De novo identification of VRC01 class HIV-1–neutralizing antibodies by next-generation sequencing of B-cell transcripts

Serum Analysis of Donor C38 Sample.

VRC01 class antibodies precisely target the CD4-binding site of the gp120 envelope glycoprotein to achieve broad neutralization of HIV-1 (8, 19). To date, heavy chains of all VRC01 class antibodies identified originate from IGHV1-2 family genes. We selected an HIV-1–infected donor (C38), whose sera exhibited broad and potent neutralization of HIV-1 (SI Appendix, Table S1). Serum neutralization of a panel of eight HIV-1 isolates was assessed in the presence of a resurfaced HIV-1 gp120 core protein (RSC3), a probe that binds VRC01 (19), and a negative control (ΔRSC3) that contains a single amino acid deletion in the VRC01 binding site. In the presence of RSC3 but not ΔRSC3, sera from donor C38 showed a mean 22% reduction in neutralization. Although this reduction was less than seen in donors 45, 74, and 0219, from which we had previously identified VRC01 class antibodies (7) (Fig. 1A), the data nonetheless suggested the presence of VRC01 class or other types of CD4-binding site-directed antibodies in this donor.

454 Pyrosequencing and Sequence-Based Identification of VRC01-Class Antibodies.

We performed next-generation sequencing of donor C38 B-cell transcripts using PCR to amplify IgG and IgM heavy chain sequences from the VH1 family. mRNA from an estimated 5 million peripheral blood mononuclear cells (PBMCs) was used for reverse transcription to produce template cDNA. We initially used primers that overlapped the end of the V-gene leader sequence and the start of the V region (H1 primers) (7, 20) and then switched to more upstream primers that annealed to the start of the V-gene leader sequence (G1 primers; SI Appendix, Table S2) (10); by avoiding the variable region, the G1 primers were better able to capture sequences with increased somatic mutation (10).

Roche 454 pyrosequencing provided 460,706 (H1 primers) and 574,027 (G1 primers) heavy chain reads for donor C38. After initial processing using our previously described bioinformatics pipeline (7, 21) (SI Appendix, Table S3), 138,523 sequences in the H1 primer data set and 168,365 sequences in the G1 primer data set were assigned to the IGHV1-2*02 allelic origin, the heavy chain germ-line gene for VRC01 class antibodies (7, 19). These sequences were then compared with a set of known template VRC01 class antibodies isolated from other donors, which included eight from our previous studies (7, 19) (VRC01-03, VRC-PG04/04b, and VRC-CH30-32) and five identified by Nussenzweig and coworkers (10) (12A12, 12A21, 3BNC60, 3BNC117, and NIH45-46) (SI Appendix, Fig. S1). Because all antibody variable domains, heavy and light chains alike, share a similar scaffold with conserved sequence motifs in the framework regions that are responsible for the heavy/light chain complexation, the sequence identity of irrelevant antibody chains is usually higher than that of functionally distinct protein domains. Therefore, an identity cutoff of 80%, instead of 30–40% as is generally used for unrelated proteins, was used here to detect sequence homology between antibody chains. No sequence from either the H1 or G1 primer data set was found to be >80% identical to any of the heavy chains from template antibodies from other donors (Fig. 1B). These results suggested that VRC01 class antibodies, if they did exist in the 454 pyrosequencing–derived repertoire of donor C38, could not be simply recognized by sequence homology to a known antibody.

Sequence Prevalence-Based Identification of VRC01 Class Heavy Chains.

Given the low homology of donor sequences to the known VRC01 class antibodies, we sought to use a prevalence-based method to interrogate the donor C38 repertoire. One implementation of antibody prevalence analysis, the lineage rank method (6), is based on the supposition that the desired antibodies form highly prevalent lineages in the total antibody repertoire (Fig. 2A). Because the heavy chains of all known VRC01 class antibodies can complement the VRC01 light chain to form neutralizing antibodies, we tested the lineage rank approach on heavy chain sequences. The heavy chain complementary-determining region 3 (CDR H3), which encompasses the site of V(D)J recombination—V-D junction, D segment, D-J junction, and part of J gene preceding the conserved WGXG motif, provides a sequence signature for antibody lineage definition. This definition of an antibody clonal lineage may be more robust than analyses based on the inferred germ-line use and junctional sequences, due to the relatively low accuracy of D gene assignment, as well as other complicating factors such as repertoire diversity and sequencing errors. Both CDR H3 similarity and VH gene characteristics were considered in our lineage definition. Within a germ-line VH gene family, sequences having a divergence of 20% or greater were first clustered into groups such that sequences within each group have no more than five nucleotide differences in the CDR H3 region. These heavy chain sequence groups are henceforth referred to as CDR H3 groups; lineages were then constructed by merging similar CDR H3 groups based on three criteria: (i) their CDR H3 sequences were of the same length; (ii) they shared >80% amino acid sequence identity; and (iii) divergence in V gene did not increase significantly on group merging (to prevent merging two different heavy chain lineages with coincidentally similar CDR H3s). Prevalent lineages (those with >1,000 sequences in the current case) could then be subjected to experimental validation. Lineage rank was tested on both H1 and G1 primer-derived data acquired from donor C38 and H1 primer-derived data from donor 74, which was acquired in our previous study (7) and served here as a benchmark. The intermediate output of lineage rank analyses, such as CDR H3 groups and lineages constructed from these groups, is detailed for donor 74 and donor C38 in the SI Appendix, Figs. S2 and S3 and Tables S4–S9.

Analysis of IGHV1-2 family heavy chain sequences derived from donor 74, from whom we previously isolated mAb VRC-PG04, revealed four CDR H3 groups (SI Appendix, Table S4). The two most prevalent of these (SI Appendix, Table S5) were found to correspond to the broadly neutralizing VRC-PG04–like antibodies designated classes 7 and 8 in our earlier paper (7) (SI Appendix, Fig. S2). Analysis of the donor C38 IGHV1-2 sequences derived from H1 primer amplification revealed a lower than expected population (<6%) of VH sequences that were more than 20% diverged from their germ-line ancestor, suggesting that H1 primers were suboptimal for amplification of matured antibodies in this donor (SI Appendix, Fig. S3 and Tables S6 and S7), and thus the lineage rank analysis focused on the mature sequences from VH1 gene families that were amplified using the G1 primers (Fig. 2B). For each identified lineage, the predicted sequence at the center of the largest sequence cluster was selected as a representative heavy chain, which was synthesized and cloned into an expression vector; the full antibody was reconstituted with the VRC01 light chain and tested for neutralization. As stated above, we chose the VRC01 light chain because prior data demonstrated that it complemented other VRC01 class heavy chain sequences (7). If total V gene variation within the lineage was 7% or greater, a second representative sequence was also tested. A total of 35 heavy chain sequences were selected from 22 lineages (Fig. 2C; SI Appendix, Tables S8 and S9), comprising >15% of the whole data set and 73% of the heavy chain population with a divergence of >20%. Thirty-three of these VH region sequences, when reconstituted as full heavy chains and paired with a VRC01 light chain, could be expressed as soluble IgGs (SI Appendix, Table S10), but none of these showed neutralization against HIV-1. Thus, analysis of the most prevalent lineages in donor C38 antibodyome did not identify VRC01 class neutralizing antibody sequences, suggesting that either this donor does not contain VRC01 class antibodies or the prevalence assumption of the lineage rank method did not apply for this donor.

Phylogeny-Based Identification of VRC01 Class Antibody Heavy Chains.

We next tested an evolution-based method for identification of VRC01 class antibody heavy chains termed “cross-donor phylogenetic analysis.” We previously demonstrated that VRC01 class antibody heavy chains from different donors evolve in a similar way to achieve precise recognition of the CD4-binding site and developed a sieving method—cross-donor phylogenetic analysis—based on phylogenetic similarity to capture heavy chain sequences with similar maturation pattern (7). In this analysis, a phylogenetic tree was rooted by the IGHV1-2*02 germ line, and donor VRC01 class heavy chain sequences segregated with exogenously added, known VRC01 class antibody heavy chains from other donors. Our prior analysis of donor 74 identified 5,047 heavy chain sequences in the VRC01-like subtree. (Note that VRC01-like refers to sequences that possess the key characteristics of VRC01 class enumerated above but have not been experimentally confirmed to have the same function.) Twenty-four of these were members of a set of sequences chosen by other means (identity/divergence analysis) that were reconstituted and shown in our prior paper to neutralize HIV-1 (7). Neutralization by these sequences suggests that the phylogeny-based method can be used to directly identify VRC01 class antibody heavy chains from a donor antibodyome. However, it is worth noting that the cross-donor analysis of donor 74 was performed with prior knowledge of VRC01 class antibodies isolated from the same donor, as opposed to the donor C38, for whom such information is not available.

To test this possibility, we performed cross-donor phylogenetic analysis on the donor C38 heavy chain sequences derived from H1 primer amplification (131,108 total sequences) and identified 11 sequences in the VRC01-like subtree. Two of these were expressed with the VRC01 light chain, and one of them (gVRC-H1_dC38) showed broad neutralization of HIV-1 (Fig. 3A). Using the G1 primers (163,108 sequences), 93 nonredundant heavy chain sequences were identified as segregating in the VRC01-like subtree, including the sequence designated gVRC-H1_dC38. Among these, 10 representative sequences were manually selected from different branches of the VRC01-like subtree for expression with the VRC01 light chain. Nine of the reconstituted antibodies displayed HIV-1 neutralization, yielding a 90% success rate in identification (Fig. 3B). In total, 10 functional VRC01 class antibody heavy chains were identified from donor C38 (Fig. 4A; SI Appendix, Table S11). Identity/divergence analysis (Fig. 4B) showed that none of these 10 heavy chains was >75% identical to any of the 13 template VRC01 class antibody heavy chains. Furthermore, a significant number of unrelated sequences with higher sequence homology to the template sequences were found in both H1 and G1 primer-derived antibodyomes. A detailed gene family and junction analysis of 10 heavy chains (SI Appendix, Figs. S4 and S5) revealed four different CDR H3 groups, suggesting that they might belong to different lineages or alternatively that they evolved from the same ancestor but diverged significantly after a long maturation process. The antibody gVRC-H1_dC38/VRC01L was tested on a panel of 153 HIV-1 strains and neutralized >80% of these at concentrations of <50 μg/mL (SI Appendix, Fig. S6 and Table S12).

Identification of VRC01 Class Antibody Light Chains from Donor C38.

An antibody consists of two paired heavy and light chains. Identification of light chain partners for the 10 neutralizing heavy chains allows for the reconstitution of functional VRC01 class antibodies from donor C38. Because the germ-line origin of C38 light chains was unclear, we designed primers to amplify both λ and κ germ-line V genes (SI Appendix, Table S2). With the first 454 pyrosequencing experiment, 211,830 full-length light chain sequences, 54,079 λ chains and 157,751 κ chains, were obtained from 257,910 raw reads (SI Appendix, Table S3). Recently, West et al. (12) analyzed structural and sequence data for known VRC01 class antibodies and found that these antibodies possess a CDR L3 loop of five amino acids and a glutamine (Q) or glutamate (E) at position 96 (Kabat numbering). Separately, we investigated the B-cell ontogeny of VRC01 class antibodies from multiple donors with next-generation sequencing and arrived at a similar conclusion to the definition of CDR L1 signature. Based on these studies, we adopted a sequence-specific motif—a CDR L3 length of five amino acids and Q or E at position 96—as a simple signature, along with a requirement of maturation from the germ line greater than 10%, to identify VRC01-like antibody light chains from the antibodyome. Four κ chains, but no λ chains, were found to meet the criteria. Given this result, we performed a second 454 pyrosequencing experiment to amplify only κ chains and obtained greater sequencing depth (SI Appendix, Table S3). Nine more κ chains with this sequence motif were identified from the total of 448,125 raw reads.

Sequence alignment of 13 light chain sequences (Fig. 5A) showed prevalent use of the IGKV3-20 germ-line gene, the same light chain variable gene used in VRC01, VRC03, and VRC-PG04. These 13 sequences, along with the two antibodyomes where they were identified, were compared with the same set of template VRC01 class antibodies used in the heavy chain analysis (Fig. 1B). None of the 13 light chains was found to be >80% identical to any of the template light chains from other donors (Fig. 5B), suggesting a lack of sequence homology in their variable genes. Furthermore, a large portion of the κ chain sequences in two antibodyomes, notably closer to germ-line genes on the identity/divergence plots (Fig. 5B), were found to be 80–90% identical to the template light chains. These results suggested that random sequences of low divergence, instead of the 13 light chains that resemble the CDR L1 signature of VRC01 class antibodies, would be selected by a pure homology-based method of identification. Collectively, we show with both heavy chains (Figs. 1B and 4B) and light chains (Fig. 5B) that sequence homology to known VRC01 class antibodies from other donors cannot identify such antibodies from donor C38.

De Novo Identification of Functional VRC01 Class Antibodies from Donor C38.

Pairing 10 neutralizing heavy chains with 13 candidate light chains may reveal their optimal combinations but would require significant experimental effort. Here we adopted a two-step approach to this problem. In the first step, we used a VRC01 heavy chain to screen 13 light chains. Six of 13 light chains were able to be reconstituted as full antibodies with a VRC01 heavy chain, but only one light chain, named gVRC-L1_dC38, showed weak neutralization of two HIV-1 isolates on a seven-virus panel (Fig. 6A). In the second step, we used the most potent C38 heavy chain obtained from cross-donor analysis, gVRC-H3_dC38, and the only effective C38 light chain from previous screening, gVRC-L1_dC38, to search for their respective partner chains. With heavy chain gVRC-H3_dC38, 8 of 13 light chains were expressed as full antibodies (Fig. 6B, Upper) compared with 6 with a VRC01 heavy chain (Fig. 6A). When tested on seven HIV-1 isolates, six of eight reconstituted antibodies showed neutralization with various breadth and potency (Fig. 6B, Upper), as opposed to a single neutralizer when paired with a VRC01 heavy chain (Fig. 6A). Interestingly, gVRC-L1_dC38 remained the most effective of C38 light chains tested, suggesting that the initial screening with the VRC01 heavy chain was unbiased. With the light chain gVRC-L1_dC38, we observed broader neutralization from most C38 heavy chains tested, with the isolate ZM109.4 neutralized by three antibodies, and identified gVRC-H3_dC38/gVRC-L1_dC38 as the optimal pair of C38 heavy and light chains (Fig. 6B, Lower).

The marked difference of breadth and potency seen in the two screening experiments of C38 light chains (Fig. 6 A and B, Upper) suggested that some unfavorable interactions between C38 light chains and the VRC01 heavy chain might underlie deteriorated function when paired. The comparison of 13 light chain sequences (Fig. 5A) offered some clues to the nonexpression of five tested light chains, four of which possess a nonphenylalanine (F) residue at position 97—the last residue of the CDR L3 loop—and the fifth sequence, with an index of 393230, is 97% identical to gVRC-L1_dC38 but with a glutamate (E) mutated to arginine (R) in the CDR L2 loop. The best C38 heavy/light chain pair, gVRC-H7_dC38/gVRC-L1_dC38, showed an ∼10-fold decrease in potency compared with gVRC-Hd_dC38/VRC01L (Fig. 6B, Lower), suggesting that the C38 light chain might be suboptimal for complementing VRC01 class antibody heavy chains, even those from the same donor. A possible explanation might involve the long CDR L1 loops. All 13 C38 light chains have one or zero deletions in the CDR L1 region (Fig. 5A), whereas VRC01 or other light chains of this class have two or more residues deleted or mutated to glycines in this region.

With the functional pairing of 10 heavy chains and 6 light chains from donor C38, we next sought to understand the evolutionary relationship of these sequences. We calculated maximum-likelihood phylogenetic trees rooted by their respective germ-line genes (Fig. 6C). Similar topology was observed for the heavy chain and light chain dendrograms, with the optimal pair, gVRC-H7_dC38/gVRC-L1_dC38, formed by sequences from two corresponding branches, suggesting that this chimera might resemble a native pair.

Human Specimens.

The sera and PBMCs were obtained from HIV-1–infected donors (SI Appendix, Table S1) enrolled in investigational review board–approved clinical protocols at the National Institute of Allergy and Infectious Diseases, including the VRC200 protocol. At the time of sampling, all HIV-1–infected donors were off antiretroviral treatment.

Antibody Expression, Purification, and Neutralization.

Similar procedures and measurements were used as described in a previous study (7, 21).

454 Sample Preparation and 454 Pyrosequencing.

The procedure was similar to ref. 7 but included new primers aimed to capture highly matured heavy chain sequences. For IGHV1 amplification, the H1 primers used in previous study extended from the end of the V-gene leader sequence into the start of the V region, whereas the G1 primers (10) used in the current study were more upstream and overlapped only with the V-gene leader sequence.

Computational Analysis of 454 Pyrosequencing Data.

The tools used for analysis were implemented as a suite of PERL scripts, named Antibodyomics1.0. The analysis consisted of two steps. In the first step, sequences encompassing antibody heavy chain variable domains generated by 454 pyrosequencing were processed using a bioinformatics pipeline (7, 21) to derive biologically defined properties (such as germ-line divergence and sequence identity), as well as to improve sequence quality by correcting the most common sequencing errors. In pipeline processing, sequences were reformatted, assigned to germ-line gene families, error-corrected using a template-based algorithm, compared with a set of user-specified antibodies, and subjected to the determination and comparison of CDR H3 regions. Full-length heavy chain variable domain sequences with detailed annotation were outputted for subsequent analyses. In the second step, various systems-level analyses were carried out to mine the antibody population, including detailed analyses of antibody properties (e.g., CDR H3 lineages and their distribution), lineage rank, and cross-donor phylogenetic analysis.

Data Deposition and Software Distribution.

The 10 functional heavy chain sequences (gVRC-H1-10_dC38) and 6 functional light chain sequences (gVRC-L1-6_dC38) identified from donor C38 antibodyomes have been deposited in GenBank under accession numbers KF306044–KF306069. The 454 pyrosequencing data sets have been deposited in the NCBI Short Reads Archives under accession number SRP026397. The software suite described here, Antibodyomics1.0, can be obtained by request from J.Z., P.D.K., or L.S.