Global mapping of antibody recognition of the hepatitis C virus E2 glycoprotein: Implications for vaccine design

Significance

Hepatitis C virus is a major public health concern, infecting approximately 3% of the world’s population, with no vaccine currently available. To enable rational vaccine design for this highly diverse and dynamic virus, we performed alanine scanning of nearly all positions of the E2 envelope protein, which is the primary target of the antibody response, using a panel of 16 human monoclonal antibodies that target a broad range of epitopes. This approach provided an unprecedented global view of the determinants of E2 stability, residue connectivity, and neutralizing antibody recognition. These insights and mapping data provide a framework to engineer E2 to modulate antibody recognition and optimize its capacity to induce broadly neutralizing antibodies in the context of a vaccine.

Abstract

The E2 envelope glycoprotein is the primary target of human neutralizing antibody response against hepatitis C virus (HCV), and is thus a major focus of vaccine and immunotherapeutics efforts. There is emerging evidence that E2 is a highly complex, dynamic protein with residues across the protein that are modulating antibody recognition, local and global E2 stability, and viral escape. To comprehensively map these determinants, we performed global E2 alanine scanning with a panel of 16 human monoclonal antibodies (hmAbs), resulting in an unprecedented dataset of the effects of individual alanine substitutions across the E2 protein (355 positions) on antibody recognition. Analysis of shared energetic effects across the antibody panel identified networks of E2 residues involved in antibody recognition and local and global E2 stability, as well as predicted contacts between residues across the entire E2 protein. Further analysis of antibody binding hotspot residues defined groups of residues essential for E2 conformation and recognition for all 14 conformationally dependent E2 antibodies and subsets thereof, as well as residues that enhance antibody recognition when mutated to alanine, providing a potential route to engineer E2 vaccine immunogens. By incorporating E2 sequence variability, we found a number of E2 polymorphic sites that are responsible for loss of neutralizing antibody binding. These data and analyses provide fundamental insights into antibody recognition of E2, highlighting the dynamic and complex nature of this viral envelope glycoprotein, and can serve as a reference for development and rational design of E2-targeting vaccines and immunotherapeutics.

Hepatitis C virus (HCV) infects ∼185 million of the world’s population, with 3–4 million new infections each year. Infection often leads to chronic hepatitis, cirrhosis, and hepatocellular carcinoma, and is a leading reason for liver transplantation (1). Despite recently developed direct-acting antiviral agents, there is a major need for a preventive HCV vaccine, because of the high cost of treatment therapies—which limit their clinical use—a high rate of asymptomatic and untreated infected individuals (over 95% of the infected population) (2, 3), concern of viral resistance to direct-acting antiviral agents (4), and that treatment-induced cure in patients with established cirrhosis does not eliminate the risk of hepatocellular carcinoma (5).

A major obstacle to HCV vaccine development efforts is the extreme diversity of the virus and its high rate of mutation, which allows it to actively evade the immune response in infected individuals. Critical to the development of an effective vaccine is the identification and characterization of conserved epitopes associated with viral neutralization. The antibody response to HCV is directed primarily against the E2 glycoprotein because E2 directly interacts with the HCV coreceptors, scavenger receptor class B type 1 (SR-B1) (6) and the tetraspanin CD81 (7), during viral entry. There is recent evidence that the E1E2 heterodimer, not E2 alone, interacts with a third coreceptor, the tight junction protein Claudin-1 (8). A number of antigenic sites have been identified over the past two decades, as summarized in a recent review (9). Varying nomenclature has been used to describe these sites [epitope I–III (10), antigenic region 1–5 (11), and antigenic domain A–E (12)], but they are largely if not wholly overlapping (e.g., epitope I and antigenic domain E both correspond to E2 residues 412–423). These have been defined in previous work by alanine scanning mutagenesis of limited sets of E2 residues, or binding of short E2 peptides, with monoclonal antibodies (mAbs) (13–16). In addition to these sites, E2 contains several hypervariable regions (HVRs) with high sequence diversity; HVR1 in particular appears to serve as a decoy epitope that elicits strain-specific antibodies (17).

Current knowledge of the E2 3D structure is primarily through two independently determined crystal structures of engineered truncations of the E2 core domain (18, 19), which comprises a central immunoglobulin β-sandwich fold flanked by two additional protein layers. These crystal structures contain numerous regions with no apparent regular secondary structure, with ∼60% of all E2 residues either disordered or in loops. Discrepancies between these two structures, including their disulfide bonding patterns (20), suggest that further work is necessary to delineate the native, immunologically relevant E2 structure. Structural and neutralization studies provide additional evidence that E2 is a highly dynamic protein with conformationally variable epitopes (21, 22) and allosteric sites, where mutations at residues distant from antibody binding sites impact E2 recognition and viral neutralization (23).

To provide a comprehensive view of the HCV E2 glycoprotein and determinants of E2 antibody recognition, we performed global E2 alanine scanning mutagenesis with a panel of 16 human mAbs (hmAbs) derived from HCV-infected individuals that target five distinct E2 regions (antigenic domains A–E). We used unsupervised learning methods to identify residue-level energetic signatures underlying the E2 recognition of this antibody panel. This approach permitted us to group these antibodies as well as the full range of E2 residues, which revealed interconnected networks of E2 residues that are in many cases distant in sequence and 3D structure, and critical for E2 stability and antibody binding. Additionally, we found that some E2 residues enhanced antibody binding for sets of neutralizing antibodies when mutated to alanine, which given their location away from antibody binding sites, provides further evidence of E2 allostery and global dynamic effects. By incorporating residue polymorphism, we observed previously described E2 antibody escape variants and explored global E2 antibody targeting and adaptability. This unprecedented dataset and analysis can serve as a reference for future studies of antibody recognition of HCV, rationally designed HCV vaccines and immunotherapeutics, and antibody recognition of viral antigens in general.

Results

Clustering of E2 Residues and Antibodies.

Global alanine scanning of E2 was performed with 16 distinct E2-binding hmAbs (Table 1), using site-directed mutagenesis of the E1E2 coding sequence and ELISA to measure antibody binding. Individual mutants were produced at 355 E2 positions, resulting in a total of 5,583 binding data points (SI Appendix, Fig. S1 and Table S1). The antibodies in our panel engage a variety of sites on the E2 glycoprotein (grouped into antigenic domains A–E) (12, 24, 25), and represent a range of HCV neutralization as well as inhibition of CD81 binding. The panel includes two antibodies (A27 and CBH-23) that have not been previously described and target antigenic domains B and C; these were cloned from HCV-infected individuals as were the other hmAbs in the panel (26, 27).

Table 1.

E2 human monoclonal antibody panel

Antibody	Antigenic domain	Neutralizing	Bind denatured E2	Source
CBH-4D	A	N	N	(26)
CBH-4G	A	N	N	(26)
CBH-4B	A	N	N	(26)
CBH-20	A	N	N	(56)
CBH-21	A	N	N	(56)
CBH-22	A	N	N	(56)
HC-1	B	Y	N	(27)
HC-11	B	Y	N	(27)
A27*	B	Y	N	—
CBH-7	C	Y	N	(26)
CBH-23*	C	Y	N	—
HC84.20	D	Y	N	(25)
HC84.24	D	Y	N	(25)
HC84.26	D	Y	N	(25)
HC33.1	E	Y	Y	(12)
HC33.4	E	Y	Y	(12)

^*A27 and CBH-23 hmAbs not previously described; neutralization and denatured E2 binding measured in the same manner as other hmAbs in the panel (25).

To probe signatures underlying global E2 antibody recognition, we used hierarchical clustering analysis to group similar E2 positions and antibodies (Fig. 1). The resultant clusters of E2 residues (Fig. 1A, Table 2, and SI Appendix, Table S2) are largely proximal residues in E2 sequence and structure (Fig. 2), grouping previously mapped epitope residues (clusters 3, 6, and 9), and delineating networks of energetically related residues from the standpoint of antibody recognition. Investigation of individual clusters revealed residues that are critical for E2 structure which lead to loss of binding for all conformation-specific hmAbs (clusters 4 and 8), and residues that result in loss of binding only for hmAbs corresponding to certain antigenic domains (clusters 1, 2, 3, and 9).

Fig. 1.

Clustering analysis of global alanine scanning data. (A) Clustering of E2 positions (y axis) according to binding profile to panel of 20 human hmAbs (x axis). Twenty clusters of E2 positions obtained by hierarchical clustering are indicated by the colored bar on the left. Cells are colored according to percent of mutant E2 binding with respect to wild-type E2: 0–20% (red), 21–40% (orange), 41–60% (yellow), 61–90% (white), 91–150% (green), and >150% (blue). Antibody names are colored according to previously determined antigenic domains, with A, B, C, D, and E colored red, magenta, cyan, green, and blue, respectively. (B) Clustering of antibodies based on binding data, with antibody names colored according to antigenic domains as in A. Antibody clusters are outlined with dotted lines and labeled with cluster P values; P values of antigenic domain B and D subclusters are 0.94 and 0.99, respectively.

Table 2.

E2 residue clusters 1–12, and average alanine mutant percent binding within antigenic domains (and average for all antigenic domains) for each cluster

Cells with average (Avg) percent binding are colored as in Fig. 1. Epitope residues for antigenic domains A–E noted in previous studies (12, 25, 39, 57) are in bold. Clusters ordered according to cluster size; clusters 13–16, each containing 30 or more residues, are omitted here for brevity and provided in SI Appendix, Table S2. Cluster numbers in bold have bootstrapping P value ≥ 95%.

Fig. 2.

Alanine scanning-based residue clusters on the E2 core structure. Front (A and C) and side (B and D) views of E2 are shown (light blue cartoon in A and B, surface in C and D), with bound AR3C hmAb (tan) shown for reference. Residues are colored according to cluster, with clusters 1 (blue), 2 (red), 4 (purple), 5 (orange), 6 (green), 7 (tan), 8 (yellow), 9 (magenta), 10 (cyan), 11 (dark green), and 12 (slate) shown as spheres. Larger clusters comprising the remainder of E2 (clusters 13–16) are omitted for clarity, and cluster 3 is not shown because its residues (L413, G418, W420) are not resolved in the E2 core structure. Previously identified epitope residues for antibodies in antigenic domains A (Y632), B (T425, L427, G530, D535), C (W549), and D (L441, F442, Y443, W616) (16, 25, 39) are labeled in A, and locations of antigenic domains A–E are shown in C. Antigenic domain E is shaded in gray in its approximate location, based on the location of residues 421–423, which are present in the structure. (E) Putative CD81 binding residues [< 20% alanine mutant CD81 binding, compared with wild-type E2, in a recent study (16)] are shown as spheres, colored according to cluster as in A–D, with residues from remaining clusters (clusters 13–16) colored gray. Prominent CD81 binding residues are labeled.

Clustering of antibodies based on global E2 binding data (Fig. 1B) also confirmed previous epitope mapping and competition studies (12, 24, 25) by separating antibodies into distinct clusters corresponding to antigenic domains A–E. As with previous antibody clustering based on HCV neutralization with a panel of genotype 1 viruses (23), antigenic domain B and D hmAbs were coclustered, as expected given their overlapping binding sites on the E2 surface. In contrast with that study, hmAbs HC-1 (antigenic domain B) and CBH-7 (antigenic domain C) were not coclustered, reflecting likely distinct binding sites on E2. This finding is corroborated by previous hierarchical clustering analysis using E2 cross-competition binding data (27).

Prediction of Residue Contacts in E2.

Given the coclustering of residues that in many cases are proximal in the 3D structure of E2 core, we tested the possibility of our immunological mapping data being used to predict pairwise residue contacts within native E2, analogous to other efforts using sequence coevolution between residues to predict contacts in protein structures (28, 29). Residue contact predictions were produced using two distance metrics and compared with the contacts observed in the crystal structure of E2 core (Fig. 3). A number of experimentally observed E2 residue contacts between residues that are nonadjacent or distant in sequence were predicted by the correlation-based distance measure. These contacts include the predicted contact between L413 and W420, which is confirmed by structures of E2 412–423 mAb complexes (30–32), as well as residue pairs T425-G530, N428-G530, and L441-W616, which are contacts in the E2 core crystal structure (18) (circled in Fig. 3). Many contacting sites within the central Ig β-sandwich of E2 (residues 492–566) were also predicted by this analysis (shown close-up in SI Appendix, Fig. S2). Other predictions involve regions not present in E2 crystal structures, including contacts within HVR1, HVR2, and the C-terminal portion of E2, as well as contacts between the latter two.

Fig. 3.

Prediction of E2 residue contacts based on alanine scanning data. Pairwise residue contacts were selected by correlation (red circles, Upper Right) or Euclidean distance (green circles, Lower Left) between positions, based on hmAb binding data. Residue contacts observed in the E2 core structure (pairs of residues within a 5 Å distance cutoff) are shown as black circles, with regions present in the E2 core structure shown in gray. Circles indicate several predicted contacts between nonadjacent residues observed in E2 crystal structures (18, 30–32): L413-W420, T425-G530, N428-G530, and L441-W616.

E2 Hotspots and Required Folding Residues.

To further investigate the E2 mutants that led to loss of binding for multiple conformation-dependent antigenic domains (A–D), we analyzed the sets of hotspot residues for each antigenic domain, defined as residues that when mutated to alanine or glycine led to 20% or lower binding compared with wild-type E2 for all hmAbs. We visualized the sets of antigenic domain-specific and shared hotspot residues using a recently described method (33) (Fig. 4). This approach revealed that domain A hmAbs are sensitive to mutants at far more E2 positions versus hmAbs targeting the other antigenic domains, and that a large number of hotspot residues (11 in total, 8 of which comprise residue cluster 8) are shared by all conformation-dependent E2 antibodies. Investigation of individual mutants comprising these sets (SI Appendix, Table S3) confirms that the vast majority of these residues (32 of 36) are buried within the E2 core structure, thus are likely to impact E2 folding or local stability versus direct antibody contact.

Fig. 4.

Shared and unique hotspot residues among E2 antigenic domains. For each antigenic domain, hotspot residues common to all antibodies were identified, and shared and domain-specific sets were computed. Hotspot residues are defined here as E2 positions that, when mutated to alanine or glycine, reduce mAb binding level to 20% or less compared with wild-type E2.

Given that disulfide bonds and glycans can impact protein folding and stability, we investigated in detail the effects of cysteine (SI Appendix, Fig. S3) and glycan sequon (SI Appendix, Fig. S4) mutants on binding of our hmAb panel. Mutagenesis of cysteine residues indicates that many are involved in disulfide bonds essential for E2 folding; however, some have a local rather than global impact on E2 structure. For example, alanine substitutions at residues C429 and C503, which form a disulfide bond in the E2 core crystal structure (18), result in largely unaffected antigenic domain C hmAbs binding, versus antigenic domains B and D hmAbs, which lose binding. Glycan sequon mutants showed that most effects of antibody binding were observed with loss of N7 (N540) and N10 (N623) glycans; however, binding for some hmAbs was also disrupted by N8 (N556) and N9 (N576) glycan sequon mutants. Given that N7 and N10 are distal from the antigenic domain B–D supersite in the E2 core structure (18), as well as their broad effects on the panel, it is likely that these glycans impact hmAb binding indirectly via E2 folding or stability; this finding is supported by a previous study where N7, N8, and N10 mutants significantly reduced formation of E1E2 complexes (34).

To further explore the impact of alanine mutants on E2 stability versus direct antibody interaction, we used the Rosetta modeling program (35) to perform in silico alanine (or glycine, in the case of alanine residues) mutagenesis of all 171 E2 residues available in the core crystal structure. The top 10 predicted destabilizing mutants (Table 3) included the four alanine mutants with lowest average domain A–D percent binding (excluding cysteine residues in disulfide bonds; all <1% average A–D binding), as well as F537A, which is at a buried hydrophobic site that also greatly impacted binding of the hmAb panel (5.1% average domain A–D binding). Interestingly, several destabilizing E2 mutants predicted by Rosetta were not found to disrupt recognition of the antibody panel overall, in particular W437A and W616A, which are contiguous and proximal to the bound AR3C hmAb in the E2 core crystal structure (18). It is not clear whether this is because of impact on local rather than global E2 stability, the specific conformation of the AR3C-bound E2 structure rather than dynamic native E2, limitations in modeling parameters, or combinations thereof; others have suggested that this is a conformationally variable region in the context of antibody recognition (21).

Table 3.

Top predicted destabilizing mutants based on the E2 core structure

Mutant	Score*	Average A-D†
Y611A	6.4	0.8
W616A	4.8	84.2
R614A	4.7	0.9
F550A	4.6	25.8
F509A	4.3	0.7
Y507A	3.8	14.1
W554A	3.8	0.4
W549A	3.8	37.8
W437A	3.8	73.0
F537A	3.5	5.2

Mutants in bold denote measured destabilizing mutants based on reactivity to conformation-dependent antibodies for antigenic domains A-D. Mutants in italics exhibit major disparity between antibody-inferred E2 stability and impact on stability predicted from the E2 core structure (18).

^*Rosetta score (35) for predicted energetic impact of mutant on E2.

^†Average percent binding versus wild-type E2 for conformation-dependent E2 antibodies.

E2 Residue Polymorphism and Antibody Escape.

Noting the capacity for HCV to readily escape from antibody targeting, at times using distal allosteric mutants (23), we searched our global mapping data for hotspots for hmAbs in our panel with the capacity to mutate. Using a multiple sequence alignment of 627 E2 sequences from the Los Alamos HCV Database (36), we computed sequence variability at all E2 positions, and compared with maximum binding impact across the neutralizing hmAbs in our panel (Fig. 5), highlighting E2 positions found in previous studies to mutate under antibody selective pressure or mediate neutralization escape (23, 37–40).

Fig. 5.

Variability at E2 neutralizing antibody binding residues. Shannon entropy, calculated from an alignment of E2 sequences (y axis) is compared with minimum percent binding, from alanine scanning of neutralizing hmAbs in our panel (x axis), for each position of E2. A dotted vertical line at 30% binding is shown for reference. Points corresponding to E2 positions previously associated with antibody escape or mutation under immune pressure (23, 37–40) are colored red, and nonsurface residues in the E2 core structure (18) (<20% side-chain solvent accessibility) are shown as squares. Outlier points critical for antibody binding, with relatively high sequence variability, are labeled by residue.

Previously identified residues L438, F442, K446, and A531 were among the more polymorphic positions associated with loss of hmAb binding in our analysis. Investigation of the sequence variability of residues 434–446 (SI Appendix, Fig. S5) confirmed that a variety of amino acids have been observed at positions 438 and 446, with limited variability at position 442. Additionally, our analysis identified K408, which is a hotspot residue for the HC33.4 hmAb and is also highly polymorphic (as is expected given its location in HVR1). Although a study reporting possible HCV resistance mutants to hmAb HC33.1 has been reported (41), an analogous study with HC33.4, to determine whether K408 mutants can mediate escape, has not been reported. The recently described crystal structure of HC33.4 in complex with its peptide epitope (42) revealed a similar epitope backbone conformation to the HC33.1-bound epitope (22), although because coordinates of E2 residue 408 were not resolved in the context of HC33.4, it is unclear whether a direct hmAb interaction is made with this residue. Several additional E2 residues, including buried residues in the E2 core structure (squares in Fig. 5), exhibit the capacity to mutate and disrupt antibody binding. Although providing a view of variability and antibody escape, this analysis omits several potentially relevant features, including nonalanine mutants and combinations of mutants. Furthermore, many positions have relatively small variability in the set of all E2 sequences, yet under immune pressure will mutate to evade antibodies, such as N415 and N417 (Shannon entropies 0.22 and 0.19, respectively), which mutated in vivo during clinical trials of a therapeutic monoclonal antibody (43).

Viral Fitness and Entry Receptor Binding Residues.

A subset of functionally critical E2 residues is responsible for binding the human CD81 protein, which is required for HCV infection. Thus, mutation of these sites has been shown to reduce viral entry and impact viral fitness (44). Comparing recently reported E2 alanine scanning data for CD81 binding (16) with data from our antibody panel (SI Appendix, Fig. S6; critical CD81 binding residues shown on E2 structure in Fig. 2E) shows that some neutralizing antibodies, such as HC-1 and A27, have significant overlap of key binding residues with this receptor. However, the HC33.1 hmAb exhibits little if any potential for antibody escape (41) and only depends on residue W420 in this set. However, as E2 mutagenesis studies for CD81 binding have only been performed on limited sets of E2 residues to date (16, 18, 44, 45) (several of these are represented in SI Appendix, Table S1), and there is no CD81-E2 complex crystal structure, it is possible that additional E2 residues may be involved in CD81 recognition.

To further explore the impact on viral fitness and CD81 binding of the alanine mutants in this panel, we generated and assessed infectivity for a large set of HCV pseudoparticles (HCVpp) representing 73 alanine mutants at selected residues proximal to the putative CD81 binding site on E2 (Fig. 6), mapping a larger region than previously reported in this regard (44). The majority of tested mutants (43 of 73) maintained at least some infectivity, including all mutants with point substitutions in HVR1 or residues 430–435. We tested the sensitivities of all HCVpp over an infectivity threshold [5 × 10³ relative light units (RLU)] using anti-CD81 mAb (SI Appendix, Table S4). This approach yielded nine mutants with a significant (approximately twofold) decrease in EC₅₀, corresponding to increased anti-CD81 sensitivity. One of these mutants (V515A) was previously reported, likewise using anti-CD81 mAb, to have approximately twofold increased sensitivity (46), and as noted in that study the increased sensitivities observed here may be because of lower CD81 affinity. Additionally, one mutant (T385A) exhibited decreased anti-CD81 neutralization sensitivity, which also suggests direct or indirect involvement in CD81 engagement.

Fig. 6.

Infectivity of E2 alanine mutants using HCVpp. Wild-type H77 and 73 E2 mutants were tested, and mean ± SD from three experiments performed in triplicate are shown. The detection limit for positive luciferase reporter protein expression (dotted line) was 3 × 10³ RLU, corresponding to the mean ± 3 SD of background levels (i.e., luciferase activity of naive uninfected cells) (55). *P < 0.001 from Mann–Whitney test.

We selected two mutants, I411A and S432A, based on loss of infectivity and altered anti-CD81 sensitivity, respectively, for direct measurement of CD81 binding as they were not previously characterized in this regard. Additionally, we omitted mutants with broad binding effects on the antibody panel, which suggest general E2 destabilization rather than direct CD81 interaction, for example Y507A in cluster 12 (Table 2). Both tested mutants had moderate effects on CD81 binding: 56% and 32% of wild-type binding levels for I411A and S432A, respectively (SI Appendix, Fig. S6), versus 0% binding for D535A, which was tested as a control. In the case of I411A, given the pronounced effect of viral viability, it is likely to have additional effects than CD81 binding, for example SR-B1 engagement. Collectively, these results provide an expanded view of residue-level viral fitness and entry receptor binding effects that complements the detailed and comprehensive assessment of hmAb binding.

Affinity-Improving Residues.

Noting that some E2 mutants appeared to improve antibody affinities from alanine scanning, we systematically identified mutants that increased affinity for sets of neutralizing hmAbs in our panel. The top three mutants, according to average percent binding for each neutralizing antigenic domain (B–E) were selected for detailed investigation (Fig. 7).

Fig. 7.

E2 alanine mutants with improved binding to neutralizing hmAbs. For each antigenic domain, the top three mutants with improved average hmAb binding are shown. Percent binding for hmAb panel and antigenic domain averages are shown; average numbers in bold denote mutant in top three for that antigenic domain. C-terminal E2 residues (after residue 661) are not shown. Cells are colored according to percent binding as in Fig. 1.

We identified a number of positions, localized to specific regions of E2, which improved average binding from 1.2-fold to over 3-fold. Two of the top three mutants for antigenic domain D (L433A, L438A) are adjacent to the domain D epitope region on E2 and led to major loss of binding by one or more antigenic domain B hmAbs. Because this region is associated with escape from certain antibodies (39), it is still possible that mutants in this region would be useful in vaccine design. Several positions within or near HVR2 (L480, P490, P491, R492, K500) seemed to impact mAb recognition when mutated to alanine, including antigenic domain E antibodies. P491A was also observed in a separate study to improve mAb binding of a panel of broadly neutralizing E2 hmAbs, altering binding of AR3 antibodies (which significantly overlap with domain B antibodies) by over 1.5-fold (13). Mutants I622A and F627A, located near and within the domain A binding site, also showed affinity improvement for certain antigenic domains, with F627A also resulting in loss of binding for nonneutralizing domain A hmAbs.

Effects of Nonalanine Substitutions.

Although alanine substitutions provide a view of E2 residue dependence for binding of the panel of hmAbs studied here, it is likely that nonalanine amino acid substitutions may provide additional information in this regard. To this end, we tested nonalanine single and double mutants at positions 436, 439, 440, and 616 for binding to a subset of the hmAb panel using ELISA (Fig. 8). Mutants were generated in E2 from a genotype 1b isolate, 1bSF (47), and were selected to add a glycan sequon (G436S, leading to possible glycan at position N434) or alter E2 mobility via proline or disulfide (A439P, A440C-W616C), using positions with limited effects on the hmAb panel, specifically domain B and D hmAbs, when mutated to alanine. Binding data for these mutants largely supported the results obtained using alanine mutants in the context of the H77 isolate E2; binding to conformationally dependent domain A (CBH-4B, CBH-4G) and C (CBH-7) hmAbs was unaffected; and the domain B and D hmAbs showed variable binding, particularly for the 440–616 cysteine double mutant for which binding to those hmAbs was eliminated. This result highlights the modular nature of the E2 antigenic surface, where disruptive mutants exert local effects on subsets of conformationally dependent antibodies.

Fig. 8.

Antibody binding effects of nonalanine E2 mutations. Selected nonalanine mutants of E2 were expressed and tested for binding to a subset of the hmAb panel using ELISA, at an antibody concentration of 5 μg/mL hmAb names colored according to antigenic domain, as in Fig. 1.

Discussion

Our analysis of the antibody binding effects of global E2 alanine substitutions provides many new insights into E2 3D structure and determinants of antibody recognition. Through comprehensively mapping the binding determinants for antibodies in our panel, we identified all critical binding residues for hmAbs for which limited epitope mapping was previously described (HC-1, HC-11, HC84.26, HC84.20, HC84.24, HC84.26, HC33.1, HC33.4) (12, 25, 39), and also present mapping data for antigenic domain A and C hmAbs, which have little or no previous epitope mapping data nor crystal structures available. Whereas some key binding residues are on the E2 surface, a large proportion of them are buried in E2 core structures (18, 19), including cysteines in disulfide bonds and large hydrophobic residues; these likely do not contact the antibodies directly and influence recognition through effects on E2 local or global stability. Previous studies have shown that escape from antibody recognition (23, 37), as well as improved antibody recognition (16, 23, 37), can be mediated through residues in E2 likely distant from antibody binding sites, including buried residues and HVRs. This study confirms these findings and delineates such residues in the context of our hmAb panel. Comparing our analysis with CD81 binding and viral infectivity measurements demonstrated the overlap of functionally critical E2 residues and antibody binding determinants, as noted by others (16, 44).

We found highly varying levels of dependence on E2 residues for binding across our hmAb panel. Antibodies associated with antigenic domains A–D were previously shown to be dependent on E2 conformation based on lack of binding to denatured E2 (24, 25). Although alanine mutagenesis confirms this, we observed varying levels of this dependence ranging from antigenic domain A hmAbs (37 shared hotspot residues) to antigenic domain C hmAbs (12 shared hotspot residues, of which 11 are common to all conformation-dependent hmAbs in the panel). Our analysis suggests that E2 is somewhat modular in terms of conformational epitope stability. Antigenic domain E hmAbs were confirmed to recognize a linear epitope that is largely uncoupled from the remainder of E2, corroborating the “flap-like” nature of this epitope outside of the globular E2 core structure, as speculated by others (30). However, several E2 mutants outside of this epitope disrupt or improve binding of these hmAbs by two- to threefold, including C652A and P490A. The observed subtle effects on binding could explain the differential neutralization of HC33 hmAbs for various HCV cell culture-derived genotypes, despite identical sequences for residues 412–423 among those isolates (12). Other studies, through deep sequencing of viral sequences in patients undergoing immune therapy with the HCV1 mAb (which also binds to this epitope) (43), or epitope mapping of mAbs that bind to portions of this epitope as well as other residues (13), have linked this epitope to other regions of E2.

Although this study provides an unprecedented global view of E2 antibody recognition, there are numerous areas where future work can provide further insights. Although other E2-binding antibodies overlap with many of the epitopes studied here (SI Appendix, Table S5), alanine scanning using other antibodies may yield distinct global E2 binding patterns because of altered specificity or binding mode. This is also the case for E1E2-binding hmAbs, at least one of which competes with the antigenic domain C hmAb CBH-7 for E1E2 binding (11). Additionally, nonalanine and nonglycine mutants and combinations of multiple mutants, shown in some cases to be used in viral escape (23, 37), were not systematically measured in this study. However, we did test a small set of nonalanine mutants (including one double mutant), showing that alanine mutant binding data can indicate which sets of hmAbs in a panel of antibodies would be affected when residues are mutated to nonalanine amino acids (Fig. 8).

These data and analysis should facilitate rational antigen design of E2 in development of an HCV vaccine to elicit broadly neutralizing antibodies. Notably, such immunogen design would not need to explicitly conserve determinants of viral infectivity or maintain binding to coreceptors. In this study, we identified a number of alanine mutants of E2 residues that specifically enhance binding for groups of hmAbs in our panel. Although individual effects were often limited to two- to threefold, combinations of these mutants may yield significant improvements via additive or cooperative allosteric effects. We also observed several residues on E2 that specifically disrupt binding by nonneutralizing hmAbs to antigenic domain A; mutations at these sites can be used to reduce or eliminate elicitation of antibodies targeting this region, thereby shifting the immune response to antigenic domains associated with broadly neutralizing antibodies. Others have noted that structure-based stabilization of viral antigens can improve antibody recognition or immunogenicity (48–50), and the observed antibody binding influence of core E2 residues, highlighting its local and global dynamics, suggests that stabilizing designs in the context of E2 may yield improved antigenic properties. The concept of engineering E2 to improve its ability to induce neutralizing antibodies to select sites has yet to be tested in vivo, although a recent review has noted its potential (9). The design principles from such an effort would likely be applicable to vaccine design for other highly variable viruses, such as influenza and HIV.

Materials and Methods

E2 Mutagenesis and Binding Measurements.

Global alanine scanning of E2 was performed using site-directed mutagenesis of E2 residues to alanine (glycine substitutions for alanine residues), with antibody binding measured by ELISA. Mutants were constructed in plasmids carrying the 1a H77C E1E2 coding sequence (GenBank ID AF009606), as described previously (39). All of the mutations were confirmed by DNA sequence analysis (Elim Biopharmaceuticals) for the desired mutations and for absence of unexpected residue changes in the full-length E1E2-encoding sequence. The resulting plasmids were transfected into HEK 293T cells for transient protein expression using the calcium-phosphate method. Each antibody was tested at a midrange concentration that was established by dose-dependent binding against wild-type recombinant E1E2 cell lysate. Individual E2 protein expression was normalized by binding of CBH-17, an HCV E2 hmAb to a linear epitope (26). Data are shown as mean values of two experiments performed in triplicate; variability between replicate measurements was less than 10%.

Nonalanine E2 mutants were expressed as soluble E2 glycoproteins (amino acids 384–662), with wild-type sequence derived from the genotype 1b isolate 1bSF (47) (GenBank ID JN118490). These were expressed in HEK 293T cells and tested for binding using ELISA, as described previously (25).

CD81 binding to E2 mutant H77 pseudoparticles (HCVpp) was measured using as described previously (39). Briefly, HCVpp were produced as described below and captured on GNA-coated plates. CD81 at 100 µg/mL was added to each well and bound CD81 was determined with alkaline phosphatase-conjugated anti-human CD81.

HCVpp Infectivity Assays.

Huh7.5.1 cells were infected with wild-type or E2 mutant H77 HCVpp. Mutant W420A was not tested because of demonstrated lack of HCVpp infectivity in several studies (44, 51). HCV infection was analyzed by luciferase reporter protein expression, and results are expressed in relative light units (RLU). For HCVpp with infectivities lower than 5 × 10³ RLU, further neutralization experiments were not performed because of infectivities too low to obtain robust data. For the remaining HCVpp, Huh7.5.1 cells were preincubated with increasing concentrations of anti-CD81 [QV-6A8-F2-C4 (52)] or control mAbs for 1 h at 37 °C before infection with wild-type and E2 alanine mutants HCVpp. HCV infection was analyzed by luciferase reporter gene expression. EC₅₀ values were obtained using GraphPad Prism software.

Clustering.

Hierarchical clustering was performed using R (www.R-project.org/), after transforming binding percentage data to log ratios, with binding values of 0% set to 0.5% to permit log calculation. E2 positions and antibodies were compared using Euclidean and correlation-based distances, respectively, and clustered using Ward’s minimum variance method. Positions that lacked binding data for one or more hmAbs (13 of 355 positions) were removed before clustering. Tree height cutoffs for clustering were selected based on known binding epitope sizes and antibody antigenic domain sizes. Cluster P values were computed using the approximate unbiased method in the pvclust R package (53), with 10,000 bootstrap replicates.

Contact Prediction.

Pairs of residues were compared based on Euclidean distance or Pearson correlation between log-transformed alanine scan binding data used for clustering. Only pairs of positions with values for all hmAbs were considered. The top 600 of a possible 58,653 pairs of positions (∼1%) were selected as predicted contacts.

Solvent Accessibility.

Solvent accessible residues were determined using NACCESS (54), and surface residues were those with 20% or higher relative side-chain accessibility. Glycan hetero atoms resolved in the E2 core crystal structure (PDB ID code 4MWF, chain D) were included in NACCESS calculations, to account for the solvent accessibility of glycosylated asparagine side chains.

Computational Mutagenesis.

Rosetta v2.3 was used to perform computational alanine scanning (35) of all individual residues in the E2 core structure (PDB ID code 4MWF, chain D). All nonprotein atoms were removed before Rosetta modeling. Minimization of backbone and side chains was performed before and after mutation (command line arguments “-min_interface -min_chi -min_bb”).