Rhesus Macaque B-Cell Responses to an HIV-1 Trimer Vaccine Revealed by Unbiased Longitudinal Repertoire Analysis

F128 MAbs and an NHP antibodyomics pipeline.

In the previous study of rhesus macaques immunized five times with a YU2 gp140 foldon (gp140-F) trimer (30), F128 was one of the animals that displayed the most potent neutralizing titers against a panel of cross-clade tier 1 isolates. A total of eight functional MAbs directed to the conserved CD4bs were isolated from F128. Among these MAbs, GE136, GE140, and GE143 neutralized SF162 and MN, while GE136 and GE143 also neutralized non-clade B viruses. Sequence analysis based on the current database of rhesus macaque germline genes (32) revealed a degree of SHM ranging from 1.4 to 8.4% at the nucleotide level, with the heavy chain complementarity-determining region 3 (HCDR3) varying between 15 and 21 amino acids (aa). In the current study, seven MAbs (GE121, GE136, GE137, GE140, GE143, GE147, and GE148) were selected to query the NGS-derived F128 antibody repertoires, with GE125 being excluded due to its similarity to GE121.

We developed an NHP antibodyomics pipeline based on the human antibodyomics pipeline that has been widely used to study HIV-1 bNAbs (4, 10, 19–21, 41). Rhesus H, κ, and λ germline gene segments—including the variable (V), diverse (D), and joining (J) segments—reported by Sundling et al. (32) (termed the “CS germline gene database” hereafter) were incorporated into the pipeline to facilitate germline gene assignment, error correction, and determination of H/LCDR3 and variable region boundaries (see Materials and Methods). Sixty-three V_H genes along with 60 D_H and 7 J_H genes were included for heavy chain analysis, while 62 V_K genes with 5 J_K genes and 50 V_L genes with 6 J_L genes were included for κ and λ light chain analysis, respectively. The resulting NHP antibodyomics pipeline was first validated using the seven F128 MAbs, resulting in an SHM level of 1.4 to 7.1% for heavy chains and 1.8 to 12.2% for light chains based on the CS germline gene database, consistent with the results from manual alignment (32). A pipeline based on the IMGT rhesus macaque germline gene database (http://www.imgt.org/), which contains 23 V_H, 32 D_H, and 8 J_H gene segments, was implemented for purposes of comparison.

Antibodyomics analysis of the 454 pyrosequencing data.

We reanalyzed the 454 sequencing data (34) using both NHP antibodyomics pipelines (see Fig. S1 in the supplemental material). Of 846,736 reads, 797,794 and 798,333 can be assigned to NHP germline V_H genes defined within the CS and IMGT germline gene databases, respectively. After error correction using the respective germline genes as templates (21), the sequences were then compared to the seven MAb heavy chains at both nucleotide and amino acid levels. After pipeline processing, an additional filter was used to eliminate sequences containing PCR errors, resulting in 657,713 and 657,603 full-length heavy chain variable region sequences for the two germline gene databases, respectively. From the data processed by the CS germline gene database, we identified 163,921 (24.9%), 220,529 (33.5%), and 201,647 (30.6%) heavy chains of VH1, VH3, and VH4 origins, with VH2 and VH7 genes accounting for 1.7 and 9.2% of the population, respectively (Fig. 1A, top). Of note, the VH1, VH3, and VH4 germline gene families obtained from this NHP antibodyomics pipeline are 2.4- to 8.1-fold larger than reported previously (34), providing a more complete data set for subsequent analysis. From the data processed by the IMGT germline gene database, a highly skewed usage of IgHV1 (24.9%) and IgHV4 (30.7%) was observed, likely due to incorrect gene assignment (Fig. 1A, bottom left). The germline divergence distributions calculated from the CS and IMGT germline gene databases showed a marked difference, with averages of 7.1 and 11.7%, respectively (Fig. 1A, bottom middle). In contrast, the choice of germline gene database appeared to have no effect on the HCDR3 distribution, with an average loop length of 12.4 aa (Kabat definition) for both data sets (Fig. 1A, bottom right).

We then searched the reprocessed 454 sequencing data to identify the somatic variants of seven MAbs using an HCDR3 identity cutoff of 95% or greater. Such analysis is a “trademark” of human bNAb studies (4, 10, 19–23, 41) but not yet demonstrated for NHP MAbs. A two-dimensional (2D) identity-divergence analysis was performed to visualize the F128 heavy chain population with the identified somatic variants shown on the plots (Fig. 1B). Irrespective of the germline gene database used, the same somatic variants were identified for GE136 (n = 1), GE137 (n = 3), GE140 (n = 75), GE143 (n = 18), GE147 (n = 8), and GE148 (n = 4), with no relatives found for GE121. Of note, the 2D plots calculated from the IMGT germline gene database appeared to be elongated along the divergence (X−) axis, consistent with the shift of germline divergence distribution (Fig. 1A, bottom middle). Overall, the identity-divergence plots validated the NHP antibodyomics pipeline and provided initial insight into the somatic population of these F128 MAbs. However, the ontogeny of these MAb lineages and their development during the course of vaccination are currently unclear, and will likely require a longitudinal repertoire analysis.

Taken together, the results of reanalysis of 454 sequencing data demonstrated a robust performance of the NHP antibodyomics pipeline, as indicated by the output during data processing as well as the distributions and 2D plots derived from the processed data (Fig. 1). More importantly, the relative completeness of the NHP germline gene database showed no effect on the HCDR3 calculation, which is the basis of antibody lineage analysis. Therefore, the NHP antibodyomics pipeline, in conjunction with the most current CS germline gene database, provides a reliable platform for both NHP B-cell repertoire profiling and lineage tracing, which will be used hereafter to analyze immune responses following F128 immunization.

Utility of 5′ RACE PCR for unbiased capture of NHP antibody repertoire.

The unbiased analysis of human B-cell repertoires has been reported (41, 42). Such methods can in principle be adapted to monitor the vaccine-induced B-cell responses in NHPs, as illustrated schematically in Fig. 2A. For F128, blood samples were collected 1 and 2 weeks after each inoculation with an additional blood draw 3 weeks after the second and fourth inoculations. Given the F128 sample availability, four time points (after immunizations 2, 3, 4, and 5; referred to as post-2, -3, -4, and -5) were chosen to investigate the memory B-cell responses to an HIV-1 Env immunogen (Fig. 2B, NGS). In template preparation, the use of 5′ RACE PCR is critical to unbiased repertoire analysis but has not been demonstrated for NHPs. Here, we designed an experimental protocol to prepare NHP antibody libraries from purified PBMCs (Fig. 2C). In brief, a 5′ RACE adaptor was attached to the 5′ end of the antibody cDNA molecules by reverse transcription (RT). Three 3′ reverse primers, the IgH inner primer, the Igκ outer primer, and the Igλ outer primer, were adapted from the nested PCR protocol (31) to amplify the 5′ RACE adaptor-tagged H, κ, and λ chains in PCRs (see Table S1 in the supplemental material). These primers have been used to isolate hundreds of NHP MAbs in previous studies (30, 32, 34). The use of a single 3′ reverse primer ensured unbiased repertoire capture for each antibody chain type. As shown previously, the 5′ RACE PCR products of human antibody transcripts are ~600 bp, exceeding the optimal sequencing length of most NGS platforms (41). The combined use of isothermal amplification (IA), a high-quality (Hi-Q) sequencing enzyme, and a modified PGM protocol has enabled a more accurate analysis of human antibody repertoires (41). However, it is unclear whether this advance is applicable to the NHP antibody libraries. To address this issue, we characterized the human and NHP antibody libraries generated by different PCR methods using gel electrophoresis and PGM sequencing (Fig. 2D). For a human donor, the multiplex PCR products showed three bands ranging from 400 to 450 bp, with an average read length of 428 bp from PGM sequencing. In contrast, the 5′ RACE PCR products of human H, κ, and λ chains yielded bands of similar size (600 bp) when prepared from an aliquot of the same PBMC sample, with an average read length of 571 bp from PGM sequencing. For F128, the multiplex PCR products showed three bands with sizes similar to those of the human donor (400 to 450 bp), with an average read length of 437 bp. The 5′ RACE PCR products of NHP H, κ, and λ chains showed three bands of 600 to 700 bp on the gel, with an average read length of 635 bp from PGM sequencing. The longer reads from the 5′ RACE PCR-prepared NHP antibody library can be attributed to the use of further downstream primers, a library barcode (10 bp), a full-length PGM P1 adaptor (18 bp longer than the truncated P1 adaptor), and an increased flow number in PGM sequencing. Nonetheless, our results highlight the importance of long-read NGS capability for an unbiased analysis of human and NHP B-cell repertoires alike.

Analysis of the F128 repertoire during immunization.

PBMCs were obtained from F128 at four different time points to prepare H, κ, and λ chain libraries from the 5′ RACE PCR products (Fig. 2C; see also Table S1 in the supplemental material) and to confirm sample-to-sample reproducibility. We pooled the libraries from two time points on a single chip, 316 or 318 v2, and repeated the PGM sequencing twice to reduce potential experimental noise. In total, four PGM experiments generated over 22 million reads (Tables 1 and 2). Each of the four PGM data sets was divided into three subsets (H, κ, and λ) based on germline V gene assignment, with each subset being processed by a chain-specific NHP antibodyomics pipeline (see Fig. S2 in the supplemental material), yielding an average read length of 606 to 645 bp. Following correction for potential PCR errors, 80 to 90% of the reads were of sufficient length to cover the entire V(D)J reading frame (Table 1). The error-corrected, annotated sequences were merged into four data sets, each representing one time point and containing the H, κ, and λ libraries (Table 2). Greater sequencing depth and unbiased repertoire capture were expected to enable a more accurate comparison of repertoire properties across multiple time points.

We first analyzed the germline gene usage of F128 memory B-cell repertoires at four time points spanning the course of immunization (Fig. 3A). Since total PBMCs were taken 2 or 3 weeks after each inoculation per the original experimental design, the germline gene usage is mostly indicative of total memory B-cell responses. Overall, the germline gene usage appeared to be relatively constant during the immunization regimen. The heavy chain repertoire exhibited more fluctuations than the two light chain repertoires, suggesting that vaccine-induced changes in heavy chain gene usage and SHM might be detectable above contributions from the total resting memory B-cell compartment. In particular, the VH4 usage increased from 44.9 to 52.4% between the post-2 and -3 time points (see Fig. S3 in the supplemental material), while VH3.44 showed a notable decrease during the same 3-week period (Fig. 3), potentially resulting from oligoclonal B-cell expansion. The frequencies of VH1, VH3, and VH4 after 5 immunizations appeared to be different than those obtained from the 454 sequencing (Fig. 1A), in which the antibody libraries were prepared from three separate gene-specific primer reactions (34). For light chains, VK1 constituted ~60% of the κ chain repertoires, whereas VL1, VL2, and VL3 each accounted for 18.3 to 37.0% of the λ chain repertoires (see Fig. S3). Of note, the κ chains showed less than a 2.5% change between any two consecutive time points, compared to maximal changes of 6.9 and 3.0% for VL3 and VL5, respectively. Given the unbiased nature of this analysis, the germline gene frequencies likely reflect the general patterns of the NHP resting memory, class-switched, B-cell repertoire. We next determined the germline gene divergence and CDR3 length distributions. Consistent with the pattern of gene usage, the germline divergence showed limited changes over time (Fig. 3B). Overall, heavy chains yielded an average divergence of 9.1 to 9.7%, while κ and λ chains showed lower divergence values of 6.4 to 7.0% and 5.7 to 6.7%, respectively. These were lower than the divergence values observed for HIV-1-infected individuals but more closely resembled those of uninfected human donors (41). We also observed an oscillating pattern of divergence over the four time points, with more mature heavy chains (≥13%) appearing after the second and fourth inoculations and low-divergence light chains (<5%) increasing by up to 5% at the other two time points. The NHP CDR3 length distributions exhibited human-like patterns (Fig. 3C), with an average HCDR3 length of 12.5 aa compared to an average of 12.5 to 15.6 aa for human donors (Kabat definition) (41), supporting the notion that NHP can produce long HCDR3 loops similar to those displayed by human MAbs.

The unbiased analysis thus provided a global view of NHP memory B-cell repertoires during the course of vaccination. The population-based survey revealed a rather stable profile of general properties such as germline gene usage, degree of SHM, and CDR3 length with relatively small but notable changes. To better understand the vaccine-induced NHP B-cell responses, we traced the lineages of three CD4bs-directed neutralizing MAbs isolated from F128 (GE136, GE140, and GE143) in the unbiased antibody repertoires of four time points.

GE136-like transient memory B-cell response.

One GE136-like heavy chain was identified in the 454 sequencing data obtained at the completion of this immunization regimen (Fig. 1B). Here we sought to determine the approximate “birth time” and lineage development of GE136 in the context of unbiased repertoires. Using GE136 as a template, we interrogated the F128 repertoire from each of the four time points (Fig. 4A). GE136-like heavy chains were not identified from the last time point, perhaps indicating a low representation of the GE136 lineage in the repertoire in general, consistent with the 454 sequencing data (34). From the repertoires of the first three time points, we could find only 57 GE136-like heavy chains within the post-3 repertoire. These results indicate that the clonal lineage of GE136 notably expanded following the third immunization but fell below the level of detection afterward, perhaps due to competing CD4bs-directed antibody lineages or lineages of other epitope specificities.

We next examined the biological function of NGS-derived GE136-like heavy chains. The bioinformatics selection (see Materials and Methods) resulted in seven sequences, four of which were subjected to phylogenetic analysis and functional validation (see Table S2A in the supplemental material). In the maximum-likelihood (ML) tree rooted by the germline gene VH4.11, the GE136 heavy chain gene was sandwiched by 1194073_1B and three other sequences (Fig. 4B). GE136-like heavy chains were paired with the GE136 light chain and tested for their binding to the gp140-F trimer. By enzyme-linked immunosorbent assays (ELISA), the reconstituted antibodies bound to the vaccine antigen with an ~10-fold range of 50% effective concentrations (EC₅₀s) (Fig. 4C). The ability of the reconstituted antibodies to neutralize HIV-1 was confirmed using a cross-clade 6-virus panel (Table 3). Similar to the parental GE136, three of four GE136 somatic variants neutralized three of six HIV-1 isolates.

GE140-like evolving memory B-cell response.

A group of diverse somatic variants (n = 75) was identified for the GE140 heavy chain from the 454 sequencing of a sample after 5 immunizations (Fig. 1B). Given the transient presence of the GE136 lineage, we asked whether we could detect a similar pattern of lineage development for GE140. We first visualized the unbiased heavy and light chain repertoires with respect to GE140 using the 2D plots (Fig. 5A). Interestingly, GE140-like heavy chains were found from all time points except for the one after 2 immunizations. This finding suggests that the GE140 lineage formed within a time frame similar to that of GE136 but undertook a different developmental pathway, such that it remained detectable over the remaining course of the vaccination regimen (Fig. 5A, row 1). It should also be noted that two GE140-like MAbs were isolated from a different post-2 sample by single B-cell sorting (our unpublished data), suggesting an earlier birth time of the GE140 lineage. GE140-like heavy chains possessed a lower divergence after the third immunization yet continued to mature and form a separate lineage over the following 2 months, as indicated by the more divergent sequences along the 100% identity line and a wedge-shaped density departing from the main population (Fig. 5A, row 1). To quantify the GE140 lineage development, we identified all heavy chains of the VH4.11 origin and determined the fraction of heavy chains with an HCDR3 identity of 95% or more to GE140 within the VH4.11 family (Fig. 5A, row 2). The GE140 lineage evolved rapidly between post-3 and -4, with the fraction increasing from 0.83 to 6.07‰ before reaching a plateau of 5.59‰ between post-4 and -5. The GE140 lineage continued to diversify during maturation, as indicated by more sequences with higher divergence but lower GE140 identity (Fig. 5A, row 2). In contrast, the light chain repertoires contained sequences that were nearly 100% identical to the GE140 light chain, displaying little change over time (Fig. 5A, row 3).

We next investigated the phylogeny of GE140-like heavy chains. A bioinformatics selection (see Materials and Methods) yielded 40 heavy chains, which are displayed in an ML tree rooted by VH4.11 (Fig. 5B). Remarkably, the tree branching pattern was consistent with the chronological order of the maturation events: the post-3 sequences were found only from the top branches (marked as “B”), the post-4 sequences were found from the middle branches (“C”), and the distant branches were occupied solely by the post-5 sequences (“D”). Of note, some post-5 heavy chains were clustered with post-3 sequences on the top branch, suggestive of a recall of post-3 memory B cells after the last inoculation. The GE140 heavy chain itself was sandwiched by two post-5 sequences, consistent with the finding that a larger number of GE140 MAbs (28) were seen from the last time point (post-5) versus the two early time points (post-3 and -4). Overall, phylogenetic analysis revealed sustained GE140 lineage development over a 2-month period. We then selected 12 NGS-derived GE140 heavy chains for pairing with the GE140 light chain (see Table S2B in the supplemental material). Six of the reconstituted antibodies were expressed and showed gp140-F binding comparable to that of GE140 by ELISA (Fig. 5C), except for 2356992_1B. Of note, the four post-5 heavy chains on the most distant phylogenetic branch were not expressed, suggesting that they are likely remote variants of the GE140 heavy chain and might not be compatible with the GE140 light chain.

We identified 2,275 light chains with the VL2.7 germline origin and an LCDR3 identity of 95% or greater to the GE140 light chain from the last time point (Fig. 5D, left). This finding agrees with the low diversity expected of light chains due to the absence of the D gene segment. We then estimated the contribution of LCDR3 to GE140-antigen interactions by testing three light chains selected from this population: one with the same LCDR3 as GE140 and two possessing an insertion in LCDR3 (see Table S2B in the supplemental material). Interestingly, although all three light chain variants were expressed when paired with the GE140 heavy chain, only the one with an identical LCDR3 behaved like GE140 in ELISA binding (Fig. 5D, right), suggesting that LCDR3 plays a critical role in CD4bs recognition. Overall, a large number of light chains with the same germline gene composition and LCDR3 will facilitate the rapid production of a pool of antigen-binding B cell receptors (BCRs) capable of further antigen selection and maturation.

All expressed GE140 variants were tested on a 6-virus panel to assess their neutralizing ability (Table 3). As expected, the gp140-F-binding heavy chain variant (1095891_2B) and the light chain variant with identical LCDR3 (1657151_2D) could neutralize the same tier 1 isolates as GE140 itself when paired with respective GE140 partner chains. In summary, the clonal lineage of GE140 exhibited a pattern of sustained development in response to the CD4bs presented on an HIV-1 Env trimer, in contrast to the transient GE136 lineage of the same epitope specificity and germline V_H origin (but a different germline V_L origin, VL5.28). Recently, Tran et al. reported a less favorable angle of approach for GE136 (43). Therefore, the rapid expansion of the GE140 lineage in comparison to GE136 can likely be explained by GE140 having more favorable Env interactions, which would lead to greater fitness of the GE140 lineage in B-cell selection against GE136 and other antibody lineages.

GE143-like oscillating memory B-cell response.

Among all the CD4bs-directed MAbs isolated from F128, GE143 showed the highest degree of SHM at 8.4% (32). Compared to the GE136 and GE140 lineages, which first became detectable after 3 immunizations, a larger set of GE143-like heavy chains can be found in the repertoires of all four time points studied (Fig. 6A, row 1), suggesting that the GE143 lineage originated from naïve B cells responding to the first two inoculations. In addition, the higher degree of SHM and larger somatic population observed for GE143 could be the result of a more extensive maturation process. Specifically, the distribution of GE143-like heavy chains on the 2D plots exhibited an oscillating pattern (Fig. 6A, row 1), suggesting—within the limits of sampling error—a unique lineage development pertaining to the regimen, vaccine antigen, and antibody-antigen interactions.

Further analysis revealed that GE143-like heavy chains comprised the sequences of two putative germline origins: VH4.34 and VH4.40 (Fig. 6A, rows 2 and 3). Although both families waxed and waned during the immunization regimen, VH4.34 dominated the oscillating pattern. VH4.34 is different from VH4.40—the putative GE143 germline gene—by 18 nucleotides (6%) and 8 amino acids (8%), suggesting that multiple sites in the sequence may contribute to the ambiguity in gene assignment (see Fig. S4 in the supplemental material). To investigate the lineage structure of two putative germline families, we plotted sequences with an HCDR3 identity of over 92.0% (Fig. 6A, rows 2 and 3, blue dots) and over 95.0% (red dots) relative to GE143. Such hierarchical HCDR3 analysis will likely reveal hidden lineage patterns that would otherwise be overlooked with a single cutoff. Within the VH4.34 family (Fig. 6A, row 2), the frequency of GE143-like sequences defined by a lower cutoff (92.0%) appeared to increase over time (from 0.95 to 2.88‰). In contrast, the more stringent cutoff (95%) identified GE143-like sequences only after 3 and 5 immunizations, with frequencies of 1.13 and 1.81‰, respectively. Within the VH4.40 family (Fig. 6A, row 3), the lower cutoff showed a response peaking after 3 immunizations (0.79‰), whereas the more stringent cutoff—similar to the case of the VH4.34 family—identified GE143-like sequences only after 3 and 5 immunizations, with a notable decrease at the later time point. Our analysis thus uncovered a memory B-cell repertoire evolving toward GE143 driven by the HCDR3-mediated interactions. The activated memory B cells, upon acquiring the GE143-like HCDR3s, as defined by the 95% cutoff, were no longer detectable in the repertoire, perhaps due to antigen adsorption. The light chain repertoire remained consistent over the four time points studied (Fig. 6A, row 4), suggesting a lesser role of light chains in GE143 antigen recognition and maturation.

We next combined phylogenetic analysis, antibody synthesis, ELISA binding, and HIV-1 neutralization to interrogate the function of GE143-like heavy chains of two putative germline origins. The ML tree of VH4.34-originated heavy chains showed three major branches, each containing sequences from multiple time points (Fig. 6B). The bioinformatics selection identified 15 heavy chains (see Materials and Methods; see also Table S2C in the supplemental material). Twelve of these heavy chains could be expressed when paired with the GE143 light chain and bind to the vaccine antigen with ELISA curves similar to those of both GE143 and a human bNAb (VRC01), with the exception of 3064746_1A (Fig. 6B, bottom right). The antibody reconstituted from 681893_2D, which has the shortest genetic distance to the GE143 heavy chain on the ML tree, bound to gp140-F and neutralized three isolates on a 6-virus panel (Table 3). Our results thus confirmed that these heavy chains, although assigned to a different germline V_H gene, are functionally GE143-like. The ML tree of VH4.40-originated heavy chains was consistent with the oscillating pattern observed from the 2D plots, with sequences from after 2 and 4 immunizations grouped closer to the germline gene and the sequences from after 3 and 5 immunizations occupying a more distant branch (Fig. 6C). Five heavy chains were selected from the phylogenetic tree and paired with the GE143 light chain for experimental validation (see Table S2C). Four reconstituted antibodies were expressed that showed binding to gp140-F (Fig. 6C, top right). When assessed for neutralization on a cross-clade 6-virus panel, they could neutralize up to 3 isolates (Table 3).

In summary, longitudinal tracing of the GE143 lineage revealed an unexpected complexity of F128 memory B-cell responses, with two related sub-lineages likely originating from the same germline gene before diverging into two distinct groups during maturation. Such intralineage diversity is not uncommon for HIV-1 bNAbs, as previously reported for the VRC01-class bNAbs targeting the same site (4, 10, 19, 23), which can develop different HCDR3 loops and up to 30% difference in sequence during maturation. However, this is the first time that such B-cell lineage diversification, albeit at a lower level than that of VRC01, was observed in NHP immunization. It is possible that the oscillating pattern may be related to the short interval in vaccination, although a more detailed comparative study is warranted to confirm this possibility. Therefore, our results suggest that unbiased longitudinal repertoire analysis may assist in the optimization of injection interval, adjuvant, and other critical variables in a vaccine regimen.

Validation of the NHP B-cell lineage patterns.

We have performed a similar lineage tracing analysis for other F128 MAbs. For GE121, an HCDR3 identity cutoff of 95% or greater did not yield any somatic variants in the unbiased repertoires of four time points except for the post-3 repertoire. Six GE121-like heavy chains were found after 3 immunizations, accounting for 0.2‰ of the VH3.8 family, similar to the GE136 lineage (see Fig. S5A in the supplemental material). No GE137 somatic variants were identified in the unbiased repertoire at any time point, indicating a low frequency of this MAb lineage (see Fig. S5B in the supplemental material). For GE147, we observed a rapid lineage expansion between the post-3 and post-4 time points, with the frequency increasing from 4.51 to 53.87‰ in the VH3.24 family (see Fig. S6A in the supplemental material). The GE147 lineage showed a notable decline after 5 immunizations, accounting for only 5.57‰ of the VH3.24 family. Overall, the lineage pattern of GE147 is reminiscent of GE140, even though these two MAbs evolved from different germline origins. For the GE148 lineage, an HCDR3 identity cutoff of 92% revealed an oscillating pattern resembling that of the GE143 lineage despite the different germline origins (see Fig. S6B). Taken together, the lineage patterns observed for MAbs GE136, GE140, and GE143 represent the general patterns of CD4bs-directed memory B-cell responses during F128 immunization.

We next evaluated how experimental factors such as library variation, sequencing variation, and sequencing depth influence the lineage patterns by sequencing different aliquots of the heavy chain cDNA libraries from four time points (see Table S3 in the supplemental material). We first examined the correlation of repertoire properties derived from two sets of PGM experiments, since the unbiased repertoires form the basis of lineage tracing (see Fig. S7A in the supplemental material). A correlation coefficient (R²) of 0.94 or greater was observed for germline gene usage, germline divergence, and HCDR3 loop length. However, we did notice that the average germline divergence (8.3 to 8.7%) is ~1% lower than that obtained previously, likely due to different library pooling strategies and upgraded PGM technologies. We then traced the GE136, GE140, and GE143 lineages in the resequenced heavy chain repertoires (see Fig. S7B to D in the supplemental material). The sequencing depth, with a reduction of 13 to 45%, showed a clear effect on the lineage analysis. The distribution of somatic variants appeared to be sparser on the 2D plots, with no somatic relatives identified for GE136 from the post-3 repertoire, consistent with a 42% reduction in the repertoire size for this time point. For GE140 and GE143, although the lineage patterns remained consistent between the two sets of PGM experiments, the frequencies became less reliable due to the reduced data size. In summary, additional NGS experiments not only confirmed the B-cell lineage patterns but also demonstrated the importance of experimental design (e.g., library pooling and sequencing depth) in antibody lineage analysis.

NHP specimens.

Peripheral blood mononuclear cells (PBMCs) were obtained from F128, a rhesus macaque that has been the subject of intensive study of NHP B-cell response to an HIV-1 trimer vaccine (30, 32, 34, 43). The animal was housed at the animal facility of the Astrid Fagraeus Laboratory at the Swedish Institute for Infectious Disease Control. Housing and care procedures were in compliance with the guidelines of the Swedish Board of Agriculture. The facility has been assigned an Animal Welfare Assurance number by the Office of Laboratory Animal Welfare at the National Institutes of Health. All procedures were approved by the Local Ethical Committee on Animal Experiments. A detailed description of the immunization experiment was previously reported (30). Essentially, the regimen consists of purified gp140-F trimers in adjuvant inoculated five times at monthly intervals.

Sample preparation for PGM sequencing.

The 5′ RACE PCR protocol developed for library preparation from human PBMCs was modified to prepare NHP antibody libraries (41). Briefly, total RNA was extracted from 10 to 20 million PBMCs into 30 µl of water with an RNeasy minikit (Qiagen). For unbiased antibody repertoire analysis, 5′ RACE was performed with a SMARTer RACE cDNA amplification kit (Clontech). The immunoglobulin (Ig) PCRs were set up with Platinum Taq high-fidelity DNA polymerase (Life Technologies) in a total volume of 50 µl, with 5 µl of cDNA as the template, 1 µl of 5′ RACE primer, and 1 µl of 10 µM reverse primer. The 5′ RACE primer contained a PGM P1 adaptor, while the reverse primer contained both a PGM A adaptor and a time point-specific barcode (see Table S1 in the supplemental material). Two Ion Xpress barcodes (Life Technologies), 1 and 2, were used to tag post-2 and -4 samples (or post-3 and -5 samples), respectively. Twenty-five cycles of PCRs were performed, and the expected PCR products (600 to 800 bp) were gel purified (Qiagen).

Ion Torrent PGM sequencing of NHP antibody libraries.

The PGM sequencing protocol used for NHP antibody libraries was similar to that for human antibody libraries (41). Briefly, the antibody heavy (H) and light (κ and λ) chain libraries were quantitated using a Qubit 2.0 fluorometer with a Qubit double-stranded DNA (dsDNA) HS (high sensitivity) assay kit and mixed at a ratio of 1:1:1, except for the validation runs, in which only heavy chains were sequenced to confirm the lineage patterns (see Table S3 in the supplemental material). Next, post-2 and -4 (or post-3 and -5) libraries with different barcodes were pooled at a 1:1 ratio. The pooled library was diluted to a concentration of 50 pM prior to the template preparation, which was performed with the isothermal amplification (IA) kit obtained from the Early Access Program. The sequencing experiment was performed on the Ion Torrent PGM sequencer with the PGM Hi-Q 400 kit using an Ion 316 or 318 v2 chip for a total of 1,200 nucleotide flows. Note that an Ion 314 chip was used to sequence the post-4 heavy chain library in the validation run 3 (see Table S3 in the supplemental material). Raw data were processed without 3′-end trimming in the base-calling process to extend the read length (41). An antibody-specific bioinformatics pipeline (see below) was used to filter, correct, and annotate the full-length reads, with a number of metrics measured to ensure the sequence quality (see Fig. S1 and S2 in the supplemental material).

Bioinformatics analysis of antibody NGS data.

An NHP antibodyomics pipeline has been developed based on the framework of a human antibodyomics pipeline (4, 10, 19–21, 41). The NHP antibodyomics pipeline consists of five consecutive steps. Given a data set of NGS-derived NHP antibody sequences, each sequence was (i) reformatted and labeled with a unique index number; (ii) assigned to the V, D (for heavy chain only), or J gene family using the current rhesus macaque germline gene database and an in-house implementation of IgBLAST, with sequences possessing an E value of >10⁻³ for V gene assignment removed from the data set; (iii) subjected to a template-based error correction procedure, in which insertion-and-deletion (indel) errors of fewer than three nucleotides in the V gene segment were detected by alignment to their respective germline gene sequences and corrected; (iv) compared to the template antibody sequences at both the nucleotide level and the amino acid level using a global alignment module in CLUSTALW2 (50); (v) subjected to a multiple-sequence-alignment (MSA)-based procedure to determine complementarity-determining region 3 (CDR3), which was further compared to a set of template CDR3 sequences at the nucleotide level, and to determine the sequence boundary of the V(D)J coding region. For this purpose, the MSAs of representative germline V genes (truncated after the CXR motif) and J genes (truncated before the F/WGXG motif) have been precalculated and calibrated. Using heavy chain analysis as an example, a PGM-derived read would be included in the sequence alignment with 63 truncated V_H genes and separately with 7 truncated J_H genes to determine HCDR3, which should lie between the last column of aligned V_H genes and the first column of aligned J_H regions, and to determine the variable region, which should lie between the first column of aligned V_H genes and the last column of aligned J_H genes. This implicit pattern-matching method was specifically designed to take into account the sequence variation in the CDR3 flanking regions and the termini of variable regions (20). At the end of pipeline processing, a bioinformatics filter was applied to detect and remove erroneous sequences that may contain swapped gene segments due to PCR errors. Specifically, a full-length read would be removed from the data set if the V-gene alignment was less than 250 bp (41). The resulting antibody chain sequences were then subjected to a more in-depth repertoire analysis.

The NHP antibodyomics pipeline is a suite of Perl scripts that can be run on a Linux system with the option for parallel computing to reduce the data processing time. It requires pre-installed BLAST (51) and CLUSTALW2 (50) modules to facilitate germline gene assignment and global sequence alignment, respectively. Pre-compiled NHP germline gene databases are required for heavy and light chain-specific pipeline processing. The software suite can be obtained by request from J. Zhu, with source codes provided for main and auxiliary functions such as sequence quality assessment (see Fig. S1 and S2 in the supplemental material) and repertoire profiling (Fig. 3).

A bioinformatics procedure was devised to select heavy chain somatic variants for phylogenetic analysis, antibody synthesis, and characterization. From the NGS data obtained for each time point, we identified heavy chains with an HCDR3 identity of 95% or greater to a given MAb and subjected them to a filter that selects only for heavy chains possessing two conserved cysteines in the framework and the CXR-WGXG motif that flanks the HCDR3 region. Due to the complexity of GE143 lineage, a lower HCDR3 identity of 92% was used to include remote somatic variants. After filtering, a clustering procedure with an identity cutoff of 99.0% was used to further reduce sequence redundancy. The representative sequences of resulting clusters from four time points, if available, were combined into a single data set for further manual inspection.

The putative germline gene sequence was added to each data set, and a multiple sequence alignment (MSA) generated by CLUSTALW2 (50) was used as the input to construct a phylogenetic tree using DNAMLK (http://evolution.genetics.washington.edu/phylip/doc/dnamlk.html) in the PHYLIP package v3.69 (http://evolution.genetics.washington.edu/phylip.html). The displayed phylogenetic trees for the three MAb lineages (Fig. 4 to 6) were generated using Dendroscope (52), ordered to ladderize right, and rooted by their respective germline genes.

Antibody expression.

Antibody production was performed as previously described (43). Briefly, the antibody sequences selected by bioinformatics were synthesized (Invitrogen, Inc.) and cloned into the cytomegalovirus R (CMV/R) expression vector containing the constant regions of NHP IgG. The heavy and light chains identified from PGM sequencing of F128 repertoires at multiple time points were paired with their respective partner chain DNAs. Full-length IgGs were expressed by transient transfection of 293F cells and purified using a recombinant protein A column (Pierce). The expression and other information for PGM-derived antibodies are summarized in Table S2 in the supplemental material.

HIV-1 neutralization assays.

Neutralization assays were performed as previously described (32). Briefly, neutralization was measured using HIV-1 Env pseudoviruses and TZM-bl reporter cells (53–55). A six-virus panel was used in the current study. Neutralization curves were fit by a nonlinear regression analysis using a 5-parameter Hill slope equation (55). The 50% inhibitory concentration (IC₅₀) is defined as the antibody concentration required to inhibit HIV-1 infection by 50%.