around a five-stranded all-parallel β-sheet (Figure 4B) with the strand order β5↑-β3↑-β4↑-β1↑-β2↑. The loops connecting β1 and β2, β3 and β4 and β4 and β5 include α-helices α1, α2 and α3, respectively. The loop connecting β2 and β3 contains a single turn of a 310-helix. Helices α1 and α2 are located on one side of the five-stranded β-sheet while α3 packs against the opposite β-sheet surface. The C-terminal domain (aa 93–184) has a globular all α-helical structure comprising α-helices α4 to α9. Both domains are tightly packed against each other. Remarkably, the entire C-terminal domain (92 aa) of the protein is threaded through the loop which connects β-strand β3 and α-helix α2 of the N-terminal domain. Thus, the VdTsr3 structure contains a deep trefoil knot. The structure of SsTsr3 in the apo state is very similar to that of VdTsr3 (Figure 4C) with an RMSD for equivalent Cα atoms of 1.1 Å. The only significant difference in the global structure of the two proteins is the presence of an extended α-helix α8 and the absence of α-helix α9 in SsTsr3. Tsr3 has a fold similar to SPOUT-class RNA methyltransferases. (A) Cartoon representation of the X-ray structure of VdTsr3 in two orientations. β-strands are colored in crimson whereas α-helices in the N-terminal domain are colored light blue and α-helices in the C-terminal domain are colored dark blue. The bound S-adenosylmethionine is shown in a stick representation and colored by atom type. A red arrow marks the location of the topological knot in the structure. (B) Secondary structure representation of the VdTsr3 structure. The color coding is the same as in (A). (C) Structural superposition of the X-ray structures of VdTsr3 in the SAM-bound state (red) and SsTsr3 (blue) in the apo state. The locations of the α-helix α8 which is longer in SsTsr3 and of α-helix α9 which is only present in VdTsr3 are indicated. (D) Secondary structure cartoon (left) of S. pombe Trm10 (pdb4jwf)—the SPOUT-class RNA methyltransferase structurally most similar to Tsr3 and superposition of the VdTsr3 and Trm10 X-ray structures (right). (E) Analytical gel filtration profiles for VdTsr3 (red) and SsTsr3 (blue) show that both proteins are monomeric in solution. Vertical lines indicate the elution volumes of molecular weight markers. Vd, Vulcanisaeta distributa; Ss, Sulfolobus solfataricus. Structure predictions suggested that Tsr3 might contain a so-called RLI domain which contains a ‘bacterial like’ ferredoxin fold and binds two iron-sulfur clusters through eight conserved cysteine residues (46). However, no structural similarity to an RLI-domain was detectable. This is in accordance with the functional analysis of alanine replacement mutations of cysteine residues in ScTsr3 (Supplementary Figure S3). The β-strand topology and the deep C-terminal trefoil knot of archaeal Tsr3 are the structural hallmarks of the SPOUT-class RNA-methyltransferase fold. The closest structural homolog identified in a DALI search is the tRNA methyltransferase Trm10 (DALI Z-score 6.8) which methylates the N1 nitrogen of G9/A9 in many archaeal and eukaryotic tRNAs by using SAM as the methyl group donor (47,48). In comparison to Tsr3 the central β-sheet element of Trm10 is extended by one additional β-strand pairing to β2. Furthermore, the trefoil knot of Trm10 is not as deep as that of Tsr3 (Figure 4D). Interestingly, Nep1—the enzyme preceding Tsr3 in the biosynthetic pathway for the synthesis of macpΨ—also belongs to the SPOUT-class of RNA methyltransferases. However, the structural similarities between Nep1 and Tsr3 (DALI Z-score 4.4) are less pronounced than between Tsr3 and Trm10. Most SPOUT-class RNA-methyltransferases are homodimers. A notable exception is Trm10. Gel filtration experiments with both VdTsr3 and SsTsr3 (Figure 4E) showed that both proteins are monomeric in solution thereby extending the structural similarities to Trm10. So far, structural information is only available for one other enzyme that transfers the acp group from SAM to an RNA nucleotide. This enzyme, Tyw2, is part of the biosynthesis pathway of wybutosine nucleotides in tRNAs. However, there are no structural similarities between Tsr3 and Tyw2, which contains an all-parallel β-sheet of a different topology and no knot structure (30). Instead, Tyw2 has a fold typical for the class-I-or Rossmann-fold class of methyltransferases (Supplementary Figure S5B). The SAM-binding site of Tsr3 is located in a deep crevice between the N- and C-terminal domains in the vicinity of the trefoil knot as typical for SPOUT-class RNA-methyltransferases (Figure 4A). The adenine base of the cofactor is recognized by hydrogen bonds between its N1 nitrogen and the backbone amide of L93 directly preceding β5 as well as between its N6-amino group and the backbone carbonyl group of Y108 located in the loop connecting β5 in the N-terminal and α4 in the C-terminal domain (Figure 5A). Furthermore, the adenine base of SAM is involved in hydrophobic packing interactions with the side chains of L45 (β3), P47 and W73 (α3) in the N-terminal domain as well as with L93, L110 (both in the loop connecting β5 and α4) and A115 (α5) in the C-terminal domain. The ribose 2′ and 3′ hydroxyl groups of SAM are hydrogen bonded to the backbone carbonyl group of I69. The acp side chain of SAM is fixed in position by hydrogen bonding of its carboxylate group to the backbone amide and the side chain hydroxyl group of T19 in α1 as well as the backbone amide group of T112 in α4 (C-terminal domain). Most importantly, the methyl group of SAM is buried in a hydrophobic pocket formed by the sidechains of W73 and A76 both located in α3 (Figure 5A and B). W73 is highly conserved in all known Tsr3 proteins, whereas A76 can be replaced by other hydrophobic amino acids. Consequently, the accessibility of this methyl group for a nucleophilic attack is strongly reduced in comparison with RNA-methyltransferases such as Trm10 (Figure 5B, C). In contrast, the acp side chain of SAM is accessible for reactions in the Tsr3-bound state (Figure 5B). SAM-binding by Tsr3. (A) Close-up view of the SAM-binding pocket of VdTsr3. Nitrogen atoms are dark blue, oxygen atoms red, sulfur atoms orange, carbon atoms of the protein light blue and carbon atoms of SAM yellow. Hydrogen bonds are indicated by dashed lines. (B) Solvent accessibility of the acp group of SAM bound to VdTsr3. The solvent accessible surface of the protein is shown in semitransparent gray whereas SAM is show in a stick representation. Atoms are colored as in (A). A red arrow indicates the reactive CH2-moiety of the acp group. (C) Solvent accessibility of the SAM methyl group for SAM bound to the RNA methyltransferase Trm10. Bound SAM was modelled based on the X-ray structure of the Trm10/SAH-complex (pdb4jwf). A red arrow indicates the SAM methyl group. (D) Binding of SAM analogs to SsTsr3. Tryptophan fluorescence quenching curves upon addition of SAM (blue), 5′-methyl-thioadenosine (red) and SAH (black). (E) Binding of C-labeled SAM to SsTsr3. Radioactively labeled SAM is retained on a filter in the presence of SsTsr3. Addition of unlabeled SAM competes with the binding of labeled SAM. A W66A-mutant of SsTsr3 (W73 in VdTsr3) does not bind SAM. (F) Primer extension (upper left) shows a strongly reduced acp modification of yeast 18S rRNA in Δtsr3 cells expressing Tsr3-S62D, -E111A or –W114A. This correlates with a 20S pre-rRNA accumulation comparable to the Δtsr3 deletion (right: northern blot). 3xHA tagged Tsr3 mutants are expressed comparable to the wild type as shown by western blot (lower left). Binding affinities for SAM and its analogs 5′-methylthioadenosin and SAH to SsTsr3 were measured using tryptophan fluorescence quenching. VdTsr3 could not be used in these experiments since we could not purify it in a stable SAM-free form. SsTsr3 bound SAM with a KD of 6.5 μM, which is similar to SAM-KD's reported for several SPOUT-class methyltransferases. 5′-methylthioadenosin—the reaction product after the acp-transfer—binds only ∼2.5-fold weaker (KD = 16.7 μM) compared to SAM. S-adenosylhomocysteine which lacks the methyl group of SAM binds with significantly lower affinity (KD = 55.5 μM) (Figure 5D). This suggests that the hydrophobic interaction between SAM's methyl group and the hydrophobic pocket of Tsr3 is thermodynamically important for the interaction. On the other hand, the loss of hydrogen bonds between the acp sidechain carboxylate group and the protein appears to be thermodynamically less important but these hydrogen bonds might play a crucial role for the proper orientation of the cofactor side chain in the substrate binding pocket. Accordingly, a W66A-mutation (W73 in VdTsr3) of SsTsr3 significantly diminished SAM-binding in a filter binding assay compared to the wild type (Figure 5E). Furthermore, a W to A mutation at the equivalent position W114 in ScTsr3 strongly reduced the in vivo acp transferase activity (Figure 5F). The side chain hydroxyl group of T19 seems of minor importance for SAM binding since mutations of T17 (T19 in VdTsr3) to either A or D did not significantly influence the SAM-binding affinity of SsTsr3 (KD's = 3.9 or 11.2 mM, respectively). Nevertheless, a mutation of the equivalent position S62 of ScTsr3 to D, but not to A, resulted in reduced acp modification in vivo, as shown by primer extension analysis (Figure 5F). The acp-transfer reaction catalyzed by Tsr3 most likely requires the presence of a catalytic base in order to abstract a proton from the N3 imino group of the modified pseudouridine. The side chain of D70 (VdTsr3) located in β4 is ∼5 Å away from the SAM sulfur atom. This residue is conserved as D or E both in archaeal and eukaryotic Tsr3 homologs. Mutations of the corresponding residue in SsTsr3 to A (D63) does not significantly alter the SAM-binding affinity of the protein (KD = 11.0 μM). However, the mutation of the corresponding residue of ScTsr3 (E111A) leads to a significant decrease of the acp transferase activity in vivo (Figure 5F). Analysis of the electrostatic surface properties of VdTsr3 clearly identified positively charged surface patches in the vicinity of the SAM-binding site suggesting a putative RNA-binding site (Figure 6A). Furthermore, a negatively charged MES-ion is found in the crystal structure of VdTsr3 complexed to the side chain of K22 in helix α1. Its negatively charged sulfate group might mimic an RNA backbone phosphate. Helix α1 contains two more positively charged amino acids K17 and R25 as does the loop preceding it (R9). A second cluster of positively charged residues is found in or near helix α3 (K74, R75, K82, R85 and K87). Some of these amino acids are conserved between archaeal and eukaryotic Tsr3 (Supplementary Figure S1A). In the C-terminal domain, the surface exposed α-helices α5 and α7 carry a significant amount of positively charged amino acids. A triple mutation of the conserved positively charged residues R60, K65 and R131 to A in ScTsr3 resulted in a protein with a significantly impaired acp transferase activity in vivo (Figure 6D) in line with an important functional role for these positively charged residues. RNA-binding of Tsr3. (A) Electrostatic charge distribution on the surface of VdTsr3. Surface areas colored in blue are positively charged whereas red areas are negatively charged. SAM is shown in a stick representation. Also shown in stick representation is a negatively charged MES ion. Conserved basic amino acids are labeled. (B) Comparison of the secondary structures of helix 31 from the small ribosomal subunit rRNAs in S. cerevisiae and S. solfataricus with the location of the hypermodified nucleotide indicated in red. For S. solfataricus the chemical identity of the hypermodified nucleotide is not known but the existence of NEP1 and TSR3 homologs suggest that it is indeed N1-methyl-N3-acp-pseudouridine. (C) Binding of SsTsr3 to RNA. 5′-fluoresceine labeled RNA oligonucleotides corresponding either to the native (20mer – see inset) or a stabilized (20mer_GC - inset) helix 31 of the small ribosomal subunit rRNA from S. solfataricus were titrated with increasing amounts of SsTsr3 and the changes in the fluoresceine fluorescence anisotropy were measured and fitted to a binding curve (20mer – red, 20mer_GC – blue). Oligo-U9-RNA was used for comparison (black). The 20mer_GC RNA was also titrated with SsTsr3 in the presence of 2 mM SAM (purple). (D) Mutants of ScTsr3 R60, K65 or R131 (equivalent to K17, K22 and R91 in VdTsr3) expressed in Δtsr3 yeast cells show a primer extension stop comparable to the wild type. Combination of the three point mutations (R60A/K65A/R131A) leads to a strongly reduced acp modification of 18S rRNA. In order to explore the RNA-ligand specificity of Tsr3 we titrated SsTsr3 prepared in RNase-free form with 5′-fluoresceine-labeled RNA and determined the affinity by fluorescence anisotropy measurements. SsTsr3 in the apo state bound a 20mer RNA corresponding to helix 31 of S. solfataricus 16S rRNA (Figure 6B) with a KD of 1.9 μM and to a version of this hairpin stabilized by additional GC base pairs (20mer-GC) with a KD of 0.6 μM (Figure 6C). A single stranded oligoU-RNA bound with a 10-fold-reduced affinity (6.0 μM). The presence of saturating amounts of SAM (2 mM) did not have a significant influence on the RNA-affinity of SsTsr3 (KD of 1.7 μM for the 20mer-GC-RNA) suggesting no cooperativity in substrate binding. U1191 is the only hypermodified base in the yeast 18S rRNA and is strongly conserved in eukaryotes (21,22,26). The formation of 1-methyl-3-(3-amino-3-carboxypropyl)-pseudouridine (macpΨ) is very complex requiring three successive modification reactions involving one H/ACA snoRNP (snR35) and two protein enzymes (Nep1/Emg1 and Tsr3). This makes it unique in eukaryotic rRNA modification. The macpΨ base is located at the tip of helix 31 on the 18S rRNA (Supplementary Figure S1B) which, together with helices 18, 24, 34 and 44, contribute to building the decoding center of the small ribosomal subunit (49). A similar modification (acpU) was identified in Haloferax volcanii (50) and corresponding modified nucleotides were also shown to occur in other archaea (45,51). As shown here TSR3 encodes the transferase catalyzing the acp modification as the last step in the biosynthesis of macpΨ in yeast and human cells. Unexpectedly, archaeal Tsr3 has a structure similar to SPOUT-class RNA methyltransferases, and it is the first example for an enzyme of this class transferring an acp group, due to a modified SAM-binding pocket that exposes the acp instead of the methyl group of SAM to its RNA substrate. Similar to the structurally unrelated Rossmann-fold Tyw2 acp transferase, the SAM methyl group of Tsr3 is bound in an inaccessible hydrophobic pocket whereas the acp side chain becomes accessible for a nucleophilic attack by the N3 of pseudouridine. In contrast, in the structurally closely related RNA methyltransferase Trm10 the methyl group of the cofactor SAM is accessible whereas its acp side chain is buried inside the protein. This suggests that enzymes with a SAM-dependent acp transferase activity might have evolved from SAM-dependent methyltransferases by slight modifications of the SAM-binding pocket. Thus, additional examples for acp transferase enzymes might be found with similarities to other structural classes of methyltransferases. In contrast to Nep1 (24), the enzyme preceding Tsr3 in the macpΨ biosynthesis pathway, Tsr3 binds rather weakly and with little specificity to its isolated substrate RNA. This suggests that Tsr3 is not stably incorporated into pre-ribosomal particles and that its binding to the nascent ribosomal subunit possibly requires additional interactions with other pre-ribosomal components. Consistently, in sucrose gradient analysis, Tsr3 was found in low-molecular weight fractions rather than with pre-ribosome containing high-molecular weight fractions. In contrast to several enzymes that catalyze base specific modifications in rRNAs Tsr3 is not an essential protein. Typically, other small subunit rRNA methyltransferases as Dim1, Bud23 and Nep1/Emg1 carry dual functions, in ribosome biogenesis and rRNA modification, and it is their involvement in pre-RNA processing that is essential rather than their RNA-methylating activity (25,52–55, discussed in 7). In contrast, for several Tsr3 mutants (SAM-binding and cysteine mutations) we found a systematic correlation between the loss of acp modification and the efficiency of 18S rRNA maturation. This demonstrates that, unlike the other small subunit rRNA base modifications, the acp modification is required for efficient pre-rRNA processing. Recently, structural, functional, and CRAC (cross-linking and cDNA analysis) experiments of late assembly factors involved in cytoplasmic processing of 40S subunits, along with cryo-EM studies of the late pre-40S subunits have provided important insights into late pre-40S processing (56–58). Apart from most of the ribosomal proteins, cytoplasmic pre-40S particles contain 20S rRNA and at least seven non-ribosomal proteins including the D-site endonuclease Nob1 as well as Tsr1, a putative GTPase and Rio2 which block the mRNA channel and the initiator tRNA binding site, respectively, thus preventing translation initiation. After structural changes, possibly driven by GTP hydrolysis, which go together with the formation of the decoding site, the 20S pre-rRNA becomes accessible for Nob1 cleavage at site D. This also involves joining of pre-40S and 60S subunits to 80S-like particles in a translation-like cycle promoted by eIF5B (59,60). The cleavage step most likely acts as a quality control check that ensures the proper 40S subunit assembly with only completely processed precursors (59). Finally, termination factor Rli1, an ATPase, promotes the dissociation of assembly factors and the 80S-like complex dissociates and releases the mature 40S subunit (60). Interestingly, differences in the level of acp modification were demonstrated for different steps of the cytoplasmic pre-40S subunit maturation after analyzing purified 20S pre-rRNAs using different purification bait proteins. Early cytoplasmic pre-40S subunits still containing the ribosome assembly factors Tsr1, Ltv1, Enp1 and Rio2 were not or only partially acp modified. In contrast, late pre-40S subunits containing Nob1 and Rio1 or already associated with 60S subunits in 80S-like particles showed acp modification levels comparable to mature 40S subunits. Thus, the acp transfer to mΨ1191 occurs during the step at which Rio2 leaves the pre-40S particle (27). These data and the finding that a missing acp modification hinders pre-20S rRNA processing, suggest that the acp modification together with the release of Rio2 promotes the formation of the decoding site and thus D-site cleavage by Nob1. The interrelation between acp modification and Rio2 release is also supported by CRAC analysis showing that Rio2 binds to helix 31 next to the Ψ1191 residue that receives the acp modification (56). Therefore, Rio2 either blocks the access of Tsr3 to helix 31, and acp modification can only occur after Rio2 is released, or the acp modification of mΨ1191 and putative subsequent conformational changes of 20S rRNA weaken the binding of Rio2 to helix 31 and support its release from the pre-rRNA. In summary, by identifying Tsr3 as the enzyme responsible for introducing the acp group to the hypermodified macpΨ nucleotide at position 1191 (yeast)/ 1248 (humans) of 18S rRNA we added one of the last remaining pieces to the puzzle of eukaryotic small ribosomal subunit rRNA modifications. The current data together with the finding that acp modification takes place at the very last step in pre-40S subunit maturation (27) indicate that the acp modification probably supports the formation of the decoding site and efficient 20S pre-rRNA D-site cleavage. Furthermore, our structural data unravelled how the regioselectivity of SAM-dependent group transfer reactions can be tuned by distinct small evolutionary adaptions of the ligand binding pocket of SAM-binding enzymes. Coordinates and structure factors have been deposited in the Protein Data Bank under accession codes PDB 5APG (VdTsr3/SAM-complex) and PDB 5AP8 (SsTsr3).
PMC4784909	The Structural Basis of Coenzyme A Recycling in a Bacterial Organelle	Bacterial Microcompartments (BMCs) are proteinaceous organelles that encapsulate critical segments of autotrophic and heterotrophic metabolic pathways; they are functionally diverse and are found across 23 different phyla. The majority of catabolic BMCs (metabolosomes) compartmentalize a common core of enzymes to metabolize compounds via a toxic and/or volatile aldehyde intermediate. The core enzyme phosphotransacylase (PTAC) recycles Coenzyme A and generates an acyl phosphate that can serve as an energy source. The PTAC predominantly associated with metabolosomes (PduL) has no sequence homology to the PTAC ubiquitous among fermentative bacteria (Pta). Here, we report two high-resolution PduL crystal structures with bound substrates. The PduL fold is unrelated to that of Pta; it contains a dimetal active site involved in a catalytic mechanism distinct from that of the housekeeping PTAC. Accordingly, PduL and Pta exemplify functional, but not structural, convergent evolution. The PduL structure, in the context of the catalytic core, completes our understanding of the structural basis of cofactor recycling in the metabolosome lumen.Bacterial Microcompartments (BMCs) are organelles that encapsulate enzymes for sequential biochemical reactions within a protein shell [1–4]. The shell is typically composed of three types of protein subunits, which form either hexagonal (BMC-H and BMC-T) or pentagonal (BMC-P) tiles that assemble into a polyhedral shell. The facets of the shell are composed primarily of hexamers that are typically perforated by pores lined with highly conserved, polar residues that presumably function as the conduits for metabolites into and out of the shell . The vitamin B12-dependent propanediol-utilizing (PDU) BMC was one of the first functionally characterized catabolic BMCs ; subsequently, other types have been implicated in the degradation of ethanolamine, choline, fucose, rhamnose, and ethanol, all of which produce different aldehyde intermediates (Table 1). More recently, bioinformatic studies have demonstrated the widespread distribution of BMCs among diverse bacterial phyla and grouped them into 23 different functional types . The reactions carried out in the majority of catabolic BMCs (also known as metabolosomes) fit a generalized biochemical paradigm for the oxidation of aldehydes (Fig 1) . This involves a BMC-encapsulated signature enzyme that generates a toxic and/or volatile aldehyde that the BMC shell sequesters from the cytosol . The aldehyde is subsequently converted into an acyl-CoA by aldehyde dehydrogenase, which uses NAD and CoA as cofactors . These two cofactors are relatively large, and their diffusion across the protein shell is thought to be restricted, necessitating their regeneration within the BMC lumen . NAD is recycled via alcohol dehydrogenase , and CoA is recycled via phosphotransacetylase (PTAC) (Fig 1). The final product of the BMC, an acyl-phosphate, can then be used to generate ATP via acyl kinase, or revert back to acyl-CoA by Pta for biosynthesis. Collectively, the aldehyde and alcohol dehydrogenases, as well as the PTAC, constitute the common metabolosome core. Substrates and cofactors involving the PTAC reaction are shown in red; other substrates and enzymes are shown in black, and other cofactors are shown in gray. * PduL from these functional types of metabolosomes were purified in this study. The activities of core enzymes are not confined to BMC-associated functions: aldehyde and alcohol dehydrogenases are utilized in diverse metabolic reactions, and PTAC catalyzes a key biochemical reaction in the process of obtaining energy during fermentation . The concerted functioning of a PTAC and an acetate kinase (Ack) is crucial for ATP generation in the fermentation of pyruvate to acetate (see Reactions 1 and 2). Both enzymes are, however, not restricted to fermentative organisms. They can also work in the reverse direction to activate acetate to the CoA-thioester. This occurs, for example, during acetoclastic methanogenesis in the archaeal Methanosarcina species . Reaction 1: acetyl-S-CoA + Pi ←→ acetyl phosphate + CoA-SH (PTAC) Reaction 2: acetyl phosphate + ADP ←→ acetate + ATP (Ack) The canonical PTAC, Pta, is an ancient enzyme found in some eukaryotes and archaea , and widespread among the bacteria; 90% of the bacterial genomes in the Integrated Microbial Genomes database contain a gene encoding the PTA_PTB phosphotransacylase (Pfam domain PF01515 ). Pta has been extensively characterized due to its key role in fermentation . More recently, a second type of PTAC without any sequence homology to Pta was identified . This protein, PduL (Pfam domain PF06130), was shown to catalyze the conversion of propionyl-CoA to propionyl-phosphate and is associated with a BMC involved in propanediol utilization, the PDU BMC . Both pduL and pta genes can be found in genetic loci of functionally distinct BMCs, although the PduL type is much more prevalent, being found in all but one type of metabolosome locus: EUT1 (Table 1) . Furthermore, in the Integrated Microbial Genomes Database , 91% of genomes that encode PF06130 also encode genes for shell proteins. As a member of the core biochemical machinery of functionally diverse aldehyde-oxidizing metabolosomes, PduL must have a certain level of substrate plasticity (see Table 1) that is not required of Pta, which has generally been observed to prefer acetyl-CoA . PduL from the PDU BMC of Salmonella enterica favors propionyl-CoA over acetyl-CoA , and it is likely that PduL orthologs in functionally diverse BMCs would have substrate preferences for other CoA derivatives. Another distinctive feature of BMC-associated PduL homologs is an N-terminal encapsulation peptide (EP) that is thought to “target” proteins for encapsulation by the BMC shell . EPs are frequently found on BMC-associated proteins and have been shown to interact with shell proteins . EPs have also been observed to cause proteins to aggregate , and this has recently been suggested to be functionally relevant as an initial step in metabolosome assembly, in which a multifunctional protein core is formed, around which the shell assembles . Of the three common metabolosome core enzymes, crystal structures are available for both the alcohol and aldehyde dehydrogenases. In contrast, the structure of PduL, the PTAC found in the vast majority of catabolic BMCs, has not been determined. This is a major gap in our understanding of metabolosome-encapsulated biochemistry and cofactor recycling. Structural information will be essential to working out how the core enzymes and their cofactors assemble and organize within the organelle lumen to enhance catalysis. Moreover, it will be useful for guiding efforts to engineer novel BMC cores for biotechnological applications . The primary structure of PduL homologs is subdivided into two PF06130 domains, each roughly 80 residues in length. No available protein structures contain the PF06130 domain, and homology searches using the primary structure of PduL do not return any significant results that would allow prediction of the structure. Moreover, the evident novelty of PduL makes its structure interesting in the context of convergent evolution of PTAC function; to-date, only the Pta active site and catalytic mechanism is known . Here we report high-resolution crystal structures of a PduL-type PTAC in both CoA- and phosphate-bound forms, completing our understanding of the structural basis of catalysis by the metabolosome common core enzymes. We propose a catalytic mechanism analogous but yet distinct from the ubiquitous Pta enzyme, highlighting the functional convergence of two enzymes with completely different structures and metal requirements. We also investigate the quaternary structures of three different PduL homologs and situate our findings in the context of organelle biogenesis in functionally diverse BMCs. We cloned, expressed, and purified three different PduL homologs from functionally distinct BMCs (Table 1): from the well-studied pdu locus in S. enterica Typhimurium LT2 (sPduL) , from the recently characterized pvm locus in Planctomyces limnophilus (pPduL) , and from the grm3 locus in Rhodopseudomonas palustris BisB18 (rPduL) . While purifying full-length sPduL, we observed a tendency to aggregation as described previously , with a large fraction of the expressed protein found in the insoluble fraction in a white, cake-like pellet. Remarkably, after removing the N-terminal putative EP (27 amino acids), most of the sPduLΔEP protein was in the soluble fraction upon cell lysis. Similar differences in solubility were observed for pPduL and rPduL when comparing EP-truncated forms to the full-length protein, but none were quite as dramatic as for sPduL. We confirmed that all homologs were active (S1a and S1b Fig). Among these, we were only able to obtain diffraction-quality crystals of rPduL after removing the N-terminal putative EP (33 amino acids, also see Fig 2a) (rPduLΔEP). Truncated rPduLΔEP had comparable enzymatic activity to the full-length enzyme (S1a Fig). (a) Primary and secondary structure of rPduL (tubes represent α-helices, arrows β-sheets and dashed line residues disordered in the structure. Blocks of ten residues are shaded alternatively black/dark gray. The first 33 amino acids are present only in the wildtype construct and contains the predicted EP alpha helix, α0); the truncated rPduLΔEP that was crystallized begins with M-G-V. Coloring is according to structural domains (domain 1 D36-N46/Q155-C224, blue; loop insertion G61-E81, grey; domain 2 R47-F60/E82-A154, red). Metal coordination residues are highlighted in light blue and CoA contacting residues in magenta, residues contacting the CoA of the other chain are also outlined. (b) Cartoon representation of the structure colored by domains and including secondary structure numbering. The N-and C-termini are in close proximity. Coenzyme A is shown in magenta sticks and Zinc (grey) as spheres. We collected a native dataset from rPduLΔEP crystals diffracting to a resolution of 1.54 Å (Table 2). Using a mercury-derivative crystal form diffracting to 1.99 Å (Table 2), we obtained high quality electron density for model building and used the initial model to refine against the native data to Rwork/Rfree values of 18.9/22.1%. There are two PduL molecules in the asymmetric unit of the P212121 unit cell. We were able to fit all of the primary structure of PduLΔEP into the electron density with the exception of three amino acids at the N-terminus and two amino acids at the C-terminus (Fig 2a); the model is of excellent quality (Table 2). A CoA cofactor as well as two metal ions are clearly resolved in the density (for omit maps of CoA see S2 Fig). *Highest resolution shell is shown in parentheses. Structurally, PduL consists of two domains (Fig 2, blue/red), each a beta-barrel that is capped on both ends by short α-helices. β-Barrel 1 consists of the N-terminal β strand and β strands from the C-terminal half of the polypeptide chain (β1, β10-β14; residues 37–46 and 155–224). β-Barrel 2 consists mainly of the central segment of primary structure (β2, β5–β9; residues 47–60 and 82–154) (Fig 2, red), but is interrupted by a short two-strand beta sheet (β3-β4, residues 61–81). This β-sheet is involved in contacts between the two domains and forms a lid over the active site. Residues in this region (Gln42, Pro43, Gly44), covering the active site, are strongly conserved (Fig 3). This structural arrangement is completely different from the functionally related Pta, which is composed of two domains, each consisting of a central flat beta sheet with alpha-helices on the top and bottom . Sequence logo calculated from the multiple sequence alignment of PduL homologs (see Materials and Methods), but not including putative EP sequences. Residues 100% conserved across all PduL homologs in our dataset are noted with an asterisk, and residues conserved in over 90% of sequences are noted with a colon. The sequences aligning to the PF06130 domain (determined by BLAST) are highlighted in red and blue. The position numbers shown correspond to the residue numbering of rPduL; note that some positions in the logo represent gaps in the rPduL sequence. There are two PduL molecules in the asymmetric unit forming a butterfly-shaped dimer (Fig 4c). Consistent with this, results from size exclusion chromatography of rPduLΔEP suggest that it is a dimer in solution (Fig 5e). The interface between the two chains buries 882 Å per monomer and is mainly formed by α-helices 2 and 4 and parts of β-sheets 12 and 14, as well as a π–π stacking of the adenine moiety of CoA with Phe116 of the adjacent chain (Fig 4c). The folds of the two chains in the asymmetric unit are very similar, superimposing with a rmsd of 0.16 Å over 2,306 aligned atom pairs. The peripheral helices and the short antiparallel β3–4 sheet mediate most of the crystal contacts. (a,b) Proposed active site of PduL with relevant residues shown as sticks in atom coloring (nitrogen blue, oxygen red, sulfur yellow), zinc as grey colored spheres and coordinating ordered water molecules in red. Distances between atom centers are indicated in Å. (a) Coenzyme A containing, (b) phosphate-bound structure. (c) View of the dimer in the asymmetric unit from the side, domains 1 and 2 colored as in Fig 2 and the two chains differentiated by blue/red versus slate/firebrick. The bottom panel shows a top view down the 2-fold axis as indicated by the arrow in the top panel. The asterisk and double arrow marks the location of the π–π interaction between F116 and the CoA base of the other dimer chain. (d) Surface representation of the structure with indicated conservation (red: high, white: intermediate, yellow: low). (a)–(c): Chromatograms of sPduL (a), rPduL (b), and pPduL (c) with (orange) or without (blue) the predicted EP, post-nickel affinity purification, applied over a preparative size exclusion column (see Materials and Methods). (d)–(f): Chromatograms of sPduL (d), rPduL (e), and pPduL (f) post-preparative size exclusion chromatography with different size fractions separated, applied over an analytical size exclusion column (see Materials and Methods). All chromatograms are cropped to show only the linear range of separation based on standard runs, shown in black squares with a dashed linear trend line. All y-axes are arbitrary absorbance units except the right-hand axes for panels (a) and (d), which is the log10(molecular weight) of the standards. CoA and the metal ions bind between the two domains, presumably in the active site (Figs 2b and 4a). To identify the bound metals, we performed an X-ray fluorescence scan on the crystals at various wavelengths (corresponding to the K-edges of Mn, Fe, Co, Ni, Cu, and Zn). There was a large signal at the zinc edge, and we tested for the presence of zinc by collecting full data sets before and after the Zn K-edge (1.2861 and 1.2822 Å, respectively). The large differences between the anomalous signals confirm the presence of zinc at both metal sites (S3 Fig). The first zinc ion (Zn1) is in a tetrahedral coordination state with His48, His50, Glu109, and the CoA sulfur (Fig 4a). The second (Zn2) is in octahedral coordination by three conserved histidine residues (His157, His159 and His204) as well as three water molecules (Fig 4a). The nitrogen atom coordinating the zinc is the Nε in each histidine residue, as is typical for this interaction . When the crystals were soaked in a sodium phosphate solution for 2 d prior to data collection, the CoA dissociates, and density for a phosphate molecule is visible at the active site (Table 2, Fig 4b). The phosphate-bound structure aligns well with the CoA-bound structure (0.43 Å rmsd over 2,361 atoms for the monomer, 0.83 Å over 5,259 aligned atoms for the dimer). The phosphate contacts both zinc atoms (Fig 4b) and replaces the coordination by CoA at Zn1; the coordination for Zn2 changes from octahedral with three bound waters to tetrahedral with a phosphate ion as one of the ligands (Fig 4b). Conserved Arg103 seems to be involved in maintaining the phosphate in that position. The two zinc atoms are slightly closer together in the phosphate-bound form (5.8 Å vs 6.3 Å), possibly due to the bridging effect of the phosphate. An additional phosphate molecule is bound at a crystal contact interface, perhaps accounting for the 14 Å shorter c-axis in the phosphate-bound crystal form (Table 2). Interestingly, some of the residues important for dimerization of rPduL, particularly Phe116, are poorly conserved across PduL homologs associated with functionally diverse BMCs (Figs 4c and 3), suggesting that they may have alternative oligomeric states. We tested this hypothesis by performing size exclusion chromatography on both full-length and truncated variants (lacking the EP, ΔEP) of sPduL, rPduL, and pPduL. These three homologs are found in functionally distinct BMCs (Table 1). Therefore, they are packaged with different signature enzymes and different ancillary proteins . It has been proposed that the catabolic BMCs may assemble in a core-first manner, with the luminal enzymes (signature enzyme, aldehyde, and alcohol dehydrogenases and the BMC PTAC) forming an initial bolus, or prometabolosome, around which a shell assembles . Given the diversity of signature enzymes (Table 1), it is plausible that PduL orthologs may adopt different oligomeric states that reflect the differences in the proteins being packaged with them in the organelle lumen. We found that not only did the different orthologs appear to assemble into different oligomeric states, but that quaternary structure was dependent on whether or not the EP was present. Full-length sPduL was unstable in solution—precipitating over time—and eluted throughout the entire volume of a size exclusion column, indicating it was nonspecifically aggregating. However, when the putative EP (residues 1–27) was removed (sPduL ΔEP), the truncated protein was stable and eluted as a single peak (Fig 5a) consistent with the size of a monomer (Fig 5d, blue curve). In contrast, both full-length rPduL and pPduL appeared to exist in two distinct oligomeric states (Fig 5b and 5c respectively, orange curves), one form of the approximate size of a dimer and the second, a higher molecular weight oligomer (~150 kDa). Upon deletion of the putative EP (residues 1–47 for rPduL, and 1–20 for pPduL), there was a distinct change in the elution profiles (Fig 5b and 5c respectively, blue curves). pPduLΔEP eluted as two smaller forms, possibly corresponding to a trimer and a monomer. In contrast, rPduLΔEP eluted as one smaller oligomer, possibly a dimer. We also analyzed purified rPduL and rPduLΔEP by size exclusion chromatography coupled with multiangle light scattering (SEC-MALS) for a complementary approach to assessing oligomeric state. SEC-MALS analysis of rPdulΔEP is consistent with a dimer (as observed in the crystal structure) with a weighted average (Mw) and number average (Mn) of the molar mass of 58.4 kDa +/− 11.2% and 58.8 kDa +/− 10.9%, respectively (S4a Fig). rPduL full length runs as Mw = 140.3 kDa +/− 1.2% and Mn = 140.5 kDa +/− 1.2%. This corresponds to an oligomeric state of six subunits (calculated molecular weight of 144 kDa). Collectively, these data strongly suggest that the N-terminal EP of PduL plays a role in defining the quaternary structure of the protein. The hallmark attribute of an organelle is that it serves as a discrete subcellular compartment functioning as an isolated microenvironment distinct from the cytosol. In order to create and preserve this microenvironment, the defining barrier (i.e., lipid bilayer membrane or microcompartment shell) must be selectively permeable. The BMC shell not only sequesters specific enzymes but also their cofactors, thereby establishing a private cofactor pool dedicated to the encapsulated reactions. In catabolic BMCs, CoA and NAD must be continually recycled within the organelle (Fig 1). Homologs of the predominant cofactor utilizer (aldehyde dehydrogenase) and NAD regenerator (alcohol dehydrogenase) have been structurally characterized, but until now structural information was lacking for PduL, which recycles CoA in the organelle lumen . Curiously, while the housekeeping Pta could provide this function, and indeed does so in the case of one type of ethanolamine-utilizing (EUT) BMC , the evolutionarily unrelated PduL fulfills this function for the majority of metabolosomes using a novel structure and active site for convergent evolution of function. The structure of PduL consists of two β-barrel domains capped by short alpha helical segments (Fig 2b). The two domains are structurally very similar (superimposing with a rmsd of 1.34 Å (over 123 out of 320/348 aligned backbone atoms, S5a Fig). However, the amino acid sequences of the two domains are only 16% identical (mainly the RHxH motif, β2 and β10), and 34% similar. Our structure reveals that the two assigned PF06130 domains (Fig 3) do not form structurally discrete units; this reduces the apparent sequence conservation at the level of primary structure. One strand of the domain 1 beta barrel (shown in blue in Fig 2) is contributed by the N-terminus, while the rest of the domain is formed by the residues from the C-terminal half of the protein. When aligned by structure, the β1 strand of the first domain (Fig 2a and 2b, blue) corresponds to the final strand of the second domain (β9), effectively making the domains continuous if the first strand was transplanted to the C-terminus. Refined domain assignment based on our structure should be able to predict domains of PF06130 homologs much more accurately. The closest structural homolog of the PduL barrel domain is a subdomain of a multienzyme complex, the alpha subunit of ethylbenzene dehydrogenase (S5b Fig, rmsd of 2.26 Å over 226 aligned atoms consisting of one beta barrel and one capping helix). In contrast to PduL, there is only one barrel present in ethylbenzene dehydrogenase, and there is no comparable active site arrangement. The PduL signature primary structure, two PF06130 domains, occurs in some multidomain proteins, most of them annotated as Acks, suggesting that PduL may also replace Pta in variants of the phosphotransacetylase-Ack pathway. These PduL homologs lack EPs, and their fusion to Ack may have evolved as a way to facilitate substrate channeling between the two enzymes. For BMC-encapsulated proteins to properly function together, they must be targeted to the lumen and assemble into an organization that facilitates substrate/product channeling among the different catalytic sites of the signature and core enzymes. The N-terminal extension on PduL homologs may serve both of these functions. The extension shares many features with previously characterized EPs : it is present only in homologs associated with BMC loci, and it is predicted to form an amphipathic α-helix. Moreover, its removal affects the oligomeric state of the protein. EP-mediated oligomerization has been observed for the signature and core BMC enzymes; for example, full-length propanediol dehydratase and ethanolamine ammonia-lyase (signature enzymes for PDU and EUT BMCs) subunits are also insoluble, but become soluble upon removal of the predicted EP . sPduL has also previously been reported to localize to inclusion bodies when overexpressed ; we show here that this is dependent on the presence of the EP. This propensity of the EP to cause proteins to form complexes (Fig 5) might not be a coincidence, but could be a necessary step in the assembly of BMCs. Structured aggregation of the core enzymes has been proposed to be the initial step in metabolosome assembly and is known to be the first step of β-carboxysome biogenesis, where the core enzyme Ribulose Bisphosphate Carboxylase/Oxygenase (RuBisCO) is aggregated by the CcmM protein . Likewise, CsoS2, a protein in the α-carboxysome core, also aggregates when purified and is proposed to facilitate the nucleation and encapsulation of RuBisCO molecules in the lumen of the organelle . Coupled with protein–protein interactions with other luminal components, the aggregation of these enzymes could lead to a densely packed organelle core. This role for EPs in BMC assembly is in addition to their interaction with shell proteins [24–26,36,38]. Moreover, the PduL crystal structures offer a clue as to how required cofactors enter the BMC lumen during assembly. Free CoA and NAD/H could potentially be bound to the enzymes as the core assembles and is encapsulated. However, this raises an issue of stoichiometry: if the ratio of cofactors to core enzymes is too low, then the sequestered metabolism would proceed at suboptimal rates. Our PduL crystals contained CoA that was captured from the Escherichia coli cytosol, indicating that the “ground state” of PduL is in the CoA-bound form; this could provide an elegantly simple means of guaranteeing a 1:1 ratio of CoA:PduL within the metabolosome lumen. The active site of PduL is formed at the interface of the two structural domains (Fig 2b). As expected, the amino acid sequence conservation is highest in the region around the proposed active site (Fig 4d); highly conserved residues are also involved in CoA binding (Figs 2a and 3, residues Ser45, Lys70, Arg97, Leu99, His204, Asn211). All of the metal-coordinating residues (Fig 2a) are absolutely conserved, implicating them in catalysis or the correct spatial orientation of the substrates. Arg103, which contacts the phosphate (Fig 4b), is present in all PduL homologs. The close resemblance between the structures binding CoA and phosphate likely indicates that no large changes in protein conformation are involved in catalysis, and that our crystal structures are representative of the active form. The native substrate for the forward reaction of rPduL and pPduL, propionyl-CoA, most likely binds to the enzyme in the same way at the observed nucleotide and pantothenic acid moiety, but the propionyl group in the CoA-thioester might point in a different direction. There is a pocket nearby the active site between the well-conserved residues Ser45 and Ala154, which could accommodate the propionyl group (S6 Fig). A homology model of sPduL indicates that the residues making up this pocket and the surrounding active site region are identical to that of rPduL, which is not surprising, because these two homologs presumably have the same propionyl-CoA substrate. The homology model of pPduL also has identical residues making up the pocket, but with a key difference in the vicinity of the active site: Gln77 of rPduL is replaced by a tyrosine (Tyr77) in pPduL. The physiological substrate of pPduL (Table 1) is thought to be lactyl-CoA, which contains an additional hydroxyl group relative to propionyl-CoA. The presence of an aromatic residue at this position may underlie the substrate preference of the PduL enzyme from the pvm locus. Indeed, in the majority of PduLs encoded in pvm loci, Gln77 is substituted by either a Tyr or Phe, whereas it is typically a Gln or Glu in PduLs in all other BMC types that degrade acetyl- or propionyl-CoA. A comparison of the PduL active site to that of the functionally identical Pta suggests that the two enzymes have distinctly different mechanisms. The catalytic mechanism of Pta involves the abstraction of a thiol hydrogen by an aspartate residue, resulting in the nucleophilic attack of thiolate upon the carbonyl carbon of acetyl-phosphate, oriented by an arginine and stabilized by a serine —there are no metals involved. In contrast, in the rPduL structure, there are no conserved aspartate residues in or around the active site, and the only well-conserved glutamate residue in the active site is involved in coordinating one of the metal ions. These observations strongly suggest that an acidic residue is not directly involved in catalysis by PduL. Instead, the dimetal active site of PduL may create a nucleophile from one of the hydroxyl groups on free phosphate to attack the carbonyl carbon of the thioester bond of an acyl-CoA. In the reverse direction, the metal ion(s) could stabilize the thiolate anion that would attack the carbonyl carbon of an acyl-phosphate; a similar mechanism has been described for phosphatases where hydroxyl groups or hydroxide ions can act as a base when coordinated by a dimetal active site . Our structures provide the foundation for studies to elucidate the details of the catalytic mechanism of PduL. Conserved residues in the active site that may contribute to substrate binding and/or transition state stabilization include Ser127, Arg103, Arg194, Gln107, Gln74, and Gln/Glu77. In the phosphate-bound crystal structure, Ser127 and Arg103 appear to position the phosphate (Fig 4b). Alternatively, Arg103 might act as a base to render the phosphate more nucleophilic. The functional groups of Gln74, Gln/Glu77, and Arg194 are directed away from the active site in both CoA and phosphate-bound crystal structures and do not appear to be involved in hydrogen bonding with these substrates, although they could be important for positioning an acyl-phosphate. The free CoA-bound form is presumably poised for attack upon an acyl-phosphate, indicating that the enzyme initially binds CoA as opposed to acyl-phosphate. This hypothesis is strengthened by the fact that the CoA-bound crystals were obtained without added CoA, indicating that the protein bound CoA from the E. coli expression strain and retained it throughout purification and crystallization. The phosphate-bound structure indicates that in the opposite reaction direction phosphate is bound first, and then an acyl-CoA enters. The two high-resolution crystal structures presented here will serve as the foundation for mechanistic studies on this noncanonical PTAC enzyme to determine how the dimetal active site functions to catalyze both forward and reverse reactions. PduL and Pta are mechanistically and structurally distinct enzymes that catalyze the same reaction , a prime example of evolutionary convergence upon a function. There are several examples of such functional convergence of enzymes, although typically the enzymes have independently evolved similar, or even identical active sites; for example, the carbonic anhydrase family . However, apparently less frequent is functional convergence that is supported by distinctly different active sites and accordingly catalytic mechanism, as revealed by comparison of the structures of Pta and PduL. One well-studied example of this is the β-lactamase family of enzymes, in which the active site of Class A and Class C enzymes involve serine-based catalysis, but Class B enzymes are metalloproteins . This is not surprising, as β-lactamases are not so widespread among bacteria and therefore would be expected to have evolved independently several times as a defense mechanism against β-lactam antibiotics. However, nearly all bacteria encode Pta, and it is not immediately clear why the Pta/PduL functional convergence should have evolved: it would seem to be evolutionarily more resourceful for the Pta-encoding gene to be duplicated and repurposed for BMCs, as is apparently the case in one type of BMC—EUT1 (Table 1). There could be some intrinsic biochemical difference between the two enzymes that renders PduL a more attractive candidate for encapsulation in a BMC—for example, PduL might be more amenable to tight packaging, or is better suited for the chemical microenvironment formed within the lumen of the BMC, which can be quite different from the cytosol . Further biochemical comparison between the two PTACs will likely yield exciting results that could answer this evolutionary question. BMCs are now known to be widespread among the bacteria and are involved in critical segments of both autotrophic and heterotrophic biochemical pathways that confer to the host organism a competitive (metabolic) advantage in select niches. As one of the three common metabolosome core enzymes, the structure of PduL provides a key missing piece to our structural picture of the shared core biochemistry (Fig 1) of functionally diverse catabolic BMCs. We have observed the oligomeric state differences of PduL to correlate with the presence of an EP, providing new insight into the function of this sequence extension in BMC assembly. Moreover, our results suggest a means for Coenzyme A incorporation during metabolosome biogenesis. A detailed understanding of the underlying principles governing the assembly and internal structural organization of BMCs is a requisite for synthetic biologists to design custom nanoreactors that use BMC architectures as a template. Furthermore, given the growing number of metabolosomes implicated in pathogenesis [46–50], the PduL structure will be useful in the development of therapeutics. It is gradually being realized that the metabolic capabilities of a pathogen are also important for virulence, along with the more traditionally cited factors like secretion systems and effector proteins . The fact that PduL is confined almost exclusively to metabolosomes can be used to develop an inhibitor that blocks only PduL and not Pta as a way to selectively disrupt BMC-based metabolism, while not affecting most commensal organisms that require PTAC activity. Genes for PduL homologs with and without the EP were amplified via PCR using the primers listed in S1 Table. sPduL was amplified using S. enterica Typhimurium LT2 genomic DNA, and pPduL and rPduL sequences were codon optimized and synthesized by GenScript with the 6xHis tag. All 5’ primers included EcoRI and BglII restriction sites, and all 3’ primers included a BamHI restriction site to facilitate cloning using the BglBricks strategy. 5’ primers also included the sequence TTTAAGAAGGAGATATACCATG downstream of the restriction sites, serving as a strong ribosome binding site. The 6x polyhistidine tag sequence was added to the 3’ end of the gene using the BglBricks strategy and was subcloned into the pETBb3 vector, a pET21b-based vector modified to be BglBricks compatible. E. coli BL21(DE3) expression strains containing the relevant PduL construct in the pETBb3 vector were grown overnight at 37°C in standard LB medium and then used to inoculate 1L of standard LB medium in 2.8 L Fernbach flasks at a 1:100 dilution, which were then incubated at 37°C shaking at 150 rpm, until the culture reached an OD600 of 0.8–1.0, at which point cultures were induced with 200 μM IPTG (isopropylthio-β-D-galactoside) and incubated at 20°C for 18 h, shaking at 150 rpm. Cells were centrifuged at 5,000 xg for 15 min, and cell pellets were frozen at –20°C. For protein purifications, cell pellets from 1–3 L cultures were resuspended in 20–30 ml buffer A (50 mM Tris-HCl pH 7.4, 300 mM NaCl) and lysed using a French pressure cell at 20,000 lb/in. The resulting cell lysate was centrifuged at 15,000 xg. 30 mM imidazole was added to the supernatant that was then applied to a 5 mL HisTrap column (GE Healthcare Bio-Sciences, Pittsburgh, PA). Protein was eluted off the column using a gradient of buffer A from 0 mM to 500 mM imidazole over 20 column volumes. Fractions corresponding to PduL were pooled and concentrated using Amicon Ultra Centrifugal filters (EMD Millipore, Billerica, MA) to a volume of no more than 2.5 mL. The protein sample was then applied to a HiLoad 26/60 Superdex 200 preparative size exclusion column (GE Healthcare Bio-Sciences, Pittsburgh, PA) and eluted with buffer B (20 mM Tris pH 7.4, 50 mM NaCl). Where applicable, fractions corresponding to different oligomeric states were pooled separately, leaving one or two fractions in between to prevent cross contamination. Pooled fractions were concentrated to 1–20 mg/mL protein as determined by the Bradford method prior to applying on a Superdex 200 10/300 GL analytical size exclusion column (GE Healthcare Bio-Sciences, Pittsburgh, PA). Size standards used were Thyroglobulin 670 kDa, γ-globulin 158 kDa, Ovalbumin 44 kDa, and Myoglobin 17 kDa. For light scattering, the proteins were measured in a Protein Solutions Dynapro dynamic light scattering instrument with an acquisition time of 5 s, averaging 10 acquisitions at a constant temperature of 25°C. The radii were calculated assuming a globular particle shape. Size exclusion chromatography coupled with SEC-MALS was performed on full-length rPduL and rPduL-ΔEP similar to Luzi et al. 2015 . A Wyatt DAWN Heleos-II 18-angle light scattering instrument was used in tandem with a GE AKTA pure FPLC with built in UV detector, and a Wyatt Optilab T-Rex refractive index detector. Detector 16 of the DAWN Heleos-II was replaced with a Wyatt Dynapro Nanostar QELS detector for dynamic light scattering. A GE Superdex S200 10/300 GL column was used, with 125–100 μl of protein sample at 1 mg/ml concentration injected, and the column run at 0.5 ml/min in 20 mM Tris, 50 mM NaCl, pH 7.4. Each detector of the DAWN-Heleos-II was plotted with the Zimm model in the Wyatt ASTRA software to calculate the molar mass. The molar mass was measured at each collected data point across the peaks at ~1 point per 8 μl eluent. Both the Mw and Mn of the molar mass calculations, as well as percent deviations, were also determined using Wyatt software program ASTRA. For preparing protein for crystallography, expression cells were grown as above, except were induced with 50 μM IPTG. Harvested cells were resuspended in buffer B and lysed using a French Press. Cleared lysate was applied on a 5 ml HisTrap HP column (GE Healthcare) and washed with buffer A containing 20 mM imidazole. Pdul-His was eluted with 2 CV buffer B containing 300 mM imidazole, concentrated and then applied on a HiLoad 26/60 Superdex 200 (GE Healthcare) column equilibrated in buffer B for final cleanup. Protein was then concentrated to 20–30 mg/ml for crystallization. Crystals were obtained from sitting drop experiments at 22°C, mixing 3 μl of protein solution with 3 μl of reservoir solution containing 39%–35% MPD. Crystals were flash frozen in liquid nitrogen after being adding 5 μl of a reservoir solution. For heavy atom derivatives, 0.2 μl of 100 mM Thiomerosal (Hampton Research) was added to the crystallization drop 36 h prior to freezing. For phosphate soaks, 5 μl reservoir and 1.5 μl 200 mM sodium phosphate solution (pH 7.0) were added 2 d prior to flash freezing. Enzyme reactions were performed in a 2 mL cuvette containing 50 mM Tris-HCl pH 7.5, 0.2 mM 5,5'-dithiobis-2-nitrobenzoic acid (DTNB; Ellman’s reagent), 0.1 mM acyl-CoA, and 0.5 μg purified PTAC, unless otherwise noted. To initiate the reaction, 5 mM NaH2PO4 was added, the cuvette was inverted to mix, and the absorbance at 412 nm was measured every 2 s over the course of four minutes in a Nanodrop 2000c, in the cuvette holder. 14,150 Mcm was used as the extinction coefficient of DTNB to determine the specific activity. A multiple sequence alignment of 228 PduL sequences associated with BMCs and 20 PduL sequences not associated with BMCs was constructed using MUSCLE . PduL sequences associated with BMCs were determined from Dataset S1 of Reference , and those not associated with BMCs were determined by searching for genomes that encoded PF06130 but not PF03319 nor PF00936 in the IMG database . The multiple sequence alignment was visualized in Jalview , and the nonconserved N- and C-terminal amino acids were deleted. This trimmed alignment was used to build the sequence logo using WebLogo . Diffraction data were collected at the Advanced Light Source at Lawrence Berkeley National Laboratory beamline 5.0.2 (100 K, 1.0000 Å wavelength for native data, 1.0093 Å for mercury derivative, 1.2861 Å for Zn pre-edge and 1.2822 Å for Zn peak). Diffraction data were integrated with XDS and scaled with SCALA (CCP4 ). The structure of PduL was solved using phenix.autosol , which found 11 heavy atom sites and produced density suitable for automatic model building. The model was refined with phenix.refine , with refinement alternating with model building using 2Fo-Fc and Fo-Fc maps visualized in COOT . Statistics for diffraction data collection, structure determination and refinement are summarized in Table 2. Figures were prepared using pymol (www.pymol.org) and Raster3D . Models of S. enterica Typhimurium LT2 and P. limnophilus PduL were generated with Modeller using the align2d and model-default scripts .
PMC4857006	Reversal of DNA damage induced Topoisomerase 2 DNA–protein crosslinks by Tdp2	Mammalian Tyrosyl-DNA phosphodiesterase 2 (Tdp2) reverses Topoisomerase 2 (Top2) DNA–protein crosslinks triggered by Top2 engagement of DNA damage or poisoning by anticancer drugs. Tdp2 deficiencies are linked to neurological disease and cellular sensitivity to Top2 poisons. Herein, we report X-ray crystal structures of ligand-free Tdp2 and Tdp2-DNA complexes with alkylated and abasic DNA that unveil a dynamic Tdp2 active site lid and deep substrate binding trench well-suited for engaging the diverse DNA damage triggers of abortive Top2 reactions. Modeling of a proposed Tdp2 reaction coordinate, combined with mutagenesis and biochemical studies support a single Mg-ion mechanism assisted by a phosphotyrosyl-arginine cation-π interface. We further identify a Tdp2 active site SNP that ablates Tdp2 Mg binding and catalytic activity, impairs Tdp2 mediated NHEJ of tyrosine blocked termini, and renders cells sensitive to the anticancer agent etoposide. Collectively, our results provide a structural mechanism for Tdp2 engagement of heterogeneous DNA damage that causes Top2 poisoning, and indicate that evaluation of Tdp2 status may be an important personalized medicine biomarker informing on individual sensitivities to chemotherapeutic Top2 poisons.Nuclear DNA compaction and the action of DNA and RNA polymerases create positive and negative DNA supercoiling—over- and under-winding of DNA strands, respectively—and the linking together (catenation) of DNA strands. Topoisomerases relieve topological DNA strain and entanglement to facilitate critical nuclear DNA transactions including DNA replication, transcription and cell division. The mammalian type II topoisomerases Top2α and Top2β enzymes generate transient, reversible DNA double strand breaks (DSBs) to drive topological transactions (1–3). Reversibility of Top2 DNA cleavage reactions is facilitated by formation of covalent enzyme phosphotyrosyl linkages between the 5′-phosphate ends of the incised duplex and an active site Top2 tyrosine, resulting in Top2 cleavage complexes (Top2cc). The Top2cc protein–DNA adduct is a unique threat to genomic integrity which must be resolved to prevent catastrophic Top2cc collisions with the cellular replication and transcription machineries (4–7). To promote cancer cell death, Top2 reactions are ‘poisoned’ by keystone pharmacological anticancer agents like etoposide, teniposide and doxorubicin. Importantly, Top2 is also poisoned when it engages abundant endogenous DNA damage not limited to but including ribonucleotides (8,9), abasic sites (10–14) and alkylation damage such as exocyclic DNA adducts arising from bioactivation of the vinyl chloride carcinogen (15,16) (Figure 1A). In the case of DNA damage-triggered Top2cc, compound DNA lesions arise that consist of the instigating lesion, and a DNA DSB bearing a bulky terminal 5′-linked Top2 DNA–protein crosslink. The chemical complexity of DNA damage-derived Top2cc necessitates that DNA repair machinery dedicated to resolving these lesions recognizes both DNA and protein, whilst accommodating diverse chemical structures that trap Top2cc. Precisely how the cellular DNA repair machinery navigates these complex lesions is an important aspect of Top2cc repair that has not yet been explored. Tdp2 processes phosphotyrosyl linkages in diverse DNA damage contexts. (A) Unrepaired DNA damage and repair intermediates such as bulky DNA adducts, ribonucleotides or abasic sites can poison Top2 and trap Top2 cleavage complex (Top2cc), resulting in a DSB with a 5′–Top2 protein adduct linked by a phosphotyrosine bond. Tdp2 hydrolyzes the 5′–phosphotyrosine adduct derived from poisoned Top2 leaving DNA ends with a 5′-phosphate, which facilitates DNA end joining through the NHEJ pathway. (B) DNA oligonucleotide substrates synthesized by EDC-imidazole coupling and used in Tdp2 enzyme assays contain deoxyadenine (dA), Ethenoadenine (ϵA) or an abasic site (THF) and a 5′–nitrophenol moiety. Phosphotyrosyl bond hydrolysis catalyzed by mTdp2 releases p-nitrophenol, which is detected by measuring absorbance at 415 nm. (C) mTdp2 reaction rates on p–nitrophenol modified DNA substrates shown in panel B. Rates are reported as molecules of PNP s produced by mTdp2. P-values calculated using two-tailed t-test; error bars, s.d. n = 4, n.s. = not statistically significant. (D) Structure of mTdp2 bound to 5′-phosphate DNA (product complex) containing ϵA (yellow). DNA binding β2Hβ–grasp (tan) and cap elements engage the 5′-nucleotide as well as the +2 and +3 nucleotides (blue) of substrate DNA. PDB entry 5HT2 is displayed, also see Table 1. (E) Structure of mTdp2 bound to 5′-phosphate DNA (product complex) containing THF (yellow). DNA binding β2Hβ–grasp (tan) and cap elements engage the 5′-nucleotide as well as the +2 and +3 nucleotides (blue) of substrate DNA. PDB entry 5INK is displayed, also see Table 1. (F) Structure of mTdp2 in the absence of DNA showing the extended 3-helix loop (tan) open-conformation of the DNA-binding grasp as seen in monomer E of the apo structure. PDB entry 5INM is displayed, also see Table 1. Tyrosyl DNA phosphodiesterase 2 (Tdp2) directly hydrolyzes 5′-phosphotyrosyl (5′-Y) linkages, and is a key modulator of cellular resistance to chemotherapeutic Top2 poisons (17–20). Tdp2 knockdown sensitizes A549 lung cancer cells to etoposide, and increases formation of nuclear γH2AX foci, a marker of DSBs (17), underlining the importance of Tdp2 in cellular Top2cc repair. Tdp2 is overexpressed in lung cancers, is transcriptionally up-regulated in mutant p53 cells and mediates mutant p53 gain of function phenotypes, which can lead to acquisition of therapy resistance during cancer progression (21). The importance of Tdp2 in mediating topoisomerase biology is further underlined by the facts that human TDP2 inactivating mutations are found in individuals with intellectual disabilities, seizures and ataxia, and at the cellular level, loss of Tdp2 inhibits Top2β-dependent transcription (18). It is possible that TDP2 single nucleotide polymorphisms (SNPs) encode mutations that impact Tdp2 function, but the molecular underpinnings for such Tdp2 deficiencies are not understood. Previously we reported high-resolution X-ray crystal structures of the minimal catalytically active endonuclease/exonuclease/phosphatase (EEP) domain of mouse Tdp2 (mTdp2) bound to a DNA substrate mimic, and a 5′-phosphorylated reaction product (20). However, important questions regarding the mechanism of Tdp2 engagement and processing of DNA damage remain. First, it is unclear if Tdp2 processes phosphotyrosyl linkages in the context of DNA damage that triggers Top2cc, and if so, how the enzyme can accommodate such complex DNA damage within its active site. Based on metal-bound Tdp2 structures (20), we also proposed a single Mg mediated catalytic mechanism, but this mechanism requires further scrutiny and characterization. Herein, we report an integrated structure-function study of the Tdp2 reaction mechanism, including a description of new X-ray structures of ligand-free Tdp2, and Tdp2 bound to abasic and alkylated (1-N-etheno-adenine) DNA damage. Our integrated results from structural analysis, mutagenesis, functional assays and quanyum mechanics/molecular mechanics (QM/MM) modeling of the Tdp2 reaction coordinate describe in detail how Tdp2 mediates a single-metal ion tyrosyl DNA phosphodiesterase reaction capable of acting on diverse DNA end damage. We further establish that DNA damage binding in the Tdp2 active site is linked to conformational change and binding of metal cofactor. Finally, we characterize a Tdp2 SNP that ablates the Tdp2 single metal binding site and Tdp2 substrate induced conformational changes, and confers Top2 drug sensitivity in mammalian cells. Oligonucleotides with a 5′-phosphate modification were obtained from IDT and diluted to a concentration of 2 mmol l in water. 1-Ethyl-3-(3-dimethylaminopropyl) carbodiimide (EDC; Pierce) was used to couple p-nitrophenol to the oligonucleotide with a modified version of the manufacturer's instructions. Briefly, 25 mg EDC were dissolved in 150 μl oligonucleotide solution. After the EDC dissolved, 400 μl DDW, 60 μl 100 mmol l Imidazole pH 6 and 100 μl of PNP from a saturated water/PNP mixture preheated to 55°C were added. Reactions were heated at 55°C for 30 min to ensure the PNP dissolved, then incubated at 37°C overnight. Reactions were quenched with 500 μl 2 mol l acetic acid and heated to 55°C for 1 h, followed by neutralization with 500 μl 2 mol l Tris base. Reactions were twice diluted to 15 ml in DDW then concentrated to 300 μl with a 3K MWT cutoff spin concentrator (Amicon), then run on 20% (w/v) 19:1 8M urea-TBE PAGE to resolve reaction products. Bands were visualized by UV shadow, excised, soaked in 15 ml DDW for 16 h at 15°C, and purified on a C18 reverse-phase Sep-Pac (Waters). Mouse Tdp2 catalytic domain (mTdp2, residues 118 to 370), human Tdp2 catalytic domain (hTdp2, residues 108–362) or full-length human Tdp2 (hTdp2, residues 1–362) were expressed and purified as previously described (20). Plasmids containing mutant Tdp2 sequences were generated using the Quickchange kit (Stratagene). Crystals containing mTdp2 and 5′-phosphorylated DNA (product) with modified 5′ nucleotides (ϵA, dA, THF) were grown and cryoprotected as described (20). Sequences, modifications and synthesis sources for oligonucleotides used for co-crystallization are indicated in Supplementary Table S1. For alternate divalent metal complexes, a 5′-phosphorylated DNA substrate (substrate dC) was co-crystallized in the presence of Mg, and divalent metals were swapped by soaking the crystals in crystallization buffer containing 5 mmol l MnCl2 or 10 mmol l Ca(OAc)2 for 1 h prior to cryoprotection. Cryoprotectant solutions contained mother liquor plus 25% PEG-3350, 8% glycerol and 5% glucose, and either 5 mmol l MnCl2 or 10 mmol l Ca(OAc)2. Crystals of hTdp2 bound to 5′-phosphate DNA (substrate dC) were prepared by mixing a 1.2-fold molar excess of DNA, and grown in 90 mmol l TRIS pH 7.0, 27% (w/v) PEG600, 9% (v/v) glycerol and 450 mmol l (NH4)2SO4. Crystals of apo-mTdp2 were grown in 14–18% (w/v) PEG3350, 100 mmol l HEPES pH 7.5, 200 mmol l Li2SO4 and 10 mmol l Mg(OAc)2, and soaked into the same buffer containing 25% PEG3350 and 12% (v/v) glycerol prior to flash-freezing in liquid nitrogen for data collection. X-ray data (Table 1) for all structures except the manganese soak (PDB entry 5INP) were collected at 100 K on beamline 22-ID of the Advanced Photon Source at a wavelength of 1.000 Å. X-ray data for the manganese soak (PDB entry 5INP) were collected on a Rigaku HF007 Cu rotating anode X-ray source at a wavelength of 1.5418 Å. X-ray diffraction data were processed and scaled using the HKL2000 suite (22). The hTdp2-DNA and mTdp2-apo crystals were phased by molecular replacement in PHASER (23) using chain A of PDB entry 4GZ1. Initial solutions were improved by iterative rounds of manual fitting in COOT (24) and refinement in PHENIX (25). Each data set was collected from a single crystal. Values in parentheses are for highest-resolution shell (10% of relections). For proteolysis experiments, 4 μl reactions containing 40 μmol l mTdp2 (aa 118 to 370) in reaction buffer (10 mmol l HEPES pH 7.5, 200 mmol l NaCl, 0.5 mmol L TCEP) with 4 mmol l Mg(OAc)2 (indicated +Mg) or 8 mmol l NaOAc (indicated -Mg), with 0 or 60 μmol l 12 nt DNA were incubated with 0, 5, 1.7 or 0.6 μg l Trypsin for 1 h at 22°C. Reactions were quenched by addition of Laemmli SDS-PAGE dye, heated at 70°C for 10 min, and analyzed by SDS-PAGE. For mass-spectrometry analysis of peptide masses, reactions were quenched with 1% (v/v) trifluoroacetic acid, purified on a C18 ZipTip (Millipore) and an ESI-MS mass measurement made on a Q-ToF Ultima/Global (Micromass/Waters) using flow injection from a pressurized bomb. The instrument was operated in the positive ion, V-mode and calibrated using the multiply-charged ion envelope of horse heart cytochrome C. The molecular ion mass was determined using the Max Ent 1 routine from the MassLynx software. Reactions contained 50 μl with 1 μmol l mTdp2 (residues 118 to 370) in buffer (6 mmol l HEPES pH 7.5, 300 mmol l NaCl, 0.3 mmol L TCEP and 0.01% (v/v) TWEEN-20) with 0–20 mmol l Mg(OAc)2 or ultrapure Ca(OAc)2 (99.9965%, Alfa Aesar) titrated against 0–40 mmol l NaOAc to maintain a constant concentration of acetate, with 0 or 1.5 μmol l 12 nt DNA. Reactions were incubated at room temperature for 20 minutes in a 96-well black plastic plate (Corning), then tryptophan fluorescence was measured in a Polarstar Omega platereader (BMG Labtech) with 4 readings of 150 pulses per well using 280/10 excitation and 350/10 emission filters. The increase in fluorescent intensity was calculated by subtracting the fluorescent intensity from samples with no divalent metal ions and plotted as a function of divalent metal ion concentration. Kd values and Hill coefficients were calculated using PRISM6. Assays on MBP-fusion proteins of the human catalytic domain (MBP-hTdp2) mutant proteins with 5′ -tyrosylated DNA substrates (Figures 5E and 6C and Supplementary Figure S5C) were performed as described (20). For experiments measuring the effect of divalent metal ions on reaction rates, 50 μl reactions contained 1 μmol l mTdp2 (residues 118 to 370, Figure 4B), hTdp2 (residues 108–362) or hTdp2 (residues 108–362) in buffer (10 mmol l HEPES pH 7.5, 100 mmol l NaCl, 0.5 mmol l TCEP, 0.4 mg ml BSA, 0.02% (v/v) TWEEN-20, 1 mmol l Thymidine 5′–p-nitrophenyl phosphate) with 0–100 mmol l Mg(OAc)2 titrated against 20–220 mmol l NaOAc to maintain a constant concentration of acetate. Reactions contained 0, 1 or 10 mmol l ultrapure Ca(OAc)2 (99.9965%, Alfa Aesar), with 0, 2 or 20 mmol l less NaOAc to maintain a constant acetate concentration. PNP formation was monitored by the absorbance at 415 nm with a background correction at 515 nm. The change in absorbance at 415 nm at 10 min as a function of Mg concentration was plotted. Oligonucleotide 300 bp substrates with 5′ phosphorylated GATC overhangs were generated as previously described (26). Comparable substrate with 5′-phophotyrosine adducted GATC overhangs were generated by annealing the modified strand to complementary strands that generate caps for substrate head and tail ends. Head and tail caps have a 5′-phosphotyrosine-GATC overhang terminus on one end; the 5′-phosphorylated, non-adducted overhangs on the other end are made complementary to the head and tail ends of a 270 bp double stranded core fragment generated by polymerase chain reaction. Ligation of an excess of these caps to the 270 bp fragment generates a 300 bp substrate with 5′-phosphotyrosine end structures as described in Figure 7. Unligated caps are removed using a Qiaquick PCR cleanup kit. Purified NHEJ proteins (Ku, XRCC4-ligase IV, XLF) were prepared as previously described (27). End joining reactions were performed using 5 nM DNA substrate, 25 nM Ku, 25 nM XRCC4/LigaseIV complex, 50 nM XLF and hTdp2 proteins as indicated. Reactions contained 25 mM Tris-HCl pH 7.5, 0.1 mM ethylenediaminetetraacetic acid (EDTA), 2 mM dithiothreitol (DTT), 125 mM KCl, 5 mM MgCl2, 100 μM ATP, 8% (w/v) polyethyleneglycol, 0.05% (w/v) Triton X-100, 50 μg ml bovine serum albumin (BSA) and 50 ng supercoiled plasmid DNA. Reactions were carried out in a final volume of 10 μl and incubated at 37°C for 5 min. Reactions were stopped by the addition of 0.1% (w/v) SDS and 5 mM EDTA and analyzed by 5% native PAGE. Mouse embryo fibroblast (MEF) cells from matched Tdp2 and Tdp2 mice (19) were a gift from F. Cortes-Ledesma, and were immortalized by transformation with a construct that expresses SV40 T-antigen (Addgene #1779). HCT-116 cells and a ligase IV deficient variant were the gift of E. A. Hendrickson (28). The cDNAs with wild-type human Tdp2 and polymorphic variants generated by mutagenesis were introduced into pLX302 (29) (Addgene#25896) to prepare lentivirus. Tdp2 cells were infected with lentivirus and bulk cell cultures expressing lentiviral delivered construct purified by treatment with puromycin. Expression of human Tdp2 (hTdp2) was validated by Western analysis (12203-1-AP; Proteintech). Fifty nanograms of the 300 bp substrates used in in vitro experiments and 2 μg of carrier supercoiled plasmid DNA were introduced into 2 × 10 MEF or HCT 116 cells by electroporation (Neon, Invitrogen) using a 10 μl chamber and a single 1350 V, 30 ms pulse. Cells were recovered 1 h later, washed with phosphate buffered saline and DNA purified using a Qiamp DNA mini kit on a Qiacube. NHEJ products were quantified by qPCR and characterized by amplicon sequencing as previously described (27). Clonogenic survival assays were carried out by treating log phase cells with Etoposide as described in the legend to Figure 7 before seeding treated and mock treated cells in 10 cm dishes. Colonies formed after 10 days post-treatment were strained using a crystal violet (0.5% w/v) solution. Plates containing a minimum of 50 colonies were counted by hand, and at least three plates were counted for each dose. In the QM/MM calculation, in addition to the water nucleophile and the putative catalytic Lewis base (Asp272), the side chains of proximal residues (Asn130, Asn274, His236, His326, His359, Ser239 and Glu162), the bound phosphate moiety, the Mg and three waters in its coordinate shell are all included in the QM sub-system. In addition, we modeled the position of the Top2 peptide tyrosine based on conformations of the substrate analog. The QM sub-system of this study consisted of 110 atoms with a zero net charge on the sub-system. QM/MM calculations are performed using Gaussian09.D01 (30). Since the QM sub-system contains a large number of buried atoms and commonly used electrostatic potential fitting schemes to obtain the charge distribution at atomic positions become unreliable for such systems with buried atoms, we have selected the CM5 charge model in the current study to calculate the charges on atomic positions at each step of the QM/MM calculation. This CM5 scheme is an extension to the Hirshfeld population analysis and is adapted to handle buried charges properly. This charge distribution of the QM sub-system was used to evaluate contributions of residues in the MM region to the net stability of the transition and product states as compared to that of the initial reactive state. As reported previously (31), this residue analysis is solely based on the electrostatic energy contributions to the initial, transition and product states. According to this estimation, the residues Arg142, Lys213, Ser235, Asp277, Glu279, Asp292, Glu295, Asp308, Asp343, Arg354 and Trp360 contribute toward the stability of both the transition and product states as compared with the reactant. However, the electrostatic contributions from residues Asp132, Asp135, Glu186, Glu242, Arg247, Thr273, Arg276, Lys299, Lys322, Arg324, Arg327 and Asp358 have the opposite effect toward the stabilities of the transition and product states. Two potent Top2 poisons include bulky alkylated DNA helix-distorting DNA base adducts (e.g. 1-N-ethenoadenine, ϵA) (16) and abundant abasic sites (13) (Figure 1A). Whether Tdp2 processes phosphotyrosyl linkages within these diverse structural contexts is not known. To test this, we adapted an EDC coupling method to generate 5′-terminal p-nitrophenol (PNP) modified oligonucleotides (32,33) that also harbored DNA damage at the 5′-nucleotide position (see Materials and Methods). We then evaluated the ability of a recombinant purified mouse Tdp2 catalytic domain (mTdp2) to release PNP (a structural mimic of a topoisomerase tyrosine) from the 5′-terminus of compound damaged DNA substrates using a colorimetric assay (Figure 1B). We observe robust Tdp2-dependent release of PNP from 5′-modified oligonucleotides in the context of dA-PNP, ϵA-PNP or the abasic-site analog tetrahydrofuran spacer (THF) (Figure 1C). Thus, Tdp2 efficiently cleaves phosphotyrosyl linkages in the context of a compound 5′ lesions composed of abasic or bulky DNA base adduct DNA damage. To understand the molecular basis for Tdp2 processing of Top2cc in the context of DNA damage, we crystallized and determined X-ray crystal structures of mTdp2 bound to 5′-phosphate DNA (product complex) with a 5′-ϵA at 1.43 Å resolution (PDB entry 5HT2) and the abasic DNA damage mimic 5′-THF at 2.15 Å resolution (PDB entry 5INK; Figure 1D and E, Table 1). In these Tdp2-DNA complex structures, mTdp2 adopts a mixed α-β fold typified by a central 12-stranded anti-parallel β-sandwich enveloped by several helical elements that mold the Tdp2 active site. One half of the molecule contributes to formation of the walls of the DNA-binding cleft that embraces the terminal position of the damaged DNA substrate. In the DNA lesion-bound state, two key DNA binding elements, the β-2-helix-β (β2Hβ) ‘grasp’, and ‘helical cap’ mold the substrate binding trench and direct the ssDNA of a 5′-overhang substrate into the active site. A comparison to an additional new structure of DNA-free Tdp2 (apo state, Figure 1F) shows that this loop is conformationally mobile and important for engaging DNA substrates. The mode of engagement of the 5′-nucleobase of the bulky ϵA adduct describes a mechanism for Tdp2 to bind 5′-tyrosylated substrates that contain diverse forms of DNA damage. The 5′-ϵA nucleobase is recognized by an extended Tdp2 van Der Waals interaction surface, referred to here as the ‘hydrophobic wall’ that is assembled with the sidechains of residues Leu315 and Ile317 (Figure 2A and B). Structures of mTdp2 bound to DNA damage that triggers Top2 poisoning. (A) Structure of mTdp2 bound to 5′-phosphate DNA (product complex) containing ϵA (yellow), Mg (magenta) and its inner-sphere waters (gray). mTdp2 is colored by electrostatic surface potential (red = negative, blue = positive, gray = neutral/hydrophobic). PDB entry 5HT2. (B) σ-A weighted 2Fo-Fc electron density map (at 1.43 Å resolution, contoured at 2.0 σ) for the ϵA DNA complex. The ϵA nucleotide is shown in yellow and a hydrogen bond from the ϵA O4′ to inner-sphere water is shown as gray dashes. (C) Structure of mTdp2 bound to 5′-phosphate DNA (product complex) containing THF (yellow), Mg (magenta) and its inner-sphere waters (gray). mTdp2 is colored with red (electronegative), blue (electropositive) and gray (hydrophobic) electrostatic surface potential displayed. PDB entry 5INK is displayed. (D) σ-A weighted 2Fo-Fc electron density map (at 2.15 Å resolution, contoured at 2.0 σ) for THF-DNA complex. The THF is shown in yellow and a hydrogen bond from the THF O4′ to inner-sphere water is shown as gray dashes. For comparison, we also determined a structure of an undamaged 5′-adenine (5′-dA) bound to Tdp2 at 1.55 Å (PDB entry 5INL). A structural overlay of damaged and undamaged nucleotides shows no major distortions to nucleotide planarity between different bound sequences and DNA damage (compare ϵA, dA and dC, Supplementary Figure S1A–D). Therefore, structurally diverse undamaged or alkylated bases (e.g. ϵG, ϵT) (34,35) could likely be accommodated in the Tdp2 active site via planar base stacking with the active site facing hydrophobic wall of the β2Hβ motif. Likewise, the abasic deoxyribose analog THF substrate binds similar to the alkylated and non-alkylated substrates, but with a slight alteration in the approach of the 5′-terminus (Figure 2C). Interestingly, in the absence of a nucleobase, O4′ of the THF ring adopts a close approach (2.8 Å) to a water molecule that directly participates in the outer sphere single Mg ion coordination shell (Figure 2D). This shift is coincident with a small adjustment in the position of the +2 and +3 nucleotides (Supplementary Figure S1E). These collective differences may explain the slight, but statistically significant elevated activity on the THF substrate (Figure 1C). An intriguing feature of the DNA-damage bound conformation of the Tdp2 active site is an underlying network of protein–water–protein contacts that span a gap between the catalytic core and the DNA binding β2Hβ-grasp (Supplementary Figure S2). In this arrangement, six solvent molecules form a channel under the β2Hβ-grasp, ending with hydrogen bonds to the peptide backbone of the Mg ligand Asp358. The paucity of hydrophobic interactions stabilizing the β2Hβ DNA-bound conformation suggests that conformational plasticity in the β2Hβ might be a feature of DNA damage and metal cofactor engagement. To test this hypothesis, we crystallized Tdp2 in the absence of DNA and determined a DNA free Tdp2 structure to 2.4 Å resolution (PDB entry 5INM; Figures 1F and 3A). Conformational plasticity in the Tdp2 active site. (A) The open, 3-helix conformation (tan) of flexible active-site loop observed in monomer E of the DNA-free mTdp2 structure (PDB entry 5INM) is supported by T309 (green), which packs against the EEP core. The β2Hβ docking pocket (circled) is unoccupied and residues N312, N314 and L315 (orange) are solvent-exposed. Wall-eyed stereo view is displayed. (B) The closed β2Hβ conformation in the mTdp2–DNA product structure containing 5′-ϵA (yellow, PDB entry 5HT2). T309 (green) is an integral part of the β2Hβ DNA-binding grasp (tan) and hydrogen bonds to the backbone of Y321, while N314 (orange) occupies the β2Hβ docking pocket. Wall-eyed stereo view is displayed. (C) Alignment of active site loop conformers observed in the 5 promoters of the DNA-free mTdp2 (PDB entry 5INM, see Table 1) crystallographic asymmetric unit (left) and sequence alignment showing residues not observed in the electron density as ‘∼’ (right). (D) Limited trypsin proteolysis probes the solvent accessibility of the flexible active-site loop. mTdp2 WT (lanes 1–13) or mTdp2 D358N (lanes 14–26) were incubated in the presence or absence of Mg and/or a 12 nt self annealing, 5′-phosphorylated DNA (substrate ‘12 nt’ in Supplementary Table S1), then reacted with 0.6, 1.7 or 5 ng μl of trypsin. Reactions were separated by SDS-PAGE and proteins visualized by staining with coomassie blue. (E) Limited chymotrypsin proteolysis probes the solvent accessibility of the flexible active-site loop. Experiments performed as in panel D for mTdp2 WT (lanes 27–39) or mTdp2 D358N (lanes 40–52), but with chymotrypsin instead of trypsin. This crystal form contains 5 Tdp2 protein molecules in the asymmetric unit, with variations in active site Mg occupancy and substrate binding loops observed for the individual protomers. The most striking feature of the DNA ligand-free state is that the active site β2Hβ-grasp can adopt alternative structures that are distinct from the DNA-bound, closed β2Hβ DNA binding grasp (Figure 3A and B). In one monomer (chain ‘E’), the grasp adopts an ‘open’ 3-helix loop conformation that projects away from the EEP catalytic core. Two monomers have variable disordered states for which much of the DNA binding loop is not visible in the electron density. The remaining two molecules in the DNA-free crystal form are closed β2Hβ conformers similar to the DNA bound structures (Figure 3C). Thus, we posit that Tdp2 DNA binding conformationally selects the closed form of the β2Hβ grasp, rather than inducing closure upon binding. A detailed analysis of the extended 3-helix conformation shows that the substrate-binding loop is able to undergo metamorphic structural changes. In this open form, residues Asn312-Leu315 are distal from the active site and solvent-exposed (orange sticks, Figure 3A), while Thr309 (green surface, Figure 3A) packs into a shallow pocket of the EEP core to anchor the loop. Burial of Thr309 is enabled by an unusual main chain cis–peptide bond between Asp308-Thr309 and disassembly of the short antiparallel beta-strand of the β2Hβ fold. By comparison, the closed β2Hβ grasp conformer is stabilized by Asn312 and Asn314 binding into two β2Hβ docking pockets, and Leu315 engagement of the 5′-terminal nucleobase (Figure 3B). To transition into the closed β2Hβ conformation, Thr309 disengages from the EEP domain pocket, flips peptide backbone conformation cis to trans, and is integral to the β2Hβ antiparallel β-sheet. Stabilization of the closed β2Hβ-grasp conformation is linked to the active site through a hydrogen bond between Trp307 and the Mg coordinating residue Asp358. Accordingly, in the DNA free structure, we observe a trend where the 2 closed monomers have an ordered Mg ion in their active sites, while the monomers with open conformations have a poorly ordered or vacant metal binding site. Overall, these observations suggest that engagement of diverse damaged DNA ends is enabled by an elaborate substrate selected stabilization of the β2Hβ DNA binding grasp, and these rearrangements are coordinated with Mg binding in the Tdp2 active site. To evaluate Mg and DNA-dependent Tdp2 structural states in solution, we probed mTdp2 conformations using limited trypsin and chymotrypsin proteolysis (Figure 3C–E). In the absence of DNA or Mg, mTdp2 is efficiently cleaved in the metamorphic DNA binding grasp at one site by trypsin (Arg316), or at two positions by chymotrypsin (Trp307 and Leu315). By comparison, Mg, and to a greater extent Mg/DNA mixtures (compare Figure 3, lanes 4, 7 and 13) protect mTdp2 from proteolytic cleavage. Interestingly, addition of Mg alone protects against proteolysis as well. This is consistent with Mg stabilizing the closed conformation of the β2Hβ-grasp through an extended hydrogen-bonding network with Asp358 and the indole ring of the β2Hβ-grasp residue Trp307 (also discussion below on Tdp2 active site SNPs). To assess structural conservation of Tdp2 conformational changes between human and mouse Tdp2, we also determined a 3.2 Å resolution structure of the human Tdp2 domain bound to a DNA 5′-PO4 terminus product complex (PDB entry 5INO). Comparisons of the human hTdp2-DNA complex structure to the mTdp2 DNA bound state show a high level of conservation of the DNA-bound conformations (Supplementary Figure S3A). Moreover, similar to mTdp2, proteolytic protection of the hTdp2 substrate binding loop occurs with addition of Mg and DNA (Supplementary Figure S3B). Thus, X-ray structures and limited proteolysis analysis indicate that DNA- and metal-induced conformational changes are a conserved feature of the vertebrate Tdp2-substrate interaction. Consistently in high-resolution X-ray structural analyses we, (8,20) and others (36) observe a single Mg metal bound in the Tdp2 active site. This includes the DNA-free (Figure 3A), DNA damage bound (Figure 3B) and reaction product-bound crystal forms of mouse, (PDB entry 4GZ1), D. rerio (PDB entry 4FPV) and C. elegans Tdp2 (PDB entry 4FVA). However, previous biochemical analysis has suggested an alternative two-metal ion mechanism for the Tdp2-phosphotyrosyl phosphodiesterase reaction (37). In these experiments, at limiting Mg concentrations, Ca addition to Tdp2 reactions stimulated activity. While this work was suggestive of a two metal ion mechanism for phosphotyrosyl bond cleavage by Tdp2, we note that second metal ion titrations can be influenced by metal ion binding sites outside of the active site (38). In fact, divalent metals have been observed in the Tdp2 protein–DNA complexes (PDB entry 4GZ2) distal to the active center (20), and we propose this might account for varied results in different studies. To further probe the metal ion dependence of the Tdp2 phosphodiesterase reaction, we performed metal ion binding assays, determined crystal structures in the presence of varied divalent metals (Mn and Ca), and analyzed metal ion dependence of the Tdp2 phosphotyrosyl phosphodiesterase reaction (Figure 4). Metal cofactor interactions with Tdp2. (A) Intrinsic tryptophan fluorescence of mTdp2 was used to monitor a conformational response to divalent metal ion binding. Either Mg or Ca were titrated in the presence or absence of 5′-P DNA, and the tryptophan fluorescence was monitored with an excitation wavelength of 280 nm and emission wavelength of 350 nm using 10 nm band pass filters. Both Mg and Ca induce a conformational change which elicits an increase in tryptophan fluorescence of mTdp2 in the presence and absence of DNA, while D358N active site mutant of mTdp2 is unresponsive to Mg. (B) mTdp2 activity assayed on a T5PNP substrate as a function of Mg and Ca concentration. PNP release (monitored by absorbance at 415 nm) as a function of Mg concentration and in the absence or presence of 1 or 10 mM Ca is shown; error bars, s.d. n = 4. (C) σ-A weighted 2Fo-Fc electron density map (blue) and model-phased anomalous difference Fourier (magenta) maps for the mTdp2–DNA–Mn complex (PDB entry 5INP) show a single Mn (cyan) is bound with expected octahedral coordination geometry. A 53σ peak in the anomalous difference Fourier map (data collected at λ = 1.5418 Å) supports Mn as the identity of this atom. (D) Comparison of Ca (green Ca ion, orange DNA) (PDB entry 5INQ), and Mg (magenta Mg ion, yellow DNA) (PDB entry 4GZ1) mTdp2–DNA structures shows that Ca distorts the 5′-phosphate binding mode. Our proteolysis results indicate a Mg-dependent Tdp2 conformational response to metal binding. The Tdp2 active site has three tryptophan residues within 10 Å of the metal binding center, so we assayed intrinsic tryptophan fluorescence to detect metal-induced conformational changes in mTdp2. These data were an excellent fit to a single-site binding model both in the presence and absence of DNA (Figure 4A). This analysis revealed Mg Kd values in the sub-millimolar range and Hill coefficients which were consistent with a single metal binding site both in the presence and absence of DNA (Supplementary Table S2). We then measured effects of metal ion concentrations on Tdp2 cleavage of p-nitrophenyl-thymidine-5′-phosphate by mTdp2 (20,39). This small molecule substrate is not expected to be influenced by metal–DNA coordination outside of the active site. Inclusion of ultrapure Ca (1 mM or 10 mM) results in a dose-dependent inhibition but not stimulation Tdp2 activity, even in conditions of limiting Mg (Figure 4B). We performed the same titrations with human hTdp2 and hTdp2 (Supplementary Figure S4), and find similar stimulation of activity by Mg and inhibition by Ca. Overall, these metal binding analyses are consistent with a single metal ion mediated reaction. To further evaluate the structural influence of divalent cations on the Tdp2 active site, we determined crystal structures by soaking crystals with metal cofactors that either support (Mn) (20) or inhibit (Ca, Figure 4B) the Tdp2 reaction (PDB entries 5INP and 5INQ). Anomalous difference Fourier maps of the Tdp2–DNA–Mn complex show a single binding site for Mn in each Tdp2 active site (Figure 4C), with octahedral coordination and bond lengths typical for Mn ligands (Supplementary Table S3). The Mn ion is positioned in the Tdp2 active site similar to the Mg-bound complex (Figure 2C), which is consistent with the ability of Mn to support robust Tdp2 catalytic activity (20). In contrast, while co-complex structures with Ca also show a single metal ion, Ca binds in a slightly different position, shifted ∼1 Å from the Mg site. Although Ca is also octahedrally coordinated, longer bond lengths for the Ca ligands (Supplementary Table S3) shift the Ca ion relative to the Mg ion site. Interestingly, bi-dentate inner sphere metal contacts from the Ca ion to Glu162 distort the active site phosphate-binding mode, and dislodge the 5′-PO4 out of the Tdp2 active site (Figure 4D). Together with results showing that under the conditions examined here, Ca inhibits rather than stimulates the Tdp2 reaction, the divalent metal bound Tdp2 structures provide a mechanism for Ca-mediated inhibition of the Tdp2 reaction. Next, to examine the feasibility of our proposed single Mg mechanism, we simulated the Tdp2 reaction coordinate with hybrid QM/MM modeling using Tdp2 substrate analog- and product-bound structures as guides. Previous structural analyses showed that the superposition of a DNA substrate mimic (5′-aminohexanol) and product (5′-PO4) complexes delineates a probable Tdp2 reaction trajectory characterized by inversion of stereochemistry about the adducted 5′-phosphorus (20). In this scheme (Figure 5A), a candidate nucleophilic water that is strongly hydrogen bonded to Asp272 and Asn274, is well positioned for the in-line nucleophilic attack ∼180° opposite of the P–O bond of the 5′-Tyr adduct. Structure-function analysis of the Tdp2 reaction mechanism. (A) Proposed mechanism for hydrolysis of phosphotyrosine bond by Tdp2. Residues in green form the binding-site for the 5′-tyrosine (red) and phosphate, yellow bind the 5′ nucleotide and blue bind nucleotides 2–3. Residue numbers shown are for the mTdp2 homolog. (B) Free energy during the QM/MM simulation as a function of distance between the nucleophilic water and 5′-phosphorus atom. Reaction proceeds from right to left. (C) Models for the mTdp2-DNA complex during the QM/MM reaction path simulation showing the substrate (left, tan), transition state intermediate (center, cyan) and product (right, pink) states. Residue numbers shown are for the mTdp2 homolog. (D) Electrostatic surface potential calculated for 5′-phosphotyrosine in isolation (upper panel) and in the presence of a cation–π interaction with the guanidinium group of Arg216 (lower panel) shows electron-withdrawing effect of this interaction. Electrostatic potential color gradient extends from positive (red) through neutral (gray), to negative (blue). (E) Bar graph displaying the relative activity of wild-type and mutant human MBP-hTdp2 fusion proteins on the three substrates. Release of PNP from PNP phosphate and T5PNP was detected as an increase in absorbance at 415 nm. Reaction rates are expressed as the percent of activity relative to wildtype MBP-hTdp2; error bars, s.d. n = 3. Mutants of hTdp2 (black) and the equivalent residue in mTdp2 (tan) are indicated. We examined the energy profile of the nucleophilic attack of the water molecule by using the distance between the water oxygen and the P atom on the phosphate moiety as the sole reaction coordinate in the present calculation (Figure 5B and C). A starting model was generated from atomic coordinates of the mTdp2 5′–aminohexanol substrate analog structure (PDB 4GZ0) with a tyrosine replacing the 5′-aminohexanol then adding the Mg and inner-sphere waters from the mTdp2-DNA product structure (PDB, 4GZ1), and running an initial round of molecular dynamics simulation (10 ns) to allow the system to reach an equilibrium. After QM/MM optimization of this model (Figure 5C, ‘i-substrate’), the O–P distance is 3.4 Å, which is in agreement with the range of distances observed in the mTdp2 5′-aminohexanol substrate analog structure (3.2–3.4 Å). No appreciable energy penalty is observed during the first 0.5 Å of the reaction coordinate. When the reaction reaches an O–P distance of 2.18 Å, formation of a transition state with an energy maximum of +7.4 kcal mol is observed. Here, the water proton and the neighboring O of Asp272 participates in a strong hydrogen bond (distance of 1.58 Å) and the phosphotyrosyl O–P distance is stretched to 1.77 Å, which is 0.1 Å beyond an equilibrium bond length. In the subsequent two steps of the simulation, as the water-phosphate O–P distance reduces to 1.98 Å, a key hydrogen bond between the nucleophilic water and Asp272 shortens to 1.38 Å as the water H–O bond approaches the point of dissociation. The second proton on the water nucleophile maintains a strong hydrogen bond with Asn274 throughout the reaction, implicating this residue in orienting the water nucleophile during the reaction. Concomitant with this, the phosphotyrosyl O–P bond weakens (d = 1.89 Å), and the formation of the penta-covalent transition state (Figure 5C ‘ii-transition state’) is observed. The final steps show inversion of stereochemistry at the phosphate, along with lengthening and breaking of the phosphotyrosyl O–P bond. Product formation is coupled to a transfer of a proton from the nucleophillic water to Asp272, consistent with the proposed function for this residue as the catalytic base. Of note, both nitrogens of the imidazole side chain of His 359 require protonation for stability of the simulation. Asp 326 makes a hydrogen bond to N∂1 of His359, suggesting that this salt bridge could stabilize the protonated form of His359 as has been demonstrated for the analogous Asp-His pair in the EEP domain of APE1, which elevates the pKa of this His above 8.0 (40). In our model, the transition state contains a hydrogen bond between the doubly protonated His359 and the phosphate oxygen that also coordinates with the single catalytic Mg, while the second His359 imidazole proton maintains a H-bond with the Asp326 residue throughout the reaction. In the final optimized structure, the observed product state (Figure 5C, ‘iii-product’) is found in a conformation that is 7.4 kcal mol more stable than the initial reactive state (Figure 5B). The tyrosine oxy-anion product is coordinated to the Mg ion with a 2.0 Å distance, which is the shortest of the six Mg ligands (including three water molecules, one of the free oxygens on the phosphate group and the Glu162 residue), indicating the single Mg greatly stabilizes the product oxy-anion. An additional striking feature gleaned from the QM/MM modeling is the putative binding mode of the Top2 tyrosine-leaving group. A trio of conserved residues (Tyr 188, Arg 216 and Ser 239) forms the walls of a conserved Top2 tyrosine binding pocket. We propose this cation–π interaction further contributes to tuned stabilization of the negatively charged phenolate reaction product. Consistent with this, analysis of electrostatic potential of the phosphotyrosyl moiety using Gaussian 09.D01 (30) in the presence and absence of the Arg216 guanidinium reveals Arg216 is strongly electron withdrawing (Figure 5D). We further examined the contribution of this cation–π interaction to the reaction chemistry by moving the guanidinium group of Arg216 from the QM system to the MM system as either a +1 or ∼0 charge species, and re-computed energy penalties for each step in the reaction coordinate (Supplementary Figure S5A). Removing Arg216 from the quantum subsystem incurs an ∼2 kcal mol penalty in the transition state and product complex. Removing the +1 charge on the Arg216 has a minimal impact on the transition state, but incurs an additional ∼2 kcal mol penalty in the product complex. Altogether, QM/MM modeling identifies new determinants of the Tdp2 reaction, and demonstrates our proposed single Mg catalyzed reaction model is a viable mechanism for Tdp2-catalyzed 5′-phosphotyrosine bond hydrolysis. To test the aspects of the Tdp2 reaction mechanism described here derived from high-resolution mouse Tdp2 crystal structures (denoted with superscript numbering ‘m’ for numbering of the mouse protein), we engineered and purified thirteen human MBP-hTdp2 mutant proteins (denoted with superscript numbering and ‘h’ for the human protein) and assayed the impacts of mutations on Tdp2 catalytic activity using three in vitro reporter substrates including a tyrosylated DNA substrate (5′-Y), p-nitrophenyl phosphate (PNPP) and thymidine 5′-monophosphate p-nitrophenyl ester (T5PNP) (20) (Figure 5E, Supplementary Figures S5B and S5C). By analyzing activities on this nested set of chemically related substrates we aimed to dissect structure-activity relationships of Tdp2 catalysis. For example, mutations impacting Tdp2 active site chemistry and phosphotyrosyl bond cleavage should similarly affect catalysis on all three substrates, but mutants impacting DNA damage binding might only impair catalysis on 5′-Y and T5PNP but not PNPP that lacks a nucleobase. Structural results and QM/MM modeling indicate Asp272 activates a water molecule for in-line nucleophilic attack of the scissile phosphotyrosyl linkage. To test if this proposed Lewis base is critical for reaction chemistry we mutated it to a His, which could alternatively support metal binding, as well as bulky hydrophobic residues (Leu and Met) that we predict would block the water-binding site. Similar to a previously characterized D262N mutation (20), all three substitutions ablate activity, supporting essential roles for Asp262 (Asp272) in catalysis. Next, we mutated key elements of the mobile loop (β2Hβ hydrophobic wall, Figure 2A and C). Mutations I307A, L305A, L305F and L305W all impaired catalysis on both nucleotide-containing substrates (<50% activity). The L305W substitution that we expect to have the most distorting impact on conformation of the β2Hβ hydrophobic wall also has the largest impact on catalysis of the DNA substrate 5′-Y. By comparison, as predicted by our model where β2Hβ dictates key interactions with undamaged and damaged nucleobases, all of these substitutions have little impact on PNPP (>90% activity). Third, we altered properties of the proposed enzyme substrate cation–π interface. No activity was detected for a mutant that removes the positive charge at this position (R206A). The precise geometry of this pocket is also critical for catalysis as replacement of Arg206 (Arg216) with a lysine also results in a profound decrease in catalysis (<5% activity on 5′-Y, no detectable activity on T5PNP or PNPP). Similarly, mutation of Tyr178 that structurally scaffolds the Arg206 (Arg216) guanidinium also significantly impacts activity, with Y178F and Y178W having <25% activity on all substrates. Fourth, we evaluated roles for the His351–Asp316 (Asp326–His359) transition state stabilization charge pair. We found that mutations that removed the charge yet retained the ability to hydrogen bond (H351Q) or should abrogate the elevated pKa of the Histidine (D316N) had severe impacts on catalysis. Thus altogether, our mutational data support key roles for the active site Lewis base aspartate, mobile substrate engagement loops, enzyme–substrate cation–π interactions, and active site transition state stabilizing charge interaction in supporting Tdp2 catalysis. Recently, it was found that inactivation of TDP2 by a splice-site mutation is associated with neurological disease and confers hypersensitivity to Top2 poisons (18). We considered whether human SNPs causing missense mutations might also impact Tdp2 DNA–protein crosslink repair functions established here as well as Tdp2-mediated NHEJ of blocked DNA termini. We identified two SNPs in human TDP2 curated in the NCBI SNP database (41) that result in missense mutations within the DNA processing active site: rs199602263 (minor allele frequency 0.0002), which substitutes Asp350 for Asn, and rs77273535 (minor allele frequency 0.004, which substitutes Ile307 for Val) (Figure 6A). We show the D350N substitution severely impairs activity on all substrates tested in vitro, whereas I307V only has a mild impact on catalysis (Figure 6B–D). To better understand the basis for the D350N catalytic defect, we analyzed the structural environment of this substitution based on the high-resolution structures of mTdp2 (Figure 6A). Interestingly, the Tdp2 single Mg ion octahedral coordination shell also involves an extended hydrogen-bonding network mediated by Asp350 (Asp358) that stabilizes the DNA-bound conformation of the β2Hβ substrate-binding loop through hydrogen bonding to Trp307. Here, Asp350 (Asp358) serves as a structural nexus linking active site metal binding to substrate binding loop conformations. Tdp2 SNPs impair function. (A) Active site residues mutated by TDP2 SNPs. D350N (mTdp2 D358N) and I307V (mTdp2 I317V) substitutions are mapped onto the Tdp2 active site of the high-resolution mTdp2 structure (4GZ1). (B) Coomassie blue stained SDS-PAGE gel of purified WT and mutant MBP-hTdp2 proteins used for assays in panels C and D. (C) Activity of WT and mutant MBP-hTdp2 proteins on a 5′–phosphotyrosyl–DNA oligonucleotides with 3′-fluorescein label. Samples were withdrawn from reactions, neutralized with TBE-urea loading dye at the indicated timepoints, and electrophoresed on a 20% TBE-urea PAGE. (D) Relative activity of WT and indicated mutant human MBP-hTdp2 fusion proteins on three model Tdp2 substrates. Quantification of percent MBP-hTdp2 activity relative to WT protein for the 5′-Y DNA oligonucleotide substrate (blue bars), T5PNP (red bars) and PNPP (green bars) is displayed. Release of PNP from PNP phosphate (PNPP) and was detected as an increase in absorbance at 415 nm, whereas the 5′-Y substrate is quantification of activity in a gel based assay shown in Figure 6C. Error bars, s.d. n = 3. To define the molecular basis for the D350N (D358N) defect, we crystallized and determined the structure of the DNA-free form of the D358N protein to 2.8Å resolution (PDB entry 5INN). This structure shows the D358N mutation disrupts the hydrogen bond between Asp358 and Trp307, shifts the position of Asn358 and destabilizes Trp307. Consequently, poor electron density is visible for the β2Hβ loop which is mostly disordered (Supplementary Figure S6). Although Mg is present at the same concentration as the WT-mTdp crystals (10 mM), we find the metal site is unoccupied in the D358N crystals. Therefore, metal-regulated opening/closure of the active site may modulate Tdp2 activity, and D350N is sufficient to block both metal binding and conformational change. In support of this, we also find that D350N (D358N) impairs Mg binding as measured by intrinsic tryptophan fluorescence (Figure 4A), and abrogates Mg-stimulated active site conformational changes detected by trypsin and chymotrypsin sensitivity of the Tdp2 metamorphic loop (Figure 3D). Overall, our Tdp2 structure/activity studies reveal a tuned, 5′-detyrosylation DNA end processing activity and it has been demonstrated that Tdp2 could enable repair of Top2 damage by the non-homologous end-joining (NHEJ) pathway (19). Accordingly, we demonstrate here that 5′-tyrosylated ends are sufficient to severely impair an in vitro reconstituted mammalian NHEJ reaction (Figure 7A, lanes 3 and 6), unless supplemented with catalytic quantities of hTdp2 (Figure 7A, lane 8). Interestingly, hTdp2 is slightly more effective than hTdp2 in promoting NHEJ of adducted ends, while a catalytically deficient E152Q mutant was inactive in this assay, supporting the notion that Tdp2 catalytic activity is required to support NHEJ of phosphotyrosyl blocked DSBs (Supplementary Figure S7A). We confirmed that efficient joining of the same tyrosine-adducted substrate in cells (Figure 7B) was dependent on both NHEJ (reduced over 10-fold in ligase IV deficient HCT 116 cells; Supplementary Figure S7B), and Tdp2 (reduced 5-fold in Tdp2 deficient MEFs; Figure 7C). Moreover, products with error (i.e. junctions have missing sequence flanking the adducted terminus) are twice as frequent in cells deficient in Tdp2 (Figure 7D). Therefore, in accord with previous work (19), joining of tyrosine adducted ends after Tdp2-mediated detyrosylation is both more efficient and more accurate than joining after endonucleolytic excision (e.g. mediated by Artemis or the Mre11/Rad50/Nbs1 complex). Effects of Tdp2 active site SNP-encoded mutants on cellular Tdp2 functions. (A) Cy5 labeled substrates with 5′-phosphate termini (Lanes 1–4) or 5′-tyrosylated termini (Lanes 5–9) were incubated with Ku, the NHEJ ligase (XRCC4, ligase IV and XLF; X-L-X) and 1 nM hTdp2 as indicated (+) for 5 min at 37°C. Concatemer ligation products were detected by 5% native PAGE. (B) Workflow diagram of cellular end joining assays. DNA substrates with 5′-phosphotyrosine adducts and 4 nucleotide 5′ overhangs were electroporated into cultured mammalian cells. After 1 h, DNA was recovered from cells and repair efficiency by qPCR or sequencing as indicated. (C) qPCR assessment of cellular end joining efficiency of the tyrosylated substrate comparing results from wildtype MEF cells to Tdp2 cells and Tdp2 cells complemented with wildtype or the noted hTDP2 variants; Joining efficiency shown is the ratio of junctions recovered relative to WT cells. Error bars, s.d, n = 3. (D) Junctions recovered from cellular end-joining assays in the noted cell types were characterized by sequencing to assess the end-joining error rate. Error bars, s.d, n = 3. (E) Clonogenic survival assay of WT, Tdp2 knockout and complemented MEF cells after treatment with indicated concentrations of etoposide for 3 h; error bars, s.d, n = 3. We next compared the ability of wild-type and mutant hTdp2 variants to complement Tdp2 deficient mouse embryonic fibroblasts (Supplementary Figure S7C). Joining of extrachromosomal DNA with phosphotyrosine blocked ends, both in terms of efficiency (Figure 7C) and fidelity (Figure 7D), was indistinguishable comparing MEFs from a wild-type mouse, MEFs from a Tdp2-/- mouse overexpressing wild-type human Tdp2, and Tdp2 -/- MEFs overexpressing the I307V variant human Tdp2. In contrast, joining of 5′ phosphotyrosine-blocked ends was reduced 5-fold in Tdp2-/- MEFs, and an equivalent defect was observed in Tdp2-/- MEFs overexpressing Tdp2 D350N. Moreover, the frequency of inaccurate repair was 2-fold higher in both Tdp2 deficient cells and Tdp2 deficient cells overexpressing D350N, relative to cells expressing wild type Tdp2 or hTdp2 I307V (Figure 7D). Expression of wild type or I307V human Tdp2 in Tdp2-/- MEFs was also sufficient to confer levels of resistance to etoposide comparable to the matched wild-type MEF line, while overexpression of human D350N Tdp2 had no apparent complementation activity (Figure 7E). The rare D350N variant is thus inactive by all metrics analyzed. By comparison the more frequent I307V has only mild effects on in vitro activity, and no detectable impact on cellular assays. Top2 chemotherapeutic agents remain frontline treatments, and exposure to the chemical and damaged DNA triggers of Top2-DNA protein crosslink formation are unavoidable (3,5,35). Understanding how cells cope with complex DNA breaks bearing topoisomerase–DNA protein crosslinks is key to deciphering individual responses to chemotherapeutic outcomes and genotoxic agents that poison Top2. Together with mutagenesis and functional assays, our new Tdp2 structures in the absence of ligands and in complex with DNA damage reveal four novel facets of Tdp2 DNA-protein conjugate processing: (i) The Tdp2 active site is well-suited for accommodating a variety of DNA structures including abasic and bulky alkylated DNA lesions that trigger Top2 poisoning, (ii) High-resolution structural analysis coupled with mutational studies and QM/MM molecular modeling of the Tdp2 reaction coordinate support a single metal-ion mechanism for the diverse clade of EEP domain catalyzed phosphoryl hydrolase reactions, (iii) The Tdp2 active site is conformationally plastic, and undergoes intricate rearrangements upon DNA and Mg cofactor binding and (iv) Naturally occurring Tdp2 variants undermine Tdp2 active site chemistry, cellular and biochemical activities. This mechanistic dissection of Tdp2 interactions with damaged DNA and metal cofactor provides a detailed molecular understanding of the mechanism of Tdp2 DNA protein crosslink processing. Tdp2 was originally identified as a protein conferring resistance to both Top1 and Top2 anti-cancer drugs (17), however it is hypothesized that the predominant natural source of substrates for Tdp2 are likely the potent DNA damage triggers of Top2 poisoning and Top2 DNA protein crosslinks encountered during transcription (18). The properties of complex DNA strand breaks bearing Top2-DNA protein crosslinks necessitate that Tdp2 accommodates both damaged nucleic acid as well as the topoisomerase protein in its active site for catalysis. The Tdp2 substrate interaction groove facilitates DNA-protein conjugate recognition in two important ways. First, the nucleic acid binding trench is assembled by a dynamic β2Hβ DNA damage-binding loop that is capable of recognizing and processing diverse phosphotyrosyl linkages even in the context of bulky adducts such as ϵA. This is achieved by binding of nucleic acid ‘bases out’ by an extended base-stacking hydrophobic wall of the β2Hβ-loop. Secondly, our QM/MM analysis further highlights an enzyme–substrate cation–π interaction as an additional key feature of the Tdp2 protein–DNA crosslink binding and reversal. The strictly conserved active site Arg216 appears optimally positioned to stabilize a delocalized charge on the phenolate product of the phosphotyrosyl cleavage reaction through molecular orbital overlap and polarization of the leaving group. To our knowledge, this is the first proposed example of a substrate cation–π interface exploited to promote a phosphoryl-transfer reaction. This unique feature likely provides an additional level of substrate-specificity for Tdp2 by restricting activity to hydrolysis of aromatic adducts characteristic of Top2cc, picornaviral protein–RNA (42) and Hepatitis B Virus (HBV) protein–DNA processing intermediates (43). By comparison, other EEP nucleases such as Ape1 (44) and Ape2 (45) have evolved robust DNA damage specific endonucleolytic and exonucleolytic activities not shared with Tdp2. The dynamic nature of the Tdp2 active site presents opportunities for enzyme regulation. However, whether additional protein factors can bind to Tdp2 and modulate assembly/disassembly of the Tdp2 β2Hβ-loop is unknown. We hypothesize that binding of the Top2 protein component of a DNA–protein crosslink and/or other protein-regulated assembly of the Tdp2 active site might also serve to regulate Tdp2 activity to restrict it from misplaced Top2 processing events, such that it cleaves only topologically trapped or poisoned Top2 molecules when needed. Furthermore, high-resolution structures of mouse (20) (Figures 3 and 4) and C. elegans (36) Tdp2 show that a single metal ion typifies the Tdp2 active site from worms to man. Herein, we report five additional lines of evidence from metal binding detected by intrinsic tryptophan fluorescence, crystallographic analysis of varied metal cofactor complexes, mutagenesis, Ca inhibition studies and QM/MM analysis that all support a feasible single Mg mediated Tdp2 catalytic mechanism. The advent of personalized medical screening opens doors for assessment of individual vulnerabilities to commonly used chemotherapeutic drugs. It would be beneficial to employ this knowledge during the early decision making processes regarding treatment. Etoposide and other Top2 poisons remain front line anti-cancer drugs (4), and Tdp2 frameshift mutations in the human population confer hypersensitivity to Top2 poisons including etoposide and doxyrubicin (18). Given Tdp2 variation in the human population, links to neurological disease (18) and viral pathogenesis (42), our finding that TDP2 SNPs ablate catalytic activity has probable implications for modulation of cancer chemotherapy, susceptibility to environmentally linked Top2 poisons, and viral infection. Lastly, Tdp2 inhibitors may synergize or potentiate cytotoxic effects of current anticancer treatments that target Tdp2. Thus, we anticipate this atomic-level and mechanistic definition of the molecular determinants of Tdp2 catalysis and conformational changes driven by DNA–protein and protein–protein interactions will foster unique strategies for the development of Tdp2 targeted small molecule interventions. Coordinates and structure factors have been deposited in the RCSB Protein Data Bank under accession code 5HT2 (mTdp2-Mg-ϵA-DNA complex), 5INK (mTdp2-Mg-THF complex), 5INL (mTdp2-Mg-dA-DNA-product complex), 5INM (mTdp2-apo structure), 5INN (mTdp2-D350N structure), 5INO (hTdp2-Mg-DNA product complex), 5INP (mTdp2-Mn-DNA product complex) and 5INQ (mTdp2-Ca-DNA product complex).
PMC4795551	Biochemistry and Crystal Structure of Ectoine Synthase: A Metal-Containing Member of the Cupin Superfamily	Ectoine is a compatible solute and chemical chaperone widely used by members of the Bacteria and a few Archaea to fend-off the detrimental effects of high external osmolarity on cellular physiology and growth. Ectoine synthase (EctC) catalyzes the last step in ectoine production and mediates the ring closure of the substrate N-gamma-acetyl-L-2,4-diaminobutyric acid through a water elimination reaction. However, the crystal structure of ectoine synthase is not known and a clear understanding of how its fold contributes to enzyme activity is thus lacking. Using the ectoine synthase from the cold-adapted marine bacterium Sphingopyxis alaskensis (Sa), we report here both a detailed biochemical characterization of the EctC enzyme and the high-resolution crystal structure of its apo-form. Structural analysis classified the (Sa)EctC protein as a member of the cupin superfamily. EctC forms a dimer with a head-to-tail arrangement, both in solution and in the crystal structure. The interface of the dimer assembly is shaped through backbone-contacts and weak hydrophobic interactions mediated by two beta-sheets within each monomer. We show for the first time that ectoine synthase harbors a catalytically important metal co-factor; metal depletion and reconstitution experiments suggest that EctC is probably an iron-dependent enzyme. We found that EctC not only effectively converts its natural substrate N-gamma-acetyl-L-2,4-diaminobutyric acid into ectoine through a cyclocondensation reaction, but that it can also use the isomer N-alpha-acetyl-L-2,4-diaminobutyric acid as its substrate, albeit with substantially reduced catalytic efficiency. Structure-guided site-directed mutagenesis experiments targeting amino acid residues that are evolutionarily highly conserved among the extended EctC protein family, including those forming the presumptive iron-binding site, were conducted to functionally analyze the properties of the resulting EctC variants. An assessment of enzyme activity and iron content of these mutants give important clues for understanding the architecture of the active site positioned within the core of the EctC cupin barrel.Compatible solutes are exploited by members of all three domains of life as versatile cyto-protectants , in particular against cellular stress elicited by high osmolarity environments [2–5]. They are especially useful for this latter purpose since their benign nature allows their amassing to exceedingly high cellular concentrations. As a result of compatible solute accumulation, dehydration of the cytoplasm of osmotically stressed cells is counteracted , and concomitantly, its solvent properties are optimized for the functioning of vital biochemical and physiological processes [8, 9]. Ectoine [(S)-2-methyl-1,4,5,6-tetrahydropyrimidine-4-carboxylic acid] and its derivative 5-hydroxyectoine [(4S,5S)-5-hydroxy-2-methyl-1,4,5,6-tetrahydropyrimidine-4-carboxylic acid] are such compatible solutes . Both marine and terrestrial microorganisms produce them widely [13, 14] in response to osmotic or temperature stress [15–17]. Synthesis of ectoine occurs from the intermediate metabolite L-aspartate-ß-semialdehyde [18, 19] and comprises the sequential activities of three enzymes: L-2,4-diaminobutyrate transaminase (EctB; EC 2.6.1.76), 2,4-diaminobutyrate acetyltransferase (EctA; EC 2.3.1.178), and ectoine synthase (EctC; EC 4.2.1.108) [20, 21] (Fig 1). The ectoine derivative 5-hydroxyectoine, a highly effective stress protectant in its own right [22–24], is synthesized by a substantial subgroup of the ectoine producers [13, 14]. This stereospecific chemical modification of ectoine (Fig 1) is catalyzed by the ectoine hydroxylase (EctD) (EC 1.14.11) [25, 26], a member of the non-heme containing iron(II) and 2-oxoglutarate-dependent dioxygenase superfamily . The remarkable function preserving effects of ectoines for macromolecules and cells [28–31], frequently also addressed as chemical chaperones, led to a substantial interest in exploiting these compounds for biotechnological purposes and medical applications [12, 32, 33]. Scheme of the ectoine and 5-hydroxyectoine biosynthetic pathway [20, 21, 25]. Here we focus on ectoine synthase (EctC), the key enzyme of the ectoine biosynthetic route [13, 14, 20, 21, 34, 35] (Fig 1). Biochemical characterizations of ectoine synthases from the extremophiles Halomonas elongata, Methylomicrobium alcaliphilum, and Acidiphilium cryptum, and from the nitrifying archaeon Nitrosopumilus maritimus have been carried out [14, 35–38]. Each of these enzymes catalyzes as their main activity the cyclization of N-γ-acetyl-L-2,4-diaminobutyric acid (N-γ-ADABA), the reaction product of the 2,4-diaminobutyrate acetyltransferase (EctA), to ectoine with the concomitant release of a water molecule (Fig 1). In side reactions, EctC can promote the formation of the synthetic compatible solute 5-amino-3,4-dihydro-2H-pyrrole-2-carboxylate (ADPC) through the cyclic condensation of two glutamine molecules and it also possesses a minor hydrolytic activity for ectoine and synthetic ectoine derivatives with either reduced or expanded ring sizes [36, 37]. Although progress has been made with respect to the biochemical characterization of ectoine synthase [14, 35–38], a clear understanding of how its structure contributes to its enzyme activity and reaction mechanism is still lacking. With this in mind, we have biochemically characterized the ectoine synthase from the cold-adapted marine bacterium Sphingopyxis alaskensis (Sa). We demonstrate here for the first time that the ectoine synthase is a metal-dependent enzyme, with iron as the most likely physiologically relevant co-factor. The EctC protein forms a dimer in solution and our structural analysis identifies it as a member of the cupin superfamily. The two crystal structures that we report here for the (Sa)EctC protein (with resolutions of 1.2 Å and 2.0 Å, respectively), and data derived from extensive site-directed mutagenesis experiments targeting evolutionarily highly conserved residues within the extended EctC protein family, provide a first view into the architecture of the catalytic core of the ectoine synthase. Ectoine [(S)-2-methyl-1,4,5,6-tetrahydropyrimidine-4-carboxylic acid] was a kind gift from bitop AG (Witten, Germany). Anhydrotetracycline (AHT), desthiobiotine and the strepavidin affinity matrix for the purification of Strep-tag II labeled proteins was purchased from IBA GmbH (Göttingen, Germany). Hydroxylamine and phenanthroline for the photometric determination of the iron-content of the recombinant (Sa)EctC proteins were purchased from Sigma-Aldrich (München, Germany). All chemicals used to synthesize the gamma and alpha forms of N-acetyl-l-2,4-diaminobutyric acid (ADABA) for EctC enzyme activity assays were purchased either from Sigma Aldrich (Steinheim, Germany), or Acros (Geel, Belgium). Alkaline hydrolysis of ectoine (284 mg, 2.0 mmol) was accomplished in aqueous KOH (50 mL, 0.1 M) for 20 h at 50°C . The reaction mixture was subsequently neutralized with perchloric acid (60% in water, 4 mL) and the precipitated potassium perchlorate was filtered off. Subsequently, the filtrate was concentrated under reduced pressure. Purification of the residue and separation of the formed compounds was then performed by repeated chromatography on a silica gel column (Merck silica gel 60) using a gradient of ethanol/25% ammonia/water 50:1:2–10:1:2 as eluent to yield pure N-γ-ADABA (192 mg, 1.20 mmol, 60%) and N-α-ADABA (32 mg, 0.20 mmol, 10%). The identity and purity of theses compounds was unequivocally established by thin-layer chromatography (TLC) and nuclear magnetic resonance (H-NMR and C-NMR) spectroscopy (S1a and S1b Fig) as described [21, 39] on a Bruker AVIII-400 or DRX-500 NMR spectrometer. (i) Analytical data for N-γ-ADABA: TLC: Rf = 0.55 (ethanol/25% ammonia/water 7:1:2); H-NMR (400 MHz, D2O): δ = 3.71 (dd, J(H,H) = 7.6 Hz, J(H,H) = 5.6 Hz, 1H, CH), 3.41–3.24 (m, 2H, CH2), 2.15–2.01 (m, 2H, CH2), 1.99 (s, 3H, CH3) ppm; C-NMR (100 MHz, D2O): δ = 177.5 (CO), 177.0 (COOH), 55.3 (CH), 38.3 (CH2), 33.0 (CH2), 24.6 (CH3) ppm. (ii) Analytical data for N-α-ADABA: TLC: Rf = 0.38 (ethanol/25% ammonia/water 7:1:2); H-NMR (400 MHz, D2O): δ = 4.24 (dd, J(H,H) = 8.8 Hz, J(H,H) = 5.1 Hz, 1H, CH), 3.07–3.02 (m, 2H, CH2), 2.22–2.11 (m, 2H, CH2), 2.04 (s, 3H, CH3) ppm; C-NMR (100 MHz, D2O): δ = 180.1 (CO), 176.7 (COOH), 55.4 (CH), 39.6 (CH2), 32.5 (CH2), 24.7 (CH3) ppm. The nucleotide sequence of the ectC gene from S. alaskensis (genome accession number: NC_008048) was used as a template to obtain a codon-optimized ectC DNA sequence (Life Technologies, Darmstadt, Germany) for its expression in E. coli. The nucleotide sequence of the synthetic ectC gene was deposited in the NCBI database under accession number KR002036. The synthetic ectC gene was used to construct an expression plasmid (pNW12) that is based on the pASG-IBA3 vector (IBA GmbH, Göttingen, Germany). In plasmid pNW12, the ectC gene is fused at its 3’ end to a short open reading frame encoding a Strep-tag II affinity peptide (NWSHPQFEK). It is transcribed from the TetR-controlled tet promoter carried by the backbone of the pASG-IBA3 expression vector. De-repression of tet promoter activity can be triggered by adding the synthetic inducer AHT for the TetR repressor to the growth medium. The details of the construction of pNW12 have been reported . Plasmids carrying ectC genes were routinely maintained in the Escherichia coli strain DH5α (Invitrogen, Karlsruhe, Germany) on LB agar plates containing ampicillin (100 μg ml). Plasmid DNA was isolated by routine procedures. Minimal medium A (MMA) containing 0.5% (w/v) glucose as the carbon source, 0.5% (w/v) casamino acids, 1 mM MgSO4, and 3 mM thiamine was used to cultivate the E. coli strain BL21 carrying pNW12 for the overproduction of the (Sa)EctC protein and its mutant derivatives. No additional metal solution was added to the components of the original recipe of MMA . Variants of the codon-optimized ectC gene from S. alaskensis present on plasmid pNW12 were prepared by site-directed mutagenesis using the QuikChange Lightning Site-Directed Mutagenesis Kit (Agilent, Waldbronn, Germany) with custom synthesized DNA primers purchased from Microsynth AG (Lindau, Germany). The DNA sequence of the entire coding region of each mutant ectC gene was determined by Eurofins MWG (Ebersberg, Germany) to ensure the presence of the desired mutation and the absence of unwanted alterations. Details on the genetic changes introduced into ectC genes are listed in Table 1. The conversion of N-γ-ADABA into ectoine by the (Sa)EctC protein and its mutant derivatives was monitored in a reaction that contained 10 mM N-γ-ADABA as the substrate, 1 mM FeSO4 and 5 μg of the EctC protein under study. The amount of ectoine formed was measured after 20 min of incubation of the enzyme-substrate mixture by HPLC analysis. The iron-content of the investigated protein preparations was determined photometrically ; note that in comparison with data obtained via ICP-MS, the colorimetric assay overestimates somewhat the iron content of the (Sa)EctC protein preparations. For the overproduction of the (Sa)EctC-Strep-tag II protein , an overnight culture of strain [BL21 (pNW12)] was prepared in MMA and used to inoculate 1 L of MMA (in a 2 L Erlenmeyer flask) to an OD578 of 0.05. The cells were grown on an aerial shaker (set to 180 rpm) at 37°C until the culture reached an OD578 of 0.5. At this time point, the growth temperature was lowered to 30°C and the speed of the shaker was reduced to 100 rpm. Growth of the culture was continued and when it reached an OD578 of 0.7, AHT was added to the growth medium at a final concentration of 0.2 mg ml to boost expression of the recombinant ectC gene. After 2 h of further incubation of the culture, the E. coli cells were harvested by centrifugation and disrupted by passing them several times through a French Pressure cell; a cleared cell lysate was prepared by ultracentrifugation (100 000 g) at 4°C for 1 h . The supernatant of this cleared lysate was then passed through a column filled with 5 ml of Strep-Tactin Superflow material (IBA GmbH, Göttingen, Germany); the column had been equilibrated with a buffer containing 200 mM NaCl and 20 mM Tris-HCl (pH 8). The (Sa)EctC-Strep-tag II protein was eluted from the affinity matrix with three column volumes of the same buffer containing 2.5 mM desthiobiotin. The recombinant (Sa)EctC-Strep-tag II protein was then concentrated to either 5 mg ml for enzymes assays or 10 mg ml for crystallization trials with Vivaspin 6 columns (Satorius Stedim Biotech GmbH, Göttingen, Germany) in the same buffer as described above. Desthiobiotin was not removed by dialysis from these protein preparations. The purified and concentrated (Sa)EctC-Strep-tag II protein was either used immediately for enzymes assays or kept at 4°C since the flash-freezing of the protein with liquid nitrogen and its subsequent storage at -80°C resulted in a rapid inactivation of ectoine synthase activity. 25 Variants of the (Sa)EctC-Strep-tag II protein carrying singe amino acid substitutions (Table 1) were overproduced and purified using the same procedure employed for the isolation of the wild-type protein. These mutant proteins behaved like the wild-type (Sa)EctC-Strep-tag II protein during the overproduction and purification procedure. Protein concentrations were determined both with a Pierce BCA Protein Assay Kit (Thermo Scientific, Schwerte, Germany) using BSA as the standard protein and spectrophotometrically by using an extinction coefficient of 15 470 M cm for the (Sa)EctC-Strep-tag II protein at a wavelength of 280 nm. The purity and integrity of the isolated (Sa)EctC-Strep-tag II proteins was inspected by SDS-polyacrylamide (15%) gel electrophoresis (SDS-PAGE). Molecular mass marker proteins for SDS-PAGE were purchased from LifeTechnologies (Darmstadt, Germany). The ectoine synthase activity of the (Sa)EctC protein was determined by HPLC-based enzyme assays. The initial enzyme activity assays were performed in a 30 μl-reaction volume for 20 min at 20°C. The used standard buffer (20 mM Tris, pH 8.0) contained 150 mM NaCl, 1 mM FeCl2, and 10 mM N-γ-ADABA. To determine optimal enzyme assay conditions for the (Sa)EctC-Strep-tag II protein, assay parameters and buffer conditions (e.g., the salt-concentrations, temperature, pH) were individually changed. The finally optimized assay buffer for ectoine synthase activity of the (Sa)EctC protein contained 20 mM Tris (pH 8.5), 200 mM NaCl, 1 mM FeCl2 and 10 mM N-γ-ADABA. Activity assays were run for 20 min at 15°C. Usually, 10 μg of the purified (Sa)EctC protein were added to start the enzyme assay. To assess the kinetic parameters of the ectoine synthase, varying concentrations of the substrates were used in the optimized assay buffer with a constant amount (10 μg) of the (Sa)EctC protein. The concentration of the natural EctC substrate N-γ-ADABA was varied between 0 and 40 mM, whereas that of N-α-ADABA was varied between 0 and 200 mM in the enzyme assays. Enzyme reactions were stopped by adding 30 μl of acetonitrile (100%) to the reaction vessel. The samples were centrifuged (13000 rpm, at room temperature for 5 min) to remove denatured proteins; the supernatant was subsequently analyzed for the formation of ectoine by HPLC analysis. Usually, 5- to 10-μl samples were injected into the HPLC system and the reaction product ectoine was analytically detected on a GROM-SIL Amino-1PR column (125 x 4 mm with a particle size of 3μm; purchased from GROM, Rottenburg-Hailfingen, Germany). Synthesis of ectoine by the purified (Sa)EctC-Strep-tag II protein and its mutant derivatives was monitored using a Infinity 1260 Diode Array Detector (DAD) (Agilent, Waldbronn, Germany) integrated into an Agilent 1260 Infinity LC system (Agilent). The ectoine content of the samples was quantified using the OpenLAB software suite (Agilent). The data shown for each ectC mutant (Table 1) were derived from two independent (Sa)EctC preparations, and each (Sa)EctC protein solution was assayed three times for its enzyme activity. To assess the dependency of the ectoine synthase for its enzyme activity on iron and other metals, purified and concentrated (Sa)EctC protein preparations (10 μM) were treated with different concentrations of EDTA for 10 minutes. They were subsequently dialyzed to remove the EDTA and the remaining (Sa)EctC enzyme activity was analyzed. To determine metal ion specificity of the ectoine synthase, 500 μl of the (Sa)EctC protein (100 μM) were initially treated with 1 mM EDTA for 10 minutes to obtain apo-(Sa)EctC protein preparations and the EDTA was then removed by dialysis. Enzyme activity assays with 10 μM of such protein preparations were then performed in the presence of either stoichiometric (10 μM) or excess amounts (1 mM) of FeCl2, FeCl3, ZnCl2, CoCl2, NiCl2, CuCl2, and MnCl2 to monitor metal ion specificity of the ectoine synthase. Prior to initiation of the enzyme reaction (by addition of the substrate), the (Sa)EctC protein solution was incubated with the different indicated metal ions for 10 minutes. To determine the oligomeric state of the (Sa)EctC protein in solution, we used high-performance liquid chromatography coupled to multi-angle light scattering detection (HPLC-MALS). A Bio SEC-5 HPLC column (Agilent Technologies Deutschland GmbH, Böblingen, Germany) with a pore size of 300 Å was equilibrated with 20 mM Tris-HCl (pH 7.5), 200 mM NaCl for high-performance liquid chromatography analysis. For these experiments, an Agilent Technologies system connected to a triple-angle light scattering detector (miniDAWN TREOS, Wyatt Technology Europe GmbH, Dernbach, Germany) followed by a differential refractive index detection system (Optilab t-rEX, Wyatt Technology) was used. Typically, 100 μl of purified (Sa)EctC protein (2 mg ml) was loaded onto the Bio SEC-5 HPLC column and the obtained data were analyzed with the ASTRA software package (Wyatt Technology). The elemental contents of P, Fe, Ni, Cu and Zn of the (Sa)EctC-Strep-Tag-II protein sample were determined by inductive-coupled plasma mass spectrometry (ICP-MS) using an Agilent 7900 ICP-MS system equipped with a HEN nebulizer and cooled scott spray chamber under standard operating conditions. The isotopes P, Fe, Fe, Ni, Ni, Ni, Cu, Cu, Zn, Zn, and Zn were measured under NoGas, He collision and H2 reaction mode conditions. Some isotopes are strongly interfered from the matrix (mainly Fe, Cu) in the NoGas mode and are therefore rejected. The (Sa)EctC protein samples and buffer blanks were diluted 100-fold with ultra pure water and spiked with 10 μg kg Y as the internal standard. The calibration of the ICP-MS was performed in the concentration range between 0.1 to 100 μg kg using a homemade P standard solution prepared from titrimetrically analyzed H3PO4 solution and from dilutions of a Merck ICP multi-element standard solution IV (Merck No. 111355, Darmstadt, Germany). To determine the iron content in our (Sa)EctC-Strep-Tag-II preparations photometrically , 10 nmol of the purified proteins were heated at 80°C for 10 min in 250 μl of a 1% HCl solution. The reaction assay was cooled down on ice and then centrifuged (13000 rpm, 10 min at room temperature). The supernatant was transferred to a new reaction tube, and 750 μl H2O, 50 μl of 10% hydroxylamine/HCl, and 250 μl of 0.1% phenanthroline were added to the reaction vessel. After 30 min of incubation at room temperature, the absorbance of the solution was measured at 512 nm. 5 to 40 nmol of ammonium iron(II) sulfate were used for calibration of the assay. Several conditions under which the (Sa)EctC protein formed crystals were found by using commercial screens (Nextal, Qiagen, Hilden, Germany; Molecular Dimensions, Suffolk, UK) in 96-well sitting drop plates (Corning 3553) at 12°C. Homogeneous (Sa)EctC protein (0.1 μl from a solution of 11 mg protein ml) was mixed with 0.1 μl reservoir solution and equilibrated against 50 μl reservoir solution. The most promising condition was found with a solution containing 0.05 M calcium acetate, 0.1 M sodium acetate (pH 4.5), and 40% (v/v) 1,2-propanediol from the Nextal Core IV suite (Qiagen, Hilden, Germany). A second condition under which (Sa)EctC crystallized was identified in microbatch setups (1 μl + 1 μl drops) using 20% (w/v) PEG 6000, 0.9 M lithium chloride, and 0.1 M citric acid (pH 5) from the Nextal Core II suite (Qiagen, Hilden, Germany). These conditions were optimized by grid screens around the initial condition and/or after the addition of tert-butanol as an additive. Large crystals were obtained either without any additive or after the addition of tert-butanol to the (Sa)EctC protein solution 30 minutes before the drops were spotted. Crystals reached their maximum dimensions of about 50 × 50 × 70 μm after 3–10 weeks. The crystals were fished after overlaying the drop with 2 μl mineral oil and flash frozen in liquid nitrogen. To obtain heavy atom derivatized crystals, methylmercury(II) chloride was added (final concentration: 0.5 mM) to the crystals in their drop for 30 minutes before they were fished and flash frozen in liquid nitrogen. Native data sets were collected from a single crystal of (Sa)EctC obtained from the various crystallization trials at the ERSF beamline ID23eh2 (Grenoble, France) at 100 K. These data sets were processed using the XDS package and scaled with XSCALE showing a maximum resolution of 1.2 Å. To obtain initial phases of (Sa)EctC, a mercury-derivatized crystal was used to collect a conservative dataset at 2.8 Å resolution. The data were processed and scaled as described above, before the program AUTORICKSHAW using single isomorphous replacement (SIRAS) , was used to localize the Hg atom, phase and built an initial model of the (Sa)EctC protein. This initial model was used as a template for molecular replacement on the 2.0 Å dataset revealing four monomers in the asymmetric unit. Once the 2.0 Å structure was refined, a single monomer of this structure was used as a template for molecular replacement to phase the 1.2 Å resolution dataset using the PHENIX software . Model building and refinement were performed using COOT , Refmac5 and Phenix_refine . Data refinement statistics and model content are summarized in Table 1. The atomic coordinates and structural factors have been deposited into the Protein Data Bank (PDB) (Brookhaven, USA) under the following accession codes: 5BXX (for the “semi-closed” (Sa)EctC structure) and 5BY5 (for the “open” (Sa)EctC structure). Figures of the crystal structures of SaEctC were prepared using the PyMol software suite (www.pymol.org) . We focused our biochemical and structural studies on the ectoine synthase from S. alaskensis [(Sa)EctC], a cold-adapted marine ultra-microbacterium , from which we recently also determined the crystal structure of the ectoine hydroxylase (EctD) in complex with either its substrate or its reaction product . We expressed a codon-optimized version of the S. alaskensis ectC gene in E. coli to produce a recombinant protein with a carboxy-terminally attached Strep-tag II affinity peptide to allow purification of the (Sa)EctC-Strep-Tag-II protein by affinity chromatography. The (Sa)EctC protein was overproduced and isolated with good yields (30–40 mg L of culture) and purity (S2a Fig). Conventional size-exclusion chromatography (SEC) has already shown that (Sa)EctC preparations produced in this fashion are homogeneous and that the protein forms dimers in solution . High performance liquid chromatography coupled with multi-angle light-scattering detection (HPLC-MALS) experiments carried out here confirmed that the purified (Sa)EctC protein was mono-disperse and possessed a molecular mass of 33.0 ± 2.3 kDa (S2b Fig). This value corresponds very well with the theoretically calculated molecular mass of an (Sa)EctC dimer (molecular mass of the monomer, including the Strep-tag II affinity peptide: 16.3 kDa). Such a quaternary assembly as dimer has also been reported for the EctC proteins from H. elongata and N. maritimus [14, 35]. The EctA-produced substrate of the ectoine synthase, N-γ-acetyl-L-2,4-diaminobutyric acid (N-γ-ADABA) (Fig 1), is commercially not available. We used alkaline hydrolysis of ectoine and subsequent chromatography on silica gel columns to obtain N-γ-ADABA in chemically highly purified form (S1a Fig). This procedure also yielded the isomer of N-γ-ADABA, N-α-acetyl-L-2,4-diaminobutyric acid (N-α-ADABA) (S1b Fig) . N-α-ADABA has so far not been considered as a substrate for EctC, but microorganisms that use ectoine as a nutrient produce it as an intermediate during catabolism . Using N-γ-ADABA as the substrate, we initially evaluated a set of biochemical parameters of the recombinant (Sa)EctC protein. S. alaskensis, from which the studied ectoine synthase was originally derived, is a microorganism that is well-adapted to a life in permanently cold ocean waters . Consistent with the physicochemical attributes of this habitat, the (Sa)EctC protein was already enzymatically active at 5°C, had a temperature optimum of 15°C and was able to function over a broad range of temperatures (S3a Fig). It possessed an alkaline pH optimum of 8.5 (S3b Fig), a value similar to the ectoine synthases from the halo-tolerant H. elongata (pH optimum of 8.5 to 9.0) , the alkaliphile M. alcaliphilum (pH optimum of 9.0) , and the acidophile Acidiphilium cryptum (pH optimum of 8.5 to 9.0) , whereas the EctC protein from N. maritimus has a neutral pH optimum (pH 7.0) . The salinity of the assay buffer had a significant influence on the maximal enzyme activity of the (Sa)EctC protein. An increase in either the NaCl or the KCl concentration led to an approximately 5-fold enhancement of the ectoine synthase activity. The maximum enzyme activity of (Sa)EctC occurred around 250 mM NaCl or KCl, respectively. (Sa)EctC is a highly salt-tolerant enzyme since it exhibited substantial enzyme activity even at NaCl and KCl concentrations of 1 M in the assay buffer (S3c and S3d Fig). The stimulation of EctC enzyme activity by salts has previously also been observed for other ectoine synthases [14, 35, 37, 38]. Considerations based on bioinformatics suggests that EctC belongs to the cupin superfamily [52–55]. Most of these proteins contain catalytically important transition state metals such as iron, copper, zinc, manganese, cobalt, or nickel [52–55]. Cupins contain two conserved motifs: G(X)5HXH(X)3,4E(X)6G and G(X)5PXG(X)2H(X)3N (the letters in bold represent those residues that often coordinate the metal). Inspection of a previous alignment of the amino acid sequences of 440 EctC-type proteins revealed that the canonical metal-binding motif(s) of cupin-type proteins [53–55] is not conserved among members of the extended ectoine synthase protein family [13, 14]. An abbreviated alignment of the amino acid sequence of EctC-type proteins is shown in Fig 2. The amino acid sequences of 20 selected EctC-type proteins are compared. Strictly conserved amino acid residues are shown in yellow. Dots shown above the (Sa)EctC protein sequence indicate residues likely to be involved in iron-binding (red), ligand-binding (green) and stabilization of the loop-architecture (blue). The conserved residue Tyr-52 with so-far undefined functions is indicated by a green dot circled in red. Secondary structural elements (α-helices and β-sheets) found in the (Sa)EctC crystal structure are projected onto the amino acid sequences of EctC-type proteins. Since variations of the above-described metal-binding motif occur frequently [52–54], we experimentally investigated the presence and nature of the metal that might be contained in the (Sa)EctC protein by inductive-coupled plasma mass spectrometry (ICP-MS). For this analysis we used recombinant (Sa)EctC preparations from three independent protein overproduction and purification experiments. The ICP-MS analyses yielded an iron content of 0.66 ± 0.06 mol iron per mol of protein and the used (Sa)EctC protein preparations also contained a minor amount of zinc (0.08 mol zinc per mol of protein). All other assayed metals (copper and nickel) were only present in trace amounts (0.01 mol metal per mol of protein, respectively). The presence of iron in these (Sa)EctC protein preparations was further confirmed by a colorimetric method that is based on an iron-complexing reagent ; this procedure yielded an iron-content of 0.84 ± 0.05 mol per mol of (Sa)EctC protein. Hence, both ICP-MS and the colorimetric method clearly established that the recombinantly produced ectoine synthase from S. alaskensis is an iron-containing protein. We note in this context, that the values obtained for the iron content of the (Sa)EctC proteins varied by approximately 10 to 20% between the two methods. The reason for this difference is not known, but indicates that the well established colorimetric assay probably overestimates the iron content of (Sa)EctC protein preparations to a certain degree. The iron detected in the (Sa)EctC protein preparations could serve a structural role, or most likely, could be critical for enzyme catalysis as is the case for many members of the cupin superfamily [52–55]. To address these questions, we incubated the (Sa)EctC enzyme with increasing concentrations of the metal chelator ethylene-diamine-tetraacetic-acid (EDTA) and subsequently assayed ectoine synthase activity. The addition of very low concentrations of EDTA (0.05 mM) to the EctC enzyme already led to a noticeable inhibition of the ectoine synthase activity and the presence of 1 mM EDTA completely inhibited the enzyme (Fig 3a). (a) Impact of the iron-chelator EDTA on the enzyme activity of the purified (Sa)EctC protein. Metal depletion and reconstitution experiments with (b) stoichiometric and (c) excess amounts of metals. The (Sa)EctC protein was present at a concentration of 10 μM. The level of enzyme activity given in (b) is benchmarked relative to that of ectoine synthase enzyme assays in which 1 mM FeCl2 was added. We then took such an inactivated enzyme preparation, removed the EDTA by dialysis, and added stoichiometric amounts (10 μM) of various metals to the (Sa)EctC enzyme. The addition of FeCl2 to the enzyme assay restored enzyme activity to about 38%, whereas the addition of ZnCl2 or CoCl2 rescued (Sa)EctC enzyme activity only to 5% and 3%, respectively. All other tested metals, including Fe, were unable to restore activity (Fig 3b). When the concentration of the various metals in the enzyme assay was increased 100-fold, Fe exhibited again the strongest stimulating effect on enzyme activity, and rescued enzyme activity to a degree similar to that exhibited by (Sa)EctC protein preparations that had not been inactivated through EDTA treatment (Fig 3c). However, a large molar excess of other transition-state metals (zinc, cobalt, nickel, copper, and manganese) typically found in members of the cupin superfamily [52–55] allowed the partial rescue of ectoine synthase activity as well (Fig 3c). This is in line with literature data showing that cupin-type enzymes are often promiscuous with respect to the use of the catalytically important metal [56–58]. Based on the data presented in S3 Fig, we formulated an optimized activity assay for the ectoine synthase of S. alaskensis and used it to determined the kinetic parameters for the (Sa)EctC enzyme for both its natural substrate N-γ-ADABA [21, 35] and the isomer N-α-ADABA. The EctC-catalyzed ring-closure of N-γ-ADABA to form ectoine exhibited Michaelis-Menten-kinetics with an apparent Km of 4.9 ± 0.5 mM, a vmax of 25.0 ± 0.8 U/mg and a kcat of 7.2 s (S4a Fig). Given the chemical relatedness of N-α-ADABA to the natural substrate (N-γ-ADABA) of the ectoine synthase (S1a and S1b Fig), we wondered whether (Sa)EctC could also use N-α-ADABA to produce ectoine. This was indeed the case. (Sa)EctC catalyzed this reaction with Michaelis-Menten-kinetics exhibiting an apparent Km of 25.4 ± 2.9 mM, a vmax of 24.6 ± 1.0 U/mg and a kcat 0.6 s (S4b Fig). Hence, N-α-ADABA is a newly recognized substrate for ectoine synthase. However, both the affinity (Km) of the (Sa)EctC protein and its catalytic efficiency (kcat/Km) were strongly reduced in comparison with N-γ-ADABA. The Km dropped fife-fold from 4.9 ± 0.5 mM to 25.4 ± 2.9 mM, and the catalytic efficiency was reduced from 1.47 mM s to 0.02 mM s, a 73-fold decrease. Both N-γ-ADABA and N-α-ADABA are concomitantly formed during the enzymatic hydrolysis of the ectoine ring during catabolism . Our finding that N-α-ADABA is a substrate for ectoine synthase has bearings for an understanding of the physiology of those microorganisms that can both synthesize and catabolize ectoine . However, these types of microorganisms should still be able to largely avoid a futile cycle since the affinity of ectoine synthase for N-γ-ADABA and N-α-ADABA, and its catalytic efficiency for the two compounds, differs substantially (S4a and S4b Fig). Since no crystal structure of ectoine synthase has been reported, we set out to crystallize the (Sa)EctC protein. Attempts to obtain crystals of (Sa)EctC in complex either with its substrate N-γ-ADABA or its reaction product ectoine were not successful. However, two crystal forms of the (Sa)EctC protein in the absence of the substrate were obtained. Crystal form A diffracted to 1.2 Å and had a unit cell of a = 72.71 b = 72.71 c = 52.33 Å and α = 90 β = 90 γ = 120° displaying a P3221 symmetry (S1 Table). Crystal form B diffracted to 2.0 Å and had a unit cell of a = 97.52 b = 43.96 c = 138.54 Å and α = 90 β = 101.5 γ = 120° and displayed a C2 symmetry (S1 Table). Attempts to solve the crystal structure of the (Sa)EctC protein by molecular replacement has previously failed . However, we were able to obtain crystals of form B that were derivatized with mercury and these diffracted up to 2.8 Å (S1 Table). This dataset was used to derive an initial structural model of the (Sa)EctC protein, which in turn was employed as a template for molecular replacement to phase the native dataset (2.0 Å) of crystal form B. After several rounds of manual model building and refinement, four monomers of (Sa)EctC were identified and the crystal structure was refined to a final Rcryst of 21.1% and an Rfree of 24.8% (S1 Table). Finally, a monomer of this structure was used as a template for molecular replacement to phase the high-resolution (1.2 Å) dataset of crystal form A, which was subsequently refined to a final Rcryst of 12.4% and an Rfree of 14.9% (S1 Table). The two EctC structures that we determined revealed that the ectoine synthase belongs to the cupin superfamily [52–55] with respect to its overall fold (Fig 4a–4c). However, they represent two different states of the 137 amino acids comprising (Sa)EctC protein (Fig 2). First, the 1.2 Å structure reveals the spatial configuration of the (Sa)EctC protein ranging from amino acid Met-1 to Glu-115; hence, it lacks 22 amino acids at the carboxy-terminus of the authentic (Sa)EctC protein. This structure adopts an open conformation with respect to the typical fold of cupin barrels [52–55] and is therefore termed in the following the “open” (Sa)EctC structure (Fig 4b). In this structure no metal co-factor was identified. The second crystal structure of the (Sa)EctC protein was solved at a resolution of 2.0 Å and contained four molecules of the protein in the asymmetric unit of which protomer A comprised amino acid Met-1 to Gly-121 and adopts a closed conformation. Hence, it still lacks 16 amino acid residues of the carboxy-terminus of the authentic 137 amino acids comprising (Sa)EctC protein (Fig 2). We therefore cannot exclude that this crystal structure does not represent the fully closed state of the ectoine synthase; consequently, we tentatively termed it the “semi-closed” (Sa)EctC structure. Interestingly, the three other monomers present in the asymmetric unit all range from Met-1 to Glu-115 and adopt a conformation similar to the “open” EctC structure. (a) The overall structure of the “semi-closed” (Sa)EctC resolved at 2.0 Å is depicted in green in a cartoon (upper panel) and surface (lower panel) representation. The β-strands are numbered β1-β11 and the helices α-I to α-II. (b) The overall structure of the “open” (Sa)EctC was resolved at 1.2 Å and is depicted in yellow in a cartoon (upper panel) and surface (lower panel) representation. The entrance to the active site of the ectoine synthase is marked. (c) Overlay of the “semi-closed” and “open” (Sa)EctC structures. The overall structure of (Sa)EctC is basically the same in both crystals except for the carboxy-terminus, which covers the entry of one side of the cupin barrel from the surroundings in monomer A in the “semi-closed” structure. This is reflected by the calculated root mean square deviation (RMSD) of the Cα atoms that was about 0.56 Å (over 117 residues) when the four “open” monomers were compared with each other. However, the “semi-closed” monomer has a slightly higher RMSD of 1.4 Å (over 117 residues) when compared with the “open” 2.0 Å structure. Therefore, we describe in the following the overall structure for the “semi-closed” form of the (Sa)EctC protein and subsequently highlight the structural differences between the “open” and “semi-closed” forms in more detail. The structure of the “semi-closed” (Sa)EctC protein consists of 11 β-strands (β1-β11) and two α-helices (α-I and α-II) (Fig 4a). The β-strands form two anti-parallel β-sheets: β2 β3, β4, β11, β6, and β9, and a smaller three-stranded β-sheet (β7, β8, and β10), respectively. These two β-sheets pack against each other, forming a cup-shaped β-sandwich with a topology characteristic for the cupin-fold [52–55]. Hence, (Sa)EctC adopts an overall bowl shape in which one side is opened towards the solvent (Fig 4a to 4c). In the “semi-closed” structure, a longer carboxy-terminal tail is visible in the electron density, folding into a small helix (α-II) that closes the active site of the (Sa)EctC protein (Fig 4a). The formation of this α-II helix induces a reorientation and shift of a long unstructured loop (as observed in the “open” structure) connecting β4 and β6, resulting in the formation of the stable β-strand β5 as observed in the “semi-closed”state of the (Sa)EctC protein (Fig 4a). Structural comparison analyses using the DALI server revealed that (Sa)EctC adopts a fold similar to other members of the cupin superfamily [52–55]. The highest structural similarities are observed for the Cupin 2 conserved barrel domain protein (YP_751781.1) from Shewanella frigidimarina (PDB accession code: 2PFW) with a Z-score of 13.1 and an RMSD of 2.2 Å over 104 Cα-atoms (structural data for this protein have been deposited in the PDB but no publication connected to this structure is currently available), a manganese-containing cupin (TM1459) from Thermotoga maritima (PDB accession code: 1VJ2) with a Z-score of 12.8 and an RMSD of 2.0 Å over 103 Cα-atoms , the cyclase RemF from Streptomyces resistomycificus (PDB accession code: 3HT1 with a Z-score of 11.9 and an RMSD of 1.9 Å over 102 Cα-atoms) , and an auxin-binding protein 1 from Zea mays (PDB accession code: 1LR5) with an Z-score of 11.8 and an RMSD of 2.8 Å over 104 Cα-atoms) . Our data classify EctC, in addition to the polyketide cyclase RemF , as the second known cupin-related enzyme that catalyze a cyclocondensation reaction. Next to RemF and the aldos-2-ulose dehydratase/isomerase , the ectoine synthase is only the third characterized dehydratase within the cupin superfamily. Both the SEC analysis and the HPLC-MALS experiments (S2b Fig) have shown that the ectoine synthase from S. alaskensis is a dimer in solution. The crystal structure of this protein reflects this quaternary arrangement. In the “semi-closed” crystal structure, (Sa)EctC has crystallized as a dimer of dimers within the asymmetric unit. This dimer (Fig 5a and 5b) is composed of two monomers arranged in a head-to-tail orientation and is stabilized via strong interactions mediated by two antiparallel β-strands, β-strand β1 (sequence MIVRN) from monomer A and β-strand β8 from monomer B (sequence GVMYAL) (Fig 5c). The strong interactions between these β-strands rely primarily on backbone contacts. In addition to these interactions, some weaker hydrophobic interactions are also observed between the two monomers in some loops connecting the β-strands. As calculated with PDBePISA , the surface area buried upon dimer formation is 1462 Å, which is 20.5% of the total accessible surface of a monomer of this protein. Both values fall within the range for known functional dimers. (a) Top-view of the dimer of the (Sa)EctC protein. The position of the water molecule, described in detail in the text, is shown in one of the monomers as an orange sphere. (b) Side-view of a (Sa)EctC dimer allowing an assessment of the dimer interface formed by two β-strands of each monomer. (c) Close-up representation of the dimer interface mediated by beta-strand β1 and β6. In the “open” (Sa)EctC structure, one monomer is present in the asymmetric unit. We therefore inspected the crystal packing and analyzed the monomer-monomer interactions with symmetry related molecules to elucidate whether a physiologically relevant dimer could be deduced from this crystal form as well. Indeed, a similar dimer configuration to the one described for the “semi-closed” (Sa)EctC structure is observed with the same monomer-monomer interactions mediated by the two β-sheets. The crystallographic two-fold axis present within the crystal symmetry is located exactly in between the two monomers, resulting in a monomer within the asymmetric unit. Hence, the same dimer observed in the “semi-closed” structure of (Sa)EctC can also be observed in the “open” structure. Interestingly, the proteins identified by the above-described DALI search not only have folds similar to EctC, but are also functional dimers that adopt similar monomer-monomer interactions within the dimer assembly as deduced from the inspection of the corresponding PDB files (2PFW, 3HT1, 1VJ2, 1LR5). The cupin core [52–55] represents the structural framework of ectoine synthase (Figs 4 and 5). The major difference in the two crystal structures of the (Sa)EctC protein reported here is the orientation of the carboxy-terminus. Some amino acids located in the carboxy-terminal region of the 137 amino acids comprising (Sa)EctC protein are highly conserved (Fig 2) within the extended EctC protein family [13, 14]. At the end of β-strand β11, two consecutive conserved proline residues (Pro-109 and Pro-110) are present that are responsible for a turn in the main chain of the (Sa)EctC protein. In the “semi-closed” (Sa)EctC structure, the visible electron density of the carboxy-terminus is extended by 7 amino acid residues and ends at position Gly-121. These additional amino acids fold into a small helix, which seals the open cavity of the cupin-fold of the (Sa)EctC protein (Fig 4a). Furthermore, this helix is stabilized via interactions with the loop region between β-strands β4 and β6, thereby inducing a structural rearrangement. This induces the formation of β-strand β5, which is not present when the small C-terminal helix is absent as observed in the “open” (Sa)EctC structure. As a result, the newly formed β-strand β5 is reoriented and moved by 2.4 Å within the “semi-closed” (Sa)EctC structure (Fig 4a to 4c). It is worth mentioning that β-strand β5 is located next to His-93, which in all likelihood involved in metal binding (see below). The position of this His residue is slightly shifted in both (Sa)EctC structures, likely the result of the formation of β-strand β5. Therefore the sealing of the cupin fold, as described above, seem to have an indirect influence on the architecture of the postulated iron-binding site. The consecutive Pro-109 and Pro-110 residues found at the end of β-strand β11are highly conserved in EctC-type proteins (Fig 2) [13, 14]. They are responsible for redirecting the main chain of the remaining carboxy-terminus (27 amino acid residues) of (Sa)EctC to close the cupin fold. In the “semi-closed” structure this results in a complete closure of the entry of the cupin barrel (Fig 4a to 4c). In the “open” (Sa)EctC structure, both proline residues are visible in the electron density; however, almost directly after Pro-110, the electron density is drastically diminished caused by the flexibility of the carboxy-terminus. A search for partners interacting with Pro-109 revealed that it interacts via its backbone oxygen with the side chain of His-55 as visible in both the “open” and “semi-closed” (Sa)EctC structures. The Pro-109/His-55 interaction ensures the stable orientation of both proline residues at the end of β-strand β11. Since these proline residues are followed by the carboxy-terminal region of the (Sa)EctC protein, the interaction of His-55 with Pro-109 will likely play a substantial role in spatially orienting this very flexible part of the protein. In addition to the interactions between Pro-109 and His-55, the carboxy-terminal region of (Sa)EctC is held in position via an interaction of Glu-115 with His-55, which stabilizes the conformation of the small helix in the carboxy-terminus further. The interaction between Glu-115 and His-55 is only visible in the “semi-closed” structure where the partially extended carboxy-terminus is resolved in the electron density. In the “open” structure of the (Sa)EctC protein, this interaction does not occur since Glu-115 is rotated outwards (Fig 6a and 6b). Hence, one might speculate that this missing interaction might be responsible for the flexibility of the carboxy-terminus in the “open” (Sa)EctC structure and consequently results in less well defined electron density in this region. (a) The described water molecule (depicted as orange sphere) is bound via interactions with the side chains of Glu-57, Tyr-85, and His-93. The position occupied by this water molecule represents probably the position of the Fe cofactor in the active side of the ectoine synthase. His-55 interacts with the double proline motif (Pro-109 and Pro-110). It is further stabilized via an interaction with the side chain of Glu-115 which is localized in the flexible carboxy-terminus (colored in orange) of (Sa)EctC that is visible in the “semi-closed” (Sa)EctC structure. (b) An overlay of the “open” (colored in light blue) and the “semi-closed” (colored in green) structure of the (Sa)EctC protein. In the “semi-closed” structure of (Sa)EctC, each of the four monomers in the asymmetric unit contains a relative strong electron density positioned within the cupin barrel. Since (Sa)EctC is a metal containing protein (Fig 3), we tried to fit either Fe, or Zn ions into this density and also refined occupancy. Only the refinement of Fe resulted in a visibly improved electron density, however with a low degree of occupancy. This possible iron molecule is bound via interactions with Glu-57, Tyr-85 and His-93 (Fig 6a and 6b). The distance between the side chains of these residues and the (putative) iron co-factor is 3.1 Å for Glu-57, 2.9 Å for Tyr-85, and 2.9 Å for His-93, respectively. These distances are to long when compared to other iron binding sites, a fact that might be caused by the absence of the proper substrate in the (Sa)EctC crystal structure. Since both the refinement and the distance did not clearly identify an iron molecule, we decided to conservatively place a water molecule at this position. The position of this water molecule is described in more detail below and is highlighted in Figs 5a and 5b and 6a and 6b as a sphere. Interestingly, all three amino acids coordinating this water molecule are strictly conserved within an alignment of 440 members of the EctC protein family (for an abbreviated alignment of EctC-type proteins see Fig 2). In the “open” structure of the (Sa)EctC protein, electron density is visible where the presumptive iron is positioned in the “semi-closed” structure. However, this electron density fits perfectly to a water molecule and not to an iron, and the water molecule was clearly visible after the refinement at this high resolution (1.2 Å) of the “open” (Sa)EctC structure. In a superimposition of both (Sa)EctC crystal structures, the spatial arrangements of the side chains of the three amino acids (Glu-57, Tyr-85, and His-93) likely to contact the iron in the “semi-closed” structure match nicely with those of the corresponding residues of the “iron-free” “open” structure (Fig 6b). Only His-93 is slightly rotated inwards in the “semi-closed” structure, most likely due to formation of β-strand β5 as described above. Taken together, this observations indicate, that the architecture of the presumptive iron-binding site is pre-set for the binding of the catalytically important metal by the ectoine synthase. Of note is the different spatial arrangement of the side-chain of Tyr-52 (located in a loop after the end of β-strand β5) in the “open” and “semi-closed” (Sa)EctC structures. In the “semi-closed” structure, the hydroxyl-group of the side-chain of Tyr-52 points towards the iron (Fig 6a and 6b), but the corresponding distance (3.9 Å) makes it highly unlikely that Tyr-52 is directly involved in metal binding. Nevertheless, its substitution by an Ala residue causes a strong decrease in iron-content and enzyme activity of the mutant protein (Table 1). It becomes apparent from an overlay of the “open” and “semi-closed” (Sa)EctC crystal structures that the side-chain of Tyr-52 rotates away from the position of the presumptive iron, whereas the side-chains of those residues that probably contacting the metal directly [Glu-57, Tyr-85, and His-93], remain in place (Fig 6a and 6b). Since Tyr-52 is strictly conserved in an alignment of 440 EctC-type proteins (Fig 2), we speculate that it might be involved in contacting the substrate of the ectoine synthase and that the absence of N-γ-ADABA in our (Sa)EctC crystal structures might endow the side chain of Tyr-52 with extra spatial flexibility. To further analyze the putative iron binding site (Fig 6a), we performed structure-guided site-directed mutagenesis and assessed the resulting (Sa)EctC variants for their iron content and studied their enzyme activity. When those three residues (Glu-57, Tyr-85, His-93) that likely form the mono-nuclear iron center in the (Sa)EctC crystal structure were individually replaced by an Ala residue, both the catalytic activity and the iron content of the mutant proteins was strongly reduced (Table 1). For some of the presumptive iron-coordinating residues, additional site-directed mutagenesis experiments were carried out. To verify the importance of the negative charge in the position of Glu-57, we created an Asp variant. This mutant protein rescued the enzyme activity and iron content of the Ala substitution substantially (Table 1). We also replaced Tyr-85 with either a Phe or a Trp residue and both mutant proteins largely lost their catalytic activity and iron content (Table 1) despite the fact that these substitutions were conservative. Collectively, these data suggest that the hydroxyl group of the Tyr-85 side chain is needed for the binding of the iron (Fig 6a). We also replaced the presumptive iron-binding residue His-93 by an Asn residue, yielding a (Sa)EctC protein variant that possessed an enzyme activity of 23% and iron content of only 14% relative to that of the wild-type protein (Table 1). Collectively, the data addressing the functionality of the putative iron-coordinating residues (Glu-57, Tyr-85, His-93) buttress our notion that the Fe present in the (Sa)EctC protein is of catalytic importance. Despite considerable efforts, either by trying co-crystallization or soaking experiments, we were not able to obtain a (Sa)EctC crystal structures that contained either the substrate N-γ-ADABA, or ectoine, the reaction product of ectoine synthase (Fig 1). However, in the “semi-closed” (Sa)EctC structure where the carboxy-terminal loop is largely resolved, a long stretched electron density feature was detected in the predicted active site of the enzyme; it remained visible after crystallographic refinement. This is in contrast to the high-resolution “open” structure of the (Sa)EctC protein where no additional electron density was observed after refinement. We tried to fit all compounds used in the buffers during purification and crystallization into the observed electron density, but none matched. This observation indicates that the chemically undefined ligand was either trapped by the (Sa)EctC protein during its heterologous production in E. coli or during crystallization. Since we used PEG molecules in the crystallization conditions, the observed density might stem from an ordered part of a PEG molecule, or low molecular weight PEG species that might have been present in the PEG preparation used in our experiments. We therefore stress that we cannot identify neither the true chemically nature of this compound, nor its precise origin. Estimating from the dimensions of the electron density feature, we modeled the chemically undefined compound trapped by the (Sa)EctC protein as a hexane-1,6-diol molecule (PDB identifier: HEZ) to best fit the observed electron density. However, to the best of our knowledge, hexane-1,6-diol is not part of the E. coli metabolome . Despite these notable limitations, we considered the serendipitously trapped compound as a mock ligand that might provide useful insights into the spatial positioning of the true EctC substrate and those residues that coordinate it within the ectoine synthase active site. We note that both N-γ-ADABA and hexane-1,6-diol are both C6-compounds and display similar length (Fig 7a). (a) The observed electron density in the active site of the “semi-closed” structure of (Sa)EctC is modeled as a hexane-1,6-diol molecule and compared with the electron density of the N-γ-ADABA substrate of the ectoine synthase to emphasize the similarity in size of these compounds. (b) The presumable binding site of the iron co-factor and of the modeled hexane-1,6-diol molecule is depicted. The amino acid side chains involved in iron-ligand binding are colored in blue and those involved in the binding of the chemically undefined ligand are colored in green using a ball and stick representation. The flexible carboxy-terminal loop of (Sa)EctC is highlighted in orange. The electron density was calculated as an omit map and contoured at 1.0 σ. We refined the (Sa)EctC structure with the trapped compound, and by doing so, the refinement parameters (especially R- and Rfree-factor) dropped by 1.5%. We also calculated an omit map and the electron density reappeared (Fig 7b). When analyzing the interactions of this compound within the (Sa)EctC protein, we found that it is bound via interactions with Trp-21 and Ser-23 of β-sheet β3, Thr-40 located in β-sheet β4, and Cys-105 and Phe-107, which are both part of β-sheet β11. Remarkably, all of these residues are highly conserved throughout the extended EctC protein family (Fig 2) [13, 14]. In a previous alignment of the amino acid sequences of 440 EctC-type proteins, 13 amino acids were identified as strictly conserved residues . These correspond to amino acids Thr-40, Tyr-52, His-55, Glu-57, Gly-64, Tyr-85- Leu-87, His-93, Phe-107, Pro-109, Gly-113, Glu-115, and His-117 in the (Sa)EctC protein (Fig 2). Amino acid residues Gly-64, Pro-109, and Gly-113 likely fulfill structural roles since they are positioned either at the end or at the beginning of β-strands and α-helices. We considered the remaining ten residues as important either for ligand binding, for catalysis, or for the structurally correct orientation of the flexible carboxy-terminus of the (Sa)EctC protein. As described above, the side chains of Glu-57, Tyr-85, and His-93 are probably involved in iron binding (Table 1 and Fig 6a). In view of the (Sa)EctC structure with the serendipitously trapped compound (Fig 7b), we probed the functional importance of the seven residues that contact this ligand by structure-guided site-directed mutagenesis (Table 1). Each of these mutant (Sa)EctC proteins was overproduced in E. coli and purified by affinity chromatography; they all yielded pure and stable protein preparations. We benchmarked the activity of the (Sa)EctC variants in a single time-point enzyme assay under conditions where 10 μM of the wild-type (Sa)EctC protein converted almost completely the supplied 10 mM N-γ-ADABA substrate to 9.33 mM ectoine within a time frame of 20 min. In addition, we determined the iron content of each of the mutant (Sa)EctC protein by a colorimetric assay (Table 1). The side chains of the evolutionarily conserved Trp-21, Ser-23, Thr-40, Cys-105, and Phe-107 residues (Fig 2) make contacts with the chemically undefined ligand that we observed in the “semi-closed” (Sa)EctC structure (Fig 7b). We replaced each of these residues with an Ala residue and found that none of them had an influence on the iron content of the mutant proteins. However, their catalytic activity was substantially impaired (Table 1). Thr-40 is positioned on β-strand β5 and its side chain protrudes into the lumen of the cupin barrel formed by the (Sa)EctC protein (Fig 7b). We also replaced Phe-107 with either an Tyr or an Trp residue: the Phe-107/Tyr substitution possessed near wild-type enzyme activity (about 95%) and the full iron content, but the Phe-107/Trp substitution possessed only 12% enzyme activity and 72% iron content compared to the wild-type protein. The properties of these mutant proteins indicate that the aromatic side chain at position 107 of (Sa)EctC is of importance but that a substitution with a bulky aromatic side chain is strongly detrimental to enzyme activity and concomitantly moderately impairs iron binding. Replacement of the only Cys residue in (Sa)EctC (Cys-105; Fig 2) by a Ser residue, a configuration that is naturally found in two EctC proteins among 440 inspected amino acid sequences , yielded a (Sa)EctC variant with 84% wild-type activity and an iron content similar to that of the wild-type protein. However, the Cys-105/Ala variant was practically catalytically inactive while largely maintaining its iron content (Table 1). Since the side-chains of Cys residues are chemically reactive and often participate in enzyme catalysis, Cys-105 (or Ser-105) might serve such a role for ectoine synthase. We observed two amino acid substitutions that simultaneously strongly affected enzyme activity and iron content; these were the Tyr-52/Ala and the His-55/Ala (Sa)EctC protein variants (Table 1). Based on the (Sa)EctC crystal structures that we present here, we can currently not firmly understand why the replacement of Tyr-52 by Ala impairs enzyme function and iron content so drastically (Table 1). This is different for the His-55/Ala substitution. The carboxy-terminal region of the (Sa)EctC protein is held in its position via an interaction of Glu-115 with His-55, where His-55 in turn interacts with Pro-110 (Fig 6a and 6b). Each of these residues is evolutionarily highly conserved . The individual substitution of either Glu-115 or His-55 by an Ala residue is predicted to disrupt this interactive network and therefore should affect enzyme activity. Indeed, the Glu-115/Ala and the His-55/Ala substitutions possessed only 21% and 16% activity of the wild-type protein, respectively (Table 1). The Glu-115/Ala mutant possessed wild-type levels of iron, whereas the iron content of the His-55/Ala substitutions dropped to 15% of the wild-type level (Table 1). We also replaced Glu-115 with a negatively charged residue (Asp); this (Sa)EctC variant possessed wild-type levels of iron and still exhibited 77% of wild-type enzyme activity. Collectively, these data suggest that the correct positioning of the carboxy-terminus of the (Sa)EctC protein is of structural and functional importance for the activity of the ectoine synthase. Residues Leu-87 and Asp-91 are highly conserved in the ectoine synthase protein family [13, 14]. The replacement of Leu-87 by Ala led to a substantial drop in enzyme activity (Table 1). Conversely, the replacement of Asp-91 by Ala and Glu, resulted in (Sa)EctC protein variants with 80% and 98% enzyme activity, respectively (Table 1). We currently cannot comment on possible functional role Asp-91. However, Leu-87 is positioned at the end of one of the β-sheets that form the dimer interface (Fig 5c) and it might therefore possess a structural role. It is also located near Tyr-85, one of the residues that probably coordinate the iron molecule with in the (Sa)EctC active site (Fig 6a) and therefore might exert indirect effects. His-117 is a strictly conserved residue and its substitution by an Ala residue results in a drop of enzyme activity (down to 44%) and an iron content of 83% (Table 1). We note that His-117 is located close to the chemically undefined ligand in the (Sa)EctC structure (Fig 7b) and might thus play a role in contacting the natural substrate of the ectoine synthase. As an internal control for our mutagenesis experiments, we also substituted Thr-41 and His-51, two residues that are not evolutionarily conserved in EctC-type proteins [13, 14] with Ala residues. Both (Sa)EctC protein variants exhibited wild-type level enzyme activities and possessed a iron content matching that of the wild-type (Table 1). This illustrates that not every amino acid substitution in the (Sa)EctC protein leads to an indiscriminate impairment of enzyme function and iron content. The crystallographic data presented here firmly identify ectoine synthase (EctC), an enzyme critical for the production of the microbial cytoprotectant and chemical chaperone ectoine [12, 14, 20, 33], as a new member of the cupin superfamily. The overall fold and bowl shape of the (Sa)EctC protein (Figs 4 and 5) with its 11 β-strands (β1-β11) and two α-helices (α-I and α-II) closely adheres to the design principles typically found in crystal structures of cupins [52–55]. In addition to the ectoine synthase, the polyketide cyclase RemF is the only other currently known cupin-related enzyme that catalyze a cyclocondensation reaction although the substrates of EctC and RemF are rather different. As a consequence of the structural relatedness of EctC and RemF and the type of chemical reaction these two enzymes catalyze, is now understandable why bona fide EctC-type proteins are frequently (mis)-annotated in microbial genome sequences as “RemF-like” proteins. The pro- and eukaryotic members of the cupin superfamily perform a variety of both enzymatic and non-enzymatic functions that are built upon a common structural scaffold [53, 55]. Most cupins contain transition state metals that can promote different types of chemical reactions [52, 54]. Except for some cupin-related proteins that seem to function as metallo-chaperones , the bound metal is typically an essential part of the active sites [55, 56, 66, 67]. We report here for the first time that the ectoine synthase is a metal-dependent enzyme. ICP-MS, metal-depletion and reconstitution experiments (Fig 3) consistently identify iron as the biologically most relevant metal for the EctC-catalyzed cyclocondensation reaction. However, as observed with other cupins [56–58], EctC is a somewhat promiscuous enzyme as far as the catalytically important metal is concerned when they are provided in large molar excess (Fig 3c). Although some uncertainty remains with respect to the precise identity of amino acid residues that participate in metal binding by (Sa)EctC, our structure-guided site-directed mutagenesis experiments targeting the presumptive iron-binding residues (Fig 6a and 6b) demonstrate that none of them can be spared (Table 1). The architecture of the metal center of ectoine synthase seems to be subjected to considerable evolutionary constraints. The three residues (Glu-57, Tyr-85, His-93) that we deem to form it (Figs 6 and 7b) are strictly conserved in a large collection of EctC-type proteins originating from 16 bacterial and three archaeal phyla (Fig 2) [13, 14]. We also show here for the first time that, in addition to its natural substrate N-γ-ADABA, EctC also converts the isomer N-α-ADABA into ectoine, albeit with a 73-fold reduced catalytic efficiency (S3a and S3b Fig). Hence, the active site of ectoine synthase must possess a certain degree of structural plasticity, a notion that is supported by the report on the EctC-catalyzed formation of the synthetic compatible solute ADPC through the cyclic condensation of two glutamine molecules . Our finding that N-α-ADABA serves as a substrate for ectoine synthase has physiologically relevant ramifications for those microorganisms that can both synthesize and catabolize ectoine , since they need to prevent a futile cycle of synthesis and degradation when N-α-ADABA is produced as an intermediate in the catabolic route. Although we cannot identify the true chemical nature of the C6 compound that was trapped in the (Sa)EctC structure nor its precise origin, we treated this compound as a proxy for the natural substrate of ectoine synthase, which is a C6 compound as well (Fig 7a). We assumed that its location and mode of binding gives, in all likelihood, clues as to the position of the true substrate N-γ-ADABA within the EctC active site. Indeed, site-directed mutagenesis of those five residues that contact the unknown C6 compound (Fig 7b) yielded (Sa)EctC variants with strongly impaired enzyme function but near wild-type levels of iron (Table 1). This set of data and the fact that the targeted residues are strongly conserved among EctC-type proteins (Fig 2) [13, 14] is consistent with their potential role in N-γ-ADABA binding or enzyme catalysis. We therefore surmise that our crystallographic data and the site-directed mutagenesis study reported here provide a structural and functional view into the architecture of the EctC active site (Fig 7b). The ectoine synthase from the cold-adapted marine bacterium S. alaskensis can be considered as a psychrophilic enzyme (S3a Fig), types of proteins with a considerable structural flexibility [68, 69]. This probably worked to the detriment of our efforts in solving crystal structures of the full-length (Sa)EctC protein in complex with either N-γ-ADABA or ectoine. Because microbial ectoine producers can colonize ecological niches with rather different physicochemical attributes, it seems promising to exploit this considerable biodiversity [13, 14] to identify EctC proteins with enhanced protein stability. It is hoped that these can be further employed to obtain EctC crystal structures with either the substrate or the reaction product. Together with our finding that ectoine synthase is metal dependent, these crystal structures should allow a more detailed understanding of the chemistry underlying the EctC-catalyzed cyclocondensation reaction.
PMC4850288	Crystal Structure and Activity Studies of the C11 Cysteine Peptidase from Parabacteroides merdae in the Human Gut Microbiome*	Clan CD cysteine peptidases, a structurally related group of peptidases that include mammalian caspases, exhibit a wide range of important functions, along with a variety of specificities and activation mechanisms. However, for the clostripain family (denoted C11), little is currently known. Here, we describe the first crystal structure of a C11 protein from the human gut bacterium, Parabacteroides merdae (PmC11), determined to 1.7-Å resolution. PmC11 is a monomeric cysteine peptidase that comprises an extended caspase-like α/β/α sandwich and an unusual C-terminal domain. It shares core structural elements with clan CD cysteine peptidases but otherwise structurally differs from the other families in the clan. These studies also revealed a well ordered break in the polypeptide chain at Lys, resulting in a large conformational rearrangement close to the active site. Biochemical and kinetic analysis revealed Lys to be an intramolecular processing site at which cleavage is required for full activation of the enzyme, suggesting an autoinhibitory mechanism for self-preservation. PmC11 has an acidic binding pocket and a preference for basic substrates, and accepts substrates with Arg and Lys in P1 and does not require Ca for activity. Collectively, these data provide insights into the mechanism and activity of PmC11 and a detailed framework for studies on C11 peptidases from other phylogenetic kingdoms.Cysteine peptidases play crucial roles in the virulence of bacterial and other eukaryotic pathogens. In the MEROPS peptidase database (1), clan CD contains groups (or families) of cysteine peptidases that share some highly conserved structural elements (2). Clan CD families are typically described using the name of their archetypal, or founding, member and also given an identification number preceded by a “C,” to denote cysteine peptidase. Although seven families (C14 is additionally split into three subfamilies) have been described for this clan, crystal structures have only been determined from four: legumain (C13) (3), caspase (C14a) (4), paracaspase (C14b(P) (5), metacaspase (C14b(M) (6), gingipain (C25) (7), and the cysteine peptidase domain (CPD) of various toxins (C80) (8). No structural information is available for clostripain (C11), separase (C50), or PrtH-peptidase (C85). Clan CD enzymes have a highly conserved His/Cys catalytic dyad and exhibit strict specificity for the P1 residue of their substrates. However, despite these similarities, clan CD forms a functionally diverse group of enzymes: the overall structural diversity between (and at times within) the various families provides these peptidases with a wide variety of substrate specificities and activation mechanisms. Several members are initially expressed as proenzymes, demonstrating self-inhibition prior to full activation (2). The archetypal and arguably most notable family in the clan is that of the mammalian caspases (C14a), although clan CD members are distributed throughout the entire phylogenetic kingdom and are often required in fundamental biological processes (2). Interestingly, little is known about the structure or function of the C11 proteins, despite their widespread distribution (1) and its archetypal member, clostripain from Clostridium histolyticum, first reported in the literature in 1938 (9). Clostripain has been described as an arginine-specific peptidase with a requirement for Ca (10) and loss of an internal nonapeptide for full activation; lack of structural information on the family appears to have prohibited further investigation. As part of an ongoing project to characterize commensal bacteria in the microbiome that inhabit the human gut, the structure of C11 peptidase, PmC11, from Parabacteroides merdae was determined using the Joint Center for Structural Genomics (JCSG) HTP structural biology pipeline (11). The structure was analyzed, and the enzyme was biochemically characterized to provide the first structure/function correlation for a C11 peptidase. Cloning, expression, purification, crystallization, and structure determination of PmC11 were carried out using standard JCSG protocols (11) as follows. Clones were generated using the polymerase incomplete primer extension (PIPE) cloning method (12). The gene encoding PmC11 (SP5111E) was amplified by polymerase chain reaction (PCR) from P. merdae genomic DNA using PfuTurbo DNA polymerase (Stratagene), using I-PIPE primers that included sequences for the predicted 5′ and 3′ ends (shown below). The expression vector, pSpeedET, which encodes an amino-terminal tobacco etch virus protease-cleavable expression and purification tag (MGSDKIHHHHHHENLYFQ/G), was PCR amplified with V-PIPE (Vector) primers. V-PIPE and I-PIPE PCR products were mixed to anneal the amplified DNA fragments together. Escherichia coli GeneHogs (Invitrogen) competent cells were transformed with the I-PIPE/V-PIPE mixture and dispensed on selective LB-agar plates. The cloning junctions were confirmed by DNA sequencing. The plasmid encoding the full-length protein was deposited in the PSI:Biology Materials Repository at the DNASU plasmid repository (PmCD00547516). For structure determination, to obtain soluble protein using the PIPE, method the gene segment encoding residues Met-Asn was deleted because these residues were predicted to correspond to a signal peptide using SignalP (13). The expression plasmid for the truncated PmC11 construct was transformed into E. coli GeneHogs competent cells and grown in minimal media supplemented with selenomethionine and 30 μg ml of kanamycin at 37 °C using a GNF fermentor (14). A methionine auxotrophic strain was not required as selenomethionine is incorporated via the inhibition of methionine biosynthesis (15, 16). Protein expression was induced using 0.1% (w/v) l-arabinose and the cells were left to grow for a further 3 h at 37 °C. At the end of the cell culture, lysozyme was added to all samples to a final concentration of 250 μg ml and the cells were harvested and stored at −20 °C, until required. Primers used in this section are as follows: I-PIPE (forward): CTGTACTTCCAGGGCGAGACTCCGGAACCCCGGACAACCCGC; I-PIPE (reverse): AATTAAGTCGCGTTATTCATAAACTGCCTTATACCAGCCGAC; V-PIPE (forward): TAACGCGACTTAATTAACTCGTTTAAACGGTCTCCAGC; and V-PIPE (reverse): GCCCTGGAAGTACAGGTTTTCGTGATGATGATGATGAT. Cells were resuspended, homogenized, and lysed by sonication in 40 mm Tris (pH 8.0), 300 mm NaCl, 10 mm imidazole, and 1 mm Tris(2-carboxyethyl)phosphine hydrochloride (TCEP) (Lysis Buffer 1) containing 0.4 mm MgSO4 and 1 μl of 250 unit/μl of benzonase (Sigma). The cell lysate was then clarified by centrifugation (32,500 × g for 25 min at 4 °C) before being passed over Ni-chelating resin equilibrated in Lysis Buffer 1 and washed in the same buffer supplemented with 40 mm imidazole and 10% (v/v) glycerol. The protein was subsequently eluted in 20 mm Tris (pH 8.0), 150 mm NaCl, 10% (v/v) glycerol, 1 mm TCEP, and 300 mm imidazole, and the fractions containing the protein were pooled. To remove the His tag, PmC11 was exchanged into 20 mm Tris (pH 8.0), 150 mm NaCl, 30 mm imidazole, and 1 mm TCEP using a PD-10 column (GE Healthcare), followed by incubation with 1 mg of His-tagged tobacco etch virus protease per 15 mg of protein for 2 h at room temperature and subsequent overnight incubation at 4 °C. The sample was centrifuged to remove any precipitated material (13,000 × g for 10 min at 4 °C) and the supernatant loaded onto Ni-chelating resin equilibrated with 20 mm Tris (pH 8.0), 150 mm NaCl, 30 mm imidazole, and 1 mm TCEP and washed with the same buffer. The flow-through and wash fractions were collected and concentrated to 13.3 mg ml using Amicon Ultra-15 5K centrifugal concentrators (Millipore). PmC11 was crystallized using the nanodroplet vapor diffusion method using standard JCSG crystallization protocols (11). Drops were comprised of 200 nl of protein solution mixed with 200 nl of crystallization solution in 96-well sitting-drop plates, equilibrated against a 50-μl reservoir. Crystals of PmC11 were grown at 4 °C in mother liquor consisting of 0.2 m NH4H2PO4, 20% PEG-3350 (JCSG Core Suite I). Crystals were flash cooled in liquid nitrogen using 10% ethylene glycol as a cryoprotectant prior to data collection and initial screening for diffraction was carried out using the Stanford Automated Mounting system (17) at the Stanford Synchrotron Radiation Lightsource (SSRL, Menlo Park, CA). Single wavelength anomalous dispersion data were collected using a wavelength of 0.9793 Å, at the Advanced Light Source (ALS, beamline 8.2.2, Berkeley, CA) on an ADSC Quantum 315 CCD detector. The data were indexed and integrated with XDS (18) and scaled using XSCALE (18). The diffraction data were indexed in space group P21 with a = 39.11, b = 108.68, c = 77.97 Å, and β = 94.32°. The unit cell contained two molecules in the asymmetric unit resulting in a solvent content of 39% (Matthews' coefficient (Vm) of 2.4 Å Da). The PmC11 structure was determined by the single wavelength anomalous dispersion method using an x-ray wavelength corresponding to the peak of the selenium K edge. Initial phases were derived using the autoSHARP interface (19), which included density modification with SOLOMON (20). Good quality electron density was obtained at 1.7-Å resolution, allowing an initial model to be obtained by automated model building with ARP/wARP (21). Model completion and refinement were iteratively performed with COOT (22) and REFMAC (23, 24) to produce a final model with an Rcryst and Rfree of 14.3 and 17.5%, respectively. The refinement included experimental phase restraints in the form of Hendrickson-Lattman coefficients, TLS refinement with one TLS group per molecule in the asymmetric unit, and NCS restraints. The refined structure contains residues 24–375 and 28–375 for the two molecules in the crystallographic asymmetric unit. Structural validation was carried using the JCSG Quality Control Server that analyzes both the coordinates and data using a variety of structural validation tools to confirm the stereochemical quality of the model (ADIT (25), MOLPROBITY (26), and WHATIF 5.0 (27)) and agreement between model and data (SGCHECK (28) and RESOLVE (29)). All of the main-chain torsion angles were in the allowed regions of the Ramachandran plot and the MolProbity overall clash score for the structure was 2.09 (within the 99th percentile for its resolution). The atomic coordinates and structure factors for PmC11 have been deposited in the Protein Data Bank (PDB) with the accession code 3UWS. Data collection, model, and refinement statistics are reported in Table 1. Crystallographic statistics for PDB code 3UWS Values in parentheses are for the highest resolution shell. Rmerge = ΣhklΣi\|Ii(hkl) − 〈I(hkl)〉\|/Σhkl Σi(hkl). Rmeas = Σhkl[N/(N-1)]Σi\|Ii(hkl) − 〈I(hkl)〉\|/ΣhklΣiIi(hkl). Rpim (precision-indicating Rmerge) = Σhkl[(1/(N-1)] Σi\|Ii (hkl) − 〈I(hkl)〉\|/ΣhklΣi Ii(hkl) (43), where n is the multiplicity of reflection hkl, and Ii(hkl) and 〈I(hkl)〉 are the intensity of the ith measurement and the average intensity of reflection hkl, respectively (44). Rcryst and Rfree = Σ‖Fobs\| − \|Fcalc‖/Σ\|Fobs\| for reflections in the working and test sets, respectively, where Fobs and Fcalc are the observed and calculated structure-factor amplitudes, respectively. Rfree is the same as Rcryst but for 5% of the total reflections chosen at random and omitted from structural refinement. ESU is the estimated standard uncertainties of atoms. The average isotropic B includes TLS and residual B components. RMSD, root-mean-square deviation. The primary sequence alignment with assigned secondary structure was prepared using CLUSTAL OMEGA (30) and ALINE (31). The topology diagram was produced with TOPDRAW (32) and all three-dimensional structural figures were prepared with PyMol (33) with the electrostatic surface potential calculated with APBS (18) and contoured at ±5 kT/e. Architectural comparisons with known structures revealed that PmC11 was most structurally similar to caspase-7, gingipain-K, and legumain (PBD codes 4hq0, 4tkx, and 4aw9, respectively). The statistical significance of the structural alignment between PmC11 and both caspase-7 and gingipain-K is equivalent (Z-score of 9.2) with legumain giving a very similar result (Z-score of 9.1). Of note, the β-strand topology of the CDP domains of Clostridium difficile toxin B (family C80; TcdB; PDB code 3pee) is identical to that observed in the PmC11 β-sheet, but the Z-score from DaliLite was notably less at 7.6. It is possible that the PmC11 structure is more closely related to the C80 family than other families in clan CD, and appear to reside on the same branch of the phylogenetic tree based on structure (2). The PmCD00547516 plasmid described above was obtained from the PSI:Biology Materials Repository and used to generate a cleavage site mutant PmC11 and an active-site mutant PmC11 using the QuikChange Site-directed Mutagenesis kit (Stratagene) as per the manufacturer's instructions using the following primers: K147A mutant (forward): CAGAATAAGCTGGCAGCGTTCGGACAGGACG, and K147A mutant (reverse): CGTCCTGTCCGAACGCTGCCAGCTTATTCTG; C179A mutant (forward): CCTGTTCGATGCCGCCTACATGGCAAGC, and C179A mutant (reverse): GCTTGCCATGTAGGCGGCATCGAACAGG. The expression plasmids containing PmC11 were transformed into E. coli BL21 Star (DE3) and grown in Luria-Bertani media containing 30 μg ml of kanamycin at 37 °C until an optical density (600 nm) of ∼0.6 was reached. l-Arabinose was added to a final concentration of 0.2% (w/v) and the cells incubated overnight at 25 °C. Compared with the protein production for crystallography, a slightly modified purification protocol was employed for biochemical assays. Initially, the cells were resuspended in 20 mm sodium phosphate (pH 7.5), 150 mm NaCl (Lysis Buffer 2) containing an EDTA-free protease inhibitor mixture (cOmplete, Roche Applied Science). Cells were disrupted by three passages (15 KPSI) through a One-Shot cell disruptor (Constant Systems) followed by centrifugation at 20,000 × g for 20 min at 4 °C. The supernatant was collected and sterile-filtered (0.2 μm) before being applied to a 5-ml HisTrap HP column (GE Healthcare) equilibrated in Lysis Buffer 2 containing 25 mm imidazole, and the protein was eluted in the same buffer containing 250 mm imidazole. The peak fractions were pooled and buffer exchanged into the assay buffer (20 mm Tris, 150 mm NaCl, pH 8.0) using a PD-10 column. When required, purified PmC11 was concentrated using Vivaspin 2 30-K centrifugal concentrators (Sartorius). Protein concentration was routinely measured using Bradford's reagent (Bio-Rad) with a BSA standard. The release of the fluorescent group AMC (7-amino-4-methylcoumarin) from potential peptide substrates was used to assess the activity of PmC11. Peptidase activity was tested using 20 μg of PmC11 and 100 μm substrate (unless otherwise stated) in assay buffer to a final reaction volume of 200 μl and all samples were incubated (without substrate) at 37 °C for 16 h prior to carrying out the assay. The substrate and plate reader were brought to 37 °C for 20 min prior to the addition of the PmC11 and samples prepared without PmC11 were used as blanks (negative controls). The curves were plotted using the blank-corrected fluorescence units against the time of acquisition (in min). The assays were carried out in black 96-well flat-bottomed plates (Greiner). AMC fluorescence was measured using a PHERAstar FS plate reader (BMG Labtech) with excitation and emission wavelengths of 355 and 460 nm, respectively. To investigate the substrate specificity of PmC11, substrates Z-GGR-AMC, Bz-R-AMC, Z-GP-AMC, Z-HGP-AMC, Ac-DEVD-AMC (all Bachem), BOC-VLK-AMC, and BOC-K-AMC (both PeptaNova) were prepared at 100 mm in 100% dimethyl sulfoxide. The amount of AMC (micromoles) released was calculated by generating an AMC standard curve (as described in Ref. 34) and the specific activity of PmC11 was calculated as picomoles of AMC released per min per mg of the protein preparation. The reaction rates (Vmax) and Km values were determined for mutants PmC11 and PmC11 by carrying out the activity assay at varying concentrations of Bz-R-AMC between 0 and 600 μm. The blank-corrected relative fluorescence units were plotted against time (min) with ΔFU/T giving the reaction rate. The Km and Vmax of PmC11 and PmC11 against an R-AMC substrate were determined from the Lineweaver-Burk plot as described (34), calculated using GraphPad Prism6 software. All experiments were carried out in triplicate. To test the effect of the inhibitor on the activity of PmC11, 25 μm Z-VRPR-FMK (100 mm stock in 100% dimethyl sulfoxide, Enzo Life Sciences), 20 μg of PmC11, 100 μm R-AMC substrate, 1 mm EGTA were prepared in the assay buffer and the activity assay carried out as described above. A gel-shift assay, to observe Z-VRPR-FMK binding to PmC11, was also set up using 20 μg of PmC11, 25 μm inhibitor, 1 mm EGTA in assay buffer. The reactions were incubated at 37 °C for 20 min before being stopped by the addition SDS-PAGE sample buffer. Samples were analyzed by loading the reaction mixture on a 10% NuPAGE BisTris gel using MES buffer. The enzyme activity of PmC11 was tested in the presence of various divalent cations: Mg, Ca, Mn, Co, Fe, Zn, and Cu. The final concentration of the salts (MgSO4, CaCl2, MnCl2, CoCl2, FeSO4, ZnCl2, and CuSO4) was 1 mm and the control was set up without divalent ions but with addition of 1 mm EGTA. The assay was set up using 20 mg of PmC11, 1 mm salts, 100 μm R-AMC substrate, and the assay buffer, and incubated at 37 °C for 16 h. The activity assay was carried out as described above. Affinity-purified PmC11 was loaded onto a HiLoad 16/60 Superdex 200 gel filtration column (GE Healthcare) equilibrated in the assay buffer. The apparent molecular weight of PmC11 was determined from calibration curves based on protein standards of known molecular weights. Autoprocessing of PmC11 was evaluated by incubating the enzyme at 37 °C and removing samples at 1-h intervals from 0 to 16 h and placing into SDS-PAGE loading buffer to stop the processing. Samples were then analyzed on a 4–12% NuPAGE (Thermo Fisher) Novex BisTris gel run in MES buffer. To investigate whether processing is a result of intra- or inter-molecular cleavage, the PmC11 mutant was incubated with increasing concentrations of activated PmC11 (0, 0.1, 0.2, 0.5, 1, 2, and 5 μg). The final assay volume was 40 μl and the proteins were incubated at 37 °C for 16 h in the PmC11 assay buffer. To stop the reaction, NuPAGE sample buffer was added to the protein samples and 20 μl was analyzed on 10% NuPAGE Novex BisTris gel using MES buffer. These studies revealed no apparent cleavage of PmC11 by the active enzyme at low concentrations of PmC11 and that only limited cleavage was observed when the ratio of active enzyme (PmC11: PmC11) was increased to ∼1:10 and 1:4. The crystal structure of the catalytically active form of PmC11 revealed an extended caspase-like α/β/α sandwich architecture comprised of a central nine-stranded β-sheet, with an unusual C-terminal domain (CTD), starting at Lys. A single cleavage was observed in the polypeptide chain at Lys (Fig. 1, A and B), where both ends of the cleavage site are fully visible and well ordered in the electron density. The central nine-stranded β-sheet (β1–β9) of PmC11 consists of six parallel and three anti-parallel β-strands with 4↑3↓2↑1↑5↑6↑7↓8↓9↑ topology (Fig. 1A) and the overall structure includes 14 α-helices with six (α1–α2 and α4–α7) closely surrounding the β-sheet in an approximately parallel orientation. Helices α1, α7, and α6 are located on one side of the β-sheet with α2, α4, and α5 on the opposite side (Fig. 1A). Helix α3 sits at the end of the loop following β5 (L5), just preceding the Lys cleavage site, with both L5 and α3 pointing away from the central β-sheet and toward the CTD, which starts with α8. The structure also includes two short β-hairpins (βA–βB and βD–βE) and a small β-sheet (βC–βF), which is formed from two distinct regions of the sequence (βC precedes α11, α12 and β9, whereas βF follows the βD-βE hairpin) in the middle of the CTD (Fig. 1B). Crystal structure of a C11 peptidase from P. merdae. A, primary sequence alignment of PmC11 (Uniprot ID A7A9N3) and clostripain (Uniprot ID P09870) from C. histolyticum with identical residues highlighted in gray shading. The secondary structure of PmC11 from the crystal structure is mapped onto its sequence with the position of the PmC11 catalytic dyad, autocatalytic cleavage site (Lys), and S1 binding pocket Asp (Asp) highlighted by a red star, a red downturned triangle, and a red upturned triangle, respectively. Connecting loops are colored gray, the main β-sheet is in orange, with other strands in olive, α-helices are in blue, and the nonapeptide linker of clostripain that is excised upon autocleavage is underlined in red. Sequences around the catalytic site of clostripain and PmC11 align well. B, topology diagram of PmC11 colored as in A except that additional (non-core) β-strands are in yellow. Helices found on either side of the central β-sheet are shown above and below the sheet, respectively. The position of the catalytic dyad (H, C) and the processing site (Lys) are highlighted. Helices (1–14) and β-strands (1–9 and A-F) are numbered from the N terminus. The core caspase-fold is highlighted in a box. C, tertiary structure of PmC11. The N and C termini (N and C) of PmC11 along with the central β-sheet (1–9), helix α5, and helices α8, α11, and α13 from the C-terminal domain, are all labeled. Loops are colored gray, the main β-sheet is in orange, with other β-strands in yellow, and α-helices are in blue. The CTD of PmC11 is composed of a tight helical bundle formed from helices α8–α14 and includes strands βC and βF, and β-hairpin βD–βE. The CTD sits entirely on one side of the enzyme interacting only with α3, α5, β9, and the loops surrounding β8. Of the interacting secondary structure elements, α5 is perhaps the most interesting. This helix makes a total of eight hydrogen bonds with the CTD, including one salt bridge (Arg-Asp) and is surrounded by the CTD on one side and the main core of the enzyme on the other, acting like a linchpin holding both components together (Fig. 1C). PmC11 is, as expected, most structurally similar to other members of clan CD with the top hits in a search of known structures being caspase-7, gingipain-K, and legumain (PBD codes 4hq0, 4tkx, and 4aw9, respectively) (Table 2). The C-terminal domain is unique to PmC11 within clan CD and structure comparisons for this domain alone does not produce any hits in the PDB (DaliLite, PDBeFold), suggesting a completely novel fold. As the archetypal and arguably most well studied member of clan CD, the caspases were used as the basis to investigate the structure/function relationships in PmC11, with caspase-7 as the representative member. Six of the central β-strands in PmC11 (β1–β2 and β5–β8) share the same topology as the six-stranded β-sheet found in caspases, with strands β3, β4, and β9 located on the outside of this core structure (2) (Fig. 1B, box). His and Cys were found at locations structurally homologous to the caspase catalytic dyad, and other clan CD structures (2), at the C termini of strands β5 and β6, respectively (Figs. 1, A and B, and 2A). A multiple sequence alignment of C11 proteins revealed that these residues are highly conserved (data not shown). Summary of PDBeFOLD (45) superposition of structures found to be most similar to PmC11 in the PBD based on DaliLite (46) The results are ordered in terms of structural homology (QH), where %SSE is the percentage of the SSEs in the PmC11 that can be identified in the target X (where X = caspase-7 (47), legumain (3), gingipain (48), and TcdB-CPD (49); % SSE is the percentage of SSEs in X (as above) that can be identified in PmC11 (as above); % sequence ID is the percentage sequence identity after structural alignment; Nalign is the number of matched residues; and r.m.s. deviation the root mean squared deviation on the Cα positions of the matched residues. Biochemical and structural characterization of PmC11. A, ribbon representation of the overall structure of PmC11 illustrating the catalytic site, cleavage site displacement, and potential S1 binding site. The overall structure of PmC11 is shown in gray, looking down into the catalytic site with the catalytic dyad in red. The two ends of the autolytic cleavage site (Lys and Ala, green) are displaced by 19.5 Å (thin black line) from one another and residues in the potential substrate binding pocket are highlighted in blue. B, size exclusion chromatography of PmC11. PmC11 migrates as a monomer with a molecular mass around 41 kDa calculated from protein standards of known molecular weights. Elution fractions across the major peak (1–6) were analyzed by SDS-PAGE on a 4–12% gel in MES buffer. C, the active form of PmC11 and two mutants, PmC11 (C) and PmC11 (K), were examined by SDS-PAGE (lane 1) and Western blot analysis using an anti-His antibody (lane 2) show that PmC11 autoprocesses, whereas mutants, PmC11 and PmC11, do not show autoprocessing in vitro. D, cysteine peptidase activity of PmC11. Km and Vmax of PmC11 and K147A mutant were determined by monitoring change in the fluorescence corresponding to AMC release from Bz-R-AMC. Reactions were performed in triplicate and representative values ± S.D. are shown. E, intermolecular processing of PmC11 by PmC11. PmC11 (20 μg) was incubated overnight at 37 °C with increasing amounts of processed PmC11 and analyzed on a 10% SDS-PAGE gel. Inactive PmC11 was not processed to a major extent by active PmC11 until around a ratio of 1:4 (5 μg of active PmC11). A single lane of 20 μg of active PmC11 (labeled 20) is shown for comparison. F, activity of PmC11 against basic substrates. Specific activity is shown ± S.D. from three independent experiments. G, electrostatic surface potential of PmC11 shown in a similar orientation, where blue and red denote positively and negatively charged surface potential, respectively, contoured at ±5 kT/e. The position of the catalytic dyad, one potential key substrate binding residue Asp, and the ends of the cleavage site Lys and Ala are indicated. Five of the α-helices surrounding the β-sheet of PmC11 (α1, α2, α4, α6, and α7) are found in similar positions to the five structurally conserved helices in caspases and other members of clan CD, apart from family C80 (2). Other than its more extended β-sheet, PmC11 differs most significantly from other clan CD members at its C terminus, where the CTD contains a further seven α-helices and four β-strands after β8. Purification of recombinant PmC11 (molecular mass = 42.6 kDa) revealed partial processing into two cleavage products of 26.4 and 16.2 kDa, related to the observed cleavage at Lys in the crystal structure (Fig. 2A). Incubation of PmC11 at 37 °C for 16 h, resulted in a fully processed enzyme that remained as an intact monomer when applied to a size-exclusion column (Fig. 2B). The single cleavage site of PmC11 at Lys is found immediately after α3, in loop L5 within the central β-sheet (Figs. 1, A and B, and 2A). The two ends of the cleavage site are remarkably well ordered in the crystal structure and displaced from one another by 19.5 Å (Fig. 2A). Moreover, the C-terminal side of the cleavage site resides near the catalytic dyad with Ala being 4.5 and 5.7 Å from His and Cys, respectively. Consequently, it appears feasible that the helix attached to Lys (α3) could be responsible for steric autoinhibition of PmC11 when Lys is covalently bonded to Ala. Thus, the cleavage would be required for full activation of PmC11. To investigate this possibility, two mutant forms of the enzyme were created: PmC11 (a catalytically inactive mutant) and PmC11 (a cleavage-site mutant). Initial SDS-PAGE and Western blot analysis of both mutants revealed no discernible processing occurred as compared with active PmC11 (Fig. 2C). The PmC11 mutant enzyme had a markedly different reaction rate (Vmax) compared with WT, where the reaction velocity of PmC11 was 10 times greater than that of PmC11 (Fig. 2D). Taken together, these data reveal that PmC11 requires processing at Lys for optimum activity. To investigate whether processing is a result of intra- or intermolecular cleavage, the PmC11 mutant was incubated with increasing concentrations of processed and activated PmC11. These studies revealed that there was no apparent cleavage of PmC11 by the active enzyme at low concentrations of PmC11 and that only limited cleavage was observed when the ratio of active enzyme (PmC11:PmC11) was increased to ∼1:10 and 1:4, with complete cleavage observed at a ratio of 1:1 (Fig. 2E). This suggests that cleavage of PmC11 was most likely an effect of the increasing concentration of PmC11 and intermolecular cleavage. Collectively, these data suggest that the pro-form of PmC11 is autoinhibited by a section of L5 blocking access to the active site, prior to intramolecular cleavage at Lys. This cleavage subsequently allows movement of the region containing Lys and the active site to open up for substrate access. The autocatalytic cleavage of PmC11 at Lys (sequence KLK∧A) demonstrates that the enzyme accepts substrates with Lys in the P1 position. The substrate specificity of the enzyme was further tested using a variety of fluorogenic substrates. As expected, PmC11 showed no activity against substrates with Pro or Asp in P1 but was active toward substrates with a basic residue in P1 such as Bz-R-AMC, Z-GGR-AMC, and BOC-VLK-AMC. The rate of cleavage was ∼3-fold greater toward the single Arg substrate Bz-R-AMC than for the other two (Fig. 2F) and, unexpectedly, PmC11 showed no activity toward BOC-K-AMC. These results confirm that PmC11 accepts substrates containing Arg or Lys in P1 with a possible preference for Arg. The catalytic dyad of PmC11 sits near the bottom of an open pocket on the surface of the enzyme at a conserved location in the clan CD family (2). The PmC11 structure reveals that the catalytic dyad forms part of a large acidic pocket (Fig. 2G), consistent with a binding site for a basic substrate. This pocket is lined with the potential functional side chains of Asn, Asp, and Thr with Gly, Asp, and Met also contributing to the pocket (Fig. 2A). Interestingly, these residues are in regions that are structurally similar to those involved in the S1 binding pockets of other clan CD members (shown in Ref. 2). Because PmC11 recognizes basic substrates, the tetrapeptide inhibitor Z-VRPR-FMK was tested as an enzyme inhibitor and was found to inhibit both the autoprocessing and activity of PmC11 (Fig. 3A). Z-VRPR-FMK was also shown to bind to the enzyme: a size-shift was observed, by SDS-PAGE analysis, in the larger processed product of PmC11 suggesting that the inhibitor bound to the active site (Fig. 3B). A structure overlay of PmC11 with the MALT1-paracacaspase (MALT1-P), in complex with Z-VRPR-FMK (35), revealed that the PmC11 dyad sits in a very similar position to that of active MALT1-P and that Asn, Asp, and Asp superimpose well with the principal MALT1-P inhibitor binding residues (Asp, Asp, and Glu, respectively (VRPR-FMK from MALT1-P with the corresponding PmC11 residues from the structural overlay is shown in Fig. 1D), as described in Ref. 5). Asp is located near the catalytic cysteine and is conserved throughout the C11 family, suggesting it is the primary S1 binding site residue. In the structure of PmC11, Asp resides on a flexible loop pointing away from the S1 binding pocket (Fig. 3C). However, this loop has been shown to be important for substrate binding in clan CD (2) and this residue could easily rotate and be involved in substrate binding in PmC11. Thus, Asn, Asp, and Asp are most likely responsible for the substrate specificity of PmC11. Asp is highly conserved throughout the clan CD C11 peptidases and is thought to be primarily responsible for substrate specificity of the clan CD enzymes, as also illustrated from the proximity of these residues relative to the inhibitor Z-VRPR-FMK when PmC11 is overlaid on the MALT1-P structure (Fig. 3C). PmC11 binds and is inhibited by Z-VRPR-FMK and does not require Ca for activity. A, PmC11 activity is inhibited by Z-VRPR-FMK. Cleavage of Bz-R-AMC by PmC11 was measured in a fluorometric activity assay with (+, purple) and without (−, red) Z-VRPR-FMK. The relative fluorescence units of AMC released are plotted against time (min) (n = 3; ±S.D.). B, gel-shift assay reveals that Z-VRPR-FMK binds to PmC11. PmC11 was incubated with (+) or without (−) Z-VRPR-FMK and the samples analyzed on a 10% SDS-PAGE gel. A size shift can be observed in the larger processed product of PmC11 (26.1 kDa). C, PmC11 with the Z-VRPR-FMK from the MALT1-paracacaspase (MALT1-P) superimposed. A three-dimensional structural overlay of Z-VRPR-FMK from the MALT1-P complex onto PmC11. The position and orientation of Z-VRPR-FMK was taken from superposition of the PmC11 and MALTI_P structures and indicates the presumed active site of PmC11. Residues surrounding the inhibitor are labeled and represent potentially important binding site residues, labeled in black and shown in an atomic representation. Carbon atoms are shown in gray, nitrogen in blue, and oxygen in red. C, divalent cations do not increase the activity of PmC11. The cleavage of Bz-R-AMC by PmC11 was measured in the presence of the cations Ca, Mn, Zn, Co, Cu, Mg, and Fe with EGTA as a negative control, and relative fluorescence measured against time (min). The addition of cations produced no improvement in activity of PmC11 when compared in the presence of EGTA, suggesting that PmC11 does not require metal ions for proteolytic activity. Furthermore, Cu, Fe, and Zn appear to inhibit PmC11. Clostripain from C. histolyticum is the founding member of the C11 family of peptidases and contains an additional 149 residues compared with PmC11. A multiple sequence alignment revealed that most of the secondary structural elements are conserved between the two enzymes, although they are only ∼23% identical (Fig. 1A). Nevertheless, PmC11 may be a good model for the core structure of clostripain. The primary structural alignment also shows that the catalytic dyad in PmC11 is structurally conserved in clostripain (36) (Fig. 1A). Unlike PmC11, clostripain has two cleavage sites (Arg and Arg), which results in the removal of a nonapeptide, and is required for full activation of the enzyme (37) (highlighted in Fig. 1A). Interestingly, Arg was found to align with Lys in PmC11. In addition, the predicted primary S1-binding residue in PmC11 Asp also overlays with the residue predicted to be the P1 specificity determining residue in clostripain (38) (Asp, Fig. 1A). As studies on clostripain revealed addition of Ca ions are required for full activation, the Ca dependence of PmC11 was examined. Surprisingly, Ca did not enhance PmC11 activity and, furthermore, other divalent cations, Mg, Mn, Co, Fe, Zn, and Cu, were not necessary for PmC11 activity (Fig. 3D). In support of these findings, EGTA did not inhibit PmC11 suggesting that, unlike clostripain, PmC11 does not require Ca or other divalent cations, for activity. The crystal structure of PmC11 now provides three-dimensional information for a member of the clostripain C11 family of cysteine peptidases. The enzyme exhibits all of the key structural elements of clan CD members, but is unusual in that it has a nine-stranded central β-sheet with a novel C-terminal domain. The structural similarity of PmC11 with its nearest structural neighbors in the PDB is decidedly low, overlaying better with six-stranded caspase-7 than any of the other larger members of the clan (Table 2). The substrate specificity of PmC11 is Arg/Lys and the crystal structure revealed an acidic pocket for specific binding of such basic substrates. In addition, the structure suggested a mechanism of self-inhibition in both PmC11 and clostripain and an activation mechanism that requires autoprocessing. PmC11 differs from clostripain in that is does not appear to require divalent cations for activation. Several other members of clan CD require processing for full activation including legumain (39), gingipain-R (40), MARTX-CPD (8), and the effector caspases, e.g. caspase-7 (41). To date, the effector caspases are the only group of enzymes that require cleavage of a loop within the central β-sheet. This is also the case in PmC11, although the cleavage loop is structurally different to that found in the caspases and follows the catalytic His (Fig. 1A), as opposed to the Cys in the caspases. All other clan CD members requiring cleavage for full activation do so at sites external to their central sheets (2). The caspases and gingipain-R both undergo intermolecular (trans) cleavage and legumain and MARTX-CPD are reported to perform intramolecular (cis) cleavage. In addition, several members of clan CD exhibit self-inhibition, whereby regions of the enzyme block access to the active site (2). Like PmC11, these structures show preformed catalytic machinery and, for a substrate to gain access, movement and/or cleavage of the blocking region is required. The structure of PmC11 gives the first insight into this class of relatively unexplored family of proteins and should allow important catalytic and substrate binding residues to be identified in a variety of orthologues. Indeed, insights gained from an analysis of the PmC11 structure revealed the identity of the Trypanosoma brucei PNT1 protein as a C11 cysteine peptidase with an essential role in organelle replication (42). The PmC11 structure should provide a good basis for structural modeling and, given the importance of other clan CD enzymes, this work should also advance the exploration of these peptidases and potentially identify new biologically important substrates.
PMC4918759	Structures of human ADAR2 bound to dsRNA reveal base-flipping mechanism and basis for site selectivity	ADARs (adenosine deaminases acting on RNA) are editing enzymes that convert adenosine (A) to inosine (I) in duplex RNA, a modification reaction with wide-ranging consequences on RNA function. Our understanding of the ADAR reaction mechanism, origin of editing site selectivity and effect of mutations is limited by the lack of high-resolution structural data for complexes of ADARs bound to substrate RNAs. Here we describe four crystal structures of the deaminase domain of human ADAR2 bound to RNA duplexes bearing a mimic of the deamination reaction intermediate. These structures, together with structure-guided mutagenesis and RNA-modification experiments, explain the basis for ADAR deaminase domain’s dsRNA specificity, its base-flipping mechanism, and nearest neighbor preferences. In addition, an ADAR2-specific RNA-binding loop was identified near the enzyme active site rationalizing differences in selectivity observed between different ADARs. Finally, our results provide a structural framework for understanding the effects of ADAR mutations associated with human disease.RNA editing reactions alter a transcript’s genomically encoded sequence by inserting, deleting or modifying nucleotides. Deamination of adenosine (A), the most common form of RNA editing in humans, generates inosine (I) at the corresponding nucleotide position. Since I base pairs with cytidine (C), it functions like guanosine (G) in cellular processes such as splicing, translation and reverse transcription. A to I editing has wide-ranging consequences on RNA function including altering miRNA recognition sites, redirecting splicing and changing the meaning of specific codons. Two different enzymes carry out A to I editing in humans; ADAR1 and ADAR2. ADAR activity is required for nervous system function and altered editing has been linked to neurological disorders such as epilepsy and Prader Willi Syndrome. In addition, mutations in the ADAR1 gene are known to cause the autoimmune disease Aicardi-Goutieres Syndrome (AGS) and the skin disorder Dyschromatosis Symmetrica Hereditaria (DSH). Hyper editing has been observed at certain sites in cancer cells, such as in the mRNA for AZIN1 (antizyme inhibitor 1). However, hypo editing also occurs in cancer-derived cell lines exemplified by reduced editing observed in the message for glioma-associated oncogene 1 (Gli1). The ADAR proteins have a modular structure with double stranded RNA binding domains (dsRBDs) and a C-terminal deaminase domain (see Fig. 1a for hADAR2 domains). ADARs efficiently deaminate specific adenosines in duplex RNA while leaving most adenosines unmodified. The mechanism of adenosine deamination requires ADAR to flip the reactive base out of an RNA double helix to access its active site. How an enzyme could accomplish this task with a duplex RNA substrate is not known. Furthermore, how an ADAR deaminase domain contributes to editing site selectivity is also unknown, since no structures of ADAR deaminase domain-RNA complexes have been reported. To address these knowledge gaps, we set out to trap the human ADAR2 deaminase domain (aa299–701, hADAR2d) bound to different duplex RNAs and solve structures for the resulting complexes using x-ray crystallography. We then evaluated the importance of protein-RNA contacts using structure-guided mutagenesis and RNA-modification experiments coupled with adenosine deamination kinetics. The ADAR reaction involves the formation of a hydrated intermediate that loses ammonia to generate the inosine-containing product RNA (for reaction scheme see Fig. 1b). The covalent hydrate of the nucleoside analog 8-azanebularine (N) mimics the proposed high-energy intermediate (for reaction scheme see Fig. 1b). For trapping hADAR2d bound to RNA for crystallography, we incorporated 8-azanebularine into duplex RNAs shown recently to be excellent substrates for deamination by hADAR2d (for duplex sequence see Fig. 1c) (for characterization of protein–RNA complex see Supplementary Fig. 1). In addition, for one of these duplexes (Bdf2), we positioned the 8-azanebularine opposite either uridine or cytidine to mimic an A-U pair or A-C mismatch at the editing site creating a total of three different RNA substrates for structural studies (Fig. 1c). The hADAR2d protein (without RNA bound) has been previously crystallized and structurally characterized revealing features of the active site including the presence of zinc. In addition, an inositol hexakisphosphate (IHP) molecule was found buried in the core of the protein hydrogen bonded to numerous conserved polar residues. For crystallization of hADAR2d-RNA complexes, we used both the wild type (WT) deaminase domain and a mutant (E488Q) that has enhanced catalytic activity. A description of the crystallization conditions, X-ray diffraction data collection and solution of the structures can be found in Online Methods. Four protein-RNA combinations generated diffracting crystals that resulted in high-resolution structures (hADAR2d WT–Bdf2-U, hADAR2d WT–Bdf2-C, hADAR2d E488Q–Bdf2-C, hADAR2d E488Q–Gli1) (Table 1). In each of these complexes, the protein binds the RNA on one face of the duplex over ~ 20 bp using a positively charged surface near the zinc-containing active site (Fig. 2, Supplementary Fig. 2a). The large binding site (1493 Å RNA surface area and 1277 Å protein surface area buried) observed for hADAR2d is consistent with recent footprinting studies. Both strands of the RNA contact the protein with the majority of these interactions mediated through the phosphodiester-ribose backbone near the editing site (Fig. 2c, Supplementary Fig. 2 b–d). The structures show a large deviation from A-form RNA conformation at the editing site (Fig. 2, Fig. 3, Supplementary Video 1). The 8-azanebularine is flipped out of the helix and bound into the active site as its covalent hydrate where it interacts with several amino acids including V351, T375, K376, E396 and R455 (Fig. 3a, Supplementary Fig. 3a). The side chain of E396 H-bonds to purine N1 and O6. This interaction was expected given the proposed role of E396 in mediating proton transfers to and from N1 of the substrate adenosine. The 2’-hydroxyl of 8-azanebularine H-bonds to the backbone carbonyl of T375 while the T375 side chain contacts its 3’-phosphodiester. R455 and K376 help position the flipped nucleotide in the active site by fastening the phosphate backbone flanking the editing site. The R455 side chain ion pairs with the 5’-phosphodiester of 8-azanebularine while the K376 side chain contacts its 3’-phosphodiester. Lastly, the side chain of V351 provides a hydrophobic surface for interaction with the nucleobase of the edited nucleotide. RNA binding does not alter IHP binding or the H-bonding network linking IHP to the active site. The ADAR2 base-flipping loop, bearing residue 488, approaches the RNA duplex from the minor groove side at the editing site. The side chain of this amino acid penetrates the helix where it occupies the space vacated by the flipped out base and H-bonds to the complementary strand orphaned base and to the 2’ hydroxyl of the nucleotide immediately 5’ to the editing site (Figs. 3b, 3c). In the four structures reported here, three different combinations of helix-penetrating residue and orphan base are observed (i.e. E488 + U, E488 + C and Q488 + C) and all three combinations show the same side chain and base positions (Figs. 3b, 3c, Supplementary Fig. 4a for overlay of all three). For instance, in the complex with hADAR2d E488Q and the Bdf2-C duplex, the protein recognizes an orphaned C by donating H-bonds from Nε2 to cytosine N3 and from its backbone NH to cytosine O2 (Fig. 3b). In the complex with hADAR2d WT and the Bdf2-U duplex, a similar interaction is observed with the E488 backbone NH hydrogen bonded to the uracil O2 and the E488 side chain H-bonded to the uracil N3H (Fig. 3c). Interestingly, the E488Q mutant was discovered in a screen for highly active ADAR2 mutants and this residue was suggested to be involved in base flipping given its effect on editing substrates with a fluorescent nucleobase at the editing site. ADARs react preferentially with adenosines in A•C mismatches and A-U pairs over A•A and A•G mismatches. A purine at the orphan base position (in its anti conformation) would clash with the 488 residue explaining the preference for pyrimidines here. The interaction of the 488 residue with the orphaned base is reminiscent of an interaction observed for Hha I DNA methyltransfersase (MTase), a duplex DNA modifying enzyme that also uses a base flipping mechanism to access 2’-deoxycytidine (dC) for methylation. For that enzyme, Q237 H-bonds to an orphaned dG while it fills the void left by the flipped out dC (Supplementary Fig. 4b). In addition, two glycine residues flank Q237 allowing the loop to adopt the conformation necessary for penetration into the helix. The flipping loop in ADAR2 (i.e. aa487–489) also has the helix-penetrating residue flanked by glycines. However, unlike the case of the DNA MTase that approaches the DNA from the major groove, the ADAR2 loop approaches the duplex from the minor groove side. Such an approach requires deeper penetration of the intercalating residue to access the H-bonding sites on the orphaned base, necessitating an additional conformational change in the RNA duplex. This change includes shifting of the base pairs immediately 5’ to the editing site toward the helical axis and a widening of the major groove opposite the editing site (Figs. 4a, 4b, Supplementary Video 1). In the case of the hADAR2d WT–Bdf2-U RNA, this shift is accompanied by a shearing of the U11-A13' base pair with U11 shifted further in the direction of the major groove creating an unusual U-A "wobble" interaction with adenine N6 and N1 within H-bonding distance to uracil N3H and O2, respectively (Fig. 4c, Supplementary Fig. 3b). This type of wobble pair has been observed before and requires either the imino tautomer of adenine or the enol tautomer of uracil. The ADAR-induced distortion in RNA conformation results in a kink in the RNA strand opposite the editing site (Fig. 4b). This kink is stabilized by interactions of the side chains of R510 and S495 with phosphodiesters in the RNA backbone of the unedited strand (Fig. 4a). Interestingly, ADAR2’s flipping loop approach from the minor groove side is like that seen with certain DNA repair glycosylases (e.g. UDG, HOGG1, and AAG) that also project intercalating residues from loops bound in the minor groove (Supplementary Fig. 5a). However, these enzymes typically bend the DNA duplex at the site of modification to allow for penetration of intercalating residues and damage recognition. While hADAR2d clearly alters the duplex conformation to gain access to the modification site from the minor groove, it does not bend the RNA duplex (Figs. 2a, 2b, 4b). Furthermore, ADARs do not modify duplex DNA. The DNA B-form helix has groove widths and depths that would prevent productive interactions with ADAR. For instance, ADAR can readily penetrate an A-form helix from the minor groove side and place the helix-penetrating residue in the space occupied by the editing site base (Supplementary Fig. 6). However, this residue cannot penetrate the minor groove enough to occupy the base position in a B-form helix (Supplementary Fig. 6). Furthermore, DNA lacks the 2’ hydroxyls that are used by ADAR for substrate recognition (Fig. 2c). Indeed, in each of the four complexes reported here, the protein contacts at least five ribose 2’ hydroxyl groups (Fig. 2c, Supplementary Fig. 2 b–d). Thus, hADAR2d uses a substrate recognition and base flipping mechanism with similarities to other known nucleic acid-modifying enzymes but uniquely suited for reaction with adenosine in the context of duplex RNA. ADARs have a preference for editing adenosines with 5’ nearest neighbor U (or A) and 3’ nearest neighbor G. The ADAR2 flipping loop occupies the minor groove spanning the three base pairs that include the nearest neighbor nucleotides flanking the edited base (Figs. 3b, 3c). As described above, the base pair including the 5’ nearest neighbor U (U11-A13’ in the Bdf2 duplex) is shifted from the position it would occupy in a typical A-form helix to accommodate the loop (Fig. 4a). Also, the minor groove edge of this pair is juxtaposed to the protein backbone at G489. Modeling a G-C or C-G pair at this position (i.e. 5’ G or 5’ C) suggests a 2-amino group in the minor groove would clash with the protein at G489 (Fig. 5a, Supplementary Fig. 7c). Indeed, replacing the U-A pair adjacent to the editing site with a C-G pair in the Gli1 duplex substrate resulted in an 80% reduction in the rate of hADAR2d-catalyzed deamination (Figs. 5b, 5c). To determine whether this effect arises from an increase in local duplex stability from the C-G for U-A substitution or from the presence of the 2-amino group, we replaced the U-A pair with a U-2-aminopurine (2AP) pair. 2AP is an adenosine analog that forms a base pair with uridine of similar stability to a U-A pair, but places an amino group in the minor groove (Fig. 5b). Importantly, this substitution also resulted in an 80% reduction in rate, illustrating the detrimental effect of the amino group in the minor groove at this location. These observations suggest that hADAR2’s 5’ nearest neighbor preference for U (or A) is due to a destabilizing clash with the protein backbone at G489 that results from the presence of an amino group in the minor groove at this location for sequences with 5’ nearest neighbor G or C. However, the observed clash is not severe and the enzyme would be able to accommodate G or C 5’ nearest neighbors by slight structural perturbations, explaining why this sequence preference is not an absolute requirement. In each of the hADAR2d-RNA structures reported here, the backbone carbonyl oxygen at S486 accepts an H-bond from the 2-amino group of the G on the 3’ side of the edited nucleotide (Fig. 5d). Guanine is the only common nucleobase that presents an H-bond donor in the RNA minor groove suggesting that other nucleotides in this position would reduce editing efficiency. Indeed, mutating this base to A, C or U, while maintaining base pairing at this position, reduced the rate of deamination by hADAR2d in Gli1 mRNA model substrates (Supplementary Fig. 7 a–b). To test the importance of the amino group on the 3’ G in the hADAR2d reaction, we prepared RNA duplex substrates with purine analogs on the 3’ side of the edited A (Fig. 5e). We tested a G analog that lacks the 2-amino group (inosine, I) and one that blocks access to this amino group (N-methylguanosine (NMeG). In addition, we compared a 3’ A to a 3’ 2AP since 2AP could form the H-bonding interaction observed with S486. We found the substrate with a 3’ NMeG to be unreactive to hADAR2d-catalyzed deamination confirming the importance of the observed close approach by the protein to the 3’ G 2-amino group (Fig. 5f). In addition, the substrate with a 3’ I displayed a reduced deamination rate compared to the substrate with a 3’ G suggesting the observed H-bond to the 2-amino group contributes to the 3’ nearest neighbor selectivity (Fig. 5f). This conclusion is further supported by the observation that deamination in the substrate with a 3’ 2AP is faster than in the substrate with a 3’ A (Fig. 5f). The structures reported here identify RNA-binding loops of the ADAR catalytic domain and suggest roles for several amino acids not previously known to be important for editing, either substrate binding or catalysis (Fig. 6). The side chain for R510 ion-pairs with the 3’ phosphodiester of the orphaned nucleotide (Figs. 3a, 3c). This residue is conserved in ADAR2s and ADAR1s, but is glutamine in the editing-inactive ADAR3s (Supplementary Table 1). Mutation of hADAR2d at this site to either glutamine (R510Q) or to alanine (R510A) reduced the measured deamination rate constant by approximately an order of magnitude (Fig. 6c). In addition, the contact point near the 5’ end of the unedited strand involves G593, K594 and R348, residues completely conserved in the family of ADAR2s (Fig. 2c, Supplementary Table 1). Mutation of any of these residues to alanine (G593A, K594A, R348A) substantially reduces editing activity (Fig. 6c). In addition, mutation of G593 to glutamic acid (G593E) resulted in a nearly two orders of magnitude reduction in rate, consistent with proximity of this residue to the negatively charged phosphodiester backbone of the RNA (Fig. 6c). RNA binding leads to an ordering of the 454–477 loop, which was disordered in the RNA-free hADAR2d structure (Fig. 1d, green) (Supplementary Video 2). This loop binds the RNA duplex contacting the minor groove near the editing site and inserting into the adjacent major groove (Fig. 6e). This loop sequence is conserved in ADAR2s but different in the family of ADAR1s (Fig. 6d). The substantial difference in sequence between the ADARs in this RNA-binding loop suggests differences in editing site selectivity between the two ADARs arise, at least in part, from differences in how this loop binds RNA substrates. Base flipping is a well-characterized mechanism by which nucleic acid modifying enzymes gain access to sites of reaction that are otherwise buried in base-paired structures. DNA methylases, DNA repair glycosylases and RNA loop modifying enzymes are known that flip a nucleotide out of a base pair. However, none of the structurally characterized base-flipping enzymes access their reactive sites from within a normal base-paired RNA duplex. We are aware of one other protein-induced nucleotide flipping from an RNA duplex region. Bacterial initiation factor 1 (IF1) binds to the 30S ribosomal subunit at helix 44 of 16S RNA with A1492 and A1493 flipped out of the helix and bound into protein pockets (Supplementary Fig. 5b). However, these nucleotides are located in a highly distorted and dynamic duplex region containing several mismatches and are predisposed to undergo this conformational change. Thus, this system is not illustrative of base flipping from a normal duplex and does not involve an enzyme that must carryout a chemical reaction on the flipped out nucleotide. Other RNA modification enzymes are known that flip nucleotides out of loops, even from base pairs in loop regions (pseudoU synthetase, tRNA transglycosylase, and restrictocin bound to sarcin/ricin loop of 28S rRNA) (Supplementary Fig. 5b). Because the modification sites are not flanked on both sides by normal duplex, these enzymes do not experience the same limits in approach to the substrate that ADARs do. The fact that ADARs must induce flipping from a normal duplex has implications on its preference for adenosines flanked by certain base pairs, a phenomenon that was not well understood prior to this work. In our structures, the flipped out 8-azanebularine is hydrated, mimicking the tetrahedral intermediate predicted for deamination of adenosine (Figs. 1b, 3a, Supplementary Fig. 3 a–b). Our use of 8-azanebularine, with its high propensity to form a covalent hydrate, allowed us to capture a true mimic of the tetrahedral intermediate and reveal the interactions between the deaminase active site and the reactive nucleotide. In addition, 8-azanebularine was found to adopt a 2’-endo sugar pucker with its 2’-hydroxyl H-bonded to the protein backbone carbonyl at T375. The 2’ endo conformation appears to facilitate access of the nucleobase to the zinc-bound water for nucleophilic attack at C6. Several other base-flipping enzymes stabilize the altered nucleic acid conformation by intercalation of an amino acid side chain into the space vacated by the flipped out base. For hADAR2, E488 serves this role. In the two structures with wild type hADAR2, the E488 residue and orphan base are in nearly identical positions (see Supplementary Fig. 4a for overlay). Thus, the E488 side chain directly contacts each orphan base, likely by accepting an H-bond from uracil N3H or by donating an H-bond to cytidine N3. The latter interaction requires E488 to be protonated. The pKa of E488 in the ADAR-RNA complex has not been measured, but proximity to H-bond acceptors, such as cytidine N3, and insertion between stacked nucleobases, would undoubtedly elevate this value and could lead to a substantial fraction in the protonated state at physiologically relevant pH. Since the glutamine side chain is fully protonated under physiologically relevant conditions, a rate enhancement for the E488Q mutant would be expected if the reaction requires E488 protonation. The interactions of hADAR2d with base pairs adjacent to the editing site adenosine explain the known 5’ and 3’ nearest neighbor preferences (Fig. 5). While these studies indicate the ADAR2 catalytic domain makes an important contact to the 3’ nearest neighbor G, Stefl et al. suggested the 3’ G preference arises from dsRBD binding selectivity for ADAR2. These authors reported a model for ADAR2’s dsRBDs bound to an editing substrate based on NMR data from the isolated dsRBDs (lacking the deaminase domain) and short RNA fragments derived from the GluR-B R/G site RNA. They describe an interaction wherein the 3’ G 2-amino group H-bonds to the backbone carbonyl of S258 found in the β1-β2 loop of ADAR2’s dsRBDII. It is not possible for the S486-3’G interaction we describe here and the S258-3’G interaction reported by Stefl et al. to exist in the same complex since both involve protein loops bound in the RNA minor groove at the same location. Because our structures have captured the edited nucleotide in the conformation required to access the active site, the interactions observed here are highly likely to occur during the deamination reaction at the editing site. However, since dsRBDs are known to bind promiscuously with duplex RNA, it is possible that the S258-3’G interaction found in a complex lacking the deaminase domain is not relevant to catalysis at the editing site. It is also possible that ADAR dsRBD and catalytic domain binding are sequential, with release of the dsRBD from the RNA taking place prior to catalytic domain engagement and base flipping. Aicardi-Goutieres Syndrome (AGS) and Dyschromatosis Symmetrica Hereditaria (DSH) are human diseases caused by mutations in the human ADAR1 gene and several of the disease-associated mutations are found in the deaminase domain. Given the conservation in RNA binding surface and active site residues, we expect the hADAR1 catalytic domain to bind RNA with a similar orientation of the helix found in our hADAR2d-RNA structures. When one maps the locations of the AGS-associated mutations onto the hADAR2d-RNA complex, two mutations involve residues in close proximity to the RNA (< 4 Å) (Supplementary Fig. 8a). G487 of hADAR2 is found on the flipping loop near the RNA (Fig. 3b). Sequence in this loop is highly conserved among ADARs and corresponds to G1007 in hADAR1 (Supplementary Table 2). An arginine at this position would preclude close approach of the flipping loop to the RNA, preventing E1008 insertion and base flipping into the active site (Supplementary Fig. 8b). This is consistent with the observation that the G1007R mutation in hADAR1 inhibits RNA editing activity. Also, K376 forms salt bridges with both the 5’ and 3’ phosphodiesters of the guanosine on the 3’ side of the editing site (Fig. 2). The corresponding residue in hADAR1 (R892) could form similar contacts and the R892H mutation would likely alter this interaction. In summary, the structures described here establish human ADAR2 as a base-flipping enzyme that uses a unique mechanism well suited for modifying duplex RNA. In addition, this work provides a basis for understanding the role of the ADAR catalytic domain in determining editing site selectivity and additional structural context to evaluate the impact of ADAR mutations associated with human disease. Unless otherwise stated, reagents were purchased from Fisher Scientific, Sigma-Aldrich, or Life Technologies. T4 polynucleotide kinase, T4 DNA ligase, molecular biology grade bovine serum albumin (BSA), and RNase inhibitor were purchased from New England Biolabs. γ-[P] ATP was purchased from Perkin-Elmer Life Sciences. The Avian Myeloblastosis Virus (AMV) reverse transcriptase, deoxynucleotide triphosphate (dNTP) mix and RQ1 RNase free DNase were purchased from Promega. Pfu Ultra II was purchased from Stratagene. Dpn 1 was purchased from Invitrogen. Quickchange XL II mutagenesis kit was purchased from Agilent Technologies. RNA oligonucleotides were synthesized at the University of Utah DNA/Peptide Core Facility or purchased from GE Healthcare Dharmacon, Inc. or Sigma Aldrich. DNA oligonucleotides were purchased from Integrated DNA Technologies. Storage phosphor imaging plates from Molecular Dynamics were imaged using Molecular Dynamics 9400 Typhoon phosphorimager. Data were analyzed using Molecular Dynamics ImageQuant 5.2 software. Electrospray Ionization (ESI) mass spectrometry of oligonucleotide samples was carried out at the Campus Mass Spectrometry Facilities, UC Davis. Oligonucleotide masses were determined using Mongo Oligo Mass Calculator v2.06. Protein expression and purification were carried out by modifying a previously reported protocol. In brief, BCY123 cells were transformed with a pSc-ADAR construct encoding either hADAR2d-WT or hADAR2d-E488Q (corresponding to the deaminase domain; residues 299–701). Cells were streaked on yeast minimal media minus uracil (CM-ura) plates. A single colony was used to inoculate a 15 mL CM-ura starter culture. After shaking at 300 rpm and 30 °C overnight, 10 mL of starter culture was used to inoculate each liter of yeast growth media. After 24 h, cells were induced with the addition of 110 mL of sterile 30% galactose per liter, and protein was expressed for 5 h. Cells were collected by centrifugation and stored at −80 °C. Cells were lysed in Buffer A (20 mM Tris-HCl pH 8.0, 5% glycerol, 35 mM imidazole, 1mM BME, 0.01% Triton × 100) with 750 mM NaCl using a microfluidizer and cell lysate clarified by centrifugation (39,000 × g for 25 min). Lysate was passed over a 5 mL Ni-NTA column; washed in three steps with 20–50mL of Lysis Buffer, Wash I buffer (Buffer A + 300 mM NaCl), and Wash II buffer (Buffer A + 100 mM NaCl); and protein eluted by a 35–300 mM imidazole gradient in Wash II over 80 min at a flow rate of 1 ml/min. Fractions containing the target protein were pooled and further purified on a 2 mL GE Healthcare Lifesciences Hi-Trap Heparin HP column in the absence of BME. The 10xHis fusion protein was cleaved with an optimized ratio of 1 mg of TEV protease for each 1 mg of protein. Cleavage was carried out for 1–2 h before passing the product over another Ni-NTA column at 0.5 mL/min. The flow-through and wash were collected; dialyzed against 20 mM Tris pH 8.0, 200 mM NaCl, 5% glycerol, and 1 mM BME; and concentrated to just under 1 mL for gel filtration on a GE Healthcare HiLoad 16/600 Superdex 200 PG column. Fractions containing purified protein were pooled and concentrated to 5–7 mg/mL for crystallography trials. The 8-azanebularine (N) phosphoramidite was synthesized as previously described and RNAs were synthesized as previously described. Single-stranded RNAs (See Supplementary Table 2 for sequences) were purified by denaturing polyacrylamide gel electrophoresis and visualized using UV shadowing. Bands were excised from the gel, crushed and soaked overnight at 4 °C in 0.5 M NH4OAc, 0.1% sodium dodecyl sulfate (SDS) and 0.1 mM EDTA. Polyacrylamide fragments were removed using a 0.2 µm filter followed by desalting on C18 Sep-Pak column. The RNA solutions were lyophilized to dryness, re-suspended in nuclease-free water, quantified by absorbance at 260 nm and stored at −70 °C. Oligonucleotide mass was confirmed by electrospray ionization mass spectrometry. Unmodified RNA stands were purchased from Dharmacon-GE Life Sciences and purified as described above. Duplex RNA was hybridized in a 1:1 ratio by heating to 95 °C for 5 min and slowly cooling to 30 °C. Crystals of hADAR2d E488Q+Bdf2-C RNA complex were grown at room temperature by the sitting drop vapor diffusion method. A solution of 0.5 µL volume containing 4.5 mg/mL protein and 70 µM of Bdf2-C 23mer RNA (1:0.7 ADAR2:RNA molar ratio) were mixed with 0.5 µL of 0.1 M MES:NaOH pH 6.5, 9% (w/v) PEG 3350, 13% glycerol, and 0.015M NAD, which was added to improve crystal growth. Crystals took several weeks to grow. A single, cube-shaped crystal about 120 µm in size was soaked briefly in a solution of mother liquor plus 30% glycerol before flash-cooling in liquid nitrogen. Data were collected via fine-phi slicing using 0.2° oscillations on beamline 24-ID-C at the Advanced Photon Source at Argonne National Laboratories in Chicago. To obtain crystals of the hADAR2d+WT:Bdf2-C RNA, an identical procedure was used as above; however, the crystallization conditions had slightly different concentrations of reagents (10% PEG 3350, 15% glycerol, 0.1 M MES:NaOH pH 6.5, no NAD). For the hADAR2d+WT:Bdf2-U construct, hanging drop vapor diffusion using 200 nL of a mixture containing 4.5 mg/mL protein and 70 µM of Bdf2-U (1:0.7 molar ratio) and 200 nL of a mother liquor (0.1 M ammonium acetate, 0.1M Bis-tris pH 5.5, 17% PEG 10,000) yielded several crystals with a morphology similar to the one described above. All wild type crystals were soaked briefly in a solution of mother liquor plus 30% glycerol before flash-cooling in liquid nitrogen. Data were collected via fine-phi slicing using 0.2° oscillations on beamline 12-2 at the Stanford Synchrotron Radiation Lightsource. Crystals of the hADAR2d E488Q+Gli1 RNA complex were grown using hanging drop vapor diffusion. A solution of volume 200 nL containing 4.5 mg/mL protein and 100 µM of Gli1 23mer RNA (1:1 ADAR2:RNA molar ratio) were mixed with 200 nL of 0.1 M MES:NaOH pH 6.5 and 12% PEG 20,000. At room temperature, a single diamond-shaped crystal about 150 µm long and 50 µm wide was observed about a week later. This crystal was soaked briefly in a solution of mother liquor plus 30% glycerol before flash-cooling in liquid nitrogen. Data were collected on beamline 12-2 at the Stanford Synchrotron Radiation Lightsource using the fine-phi splicing described above. Data for the E488Q Bdf2-C-bound and Gli1-bound structures were processed using XDS and scaled with Aimless (CCP4 1994). Diffraction data for hADAR2d wild type structures were processed with XDS and scaled with SCALA (Kabsch, 2010). The RNA-free hADAR2d crystal structure (PDB ID: 1ZY7) was used as a model for molecular replacement using PHENIX. The structures were refined using PHENIX including TLS parameters and Zn coordination restraints. Ideal Zn-ligand distances were determined using average distances found for similar coordination models in the PDB database. Table 1 gives the statistics in data processing and model refinement. The asymmetric unit for Gli1-bound hADAR2d E488Q includes two complexes of protein:RNA. In each of these complexes, the first 17 residues of the deaminase domain (residues 299–316) as well as a C-terminal proline (Pro701) are disordered and were therefore not included in the model. However, although the RNA-free ADAR2 structure (PDB ID: 1ZY7) lacked electron density for residues 457–475, we observed density for the backbone atoms of these residues. These residues were initially modeled as polyalanine. After a several rounds of refinement, electron density revealed the location of some side chains. Residues whose basic side chains interact with the RNA backbone are clearly defined in the final density map. Although some non-RNA-binding side chains show only weak density, backbone density is strong. As observed in the original hADAR2d RNA-free structure, inositol hexakisphosphate (IHP) was buried in the enzyme core. The asymmetric units for Bdf2-bound ADARs contain one ADAR2d:RNA complex (protein chain A) and one RNA-free ADAR2d monomer (chain D). The N-terminus of the Bdf2-bound structures include more residues than Gli1-bound, beginning at Pro305 in chain A and Thr304 for chain D in the mutant structure, and beginning at Arg307 in chain A and Thr304 or Pro305 in chain D in the wild type structures. The first few residues (in structures in which the specified residues are modeled) had weak side chain density, including residues 305 and 307 in chain A, and residues 304–307 in chain D, and are modeled in the structure as alanine. The last residue of E488Q+Bdf2-C, Pro701, had very weak electron density for both protein subunits in the asymmetric unit. Unlike the E488Q+Gli1 structure, electron density was defined better in the originally disordered loop (residues 457–475) for most residues in the Bdf2-bound structures. With the exception of Glu466 in the wildtype structures, we were able to model-build in main chain and side chain atoms for all residues of this loop in the ADAR subunit complexed to the Bdf2 RNA duplexes. In the RNA-free subunit (chain D) of E488Q+Bdf2-C, a crystal contact stabilized this flexible loop so that we were able to model in the backbone for residues 457–475, but residues 465–475 were modeled as alanine because of poorly defined side chain density. An identical crystal contact was observed in the wildtype structures. In the wt+Bdf2-C complex, density for residues 467–470 was strong enough for side chains to be included in these structures; however, side chain density was not strong for residues 465, 466, 471, 473–475, and 477. Therefore, these side chains were not included in the model. In wt+Bdf2-U, density for side chains 465, 466, 470, 471, and 473–475 was too weak to model. IHP was observed in all ADAR deaminase domains in the asymmetric unit. To model the hydrated 8-azanebularine nucleotide in all RNAs, a CCP4 dictionary file for adenosine (A) was modified to replace the 6-amino group with hydrogen, to change atom 8 to nitrogen and to include an additional hydroxyl group off carbon 6. Additionally, an energy minimization calculated idealized structure was used to determine ideal bond angles and distances for the modified base of the hydrated 8-azanebularine (unpublished data, Professor Dean Tantillo, University of California- Davis). The refinement restraint dictionary file was edited to match these parameters. Histidine-tagged human ADAR2 deaminase domain (hADAR2d) and hADAR2d mutant proteins were expressed in S. cerevisiae strain BCY123 and purified as described above with the following modifications. Cell lysate was 0.45 µm filtered after centrifugation and loaded 3 times through 5 mL Ni-NTA Superflow (Qiagen) at 3 mL/min. Washes of 50 ml with buffer 1, 2 and 3 at 4 mL/min followed by elution with 35 mL gradient from Buffer 3 to elution buffer. Selected elution fractions from the Ni NTA column were pooled and loaded at 0.5 mL/min on 1 mL HiTrap Heparin HP column from GE. The column was washed with 10 mL of Heparin 1 buffer at 0.5 mL/min and eluted with a 12 mL gradient from Heparin 1 to Heparin 2 buffer. Selected elution fractions from the Heparin column were pooled and concentrated to <300 µL in 10,000 MWCO Amicon Ultra 4 centrifugal filter at 6500 RCF and 4 °C. TEV protease cleavage and gel filtration steps were omitted. Buffer exchange was accomplished via 3 rounds of concentration to <300 µL followed by addition of 3 mL of Storage buffer. After final concentration, protein concentrations were determined using BSA standards visualized by SYPRO Orange staining on SDS-polyacrylamide gels and the purified proteins were stored at −70 °C. Mutagenesis of hADAR2 catalytic domain was carried out via PCR site directed mutagenesis using the primers listed in Supplementary Table 2. All primers were purchased from IDT and PAGE purified as described above but were desalted by phenol chloroform extraction, ethanol precipitation and 70% ethanol wash instead of C18 Sep-Pack. Sequences for mutant plasmids were confirmed by Sanger sequencing. Oligonucleotides were purified as described above but were desalted by phenol chloroform extraction, ethanol precipitation and 70% ethanol wash. The 3’ GLI1 top strand 12 mer RNAs were radiolabeled with [γ-P] at the 5’ end with T4 PNK as described previously. Labeled 3’ GLI1 top strand 12 mer RNAs were ligated as previously described to give an internally labeled RNA. The splint ligation products were PAGE purified as described above. Labeled RNAs were hybridized with the complementary GLI1 bottom strand 24 mer RNA (Y is chosen based on the identity of X, see Fig. 3b) in 10 mM Tris-HCl, 0.1 mM EDTA pH 7.5 and 100 mM NaCl. See Supplementary Table 2 for RNA sequences. Deamination kinetics of analog containing RNAs were carried out as previously described but with the following modifications. The final reaction volume was 10 µL. Final enzyme concentration was 300 nM. Final RNA concentration was 10 nM. Final reaction conditions were: 16 mM Tris HCl pH 7.4, 3.3% glycerol, 1.6 mM EDTA, 0.003% Nonidet NP-40, 60 mM KCl, 7.1 mM NaCl, 0.5 mM DTT, 160 units/mL Rnasin, 1 µg/mL yeast tRNA. Reactions were quenched by adding 190 µL 95 °C nuclease-free water followed by incubation at 95 °C for 5 min or by 10 µL 0.5% SDS at 95 °C followed by incubation at 95 °C for 5 min. Each experiment was carried out in triplicate, and the rate constants reported in the text are average values ± standard deviations. Sequences of RNAs used to prepare internally labeled substrates are shown in Supplementary Table 2. For comparison of hADAR2-D mutants, deamination kinetics were carried out as described above with the following modifications. Final reaction conditions were 300 nM hADAR2d, 10 nM RNA, 16 mM Tris HCl pH 7.4, 3.6% glycerol, 1.6 mM EDTA, 0.003% Nonidet NP-40, 60 mM KCl, 8.6 mM NaCl, 0.5 mM DTT, 160 units/mL Rnasin, 1 µg/mL yeast tRNA. Duplex RNAs containing 8-azanebularine andP labeled were prepared as previously described. Samples containing 0.25 nM RNA and different concentrations of hADAR2d E488Q (128, 64, 32, 16, 8, 4, 2, 1, 0.5, 0.25 and 0 nM) were equilibrated in 20 mM Tris-HCl, pH 7, 6% glycerol, 0.5 mM DTT, 60 mM KCl, 20 mM NaCl, 0.1 mM BME, 1.5 mM EDTA, 0.003% NP-40, 160 units/ml RNasin, 100 µg/ml BSA and 1.0 µg/ml yeast tRNA for 30 min at 30 °C. Assay and data analysis were carried out as previously described. See Supplementary Table 2 for RNA sequences. A truncation of hGLI1 mRNA incorporating 81 nucleotides upstream and 65 nucleotides downstream of the edited site was transcribed and purified as previously described. 3’ Nearest neighbor mutants of hGLI1 RNA were generated by site directed mutagenesis to generate G to A, G to C and G to U nearest neighbor mutants. A second site −32 bases from the edit site was mutated to maintain the original secondary structure of the RNA. See Supplementary Table 2 for primers used for mutagenesis. Deamination kinetics of transcribed RNAs were carried out as previously described but with the following modifications. Final reaction volume was 20 µL. Final enzyme concentrations was 10 nM. Final RNA concentration was 2 nM. Final reaction conditions were: 17 mM Tris HCl pH 7.4, 5.0% glycerol, 1.6 mM EDTA, 0.003% Nonidet NP-40, 60 mM KCl, 15.6 mM NaCl, 0.5 mM DTT, 160 units/mL RNasin, 1 µg/mL yeast tRNA. Reactions were quenched by adding 10 µL 0.5% SDS at 95 °C followed by incubation at 95 °C for 5 min. cDNA was generated from RNA via RT-PCR, Sanger sequenced and quantified using SeqScanner 2 software from Applied Biosystems. The kobs (min) of each assay was calculated as described previously.
PMC4820378	Template-dependent nucleotide addition in the reverse (3′-5′) direction by Thg1-like protein	Structures of Thg1-like proteins provide insight into the template-dependent nucleotide addition in the reverse (3′-5′) direction.All polynucleotide chain elongation reactions, whether with DNA or RNA, proceed in the 5′-3′ direction. This reaction involves the nucleophilic attack of a 3′-OH of the terminal nucleotide in the elongating chain on the α-phosphate of an incoming nucleotide. The energy in the high-energy bond of the incoming nucleotide is used for its addition [termed tail polymerization (1)]. This elongation reaction of DNA/RNA chains is in clear contrast to the elongation of protein chains in which the high energy of the incoming aminoacyl-tRNA is not used for its own addition but for the addition of the next monomer (termed head polymerization). However, recent studies have shown that the Thg1/Thg1-like protein (TLP) family of proteins extends tRNA nucleotide chains in the reverse (3′-5′) direction (2, 3). In this case, the 5′-end of tRNA is first activated using adenosine 5′-triphosphate (ATP)/guanosine 5′-triphosphate (GTP), followed by nucleophilic attack of a 3′-OH on the incoming nucleotide [nucleoside 5′-triphosphate (NTP)] to yield pppN-tRNA. Thus, the energy in the triphosphate bond of the incoming nucleotide is not used for its own addition but is reserved for subsequent polymerization (that is, head polymerization) (Fig. 1). Top: Reaction scheme of 3′-5′ elongation by Thg1/TLP family proteins. Bottom: Reaction scheme of 5′-3′ elongation by DNA/RNA polymerases. In 3′-5′ elongation by Thg1/TLP family proteins, the 5′-monophosphate of the tRNA is first activated by ATP/GTP, followed by the actual elongation reaction. The energy of the incoming nucleotide is not used for its own addition but is reserved for the subsequent addition (head polymerization). In 5′-3′ elongation by DNA/RNA polymerases, the energy of the incoming nucleotide is used for its own addition (tail polymerization). The best-characterized member of this family of proteins is eukaryotic Thg1 (tRNA guanylyltransferase), which catalyzes the nontemplated addition of a guanylate to the 5′-end of immature tRNA (4). This guanosine at position −1 (G−1) of tRNA is a critical identity element for recognition by the histidyl-tRNA synthase (5–8). Therefore, Thg1 is essential to the fidelity of protein synthesis in eukaryotes. However, Thg1 homologs or TLPs are found in organisms in which G−1 is genetically encoded, and thus, posttranscriptional modification is not required (9, 10). This finding suggests that TLPs may have potential functions other than tRNA maturation. TLPs have been shown to catalyze 5′-end nucleotide addition to truncated tRNA species in vitro in a Watson-Crick template–dependent manner (11). This function of TLPs is not limited to tRNA but occurs efficiently with other tRNAs. Furthermore, the yeast homolog (Thg1p) has been shown to interact with the replication origin recognition complex for DNA replication (12), and the plant homolog (ICA1) was identified as a protein affecting the capacity to repair DNA damage (13). These observations suggest that TLPs may have more general functions in DNA/RNA repair (3, 9, 11). The 3′-5′ addition reaction catalyzed by Thg1 occurs through three reaction steps (4, 14). In the first step, the 5′-monophosphorylated tRNA, which is cleaved by ribonuclease P from pre-tRNA, is activated by ATP, creating a 5′-adenylylated tRNA intermediate. In the second step, the 3′-OH of the incoming GTP attacks the activated intermediate, yielding pppG−1-tRNA. Finally, the pyrophosphate is removed, and mature pG−1-tRNA is created. The crystal structure of human Thg1 (HsThg1) (15) shows that its catalytic core shares structural homology with canonical 5′-3′ nucleotide polymerases, such as T7 DNA/RNA polymerases (16, 17). This finding suggests that 3′-5′ elongation enzymes are related to 5′-3′ polymerases and raises important questions on why 5′-3′ polymerases predominate in nature. The crystal structure of TLP from Bacillus thuringiensis shows that it shares a similar tetrameric assembly and active-site architecture with HsThg1 (18). Furthermore, the structure of Candida albicans Thg1 (CaThg1) complexed with tRNA reveals that the tRNA substrate accesses the reaction center from a direction opposite to that of canonical DNA/RNA polymerase (19). However, in this structural analysis, the 5′-end of tRNA was not activated and the second substrate (GTP) was not bound. Thus, a detailed reaction mechanism remains unknown. Here, we successfully solved the structure of TLP from the methanogenic archaeon Methanosarcina acetivorans (MaTLP) in complex with ppptRNAΔ1, which mimics the activated intermediate of the repair substrate. Although TLP and Thg1 have similar tetrameric organization, the mode of tRNA binding is different in TLP. Furthermore, we obtained the structure in which the GTP analog (GDPNP) was inserted into this complex to form a Watson-Crick base pair with C72 at the 3′-end region of the tRNA. On the basis of these structures, we discuss the reaction mechanism of template-dependent reverse (3′-5′) polymerization in comparison with canonical 5′-3′ polymerization. Previous biochemical experiments have suggested that ppptRNAΔ1, in which the 5′-end of tRNA was triphosphorylated and G1 was deleted, can be an efficient substrate for the repair reaction (guanylyl transfer) of Thg1/TLP (4, 10, 20). Therefore, we prepared a crystal of MaTLP complexed with ppptRNAΔ1 and solved its structure to study the template-directed 3′-5′ elongation reaction by TLP (fig. S1). The crystal contained a dimer of TLP (A and B molecules) and one tRNA in an asymmetric unit. Two dimers in the crystal further assembled as a dimer of dimers by the crystallographic twofold axis (Fig. 2). This tetrameric structure and 4:2 stoichiometry of the TLP-tRNA complex are the same as those of the CaThg1-tRNA complex (19). However, whereas the AB and CD dimers of tetrameric CaThg1 play different roles, respectively recognizing the accepter stem and anticodon of tRNA (19), the AB dimer and its symmetry mate (CD dimer) on tetrameric MaTLP independently bind one molecule of tRNA (fig. S2), recognizing the tRNA accepter stem and elbow region. Thus, consistent with the notion that MaTLP is an anticodon-independent repair enzyme (11), the anticodon was not recognized in the MaTLP-tRNA complex, whereas the binding mode of CaThg1 is for the G−1 addition reaction, therefore the His anticodon has to be recognized (see “Dual binding mode for tRNA repair”). Left: One molecule of the tRNA substrate (ppptRNAΔ1) is bound to the MaTLP dimer. The AB and CD dimers are further dimerized by the crystallographic twofold axis to form a tetrameric structure (dimer of dimers). Right: Left figure rotated by 90. The CD dimer is omitted for clarity. The accepter stem of the tRNA is recognized by molecule A (yellow), and the elbow region by molecule B (blue). Residues important for binding are depicted in stick form. The β-hairpin region of molecule B is shown in red. The elbow region of the tRNA substrate was recognized by the β-hairpin of molecule B of MaTLP. The N atoms in the side chain of R215 in the β-hairpin region of MaTLP were hydrogen-bonded to the phosphate groups of U55 and G57. The O atom on the S213 side chain was also hydrogen-bonded to the phosphate moiety of G57 of the tRNA (Fig. 2). This β-hairpin region was disordered in the crystal structure of the CaThg1-tRNA complex. The accepter stem of the tRNA substrate was recognized by molecule A of MaTLP. The N7 atom of G2 at the 5′-end was hydrogen-bonded to the N atom of the R136 side chain, whereas the α-phosphate was bonded to the N137 side chain (Fig. 2). R136 was also hydrogen-bonded to the base of C72 (the Watson-Crick bond partner of ΔG1). The triphosphate moiety at the 5′-end of the tRNA was bonded to the D21-K26 region. These phosphates were also coordinated to two metal ions, presumably Mg (MgA and MgB) because they were observed at the same positions (figs. S3 and S4) previously identified by CaThg1 (19) and HsThg1 structures (15). These ions were in turn coordinated by the O atoms of the side chains of D21 and D69 and the main-chain O of G22 (fig. S3A). Mutation of D29 and D76 in HsThg1 (corresponding to D21 and D69 of MaTLP) has been shown to markedly decrease G−1 addition activity (15). Here, we successfully obtained the structure of the ternary complex of MaTLP, 5′-activated tRNA (ppptRNAΔ1), and the GTP analog (GDPNP) (Fig. 3 and fig. S4) by soaking the MaTLP-ppptRNAΔ1 complex crystal in a solution containing GDPNP. The obtained structure showed that the guanine base of the incoming GDPNP formed Watson-Crick hydrogen bonds with C72 and accompanied base-stacking interactions with G2 of the tRNA (Fig. 3B), whereas no interaction was observed between the guanine base and the enzyme. These features are consistent with the fact that this elongation reaction is template-dependent. The 5′-end (position 2) of the tRNA moved significantly (Fig. 3C) due to the insertion of GDPNP. Surprisingly, the 5′-triphosphate moiety after movement occupied the GTP/ATP triphosphate position during the activation step (Fig. 3D). Together with the observation that the 3′-OH of the incoming GTP analog was within coordination distance (2.8 Å) to MgA (fig. S3B) and was able to execute a nucleophilic attack on the α-phosphate of the 5′-end, this structure indicates that the elongation reaction (second reaction) takes place at the same reaction center where the activation reaction (first reaction) occurs. Structural change of the tRNA (ppptRNAΔ1) accepter stem in MaTLP caused by insertion of GDPNP. (A) Structure before GDPNP binding. (B) Structure after GDPNP binding. (C) Superposition of the two structures showing movement of the 5′-end of the tRNA before (blue) and after (red) insertion of GDPNP. (D) Superposition of the 5′-end of the tRNA after GDPNP insertion (red) with GTP at the activation step (green), showing that both triphosphate moieties superpose well. The triphosphate moiety of GDPNP was at the interface between molecules A and B and was recognized by the side chains of both molecules, including R19 (molecule A), R83 (molecule B), K86 (molecule B), and R114 (molecule A) (Fig. 3B). All of these residues are well conserved (fig. S5), and mutation of corresponding residues in ScThg1 (R27, R93, K96, and R133) decreased the catalytic efficiency of G−1 addition (21). The triphosphate of the GDPNP was also bonded to the third Mg (MgC), which, unlike MgA and MgB, is not coordinated by the TLP molecule (fig. S3B). This triphosphate binding mode is the same as that for the second nucleotide binding site in Thg1. However, in previous analyses, the base moiety at the second site was either invisible (15) or far beyond the reaction distance of the phosphate, and therefore, flipping of the base was expected to occur (19). To confirm tRNA recognition by the β-hairpin, we created mutation variants with altered residues in the β-hairpin region. Then, tRNA binding and enzymatic activities were measured. β-Hairpin deletion variant delR198-R215 almost completely abolished the binding of tRNAΔ1 (fig. S6). Furthermore, the enzymatic activities of delR198-R215 and delG202-E210 were very weak (5.2 and 13.5%, respectively) compared with wild type, whereas mutations (N179A and F174A/N179A/R188A) on the anticodon recognition site [deduced from the Thg1-tRNA complex structure (19)] had no effect on the catalytic activity (Fig. 4A). Experiments using the tRNAΔ1 substrate gave similar results (Fig. 4A). All these results are consistent with the crystal structure and suggest that the β-hairpin plays an important role in anticodon-independent binding of the tRNA substrate. Residues in the β-hairpin are not well conserved, except for R215 (fig. S5). Mutants R215A and R215A/S213A, in which the completely conserved R215 was changed to alanine, showed a moderate effect on the activity (27.3 and 16.3%, respectively). Thus, specific interactions with the conserved R215 and van der Waals contacts to residues in the β-hairpin would be important for tRNA recognition. The rates of guanylylation by various mutants were measured. Error bars represent the SD of three independent experiments. (A) Guanylylation of ppptRNAΔ1 and ppptRNAΔ1 by various TLP mutants. The activity using [α-P]GTP, wild-type MaTLP, and ppptRNAΔ1 is denoted as 100. (B) Guanylylation of tRNAΔ1, tRNA, and tRNAΔ−1 by various TLP mutants. The activity to tRNAΔ1 is about 10% of ppptRNAΔ1. TLPs catalyze the Watson-Crick template–dependent elongation or repair reaction for 5′-end truncated tRNA substrates lacking G1 only (tRNAΔ1), or lacking both G1 and G2 (tRNAΔ1,2) (11), whereas they do not show any activity with intact tRNA (thus, repair is unnecessary). How TLP distinguishes between tRNAs that need 5′-end repair from ones that do not, or in other words, how the elongation reaction is properly terminated, remains unknown. The present structure of the MaTLP-ppptRNAΔ1 complex shows that, unlike Thg1, the TLP dimer binds one molecule of tRNA by recognizing the elbow region by the β-hairpin of molecule B and the 5′-end by molecule A. Therefore, we speculated that the flexible nature of the β-hairpin enables the recognition of tRNA substrates with different accepter stem lengths. To confirm this speculation, we used computer graphics to examine whether the β-hairpin region was able to bind tRNA substrates with different accepter stem lengths when the 5′-end was properly placed in the reaction site. When the 5′-end was placed in the reaction site, the body of the tRNA molecule shifted in a manner dependent on the accepter stem length. The tRNA body also rotated because of the helical nature of the accepter stem (fig. S7). This model structure showed that the accepter stem of intact tRNA was too long for the β-hairpin to recognize its elbow region, whereas tRNAΔ1 and tRNAΔ1,2 were recognized by the β-hairpin region (fig. S7), which is consistent with previous experiments (11). On the basis of these model structures, we concluded that the TLP molecule can properly terminate elongation by measuring the accepter stem length of tRNA substrates. The present structural analysis revealed that although TLP and Thg1 have a similar tetrameric architecture, they have different binding modes for tRNAs: Thg1 is bound to tRNA as a tetramer, whereas TLP is bound to tRNA as a dimer. This difference in the tRNA binding modes is closely related to their enzymatic functions. The tRNA-specific G−1 addition enzyme Thg1 needs to recognize both the accepter stem and anticodon of tRNA. The tetrameric architecture of the Thg1 molecule allows it to access both regions located at the opposite side of the tRNA molecule [the AB dimer recognizes the accepter stem and CD dimer anticodon (19)]. In contrast, the binding mode of TLP corresponds to the anticodon-independent repair reactions of 5′-truncated general tRNAs. This binding mode is also suitable for the correct termination of the elongation or repair reaction by measuring the length of the accepter stem by the flexible β-hairpin. Because tRNA requires an extra guanosine (G−1) at the 5′-end, the repair enzyme has to extend the 5′-end by one more nucleotide than other tRNAs. TLP has been shown to confer such catalytic activity on tRNAΔ−1 (Fig. 4B) (11). Here, we showed that the TLP mutants, wherein the β-hairpin is truncated and tRNAΔ1 binding ability is lost, can still bind to tRNA (GUG) whose anticodon is changed to that for His (fig. S6, C, H, and I). Also, the intact tRNA, which is not recognized by TLP (Fig. 4B and fig. S6E), can be recognized when its anticodon is changed to that for His (fig. S6D). Furthermore, the TLP variant (F174A/N179A/R188A) whose anticodon recognition site [deduced from the Thg1-tRNA complex structure (19)] is disrupted has been shown to have a reduced catalytic activity to tRNAΔ−1 (Fig. 4B). All these experimental results indicate that TLP recognizes and binds tRNAs carrying the His anticodon in the same way that Thg1 recognizes tRNA. Thus, we concluded that TLP has two tRNA binding modes that are selectively used, depending on both the length of the accepter stem and the anticodon. The elongation or repair reaction normally terminates when the 5′-end reaches position 1, but when the His anticodon is present, TLP binds the tRNA in the second mode by recognizing the anticodon to execute the G−1 addition reaction. By having two different binding modes, TLP can manage this special feature of tRNA. The Thg1/TLP family of proteins extends tRNA chains in the 3′-5′ direction. The reaction involves two steps. First, the 5′-phosphate is activated by GTP/ATP. Then, the activated phosphate is attacked by the incoming nucleotide, resulting in an extension by one nucleotide at the 5′-end (4, 14). Here, we successfully solved for the first time the intermediate structures of the template-dependent 3′-5′ elongation complex of MaTLP. On the basis of these structures, we will discuss the 3′-5′ addition reaction compared with canonical 5′-3′ elongation by DNA/RNA polymerases. Figure 5 is a schematic diagram of the 3′-5′ addition reaction of TLP. This enzyme has two triphosphate binding sites and one reaction center at the position overlapping these two binding sites (Fig. 5A). In the first activation step, when GTP/ATP is bound to site 1 (Fig. 5B), the 5′-phosphate of the tRNA is deprotonated by MgA and attacks the α-phosphate of the GTP/ATP, resulting in an activated intermediate (Fig. 5C). The structure of the MaTLP-ppptRNAΔ1 complex, wherein β- and γ-phosphates coordinate with MgA and MgB, respectively (Figs. 3A and 5C′), may represent this activated intermediate. Subsequent binding of an incoming nucleotide to site 2 followed by formation of the Watson-Crick base pair with a nucleotide in the template strand conveys the 3′-OH of the incoming nucleotide to the position of deprotonation by MgA and the 5′-triphosphate of the tRNA to the reaction center (Figs. 3B and 5D). Then, the elongation reaction of step 2 occurs (Fig. 5E). Thus, the present structure shows that this 3′-5′ elongation enzyme utilizes a reaction center homologous to that of 5′-3′ elongation enzymes for both activation and elongation in a stepwise fashion. Although these two reactions are similar in chemistry, their substrate characteristics are very different. It should be noted that TLP has evolved to allow the occurrence of these two elaborate reaction steps within one reaction center. (A) The reaction center overlapped with two triphosphate binding sites. A, B, and C (in green) represent binding sites for MgA, MgB, and MgC. P (in blue) represents the phosphate binding sites; O (in red) is the binding site for the deprotonated OH group. Important TLP residues for tRNA and Mg binding are also shown. (B) Structure of the activation complex (corresponding to fig. S8). GTP/ATP binds to triphosphate binding site 1; the deprotonated OH group of the 5′-phosphate attacks the α-phosphate of GTP/ATP, and PPi (inorganic pyrophosphate) is released. (C) Possible structure after the activation step as suggested from the structure of (C′). (C′) Structure before the elongation reaction (corresponding to Fig. 3A). The 5′-triphosphate of the tRNA binds to the same site as for activation of the 5′-terminus of the tRNA in (B). (D) Structure of initiation of the elongation reaction (corresponding to Fig. 3B). The base of the incoming GTP forms a Watson-Crick hydrogen bond with the nucleotide at position 72 in the template chain and a base-stacking interaction with a neighboring base (G2). Movement of the 5′-terminal chain leaves the 5′-triphosphate of the tRNA in the same site as the activation step in (B). The 3′-OH of the incoming GTP is deprotonated by MgA and attacks the α-phosphate to form a covalent bond. (E) After the elongation reaction, the triphosphate of the new nucleotide is placed on site 1, as in (C′), and is ready for the next reaction. Figure 6 compares the 3′-5′ and 5′-3′ elongation mechanisms, showing the symmetrical nature of both elongation reactions using a similar reaction center composed of MgA and MgB in the conserved catalytic core. In TLP, which carries out 3′-5′ elongation, the 3′-OH of the incoming nucleotide attacks the 5′-activated phosphate of the tRNA to form a phosphodiester bond, whereas in the T7 RNA polymerase, a representative 5′-3′ DNA/RNA polymerase, the 3′-OH of the 3′-terminal nucleotide of the RNA attacks the activated phosphate of the incoming nucleotide to form a phosphodiester bond. In these reactions, the roles of the two Mg ions are identical. MgA activates the 3′-OH of the incoming nucleotide in TLP and the 3′-OH of the 3′-end of the RNA chain in T7 RNA polymerase. The role of MgB is to position the 5′-triphosphate of the tRNA in TLP and the incoming nucleotide in T7 RNA polymerase. These two Mg ions are coordinated by a conserved Asp (D21 and D69 in TLP) in the conserved catalytic core. Symmetrical relationship between 3′-5′ elongation by TLP (this study) (left) and 5′-3′ elongation by T7 RNA polymerase [Protein Data Bank (PDB) ID: 1S76] (right). Red arrows represent elongation directions. In the 3′-5′ elongation reaction, the 3′-OH of the incoming nucleotide attacks the 5′-activated phosphate of the tRNA to form a phosphodiester bond, whereas in the 5′-3′ elongation reaction, the 3′-OH of the 3′-terminal nucleotide of the RNA attacks the activated phosphate of the incoming nucleotide to form a phosphodiester bond. Green spheres represent Mg ions. Because the chemical roles of tRNA and the incoming nucleotide are reversed in these two reactions, these two substrates are inserted into a similar reaction center from opposite directions (Fig. 6). In spite of this difference, their fundamental reaction scheme is conserved. However, from an energetic viewpoint, these two reactions are clearly different: Whereas the high energy of the incoming nucleotide is used for its own addition in DNA/RNA polymerases, the high energy of the incoming nucleotide is used for subsequent addition in TLP. For this reason, TLP requires a mechanism that activates the 5′-terminus of the tRNA during the initial step of the reaction. Our analysis showed that the initial activation and subsequent elongation reactions occur sequentially at one reaction center. In this case, the enzyme needs to create two substrate binding sites for two different reactions in the vicinities of one reaction center. TLP has successfully created such sites by utilizing a conformational change in the tRNA through Watson-Crick base pairing (Fig. 3). These structural features of the TLP molecule suggest that development of an activation reaction site is a prerequisite for developing the 3′-5′ elongation enzyme. This is clearly more difficult than developing the 5′-3′ elongation enzyme, wherein the activation reaction site is not necessary, and which may be the primary reason why the 5′-3′ elongation enzyme has been exclusively developed. Here, we established a structural basis for 3′-5′ nucleotide elongation and showed that TLP has evolved to acquire a two-step Watson-Crick template–dependent 3′-5′ elongation reaction using the catalytic center homologous to 5′-3′ elongation enzymes. The active site of this enzyme is created at the dimerization interface. The dimerization also endows this protein with the ability to measure the length of the accepter stem of the tRNA substrate, so that the enzyme can properly terminate the elongation reaction. Furthermore, the dual binding mode of this protein suggests that it has further evolved to cover G−1 addition of tRNA by additional dimerization (dimer of dimers). Thus, the present structural analysis is consistent with the scenario in which TLP began as a 5′-end repair enzyme and evolved into a tRNA-specific G−1 addition enzyme. The detailed molecular mechanism of the Thg1/TLP family established by our analysis will open up new perspectives in our understanding of 3′-5′ versus 5′-3′ polymerization and the molecular evolution of template-dependent polymerases. Genomic DNA from M. acetivorans NBRC100939 was obtained from the NITE Biological Resource Center. The MaTLP gene was amplified by polymerase chain reaction from genomic DNA. The DNA fragment encoding MaTLP was then cloned between the Nde I and Xho I restriction sites in a pET26b vector with a C-terminal His tag. In the MaTLP gene, the amber stop codon (UAG) at position 142 was translated as Pyl. To express the full-length MaTLP in Escherichia coli, the TAG codon was altered to TGG (encoding Trp) with the QuikChange Site-Directed Mutagenesis Kit (Agilent Technologies) as previously described (9, 22). The inserted sequence was verified by DNA sequencing. Plasmids were transformed into E. coli strain BL21 (DE3) pLysSRARE by electroporation, and cells were grown in LB medium containing kanamycin (25 μg/ml) and chloramphenicol (34 μg/ml) at 37°C until reaching an optical density at 600 nm (OD600) of 0.45. The cells were then induced by the addition of isopropyl-β-d-thiogalactopyranoside to a final concentration of 250 μM and shifted to 18°C for approximately 20 hours before harvest. The cells were harvested and resuspended in buffer A [50 mM Hepes-NaOH (pH 7.5), 1 M NaCl, 4 mM MgCl2, 10% glycerol, 0.5 mM β-mercaptoethanol, lysozyme (0.5 mg/ml), and deoxyibonuclease (0.1 mg/ml)]. After sonication and centrifugation, the His6-tagged protein was purified by immobilized metal-ion affinity chromatography using a HisTrap HP column (GE Healthcare). The sample was washed with 75 mM imidazole and eluted with a 75 to 400 mM imidazole gradient in buffer B [50 mM tris-HCl (pH 7.5), 500 mM NaCl, 4 mM MgCl2, 20% glycerol, and 0.5 mM β-mercaptoethanol]. Then, the collected fractions were diluted in 300 mM NaCl with buffer C [25 mM tris-HCl (pH 7.5), 10% glycerol, 5 mM MgCl2, and 1 mM dithiothreitol (DTT)] and further purified on a HiTrap Heparin HP column (GE Healthcare) by elution with a 300 to 1000 mM NaCl gradient in buffer C. Finally, the protein was loaded onto a HiLoad 16/60 Superdex 200 prep grade column (GE Healthcare) equilibrated with buffer D [20 mM Hepes-NaOH (pH 7.5), 500 mM NaCl, 5 mM MgCl2, 10% glycerol, and 1 mM DTT]. The protein was concentrated to 3.9 mg/ml by ultrafiltration. All MaTLP mutants were constructed with the QuikChange Site-Directed Mutagenesis Kit. MaTLP mutants were purified by a HisTrap HP column for RNA binding assay and further purified by a HiLoad 16/60 Superdex 200 prep grade column for 3′-5′ nucleotide addition assay. tRNA transcripts derived from yeast tRNA and tRNA were prepared using T7 RNA polymerase as previously described (19). ppptRNA transcripts were prepared by excluding guanosine 5′-monophosphate (GMP) from the reaction mixture. Transcribed tRNAs were purified by a HiTrap DEAE FF column (GE Healthcare) as previously described (23). Pooled tRNAs were precipitated with isopropanol and dissolved in buffer E [20 mM Hepes-NaOH (pH 7.5), 100 mM NaCl, and 10 mM MgCl2]. MaTLP and ppptRNAΔ1 (tRNA with a triphosphorylated 5′-end and deleted G1) were mixed in a molar ratio of 1.7:1 and incubated for 30 min at room temperature. The mixture was then loaded onto a HiLoad 16/60 Superdex 200 prep grade column equilibrated with buffer F [20 mM Hepes-NaOH (pH 7.5), 400 mM NaCl, 5 mM MgCl2, 10% glycerol, and 1 mM DTT]. Fractions containing the MaTLP-ppptRNAΔ1 complex were mixed with 1 mM spermine and concentrated to an OD280 of 16 by ultrafiltration. All crystallization experiments were performed with the sitting-drop vapor diffusion method at 293 K. Initial crystals of MaTLP were obtained by mixing 1 μl of protein solution (3.9 mg/ml) with 1 μl of a reservoir solution containing 0.1 M Hepes-NaOH buffer (pH 7.5), 0.2 M magnesium chloride, and 30% polyethylene glycol 400 (PEG 400). MaTLP-GTP complex crystals were obtained by soaking the MaTLP crystals in the above reservoir solution supplemented with 1 mM GTP overnight. High-resolution crystals of MaTLP in apo form (MaTLP-apo) were obtained unexpectedly by mixing MaTLP with tRNA in 0.1 M sodium/potassium phosphate (pH 6.2) containing 2.5 M NaCl. Crystals of the MaTLP-ppptRNAΔ1 complex were obtained from a solution containing 0.2 M tripotassium citrate, 0.1 M tris (pH 8.0), 37% PEG3350, and 10 mM praseodymium (III) acetate. Crystals of the MaTLP-ppptRNAΔ1-GDPNP complex were obtained by soaking MaTLP-ppptRNAΔ1 complex crystals in a reservoir solution containing 0.2 M tripotassium citrate, 0.1 M tris (pH 8.0), 30% PEG3350, 5% glycerol, and 15 mM GDPNP overnight. Crystals of MaTLP-apo and MaTLP-GTP were cryoprotected with a reservoir solution containing 50% PEG400 before flash-cooling, whereas crystals of the MaTLP-ppptRNAΔ1-GDPNP and MaTLP-ppptRNAΔ1 complexes were flash-cooled without any cryoprotectant under a stream of liquid nitrogen at 100 K. X-ray diffraction data were collected from beamline BL41XU at SPring-8 (Hyogo, Japan) and beamlines BL5A and BL17A at Photon Factory (Ibaraki, Japan). All diffraction data were indexed, integrated, scaled, and merged using XDS (24). The crystal structure of MaTLP-apo was determined by the molecular replacement (MR) method with Molrep (25, 26), using the protomer structure of CaThg1 (PDB ID: 3WBZ) (19) as a search model. The protomer structure of MaTLP-apo was then used as a search model to solve the structures of MaTLP-GTP. The crystal structure of the MaTLP-ppptRNAΔ1 complex was determined by the MR method with PHASER (27), using the protomer structures of MaTLP-apo and tRNA from Saccharomyces cerevisiae (PDB ID: 1EHZ) (28) as search models. The structure of the MaTLP-ppptRNAΔ1 complex was then used as a search model to solve the MaTLP-ppptRNAΔ1-GDPNP complex structure. Initial protein models were fitted manually using Coot (29), and tRNA models were automatically rebuilt by LAFIRE_NAFIT (30); these models were then refined using phenix.refine (31). The data collection and refinement statistics are summarized in Table 1. All structure figures were generated by PyMol (32). Values in parentheses are for the highest-resolution shell. PF, Photon Factory; Rmsd, root-mean-square deviation. *Rmeas = Σhkl Σi \| Ii(hkl) − 〈I(hkl)〉 \|/Σhkl Σi Ii(hkl), where 〈I(hkl)〉 and N(hkl) are the mean intensity of a set of equivalent reflections and the multiplicity, respectively. †Rwork = Σhkl \|\|Fobs\| − \|Fcalc\|\|/Σhkl \|Fobs\|; Rfree was calculated for 5% randomly selected test sets that were not used in the refinement. Nucleotide addition assays were performed as previously described (4). A reaction mixture containing 25 mM Hepes-NaOH (pH 7.5), 400 mM NaCl, 10 mM MgCl2, 3 mM DTT, 5% glycerol, 0.1 μM [α-P]GTP, 100 μM GTP, 1 μM MaTLP variants, and 10 μM tRNA transcript was incubated at 30°C for 2 hours. Then, the reaction was quenched with phenol/chloroform, and the supernatant was resolved on a 10% polyacrylamide gel containing 8 M urea. The radioactivity was visualized with a BAS-1800 II bioimaging analyzer (Fujifilm). A reaction mixture containing 34 μM MaTLP variants and 20 μM tRNA transcript was incubated in buffer F at room temperature for 30 min. Then, the mixture was loaded onto a Superdex 200 10/300 GL column (GE Healthcare) equilibrated with the same buffer.
PMC4887326	Structural insights into the regulatory mechanism of the Pseudomonas aeruginosa YfiBNR system	YfiBNR is a recently identified bis-(3’-5’)-cyclic dimeric GMP (c-di-GMP) signaling system in opportunistic pathogens. It is a key regulator of biofilm formation, which is correlated with prolonged persistence of infection and antibiotic drug resistance. In response to cell stress, YfiB in the outer membrane can sequester the periplasmic protein YfiR, releasing its inhibition of YfiN on the inner membrane and thus provoking the diguanylate cyclase activity of YfiN to induce c-di-GMP production. However, the detailed regulatory mechanism remains elusive. Here, we report the crystal structures of YfiB alone and of an active mutant YfiB complexed with YfiR with 2:2 stoichiometry. Structural analyses revealed that in contrast to the compact conformation of the dimeric YfiB alone, YfiB adopts a stretched conformation allowing activated YfiB to penetrate the peptidoglycan (PG) layer and access YfiR. YfiB shows a more compact PG-binding pocket and much higher PG binding affinity than wild-type YfiB, suggesting a tight correlation between PG binding and YfiB activation. In addition, our crystallographic analyses revealed that YfiR binds Vitamin B6 (VB6) or L-Trp at a YfiB-binding site and that both VB6 and L-Trp are able to reduce YfiB-induced biofilm formation. Based on the structural and biochemical data, we propose an updated regulatory model of the YfiBNR system.Bis-(3’-5’)-cyclic dimeric GMP (c-di-GMP) is a ubiquitous second messenger that bacteria use to facilitate behavioral adaptations to their ever-changing environment. An increase in c-di-GMP promotes biofilm formation, and a decrease results in biofilm degradation (Boehm et al., 2010; Duerig et al., 2009; Hickman et al., 2005; Jenal, 2004; Romling et al., 2013). The c-di-GMP level is regulated by two reciprocal enzyme systems, namely, diguanylate cyclases (DGCs) that synthesize c-di-GMP and phosphodiesterases (PDEs) that hydrolyze c-di-GMP (Kulasakara et al., 2006; Ross et al., 1991; Ross et al., 1987). Many of these enzymes are multiple-domain proteins containing a variable N-terminal domain that commonly acts as a signal sensor or transduction module, followed by the relatively conserved GGDEF motif in DGCs or EAL/HD-GYP domains in PDEs (Hengge, 2009; Navarro et al., 2011; Schirmer and Jenal, 2009). Intriguingly, studies in diverse species have revealed that a single bacterium can have dozens of DGCs and PDEs (Hickman et al., 2005; Kirillina et al., 2004; Kulasakara et al., 2006; Tamayo et al., 2005). In Pseudomonas aeruginosa in particular, 42 genes containing putative DGCs and/or PDEs were identified (Kulasakara et al., 2006). The functional role of a number of downstream effectors of c-di-GMP has been characterized as affecting exopolysaccharide (EPS) production, transcription, motility, and surface attachment (Caly et al., 2015; Camilli and Bassler, 2006; Ha and O’Toole, 2015; Pesavento and Hengge, 2009). However, due to the intricacy of c-di-GMP signaling networks and the diversity of experimental cues, the detailed mechanisms by which these signaling pathways specifically sense and integrate different inputs remain largely elusive. Biofilm formation protects pathogenic bacteria from antibiotic treatment, and c-di-GMP-regulated biofilm formation has been extensively studied in P. aeruginosa (Evans, 2015; Kirisits et al., 2005; Malone, 2015; Reinhardt et al., 2007). In the lungs of cystic fibrosis (CF) patients, adherent biofilm formation and the appearance of small colony variant (SCV) morphologies of P. aeruginosa correlate with prolonged persistence of infection and poor lung function (Govan and Deretic, 1996; Haussler et al., 1999; Haussler et al., 2003; Parsek and Singh, 2003; Smith et al., 2006). Recently, Malone and coworkers identified the tripartite c-di-GMP signaling module system YfiBNR (also known as AwsXRO (Beaumont et al., 2009; Giddens et al., 2007) or Tbp (Ueda and Wood, 2009)) by genetic screening for mutants that displayed SCV phenotypes in P. aeruginosa PAO1 (Malone et al., 2012; Malone et al., 2010). The YfiBNR system contains three protein members and modulates intracellular c-di-GMP levels in response to signals received in the periplasm (Malone et al., 2010). More recently, this system was also reported in other Gram-negative bacteria, such as Escherichia coli (Hufnagel et al., 2014; Raterman et al., 2013; Sanchez-Torres et al., 2011), Klebsiella pneumonia (Huertas et al., 2014) and Yersinia pestis (Ren et al., 2014). YfiN is an integral inner-membrane protein with two potential transmembrane helices, a periplasmic Per-Arnt-Sim (PAS) domain, and cytosolic domains containing a HAMP domain (mediate input-output signaling in histidine kinases, adenylyl cyclases, methyl-accepting chemotaxis proteins, and phosphatases) and a C-terminal GGDEF domain indicating a DGC’s function (Giardina et al., 2013; Malone et al., 2010). YfiN is repressed by specific interaction between its periplasmic PAS domain and the periplasmic protein YfiR (Malone et al., 2010). YfiB is an OmpA/Pal-like outer-membrane lipoprotein (Parsons et al., 2006) that can activate YfiN by sequestering YfiR (Malone et al., 2010) in an unknown manner. Whether YfiB directly recruits YfiR or recruits YfiR via a third partner is an open question. After the sequestration of YfiR by YfiB, the c-di-GMP produced by activated YfiN increases the biosynthesis of the Pel and Psl EPSs, resulting in the appearance of the SCV phenotype, which indicates enhanced biofilm formation (Malone et al., 2010). It has been reported that the activation of YfiN may be induced by redox-driven misfolding of YfiR (Giardina et al., 2013; Malone et al., 2012; Malone et al., 2010). It is also proposed that the sequestration of YfiR by YfiB can be induced by certain YfiB-mediated cell wall stress, and mutagenesis studies revealed a number of activation residues of YfiB that were located in close proximity to the predicted first helix of the periplasmic domain (Malone et al., 2012). In addition, quorum sensing-related dephosphorylation of the PAS domain of YfiN may also be involved in the regulation (Ueda and Wood, 2009; Xu et al., 2015). Recently, we solved the crystal structure of YfiR in both the non-oxidized and the oxidized states, revealing breakage/formation of one disulfide bond (Cys71-Cys110) and local conformational change around the other one (Cys145-Cys152), indicating that Cys145-Cys152 plays an important role in maintaining the correct folding of YfiR (Yang et al., 2015). In the present study, we solved the crystal structures of an N-terminal truncated form of YfiB (34–168) and YfiR in complex with an active mutant YfiB. Most recently, Li and coworkers reported the crystal structures of YfiB (27–168) alone and YfiR in complex with YfiB (59–168) (Li et al., 2015). Compared with the reported complex structure, YfiB in our YfiB-YfiR complex structure has additional visible N-terminal residues 44–58 that are shown to play essential roles in YfiB activation and biofilm formation. Therefore, we are able to visualize the detailed allosteric arrangement of the N-terminal structure of YfiB and its important role in YfiB-YfiR interaction. In addition, we found that the YfiB shows a much higher PG-binding affinity than wild-type YfiB, most likely due to its more compact PG-binding pocket. Moreover, we found that Vitamin B6 (VB6) or L-Trp can bind YfiR with an affinity in the ten millimolar range. Together with functional data, these results provide new mechanistic insights into how activated YfiB sequesters YfiR and releases the suppression of YfiN. These findings may facilitate the development and optimization of anti-biofilm drugs for the treatment of chronic infections. We obtained two crystal forms of YfiB (residues 34–168, lacking the signal peptide from residues 1–26 and periplasmic residues 27–33), crystal forms I and II, belonging to space groups P21 and P41, respectively. The crystal structure of YfiB monomer consists of a five-stranded β-sheet (β1-2-5-3-4) flanked by five α-helices (α1–5) on one side. In addition, there is a short helix turn connecting the β4 strand and α4 helix (Fig. 1A and 1B). Each crystal form contains three different dimeric types of YfiB, two of which are present in both, suggesting that the rest of the dimeric types may result from crystal packing. Here, we refer to the two dimeric types as “head to head” and “back to back” according to the interacting mode (Fig. 2A and 2E), with the total buried surface areas being 316.8 Å and 554.3 Å, respectively.Figure 1Overall structure of YfiB. (A) The overall structure of the YfiB monomer. (B) A topology diagram of the YfiB monomer. (C and D) The analytical ultracentrifugation experiment results for the wild-type YfiB and YfiB Figure 2Two dimeric types of YfiB dimer. (A–C) The “head to head” dimer. (D–F) The “back to back” dimer. (A) and (E) indicate the front views of the two dimers, (B) and (F) indicate the top views of the two dimers, and (C) and (D) indicate the details of the two dimeric interfaces Overall structure of YfiB. (A) The overall structure of the YfiB monomer. (B) A topology diagram of the YfiB monomer. (C and D) The analytical ultracentrifugation experiment results for the wild-type YfiB and YfiB Two dimeric types of YfiB dimer. (A–C) The “head to head” dimer. (D–F) The “back to back” dimer. (A) and (E) indicate the front views of the two dimers, (B) and (F) indicate the top views of the two dimers, and (C) and (D) indicate the details of the two dimeric interfaces The “head to head” dimer exhibits a clamp shape. The dimerization occurs mainly via hydrophobic interactions formed by A37 and I40 on the α1 helices, L50 on the β1 strands, and W55 on the β2 strands of both molecules, making a hydrophobic interacting core (Fig. 2A–C). The “back to back” dimer presents a Y shape. The dimeric interaction is mainly hydrophilic, occurring among the main-chain and side-chain atoms of N68, L69, D70 and R71 on the α2-α3 loops and R116 and S120 on the α4 helices of both molecules, resulting in a complex hydrogen bond network (Fig. 2D–F). To gain structural insights into the YfiB-YfiR interaction, we co-expressed YfiB (residues 34–168) and YfiR (residues 35–190, lacking the signal peptide), but failed to obtain the complex, in accordance with a previous report in which no stable complex of YfiB-YfiR was observed (Malone et al., 2012). It has been reported that single mutants of Q39, L43, F48 and W55 contribute to YfiB activation leading to the induction of the SCV phenotype in P. aeruginosa PAO1 (Malone et al., 2012). It is likely that these residues may be involved in the conformational changes of YfiB that are related to YfiR sequestration (Fig. 3C). Therefore, we constructed two such single mutants of YfiB (YfiB and YfiB). As expected, both mutants form a stable complex with YfiR. Finally, we crystalized YfiR in complex with the YfiB mutant and solved the structure at 1.78 Å resolution by molecular replacement using YfiR and YfiB as models.Figure 3Overall structure of the YfiB-YfiR complex and the conserved surface in YfiR. (A) The overall structure of the YfiB-YfiR complex. The YfiB molecules are shown in cyan and yellow. The YfiR molecules are shown in green and magenta. Two interacting regions are highlighted by red rectangles. (B) Structural superposition of apo YfiB and YfiR-bound YfiB. To illustrate the differences between apo YfiB and YfiR-bound YfiB, the apo YfiB is shown in pink, except residues 34–70 are shown in red, whereas the YfiR-bound YfiB is shown in cyan, except residues 44–70 are shown in blue. (C) Close-up view of the differences between apo YfiB and YfiR-bound YfiB. The residues proposed to contribute to YfiB activation are illustrated in sticks. The key residues in apo YfiB are shown in red and those in YfiB are shown in blue. (D) Close-up views showing interactions in regions I and II. YfiB and YfiR are shown in cyan and green, respectively. (E and F) The conserved surface in YfiR contributes to the interaction with YfiB. (G) The residues of YfiR responsible for interacting with YfiB are shown in green sticks, and the proposed YfiN-interacting residues are shown in yellow sticks. The red sticks, which represent the YfiB-interacting residues, are also responsible for the proposed interactions with YfiN Overall structure of the YfiB-YfiR complex and the conserved surface in YfiR. (A) The overall structure of the YfiB-YfiR complex. The YfiB molecules are shown in cyan and yellow. The YfiR molecules are shown in green and magenta. Two interacting regions are highlighted by red rectangles. (B) Structural superposition of apo YfiB and YfiR-bound YfiB. To illustrate the differences between apo YfiB and YfiR-bound YfiB, the apo YfiB is shown in pink, except residues 34–70 are shown in red, whereas the YfiR-bound YfiB is shown in cyan, except residues 44–70 are shown in blue. (C) Close-up view of the differences between apo YfiB and YfiR-bound YfiB. The residues proposed to contribute to YfiB activation are illustrated in sticks. The key residues in apo YfiB are shown in red and those in YfiB are shown in blue. (D) Close-up views showing interactions in regions I and II. YfiB and YfiR are shown in cyan and green, respectively. (E and F) The conserved surface in YfiR contributes to the interaction with YfiB. (G) The residues of YfiR responsible for interacting with YfiB are shown in green sticks, and the proposed YfiN-interacting residues are shown in yellow sticks. The red sticks, which represent the YfiB-interacting residues, are also responsible for the proposed interactions with YfiN The YfiB-YfiR complex is a 2:2 heterotetramer (Fig. 3A) in which the YfiR dimer is clamped by two separated YfiB molecules with a total buried surface area of 3161.2 Å. The YfiR dimer in the complex is identical to the non-oxidized YfiR dimer alone (Yang et al., 2015), with only Cys145-Cys152 of the two disulfide bonds well formed, suggesting Cys71-Cys110 disulfide bond formation is not essential for forming YfiB-YfiR complex. The N-terminal structural conformation of YfiB, from the foremost N-terminus to residue D70, is significantly altered compared with that of the apo YfiB. The majority of the α1 helix (residues 34–43) is invisible on the electron density map, and the α2 helix and β1 and β2 strands are rearranged to form a long loop containing two short α-helix turns (Fig. 3B and 3C), thus embracing the YfiR dimer. The observed changes in conformation of YfiB and the results of mutagenesis suggest a mechanism by which YfiB sequesters YfiR. The YfiB-YfiR interface can be divided into two regions (Fig. 3A and 3D). Region I is formed by numerous main-chain and side-chain hydrophilic interactions between residues E45, G47 and E53 from the N-terminal extended loop of YfiB and residues S57, R60, A89 and H177 from YfiR (Fig. 3D-I(i)). Additionally, three hydrophobic anchoring sites exist in region I. The residues F48 and W55 of YfiB are inserted into the hydrophobic cores mainly formed by the main chain and side chain carbon atoms of residues S57/Q88/A89/N90 and R60/R175/H177 of YfiR, respectively; and F57 of YfiB is inserted into the hydrophobic pocket formed by L166/I169/V176/P178/L181 of YfiR (Fig. 3D-I(ii)). In region II, the side chains of R96, E98 and E157 from YfiB interact with the side chains of E163, S146 and R171 from YfiR, respectively. Additionally, the main chains of I163 and V165 from YfiB form hydrogen bonds with the main chains of L166 and A164 from YfiR, respectively, and the main chain of P166 from YfiB interacts with the side chain of R185 from YfiR (Fig. 3D-II). These two regions contribute a robust hydrogen-bonding network to the YfiB-YfiR interface, resulting in a tightly bound complex. Based on the observations that two separated YfiB molecules form a 2:2 complex structure with YfiR dimer, we performed an analytical ultracentrifugation experiment to check the oligomeric states of wild-type YfiB and YfiB. The results showed that wild-type YfiB exists in both monomeric and dimeric states in solution, while YfiB primarily adopts the monomer state in solution (Fig. 1C–D). This suggests that the N-terminus of YfiB plays an important role in forming the dimeric YfiB in solution and that the conformational change of residue L43 is associated with the stretch of the N-terminus and opening of the dimer. Therefore, it is possible that both dimeric types might exist in solution. For simplicity, we only discuss the “head to head” dimer in the following text. PG-associated lipoprotein (Pal) is highly conserved in Gram-negative bacteria and anchors to the outer membrane through an N-terminal lipid attachment and to PG layer through its periplasmic domain, which is implicated in maintaining outer membrane integrity. Previous homology modeling studies suggested that YfiB contains a Pal-like PG-binding site (Parsons et al., 2006), and the mutation of two residues at this site, D102 and G105, reduces the ability for biofilm formation and surface attachment (Malone et al., 2012). In the YfiB-YfiR complex, one sulfate ion is found at the bottom of each YfiB molecule (Fig. 3A) and forms a strong hydrogen bond with D102 of YfiB (Fig. 4A and 4C). Structural superposition between YfiB and Haemophilus influenzae Pal complexed with biosynthetic peptidoglycan precursor (PG-P), UDP-N-acetylmuramyl-L-Ala-α-D-Glu-m-Dap-D-Ala-D-Ala (m-Dap is meso-diaminopimelate) (PDB code: 2aiz) (Parsons et al., 2006), revealed that the sulfate ion is located at the position of the m-Dap5 ϵ-carboxylate group in the Pal/PG-P complex (Fig. 4A). In the Pal/PG-P complex structure, the m-Dap5 ϵ-carboxylate group interacts with the side-chain atoms of D71 and the main-chain amide of D37 (Fig. 4B). Similarly, in the YfiR-bound YfiB structure, the sulfate ion interacts with the side-chain atoms of D102 (corresponding to D71 in Pal) and R117 (corresponding to R86 in Pal) and the main-chain amide of N68 (corresponding to D37 in Pal). Moreover, a water molecule was found to bridge the sulfate ion and the side chains of N67 and D102, strengthening the hydrogen bond network (Fig. 4C). In addition, sequence alignment of YfiB with Pal and the periplasmic domain of OmpA (proteins containing PG-binding site) showed that N68 and D102 are highly conserved (Fig. 4G, blue stars), suggesting that these residues contribute to the PG-binding ability of YfiB.Figure 4The PG-binding site in YfiB. (A) Structural superposition of the PG-binding sites of the H. influenzae Pal/PG-P complex and YfiR-bound YfiB complexed with sulfate ions. (B) Close-up view showing the key residues of Pal interacting with the m-Dap5 ε-carboxylate group of PG-P. Pal is shown in wheat and PG-P is in magenta. (C) Close-up view showing the key residues of YfiR-bound YfiB interacting with a sulfate ion. YfiR-bound YfiB is shown in cyan; the sulfate ion, in green; and the water molecule, in yellow. (D) Structural superposition of the PG-binding sites of apo YfiB and YfiR-bound YfiB, the key residues are shown in stick. Apo YfiB is shown in yellow and YfiR-bound YfiB in cyan. (E and F) MST data and analysis for binding affinities of (E) YfiB wild-type and (F) YfiB with PG. (G) The sequence alignment of P. aeruginosa and E. coli sources of YfiB, Pal and the periplasmic domain of OmpA The PG-binding site in YfiB. (A) Structural superposition of the PG-binding sites of the H. influenzae Pal/PG-P complex and YfiR-bound YfiB complexed with sulfate ions. (B) Close-up view showing the key residues of Pal interacting with the m-Dap5 ε-carboxylate group of PG-P. Pal is shown in wheat and PG-P is in magenta. (C) Close-up view showing the key residues of YfiR-bound YfiB interacting with a sulfate ion. YfiR-bound YfiB is shown in cyan; the sulfate ion, in green; and the water molecule, in yellow. (D) Structural superposition of the PG-binding sites of apo YfiB and YfiR-bound YfiB, the key residues are shown in stick. Apo YfiB is shown in yellow and YfiR-bound YfiB in cyan. (E and F) MST data and analysis for binding affinities of (E) YfiB wild-type and (F) YfiB with PG. (G) The sequence alignment of P. aeruginosa and E. coli sources of YfiB, Pal and the periplasmic domain of OmpA Interestingly, superposition of apo YfiB with YfiR-bound YfiB revealed that the PG-binding region is largely altered mainly due to different conformation of the N68 containing loop. Compared to YfiB, the N68-containing loop of the apo YfiB flips away about 7 Å, and D102 and R117 swing slightly outward; thus, the PG-binding pocket is enlarged with no sulfate ion or water bound (Fig. 4D). Therefore, we proposed that the PG-binding ability of inactive YfiB might be weaker than that of active YfiB. To validate this, we performed a microscale thermophoresis (MST) assay to measure the binding affinities of PG to wild-type YfiB and YfiB, respectively. The results indicated that the PG-binding affinity of YfiB is 65.5 μmol/L, which is about 16-fold stronger than that of wild-type YfiB (Kd = 1.1 mmol/L) (Fig. 4E–F). As the experiment is performed in the absence of YfiR, it suggests that an increase in the PG-binding affinity of YfiB is not a result of YfiB-YfiR interaction and is highly coupled to the activation of YfiB characterized by a stretched N-terminal conformation. Calculation using the ConSurf Server (http://consurf.tau.ac.il/), which estimates the evolutionary conservation of amino acid positions and visualizes information on the structure surface, revealed a conserved surface on YfiR that contributes to the interaction with YfiB (Fig. 3E and 3F). Interestingly, the majority of this conserved surface contributes to the interaction with YfiB (Fig. 3E and 3F). Malone JG et al. have reported that F151, E163, I169 and Q187, located near the C-terminus of YfiR, comprise a putative YfiN binding site (Malone et al., 2012). Interestingly, these residues are part of the conserved surface of YfiR (Fig. 3G). F151, E163 and I169 form a hydrophobic core while, Q187 is located at the end of the α6 helix. E163 and I169 are YfiB-interacting residues of YfiR, in which E163 forms a hydrogen bond with R96 of YfiB (Fig. 3D-II) and I169 is involved in forming the L166/I169/V176/P178/L181 hydrophobic core for anchoring F57 of YfiB (Fig. 3D-I(ii)). Collectively, a part of the YfiB-YfiR interface overlaps with the proposed YfiR-YfiN interface, suggesting alteration in the association-disassociation equilibrium of YfiR-YfiN and hence the ability of YfiB to sequester YfiR. Previous studies indicated that YfiR constitutes a YfiB-independent sensing device that can activate YfiN in response to the redox status of the periplasm, and we have reported YfiR structures in both the non-oxidized and the oxidized states earlier, revealing that the Cys145-Cys152 disulfide bond plays an essential role in maintaining the correct folding of YfiR (Yang et al., 2015). However, whether YfiR is involved in other regulatory mechanisms is still an open question. Intriguingly, a Dali search (Holm and Rosenstrom, 2010) indicated that the closest homologs of YfiR shared the characteristic of being able to bind several structurally similar small molecules, such as L-Trp, L-Phe, B-group vitamins and their analogs, encouraging us to test whether YfiR can recognize these molecules. For this purpose, we co-crystallized YfiR or soaked YfiR crystals with different small molecules, including L-Trp and B-group vitamins. Fortunately, we found obvious small-molecule density in the VB6-bound and Trp-bound YfiR crystal structures (Fig. 5A and 5B), and in both structures, the YfiR dimers resemble the oxidized YfiR structure in which both two disulfide bonds are well formed (Yang et al., 2015).Figure 5Overall Structures of VB6-bound and Trp-bound YfiR. (A) Superposition of the overall structures of VB6-bound and Trp-bound YfiR. (B) Close-up views showing the key residues of YfiR that bind VB6 and L-Trp. The electron densities of VB6 and Trp are countered at 3.0σ and 2.3σ, respectively, in \|Fo\|-\|Fc\| maps. (C) Superposition of the hydrophobic pocket of YfiR with VB6, L-Trp and F57 of YfiB Overall Structures of VB6-bound and Trp-bound YfiR. (A) Superposition of the overall structures of VB6-bound and Trp-bound YfiR. (B) Close-up views showing the key residues of YfiR that bind VB6 and L-Trp. The electron densities of VB6 and Trp are countered at 3.0σ and 2.3σ, respectively, in \|Fo\|-\|Fc\| maps. (C) Superposition of the hydrophobic pocket of YfiR with VB6, L-Trp and F57 of YfiB Structural analyses revealed that the VB6 and L-Trp molecules are bound at the periphery of the YfiR dimer, but not at the dimer interface. Interestingly, VB6 and L-Trp were found to occupy the same hydrophobic pocket, formed by L166/I169/V176/P178/L181 of YfiR, which is also a binding pocket for F57 of YfiB, as observed in the YfiB-YfiR complex (Fig. 5C). To evaluate the importance of F57 in YfiB-YfiR interaction, the binding affinities of YfiB and YfiB for YfiR were measured by isothermal titration calorimetry (ITC). The results showed Kd values of 1.4 × 10 mol/L and 5.3 × 10 mol/L for YfiB and YfiB, respectively, revealing that the YfiB mutant caused a 3.8-fold reduction in the binding affinity compared with the YfiB mutant (Fig. 6F and 6G).Figure 6Functional analysis of VB6 and L-Trp. (A and B) The effect of increasing concentrations of VB6 or L-Trp on YfiB-induced attachment (bars). The relative optical density is represented as curves. Wild-type YfiB is used as negative control. (C and D) BIAcore data and analysis for binding affinities of (C) VB6 and (D) L-Trp with YfiR. (E–G) ITC data and analysis for titration of (E) YfiB wild-type, (F) YfiB, and (G) YfiB into YfiR Functional analysis of VB6 and L-Trp. (A and B) The effect of increasing concentrations of VB6 or L-Trp on YfiB-induced attachment (bars). The relative optical density is represented as curves. Wild-type YfiB is used as negative control. (C and D) BIAcore data and analysis for binding affinities of (C) VB6 and (D) L-Trp with YfiR. (E–G) ITC data and analysis for titration of (E) YfiB wild-type, (F) YfiB, and (G) YfiB into YfiR In parallel, to better understand the putative functional role of VB6 and L-Trp, yfiB was deleted in a PAO1 wild-type strain, and a construct expressing the YfiB mutant was transformed into the PAO1 ΔyfiB strain to trigger YfiB-induced biofilm formation. Growth and surface attachment assays were carried out for the yfiB-L43P strain in the presence of increasing concentrations of VB6 or L-Trp. As shown in Fig. 6A and 6B, the over-expression of YfiB induced strong surface attachment and much slower growth of the yfiB-L43P strain, and as expected, a certain amount of VB6 or L-Trp (4–6 mmol/L for VB6 and 6–10 mmol/L for L-Trp) could reduce the surface attachment. Interestingly, at a concentration higher than 8 mmol/L, VB6 lost its ability to inhibit biofilm formation, implying that the VB6-involving regulatory mechanism is highly complicated and remains to be further investigated. Of note, both VB6 and L-Trp have been reported to correlate with biofilm formation in certain Gram-negative bacteria (Grubman et al., 2010; Shimazaki et al., 2012). In Helicobacter pylori in particular, VB6 biosynthetic enzymes act as novel virulence factors, and VB6 is required for full motility and virulence (Grubman et al., 2010). In E. coli, mutants with decreased tryptophan synthesis show greater biofilm formation, and matured biofilm is degraded by L-tryptophan addition (Shimazaki et al., 2012). However, the detailed mechanism remains elusive. To answer the question whether competition of VB6 or L-Trp for the YfiB F57-binding pocket of YfiR plays an essential role in inhibiting biofilm formation, we measured the binding affinities of VB6 and L-Trp for YfiR via BIAcore experiments. The results showed relatively weak Kd values of 35.2 mmol/L and 76.9 mmol/L for VB6 and L-Trp, respectively (Fig. 6C and 6D). Based on our results, we concluded that VB6 or L-Trp can bind to YfiR, however, VB6 or L-Trp alone may have little effects in interrupting the YfiB-YfiR interaction, the mechanism by which VB6 or L-Trp inhibits biofilm formation remains unclear and requires further investigation. Previous studies suggested that in response to cell stress, YfiB in the outer membrane sequesters the periplasmic protein YfiR, releasing its inhibition of YfiN on the inner membrane and thus inducing the diguanylate cyclase activity of YfiN to allow c-di-GMP production (Giardina et al., 2013; Malone et al., 2012; Malone et al., 2010). However, the pattern of interaction between these proteins and the detailed regulatory mechanism remain unknown due to a lack of structural information. Here, we report the crystal structures of YfiB alone and an active mutant YfiB in complex with YfiR, indicating that YfiR forms a 2:2 complex with YfiB via a region composed of conserved residues. Our structural data analysis shows that the activated YfiB has an N-terminal portion that is largely altered, adopting a stretched conformation compared with the compact conformation of the apo YfiB. The apo YfiB structure constructed beginning at residue 34 has a compact conformation of approximately 45 Å in length. In addition to the preceding 8 aa loop (from the lipid acceptor Cys26 to Gly34), the full length of the periplasmic portion of apo YfiB can reach approximately 60 Å. It was reported that the distance between the outer membrane and the cell wall is approximately 50 Å and that the thickness of the PG layer is approximately 70 Å (Matias et al., 2003). Thus, YfiB alone represents an inactive form that may only partially insert into the PG matrix. By contrast, YfiR-bound YfiB (residues 44–168) has a stretched conformation of approximately 55 Å in length. In addition to the 17 preceding intracellular residues (from the lipid acceptor Cys26 to Leu43), the length of the intracellular portion of active YfiB may extend over 100 Å, assuming a fully stretched conformation. Provided that the diameter of the widest part of the YfiB dimer is approximately 64 Å, which is slightly smaller than the smallest diameter of the PG pore (70 Å) (Meroueh et al., 2006), the YfiB dimer should be able to penetrate the PG layer. These results, together with our observation that activated YfiB has a much higher cell wall binding affinity, and previous mutagenesis data showing that (1) both PG binding and membrane anchoring are required for YfiB activity and (2) activating mutations possessing an altered N-terminal loop length are dominant over the loss of PG binding (Malone et al., 2012), suggest an updated regulatory model of the YfiBNR system (Fig. 7). In this model, in response to a particular cell stress that is yet to be identified, the dimeric YfiB is activated from a compact, inactive conformation to a stretched conformation, which possesses increased PG binding affinity. This allows the C-terminal portion of the membrane-anchored YfiB to reach, bind and penetrate the cell wall and sequester the YfiR dimer. The YfiBNR system provides a good example of a delicate homeostatic system that integrates multiple signals to regulate the c-di-GMP level. Homologs of the YfiBNR system are functionally conserved in P. aeruginosa (Malone et al., 2012; Malone et al., 2010), E. coli (Hufnagel et al., 2014; Raterman et al., 2013; Sanchez-Torres et al., 2011), K. pneumonia (Huertas et al., 2014) and Y. pestis (Ren et al., 2014), where they affect c-di-GMP production and biofilm formation. The mechanism by which activated YfiB relieves the repression of YfiN may be applicable to the YfiBNR system in other bacteria and to analogous outside-in signaling for c-di-GMP production, which in turn may be relevant to the development of drugs that can circumvent complicated antibiotic resistance.Figure 7Regulatory model of the YfiBNR tripartite system. The periplasmic domain of YfiB and the YfiB-YfiR complex are depicted according to the crystal structures. The lipid acceptor Cys26 is indicated as blue ball. The loop connecting Cys26 and Gly34 of YfiB is modeled. The PAS domain of YfiN is shown as pink oval. Once activated by certain cell stress, the dimeric YfiB transforms from a compact conformation to a stretched conformation, allowing the periplasmic domain of the membrane-anchored YfiB to penetrate the cell wall and sequester the YfiR dimer, thus relieving the repression of YfiN Regulatory model of the YfiBNR tripartite system. The periplasmic domain of YfiB and the YfiB-YfiR complex are depicted according to the crystal structures. The lipid acceptor Cys26 is indicated as blue ball. The loop connecting Cys26 and Gly34 of YfiB is modeled. The PAS domain of YfiN is shown as pink oval. Once activated by certain cell stress, the dimeric YfiB transforms from a compact conformation to a stretched conformation, allowing the periplasmic domain of the membrane-anchored YfiB to penetrate the cell wall and sequester the YfiR dimer, thus relieving the repression of YfiN P. aeruginosa YfiR (residues 35–190, lacking the predicted N-terminal periplasmic localization signaling peptide) and YfiB (residues 34–168, lacking the signal peptide from residues 1–26 and periplasmic residues 27–33) were cloned into ORF1 of the pETDuet-1 (Merck Millipore, Darmstadt, Germany) vector via the BamHI and HindIII restriction sites, with a constructed N-terminal His6 and a TEV cleavage site, respectively. In addition, YfiB (residues 34–168) was ligated into the NdeI and XhoI restriction sites of ORF2 in the previously constructed YfiR expression vector. Site-directed mutagenesis was carried out using a QuikChange kit (Agilent Technologies, Santa Clara, CA), following the manufacturer’s instructions. The proteins were over-expressed in the E. coli BL21-CodonPlus(DE3)-RIPL strain. Protein expression was induced by adding 0.5–1 mmol/L IPTG at an OD600 of approximately 0.8. The cell cultures were then incubated for an additional 4.5 h at 37°C. The cells were subsequently harvested by centrifugation and stored at −80°C. Cell suspensions were thawed and homogenized using a high-pressure homogenizer (JNBIO, Beijing, China). YfiR was first purified by Ni affinity chromatography and then incubated with His6-tagged TEV protease overnight. The His6-TEV cleavage site was subsequently removed by incubation with Ni-NTA resin. Finally, YfiR was purified with a HiTrap S column (GE Healthcare), followed by a Superdex 200 (GE Healthcare) column. YfiB was purified with Ni affinity chromatography, followed by a Superdex 200 (GE Healthcare) column. The YfiB-YfiR complex was first purified by Ni affinity chromatography, then by a Superdex 200 (GE Healthcare) column, and finally by a HiTrap S column (GE Healthcare). All of the purified fractions were collected and concentrated to ~40 mg/mL in 20 mmol/L Tris-HCl (pH 8.0) and 200 mmol/L NaCl, frozen in liquid nitrogen and stored at −80°C. Crystal screening was performed with commercial screening kits (Hampton Research, CA, USA) using the sitting-drop vapor diffusion method, and positive hits were optimized using the hanging-drop vapor diffusion method at 293 K. Crystals of the YfiB protein were obtained and optimized in buffer containing 0.2 mol/L lithium sulfate monohydrate, 0.1 mol/L Tris-HCl (pH 8.0) and 30% w/v polyethylene glycol 4000. After being soaked for a few seconds in cryoprotection solution (well solution complemented with 25% xylitol), the crystals were cooled by plunging them into liquid nitrogen. Diffraction-quality crystals of the YfiB-YfiR complex were grown in buffer containing 0.2 mol/L ammonium sulfate, 0.1 mol/L Tris-HCl (pH 8.0) and 12% w/v polyethylene glycol 8000. The crystals were cryoprotected with 8% (w/v) polyethylene glycol 8000 and 0.1 mol/L Tris-HCl (pH 7.5) supplemented with saturated sucrose prior to being flash frozen. Crystals of the native YfiR were obtained and optimized in 0.1 mol/L HEPES (pH 7.5) and 1.8 mol/L ammonium sulfate. VB6-bound YfiR crystals were obtained by soaking the native YfiR crystals in 2 mmol/L VB6 molecules. Trp-bound YfiR crystals were obtained by co-crystalizing the YfiR protein and 4 mmol/L L-Trp molecules in 0.2 mol/L NaCl, 0.1 mol/L BIS-TRIS (pH 5.5), and 25% w/v polyethylene glycol 3350. For cryoprotection, both the VB6-bound and the L-Trp-bound YfiR crystals were soaked in 2.5 mol/L lithium sulfate monohydrate for a few seconds before data collection. Diffraction data for the YfiB crystal belonging to space group P21 was collected in house, the data for the YfiB crystal belonging to space group P41 and for the Trp-bound YfiR crystal were collected on beamline BL17U at the Shanghai Synchrotron Radiation Facility (SSRF), and the data for the VB6-bound YfiR crystal were collected on beamline BL18U at SSRF. Finally, the data for the YfiB-YfiR complex crystal were collected on beamline BL-1A at the Photon Factory in Japan. The diffraction data were processed with the HKL2000 software program (Otwinowski and Minor, 1997). The two YfiB crystal structures respectively belonging to space groups P21 and P41 were both solved by molecular replacement (Lebedev et al., 2008) using the putative MotB-like protein DVU_2228 from D. vulgaris as a model (PDB code: 3khn) at 2.15 Å and 2.8 Å resolution, respectively. Both the VB6-bound and the Trp-bound YfiR crystals belonging to space group P43212, with a dimer in the asymmetric unit, were solved by molecular replacement (Lebedev et al., 2008) using native YfiR as a model (PDB code: 4YN7) at 2.4 Å and 2.5 Å resolution, respectively. The YfiB-YfiR crystal belonging to space group P1, with a 2:2 heterotetramer in the asymmetric unit, was solved by molecular replacement using YfiR and YfiB as models. Electron density maps were calculated using PHENIX (Adams et al., 2010). Model building was performed using COOT (Emsley et al., 2010) and refined with PHENIX (Adams et al., 2010; Afonine et al., 2012). The final structures were analyzed with PROCHECK (Laskowski et al., 1993). Data collection and refinement statistics are presented in Table 1. The figures depicting structures were prepared using PyMOL (http://www.pymol.org). Atomic coordinates and structure factors have been deposited in the RCSB Protein Data Bank (http://www.pdb.org) under accession codes 5EAZ, 5EB0, 5EB1, 5EB2 and 5EB3.Table 1Data collection, phasing and refinement statistics Data collection YfiB (crystal form I)YfiB (crystal form II)VB6-bound YfiRTrp-bound YfiRYfiBL43P-YfiRSpace group P21 P41 P43212 P43212 P1Wavelength (Å)1.541870.97910.978610.97911.10000Resolution (Å) 50.0–2.15 (2.19–2.15)50.0–2.80 (2.85–2.8)50.0–2.4 (2.44–2.4)50.0–2.5 (2.54–2.5)50–1.78 (1.86–1.78)Cell dimensions a, b, c (Å)65.85, 90.45, 66.3046.95, 46.95, 154.24120.24, 120.24, 84.99120.88, 120.88, 88.4649.50, 58.57, 69.86 α, β, γ (°)90, 113.87, 9090, 90, 9090, 90, 9090, 90, 9072.93, 96.98, 90.19 Unique reflections37,625 (1866)8,105 (412)24,776 (1202)23170 (1132)67,774 (6615) I/σI 19.59 (2.62)12.36 (4.15)20.17 (2.4)39.5 (4.68)17.75 (1.89) Completeness (%)97.1 (95.4)97.8 (100)99.1 (98.8)99.9 (100)96.5 (94.6) R merge (%)6.5 (44.5)14.6 (49.7)8.9 (56.8)9.4 (89.2)5.6 (46.3) R meas (%)7.4 (51.6)15.4 (52.0)9.6 (61.7)9.6 (90.8)6.6 (55.1) CC1/2 0.7470.9520.8990.9740.849 Refinement R work (%)20.1419.1717.8218.6617.90 R free(%)26.2926.4919.8123.0520.61Average B factors (Å) Protein25.5442.7038.6835.0332.54 VB6--44.08-- Trp---87.51- SO4 37.1666.5251.5541.9345.51 H2O32.9136.0940.5834.7543.52Root mean square deviations Bond lengths (Å)0.0090.0090.0070.0070.007 Bond angles (°)1.0851.1321.0210.9771.110Ramachandran plot Most favored (%)92.687.796.598.194.2 Additionally allowed (%)7.412.33.51.95.8 Generously allowed (%)00000 Disallowed00000 Numbers in parentheses are for the highest resolution shell The values of CC1/2 are for the highest resolution shell Data collection, phasing and refinement statistics Numbers in parentheses are for the highest resolution shell The values of CC1/2 are for the highest resolution shell Sedimentation velocity measurements were performed on a Beckman ProteomeLab XL-I at 25°C. All protein samples were diluted to an OD280 of 0.7 in 20 mmol/L Tris (pH 8.0) and 200 mmol/L NaCl. Data were collected at 60,000 rpm. (262,000 ×g) every 3 min at a wavelength of 280 nm. Interference sedimentation coefficient distributions, or c(M), were calculated from the sedimentation velocity data using SEDFIT (Schuck, 2000). PG was extracted from the E. coli DH5α strain by following a method described previously (Desmarais et al., 2014). Briefly, cells were cultured until they reached an OD600 of 0.7–0.8 and then collected at 5,000 ×g, 4°C. The collected bacteria were dripped into the boiling 6% (w/v) SDS and stirred at 500 rpm in a boiling water bath for 3 h before incubating overnight at room temperature. The large PG polymers were collected by ultracentrifugation at 130,000 ×g for 1 h at room temperature and washed repeatedly to remove SDS. The pellet was treated with Pronase E (200 μg/mL final concentration) for 3 h at 60°C followed by SDS to remove contaminating proteins and washed three times to remove the SDS by ultracentrifugation. Next, the samples were treated with lysozyme (200 μg/mL final concentration) for 16 h at 37°C. Finally, the purified PG is obtained by treating the samples in a boiling water bath for 10 min and centrifuging it at 13,000 ×g to remove the contaminating lysozyme. Purified YfiB wild-type and it mutant YfiB were fluorescently labeled using the NanoTemper blue protein-labeling kit according to the manufacturer’s protocol. This resulted in coupling of the fluorescent dye NT-495. PG was titrated in 1:1 dilutions starting at 1 mmol/L. To determine of the Kd values, 10 μL labeled protein was mixed with 10 μL PG at various concentrations in Hepes buffer (20 mmol/L Hepes, 200 mmol/L NaCl, 0.005% Tween-20, pH 7.5). After 10 min of incubation, all binding reaction mixtures were loaded into the MST-grade glass capillaries (NanoTemper Technologies), and thermophoresis was measured with a NanoTemper Monolith-NT115 system (20% light-emitting diode, 20% IR laser power). The yfiB deletion construct was produced by SOE-PCR (Hmelo et al., 2015) and contained homologous flanking regions to the target gene. This construct was ligated into the pEX18Gm vector between the HindIII and the KpnI sites. The resulting vector was then used to delete yfiB by two-step allelic exchange (Hmelo et al., 2015). After being introduced into PAO1 via biparental mating with E. coli SM10 (λpir), single crossovers were selected on Vogel-Bonner Minimal Medium (VBMM), which was used for counter-selection against E. coli (P. aeruginosa can utilize citrate as a sole carbon source and energy source, whereas E. coli cannot), containing 50 μg/mL gentamycin. Restreaking was then performed on no-salt Luria-Bertani (NSLB) agar that contained 15% sucrose to force the resolution of double crossovers. Deletion of yfiB in the strains was confirmed by colony PCR. For complementation experiments, yfiB wild-type and L43P mutant genes were cloned into the pJN105 vector via the EcoRI and XbaI restriction sites, respectively. The plasmids were then individually transformed into the PAO1 ΔyfiB strain using the rapid electroporation method described in Choi KH et al. (Choi et al., 2006). Transformants were selected on LB plates containing 50 μg/mL gentamycin. For induction, arabinose was added to a final concentration of 0.2%. The attachment assays were carried out using the MBEC (Minimum Biofilm Eradication Concentration, Innovotech, Inc.) biofilm inoculator, which consists of a plastic lid with 96 pegs and 96 individual wells. The MBEC plates containing 150 μL LB medium/well were inoculated with 1% overnight cultures of the yfiB-L43P strain and incubated overnight at 37°C without shaking. VB6, L-Trp and arabinose were added as appropriate. The peg lids were washed with distilled water, and the attached cell material was then stained with 0.1% crystal violet solution (5% methanol, 5% isopropanol) before further washing to remove excess dye. The crystal violet was re-dissolved in 20% acetic acid solution, and the absorbance was measured at 600 nm. Assays were performed with 12 wells/strain and repeated independently for each experiment. The interaction kinetics of YfiR with VB6 and L-Trp were examined on a SPR machine Biacore 3000 (GE Healthcare) at 25°C. The running buffer (20 mmol/L HEPES, pH 7.5, 150 mmol/L NaCl, 0.005% (v/v) Tween-20) was vacuum filtered, and degassed immediately prior to use. YfiR at 10 μg/mL in 10 mmol/L sodium acetate (pH 5.5) was immobilized to 3000 response units on the carboxymethylated dextran surface-modified chip (CM5 chip). The binding affinities were evaluated over a range of 2.5–40 mmol/L concentrations. Meanwhile, for both binding assays, the concentration of 10 mmol/L was repeated as an internal control. All of the data collected were analyzed using BIAevaluation software version 4.1. ITC experiments were performed in a buffer composed of 20 mmol/L Tris (pH 8.0) and 150 mmol/L NaCl at 25°C using an iTC200 calorimeter (GE Healthcare). YfiB wild-type or its mutants (YfiB, YfiB) (0.4 mmol/L, in the syringe) was titrated into YfiR (0.04 mmol/L, in the cell), respectively. The titration sequence included a single 0.5 µL injection, followed by 19 injections of 2 µL each, with a 2-min interval between injections and a stirring rate of 1000 rpm. The calorimetric data were then analyzed with OriginLab software (GE Healthcare).
PMC4746701	Crystal structure of SEL1L: Insight into the roles of SLR motifs in ERAD pathway	Terminally misfolded proteins are selectively recognized and cleared by the endoplasmic reticulum-associated degradation (ERAD) pathway. SEL1L, a component of the ERAD machinery, plays an important role in selecting and transporting ERAD substrates for degradation. We have determined the crystal structure of the mouse SEL1L central domain comprising five Sel1-Like Repeats (SLR motifs 5 to 9; hereafter called SEL1L). Strikingly, SEL1L forms a homodimer with two-fold symmetry in a head-to-tail manner. Particularly, the SLR motif 9 plays an important role in dimer formation by adopting a domain-swapped structure and providing an extensive dimeric interface. We identified that the full-length SEL1L forms a self-oligomer through the SEL1L domain in mammalian cells. Furthermore, we discovered that the SLR-C, comprising SLR motifs 10 and 11, of SEL1L directly interacts with the N-terminus luminal loops of HRD1. Therefore, we propose that certain SLR motifs of SEL1L play a unique role in membrane bound ERAD machinery.Protein quality control in the endoplasmic reticulum (ER) is essential for maintenance of cellular homeostasis in eukaryotes and is implicated in many severe diseases1. Terminally misfolded proteins in the lumen or membrane of the ER are retrotranslocated into the cytosol, polyubiquitinated, and degraded by the proteasome. The process is called ER-associated protein degradation (ERAD) and is conserved in all eukaryotes234. Accumulating studies have identified key components for ERAD, including HRD1, SEL1L (Hrd3p), Derlin-1, -2, -3 (Der1p), HERP-1, -2 (Usa1p), OS9 (Yos9), XTP-B, and Grp94, all of which are involved in the recognition and translocation of the ERAD substrates in yeast and metazoans56789. The components are differentially localized from the lumen and membrane of the ER to the cytosol, and have different functions in the ERAD process. Yeast ERAD components, which have been extensively characterized through genetic and biochemical studies, are comparable with mammalian ERAD components, sharing similar molecular functions and structural composition. The HRD1 E3 ubiquitin ligase, which is embedded in the ER membrane, is involved in translocating ERAD substrates across the ER membrane and catalyzing substrate ubiquitination via its cytosolic RING finger domain10. SEL1L, the mammalian homolog of Hrd3p, associates with HRD1, mediates HRD1 interactions with the ER luminal lectin OS9, and recognizes substrates to be degraded61112131415. In particular, SEL1L is crucial for translocation of Class I major histocompatibility complex (MHC) heavy chains (HCs)1415. Recent research based on the inducible Sel1l knockout mouse model highlights the physiological functions of SEL1L1617. SEL1L is required for ER homeostasis, which is essential for protein translation, pancreatic function, and cellular and organismal survival. However, despite the functional importance of SEL1L, the molecular structure of SEL1L has not been solved. Previous biochemical studies reveal that SEL1L is a type I transmembrane protein and has a large luminal domain comprising sets of repeated Sel1-like (SLR) motifs18. The SLR motif is a structural motif that closely resembles the tetratricopeptide-repeat (TPR) motif, which is a protein-protein interaction module19. Although there is evidence that the luminal domain of SEL1L is involved in substrate recognition or in forming complexes with chaperones12, it is not known how the unique structure of the repeated SLR motifs contributes to the molecular function of the HRD1-SEL1L E3 ligase complex and affects ERAD at the molecular level. Furthermore, while repeated SLR motifs are commonly found in tandem arrays, the SLR motifs in SEL1L are, according to the primary structure prediction of SEL1L, interspersed among other sequences in the luminal domain and form three SLR domain clusters. Therefore, the way in which these unique structural features of SEL1L are related to its critical function in ERAD remains to be elucidated. To clearly understand the biochemical role of the SLR domains of SEL1L in ERAD, we determined the crystal structure of the central SLR domain of SEL1L. We found that the central domain of SEL1L, comprising SLR motifs 5 through 9 (SEL1L), forms a tight dimer with two-fold symmetry due to domain swapping of the SLR motif 9. We also found that SLR-C, consisting of SLR motifs 10 and 11, directly interacts with the N-terminus luminal loop of HRD1. Based on these observations, we propose a model wherein the SLR domains of SEL1L contribute to the formation of stable oligomers of the ERAD translocation machinery, which is indispensable for ERAD. The Mus musculus SEL1L protein contains 790 amino acids and has 17% sequence identity to its yeast homolog, Hrd3p. Mouse SEL1L contains a fibronectin type II domain at the N-terminus, followed by 11 SLR motifs and a single transmembrane domain at the C-terminus (Fig. 1A)18. The 11 SLR motifs are located in the ER lumen and account for more than two thirds of the mass of full-length SEL1L. The SLR motifs can be grouped into three regions due to the presence of linker sequences among the groups of SLR motifs: SLR-N (SLR motifs 1 to 4), SLR-M (SLR motifs 5 to 9), and SLR-C (SLR motifs 10 to 11) (Fig. 1A). Sequence alignment of the SLR motifs revealed that there is a short linker sequence (residues 336–345) between SLR-N and SLR-M and a long linker sequence (residues 528–635) between SLR-M and SLR-C (Fig. 1A). We first tried to prepare the full-length mouse SEL1L protein, excluding the transmembrane domain at the C-terminus (residues 735–755), by expression in bacteria. However, the full-length SEL1L protein aggregated in solution and produced no soluble protein. To identify a soluble form of SEL1L, we generated serial truncation constructs of SEL1L based on the predicted SLR motifs and highly conserved regions across several different species. Both SLR-N (residues 194–343) and SLR-C (residues 639–719) alone could be solubilized with an MBP tag at the N-terminus, but appeared to be polydisperse when analyzed by size-exclusion chromatography. However, the central region of SEL1L, comprising residues 337–554, was soluble and homogenous in size, as determined by size-exclusion chromatography. To define compact domain boundaries for the central region of SEL1L, we digested the protein with trypsin and analyzed the proteolysis products by SDS-PAGE and N-terminal sequencing. The results of this preliminary biochemical analysis suggested that SEL1L residues 348–533 (SEL1L) would be amenable to structural analysis (Fig. 1A). Crystals of SEL1L grew in space group P21 with four copies of SEL1L (a total of 82 kDa) in the asymmetric unit. The structure was determined by the single-wavelength anomalous diffraction (SAD) method using selenium as the anomalous scatterer (Table 1 and Methods). The assignment of residues during model building was aided by the selenium atom positions, and the structure was refined with native data to 2.6 Å resolution with Rwork/Rfree values of 20.7/27.7%. Statistics for data collection and refinement are presented in Table 1. The mouse SEL1L crystallized as a homodimer, and there were two homodimers in the crystal asymmetric unit (Fig. 1B,C, Supplementary Fig. 1). The two SEL1L molecules dimerize in a head-to-tail manner through a two-fold symmetry interface resulting in a cosmos-like shaped structure (Fig. 1B). The resulting structure resembles the yin-yang symbol with overall dimensions of 60 × 60 × 25 Å, where a SEL1L monomer corresponds to half the symbol. The dimer formation buries a surface area of 1670 Å for each monomer, and no significant differences between the protomers were displayed (final root mean square deviation (RMSD) of 0.6 Å for all Cα atoms). Each protomer is composed of ten α-helices, which form the five SLRs, resulting in an elongated curved structure, confirming the primary structure prediction (Fig. 1D). The α-helices subdivide the structure into five pairs (A and B) as shown in a number of TPRs19 and SLRs2021. Helices A and B are 14 and 13 residues long, respectively, and the two helices are connected by a short turn and loop (Fig. 1D). In addition, a longer loop, consisting of approximately eight amino acids, is inserted between helix B of one SLR and helix A of the next SLR. This arrangement is a unique feature for SLRs among the major classes of repeats containing an α-solenoid. Starting from its N-terminus, the α-solenoid of SEL1L extends across a semi-circle in a right-handed superhelix fashion along the rotation axis of the yin-yang circle. However, the last helix, 9B, at the C-terminus adopts a different conformation, lying parallel to the long axis of helix 9A instead of forming an antiparallel SLR. This unique conformation of helix 9B most likely contributes to formation of the dimer structure of SEL1L, as detailed below. With the exception of the last SLR, the four α-helix pairs possess similar conformations, with RMSD values of 0.7 Å for all Cα atoms. Although the sequence similarity for the pairwise alignments varies between 25% and 35%, all the residues present in the SLR motifs are conserved among the five pairs. The SLR domain of SLR-M ends at residue 524, and C-terminal amino acids 525–533 of the protein are not visible in the electron density map, suggesting that this region is highly flexible. Since no information regarding dimer formation by SEL1L through its SLR motifs is available, we tested whether the SEL1L dimer shown in our crystal structure is a biological unit. First, we cross-linked SEL1L or a longer construct of SEL1L (SEL1L, residues 337–554) using various concentrations of glutaraldehyde (GA) or dimethyl suberimidate (DMS) and analyzed the products by SDS-PAGE. We detected bands at the mass of a dimer for both SEL1L and SEL1L when cross-linked with low concentrations of GA (0.005%) or DMS (0.3 mM) (Supplementary Fig. 2A,B). Next, we conducted analytical ultracentrifugation of SEL1L. Consistent with the cross-linking data, analytical ultracentrifugation revealed that the molecular weight of SEL1L corresponds to a dimer (Supplementary Fig. 2C). Taken together, these data indicate that some kind of dimer is formed in solution. In contrast to a previously described SLR motif containing proteins that exist as monomers in solution2021, SEL1L forms an intimate two-fold homotypic dimer interface (Figs 1B and 2A). The concave surface of each SEL1L domain comprising helix 5A to 9A encircles its dimer counterpart in an interlocking clasp-like arrangement. However, no interactions were seen between the two-fold-related protomers through the concave inner surfaces themselves. Rather, the unique structure of SLR motif 9, consisting of two parallel helices (9A and 9B), is located in the space generated by the concave surface and provides an extensive dimerization interface between the two-fold-related molecules (Fig. 2A). Helix 9B from one protomer inserts into the empty space surrounded by the concave region in the other monomer, forming a domain-swapped conformation. Three major contact interfaces are involved in the interactions, and all interfaces are symmetrically related between the dimer subunits (Fig. 2A). Structure-based sequence alignment of 135 SEL1L phylogenetic sequences using a ConSurf server revealed that the surface residues in the dimer interfaces were highly conserved among the SEL1L orthologs (Fig. 1E)22. First, helix 9B of each SEL1L subunit interacts with residues lining the inner groove from the SLR α-helices (5B, 6B, 7B, and 8B) from its counterpart. In this interface, Leu 516 and Tyr 519 on helix 9B are located in the center, making hydrophobic interactions with Trp 478 on helix 8B, Val 444 on helix 7B, Phe 411 on helix 6B, and Leu 380 on helix 5B from the SEL1L counterpart (Fig. 2A, Interface 1 detail). In addition to hydrophobic interactions, the side chain hydroxyl group of Tyr 519 and the main-chain oxygen of Ile 515 form H-bonds to the side chain of the conserved Gln 377 and His 381 on helix 5B of the two-fold-related protomer. The side chain of Gln 523 forms an H-bond to the side chain of Asp 480 on the two-fold-related protomer (Fig. 2A, Interface 1 detail). Second, the residues from helix 9A interact with the residues from helix 5A of its counterpart in a head-to-tail orientation. In this interface, the interacting residues on helix 9A, including Leu 503, Tyr 499, and the aliphatic side chain of Lys 500, form an extensive network of van der Waals contacts with the hydrophobic residues of the counterpart helix 5A, including Tyr 360, Leu 356, Tyr 359, and Leu 363. In addition to hydrophobic interactions, the side chains of Asn 507 and Ser 510 on helix 9A make H-bonds with highly conserved Arg 384 in the loop between helix 5B and 6A from the two-fold-related protomer (Fig. 2A, Interface 2 detail). Third, the helix 9B from each protomer is involved in the dimer interaction by forming a two-fold antiparallel symmetry. In particular, the side chains of hydrophobic residues, including Phe 518, Leu 521, and Met 524, are directed toward each other, where they make both inter- and intramolecular contacts (Fig. 2A, Interface 3 detail). To further investigate the interactions observed in our crystal structure, we generated a C-terminal deletion mutant (SEL1L) lacking SLR motif 9 (helix 9A and 9B) from SEL1L for comparative analysis. The deletion mutant and the wild-type SEL1L showed no difference in spectra by CD spectroscopy, indicating that the deletion of the SLR motif 9 did not affect the secondary structure of SEL1L (Supplementary Fig. 3). However, the mutant behaved as a monomer in size-exclusion chromatography and analytical ultracentrifugation experiments (Fig. 2B, Supplementary Fig. 2C). Additionally, to further validate the key residues involved in dimer formation, we generated a triple point mutant (Interface 1, I515A, L516A, and Y519A) of the hydrophobic residues that are involved in dimerization. The triple point mutant eluted at the monomer position upon size-exclusion chromatography, while the negative control point mutant (Q460A) eluted at the same position as the wild-type. Notably, a single-residue mutation (L521A in interface 3) abolished the dimerization of SEL1L (Fig. 2B). Leu 521 is located in the dimerization center of the antiparallel 9B helices in the SEL1L dimer. Taken together, these structural and biochemical data demonstrate that SEL1L exists as a dimer in solution and that SLR motif 9 in SEL1L plays an important role in generating a two-fold dimerization interface. SLRs of mouse SEL1L were predicted using the TPRpred server23. Based on the prediction, full-length SEL1L contains a total of 11 SLR motifs, and our construct corresponds to SLR motifs 5 through 9. Although amino acid sequences from helix 9A and 9B correctly aligned with the regular SLR repeats and corresponded to SLR motif 9 (Fig. 3A), the structural arrangement of the two helices deviated from the common structure for the SLR motif. According to our crystal structure, the central axis of helix 9B is almost parallel to that of helix 9A (Fig. 3B). However, this unusual conformation of SLR motif 9 seems to be essential for dimer formation, as described earlier. For this structural geometry, two adjacent residues, Gly 512 and Gly 513, in SEL1L confer flexibility at this position by adopting main-chain dihedral angles that are disallowed for non-glycine residues. The phi and psi dihedrals are 100° and 20° for Gly 512, and 110° and −20° for Gly 513, respectively (Fig. 3C). Gly 513 is conserved among other SLR motifs in the SEL1L, but Gly 512 is present only in the SLR motif 9 of SEL1L (Fig. 3A). Thus, the Gly-Gly residues generate an unusual sharp bend at the C-terminal SLR motif 9. The involvement of a glycine residue in forming a hinge for domain swapping has been reported previously24. The significance of Gly 513 is further highlighted by its absolute conservation among different species, including the budding yeast homolog Hrd3p. To further investigate the importance of Gly 512 and Gly 513 in the unusual SLR motif geometry, we generated a point mutation (Gly to Ala), which restricts the flexibility. Although the Gly 512 and Gly 513 residues are closely surrounded by helix 9B from the counter protomer, there is enough space for the side chain of alanine, suggesting that no steric hindrance would be caused by the mutation (Fig. 3C). This means that the effect of the mutation is mainly to generate a more restricted geometry at the hinge region. G512A or G513A alone showed no differences from wild-type in terms of the size-exclusion chromatography elution profile (Fig. 3D), suggesting that the restriction for single glycine flexibility would not be enough to break the swapped structure of helix 9B. However, the double mutant (G512A/G513A) eluted over a broad range and much earlier than the wild-type, suggesting that mutation of the residues involved in the hinge linking helix 9A and 9B significantly affected the geometry of helix 9B in generating domain swapping, and eventually altered the overall oligomeric state of SEL1L into a polydisperse pattern (Fig. 3D, Supplementary Fig. 6). When the residues were mutated to lysine (G512K/G513K), the mutant not only restricted the geometry of residues at the hinge but also generated steric hindrance during interaction with the counter protomer of SEL1L, thereby inhibiting self-association of SEL1L completely. The G512K/G513K double mutant eluted at the monomer position in size-exclusion chromatography (Fig. 3D). A previous study shows that induction of steric hindrance by mutation destabilizes the dimerization interface of a different protein, ClC transporter25. Collectively, these data suggest that the Gly 512 and Gly 513 at the connection between helix 9A and 9B play a crucial role in forming the domain-swapped conformation that enables dimer formation. Next, we examined if SEL1L also forms self-oligomers in vivo using HEK293T cells. We generated full-length SEL1L-HA and SEL1L-FLAG fusion constructs and co-transfected the constructs into HEK293T cells. A co-immunoprecipitation assay using an anti-FLAG antibody followed by Western blot analysis using an anti-HA antibody showed that full-length SEL1L forms self-oligomers in vivo (Fig. 4A). To further examine whether the SEL1L domain is sufficient to physically interact with full-length SEL1L, we generated SEL1L and SLR motif 9 deletion (SEL1L) construct, which were fused to the C-terminus of SEL1L signal peptides. Co-immunoprecipitation analysis showed that the SEL1L was sufficient to physically interact with the full-length SEL1L, while SEL1L failed to do so (Fig. 4A). Interestingly, however, the expression level of SEL1L was consistently lower than that of SEL1L (Fig. 4A,B). Semi-quantitative RT-PCR revealed no significant difference in transcriptional levels of the two constructs (data not shown). We speculated that SEL1L could be secreted while the SEL1L is retained in the ER by association with the endogenous ERAD complex. Indeed, immunoprecipitation followed by western blot analysis using the culture medium detected secreted SEL1L fragment, but not SEL1L (Fig. 4B). We next examined if the reason why SEL1L failed to bind to the full-length SEL1L may be because of the lower level of SEL1L in the ER lumen compared to SEL1L fragment. In order to retain two SEL1L fragments in the ER lumen, we added KDEL ER retention sequence to the C-terminus of both fragments. Indeed, the addition of KDEL peptide increased the level of SEL1L in the ER lumen (Fig. 4D,E) and the immunostaining analysis showed both constructs were well localized to the ER (Fig. 4C). We further analyzed whether SEL1L may competitively inhibit the self-oligomerization of SEL1L in vivo. To this end, we co-transfected the differentially tagged full-length SEL1L (SEL1L-HA and SEL1L-FLAG) and increasing doses of SEL1L-KDEL, SEL1L-KDEL or SEL1L (L521A)-KDEL, respectively. Co-immunoprecipitation assay revealed that wild-type SEL1L-KDEL, indeed, competitively disrupted the self-association of the full-length SEL1L (Fig. 4E). In contrast, SEL1L-KDEL and the single-residue mutation L521A in SEL1L did not competitively inhibit the self-association of full-length SEL1L (Fig. 4E,F). These data suggest that the SEL1L forms self-oligomers and the oligomerization is mediated by the SEL1L domain in vivo. Previous studies reveal that TPRs and SLRs have similar consensus sequences, suggesting that their three-dimensional structures are also similar18. The superposition of isolated TPRs from Cdc23 (S. pombe, cell division cycle 23 homolog, PDB code 3ZN3) and SLRs from HcpC (Helicobacter Cysteine-rich Protein C, PDB code 1OUV) yields RMSDs below 1 Å, confirming that the isolated repeats are indeed similar20. This is relevant to SLR motifs in SEL1L, as isolated SLR motifs from SEL1L showed good structural alignment with isolated TPRs (RMSD 1.6 Å for all Cα chains) from Cdc23 and SLRs (RMSD 0.6 Å for all Cα chains) from HcpC (Fig. 5A). However, superimposing the structure of SLR motifs 5 to 9 from SEL1L onto the overall Cdc23 or full-length HcpC structures revealed that SLR motifs 5 to 9 in SEL1L have a different superhelical structure than either Cdc23 or HcpC (RMSD values of >2.5 Å for Cα atoms) (Fig. 5B). The differences may result from the differing numbers of residues in the loops and differences in antiparallel helix packing. Moreover, there are conserved disulfide bonds in the SLR motifs of HcpC and HcpB, but no such bonds are observed in SEL1L. These factors contribute to the differences in the overall conformation of the SLR motifs in SEL1L and other SLR or TPR motif-containing proteins. Another major difference in the structure of SLR motifs between SEL1L and HcpC is the oligomeric state of proteins. The TPR motif is involved in the dimerization of proteins such as Cdc23, Cdc16, and Cdc2726. In particular, the N-terminal domain of Cdc23 (Cdc23) has a TPR-motif organization similar to that of the SLR motif in SEL1L. The seven TPR motifs of Cdc23 are assembled into a superhelical structure, generating a hollow surface and encircling its dimer counterpart in an interlocking clasp-like arrangement (Fig. 5C)26. The TPR motif 1 (TPR1) of each Cdc23 subunit is located in the hollow surface of the counter subunit and interacts with residues lining the inner groove TPR α-helices, generating two-fold symmetry homotype interactions. However, in this structure, a conformational change in the TPR motif itself is not observed. Self-association of HcpC has not been reported, and there is no domain-swapped structure in the SLR motifs of HcpC, in contrast to that observed in SEL1L. Although SEL1L contains a number of SLR motifs comparable to HcpC, the SLR motifs in SEL1L are interrupted by other sequences, making three SLR motif clusters (Fig. 1A). The interrupted SLR motifs may be required for dimerization of SEL1L, as five SLR motifs are more than enough to form the semicircle of the yin-yang symbol (Fig. 1B). Helix 5A from SLR motif 5 meets helix 9A from SLR motif 9 of the counterpart SEL1L. If the SLR motifs 5 to 9 were not isolated from other SLR motifs, steric hindrance could interfere with dimerization of SEL1L. This is one of the biggest differences from TPRs in Cdc23 and from the SLRs in HcpC, where the motifs exist in tandem. TPR and SLR motifs are generally involved in protein-protein interaction modules, and the sequences between the SLR motifs of SEL1L might actually facilitate the self-association of this protein. Based on the structural data presented herein, a possible arrangement of membrane-associated ERAD components in mammals, highlighting the molecular functions of SLR domains in SEL1L, is shown in Fig. 6C. We suggest that the middle SLR domains are involved in the dimerization of SEL1L based on the crystal structure and biochemical data. SLR-C, which contains SLR motifs 10 to 11, might be involved in the interaction with HRD1. Indirect evidence from a previous yeast study shows that the circumscribed region of C-terminal Hrd3p, specifically residues 664–695, forms contacts with the Hrd1 luminal loops12. The Hrd3p residues 664–695 correspond to mouse SEL1L residues 696–727, which include the entire helix 11B (residue 697–709) of SLR motif 11 and a well-conserved adjacent region (Supplementary Fig. 4). This observation is supported by the following: (1) the meticulous range of SLR motif 10 to 11 is newly established from a structure-guided SLR motif alignment, based on the present structure study, and (2) the relatively high sequence conservation between mammalian SEL1L and yeast Hrd3p around SLR motifs 10 to 11, which contain contact regions with HRD1 (Hrd1p) (Supplementary Figs. 4 and 5). To address this hypothesis, we prepared constructs encoding mouse HRD1 luminal fragments fused to GST as shown in Fig. 6A, and tested their ability to bind certain SLR motifs in SEL1L. The fusion proteins were immobilized on glutathione-Sepharose beads and probed for binding to SLR-N, SLR-M, SLR-C, and monomer form of SLR-M (SLR-M). Figure 6B shows that the SLR-C, consisting of SLR motifs 10 and 11, exclusively interacts with N-terminal luminal loop (residues 21–42) of HRD1. The molecular functions of SLR-N are unclear. One possibility is that SLR-N contributes to substrate recognition of proteins to be degraded because there are a couple of putative glycosylation sites within the SLR-N domain (Fig. 1A). SEL1L contains a putative N-glycosylation site, Asn 427, which is highly conserved among different species and structurally exposed to the surface of the SEL1L dimer according to the crystal structure (Fig. 6C). Many reports demonstrate that membrane-bound ERAD machinery proteins in yeast, such as Hrd1p, Der1p, and Usa1p, are involved in oligomerization of ERAD components272829. The Hrd1p complex forms dimers upon sucrose gradient sedimentation530 and size-exclusion chromatography30. Previous data show that HA-epitope-tagged Hrd3p or Hrd1p efficiently co-precipitate with unmodified Hrd3p and Hrd1p, respectively, suggesting that both Hrd1p and Hrd3p homodimers are involved in self-association of the Hrd complex. Considering that the functional and structural composition of ERAD components are conserved in both yeast and mammals, we propose that the mammalian ERAD components also form self-associating oligomers. This hypothesis is supported by cross-linking data suggesting that human HRD1 forms a homodimer31. Consistent with the previous data, our crystal structure and biochemical data demonstrate that mouse SEL1L exists as a homodimer in the ER lumen via domain swapping of SLR motif 9. We need to further test whether there are contacts involved in dimer formation in SEL1L in addition to those in the SLR-M region. In yeast, Usa1p acts as a scaffold for Hrd1p and Der1p, in which the N-terminus of Usa1p interacts with the C-terminal 34 amino acids of Hrd1p in the cytosol to induce oligomerization of Hrd1p, which is essential for its activity3032. However, metazoans lack a clear Usa1p homolog. Although mammalian HERP has sequences and domains that are conserved in Usa1p, the molecular function of HERP is not clearly related to that of Usa1p431. Rather, recent research shows that a transiently expressed HRD1-SEL1L complex alone associates with the ERAD lectins OS9 or XTP-B and is sufficient to facilitate the retrotranslocation and degradation of the model ERAD substrate α-antitrypsin null Hong-Kong (NHK) and its variant, NHK-QQQ, which lacks the N-glycosylation sites33. Assuming that the correct oligomerization of ERAD components may be critical for their function, we hypothesize that homodimer formation of SEL1L in the ER lumen may stabilize oligomerization of the HRD complex, given that SEL1L forms a stoichiometric complex with HRD110131530. This is further supported by our data showing that the SLR-C of SEL1L directly interacts with the luminal fragment of HRD1 in the ER lumen. Although the organization of membrane-bound HRD complex components may be very similar between metazoans and yeast, the molecular details of interactions between the components may not necessarily be conserved. In yeast, it is unclear whether self-association of Hrd3p is due to SLR motifs because the sequence of Hrd3p does not align precisely with the SLR motifs in SEL1L18. Furthermore, we are uncertain whether self-association of Hrd3p contributes to formation of the active form of the Hrd1p complex. Recently, a truncated version of Yos9 was shown to form a dimer in the ER lumen and to contribute to the dimeric state of the Hrd1p complex34. This interaction seems to be weak because direct Yos9-Yos9 interactions were not detected in immunoprecipitation experiments from yeast cell extracts containing different epitope-tagged variants of Yos9. However, the dimerization of Yos9 could provide a higher stability for the Hrd1p complex oligomer. Likewise, the dimerization of SEL1L might provide stability for the mammalian HRD oligomer complex. Further cell biological studies are required to clarify whether SEL1L (Hrd3p) dimerization could be cooperative with the oligomerization of the HRD complex. Considering that it is very important for the function of the HRD complex that the components assemble as oligomers, we believe that the self-association of SEL1L strongly contributes to generating active forms of the HRD complex, even in the absence of Usa1p, in metazoans. These findings should provide a foundation for molecular-level studies to understand the membrane-associated HRD complex assembly in ERAD. The expression and purification of SEL1L was performed as described previously35. Crystals were grown using the hanging-drop vapor diffusion method at 4 °C. For crystallization of the M. musculus SEL1L, 1 μl of protein solution (in 25 mM Tris-HCl, 150 mM NaCl, and 5 mM DTT, pH 7.5) was equilibrated with 1 μl of well solution (30% isopropanol, 100 mM NaCl, 100 mM Tris, 5 mM DTT, and 20 mM phenol, pH 8.5). The crystals, which appeared after 4 days, contain two SEL1L dimers in the asymmetric unit (space group P21, a = 29.13, b = 110.52, c = 109.81 Å, α = 90.00, β = 90.61, γ = 90.00, 44% solvent). For X-ray diffraction experiments, crystals were transferred to well solution plus paraffin-oil, then flash frozen in liquid nitrogen. SAD data were collected with a Se-Met crystal at beamline 7A of the Pohang Accelerator Laboratory (PAL) and processed using HKL2000 software36. Native data (2.6 Å resolution) were collected from a single frozen crystal at the same beamline of PAL and were integrated and scaled as described above. The SAD data analysis was performed using Phenix software37 using data between 50 and 2.9 Å resolution. Phenix identified 31 of the 32 selenium sites and refined these to give a mean f.o.m. = 0.472. Electron density modification, including non-crystallographic symmetry (NCS) averaging, using the RESOLVE software38 yielded an initial electron density map of excellent quality. Model building and refinement were carried out with the Coot39 and Phenix programs, respectively. The final model was refined to an R factor of 20.7% (Rfree = 27.7%) for native data between 30 and 2.6 Å resolution (Table 1). The final model consisted of 5402 protein atoms and 47 water molecules. There were no outliers in a Ramachandran plot of the final model. The model contained four copies of SEL1L (residues 348–533) in the asymmetric unit. Of these, the following residues were not modeled due to weak electron densities: SEL1L residues 348–351, 420, 421, and 525–533 in the first copy; residues 348–351 and 525–533 in the second and third copies; and residues 348–352 and 525–533 in the fourth copy. The X-ray data and refinement statistics are summarized in Table 1. HEK293T cells were cultured in DMEM (Gibco) supplemented with 10% FBS. The mouse Sel1L gene was cloned into pCS108 and the 3 × HA or 3 × FLAG tag was fused to the C-terminus of SEL1L. The signal peptide from Xenopus Sel1L was cloned into pCS108 and the mouse SEL1L domain, SEL1L (348-497) fragments, and SEL1L (L521A) were fused to the C-terminus of the signal peptide. Then, a 3 × HA or a 3 × FLAG tag was fused to the C-terminus of the constructs. For the ER retention signal, the KDEL sequences were added to the C-terminus of the fragments. The plasmids were transfected using Lipofectamine 2000 (Life Technologies) according to the manufacturer’s manual. For western blot analysis, HEK293T cells were transfected with the indicated construct and harvested after washing in PBS. The cells were homogenized in lysis buffer (50 mM Tris, pH 7.4, 150 mM NaCl, 0.1% Triton X-100, 5% glycerol), supplemented with protease and phosphatase inhibitor cocktails. Homogenates were cleared by centrifugation at 13,200 rpm for 15 minutes at 4 °C. The lysates were subsequently used for either co-immunoprecipitation experiment or western blot analysis. For the western blot analysis, the samples were run onto 6–12% polyacrylamide gel. Blots were blocked in 5% TBS + 0.05% Tween 20 and incubated with anti-DDDD-K (Abcam) or anti-HA (Roche) antibodies. Proteins were visualized using HRP-conjugated secondary antibodies (1:4000) and SuperSignal West Pico Chemiluminescent Substrate or SuperSignal West Dura Extended Duration Substrate (Thermo) and exposed to ChemiDoc MP (Bio-Rad). For immunostaining, the cells were fixed in 4% formaldehyde and incubated with the indicated antibodies. The coverslips were incubated in blocking solution (10% FBS + 2% DMSO in TBS + 0.1% Triton X-100) at room temperature for 30 minutes to block non-specific binding. Fluorescent labeling was performed using Alexa Fluor 555 or 488-conjugated secondary antibodies and nuclei were stained with DAPI. The samples were mounted and confocal images were obtained using a Zeiss LSM700. For pull-down experiments, 400 μg of HRD1 luminal fragment GST-fusion proteins were incubated with 5 μl of a 50% (v/v) slurry of glutathione sepharose 4B beads (GE Healthcare) for 50 min at 4 °C. Beads were washed twice with buffer A (150 mM NaCl, 25 mM sodium phosphate pH 7.5, 5 mM DTT), and then mixed with 100 μg of MBP-SEL1L protein (SLR-N, SLR-M, SLR-C, and SLR-M) in buffer A, in a total assay volume of 500 μl. The assay mix was incubated at 4 °C for 15 minutes, and beads were washed twice with 500 μl buffer A. Proteins were eluted with SDS sample buffer, and analyzed by SDS-PAGE. Accession Numbers: The coordinates and structure factors have been deposited in the Protein Data Bank with the accession code of 5B26. How to cite this article: Jeong, H. et al. Crystal structure of SEL1L: Insight into the roles of SLR motifs in ERAD pathway. Sci. Rep. 6, 20261; doi: 10.1038/srep20261 (2016).
PMC4918766	Mechanism of extracellular ion exchange and binding-site occlusion in the sodium-calcium exchanger	Na/Ca exchangers utilize the Na electrochemical gradient across the plasma membrane to extrude intracellular Ca, and play a central role in Ca homeostasis. Here, we elucidate their mechanisms of extracellular ion recognition and exchange through a structural analysis of the exchanger from Methanococcus jannaschii (NCX_Mj) bound to Na, Ca or Sr in various occupancies and in an apo state. This analysis defines the binding mode and relative affinity of these ions, establishes the structural basis for the anticipated 3Na:1Ca exchange stoichiometry, and reveals the conformational changes at the onset of the alternating-access transport mechanism. An independent analysis of the dynamics and conformational free-energy landscape of NCX_Mj in different ion-occupancy states, based on enhanced-sampling molecular-dynamics simulations, demonstrates that the crystal structures reflect mechanistically relevant, interconverting conformations. These calculations also reveal the mechanism by which the outward-to-inward transition is controlled by the ion-occupancy state, thereby explaining the emergence of strictly-coupled Na/Ca antiport.Na/Ca exchangers (NCX) play physiologically essential roles in Ca signaling and homeostasis. NCX catalyzes the uphill extrusion of intracellular Ca across the cell membrane, by coupling this process to the downhill permeation of Na into the cell, with a 3 Na to 1 Ca stoichiometry. This reaction is, however, inherently reversible, its direction being dictated only by the transmembrane electrochemical ion gradients . The mechanism of NCX proteins is therefore highly likely to be consistent with the alternating-access model of secondary-active transport. The basic functional unit for ion transport in NCX consists of ten membrane-spanning segments, comprising two homologous halves. Each of these halves contains a highly conserved region, referred to as α-repeat, known to be important for ion binding and translocation; in eukaryotic NCX, the two halves are connected by a large intracellular regulatory domain, which is absent in microbial NCX (Supplementary Fig. 1). Despite a long history of physiological and functional studies, the molecular mechanism of NCX has been elusive, owing to the lack of structural information. Our recent atomic-resolution structure of NCX_Mj from Methanococcus jannaschii provided the first view of the basic functional unit of an NCX protein. This structure shows the exchanger in an outward-facing conformation and reveals four putative ion-binding sites, denominated internal (Sint), external (Sext), Ca-binding (SCa) and middle (Smid), clustered in the center of the protein and occluded from the solvent (Fig. 1a-b). With similar ion exchange properties to those of its eukaryotic counterparts, NCX_Mj provides a compelling model system to investigate the structural basis for the specificity, stoichiometry and mechanism of the ion-exchange reaction catalyzed by NCX. In this study, we set out to determine the structures of outward-facing wild-type NCX_Mj in complex with Na, Ca and Sr, at various concentrations. These structures reveal the mode of recognition of these ions, their relative affinities, and the mechanism of extracellular ion exchange, for a well-defined, functional conformation in a membrane-like environment. An independent analysis based on molecular-dynamics simulations demonstrates that the structures capture mechanistically relevant states. These calculations also reveal how the ion occupancy state of the outward-facing exchanger determines the feasibility of the transition to the inward-facing conformation, thereby addressing a key outstanding question in secondary-active transport, namely how the transported substrates control the alternating-access mechanism. The assignment of the four central binding sites identified in the previously reported NCX_Mj structure was hampered by the presence of both Na and Ca in the protein crystals. To conclusively clarify this assignment, we first set out to examine the Na occupancy of these sites without Ca. Crystals were grown in 150 mM NaCl using the lipidic cubic phase (LCP) technique. The crystallization solutions around the LCP droplets were then slowly replaced by solutions containing different concentrations of NaCl and EGTA (Methods). X-ray diffraction of these soaked crystals revealed a Na-dependent variation in the electron-density distribution at sites Sext, SCa and Sint, indicating a Na occupancy change (Fig. 1c). Occupancy refinement indicated two Na ions bind to Sint and SCa at low Na concentrations (Fig. 1c), with a slight preference for Sint (Table 1). Binding of a third Na to Sext occurs at higher concentrations, as no density was observed there at 10 mM Na or lower (Fig. 1c); Sext is however partially occupied at 20 mM Na, and fully occupied at 150 mM (Fig. 1c). The Na occupation at SCa, compounded with the expected 3Na:1Ca stoichiometry, implies our previous assignment of the Smid site must be re-evaluated. Indeed, two observations indicate that a water molecule rather than a Na ion occupies Smid, as was predicted in a recent simulation study. First, the electron density at Smid does not depend significantly on the Na concentration. Second, the protein coordination geometry at Smid is clearly suboptimal for Na (Supplementary Fig. 1d). The water molecule at Smid forms hydrogen-bonds with the highly conserved Glu54 and Glu213 (Supplementary Fig. 1d), stabilizing their orientation to properly coordinate multiple Na ions at Sext, SCa and Sint. It can be inferred from this assignment that Glu54 and Glu213 are ionized, while Asp240, which flanks Smid (and is replaced by Asn in eukaryotic NCX) would be protonated, as indicated by the abovementioned simulation study. The NCX_Mj structures in various Na concentrations also reveal that Na binding to Sext is coupled to a subtle but important conformational change (Fig. 2). When Na binds to Sext at high concentrations, the N-terminal half of TM7 is bent into two short helices, TM7a and TM7b (Fig. 2a). TM7b occludes the four central binding sites from the external solution, with the backbone carbonyl of Ala206 coordinating the Na ion (Fig. 2b-d). However, when Sext becomes empty at low Na concentrations, TM7a and TM7b become a continuous straight helix (Fig. 2a), and the carbonyl group of Ala206 retracts away (Fig. 2b-d). TM7a also forms hydrophobic contacts with the C-terminal half of TM6. These contacts are absent in the structure with Na at Sext, in which there is an open gap between the two helices (Fig. 2b). This difference is noteworthy because TM6 and TM1 are believed to undergo a sliding motion, relative to the rest of the protein, when the transporter switches to the inward-facing conformation. The straightening of TM7ab also opens up a passageway from the external solution to Sext and Smid, while SCa and Sint remain occluded (Fig. 2d). Thus, the structures at high and low Na concentrations represent the outward-facing occluded and partially open states, respectively. This conformational change is dependent on the Na occupancy of Sext and occurs when Na already occupies Sint and SCa. Our crystallographic titration experiment indicates that the K1/2 of this Na-driven conformational transition is ~20 mM. At this concentration, Sext is partially occupied and the NCX_Mj crystal is a mixture of both the occluded and partially open conformations. This structurally-derived Na affinity agrees well with the external Na concentration required for NCX activation in eukaryotes. The finding that the Na occupancy change from 2 to 3 ions coincides with a conformational change of the transporter also provides a rationale to the Hill coefficient of the Na-dependent activation process in eukaryotic NCX. To determine how Ca binds to NCX_Mj and competes with Na, we first titrated the crystals with Sr (Methods). Sr is transported by NCX similarly to Ca, and is distinguishable from Na by its greater electron-density intensity. Protein crystals soaked with 10 mM Sr and 2.5 mM Na revealed a strong electron-density peak at site SCa, indicating binding of a single Sr ion (Fig. 3a). The Sr-loaded NCX_Mj structure adopts the partially open conformation observed at low Na concentrations. Binding of Sr, however, excludes Na entirely. Crystal titrations with decreasing Sr or increasing Na demonstrated that Sr binds to the outward-facing NCX_Mj with low affinity, and that it can be out-competed by Na even at low concentrations (Supplementary Note 1 and Supplementary Fig. 2a-b). Thus, in 100 mM Na and 10 mM Sr, Na completely replaced Sr (Fig. 3a) and reverted NCX_Mj to the Na-loaded, fully occluded state. Similar titration experiments showed that Ca and Sr binding to NCX_Mj are not exactly alike The electron density distribution from crystals soaked in high Ca and low Na, indicates that Ca can bind to Smid as well as SCa, with a preference for SCa (Fig. 3b). Binding of Ca to both sites simultaneously is highly improbable due to their close proximity, and at least one water molecule can be discerned coordinating the ion (Fig. 3b). The partial Ca occupancy at Smid is likely caused by Asp240, which flanks this site and can in principle coordinate Ca. Previous functional and computational studies, however, indicate Asp240 becomes protonated during transport. Indeed, in most NCX proteins Asp240 is substituted by Asn, which would likely weaken or abrogate Ca binding to Smid. SCa is therefore the functional Ca site. Similarly to Sr, Ca binds with low affinity to outward-facing NCX_Mj and can be readily displaced by Na (Supplementary Note 1 and Supplementary Fig. 2c). This finding is consistent with physiological and biochemical data for both eukaryotic NCX and NCX_Mj indicating that the apparent Ca affinity is much lower on the extracellular than the cytoplasmic side. Specifically, our crystallographic titration assay indicates Ca binds with sub-millimolar affinity, in good agreement with the external apparent Ca affinities deduced functionally for cardiac NCX (Km ~ 0.32 mM) and NCX_Mj (Km ~ 0.175 mM). Taken together, these crystal titration experiments demonstrate that the four binding sites in outward-facing NCX_Mj exhibit different specificity: Sint and Sext are Na specific whereas SCa, previously hypothesized to be Ca specific, can also bind Na, confirming our earlier simulation study, as well as Sr; Smid can also transiently accommodate Ca but during transport Smid is most likely occupied by water. The ion-binding sites in NCX_Mj can therefore accommodate up to three Na ions or a single divalent ion, and occupancy by Na and Ca (or Sr) are mutually exclusive, as was deduced for eukaryotic exchangers. An apo state of outward-facing NCX_Mj is likely to exist transiently in physiological conditions, despite the high amounts of extracellular Na (~150 mM) and Ca (~2 mM). We were able to determine an apo-state structure of NCX_Mj, by crystallizing the protein at lower pH and in the absence of Na (Methods). This structure is similar to the partially open structure with two Na or either one Ca or one Sr ion, with two noticeable differences. First, TM7ab along with the extracellular half of the TM6 and TM1 swing further away from the protein core (Fig. 3c), resulting in a slightly wider passageway into the binding sites. Second, Glu54 and Glu213 side chains rotate away from the binding sites and appear to form hydrogen-bonds with residues involved in ion coordination in the fully Na-loaded structure (Fig. 3d). Although the binding sites are thus fully accessible to the external solution (Fig. 3e), the lack of electron density therein indicates no ions or ordered solvent molecules. This apo structure might therefore represent the unloaded, open state of outward-facing NCX_Mj. Alternatively, this structure might capture a fully protonated state of the transporter, to which Na and Ca cannot bind. Such interpretation would be consistent with the computer simulations reported below. Indeed, transport assays of NCX_Mj have shown that even in the presence of Na or Ca, low pH inactivates the transport cycle. That secondary-active transporters are able to harness an electrochemical gradient of one substrate to power the uphill transport of another relies on a seemingly simple principle: they must not transition between outward- and inward-open conformations unless in two precise substrate occupancy states. NCX must be loaded either with 3 Na or 1 Ca, and therefore functions as an antiporter; symporters, by contrast, undergo the alternating-access transition only when all substrates and coupling ions are concurrently bound, or in the apo state. The reason why only specific occupancy states permit this transition in a given system, thereby determining its biological function, remains unclear. To examine this central question, we sought to characterize the conformational free-energy landscape of NCX_Mj and to examine its dependence on the ion-occupancy state, using molecular dynamics (MD) simulations. This computational analysis was based solely on the published structure of NCX_Mj, independently of the crystallographic studies described above. As it happens, the results confirm that the structures now available are representing interconverting states of the functional cycle of NCX_Mj, while revealing how the alternating-access mechanism is controlled by the ion-occupancy state. A series of exploratory MD simulations was initially carried out to examine what features of the NCX_Mj structure might depend on the ion-binding sites occupancy. Specifically, we first simulated the outward-occluded form, in the ion configuration we previously predicted, now confirmed by the high-Na crystal structure described above (Fig. 1b). That is, Na ions occupy Sext, SCa, and Sint, while D240 is protonated and a water molecule occupies Smid. The Na ion at Sext was then relocated from the site to the bulk solution (Methods), and this system was then allowed to evolve freely in time. The Na ions at SCa and Sint were displaced subsequently, and an analogous simulation was then carried out. These initial simulations revealed noticeable changes in the transporter, consistent with those observed in the new crystal structures. The most notable change upon displacement of Na from Sext was the straightening of TM7ab (Fig. 4a). When 3 Na ions are bound, TM7ab primarily folds as two distinct, non-collinear α-helical fragments, owing to the loss of the backbone carbonyl-amide hydrogen-bonds between F202 and A206, and T203 and F207 (Fig. 4b). This distortion occludes Sext from the exterior (Fig. 4d, 4h-i) and appears to be induced by the Na ion itself, which pulls the carbonyl group of A206 into its coordination sphere (Fig. 4g). With Sext empty, however, TM7ab forms a canonical α-helix (Fig. 4a-b, 4g), thereby creating an opening between TM3 and TM7, which in turn allows water molecules from the external solution to reach into Sext (Fig. 4e, 4h-i), i.e. the transporter is no longer occluded. Displacement of Na from SCa and Sint induces further changes (Fig. 4c). The most noticeable is an increased separation between TM7 and TM2 (Fig. 4f), previously brought together by concurrent backbone interactions with the Na ion at SCa (Fig. 4d-e). TM1 and TM6 also slide further towards the membrane center, relative to the outward-occluded state (Fig. 4c). Together, these changes open a second aqueous channel leading directly into SCa and Sint (Fig. 4f, Fig. 4h-i). The transporter thus becomes fully outward-open. To more rigorously characterize the influence of the ion-occupancy state on the conformational dynamics of the exchanger, we carried out a series of enhanced-sampling MD calculations designed to reversibly simulate the transition between the outward-occluded and fully outward-open states, and thus quantify the free-energy landscape encompassing these states (Methods). As above, we initially examined three occupancy states, namely with Na in Sext, SCa and Sint, with Na only at SCa and Sint, and without Na. These calculations demonstrate that the Na occupancy state of the transporter has a profound effect on its conformational free-energy landscape. When all Na sites are occupied, the global free-energy minimum corresponds to a conformation in which the ions are maximally coordinated by the protein (Fig. 5a, 5c); TM7ab is bent and packs closely with TM2 and TM3, and so the binding sites are occluded from the solvent (Fig. 5b). At a small energetic cost, however, the transporter can adopt a metastable ‘half-open’ conformation in which TM7ab is completely straight and Sext is open to the exterior (Fig. 5a, 5b). The Na ion at Sext remains fully coordinated, but an ordered water molecule now mediates its interaction with A206:O, relieving the strain on the F202:O–A206:N hydrogen-bond (Fig. 5c). This semi-open conformation is nearly identical to that found to be the most probable when Na occupies only SCa and Sint (2 × Na, Fig. 5a), demonstrating that binding (or release) of Na to Sext occurs in this metastable conformation. Interestingly, this doubly occupied state can also access conformations in which the second aqueous channel mentioned above, i.e. leading to SCa between TM7 and TM2 and over the gating helices TM1 and TM6, also becomes open (Fig. 5b-c). Crucially, though, the free-energy landscape for this partially occupied state demonstrates that the occluded conformation is no longer energetically feasible (Fig. 5a). Displacement of the two remaining Na ions from SCa and Sint further reshapes the free-energy landscape of the transporter (No ions, Fig. 5a), which now can only adopt a fully open state featuring the two aqueous channels (Fig. 5b-c). The transition to the occluded state in this apo state is again energetically unfeasible. From a mechanistic standpoint, it is satisfying to observe how the open and semi-open states are each compatible with two different Na occupancies, explaining how sequential Na binding to energetically accessible conformations (prior to those binding events) progressively reshape the free-energy landscape of the transporter; by contrast, the occluded conformation is forbidden unless the Na occupancy is complete. This processivity is logical since three Na ions are involved, but also implies that in the Ca-bound state, which includes a single ion, the transporter ought to be able to access all three major conformations, i.e. the outward-open state, in order to release (or re-bind) Ca, but also the occluded conformation, and thus the semi-open intermediate, in order to transition to the inward-open state. By contrast, occupancy by H, which as mentioned are not transported, might be compatible with a semi-open state as well as with the fully open conformation, but should not be conducive to occlusion. To assess this hypothesis, we carried out enhanced-sampling simulations for the Ca and H-bound states of outward-facing NCX_Mj analogous to those described above for Na (see Supplementary Note 2 and Supplementary Fig. 3-4 for details on how the structures of the Ca-bound state was predicted). The calculated free-energy landscape for Ca-bound NCX_Mj confirms the hypothesis outlined above (1 × Ca, Fig. 6a): consistent with the fact that NCX_Mj transports a single Ca, the occluded, dehydrated conformation is one of the major energetic minima, but clearly the exchanger can also adopt the semi-open and open states that would be required for Ca release and Na entry, via either of the aqueous access channels that lead to Sext and SCa (Fig. 6b-c). By contrast, protonation of Glu54 and Glu213 makes the occluded conformation energetically unfeasible, consistent with the fact that NCX_Mj does not transport protons; in this H-bound state, though, the exchanger can adopt the semi-open conformation captured in the low pH, apo crystal structure (2 × H, Fig. 6a-c). Taken together, this systematic computational analysis of outward-facing NCX_Mj clearly demonstrates that the alternating-access and ion-recognition mechanisms in this Na/Ca exchanger are coupled through the influence that the bound ions have on the free-energy landscape of the protein, which in turn determines whether or not the occluded conformation is energetically feasible. This occluded conformation, which is a necessary intermediate between the outward and inward-open states, and which entails the internal dehydration of the protein, is only attainable upon complete occupancy of the binding sites. The alternating-access hypothesis implicitly dictates that the switch between outward- and inward-open conformations of a given secondary-active transporter must not occur unless the appropriate type and number of substrates are recognized. This control mechanism is functionally crucial, as it precludes the backflow of the species that is transported uphill, and also prevents the dissipation of the driving electrochemical gradients. It is however also non-trivial: antiporters, for example, do not undergo the alternating-access transition without a cargo, but this is precisely how membrane symporters reset their transport cycles. Similarly puzzling is that a given antiporter will undergo this transition upon recognition of substrates of different charge, size and number. Yet, when multiple species are to be co-translocated, by either an antiporter or a symporter, partial occupancies must not be conducive to the alternating-access switch. Here, we have provided novel insights into this intriguing mechanism of conformational control through structural studies and quantitative molecular simulations of a Na/Ca exchanger. Specifically, our studies of NCX_Mj reveal the mechanism of forward ion exchange (Fig. 7). The internal symmetry of outward-facing NCX_Mj and the inward-facing crystal structures of several Ca/H exchangers indicate that the alternating-access mechanism of NCX proteins entails a sliding motion of TM1 and TM6 relative to the rest of the transporter. Here, we demonstrate that conformational changes in the extracellular region of the TM2-TM3 and TM7-TM8 bundle precede and are necessary for the transition, and are associated with ion recognition and/or release. The most apparent of these changes involves the N-terminal half of TM7 (TM7ab); together with more subtle displacements in TM2 and TM3, this change in TM7ab correlates with the opening and closing of two distinct aqueous channels leading into the ion-binding sites from the extracellular solution. Interestingly, the bending of TM7 associated with the occlusion of the ion-binding sites also unlocks its interaction with TM6, and thus enables TM6 and TM1 to freely slide to the inward-facing conformation. We anticipate that the intracellular ion-exchange process involves analogous conformational changes. The crystal structures of NCX_Mj reported here, with either Na, Ca, Sr or H bound, capture the exchanger in different conformational states. These states can only represent a subset among all possible, but they ought to reflect inherent preferences of the transporter, modulated by the experimental conditions. For example, in the crystal of NCX_Mj in LCP, the extracellular half of the gating helices (TM6 and TM1) form a lattice contact, which might ultimately restrict the degree of opening of the ion-binding sites in some cases (e.g. in the apo, low pH structure). Nonetheless, the calculated free-energy landscapes, derived without knowledge of the experimental data, reassuringly confirm that the crystallized structures correspond to mechanistically relevant, interconverting states. The simulations also demonstrate how this landscape is drastically re-shaped upon each ion-binding event. Indeed, we show that it is the presence or absence of the occluded state in this landscape that explains the antiport function of NCX_Mj and its 3Na:1Ca stoichiometry. We posit that a similar principle might govern the alternating-access mechanism in other transporters; that is, we anticipate that for both symporters and antiporters, it is the feasibility of the occluded state, encoded in the protein conformational free-energy landscape and its dependence on substrate binding, that ultimately explains their specific coupling mechanisms. In multiple ways, our findings provide an explanation for, existing functional, biochemical and biophysical data for both NCX_Mj and its eukaryotic homologues. The striking quantitative agreement between the ion-binding affinities inferred from our crystallographic titrations and the Km and K1/2 values previously deduced from functional assays has been discussed above. Consistent with that finding, mutations that have been shown to inactivate or diminish the transport activity of NCX_Mj and cardiac NCX perfectly map to the first ion-coordination shell in our NCX_Mj structures (Supplementary Fig. 4c-d). The crystallographic data also provides the long-sought structural basis for the ‘two-site’ model proposed to describe competitive cation binding in eukaryotic NCX, underscoring the relevance of these studies of NCX_Mj as a prototypical Na/Ca exchanger. Specifically, our crystal titrations suggest that, during forward Na/Ca exchange, sites Sint and SCa, which Ca and Na compete for, can be grouped into one; Na binding to these sites does not require high Na concentrations, and two Na ions along with a water molecule (at Smid) are sufficient to displace Ca, explaining the Hill coefficient of ~2 for Na-dependent inhibition of Ca fluxes. The Sext site, by contrast, might be thought as an activation site for inward Na translocation, since this is where the third Na ion binds at high Na concentration, enabling the transition to the occluded state. Interestingly, binding of Ca to Smid appears to be also possible, but available evidence indicates that this event transiently blocks the exchange cycle. Indeed, structures of NCX_Mj bound to Cd or Mn, both of which inhibit transport, show these ions at Smid; by contrast, Sr binds only to SCa, and accordingly, is transported by NCX similarly to calcium. Lastly, our theory that occlusion of NCX_Mj is selectively induced upon Ca or Na recognition is consonant with a recent analysis of the rate of hydrogen-deuterium exchange (HDX) in NCX_Mj, in the presence or absence of these ions, in conditions that favor outward-facing conformations. Specifically, saturating amounts of Ca or Na resulted in a noticeable slowdown in the HDX rate for extracellular portions of the α-repeat helices. We interpret these observations as reflecting that the solvent accessibility of the protein interior is diminished upon ion recognition, consistent with our finding that opening and closing of extracellular aqueous pathways to the ion-binding sites depend on ion occupancy state. In addition, the increased compactness of the protein tertiary structure in the occluded state would also slow down the dynamics of the secondary-structure elements, and thus further reduce the HDX rate. Our data would also explain the observation that the reduction in the HDX rate is comparable for Na and Ca, as well as the finding that the degree of deuterium incorporation remains non-negligible even under saturating ion concentrations. As the calculated free-energy landscapes show, Na and Ca induce the occlusion of the transporter in a comparable manner, and yet the ion-bound states retain the ability to explore conformations that are partially or fully open to the extracellular solution, precisely so as to be able to unload and re-load the substrates. NCX_Mj was expressed, purified and crystallized as previously described. Briefly, the NCX_Mj gene with a C-terminal hexa-histidine tag was subcloned into the pQE60 vector and expressed in Escherichia coli BL21(DE3)plysS. Harvested cells were homogenized and incubated in buffer containing 50 mM HEPES pH 7.2, 50 mM NaCl, 12 mM KCl, 10 mM CaCl2, 40 mM DDM. After incubation at room temperature (RT) for 3.5 hours, the supernatant was collected by centrifugation and loaded onto a Talon Co affinity column (Clontech). The non-specifically bound contaminates on the column were washed with buffer containing 50 mM HEPES pH 7.2, 50 mM NaCl, 12 mM KCl, 10 mM CaCl2, 15 mM imidazole, and 1 mM DDM. The bound NCX_Mj was eluted by increasing the imidazole concentration to 300 mM. The eluate was treated with thrombin to remove the hexa-histidine tag and dialyzed against 20 mM HEPES pH 7.2, 50 mM NaCl, 12 mM KCl, 10 mM CaCl2, and 1 mM DDM at RT overnight. After overnight digestion the sample was loaded onto a second Co affinity column to remove any free hexa-histidine tag and contaminant proteins. NCX_Mj in the flow-through was collected and further purified by gel filtration using a Superdex-200 (10/300) column (GE Healthcare) in 20 mM HEPES pH 7.2, 50 mM NaCl, 12 mM KCl, 10 mM CaCl2 and 0.5 mM DDM. The purified protein was then concentrated to 40 mg/ml for crystallization. Native NCX_Mj was crystallized using the lipidic cubic phase (LCP) technique, as previously described. Concentrated NCX_Mj was first reconstituted into 1-oleoyl-rac-glycerol (Sigma) in a protein:lipid weight ratio of 1:1.5, using the two-syringe method. Protein-laden LCP droplets of 35 nL were dispensed onto Corning 96-well protein-crystallization plates and overlaid with 5 μL of precipitant solution containing 40-42% PEG 400, 100 mM MES pH 6.5, 100 mM NaAc. Crystals were observed after 48 hours and grew to full size after 2 weeks. The native crystals belong to space group P212121 with a cell dimension of a=49.5Å, b=72.9Å and c=96.2Å, and contain one subunit per asymmetric unit. As the LCP droplet accounts for less than 1% of the total crystallization volume, the salt composition in the crystallization condition was determined mainly by the overlaying solution, and estimated to have 150 mM Na (from MES buffer and NaAc) and 30 μM Ca (from LCP droplet). In these concentration conditions (high Na and low Ca) Ca does not bind to NCX_Mj (as shown in our crystallographic titration experiments) and thus this native crystal structure represents NCX_Mj in 150 mM Na. The native crystals were used in all subsequent titration experiments to define low-Na, Ca and Sr-loaded structures. To obtain the apo crystal form, the protein was first purified in a solution containing 20 mM Hepes-Tris pH 7.2, 100 mM NMDG, 10 mM CaCl2 and 0.5 mM DDM. The crystals were obtained in LCP with crystallization solution containing 200 mM KAc, pH 4.0, 35% PEG400. The apo NCX_Mj crystals belong to space group C2 with a cell dimension of a=164.2Å, b=46.8Å, c=97.0 Å and β=106.2°, and contain two protein subunits per asymmetric unit. Once the native crystals reached their full size, the crystallization solutions overlaying lipid/protein droplets were gradually replaced by titration solutions through multiple steps of solution exchange. In general, 2-3 μL of existing crystallization solutions (normally in 5 μL) were replaced by the same volume of titration solutions, followed by overnight equilibration. The same procedures were repeated 6-10 times until the ion components in the crystal drops reached the targeted concentrations. For titration experiments to define concentration-dependent Na binding, the titration solutions contained 100 mM MES-Tris pH 6.5, 44% PEG400, 10 mM EGTA and a 100 mM mixture of NaAc and CsAc, in the following proportions: 100 mM CsAc; 90 mM CsAc and 10 mM NaAc; 80 mM CsAc and 20 mM NaAc; and 100 mM NaAc. Note that Cs does not bind NCX proteins and is commonly used as a Na substituent to maintain the ionic strength of the solutions. As complete removal of Na would deteriorate the crystals, we had to maintain a minimum Na concentration of about 2.5 mM in the crystal drops. The final Na concentrations in this set of titration experiments were about 2.5, 10, 20 and 100 mM, respectively. It is worth noting that the observed Na-dependent conformational change occurs while the proteins are in crystal form and embedded in lipid. In the titration experiments carried out to define the mode of divalent cation binding and competition with Na, the soaking solutions contained 100 mM MES-Tris pH 6.5, 44% PEG400, 100 mM mixture of CsAc and NaAc and various concentrations of XCl2, where X=Ca or Sr, in the following proportions: 100 mM CsAc and 10 mM XCl2; 100 mM CsAc and 1 mM XCl2; 100 mM CsAc and 0.1 mM XCl2; 90 mM CsAc, 10mM NaAc and 10mM XCl2; and 100 mM NaAc and 10 mM XCl2. After multiple steps of solution exchanges, the final soaking conditions contained 0.1, 1, or 10 mM of X together with 2.5 mM Na; or 10 mM X together with 2.5, 10 or 100 mM Na. After soaking crystals were mounted on 100-μm Mitegen Microloops and frozen in liquid nitrogen. All diffraction data were collected at the Advanced Photon Source (APS) GM/CA-CAT beamlines 23ID-B or 23ID-D using a beam size of 35 μm × 50 μm. Data were processed and scaled using HKL2000 and the structures were determined by molecular replacement in PHASER using our previously published NCX_Mj structure (PDB code 3V5U) as a search model. Model building was completed using COOT and structure refinement was performed with PHENIX. The data sets from crystals soaked in solutions containing 2.5 to 100 mM Na were collected using an X-ray wavelength of 1.033Å; the crystal grown with 150 mM Na, and those soaked with Ca and Sr solutions, were obtained with a wavelength of 0.9793 Å. Lastly, the data from the crystal grown at low pH with no Na or Ca were collected with a 2.0-Å wavelength beam. The resulting statistics for data collection and refinement are shown in Tables 2-4. All structure figures were prepared in PyMOL (The PyMOL Molecular Graphics System, Version 1.5.0.4 Schrödinger, LLC.). The ion passageways in low- and high-Na structures as well as the apo state were analyzed using the program CAVER. Due to the variation in diffraction resolution and intensity among crystals, ion-occupancy comparisons were made on the basis of the diffraction data obtained in the titration experiments scaled against a common reference data before map calculation. The NCX_Mj crystal obtained with 2.5 mm Na only was used as the reference. Conventional (i.e. not enhanced) MD simulations were carried out with NAMD 2.7-2.9 at constant temperature (298 K), pressure (1 atm), and membrane surface area (~69 Å per lipid),and with periodic boundaries in all directions. All calculations used the standard CHARMM27/CMAP force field, except for NBFIX corrections for the interaction between carboxylate-oxygens and Na interaction or Ca (Supplementary Note 3, Supplementary Fig. 5). Electrostatic interactions were calculated using PME with a real-space cut-off of 12 Å; the same cut-off distance was used for all van der Waals interactions. Five ion-occupancy states of the transporter were considered, namely with 3 Na, 2 Na, 2 H or 1 Ca, and with no ions bound; in all cases Asp240 is protonated. For the 3×Na state, we reanalyzed a 200-ns trajectory of NCX_Mj reported previously. NCX_Mj had been embedded in a POPC lipid membrane using GRIFFIN. The initial configuration of the 2×Na state was generated from an equilibrated configuration of 3×Na state, from which the Na ion at Sext was displaced by means of a slow alchemical transformation that annihilates the bound ion and recreates it in the bulk solution (in the same simulation box). The resulting 2×Na state was simulated for 250 ns. Similarly, the state with no Na bound was generated from an equilibrated configuration of the 2×Na state, from which the remaining Na ions were displaced; this state was again simulated for 250 ns. For the 2×H state, an initial configuration was generated from an equilibrated configuration of the 3×Na state, by gradually annihilating the Na ions from the binding sites and creating protonated E54 and E213 side chains; concurrently, acetic acid molecules in the bulk solution (in the same simulation box) were deprotonated and Na ions introduced. A second initial configuration of the 2×H state was obtained from an equilibrated configuration of the simulation with no ions bound, by slowly transforming deprotonated E54 and E213 into their protonated form, while doing the opposite to acetic acid molecules in the bulk solution. These two initial configurations of the 2×H state were then equilibrated for 800 ns. All annihilation/creation simulations were carried out using the FEP module of NAMD; and comprised 32-50 intermediate simulations of 400 ps each, for each transformation. A soft-core van der Waals potential with a radius-shifting coefficient of 2 Å was used. The annihilated Na ions were confined within their corresponding binding sites using flat-bottom distance restraints. Specifically, the Na in Sext was concurrently maintained within 4 Å of E54:Cδ, A206:C, S77:Cβ, T209:Cβ and S210:Cβ. The Na ions in SCa and Sint were concurrently kept within 4 Å of the E213:Cδ and A47:C, respectively. The Na ions and acetic acid molecules in the bulk solution were kept at a distance greater than 37 Å from the membrane center. Finally, the initial configuration of the Ca state was generated on the basis of the published NCX_Mj X-ray structure by placing Ca in the SCa site and two water molecules coordinating Ca at and near the Smid site, so as to satisfy the expected coordination geometry (see Supplementary Note 2, Supplementary Fig. 3-4). This configuration was initially equilibrated through a series of simulations in which RMSD-based restraints of gradually diminishing strength were applied to the protein Cα atoms as well the side-chains involved in Ca coordination. A 250 ns equilibration was then carried without any restraints. Free-energy landscapes were calculated using Bias-Exchange Well-Tempered Metadynamics (BE-WT-MetaD), using GROMACS4.5.5/PLUMED. The force-field and simulation conditions were equivalent to those employed in the unbiased MD simulations. The accumulated simulation time for each of the ion-occupancy states studied was 1.6 μs. Each of these calculations consisted of 16 concurrent, interdependent simulations (or replicas); in 15 of these replicas, a WT-MetaD biasing potential was applied on different subsets of collective variables, as specified below, while the remaining replica was unbiased. Attempts to exchange coordinate configurations among replicas were made every 2-5 ps, using the Metropolis criterion. The inputs for each calculation were equilibrated configurations extracted from the unbiased MD simulations. The choice of collective variables to be biased in the BE-WT-MetaD simulations was also based on analysis of the unbiased MD trajectories (Fig. 4g-i). Specifically, to enhance the reversible opening and closing of the water channels reaching from the extracellular bulk solution into either the Sext or the SCa binding sites, we employed the following time-dependent collective variable (Fig. 4h-i): (Eq. 1)V1,2(t)=βln∑iexp where ri denotes the distance between the oxygen atom of each water molecule in the system i and the center of the binding site considered (V1 for Sext,V2 for SCa), and β is 10-100 nm. When the binding site was occupied, the ion was used to define its center. If the site was empty, its center was defined as the center-of-mass of the oxygen atoms coordinating the ion if bound. Bound water molecules at or near Smid (coordinating the Na or Ca ions) were not considered. To enhance the reversible formation and disruption of selected backbone hydrogen bonds in TM7ab (Fig. 4g), we used an analogous collective variable: (Eq. 2)V3(t)=βln∑ijexp In this case, the index i denotes atoms P202:O and T203:O, while the index j denotes atoms A206:N and P207:N. To preclude the artificial unraveling of TM7ab driven by this bias, an upper-bound V3 equal to 0.7 nm was imposed with a boundary potential of the form k (V3(t) – V3) if V3(t) > V3, where k = 10 nm kJ/mol. In addition, to control the bending and straightening of TM7ab more globally, we used the following path-collective variables: (Eq. 3)V4(t)=exp+2expexp+exp(Eq. 4)V5(t)=−1λln[exp+exp] where d1 and d2 denote the mean-square-differences between the conformation of TM7ab and either the straight or bent conformations, respectively, and λ = 100 nm. Note that V 4 is by definition confined between a lower (V4 ~ 1) and upper bound (V4 ~ 2); to confine the exploration of V5, an upper value V5 of 0.020-0.025 nm was imposed with boundary potential of the form k (V5(t) – V5) if V5(t) > V5, where k = 10 nm kJ/mol. The mean-square-differences d1 and d2 comprise the backbone atoms of residues 198-211 as well as the side chain carbon atoms of residues P207 and L211. A boundary potential was also applied to confine the ions and water molecule bound to Sext, SCa, Sint and Smid to their corresponding binding sites. Specifically, the variable confined was: (Eq. 5)Vc(t)=∑i1−[ri(t)∕r0]81−[ri(t)∕r0]10 where ri denotes the distance between the ion and each of its coordinating oxygen atoms; r0 was set to 0.24 nm for the Na ions, and to 0.30 nm for Ca. Note that the upper-bound value of Vc is, by definition, approximately the coordination number in the bound state, whereas Vc becomes 0 as the ion becomes unbound. For the Na ion bound to Sext and Sint, therefore, a lower bound value Vc = 4.3 was imposed with potential of the form k (Vc(t) – Vc) when Vc(t) < Vc, where k = 2500 kJ/mol. An analogous restraint was used for the Na ion at SCa, with Vc = 4.75. Similarly, for the Ca ion at SCa a lower bound value Vc = 7.4 was imposed with a potential of the form k (Vc(t) – Vc) when Vc(t) < Vc, with k = 400 kJ/mol. Note that these restraints do not perturb the chemical structure of the ion-coordination sphere when the ion is bound, i.e. Vc(t) > Vc. The displacement of the bound water molecules in the ion-coordination sphere by equivalent water molecules in the solvent was prevented similarly. The specific sets of collective variables biased in each of the replica WT-MetaD simulations, as well as further details on the biasing potentials introduced, are specified in Supplementary Table 1. To translate the data gathered in the BE-WT-MetaD simulations into conformational free-energy landscapes, we sought to identify a low-dimensional representation of the data that is nevertheless also intuitive and representative. We ultimately settled on two structure-based descriptors of the degree of opening of the each of the aqueous channels leading to the ion-binding sites (Fig. 4), defined as: (Eq. 6)S1,2=βln∑ijexp where rij denotes a set of pairwise distances for specific Cα atoms in the protein, for a given simulation snapshot. For S1, index i refers to the Cα of residues 198 to 209 in TM7, while index j refers to those in residues 66 to 80 and 290 to 297, in TM3 and TM10, respectively. For S2, index i refers to the Cα of residues 51 to 64 in TM2, while index j refers to those in residues 177 to 193 and 198 to 209, in TM6 and TM7, respectively. Therefore, S1 describes the effective separation between TM7 and TM3/TM10, on the extracellular half of the protein, and thus reports on the accessibility to the Sext site. Analogously, S2 measures the separation between TM2 and TM6-TM7, also on the extracellular side, and thus reports on the accessibility to SCa. The conformational free energy of NCX_Mj as a function of S1 and S2 was then computed for each ion-occupancy state separately (Figs. 5a, 6a). These landscapes were obtained through reweighting of the biased probability distribution from the BE-WT-MetaD sampling, using the WHAM method; through this approach we combine the statistics gathered in all replicas, and can consider alternate free-energy projections. To correct the landscape calculated for the Ca-bound state (Fig. 6a) on account of the excess amount of charge transferred from the ion to the protein (Supplementary Note 4, Supplementary Fig. 6), we reprocessed all the sampling obtained during the original BE-WT-MetaD simulations, introducing in the WHAM equations a re-weighting factor w for each configuration X: (Eq. 7)w(X)∝exp where U denotes the ‘uncorrected’ CHARMM27/NBFIX potential-energy function, and Uc denotes the corrected function. To calculate Uc (X), the charge of Ca was reduced to +1.8e from its standard value of +2e, and the difference was distributed among the surrounding protein residues (as specified in Supplementary Note 4). To minimally alter the charge-distribution used in the original CHARMM27 force-field, the charge added to each protein atom was proportional to the absolute value of its uncorrected charge. The statistical errors for all free-energy landscapes are provided in Supplementary Fig. 7. Representative structures and water-density iso-surfaces (Fig. 5b-c, Fig. 6b-c) were derived for each BE-WT-MetaD simulation by clustering all sampling in the multi-dimensional space of V1, V2, and V4 (Eq. 1-3), plus a descriptor S3 of the proximity between TM6-TM7 and TM2-TM3/TM10, on the extracellular side of the protein. More precisely: (Eq. 8)S3=∑ij1−[rij(t)∕r0]81−[rij(t)∕r0]12 where r0 = 7.5 Å, and rij denotes a specific set of pairwise Cα distances, for a given simulation snapshot. Similarly to S1 and S2 (Eq. 6), index i refers to Cα atoms in the extracellular halves of TM6 and TM7, while index j refers to Cα atoms in the extracellular halves of TM2, TM3, and TM10. We thus obtained ~2,000 clusters for each of the simulation systems (using RMSD cut-off values of 1.3 Å for V1 and V2, 0.1 Å for V4; and 6.25 Å for S3). Using the WHAM equations, we calculated the relative free energy of each of these clusters, and then identified the major basins in this space with the MCL method (with p = 1.4.) Water occupancy maps were calculated for each of these major free-energy basins, only using the sampling gathered by the unbiased replicas.
PMC4822050	Hemi-methylated DNA opens a closed conformation of UHRF1 to facilitate its histone recognition	UHRF1 is an important epigenetic regulator for maintenance DNA methylation. UHRF1 recognizes hemi-methylated DNA (hm-DNA) and trimethylation of histone H3K9 (H3K9me3), but the regulatory mechanism remains unknown. Here we show that UHRF1 adopts a closed conformation, in which a C-terminal region (Spacer) binds to the tandem Tudor domain (TTD) and inhibits H3K9me3 recognition, whereas the SET-and-RING-associated (SRA) domain binds to the plant homeodomain (PHD) and inhibits H3R2 recognition. Hm-DNA impairs the intramolecular interactions and promotes H3K9me3 recognition by TTD–PHD. The Spacer also facilitates UHRF1–DNMT1 interaction and enhances hm-DNA-binding affinity of the SRA. When TTD–PHD binds to H3K9me3, SRA-Spacer may exist in a dynamic equilibrium: either recognizes hm-DNA or recruits DNMT1 to chromatin. Our study reveals the mechanism for regulation of H3K9me3 and hm-DNA recognition by URHF1.DNA methylation is an important epigenetic modification for gene repression, X-chromosome inactivation, genome imprinting and maintenance of genome stability12345. Mammalian DNA methylation is established by de novo DNA methyltransferases DNMT3A/3B, and DNA methylation patterns are maintained by maintenance DNA methyltransferase 1 (DNMT1) during DNA replication678. Ubiquitin-like, containing PHD and RING fingers domains, 1 (UHRF1, also known as ICBP90 and NP95 in mouse) was shown to be essential for maintenance DNA methylation through recruiting DNMT1 to replication forks in S phase of the cell cycle910. UHRF1 is essential for S phase entry1112 and is involved in heterochromatin formation13. UHRF1 also plays an important role in promoting proliferation and is shown to be upregulated in a number of cancers, suggesting that UHRF1 may serve as a potential drug target for therapeutic applications14151617. UHRF1 is a multi-domain containing protein connecting histone modification and DNA methylation. As shown in Fig. 1a, UHRF1 is comprised of an N-terminal ubiquitin-like domain, followed by a tandem Tudor domain (TTD containing TTD and TTD sub-domains), a plant homeodomain (PHD), a SET-and-RING-associated (SRA) domain, and a C-terminal really interesting new gene (RING) domain. We and other groups demonstrated that the TTD and the PHD coordinately recognize histone H3K9me3, in which residue R2 is recognized by the PHD and tri-methylation of residue K9 (K9me3) is recognized by the TTD18192021222324. The SRA preferentially binds to hemi-methylated DNA (hm-DNA)252627. Recent studies show that the SRA directly binds to replication focus targeting sequence (RFTS) of DNMT1 (RFTS)282930. A spacer region (Fig. 1a, designated Spacer hereafter) connecting the SRA and the RING is rich in basic residues and predicted to be unstructured for unknown function. Recent study shows that phosphatidylinostiol phosphate PI5P binds to the Spacer and induces a conformational change of UHRF1 to allow the TTD to recognize H3K9me3 (ref. 31). These studies indicate that UHRF1 connects dynamic regulation of DNA methylation and H3K9me3, which are positively correlated in human genome. However, how UHRF1 regulates the recognition of these two repressive epigenetic marks and recruits DNMT1 for chromatin localization remain largely unknown. Here we report that UHRF1 adopts a closed conformation, in which the C-terminal Spacer binds to the TTD and inhibits its recognition of H3K9me3, whereas the SRA binds to the PHD and inhibits its recognition of H3R2 (unmethylated histone H3 at residue R2). Upon binding to hm-DNA, UHRF1 impairs the intramolecular interactions and promotes the H3K9me3 recognition by TTD–PHD, which may further enhance its genomic localization. As a result, UHRF1 is locked in the open conformation by the association of H3K9me3 by TTD–PHD, and thus SRA-Spacer either recognizes hm-DNA or recruits DNMT1 for DNA methylation. Therefore, UHRF1 may engage in a sophisticated regulation for its chromatin localization and recruitment of DNMT1 through a mechanism yet to be fully elucidated. Our study reveals the mechanism for regulation of H3K9me3 and hm-DNA recognition by UHRF1. To investigate how UHRF1 coordinates the recognition of H3K9me3 and hm-DNA, we purified recombinant UHRF1 (truncations and mutations) proteins from bacteria. We first performed an in vitro pull-down assay using biotinylated histone H3 peptides and hm-DNA (Supplementary Table 1). As shown in Fig. 1b, hm-DNA largely enhanced the interaction between full-length UHRF1 and unmethylated histone H3 (H3K9me0) or H3K9me3 peptide. Compared with hm-DNA, um-DNA (unmethylated DNA) or fm-DNA (fully methylated DNA) showed marginal effect on facilitating the interaction between UHRF1 and histone peptides, which is consistent with previous studies that UHRF1 prefers hm-DNA for chromatin association (Supplementary Fig. 1a)252627. In contrast, histone peptides showed no enhancement on the interaction between hm-DNA and UHRF1 (Fig. 1c). These results suggest that hm-DNA facilitates histone recognition by UHRF1. Our previous studies show that the PHD recognizes H3K9me0 and the TTD and the PHD together (TTD–PHD) coordinately recognize H3K9me3 (refs. 19, 20). We noticed that the isolated TTD–PHD showed much higher (∼31-fold) binding affinity to H3K9me3 peptide than full-length UHRF1 (Fig. 1d and Supplementary Table 2), and the isolated PHD showed much higher (∼34-fold) binding affinity to H3K9me0 peptide than full-length UHRF1 (Fig. 1e). The gel filtration analysis showed that UHRF1 is a monomer in solution (Supplementary Fig. 1b), indicating that the intramolecular (not intermolecular) interaction of UHRF1 regulates histone recognition. These results suggest that UHRF1 adopts an unfavourable conformation for histone H3 tails recognition, in which TTD–PHD might be blocked by other regions of UHRF1, and hm-DNA impairs this intramolecular interaction to facilitate its recognition of histone H3 tails. To test above hypothesis, we performed glutathione S-transferase (GST) pull-down assay using various truncations of UHRF1. Interestingly, the TTD directly bound to SRA-Spacer but not the SRA, suggesting that the Spacer (residues 587–674) is important for the intramolecular interaction (Fig. 2a). The isothermal titration calorimetry (ITC) measurements show that the TTD bound to the Spacer (but not the SRA) in a 1:1 stoichiometry with a binding affinity (KD) of 1.59 μM (Fig. 2b). The presence of the Spacer markedly impaired the interaction between TTD–PHD and H3K9me3 (Fig. 2c). The results indicate that the Spacer directly binds to the TTD and inhibits its interaction with H3K9me3. The GST pull-down assay also shows that the PHD bound to the SRA, which was further confirmed by the ITC measurements (KD=26.7 μM; Fig. 2a,d). Compared with the PHD alone, PHD-SRA showed decreased binding affinity to H3K9me0 peptide by a factor of eight (Fig. 2e). Pre-incubation of the SRA also modestly impaired PHD–H3K9me0 interaction. These results indicate that the SRA directly binds to the PHD and inhibits its binding affinity to H3K9me0. Taken together, UHRF1 seems to adopt a closed form through intramolecular interactions (TTD–Spacer and PHD-SRA), which inhibit histone H3 tail recognition by UHRF1. To investigate the intramolecular interaction within UHRF1, we first mapped the minimal regions within the Spacer for the interaction with the TTD (Supplementary Fig. 2a). Internal deletions of the Spacer, including Spacer, Spacer, Spacer and Spacer, bound to the TTD with comparable binding affinities to that of the Spacer, whereas Spacer showed no detectable interaction. Spacer, Spacer and Spacer also decreased binding affinities, indicating that residues 642–674 are important for TTD–Spacer interaction. We next determined the solution structure of the TTD (residues 134–285) bound to Spacer by conventional NMR techniques (Supplementary Table 3 and Supplementary Fig. 3a,b). In the complex structure, each Tudor domain adopts a ‘Royal' fold containing a characteristic five-stranded β-sheet and the two Tudor domains tightly pack against each other with a buried area of 573 Å (Fig. 3a). The TTD adopts similar fold to that in TTD–PHD–H3K9me3 complex structure (PDB: 4GY5)19 with a root-mean-square deviation of 1.09 Å for 128 Cα atoms, indicating that the Spacer does not result in obvious conformational change of the TTD (Fig. 3b). The Spacer (residues 643–655 were built in the model) adopts an extended conformation and binds to an acidic groove on the TTD (Fig. 3c). The TTD–Spacer interaction is mediated by a number of hydrogen bonds (Fig. 3d). The side chain of residue K648 forms hydrogen bonds with the carbonyl oxygen atom of D189 and side chain of D190 of the TTD. The side chain of residue R649 packs against an acidic surface mainly formed by residues D142 and E153. Residue S651 forms hydrogen bonds with the main chain of residues G236 and W238. The interaction is further supported by hydrogen bonds formed between residues K650, A652, G653 and G654 of the Spacer and residues N228, G236 and W238 of the TTD, respectively. In support of above structural analyses, mutation D142A/E153A of the TTD abolished its interaction with the Spacer (Fig. 3e). Mutations K648D and S651D of the Spacer decreased their binding affinities to the TTD, and mutation R649A of the Spacer showed more significant decrease (∼13-fold) in the binding affinity (Fig. 3f). As negative control, mutations S639D and S666D of the Spacer showed little effect on the interaction. Interestingly, phosphorylation at residue S651 of UHRF1 was observed in previous mass-spectrometry analyses32. Compared with the unmodified peptide of Spacer, a phosphorylation at S651 markedly decreased the binding affinity to the TTD (Supplementary Fig. 2b), suggesting that the phosphorylation may regulate the intramolecular interaction within UHRF1. Previous studies indicate that the TTD binds to a linker region connecting the TTD and PHD (residues 286–306, designated Linker, Fig. 1a), and TTD–Linker interaction is essential for H3K9me3 recognition by TTD–PHD181923. Comparison of TTD–Spacer and TTD–PHD–H3K9me3 (PDB: 4GY5) structures indicates that the Spacer and the Linker bind to the TTD in a similar manner in the two complexes (Fig. 3b). In TTD–PHD–H3K9me3 structure, residues R295, R296 and S298 of the Linker adopt almost identical conformation to residues K648, R649 and S651 of the Spacer in TTD–Spacer structure, respectively. Similar intramolecular contacts (TTD–Linker and TTD–Spacer) were observed in the two structures (Fig. 3b,d and Supplementary Fig. 4a). Thus, the Spacer may disrupt the TTD–Linker interaction and inhibits the recognition of H3K9me3 by TTD–PHD. To test this hypothesis, we first investigated the potential competition between the Linker and the Spacer for their interaction with the TTD. The ITC experiment shows that the Linker peptide (289–306) bound to the TTD with a binding affinity of 24.04 μM (Supplementary Fig. 4b), ∼15-fold lower than that of the Spacer peptide (KD=1.59 μM, Fig. 3e). The competitive ITC experiments show that TTD–Spacer binding affinity decreased by a factor of two in the presence of the Linker, whereas TTD–Linker interaction was abolished in the presence of the Spacer (Supplementary Fig. 4c). Compared with TTD–Spacer interaction (KD=1.48 μM), TTD–PHD decreased the binding affinity to the Spacer (KD=10.68 μM), whereas mutation R295D/R296D (within the Linker and important for TTD–Linker interaction) of TTD–PHD showed minor decrease in the binding affinity (KD=2.69 μM; Fig. 3g), indicating a competition between the Spacer and the Linker on the same binding site of the TTD. Notably, although the Linker (in the context of TTD-PHD) impairs the TTD–Spacer interaction to some extent, the isolated Spacer could still bind to TTD–PHD with moderate binding affinity (KD=10.68 μM), supporting the existence of the intramolecular interaction within UHRF1. To test whether TTD–Spacer association exists in the context of full-length UHRF1, we used various truncations of UHRF1 in the GST pull-down assay. As indicated in Fig. 3h, full-length UHRF1 and UHRF1 showed no interaction with GST-tagged TTD, Linker or Spacer, suggesting that TTD–Spacer interaction in-cis within full-length UHRF1 or UHRF1 prohibits TTD–Spacer complex formation in-trans. In contrast, UHRF1 bound to GST-TTD, and UHRF1 bound to GST-Spacer, indicating that lack of TTD–Spacer interaction in-cis, TTD–Spacer complex could form in-trans, supporting that the TTD binds to the Spacer in the context of full-length UHRF1. Moreover, GST-Linker showed very weak if not undetectable interaction with wild-type or deletions of UHRF1, suggesting that TTD–Linker interaction is much weaker than that of TTD–Spacer. Taken together, UHRF1 adopts a closed conformation, in which the Spacer binds to the TTD through competing with the Linker, and therefore inhibits H3K9me3 recognition by UHRF1. Our previous study indicates that H3K9me3 binds to the TTD in different manner in TTD–PHD–H3K9me3 (ref. 19) and TTD-H3K9me3 (PDB: 2L3R)21 structures. Because the TTD is always associated with the PHD, whether the pattern of TTD–H3K9me3 interaction exists in vivo remains unknown. Nevertheless, comparison of TTD–H3K9me3 and TTD–Spacer structures indicates that H3K9me3 and the Spacer overlap on the surface of the TTD (Supplementary Fig. 4d), suggesting that the Spacer might block the H3K9me3 recognition by the isolated TTD. As shown in Supplementary Fig. 4e, the Spacer inhibited TTD–H3K9me3 interaction, whereas its TTD-binding defective mutants of the Spacer or the SRA (a negative control) markedly decreased the inhibition. We next tested whether such inhibition also occurs in the context of full-length UHRF1. Compared with full-length UHRF1, UHRF1 enhanced H3K9me3-binding affinity by a factor of four (Supplementary Fig. 4f). The restoration of H3K9me3-binding affinity is not dramatic because the PHD still binds to histone H3 in both proteins. To exclude this effect, we performed the assay using UHRF1, which abolishes H3R2-binding affinity of the PHD1920. UHRF1 showed undetectable H3K9me3-binding affinity, whereas UHRF1 dramatically restored its H3K9me3-binding affinity (KD=8.69 μM; Supplementary Fig. 4f), indicating that H3K9me3 recognition by the TTD is blocked by the Spacer through competitive interaction with the TTD. Moreover, the R295D/R296D mutant of full-length UHRF1 showed decreased binding affinity to H3K9me3 (eightfold lower than wild type), suggesting that mutation of R295D/R296D favours TTD–Spacer interaction and therefore promotes UHRF1 to exhibit a more stable closed conformation (Supplementary Fig. 4g). Taken together, the Spacer binds to the TTD and inhibits H3K9me3 recognition by UHRF1 through (i) disrupting TTD–Linker interaction, which is essential for H3K9me3 recognition by TTD–PHD, (ii) prohibiting H3K9me3 binding to the isolated TTD. Interestingly, pre-incubation of H3K9me3 peptide completely blocked the interaction between the Spacer and the TTD alone or TTD–PHD (Supplementary Fig. 4h), whereas the presence of the Spacer partially impaired the interaction between TTD–PHD and H3K9me3 (Fig. 2c). The results are also consistent with the previous observation that the interaction between TTD–PHD and the Spacer is much weaker (KD=10.68 μM, Fig. 3g) than that between TTD–PHD and H3K9me3 (KD=0.15 μM, Fig. 1d). These results suggest that once TTD–PHD binds to H3K9me3, UHRF1 will be locked by H3K9me3 and the Spacer is unlikely to fold back for the intramolecular interaction. To investigate whether hm-DNA could open the closed conformation of UHRF1, we first measured the intramolecular interaction using UHRF1 truncations in the presence or absence of hm-DNA. The GST pull-down assays show that the PHD bound to the SRA and such interaction was impaired by the addition of hm-DNA (Fig. 4a). H3 peptide pull-down assays show that hm-DNA only enhanced the H3K9me0-binding affinities of UHRF1 truncations containing PHD-SRA, such as PHD-SRA, TTD-PHD-SRA, TTD-PHD-SRA-Spacer, UHRF1 and UHRF1 (Fig. 4b). The result indicates that hm-DNA disrupts PHD–SRA interaction and facilitates H3K9me0-binding affinity of the PHD in a manner independent on the TTD or the Spacer. Moreover, the TTD or TTD–PHD bound to SRA–Spacer and the interaction was impaired by the addition of hm-DNA (Fig. 4c). The ITC measurements show that the presence of hm-DNA markedly impaired the interaction between the TTD and SRA–Spacer (Supplementary Fig. 5a). However, the TTD–Spacer interaction was not affected by the presence of the hm-DNA, indicating that hm-DNA displaces the Spacer from the TTD in a SRA-dependent manner (Supplementary Fig. 5b). To investigate whether hm-DNA disrupts TTD–Spacer interaction in the context of full-length UHRF1, we monitored the conformational changes of UHRF1 using its histone-binding affinity as read-out. UHRF1 was used to exclude the effect of H3K9me0 recognition by the PHD. As expected, all D334A-containing mutants showed undetectable interaction with H3K9me0 (Fig. 4d). UHRF1 bound to H3K9me3 peptide in the presence of hm-DNA, but showed no interaction in the absence of hm-DNA, which is consistent with the ITC experiments (Supplementary Fig. 4f). In contrast, UHRF1 strongly bound to H3K9me3 even in the absence of hm-DNA (Fig. 4d), indicating that the deletion of the Spacer releases otherwise blocked TTD–PHD for H3K9me3 recognition. The results further support the conclusion that the Spacer binds to the TTD in the context of full-length UHRF1 and the intramolecular interactions are disrupted by hm-DNA. We next performed similar peptide pull-down assay using two mutants (N228C/G653C and R235C/G654C) generated on UHRF1. Residues N228/R235 from the TTD and G653/G654 from the Spacer were chosen according to the TTD–Spacer complex structure (Supplementary Fig. 5c) so that the replaced Cysteine residues (one from the TTD and one from the Spacer) are physically close enough to each other to form a disulphide bond in the absence of reducing reagent (dithiothreitol, DTT). As shown in Fig. 4d, hm-DNA largely enhanced the H3K9me3-binding affinities of both mutants in the presence of DTT, but not in the absence of DTT, indicating that the disulphide bond formation (in the absence of DTT) disallows hm-DNA to disrupt TTD–Spacer interaction for H3K9me3 recognition. As negative controls, H3K9me3 recognition by UHRF1 or UHRF1 is not affected by DTT. The above results collectively demonstrate that (i) full-length UHRF1 adopts a closed form, in which the Spacer binds to the TTD and H3K9me3 recognition is inhibited; (ii) hm-DNA displaces the Spacer from the TTD in the context of full-length UHRF1 and therefore largely enhances its histone H3K9me3-binding activity in a manner independent on the PHD (SRA is required). We have previously demonstrated that hm-DNA also disrupts PHD–SRA interaction and facilitates H3K9me0-binding affinity of the PHD in a manner independent on the TTD or the Spacer. Taken together, hm-DNA disrupts the intramolecular interactions within UHRF1, and therefore facilitates the coordinate recognition of H3K9me3 by TTD–PHD. To investigate how hm-DNA impairs TTD–Spacer interaction, we tested whether the Spacer is involved in hm-DNA recognition by the SRA, which is the only known domain for hm-DNA recognition within UHRF1. In the electrophoretic mobility-shift assay, SRA–Spacer showed higher hm-DNA-binding affinity than the SRA alone (Supplementary Fig. 6a). ITC measurements show that SRA–Spacer bound to hm-DNA with a much higher binding affinity (KD=1.75 μM) than the SRA (KD=25.12 μM), whereas the Spacer alone showed no interaction with hm-DNA (Fig. 5a). In the fluorescence polarization (FP) measurements, SRA–Spacer, full-length UHRF1 and UHRF1 showed comparable hm-DNA-binding affinities (Fig. 5b and Supplementary Table 4), suggesting that UHRF1 binds to hm-DNA no matter UHRF1 adopts a closed form or not. In contrast, UHRF1 abolished hm-DNA-binding affinity, indicating that the SRA is essential for hm-DNA recognition. Compared with full-length UHRF1, UHRF1 decreased the hm-DNA-binding affinity by a factor of 14 (Fig. 5b), further supporting that the Spacer plays an important role in hm-DNA recognition in the context of full-length UHRF1. In addition, hm-DNA-binding affinities of SRA or SRA–Spacer did not obviously vary upon the change of DNA lengths but did decrease with the increasing salt concentrations (Supplementary Fig. 6b,c and Supplementary Table 5). These results indicate that the Spacer not only binds to the TTD and inhibits H3K9me3 recognition when UHRF1 adopts closed conformation, but also facilitates hm-DNA recognition by the SRA when UHRF1 binds to hm-DNA. We next mapped the minimal region of the Spacer for the enhancement of hm-DNA-binding affinity. SRA–Spacer-661 (residues 414–661) still maintained strong hm-DNA-binding affinity comparable to that of SRA–Spacer (residues 414–674), whereas SRA–Spacer-652 and SRA–Spacer-642 markedly decreased their hm-DNA-binding affinities (Fig. 5c), indicating that residues 642–661 are important for enhancing hm-DNA-binding affinity of the SRA. This minimal region largely overlaps with the Spacer region (643–655) essential for TTD interaction. We also determined the crystal structure of SRA–Spacer bound to hm-DNA at 3.15 Å resolution (Supplementary Table 6 and Supplementary Fig. 7a). The structure shows that the SRA binds to hm-DNA in a manner similar to that observed in the previously reported SRA-hm-DNA structures182627. Intriguingly, no electron density was observed for the Spacer. A possible explanation is that the Spacer facilitates SRA–hm-DNA interaction through nonspecific salt bridge contacts because DNA is rich in acidic groups and the Spacer is rich in basic residues (Supplementary Fig. 7b). The nonspecific interaction is consistent with the previous observation that UHRF1 has no DNA sequence selectivity besides hm-CpG dinucleotide. To investigate the role of the Spacer in the regulation of UHRF1 function, we transiently overexpressed GFP-tagged wild type or mutants of UHRF1 in NIH3T3 cells to determine their subcellular localization. For the NIH3T3 cells expressing wild-type UHRF1, most cells (∼74.6%) showed a focal pattern of protein that is co-localized with 4,6-diamidino-2-phenylindole (DAPI) foci (Fig. 5d), whereas the rest cells showed a diffuse nuclear staining pattern. The result is consistent with the previous studies that UHRF1 is mainly localized to highly methylated pericentromeric heterochromatin (PCH)33. In contrast, for the cells expressing UHRF1, a spacer deletion mutant with decreased hm-DNA-binding affinity (Fig. 5b), only ∼22.1% cells showed co-localization with DAPI. Previous reports have shown that the H3K9me3 recognition of UHRF1 also plays an important role in its heterochromatin localization. For example, UHRF1 mutant (within TTD domain) lacking H3K9me3-binding affinity largely reduces its co-localization with heterochromatin13213134. Because manipulation of endogenous hm-DNA in cells is technically challenging, we used UHRF1 (lacks hm-DNA-binding affinity but maintains closed conformation, Figs 3h and 5b) to test whether closed conformation of UHRF1 exists in vivo. In NIH3T3 cells, UHRF1 largely decreased chromatin association (Fig. 5d). Only ∼4.8% cells expressing UHRF1 showed an intermediate enrichment, but not characteristic focal pattern, at DAPI foci, whereas the majority of the cells showed a diffuse nuclear staining pattern. The results suggest that UHRF1 adopts closed conformation so that H3K9me3 recognition by TTD–PHD is blocked by the intramolecular interaction, and support the regulatory role of the Spacer in PCH localization of UHRF1 in vivo. Previous studies show that UHRF1 recruits DNMT1 to hm-DNA for maintenance DNA methylation through the interaction between the SRA and RFTS (refs 9, 10, 28, 29, 30). We confirmed the direct interaction between RFTS and the SRA in a solution with low salt concentration (50 mM NaCl), but observed weak or undetectable interaction in a solution with higher salt concentrations (100 or 150 mM NaCl) (Supplementary Fig. 8a). Compared with the SRA, SRA–Spacer exhibited stronger interaction with RFTS. In addition, RFTS bound to SRA–Spacer with a binding affinity of 7.09 μM, but showed no detectable interaction with the SRA (Supplementary Fig. 8b). Interestingly, the addition of hm-DNA abolished the interaction between RFTS and SRA–Spacer, suggesting that hm-DNA also regulates UHRF1–DNMT1 interaction (Supplementary Fig. 8c). These results indicate that the Spacer facilitates the interaction between RFTS and the SRA, and the interaction is impaired by the presence of hm-DNA. We next tested whether the UHRF1–DNMT1 interaction is regulated by the conformational change of UHRF1. Because the addition of hm-DNA disrupts the interaction between the SRA–Spacer and RFTS, we used various truncations to mimic open and closed forms of UHRF1. In the absence of hm-DNA, only UHRF1 bound to RFTS, whereas full-length UHRF1, UHRF1 and UHRF1 showed undetectable interaction (Fig. 5e). As the deletion of the TTD allows UHRF1 to adopt an open conformation, the results suggest that RFTS binds to SRA–Spacer when UHRF1 adopts an open conformation in the absence of hm-DNA. In support of above observations, the addition of large amount of RFTS impaired the interaction between UHRF1 and hm-DNA (Supplementary Fig. 8d), suggesting an existence of dynamic equilibrium between UHRF1–hm-DNA and UHRF1–DNMT1 complexes. According to the above results, we here proposed a working model for hm-DNA-mediated regulation of UHRF1 conformation (Fig. 5f). In the absence of hm-DNA (A), UHRF1 prefers a closed conformation, in which the Spacer binds to the TTD by competing with the Linker and the SRA binds to the PHD. As a result, the recognition of histone H3K9me3 by the TTD is blocked by the Spacer, and recognition of unmodified histone H3 (H3R2) by the PHD is inhibited by the SRA. The interaction between UHRF1 and DNMT1 is also weak because the Spacer is unable to facilitate the intermolecular interaction. In the presence of hm-DNA (B), UHRF1 prefers an open conformation, in which the SRA binds to the hm-DNA; the Spacer dissociates from the TTD and facilitates the interaction between the SRA and hm-DNA; the Linker binds to the TTD and allows TTD–PHD to recognize histone H3K9me3. When UHRF1 adopts an open conformation and has already bound to H3K9me3 (B), the interaction between H3K9me3 and TTD–PHD further prevents the Spacer from folding back to interact with the TTD, and therefore locks UHRF1 in an open conformation. The association of UHRF1 to the histone may facilitate the ubiquitination of histone tail (mediated by RING domain) for DNMT1 targeting3536. Moreover, through a mechanism yet to be fully elucidated, DNMT1 targets hm-DNA for maintenance DNA methylation, probably through interaction with the histone ubiquitylation and/or SRA-Spacer. This cartoon summarizes our findings in this study. The P(r) function obtained from small-angle X-ray scattering (SAXS) measurements of TTD–PHD–SRA–Spacer–hm-DNA complex showed a broader distribution than that of the TTD–PHD–SRA–Spacer alone, supporting the proposed model that UHRF1 adopts an open conformation in the presence of hm-DNA (Supplementary Fig. 8e). Many questions need to be further clarified. We have tried crystallizing more than three sub-constructs with and without DNA across over 1,200 crystallization conditions but failed to determine the structure of TTD–PHD–SRA–Spacer in the absence or presence of hm-DNA. Getting these structures would greatly help for understanding the hm-DNA-mediated regulation of UHRF1. In addition, this regulatory process should be further characterized using advanced techniques, such as single molecular measurement. Our previous studies show that phosphorylation at S639 within the Spacer disrupts interaction between UHRF1 and deubiquitylase USP7 and decreases UHRF1 stability in the M phase of the cell cycle37. The Spacer was predicted to contain two nuclear localization signals, residues 581–600 and 648-670 (ref. 38). In this report, we found that the Spacer (i) binds to the TTD in the closed form of UHRF1 and inhibits its interaction with H3K9me3; (ii) facilitates hm-DNA recognition by the SRA and (iii) facilitates the interaction between the SRA and RFTS. These findings together indicate that the Spacer plays a very important role in the dynamic regulation of UHRF1. When our manuscript was in preparation, Gelato et al. reported that binding of PI5P to the Spacer opens the closed conformation of UHRF1 and increases H3K9me3-binding affinity of the TTD31. The result suggests that PI5P may facilitate the conformational change of UHRF1 induced by hm-DNA when UHRF1 is recruited to chromatin. In addition, mass-spectrometry analyses have identified several phosphorylation sites (S639, S651, S661) within the Spacer, suggesting that post-translational modification may add another layer of regulation of UHRF1 (refs 32, 37, 39, 40). It has been well characterized that the SRA of UHRF1 preferentially recognizes hm-DNA through a base-flipping mechanism182627. Our study demonstrates that the Spacer markedly enhances the hm-DNA-binding affinity of the SRA and the deletion of the Spacer impairs heterochromatin localization of UHRF1, indicating that the Spacer is essential for recognition of hm-DNA in the context of full-length UHRF1. Interestingly, variant in methylation 1 (VIM1, a UHRF1 homologue in Arabidopsis) contains an equivalent spacer region, which was shown to be required for hm-DNA recognition by its SRA domain941, suggesting a conserved regulatory mechanism in SRA domain-containing proteins. Intriguingly, UHRF2 (the only mammalian homologue of UHRF1) and UHRF1 show very high sequence similarities for all the domains but very low similarity for the Spacer (Supplementary Fig. 7c). Thus, although UHRF2 exhibits the histone- and hm-DNA-binding activities, the difference in the Spacer region may contribute to the functional differences between UHRF1 and UHRF2. This is also consistent with previous finding that UHRF2 is unable to replace UHRF1 to maintain the DNA methylation144243. One of the key questions in the field of DNA methylation is why UHRF1 contains modules recognizing two repressive epigenetic marks: H3K9me3 (by TTD–PHD) and hm-DNA (by the SRA). Previous studies show that chromatin localization of UHRF1 is dependent on hm-DNA10, whereas other studies indicate that histone H3K9me3 recognition and hm-DNA association are both required for UHRF1-mediated maintenance DNA methylation2334. However, little is known about the crosstalk between these two epigenetic marks within UHRF1. In this study, we provide an explanation. As shown in the proposed model, recognition of H3K9me3 by full-length UHRF1 is blocked to avoid its miss-localization to unmethylated genomic region, in which chromatin contains H3K9me3 (KD=4.61 μM) or H3K9me0 (KD=25.99 μM). We have shown that full-length UHRF1 and SRA–Spacer strongly bind to hm-DNA (0.35 and 0.49 μM, respectively) and the Spacer plays an important role in PCH localization (Fig. 5d). Therefore, genomic localization of UHRF1 is primarily determined by its recognition of hm-DNA, which allows UHRF1 to adopt an open form and promotes its histone tail recognition for proper genomic localization and function. As a result, when SRA–Spacer dissociates from hm-DNA and binds to DNMT1 with a currently unknown mechanism, UHRF1 may keep the complex associated with chromatin through the interaction between TTD–PHD and H3K9me3 (or PHD-H3), and make it possible for DNMT1 to target proper DNA substrate for methylation. This explanation agrees nicely with previous observations and clarifies the importance of coordinate recognition of H3K9me3 and hm-DNA by UHRF1 for maintenance DNA methylation. UHRF1 is essential for maintenance DNA methylation through recruiting DNMT1 to DNA replication forks during S phase9103435. This function is probably induced by a direct interaction between the SRA and RFTS (refs 28, 29, 30) or interaction between DNMT1 and ubiquitylation of histione tail3544. Recent study indicates that histone tail association of UHRF1 (by the PHD domain) is required for histone H3 ubiquitylation, which is dependent on ubiquitin ligase activity of the RING domain of UHRF1 (ref. 44). DNMT1 binds to ubiquitylated histone H3 and ubiquitylation is required for maintenance of DNA methylation in vivo. In this study, we found that both TTD and PHD are regulated by hm-DNA to recognize histone tail. Thus, the closed form UHRF1 may prevent miss localization of URHF1, whereas only the UHRF1 in open conformation (induced by hm-DNA) could properly binds to histone tail for ubiquitylation and subsequent DNA methylation. Moreover, structural analyses of DNMT1–DNA4546 and SRA–DNA182627 complexes also indicate that it is impossible for DNMT1 to methylate the hm-DNA that UHRF1 binds to because of steric hindrance. In our in vitro assays, we could detect interaction between SRA–Spacer and RFTS, but not the interaction between full-length UHRF1 and RFTS (Supplementary Fig. 8a,b and Fig. 5e). The results suggest that UHRF1 adopts multiple conformations. Binding of UHRF1 to hm-DNA may serve as a switch for its recruitment of DNMT1. The S phase-dependent interaction between UHRF1 and DNMT1 (refs 9, 10, 43) suggest that DNMT1 may also undergo conformation changes so that RFTS binds to UHRF1 and the catalytic domain of DNMT1 binds to hm-DNA for reaction. The ubiquitin-like domain (residues 1–133), TTD (residues 134–285), PHD (residues 307–366), TTD–PHD (residues 134–366), SRA (residues 414–617), SRA–Spacer (residues 414–674), Spacer (residues 587–674), RING (residues 675–793), UHRF1 (residues 1–793) and other mutants or truncations of human UHRF1 were sub-cloned in a pGEX-6p-1 derivative vector. The truncated Spacer (residues 627–674) used for NMR analyses was inserted into modified pRSF-Duet-1 vector. All the proteins were expressed in E. coli strain BL21 (DE3) and purified as described previously1920. In brief, the transformants were grown at 37 °C in 2X YT medium and induced by adding isopropyl-β-D-thiogalactopyranoside (IPTG) to 0.1 mM when the OD600 reached 0.6 and further incubated at 15 °C overnight. The cells were harvested and disrupted. After centrifugation, the supernatant of GST-tagged proteins was purified by GST affinity column (GE Healthcare) and the His-tagged truncated Spacer (residues 627–674) was purified by Nickel Nitrilotriacetic Acid affinity chromatography (GE Healthcare). The GST-tagged proteins used for GST pull-down experiment were eluted directly. The fusion proteins were digested with PreScission protease and further purified by ion exchange and gel filtration chromatography. The proteins were concentrated to 5–20 mg ml for the following biochemical and structural analyses. To purify N- and C-labelled proteins, the transformants were grown in M9 medium containing N-labelled NH4Cl (1 g l) and C-labelled glucose (2 g l). The isotope-labelled TTD and truncated Spacer were purified as described above. For GST pull-down assays, 15 μg GST-tagged proteins were incubated with 40 μg recombinant proteins in 500 μl pull-down buffer (20 mM HEPES-NaOH, pH 8.0, 100 mM NaCl, 5% glycerol and 0.1% Triton X-100) for 1 h at 4 °C. Glutathione resins (GE Healthcare) were washed six times with pull-down buffer then mixed with the proteins for 1 h at 4 °C. After washed three times with pull-down buffer, the bound proteins were analysed by SDS–PAGE Coomassie blue staining. For competitive pull-down experiments: purified proteins were pre-incubated with hemi-methylated-DNA (12 bp, upper strand: 5′-GGGCCXGCAGGG-3′, X=5-methyldeoxycytosine) at the indicated molar ratios for 10 min at 4 °C. For salt concentration-dependent pull-down experiments, the pull-down buffer contains 50, 100 or 150 mM NaCl, respectively. For histone peptide or hm-DNA pull-down, 1 μg biotinylated histone H3 peptide (residues 1–21) or hm-DNA (12-bp, upper strand 5′-GAGGCXGCCTGC-3′ and lower strand 5′-biotin-GCAGGCGGCCTC-3′, X=5-methyldeoxycytosine) were incubated with 20 μg wild type or mutants of UHRF1 proteins in 500 μl pull-down buffer (20 mM HEPES-NaOH, pH 8.0, 100 mM NaCl, 5% glycerol and 0.1% Triton X-100) for 1 h at 4 °C. The proteins were pre-incubated with hm-DNA (12- bp, upper strand: 5′-GGGCCXGCAGGG-3′, X=5-methyldeoxycytosine) or indicated H3 peptide, or binding buffer as a control, at 1:2 molar ratios for 10 min at 4 °C. Then, 20 μl streptavidin beads were washed six times with pull-down buffer and incubated with the mixture for 1 h at 4 °C. The bound proteins were analysed as described above. The results are summarized in Supplementary Fig. 10. The binding affinity of protein/protein, protein/peptide or protein/DNA was measured by adding 0.05 mM protein in cell and titrated with 0.5 mM protein, peptide or hm-DNA (12 bp, upper strand: 5′-GGGCCXGCAGGG-3′, X=5-methyldeoxycytosine) in the syringe using iTC200 microcalorimeter (GE Healthcare) at 18 °C. For competition ITC experiments: the indicated proteins were pre-incubated with competitive peptide, protein or hm-DNA (at 1:2 molar ratio if not specified) for 10 min followed by ITC measurements as described above. Proteins, DNA and peptides were prepared within ITC buffer containing 10 mM HEPES, pH 8.0, 100 mM NaCl. The data were fitted by software Origin 7.0. All ITC results were summarized in Supplementary Table 2 and raw data were shown in Supplementary Fig. 9. To determine the mole ratio of TTD versus Spacer peptide (residues 627–674) in the complex for NMR studies, NMR stepwise titration assay was performed at 20 °C in a PBS buffer supplemented with 0.01% NaN3, pH 7.4 and 10% D2O. The Spacer peptide was added into N-labelled TTD solution with an increasing molar ratio of TTD/Spacer as follows: 1:0.0, 1:0.2, 1:0.4, 1:0.6, 1:0.8, 1:1.2 and 1:1.5. The H–N heteronuclear single-quantum correlation (HSQC) spectra of the TTD were collected after each addition. Two NMR samples were prepared in a mole ratio of 1:1.2 (TTD/Spacer). One is 0.7 mM uniformly C/N-labelled TTD in complex with unlabelled Spacer peptide in NMR buffer (PBS buffer, 0.01% NaN3, pH 7.4 and 10% D2O). The other is 0.5 mM N-labelled Spacer peptide mixed with unlabelled TTD protein. All NMR experiments were performed at 20 °C on a Varian Unity Inova 600 NMR spectrometer equipped with a triple resonances cryoprobe and pulsed field gradients. The standard suite of experiments for assigning the H, C and N backbone, determining the side-chain chemical shifts of the TTD in complex with the Spacer peptide and collecting the Nuclear Overhauser effect (NOE)-based distance restraints were measured47, including two-dimensional (2D) C-edited HSQC and N-edited HSQC; three-dimensional (3D) HNCA, HNCO, HN(CO)CA, HNCACB, CBCA(CO)NH, N-resolved HSQC-total correlation spectroscopy (TOCSY) and C-resolved HSQC-TOCSY in both aliphatic and aromatic regions; N-resolved HSQC-NOESY; C-resolved HSQC-NOESY for both aliphatic and aromatic resonances and 2D hbcbcgcdceheA and hbcbcgcdhdA spectra for the correlation of Cβ and Hδ or Hɛ in the aromatic ring that is used for aromatic proton assignment48. The NMR signals of bound TTD were assigned according to the previously report21. The proton NMR signals of the bound Spacer peptide were assigned by analysing the 2D C-filtered, N-filtered and J-resolved NOE spectroscopy (NOESY) and TOCSY spectra recorded for the C- and N-labelled protein with the unlabelled Spacer peptide and the 2D H–H COSY, NOESY and TOCSY spectra recorded for the unlabelled free Spacer peptide, and N-edited HSQC, 3D N-resolved HSQC-TOCSY for the N-labelled Spacer in complex with the TTD protein in the NMR buffer described above, respectively. The intermolecular NOEs between the labelled protein and the unlabelled Spacer peptide were obtained by analysing the 3D C-F1-edited and C/N-F3-filtered NOESY spectra. The spectra were processed with the NMRPipe programme49 and analysed using Sparky 3 (http://www.cgl.ucsf.edu/home/sparky/). The calculations were performed using a standard simulated annealing protocol implemented in the XPLOR-2.29 programme (NIH version)50. The inter-proton distance restraints derived from the NOE intensities were grouped into three distance ranges, namely 1.8–2.9, 1.8–3.5and 1.8–6.0 Å, which corresponds to strong, medium and weak NOEs, respectively. The dihedral angles phi and psi were derived from the backbone chemical shifts (HN, HA, CO and CA) using the programme TALOS51. The hydrogen-bond constraints were generated based on the observed NOE pattern between anti-β-sheets in the TTD domain, confirmed by H-D exchange experiments, and used in structural calculation. A total of ten iterations were performed (50 structures in the initial eight iterations). In total, 100 structures were computed during the last two iterations, and the 20 conformers with the lowest energy were used to represent the 3D structures. The conformers of these bundles (TTD in complex with the Spacer peptide) do not violate the following constraints: NOE >0.3 Å and dihedral angle >3. The entire structure statistics were evaluated with PROCHECK52 and PROCHECK-NMR53 and are summarized in Supplementary Table 3. All of the structure figures were generated using the PyMOL54 and MOLMOL programmes55. A 6-carboxy-fluorescein (FAM)-labelled primer, 5′-CCATGCGCTGAC-3′, was annealed to a primer 5′-GTCAGXGCATGG-3′ (X=5-methyldeoxycytosine). The hemi-methylated double-strand DNA was used in both electrophoretic mobility-shift assay and FP assays. 50 nM FAM-hm-DNA (1 pmole per lane) was pre-incubated with indicated amount of proteins in reaction buffer (20 mM HEPES, pH 7.5, 100 mM NaCl, 8% glycerol and 1 mM DTT) for 20 min on ice. The samples were subjected to a 10% polyacrylamide gel electrophoresis and run in 0.5 × Tris-borate-EDTA buffer at 100 V for 1 h at 4 °C. The results were visualized on Tanon-5200 Chemiluminescent Imaging System (Tanon Science & Technology Co., Ltd). The 12-bp FAM-labelled hm-DNA (as described above) was incubated with increasing amount of indicated proteins for 20 min at 25 °C in reaction buffer containing 20 mM HEPES, pH 7.5, 175 mM NaCl, 8% glycerol and 1 mM DTT. FP measurements were performed at 25 °C on Synergy 4 Microplate Reader (BioTek). The 16-bp and 20-bp FAM-labelled hm-DNA (lower strand: 5′-GTGTCAGXGCATGGCC-3′ and 5′-CCGTGTCAGXGCATGGCCAT-3′, respectively. X=5-methyldeoxycytosine) were used in the FP experiment to test the effect of DNA length on the protein/DNA interaction. All experiments were performed in triplicate. The curves were fitted by GraphPad Prism 5. For salt concentration-dependent FP experiments, the reaction buffer contains 50 mM or 150 mM NaCl, respectively. Crystals of SRA–Spacer in complex with an 18-bp hm-DNA (upper strand: 5′-CATCGTCCCTGCGGGCCC-3′, lower strand: 5′-GGGCCXGCAGGGACGATG-3′. X=5-methyldeoxycytosine) were grown at 18 °C using the hanging drop vapour diffusion method by mixing an equal volume of protein–DNA complex and crystallization buffer containing 12% PEG 3350, 45 mM citric acid/55 mM BIS-TRIS propane (pH 6.9). Protein and hm-DNA were mixed at the molar ratio of 1:1.5 and incubated for 0.5 h on ice before crystallization. Crystals were flash frozen in a cold nitrogen stream at −173 °C. All data sets were collected on beamline BL17U at the SSRF (Shanghai Synchrotron Radiation Facility, China). The data were processed using the programme HKL2000 (ref. 56). The structure of SRA–Spacer–hm-DNA complex was determined by molecular replacement using structure of the SRA (PDB:3BI7)26 as a searching model. Rotation and translation function searches were performed with the programme PHASER57. The model was manually built with COOT58. All refinements were performed using the refinement module phenix.refine of PHENIX package59. The model quality was checked with the PROCHECK programme52 and all structure figures were generated by PyMol54. NIH3T3 cells were obtained from the Shanghai Institute of Biochemistry and Cell Biology. Wild type and mutants of UHRF1 were sub-cloned into pEGFP-C1 vector. Transient transfections of NIH3T3 cells were carried out using Lipofectamine 2000 (Invitrogen). The NIH3T3 cells were grown on glass coverslips and harvested in 36 h after transfection. The images were acquired and examined as previously described34. Briefly, cells were fixed with 4% paraformaldehyde for 25 min, then washed with PBS three times. Coverslips were mounted with Antifade reagent containing DAPI (Molecular Probes) on slides and examined with a confocal microscopy. SAXS measurements were performed with Anton Paar SAXSess mc2 instrument with linecolimation and charge-coupled-device detection. The X-ray wavelength was 1.5418 Å (CuKα), the sample to detector was 306.8 mm and the sample slit width was 10 mm. Each sample was prepared in 300 μl solution contained 150 mM NaCl, 10 mM HEPES, pH=8.0, 5 mM DTT and 5% glycerol. For the hm-DNA-bound form, the protein was pre-incubated with hm-DNA (12-bp, upper strand: 5′-GGGCCmCGCAGGG-3′, mC=5-methyldeoxycytosine) at 1:1.2 molar ratio for 10 min on ice. To correct for interparticle interference, the data of protein sample were collected twice and each time for 1 h. The solution containing no protein sample was also tested as background. The initial data were first processed using SAXSquant and the further analysis with ATSAS software. The SAXS data were only analysed these were collected in the first hour because there was no time effect on the samples. The radius of gyration Rg was estimated from primus. The distance distribution function P(r) was calculated in PCG package. The maximum particle dimension Dmax was estimated from the P(r) function as the r for which P(r)=0. Accession codes: The coordinate and structure factor for the TTD–Spacer complex structure have been deposited in the Protein Data Bank under accession code 5IAY. The chemical shift assignment of TTD–Spacer was deposited with BMRB ADIT-NMR online deposition system under the accession number 30019. How to cite this article: Fang, J. et al. Hemi-methylated DNA opens a closed conformation of UHRF1 to facilitate its histone recognition. Nat. Commun. 7:11197 doi: 10.1038/ncomms11197 (2016).
PMC4772114	Structural basis for the regulation of enzymatic activity of Regnase-1 by domain-domain interactions	Regnase-1 is an RNase that directly cleaves mRNAs of inflammatory genes such as IL-6 and IL-12p40, and negatively regulates cellular inflammatory responses. Here, we report the structures of four domains of Regnase-1 from Mus musculus—the N-terminal domain (NTD), PilT N-terminus like (PIN) domain, zinc finger (ZF) domain and C-terminal domain (CTD). The PIN domain harbors the RNase catalytic center; however, it is insufficient for enzymatic activity. We found that the NTD associates with the PIN domain and significantly enhances its RNase activity. The PIN domain forms a head-to-tail oligomer and the dimer interface overlaps with the NTD binding site. Interestingly, mutations blocking PIN oligomerization had no RNase activity, indicating that both oligomerization and NTD binding are crucial for RNase activity in vitro. These results suggest that Regnase-1 RNase activity is tightly controlled by both intramolecular (NTD-PIN) and intermolecular (PIN-PIN) interactions.The initial sensing of infection is mediated by a set of pattern-recognition receptors (PRRs) such Toll-like receptors (TLRs) and the intracellular signaling cascades triggered by TLRs evoke transcriptional expression of inflammatory mediators that coordinate the elimination of pathogens and infected cells123. Since aberrant activation of this system leads to auto immune disorders, it must be tightly regulated. Regnase-1 (also known as Zc3h12a and MCPIP1) is an RNase whose expression level is stimulated by lipopolysaccharides and prevents autoimmune diseases by directly controlling the stability of mRNAs of inflammatory genes such as interleukin (IL)-6, IL-1β, IL-2, and IL-12p404567. Regnase-1 accelerates target mRNA degradation via their 3′-terminal untranslated region (3′UTR), and also degrades its own mRNA8. Regnase-1 is a member of Regnase family and is composed of a PilT N-terminus like (PIN) domain followed by a CCCH-type zinc–finger (ZF) domain, which are conserved among Regnase family members7910. Recently, the crystal structure of the Regnase-1 PIN domain derived from Homo sapiens was reported11. The structure combined with functional analyses revealed that four catalytically important Asp residues form the catalytic center and stabilize Mg binding that is crucial for RNase activity. Several CCCH-type ZF motifs in RNA-binding proteins have been reported to directly bind RNA12131415. In addition, Regnase-1 has been predicted to possess other domains in the N- and C- terminal regions1617. However, the structure and function of the ZF domain, N-terminal domain (NTD) and C-terminal domain (CTD) of Regnase-1 have not been solved. Here, we performed structural and functional analyses of individual domains of Regnase-1 derived from Mus musculus in order to understand the catalytic activity in vitro. Our data revealed that the catalytic activity of Regnase-1 is regulated through both intra and intermolecular domain interactions in vitro. The NTD plays a crucial role in efficient cleavage of target mRNA, through intramolecular NTD-PIN interactions. Moreover, Regnase-1 functions as a dimer through intermolecular PIN-PIN interactions during cleavage of target mRNA. Our findings suggest that Regnase-1 cleaves its target mRNA by an NTD-activated functional PIN dimer, while the ZF increases RNA affinity in the vicinity of the PIN dimer. We analyzed Rengase-1 derived from Mus musculus and solved the structures of the four domains; NTD, PIN, ZF, and CTD individually by X-ray crystallography or NMR (Fig. 1a–e). X-ray crystallography was attempted for the fragment containing both the PIN and ZF domains, however, electron density was observed only for the PIN domain (Fig. 1c), consistent with a previous report on Regnase-1 derived from Homo sapiens11. This suggests that the PIN and ZF domains exist independently without interacting with each other. The domain structures of NTD, ZF, and CTD were determined by NMR (Fig. 1b,d,e). The NTD and CTD are both composed of three α helices, and structurally resemble ubiquitin conjugating enzyme E2 K (PDB ID: 3K9O) and ubiquitin associated protein 1 (PDB ID: 4AE4), respectively, according to the Dali server18. Although the PIN domain is responsible for the catalytic activity of Regnase-14, the roles of the other domains are largely unknown. First, we evaluated a role of the NTD and ZF domains for mRNA binding by an in vitro gel shift assay (Fig. 1f). Fluorescently 5′-labeled RNA corresponding to nucleotides 82–106 of the IL-6 mRNA 3′UTR and the catalytically inactive mutant (D226N and D244N) of Regnase-1—hereafter referred to as the DDNN mutant—were utilized. Upon addition of a larger amount of Regnase-1, the fluorescence of free RNA decreased, indicating that Regnase-1 bound to the RNA. Based on the decrease in the free RNA fluorescence band, we evaluated the contribution of each domain of Regnase-1 to RNA binding. While the RNA binding ability was not significantly changed in the presence of NTD, it increased in the presence of the ZF domain (Fig. 1f,g and Supplementary Fig. 1). Direct binding of the ZF domain and RNA were confirmed by NMR spectral changes. The fitting of the titration curve of Y314 resulted in an apparent dissociation constant (Kd) of 10 ± 1.1 μM (Supplementary Fig. 2). These results indicate that not only the PIN but also the ZF domain contribute to RNA binding, while the NTD is not likely to be involved in direct interaction with RNA. In order to characterize the role of each domain in the RNase activity of Regnase-1, we performed an in vitro cleavage assay using fluorescently 5′-labeled RNA corresponding to nucleotides 82–106 of the IL-6 mRNA 3′UTR (Fig. 1g). Regnase-1 constructs consisting of NTD-PIN-ZF completely cleaved the target mRNA and generated the cleaved products. The apparent half-life (T1/2) of the RNase activity was about 20 minutes. Regnase-1 lacking the ZF domain generated a smaller but appreciable amount of cleaved product (T1/2 ~ 70 minutes), while those lacking the NTD did not generate cleaved products (T1/2 > 90 minutes). It should be noted that NTD-PIN(DDNN)-ZF, which possesses the NTD but lacks the catalytic residues in PIN, completely lost all RNase activity (Fig. 1g, right panel), as expected, confirming that the RNase catalytic center is located in the PIN domain. Taken together with the results in the previous section, we conclude that the NTD is crucial for the RNase activity of Regnase-1 in vitro, although it does not contribute to the direct mRNA binding. During purification by gel filtration, the PIN domain exhibited extremely asymmetric elution peaks in a concentration dependent manner (Fig. 2a). By comparison with the elution volume of standard marker proteins, the PIN domain was assumed to be in equilibrium between a monomer and a dimer in solution at concentrations in the 20–200 μM range. The crystal structure of the PIN domain has been determined in three distinct crystal forms with a space group of P3121 (form I in this study and PDB ID 3V33), P3221 (form II in this study), and P41 (PDB ID 3V32 and 3V34), respectively11. We found that the PIN domain formed a head-to-tail oligomer that was commonly observed in all three crystal forms in spite of the different crystallization conditions (Supplementary Fig. 3). Mutation of Arg215, whose side chain faces to the opposite side of the oligomeric surface, to Glu preserved the monomer/dimer equilibrium, similar to the wild type. On the other hand, single mutations of side chains involved in the PIN–PIN oligomeric interaction resulted in monomer formation, judging from gel filtration (Fig. 2a,b). Wild type and monomeric PIN mutants (P212A and D278R) were also analyzed by NMR. The spectra indicate that the dimer interface of the wild type PIN domain were significantly broadened compared to the monomeric mutants (Supplementary Fig. 4). These results indicate that the PIN domain forms a head-to-tail oligomer in solution similar to the crystal structure. Interestingly, the monomeric PIN mutants P212A, R214A, and D278R had no significant RNase activity for IL-6 mRNA in vitro (Fig. 2c). The side chains of these residues point away from the catalytic center on the same molecule (Fig. 2b). Therefore, we concluded that head-to-tail PIN dimerization, together with the NTD, are required for Regnase-1 RNase activity in vitro. While the NTD does not contribute to RNA binding (Fig. 1f,g, and Supplementary Fig. 1), it increases the RNase activity of Regnase-1 (Fig. 1h). In order to gain insight into the molecular mechanism of the NTD-mediated enhancement of Regnase-1 RNase activity, we further investigated the domain-domain interaction between the NTD and the PIN domain using NMR. We used the catalytically inactive monomeric PIN mutant possessing both the DDNN and D278R mutations to avoid dimer formation of the PIN domain. The NMR signals from the PIN domain (residues V177, F210-T211, R214, F228-L232, and F234-S236) exhibited significant chemical shift changes upon addition of the NTD (Fig. 3a). Likewise, upon addition of the PIN domain, NMR signals derived from R56, L58-G59, and V86-H88 in the NTD exhibited large chemical shift changes and residues D53, F55, K57, Y60-S61, V68, T80-G83, L85, and G89 of the NTD as well as side chain amide signals of N79 exhibited small but appreciable chemical shift changes (Fig. 3b and Supplementary Fig. 5). These results clearly indicate a direct interaction between the PIN domain and the NTD. Based on the titration curve for the chemical shift changes of L58, the apparent Kd between the isolated NTD and PIN was estimated to be 110 ± 5.8 μM. Considering the fact that the NTD and PIN domains are attached by a linker, the actual binding affinity is expected much higher in the native protein. Mapping the residues with chemical shift changes reveals the putative PIN/NTD interface, which includes a helix that harbors catalytic residues D225 and D226 on the PIN domain (Fig. 3a). Interestingly, the putative binding site for the NTD overlaps with the PIN-PIN dimer interface, implying that NTD binding can “terminate” PIN-PIN oligomerization (Fig. 2b). An in silico docking of the NTD and PIN domains using chemical shift restraints provided a model consistent with the NMR experiments (Fig. 3c). To gain insight into the residues critical for Regnase-1 RNase activity, each basic or aromatic residue located around the catalytic site of the PIN oligomer was mutated to alanine, and the oligomerization and RNase activity were investigated (Fig. 4). From the gel filtration assays, all mutants except R214A formed dimers, suggesting that any lack of RNase activity in the mutants, except R214A, was directly due to mutational effects of the specific residues and not to abrogation of dimer formation. The W182A, R183A, and R214A mutants markedly lost cleavage activity for IL-6 mRNA as well as for Regnase-1 mRNA. The K184A, R215A, and R220A mutants moderately but significantly decreased the cleavage activity for both target mRNAs. The importance of K219 and R247 was slightly different for IL-6 and Regnase-1 mRNA; both K219 and R247 were more important in the cleavage of IL-6 mRNA than for Regnase-1 mRNA. The other mutated residues—K152, R158, R188, R200, K204, K206, K257, and R258—were not critical for RNase activity. The importance of residues W182 and R183 can readily be understood in terms of the monomeric PIN structure as they are located near to the RNase catalytic site; however, the importance of residue K184, which points away from the active site is more easily rationalized in terms of the oligomeric structure, in which the “secondary” chain’s residue K184 is positioned near the “primary” chain’s catalytic site (Fig. 4). In contrast, R214 is important for oligomerization of the PIN domain and the “secondary” chain’s residue R214 is also positioned near the “primary” chain’s active site within the dimer interface. It should be noted that the putative-RNA binding residues K184 and R214 are unique to Regnase-1 among PIN domains. Our mutational experiments indicated that the observed dimer is functional and that the role of the secondary PIN domain is to position Regnase-1-unique RNA binding residues near the active site of the primary PIN domain. If this model is correct, then we reasoned that a catalytically inactive PIN and a PIN lacking the putative RNA-binding residues ought to be inactive in isolation but become active when mixed together. In order to test this hypothesis, we performed in vitro cleavage assays using combinations of Regnase-1 mutants that had no or decreased RNase activities by themselves (Fig. 5). One group consisted of catalytically active PIN domains with mutation of basic residues found in the previous section to confer decreased RNase activity (Fig. 4). These were paired with a DDNN mutant that had no RNase activity by itself. When any members of the two groups are mixed, two kinds of heterodimers can be formed: one is composed of a DDNN primary PIN and a basic residue mutant secondary PIN and is expected to exhibit no RNase activity; the other is composed of a basic residue mutant primary PIN and a DDNN secondary PIN and is predicted to rescue RNase activity (Fig. 5a). When we compared the fluorescence intensity of uncleaved IL-6 mRNA, basic residue mutants W182A, K184A, R214A, and R220A were rescued upon addition of the DDNN mutant (Fig. 5b). Consistently, when we compared the fluorescence intensity of the uncleaved Regnase-1 mRNA, basic residue mutants K184A and R214A were rescued upon addition of the DDNN mutant (Fig. 5c). Rescue of K184A and R214A by the DDNN mutant was also confirmed by a significant increase in the cleaved products. This is particularly significant because the side chains of K184 and R214 in the primary PIN are oriented away from their own catalytic center, while those in the secondary PIN face toward the catalytic center of the primary PIN. R214 is an important residue for dimer formation as shown in Fig. 2, therefore, R214A in the secondary PIN cannot dimerize. According to the proposed model, an R214A PIN domain can only form a dimer when the DDNN PIN acts as the secondary PIN. Taken together, the rescue experiments above support the proposed model in which the head-to-tail dimer is functional in vitro. We determined the individual domain structures of Regnase-1 by NMR and X-ray crystallography. Although the function of the CTD remains elusive, we revealed the functions of the NTD, PIN, and ZF domains. A Regnase-1 construct consisting of PIN and ZF domains derived from Mus musculus was crystallized; however, the electron density of the ZF domain was low, indicating that the ZF domain is highly mobile in the absence of target mRNA or possibly other protein-protein interactions. Our NMR experiments confirmed direct binding of the ZF domain to IL-6 mRNA with a Kd of 10 ± 1.1 μM. Furthermore, an in vitro gel shift assay indicated that Regnase-1 containing the ZF domain enhanced target mRNA-binding, but the protein-RNA complex remained in the bottom of the well without entering into the polyacrylamide gel. These results indicate that Regnase-1 directly binds to RNA and precipitates under such experimental conditions. Due to this limitation, it is difficult to perform further structural analyses of mRNA-Regnase-1 complexes by X-ray crystallography or NMR. The previously reported crystal structure of the Regnase-1 PIN domain derived from Homo sapiens is nearly identical to the one derived from Mus musculus in this study, with a backbone RMSD of 0.2 Å. The amino acid sequences corresponding to PIN (residues 134–295) are the two non-identical residues are substituted with similar amino acids. Both the mouse and human PIN domains form head-to-tail oligomers in three distinct crystal forms. Rao and co-workers previously argued that PIN dimerization is likely to be a crystallographic artifact with no physiological significance, since monomers were dominant in their analytical ultra-centrifugation experiments11. In contrast, our gel filtration data, mutational analyses, and NMR spectra all indicate that the PIN domain forms a head-to-tail dimer in solution in a manner similar to the crystal structure. This inconsistency might be due to difference in the analytical methods and/or protein concentrations used in each experiment, since the oligomer formation of PIN was dependent on the protein concentration in our study. Single mutations to residues involved in the putative oligomeric interaction of PIN monomerized as expected and these mutants lost their RNase activity as well. Since the NMR spectra of monomeric mutants overlaps with those of the oligomeric forms, it is unlikely that the tertiary structure of the monomeric mutants were affected by the mutations. (Supplementary Fig. 4b,c). Based on these observations, we concluded that PIN-PIN dimer formation is critical for Regnase-1 RNase activity in vitro. Within the crystal structure of the PIN dimer, the Regnase-1 specific basic regions in both the “primary” and “secondary” PINs are located around the catalytic site of the primary PIN (Supplementary Fig. 6). Moreover, our structure-based mutational analyses showed these two Regnase-1 specific basic regions were essential for target mRNA cleavage in vitro. The cleavage assay also showed that the NTD is crucial for efficient mRNA cleavage. Moreover, we found that the NTD associates with the oligomeric surface of the primary PIN, docking to a helix that harbors its catalytic residues (Figs 2b and 3a). Taken together, this suggests that the NTD and the PIN domain compete for a common binding site. The affinity of the domain-domain interaction between two PIN domains (Kd = ~10 M) is similar to that of the NTD-PIN (Kd = 110 ± 5.8 μM) interactions; however, the covalent connection corresponding to residues 90–133 between the NTD and the primary PIN will greatly enhance the intramolecular domain interaction in the case of full-length Regnase-1. While further analyses are necessary to prove this point, our preliminary docking and molecular dynamics simulations indicate that NTD-binding rearranges the catalytic residues of the PIN domain toward an active conformation suitable for binding Mg. In this context, it is interesting that, in response to TCR stimulation, Malt1 cleaves Regnase-1 at R111 to control immune responses in vivo19. This result is consistent with a model in which the NTD acts as an enhancer, and cleavage of the linker lowers enzymatic activity dramatically. Based on these structural and functional analyses of Regnase-1 domain-domain interactions, we performed docking simulations of the NTD, PIN dimer, and IL-6 mRNA. We incorporated information from the cleavage site of IL-6 mRNA in vitro is indicated by denaturing polyacrylamide gel electrophoresis (Supplementary Fig. 7a,b). The docking result revealed multiple RNA binding modes that satisfied the experimental results in vitro (Supplementary Fig. 7c,d), however, it should be noted that, in vivo, there would likely be many other RNA-binding proteins that would protect loop regions from cleavage by Regnase-1. The overall model of regulation of Regnase-1 RNase activity through domain-domain interactions in vitro is summarized in Fig. 6. In the absence of target mRNA, the PIN domain forms head-to-tail oligomers at high concentration. A fully active catalytic center can be formed only when the NTD associates with the oligomer surface of the PIN domain, which terminates the head-to-tail oligomer formation in one direction (primary PIN), and forms a functional dimer together with the neighboring PIN (secondary PIN). While further investigations on the domain-domain interactions of Regnase-1 in vivo are necessary, these intramolecular and intermolecular domain interactions of Regnase-1 appear to structurally constrain Regnase-1activity, which, in turn, enables tight regulation of immune responses. The DNA fragment encoding Regnase-1 derived from Mus musculus was cloned into pGEX6p vector (GE Healthcare). All the mutants were generated by PCR-mediated site-directed mutagenesis and confirmed by the DNA sequence analyses. As a catalytically deficient mutant, both Asp226 and Asp244 at the catalytic center of PIN were mutated to Asn, which is referred to as DDNN mutant. Regnase-1 was expressed at 16 °C using the Escherichia coli Rosetta(DE3)pLysS strain. After purification with a GST-affinity resin, an N-terminal GST tag was digested by HRV-3 C protease. NTD was further purified by gel filtration chromatography using a HiLoad 16/60 Superdex 75 pg (GE Healthcare). The other domains were further purified by cation exchange chromatography using Resource S (GE Healthcare), followed by gel filtration chromatography using a HiLoad 16/60 Superdex 75 pg (GE Healthcare). Uniformly N or C, N-double labeled proteins for NMR experiments were prepared by growing E. coli host in M9 minimal medium containing NH4Cl, unlabeled glucose and N CELTONE Base Powder (CIL) or NH4Cl, C6-glucose, andC, N CELTONE Base Powder (CIL), respectively. Crystallization was performed using the sitting drop vapor diffusion method at 20 °C and two crystal forms (I and II) were obtained. In the case of form I crystals, drops (0.5 μl) of 6 mg/ml selenomethionine-labeled Regnase-1 PIN-ZF (residues 134–339 derived from Mus musculus) in 20 mM HEPES-NaOH (pH 6.8), 200 mM NaCl and 5 mM DTT were mixed with reservoir solution consisting of 1 M (NH4)2HPO4, 200 mM NaCl and 100 mM sodium citrate (pH 5.5) whereas in the case of form II crystals, drops (0.5 μl) of 6 mg/ml native Regnase-1 PIN-ZF (residues 134–339) in 20 mM HEPES-NaOH (pH 6.8), 200 mM NaCl and 5 mM DTT were mixed with reservoir solution consisting of 1.7 M NaCl and 100 mM HEPES-NaOH (pH 7.0). Diffraction data were collected at a Photon Factory Advanced Ring beamline NE3A (form I) or at a SPring-8 beamline BL41XU (form II), and were processed with HKL200020. The structure of the form I crystal was determined by the multiple anomalous dispersion (MAD) method. Nine Se sites were found using the program SOLVE21; however, the electron density obtained by MAD phases calculated using SOLVE was not good enough to build a model even after density modification using the program RESOLVE22. Then the program CNS23 was used to find additional three Se sites and calculate MAD phases using 12 Se sites. The electron density after density modification using CNS was good enough to build a model. Structure of the form II crystal was determined by the molecular replacement method using CNS and using the structure of the form I crystal as a search model. For all structures, further model building was performed manually with COOT24, and TLS and restrained refinement using isotropic individual B factors was performed with REFMAC525 in the CCP4 program suite26. Crystallographic parameters are summarized in Supplementary Table 1. All NMR experiments were carried out at 298 K on Inova 500-MHz, 600-MHz, and 800-MHz spectrometer (Agilent). The NMR data were processed using the NMRPipe27, the Olivia (fermi.pharm.hokudai.ac.jp/olivia/), and the Sparky program (Sparky3, University of California, San Francisco). For structure calculation, NOE distance restraints were obtained from 3D N-NOESY-HSQC (100 ms mixing time for the NTD, 75 ms mixing time for the ZF domain and the CTD) and C-NOESY-HSQC spectra (100 ms mixing time for the NTD, 75 ms mixing time for the ZF domain and the CTD). The NMR structures were determined using the CANDID/CYANA2.128. Dihedral restraints were derived from backbone chemical shifts using TALOS29. For the domain-domain interaction analyses between the NTD and the PIN domain, H-N HSQC spectra of uniformly N-labeled proteins in the concentration of 100 μM were obtained in the presence of 3 or 6 molar equivalents of unlabeled proteins. The fluorescently labeled RNAs at the 5′-end by 6-FAM were purchased from SIGMA-ALDORICH. The RNA sequences used in this study were shown below. IL-6 mRNA 3′UTR (82–106): 5′-UGUUGUUCUCUACGAAGAACUGACA-3′ (25 nts) Regnase-1 mRNA 3′UTR (191–211): 5′- CUGUUGAUACACAUUGUAUCU-3′ (21 nts) Catalytically deficient Regnase-1 proteins, containing DDNN mutations, and 5′-terminally 6-FAM labeled RNAs were incubated in the RNA-binding buffer (20 mM HEPES-NaOH (pH 6.8), 150 mM NaCl, 1 mM DTT, 10% glycerol (v/v), and 0.1% NP-40 (v/v)) at 4 °C for 30 minutes, then analyzed by non-denaturing polyacrylamide gel electrophoresis. The electrophoreses were performed at 4 °C using the 7.5% polyacrylamide (w/v) gel (monomer : bis = 29 : 1) in the electrophoresis buffer (25 mM Tris-HCl (pH 7.5) and 200 mM glycine). The fluorescence of 6-FAM labeled RNA was directly detected at the excitation wavelength of 460 nm with a fluorescence filter (Y515-Di) using a fluoroimaging analyzer (LAS-4000 (FUJIFILM)). The fluorescence intensity of each sample was quantified using ImageJ software. Regnase-1 (2 μM) and 5′-terminally 6-FAM labeled RNA (1 μM) were incubated in the RNA-cleavage buffer (20 mM Tris-HCl (pH 7.5), 150 mM NaCl, 5 mM MgCl2, and 1 mM DTT) at 37 °C. For the assay using combinations of Regnase-1 mutants, equimolar amounts of Regnase-1 mutants (2 μM each) were mixed with fluorescently labeled RNA (1 μM). After incubation for 30–120 minutes, the reaction was stopped by the addition of 1.5-fold volume of denaturing buffer containing 8 M urea and 100 mM EDTA, and samples were boiled. The electrophoreses were performed at room temperature using the 8 M urea containing denaturing gel with 20% polyacrylamide (w/v) (monomer : bis = 19 : 1) in 0.5 × TBE as the electrophoresis buffer. For docking NTD to PIN, OSCAR-star30 was first used to rebuild sidechains in the head-to-tail PIN dimer. Docking was carried out by surFit (http://sysimm.ifrec.osaka-u.ac.jp/docking/main/) with restraints obtained from NMR data (Fig. 3a,b) as follows. NTD: R56, L58, G59, V86, K87, H88; PIN: V177, F210, T211, R214, F228, I229, V230, K231, L232, F234, D235, S236. Top-scoring model was selected. For docking IL-6 mRNA 3′UTR to the PIN dimer, each domain of the PIN dimer structure was superimposed onto the PIN dimer of the human X-ray structure (PDB ID: 3V34) in order to graft both water molecules and Mg ions to the mouse model. Each IL-6 representative structure was submitted to the HADDOCK 2.0 server, for total of 10 independent jobs. In order to be consistent with the cleavage assay, active residues consisted of all nucleotides in RNA, Mg and W182, R183, K184, R188, R214, R215, K219, R220, and R247 in the protein. Docked models were selected based on the following criteria: one heavy atom within 7, 8, or 9 nucleotide from the 5′ end was <5 Å from the Mg ion on the primary PIN. Further classification was done manually in order to divide the selected models into two clusters. Accession codes: The crystal structure of the Regnase-1 PIN domain has been deposited in the Protein Data Bank (accession codes: 5H9V (Form I) and 5H9W (Form II)). The chemical shift assignments of the NTD, the ZF domain, and the CTD have been deposited at Biological Magnetic Resonance Bank (accession codes: 25718, 25719, and 25720, respectively), and the coordinates for the ensemble have been deposited in the Protein Data Bank (accession codes: 2N5J, 2N5K, and 2N5L, respectively). How to cite this article: Yokogawa, M. et al. Structural basis for the regulation of enzymatic activity of Regnase-1 by domain-domain interactions. Sci. Rep. 6, 22324; doi: 10.1038/srep22324 (2016).
PMC4831588	X-ray Crystallographic Structures of a Trimer, Dodecamer, and Annular Pore Formed by an Aβ17–36 β-Hairpin	High-resolution structures of oligomers formed by the β-amyloid peptide Aβ are needed to understand the molecular basis of Alzheimer’s disease and develop therapies. This paper presents the X-ray crystallographic structures of oligomers formed by a 20-residue peptide segment derived from Aβ. The development of a peptide in which Aβ17–36 is stabilized as a β-hairpin is described, and the X-ray crystallographic structures of oligomers it forms are reported. Two covalent constraints act in tandem to stabilize the Aβ17–36 peptide in a hairpin conformation: a δ-linked ornithine turn connecting positions 17 and 36 to create a macrocycle and an intramolecular disulfide linkage between positions 24 and 29. An N-methyl group at position 33 blocks uncontrolled aggregation. The peptide readily crystallizes as a folded β-hairpin, which assembles hierarchically in the crystal lattice. Three β-hairpin monomers assemble to form a triangular trimer, four trimers assemble in a tetrahedral arrangement to form a dodecamer, and five dodecamers pack together to form an annular pore. This hierarchical assembly provides a model, in which full-length Aβ transitions from an unfolded monomer to a folded β-hairpin, which assembles to form oligomers that further pack to form an annular pore. This model may provide a better understanding of the molecular basis of Alzheimer’s disease at atomic resolution.High-resolution structures of oligomers formed by the β-amyloid peptide Aβ are desperately needed to understand the molecular basis of Alzheimer’s disease and ultimately develop preventions or treatments. In Alzheimer’s disease, monomeric Aβ aggregates to form soluble low molecular weight oligomers, such as dimers, trimers, tetramers, hexamers, nonamers, and dodecamers, as well as high molecular weight aggregates, such as annular protofibrils. Over the last two decades the role of Aβ oligomers in the pathophysiology of Alzheimer’s disease has begun to unfold. Mouse models for Alzheimer’s disease have helped shape our current understanding about the Aβ oligomerization that precedes neurodegeneration. Aβ isolated from the brains of young plaque-free Tg2576 mice forms a mixture of low molecular weight oligomers. A 56 kDa soluble oligomer identified by SDS-PAGE was found to be especially important within this mixture. This oligomer was termed Aβ56 and appears to be a dodecamer of Aβ. Purified Aβ56 injected intercranially into healthy rats was found to impair memory, providing evidence that this Aβ oligomer may cause memory loss in Alzheimer’s disease. Smaller oligomers with molecular weights consistent with trimers, hexamers, and nonamers were also identified within the mixture of low molecular weight oligomers. Treatment of the mixture of low molecular weight oligomers with hexafluoroisopropanol resulted in the dissociation of the putative dodecamers, nonamers, and hexamers into trimers and monomers, suggesting that trimers may be the building block of the dodecamers, nonamers, and hexamers. Recently, Aβ trimers and Aβ56 were identified in the brains of cognitively normal humans and were found to increase with age. A type of large oligomers called annular protofibrils (APFs) have also been observed in the brains of transgenic mice and isolated from the brains of Alzheimer’s patients. APFs were first discovered in vitro using chemically synthesized Aβ that aggregated into porelike structures that could be observed by atomic force microscopy (AFM) and transmission electron microscopy (TEM). The sizes of APFs prepared in vitro vary among different studies. Lashuel et al. observed APFs with an outer diameter that ranged from 7–10 nm and an inner diameter that ranged from 1.5–2 nm, consistent with molecular weights of 150–250 kDa. Quist et al. observed APFs with an outer diameter of 16 nm embedded in a lipid bilayer. Kayed et al. observed APFs with an outer diameter that ranged from 8–25 nm, which were composed of small spherical Aβ oligomers, 3–5 nm in diameter. Although the APFs in these studies differ in size, they share a similar annular morphology and appear to be composed of smaller oligomers. APFs have also been observed in the brains of APP23 transgenic mice by immunofluorescence with an anti-APF antibody and were found to accumulate in neuronal processes and synapses. In a subsequent study, APFs were isolated from the brains of Alzheimer’s patients by immunoprecipitation with an anti-APF antibody. These APFs had an outer diameter that ranged from 11–14 nm and an inner diameter that ranged from 2.5–4 nm. Dimers of Aβ have also been isolated from the brains of Alzheimer’s patients. Aβ dimers inhibit long-term potentiation in mice and promote hyperphosphorylation of the microtubule-associated protein tau, leading to neuritic damage. Aβ dimers have only been isolated from human or transgenic mouse brains that contain the pathognomonic fibrillar Aβ plaques associated with Alzheimer’s disease. Furthermore, the endogenous rise of Aβ dimers in the brains of Tg2576 and J20 transgenic mice coincides with the deposition of Aβ plaques. These observations suggest that the Aβ trimers, hexamers, dodecamers, and related assemblies may be associated with presymptomatic neurodegeneration, while Aβ dimers are more closely associated with fibril formation and plaque deposition during symptomatic Alzheimer’s disease. The approach of isolating and characterizing Aβ oligomers has not provided any high-resolution structures of Aβ oligomers. Techniques such as SDS-PAGE, TEM, and AFM have only provided information about the molecular weights, sizes, morphologies, and stoichiometry of Aβ oligomers. High-resolution structural studies of Aβ have primarily focused on Aβ fibrils and Aβ monomers. Solid-state NMR spectroscopy studies of Aβ fibrils revealed that Aβ fibrils are generally composed of extended networks of in-register parallel β-sheets. X-ray crystallographic studies using fragments of Aβ have provided additional information about how Aβ fibrils pack. Solution-phase NMR and solid-state NMR have been used to study the structures of the Aβ monomers within oligomeric assemblies. A major finding from these studies is that oligomeric assemblies of Aβ are primarily composed of antiparallel β-sheets. Many of these studies have reported the monomer subunit as adopting a β-hairpin conformation, in which the hydrophobic central and C-terminal regions form an antiparallel β-sheet. In 2008, Hoyer et al. reported the NMR structure of an Aβ monomer bound to an artificial binding protein called an affibody (PDB 2OTK). The structure revealed that monomeric Aβ forms a β-hairpin when bound to the affibody. This Aβ β-hairpin encompasses residues 17–37 and contains two β-strands comprising Aβ17–24 and Aβ30–37 connected by an Aβ25–29 loop. Sequestering Aβ within the affibody prevents its fibrilization and reduces its neurotoxicity, providing evidence that the β-hairpin structure may contribute to the ability of Aβ to form neurotoxic oligomers. In a related study, Sandberg et al. constrained Aβ in a β-hairpin conformation by mutating residues A21 and A30 to cysteine and forming an intramolecular disulfide bond. Locking Aβ into a β-hairpin structure resulted in the formation Aβ oligomers, which were observed by size exclusion chromatography (SEC) and SDS-PAGE. The oligomers with a molecular weight of ∼100 kDa that were isolated by SEC were toxic toward neuronally derived SH-SY5Y cells. This study provides evidence for the role of β-hairpin structure in Aβ oligomerization and neurotoxicity. Inspired by these β-hairpin structures, our laboratory developed a macrocyclic β-sheet peptide derived from Aβ17–36 designed to mimic an Aβ β-hairpin and reported its X-ray crystallographic structure. This peptide (peptide 1) consists of two β-strands comprising Aβ17–23 and Aβ30–36 covalently linked by two δ-linked ornithine (Orn) β-turn mimics. The Orn that connects residues D23 and A30 replaces the Aβ24–29 loop. The Orn that connects residues L17 and V36 enforces β-hairpin structure. We incorporated an N-methyl group at position G33 to prevent uncontrolled aggregation and precipitation of the peptide. To improve the solubility of the peptide we replaced M35 with the hydrophilic isostere of methionine, ornithine (α-linked) (Figure 1B). The X-ray crystallographic structure of peptide 1 reveals that it folds to form a β-hairpin that assembles to form trimers and that the trimers further assemble to form hexamers and dodecamers. (A) Cartoon illustrating the design of peptides 1 and 2 and their relationship to an Aβ17–36 β-hairpin. (B) Chemical structure of peptide 1 illustrating Aβ17–23 and Aβ30–36, M35Orn, the N-methyl group, and the δ-linked ornithine turns. (C) Chemical structure of peptide 2 illustrating Aβ17–36, the N-methyl group, the disulfide bond across positions 24 and 29, and the δ-linked ornithine turn. Our design of peptide 1 omitted the Aβ24–29 loop. To visualize the Aβ24–29 loop, we performed replica-exchange molecular dynamics (REMD) simulations on Aβ17–36 using the X-ray crystallographic coordinates of Aβ17–23 and Aβ30–36 from peptide 1. These studies provided a working model for a trimer of Aβ17–36 β-hairpins and demonstrated that the trimer should be capable of accommodating the Aβ24–29 loop. In the current study we set out to restore the Aβ24–29 loop, reintroduce the methionine residue at position 35, and determine the X-ray crystallographic structures of oligomers that form. We designed peptide 2 as a homologue of peptide 1 that embodies these ideas. Peptide 2 contains a methionine residue at position 35 and an Aβ24–29 loop with residues 24 and 29 (Val and Gly) mutated to cysteine and linked by a disulfide bond (Figure 1C). Here, we describe the development of peptide 2 and report the X-ray crystallographic structures of the trimer, dodecamer, and annular pore observed within the crystal structure. We developed peptide 2 from peptide 1 by an iterative process, in which we first attempted to restore the Aβ24–29 loop without a disulfide linkage. We envisioned peptide 3 as a homologue of peptide 1 with the Aβ24–29 loop in place of the Orn that connects D23 and A30 and p-iodophenylalanine (F) in place of F19. We routinely use p-iodophenylalanine to determine the X-ray crystallographic phases. After determining the X-ray crystallographic structure of the p-iodophenylalanine variant we attempt to determine the structure of the native phenylalanine compound by isomorphous replacement. Upon synthesizing peptide 3, we found that it formed an amorphous precipitate in most crystallization conditions screened and failed to afford crystals in any condition. We postulate that the loss of the Orn constraint leads to conformational heterogeneity that prevents peptide 3 from crystallizing. To address this issue, we next incorporated a disulfide bond between residues 24 and 29 as a conformational constraint that serves as a surrogate for Orn. We designed peptide 4 to embody this idea, mutating Val24 and Gly29 to cysteine and forming an interstrand disulfide linkage. We mutated these residues because they occupy the same position as the Orn that connects D23 and A30 in peptide 1. Residues V24 and G29 form a non-hydrogen-bonded pair, which can readily accommodate disulfide linkages in antiparallel β-sheets. Disulfide bonds across non-hydrogen-bonded pairs stabilize β-hairpins, while disulfide bonds across hydrogen-bonded pairs do not. Although the disulfide bond between positions 24 and 29 helps stabilize the β-hairpin, it does not alter the charge or substantially change the hydrophobicity of the Aβ17–36 β-hairpin. We were gratified to find that peptide 4 afforded crystals suitable for X-ray crystallography. As the next step in the iterative process, we determined the X-ray crystallographic structure of this peptide (PDB 5HOW). After determining the X-ray crystallographic structure of peptide 4 we reintroduced the native phenylalanine at position 19 and the methionine at position 35 to afford peptide 2. We completed the iterative process—from 1 to 3 to 4 to 2—by successfully determining the X-ray crystallographic structure of peptide 2 (PDB 5HOX and 5HOY). The following sections describe the synthesis of peptides 2–4 and the X-ray crystallographic structure of peptide 2. We synthesized peptides 2–4 by similar procedures to those we have developed for other macrocyclic peptides. Our laboratory routinely prepares macrocyclic peptides by solid-phase synthesis of the corresponding linear peptide on 2-chlorotrityl resin, followed by cleavage of the protected linear peptide from the resin, solution-phase macrolactamization, and deprotection of the resulting macrocyclic peptide. In synthesizing peptides 2 and 4 we formed the disulfide linkage after macrolactamization and deprotection of the acid-labile side chain protecting groups. We used acid-stable Acm-protected cysteine residues at positions 24 and 29 and removed the Acm groups by oxidation with I2 in aqueous acetic acid to afford the disulfide linkage. Peptides 2–4 were purified by RP-HPLC. We screened crystallization conditions for peptide 4 in a 96-well-plate format using three different Hampton Research crystallization kits (Crystal Screen, Index, and PEG/Ion) with three ratios of peptide and mother liquor per condition (864 experiments). Peptide 4 afforded crystals in a single set of conditions containing HEPES buffer and Jeffamine M-600—the same crystallization conditions that afforded crystals of peptide 1. Peptide 2 also afforded crystals in these conditions. We further optimized these conditions to rapidly (∼72 h) yield crystals suitable for X-ray crystallography. The optimized conditions consist of 0.1 M HEPES at pH 6.4 with 31% Jeffamine M-600 for peptide 4 and 0.1 M HEPES pH 7.1 with 29% Jeffamine M-600 for peptide 2. Crystal diffraction data for peptides 4 and 2 were collected in-house with a Rigaku MicroMax 007HF X-ray diffractometer at 1.54 Å wavelength. Crystal diffraction data for peptide 2 were also collected at the Advanced Light Source at Lawrence Berkeley National Laboratory with a synchrotron source at 1.00 Å wavelength to achieve higher resolution. Data from peptides 4 and 2 suitable for refinement at 2.30 Å were obtained from the diffractometer; data from peptide 2 suitable for refinement at 1.90 Å were obtained from the synchrotron. Data for peptides 4 and 2 were scaled and merged using XDS. Phases for peptide 4 were determined by single-wavelength anomalous diffraction (SAD) phasing by using the coordinates of the iodine anomalous signal from p-iodophenylalanine. Phases for peptide 2 were determined by isomorphous replacement of peptide 4. The structures of peptides 2 and 4 were solved and refined in the P6122 space group. Coordinates for hydrogens were generated by phenix.refine during refinement. The asymmetric unit of each peptide consists of six monomers, arranged as two trimers. Peptides 2 and 4 form morphologically identical structures and assemblies in the crystal lattice. The X-ray crystallographic structure of peptide 2 reveals that it folds to form a twisted β-hairpin comprising two β-strands connected by a loop (Figure 2A). Eight residues make up each surface of the β-hairpin: L17, F19, A21, D23, A30, I32, L34, and V36 make up one surface; V18, F20, E22, C24, C29, I31, G33, and M35 make up the other surface. The β-strands of the monomers in the asymmetric unit are virtually identical, differing primarily in rotamers of F20, E22, C24, C29, I31, and M35 (Figure S1). The disulfide linkages suffered radiation damage under synchrotron radiation. We refined three of the β-hairpins with intact disulfide linkages and three with thiols to represent cleaved disulfide linkages in the synchrotron data set (PDB 5HOX). No evidence for cleavage of the disulfides was observed in the refinement of the data set collected on the X-ray diffractometer, and we refined all disulfide linkages as intact (PDB 5HOY). X-ray crystallographic structure of peptide 2 (PDB 5HOX, synchrotron data set). (A) X-ray crystallographic structure of a representative β-hairpin monomer formed by peptide 2. (B) Overlay of the six β-hairpin monomers in the asymmetric unit. The β-hairpins are shown as cartoons to illustrate the differences in the Aβ25–28 loops. The Aβ25–28 loops of the six monomers within the asymmetric unit vary substantially in backbone geometry and side chain rotamers (Figures 2B and S1). The electron density for the loops is weak and diffuse compared to the electron density for the β-strands. The B values for the loops are large, indicating that the loops are dynamic and not well ordered. Thus, the differences in backbone geometry and side chain rotamers among the loops are likely of little significance and should be interpreted with caution. Peptide 2 assembles into oligomers similar in morphology to those formed by peptide 1. Like peptide 1, peptide 2 forms a triangular trimer, and four trimers assemble to form a dodecamer. In the higher-order assembly of the dodecamers formed by peptide 2 a new structure emerges, not seen in peptide 1, an annular pore consisting of five dodecamers. Peptide 2 forms a trimer, much like that which we observed previously for peptide 1, in which three β-hairpins assemble to form an equilateral triangle (Figure 3A). The trimer maintains all of the same stabilizing contacts as those of peptide 1. Hydrogen bonding and hydrophobic interactions between residues on the β-strands comprising Aβ17–23 and Aβ30–36 stabilize the core of the trimer. The disulfide bonds between residues 24 and 29 are adjacent to the structural core of the trimer and do not make any substantial intermolecular contacts. Two crystallographically distinct trimers comprise the peptide portion of the asymmetric unit. The two trimers are almost identical in structure, differing slightly among side chain rotamers and loop conformations. X-ray crystallographic structure of the trimer formed by peptide 2. (A) Triangular trimer. The three water molecules in the center hole of the trimer are shown as spheres. (B) Detailed view of the intermolecular hydrogen bonds between the main chains of V18 and E22 and Orn and C24, at the three corners of the triangular trimer. (C) The F19 face of the trimer, with key side chains shown as spheres. (D) The F20 face of the trimer, with key side chains as spheres. A network of 18 intermolecular hydrogen bonds helps stabilize the trimer. At the corners of the trimer, the pairs of β-hairpin monomers form four hydrogen bonds: two between the main chains of V18 and E22 and two between Orn and the main chain of C24 (Figure 3B). Three ordered water molecules fill the hole in the center of the trimer, hydrogen bonding to each other and to the main chain of F20 (Figure 3A). Hydrophobic contacts between residues at the three corners of the trimer, where the β-hairpins meet, further stabilize the trimer. At each corner, the side chains of residues L17, F19, and V36 of one β-hairpin pack against the side chains of residues A21, I32, L34, and also D23 of the adjacent β-hairpin to create a hydrophobic cluster (Figure 3C). The three hydrophobic clusters create a large hydrophobic surface on one face of the trimer. The other face of the trimer displays a smaller hydrophobic surface, which includes the side chains of residues V18, F20, and I31 of the three β-hairpins (Figure 3D). In subsequent discussion, we designate the former surface the “F19 face” and the latter surface the “F20 face”. Four trimers assemble to form a dodecamer. The four trimers arrange in a tetrahedral fashion, creating a central cavity inside the dodecamer. Because each trimer is triangular, the resulting arrangement resembles an octahedron. Each of the 12 β-hairpins constitutes an edge of the octahedron, and the triangular trimers occupy four of the eight faces of the octahedron. Figure 4A illustrates the octahedral shape of the dodecamer. Figure 4B illustrates the tetrahedral arrangement of the four trimers. X-ray crystallographic structure of the dodecamer formed by peptide 2. (A) View of the dodecamer that illustrates the octahedral shape. (B) View of the dodecamer that illustrates the tetrahedral arrangement of the four trimers that comprise the dodecamer. (C) View of two trimer subunits from inside the cavity of the dodecamer. Residues L17, L34, and V36 are shown as spheres, illustrating the hydrophobic packing that occurs at the six vertices of the dodecamer. (D) Detailed view of one of the six vertices of the dodecamer. The F19 faces of the trimers line the interior of the dodecamer. At the six vertices, hydrophobic packing between the side chains of L17, L34, and V36 helps stabilize the dodecamer (Figures 4C and D). Salt bridges between the side chains of D23 and Orn at the vertices further stabilize the dodecamer. Each of the six vertices includes two Aβ25–28 loops that extend past the core of the dodecamer without making any substantial intermolecular contacts. The exterior of the dodecamer displays four F20 faces (Figure S3). In the crystal lattice, each F20 face of one dodecamer packs against an F20 face of another dodecamer. Although the asymmetric unit comprises half a dodecamer, the crystal lattice may be thought of as being built of dodecamers. The electron density map for the X-ray crystallographic structure of peptide 2 has long tubes of electron density inside the central cavity of the dodecamer. The shape and length of the electron density is consistent with the structure of Jeffamine M-600, which is an essential component of the crystallization conditions. Jeffamine M-600 is a polypropylene glycol derivative with a 2-methoxyethoxy unit at one end and a 2-aminopropyl unit at the other end. Its average molecular weight is about 600 Da, which corresponds to nine propylene glycol units. Although Jeffamine M-600 is a heterogeneous mixture with varying chain lengths and stereochemistry, we modeled a single stereoisomer with nine propylene glycol units (n = 9) to fit the electron density. The Jeffamine M-600 appears to stabilize the dodecamer by occupying the central cavity and making hydrophobic contacts with residues lining the cavity (Figure S3). In a dodecamer formed by full-length Aβ, the hydrophobic C-terminal residues (Aβ37–40 or Aβ37–42) might play a similar role in filling the dodecamer and thus create a packed hydrophobic core within the central cavity of the dodecamer. Five dodecamers assemble to form an annular porelike structure (Figure 5A). Hydrophobic packing between the F20 faces of trimers displayed on the outer surface of each dodecamer stabilizes the porelike assembly. Two morphologically distinct interactions between trimers occur at the interfaces of the five dodecamers: one in which the trimers are eclipsed (Figure 5B), and one in which the trimers are staggered (Figure 5C). Hydrophobic packing between the side chains of F20, I31, and E22 stabilizes these interfaces (Figure 5D and E). The annular pore contains three eclipsed interfaces and two staggered interfaces. The eclipsed interfaces occur between dodecamers 1 and 2, 1 and 5, and 3 and 4, as shown in Figure 5A. The staggered interfaces occur between dodecamers 2 and 3 and 4 and 5. The annular pore is not completely flat, instead, adopting a slightly puckered shape, which accommodates the eclipsed and staggered interfaces. Ten Aβ25–28 loops from the vertices of the five dodecamers line the hole in the center of the pore. The hydrophilic side chains of S26, N27, and K28 decorate the hole. X-ray crystallographic structure of the annular pore formed by peptide 2. (A) Annular porelike structure illustrating the relationship of the five dodecamers that form the pore (top view). (B) Eclipsed interface between dodecamers 1 and 2 (side view). The same eclipsed interface also occurs between dodecamers 1 and 5 and 3 and 4. (C) Staggered interface between dodecamers 2 and 3 (side view). The same staggered interface also occurs between dodecamers 4 and 5. (D) Eclipsed interface between dodecamers 1 and 5 (top view). Residues F20, I31, and E22 are shown as spheres to detail the hydrophobic packing. (E) Staggered interface between dodecamers 2 and 3 (top view). Residues F20, I31, and E22 are shown as spheres to detail the hydrophobic packing. The annular pore is comparable in size to other large protein assemblies. The outer diameter is ∼11–12 nm. The diameter of the hole in the center of the pore is ∼2 nm. The thickness of the pore is ∼5 nm, which is comparable to that of a lipid bilayer membrane. It is important to note that the annular pore formed by peptide 2 is not a discrete unit in the crystal lattice. Rather, the crystal lattice is composed of conjoined annular pores in which all four F20 faces on the surface of each dodecamer contact F20 faces on other dodecamers (Figure S4). The crystal lattice shows how the dodecamers can further assemble to form larger structures. Each dodecamer may be thought of as a tetravalent building block with the potential to assemble on all four faces to form higher-order supramolecular assemblies. The X-ray crystallographic study of peptide 2 described here provides high-resolution structures of oligomers formed by an Aβ17–36 β-hairpin. The crystallographic assembly of peptide 2 into a trimer, dodecamer, and annular pore provides a model for the assembly of the full-length Aβ peptide to form oligomers. In this model Aβ folds to form a β-hairpin comprising the hydrophobic central and C-terminal regions. Three β-hairpins assemble to form a trimer, and four trimers assemble to form a dodecamer. The dodecamers further assemble to form an annular pore (Figure 6). Model for the hierarchical assembly of an Aβ β-hairpin into a trimer, dodecamer, and annular pore based on the crystallographic assembly of peptide 2. Monomeric Aβ folds to form a β-hairpin in which the hydrophobic central and C-terminal regions form an antiparallel β-sheet. Three β-hairpin monomers assemble to form a triangular trimer. Four triangular trimers assemble to form a dodecamer. Five dodecamers assemble to form an annular pore. The molecular weights shown correspond to an Aβ42 monomer (∼4.5 kDa), an Aβ42 trimer (∼13.5 kDa), an Aβ42 dodecamer (∼54 kDa), and an Aβ42 annular pore composed of five dodecamers (∼270 kDa). The model put forth in Figure 6 is consistent with the current understanding of endogenous Aβ oligomerization and explains at atomic resolution many key observations about Aβ oligomers. Two general types of endogenous Aβ oligomers have been observed: Aβ oligomers that occur on a pathway to fibrils, or “fibrillar oligomers”, and Aβ oligomers that evade a fibrillar fate, or “nonfibrillar oligomers”. Fibrillar oligomers accumulate in Alzheimer’s disease later than nonfibrillar oligomers and coincide with the deposition of plaques. Nonfibrillar oligomers accumulate early in Alzheimer’s disease before plaque deposition. Fibrillar and nonfibrillar oligomers have structurally distinct characteristics, which are reflected in their reactivity with the fibril-specific OC antibody and the oligomer-specific A11 antibody. Fibrillar oligomers are recognized by the OC antibody but not the A11 antibody, whereas nonfibrillar oligomers are recognized by the A11 antibody but not the OC antibody. These criteria have been used to classify the Aβ oligomers that accumulate in vivo. Aβ dimers have been classified as fibrillar oligomers, whereas Aβ trimers, Aβ56, and APFs have been classified as nonfibrillar oligomers. Larson and Lesné proposed a model for the endogenous production of nonfibrillar oligomers that explains these observations. In this model, folded Aβ monomer assembles into a trimer, the trimer further assembles into hexamers and dodecamers, and the dodecamers further assemble to form annular protofibrils. The hierarchical assembly of peptide 2 is consistent with this model; and the trimer, dodecamer, and annular pore formed by peptide 2 may share similarities to the trimers, Aβ56, and APFs observed in vivo. At this point, we can only speculate whether the trimer and dodecamer formed by peptide 2 share structural similarities to Aβ trimers and Aβ56, as little is known about the structure of Aβ trimers and Aβ56. The crystallographically observed annular pore formed by peptide 2 is morphologically similar to the APFs formed by full-length Aβ. The annular pore formed by peptide 2 is comparable in size to the APFs prepared in vitro or isolated from Alzheimer’s brains (Figure 7 and Table 1). The varying sizes of APFs formed by full-length Aβ might result from differences in the number of oligomer subunits comprising each APF. Although the annular pore formed by peptide 2 contains five dodecamer subunits, pores containing fewer or more subunits can easily be envisioned. The dodecamers that comprise the annular pore exhibit two modes of assembly—eclipsed interactions and staggered interactions between the F20 faces of trimers within dodecamers. These two modes of assembly might reflect a dynamic interaction between dodecamers, which could permit assemblies of more dodecamers into larger annular pores. Surface views of the annular pore formed by peptide 2. (A) Top view. (B) Side view. Dot blot analysis shows that peptide 2 is reactive toward the A11 antibody (Figure S5). This reactivity suggests that peptide 2 forms oligomers in solution that share structural similarities to the nonfibrillar oligomers formed by full-length Aβ. Further studies are needed to elucidate the species that peptide 2 forms in solution and to study their biological properties. This is an active area of research in our laboratory. Preliminary attempts to study these species by SEC and SDS-PAGE have not provided a clear measure of the structures formed in solution. The difficulty in studying the oligomers formed in solution may reflect the propensity of the dodecamer to assemble on all four F20 faces. The X-ray crystallographic structure and A11 reactivity of peptide 2 support the model proposed by Larsen and Lesné and suggest that β-hairpins constitute a fundamental building block for nonfibrillar oligomers. What makes β-hairpins special is that three β-hairpins can nestle together to form trimers, stabilized by a network of hydrogen bonds and hydrophobic interactions. This mode of assembly is not unique to Aβ. The foldon domain of bacteriophage T4 fibritin is composed of three β-hairpins that assemble into a triangular trimer similar to the triangular trimer formed by peptide 2. Additionally, our research group has observed a similar assembly of a β-hairpin peptide derived from β2-microglobulin. Although we began these studies with a relatively simple hypothesis—that the trimers and dodecamers formed by peptide 1 could accommodate the Aβ24–29 loop—an even more exciting finding has emerged—that the dodecamers can assemble to form annular pores. This finding could not have been anticipated from the X-ray crystallographic structure of peptide 1 and reveals a new level of hierarchical assembly that recapitulates micrographic observations of annular protofibrils. The crystallographically observed dodecamer, in turn, recapitulates the observation of Aβ56, which appears to be a dodecamer of Aβ. The crystallographically observed trimer recapitulates the Aβ trimers that are observed even before the onset of symptoms in Alzheimer’s disease. Our approach of constraining Aβ17–36 into a β-hairpin conformation and blocking aggregation with an N-methyl group has allowed us to crystallize a large fragment of what is generally considered to be an uncrystallizable peptide. We believe this iterative, “bottom up” approach of identifying the minimal modification required to crystallize Aβ peptides will ultimately allow larger fragments of Aβ to be crystallized, thus providing greater insights into the structures of Aβ oligomers.
PMC4871749	The Taf14 YEATS domain is a reader of histone crotonylation	The discovery of new histone modifications is unfolding at startling rates, however, the identification of effectors capable of interpreting these modifications has lagged behind. Here we report the YEATS domain as an effective reader of histone lysine crotonylation – an epigenetic signature associated with active transcription. We show that the Taf14 YEATS domain engages crotonyllysine via a unique π-π-π-stacking mechanism and that other YEATS domains have crotonyllysine binding activity.Crotonylation of lysine residues (crotonyllysine, Kcr) has emerged as one of the fundamental histone post-translational modifications (PTMs) found in mammalian chromatin. This epigenetic PTM is widespread and enriched at active gene promoters and potentially enhancers. The crotonyllysine mark on histone H3K18 is produced by p300, a histone acetyltransferase also responsible for acetylation of histones. Owing to some differences in their genomic distribution, the crotonyllysine and acetyllysine (Kac) modifications have been linked to distinct functional outcomes. p300-catalyzed histone crotonylation, which is likely metabolically regulated, stimulates transcription to a greater degree than p300-catalyzed acetylation. The discovery of individual biological roles for the crotonyllysine and acetyllysine marks suggests that these PTMs can be read by distinct readers. While a number of acetyllysine readers have been identified and characterized, a specific reader of the crotonyllysine mark remains unknown (reviewed in). A recent survey of bromodomains (BDs) demonstrates that only one BD associates very weakly with a crotonylated peptide, however it binds more tightly to acetylated peptides, inferring that bromodomains do not possess physiologically relevant crotonyllysine binding activity. The family of acetyllysine readers has been expanded with the discovery that the YEATS (Yaf9, ENL, AF9, Taf14, Sas5) domains of human AF9 and yeast Taf14 are capable of recognizing the histone mark H3K9ac. The acetyllysine binding function of the AF9 YEATS domain is essential for the recruitment of the histone methyltransferase DOT1L to H3K9ac-containing chromatin and for DOT1L-mediated H3K79 methylation and transcription. Similarly, activation of a subset of genes and DNA damage repair in yeast require the acetyllysine binding activity of the Taf14 YEATS domain. Consistent with its role in gene regulation, Taf14 was identified as a core component of the transcription factor complexes TFIID and TFIIF. However, Taf14 is also found in a number of chromatin-remodeling complexes (i.e., INO80, SWI/SNF and RSC) and the histone acetyltransferase complex NuA3, indicating a multifaceted role of Taf14 in transcriptional regulation and chromatin biology. In this study, we identified the Taf14 YEATS domain as a reader of crotonyllysine that binds to histone H3 crotonylated at lysine 9 (H3K9cr) via a distinctive binding mechanism. We found that H3K9cr is present in yeast and is dynamically regulated. To elucidate the molecular basis for recognition of the H3K9cr mark, we obtained a crystal structure of the Taf14 YEATS domain in complex with H3K9cr5-13 (residues 5–13 of H3) peptide (Fig. 1, Supplementary Results, Supplementary Fig. 1 and Supplementary Table 1). The Taf14 YEATS domain adopts an immunoglobin-like β sandwich fold containing eight anti-parallel β strands linked by short loops that form a binding site for H3K9cr (Fig. 1b). The H3K9cr peptide lays in an extended conformation in an orientation orthogonal to the β strands and is stabilized through an extensive network of direct and water-mediated hydrogen bonds and a salt bridge (Fig. 1c). The most striking feature of the crotonyllysine recognition mechanism is the unique coordination of crotonylated lysine residue. The fully extended side chain of K9cr transverses the narrow tunnel, crossing the β sandwich at right angle in a corkscrew-like manner (Fig. 1b and Supplementary Figure 1b). The planar crotonyl group is inserted between Trp81 and Phe62 of the protein, the aromatic rings of which are positioned strictly parallel to each other and at equal distance from the crotonyl group, yielding a novel aromatic-amide/aliphatic-aromatic π-π-π-stacking system that, to our knowledge, has not been reported previously for any protein-protein interaction (Fig. 1d and Supplementary Fig. 1c). The side chain of Trp81 appears to adopt two conformations, one of which provides maximum π-stacking with the alkene functional group while the other rotamer affords maximum π-stacking with the amide π electrons (Supplementary Fig. 1c). The dual conformation of Trp81 is likely due to the conjugated nature of the C=C and C=O π-orbitals within the crotonyl functional group. In addition to π-π-π stacking, the crotonyl group is stabilized by a set of hydrogen bonds and electrostatic interactions. The π bond conjugation of the crotonyl group gives rise to a dipole moment of the alkene moiety, resulting in a partial positive charge on the β-carbon (Cβ) and a partial negative charge on the α-carbon (Cα). This provides the capability for the alkene moiety to form electrostatic contacts, as Cα and Cβ lay within electrostatic interaction distances of the carbonyl oxygen of Gln79 and of the hydroxyl group of Thr61, respectively. The hydroxyl group of Thr61 also participates in a hydrogen bond with the amide nitrogen of the K9cr side chain (Fig. 1d). The fixed position of the Thr61 hydroxyl group, which facilitates interactions with both the amide and Cα of K9cr, is achieved through a hydrogen bond with imidazole ring of His59. Extra stabilization of K9cr is attained by a hydrogen bond formed between its carbonyl oxygen and the backbone nitrogen of Trp81, as well as a water-mediated hydrogen bond with the backbone carbonyl group of Gly82 (Fig 1d). This distinctive mechanism was corroborated through mapping the Taf14 YEATS-H3K9cr binding interface in solution using NMR chemical shift perturbation analysis (Supplementary Fig. 2a, b). Binding of the Taf14 YEATS domain to H3K9cr is robust. The dissociation constant (Kd) for the Taf14 YEATS-H3K9cr5-13 complex was found to be 9.5 μM, as measured by fluorescence spectroscopy (Supplementary Fig. 2c). This value is in the range of binding affinities exhibited by the majority of histone readers, thus attesting to the physiological relevance of the H3K9cr recognition by Taf14. To determine whether H3K9cr is present in yeast, we generated whole cell extracts from logarithmically growing yeast cells and subjected them to Western blot analysis using antibodies directed towards H3K9cr, H3K9ac and H3 (Fig. 2a, b, Supplementary Fig. 3 and Supplementary Table 2). Both H3K9cr and H3K9ac were detected in yeast histones; to our knowledge, this is the first report of H3K9cr occurring in yeast. We next asked if H3K9cr is regulated by the actions of histone acetyltransferases (HATs) and histone deacetylases (HDACs). Towards this end, we probed extracts derived from yeast cells in which major yeast HATs (HAT1, Gcn5, and Rtt109) or HDACs (Rpd3, Hos1, and Hos2) were deleted. As shown in Figure 2a, b and Supplementary Fig. 3e, H3K9cr levels were abolished or reduced considerably in the HAT deletion strains, whereas they were dramatically increased in the HDAC deletion strains. Furthermore, fluctuations in the H3K9cr levels were more substantial than fluctuations in the corresponding H3K9ac levels. Together, these results reveal that H3K9cr is a dynamic mark of chromatin in yeast and suggest an important role for this modification in transcription as it is regulated by HATs and HDACs. We have previously shown that among acetylated histone marks, the Taf14 YEATS domain prefers acetylated H3K9 (also see Supplementary Fig. 3b), however it binds to H3K9cr tighter. The selectivity of Taf14 towards crotonyllysine was substantiated by H,N HSQC experiments, in which either H3K9cr5-13 or H3K9ac5-13 peptide was titrated into the N-labeled Taf14 YEATS domain (Fig. 2c and Supplementary Fig. 4a, b). Binding of H3K9cr induced resonance changes in slow exchange regime on the NMR time scale, indicative of strong interaction. In contrast, binding of H3K9ac resulted in an intermediate exchange, which is characteristic of a weaker association. Furthermore, crosspeaks of Gly80 and Trp81 of the YEATS domain were uniquely perturbed by H3K9cr and H3K9ac, indicating a different chemical environment in the respective crotonyllysine and acetyllysine binding pockets (Supplementary Fig. 4a). These differences support our model that Trp81 adopts two conformations upon complex formation with the H3K9cr mark as compared to H3K9ac (Supplementary Figs. 1c, d and 4c). One of the conformations, characterized by the π stacking involving two aromatic residues and the alkene group, is observed only in the YEATS-H3K9cr complex. To establish whether the Taf14 YEATS domain is able to recognize other recently identified acyllysine marks, we performed solution pull-down assays using H3 peptides acetylated, propionylated, butyrylated, and crotonylated at lysine 9 (residues 1–20 of H3). As shown in Figure 2d and Supplementary Fig. 5a, the Taf14 YEATS domain binds more strongly to H3K9cr1-20, as compared to other acylated histone peptides. The preference for H3K9cr over H3K9ac, H3K9pr and H3K9bu was supported by H,N HSQC titration experiments. Addition of H3K9ac1-20, H3K9pr1-20, and H3K9bu1-20 peptides caused chemical shift perturbations in the Taf14 YEATS domain in intermediate exchange regime, implying that these interactions are weaker compared to the interaction with the H3K9cr1-20 peptide (Supplementary Fig. 5b). We concluded that H3K9cr is the preferred target of this domain. From comparative structural analysis of the YEATS complexes, Gly80 emerged as candidate residue potentially responsible for the preference for crotonyllysine. In attempt to generate a mutant capable of accommodating a short acetyl moiety but discriminating against a longer, planar crotonyl moiety, we mutated Gly80 to more bulky residues, however all mutants of Gly80 lost their binding activities towards either acylated peptide, suggesting that Gly80 is absolutely required for the interaction. In contrast, mutation of Val24, a residue located on another side of Trp81, had no effect on binding (Fig. 2d and Supplementary Fig. 5a, c). To determine if the binding to crotonyllysine is conserved, we tested human YEATS domains by pull-down experiments using singly and multiply acetylated, propionylated, butyrylated, and crotonylated histone peptides (Supplementary Fig. 6). We found that all YEATS domains tested are capable of binding to crotonyllysine peptides, though they display variable preferences for the acyl moieties. While YEATS2 and ENL showed selectivity for the crotonylated peptides, GAS41 and AF9 bound acylated peptides almost equally well. Unlike the YEATS domain, a known acetyllysine reader, bromodomain, does not recognize crotonyllysine. We assayed a large set of BDs in pull-down experiments and found that this module is highly specific for acetyllysine and propionyllysine containing peptides (Supplementary Fig. 7). However, bromodomains did not interact (or associated very weakly) with longer acyl modifications, including crotonyllysine, as in the case of BDs of TAF1 and BRD2, supporting recent reports. These results demonstrate that the YEATS domain is currently the sole reader of crotonyllysine. In conclusion, we have identified the YEATS domain of Taf14 as the first reader of histone crotonylation. The unique and previously unobserved aromatic-amide/aliphatic-aromatic π-π-π-stacking mechanism facilitates the specific recognition of the crotonyl moiety. We further demonstrate that H3K9cr exists in yeast and is dynamically regulated by HATs and HDACs. As we previously showed the importance of acyllysine binding by the Taf14 YEATS domain for the DNA damage response and gene transcription, it will be essential in the future to define the physiological role of crotonyllysine recognition and to differentiate the activities of Taf14 that are due to binding to crotonyllysine and acetyllysine modifications. Furthermore, the functional significance of crotonyllysine recognition by other YEATS proteins will be of great importance to elucidate and compare. The Taf14 YEATS constructs (residues 1–132 or 1–137) were expressed in E. coli BL21 (DE3) RIL in either Luria Broth or M19 minimal media supplemented with NH4Cl and purified as N-terminal GST fusion proteins. Cells were harvested by centrifugation and resuspended in 50 mM HEPES (pH 7.5) supplemented with 150 mM NaCl and 1 mM TCEP. Cells are lysed by freeze-thaw followed by sonication. Proteins were purified on glutathione Sepharose 4B beads and the GST tag was cleaved with PreScission protease. Taf14 YEATS (residues 1–137) was concentrated to 9 mg/mL in 25 mM MES (pH 6.5) and incubated with 2 molar equivalence of the H3K9cr5-13 at RT for 30 mins prior to crystallization. Crystals were obtain via sitting drop diffusion method at 18°C by mixing 800 nL of protein/peptide solution with 800 nL of well solution composed of 44% PEG600 (v/v) and 0.2 M citric acid (pH 6.0). X-ray diffraction data was collected at a wavelength of 1.54 Å at 100 K from a single crystal on the UC Denver Biophysical Core home source composed of a Rigaku Micromax 007 high frequency microfocus X-ray generator with a Pilatus 200K 2D area detector. HKL3000 was used for indexing, scaling, and data reduction. Solution was solved via molecular replacement with Phaser using the Taf14 YEATS domain (PDB 5D7E) as search model with waters, ligands, and peptide removed. Phenix was used for refinement of structure and waters were manually placed by inception of difference maps in Coot. Ramachandran plot indicates good stereochemistry of the three-dimensional structure with 100% of all residues falling within the favored (98%) and allowed (2%) regions. The crystallographic statistics are shown in Supplementary Table 1. NMR spectroscopy was carried out on a Varian INOVA 600 MHz spectrometer outfitted with a cryogenic probe. Chemical shift perturbation (CSP) analysis was performed using uniformly N-labeled Taf14 (1–132). H,N heteronuclear single quantum coherence (HSQC) spectra of the Taf14 YEATS domain were collected in the presence of increasing concentrations of either H3K9cr5-13, H3K9ac5-13, H3K9cr1-20, H3K9ac1-20 H3K9pr1-20, H3K9bu1-20 or free Kcr in PBS buffer pH 6.8, 8% D2O. Tryptophan fluorescence measurements were performed on a Fluorolog spectrofluorometer at room temperature as described. The samples containing 2 μM of Taf14 YEATS in PBS (pH 7.4) and increasing concentrations of H3K9cr5-13 were excited at 295 nm. Emission spectra were recorded from 310 to 340 nm with a 1 nm step size and a 0.5 sec integration time. The Kd value was determined using a nonlinear least-squares analysis and the equation: ΔI=ΔImax(([L]+[P]+Kd)-([L]+[P]+Kd)2-4[P][L]))2[P] where [L] is the concentration of the peptide, [P] is the concentration of the protein, ΔI is the observed change of signal intensity, and ΔImax is the difference in signal intensity of the free and bound states. The Kd values were averaged over 3 separate experiments, with error calculated as the standard deviation (SD). YEATS domains in pGEX vectors were expressed in SoluBL21 cells (Amsbio) by induction with 1 mM IPTG at 16–18°C overnight with shaking. Cells were lysed by freeze-thaw and sonication then purified over glutathione agarose (Pierce) in a buffer containing 50 mM Tris pH 8.0, 500 mM NaCl, 20% glycerol (v/v) and 1 mM dithiothreitol (DTT). Peptide pull-downs were performed essentially as described except that the assay buffer contained 50 mM Tris pH 8.0, 500 mM NaCl, and 0.1% NP-40, and 500 pmols of biotinylated histone peptides were loaded onto streptavidin coated magnetic beads before incubation with 40 pmols of protein. Bound proteins were detected with rabbit GST antibody (Sigma, G7781). Point mutants were generated by site-directed mutagenesis and purified/assayed as described above. The YEATS domains of Taf14, AF9, ENL, and GAS41 were previously described. Yeast cultures were grown in YPD media at 30°C to mid-log phase and extracts were prepared as previously described. Proteins from cell lysates were separated by SDS-PAGE and transferred to a PVDF membrane. Anti-H3K9ac (Millipore, 07-352) and anti-H3K9cr (PTM Biolabs, PTM-516) were diluted to 1:2000 and 1:1000, respectively, in 1x Superblock (ThermoScientific). An HRP-conjugated anti-rabbit (GE Healthcare) was used for detection. Bands were quantified using the ImageJ program. Increasing concentrations of biotinylated histone peptides (0.06–1.5 μg) were spotted onto a PVDF membrane then probed with the anti-H3K9ac (Millipore, 07-352) or H3K9cr (PTM Biolabs, PTM-516) at 1:2000 in a 5% non-fat milk solution and detected with an HRP-conjugated anti-rabbit by enhanced chemiluminesence (ECL). cDNAs of GST-fused bromodomains were obtained either from EpiCypher Inc. or as a kind gift from Katrin Chua (Stanford University). GST fusions were expressed as described above except that the preparation buffer contained 50 mM Tris (pH 7.5), 150 mM NaCl, 10% glycerol (v/v), and 1 mM DTT. Pull-down assays were preformed as described above except that the assay buffer contained 50 mM Tris (pH 8.0), 300 mM NaCl, and 0.1% NP-40.
PMC4888278	Structural determinant for inducing RORgamma specific inverse agonism triggered by a synthetic benzoxazinone ligand	The nuclear hormone receptor RORγ regulates transcriptional genes involved in the production of the pro-inflammatory interleukin IL-17 which has been linked to autoimmune diseases such as rheumatoid arthritis, multiple sclerosis and inflammatory bowel disease. This transcriptional activity of RORγ is modulated through a protein-protein interaction involving the activation function 2 (AF2) helix on the ligand binding domain of RORγ and a conserved LXXLL helix motif on coactivator proteins. Our goal was to develop a RORγ specific inverse agonist that would help down regulate pro-inflammatory gene transcription by disrupting the protein protein interaction with coactivator proteins as a therapeutic agent. We identified a novel series of synthetic benzoxazinone ligands having an agonist (BIO592) and inverse agonist (BIO399) mode of action in a FRET based assay. We show that the AF2 helix of RORγ is proteolytically sensitive when inverse agonist BIO399 binds. Using x-ray crystallography we show how small modifications on the benzoxazinone agonist BIO592 trigger inverse agonism of RORγ. Using an in vivo reporter assay, we show that the inverse agonist BIO399 displayed specificity for RORγ over ROR sub-family members α and β. The synthetic benzoxazinone ligands identified in our FRET assay have an agonist (BIO592) or inverse agonist (BIO399) effect by stabilizing or destabilizing the agonist conformation of RORγ. The proteolytic sensitivity of the AF2 helix of RORγ demonstrates that it destabilizes upon BIO399 inverse agonist binding perturbing the coactivator protein binding site. Our structural investigation of the BIO592 agonist and BIO399 inverse agonist structures identified residue Met358 on RORγ as the trigger for RORγ specific inverse agonism.Retinoid-related orphan receptor gamma (RORγ) is a transcription factor belonging to a sub-family of nuclear receptors that includes two closely related members RORα and RORβ. Even though a high degree of sequence similarity exists between the RORs, their functional roles in regulation for physiological processes involved in development and immunity are distinct . During development, RORγ regulates the transcriptional genes involved in the functioning of multiple pro-inflammatory lymphocyte lineages including T helper cells (TH17cells) which are necessary for IL-17 production . IL-17 is a pro-inflammatory interleukin linked to autoimmune diseases such as rheumatoid arthritis, multiple sclerosis and inflammatory bowel disease; making its transcriptional regulation through RORγ an attractive therapeutic target [3–5]. RORγ consists of an N-terminal DNA binding domain (DBD) connected to a C-terminal ligand binding domain (LBD) via a flexible hinge region. The DBD is composed of two zinc fingers that allow it to interact with specifically encoded regions on the DNA called the nuclear receptor response elements. The LBD consists of a coactivator protein binding pocket and a hydrophobic ligand binding site (LBS) which are responsible for regulating transcription. The coactivator binding pocket of RORγ recognizes a conserved helix motif LXXLL (where X can be any amino acid) on transcriptional coactivator complexes and recruits it to activate transcription . Like other nuclear hormone receptors, RORγ’s helix12 which makes up the C-termini of the LBD is an essential part of the coactivator binding pocket and is commonly referred to as the activation function helix 2 (AF2) . In RORγ, the conformation of the AF2 helix required to form the coactivator binding pocket is mediated by a salt bridge between His479 and Tyr502 in addition to π- π interactions between Tyr502 and Phe506 . The conformation of the AF2 helix can be modulated through targeted ligands which bind the LBS and increase the binding of the coactivator protein (agonists) or disrupt binding (inverse agonists) thereby enhancing or inhibiting transcription [1, 6]. Since RORγ has been demonstrated to play an important role in pro-inflammatory gene expression patterns implicated in several major autoimmune diseases, our aim was to develop RORγ inverse agonists that would help down regulate pro-inflammatory gene transcription [1–5, 9–12]. Here we present the identification of two synthetic benzoxazinone RORγ ligands, a weak agonist BIO592 (Fig. 1a) and an inverse agonist BIO399 (Fig. 1b) which were identified using a Fluorescence Resonance Energy transfer (FRET) based assay that monitored coactivator peptide recruitment. Using partial proteolysis in combination with mass spectrometry analysis we demonstrate that the AF2 helix of RORγ destabilizes upon BIO399 (inverse agonist) binding. Finally, comparing binding modes of our benzoxazinone RORγ crystal structures to other ROR structures, we hypothesize a new mode of action for achieving inverse agonism and selectivity.Fig. 1FRET results for agonist BIO592 (a) and Inverse Agonist BIO399 (b) FRET results for agonist BIO592 (a) and Inverse Agonist BIO399 (b) GST-RORγ518 was constructed by sub-cloning residues 259 to 518 of a human RORγ cDNA into a pGEX-6P vector with a cleavable N-terminal GST fusion tag. BL21 (DE3) Escherichia coli cells were transformed with the plasmid encoding the GST-PreScission-hRORgamma 259–518 protein (GST-RORγ518) and were grown at 37 °C in LB media supplemented with ampicillin to an OD of 1. The temperature was reduced to 18 °C and protein expression was induced by adding 1 mM IPTG and was shaking for an additional 16 h. The cells were harvested and resuspended in lysis buffer (25 mM TRIS pH 8.0, 250 mM NaCl, 10 % Glycerol, 5 mM DTT and Roche EDTA-free protease inhibitor cocktail) and were lysed using a microfluidizer. The lysate was clarified by centrifugation at 20,000 × g for 1 h at 4 °C and GST-RORγ518 was captured by batch binding to Glutathione Sepharose resin overnight at 4 °C. The resin was washed with buffer A (25 mM TRIS pH 8.0, 250 mM NaCl, 10 % glycerol, 5 mM DTT) and loaded onto a XK column and washed until no non-specific unbound protein was detected. GST- RORγ518 was eluted from the column using buffer A supplemented with 10 mM Glutathione pH 8.0 and analyzed by SDS-PAGE. The eluate was then treated with PreScission Protease (10units/mg of protein) and further purified on a Superdex 75 column equilibrated in buffer B (25 mM TRIS pH 8.0, 250 mM NaCl, 5 % glycerol and 2 mM DTT). RORγ518 eluted as a monomer and was approximately 95 % pure as observed by SDS-PAGE. Additional constructs including c-terminal truncations, surface entropy reduction and cysteine scrubbed mutations were also expressed and purified in the same manner as RORγ518 if an expression level of >1 mg/L was achieved. FRET-based (Fluorescence Resonance Energy Transfer) assay and the GAL4 Reporter assay were performed as described previously . BIO592 and BIO399 were synthesized (Additional file 1) and belonged to a proprietary library where they were identified as RORγ activity modulators using the FRET-based assay. RORγ518 at 8 mg/ml or in complex with 1 mM BIO399 or 1 mM BIO592 and 0.5 mM coactivator peptide EBI96 EFPYLLSLLGEVSPQ (New England Peptide) were treated with Actinase E (Hampton Research) added at a ratio of 1.25ugs of protease/1 mg of RORγ518 for 6 h at 4 °C . The reactions were quenched using 1X Protease inhibitor cocktail (Roche) + 1 mM EDTA and subjected to mass spectrometry analysis. Proteolyzed RORγ518 samples were reduced with 50 mM dithiothreitol in 50 mM Tris pH 8.0, 150 mM NaCl containing 4 M urea and 5 mM EDTA. The sample was then analyzed on a LC-MS system comprised of a UPLC (ACQUITY, Waters Corp.), a TUV dual-wavelength UV detector (Waters Corp.), and a ZQ mass spectrometer (Waters Corp.). A Vydac C4 cartridge was used for desalting. Molecular masses for the Actinase E treated RORγ518 samples were obtained by deconvoluting the raw mass spectra using MaxLynx 4.1 software (Waters Corp.). RORγ518 was concentrated to 8 mg/ml and EBI96 was added to a final concentration of 0.5 mM and agonist BIO592 to 1 mM and incubated on ice for 1 h. The coactivator peptide EBI96 which was identified by phage display was chosen for crystallization because of its strong interaction with RORγ in a mammalian two-hybrid analysis system that assessed the transactivation of RORγ . Diffraction quality crystals were grown through vapor diffusion in a buffer containing 0.1 M HEPES pH 8.0, 25 % PEG3350 and 0.2 M NaCl at 18 °C. Crystals were cryoprotected in the mother liquor containing 20 % glycerol as cryoprotectant prior to being frozen in liquid nitrogen for data collection. Actinase E proteolyzed RORγ518 BIO399 concentrated to 8 mg/ml was crystallized using vapor diffusion in a buffer containing 0.1 M BisTRIS pH 5.5, 0.2 M ammonium acetate and 15 % PEG3350 at 18 °C. Crystals were cryoprotected for data collection by transferring them to a mother liquor containing 15 % PEG400 prior to being frozen in liquid nitrogen. X-ray diffraction data for all the crystals were measured at beam line ID31 at the Argonne Photon Source. The data were processed with Mosflm in case of the RORγ518-BIO592-EBI96 ternary complex and with HKL2000 in the case of the Actinase E treated aeRORγ518/BIO399 complex. For both datasets, PDB ID: 3LOL was used as the search model, and the molecular replacement solutions were determined using MOLREP . The refinement was carried out using Refmac5 and model building was carried out in Coot . The data processing and refinement statistics are provided in Additional file 2. RORγ518-BIO592-EBI96 ternary complex: The data for the ternary complex were measured to 2.63 Å. It crystallized in a P21 space group with four molecules of the ternary complex in the asymmetric unit. The final model was refined to a Rcryst of 19.9 % and Rfree of 25.5 %. aeRORγ518/BIO399 complex: Diffraction data for the aeRORγ518-BIO399 complex were measured to 2.35 Å. It crystallized in C2 space group with two molecules in the asymmetric unit. The final model was refined to a Rcryst of 21.1 % and Rfree of 26.3 %. Using a FRET based assay we discovered agonist BIO592 (Fig. 1a) which increased the coactivator peptide TRAP220 recruitment to RORγ (EC50 0f 58nM and Emax of 130 %) and a potent inverse agonist BIO399 (Fig. 1b) which inhibited coactivator recruitment (IC50: 4.7nM). Interestingly, the structural difference between the agonist BIO592 and inverse agonist BIO399 was minor; with the 2,3-dihydrobenzooxazepin-4-one ring system of BIO399 being 3 atoms larger than the benzooxazine-3-one ring system of BIO592. In order to understand how small changes in the core ring system leads to inverse agonism, we wanted to structurally determine the binding mode of both BIO592 and BIO399 in the LBS of RORγ using x-ray crystallography. RORγ518 bound to agonist BIO592 was crystallized with a truncated form of the coactivator peptide EBI96 to a resolution of 2.6 Å (Fig. 2a). The structure of the ternary complex had features similar to other ROR agonist coactivator structures in a transcriptionally active canonical three layer helix fold with the AF2 helix in the agonist conformation . The agonist conformation is stabilized by a hydrogen bond between His479 and Tyr502, in addition to π-π interactions between His479, Tyr502 and Phe506 (Fig. 2b). The hydrogen bond between His479 and Tyr502 has been reported to be critical for RORγ agonist activity. Disrupting this interaction through mutagenesis reduced transcriptional activity of RORγ . This reduced transcriptional activity has been attributed to the inability of the AF2 helix to complete the formation of the coactivator binding pocket necessary for coactivator proteins to bind.Fig. 2 a The ternary structure of RORγ518 BIO592 and EBI96. b RORγ AF2 helix in the agonist conformation. c EBI96 coactivator peptide bound in the coactivator pocket of RORγ a The ternary structure of RORγ518 BIO592 and EBI96. b RORγ AF2 helix in the agonist conformation. c EBI96 coactivator peptide bound in the coactivator pocket of RORγ Electron density for the coactivator peptide EBI96 was observed for residues EFPYLLSLLG which formed a α-helix stabilized through hydrophobic interactions with the coactivator binding pocket on RORγ (Fig. 2c). This interaction is further stabilized through a conserved charged clamp wherein the backbone amide of Tyr7 and carbonyl of Leu11 of EBI96 form hydrogen bonds with Glu504 (helix12) and Lys336 (helix3) of RORγ. Formation of this charged clamp is essential for RORγ’s function for playing a role in transcriptional activation and this has been corroborated through mutagenic studies in this region [14, 17]. BIO592 bound in a collapsed conformational state in the LBS of RORγ with the xylene ring positioned at the bottom of the pocket making hydrophobic interactions with Val376, Phe378, Phe388 and Phe401, with the ethyl-benzoxazinone ring making several hydrophobic interactions with Trp317, Leu324, Met358, Leu391, Ile 400 and His479 (Fig. 3a, Additional file 3). The sulfonyl group faces the entrance of the pocket, while the CF3 makes a hydrophobic contact with Ala327. Hydrophobic interaction between the ethyl group of the benzoxazinone and His479 reinforce the His479 sidechain position for making the hydrogen bond with Tyr502 thereby stabilizing the agonist conformation (Fig. 3b).Fig. 3 a Collapsed binding mode of agonist BIO592 in the hydrophobic LBS of RORγ. b Benzoxazinone ring system of agonist BIO592 packing against His479 of RORγ stabilizing agonist conformation of the AF2 helix a Collapsed binding mode of agonist BIO592 in the hydrophobic LBS of RORγ. b Benzoxazinone ring system of agonist BIO592 packing against His479 of RORγ stabilizing agonist conformation of the AF2 helix Next, we attempted co-crystallization with the inverse agonist BIO399. However, extensive crystallization efforts with BIO399 and RORγ518 or other AF2 intact constructs did not produce crystals. We hypothesized that the RORγ518 coactivator peptide interaction in the FRET assay was disrupted upon BIO399 binding and that a conformational rearrangement of the AF2 helix could have occurred, hindering crystallization. The unfolding of the AF2 helix has been observed for other nuclear hormone receptors when bound to an inverse agonist or antagonist [22–24]. We used partial proteolysis in combination with mass spectrometry to determine if BIO399 was causing the AF2 helix to unfold . Results of the Actinase E proteolysis experiments on RORγ518, the ternary complex of RORγ518 with agonist BIO592 and coactivator EBI96, or in the presence of inverse agonist BIO399 supported our hypothesis. Analysis of the fragmentation pattern showed minimal proteolytic removal of the AF2 helix by Actinase E on RORγ518 alone (ending at 504 to 506) and the ternary complex remained primarily intact (ending at 515/518) (Additional file 4). However, in the presence of inverse agonist BIO399, the proteolytic pattern showed significantly less protection, albeit the products were more heterogeneous (majority ending at 494/495), indicating the destabilization of the AF2 helix compared to either the APO or ternary agonist complex (Fig. 4, Additional file 5).Fig. 4Specific proteolytic positions on RORγ518 when treated with Actinase E alone (Green) or in the presence of BIO399 (Red) and shared proteolytic sites (Yellow) Specific proteolytic positions on RORγ518 when treated with Actinase E alone (Green) or in the presence of BIO399 (Red) and shared proteolytic sites (Yellow) Several rounds of cocrystallization attempts with RORγ518 or other RORγ AF2 helix containing constructs complexed with BIO399 had not produced crystals. We attributed the inability to form crystals to the unfolding of the AF2 helix induced by BIO399. We reasoned that if we could remove the unfolded AF2 helix using proteolysis we could produce a binary complex more amenable to crystallization. The Actinase E treated RORγ518 BIO399 ternary complex (aeRORγ493/4) co-crystallized readily in several PEG based conditions. The structure of aeRORγ493/4 BIO399 complex was solved to 2.3 Å and adopted a similar core fold to the BIO592 agonist crystal structure (Fig. 5a, Additional file 3). The aeRORγ493/4 BIO399 structure diverged at the c-terminal end of Helix 11 from the RORγ518 BIO592 EBI96 structure, where helix 11 unwinds into a random coil after residue L475.Fig. 5 a The binary structure of AF2-truncated RORγ and BIO399. b The superposition of inverse agonist BIO399 (Cyan) and agonist BIO592 (Green). c Movement of Met358 and His479 in the BIO399 (Cyan) and BIO592 (Green) structures a The binary structure of AF2-truncated RORγ and BIO399. b The superposition of inverse agonist BIO399 (Cyan) and agonist BIO592 (Green). c Movement of Met358 and His479 in the BIO399 (Cyan) and BIO592 (Green) structures BIO399 binds to the ligand binding site of RORγ adopting a collapsed conformation as seen with BIO592 where the two compounds superimpose with an RMSD of 0.72 Å (Fig. 5b). The majority of the side chains within 4 Å of BIO399 and BIO592 adopt similar rotomer conformations with the exceptions of Met358 and His479 (Fig. 5c). The difference density map showed clear positive density for Met358 in an alternate rotomer conformation compared to the one observed in the molecular replacement model or the other agonist containing models (Additional file 6). We tried to refine Met358 in the same conformation as the molecular replacement model or the other agonist containing models, but the results clearly indicated that this was not possible, thus confirming the new rotamer conformation for the Met358 sidechain in the inverse agonist bound structure. The change in rotomer conformation of Met358 between the agonist and inverse agonist structures is attributed to the gem-dimethyl group on the larger 7 membered benzoxazinone ring system of BIO399. The comparison of the two structures shows that the agonist conformation observed in the BIO592 structure would be perturbed by BIO399 pushing Met358 into Phe506 of the AF2 helix indicating that Met358 is a trigger for inducing inverse agonism in RORγ (Fig. 5c). The co-crystal structure of RORγ with T0901317 (PDB code: 4NB6), an inverse agonist of RORγ (IC50 of 54nM in an SRC1 displacement FRET assay and an IC50 of 59nM in our FRET assay (Additional file 7)) shows that it adopts a collapsed conformation similar to the structure of BIO399 described here . The two compounds superimpose with an RMSD of 0.81 Å (Fig. 6a). The CF3 group on the hexafluoropropanol group of T0901317 was reported to fit the electron density in two conformations one of which pushes Met358 into the vicinity of Phe506 in the RORγ BIO592 agonist structure. We hypothesize that since the Met358 sidechain conformation in the T0901317 RORγ structure is not in the BIO399 conformation, this difference could account for the 10-fold reduction in the inverse agonism for T0901317 compared to BIO399 in the FRET assay.Fig. 6 a Overlay of RORγ structures bound to BIO596 (Green), BIO399 (Cyan) and T0901317 (Pink). b Overlay of M358 in RORγ structure BIO596 (Green), BIO399 (Cyan), Digoxin (Yellow), Compound 2 (Grey), Compound 48 (Salmon) and Compound 4j (Orange) a Overlay of RORγ structures bound to BIO596 (Green), BIO399 (Cyan) and T0901317 (Pink). b Overlay of M358 in RORγ structure BIO596 (Green), BIO399 (Cyan), Digoxin (Yellow), Compound 2 (Grey), Compound 48 (Salmon) and Compound 4j (Orange) Co-crystal structures of RORγ have been generated with several potent inverse agonists adopting a linear conformation distinct from the collapsed conformations seen for BIO399 and T090131718 [27–31]. The inverse agonist activity for these compounds has been attributed to orientating Trp317 to clash with Tyr502 or a direct inverse agonist hydrogen bonding event with His479, both of which would perturb the agonist conformation of RORγ. BIO399 neither orients the sidechain of Trp317 toward Tyr502 nor forms a hydrogen bond with His479 suggesting its mode of action is distinct from linear inverse agonists (Additional file 8). In the linear inverse agonist crystal structures the side chain of Met358 resides in a similar position as the rotomer observed in RORγ agonist structures with BIO592 described here or as observed in the hydroxycholesterol derivatives and therefore would not trigger inverse agonism with these ligands (Fig. 6b) . In order to assess the in vivo selectivity profile of BIO399 a cellular reporter assay was implemented where the ligand binding domains of ROR α, β and γ were fused to the DNA binding domain of the transcriptional factor GAL4. The ROR-GAL4 fusion proteins were expressed in cells with the luciferase reporter gene under the control of a GAL4 promoter . BIO399 inhibited the luciferase activity when added to the cells expressing the RORγ-GAL4 fusion with an in vivo IC50 of 42.5nM while showing >235 and 28 fold selectivity over cells expressing GAL4 fused to the LBD of ROR α or β, respectively (Table 1). The LBS of RORs share a high degree of similarity. However, the inverse agonism trigger of BIO399, residue Met358, is a leucine in both RORα and β. This selectivity profile for BIO399 is attributed to the shorter leucine side chain in RORα and β which would not reach the phenylalanine on the AF2 helix further underscoring the role of Met358 as a trigger for RORγ specific inverse agonism (Fig. 7a). Furthermore, RORα contains two phenylalanine residues in its LBS whereas RORβ and γ have a leucine in the same position (Fig. 6b). We hypothesize that the two phenylalanine residues in the LBS of RORα occlude the dihydrobenzoxazepinone ring system of BIO399 from binding it and responsible for the increase in selectivity for RORα over β.Table 1GAL4 cell assay selectivity profile for BIO399 toward RORα and RORβ in GAL4RORγαβIC50 (uM)0.043 (+/− 0.01uM; N = 6)>10 (N = 2)>1.2 (N = 2)Selectivity (X)->235>28.2Fig. 7 a Overlay of RORα (yellow), β (pink) and γ (cyan) showing side chain differences at Met358 inverse agonism trigger position and (b) around the benzoxazinone ring system of BIO399 GAL4 cell assay selectivity profile for BIO399 toward RORα and RORβ in GAL4 a Overlay of RORα (yellow), β (pink) and γ (cyan) showing side chain differences at Met358 inverse agonism trigger position and (b) around the benzoxazinone ring system of BIO399 We have identified a novel series of synthetic benzoxazinone ligands which modulate the transcriptional activity of RORγ in a FRET based assay. Using partial proteolysis we show a conformational change which destabilizes the AF2 helix of RORγ when the inverse agonist BIO399 binds. The two RORγ co-crystal structures reported here show how a small change to the core ring system can modulate the mode of action from agonist (BIO592) to inverse agonism (BIO399). Finally, we are reporting a newly identified trigger for achieving RORγ specific inverse agonism in an in vivo setting through Met358 which perturbs the agonist conformation of the AF2 helix and prevents coactivator protein binding. AF2, activation function 2; BisTRIS, 2-[Bis(2-hydroxyethyl)amino]-a-(hydroxymethyl)propane-1,3-diol; DND, DNA binding domain; DTT, 1,4-Dithiothreitol; EDTA, 2-((carboxymethyl)amino)acetic acid; FRET, fluorescence resonance energy transfer; GST, Glutathione-S-Transferase; HEPES, 2-[4(2-hydroxyethyl)-1-piperazineethanesulfonic acid; IC50, half maximal inhibitory concentration; IL-17, Interleukin-17; IPTG, isopropyl β-D-1-thiogalactopyranoside; LBD, Ligand Binding Domain; LBS, ligand binding site; LC-MS, liquid chromatography/mass spectrometry; PDB, Protein Data Bank; ROR, retinoid orphan receptor; SRC-1, steroid receptor coactivator-1; TH17 Cells, T helper cells; TRIS, 2-amino-2-hydroxymethyl-propane-1,3,diol.
PMC4786784	An extended U2AF–RNA-binding domain recognizes the 3′ splice site signal	How the essential pre-mRNA splicing factor U2AF recognizes the polypyrimidine (Py) signals of the major class of 3′ splice sites in human gene transcripts remains incompletely understood. We determined four structures of an extended U2AF–RNA-binding domain bound to Py-tract oligonucleotides at resolutions between 2.0 and 1.5 Å. These structures together with RNA binding and splicing assays reveal unforeseen roles for U2AF inter-domain residues in recognizing a contiguous, nine-nucleotide Py tract. The U2AF linker residues between the dual RNA recognition motifs (RRMs) recognize the central nucleotide, whereas the N- and C-terminal RRM extensions recognize the 3′ terminus and third nucleotide. Single-molecule FRET experiments suggest that conformational selection and induced fit of the U2AF RRMs are complementary mechanisms for Py-tract association. Altogether, these results advance the mechanistic understanding of molecular recognition for a major class of splice site signals.The differential skipping or inclusion of alternatively spliced pre-mRNA regions is a major source of diversity for nearly all human gene transcripts1. The splice sites are marked by relatively short consensus sequences and are regulated by additional pre-mRNA motifs (reviewed in ref. 2). At the 3′ splice site of the major intron class, these include a polypyrimidine (Py) tract comprising primarily Us or Cs, which is preceded by a branch point sequence (BPS) that ultimately serves as the nucleophile in the splicing reaction and an AG-dinucleotide at the 3′ splice site junction. Disease-causing mutations often compromise pre-mRNA splicing (reviewed in refs 3, 4), yet a priori predictions of splice sites and the consequences of their mutations are challenged by the brevity and degeneracy of known splice site sequences. High-resolution structures of intact splicing factor–RNA complexes would offer key insights regarding the juxtaposition of the distinct splice site consensus sequences and their relationship to disease-causing point mutations. The early-stage pre-mRNA splicing factor U2AF is essential for viability in vertebrates and other model organisms (for example, ref. 5). A tightly controlled assembly among U2AF, the pre-mRNA, and partner proteins sequentially identifies the 3′ splice site and promotes association of the spliceosome, which ultimately accomplishes the task of splicing2. Initially U2AF recognizes the Py-tract splice site signal6. In turn, the ternary complex of U2AF with SF1 and U2AF identifies the surrounding BPS78 and 3′ splice site junctions91011. Subsequently U2AF recruits the U2 small nuclear ribonucleoprotein particle (snRNP) and ultimately dissociates from the active spliceosome. Biochemical characterizations of U2AF demonstrated that tandem RNA recognition motifs (RRM1 and RRM2) recognize the Py tract1213 (Fig. 1a). Milestone crystal structures of the core U2AF RRM1 and RRM2 connected by a shortened inter-RRM linker (dU2AF1,2) detailed a subset of nucleotide interactions with the individual U2AF RRMs1415. A subsequent NMR structure16 characterized the side-by-side arrangement of the minimal U2AF RRM1 and RRM2 connected by a linker of natural length (U2AF1,2), yet depended on the dU2AF1,2 crystal structures for RNA interactions and an ab initio model for the inter-RRM linker conformation. As such, the molecular mechanisms for Py-tract recognition by the intact U2AF–RNA-binding domain remained unknown. Here, we use X-ray crystallography and biochemical studies to reveal new roles in Py-tract recognition for the inter-RRM linker and key residues surrounding the core U2AF RRMs. We use single-molecule Förster resonance energy transfer (smFRET) to characterize the conformational dynamics of this extended U2AF–RNA-binding domain during Py-tract recognition. The RNA affinity of the minimal U2AF1,2 domain comprising the core RRM1–RRM2 folds (U2AF1,2, residues 148–336) is relatively weak compared with full-length U2AF (Fig. 1a,b; Supplementary Fig. 1). Historically, this difference was attributed to the U2AF arginine–serine rich domain, which contacts pre-mRNA–U2 snRNA duplexes outside of the Py tract171819. We noticed that the RNA-binding affinity of the U2AF1,2 domain was greatly enhanced by the addition of seven and six residues at the respective N and C termini of the minimal RRM1 and RRM2 (U2AF1,2L, residues 141–342; Fig. 1a). In a fluorescence anisotropy assay for binding a representative Py tract derived from the well-characterized splice site of the adenovirus major late promoter (AdML), the RNA affinity of U2AF1,2L increased by 100-fold relative to U2AF1,2 to comparable levels as full-length U2AF (Fig. 1b; Supplementary Fig. 1a–d). Likewise, both U2AF1,2L and full-length U2AF showed similar sequence specificity for U-rich stretches in the 5′-region of the Py tract and promiscuity for C-rich regions in the 3′-region (Fig. 1c, Supplementary Fig. 1e–h). To investigate the structural basis for cognate U2AF recognition of a contiguous Py tract, we determined four crystal structures of U2AF1,2L bound to Py-tract oligonucleotides (Fig. 2a; Table 1). By sequential boot strapping (Methods), we optimized the oligonucleotide length, the position of a Br-dU, and the identity of the terminal nucleotide (rU, dU and rC) to achieve full views of U2AF1,2L bound to contiguous Py tracts at up to 1.5 Å resolution. The protein and oligonucleotide conformations are nearly identical among the four new U2AF1,2L structures (Supplementary Fig. 2a). The U2AF1,2L RRM1 and RRM2 associate with the Py tract in a parallel, side-by-side arrangement (shown for representative structure iv in Fig. 2b,c; Supplementary Movie 1). An extended conformation of the U2AF inter-RRM linker traverses across the α-helical surface of RRM1 and the central β-strands of RRM2 and is well defined in the electron density (Fig. 2b). The extensions at the N terminus of RRM1 and C terminus of RRM2 adopt well-ordered α-helices. Both RRM1/RRM2 extensions and the inter-RRM linker of U2AF1,2L directly recognize the bound oligonucleotide. We compare the global conformation of the U2AF1,2L structures with the prior dU2AF1,2 crystal structure15 and U2AF1,2 NMR structure16 in the Supplementary Discussion and Supplementary Fig. 2. The discovery of nine U2AF-binding sites for contiguous Py-tract nucleotides was unexpected. Based on dU2AF1,2 structures141520, we originally hypothesized that the U2AF RRMs would bind the minimal seven nucleotides observed in these structures. Surprisingly, the RRM2 extension/inter-RRM linker contribute new central nucleotide-binding sites near the RRM1/RRM2 junction and the RRM1 extension recognizes the 3′-terminal nucleotide (Fig. 2c; Supplementary Movie 1). The U2AF1,2L structures characterize ribose (r) nucleotides at all of the binding sites except the seventh and eighth deoxy-(d)U, which are likely to lack 2′-hydroxyl contacts based on the RNA-bound dU2AF1,2 structure15. Qualitatively, a subset of the U2AF1,2L-nucleotide-binding sites (sites 1–3 and 7–9) share similar locations to those of the dU2AF1,2 structures (Supplementary Figs 2c,d and 3). Yet, only the U2AF1,2L interactions at sites 1 and 7 are nearly identical to those of the dU2AF1,2 structures (Supplementary Fig. 3a,f). In striking departures from prior partial views, the U2AF1,2L structures reveal three unanticipated nucleotide-binding sites at the centre of the Py tract, as well as numerous new interactions that underlie cognate recognition of the Py tract (Fig. 3a–h). The U2AF1,2L RRM2, the inter-RRM linker and RRM1 concomitantly recognize the three central nucleotides of the Py tract, which are likely to coordinate the conformational arrangement of these disparate portions of the protein. Residues in the C-terminal region of the U2AF inter-RRM linker comprise a centrally located binding site for the fifth nucleotide on the RRM2 surface and abutting the RRM1/RRM2 interface (Fig. 3d). The backbone amide of the linker V254 and the carbonyl of T252 engage in hydrogen bonds with the rU5-O4 and -N3H atoms. In the C-terminal β-strand of RRM1, the side chains of K225 and R227 donate additional hydrogen bonds to the rU5-O2 lone pair electrons. The C-terminal region of the inter-RRM linker also participates in the preceding rU4-binding site, where the V254 backbone carbonyl and D256 carboxylate position the K260 side chain to hydrogen bond with the rU4-O4 (Fig. 3c). Otherwise, the rU4 nucleotide packs against F304 in the signature ribonucleoprotein consensus motif (RNP)-2 of RRM2. At the opposite side of the central fifth nucleotide, the sixth rU6 nucleotide is located at the inter-RRM1/RRM2 interface (Fig. 3e; Supplementary Movie 1). This nucleotide twists to face away from the U2AF linker and instead inserts the rU6-uracil into a sandwich between the β2/β3 loops of RRM1 and RRM2. The rU6 base edge is relatively solvent exposed; accordingly, the rU6 hydrogen bonds with U2AF are water mediated apart from a single direct interaction by the RRM1-N196 side chain. We tested the contribution of the U2AF1,2L interactions with the new central nucleotide to Py-tract affinity (Fig. 3i; Supplementary Fig. 4a,b). Mutagenesis of either V254 in the U2AF inter-RRM linker to proline or RRM1–R227 to alanine, which remove the hydrogen bond with the fifth uracil-O4 or -O2, reduced the affinities of U2AF1,2L for the representative AdML Py tract by four- or five-fold, respectively. The energetic penalties due to these mutations (ΔΔG 0.8–0.9 kcal mol) are consistent with the loss of each hydrogen bond with the rU5 base and support the relevance of the central nucleotide interactions observed in the U2AF1,2L structures. The N- and C-terminal extensions of the U2AF RRM1 and RRM2 directly contact the bound Py tract. Rather than interacting with a new 5′-terminal nucleotide as we had hypothesized, the C-terminal α-helix of RRM2 instead folds across one surface of rU3 in the third binding site (Fig. 3b). There, a salt bridge between the K340 side chain and nucleotide phosphate, as well as G338-base stacking and a hydrogen bond between the backbone amide of G338 and the rU3-O4, secure the RRM2 extension. Indirectly, the additional contacts with the third nucleotide shift the rU2 nucleotide in the second binding site closer to the C-terminal β-strand of RRM2. Consequently, the U2AF1,2L-bound rU2-O4 and -N3H form dual hydrogen bonds with the K329 backbone atoms (Fig. 3a), rather than a single hydrogen bond with the K329 side chain as in the prior dU2AF1,2 structure15 (Supplementary Fig. 3b). At the N terminus, the α-helical extension of U2AF RRM1 positions the Q147 side chain to bridge the eighth and ninth nucleotides at the 3′ terminus of the Py tract (Fig. 3f–h). The Q147 residue participates in hydrogen bonds with the -N3H of the eighth uracil and -O2 of the ninth pyrimidine. The adjacent R146 guanidinium group donates hydrogen bonds to the 3′-terminal ribose-O2′ and O3′ atoms, where it could form a salt bridge with a phospho-diester group in the context of a longer pre-mRNA. Consistent with loss of a hydrogen bond with the ninth pyrimidine-O2 (ΔΔG 1.0 kcal mol), mutation of the Q147 to an alanine reduced U2AF1,2L affinity for the AdML Py tract by five-fold (Fig. 3i; Supplementary Fig. 4c). We compare U2AF interactions with uracil relative to cytosine pyrimidines at the ninth binding site in Fig. 3g,h and the Supplementary Discussion. The U2AF1,2L structures reveal that the inter-RRM linker mediates an extensive interface with the second α-helix of RRM1, the β2/β3 strands of RRM2 and the N-terminal α-helical extension of RRM1. Altogether, the U2AF inter-RRM linker residues (R228–K260) bury 2,800 Å of surface area in the U2AF1,2L holo-protein, suggestive of a cognate interface compared with 1,900 Å for a typical protein–protein complex21. The path of the linker initiates at P229 following the core RRM1 β-strand, in a kink that is positioned by intra-molecular stacking among the consecutive R228, Y232 and P234 side chains (Fig. 4a, lower right). A second kink at P236, coupled with respective packing of the L235 and M238 side chains on the N-terminal α-helical RRM1 extension and the core RRM1 α2-helix, reverses the direction of the inter-RRM linker towards the RRM1/RRM2 interface and away from the RNA-binding site. In the neighbouring apical region of the linker, the V244 and V246 side chains pack in a hydrophobic pocket between two α-helices of the core RRM1. The adjacent V249 and V250 are notable for their respective interactions that connect RRM1 and RRM2 at this distal interface from the RNA-binding site (Fig. 4a, top). A third kink stacks P247 and G248 with Y245 and re-orients the C-terminal region of the linker towards the RRM2 and bound RNA. At the RNA surface, the key V254 that recognizes the fifth uracil is secured via hydrophobic contacts between its side chain and the β-sheet surface of RRM2, chiefly the consensus RNP1-F304 residue that stacks with the fourth uracil (Fig. 4a, lower left). Few direct contacts are made between the remaining residues of the linker and the U2AF RRM2; instead, the C-terminal conformation of the linker appears primarily RNA mediated (Fig. 3c,d). We investigated whether the observed contacts between the RRMs and linker were critical for RNA binding by structure-guided mutagenesis (Fig. 4b). We titrated these mutant U2AF1,2L proteins into fluorescein-labelled AdML Py-tract RNA and fit the fluorescence anisotropy changes to obtain the apparent equilibrium affinities (Supplementary Fig. 4d–h). We introduced glycine substitutions to maximally reduce the buried surface area without directly interfering with its hydrogen bonds between backbone atoms and the base. First, we replaced V249 and V250 at the RRM1/RRM2 interface and V254 at the bound RNA site with glycine (3Gly). However, the resulting decrease in the AdML RNA affinity of the U2AF1,2L-3Gly mutant relative to wild-type protein was not significant (Fig. 4b). In parallel, we replaced five linker residues (S251, T252, V253, V254 and P255) at the fifth nucleotide-binding site with glycines (5Gly) and also found that the RNA affinity of the U2AF1,2L-5Gly mutant likewise decreased only slightly relative to wild-type protein. A more conservative substitution of these five residues (251–255) with an unrelated sequence capable of backbone-mediated hydrogen bonds (STVVP>NLALA) confirmed the subtle impact of this versatile inter-RRM sequence on affinity for the AdML Py tract. Finally, to ensure that these selective mutations were sufficient to disrupt the linker/RRM contacts, we substituted glycine for the majority of buried hydrophobic residues in the inter-RRM linker (including M144, L235, M238, V244, V246, V249, V250, S251, T252, V253, V254, P255; called 12Gly). Despite 12 concurrent mutations, the AdML RNA affinity of the U2AF1,2L-12Gly variant was reduced by only three-fold relative to the unmodified protein (Fig. 4b), which is less than the penalty of the V254P mutation that disrupts the rU5 hydrogen bond (Fig. 3d,i). To test the interplay of the U2AF inter-RRM linker with its N- and C-terminal RRM extensions, we constructed an internal linker deletion of 20-residues within the extended RNA-binding domain (dU2AF1,2L). We found that the affinity of dU2AF1,2L for the AdML RNA was significantly reduced relative to U2AF1,2L (four-fold, Figs 1b and 4b; Supplementary Fig. 4i). Yet, it is well known that the linker deletion in the context of the minimal RRM1–RRM2 boundaries151622 has no detectable effect on the RNA affinities of dU2AF1,2 compared with U2AF1,2 (refs 15, 16, 23; Figs 1b and 4b; Supplementary Fig. 4j). The U2AF1,2L structures suggest that an extended conformation of the truncated dU2AF1,2 inter-RRM linker would suffice to connect the U2AF1,2L RRM1 C terminus to the N terminus of RRM2 (24 Å distance between U2AF1,2L R227-Cα–H259-Cα atoms), which agrees with the greater RNA affinities of dU2AF1,2 and U2AF1,2 dual RRMs compared with the individual U2AF RRMs23. However, stretching of the truncated dU2AF1,2L linker to connect the RRM termini is expected to disrupt its nucleotide interactions. Likewise, deletion of the N-terminal RRM1 extension in the shortened constructs would remove packing interactions that position the linker in a kinked turn following P229 (Fig. 4a), consistent with the lower RNA affinities of dU2AF1,2L, dU2AF1,2 and U2AF1,2 compared with U2AF1,2L. To further test cooperation among the U2AF65 RRM extensions and inter-RRM linker for RNA recognition, we tested the impact of a triple Q147A/V254P/R227A mutation (U2AF1,2L-3Mut) for RNA binding (Fig. 4b; Supplementary Fig. 4d). Notably, the Q147A/V254P/R227A mutation reduced the RNA affinity of the U2AF1,2L-3Mut protein by 30-fold more than would be expected based on simple addition of the ΔΔG's for the single mutations. This difference indicates that the linearly distant regions of the U2AF primary sequence, including Q147 in the N-terminal RRM1 extension and R227/V254 in the N-/C-terminal linker regions at the fifth nucleotide site, cooperatively recognize the Py tract. Altogether, we conclude that the conformation of the U2AF inter-RRM linker is key for recognizing RNA and is positioned by the RRM extension but otherwise relatively independent of the side chain composition. The non-additive effects of the Q147A/V254P/R227A triple mutation, coupled with the context-dependent penalties of an internal U2AF linker deletion, highlights the importance of the structural interplay among the U2AF linker and the N- and C-terminal extensions flanking the core RRMs. We proceeded to test the importance of new U2AF–Py-tract interactions for splicing of a model pre-mRNA substrate in a human cell line (Fig. 5; Supplementary Fig. 5). As a representative splicing substrate, we utilized a well-characterized minigene splicing reporter (called pyPY) comprising a weak (that is, degenerate, py) and strong (that is, U-rich, PY) polypyrimidine tracts preceding two alternative splice sites24 (Fig. 5a). When transfected into HEK293T cells containing only endogenous U2AF, the PY splice site is used and the remaining transcript remains unspliced. When co-transfected with an expression plasmid for wild-type U2AF, use of the py splice site significantly increases (by more than five-fold) and as documented24 converts a fraction of the unspliced to spliced transcript. The strong PY splice site is insensitive to added U2AF, suggesting that endogenous U2AF levels are sufficient to saturate this site (Supplementary Fig. 5b). We introduced the triple mutation (V254P/R227A/Q147A) that significantly reduced U2AF1,2L association with the Py tract (Fig. 4b) in the context of full-length U2AF (U2AF-3Mut). Co-transfection of the U2AF-3Mut with the pyPY splicing substrate significantly reduced splicing of the weak ‘py' splice site relative to wild-type U2AF (Fig. 5b,c). We conclude that the Py-tract interactions with these residues of the U2AF inter-RRM linker and RRM extensions are important for splicing as well as for binding a representative of the major U2-class of splice sites. The direct interface between U2AF1,2L RRM1 and RRM2 is minor, burying 265 Å of solvent accessible surface area compared with 570 Å on average for a crystal packing interface21. A handful of inter-RRM hydrogen bonds are apparent between the side chains of RRM1-N155 and RRM2-K292, RRM1-N155 and RRM2-D272 as well as the backbone atoms of RRM1-G221 and RRM2-D273 (Fig. 4c). This minor U2AF RRM1/RRM2 interface, coupled with the versatile sequence of the inter-RRM linker, highlighted the potential role for inter-RRM conformational dynamics in U2AF-splice site recognition. Paramagnetic resonance enhancement (PRE) measurements previously had suggested a predominant back-to-back, or ‘closed' conformation of the apo-U2AF1,2 RRM1 and RRM2 in equilibrium with a minor ‘open' conformation resembling the RNA-bound inter-RRM arrangement16. Yet, small-angle X-ray scattering (SAXS) data indicated that both the minimal U2AF1,2 and longer constructs comprise a highly diverse continuum of conformations in the absence of RNA that includes the ‘closed' and ‘open' conformations2526. To complement the static portraits of U2AF1,2L structure that we had determined by X-ray crystallography, we used smFRET to characterize the probability distribution functions and time dependence of U2AF inter-RRM conformational dynamics in solution. The inter-RRM dynamics of U2AF were followed using FRET between fluorophores attached to RRM1 and RRM2 (Fig. 6a,b, Methods). The positions of single cysteine mutations for fluorophore attachment (A181C in RRM1 and Q324C in RRM2) were chosen based on inspection of the U2AF1,2L structures and the ‘closed' model of apo-U2AF1,2. Criteria included (i) residue locations that are distant from and hence not expected to interfere with the RRM/RNA or inter-RRM interfaces, (ii) inter-dye distances (50 Å for U2AF1,2L–Py tract and 30 Å for the closed apo-model) that are expected to be near the Förster radius (Ro) for the Cy3/Cy5 pair (56 Å)27, where changes in the efficiency of energy transfer are most sensitive to distance, and (iii) FRET efficiencies that are calculated to be significantly greater for the ‘closed' apo-model as opposed to the ‘open' RNA-bound structures (by ∼30%). The FRET efficiencies of either of these structurally characterized conformations also are expected to be significantly greater than elongated U2AF conformations that lack inter-RRM contacts. Double-cysteine variant of U2AF1,2 was modified with equimolar amount of Cy3 and Cy5. Only traces that showed single photobleaching events for both donor and acceptor dyes and anti-correlated changes in acceptor and donor fluorescence were included in smFRET data analysis. Hence, molecules that were conjugated to two donor or two acceptor fluorophores were excluded from analysis. We first characterized the conformational dynamics spectrum of U2AF in the absence of RNA (Fig. 6c,d; Supplementary Fig. 7a,b). The double-labelled U2AF1,2L(Cy3/Cy5) protein was tethered to a slide via biotin-NTA/Ni resin. Virtually no fluorescent molecules were detected in the absence of biotin-NTA/Ni, which demonstrates the absence of detectable non-specific binding of U2AF1,2L to the slide. The FRET distribution histogram built from more than a thousand traces of U2AF1,2L(Cy3/Cy5) in the absence of ligand showed an extremely broad distribution centred at a FRET efficiency of ∼0.4 (Fig. 6d). Approximately 40% of the smFRET traces showed apparent transitions between multiple FRET values (for example, Fig. 6c). Despite the large width of the FRET-distribution histogram, the majority (80%) of traces that showed fluctuations sampled only two distinct FRET states (for example, Supplementary Fig. 7a). Approximately 70% of observed fluctuations were interchanges between the ∼0.65 and ∼0.45 FRET values (Supplementary Fig. 7b). We cannot exclude a possibility that tethering of U2AF1,2L(Cy3/Cy5) to the microscope slide introduces structural heterogeneity into the protein and, thus, contributes to the breadth of the FRET distribution histogram. However, the presence of repetitive fluctuations between particular FRET values supports the hypothesis that RNA-free U2AF samples several distinct conformations. This result is consistent with the broad ensembles of extended solution conformations that best fit the SAXS data collected for U2AF1,2 as well as for a longer construct (residues 136–347)25. We conclude that weak contacts between the U2AF RRM1 and RRM2 permit dissociation of these RRMs in the absence of RNA. We next used smFRET to probe the conformational selection of distinct inter-RRM arrangements following association of U2AF with the AdML Py-tract prototype. Addition of the AdML RNA to tethered U2AF1,2L(Cy3/Cy5) selectively increases a fraction of molecules showing an ∼0.45 apparent FRET efficiency, suggesting that RNA binding stabilizes a single conformation, which corresponds to the 0.45 FRET state (Fig. 6e,f). To assess the possible contributions of RNA-free conformations of U2AF and/or structural heterogeneity introduced by tethering of U2AF1,2L(Cy3/Cy5) to the slide to the observed distribution of FRET values, we reversed the immobilization scheme. We tethered the AdML RNA to the slide via a biotinylated oligonucleotide DNA handle and added U2AF1,2L(Cy3/Cy5) in the absence of biotin-NTA resin (Fig. 6g,h; Supplementary Fig. 7c–g). A 0.45 FRET value was again predominant, indicating a similar RNA-bound conformation and structural dynamics for the untethered and tethered U2AF1,2L(Cy3/Cy5). We examined the effect on U2AF1,2L conformations of purine interruptions that often occur in relatively degenerate human Py tracts2. We introduced an rArA purine dinucleotide within a variant of the AdML Py tract (detailed in Methods). Insertion of adenine nucleotides decreased binding affinity of U2AF65 to RNA by approximately five-fold. Nevertheless, in the presence of saturating concentrations of rArA-interrupted RNA slide-tethered U2AF651,2LFRET(Cy3/Cy5) showed a prevalent ∼0.45 apparent FRET value (Fig. 6i,j), which was also predominant in the presence of continuous Py tract. Therefore, RRM1-to-RRM2 distance remains similar regardless of whether U2AF is bound to interrupted or continuous Py tract. The inter-fluorophore distances derived from the observed 0.45 FRET state agree with the distances between the α-carbon atoms of the respective residues in the crystal structures of U2AF1,2L bound to Py-tract oligonucleotides. It should be noted that inferring distances from FRET values is prone to significant error because of uncertainties in the determination of fluorophore orientation factor κ and Förster radius R0, the parameters used in distance calculations28. Nevertheless, the predominant 0.45 FRET state in the presence of RNA agrees with the Py-tract-bound crystal structure of U2AF1,2L. Importantly, the majority of traces (∼70%) of U2AF1,2L(Cy3/Cy5) bound to the slide-tethered RNA lacked FRET fluctuations and predominately exhibited a ∼0.45 FRET value (for example, Fig. 6g). The remaining ∼30% of traces for U2AF1,2L(Cy3/Cy5) bound to the slide-tethered RNA showed fluctuations between distinct FRET values. The majority of traces that show fluctuations began at high (0.65–0.8) FRET value and transitioned to a ∼0.45 FRET value (Supplementary Fig. 7c–g). Hidden Markov modelling analysis of smFRET traces suggests that RNA-bound U2AF1,2L can sample at least two other conformations corresponding to ∼0.7–0.8 and ∼0.3 FRET values in addition to the predominant conformation corresponding to the 0.45 FRET state. Although a compact conformation (or multiple conformations) of U2AF1,2L corresponding to ∼0.7–0.8 FRET values can bind RNA, on RNA binding, these compact conformations of U2AF1,2L transition into a more stable structural state that corresponds to ∼0.45 FRET value and is likely similar to the side-by-side inter-RRM-arrangement of the U2AF1,2L crystal structures. Thus, the sequence of structural rearrangements of U2AF observed in smFRET traces (Supplementary Fig. 7c–g) suggests that a ‘conformational selection' mechanism of Py-tract recognition (that is, RNA ligand stabilization of a pre-configured U2AF conformation) is complemented by ‘induced fit' (that is, RNA-induced rearrangement of the U2AF RRMs to achieve the final ‘side-by-side' conformation), as discussed below. The U2AF structures and analyses presented here represent a successful step towards defining a molecular map of the 3′ splice site. Several observations indicate that the numerous intramolecular contacts, here revealed among the inter-RRM linker and RRM1, RRM2, and the N-terminal RRM1 extension, synergistically coordinate U2AF–Py-tract recognition. Truncation of U2AF to the core RRM1–RRM2 region reduces its RNA affinity by 100-fold. Likewise, deletion of 20 inter-RRM linker residues significantly reduces U2AF–RNA binding only when introduced in the context of the longer U2AF1,2L construct comprising the RRM extensions, which in turn position the linker for RNA interactions. Notably, a triple mutation of three residues (V254P, Q147A and R227A) in the respective inter-RRM linker, N- and C-terminal extensions non-additively reduce RNA binding by 150-fold. Altogether, these data indicate that interactions among the U2AF RRM1/RRM2, inter-RRM linker, N-and C-terminal extensions are mutually inter-dependent for cognate Py-tract recognition. The implications of this finding for U2AF conservation and Py-tract recognition are detailed in the Supplementary Discussion. Recently, high-throughput sequencing studies have shown that somatic mutations in pre-mRNA splicing factors occur in the majority of patients with myelodysplastic syndrome (MDS)29. MDS-relevant mutations are common in the small U2AF subunit (U2AF, or U2AF1), yet such mutations are rare in the large U2AF subunit (also called U2AF2)—possibly due to the selective versus nearly universal requirements of these factors for splicing. A confirmed somatic mutation of U2AF in patients with MDS, L187V30, is located on a solvent-exposed surface of RRM1 that is distinct from the RNA interface (Fig. 7a). This L187 surface is oriented towards the N terminus of the U2AF1,2L construct, where it is expected to abut the U2AF-binding site in the context of the full-length U2AF heterodimer. Likewise, an unconfirmed M144I mutation reported by the same group corresponds to the N-terminal residue of U2AF1,2L, which is separated by only ∼20 residues from the U2AF-binding site31. As such, we suggest that the MDS-relevant U2AF mutations contribute to MDS progression indirectly, by destabilizing a relevant conformation of the conjoined U2AF subunit rather than affecting U2AF functions in RNA binding or spliceosome recruitment per se. Our smFRET results agree with prior NMR/PRE evidence for multi-domain conformational selection16 as one mechanistic basis for U2AF–RNA association (Fig. 7b). The ‘induced fit' versus ‘conformational selection' models are the prevailing views of the mechanisms underlying bio-molecular interactions (reviewed in ref. 32). In the former, ligand binding promotes a subsequent conformational change in the protein, whereas in the latter, the ligand selects a protein conformation from a pre-existing ensemble and thereby shifts the population towards that state. An ∼0.45 FRET value is likely to correspond to the U2AF conformation visualized in our U2AF1,2L crystal structures, in which the RRM1 and RRM2 bind side-by-side to the Py-tract oligonucleotide. The lesser 0.65–0.8 and 0.2–0.3 FRET values in the untethered U2AF1,2LFRET(Cy3/Cy5) experiment could correspond to respective variants of the ‘closed', back-to-back U2AF conformations characterized by NMR/PRE data16, or to extended U2AF conformations, in which the intramolecular RRM1/RRM2 interactions have dissociated the protein is bound to RNA via single RRMs. An increased prevalence of the ∼0.45 FRET value following U2AF–RNA binding, coupled with the apparent absence of transitions in many ∼0.45-value single molecule traces (for example, Fig. 6e), suggests a population shift in which RNA binds to (and draws the equilibrium towards) a pre-configured inter-RRM proximity that most often corresponds to the ∼0.45 FRET value. Notably, our smFRET results reveal that U2AF–Py-tract recognition can be characterized by an ‘extended conformational selection' model (Fig. 7b). In this recent model for macromolecular interactions3233, the pure ‘conformational selection' and ‘induced fit' scenarios represent the limits of a mechanistic spectrum and may compete or occur sequentially. Examples of ‘extended conformational selection' during ligand binding have been characterized for a growing number of macromolecules (for example, adenylate kinase3435, LAO-binding protein36, poly-ubiquitin37, maltose-binding protein38 and the preQ1 riboswitch39, among others). Here, the majority of changes in smFRET traces for U2AF1,2L(Cy3/Cy5) bound to slide-tethered RNA began at high (0.65–0.8) FRET value and transition to the predominant 0.45 FRET value (Supplementary Fig. 7c–g). These transitions could correspond to rearrangement from the ‘closed' NMR/PRE-based U2AF conformation in which the RNA-binding surface of only a single RRM is exposed and available for RNA binding16, to the structural state seen in the side-by-side, RNA-bound crystal structure. As such, the smFRET approach reconciles prior inconsistencies between two major conformations that were detected by NMR/PRE experiments16 and a broad ensemble of diverse inter-RRM arrangements that fit the SAXS data for the apo-protein2526. Similar interdisciplinary structural approaches are likely to illuminate whether similar mechanistic bases for RNA binding are widespread among other members of the vast multi-RRM family. The finding that U2AF recognizes a nine base pair Py tract contributes to an elusive ‘code' for predicting splicing patterns from primary sequences in the post-genomic era (reviewed in ref. 40). Based on (i) similar RNA affinities of U2AF and U2AF1,2L, (ii) indistinguishable conformations among four U2AF1,2L structures in two different crystal packing arrangements and (iii) penalties of structure-guided mutations in RNA binding and splicing assays, we suggest that the extended inter-RRM regions of the U2AF1,2L structures underlie cognate Py-tract recognition by the full-length U2AF protein. Further research will be needed to understand the roles of SF1 and U2AF subunits in the conformational equilibria underlying U2AF association with Py tracts. Moreover, structural differences among U2AF homologues and paralogues may regulate splice site selection. Ultimately, these guidelines will assist the identification of 3′ splice sites and the relationship of disease-causing mutations to penalties for U2AF association. For crystallization and RNA-binding experiments, human U2AF1,2L (residues 141–342 of NCBI RefSeq NP_009210) was expressed in Escherichia coli strain BL21 Rosetta-2 as a GST-fusion protein in the vector pGEX6P-2 and purified by glutathione affinity, followed by anion exchange and gel filtration chromatography. The GST-tagged protein was bound to a GSTrap column (GE Healthcare) in 1 M NaCl, 25 mM HEPES, pH 7.4 and eluted using 150 mM NaCl, 100 mM Tris, pH 8 containing 10 mM glutathione. The GST tag was cleaved from the protein by treatment with PreScission Protease during dialysis against a buffer containing 100 mM NaCl, 25 mM HEPES, pH 8, 5% (v/v) glycerol, 5 mM DTT, 0.25 mM EDTA and 0.1 mM PMSF. Cleaved GST was separated from the U2AF1,2L by subtractive glutathione affinity chromatography in 100 mM NaCl, 25 mM Tris, pH 8, 0.2 mM TCEP followed by subtractive anion-exchange chromatography with a HiTrap Q column (GE Healthcare). The final purification step was size-exclusion chromatography on a Superdex-75 prep-grade column (GE Healthcare) that had been previously equilibrated with 100 mM NaCl, 15 mM HEPES, pH 6.8, 0.2 mM tris(2-carboxy-ethyl)phosphine (TCEP). The purified U2AF1,2L was concentrated using a Vivaspin 15 R (Sartorius) centrifugal concentrator with 10 kDa MWCO, and the protein concentration was estimated using the calculated extinction coefficient of 8,940 Mcm and absorbance at 280 nm. Shorter constructs (U2AF1,2, residues 148–336; dU2AF1,2, residues 148–237, 258–336; dU2AF1,2L, residues 141–237, 258–342) (Fig. 1a) and individual U2AF1,2L Q147A, R227A, V254P mutants used for RNA-binding experiments were purified similarly. For comparative RNA-binding experiments, full-length human U2AF (residues 1–475) and the U2AF-UHM (U2AF homology motif; residues 43–146, NCBI RefSeq NP_006749) initially were expressed and purified separately as GST fusion proteins. Following GST cleavage and ion-exchange chromatography (SP-HiTrap and Q-HiTrap, respectively), U2AF was combined with slight excess U2AF-UHM (in stoichiometric ratio of 1:1.2) and dialysed overnight. The final U2AF heterodimer was purified by size-exclusion chromatography using a Superdex-200 prep-grade column (GE Healthcare) pre-equilibrated with 150 mM NaCl, 25 mM HEPES, pH 6.8, 0.2 mM TCEP. Representative purified U2AF1,2L and U2AF65–U2AF-UHM proteins are shown in Supplementary Fig. 1a. High-performance liquid chromatography-purified oligonucleotides (sequences shown in Supplementary Fig. 2a) were purchased for crystallization (Integrated DNA Technologies, Inc.). The lyophilized oligonucleotides were diluted in gel filtration buffer for crystallization experiments. The 5′-fluorescein (Fl)-labelled RNAs (AdML: 5′-Fl-CCCUUUUUUUUCC-3′, Py tract of the AdML splicing substrate; 5′-4rU: 5′-Fl-CCUUUUCCCCCCC-3′; 3′-4rU: 5′-Fl-CCCCCCCUUUUCC-3′) for RNA-binding experiments (Dharmacon Research, Inc., Thermo Scientific) was deprotected according to the manufacturer's protocol, vacuum dried and resuspended in nuclease-free water. RNA and RNA–DNA concentrations were calculated using the calculated molar extinction coefficients41 and absorbance at 260 nm. For RNA-binding experiments, purified proteins and RNA were diluted separately >100-fold in binding buffer (100 mM NaCl, 15 mM HEPES, pH 6.8, 0.2 mM TCEP, 0.1 U μl Superase-In (Ambion Life Technologies)). The final RNA concentration in the cuvette was 30 nM. Volume changes during addition of the protein were <10% to minimize dilution effects. The fluorescence anisotropy changes during titration were measured using a FluoroMax-3 spectrophotometer temperature controlled by a circulating water bath at 23 °C. Samples were excited at 490 nm and emission intensities recorded at 520 nm with a slit width of 5 nm. The titrations were repeated three times in succession. Each titration was fit with Graphpad Prism v4.0 to obtain the apparent equilibrium dissociation constant (KD)22. The apparent equilibrium affinities (KA) are the reciprocal of the KD. The average KD's or KA's and s.e.m. among the three replicates were calculated using Excel and are reported in Figs 3 and 4; Supplementary Figs 1 and 4. The P values from a two-tailed unpaired t-test were calculated using Graphpad Prism v4.0. For transfection experiments, the full-length human U2AF cDNA in pCMV6-XL5 (Origene Tech. Inc., clone ID BC008740) was used (WT U2AF) and in parallel mutated to encode the Q147A/R227A/V254P triple-mutant protein (Mut U2AF). The pyPY minigene was a gift from M. Carmo-Fonseca (University of Lisbon, Portugal)24. HEK293T cells (kindly provided by Dr Lata Balakrishnan, originally purchased from ATCC, cat. no. CRL3216) were seeded into 12-well plates (2–4 × 10 cells per well) and grown as monolayers in MEM (Gibco Life Technologies) supplemented with 10% (v/v) of heat-inactivated fetal bovine serum, 1% (v/v) L-glutamine and 1% (v/v) penicillin–streptomycin. After 1 day, the cells were transiently transfected with either 0.5 μg of pyPY plasmid or a mixture of 0.5 μg of U2AF variant and 0.5 μg of pyPY plasmid per well using appropriately adjusted Lipofectamine 2000 (Invitrogen Life Technologies) ratio according to the manufacturer's instructions. For immunoblots of WT U2AF and Mut U2AF expression levels (Supplementary Fig. 5a), transfected or control cells were lysed in radioimmunoprecipitation assay buffer with proteinase and kinase inhibitors. Total protein (20 μg) was separated by SDS–PAGE, and transferred onto polyvinylidene difluoride membranes (Millipore Corp., Billerica, MA, USA) and immunoblotted using mouse monoclonal antibodies directed against U2AF (ref. 42) (MC3, cat. no. U4758 Sigma-Aldrich at 1:500 dilution) or as a control for comparison, GAPDH (glyceraldehyde-3-phosphate dehydrogenase; monoclonal clone 71.1, cat. no. G8795 Sigma-Aldrich at 1:5,000 dilution). Immunoblots were developed using anti-mouse horseradish peroxidase-conjugates (cat. no. U4758 Sigma-Aldrich, Co. at 1:2,500 or 1:10,000 dilutions for GAPDH and U2AF, respectively) and detected using SuperSignal WestPico chemi-luminescent substrate (Pierce Thermo Scientific Inc.). Blots were imaged using a IS4000MM system (Carestream, Rochester, NY, USA). For size analysis, fluorescent images of the BioRad Precision Plus Dual Color Standards were overlaid directly. For reverse transcription PCR (RT-PCR), the total RNA was isolated 2 days post transfection using the Cells-to-cDNA II kit (Ambion Life Technologies). The RT-PCR reaction comprised 35 cycles (94 °C per 60 s—60 °C per 50 s—72 °C per 60 s) with forward (5′-TGAGGGGAGGTGAATGAGGAG-3′) and reverse (5′-TCCACTGGAAAGACCGCGAAG-3′) primers for the pyPY product or forward (5′-CATGTTCGTCATGGGTGTGAACCA-3′) and reverse (5′-ATGGCATGGACTGTGGTCATGAGT-3′) primers for a GAPDH control. The RT-PCR products were separated by 2% agarose gel electrophoresis and stained with ethidium bromide. The percentages of splice site use were calculated from the background corrected intensities I using the formula 100% × I(py)/[I(py)+I(PY)+I(unspliced)] for py spliced (Fig. 5b,c) or 100% × I(PY)/[I(py)+I(PY)+I(unspliced)] for PY spliced (Supplementary Fig. 5b). The band intensities of four independent biological replicates were measured using ImageQuant software. Before crystallization, the purified U2AF1,2L and given oligonucleotide were mixed to achieve respective final concentrations of 1.0 and 1.1 mM and incubated on ice for 20–30 min. For each oligonucleotide, sparse matrix screens of the Jancarik and Kim Crystal Screen43(in hanging drop format; Hampton Research, Corp.) and JCSG-Plus (in sitting drop format; Molecular Dimensions) were used to identify initial crystallization conditions, which were obtained from the latter screen and further optimized in hanging drop format. In optimized crystallization experiments, a mixture of sample and reservoir solution (1.2:1 μl) was equilibrated against 700 μl reservoir solution at 4 °C. The oligonucleotide sequences were optimized and the structures were determined as follows: in addition to the previously characterized dU2AF1,2-binding sites for seven nucleotides, the new terminal residues of the U2AF1,2L construct were presumed to contact an additional nucleotide and the crystal packing of a central nucleotide between the RRM1/RRM2 of dU2AF1,2 was presumed to represent one nucleotide. Also considering the known proclivity for deoxy(d)U to co-crystallize with dU2AF1,2 (ref. 44) and for 5-bromo-dU (5BrdU) to bind a given site of dU2AF1,2 (ref. 14), we initially designed two 9-mer oligonucleotides (5′-ribose (r)UrUrUrUrU(5BrdU)dUrUrU and 5′-rUrUrUdUdU(5BrdU)dUrUrU) and screened for co-crystallization with U2AF1,2L. The former oligonucleotide failed to produce crystals in these screens. The latter oligonucleotide comprising central dU nucleotides produced diffracting crystals, which were frozen directly from a reservoir comprising 100 mM phosphate–citrate buffer pH 4.2, 40% Peg 300. The structure determined by molecular replacement using Phenix45 with a data set collected at beamline (BL) 12-2 of the Stanford Synchrotron Radiation Lightsource (SSRL; Menlo Park, CA, USA) (Table 1). The search models comprising each of the individual RRMs bound to two nucleotides were derived from the dU2AF1,2 structure (PDB ID 2G4B) (translation function Z-score equivalent 12.9, log-likelihood gain 528). For comparison, searches with the NMR structure (PDB ID 2YH1) as a search model failed to find a solution. The initial structure revealed a greater number of central nucleotide-binding sites than expected. The oligonucleotide binding register had slipped to place the BrdU in the preferred site, leave the 5′ terminal-binding sites empty, and the terminal nucleotide unbound and disordered. Subsequent oligonucleotides were designed to place BrdU in the preferred site, fill the unoccupied 5′ terminal sites, capture rU at the central sites, and compare rC at the terminal site. The U2AF1,2L protein co-crystals with oligonucleotide 5′-phosphorylated (P)-rUrUdUdUrUdU(BrdU)dU were obtained using a reservoir of 200 mM LiCl, 100 mM sodium citrate pH 4.0, 8% (w/v) polyethylene glycol (PEG) 6,000, 10% (v/v) PEG 300, 10% (v/v) dioxane with 0.1 μl of N,N-bis[3-(D-gluconamido)propyl]deoxy-cholamide (deoxy-BigCHAP) (14 mM) added to the hanging drop and cryoprotected by sequential layering with reservoir solution supplemented with increasing PEG 300 to a final concentration of 26%. Co-crystals with either 5′-(P)rUrUdUrUrU(BrdU)dUdU or 5′-(P)rUrUrUdUrUrU(BrdU)dUrC were obtained from 1 M succinate, 100 mM HEPES, pH 7.0, 1–3% (w/v) PEG monomethylether 2,000. The former was cryoprotected by coating with a 1:1 (v/v) mixture of silicon oil and Paratone-N and the latter by sequential transfer to 21% (v/v) glycerol. Data sets for flash-cooled crystals were collected at 100 K using remote access to SSRL BL12-2. Structures were determined by molecular replacement using the initial U2AF1,2L/rUrUrUdUdU(BrdU)dUrUrU structure as a search model. Consistent sets of free-R reflections were maintained (6% of the total reflections). Models were built using COOT46 and refined with PHENIX45. No non-glycine/non-proline residues were found in the disallowed regions of the Ramachandran plots. Clash scores and Molprobity scores calculated using the program Molprobity47 were above average. Structure illustrations were prepared using PYMOL48. Crystallographic data and refinement statistics are given in Table 1. The U2AF1,2L construct used for smFRET comprises the six histidine and T7 tags from the pET28a vector (Merck), a GGGS linker and U2AF residues 113–343. The single cysteine of human U2AF was replaced by alanine (C305A), which is a natural amino-acid variation among U2AF homologues. Single A181C and Q324C mutations were introduced in each RRM for fluorophore attachment at residues that were carefully selected to meet experimental criteria described in the Results. The U2AF1,2L was purified by the same method as described above for U2AF1,2L protein and binds RNA with similar affinity as U2AF1,2L (Supplementary Fig. 6a). Before labelling, the purified U2AF1,2L protein was incubated with 10 mM DTT on ice for 30 min and then buffer exchanged into Labelling Buffer (100 mM NaCl, 25 mM HEPES pH 7.0, 5 mM EDTA, 0.5 mM tris(2-carboxy-ethyl)phosphine (TCEP)) using Zeba Spin Desalting Columns 7K MWCO (Pierce, ThermoFisher Scientific). To initiate the labelling reaction, 4 μl each of cyanine (Cy)3-Maleimide and Cy5-Maleimide (Combinix, Inc.) stock solutions (10 mM in DMSO) were pre-mixed (total volume 8 μl) and then added to 200 μl of 20 μM protein (final 20:1 molar ratio of dye:protein). The labelling reaction was incubated at room temperature in the dark for 2 h and then quenched by the addition of 10 mM DTT. The labelled protein was separated from excess dye using a Zeba Spin Desalting Column followed by size exclusion chromatography using a pre-packed Superdex-75 10/300 GL (GE Healthcare) column in Labelling Buffer. Our previous experience of conjugating cysteines with maleimide derivatives of fluorophores and suggests that nonspecific modification of aminogroups of proteins with fluorescent dyes under the employed experimental conditions is negligible. Consistent with specific labelling of A181C and Q324C, the labelling efficiencies were ∼60% each for Cy3 and Cy5 as estimated using the dye extinction coefficients (ɛ=150,000 M cm at 550 nm, ɛ=170,000 M cm at 650 nm) and the calculated extinction coefficient of the U2AF1,2L protein (ɛ=8,940 M cm at 280 nm), and correcting for the absorbance (A) of the dyes at 280 nm (GE Healthcare, Amersham CyDye Maleimide product booklet): For smFRET experiments with a ‘strong', homogeneous Py tract, we used the prototypical AdML sequence (5′-CCUUUUUUUUCC-3′). To investigate the inter-RRM separation in the presence of a ‘weak' Py tract interrupted by purines, we compared the U2AF1,2L affinity for a purine-interrupted Py tract comprising an rUrUrUrUrU tract that is expected to bind U2AF RRM2/inter-RRM linker, a central rArA and an rUrUrUrCrC tract that is expected to bind RRM1. The tandem purines represent a compromise between significant inhibition of U2AF binding by longer A interruptions16 and an approximately five-fold penalty for the rArA mutation in the AdML Py tract (Supplementary Fig. 6b,c). To maintain avidity and provide flanking phosphoryl groups in case of inter-RRM adjustment, we included the 5′-C and 3′-A of parent AdML sequence, which are respective low-affinity nucleotides for binding RRM2 and RRM1 (ref. 14), in the final rArA-interrupted RNA oligonucleotide (5′-rCrUrUrUrUrUrArArUrUrUrCrCrA-3′). For the reversed immobilization of RNA via a complementary biotinyl-DNA primer experiment, the AdML Py-tract RNA was extended to include the DNA counterpart of downstream AdML intron/exon sequences that were complementary to the biotinyl-DNA primer. To increase separation from the slide surface, a hexaethylene glycol linker (18PEG) was inserted between the AdML Py-tract RNA and the tethered DNA duplex. The tethered oligonucleotide sequences included: 5′-rCrCrUrUrUrUrUrUrUrUrCrC/18PEG/dAdCdAdGdCdTdCdGdCdG-dGdTdTdGdAdGdGdAdCdAdA-3′ annealed to 5′-biotinyl-dTdTdGdTdCdCdTdCdAdA-dCdCdGdCdGdAdGdCdTdGdT-3' (purchased with high-performance liquid chromatography purification from Integrated DNA Technologies). The smFRET measurements were carried out at room temperature in 50 mM HEPES, pH 7.4, 100 mM NaCl. The imaging buffer also contained an oxygen-scavenging system (0.8 mg ml glucose oxidase, 0.625% glucose, 0.02 mg ml catalase), 1.5 mM Trolox (used to eliminate Cy5 blinking) and 6 mM β-mercaptoethanol. The sample chamber was assembled from quartz microscope slides and glass cover slips coated with a mixture of m-PEG and biotin-PEG and pre-treated with neutravidin (0.2 mg ml). Surface tethering of doubly labelled U2AF1,2L(Cy3/Cy5) via its His-tag (Fig. 6c–f,i,j; Supplementary Fig. 7a,b) was achieved by pre-incubating the sample chamber with 50 nM biotinyl-NTA resin (Biotin-X NTA, Biotium), pre-loaded with three-fold excess NiSO4) for 20 min before addition of 5 nM U2AF1,2L(Cy3/Cy5). After 10 min, unbound sample was removed by washing the sample chamber with imaging buffer. The AdML RNA ligand was added to the imaging buffer at a concentration of 5 μM (100-fold higher than the measured KD value), whereas the rArA-interrupted RNA was added at a concentration of 10 μM. Alternatively, to detect binding of doubly labelled U2AF1,2L(Cy3/Cy5) to surface-tethered RNA ligand (Fig. 6g,h Supplementary Fig. 7c–g), 10 nM AdML RNA (pre-annealed to biotinyl-DNA primer) was incubated in the neutravidin-treated sample chamber for 20 min, and 1 nM U2AF1,2L(Cy3/Cy5) was then added to the imaging buffer. Single-molecule FRET measurements were taken as previously described2849. An Olympus IX71 inverted microscope, equipped with a UPlanApo 60x/1.20w objective lens, a 532 nm laser (Spectra-Physics) for excitation of Cy3 dyes, and a 642 nm laser (Spectra-Physics) for excitation of Cy5 dyes was used. Total internal reflection (TIR) was obtained by a quartz prism (ESKMA Optics). Fluorescence emission was split into Cy3 and Cy5 fluorescence using a dual view imaging system DV2 (Photometrics) equipped with a 630 nm dichroic mirror and recorded via an Andor iXon+ EMCCD camera. Movies were recorded using the Single software (downloaded from Prof. Taekjip Ha's laboratory website at the University of Illinois at Urbana-Champaign, physics.illinois.edu/cplc/software), with the exposure time set at 100 ms. We typically took up to five 5-minute-long movies while imaging different sections of the slide for each sample. Before each measurement, we checked for non-specific binding by adding doubly-labeled U2Fret to the slide in the absence of neutravidin and imaging the slide. Non-specific binding was virtually absent. Collected data sets were processed with IDL and Matlab softwares, using scripts downloaded from a freely available source: physics.illinois.edu/cplc/software. Apparent FRET efficiencies (Eapp) were calculated from the emission intensities of donor (ICy3) and acceptor (ICy5) as follows: Eapp=ICy5/(ICy5+ICy3). The FRET distribution histograms were built from traces that showed single-step photobleaching in both Cy3 and Cy5 signals using a Matlab script generously provided by Prof. Peter Cornish (University of Missouri, Columbia). Anti-correlated changes in donor and acceptor intensities with constant sum of intensities indicated the presence of an energy transfer in single molecules labelled with one donor and one acceptor dye. All histograms were smoothed with a five-point window and plotted using Origin software (Origin Lab Co). Idealization of FRET trajectories was done using the hidden Markov model algorithms via HaMMy software (http://bio.physics.illinois.edu/HaMMy.asp)50. Transition density plots were generated from transitions detected in idealized FRET trajectories obtained by HaMMy fit of raw FRET traces via Matlab. Frequency of transitions from starting FRET efficiency value (x-axis) to ending FRET efficiency value (y-axis) was represented by a heat map. The range of FRET efficiencies from 0 to 1 was separated in 200 bins. The resulting heat map was normalized to the most populated bin in the plot; the lower- and upper-bound thresholds were set to 20% and 100% of the most populated bin, respectively. The surface contour plots were generated as follows: the individual single-molecule FRET traces (for example, Fig. 6g of the main text and Supplementary Fig. 7e,f) were post synchronized at the first time point showing non-zero (>0.15) FRET efficiency, corresponding to binding. The time range (x-axis, 0–10 s) was separated into 100 bins. The FRET efficiency range (y-axis, 0–1 FRET) was separated into 100 bins. A heat map is used to represent the frequency of sampling of each FRET state over time; frequency in each bin was normalized to the most populated bin in the plot with lower- and upper-bound thresholds set at 10% and 80% of the most populated bin, respectively. Accession codes: Coordinates and structure factors have been deposited in the Protein Data Bank with accession codes 5EV1, 5EV2, 5EV3 and 5EV4 for respective U2AF1,2L-oligonucleotide structures (i)–(iv). How to cite this article: Agrawal, A. A. et al. An extended U2AF–RNA-binding domain recognizes the 3′ splice site signal. Nat. Commun. 7:10950 doi: 10.1038/ncomms10950 (2016).
PMC4880283	Crystal Structures of Putative Sugar Kinases from Synechococcus Elongatus PCC 7942 and Arabidopsis Thaliana	The genome of the Synechococcus elongatus strain PCC 7942 encodes a putative sugar kinase (SePSK), which shares 44.9% sequence identity with the xylulose kinase-1 (AtXK-1) from Arabidopsis thaliana. Sequence alignment suggests that both kinases belong to the ribulokinase-like carbohydrate kinases, a sub-family of FGGY family carbohydrate kinases. However, their exact physiological function and real substrates remain unknown. Here we solved the structures of SePSK and AtXK-1 in both their apo forms and in complex with nucleotide substrates. The two kinases exhibit nearly identical overall architecture, with both kinases possessing ATP hydrolysis activity in the absence of substrates. In addition, our enzymatic assays suggested that SePSK has the capability to phosphorylate D-ribulose. In order to understand the catalytic mechanism of SePSK, we solved the structure of SePSK in complex with D-ribulose and found two potential substrate binding pockets in SePSK. Using mutation and activity analysis, we further verified the key residues important for its catalytic activity. Moreover, our structural comparison with other family members suggests that there are major conformational changes in SePSK upon substrate binding, facilitating the catalytic process. Together, these results provide important information for a more detailed understanding of the cofactor and substrate binding mode as well as the catalytic mechanism of SePSK, and possible similarities with its plant homologue AtXK-1.Carbohydrates are essential cellular compounds involved in the metabolic processes present in all organisms. Phosphorylation is one of the various pivotal modifications of carbohydrates, and is catalyzed by specific sugar kinases. These kinases exhibit considerable differences in their folding pattern and substrate specificity. Based on sequence analysis, they can be divided into four families, namely HSP 70_NBD family, FGGY family, Mer_B like family and Parm_like family. The FGGY family carbohydrate kinases contain different types of sugar kinases, all of which possess different catalytic substrates with preferences for short-chained sugar substrates, ranging from triose to heptose. These sugar substrates include L-ribulose, erythritol, L-fuculose, D-glycerol, D-gluconate, L-xylulose, D-ribulose, L-rhamnulose and D-xylulose . Structures reported in the Protein Data Bank of the FGGY family carbohydrate kinases exhibit a similar overall architecture containing two protein domains, one of which is responsible for the binding of substrate, while the second is used for binding cofactor ATP. While the binding pockets for substrates are at the same position, each FGGY family carbohydrate kinases uses different substrate-binding residues, resulting in high substrate specificity [2–4]. Synpcc7942_2462 from the cyanobacteria Synechococcus elongatus PCC 7942 encodes a putative sugar kinase (SePSK), and this kinase contains 426 amino acids. The At2g21370 gene product from Arabidopsis thaliana, xylulose kinase-1 (AtXK-1), whose mature form contains 436 amino acids, is located in the chloroplast (ChloroP 1.1 Server) [5, 6]. SePSK and AtXK-1 display a sequence identity of 44.9%, and belong to the ribulokinase-like carbohydrate kinases, a sub-family of FGGY family carbohydrate kinases. Members of this sub-family are responsible for the phosphorylation of sugars similar to L-ribulose and D-ribulose. The sequence and the substrate specificity of ribulokinase-like carbohydrate kinases are different, but they share the common folding feature with two domains. Domain I exhibits a ribonuclease H-like folding pattern, and is responsible for the substrate binding, while domain II possesses an actin-like ATPase domain that binds cofactor ATP [1–4, 7–9]. Two possible xylulose kinases (xylulose kinase-1: XK-1 and xylulose kinase-2: XK-2) from Arabidopsis thaliana were previously proposed . It was shown that XK-2 (At5g49650) located in the cytosol is indeed xylulose kinase. However, the function of XK-1 (At2g21370) inside the chloroplast stroma has remained unknown. SePSK from Synechococcus elongatus strain PCC 7942 is the homolog of AtXK-1, though its physiological function and substrates remain unclear. In order to obtain functional and structural information about these two proteins, here we reported the crystal structures of SePSK and AtXK-1. Our findings provide new details of the catalytic mechanism of SePSK and lay the foundation for future studies into its homologs in eukaryotes. The attempt to solve the SePSK structure by molecular replacement method failed with ribulokinase from Bacillus halodurans (PDB code: 3QDK, 15.7% sequence identity) as an initial model . We therefore used single isomorphous replacement anomalous scattering method (SIRAS) for successful solution of the apo-SePSK structure at a resolution of 2.3 Å. Subsequently, the apo-SePSK structure was used as molecular replacement model to solve all other structures identified in this study. Our structural analysis showed that apo-SePSK consists of one SePSK protein molecule in an asymmetric unit. The amino-acid residues were traced from Val2 to His419, except for the Met1 residue and the seven residues at the C-termini. Apo-SePSK contains two domains referred to further on as domain I and domain II (Fig 1A). Domain I consists of non-contiguous portions of the polypeptide chains (aa. 2–228 and aa. 402–419), exhibiting 11 α-helices and 11 β-sheets. Among all these structural elements, α4/α5/α11/α18, β3/β2/β1/β6/β19/β20/β17 and α21/α32 form three patches, referred to as A1, B1 and A2, exhibiting the core region. In addition, four β-sheets (β7, β10, β12 and β16) and five α-helices (α8, α9, α13, α14 and α15) flank the left side of the core region. Domain II is comprised of aa. 229–401 and classified into B2 (β31/β29/β22/β23/β25/β24) and A3 (α26/α27/α28/α30) (Fig 1A and S1 Fig). In the SePSK structure, B1 and B2 are sandwiched by A1, A2 and A3, and the whole structure shows the A1/B1/A2/B2/A3 (α/β/α/β/α) folding pattern, which is in common with other members of FGGY family carbohydrate kinases (S2 Fig) [3, 4, 10]. The overall folding of SePSK resembles a clip, with A2 of domain I acting as a hinge region. As a consequence, a deep cleft is formed between the two domains. (A) Three-dimensional structure of apo-SePSK. The secondary structural elements are indicated (α-helix: cyan, β-sheet: yellow). (B) Three-dimensional structure of apo-AtXK-1. The secondary structural elements are indicated (α-helix: green, β-sheet: wheat). Apo-AtXK-1 exhibits a folding pattern similar to that of SePSK in line with their high sequence identity (Fig 1B and S1 Fig). However, superposition of structures of AtXK-1 and SePSK shows some differences, especially at the loop regions. A considerable difference is found in the loop3 linking β3 and α4, which is stretched out in the AtXK-1 structure, while in the SePSK structure, it is bent back towards the inner part. The corresponding residues between these two structures (SePSK-Lys35 and AtXK-1-Lys48) have a distance of 15.4 Å (S3 Fig). In order to understand the function of these two kinases, we performed structural comparison using Dali server. The structures most closely related to SePSK are xylulose kinase, glycerol kinase and ribulose kinase, implying that SePSK and AtXK-1 might function similarly to these kinases. We first tested whether both enzymes possessed ATP hydrolysis activity in the absence of substrates. As shown in Fig 2A, both SePSK and AtXK-1 exhibited ATP hydrolysis activity. This finding is in agreement with a previous result showing that xylulose kinase (PDB code: 2ITM) possessed ATP hydrolysis activity without adding substrate . To further identify the actual substrate of SePSK and AtXK-1, five different sugar molecules, including D-ribulose, L-ribulose, D-xylulose, L-xylulose and Glycerol, were used in enzymatic activity assays. As shown in Fig 2B, the ATP hydrolysis activity of SePSK greatly increased upon adding D-ribulose than adding other potential substrates, suggesting that it has D-ribulose kinase activity. In contrary, limited increasing of ATP hydrolysis activity was detected for AtXK-1 upon addition of D-ribulose (Fig 2C), despite its structural similarity with SePSK. (A) The ATP hydrolysis activity of SePSK and AtXK-1. Both SePSK and AtXK-1 showed ATP hydrolysis activity in the absence of substrate. While the ATP hydrolysis activity of SePSK greatly increases upon addition of D-ribulose (DR). (B) The ATP hydrolysis activity of SePSK with addition of five different substrates. The substrates are DR (D-ribulose), LR (L-ribulose), DX (D-xylulose), LX (L-xylulose) and GLY (Glycerol). (C) The ATP hydrolysis activity of SePSK and AtXK-1 with or without D-ribulose. (D) The ATP hydrolysis activity of wild-type (WT) and single-site mutants of SePSK. Three single-site mutants of SePSK are D8A-SePSK, T11A-SePSK and D221A-SePSK. The ATP hydrolysis activity measured via luminescent ADP-Glo assay (Promega). To understand the catalytic mechanism of SePSK, we performed structural comparisons among xylulose kinase, glycerol kinase, ribulose kinase and SePSK. Our results suggested that three conserved residues (D8, T11 and D221 of SePSK) play an important role in SePSK function. Mutations of the corresponding residue in xylulose kinase and glycerol kinase from Escherichia coli greatly reduced their activity [4, 11]. To identify the function of these three residues of SePSK, we constructed D8A, T11A and D221A mutants. Using enzymatic activity assays, we found that all of these mutants exhibit much lower activity of ATP hydrolysis after adding D-ribulose than that of wild type, indicating the possibility that these three residues are involved in the catalytic process of phosphorylation D-ribulose and are vital for the function of SePSK (Fig 2D). To obtain more detailed information of SePSK and AtXK-1 in complex with ATP, we soaked the apo-crystals in the reservoir adding cofactor ATP, and obtained the structures of SePSK and AtXK-1 bound with ATP at the resolution of 2.3 Å and 1.8 Å, respectively. In both structures, a strong electron density was found in the conserved ATP binding pocket, but can only be fitted with an ADP molecule (S4 Fig). Thus the two structures were named ADP-SePSK and ADP-AtXK-1, respectively. The extremely weak electron densities of ATP γ-phosphate in both structures suggest that the γ-phosphate group of ATP is either flexible or hydrolyzed by SePSK and AtXK-1. This result was consistent with our enzymatic activity assays where SePSK and AtXK-1 showed ATP hydrolysis activity without adding any substrates (Fig 2A and 2C). To avoid hydrolysis of ATP, we soaked the crystals of apo-SePSK and apo-AtXK-1 into the reservoir adding AMP-PNP. However, we found that the electron densities of γ-phosphate group of AMP-PNP (AMP-PNP γ-phosphate) are still weak in the AMP-PNP-SePSK and AMP-PNP-AtXK-1 structures, suggesting high flexibility of ATP-γ-phosphate. The γ-phosphate group of ATP is transferred to the sugar substrate during the reaction process, so this flexibility might be important for the ability of these kinases. The overall structures as well as the coordination modes of ADP and AMP-PNP in the AMP-PNP-AtXK-1, ADP-AtXK-1, ADP-SePSK and AMP-PNP-SePSK structures are nearly identical (S5 Fig), therefore the structure of AMP-PNP-SePSK is used here to describe the structural details and to compare with those of other family members. As shown in Fig 3A, one SePSK protein molecule is in an asymmetric unit with one AMP-PNP molecule. The AMP-PNP is bound at the domain II, where it fits well inside a positively charged groove. The AMP-PNP binding pocket consists of four α-helices (α26, α28, α27 and α30) and forms a shape resembling a half-fist (Fig 3A and 3B). The head group of the AMP-PNP is embedded in a pocket surrounded by Trp383, Asn380, Gly376 and Gly377. The purine ring of AMP-PNP is positioned in parallel to the indole ring of Trp383. In addition, it is hydrogen-bonded with the side chain amide of Asn380 (Fig 3B). The tail of AMP-PNP points to the hinge region of SePSK, and its α-phosphate and β-phosphate groups are stabilized by Gly376 and Ser243, respectively. Together, this structure clearly shows that the AMP-PNP-β-phosphate is sticking out of the ATP binding pocket, thus the γ-phosphate group is at the empty space between domain I and domain II and is unconstrained in its movement by the protein. (A) The electron density of AMP-PNP. The SePSK structure is shown in the electrostatic potential surface mode. The AMP-PNP is depicted as sticks with its ǀFoǀ-ǀFcǀ map contoured at 3 σ shown as cyan mesh. (B) The AMP-PNP binding pocket. The head of AMP-PNP is sandwiched by four residues (Leu293, Gly376, Gly377 and Trp383). The protein skeleton is shown as cartoon (cyan). The four α-helices (α26, α28, α27 and α30) are labeled in red. The AMP-PNP and coordinated residues are shown as sticks. The interactions between them are represented as black dashed lines. The numerical note near the black dashed line indicates the distance (Å). The results from our activity assays suggested that SePSK has D-ribulose kinase activity. To better understand the interaction pattern between SePSK and D-ribulose, the apo-SePSK crystals were soaked into the reservoir with 10 mM D-ribulose (RBL) and the RBL-SePSK structure was solved. As shown in S6 Fig, two residual electron densities are visible in domain I, which can be interpreted as two D-ribulose molecules with reasonable fit. As shown in Fig 4A, the nearest distance between the carbon skeleton of two D-ribulose molecules are approx. 7.1 Å (RBL1-C4 and RBL2-C1). RBL1 is located in the pocket consisting of α21 and the loop between β6 and β7. The O4 and O5 of RBL1 are coordinated with the side chain carboxyl group of Asp221. Furthermore, the O2 of RBL1 interacts with the main chain amide nitrogen of Ser72 (Fig 4B). This pocket is at a similar position of substrate binding site of other sugar kinase, such as L-ribulokinase (PDB code: 3QDK) (S7 Fig). However, structural comparison shows that the substrate ligating residues between the two structures are not strictly conserved. Based on the structures, the ligating residues of RBL1 in RBL-SePSK structure are Ser72, Asp221 and Ser222, and the interacting residues of L-ribulose with L-ribulokinase are Ala96, Lys208, Asp274 and Glu329 (S7 Fig). Glu329 in 3QDK has no counterpart in RBL-SePSK structure. In addition, although Lys208 of L-ribulokinase has the corresponding residue (Lys163) in RBL-SePSK structure, the hydrogen bond of Lys163 is broken because of the conformational change of two α-helices (α9 and α13) of SePSK. These differences might account for their different substrate specificity. (A) The electrostatic potential surface map of RBL-SePSK and a zoom-in view of RBL binding site. The RBL1 and RBL2 are depicted as sticks. (B) Interaction of two D-ribulose molecules (RBL1 and RBL2) with SePSK. The RBL molecules (carbon atoms colored yellow) and amino acid residues of SePSK (carbon atoms colored green) involved in RBL interaction are shown as sticks. The hydrogen bonds are indicated by the black dashed lines and the numbers near the dashed lines are the distances (Å). (C) The binding affinity assays of SePSK with D-ribulose. Single-cycle kinetic data are reflecting the interaction of SePSK and D8A-SePSK with D-ribulose. It shows two experimental sensorgrams after minus the empty sensorgrams. The original data is shown as black curve, and the fitted data is shown as different color (wild type SePSK: red curve, D8A-SePSK: green curve). Dissociation rate constant of wild type and D8A-SePSK are 3 ms and 9 ms, respectively. The binding pocket of RBL2 with relatively weak electron density is near the N-terminal region of SePSK and is negatively charged. The side chain of Asp8 interacts strongly with O3 and O4 of RBL2. The hydroxyl group of Ser12 coordinates with O2 of RBL2. The backbone amide nitrogens of Gly13 and Arg15 also keep hydrogen bonds with RBL2 (Fig 4B). Structural comparison of SePSK and AtXK-1 showed that while the RBL1 binding pocket is conserved, the RBL2 pocket is disrupted in AtXK-1 structure, despite the fact that the residues interacting with RBL2 are highly conserved between the two proteins. In the RBL-SePSK structure, a 2.6 Å hydrogen bond is present between RBL2 and Ser12 (Fig 4B), while in the AtXK-1 structure this hydrogen bond with the corresponding residue (Ser22) is broken. This break is probably induced by the conformational change of the two β-sheets (β1 and β2), with the result that the linking loop (loop 1) is located further away from the RBL2 binding site. This change might be the reason that AtXK-1 only shows limited increasing in its ATP hydrolysis ability upon adding D-ribulose as a substrate after comparing with SePSK (Fig 2C). Our SePSK structure shows that the Asp8 residue forms strong hydrogen bond with RBL2 (Fig 4B). In addition, our enzymatic assays indicated that Asp8 is important for the activity of SePSK (Fig 2D). To further verified this result, we measured the binding affinity for D-ribulose of both wild type (WT) and D8A mutant of SePSK using a surface plasmon resonance method . The results showed that the affinity of D8A-SePSK with D-ribulose is weaker than that of WT with a reduction of approx. two third (Fig 4C). Dissociation rate constant (Kd) of wild type and D8A-SePSK are 3 ms and 9 ms, respectively. The results implied that the second RBL binding site plays a role in the D-ribulose kinase function of SePSK. However, considering the high concentration of D-ribulose used for crystal soaking, as well as the relatively weak electron density of RBL2, it is also possible that the second binding site of D-ribulose in SePSK is an artifact. It was reported earlier that the crossing angle between the domain I and domain II in FGGY family carbohydrate kinases is different [2, 4, 8, 13, 14]. In addition, this difference may be caused by the binding of substrates and/or ATP. As reported previously, members of the sugar kinase family undergo a conformational change to narrow the crossing angle between two domains and reduce the distance between substrate and ATP in order to facilitate the catalytic reaction of phosphorylation of sugar substrates. After comparing structures of apo-SePSK, RBL-SePSK and AMP-PNP-SePSK, we noticed that these structures presented here are similar. Superposing the structures of RBL-SePSK and AMP-PNP-SePSK, the results show that the nearest distance between AMP-PNP γ-phosphate and RBL1/RBL2 is 7.5 Å (RBL1-O5)/6.7 Å (RBL2-O1) (S8 Fig). This distance is too long to transfer the γ-phosphate group from ATP to the substrate. Since the two domains of SePSK are widely separated in this structure, we hypothesize that our structures of SePSK represent its open form, and that a conformational rearrangement must occur to switch to the closed state in order to facilitate the catalytic process of phosphorylation of sugar substrates. For studying such potential conformational change, a simulation on the Hingeprot Server was performed to predict the movement of different SePSK domains . The results showed that domain I and domain II are closer to each other with Ala228 and Thr401 in A2 as Hinge-residues. Based on the above results, SePSK is divided into two rigid parts. The domain I of RBL-SePSK (aa. 1–228, aa. 402–421) and the domain II of AMP-PNP-SePSK (aa. 229–401) were superposed with structures, including apo-AtXK-1, apo-SePSK, xylulose kinase from Lactobacillus acidophilus (PDB code: 3LL3) and the S58W mutant form of glycerol kinase from Escherichia coli (PDB code: 1GLJ). The results of superposition displayed different crossing angle between these two domains. After superposition, the distances of AMP-PNP γ-phosphate and the fifth hydroxyl group of RBL1 are 7.9 Å (superposed with AtXK-1), 7.4 Å (superposed with SePSK), 6.6 Å (superposed with 3LL3) and 6.1 Å (superposed with 1GLJ). Meanwhile, the distances of AMP-PNP γ-phosphate and the first hydroxyl group of RBL2 are 7.2 Å (superposed with AtXK-1), 6.7 Å (superposed with SePSK), 3.7 Å (superposed with 3LL3), until AMP-PNP γ-phosphate fully contacts RBL2 after superposition with 1GLJ (Fig 5). This distance between RBL2 and AMP-PNP-γ-phosphate is close enough to facilitate phosphate transferring. Together, our superposition results provided snapshots of the conformational changes at different catalytic stages of SePSK and potentially revealed the closed form of SePSK. The structures are shown as cartoon and the ligands are shown as sticks. Domain I from D-ribulose-SePSK (green) and Domain II from AMP-PNP-SePSK (cyan) are superposed with apo-AtXK-1 (1), apo-SePSK (2), 3LL3 (3) and 1GLJ (4), respectively. The numbers near the black dashed lines show the distances (Å) between two nearest atoms of RBL and AMP-PNP. In summary, our structural and enzymatic analyses provide evidence that SePSK shows D-ribulose kinase activity, and exhibits the conserved features of FGGY family carbohydrate kinases. Three conserved residues in SePSK were identified to be essential for this function. Our results provide the detailed information about the interaction of SePSK with ATP and substrates. Moreover, structural superposition results enable us to visualize the conformational change of SePSK during the catalytic process. In conclusion, our results provide important information for a more detailed understanding of the mechanisms of SePSK and other members of FGGY family carbohydrate kinases. The gene encoding SePSK was amplified by polymerase chain reaction (PCR) with forward primer 5' CATGCCATGGGCATGGTCGTTGCACTTGGCCTCG 3' containing an internal Nco I restriction site (underlined) and reverse primer 5' CCGCTCGAGGGTTCTCTTTAACCCCGCCG 3' including an internal Xho I restriction site (underlined) from Synechococcus elongatus PCC 7942 genomic DNA. The amplified PCR product was digested with Nco I and Xho I (Takara) and ligated into linearized pET28-a vector (Novagen) between Nco I and Xho I restriction sites with a C-terminal his6 tag. The recombinant plasmids were transformed into competent Escherichia coli Trans10 cells for DNA production and purification, and the final constructs were verified by sequencing. The recombinant vectors were transformed into Escherichia coli BL21 (DE3) to express the protein. After induction with the 1 mM IPTG at 289 K in Luria-Bertani medium until the cell density reached an OD 600 nm of 0.6–0.8, the cells were harvested by centrifugation at 6,000 g at 4°C for 15 min, re-suspended in buffer A (20 mM Tris-HCl, pH 8.0, 500 mM NaCl, 5 mM imidazole) and disrupted by sonication. After centrifuge 40,000 g for 30 min, the protein was purified by passage through a Ni affinity column in buffer A, and then washed the unbound protein with buffer B (20 mM Tris-HCl, pH 8.0, 500 mM NaCl, 60 mM imidazole), and eluted the fraction with the buffer C (20 mM Tris-HCl, pH 8.0, 500 mM NaCl, 500 mM imidazole). After that, the protein was further purified by size exclusion chromatography with Superdex 200 10/300 GL (GE Healthcare) equilibrated with the buffer D (20 mM Tris-HCl, pH 8.0, 300 mM NaCl). The eluted major peak fraction was concentrated to 20 mg/mL protein using 10,000 MCWO centrifugal filter units (Millipore) and stored at -80°C for crystallization trials. The purified product was analyzed by SDS-PAGE with a single band visible only. The gene encoding AtXK-1 was amplified by PCR using a forward primer 5' TACTTCCAATCCAATGCTGTTATGAGTGGCAATAAAGGAACGA 3' and reverse primer 5' TTATCCACTTCCAATGTTACAAACCACTGTTCTGTTTTGCGCCC 3' from cDNA library of Arabidopsis thaliana. The underlined nucleotides were used for the ligation-independent cloning. The PCR product was treated by T4 DNA polymerase (LIC-qualified, Novagen) and then cloned into linearized pMCSG7 vector treated by T4 DNA polymerase (LIC-qualified, Novagen) with an N-terminal his6 tag though ligation-independent cloning method . The final construct was confirmed by DNA sequencing after amplified in competent Escherichia coli Trans10 cells. The recombinant vectors were transformed into Escherichia coli BL21 (DE3) for protein expression. After induction with 1 mM IPTG at 289 K in Luria-Bertani medium, cells were grown until the cell density reached an OD 600 nm of 0.6–0.8. Subsequent purification was identical to that used for SePSK, except that there was one additional step, during which tobacco etch virus protease was used to digest the crude AtXK-1 protein for removal of the N-terminal his6 tag following Ni affinity purification. Ni affinity column buffer contained extra 20% glycerol. The protein was further purified by size exclusion chromatography with Superdex 200 10/300 GL (GE Healthcare) in elution buffer consisting of 20 mM HEPES, pH 7.5, 100 mM NaCl. Finally, AtXK-1 protein was concentrated to 40 mg/mL protein using 10,000 MCWO centrifugal filter units (Millipore) and stored at -80°C prior to crystallization trials. Purity was verified by SDS-PAGE with a single band visible only. The gene of D8A and T11A mutations were amplified by PCR with the forward primer 5' CATGCCATGGGCATGGTCGTTGCACTTGGCCTCGCCTTCGGCAC 3' and forward primer 5' CATGCCATGGGCATGGTCGTTGCACTTGGCCTCGACTTCGGCGCCTCTGGAGCCC 3' (mismatched base pairs are underlined). The reverse primers of D8A and T11A mutants, the further constructions and purification procedures were identical with those used for wild type SePSK. The N-terminal sequence of D221A was amplified with forward primer (T7 promoter primer) 5' TAATACGACTCACTATA 3' and reverse primer 5' AGCAGCAATGCTAGCCGTTGTACCG 3’, and the C-terminal sequence of D221A was amplified with forward primer 5' TGCCGGTACAACGGCTAGCATTGCT 3' and reverse primer (T7 terminator primer) CGATCAATAACGAGTCGCC (mismatched base pairs are underlined). The second cycle PCR used the above PCR products as templates, and the construction and purification procedures were identical to those used for wild type SePSK. Crystallization trials of SePSK and AtXK-1 were carried out at 281 K by mixing equal volume of 20 mg/ml protein and reservoir solution with the sitting-drop vapor diffusion method. The reservoir solution was PEG Rx I-35 (0.1 M BIS-TRIS pH 6.5, 20% w/v Polyethylene glycol monomethyl ether 5,000) (Hampton research). After 2 or 3 days, the rod-like crystals could be observed. For phasing, the high-quality apo-SePSK crystals were soaked in mother liquor containing 1 mM ethylmercuricthiosalicylic acid, sodium salt (Hampton research, heavy atom kit) overnight at 281 K. In order to get the complexes with ATP and AMP-PNP, the crystals of apo-SePSK and apo-AtXK-1 were incubated with the reservoir including 10 mM ATP and 20 mM MgCl2 as well as 10 mM AMP-PNP and 20 mM MgCl2, respectively. The apo-SePSK crystals were incubated with the reservoir including 10 mM D-ribulose in order to obtain the complex D-ribulose-bound SePSK (RBL-SePSK). The crystals of three mutants (D8A, T11A and D221A) grew in the same condition as that of the wild type SePSK. The crystals were dipped into reservoir solution supplemented with 15% glycerol and then flash frozen in a nitrogen gas stream at 100 K. All data sets were collected at Shanghai Synchrotron Radiation Facility, Photo Factory in Japan and Institute of Biophysics, Chinese Academy of Sciences. Diffraction data were processed using the HKL2000 package . The initial phases of SePSK were obtained from the Hg-derivative crystals by single isomorphous replacement anomalous scattering (SIRAS) using AutoSol from the PHENIX suite . AutoBuild from the PHENIX suite was used to build 75% of the main chain of apo-SePSK, and the remaining residues were built manually by Coot . All other structures were solved by molecular replacement method using apo-SePSK as an initial model. The model was refined using phenix.refine and REFMAC5 . The final model was checked with PROCHECK . All structural figures were prepared by PyMOL . The summary of the data-collection and structure-refinement statistics is shown in Table 1 and S1 Table. Atomic coordinates and structure factors in this article have been deposited in the Protein Data Bank. The deposited codes of all structures listed in the Table 1 and S1 Table. The values in parentheses correspond to the highest resolution shell. Rmerge = ∑j∑h\|Ij,h-<Ih>\|/∑j∑h<Ih> where h are unique reflection indices and Ij,h are intensities of symmetry-related reflections and <Ih> is the mean intensity. R-work and R-free were calculated as follows: R = Σ (\|Fobs-Fcalc\|)/Σ \|Fobs\| ×100, where Fobs and Fcalc are the observed and calculated structure factor amplitudes, respectively. Root mean square deviations (r.m.s.d.) from standard values. ADP-Glo kinase assay was used according to the manufacturer’s instructions (Promega) . Each reaction mixture system consisted of 8 uM enzyme, 100 uM ATP, 1 mM MgCl2, 20 mM HEPES (pH 7.4), 5 mM substrate. The reaction was initiated by adding the purified enzyme into the reaction system. After incubation at 298 K for different time, equal volume ADP-Glo™ reagent was added to terminate the kinase reaction and to deplete any remaining ATP. Subsequently, kinase detection reagent with double volume of reaction system was added to convert ADP to ATP and allowed the newly synthesized ATP to be measured using a luciferase/luciferin reaction which produced luminescence signal and could be recorded. After incubation at room temperature for another 60 min, luminescence was detected by Varioskan Flash Multimode Reader (Thermo). The reference experiment was carried out in the same reaction system without the enzyme. For each assay, at least three repeats were performed for the calculation of mean values and standard deviations (SDs). The purity of five substrates in the activity assays was ≥98% (D-ribulose, Santa cruz), 99.7% (L-ribulose, Carbosynth), 99.3% (D-xylulose, Carbosynth), 99.5% (L-xylulose, Carbosynth) and 99.0% (Glycerol, AMRESCO). Surface plasmon resonance (SPR) was used to analyze the interaction of SePSK and D-ribulose. The SPR experiments were performed on a Biacore T100 system using series S CM5 sensor chips (GE Healthcare). All sensorgrams were recorded at 298 K. The proteins in buffer containing 20 mM HEPES, pH 7.5, 100 mM NaCl, was diluted to 40 ug/ml by 10 mM sodium acetate buffer at pH 4.5. All flow cells on a CM5 sensor chip were activated with a freshly prepared solution of 0.2 M 1-ethyl-3-(3-dimethylaminopropyl)-carbodiimide (EDC) and 0.05 M N-hydroxysuccinimide (NHS) in a ratio of 1:1 at a constant flow rate of 10 ul/min for 420 s. Deactivation of the surface was performed with an injection of a 1 M solution of ethanolamine-HCl (pH 8.5) using the same flow rate and duration. Kinetic parameters were derived from data sets acquired in single-cycle mode. Each run consisted of five consecutive analytic injections at 125, 250, 500, 1000 and 2000 uM. Analytic injections lasted for 60 s, separated by 30 s dissociation periods. Each cycle was completed with an extended dissociation period of 300 s. The specific binding to a blank flow cell was subtracted to obtain corrected sensorgrams. Biacore data were analyzed using BiaEvaluation software (GE Healthcare) by fitting to a 1:1 Langmuir binding fitting model . Coordinates and structure factors for all the structures in this article have been deposited in the Protein Data Bank. These accession codes are 5HTN, 5HTP, 5HUX, 5HV7, 5HTJ, 5HU2, 5HTY, 5HTR, 5HTV and 5HTX. The corresponding-structures are apo-SePSK, AMP-PNP-SePSK, ADP-SePSK, RBL-SePSK, D8A-SePSK, T11A-SePSK, D221A-SePSK, apo-AtXK1, AMP-PNP-AtXK1 and ADP-AtXK1, respectively.
PMC4887163	Hotspot autoimmune T cell receptor binding underlies pathogen and insulin peptide cross-reactivity	The cross-reactivity of T cells with pathogen- and self-derived peptides has been implicated as a pathway involved in the development of autoimmunity. However, the mechanisms that allow the clonal T cell antigen receptor (TCR) to functionally engage multiple peptide–major histocompatibility complexes (pMHC) are unclear. Here, we studied multiligand discrimination by a human, preproinsulin reactive, MHC class-I–restricted CD8 T cell clone (1E6) that can recognize over 1 million different peptides. We generated high-resolution structures of the 1E6 TCR bound to 7 altered peptide ligands, including a pathogen-derived peptide that was an order of magnitude more potent than the natural self-peptide. Evaluation of these structures demonstrated that binding was stabilized through a conserved lock-and-key–like minimal binding footprint that enables 1E6 TCR to tolerate vast numbers of substitutions outside of this so-called hotspot. Highly potent antigens of the 1E6 TCR engaged with a strong antipathogen-like binding affinity; this engagement was governed though an energetic switch from an enthalpically to entropically driven interaction compared with the natural autoimmune ligand. Together, these data highlight how T cell cross-reactivity with pathogen-derived antigens might break self-tolerance to induce autoimmune disease.T cells perform an essential role in adaptive immunity by interrogating the host proteome for anomalies, classically by recognizing peptides bound in major histocompatibility (MHC) molecules at the cell surface. Recent data (1–3) supports the notion that, to perform this role, the highly variable αβ T cell antigen receptor (TCR) must be able to recognize thousands, if not millions, of different peptide ligands (4, 5). This ability is required to enable the estimated 25 million distinct TCRs expressed in humans (6) to provide effective immune coverage against all possible foreign peptide antigens (5). Although essential to avoid blind spots during pathogen recognition, T cell cross-reactivity has also been implicated as a pathway to autoimmunity, possibly mediated by highly reactive pathogen-specific T cells weakly recognizing self-ligands (7–10). Several mechanisms, by which TCRs could bind to a large number of different peptide-MHC (pMHC), have been proposed (5). Structures of unligated and ligated TCRs have shown that the TCR complementarity determining region (CDR) loops can be flexible, perhaps enabling peptide binding using different loop conformations (11, 12). Both MHC and peptide have also been shown to undergo structural changes upon TCR binding, mediating an induced fit between the TCR and pMHC (13–16). Other studies, mainly in the murine system, have demonstrated that the same TCR can interact with different pMHCs using a common (1, 11, 14, 17–20) or divergent modality (21). Recent studies in model murine systems demonstrate that TCR cross-reactivity can be governed by recognition of a conserved region in the peptide that allows tolerance of peptide sequence variation outside of this hotspot (1, 22). We recently reported that the 1E6 human CD8 T cell clone — which mediates the destruction of β cells through the recognition of a major, HLA-A0201–restricted, preproinsulin signal peptide (ALWGPDPAAA15–24) (23–25) — can recognize upwards of 1 million different peptides (3). CD8 T cells that recognize HLA-A0201–ALWGPDPAAA have been shown to populate insulitic lesions in patients with type 1 diabetes (T1D) (26). We demonstrated that the TCR from the 1E6 T cell clone bound to HLA-A0201–ALWGPDPAAA using a limited footprint and very weak binding affinity (23). This first experimental evidence of a high level of CD8 T cell cross-reactivity in a human autoimmune disease system hinted toward molecular mimicry by a more potent pathogenic peptide as a potential mechanism leading to β cell destruction (8, 24). Here, we solved the structure of the 1E6 TCR with 7 altered peptide ligands (APLs) determined by our previously published combinatorial peptide library (CPL) screening (3), 2 of which mapped within human pathogens. These APLs differed from the natural preproinsulin peptide by up to 7 of 10 residues. We also solved the structure of each unligated APL to investigate whether structural changes occurred before or after binding — which, combined with an in-depth cellular and biophysical analysis of the 1E6 interaction with each APL, demonstrated the molecular mechanism mediating the high level of cross-reactivity exhibited by this preproinsulin-reactive human CD8 T cell clone. We have previously demonstrated that the 1E6 T cell clone can recognize over 1 million different peptides with a potency comparable with, or better than, the cognate preproinsulin peptide ALWGPDPAAA (3). From this large functional scan, we selected 7 different APLs that activated the 1E6 T cell clone across a wide (4-log) functional range (Table 1). Two of these peptides, MVWGPDPLYV and RQFGPDWIVA (bold text signifies amino acids that are different from the index preproinsulin–derived sequence), are contained within the proteomes of the human pathogens Bacteroides fragilis/thetaiotaomicron and Clostridium asparagiforme, respectively. Competitive functional testing revealed that the preproinsulin-derived sequence ALWGPDPAAA was one of the least potent targets for 1E6, with only the MVWGPDPLYV and YLGGPDFPTI demonstrating a similar low-activity profile in MIP-1β secretion and target killing assays (Figure 1, A and B). The RQFGPDWIVA sequence (present in C. asparagiforme) activated the 1E6 T cell with around 1 log–greater potency compared with ALWGPDPAAA. At the other end of the spectrum, the RQFGPDFPTI peptide stimulated MIP-1β release and killing by 1E6 at an exogenous peptide concentration 2–3 logs lower compared with ALWGPDPAAA. The pattern of peptide potency was closely mirrored by pMHC tetramer staining experiments (Figure 1C and plots shown in Supplemental Figure 1; supplemental material available online with this article; doi:10.1172/JCI85679DS1). Here, the A2-RQFGPDFPTI tetramer stained 1E6 with the greatest MFI, gradually decreasing to the weakest tetramers: A2-MVWGPDPLYV and -YLGGPDFPTI. To parallel the functional analysis, we also performed thermal melt (Tm) experiments using synchrotron radiation circular dichroism (SRCD) to investigate the stability of each APL (Figure 1D). The range of Tm was between 49.4°C (RQFGPDWIVA) and 60.3°C (YQFGPDFPIA), with an average approximately 55°C, similar to our previous findings (27). This pattern of stability did not correlate with the T cell activation or tetramer staining experiments, indicating that peptide binding to the MHC do not explain ligand potency. We, and others, have previously demonstrated that antipathogenic TCRs tend to bind with stronger affinity compared with self-reactive TCRs (28), likely a consequence of the deletion of T cells with high-affinity self-reactive TCR during thymic selection. In accordance with this trend, the 1E6 TCR bound the natural preproinsulin peptide, ALWGPDPAAA, with the weakest affinity currently published for a human CD8 T cell–derived TCR with a biologically relevant ligand (KD > 200 μM; KD, equilibrium binding constant) (23). Surface plasmon resonance (SPR) analysis of the 1E6 TCR–pMHC interaction for all 7 APLs (Figure 2, A–H) demonstrated that stronger binding affinity (represented as ΔG°, kcal/mol) correlated well with the EC50 values (peptide concentration required to reach half-maximal 1E6 T cell killing) for each ligand, demonstrated by a Pearson’s correlation analysis value of 0.8 (P = 0.01) (Figure 2I). It should be noted that this correlation, although consistent with the T cell killing experiments, uses only approximate affinities calculated for the 2 weakest ligands. These experiments revealed 4 important findings. First, the 1E6 T cell could still functionally respond to peptide when the TCR binding affinity was extremely weak, e.g., the 1E6 TCR binding affinity for the A2-MVWGPDPLYV peptide was KD = ~600 μM. Second, the 1E6 TCR bound to A2-RQFGPDFPTI with KD = 0.5 μM, equivalent to the binding affinity of the very strongest antipathogen TCRs (29). Third, the 1E6 TCR bound to A2-RQFGPDWIVA peptide, within the C. asparagiforme proteome, with approximately 4-fold stronger affinity than A2-ALWGPDPAAA, demonstrating the potential for a pathogen-derived antigen to initiate a response to the self-derived sequence. Finally, these data demonstrate the largest range of binding affinities reported for a natural, endogenous human TCR of more than 3 logs of magnitude (A2-MVWGPDPLYV vs. A2-RQFGPDFPTI). To confirm the affinity spread detected by SPR, and to evaluate whether experiments performed using soluble molecules were biologically relevant to events at the T cell surface, we determined the effective 2D affinity of each APL using an adhesion frequency assay in which a human rbc coated in pMHC acted as an adhesion sensor (30, 31). In agreement with SPR experiments, the range of 2D affinities we detected differed by around 3 logs, with the A2-MVWGPDPLYV generating the weakest 2D affinity (2.6 × 10 AcKa μm) and A2-RQFGPDFPTI the strongest (4.5 × 10 AcKa μm) (Figure 2J). As with the 3D affinity measurements, the 2D affinity measurements correlated well with the EC50 values for each ligand (Figure 2K) demonstrating a strong correlation (Pearson’s correlation = 0.8, P = 0.01) between T cell antigen sensitivity and TCR binding affinity. Of note, these data demonstrate a close agreement between the 3D affinity values generated using SPR and 2D affinity values generated using adhesion frequency assays. Our previous structure of the 1E6-A2-ALWGPDPAAA complex demonstrated a limited binding footprint between the TCR and pMHC (23). The low number of contacts between the 2 molecules most likely contributed to the weak binding affinity of the interaction. In order to examine the mechanism by which the 1E6 TCR engaged a wide range of peptides with divergent binding affinities, we solved the structure of the 1E6 TCR in complex with all 7 APLs used in Figure 2. All structures were solved in space group P1 to 2–3 Å resolution with crystallographic Rwork/Rfree ratios within accepted limits as shown in the theoretically expected distribution (ref. 32 and Supplemental Table 1). The 1E6 TCR used a very similar overall binding modality to engage all of the APLs, with root mean square deviation ranging between 0.81 and 1.12 Å (compared with 1E6-A2-ALWGPDPAAA). The relatively broad range of buried surface areas (1,670–1,920 Å) did not correlate well with TCR binding affinity (Pearson’s correlation = 0.45, P = 0.2). The surface complementarity values (0.52–0.7) correlated slightly with affinity (Pearson’s correlation = 0.7, P = 0.05) but could not explain all differences in binding (Figure 3A and Table 2). The TCR CDR loops were in a very similar position in all complexes, apart from some slight deviations in the TCR β-chain (Figure 3B); the peptides were all presented in a similar conformation (Figure 3C); and there was minimal variation in crossing angles of the TCR (42.3°–45.6°) (Figure 3D). Overall, the 1E6 TCR used a canonical binding mode to engage each APL with the TCR α-chain positioned over the MHC class I (MHCI) α2-helix and the TCR β-chain over the MHCI α-1 helix, straddling the peptide cargo (29). However, subtle differences in the respective interfaces were apparent (discussed below) and resulted in altered binding affinities of the respective complexes. We next performed an in-depth atomic analysis of the contacts between the 1E6 TCR and each APL to determine the structural basis for the altered T cell peptide sensitivities and TCR binding affinities (Table 2). Concomitant with our global analysis of 1E6 TCR binding to the APLs, we observed a common interaction element, consistent with our previous findings (23), that utilized TCR residues Tyr97α and Trp97β, forming an aromatic cap over a central GPD motif that was present in all of the APLs (Figure 4). Interactions between these 2 TCR and 3 peptide residues accounted for 41%–50% of the total contacts across all complexes (Table 2), demonstrating the conserved peptide centric binding mode utilized by the 1E6 TCR. This fixed anchoring between the 2 molecules was important for stabilization of the TCR-pMHC complex, as — although other peptides without the ‘GDP’ motif were tested and shown to activate the 1E6 T cell clone (3) — we were unable to measure robust affinities using SPR (data not shown). These data support the requirement for a conserved interaction between the 1E6 TCR and the GPD motif, as we observed in our previously published 1E6-A2-ALWGPDPAAA structure (23). Although the 1E6 TCR formed a similar overall interaction with each APL, the stabilization between the TCR and the GPD motif enabled fine differences in the contact network with both the peptide and MHC surface that allowed discrimination between each ligand (Figure 5). For example, the 1E6 TCR made only 47 peptide contacts with A2-MVWGPDPLYV (KD = ~600 μM) compared with 63 and 57 contacts with A2-YQFGPDFPIA (KD = 7.4 μM) and A2-RQFGPDFPTI (KD = 0.5 μM), respectively. Although the number of peptide contacts was a good predictor of TCR binding affinity for some of the APLs, for others, the correlation was poor (Pearson’s correlation = 0.045, P = 0.92), possibly because of different resolutions for each complex structure. For example, the 1E6 TCR made 64 peptide contacts with A2-YLGGPDFPTI (KD = ~400 μM) compared with 43 contacts with A2-RQWGPDPAAV (KD = 7.8 μM). The most important peptide modification in terms of generating new contacts was peptide position 1. The stronger ligands all encoded larger side chains (Arg or Tyr) at peptide position 1 (Figure 5, E–H), enabling interactions with 1E6 that were not present in the weaker APLs that lacked large side chains in this position (Figure 5, A, C, and D). We have previously shown that the 1E6 TCR uses a rigid lock-and-key mechanism during binding to A2-ALWGPDPAAA (23). These data demonstrated that the unligated structure of the 1E6 TCR was virtually identical to its ligated counterparts. In order to determine whether any of the APLs required an induced fit mechanism during binding that could explain the difference in free binding energy (ΔG) between each complex (Table 2), we solved the unligated structures of all 7 APLs (the A2-ALWGPDPAAA structure has been previously published and was used in this comparison, ref. 23) (Figure 6 and Supplemental Table 2). The unligated A2-MVWGPDPLYV (KD = ~600 μM) structure revealed that the side chain Tyr9 swung around 8 Å in the complex structure, subsequently making contacts with TCR residues Asp30β and Asn51β (Figure 6A and Figure 5A, respectively). This movement could result in an entropic penalty contributing to the weak TCR binding affinity we observed for this ligand. Additional small movements in the Cα backbone of the peptide around peptide residue Asp6 were apparent in the A2-YLGGPDFPTI (KD = ~400 μM), A2-ALWGPDPAAA (KD = ~208 μM), and A2-RQFGPDWIVA (KD = 44.4 μM) structures (Figure 6, B, C, and E). The unligated structures of A2-AQWGPDAAA, A2-RQWGPDPAAV, A2-YQFGPDFPIA, and A2-RQFGPDFPTI were virtually identical when in complex with 1E6 (Figure 6, D and F–H). Apart from the case of A2-AQWGPDAAA (KD = 61.9 μM), these observations support the conclusion that the higher-affinity ligands required less conformational melding during binding, which could be energetically beneficial (lower entopic cost) during ligation with the 1E6 TCR. In addition to changes between the TCR and peptide component, we also observed that different APLs had different knock-on effects between the TCR and MHC. MHC residue Arg65 that forms part of the MHC restriction triad (Arg65, Ala69, and Gln155) (15, 33) played a central role in TCR-MHC contacts, with Gln155 playing a less important role and Ala69 playing no role in binding at the interface (Figure 7). Generally, the weaker-affinity APLs made fewer contacts with the MHC surface (27–29 interactions) compared with the stronger-affinity APLs (29–35 contacts), consistent with a better Pearson’s correlation value (0.55) compared with TCR-peptide interactions versus affinity (0.045). For instance, contacts were made between TCR residue Val53β and MHC residue Gln72 in all APLs except for in the weakest affinity ligand pair, 1E6-A2-MVWGPDPLYV, in which a subtle change in TCR conformation — probably mediated by different peptide contacts — abrogated this interaction (Figure 7A). Our analysis of the contact network provided some clues that could explain the different antigen potencies and binding affinities between the 1E6 TCR and the different APLs. However, there were clear outliers in which the number of contacts did not match with the strength/potency of the interaction. For example, the 1E6 TCR bound to A2-RQWGPDPAAV with the third strongest affinity (KD = 7.8 μM) but made fewer contacts than with A2-ALWGPDPAAA (KD = ~208 μM) (Table 2). However, it is not necessarily the quantity of contacts that determines the strength of an interaction, but the quality of the contacts. Thus, we performed an in-depth thermodynamic analysis of 6 of the ligands under investigation (Figure 8 and Supplemental Table 3). The weak binding affinity between 1E6 and A2-MVWGPDPLYV and A2-YLGGPDFPTI generated thermodynamic data that were not robust enough to gain insight into the enthalpic (ΔH°) and entropic (TΔS°) changes that contributed to the different binding affinities/potencies for each APL. The overall free binding energies (ΔG°) were between –4.4 and –8.6 kcal/mol, reflecting the wide range of TCR binding affinities we observed for the different APLs. The enthalpic contribution in each complex did not follow a clear trend with affinity, with all but the 1E6-A2-RQFGPDFPTI interaction (ΔH° = 6.3 kcal/mol) generating an energetically favorable enthalpy value (ΔH° = –3.7 to –11.4 kcal/mol); this indicated a net gain in electrostatic interactions during complex formation. However, there was a clear switch in entropy between the weaker-affinity and stronger-affinity ligands, indicated by a strong Pearson’s correlation value between entropy and affinity (Pearson’s correlation value 0.93, P =0.007). For instance, the A2-ALWGPDPAAA, A2-AQWGPDAAA, and A2-RQFGPDWIVA (KD = ~208 μM, KD = 61.9 μM, and KD = 44.4 μM, respectively) were all entropically unfavorable (TΔS° = –2.9 to –5.6 kcal/mol), indicating a net change from disorder to order. Conversely, the stronger-affinity ligands A2-RQWGPDPAAV (KD = 7.8 μM), A2-YQFGPDFPIA (KD = 7.4 μM), and A2-RQFGPDFPTI (KD = 0.5 μM) exhibited favorable entropy (TΔS° = 2.2 to 14.9 kcal/mol), indicating an order-to-disorder change during binding, possibly through the expulsion of ordered water molecules. Furthermore, the structures of the unligated pMHCs demonstrated that, for these stronger-affinity ligands, there was less conformational difference between the TCR ligated pMHCs compared with the weaker-affinity ligands (Figure 6). The potential requirement for a larger degree of induced fit during binding to these weaker-affinity ligands is consistent with the larger entropic penalties observed for these interactions. We searched a database of over 1,924,572 unique decamer peptides from the proteome of viral pathogens that are known, or strongly suspected, to infect humans. Three hundred forty-two of these decamers conformed to the motif xxxGPDxxxx. Of these, 53 peptides contained the motif xOxGPDxxxO, where O is one of the hydrophobic amino acid residues A,V, I, L, M, Y, F, and W that might allow binding to HLA-A0201 (Supplemental Table 4). Thus, there are many pathogen-encoded peptides that could act as agonists for the 1E6 T cell beyond the MVWGPDPLYV and RQFGPDWIVA sequences studied here. Extension of these analyses to include the larger genomes of bacterial pathogens would be expected to considerably increase these numbers. The binding affinity of the 1E6 TCR interaction with A2-RQFGPDWIVA is considerably higher than with the disease-implicated A2-ALWGPDPAAA sequence (KD = 44.4 μM and KD > 200 μM, respectively), highlighting how a pathogen-derived sequence might be capable of priming a 1E6-like T cell. T cell antigen discrimination is governed by an interaction between the clonally expressed TCR and pMHC, mediated by the chemical characteristics of the interacting molecules. It has recently become clear that TCR cross-reactivity with large numbers of different pMHC ligands is essential to plug holes in T cell immune coverage that pathogens could exploit (5). Flexibility at the interface between the TCR and pMHC, demonstrated in various studies (29), has been suggested as a mechanism mediating T cell cross-reactivity with multiple distinct epitopes. This notion is attractive because the CDR loops, which form the TCR antigen-binding site, are usually the most flexible part of the TCR and have the ability to mold around differently shaped ligands. Focused binding around a minimal peptide motif has also been implicated as an alternative mechanism enabling TCR cross-reactivity (1, 11–14, 16, 22, 34). Notably among these studies, Garcia and colleagues recently used the alloreactive murine TCR-MHC pair of the 42F3 TCR and H2-L to demonstrate recognition of a large number of different peptides via conserved hotspot contacts with prominent up-facing peptide residues (22). Sethi and colleagues recently demonstrated that the MHCII-restricted Hy.1B11 TCR, which was isolated from a patient with multiple sclerosis, could anchor into a deep pocket formed from peptide residues 2, 3, and 5 (from MBP85–99 bound to HLA-DQ1) (19). This motif was conserved in at least 2 potential foreign peptides, originating from Herpes simplex virus and Pseudomonas aeruginosa, enabling TCR recognition of foreign epitopes (19). Although these data provided some clues into the molecular mechanism of T cell recognition, there still remain several gaps in our understanding. First, we currently know nothing about how human MHCI–restricted TCRs mediate cross-reactivity in the context of a clinically relevant model of autoimmunity, thought to be a major pathway of disease initiation in several autoimmune diseases (35). Second, molecular studies have not yet revealed a broad set of rules that determine TCR cross-reactivity because, with the exception of the allo–TCR-MHC pair of the 42F3 TCR and H2-L that did not encounter each other during T cell development (22), studies have been limited to structures of a TCR with only 2 or 3 different ligands (10, 14, 21, 36–40). Finally, no studies have included characterization of pathogen-derived ligands recognized by self-reactive T cells with greater potency than the autoantigen, a potentially important facet to break self-tolerance. Here, we investigated a highly cross-reactive MHCI-restricted TCR isolated from a patient with T1D that recognizes an HLA-A0201–restricted preproinsulin signal peptide (ALWGPDPAAA15–24) (3, 23, 25). Human CD8 T cell clones expressing TCRs with this specificity mediate the destruction of β cells, have been found in islets early in infection, and are proposed to be a major driver of disease (8, 26). We solved the structure of the 1E6 TCR with 7 APLs to enable a comprehensive analysis of the molecular basis of TCR degeneracy. The epitopes we selected exhibited a broad range of potencies and could activate the 1E6 T cell clone at exogenously supplied concentrations more than 4 logs apart. Overall, the difference in antigen potency correlated well with the binding energy (ΔG° kcal/mol) of the 1E6 TCR for the different epitopes, which ranged from values of ΔG° = ~–4.4 to –8.6 kcal/mol (calculated from 3D affinity data) or 2D affinity values of AcKa = 2.5 × 10 to 4.4 × 10 μm. The weaker end of this spectrum extends our understanding of the limits in which T cells can functionally operate in terms of TCR 3D binding affinity and is in line with the types of very low affinity, yet fully functional self-reactive CD8 T cells we have observed in tumor-infiltrating lymphocytes (41–43). Previous studies of autoreactive TCRs have shown that their binding mode is generally atypical, either due to an unusual binding manner (19, 44–47), weak TCR binding affinity (23, 36), an unstable pMHC (48, 49), or a combination of these factors. Our data demonstrate the potential for an autoreactive TCR to bind with a conventional binding mode to a stable pMHC with antipathogen-like affinity (KD = 0.5 μM) depending on the peptide sequence. Our structural analysis revealed that the 1E6 TCR bound with a conserved conformation across all APLs investigated. This binding orientation was mediated through a focused interaction with TCR residues Tyr97α and Trp97β that formed an aromatic cap over a central ‘GDP’ motif that was common to all APLs. We have previously demonstrated the importance of the GPD motif using a peptide library scan (23), as well as a CPL scan approach (3). Although the 1E6 T cell was able to activate weakly with peptides that lacked this motif, we were unable to robustly measure binding affinities or generate complex structures with these ligands, highlighting the central role of this interaction during 1E6 T cell antigen recognition. This hotspot binding, defined as a localized cluster of interactions that dominate binding energy during protein-protein interactions (50), has been previously shown to contribute to TCR recognition of MHC as a mechanism that tunes T cell cross-reactivity by providing fixed anchor points that enable TCRs to tolerate a variable peptide cargo (1, 51–53). Alternatively, interactions between the TCR and peptide have been shown to dominate the energetic landscape during ligand engagement, ensuring that T cells retain peptide specificity (54, 55). The binding mechanism utilized by the 1E6 TCR during pMHC recognition is consistent with both of these models. Ligand engagement is dominated by peptide interactions, but hotspot-like interactions with the central GPD motif enable the 1E6 TCR to tolerate peptide residues that vary outside of this region, explaining how T cells expressing this TCR may cross-react with a large number of different peptides (3). These findings are also analogous to the observed binding mode of the Hy.1B11 TCR, in which one aromatic residue of the TCR CDR3α loop anchored into a pocket created by a conserved peptide motif (19). In both of these examples, self-recognition is mediated by TCR residues with aromatic side chains. These large, generally hydrophobic amino acids can form strong interactions with other residues through π-π stacking. Combined with evidence demonstrating that aromatic side chains are conserved in the CDR2 loops of TCRs from many species (56), we speculate that these aromatic residues could impart a level of “stickiness” to TCRs, which might be enriched in an autoimmune setting when the TCR often binds in a nonoptimal fashion. Despite some weak statistical correlation between the surface complementarity (SC) and affinity, closer inspection of the interface revealed no obvious structural signature that could definitively explain the differences in antigen potency and TCR binding strength between the different ligands. However, similar to our findings in other systems (57–59), modifications to residues outside of the canonical central peptide bulge were important for generating new interactions. For example, all of the stronger ligands encoded larger side chains (Arg or Tyr) at peptide position 1 that enabled new interactions with 1E6 not present with the Ala at this position in the natural preproinsulin peptide. These data also explain our previous findings that alteration of the anchor residue at peptide position 2 (Leu-Gln) has a direct effect on 1E6 TCR binding affinity (60) because our structural analysis demonstrated that 1E6 made 3 additional bonds with A2-AQWGPDPAAA compared with A2-ALWGPDPAAA, consistent with the >3-fold stronger binding affinity. We have recently demonstrated how a suboptimal position 2 anchor in a melanoma-derived antigen can improve TCR binding through a similar mechanism (58). These results challenge the notion that the most potent peptide antigens exhibit the greatest pMHC stability and have implications for the design of anchor residue–modified heteroclitic peptides for vaccination. Early thermodynamic analysis of TCR-pMHC interactions suggested a common energetic signature, driven by favorable enthalpy (generally mediated through an increase in electrostatic interactions) and unfavorable entropy (changes from disorder to order) (61, 62). These parameters aligned well with structural data, demonstrating that TCRs engaged pMHC using an induced fit binding mode (63). However, more recent data have shown that TCRs can utilize a range of energetic strategies during pMHC binding, currently with no obvious pattern in terms of TCR affinity, binding mechanism, or specificity (pathogen, cancer, or self-ligands) (12, 57, 64, 65). Although no energetic signature appears to exist for different TCRs, we used thermodynamic analysis here to explore whether changes in energetics could help explain ligand discrimination by a single TCR. This analysis demonstrated a strong relationship (according to the Pearson’s correlation analysis) between the energetic signature used by the 1E6 TCR and the sensitivity of the 1E6 T cell clone to different APLs. The weaker APL ligands were characterized by favorable enthalpy and unfavorable entropy, whereas the stronger ligands progressively shifted to favorable entropy. These differences were consistent with a greater degree of movement between the unligated and ligated pMHCs for the weaker ligands, suggesting a greater requirement for disorder-to-order changes during TCR binding (54, 66, 67). Thus, the enhanced antigen potency was probably mediated through a shift from an induced fit to a lock-and-key interaction between the stronger ligands (less requirement for energetically unfavorable disorder-to-order changes), resulting in a more energetically favorable ΔG value. Importantly, the preproinsulin-derived epitope was one of the least potent peptides, demonstrating that the 1E6 T cell clone had the ability to respond to different peptide sequences with far greater potency. The RQFGPDWIVA peptide, which was substantially more potent than the preproinsulin peptide, is within the proteome of a common human pathogen (C. asparagiforme), demonstrating the potential for an encounter between a naive 1E6-like T cell and a foreign peptide with a more potent ligand that might then break self-tolerance. Indeed, we found over 50 decamer peptides from the proteome of likely, or known, human viral pathogens alone that contained both the conserved central GPD motif and anchor residues at positions 2 and 10 that would enable binding to HLA-A02:01. Further experiments will be required to determine whether any naturally presented, human pathogen–derived peptides act as active ligands for 1E6, but our work presented here demonstrates that it is at least feasible for an autoimmune TCR to bind to a different peptide sequence that could be present in a pathogen proteome with substantially higher affinity and potency than the interaction it might use to attack self-tissue. In summary, this investigation into the molecular basis of T cell cross-reactivity using a clinically relevant cytotoxic CD8 T cell clone that kills human pancreatic β cells (24, 25) provides answers to a number of previously outstanding questions. First, our data shows that a single TCR has the potential to functionally (assessed through T cell activation) bind to different ligands with affinities ranging across 3 orders of magnitude. Second, this is the first example in which ligands have been identified and characterized for a human autoreactive TCR that are substantially more potent than the natural self-ligand, demonstrating the potential for a pathogenic ligand to break self-tolerance and prime self-reactive T cells. Third, this first structural analysis of a cross-reactive human MHCI–restricted autoimmune TCR showed that degeneracy was mediated through TCR-pMHC anchoring by a conserved minimal binding peptide motif. Finally, TCR ligand discrimination was characterized by an energetic shift from an enthalpically to entropically driven interaction. Our demonstration of the molecular mechanism governing cross-reactivity by this preproinsulin reactive human CD8 T cell clone supports the notion first put forward by Wucherpfennig and Strominger that molecular mimicry could mediate autoimmunity (7–9) and has far-reaching implications for the complex nature of T cell antigen discrimination. The 1E6 T cell clone was generated as previously described (25) and stored in vapor phase liquid nitrogen in freezing buffer (90% FCS and 10% DMSO). Cells were defrosted rapidly in a 37°C water bath until a small amount of frozen cells were left and then immediately washed in 15–20 ml of R10 media (RPMI 1640 with 10% FCS, 100 IU/ml penicillin, 100 μg/ml streptomycin, 2 mM l-glutamine) by centrifuging at 300 g for 5 minutes. Defrosted cells were cultured in T cell media: R10 with 1× nonessential amino acids, 1 mM sodium pyruvate, 10 mM HEPES buffer (all from Invitrogen), 20 IU/ml of IL-2 (aldesleukin, brand name Proleukin, Prometheus) and 25 ng/ml IL-15 (PeproTech), for 24 hours; then, 0.75 × 10 to 1.5 × 10 cells expanded by coculture with 15 × 10 irradiated (3,100 cGy) PBMCs from 3 donors in a 25 cm tissue culture flask with 1 μg/ml of phytoheamagglutinin (Alere Inc.) and T cell media as above. The clone was transferred to 24-well tissue culture plates (3 × 10 to 4 × 10 per well in 2 ml) at day 7, and the IL-2 increased to 200 IU/ml. For the purpose of this study, the clone was passaged 3 times and used between weeks 2 and 4 after expansion. The [Cr] release cytotoxicity assay was performed as previously described (43). Target A2 CIR cells were labeled for 1 hour at 37°C with 30 μCi chromium (sodium chromate in normal saline, PerkinElmer) per 1 × 10 cells, washed with R10, and allowed to leach for a further hour at 37°C in R10 to remove any excess chromium from the cells. After chromium labeling, target cells were washed and plated at 1,000 cells/well in 96-well tissue culture plates and pulsed with peptide at the indicated concentrations for 2 hours at 37°C. T cells were added to give the desired T cell/target cell (5:1) ratio and a final volume of 150 μl R10. Target cells were also incubated alone or with 1% Triton X-100 detergent (Sigma-Aldrich) to give the spontaneous and total chromium released from the target cells, respectively. After overnight incubation, at 37°C and 5% CO2, the supernatants were both (i) assayed for MIP1β by ELISA (R&D Systems) and (ii) harvested (10% of total volume), mixed with 150 μl Optipahse supermix scintillation mixture (PerkinElmer) in 96-well polyethylene terephthalate plates (PerkinElmer), and sealed; the amount of released chromium was measured indirectly on a 1450 Microbeta counter (PerkinElmer). The percentage of specific target cell lysis by T cells was calculated according to the following formula: (experimental release [with T cells and target cells] − spontaneous release from target cells)/(total release from target cells − spontaneous release from target cells) × 100. Experiments were independently completed in triplicate. Tetrameric pMHCI reagents (tetramers) were constructed by the addition of PE-conjugated streptavidin (Invitrogen) at a pMHCI/streptavidin molar ratio of 4:1. A total of 50,000 T cells were stained with PE-conjugated tetramer (25 μg/ml) folded around the indicated peptides for 30 minutes on ice and washed with PBS before staining with 2 μl (1:40 dilution of the DMSO stock in PBS) of the violet LIVE/DEAD fixable dead cell stain Vivid (Invitrogen) for 5 minutes at room temperature before direct addition of 2 μl of anti–CD8-APC antibody (clone BW135/80, Miltenyi Biotec) and incubated for a further 20 minutes on ice before being washed in FACS buffer (2% FCS in PBS). Data were acquired using a FACSCanto II flow cytometer (BD Biosciences) and analyzed with FlowJo software (Tree Star Inc.). The 1E6 TCR, HLA-A0201, and human β2m chain were generated as described previously (23). The 1E6 TCR and HLA-A0201 peptide variants were refolded and purified as described previously (23). Biotinylated pMHCI and pMHC tetramer were prepared as previously described (68). Thermal stability of the HLA-A0201–peptide complexes was assessed by circular dichroism (CD) spectroscopy monitoring the change in ellipticities at 218 nm. Data were collected, in duplicate, using a nitrogen-flushed Module B end-station spectrophotometer at the B23 Synchrotron Radiation CD Beamline at the Diamond Light Source (DLS) (69). Samples were prepared in phosphate buffered saline, pH 7.4, and concentrated to ~10 mM. Spectra were measured every 5°C over a temperature range between 5°C and 90°C with 5 minutes of equilibration time for each temperature. Four scans were acquired using an integration time of 1 second, a path length of 0.02 cm, and a slit width of 0.5 mm equivalent to a 1.2-nm bandwidth. Reversibility was monitored by measuring the spectrum at 20°C after cooling from 90°C with 30 minutes of incubation. Melting curves were analyzed assuming a 2-state trimer-to-monomer transition from the native (N) to unfolded (U) conformation N3 ↔ 3U with an equilibrium constant K = (U)/(N3) = F/(3c[1-F]), where F and c are the degree of folding and protein concentration, respectively. Data were fitted as described (70). Fitted parameters were the melting temperature Tm, van’t Hoff’s enthalpy ΔHvH, and the slope and intercept of the native baseline. As all protein complexes aggregated to various degrees upon unfolding, the ellipticity of the unfolded state was set as a constant of –4,500 deg cm/dmol (71, 72). Binding analysis was performed using a BIAcore T200 equipped with a CM5 sensor chip as previously described (73). Binding analysis was performed 3× in independent experiments using pMHC monomers generated in-house. Approximately 200–500 RU of each HLA-A0201–peptide complex was attached to the CM5 sensor chip at a slow flow rate of 10 μl/min to ensure uniform distribution on the chip surface. The 1E6 TCR was purified and concentrated to approximately 40–350 μM on the same day of SPR analysis to reduce the likelihood of TCR aggregation. For equilibrium analysis, 10 serial dilutions were prepared in triplicate for each sample and injected over the relevant sensor chips at 25°C. TCR was injected over the chip surface using kinetic injections at a flow rate of 45 μl/min using HLA-A0201–ELAGIGILTV as a negative control surface on flow cell 1. For the thermodynamics experiments, this method was repeated at the following temperatures: 5°C, 13°C, 18°C, 25°C, 30°C, and 37°C. Results were analyzed using BIAevaluation 3.1, Excel, and Origin 6.0 software. The KD values were calculated assuming a 1:1 interaction by plotting specific equilibrium-binding responses against protein concentrations, followed by nonlinear least squares fitting of the Langmuir binding equation. The thermodynamic parameters were calculated using the nonlinear van’t Hoff equation (RT ln KD = ΔH° –TΔS° + ΔCp°[T-T0] – TΔCp° ln [T/T0]) with T0=298 K. We used an adhesion frequency assay to measure the 2D affinity of TCR-pMHC interactions at the cell membrane as previously described (30). Briefly, human 1E6 T cells were mounted onto 1 micropipette, and, on the other pipette, human rbcs coated with pMHC by biotin-streptavidin coupling served as both a surrogate APC and an adhesion sensor for detecting the TCR-pMHC interaction. Site densities of TCR and pMHC were measured by flow cytometry as previously described (74). All assays were performed using at least 5 cell pairs and calculated as an average of 100 cell-cell contacts. All protein crystals were grown at 18°C by vapor diffusion via the sitting drop technique. Each pMHCI (200 nl, 10 mg/ml) in crystallization buffer (10 mM TRIS [pH 8.1] and 10 mM NaCl) was added to 200 nl of reservoir solution. HLA-A0201–MVWGPDPLYV (A2-MVW) crystals were grown in 0.2 M ammonium chloride, 0.1 M TRIS (pH 8), 20% PEG 6000; HLA-A0201–YLGGPDFPTI (A2-YLG) crystals were grown in 0.2 M sodium nitrate, 0.1 M BIS TRIS propane (pH 6.5), 20% PEG 3350; HLA-A0201–AQWGPDPAAA (A2-AQW) crystals were grown in 0.2 M sodium malonate, 0.1 M BIS TRIS propane (pH 6.5), 20% PEG3350; HLA-A0201–RQFGPDWIVA (A2-RQF[A]) crystals were grown in 0.2 M sodium sulphate, 0.1 M BIS TRIS propane (pH 6.5), 20% PEG 3350; HLA-A0201–RQWGPDPAAV (A2-RQW) crystals were grown in 0.1 M TRIS (pH 8), 20% PEG 8000, 15% glycerol; HLA-A0201–YQFGPDFPTA (A2-YQF) crystals were grown in 0.1 M TRIS (pH 8), 25% PEG 4000, 15% glycerol; HLA-A0201–RQFGPDFPTI (A2-RQF[I]) crystals were grown in 0.2 M potassium/sodium tartrate, 0.1 M BIS TRIS propane (pH 8.5), 20% PEG 3350; 1E6-A2-MVW crystals were grown in 0.1 M HEPES (pH 7.5), 15% PEG 4000, 0.2 M sodium acetate; 1E6-A2-YLG crystals were grown in 0.1 M sodium cacodylate (pH 6.5), 15% PEG 4000, 0.2 M sodium acetate; 1E6-A2-AQW crystals were grown in 0.2 M sodium citrate, 0.1 M BIS TRIS propane (pH 6.5), 20% PEG 3350; 1E6-A2-RQF(A) crystals were grown in 0.1 M HEPES (pH 7), 15% PEG 4000, 0.2 M sodium acetate; 1E6-A2-RQW crystals were grown in 0.2 M sodium cholride, 0.1 M MES (pH 6), 20% PEG 6000; 1E6-A2-YQF crystals were grown in 0.2 M sodium cholride, 0.1 M HEPES (pH 7), 20% PEG 3350; and 1E6-A2-RQF(I) crystals were grown in 0.1 M HEPES (pH 7.5), 15% PEG 4000, 0.2 M sodium acetate. Crystallization screens were conducted using an Art-Robbins Phoenix dispensing robot (Alpha Biotech Ltd.), and data were collected at 100 K at the DLS at a wavelength of 0.98 Å using an ADSC Q315 CCD detector. Reflection intensities were estimated using XIA2 (75), and the data were analyzed with Scala and the CCP4 package (76). Structures were solved with molecular replacement using Phaser (77). Sequences were adjusted with Coot (78), and the models were refined with REFMAC5. Graphical representations were prepared with PyMOL (79). The reflection data and final model coordinates were deposited with the PDB database (A2-MVW PDB: 5C0H; A2-YLG PDB: 5C0G; A2-AQW PDB: 5C0D; A2-RQF[A] PDB: 5C0J; A2-RQW PDB: 5C0F; A2-YQF PDB: 5C0E; A2-RQF[I] PDB: 5C0I; 1E6-A2-MVW PDB: 5C0A; 1E6-A2-YLG PDB: 5C09; 1E6-A2-AQW PDB: 5HYJ; 1E6-A2-RQF[A] PDB: 5C0C; 1E6-A2-RQW PDB: 5C08; 1E6-A2-YQF PDB: 5C07; and 1E6-A2-RQF[I] PDB: 5C0B). Peptide motif predictions were performed by searching a viral database compiled using publicly available protein sequences of over 1,924,572 unique decamer peptides from the proteome of viral pathogens (80). The motif xOxGPDxxxO — where O is anyone of the hydrophobic amino acid residues A,V, I, L, M, Y, F, and W that might allow binding to HLA-A*0201 — was used as the search parameter. Pearson’s correlation analysis was performed to determine the relationship between TCR binding affinity and antigen potency, structural correlates, or thermodynamics using Origin Lab 9.0 pro.
PMC4774019	The immunity-related GTPase Irga6 dimerizes in a parallel head-to-head fashion	The immunity-related GTPases (IRGs) constitute a powerful cell-autonomous resistance system against several intracellular pathogens. Irga6 is a dynamin-like protein that oligomerizes at the parasitophorous vacuolar membrane (PVM) of Toxoplasma gondii leading to its vesiculation. Based on a previous biochemical analysis, it has been proposed that the GTPase domains of Irga6 dimerize in an antiparallel fashion during oligomerization. We determined the crystal structure of an oligomerization-impaired Irga6 mutant bound to a non-hydrolyzable GTP analog. Contrary to the previous model, the structure shows that the GTPase domains dimerize in a parallel fashion. The nucleotides in the center of the interface participate in dimerization by forming symmetric contacts with each other and with the switch I region of the opposing Irga6 molecule. The latter contact appears to activate GTP hydrolysis by stabilizing the position of the catalytic glutamate 106 in switch I close to the active site. Further dimerization contacts involve switch II, the G4 helix and the trans stabilizing loop. The Irga6 structure features a parallel GTPase domain dimer, which appears to be a unifying feature of all dynamin and septin superfamily members. This study contributes important insights into the assembly and catalytic mechanisms of IRG proteins as prerequisite to understand their anti-microbial action.Immunity-related GTPases (IRGs) comprise a family of dynamin-related cell-autonomous resistance proteins targeting intracellular pathogens, such as Mycobacterium tuberculosis , Mycobacterium avium , Listeria monocytogenes , Trypanosoma cruzi , and Toxoplasma gondii [3, 5–11]. In mice, the 23 IRG members are induced by interferons, whereas the single human homologue is constitutively expressed in some tissues, especially in testis . In non-infected cells, most IRGs are largely cytosolic. However, members of a small sub-family with regulatory function associate with specific intracellular membranes, with one member favoring the endoplasmic reticulum [13, 14] and others the Golgi membrane [7, 14] and the endolysosomal system . Infection by certain intracellular pathogens initiates the redistribution of several effector members to the parasitophorous vacuole, followed by its disruption [7, 14, 16, 17]. In this way, IRGs contribute to the release of the pathogen into the cytoplasm and its subsequent destruction. Irga6, one of the effector IRG proteins, localizes to the intact parasitophorous vacuole membrane (PVM) and, after disruption of the PVM, is found associated with vesicular accumulations, presumably derived from the PVM [7, 15, 18, 19]. A myristoylation site at Gly2 is necessary for the recruitment to the PVM but not for the weak constitutive binding to the ER membrane [14, 20]. An internally oriented antibody epitope on helix A between positions 20 and 24 was demonstrated to be accessible in the GTP-, but not in the GDP-bound state [20, 21]. This indicates large-scale structural changes upon GTP binding that probably include exposure of the myristoyl group, enhancing binding to the PVM. Biochemical studies indicated that Irga6 hydrolyses GTP in a cooperative manner and forms GTP-dependent oligomers in vitro and in vivo [20, 22]. Crystal structures of Irga6 in various nucleotide-loaded states revealed the basic architecture of IRG proteins, including a GTPase domain and a composite helical domain . These studies additionally showed a dimerization interface in the nucleotide-free protein as well as in all nucleotide-bound states. It involves a GTPase domain surface, which is located at the opposite side of the nucleotide, and an interface in the helical domain, with a water-filled gap between the two contact surfaces. Mutagenesis of the contact surfaces suggests that this "backside" interface is not required for GTP-dependent oligomerization or cooperative hydrolysis, despite an earlier suggestion to the contrary . Extensive biochemical studies suggested that GTP-induced oligomerization of Irga6 requires an interface in the GTPase domain across the nucleotide-binding site . Recent structural studies indicated that a 'G interface' is typical of dynamin superfamily members, such as dynamin [25, 26], MxA [27, 28], the guanylate binding protein-1 (GBP-1) , atlastin [30, 31] and the bacterial dynamin-like proteins (BDLP) [32, 33]. For several of these proteins, formation of the G interface was shown to trigger GTP hydrolysis by inducing rearrangements of catalytic residues in cis. In dynamin, the G interface includes residues in the phosphate binding loop, the two switch regions, the 'trans stabilizing loop' and the 'G4 loop'. For Irga6, it was demonstrated that besides residues in the switch I and switch II regions, the 3'-OH group of the ribose participates in this interface . Since the signal recognition particle GTPase and its homologous receptor (called FfH and FtsY in bacteria) also employ the 3'-OH ribose group to dimerize in an anti-parallel orientation therefore activating its GTPase , an analogous dimerization model was proposed for Irga6 . However, the crystal structure of Irga6 in the presence of the non-hydrolyzable GTP analogue 5'-guanylyl imidodiphosphate (GMPPNP) showed only subtle differences relative to the apo or GDP-bound protein and did not reveal a new dimer interface associated with the GTPase domain . This structure was obtained by soaking GMPPNP in nucleotide-free crystals of Irga6, an approach which may have interfered with nucleotide-induced domain rearrangements. To clarify the dimerization mode via the G interface, we determined the GMPPNP-bound crystal structure of a non-oligomerizing Irga6 variant. The structure revealed that Irga6 can dimerize via the G interface in a parallel head-to-head fashion. This dimerization mode explains previously published biochemical data, and shows in particular how the 3'-OH group of the ribose participates in the assembly. Our data suggest that a parallel dimerization mode may be a unifying feature in all dynamin and septin superfamily proteins. Previous results indicated that Irga6 mutations in a loosely defined surface region (the "secondary patch"), which is distant from the G-interface and only slightly overlapping with the backside interface (see below), individually reduced GTP-dependent oligomerization . A combination of four of these mutations (R31E, K32E, K176E, and K246E) essentially eliminated GTP-dependent assembly (Additional file 1: Figure S1) and allowed crystallization of Irga6 in the presence of GMPPNP. Crystals diffracted to 3.2 Å resolution and displayed one exceptionally long unit cell axis of 1289 Å (Additional file 1: Table S1). The structure was solved by molecular replacement and refined to Rwork/Rfree of 29.7 %/31.7 % (Additional file 1: Table S2). The asymmetric unit contained seven Irga6 molecules that were arranged in a helical pattern along the long cell axis (Additional file 1: Figure S2). Like other dynamin superfamily members, the GTPase domain of Irga6 comprises a canonical GTPase domain fold, with a central β-sheet surrounded by helices on both sides (Fig. 1a-c). The helical domain is a bipartite structure composed of helices αA-C at the N-terminus and helix αF-L at the C-terminus of the GTPase domain. Overall, the seven molecules in the asymmetric unit are very similar to each other, with root mean square deviations (rmsd) ranging from 0.32 – 0.45 Å over all Cα atoms. The structures of the seven molecules also agree well with the previously determined structure of native GMPPNP-bound Irga6 (PDB: 1TQ6; rmsd of 1.00-1.13 Å over all Cα atoms).Fig. 1Structure of the Irga6 dimer. a Schematic view of the domain architecture of mouse Irga6. The first and last amino acids of each domain are indicated. b Ribbon-type representation of the Irga6 dimer. In the left molecule, domains are colored according to the domain architecture, the right molecule is colored in grey. The nucleotide and Mg ion (green) are shown in sphere representation. The GTPase domain dimer is boxed. The dotted line indicates a 2-fold axis. Secondary structure was numbered according to ref. . c Top view on the GTPase domain dimer. d Magnification of the contact sites. Dotted lines indicate interactions. e Superposition of different switch I conformations in the asymmetric unit; the same colors as in Additional file 1: Figure S2 are used for the switch I regions of the individual subunits. Switch I residues of subunit A (yellow) involved in ribose binding are labelled and shown in stick representation. Irga6 immunity-related GTPase 6 Structure of the Irga6 dimer. a Schematic view of the domain architecture of mouse Irga6. The first and last amino acids of each domain are indicated. b Ribbon-type representation of the Irga6 dimer. In the left molecule, domains are colored according to the domain architecture, the right molecule is colored in grey. The nucleotide and Mg ion (green) are shown in sphere representation. The GTPase domain dimer is boxed. The dotted line indicates a 2-fold axis. Secondary structure was numbered according to ref. . c Top view on the GTPase domain dimer. d Magnification of the contact sites. Dotted lines indicate interactions. e Superposition of different switch I conformations in the asymmetric unit; the same colors as in Additional file 1: Figure S2 are used for the switch I regions of the individual subunits. Switch I residues of subunit A (yellow) involved in ribose binding are labelled and shown in stick representation. Irga6 immunity-related GTPase 6 The seven Irga6 molecules in the asymmetric unit form various higher order contacts in the crystals. Within the asymmetric unit, six molecules dimerize via the symmetric backside dimer interface (buried surface area 930 Å), and the remaining seventh molecule forms the same type of interaction with its symmetry mate of the adjacent asymmetric unit (Additional file 1: Figure S2a, b, Figure S3). This indicates that the introduced mutations in the secondary patch, from which only Lys176 is part of the backside interface, do, in fact, not prevent this interaction. Another assembly interface with a buried surface area of 450 Å, which we call the “tertiary patch”, was formed via two interaction sites in the helical domains (Additional file 1: Figure S2c, d, S3). In this interface, helices αK from two adjacent molecules form a hydrogen bonding network involving residues 373-376. Furthermore, two adjacent helices αA form hydrophobic contacts. It was previously shown that the double mutation L372R/A373R did not prevent GTP-induced assembly , so there is currently no evidence supporting an involvement of this interface in higher-order oligomerization. Strikingly, molecule A of one asymmetric unit assembled with an equivalent molecule of the adjacent asymmetric unit via the G-interface in a symmetric parallel fashion via a 470 Å interface. This assembly results in a butterfly-shaped Irga6 dimer in which the helical domains protrude in parallel orientations (Fig. 1b, Additional file 1: Figure S3). In contrast, the other six molecules in the asymmetric unit do not assemble via the G interface. The G interface in molecule A can be subdivided into three distinct contact sites (Fig. 1c, d). Contact site I is formed between R159 and K161 in the trans stabilizing loops, and S132 in the switch II regions of the opposing molecules. Contact site II features polar and hydrophobic interactions formed by switch I (V104, V107) with a helix following the guanine specificity motif (G4 helix, K184 and S187) and the trans stabilizing loop (T158) of the opposing GTPase domain. In contact site III, G103 of switch I interacts via its main chain nitrogen with the exocyclic 2’-OH and 3’-OH groups of the opposing ribose in trans, whereas the two opposing exocyclic 3’-OH group of the ribose form hydrogen bonds with each other. Via the ribose contact, switch I is pulled towards the opposing nucleotide (Fig. 1e). In turn, E106 of switch I reorients towards the nucleotide and now participates in the coordination of the Mg ion (Fig. 1e, Additional file 1: Figure S4). E106 was previously shown to be essential for catalysis , and the observed interactions in contact site III explain how dimerization via the ribose is directly coupled to the activation of GTP hydrolysis. The G interface is in full agreement with previously published biochemical data that indicate crucial roles of E77, G103, E106, S132, R159, K161, K162, D164, N191, and K196 for oligomerization and oligomerization-induced GTP hydrolysis . All of these residues directly participate in contacts (G103, S132, R159, and K161) or are in direct vicinity to the interface (E77, E106, K162, D164, and N191). Residues E77, K162, and D164 appear to orient the trans stabilizing loop which is involved in interface formation in contact site II. In the earlier model of an anti-parallel G interface, it was not possible to position the side chain of R159 to avoid steric conflict . In the present structure, the side-chain of R159 projects laterally along the G interface and, therefore, does not cause a steric conflict. The buried surface area per molecule (BSA) of the G interface in Irga6 is relatively small (470 Å) compared to that of other dynamin superfamily members, such as dynamin (BSA: 1400 Å), atlastin (BSA: 820 Å), GBP-1 (BSA: 2060 Å), BDLP (BSA: 2300 Å) or the septin-related GTPase of immunity associated protein 2 (GIMAP2) (BSA: 590 Å) (Fig. 2). However, the relative orientations of the GTPase domains in these dimers are strikingly similar, and the same elements, such as switch I, switch II, the trans activating and G4 loops are involved in the parallel dimerization mode in all of these GTPase families.Fig. 2A conserved dimerization mode via the G interface in dynamin and septin GTPases. The overall architecture of the parallel GTPase domain dimer of Irga6 is related to that of other dynamin and septin superfamily proteins. The following structures are shown in cylinder representations, in similar orientations of their GTPase domains: a the GMPPNP-bound Irga6 dimer, b the GDP-AlF4 -bound dynamin 1 GTPase-minimal BSE construct [pdb 2X2E], c the GDP-bound atlastin 1 dimer [pdb 3Q5E], d the GDP-AlF3- bound GBP1 GTPase domain dimer [pdb 2B92], e the BDLP dimer bound to GDP [pdb 2J68] and f the GTP-bound GIMAP2 dimer [pdb 2XTN]. The GTPase domains of the left molecules are shown in orange, helical domains or extensions in blue. Nucleotide, Mg (green) and AlF4 are shown in sphere representation, the buried interface sizes per molecule are indicated on the right. Irga6 immunity-related GTPase 6, GMPPNP 5'-guanylyl imidodiphosphate, GTP guanosine-triphosphate, BDLP bacterial dynamin like protein, GIMAP2, GTPase of immunity associated protein 2 A conserved dimerization mode via the G interface in dynamin and septin GTPases. The overall architecture of the parallel GTPase domain dimer of Irga6 is related to that of other dynamin and septin superfamily proteins. The following structures are shown in cylinder representations, in similar orientations of their GTPase domains: a the GMPPNP-bound Irga6 dimer, b the GDP-AlF4 -bound dynamin 1 GTPase-minimal BSE construct [pdb 2X2E], c the GDP-bound atlastin 1 dimer [pdb 3Q5E], d the GDP-AlF3- bound GBP1 GTPase domain dimer [pdb 2B92], e the BDLP dimer bound to GDP [pdb 2J68] and f the GTP-bound GIMAP2 dimer [pdb 2XTN]. The GTPase domains of the left molecules are shown in orange, helical domains or extensions in blue. Nucleotide, Mg (green) and AlF4 are shown in sphere representation, the buried interface sizes per molecule are indicated on the right. Irga6 immunity-related GTPase 6, GMPPNP 5'-guanylyl imidodiphosphate, GTP guanosine-triphosphate, BDLP bacterial dynamin like protein, GIMAP2, GTPase of immunity associated protein 2 IRG proteins are crucial mediators of the innate immune response in mice against a specific subset of intracellular pathogens, all of which enter the cell to form a membrane-bounded vacuole without engagement of the phagocytic machinery. As members of the dynamin superfamily, IRGs oligomerize at cellular membranes in response to GTP binding. Oligomerization and oligomerization-induced GTP hydrolysis are thought to induce membrane remodeling events ultimately leading to disruption of the PVM. Recent structural and mechanistic analyses have begun to unravel the molecular basis for the membrane-remodeling activity and mechano-chemical function of some members (reviewed in ). For example, for dynamin and atlastin, it was shown that GTP binding and/or hydrolysis leads to dimerization of the GTPase domains and to the reorientation of the adjacent helical domains. The resulting domain movement was suggested to act as a “power stroke” during membrane remodeling events . However, for other dynamin superfamily members such as IRGs, the molecular basis for GTP hydrolysis and the exact role of the mechano-chemical function are still unclear. Our structural analysis of an oligomerization- and GTPase-defective Irga6 mutant indicates that Irga6 dimerizes via the G interface in a parallel orientation. Only one of the seven Irga6 molecules in the asymmetric unit formed this contact pointing to a low affinity interaction via the G interface, which is in agreement with its small size. In the crystals, dimerization via the G interface is promoted by the high protein concentrations which may mimic a situation when Irga6 oligomerizes on a membrane surface. Such a low affinity interaction mode may allow reversibility of oligomerization following GTP hydrolysis. Similar low affinity G interface interactions were reported for dynamin and MxA . The dimerization mode is strikingly different from the previously proposed anti-parallel model that was based on the crystal structure of the signal recognition particle GTPase, SRP54 and its homologous receptor . However, the G dimer interface is reminiscent of the GTPase domain dimers observed for several other dynamin superfamily members, such as dynamin, GBP1, atlastin, and BDLP. It was recently shown that septin and septin-related GTPases, such as the Tocs GTPases or GTPases of immunity related proteins (GIMAPs) , also employ a GTP-dependent parallel dimerization mode. Based on phylogenetic and structural analysis, these observations suggest that dynamin and septin superfamilies are derived from a common ancestral membrane-associated GTPase that featured a GTP-dependent parallel dimerization mode . Importantly, our analysis indicates that IRGs are not outliers, but bona-fide representatives of the dynamin superfamily. Whereas the overall dimerization mode is similar in septin and dynamin GTPases, family-specific differences in the G interface and the oligomerization interfaces exist. For example, the involvement of the 2’ and 3’-OH groups of the ribose in the dimerization interface of Irga6 has not been observed for other dynamin and septin superfamily members. The surface-exposed location of the ribose in the Irga6 structure, with a wide-open nucleotide-binding pocket, facilitates its engagement in the dimerization interface. This contact, in turn, appears to activate GTP hydrolysis by inducing rearrangements in switch I and the positioning of the catalytic E106. During dimerization of GBP1, an arginine finger from the P loop reorients towards the nucleotide in cis to trigger GTP hydrolysis . In dynamin, the corresponding serine residue coordinates a sodium ion that is crucial for GTP hydrolysis . Irga6 bears Gly79 at this position, which in the dimerizing molecule A appears to approach the bridging imido group of GMPPNP via a main chain hydrogen bond. Higher resolution structures of the Irga6 dimer in the presence of a transition state analogue are required to show whether Gly79 directly participates in GTP hydrolysis or whether it may also position a catalytic cation. In dynamin, further assembly sites are provided by the helical domains which assemble in a criss-cross fashion to form a helical filament. In dynamin-related Eps15 homology domain containing proteins (EHDs), a second assembly interface is present in the GTPase domain . For Irga6, additional interfaces in the helical domain are presumably involved in oligomerization, such as the secondary patch residues whose mutation prevented oligomerization in the crystallized mutant. Further structural studies, especially electron microscopy analysis of the Irga6 oligomers, are required to clarify the assembly mode via the helical domains and to show how these interfaces cooperate with the G interface to mediate the regulated assembly on a membrane surface. Notably, we did not observe major rearrangements of the helical domain versus the GTPase domain in the Irga6 molecules that dimerized via the G interface. In a manner similar to BDLP , such large-scale conformational changes may be induced by membrane binding. Our structural analysis and the identification of the G-interface paves the way for determining the specific assembly of Irga6 into a membrane-associated scaffold as the prerequisite to understand its action as an anti-parasitic machine. Selenomethionine-substituted Mus musculus Irga6 was expressed as a GST-fusion from the vector pGEX-4T-2 in BL21 Rosetta2(DE3) cells according to reference . Protein was purified as previously described and the protein stored in small aliquots at a concentration of 118 mg/mL in 50 mM Tris-HCl, pH 7.4, 5 mM MgCl2, 2 mM DTT. Oligomerization and GTPase assays for the Irga6 mutant were carried out as described in . The protein was gently thawed on ice and diluted to a final concentration of 10 mg/mL with buffer containing 20 mM Tris-HCl, pH 7.5, 8 mM MgCl2, 3 mM DTT. GMPPNP was added to a final concentration of 2 mM. Crystallization was carried out in a 96 well format using the sitting drop vapor diffusion method. The reservoir contained 100 mM HEPES-NaOH pH 7.0, 9 % PEG4000, 6 % isopropanol. The sitting drop was set up using an Art Robbins Gryphon system and consisted of 200 nL protein solution and 200 nL reservoir solution. For cryo-protection, crystals were transferred into a cryo solution containing 33 % PEG4000, 3 % isopropanol, 50 mM HEPES pH 7.0, 4 mM MgCl2, 2 mM DTT, and 2 mM GMPPNP at 4 °C for at least 5 sec. Crystals were screened for diffraction at beamline BL 14.1 at BESSY II, Berlin, Germany. All data were recorded at beamline P11 at PETRA III, DESY Hamburg, Germany using a PILATUS 6 M detector. To achieve spot separation along the long cell axis, three data sets were collected with a φ increment of 0.05/0.1° at a temperature of 100 K using detector distances between 1300 and 598.5 mm (Additional file 1: Table S1). The wavelength was 0.972/0.979 Å. Calculation of an optimal data collection strategy was done with the Mosflm software . The high- and low-resolution datasets were processed and merged using the XDS program suite . Structure solution was done by molecular replacement with Phaser employing the structure of Irga6 without nucleotide [PDB: 1TQ2] as search model . Atomic model building was done by Coot . Iterative refinement was done using Phenix at a maximum resolution of 3.2 Å . For the refinement strategy, a seven-fold non-crystallographic symmetry as well as one molecule of Irga6 [PDB: 1TQ4] as high resolution reference structure was chosen. Five percent of the measured X-ray intensities were set aside from the refinement as cross-validation . Methionine sites in the protein were confirmed by the anomalous signal of the selenium atoms. Protein superposition was done with lsqkab and the PyMol Molecular Graphics System, Version 1.3 Schrödinger, LLC. Figures were prepared using the PyMOL Molecular Graphics System, Version 1.7.4 Schrödinger, LLC. Evaluation of atom contacts and geometry of the atomic model was done by the Molprobity server . Interface sizes were calculated by the PISA server . The Irga6 coordinates were submitted to the Protein Data Bank (pdb) database with accession code 5fph. http://www.rcsb.org/pdb/explore/explore.do?structureId=5fph. Our study indicates that Irg proteins dimerize via the G interface in a parallel head-to-head fashion thereby facilitating GTPase activation. These findings contribute to a molecular understanding of the anti-parasitic action of the Irg protein family and suggest that Irgs are bona-fide members of the dynamin superfamily.

Subsets and Splits

No community queries yet

The top public SQL queries from the community will appear here once available.