Cryo-EM structures of HCV E2 glycoprotein bound to neutralizing and non-neutralizing antibodies determined using bivalent Fabs as fiducial markers

Categories: Disease & Virus

May 30, 2025

Cryo-EM structure determination of Fab–E2 complexes using bivalent Fabs

Conventional Fabs are monovalent monomers. We recently reported the crystal structure of a bivalent dimeric form of Fab HC84.26.5D created by deleting a single residue, V_HSer113 (Kabat numbering), in the elbow region linking the V_H and C_H1 domains⁴⁰. Three-dimensional domain swapping of V_L and V_H domains with the corresponding V domains of a symmetry-related Fab molecule in the crystal resulted in formation of a doughnut-shaped Fab dimer with a rectangular hole in which the two V_L/V_H modules are separated by ~25 Å. In agreement with the crystal structure, a low resolution (7 Å) cryo-EM reconstruction of the Fab HC84.26.5D–E2 complex showed that the dimeric Fab bound two E2 molecules (~220 kDa)⁴⁰. However, in the present study, at a global resolution of 3.8 Å, the Fab HC84.26.5D–E2 complex formed a superassembly of ~440 kDa composed of four E2 molecules bound to a dimer of Fab dimers. Formation of this dimer of dimers is facilitated by contacts between the C_L domains of two domain-swapped dimers arranged in parallel. This arrangement is stabilized by hydrogen bonds between backbone atoms of C_LLeu153 from each dimer (Supplementary Fig. 1). Based on local resolution analysis and side-chain features in the Fab–E2 interfaces, a resolution of 3.8 Å was achieved for two locally refined cryo-EM maps of the Fab HC84.26.5D–E2 complex as determined from gold standard FSC curves (Table 1) (Supplementary Fig. 2) (Fig. 1b).

Table 1 Cryo-EM data collection, refinement, and validation statistics for Fab HC84.26.5D–E2 complex

To define the epitope recognized by antibody CBH7, we solved the cryo-EM structure of a ternary complex of monomeric Fab CBH7, dimeric Fab HC84.26.5D, and E2 to 3.3 Å resolution using dimeric Fab HC84.26.5D as a fiducial marker (Table 2) (Supplementary Fig. 3) (Fig. 1c). In contrast to Fab HC84.26.5D, deletion of V_HSer113 alone was insufficient to convert monomeric Fab CBH4B to a dimer, which instead required deletion of four elbow region residues, V_HSer112–C_HSer115. We determined the structure of the Fab CBH4B–E2 complex using dimeric Fab CBH4B as a fiducial marker and locally refined the Fab–E2 interface to 3.7 Å resolution based on gold standard FSC curves (Table 3) (Supplementary Fig. 4) (Fig. 1d).

Table 2 Cryo-EM data collection, refinement, and validation statistics for Fab CBH7–E2–Fab HC84.26.5D complex

Table 3 Cryo-EM data collection, refinement, and validation statistics for Fab CBH4B–E2 complex

To examine rotameric outliers in the Fab HC84.26.5D–E2, Fab CBH7–E2, and Fab CBH4B–E2 structural models (Tables 1–3), we analyzed rotamers for contact residues from E2 and the CDR loops (Supplementary Table 1). We focused on the locally refined models used for interface evaluation. No rotameric outliers were observed among the contact residues in any of the three structures. To examine the Ramachandran statistics, we analyzed dihedral angles for interfacial residues in the Fab CBH7–E2 and Fab CBH4B–E2 complexes (Supplementary Table 1). In the Fab CBH7–E2 and Fab CBH4B–E2 interfaces, only two (non-glycine) and one residue, respectively, fall within the allowed region, while the remaining residues are in the favored region of the Ramachandran plot.

We previously reported the affinity of monomeric versus dimeric forms of Fab HC84.26.5D for E2⁴⁰. As measured by biolayer interferometry (BLI), the affinity of monomeric Fab HC84.26.5D for E2 (K_D = 3.8 nM) was essentially identical to that of dimeric Fab HC84.26.5D (4.9 nM). We performed similar BLI affinity measurements for Fab CBH4B (Supplementary Fig. 5). The affinity of monomeric Fab CBH4B for E2 (K_D = 19 nM) is comparable to that of dimeric Fab CBH4B (K_D = 27 nM). Thus, dimerization of Fab HC84.46.5D and Fab CBH4B did not have a major impact on affinity for E2.

Although dimeric Fab HC84.26.5D and Fab CBH4B display similar overall architectures in the HC84.26.5D–E2 and CBH4B–E2 complexes, the disposition of V and C domains is different in the two doughnut-like structures, with central holes of different shapes and dimensions: a diamond-shaped hole of 38 × 46 Å for HC84.26.5D versus a rectangular-shaped hole of 46 × 47 Å for CBH4B (Supplementary Fig. 6a, b). Of note, the shape of dimeric Fab HC84.26.5D bound to E2 in the cryo-EM structure is significantly different from that of unbound dimeric Fab HC84.26.5D in the crystal structure (Supplementary Fig. 6c)⁴⁰, indicating flexibility. To understand the domain movements underpinning dimerization of Fab CBH4B and Fab HC84.26.5D, we compared dimeric Fab CBH4B and Fab HC84.26.5D with monomeric Fab CBH7, whose crystal structure we determined to 1.6 Å resolution (Table 4). As calculated using the program DynDom (https://dyndom.cmp.uea.ac.uk/dyndom)⁴¹, the angle of rotation around the switch peptide of the CBH4B H chain (residues 110–113), which links V_H and C_H1, is 76° with respect to CBH7 (Fig. 2a). The angle of rotation around the switch peptide of the CBH4B L chain (residues 105–110), which links V_L and C_L, is 109° with respect to CBH7. Likewise, the angles of rotation around the switch peptides of the Fab HC84.26.5D H (residues 110–113) and L chains (106–109) are 112° and 68° with respect to the corresponding H and L chains of CBH7 (Fig. 2a). These rotations result in more extended conformations of the H and L chains of dimeric CBH4B and HC84.26.5D compared to the H and L chains of monomeric CBH7. This is reflected in reductions in total buried surface area between V_H and C_H1 from 1670 Å² for CBH7 to 643 Å² for CBH4B (745 Å² for HC84.26.5D) and in total buried surface area between V_L and C_L from 1724 Å² for CBH7 to 621 Å² for CBH4B (742 Å² for HC84.26.5D), as analyzed by PDBePISA (https://www.ebi.ac.uk/pdbe/pisa/)⁴².

Table 4 X-ray data collection and refinement statistics for Fab CBH7

**Fig. 2: Domain movements and elbow regions in dimeric Fabs.**

The angles of rotation around the switch peptides of the H and L chains of dimeric Fab CBH4B with respect to monomeric Fab CBH7 (76° and 109°, respectively) differ significantly from the corresponding angles of rotation for dimeric Fab HC84.26.5D (112° and 68°, respectively) (Fig. 2a). The different geometries of the two domain-swapped Fab dimers are most likely a consequence of the different elbow region deletions: V_HSer112–C_HSer115 in Fab CBH4B versus only V_HSer113 in Fab HC84.26.5D. Antibody molecules contain a highly conserved ball-and-socket joint between V_H and C_H1 that, in monomeric Fab CBH7, comprises V_H residues Val11, Thr110, and Ser112, which form the socket, and C_H1 residues Phe145 and Pro146, which form the ball (Fig. 2b)⁴³. Although these five residues are present in dimeric Fab CBH4B and Fab HC84.26.5D, they do not form a canonical ball-and-socket joint. Conformational differences in the elbow region between V_L and C_L (residues 106–110) are shown in Fig. 2b.

Structure of the Fab HC84.26.5D–E2 complex

Superposition of two locally refined complexes in the Fab HC84.26.5D–E2 superassembly gave a root-mean-square difference (RMSD) of 0.4 Å for 818 α-carbon atoms, indicating close similarity. Therefore, the following description of Fab–E2 interactions applies to both refined complexes. E2 and HC84.26.5D adopt similar overall folds in the cryo-EM and crystal structures^20,40, with RMSD of 1.5 Å for superposition of 215 α-carbons of E2 (Supplementary Fig. 7a) and RMSD of 0.9 Å for superposition of 196 α-carbons of the Fab V_LV_H module (Supplementary Fig. 7b). We were able to model the complete interface between E2 and HC84.26.5D, including all six complementarity-determining regions (CDRs) of the antibody. By contrast, in the crystal structure of Fab HC84.26.5D⁴⁰, no density feature was detected for V_HCDR2 residues Ser25–His29, implying flexibility. In the crystal structure of E2 from HCV strain 1b09²⁰, E2 contains eight disulfide bonds formed by 16 highly conserved cysteine residues that impart to E2 its globular conformation. We could model all eight disulfides in the cryo-EM structure of E2 from this strain (Fig. 1b–d), in agreement with the crystal structure²⁰. We were also able to trace structural features for at least two N-acetylglucosamine residues at six of 10 potential N-glycosylation sites on the neutralizing face of E2, with additional sugar moieties visible at three of these sites (Supplementary Fig. 8).

Neutralizing epitope recognized by HC84.26.5D

Most broadly neutralizing antibodies targeting E2, including HC84.26.5D, contain an H chain encoded by the V_H1-69 gene segment⁴⁴. Previous structural studies of V_H1-69-encoded neutralizing antibodies bound to E2 revealed that they are characterized by a hydrophobic tip of their germline-encoded V_HCDR2 loop and a long somatically-generated V_HCDR3 loop (15–22 residues)^{8,20,21,22,23,24,26,44}. HC84.26.5D has a considerably shorter V_HCDR3 (10 residues), which lacks the intraloop disulfide that stabilizes V_HCDR3 in several other V_H1-69 broadly neutralizing antibodies. In the HC84.26.5D–E2 complex, HC84.26.5D contacts antigenic site 412 (AS412; residues Gln412–Arg424), front layer, CD81 binding loop, central β-sandwich, and back layer of E2 (Supplementary Table 2) (Fig. 3a, b). HC84.26.5D buries 1124 Å² of solvent-accessible surface on E2, compared to 633–1204 Å² for other front-layer-specific antibodies. Moreover, most E2 residues that contact HC84.26.5D are 90–100% conserved among all HCV genotypes, in agreement with the broad neutralizing capacity of this antibody. V_H dominates the interactions of HC84.26.5D with E2, accounting for 71% (868 Å²) of the surface area on E2 buried at the interface, compared to 563–1192 Å² for other front-layer-specific broadly neutralizing antibodies. Superposition of the HC84.26.5D–E2 complex onto complexes involving other V_H1-69-encoded neutralizing antibodies (AR3X, HEPC3, RM2-01, and 1382_01_H05)^{20,21,22,23,24,26} revealed that the antibodies engage E2 with different angles of approach (Fig. 3a) and with footprints that map to different but overlapping regions of the antigenic surface (Fig. 3b).

**Fig. 3: Comparison of neutralizing antibodies targeting the E2 front layer.**

The surface area on E2 buried by V_HCDR3 of HC84.26.5D (258 Å²) is considerably less than the surfaces buried by V_HCDR3 of AR3X (486 Å²), HEPC3 (580 Å²), and 1382_01_H05 (532 Å²), but comparable to the surface buried by V_HCDR3 of macaque RM2-01 (257 Å²) (Fig. 4a). Therefore, V_HCDR3 of HC84.26.5D plays a less prominent role in E2 recognition than V_HCDR3 of most other front layer-specific neutralizing antibodies. This is attributable to the particular angle of approach of HC84.26.5D rather than to its short V_HCDR3. Indeed, HC84.26.5D V_HCDR1 and V_HCDR2, which bury 220 Å² (28%) and 315 Å² (40%), respectively, of E2 surface, make contributions similar to V_HCDR3 258 Å² (32%) (Fig. 4a). Whereas other V_H1-69 antibodies contact E2 almost entirely via V_H¹⁹, V_L of HC84.26.5D accounts for 29% (330 Å²) of E2 surface buried at the interface. The surface area on E2 buried by HC84.26.5D (1122 Å²) is greater than the surfaces buried by HEPC3 (972 Å²), RM2-01 (658 Å²), and 1382_01_H05 (1014 Å²), but less than that by AR3X (1181 Å²). The larger surface buried by AR3X is attributable to an unusual 14 amino acid insertion in V_HCDR2²⁶.

**Fig. 4: Binding mode and interactions between E2 and CDR loops.**

In the HC84.26.5D–E2 structure, the V_HCDR1 is loop oriented towards the C-terminal portion of the highly conserved AS412 epitope (residues 412–423), with which it makes two hydrogen bonds (V_HCDR1 Thr28 Oγ1–O His421 E2 and V_HCDR1 His29 N–O His421 E2) and multiple hydrophobic contacts with Trp420 and Ile422 (Supplementary Table 2) (Fig. 4b). The tip of V_HCDR2 of HC84.26.5D (I52-P52a-D53-F54; Kabat numbering) conforms to the hydrophobic motif (I/V52-P52a-X53-F54) characteristic of other V_H1-69-encoded broadly neutralizing antibodies^{8,20,21,22,23,24,26}. This tip interacts extensively with a highly conserved hydrophobic groove between the front layer and the CD81 binding loop, with which it makes two hydrogen bonds and a π–cation interaction with Trp529 via V_HCDR2 Arg55 (Supplementary Table 2) (Fig. 4b). In addition, V_HCDR2 contacts back layer residue Tyr613, in agreement with previous epitope mapping²⁸. V_HCDR3 packs against a hydrophobic surface on E2 formed by two 3¹⁰ helices in the front layer, helix α1 (residues 438–442) and a short helix η2 (residues 447–449), and one 3¹⁰ helix in the back layer, helix α2 (residues 614–617). Further linking V_HCDR3 to the front layer are two hydrogen with helix α1 residues Leu441 and Phe442 (V_HCDR3 Ser98 Oγ–O Leu441 E2 and V_HCDR3 Gly100 N–O Phe442 E2), which also contact V_HCDR2 (Supplementary Table 2) (Fig. 4b). In agreement with the cryo-EM structure, a previous mutagenesis study identified Leu441 and Phe442 as key residues for HC84.26.5D binding and HCV neutralization by this broadly neutralizing antibody²⁸. Of note, neutralization of genotype 2b, where Phe442 is mutated to Leu442, was reduced 10-fold compared to other HCV genotypes.

Comparison of HC84.26.5D with other E2 front layer-specific neutralizing antibodies

Human neutralizing antibodies encoded by V_H1-69 and rhesus neutralizing antibodies encoded by V_H1.36, an ortholog of the human V_H1-69 gene⁴⁵, make similar overall footprints on E2 but differ in the position of their CDR loops on the antigen surface, as shown by a comparison of E2 complexes with HC84.26.5D, AR3X, HEPC3, 1382_01_H05, and RM2-01 (Figs. 3b and 4c)^20,23,24,26. The binding mode of HC84.26.5D, in which hydrophobic tip of V_HCDR2 interacts with the E2 front layer near Cys429 and helix α1, most closely resembles the binding modes of human neutralizing antibody 1382_01_H05²⁴ and rhesus neutralizing antibody RM2-01²³ (Fig. 4c). By contrast, V_HCDR2 of AR3X and HEPC3 interacts with a hydrophobic pocket formed by front layer helices α1 and η2 and back layer helix α2 (Fig. 4c)^20,26. This pocket is occupied by V_HCDR3 of HC84.26.5D, whose short length (10 residues) compared to V_HCDR3 of 1382_01_H05 and RM2-01 (17 and 15 residues, respectively) allows an angle of approach that orients V_HCDR3 towards helix η2. In addition, V_HCDR1 of HC84.26.5D interacts with the AS412 epitope and back layer residue Pro612, whereas V_HCDR1 of 1382_01_H05 and RM2-01 contacts Trp529 at the tip of the CD81 binding loop. Interactions with E2 may stabilize V_HCDR1 of HC84.26.5D, which is disordered in the crystal structure of the unbound antibody⁴⁰.

Structural analyses of HCV E2 bound to different broadly neutralizing antibodies have shown that the E2 front layer can adopt at least two distinct conformations, designated A and B²². In the A conformation (e.g., in complexes with AR3X and HEPC3), large sections of the E2 back layer are obscured by the front layer, whereas in the B conformation (e.g., in complexes with 212.1.1 and HC1AM), back layer residues are more accessible for direct interaction with broadly neutralizing antibodies. E2 in complex with HC84.26.5D clearly adopts the A conformation in which front layer helix α1 has undergone a substantial movement relative to its position in the B conformation (Supplementary Fig. 9). The flexibility of the E2 front layer is further underscored by examination of the recent cryo-EM structure of the E1/E2 homodimeric complex⁴⁶, which does not include bound antibodies. As illustrated in Supplementary Fig. 9, residues 426–459 of the E2 front layer adopt a conformation in the E1/E2 homodimeric complex that is distinct from the A and B conformations of E2 in antibody-bound structures²².

Structure of the CBH7–E2 complex identifies a new neutralizing epitope

CBH7 is a neutralizing antibody that targets antigenic domain C and blocks E2 binding to CD81^30,35. To examine recognition of domain C by this antibody, we determined the cryo-EM structure of E2 bound to CBH7 to 3.3 Å resolution (Table 2) (Fig. 5a). Similar to the HC84.26.5D–E2 structure, we were able to model the entire E2 molecule in the CBH7–E2 complex, except hypervariable domain 1 (HVR1). In the CBH4B–E2 complex, antigenic site AS412 (residues Gln412–Arg424), part of the front layer (residues Thr425–Cys452), and most of the CD81 binding loop (residues Gly523–Val536) were disordered. Superposition of E2 in the CBH4B–E2 and HC84.26.5D–E2 structures gave an RMSD of 1.1 Å for 217 α-carbon atoms, while superposition of E2 in the CBH7–E2 and HC84.26.5D–E2 structures gave an RMSD of 1.1 Å for 166 α-carbon atoms. Thus, E2 adopts a similar fold in complexes with all three antibodies. The E2 front layer in the CBH7–E2 complex is in the A conformation, but is disordered in the CBH4B–E2 complex.

**Fig. 5: Structures of CBH7–E2 and CBH4B–E2 complexes.**

CBH7 buries 1192 Å² of solvent-accessible surface on E2 that straddles both neutralizing and non-neutralizing faces, with 13 antibody residues contacting 21 E2 residues. The CBH7–E2 interface exhibits moderate shape complementarity, based on a shape correlation statistic (S_c)⁴⁷ of 0.60 (S_c = 1.0 for interfaces with geometrically perfect fits) compared to 0.54 for the HC84.26.5D–E2 interface. The relatively short V_HCDR3 (13 residues) of CBH7 plays a less dominant role in E2 recognition than the longer V_HCDR3 of most other V_H1-69 HCV antibodies, accounting for 25% (316 Å²) of total Fab buried surface area at the interface with E2 compared to 42% (462 Å²), 49% (497 Å²), 54% (416 Å²), 53% (556 Å²), and 39% (302 Å²) for 1198_05_G10 (20 residues), AR3B (19 residues), AR3C (18 residues), HEPC3 (17 residues), and 212.1.1 (16 residues), respectively. Similar to the HC84.26.5D–E2 complex, V_L of CBH7 contributes significantly to contacts with E2 and accounts for 30% (377 Å²) of total Fab buried surface for CBH7 compared to 27% (332 Å²) for HC84.26.5D. Other V_H1-69 antibodies contact E2 almost exclusively through V_H²⁰.

Structural studies have shown that the CD81 binding site of E2 comprises the AS412 epitope (residues 412–423), the front layer (residues 424–459), the CD81 binding loop (residues 519–535), and residues 616–617 of the back layer⁴⁸. The CD81 binding loop transitions from a retracted to an extended conformation upon CD81–E2 complex formation. All the determined neutralizing antibody–E2 structures, including HC84.26.5D–E2 and CBH7–E2, have the CD81 binding loop in the retracted position^{8,20,21,22,23,24,25,26}. Binding of CBH7 to E2 alters the conformation of the AS412 epitope from that found in the CD81–E2 complex, shifting it towards the CD81 binding loop and creating steric clashes with CD81 (Fig. 5b). Repositioning of the AS412 epitope is driven by hydrophobic contacts between V_HCDR1 Tyr32 and AS412 Trp420 combined with a side-chain–side-chain hydrogen bond between V_HCDR1 Asn31 and AS412 His421 (Supplementary Table 3) (Fig. 5a). Residues 56–59 of V_HCDR2 form a β-strand that runs anti-parallel to β-strand 8 on the back layer of E2 (residues 637–644) in the CBH7–E2 complex. This anti-parallel arrangement is stabilized by four main-chain–main-chain hydrogen bonds (V_HCDR2 Glu57 N–O Arg639 E2, V_HCDR2 Glu57 O–N Arg639 E2, V_HCDR2 Tyr59 N–O Glu637 E2, and V_HCDR2 Tyr59 O–N Glu637 E2) and a salt bridge (V_HCDR2 Glu57 Oε1–Nη1 Arg639 E2) (Supplementary Table 3) (Fig. 5a). V_HCDR2 Phe54, located at the hydrophobic tip of the V_HCDR2 loop characteristic of V_H1-69-encoded neutralizing antibodies, interacts with a highly conserved hydrophobic patch formed by residues Tyr507, Pro513, and Val515 of the E2 β-sandwich core (Fig. 5a). V_HCDR3 Tyr99 mediates additional hydrophobic interactions with this patch, while V_HCDR3 Gly96, Tyr97, and Ile98 contact AS412 Trp420 and His421.

The epitope recognized by CBH7 is distinct from the epitopes recognized by HC84.26.5D and CBH4B, as illustrated in Fig. 6. It is also distinct from the epitopes described in previous structural studies of antibody–E2^{8,9,20,21,22,23,24,25,26} or antibody–E1E2^49,50 complexes. The CBH7–E2 structure, therefore, defines a new neutralizing epitope on E2. Figure 6 shows the footprint on E2 of CBH7 and its relation to the footprints of HC84.26.5D and CBH4B. Whereas HC84.26.5D targets the E2 front layer (Fig. 3a) and CBH4B targets the back layer (Fig. 5c), CBH7 straddles both front and back layers, such that the CBH7 epitope is clearly distinct from the HC84.26.D and CBH4B epitopes, despite some overlap. HC84.26.5D is representative of neutralizing antibodies that target the E2 front layer, as is evident by comparing the footprints on E2 of HC84.26.5D, AR3X, HEPC3, RM2-01, and 1382_01_H05 (Fig. 3b). CBH4B is representative of non-neutralizing antibodies, such as 2A12⁹, that target the E2 back layer (Fig. 7a) (see below).

**Fig. 6: Comparison of HC84.26.5D, CBH7, and CBH4B epitopes on E2.**

**Fig. 7: Comparison of non-neutralizing antibodies targeting E2.**

To better understand the newly defined CBH7 epitope, we used molecular dynamics (MD) simulations of glycosylated unbound E2 to assess the effects of selected E2 variants on epitope accessibility and mobility. We calculated epitope accessible surface area (ASA) and mobility (root-mean-square fluctuation; RMSF) for E2 with deleted HVR1 (ΔHVR1), E2 with deleted N-glycans 1, 4, and 6 (ΔN1/N4/N6), and a triple mutant representing E2 polymorphisms (I538V/Q546L/T563V), in comparison with wild-type 1b09 E2 (Supplementary Table 4). Each of the E2 variants has been previously shown to affect broadly neutralizing antibody neutralization, leading to increased antibody sensitivity (ΔHVR1, ΔN1/N4/N6)^51,52 or increased antibody resistance (I538V/Q546L/T563V)⁵³, although effects varied depending on the HCV isolate context and antibody. Average epitope surface accessibility increased for the E2 variants in comparison with the wild-type 1b09 for both probe sizes, with probes reflecting a water molecule (1.4 Å radius) or antibody CDR loops for a heavy or light chain V domain (10 Å radius). The largest epitope accessibility increases were observed for the E2 polymorphism triple mutant for both probe sizes, which increased the average accessible surface area by ~10–15%. Predicted epitope mobility was affected most for the ΔHVR1 mutant, which resulted in a ~30% reduction in average epitope residue mobility (1.39 Å RMSF, versus 1.99 Å average RMSF for wild-type). While these predicted epitope effects are intriguing, experimental studies to assess CBH7 binding and neutralization effects of these variants would be needed, as CBH7 was not tested with those variants previously^51,52,53, and possible comparative MD simulations with longer time scales (to better predict larger-scale E2 motions) and full E1E2 context would additionally be useful.

Comparison of the crystal structure of unbound Fab CBH7 with the cryo-EM structure of Fab CBH7 bound to E2 showed small yet relevant differences in CDR loop conformation (Supplementary Fig. 10a). The RMSD for the superposition of V domains, based on 217 Cα atoms, is 0.8 Å, indicating that the overall topology of the bound and unbound states is conserved, with the exception of minor rotameric differences in the CDR loops. For example, upon complex formation, the side chains of V_LCDR3 Phe94 and Leu95 reorient towards a hydrophobic groove on E2 formed by turns 510–513 and 544–546 of the central β-sandwich and the back layer 3¹⁰ helix (residues 633–635). Notably, the inherent flexibility of the elbow region connecting the V and C domains permits changes in the rotational angles of the switch peptides between E2-bound and unbound CBH7, as calculated using DynDom⁴¹ (Supplementary Fig. 10b). In E2-bound CBH7, the rotational angles of the switch peptides around the H chain (residues 112–116) and the L chain (residues 105–106) are altered by 27° and 32°, respectively, relative to unbound CBH7 (Supplementary Fig. 10b). These rotations facilitate the movement of the V domains toward the antigen, resulting in elbow angle shifts from 133° to 166°, while maintaining the local conformations of the CDRs (Supplementary Fig. 10a, c, d).

Structure of the CBH4B–E2 complex

CBH4B is a non-neutralizing antibody targeting antigenic domain A of E2 that does not block E2 binding to CD81^{30,31,32,33,34}. To examine recognition of domain A by this antibody, we determined the cryo-EM structure of E2 bound to CBH4B to 3.7 Å resolution (Table 3). CBH4B binds to the non-neutralizing face of E2 at a location distant from the CD81 binding site comprising the back layer (residues 597–646) and post-variable 3 region (pVR3; residues 581–596) (Fig. 5c). The CBH4B–E2 structure agrees with previous epitope mapping³⁴. Surprisingly, antigenic site AS412 and the front layer, which are well defined in the CBH7–E2 and HC84.26.5D–E2 complexes, are partly disordered in the CBH4B–E2 complex. It may be that CBH4B binding to the back layer of E2 destabilizes the front layer. Alternatively, the front layer may be intrinsically flexible and require binding of antibodies like CBH7 or HC84.26.5D that target the neutralizing face of E2 for stabilization. Indeed, all the determined antibody–E2^{8,20,21,22,23,24,25,26} and antibody–E1E2^49,50 structures in which the front layer and AS412 epitope are clearly defined involve antibodies bound to the neutralizing face.

The CBH4B–E2 interface exhibits moderate shape complementarity (S_c = 0.58)⁴⁷, similar to the CBH7–E2 and HC84.26.5D–E2 interfaces (0.60 and 0.54, respectively). CBH4B buries 813 Å² of solvent-accessible surface on E2, with V_L contributing only 120 Å² (15%). V_HCDR2, which contacts pVR3, plays a dominant role in E2 recognition, accounting for 51% (416 Å²) of total Fab buried surface with E2 compared to only 30% (240 Å²) for the relatively long V_HCDR3 (16 residues), which contacts the back layer. CBH4B forms a dense network of five hydrogen bonds with back layer residues Arg630, Tyr632, and Arg639: V_HCDR3 Asp97 Oδ2–Nη1 Arg630 E2, V_HCDR3 Asp97 Oδ2–Nη2 Arg630 E2, V_HCDR3 Val100 N–O Arg630 E2, V_HCDR3 Val100 O–N Tyr632 E2, and V_HCDR1 Tyr32 Oη–Nη1 Arg639 E2 (Supplementary Table 5) (Fig. 5c). CBH4B contacts β-strand 7 on the back layer of E2 (residues 625–632), whereas CBH7 contacts β-strand 8 (residues 637–644). The β-hairpin formed by these two strands separates antigenic domain A (CBH4B) from antigenic domain C (CBH7) (Fig. 5d). In agreement with the CBH4B–E2 and CBH7–E2 structures, CBH4B does not interfere with HCV neutralization by CBH7³⁵. V_HCDR2 Phe54, located at the tip of the V_HCDR2 loop, interacts with a hydrophobic groove on the back layer of E2 formed by Leu603, Cys607, Val629, Ala642, and Cys644 (Fig. 5c). Collectively, these residues are only moderately conserved across HCV genotypes: 89% for Leu603, 100% for Cys607, 70% for Ile626, 56% for Phe627, 74% for Cys629, and 54% for Leu640.

Previous studies have used non-neutralizing antibodies directed against the central β-sandwich (HEPC46 and E1) or back layer (2A12) of E2 to facilitate crystallization of E2 bound to front layer-specific neutralizing antibodies^8,9,20,22. The H chains of HEPC46 and 2A12 are encoded by the human IGHV1-18 and mouse IGHV14-3 genes, respectively, while the H chain of E1, like the H chains of CBH4B and CBH7, is encoded by the human V_H1-69 gene. Although both CBH4B and 2A12 bind to the back layer, their different angles of approach result in different sets of interactions, such that CBH4B contacts pVR3 as well as the back layer, whereas 2A12 contacts only the back layer (Fig. 7a–c). Likewise, CBH7, E1, and HEPC46 bind distinct, but overlapping, epitopes on E2. E1 engages the CD81 binding loop via V_HCDR1 and β-sandwich turn 542–548 via V_HCDR3²¹. This β-sandwich turn is anchored by all three V_HCDR loops of HEPC46 (Fig. 7d). Interestingly, the tips of the V_HCDR2 loops of E1 (Val54) and HEPC46 (Tyr53) interact with the same hydrophobic patch on E2 formed by residues 513–516 of the central β-sandwich as the tip of the V_HCDR3 loop (Tyr99) of CBH7 (Fig. 7d).

Surprisingly, superposition of the CBH4B–E2 structure onto the existing E1E2 structures^49,50 revealed a partial overlap of CBH4B with E1 (Supplementary Fig. 11a), even though this antibody binds HCV virions^{30,31,32,33,34}. Moreover, CBH4B bound two different recombinant forms of E1E2, one soluble and the other membrane-bound, albeit with 240-fold lower affinity compared to E2 alone for soluble E1E2 (K_d = 19 nM versus 0.08 nM) and 40-fold lower affinity compared to E2 alone for membrane-bound E1E2 (K_d = 3.1 nM versus 0.08 nM) (Supplementary Fig. 11b). Such affinity reductions were not observed for other E2-specific antibodies tested, including HC84.26.5D. One interpretation of these results is that binding of CBH4B to E1E2 requires a significant rearrangement of E1 and E2 subunits from the arrangement seen in current E1E2 structures^46,49,50, and that this rearrangement is associated with an energetic cost that lowers affinity. If so, this would imply that the E1E2 heterodimer can adopt multiple conformations, as demonstrated for E2 alone²².

To further examine E2 recognition by CBH4B, we calculated the mean Shannon sequence entropy across residues of the CBH4B epitope on E2, weighted by residue buried surface area, and compared it against the respective values for HC84.26.5D and CBH7. This analysis revealed that while all three epitopes have high residue conservation (>80% weighted mean), the CBH4B epitope is slightly more conserved than the HC84.26.5D and CBH7 epitopes, with a weighted mean sequence entropy of 13 and a weighted mean residue conservation of 86%. The HC84.26.5D epitope yielded weighted mean sequence entropy and conservation values of 20 and 82%, respectively, while for the CBH7 epitope these values were 17 and 84%, respectively. For comparison, the HEPC46 epitope is considerably more variable than the CBH4B, HC84.26.5D, or CBH7 epitopes, with mean sequence entropy and conservation values of 51 and 55%, respectively. As exemplified by CBH4B, targeting a conserved epitope is necessary but not sufficient for broad neutralization; non-neutralizing antibodies have been observed targeting conserved sites on other viruses, such as HIV gp41 and the SARS-CoV-2 fusion peptide⁵⁴, and such antibodies can play a role in viral control outside of direct neutralization.

AlphaFold predictions of antibody–E2 complexes

We used the AlphaFold v2.3 deep learning modeling tool to predict the antigen-binding mode for antibodies HC84.26.5D, CBH7, and CBH4B, and compared the models to the experimentally determined cryo-EM structures. The AlphaFold models of the HC84.26.5D–E2 (Supplementary Fig. 12), CBH7–E2 (Supplementary Fig. 13), and CBH4B–E2 (Supplementary Fig. 14) complexes superimposed very well to the respective cryo-EM structures, and the AlphaFold predicted antigenic site is nearly identical. The models have relatively high AlphaFold confidence scores, 0.88, 0.91, and 0.85, for HC84.26.5D–E2, CBH7–E2, and CBH4B–E2, respectively, which is indicative of AlphaFold antibody–antigen modeling success based on previous benchmarking⁵⁵, and all three modeled complexes have medium accuracy based on CAPRI criteria⁵⁶ and ligand RMSDs of <4.0 Å (Supplementary Figs. 12–14).

Superposition of E2 from the AlphaFold models and cryo-EM for the HC84.26.5D–E2, CBH7–E2, and CBH4B–E2 complexes gives RMSDs of 1.12 Å, 1.41 Å, and 1.24 Å for 215, 202, and 169 Cα atoms, respectively. The predicted AlphaFold models for E2 in all the three complexes (Supplementary Figs. 12b, 13b, and 14b) recapitulate the α/β folding and overall architecture of the cryo-EM structures. However, structural alignments between cryo-EM and AlphaFold models of the CBH7–E2 complex revealed significant differences in the conformation of the E2 front layer (Thr425–Cys459) (Supplementary Fig. 13e). The predicted structure of the front layer in CBH7–E2 adopts a unique conformation of peptide downstream of helix α1 (Phe442–His445), where there is a turn at residues His434–Gly436 toward helix α2 (Arg614–His617), while in the cryo-EM structure the front layer is extended toward the CD81 binding loop (Thr519–Val536). This conformational variation in the front layer may be attributed to HC84.26.5D bound to the front layer of E2 in the cryo-EM HC84.26.5D–E2–CBH7 ternary complex structure, and no HC84.26.5D present in the CBH7–E2 complex AlphaFold model. In the HC84.26.5D–E2 and CBH7–E2 complexes, AlphaFold models have AS412 (Gln412–Asn423) in a β-hairpin conformation turned at residues Thr416–Ser419, as observed in E2_412–424 peptide bound to HCV1⁵⁷, AP33⁵⁸, 19B3, or 22D11⁵⁹ (Supplementary Figs. 12c and 13e). In contrast, AS412 in the cryo-EM structures adopts extended coils with partially open hairpin conformations, containing a turn (HC84.26.5D–E2) that is also observed in E2_412–424 bound with HC33.1⁶⁰ and 3/11⁶¹. In the cryo-EM HC84.26.5D–E2 structure, the N-terminal region of AS412 (Gln412–Asn415) forms a turn and moves toward the C-terminus of a proximal β-sandwich (Val516–Thr518). Superposition of V_L/V_H domains for the HC84.26.5D–E2, CBH7–E2, and CBH4B–E2 complexes gives RMSDs of 0.86 Å, 0.73 Å, and 1.00 Å for 199, 212, and 217 Cα atoms, respectively. Thus, the AlphaFold model for V_L/V_H is similar to the cryo-EM structures, and residues in the predicted structured region have high confidence scores (average pLDDT scores of V_L/V_H regions are above 92) that correlate with model accuracy. Buried surface area and antigen–antibody interaction analysis predict models of CDR loops of comparable quality to the cryo-EM structures, except the V_HCDR1 loop (pLDDT = 85.6) in the HC84.26.5D–E2 AlphaFold model (Supplementary Fig. 12e), where the tip of V_HCDR1 moves toward the β-hairpin turn (Thr416–Ser419). The extended conformation of AS412 in cryo-EM exposed the back layer, allowing V_HHis29 to interact with Pro612 of the back layer and π–π stacking interaction with His412.