Early Antibody Lineage Diversification and Independent Limb Maturation Lead to Broad HIV-1 Neutralization Targeting the Env High-Mannose Patch

A Broad Neutralizing Response Targeting the N332 Supersite Was First Detected at 33 Months Post-Infection in Donor PC76

We characterized the breadth and potency of the serum neutralizing activity in 439 of a total of 613 IAVI (International Aids Vaccine Initiative) Protocol C participants (Landais et al., 2016). Development of neutralizing activity in donor PC76 was examined with serum samples from 5 to 38 months post-infection (mpi) tested against a cross-subtype panel of 37 viruses. Neutralization of heterologous viruses was observed starting at 5 mpi and reached a peak of 68% of viruses neutralized by 33 mpi (Figures 1A and S1A). Serum neutralizing activity was N332 dependent for most of the viruses tested (Figure S1B) and throughout this time course (Figure 1B), confirming that N332-dependent Ab responses were predominantly responsible for the broad neutralizing activity developing in this donor.

Next, serum from 33 mpi was adsorbed on wild-type (WT) recombinant gp120 (rgp120) and rgp120 with a point mutation eliminating the glycan at position N332 (N332A) (Figures 1C and S1C). Serum depleted of Abs that bound WT rgp120 lost neutralizing activity, whereas N332A-rgp120 adsorbed serum was comparable to the negative control BSA adsorbed serum. This result confirmed that neutralizing Abs present in the serum exhibited differential binding to WT and N332A rgp120 proteins.

A Lineage of N332-Glycan-Targeting Abs

On the basis of the above results, we designed a strategy to isolate B cells expressing B cell receptors specific to the N332 glycan from peripheral blood mononuclear cell (PBMC) samples of donor PC76 (Figure S2). To exclude strain-specific or non-specific binding, WT rgp120 probes from virus strains 92BR020 (subtype B) and IAVIC22 (subtype C), which both depleted neutralizing Abs from the PC76 serum (Figures 1C and S1C), were used to identify B cells binding to both proteins. The 92BR020 N332A rgp120 probe was used to isolate cells that bound only when the N332 glycan was present. Single live N332-specific IgG⁺ B cells identified as described in Figure S2 were sorted into plates for single-cell PCR amplification and sequencing.

A lineage of 12 somatically related monoclonal Abs (mAbs) from a total of five time points from 16 to 38 mpi was identified from donor PC76. Selected features of PCDN (Protocol C, lab code DN) mAbs are summarized in Figure 2A. This Ab lineage used the IGHV4–34*01 heavy chain (HC) variable gene segment (V-gene) and IGHJ5*01 joining gene segment (J-gene). The diversity gene segment (D-gene) usage could not be reliably determined. The isolated mAbs had CDRH3 lengths of 22 amino acids (aa). The HCs of the mAbs showed ~4%–5% nucleotide (nt) mutation from the GL (~8% aa) at 16 mpi and matured to ~16% nt mutation (~26% aa) for a mAb isolated at 38 mpi. Light chains (LCs) used genes IGKV3–20*01 and IGKJ1*01 and ranged from ~5%–11% nt mutation. Phylogenetic analysis of the mAb HC and LC sequences (Figure 2B) and aa sequence alignments (Figure S3) showed the sequences maturing correspondingly over time. In contrast to all previously isolated bnAbs that target the high-mannose patch, no indels were introduced into either the HC or LC sequences during affinity maturation.

The PCDN Ab Lineage Matures toward Heterologous Neutralization and is Dependent on Specific Glycans

The PCDN mAbs were characterized for binding to rgp120s of various isolates and different subtypes. All mAbs exhibited preferential binding to WT rgp120 and reduced affinity for N332A rgp120, as measured by ELISA (Figure 3A). The Abs were then tested for neutralization of a cross-subtype panel of 37 pseudoviruses (Figures 3B and S4A). As anticipated, Abs isolated at the earliest time point, PCDN-16A and 16B, had limited neutralization breadth, 5% and 11% of viruses, respectively. However, consistent with the kinetics of the development of serum neutralization (Figure 1A), Abs isolated at 33 and 38 mpi neutralized 65%–70% of HIV-1 strains.

We tested the three most mature Abs, PCDN-33A and PCDN-38A and 38B, on a 110-virus panel for a better representation of globally circulating HIV-1 strains. Potency and breadth were reduced overall, primarily due to the different composition of the large panel (Figures 3C and S4B), which is more diverse and includes a greater number of subtype AE recombinants. The majority of subtype AE viruses (91%) have an N334-glycan site instead of N332 (Sok et al., 2014a); <10% were neutralized by PCDN mAbs. PCDN Abs failed to neutralize all viruses that lacked the N332 glycan site, and viruses with a Tyr at position 330, instead of the more common His, were also not neutralized, pointing to N332 and H330 as critical residues for neutralization by these mAbs.

Specific glycan types are essential for binding of Abs targeting the N332 glycan region, and promiscuity in glycan usage has been observed for some bnAbs (Mouquet et al., 2012; Sok et al., 2014a). PCDN-33A and PCDN-38A and 38B were tested against pseudoviruses produced in cells treated with glycosidase inhibitors, kifunensine or swainsonine, or prepared in UDP-N-acetyl-D-glucosamine: α-3-D-mannoside β1 → 2-N-ace-tylglucosaminyltransferase I (GnTI)-deficient HEK293S cells to determine their dependency on certain glycan types. Although all three Abs were sensitive to kifunensine, neutralization by PCDN-38A and PCDN-38B was more severely affected than that by PCDN-33A (Figure 3D). Neutralization was lost for most HEK293S-produced viruses for all three Abs. Swainsonine treatments mostly had no effect, except on neutralization of a few viruses by PCDN-38A and PCDN-38B (Figure 3D). Swainsonine does not affect hybrid glycans but blocks processing to complex glycans, and GnTI deficiency leads to loss of both hybrid and complex glycans (Doores and Burton, 2010). Therefore, in most cases, PCDN mAbs seem to require hybrid glycans for neutralization. The direct binding of PCDN mAbs to glycans was also tested with a glycan array. Of the tested glycans, all PCDN mAbs bound only to Man₉GlcNAc₂ and no other detectable binding was observed (Figure 3E), whereas PGT128 bound to Man₉GlcNAc₂ and Man₈ as previously reported (Walker et al., 2011). No binding to hybrid glycans was observed, suggesting that although they are required for neutralization, the affinity of the Abs for hybrid glycans is not high enough to detect their binding to the array. Thus, both high-mannose and hybrid glycans appear to be important for binding of the PCDN mAbs.

To map the residues involved in recognition of Env, Ala substitution mutants of JR-CSF Env were made and neutralization was assessed with the three most mature PCDN Abs because only these were capable of neutralizing JR-CSF. Mutations to the potential N-glycosylation sites (PNGSs) of N332 and N301 glycans completely abolished neutralization (Figure S4C). N156A, S158A, and F159A also led to loss of neutralization, suggesting that the N156 glycan is involved in Ab binding. In contrast, little to no sensitivity to mutations at the N295 PNGS was observed for JR-CSF. Similar glycan-site dependencies were observed for other heterologous viruses (Figure S4D). Mutations to Y177 and L179 in V2, as well as a number of residues in V3, also had moderate effects, including those of the ³²⁴GDIR³²⁷ motif, which has been reported to make extensive contacts with PGT124 and PGT128 (Garces et al., 2014), and nearby residues. Mutating I307 had a strong impact in particular on PCDN-38A and PCDN-38B; this residue has been suggested to be part of a conserved hydrophobic core structure with I309 and F317 (Jiang et al., 2010). The Ala scan results show that mutation of these residues can affect the ability of PCDN Abs to recognize Env.

Some HIV bnAbs have been shown to exhibit polyreactivity and/or autoreactivity, and it has been proposed that this unusual feature contributes to the development of broadly neutralizing activity (Haynes et al., 2005; Liu et al., 2015). We sought to determine whether the PCDN Abs exhibited auto- or polyreactivity as measured by ELISA against a panel of antigens previously described (Wardemann et al., 2003). Histones were bound by a few of the mAbs (22A and 27C), but the 33 mpi and later Abs did not show any detectable polyreactivity (Figure S4E). We also measured binding to HEp-2 cells, and PCDN-27D was the only one that showed detectable fluorescent signals (Figure S4F). Overall, the PCDN mAbs demonstrated weak to no auto- or poly-reactivity, with the exceptions described.

Autologous Virus and the bnAb Lineage Co-evolved within Donor PC76

Changes in Env sequence can provide information as to how the virus adapted to the immune response after infection and might have driven the maturation of a bnAb lineage. A total of 76 autologous env genes were cloned and sequenced from PC76 plasma at multiple time points. A phylogenetic analysis of the virus full-length env sequences shows their course of evolution (Figure S5A).

Viruses pseudotyped with the autologous env were tested for neutralization by the 12 mAbs (Figure 4A). The viruses from 5 mpi were not neutralized by any of the earliest mAbs (from months 16 and 22), which neutralized only viruses from 10 mpi. The month-16 viruses were no longer neutralized by these Abs, demonstrating viral escape. This suggests that the lineage was triggered by viruses emerging between months 5 and 10. When the 5 mpi virus was tested for neutralization by serum, all but the 5 mpi serum were able to neutralize the virus, supporting this time frame (Figure S5B). All Abs isolated past 33 mpi displayed both strong autologous breadth and potency, neutralizing almost all of the viruses tested up to month 22. The first mAb that showed autologous breadth but weak potency was isolated at 27 mpi (PCDN-27B). Interestingly, these Abs were able to neutralize the 5 mpi viruses, in contrast to the earliest Abs. However, from 27 mpi on, most of the viruses fully escaped all PCDN mAbs.

When the Env aa changes were examined, seven out of nine clones from 10 mpi simultaneously mutated E328 to a Lys while deleting a glycan site at N335 by mutating T337 to an Ala (Figure 4B). The N335 PNGS is present in <1% of virus strains on the Los Alamos HIV Sequence Database, making this PNGS a unique feature of the PC76 autologous virus. In contrast to N334, N335 could potentially be glycosylated together with N332. This E328K mutation was then further mutated to an Ala in all of the viruses isolated from 16 mpi and later. The N335 PNGS was initially reverted back in most clones at 16 and 19 mpi and then removed again in 26 of 28 virus clones from 22 mpi and later. All of the viruses from 27 mpi and later that were not neutralized by any of the PCDN Abs had removed the N332 PNGS (N332K or S334N), which was likely the final route of escape.

To test whether these residues indeed had a role, mutagenesis of the autologous virus from 5, 10, and 19 mpi was performed (Figure 5A). When the 5 mpi virus was mutated to have both E328K and N335A mutations, some early Abs were now able to neutralize the virus. When the mutations were introduced individually, this increased sensitivity was not observed, suggesting the combination of mutations contributed to the recognition of the virus by the early PCDN lineage Abs. If the 10 mpi virus was either reverted to the 5 mpi form, K328E, or mutated to the later form, K328A, neutralization by all the early Abs was knocked out, demonstrating escape, whereas PCDN-27B and later Abs were still able to neutralize these viruses. However, when the 19 mpi virus was reverted to A328K with the N335 glycan, neutralization was not restored. Reversal of an escape mutation is more difficult because there might be additional mutations involved that are not in the vicinity of the epitope and harder to pinpoint. However, when the N332 PNGS was removed from the 19 mpi virus, neutralization by all the PCDN Abs was completely lost, which was also the case for N332 PNGS mutations of earlier viruses. Once the N332 PNGS was eliminated, the virus completely escaped neutralization by the PCDN Ab lineage, consistent with all the heterologous virus neutralization results (Figure S4B).

The 10 mpi virus was neutralized by most of the PCDN Abs, allowing for mutagenesis studies to map the change in contact residues over time. Mutations to PNGSs showed that N295, N301, and N332 were critical (Figure 5B). Loss of glycosylation sites at N301 and N332 knocked out neutralization for all the Abs, although N295 PNGS mutation did not greatly affect PCDN-27B and late Abs. Loss of the N137 glycan site in the 10 mpi virus potentiated all the Abs except those that didn’t neutralize this virus to start with. An Ala scan of the N332-proximal V3 region was also performed. Mutations to residues D322 to K328 knocked out or weakened neutralization by the majority of the early Abs, with the exception of I323A and I326A, which had little to no effect. The 33 mpi and later Abs were largely unaffected by these mutations. The H330A mutation knocked out PCDN-27B neutralization and reduced the potency of later Abs. Similarly to the results from the 110-virus panel and JR-CSF alanine scan (Figures S4B and S4C), this suggests that the later Abs matured to become less dependent on residues in the GDIR motif but still rely on N332 and N301 glycans and H330 to interact with Env.

NGS of the PCDN Ab Lineage

To examine the longitudinal maturation of the PCDN Ab lineage, we performed NGS of Ab HCs by using RNA isolated from approximately 10⁶ PBMCs from time points 5–66 mpi. Experimental details can be found in Supplemental Experimental Procedures and Figure S6A. Raw Ab sequences were corrected with molecular barcodes and GL gene assignments were made with AbStar (data not shown). Ab lineages were assigned with Clonify (Briney et al., 2016), and the PCDN lineage was identified by a homology search with the panel of PCDN mAbs.

Although lineage members were identified at all time points between 16 and 44 mpi, there was a pronounced peak at 22 mpi, immediately prior to the development of neutralization breadth and potency (Figure 6A). The subsequent drop might be due to the virus escaping neutralization by this lineage around 27 mpi (Figure 4A). To verify that this frequency peak was not an artifact of sequencing or pre-sequencing amplification, we amplified and sequenced additional longitudinal samples by using different sample preparation methods (5′ rapid amplification of cDNA ends [RACE] and alternate gene-specific primers) and observed the same frequency pattern in each case (Figure S6B).

NGS identified a PCDN lineage sequence with high homology to the GL V- and J-genes, bearing a single aa difference in the J-gene. This least mutated sequence was selected as the unmutated common ancestor (UCA) for HC (Figure S3). A UCA IgG was expressed by pairing the HC with an inferred LC that was reverted directly to GL. The UCA IgG had only minimal binding to some heterologous rgp120 proteins and no binding to the autologous rgp120s tested by ELISA (Figures S6C and S6D). In addition, no neutralization was observed for the UCA for any of the viruses of the 37-virus panel or any autologous viruses tested (data not shown).

Phylogenetic analysis of the PCDN lineage (Figure 6B) revealed multiple large limbs that diverged early in the lineage lifespan, almost as soon as the lineage was initiated, and independently matured in parallel. A genetic divide was observed in the lineage phylogeny, beyond the peak of frequency described above, where most branches did not develop beyond 27 mpi and seemingly reached a dead end. This suggests the occurrence of a virus-induced lineage-sieving event near 27 mpi due to viral escape and a need for early intra-lineage diversification to achieve lineage longevity. Interestingly, more than one of the limbs led to Abs capable of heterologous neutralization, including the ones leading to the broad and potent Abs, PCDN-33A and PCDN-38A and 38B. In addition, sequencing identified another long-lived lineage branch that might have developed breadth and from which we were unable to isolate Abs by B cell sorting. Therefore, within this single donor, multiple developmental pathways led to Abs that acquired breadth. Together, early diversification and parallel maturation appear to be central to the development of a broad Ab response.

When analyzing the lineage phylogeny, we noticed a substantial discrepancy between the phylogenetic evolution of the lineage and the divergence from GL over time. Although GL divergence (GD), measured as the percent nucleotide change from UCA, appeared to plateau at 33 mpi (top panel, Figure 6C), the tree-adjusted evolutionary distance (ED) from UCA continued to increase after 33 mpi (middle panel, Figure 6C). To quantify this discrepancy, we calculated the ratio between the GD rate of mutation and tree-adjusted ED rate of increase (bottom panel, Figure 6C). Until 33 mpi, the GD:ED ratio remained relatively constant at approximately 0.01, suggesting that GD roughly approximated total ED. After 33 mpi, however, there was a sharp increase in the ED:GD ratio, which indicated that GD was underestimating total ED. Because both were calculated at the nucleotide level, this result suggests repeated somatic mutations at the same position; re-mutating a previously mutated site would increase ED but would not change GD. Presumably, the number of activation-induced cytidine deaminase (AID) hotspot sequences, WRCY (Martin et al., 2002), should decrease with continued SHM. When we assessed the number of AID hotspots for the PCDN lineage over time, we observed a steady decrease that was not seen for a random sample of Ab sequences (Figure S6E). Therefore, as an Ab lineage matures, SHM appears to continue to occur at previously mutated sites despite the loss of hotspot motifs, suggesting a strong selection pressure for these positions.

Crystal Structures of PCDN Fabs

The structural evolution of the PCDN lineage Abs could provide explanation of how they mature to gain breadth or potency. PCDN-27A and PCDN-27B are situated on the same branch of Ab maturation (Figure 6B), yet 27A has limited heterologous neutralization in comparison to 27B (Figure 3B).

To understand this functional leap, we determined the crystal structures of PCDN-27A and PCDN-27B Fabs (Figure 7A). Both Fabs have a protruding “dagger”-like shape with the CDRH3 “blade” rising ~16 Å above the other CDR loops (Figure 7A). The extended CDRH3 will most likely penetrate the glycan shield and establish contacts with the underlying protein motif. The aromatic residues positioned on both sides of the CDRH3 loop (Figure S7A), Y103, Y106, and W107, could be interacting with the Env ³²⁴GDIR³²⁷ residues or the carbohydrates at the high-mannose patch through stacking interactions similar to PGT124 (Garces et al., 2014). The variable HC and LC regions of both PCDN-27A and PCDN-27B Abs align well with a Cα RMSD (root-mean-square deviation) of 0.49 Å (Figure 7A). However, the tip of the CDRH3 in PCDN-27B appears to be tilted ~10 Å in comparison to PCDN-27A. Before relating this difference to functionality, we investigated the crystal packing of both Fabs. When the structure of PCDN-27A is superimposed on that of PCDN-27B within the crystal lattice, the CDRH3 loop of 27A would clash with the constant region of a symmetry-related molecule of 27B (Figure S7B). Thus, the slight variation in the CDRH3 loops between the two Fabs is probably due to crystal packing and is likely not the cause of the observed greater breadth and potency of PCDN-27B.

According to the sequence alignments for PCDN-27A and PCDN-27B (Figure S3), there were no differences between the two in CDRH1 or CDRH3, whereas the FRH3s were distinct. Although this sequence variability was not translated into structural variation, the electrostatic surface indicated a shift from residues with negative charges in PCDN-27A to positive charges in PCDN-27B (Figure 7B). This suggests that FRH3 might be important for neutralization of both autologous and heterologous viruses. In addition, we noticed two PNGSs in FRH1 (N26) and FRH3 (N68) of PCDN-27B. Density was observed in the crystal structure at N26, confirming the presence of an N-linked glycan, but not at N68, due to either disorder or to non-use of the PNGS (Figure S7C). However, from the position of the N26 glycan in the Fab structure, it seems unlikely that this glycan would contribute to the greater breadth of PCDN-27B than PCDN-27A.

We compared the structural features of PCDN-27A against the other N332-targeting bnAbs for which crystal structures have been solved (Garces et al., 2014; Kong et al., 2013; Pejchal et al., 2011). The Cα RMSD was lowest for PGT135, showing overall the greatest structural similarity (Figure S7D). However, the CDRH3 of PCDN-27A (22 aa) is longer than that of PGT135 (18 aa), and PCDN-27A does not have any insertions, unlike PGT135, which has a 5 aa insertion in CDRH1.

Human Specimens

Serum and PBMC samples were obtained from donor PC76, enrolled in the IAVI Protocol C longitudinal HIV-1 infection cohort. IAVI Protocol C enrolled 613 individuals from sub-Saharan Africa who were rapidly screened for HIV Abs after recent exposure to HIV-1. Samples were collected with written informed consent, and clinical protocols were approved by the appropriate ethics and research committees.

Neutralization Assays

Neutralization assays were performed by incubating Abs or sera with HIV-1 pseudoviruses that are capable of one round of infection to TZM-BL cells (Walker et al., 2009). Serum depletion was performed with Tosylactivated Dynabeads (Life Technologies) conjugated with BSA or with WT and N332A 92BR020 rgp120 proteins. Serum samples were depleted of Abs binding to these proteins through multiple rounds of immunoprecipitation (Li et al., 2009) as confirmed by ELISA, prior to use in neutralization assays.

Single Memory B Cell Sorting and Ab Isolation

Single antigen- and epitope-specific memory B cells were sorted into wells of 96-well PCR plates, RT-PCR was performed with multiplex PCR methods and primers, then products were sequenced and cloned into HC or LC constant-region-encoding vectors (Sok et al., 2014b; Tiller et al., 2008; Wu et al., 2010).

IgG Ab and Fab Fragment Production

Full-length or truncated Fab HC constructs and LC expression plasmids were transfected into HEK293F cells with the FreeStyle 293 Expression System (Life Technologies) to produce Abs and Fab fragments and were purified as described in the Supplemental Experimental Procedures.

Glycan Array Preparation and Ab Hybridization

40 glycans were printed by a robotic pin (SMP3; TeleChem International) and 0.6 nL of 100 µM concentration of amine-functionalized glycans in printing buffer ([pH 8.5] 300 mM phosphate buffer, 0.05% Tween-20) from a 384-well plate was deposited onto NHS-coated glass slides that were then blocked and washed before being incubated with Abs (10 µg in 50 µl PBS) pre-complexed with secondary Ab. The slides were washed and scanned on an Array-Worx microarray reader. Images were analyzed with Genepix Pro 6.0 (Molecular Devices Corporation).

Ab NGS and Computational Analysis

RNA was prepared either from 1 × 10⁷ whole PBMCs or IgG⁺ memory B cells separated from PBMCs (Switched memory B cell isolation kit; Miltenyi Biotec). RNA was subjected to RT-PCR with barcoding primers that contain unique Ab identifiers, and Illumina sequencing adapters and sample-specific indexes were added during a second round of PCR; primer sequences can be found in Table S1. Samples were quantified with fluorometry (Qubit; Life Technologies) and pooled at approximately equimolar concentrations, and the sample pool was requantified to be loaded onto an Illumina MiSeq. Paired-end MiSeq reads were merged with PANDAseq (Masella et al., 2012). GL assignment, junction identification, and other basic Ab information was determined with AbStar, an antibody analysis package that uses BLASTn and Smith-Waterman alignment to calculate the most likely GL genes, identify the junctional region, and annotate SHM-induced substitutions and indels (B.B., unpublished data). Sequences were assigned to clonal lineages with Clonify, a software package specifically developed for Ab lineage assignment (Briney et al., 2016). We constructed the phylogenetic trees with FastTree by using maximum likelihood.

Crystal Diffraction Data Collection and Computational Analysis

Data were collected at Stanford Synchrotron Radiation Lightsource (SSRL) 12-2 beamline. The best dataset for the PCDN-27A crystal was collected at 2.7 Å resolution, and the diffraction data were processed with HKL-2000 (Otwinowski and Minor, 1997) to an overall R_sym of 0.12% and completeness of 93.9% (Table S2) in space group P2₁ with unit cell parameters a = 41.8 Å, b = 120.4 Å, c = 111.1 Å, and β = 96.4°. Data were collected from a PCDN-27B crystal to 2.9 Å resolution and processed with HKL-2000 (Otwinowski and Minor, 1997) to an overall R_sym of 0.28% and completeness of 99.5% in space group P6₁22 with unit cell parameters a = b = 89.3 Å, c = 211.1 Å. The PCDN-27A Fab structure was solved by molecular replacement with Phaser and a complementarity determining region (CDR)-loop-deleted X5 Fab structure (PDB: 1RRH) was used as the initial model, whereas the PCDN-27B Fab structure was solved with the family-related PCDN-27A Fab structure. Model building was carried out as previously described (Garces et al., 2014). Final R_cryst and R_free values for PCDN-27A and PCDN-27B Fabs structure are 21.3% and 24.7%, and 23.3% and 29.7%, respectively.

Detailed additional experimental procedures can be found in the Supplemental Experimental Procedures.