Cryo-EM structure of coagulation factor Va bound to activated protein C Open Access

Coagulation factor Va (FVa) is generated from the inactive precursor FV during the initiation phase of the blood coagulation cascade in which small amounts of thrombin and FXa are produced by the action of the tissue factor–FVIIa complex in the extrinsic pathway triggered by vascular injury.^1-3 The architecture of the multidomain A1-A2-B-A3-C1-C2 assembly of FV has recently been elucidated by cryogenic electron microscopy (cryo-EM).^4-8 The C domains provide the locale for interaction with membranes⁹ and support the A1 and A3 domains, with the A2 domain wedged between them. The large B domain is highly disordered^4,8 and keeps FV in its inactive form.¹⁰ Cleavage by thrombin at R709 in the A2 domain and R1545 in the A3 domain removes the B domain in its entirety and generates FVa as the active cofactor of the prothrombinase complex, comprising also FXa, Ca²⁺, and phospholipids.¹¹ In the common pathway of the coagulation cascade, the prothrombinase complex efficiently converts the zymogen prothrombin to the active protease thrombin that functions as a procoagulant, prothrombotic, and proinflammatory factor by cleaving fibrinogen and protease activated receptor 1 (PAR1).¹² In addition, thrombin regulates its own generation by activating protein C (PC), a vitamin K–dependent zymogen comprising an N-terminal γ-carboxyglutamic acid (Gla) domain responsible for interaction with membranes, 2 epidermal growth factor (EGF)–like domains, and a C-terminal protease domain (PD) containing the catalytic triad.¹³ Thrombin cleaves PC in the microvasculature with the assistance of the endothelial receptor thrombomodulin¹⁴ and generates activated PC (APC) that inactivates FVa by cleaving first at R506 in the A2 domain and then at R306 in the A1 domain with the assistance of protein S (PS).^15,16 The physiologic importance of this downregulation is well established. Severe deficiency of PC causes neonatal purpura fulminans,¹⁷ which is fatal unless treated by replacement therapy, and the FV variant R506Q or FV^Leiden is a common genetic risk factor for venous thrombosis in humans.^18-20 Mild deficiency²¹ or impaired activation of PC owing to genetic mutations²² also increases the risk of venous thromboembolism, and low levels of APC are linked to atherosclerosis, stroke, sepsis, and disseminated intravascular coagulation.^23-26 

Coagulation FVa promotes the response to vascular injury through its role as cofactor in the prothrombinase complex and downregulates this response as a target of the anticoagulant APC.^1,3 Biochemical studies have shown that assembly of prothrombinase protects FVa from APC inactivation,²⁷ suggesting that FXa shares epitopes of recognition with APC on FVa. Structural information on free FVa⁴ or bound to FXa in the prothrombinase complex^5,7 has been obtained by cryo-EM and the structure of APC has been solved by X-ray crystallography.^28,29 However, the molecular organization of the FVa-APC complex remains unknown. Information on this complex would advance basic knowledge and support studies aimed at enhancing the anticoagulant activity of APC triggered by cleavage of FVa,^30,31 or at dissociating this activity from the anti-inflammatory and cytoprotective functions triggered by cleavage of PAR1.^32,33 Efforts to dissociate these functions for therapeutic intervention have enjoyed some success^30,33-35 and would benefit from the availability of structural information.

In this study, we present a 3.15 Å–resolution cryo-EM structure of the FVa-APC complex, with APC bearing an Ala substitution of the catalytic Ser to prevent hydrolysis. The structure adds to recent cryo-EM successes in the analysis of coagulation factors and their complexes^4-6,8,36-45 and offers valuable insights into the molecular determinants of a key regulatory interaction of the coagulation cascade.

Materials obtained commercially were plasma human FVa (HCVA-0110; Prolytix, Essex Junction, VT), Quantifoil R 1.2/1.3, Cu 300 mesh grids (Q350CR1.3; Electron Microscopy Sciences, Hatfield, PA), pRC/CMV/Neo (Invitrogen, Waltham, MA), QuikChange Lightning site-directed mutagenesis kit (Agilent Technologies), X-tremeGENE 9 DNA transfection reagent (Roche Applied Science), and SimplyBlue SafeStain (LC6065; Invitrogen, Waltham, MA). PC with the catalytic Ser inactivated with Ala (S360A) to prevent hydrolysis during cryo-EM studies and a C-terminal HPC-4 tag (EDQVDPRLIDGK) was expressed in adherent baby hamster kidney cells and purified as described previously.⁴⁶ Briefly, conditioned media was collected over a period of several weeks and purified over an HPC-4 antibody immunoaffinity column. The mutant was then activated with thrombin and thrombomodulin at 37°C for 3 to 4 hours in 20 mM tris(hydroxymethyl)aminomethane–HCl, 150 mM NaCl, 5 mM CaCl₂, pH 7.4, and passed over Q Sepharose (HiTrap Q Sepharose, Cytiva) and heparin column (Cytiva) in tandem to remove thrombin and thrombomodulin. The Q Sepharose column was then detached, and APC S360A was eluted using a NaCl gradient.

The S360A mutant of APC was concentrated and buffer exchanged at pH 7.4 into 20 mM HEPES (N-2-hydroxyethylpiperazine-N′-2-ethanesulfonic acid), 150 mM NaCl, and 5 mM CaCl₂ by size exclusion chromatography using a Superdex 200 Increase 10/300 GL column. Peak fractions were collected and concentrated. At time of freezing, grids were glow discharged on a GloQube (Quorum) for 20 or 40 seconds (described hereafter) and 3 μL FVa (0.1 mg/mL) mixed with APC at 1:3 molar ratio was applied onto each grid, then immediately plunge frozen in liquid ethane on an FEI Vitrobot Mark IV (ThermoFisher Scientific). Grids were loaded onto a Titan Krios G3 cryo–transmission EM (ThermoFisher Scientific) operating at 300 kV, equipped with a Falcon IV (4096 × 4096) direct electron detector (ThermoFisher Scientific). Data were collected with 2 separate calibrated pixel sizes because of microscope upgrades during the study and scaled to a final 0.8701 Å by Fourier cropping. A portion of the data was collected at 30° tilt to enhance the angular distribution. A set defocus range of −1 to −2.4 mm was used during collection and the individual e⁻/Ų for each data set is outlined in Table 1 and the Protein Data Bank (PDB) deposition. Exposures were collected from 3 grids treated as follows: grid 1, glow discharge for 40 seconds, polylysine treatment for 30 minutes, followed by 2 washes using sample buffer; grid 2, glow discharge for 40 seconds, standard amylamine vapor treatment, and sample spiked with 40 μM n-dodecyl-β-D-maltoside 30 minutes before application; and grid 3, glow discharge for 20 seconds, followed by sample application. A total of 12 457 micrographs were collected. Relevant data are summarized in Table 1.

Patch-based motion correction, dose weighting, and patch-based contrast transfer function estimation were performed sequentially in cryoSPARC.⁴⁷ Particles were initially picked using a Blob Picker–type job and cleaned by iterative rounds of 2-dimensional (2D) classification. Good particles were used to create multiple ab initio 3D construction volumes and similar initial models were combined into a single volume to create 2D templates for template picker. Particles selected by template picker were cleaned as described previously and the cleanest particle data set was used to train Topaz neural network–based picker.⁴⁸ Particles from different types of particle picker jobs were combined and duplicate particles removed (using a 0.75 particle diameter as threshold), then cleaned iteratively with 2D classification and 3D heterogenous refinement at later stages. Particles were reextracted with recentering using a 420-pixel box. New ab initio 3D construction with multiple classes implementing a high similarity score, followed by heterogenous refinement and 3D classification, was carried out to capture potential heterogeneity. 3D volumes were compared in University of California San Francisco (UCSF) ChimeraX.⁴⁹ The resultant 3D volumes were deemed similar and were all pooled into single nonuniform refinement giving a consensus refinement with an overall masked resolution of 2.88 Å (EMD-48439) and 2.79 Å (EMD-48461). Masked local refinement of the FVa domains A1 (EMD-48465), A2 (EMD-48466), A3 (EMD-48472), C1 (EMD-48473), and C2 (EMD-48474), and APC (EMD-48475) was carried out to improve density of mobile components and disordered loops. These maps were used to guide the initial backbone trace of the new FVa and APC domains in the complex using PDB 9CTH and PDB 1AUT as initial models for FVa and APC, respectively. The consensus maps (EMD-48461 and EMD-48439) were obtained using a different initial random particle seed and guided assignment of the loop connecting the A1 and A2 domains with no sidechain resolution. 3D flexibility refinement (EMD-48481) with custom mesh defining each domain and its interface of connectivity was used to model continuous heterogeneity of the consensus map. In our recent experience, proteins containing membrane-binding domains exhibit significant orientation bias in the absence of lipids. Although the use of various freezing conditions and tilt aided diversification of 2D orientations and ameliorated initial observed anisotropy because of preferred orientation, some views of the complex remained only available at low resolution. Combined with small size, extensive glycosylation, and flexibility some degree of anisotropy was still present in the final maps (supplemental Figure 1, available on the Blood website). Further particle classification and subsetting to limit resolution spread across views while maintaining unique views allowed the selection of 384 239 particle stack for final processing. Two models were deposited: PDB 9MOT (ab initio model associated with EMD-48481) and PDB 9MOV (flexible-fit model associated with EMD-48439). PDB 9MOV includes the EGF and Gla domains of APC flexibly fit into the density map based on prominent posttranslational modification and secondary structural features as refinement in that region did not allow assignment of individual residues. A combination of PDB 9MOT (ab initio model), 1LQV (for APC Gla domain), and AlphaFold AF-P04070 (for APC EGF domains) was used for the flexible model reconstruction. All maps were sharpened by DeepEMhancer.⁵⁰ Map resolution was estimated using gold-standard Fourier shell correlation at a Fourier shell correlation threshold of 0.143. UCSF ChimeraX,⁴⁹ Coot,⁵¹ and Phenix⁵² were used for model building and refinement. Relevant parameters are summarized in Table 1. Representative 2D class averages are reported in Figure 1C.

Deposited sequences are numbered sequentially for the mature forms of both FVa and APC. Residues of the PD of APC are also numbered according to chymotrypsin, as usually done for serine proteases.^53,54 For example, E357(c192) in the PD refers to residue 357 in the deposited sequence of APC and c192 indicates the same residue in the chymotrypsin numbering. Residues inserted relative to the sequence of chymotrypsin bear a lowercase letter after the number, for example, K311(c149c) indicates the third insertion after residue 149 and E215(c60a) indicates the first insertion after residue 60.

The recent cryo-EM structure of FVa⁴ sets the stage for the analysis of the physiologically relevant interaction with APC. Especially important is the information on the A1-A2-A3-C1-C2 assembly of FVa, the sites of thrombin and APC cleavage, and the putative epitope for APC binding in the A2 domain. The sites of APC cleavage at R306 and R506 linked to FVa inactivation are 60% exposed to solvent, which supports a scenario in which APC could cleave at either site and with different specificities.⁵⁵ However, the conditions used in this study are expected to promote cleavage at R506^1,56 because cleavage at R306 requires the presence of PS and phospholipids.⁵⁷ 

The cryo-EM structure of the FVa-APC complex was solved at 3.15 Å resolution and shows the 2 proteins interacting only through the PD of APC and the A2 domain of FVa (Figure 1) via strong electrostatic coupling (Figure 2), with R506 penetrating the primary specificity (S1) site⁵⁸ of APC. The auxiliary Gla and EGF domains of APC are highly mobile and remain separated from FVa. The A1-A2-A3-C1-C2 assembly of FVa shows notable differences with FVa in the prothrombinase complex^5,7 (root mean square deviation [RMSD] of 3.97 Å over 1197 Cα atoms) and FVa free⁴ (RMSD of 2.64 Å over 985 Cα atoms), whereas the PD of APC is very similar (RMSD of 0.81 Å over 197 Cα atoms) to the only available X-ray structure of APC, which lacks the Gla domain^28,29 (Figure 3). The C domains of FVa retain the distorted jelly-roll β-barrel architecture reported in previous cryo-EM structures^4,5,7 and the X-ray structures of bovine FVai⁵⁹ and the recombinant C2 domain of human FVa and FVIIIa^60,61 (Figure 3A; supplemental Figure 2). The C1 domain (residues 1878-2036) differs from that of FVa in the prothrombinase complex (RMSD of 2.68 Å over 144 Cα atoms) insofar as residues at the tip of the loops facing the membrane shift upward, with the Cα atoms of residues G1915, Y1956, and S1997 moving up to 8.0 Å. The C2 domain (residues 2037-2196) also differs from that of FVa in the prothrombinase complex (RMSD of 2.50 Å over 130 Cα atoms), with residues W2064, R2080, and L2116 at the tip of the loops featuring a similar upward shift of up to 13 Å. The changes are comparable with those observed with free FVa (Figure 3A) for both the C1 (RMSD of 2.01 Å over 149 Cα atoms) and C2 (RMSD of 3.53 Å over 131 Cα atoms) domains.

The A1 domain (residues 1-316) is similar to that of FVa in the prothrombinase complex (RMSD of 1.83 Å over 272 Cα atoms) or free (RMSD of 1.66 Å over 253 Cα atoms), with the site of APC cleavage at R306 becoming more exposed (72% vs 32%) to solvent. Importantly, the entire segment ³⁰⁵TRNLKKITREQRRHM³¹⁹ containing the potential epitope for APC binding, that is, residues 311 through 325,⁶⁴ is pushed upward >4 Å because of interaction with APC (Figure 3A). The critical A2 domain (residues 317-709) contains the site of APC cleavage at R506 that is 97% buried inside the S1 site of APC. The A2 domain features important differences with that of FVa in the prothrombinase complex (RMSD of 3.11 Å over 296 Cα atoms) or free (RMSD of 2.55 Å over 255 Cα atoms). The ⁶⁵⁴VKCIPDDDEDSYEIFEP⁶⁷⁰ segment resolved at the terminal end of this domain is displaced by APC binding and changes orientation (RMSD of 8.43 Å over 17 Cα atoms), with P670 moving 14 Å. The segment positions itself like a “latch” over the PD of APC (Figure 2; supplemental Figures 1 and 2) and functions as an “exosite” ligand that directs the site of cleavage at R506 to the active site of APC.

The remaining 39 residues of the A2 domain, from P671 to the site of thrombin activation at R709, are displaced by APC binding and are missing in the density map because of considerable disorder. This precludes assignment of the gate (⁶⁹⁶YDYQNRL⁷⁰²) and the lid (⁶⁷²ESTVMATRKMHDRLEPEDEE⁶⁹¹) that are instead well defined in the structure of prothrombinase and play a key role in the recognition of prothrombin.^5,7 The A3 domain (residues 1546-1877) also shows some differences to that of FVa in the prothrombinase complex (RMSD of 2.28 Å over 284 Cα atoms) or free (RMSD of 2.14 Å over 268 Cα atoms). The ¹⁵⁴⁶SNNGNRRYY¹⁵⁵⁵ segment, not resolved in the prothrombinase complex, is very similar to that of free FVa and remains widely separated from APC.

The PD of APC changes little upon binding to FVa compared to that of the H-D-Phe-Pro-Arg-CH₂Cl (PPACK)-bound form^28,29 (Figure 3B). A notable difference involves a rearrangement of the segment ³⁰⁴SSREKEAKRNRTF³¹⁶ (c145-c153; RMSD, 4.39 Å over 13 Cα atoms) analogous to the flexible autolysis loop of thrombin and lining the south rim of the active site entrance. Interestingly, the segment ³¹¹KRNRTFVLNFIKIPV³²⁵ (c149c-c162) containing parts of this region and the segment ¹⁴²GRPWKRMEKKRSHL¹⁵⁵ connecting EGF2 to the PD right through the activation peptide have been implicated in direct binding to FVa from competition assays.^65,66 The former segment bears a glycosylation site at N313(c150) that extends upward next to the latch but without contacting it (supplemental Figure 2). There are few contacts between K311(c149c) with N382 and K308(c149) with H379 (Figure 4; Table 2) that confirm evidence of the autolysis loop making a modest energetic contribution to FVa recognition⁶⁷ but a critical role in directing cleavage of FVa to R506 instead of R306.⁶⁸ The poor resolution of the latch (∼6 Å; supplemental Figure 1) does not document interactions with the distal portion, yet it is evident that a peptide containing the ³¹¹KRNRTFVLNFIKIPV³²⁵ (c149c-c162) segment would interfere with binding. The negatively charged stretch ⁶⁵⁹DDDED⁶⁶³ of the latch is well positioned to dock electrostatically on the 70-loop of APC (supplemental Figure 2; Figure 2B), as seen in the interaction of the negatively charged C-tail of the inhibitor hirudin with the 70-loop of thrombin.^12,69 Indeed, the density supports a strong H-bond between D660 and R230(c75) and possibly more ionic interactions (Figures 4 and 5E; Table 2). The importance of R229(c74) and R230(c75) in FVa recognition has been documented by biochemical studies.³³ The latter segment, ¹⁴²GRPWKRMEKKRSHL¹⁵⁵, remains separated from FVa. Interestingly, mutation of E149 to Ala in this segment enhances the anticoagulant activity of APC without affecting cleavage of chromogenic substrates and PAR1.³⁰ The underlying mechanism remains unresolved structurally because residue E149 is disordered in both the PPACK-bound X-ray structure of APC^28,29 and the cryo-EM structure of the FVa-APC complex reported here. Incidentally, the analogous region in thrombin comprises the A chain⁷⁰ in which mutation of charged residues compromises activity toward chromogenic and natural substrates due to long-range perturbation of the active site.⁷¹ 

The rest of the APC structure aligns well with the PPACK-bound form^28,29 up to the EGF2 domain and then diverges drastically at the level of the EGF1 domain that points to a diametrically opposite direction (Figure 3B). The Gla domain of APC is resolved, to our knowledge, for the first time and is not aligned with the main axis of the protein, which forces APC to assume a curved conformation as recently reported for FXa in the prothrombinase complex^5,7 or free in solution⁶² and for the closed form of prothrombin.^5,7,63 An important validation of the conformation of APC in the FVa-APC complex comes from the Cα-Cα distance of 77 Å between residue S12 in the Gla domain and R312(c149d) in the autolysis loop of the PD, which agrees well with the interprobe distance of 76 Å measured independently by single molecule spectroscopy.⁴⁶ 

The interface of the FVa-APC complex (Figure 4) contains 20 H-bonds, 5 salt bridges, and 25 hydrophobic interactions involving distinct pairs of residues (Figure 5; Table 2), as assessed by application of UCSF ChimeraX⁴⁹ and the Protein Interfaces, Surfaces, and Assemblies service at the European Bioinformatics Institute.⁷² The active site of APC is well organized for catalysis (Figure 6A). A 3.2-Å H-bond between D359(c194) and the N terminus of L170(c16) folds the oxyanion hole with the N atoms of G358(c193) and A360(c195), replacing the catalytic S360(c195), correctly pointing in the same direction toward the peptide bond to be cleaved between R506 and G507 (Table 2). The latch plays a key role in directing the site of cleavage at R506 into the active site of APC. Displacement of the latch by APC moves the segment containing R506 down to facilitate docking into the S1 pocket. At the same time, the latch occupies a position in direct collision with R312(c149d) in the autolysis loop of the PPACK-bound structure of APC (Figure 6B), thereby forcing the guanidinium group of this residue to relocate 26 Å toward the active site entrance and engage the side chain of E357(c192). As a result of this latch-induced shift, the autolysis loop in the FVa-bound structure of APC becomes more compact and the Cα-Cα distance between S304(c145) and R312(c149d) shrinks from 19 Å to 6.2 Å (Figure 6). The conformational change is further stabilized by polar interactions of R306(c147) with E357(c192) and Q509 (Table 2) at the recognition site P3′⁵⁸ in the P3-P4′ ⁵⁰⁴DRRGIQR⁵¹⁰ segment of FVa, containing the cleavage site at R506. As a result of these interactions, E357(c192) points away from D504 at P3 (Figure 6A) and provides no electrostatic hindrance to binding of FVa to the active site of APC. This explains the modest effect of replacing E357(c192) with Gln in the interaction with FVa and supports the conclusion that E357(c192) is an evolutionary adaptation to slow inhibition by plasma protease inhibitors.⁷³ The drastic conformational change in the autolysis loop induced by the latch promotes productive formation of the FVa-APC complex in which R506 penetrates the S1 site to engage D354(c189) in a strong salt bridge (Figure 6A) and several other residues lining the walls of the S1 site with hydrophobic interactions (Table 2). Additional hydrophobic contacts in the active site involve R505 at P2 with W380(c215) and the catalytic H211(c57). The docking requires a shift of nearly 10 Å of the Cα atoms of R505 and R506 and a flip of the 2 side chains that reaches almost 180° for R506. I508 at P2′ is in hydrophobic contact with the phenyl ring of Y302(c143) and R510 at P4′ makes a strong salt-bridge interaction with E215(c60a) in the 60-loop defining the upper rim of the active site (Figure 5C).

Other strong contacts involved in the FVa-APC interaction include salt bridges of E333(c170) in the short 170-helix with H318, K193(c39) in the 30-loop with D513 at P7, and R177(c23) with E669 in the latch (Figures 4 and 5; Table 2). Residues E330(c167) and E333(c170) are part of an epitope for PAR1 binding according to mutagenesis studies.³⁵ The small overlap with the epitope of FVa revealed by the cryo-EM structure reported here should guide future mutagenesis attempts at completely dissociating the anticoagulant and cytoprotective functions of APC. Mutation of residues K191(c37), K192(c38), and K193(c39) in the 30-loop, together with R229(c74) and R230(c75) in the 70-loop (Figures 4 and 5B,E) produces APC variants with greatly reduced anticoagulant activity but normal cytoprotective function.³³ Involvement of the 30-loop and a strong H-bond between D660 in the latch and R230(c75) in the 70-loop rationalize the properties of this mutant (Figure 5E; Table 2).

The cryo-EM structure of the FVa-APC complex advances basic knowledge on a key step of the PC pathway that downregulates the activation of prothrombin by prothrombinase.¹⁴ The structure is relevant to hematology because it reveals the molecular basis of an interaction that, when compromised, results in venous thrombosis found in many patients with thrombophilia.^18-20,74,75 APC docks on R506 driven by electrostatic coupling of positively charged residues in the 30-loop and 70-loop and negatively charged regions of the A2 domain of FVa, with the autolysis loop playing a major role. These, and other epitopes, documented by the cryo-EM structure largely confirm the results of biochemical studies.^{30,33,35,64,66,67,68} Importantly, the FVa-APC complex overlaid on the recent cryo-EM structure of the prothrombin-prothrombinase complex^5,7 (Figure 7) explains why cleavage at R506 and inactivation of FVa are impeded when FXa is bound to FVa.²⁷ APC and FXa are in direct competition for binding to the negatively charged region of the A2 domain of FVa. Binding of APC would also clash with the PD of prothrombin but would remain separated from the Kringle and Gla domain of this zymogen, contrary to previous claims.⁷⁶ Indeed, the architecture of the FVa-APC complex adds support to the recent notion that protein-protein interactions in the coagulation cascade involve mainly the PD of the enzyme and its target, with auxiliary EGF, Kringle, or Gla domains functioning as scaffolding needed to optimally align the enzyme-substrate complex.^5-7 The general validity of this new paradigm will await solution of more coagulation complexes by cryo-EM.

The auxiliary EGF and Gla domains of APC are oriented away from FVa and curve downward toward the plane of the membrane (Figures 1 and 2) where the Gla domain could align with the C domains of FVa. This alignment would make the 2 proteins interact with their vertical axis bent over the plane of the membrane and would require the Gla domain to efficiently process FVa, in agreement with biochemical studies.^55,76 However, the alternative scenario in which only FVa docks on the membrane and APC remains “suspended” (Figures 1 and 2) would be in accord with phospholipids providing a modest enhancement of FVa cleavage by APC at R506 and being required only for the PS-assisted cleavage at R306⁵⁷. To that end, the structure of the FVa-APC complex reported here affords an important starting point for future cryo-EM studies aimed at elucidating the cross talk between PS and the auxiliary domains of APC in the mechanism of FVa inactivation. Such studies are ongoing and may also reveal the architecture of PS and its cofactor role in another physiologically important interaction that involves FV-short, a splice variant of FV responsible for the mild bleeding phenotype associated with the East Texas bleeding disorder.^44,77-80 

The authors gratefully acknowledge Tracey Baird for her assistance with illustrations.

This study was supported, in part, by the Doisy Fund of the Edward A. Doisy Department of Biochemistry and Molecular Biology at Saint Louis University School of Medicine and by the National Institutes of Health research grants from the National Heart, Lung, and Blood Institute (HL049413, HL139554, and HL147821 [E.D.C.]). K.B. is supported by the Washington University Center for Cellular Imaging, which is funded, in part, by Washington University School of Medicine, the Washington University Diabetes Research Center (P30DK020579), and the Alvin J. Siteman Cancer Center at Barnes-Jewish Hospital and Washington University School of Medicine (P30CA091842).

Contribution: B.M.M. performed biochemical preparations, structure determination, and analysis; K.B. performed cryogenic electron microscopy data acquisition; K.B., B.M.M., and E.D.C. analyzed the results and prepared the manuscript.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Enrico Di Cera, Edward A. Doisy Department of Biochemistry and Molecular Biology, Saint Louis University School of Medicine, 1100 S Grand Blvd, St. Louis, MO 63104; email: enrico@slu.edu.