Multi-Donor Longitudinal Antibody Repertoire Sequencing Reveals the Existence of Public Antibody Clonotypes in HIV-1 Infection

1The HIV-1 envelope glycoprotein (Env) mediates receptor recognition and viral fusion and serves as the sole target of the neutralizing antibody response. The developmental pathway of Env-specific antibodies has been probed previously using high-throughput sequencing, but such analyses have focused on single broadly neutralizing antibody (bNAb) lineages after infection. However, bNAbs comprise only a fraction of the antibody response within a given individual, which also includes antibodies with limited or no breadth. These diverse antibodies are subject to viral selection pressures and host constraints, target a variety of epitopes on Env, and potentially possess functions other than neutralization. More generally, thorough and large-scale profiling of the repertoire-wide antibody response during the course of natural infection remains a predominantly unexplored area of investigation and an unmet need in HIV-1 research. Indeed, the extensive evidence of the global effects that HIV-1 has on the adaptive immune system, including hypergammaglobulinemia, CD4+ T cell abnormalities, and defective CD8+ T cell function, motivates efforts to understand the dynamics of the antibody repertoires of HIV-infected individuals.

2Although putative bNAb precursors have been discovered in HIV-naive repertoires, it is unclear how the antibody repertoires of HIV-infected individuals change from the time before infection through different stages of infection. Furthermore, while ontogeny and structural studies of HIV-reactive antibodies have revealed convergence at the structural level in multiple donors, the overall differences and similarities in the antibody repertoires of HIV-infected donors have not been characterized. Due to the diversity of potential target epitopes on Env, as well as the potentially infinite antibody sequence space resulting from gene recombination and affinity maturation, it could be expected that the antibody repertoire of each individual might be unique. Yet public antibody clonotypes that are shared among multiple individuals have been observed previously for dengue infection, after influenza vaccination, and in other immune settings. However, in the context of HIV-1 infection the potential for public antibodies has not been explored.

3To better understand antibody repertoire dynamics throughout HIV-1 infection, we performed antibody repertoire sequence analysis to examine characteristics of the pre- and post-infection repertoires of multiple donors. To that end, we longitudinally sequenced the global immunoglobulin heavy chain repertoires of six South African donors from the Centre for the AIDS Programme of Research in South Africa (CAPRISA) from before infection through acute and chronic infection. We also performed paired heavy and light chain sequencing of the Env-specific post-infection repertoires of two additional CAPRISA donors. The resulting analysis provides insights into how antibody repertoires of different individuals are reshaped during the course of HIV-1 infection.

4Antibody variable genes in peripheral blood cell samples from three time points, categorized as pre-infection, 6 months post infection (mpi), or 3 years post infection (ypi), were sequenced for each of six CAPRISA donors (Table S1). The pre-infection time points ranged from 30 to 2 weeks before infection, with the exception of donor CAP322, for whom the earliest available sample was at 2 weeks post infection. All CAPRISA donors were infected with clade C viruses but exhibited diverse neutralization phenotypes, including substantial variation in neutralization breadth between 0% and 61% on a representative panel of diverse HIV-1 strains (Table S1). For the three donors with demonstrable neutralization breadth (CAP287, CAP312, and CAP322), we also performed neutralization fingerprinting analysis to delineate the epitope specificities of broadly neutralizing antibodies at the 3 ypi time points. These three donors were predicted to possess different types of antibody specificities (Table S2). Taken together, the observed differences in neutralization phenotypes for the six donors indicated diversity in the types of antibody specificities present in each donor.

5After sequencing the antibody heavy chain variable gene regions from all three time points for each of the six donors (STAR Methods), we first investigated how repertoire composition changed over time. We began by determining clonal family membership for each observed V(D)J sequence using V-gene assignment, J-gene assignment, junction length, and junction identity (STAR Methods). All time points from each donor were included during clonal family assignment so that closely related sequences from different longitudinal samples could be assigned to the same clonotype. Clonal family assignment revealed the number of clonotypes (groups of sequences resulting from clonal family assignment) in each donor to range from 9,141 to 26,777, with a total of 103,475 unique clonotypes across all six donors (Table S3). Few clonotypes spanned multiple time points within a donor, with the majority of clonotypes belonging to a single time point (Figure 1A). While some clonotypes were present in all three time points within individual donors, this finding was rare, typically representing only ∼0.08%–1.16% of all clonotypes, where clonotypes observed in pre-infection and 3ypi samples, but not 6mpi, were counted toward the number of clonotypes across all three timepoints (Figure 1A). Clonotypes spanning any two adjacent time points were also rare, although somewhat more frequent than those that spanned all three time points, representing ∼0.52%–1.78% of all clonotypes (Figure 1A). Interestingly, although sequence turnover was high, antibody heavy chain variable gene usage distributions remained predominantly constant over the course of infection (Figures 1B and S1). While some genes appeared to be used preferentially within each sample (Figure S1), there was no discernable pattern across all donors of longitudinal enrichment of particular genes. Similarly, the length distributions for the third heavy chain complementarity-determining region (CDRH3) remained relatively unchanged throughout infection for each donor (p > 0.1 for all CDRH3 lengths in all samples, Z test with Benjamini-Hochberg correction) (Figure 1C). Taken together, these results indicated that, while systems-level repertoire features were conserved over time, antibody sequence retention over the course of HIV-1 infection was low; rather, each donor was associated with virtually non-overlapping repertoires at the three different time points.

6Since recent studies have indicated the presence of public antibody clonotypes shared among individuals, including in the antigen-directed response, we set out to explore the presence of public antibody sequences in the case of HIV-1 infection. To identify public antibodies, we performed clonal family assignment on all samples from all donors simultaneously (STAR Methods). As expected, this approach resulted in most clonotypes consisting of sequences from single donors. However, we also identified a number of clonotypes with sequences from multiple donors (Figure 2), which we designated public clonotypes. Although these sequence groups are not technically biological clones (as they were derived from multiple individuals and thus are not derived from a single B cell), public clonotypes were defined as groups of sequences with the same VH gene, the same JH gene, the same junction length, and CDRH3 amino acid sequences of high identity between donors (STAR Methods).

7We analyzed the antibody sequences using a range of junction region identity thresholds during the clonal assignment procedure (Figure S3A). Of note, public clonotypes were identified for all threshold values, including at 100% junction identity. Although identities between members of known bNAb lineages can be as low as ∼30% (Table S4), for further analysis we selected a conservative threshold of 70%, in which all members of a clonotype would have at least 70% junction region identity to all other members of the clonotype. This threshold aimed at allowing reasonable inclusion of intra-clonal evolution without allowing highly divergent sequences from different donors to be grouped together.

8Comparison of the antibody repertoires between all pairs of donors revealed that public antibody clonotypes existed at each of the pre-infection, 6 mpi, and 3 ypi time points (Figure 2A). Intriguingly, at the 70% junction identity threshold used for this analysis, the number of public clonotypes significantly increased at 3 ypi (average number of public clonotypes from pairwise comparisons of 69.1) relative to pre-infection (average number of public clonotypes from pairwise comparisons of 21.5) (p < 0.0001, linear mixed effects model) (Figures 2A and 2B). Although the difference between the numbers of public clonotypes at pre-infection and 6 mpi was not significant (p = 0.69, linear mixed effects model), the difference between 6 mpi and 3 ypi was also significant (p = 0.0002, linear mixed effects model). These public clonotypes were not restricted to germline sequences and exhibited a wide distribution of V-gene deviation from germline (Figure S2).

9To determine whether public clonotypes could be identified within larger subsets of donors, we compared the antibody repertoires for all combinations of two to six donors for all three time points (Figure 2C). Smaller numbers of clonotypes were detected as the number of donors between whom a clonotype was shared increased, but 27 clonotypes shared by at least four of the six donors also were identified (Figures 2C, S3B, and S3C). Interestingly, public antibodies present 3 ypi were found predominantly only after infection, with at most ∼7.5% of 3 ypi antibodies shared by any pair of donors present in the pre-infection repertoire of either donor (Figure S3D). Taken together, these results suggested that a non-negligible fraction of public antibodies appeared to emerge after HIV-1 infection (Figures S3D and S3E) alongside the highly diverse private antibody response.

10We performed paired heavy-light chain sequencing of the antigen-specific repertoire of two additional donors, CAP248 and CAP314 (STAR Methods, Table S1), to explore the potential of public antibodies to be HIV-1 reactive. Comparison of these sequences with the sequences from each of the other six CAPRISA donors revealed a large public clonotype between CAP248, CAP314, and CAP351 (Figure 3). Examples of representative antibodies from this public clonotype, including CAP351 CDRH3 sequences within a 1–2 amino acid difference from each of antibodies CAP248_#30 and CAP314_#30, are shown in Figure 3A. The maximum junction difference among these sequences (Figures 3A and S3F) was three amino acids, or ∼81% identity, for CAP248_#30 and CAP314_#30. In addition to the high CDRH3 identity, 4 of 13 somatic mutation changes from the IGHV1-69 germline gene in the CAP248 antibody were identical in one of the representative CAP351 antibodies. Remarkably, CAP314_#30 and CAP248_#30 each used the same light chain germline gene, IGKV1-27, with CDRL3 sequences differing by just one amino acid (Figure S3F). As paired heavy and light chain sequence information was available for CAP248_#30 and CAP314_#30, we produced these antibodies as recombinant immunoglobulin G proteins and tested their reactivity against HIV-1. The two antibodies bound Consensus C (ConC) gp120 (Figure 3B) and neutralized tier 1 viruses MN.3 and MW965 but did not neutralize tier 2 viruses (Figure S3G). Epitope mapping revealed that antigen binding by both antibodies was affected by the D368R mutation but not by the N332A mutation (Figure 3C), potentially suggesting a CD4-binding site epitope specificity. Overall, these results demonstrate that public HIV-reactive antibodies can emerge during the course of natural HIV-1 infection.

11In general, while HIV-1 reactivity for an antibody cannot be determined solely from its sequence, it is informative to determine whether antibodies that are similar to known HIV-reactive sequences exist at the population level, as a way to assess the uniqueness of the space of antibody HIV-1 reactivity. We compared over 18 million heavy chain antibody sequences from approximately 250 publicly available next-generation sequencing samples, as well as from the samples presented in this study, to the heavy chain sequences of known HIV-reactive antibodies. To that end, we collected and curated a set of heavy chain sequences from HIV-reactive antibodies found in the PDB, CATNAP (Compile, Analyze and Tally NAb Panels) and literature (STAR Methods). The sequencing data were derived from infection, vaccination, and autoimmunity studies, as well as from CAPRISA samples sequenced as part of this study. We then compared the CDRH3 amino acid sequences and V-gene assignments from all sequences in each sample with CDRH3 sequences and V-gene assignments encoding HIV-reactive antibodies (Figure 4). Specific sequence features known to be important for HIV-1 recognition, such as a requisite tryptophan in the fifth position preceding the framework 4 region for 3BNC60 and gp120-binding putative VRC01 precursors, were included as additional constraints in the comparison. As expected, many observed sequences had low identity to known HIV-1 antibodies. However, 35 known HIV-reactive antibodies had matches with at least 70% CDRH3 identity, in addition to matching V-gene and sequence feature requirements, to antibodies in the examined samples (Figure 4A). With these search parameters, as many as ∼70 donors were matched to a given HIV-reactive antibody. In addition, nine antibodies had matches within a difference of no more than two amino acids in the CDRH3 region, including an exact match for antibody 02-o (Figure 4B). The nine known HIV-reactive antibodies (Figure 4B) included antibodies with extra-neutralization functions, weakly neutralizing antibodies, putative vaccine-elicited antibody precursors, and putative bNAb precursors, and spanned a range of CDRH3 lengths from 10 to 19 amino acids. These results suggest that sequences within a short distance from HIV-reactive antibodies, including some that could be possible vaccine templates, exist at the population level.

12Here, we interrogated the antibody heavy chain repertoires of multiple HIV-infected donors at multiple time points through next-generation sequencing and systems immunology analysis. Antibody sequence repertoires in chronic HIV-1 infection appeared to diverge from their pre-infection repertoires. While we cannot exclude sampling depth as a possible explanation for the observed repertoire turnover, virus-induced turnover or a more general property of B cell repertoire dynamics could also explain this observation.

13The discovery of public antibody clonotypes in the setting of HIV-1 infection is intriguing, especially given the rapid viral evolution that occurs during HIV-1 infection. While previous studies have observed public antibodies following acute viral infections, these antigens are far less diverse than HIV-1. It is plausible that some of the identified public antibody clonotypes are not HIV specific. In particular, existence of public clonotypes with affinity-matured sequences in pre-infection repertoires (Figure S2) could indicate shared modes of recognition at the antibody sequence level in response to common, previously encountered antigens. However, public clonotypes that are HIV reactive also exist (Figure 3). Furthermore, public clonotypes with high sequence identity to known HIV-reactive antibodies also were identified in a variety of immune settings, including in HIV-naive individuals (Figure 4). Our analysis identified public clonotypes for potential HIV-1 bNAb precursors but less frequently for public mature bNAbs (Figure 4), suggesting the possibility that bNAbs may evolve from virtually identical recombination events that subsequently take divergent evolutionary pathways in different individuals. Taken together, our findings suggest that a number of HIV-reactive antibody sequences may be readily accessible in multiple individuals, or even at the population level. Future vaccine development strategies may therefore benefit from specifically avoiding immunogen engagement with non-neutralizing public antibody clonotypes, such as the three-donor IGHV1-69 public clonotype characterized here, and aim to target engagement of bNAb or bNAb-precursor public clonotypes.

14The complexity of the humoral response to HIV-1 infection warrants large-scale analyses, such as those described here. Future juxtaposition of the antibody repertoires of broad and weak neutralizers, and from before and after infection or vaccination, could prove to be critical for identifying key correlates of protection and for iterative improvement of vaccine strategies. Such vaccine strategies could seek to recapitulate repertoire features observed to be conducive to protection or aim to interact with specific and common features of the antibody repertoire, such as public sequences. Further exploration of these possibilities with larger infection cohorts representative of global HIV-1 infection diversity is therefore needed.

15Further information and requests for resources and reagents should be directed to the Lead Contact, Ivelin Georgiev (ivelin.georgiev@vanderbilt.edu).

16Participants in the CAPRISA study with stored peripheral blood mononuclear cells (PBMCs) from pre-infection and post-infection time points were selected for this study. The CAPRISA 002 Acute Infection study was approved by the ethics committees of the University of KwaZulu-Natal (E013/04), the University of Cape Town (025/2004) and the University of the Witwatersrand (MM040202). All 8 donors provided written consent for the use of stored samples for HIV-related research projects. See Table S1 for further details. Gender and age of the donors was not a consideration in this study.

17For sequencing of global antibody repertoires, total RNA was extracted from PBMCs and an RT-PCR reaction was performed using a primer mix (BIOMED2 to framework 1 and framework 4) designed to amplify all heavy-chain antibody variable gene regions in an unbiased fashion. The resulting amplicons were purified, and agarose gel electrophoresis was used to confirm complete primer removal and appropriate amplicon size. The amplicon sample was quantified prior to submission to the Vanderbilt Technologies for Advanced Genomics (VANTAGE) core for paired-end (2 x 300 bp) Illumina MiSeq sequencing. The 18 samples from 6 donors were multiplexed across 3 MiSeq runs. FASTQ files from Illumina MiSeq sequencing served as the main input for subsequent data analysis. As a control, a second sequencing run was conducted with all 18 samples on one chip (Figures S4A and S4B), from which we identified ∼0.55 million V(D)J sequences across the 18 samples. Public sequences from run 1 were frequently recovered in the second run (Figure S4B).

18Preprocessing was carried out using pRESTO as follows: 1) Paired-end reads were interleaved and reads with a mean Phred quality score below 20 were removed. 2) Orientation of sequences was corrected to the forward orientation (V to J) as necessary. 3) V-region primers were masked and C-region primers were cut. Sequences with no match to primers were discarded. 4) Duplicate sequences were removed and a duplication count of each sequence was annotated. 5) All sequences with duplication count of 1 were removed. A wide distribution of duplication counts was observed (Figure S4C). 6) Each sequence was annotated for V, D, and J gene usage using IgBLAST, using reference sequences from IMGT. After removing non-functional sequences and sequences with CDRH3 lengths of under 5 amino acids, clonal clustering of all 1316148 V(D)J sequences from all donors was performed using Change-O (Figure S4D). Functional VH V(D)J sequences were assigned to clonal groups by first grouping sequences based on common IGHV gene annotation, IGHJ gene annotation and junction region lengths. IGHV and IGHJ gene annotations for each group of sequences were determined by the first gene assignment of gene assignments within each junction length. Within these larger groups, sequences differing from one another by a threshold distance within the junction region were defined as clonotypes by complete-linkage clustering. Distance was determined using an amino acid Hamming distance normalized to the length of the junction. We used amino acid distance during clustering because this is the determinant of antibody molecular recognition; however, nucleotide and amino acid identities were generally very similar (Figure S4E).

19Paired chain antibody sequencing for CAP248 and CAP314 was carried out by Atreca (Redwood City, CA) on IgG cells sorted into microtiter plates at one cell per well by FACS. Briefly, cryopreserved PBMCs were stained with the following antibodies: CD14-FITC (HCD14), CD3-FITC (UCHT1), IgM-A488 (MHM-88), IgD-A488 (IA6-2), CD20-PECy7 or CD20-BV711 (2H7), CD38-PECy7 or CD38-PerCPCy5.5 (HIT2), CD27-BV510 (O323) from BioLegend, CD19-BV421 (HIB19) from BioLegend or CD19-PE (SJ25C1) from eBiosciences, and IgA-FITC (IS11-8E10) from Miltenyi. For CAP248 PBMCs, antigen-specific cells were isolated using CAP45 SOSIP.664 trimeric protein. For CAP314 PBMCs, antigen-specific cells were isolated using ConC gp120. The sorted antigen-specific B cells were cultured for 4 days in IMDM medium (Invitrogen) in the presence of FBS, Normocin, IL-2 (PeproTech), IL-21 (PeproTech), rCD40 ligand (R&D Systems), and His-Tag antibodies (R&D Systems), prior to single cell sequencing.

20Generation of barcoded cDNA, PCR amplification, and next-generation sequencing of paired IgG heavy & light chains were performed as described in, with the following modifications: desthiobiotinylated oligo (dT) and Maxima H Minus RT (Fisher Scientific Company) were used for reverse transcription, DynaBeads™ MyOne™ Streptavidin C1 (Life Technologies) was used to isolate desthiobiotinylated cDNA, PCR amplicon concentrations were determined using qPCR (KAPA SYBR® FAST qPCR Kit for Illumina, Kapabiosystems), and amplicons were sequenced on an Illumina MiSeq instrument.

21FASTQ output files were grouped and parsed into separate FASTQ files on the basis of their compound ID (plate-ID + well-ID). We used Atreca proprietary software to assemble paired end reads into consensus sequences, requiring a minimum coverage of 30 reads for each heavy and each light chain assembly. Wells with more than one contig for a given chain type were rejected from consideration unless one of the contigs included at least 90% of the reads. V(D)J assignment and mutation identification was performed using IgBLAST. Allele assignments (Figures 3A and S3F) were determined using IMGT/V-Quest.

22A list of HIV-reactive antibodies was curated manually from the Protein Databank, CATNAP, and literature. CDRH3 sequence and V-gene usage of each antibody was determined using ANARCI with IMGT numbering. Publicly available antibody sequencing datasets collected from the Short Read Archive and the European Nucleotide Archive, in addition to the samples presented in this study, were processed via MiXCR. Briefly, reads in each sample were aligned with the mixcr align command and clustered by junction sequence and V-gene assignment using the mixcr assemble command with the –OseparateByV=true flag. First and last residues of the junction were trimmed to match the IMGT CDRH3 definition, and the first V-gene assignment in the alignment score-ranked list of possible gene assignments was used for comparison to the list of HIV-reactive antibodies. Within each sample, only sequences with a matching V-gene assignment and CDRH3 length to each HIV-1 antibody were compared for CDRH3 identity. Sequence identities were calculated using the editdistance library in Python and normalized to the length of the CDRH3 sequence. For antibodies 3BNC60, 02-cb, 02-o, 02-s, 02-u, 02-k, 02-t, 02-v, and 04-k, the requirement of a tryptophan in the fifth position preceding the framework 4 region also was imposed. Each accession number used in this study was manually annotated for donor ID in order to count the number of donors with matches to each HIV-reactive antibody (Figure 4). To retrieve identity to germline of MiXCR-processed data, alignments were exported using the exportAlignments command including the -vBestIdentityPercent flag. This calculates the fraction of matching nucleotides with the germline V-gene divided by the alignment length, which varied among publicly available samples due to differing sample preparation methods and sequencing strategies.

23Neutralization fingerprinting analysis of donor sera was performed as described previously. Briefly, epitope-specific neutralization fingerprints were constructed for ten antibody specificities against a panel of 21 diverse HIV-1 strains. For each serum, the serum-virus neutralization data against the same panel of HIV-1 strains was compared to the epitope-specific antibody fingerprints, in order to estimate the relative contribution of each of these reference antibody specificities to the polyclonal serum neutralization. For each serum, the estimated neutralization contributions by each of the ten reference antibody specificities were reported on a scale of 0 to 1, with all specificities adding up to 1.

24To determine statistical significance of pairwise donor overlap across timepoints (Figure 2), a linear mixed effects model was fit with timepoints designated as fixed effects and donors treated as random effects, thereby accounting for any correlation resulting from each donor being part of 5 pairwise comparisons per timepoint. Briefly, each donor was represented by a 45-dimensional vector, with each dimension having a value of 1 or 0 based on if that donor was part of each of the 45 pairwise comparisons that occurred across the 3 timepoints. The linear mixed effects model was then fit using the lmer function from the lme4 package in R. P values were determined using the lmerTest package in R. Other statistical tests were performed using R version 3.4.1.

25Sequencing data of global B-cell repertoires generated in this study have been deposited for public access under BioProject PRJNA415492.