Structural Basis of HCV Neutralization by Human Monoclonal Antibodies Resistant to Viral Neutralization Escape

Crystallization of Fab/peptide complexes and structure determination

We had previously reported that the human monoclonal antibodies HC84-1 and HC84-27 bind to a peptide encompassing residues 434–446 of the precursor polyprotein of the H77 strain, although the interaction depends on the epitope conformation [21]. Therefore, we performed co-crystallization trials at 20°C as described in Materials and Methods. Both complexes crystallized and diffracted to at least 2.2 Å resolution (Table 1). The structures were determined by the molecular replacement method using the variable and constant regions of an unrelated human Fab fragment as separate search models (see Materials and Methods). Difference maps calculated after refinement of the recombinant Fab molecules revealed well-defined electron density for the peptide in both structures (Figure S1), which we used to manually build an atomic model. The resulting structures of the two complexes are displayed in Figure 1. Because the observed peptide conformation could be influenced by its crystalline environment, we analyzed the packing contacts in both crystals. The HC84-1 complex crystals (space group C222₁) showed that the epitope II peptide packs about a 2-fold axis of the crystal against its counterpart from a symmetry related complex (grey arrow, see Figure S2A), implying that crystal contacts could affect its conformation. However, in the complex with HC84-27 (space group P1), the peptide makes no crystal packing contacts and yet it adopts a very similar conformation (Figure S2B). These observations are strong indications that the peptide conformation observed in both structures corresponds to the one present in the native polypeptide chain. Given that receptor-blocking neutralizing antibodies like the HC84 antibodies bind to the surface of the native virion, we conclude that this peptide conformation is similar to the one adopted by the corresponding segment of E2 at the virion surface.

HC84 Fabs in complex with epitope II

The peptide corresponding to epitope II forms a 1.5 α-helical turn spanning residues W⁴³⁷-F⁴⁴² displaying the typical extensive main chain hydrogen bonding pattern of an α-helix (Table S4). It continues at the C-terminal side of the helix in an extended conformation comprising residues Y⁴⁴³-K⁴⁴⁶ (Figure 2A+D). No electron density was observed for the N-terminal residues (one in HC84-1 and two in HC84-27), indicating that they are disordered in the complex. These residues therefore likely do not participate in the epitope.

A comparison of the structures of HC84-1 and HC84-27 showed important differences in the elbow angle between variable and constant domains of the Fabs (132° and 145° for HC84-1 and HC84-27, respectively). In contrast, a superposition of the variable domains and the peptides of both complexes revealed only small differences, indicating a very similar antigen-binding mode for both Fabs. This was expected, because the two heavy chains are essentially identical (apart from three residues at the N-terminus) and the light chains share ∼76% identical amino acids in the variable region (Figure 1C+D). Therefore, we will discuss the binding mode that is common to both Fabs and highlight differences that are due to sequence variation between the two antibodies.

Molecular determinants of epitope II recognition by HC84 antibodies

Antigen binding buries an area of 588.7 Ų and 725.3 Ų on HC84-1 and HC84-27 (Table S1), with shape complementarity indexes of 0.74 and 0.81, respectively (Table S2) [32]. The peptide is bound such that the N- and C-terminal ends interact with the heavy and light chain, respectively. At the heavy-chain side, the paratope forms a hydrophobic binding surface with contacts to W⁴³⁷-L⁴⁴¹ within the short α-helical turn. This surface extends into a hydrophobic cavity into which the aromatic side chain of F⁴⁴² at the C-terminal end of the epitope helix inserts (Figures 1A–B and 3A–B). The walls of the cavity are formed by residues from all three heavy-chain complementarity determining regions (CDRs) and framework residues around the CDR-H2 loop. The peptide makes essentially hydrophobic interactions with the heavy chain (Table S3), as highlighted by the hydrophobicity pattern of the paratopes (Figure 3A–B). The peptide/Fab complexes are further stabilized by hydrogen bonds between the peptide main chain and the CDR-H3 loop (Table S3). On the peptide side, analysis of the solvent-accessible surface area that is buried in the complex shows that mainly residues L⁴⁴¹ and F⁴⁴² interact with the heavy chain, with 60–90% of their solvent accessible surface area buried by heavy chain binding.

The interactions with the light chain include an extensive hydrogen-bonding network by the side chain of K⁴⁴⁶ to the CDR-L1 and CDR-L2 loops and in case of HC84-1 also the CDR-L3 loop. This is further stabilized by a hydrogen bond between Y⁴⁴³ at the N-terminal part of the extended segment and S^L93 within the CDR-L3 loop. In addition, there are stacking interactions between W^L90 and Y⁴⁴³. The C-terminal extended segment of the peptide crosses the edge of the cavity with the side chain of Q⁴⁴⁴ forming a hydrogen bond to Q^L49 and van der Waals interactions with Y^L31 of HC84-27. It terminates with the side chain of K⁴⁴⁶ forming a salt bridge with D^L50. In mAb HC84-1 Y^L31 is replaced by serine and Q^L49 is replaced by aspartic acid, leading to a change of the Q⁴⁴⁴ side chain conformation when compared to the HC84-27 complex and loss of the respective side chain interactions. This altered conformation results in a drastically decreased buried surface area of Q⁴⁴⁴ (20.06 Ų and 133.76 Ų for HC84-1 and HC84-27, respectively; Figure 2F–G). Likely, the loss of these interactions is the main reason for the considerably smaller solvent-accessible surface area on the light chain buried by peptide binding (242.7 Ų and 337.8 Ų for HC84-1 and HC84-27, respectively). This is also reflected in the total buried solvent-accessible surface area on the Fab (Table S1).

The main contact residues for HCV E2 binding of both monoclonal antibodies have been mapped to L⁴⁴¹ and F⁴⁴² by alanine scanning mutagenesis [21]. In addition, HC84-27 binding is impaired by substitution of residues Y⁴⁴³ and K⁴⁴⁶ as well as one residue located in region II (W⁶¹⁶). In line with these epitope mapping data the structure of the HC84-27 complex revealed additional electron density close to a distinct region of the paratope with contacts to the CDR-L3 loop and possible main chain interactions with the C″-strand of the HC84-27 heavy chain (Figure 4A–D). This electron density was clear enough to trace the main chain, but not the side chains and suggested a second binding site, which was not present in the corresponding HC84-1 complex, thereby correlating with an additional contact residue for HC84-27 at W⁶¹⁶. The presence of this extra density in a binary complex of HC84-27 and the epitope II peptide suggested a binding event independent of the amino acid sequence. We therefore also determined the structure of a ternary complex HC84-27 with two peptides (epitope II and a peptide corresponding to aa610–619; peptide 2), but although density in the second binding site appeared better defined, we were still unable to unambiguously assign side chains. Extensive experiments to characterize the Fab/peptide interaction at this second binding site (e.g., cocrystallization of a HC84-27 in complex with peptide aa610–619) did not further clarify the interactions between peptide 2 and the Fab fragment at this binding site. Further biochemical and structural studies will be required to unambiguously identify the binding mode at this second binding site of HC84-27.

Superposition of the epitope II peptide from the two complexes revealed a root mean square deviation (rmsd) per residue of 0.2–0.9 Å for the segment A⁴³⁹-Y⁴⁴³ calculated over all atoms (Figure 2B, dark grey), which is close to the positional error of the coordinates. In contrast, for residues outside this central helical region the rmsd was 1.6–3.9 Å. The amino acid backbone of the framing residues (L⁴³⁸, Q⁴⁴⁴ and H⁴⁴⁵) is also structurally very similar as revealed by calculation of the rmsd over the main chain (Figure 2B, light grey). The marked difference between the two rms deviations observed for those residues (calculated over the main chain and over all atoms, respectively) showed that the main chain adopts a defined conformation, while the corresponding side chains are more flexible. We also calculated the mean temperature factors (B-factors) per residue for both peptides and observed a similar distribution over the peptide. B-factors for A⁴³⁹-Y⁴⁴³ were at least ∼10 Ų lower than for residues outside of this region (Figure 2C). Together with the contact analysis, these results strongly suggest a highly ordered and stable antibody/antigen complex interface and an interaction that is dominated by hydrophobic contacts between the heavy chain CDRs (and framework residues surrounding the CDR-H2) and the short α-helical turn within epitope II.

Germline analysis, antibody maturation and effect on antigen binding

Antibody diversity is generated by the combinatorial association of V, D, and J germline segments, and is further diversified at the actual junctions (VL–JL, VH–D, and D–JH) due to imprecise joining and addition of “N region” nucleotides. Somatic mutations, possibly driven by antigenic selection, can further contribute to the diversity of antibodies and lead to increased affinity and specificity during antibody maturation [33], [34]. Germline analysis using IMGT V-QUEST and junction analysis [35] of the HC84-1 and -27 antibodies revealed the closest homologous germ-line genes to be IGHV1-69*01 F, IGHJ4*02 F and IGHD3-22*01 for the almost identical heavy chains. The light chain of HC84-1 uses IGLV3-1*01 F and IGLJ1*01 F and the light chain of HC84-27 is derived from IGLV3-21*03 F and IGLJ2*01 F, the two VL genes being 75% identical at the amino acid level. Heavy chains derived from IGHV1-69*01 F are frequently found in antibodies directed against HCV E2 [36], [37], indicating that this VH gene is preferentially used in the specific immune response to HCV E2. Interestingly, a group of antibodies recognizing the coreceptor-binding site on human immunodeficiency virus (HIV) gp120 selectively uses the same VH gene. This has been attributed to a strict dependence on a hydrophobic patch in the CDR-H2, given that IGHV1-69 is the only VH gene with hydrophobic residues at specific positions forming contacts that are conserved within this group of antibodies [38]. These residues (I^H52 and F^H55) are also conserved in HC84-1 and -27 and contribute to the important hydrophobic interactions with L⁴⁴¹ and F⁴⁴². A more exhaustive sequence analysis revealed that all nine HC84 antibodies use the same VH gene and while I^H52 is conserved across all heavy chains, F^H55 is in some antibodies replaced by other hydrophobic amino acids thereby maintaining the hydrophobic surface. In contrast, the VL gene usage within the HC84 antibody group is much less conserved as indicated by the usage of three different kappa VL genes and three different lambda VL genes. While we cannot fully exclude that the LC of HC84-1 and -27 derive from the same germline rearrangement and hence represent two (closely related) outcomes of a single clonal lineage, the LC diversity observed in the HC84 group shows no requirement for a particular lineage within the HC84 group. Given that the HC84 group has been isolated from a combinatorial library it remains unknown, if heavy and light chains of the two antibodies were paired in the original patient, however, peptide-reactive antibodies against epitope II have been found with high prevalence in patient sera compared to antibodies of other specificities [28]. Together with the neutralizing activity of these patient antibodies [28], the specific binding mode that is dominated by the heavy chain and the similar biological activities and epitope mapping results for the nine HC84 antibodies, these observations suggest that HC84-1 and HC84-27 could be found in HCV infected patients.

We aligned the HC84-1 and -27 antibodies to their closest homologous germ-line genes and identified somatic mutations in this alignment (Figure 1C+D, asterisks). Antibodies using the IGHV1-69 gene often acquire somatic mutations during antigenic selection that are located in or close to the CDRs, indicating a positive selection for antigen binding [37]. Mapping of the junction in CDR-H3 and the somatic mutations onto the molecular surface of the paratope revealed that this junction encloses the hydrophobic cavity at one side (Figures 4D+E). The ridge on the opposite side of the cavity is formed by M^H59 downstream of CDR-H2 and S^L93 and N^L93 in HC84-1 and HC84-27, respectively, both of which make a hydrogen bond to the OH atom of Y⁴⁴³. Together this demonstrates that the junction in CDR-H3 and somatic mutations acquired during antibody maturation contribute to the formation of the hydrophobic surface binding the epitope II peptide.

Structural explanation for broadly neutralizing activity of mAbs HC84-1 and -27

One major challenge for neutralizing antibodies in the course of HCV infection is the highly diverse population of viral variants that is found in patients. In the presence of neutralizing antibodies escape variants within this virus population will have a selective advantage over neutralization-sensitive variants. In spite of this, mAbs HC84-1 and HC84-27 neutralize a broad spectrum of HCV genotypes and do not allow neutralization escape [21]. To understand the structural basis of this broadly neutralizing activity and the resistance to neutralization escape we analyzed ∼7000 epitope II sequences from the Los Alamos HCV database (http://hcv.lanl.gov). While T⁴³⁵, G⁴³⁶, A⁴³⁹ and Y⁴⁴³ are highly conserved, Q⁴⁴⁴ and H⁴⁴⁵ are much less conserved (Figure 2E). Sequence analysis of position 446 revealed that a large majority of sequences carry either lysine or arginine, indicating a requirement for a positively charged residue at this position. F⁴⁴², which is a key determinant of binding to both broadly neutralizing antibodies, is surprisingly poorly conserved (∼60% of the sequences). Notably, sequence analysis revealed that F⁴⁴² can be replaced only by a bulky hydrophobic residue in the remaining sequences (e.g., isoleucine or leucine), which is likely to also insert into the hydrophobic pocket. Similar effects were observed for L⁴³⁸ and G⁴⁴⁰, which are also less conserved, but are replaced in the majority of the cases by isoleucine and alanine, demonstrating the requirement for a hydrophobic and small amino acid, respectively (Figure 2E).

Structure and function of epitope II in HCV E2

In solution, the isolated epitope II peptide is likely to adopt multiple conformations that are in equilibrium with each other, and the antibodies select the observed conformation for binding. Secondary structure predictions of E2 using different algorithms indeed provide contradictory results for this segment (Figure S3) with only a minority of the algorithms predicting the observed α-helical turn. Within folded E2, this segment would not be free to adopt multiple conformations, and the fact that the antibodies recognize folded E2 means that the binding determinants are presented in the same way at the protein surface. Our results imply that the aromatic side chains of F⁴⁴² and Y⁴⁴³ are both exposed on one side of the helix, and to a lesser extent also the aliphatic side chains of L⁴³⁸, A⁴³⁹ and L⁴⁴¹. Of these amino acids, systematic mutagenesis studies have shown that only L⁴³⁸ and A⁴³⁹ tolerate mutations without compromising CD81 binding and thus viability of the virus [23], suggesting that the other exposed residues within this region directly participate in receptor binding. The organization of the epitope II region is therefore compatible with the hydrophobic patch in CD81 that was identified as the target pocket [39]. Such patches on the surface of glycoproteins are often shielded from solvent by N-linked glycans. The protruding patch formed by residues F⁴⁴² and Y⁴⁴³ (Figure 3C+F) is very close in sequence to glycan N4 (corresponding to N⁴⁴⁸), suggesting a role of this glycan in masking this region of the protein. This is also in line with the reported glycan shielding of the CD81 binding site, with glycans attached to N⁴¹⁷ (N1), N⁴²³ (N2), N⁴⁴⁸ (N4), N⁵³² (N6) and N⁶⁴⁵ (N11) modulating the sensitivity of HCV infectious particles (HCVcc) to neutralizing antibodies [40]. While the presence of glycan N4 thus potentially complicates the design of an efficient E2 vaccine targeting the hydrophobic protrusion within epitope II, the viability of a mutant N448Q HCVcc JFH-1 virus is not compromised [40], indicating a valid alternative for both life vaccines and subunit vaccines consisting of recombinant HCV E2.

It is also worth noting that the observed organization of the epitope II region is reminiscent, as pointed out by Drummer and colleagues [23], of the aromatic and glycine rich fusion loop of class II fusion proteins. In the case of the flavivirus envelope protein, the fusion loop also features an α-helical turn exposing aromatic and aliphatic side chains, similar to the observed conformation of epitope II in our structures.

Our previous model for the domain organization of the HCV E2 ectodomain used the experimental disulfide connectivity and biochemical data on the composite nature of the CD81 binding site, but it also relied heavily on the assumption that the members of the various genera of the Flaviviridae family would have structurally homologous membrane fusion proteins in spite of a lack of sequence conservation [24]. This was strengthened by the observation that the membrane fusion proteins from the alphaviruses (Togaviridae family), and now also from the phleboviruses (a genus within the Bunyaviridae family of negative sense RNA viruses; [41]) share the common “class II” fusion protein fold in the absence of any sequence conservation. The recently reported structure of the E2 glycoprotein of bovine viral diarrhea virus (BVDV-1; PDB accession code 2YQ2), a member of the closely related pestivirus genus within the Flaviviridae family, demonstrated that pestivirus E2 does not have a class II fold, and may have receptor binding function but not be responsible for membrane fusion. BVDV E2 is an elongated molecule consisting of four sequential β-sandwich domains arranged linearly and named A through D from N- to C-terminus [25].

This structure gives space for three alternative models for the domain organization of the HCV E2 ectodomain. If HCV E2 were homologous to BVDV E2, the data from the antibody/peptide complexes, using the structures of epitope 413–423 [29], [30], [31] and now of epitope II, would suggest that domain A would be relatively unstructured, with the hypervariable region 1 (HVR1) at the N-terminus, followed by the flap formed by aa412–423 that is not part of a folded domain and extending to the epitope II region that can be mimicked by an isolated peptide. This interpretation is in line with the observed disulfide bond that links a cysteine at the very N-terminus of pestivirus E2 with another one further downstream, stabilizing the fold of pestiviral domain A; this cysteine is absent in HCV E2 allowing for increased structural flexibility of the HVR1. If such a model were correct, the data concerning the CD81 contact residues would suggest that the CD81 binding domain would be centered in domain B, and parts of domain A (aa412–423 and the epitope II region) and domain C (the 613–618 region) would contribute to receptor binding.

The current evidence does not rule out, however, that E2 HCV may indeed be a class II fusion protein. The lipid binding properties of CD81-primed, acid-treated HCV E2 [42] would be in line with a role in interactions with cellular membranes for virus entry. In this case, the recent structures of Fab/peptide complexes indicate that the assignment of domain I would have to be revised to begin after the aa413–123 “flap” region, and the strands reassigned such that the 1.5 helix of epitope II would be at the interface with domain III, indicating that residues currently assigned to domain II would have to complete domain I. Domain II would thus be smaller by about 6 residues.

Finally, the last hypothesis is that the HCV E2 glycoprotein adopts an entirely unrelated fold. Notably, the evolutionary constraints on the overall fold of an attachment glycoprotein are more relaxed than those for a fusion protein, which potentially leads to less structural similarity, and in some cases altogether different genes may be used for otherwise related viruses, like is the case between the arenaviruses and the bunyaviruses.

In conclusion, there are too many uncertainties left and only a structure of the HCV E2 ectodomain - or at least an isolated domain - can provide a definite answer. Nevertheless, the crystal structure of the epitope II in complex with two broadly neutralizing human antibodies provide structural insight into the neutralization mechanism of two antibodies of the HC84 group. Together with the sequence analysis of epitope II, it also provides evidence to explain why these antibodies are resistant to neutralization escape. Structural knowledge about interactions between broadly neutralizing antibodies – in particular those that are resistant to neutralization escape – is essential to understand the immune response against HCV E2 and to employ structure-based vaccine design to elicit neutralizing antibodies of similar specificity and efficiency [43].

Production and purification of recombinant Fab molecules

Synthetic genes that were codon optimized for Drosophila melanogaster coding for heavy and light chains of the Fab regions of each antibody were cloned into a Drosophila S2 Fab expression vector described previously [44] containing a double Strep tag for efficient affinity purification. Drosophila S2 cells were transfected as reported previously [45]. For large-scale production cells were induced with 4 µM CdCl2 at a density of approximately 7×106 cells/ml per ml for 8 days, pelleted and Fabs were purified by affinity chromatography from the supernatant using a StrepTactin Superflow column followed by size exclusion chromatography using a Superdex200 column. Pure monomeric Fab was concentrated to approximately 20 mg/ml.

Peptides and complex formation

Synthetic peptides comprising either the epitope II region (residues 434–446 - NTGWLAGLFYQHK) or a region located further downstream (residues 610–619 - DYPYRLWHYP) of the H77 strain were synthesized by GenScript (>98% purity) and dissolved in water at 10 mg/ml. A complex was formed overnight at 277 K containing 10 mg/ml Fab+0.9 mg/ml epitope II peptide (HC84-1), 10 mg/ml Fab+1.5 mg/ml epitope II peptide (HC84-27+one peptide) or 10 mg/ml Fab+0.9 mg/ml of each peptide (HC84-27+two peptides).

Crystallization, data collection, structure determination and refinement

Fab crystals in complex with were grown at 293 K using the hanging-drop vapor-diffusion method in drops containing 1 µl complex solution (10.9–11.8 mg/ml in 10 mM TRIS pH 8.0, 150 mM NaCl) mixed with 1 µl reservoir solution containing 100 mM TRIS pH 8.0, 19% PEG4000, 170 mM Lithium Sulfate and 15% Glycerol (HC84-1) or 22–24.6% PEG 3350 and 250–300 mM Sodium Thiocyanate (HC84-27). Diffraction quality plates (HC84-1) or rock-like (HC84-27) crystals appeared after one week and were flash-frozen in mother liquor with (HC84-27) or without (HC84-1) 22% PEG400. Spacegroups and cell dimensions of the crystals, resolution limits, data collection details and refinement statistics are summarized in Table 1.

Data were collected at the SLS (PX I) and the Synchrotron Soleil (Proxima1). Data were processed with Autoproc [46] using XDS [47] and scaling and reduction was performed using Pointless [48] and programs from the CCP4 suite [49]. The crystal structures of the Fab complexes were determined by the molecular replacement method using Phaser [50]. The molecular replacement for Fab HC84-27 was performed using separate variable and constant regions of a hypothetical Fab fragment assembled from the LC of PDB accession code 2XZA (81% aa identity) and the HC of PDB accession code 3QOT (89% aa identity) as search model. The molecular replacement for Fab HC84-1 was performed using Fab HC84-27 as search model. Model building was performed using Coot [51] and refinement was done using AutoBuster [52].

Crystal structure analysis

The two peptides derived from the complexes of HC84-1 and -27 in complex with only epitope II were aligned using the MatchMaker algorithm implemented in Chimera and an iterative alignment process pruning long atom pairs until no pair exceeds 1 Å. Root mean square deviations were calculated between the two epitope II peptides either over all atoms per residue or taking into account only the main chain atoms (N, CA, C, O) using Chimera [53].

Buried solvent accessible surface areas for the interfaces as well as for individual residues within the peptides were calculated using the PISA server [54]. Shape complementarity was calculated using programs of the CCP4 suite [49]. Interactions were determined using the protein interactions calculator (PIC; [55]). Figures were prepared with Pymol (http://www.pymol.org).

Epitope II sequence analysis and secondary structure prediction

6998 sequences were taken from the Los Alamos HCV sequence database and (http://hcv.lanl.gov) analysed using the tools of the ViPR database (http://www.viprbrc.org). Secondary structure prediction was performed using all algorithms on the Network sequence analysis server (NPS@, Network Protein Sequence Analysis, http://pbil.ibcp.fr/NPSA; [56]).

Data deposition

The atomic coordinates and structure factors for two crystal structures have been deposited in the Protein Data Bank, www.pdb.org, under the accession numbers 4JZN and 4JZO.