Mapping the immunogenic landscape of near-native HIV-1 envelope trimers in non-human primates

The induction of broad and potent immunity by vaccines is the key focus of research efforts aimed at protecting against HIV-1 infection. Soluble native-like HIV-1 envelope glycoproteins have shown promise as vaccine candidates as they can induce potent autologous neutralizing responses in rabbits and non-human primates. In this study, monoclonal antibodies were isolated and characterized from rhesus macaques immunized with the BG505 SOSIP.664 trimer to better understand vaccine-induced antibody responses. Our studies reveal a diverse landscape of antibodies recognizing immunodominant strain-specific epitopes and non-neutralizing neo-epitopes. Additionally, we isolated a subset of mAbs against an epitope cluster at the gp120-gp41 interface that recognize the highly conserved fusion peptide and the glycan at position 88 and have characteristics akin to several human-derived broadly neutralizing antibodies.


Introduction 46
HIV-1 continues to cause significant morbidity and mortality around the world with an estimated 47 1.7 million new infections in 2018 1 , which emphasizes the need for an effective prophylactic 48 vaccine. The HIV-1 envelope glycoprotein (Env) is the sole target for neutralizing antibody (NAb) 49 responses. Studies of infected patients have led to the isolation of NAbs against multiple different 50 epitopes on the Env surface that are capable of both neutralizing most circulating strains and 51 providing passive protection against repeated viral challenges in non-human primates (NHPs) 2-5 . 52 Extensive research efforts, including structure-based engineering of Env immunogens, are 53 currently directed towards developing vaccine strategies to successfully elicit broadly 54 neutralizing antibodies (bNAbs) against specific Env epitopes 6-12 . The development and structural 55 determination of soluble native-like Env trimer mimics, particularly ones based on the SOSIP 56 technology, has provided a platform for structure-based immunogen design [13][14][15][16] . NAbs induced 57 by SOSIP trimers in NHPs can protect against challenge with an autologous Simian-Human 58 Immunodeficiency virus (SHIV) 17,18 . However, as NAbs with the required breadth of activity have 59 not yet been induced in trimer-immunized animals, improvements to current vaccine design and 60 delivery strategies are clearly needed. 61 Characterizing the antibody response to SOSIP trimers may provide useful information for 62 guiding immunogen design. Initial analyses of rabbits immunized with BG505 SOSIP.664 trimers, 63 including studies of isolated monoclonal antibodies (mAbs), showed that autologous NAbs 64 targeted a large hole in the glycan shield of the BG505 virus caused by the absence of 65 glycosylation sites at positions 241 and 289 19,20 . The 241 glycan is highly conserved (97%) among 66 HIV-1 strains and while the 289 glycan is less conserved, it is still present in 79% of viruses. The (nucleotide level) was 6.4% (range: 2.1%-10.2%) with an average HC complementarity-126 determining region 3 (CDR-H3) length of 15 amino acids (aa) (range: 7-23) ( Table S2). The rh1987 127 KC mAbs utilized HC variable genes from the IGHV3 and IGHV4 families and predominantly used 128 KC V genes from the IGKV1 family (Table S2). All of the rh1987 KC mAbs had a CDR-L3 length of 9 129 aa and their average KC SHM (nucleotide level) was 4.7% (range: 2.6%-6.0%) (Table S2). A single 130 clonal family with 4 members (RM19A) was detected among rh1987 KC mAbs with members 131 isolated from both week 22 and week 25 samples (Table S2). The rh1987 LC mAbs used HC V 132 genes from the IGHV1, IGHV3 and IGHV4 families and LC V genes mainly from the IGLV2 gene 133 family (Table S2). The rh1987 LC mAbs had an average CDR-L3 length of 10 aa (range: 9-11) with 134 an average LC SHM (nucleotide level) of 3.8% (range: 0.9%-10.6%) (Table S2) (Table S2). 137 For animal rh2011, 8 KC and 9 LC mAbs were isolated. Their average HC SHM rate 138 (nucleotide level) was 6.1% (range: 3.0%-9.1%) and they had an average CDR-H3 length of 17 aa 139 (range 10-20) (Table S2). Half of the rh2011 KC mAbs belonged to the RM20E clonal family, 140 isolated from the week 53 sample. The RM20E clonal family utilized the HC V gene IGHV5-141 ABI*01_S2502 and the KC V gene LJI.Rh_IGKV2.71 (Table S2). Overall, the rh2011 KC mAbs had 142 an average KC SHM rate (nucleotide level) of 4.1% (range: 3.0%-5.3%) and a CDR-L3 length of 9 143 aa (Table S2). Among the rh2011 LC mAbs, two clonal families (RM20A [4 members] and RM20B 144 [2 members]) were isolated from weeks 22 and 25 samples (Table S2). Overall, the rh2011 mAbs 145 had an average LC SHM rate (nucleotide level) of 4.4% (range: 2.1%-6.4%) and an average CDR 146 L3 length of 10.6 aa (range: 9-11) (Table S2). 147

BG505-specific mAbs recognize multiple Env regions 149
All 42 mAbs bound to the BG505 SOSIP.664 trimer in ELISA, but only 11 of the 25 mAbs from 150 rh1987 and 9 of the 17 from rh2011 bound the corresponding gp120 monomer (Fig S1). The mAbs 151 were tested for neutralization activity against the autologous BG505 clade A Tier 2 virus, its 152 glycan-611 knockout variant (N611A), and the heterologous SF162 clade B Tier 1A virus. Only a 153 few mAbs, 4 from rh1987 and 2 from rh2011, neutralized the BG505.T332N pseudovirus but one 154 of them, RM20F from rh2011, did so potently ( Fig 1C). Two mAbs from rh1987 and 4 from rh2011 155 were able to potently neutralize the N611A-variant despite lacking activity against the autologous 156 BG505.T332N pseudovirus (Fig 1C and D). None of the 42 mAbs neutralized the easy-to-neutralize 157 heterologous SF162 virus (Fig 1C and D). 158 159 EM-based epitope mapping revealed mAbs isolated from both animals target 4 distinct, but 160 somewhat overlapping epitopes 161 We used low resolution, negative stain, single particle electron microscopy (EM) to visualize 162 where a representative subset of the isolated mAbs bound on the surface of the BG505 SOSIP 163 trimer. The majority (55%) of mAbs isolated were non-neutralizing antibodies that bound to the 164 base of the BG505 SOSIP trimer (Figs 2A, S2, and Table S2) at a neo-epitope cluster that is 165 occluded by the viral membrane on HIV-1 virions. Fabs bound to the base of the soluble trimer 166 via multiple angles of approach and utilized a variety of heavy and light chain genes/alleles to do 167 so (Fig 2A and Table S2). The extent of SHM in the base-targeting mAbs ranged from 2-10% in the 168 responses elicited by the BG505 SOSIP trimers in rabbits and RMs have shown that epitopes at 170 the base of the soluble trimer were targeted in every single animal analyzed 21 Table S2). Multiple germline genes/alleles were used 182 to target the same 289-glycan hole epitope (Table S2). 183 To further assess the epitopes targeted following immunization with the BG505 184 SOSIP.664 trimer and verify that we isolated mAbs representative of the full serum antibody 185 response, we performed electron microscopy polyclonal epitope mapping (EMPEM) 36 using week 186 28 serum. EMPEM revealed that similar epitopes were targeted in both animals and that the 187 epitope assignments correlated well with the epitopes ascribed to mAbs generated by antigen-188 specific B-cell sorting (Fig 2E versus S2). 189 A previous analysis of purified serum IgGs from RMs rh1987 and rh2011 identified the 190 C3/V5 epitope as a major target for neutralization activity 20 ; however, none of the mAbs isolated here targeted the C3/V5 epitope nor were they detected by EMPEM. While low resolution, 192 negative stain EM provides valuable information on where mAbs bind on the surface of HIV-1 193 Env, the molecular detail necessary to guide structure-based immunogen design requires high-194 resolution structural data obtained by cryoEM and x-ray crystallography. We therefore selected 195 three Fabs (RM20J, RM20F, and RM20E1) that bound to different epitopes for high-resolution 196 structural determination. We solved a 2.3 Å crystal structure of unliganded RM20J Fab and a 3.9 Å cryoEM structure 201 of RM20J Fab bound to the BG505 SOSIP Env trimer (Figs 3A and S3; Tables S4 and S5.). Together 202 these structures revealed the RM20J Fab binds to an epitope on a single gp120 protomer with 203 982 Å² of buried surface area (BSA). The CDR-H1 and CDR-H2 make contact with the C2 region of 204 gp120 including residues N289 and T290 (Fig 3B). A glycan at position N289 would directly clash 205 with both the CDR-H1 and CDR-H2 of RM20J ( Fig 3B). CDR-L2 makes contact with the first N-206 acetyl glucosamine sugar of the N355 glycan ( Fig 3C). Additional contacts are made to the α2 207 helix of gp120 by RM20J CDR-H3 and CDR-L2 ( Fig 3C). When compared to 10A, a previously 208 characterized 241/289 glycan hole targeting NAb isolated from a BG505 SOSIP.664-immunized 209 rabbit 19,36 , RM20J binds to an epitope biased more towards 289 and away from 241 in the 210 241/289 glycan hole, revealing subtle differences in the recognition of the epitope (Figs 3D and 211 3E). Despite binding to the BG505 SOSIP trimer with high affinity (Table S3), RM20J was not able 212 to neutralize the autologous BG505.T332N pseudovirus (Fig 1C). Although the hypervariable region of V4 was not resolved in the trimer structure, it lies directly above the RM20J epitope 214 and contains two additional glycans (N406 and N411) that may affect RM20J binding. 215 Comparisons between the glycosylation profiles of the BG505 viral Env and the SOSIP.664 trimer 216 revealed differences in the glycoforms present at positions N355, N406, and N411 37,38 with more 217 complex glycans being found on the viral Env that could hinder the ability of RM20J to bind on 218 the surface of the virus and, therefore, render it incapable of neutralization. Several of the mAbs 219 isolated from rh1987, including the RM19A clonal family, bind to a similar epitope as RM20J (Fig  220   2D, Fig S2, Table S2) and either fail to neutralize the autologous BG505 virus or do so with weak 221 potency ( Fig 1C and D). 222 223 MAb RM20F binds to a quaternary epitope at the gp120/gp41 interface that includes elements 224 of the fusion peptide and the N88 glycan 225 For a more detailed view of the mode of RM20F recognition, we solved a 2.2 Å crystal 226 structure of unliganded RM20F Fab and a 4.3 Å cryoEM structure of RM20F Fab bound to BG505 227 SOSIP trimer (Figs 4A and S3; Tables S4 and S5). RM20F recognizes an epitope spanning two gp41 228 protomers and a single gp120 protomer that has 1126 Å² of BSA at the interface. The RM20F LC 229 contributes 22% of the paratope surface area (250 Å²) and makes contact with the poorly 230 conserved residues H85 (8.1% prevalence among global strains) and K229 (12.5% prevalence) in 231 the C1 and C2 regions of gp120 respectively (Fig 4B). The RM20F HC contributes the remaining 232 78% of the paratope surface area (876 Å²) and uses its 20 residue CDR-H3 to wedge between the 233 FP of the primary gp41 protomer and the HR2 helix of the adjacent gp41 protomer (Fig 4C). 234 Additional contacts with the fusion peptide proximal region (FPPR) of the primary gp41 protomer 235 are made by residues at the tip of CDR-H2 ( Fig 4B). The N88 glycan accounts for 18% (198 Å²) of 236 the epitope BSA and makes contact with the CDR-H2 and FR-H3 regions of RM20F ( Fig 4B). The 237 lack of connecting density, even at lower contour, between RM20F and the glycans at N611 and 238 N637 suggests these glycans do not substantially contribute to the epitope. Epitope mapping 239 using BG505.T332N mutant pseudoviruses showed that knocking out the N611 glycan (N611Q 240 mutant) substantially enhanced neutralization by RM20F, while knocking out the N637 glycan 241 (N637Q mutant) had no effect. (Fig 4D). Other virus mutants revealed that neutralization by 242 RM20F was sensitive to various sequence changes within the epitope, particularly at residues 243 H85 and E647 (84.3% prevalence) and N88 (N88 glycan knock out) ( Fig 4D). The N88 glycan knock 244 out and the H85A mutation (to a lesser extent) significantly reduced neutralization activity of the 245 FP-targeting bNAb VRC34, but no effect on the CD4 binding site targeting bNAb VRC01 ( Fig 4D). 246 Introducing the 241 or 289 glycans (S241N and P291T, respectively) modestly reduced the 247 neutralization activity of RM20F ( Fig 4D). In comparison to the FP-targeting bNAbs VRC34 and 248 ACS202, RM20F lacked neutralization breadth when tested against a panel that included multiple 249 heterologous viruses ( Fig S4) likely due to the dependency on poorly conserved residues. 250 251

MAb RM20E1 binds to the fusion peptide and makes contact with two adjacent protomers 252
We solved a 2.3 Å crystal structure of the unliganded RM20E1 Fab and 4.4 Å cryoEM 253 structure of RM20E1 Fab bound to BG505 SOSIP trimer and Fab PGT122 (Figs 5A and S3; Tables 254 S4 and S5). RM20E1 binds to an epitope composed of one gp120 and two gp41 protomers with 255 1178 Å² of BSA. Residues 515 to 520 of the FP in the primary gp41 protomer are sandwiched 256 between CDR-H3, CDR-L1, and CDR-L3 of RM20E1 ( Fig 5B). CDR-H3 and FR-H1 make contact with 257 HR2 in the adjacent gp41 protomer ( Fig 5C). Additionally, the FR-H1 makes contact near the 258 N611-glycan site in the adjacent gp41 protomer, but we observed no connecting density that 259 could be attributed to the N611-glycan itself ( Fig 5C). RM20E1 avoids the N88 glycan but does 260 interact with residues in the C1 region of gp120, including H85, via its CDR-L1 ( Fig 5B). Despite 261 recognition of the conserved FP, RM20E1 did not neutralize the autologous BG505.T332N 262 pseudovirus. The antibody did however potently neutralize the N611A glycan KO BG505 263 pseudovirus ( Fig 1C) suggesting the epitope is shielded by the N611 glycan. The epitopes of the 264 FP-targeting bNAbs VRC34, ACS202, and DFPH-a.15 overlap to a large extent with the epitopes 265 of RM20E1 and RM20F ( Fig 5D), with DFPH-a.15 and VRC34 also neutralizing more potently in 266 the absence of the N611 glycan 9,25,39,40 . The RM20E1-bound FP conformation is similar to the FP 267 conformation when bound by the bNAb VRC34 ( Fig S5); however, the inability of RM20E1 to 268 accommodate the N611 glycan likely results in the lack of neutralization of the wild-type virus. in RMs can yield valuable information because of the close genetic relationship between RMs 278 and humans. Here, we isolated mAbs from two BG505 SOSIP.664 trimer-immunized RMs to better understand how the immune system recognizes the trimers and the epitopes associated 280 with the potent, but limited, HIV-1 Tier 2 neutralization. 281 By mapping the epitopes of all of the mAbs isolated from the immunized RMs, rather than 282 focusing only on NAbs, we were able to identify several non-neutralizing and potentially 283 immunodominant epitopes that would ideally be eliminated in future immunization studies. As 284 shown previously with mAbs from rabbits immunized with BG505 SOSIP.664 trimers 19 , the lack 285 of glycans at positions 241 and 289 in BG505 creates a large glycan hole which is targeted by 286 mAbs from both RMs. The mAbs isolated from RMs that target the 241/289 glycan hole are more 287 biased towards the 289-site compared to the previously characterized rabbit mAbs. This 288 difference may be attributed to the underlying differences in BCR repertoires between the two 289 animal models. In addition to the lack of glycans due to missing sequons that encode for 290 glycosylation, the recombinant BG505 SOSIP.664 trimer may also contain missing glycans even 291 when the correct sequon is present as previously observed by mass spectrometry studies of 292 glycopeptides 37,43 . Our study identified gp120/gp41 interface antibodies whose neutralization 293 was enhanced in the absence of the N611 glycan, suggesting that the BG505 SOSIP.664 trimer 294 may have sub-stoichiometric glycan occupancy in gp41 at this position, creating an unexpected 295 but immunogenic glycan hole. The elicitation of FP targeting mAbs in RMs with the BG505 296 SOSIP.664 trimer provides evidence that the FP bNAb epitope is accessible and immunogenic on 297 soluble Env trimer immunogens. Recent studies in mice, guinea pigs and RMs using synthetically 298 produced HIV fusion peptides covalently attached to carrier proteins as priming immunogens 299 followed by boosts with soluble Env trimer immunogens have also elicited FP specific antibodies 300 including some mAbs with neutralization breadth 9,44,45 . However, the majority of the animals immunized in these studies do not develop neutralization breadth and instead develop potent 302 neutralization against the BG505 pseudovirus with the N611 glycan KO 9,44,45 . Given the 303 consistency across studies and animal models in eliciting potent NAbs that target the FP epitope 304 and require the absence of the N611 glycan, investing in strategies to quantify and enhance the 305 N611 glycan occupancy in soluble Env trimer immunogens, particularly for boost immunogens, 306 may improve the neutralization breadth elicited by FP-targeting immunization protocols. 307 The sorting probe used to isolate BG505 Env-specific B-cells was a C-terminally 308 biotinylated BG505 SOSIP.664 trimer bound to a fluorescent streptavidin tetramer. Steric 309 constraints between the base of the trimer and the streptavidin tetramer likely resulted in a 310 lower recovery of base epitope-specific B-cells. Despite this potential selection bias, 55% of the 311 mAbs isolated from the two immunized RMs bound to the base of the BG505 SOSIP.664 trimer, 312 indicating the base of the soluble trimer is the major target for antibody responses during 313 immunization. To reduce the immunogenicity of this epitope, glycans can be introduced to shield 314 this site or the native-like trimers could be constructed onto scaffolds or particles 46-49 . 315 We were unable to construct individual germline BCR databases from RMs rh1987 and 316 rh2011 to precisely determine the SHM and gene/alleles usage as additional PBMC samples were 317 no longer available. Instead, we constructed a germline database containing BCR gene/alleles 318 from multiple Indian origin RMs that allowed us to measure SHM levels in the mAbs we isolated 319 from RMs. The new database provides web-based access (http://ward.scripps.edu/gld/) to a 320 curated and highly annotated general resource for examining BCR gene/alleles from Indian origin 321 RMs. Previous estimates of average SHM rates in mAb sequences from Env-immunized RMs were 322 8.9% and 6.1% for the HC and LC, respectively 50 . These apparently high levels of SHM following repeated immunization with the exact same soluble Env immunogens were likely due to missing 324 germline gene/alleles from the database used to calculated SHM. Using the germline database 325 reported in this study, we were able to assign vaccine-elicited antibody sequences to specific 326 germline genes/alleles and determine levels of SHM much more accurately. The average levels 327 of SHM reported in this study (6.3% for HC and 4.2% for LC) are comparable to the average levels 328 of SHM reported in similar immunization studies where per animal germline BCR databases were 329 inferred using IgDiscover 51,52 . Our germline database provides a resource for assigning germline 330 genes/alleles and accurately calculating rates of SHM when inferring individual germline 331 databases for each animal is logistically impractical. 332 In conclusion, in this study, neutralizing and non-neutralizing mAbs with distinctive genome assembly using BLAST 29,30 . Sequences that were not identical to the reference genome 383 were eliminated. Additional full-length genes/alleles from available Indian origin RM genomic 384 DNA sequencing datasets 21,31 were added to the database. Duplicates and sequences containing 385 ambiguous bases were removed. The resulting initial Indian origin RM germline BCR database 386 (Table S1) was used for running IgDiscover on additional NGS datasets that were obtained during 387 this study as described above, downloaded from the NCBI SRA 27,34 , or obtained directly from the 388 study authors 28 . Paired sequence reads were aligned and filtered for length and quality using germline_filter parameters "unique_cdr3s" and "unique_js" set to 10 and 4 respectively to 391 reduce the rate of false positives 27 . Inferred genes/alleles were kept if they were detected in 392 more than one animal or if they were identical to RM genes/alleles that were previously 393 deposited in NCBI. The resulting germline BCR database (available at 394 http://ward.scripps.edu/gld/) was converted into a custom IgBLAST database and subsequently 395 used to analyze the BG505-specific mAb sequences 35 . 396 Supernatants were harvested 4-6 days following transfection and passed through a 0.45 µm filter. 437 MAbs were purified using Protein A/G (ThermoFisher) or MAbSelect™ (GE Healthcare) affinity 438 chromatography. Fabs were purified using CaptureSelect™ CH1-XL (ThermoFisher) affinity 439 chromatography. 440

TZM-bl cell-based neutralization assays 443
Neutralization assays using the autologous BG505.T332N virus and mutants, and the 444 heterologous SF162 virus, were carried out as described previously 58  concentrated to 4 mg/mL using a 10 kDa Amicon® spin concentrator (Millipore). 3 μL of the 512 complex was mixed with 1 μL of a n-Dodecyl-β-D-Maltopyranoside (DDM) solution to a final DDM 513 concentration of 0.06 mM and applied to a grid (Quantifoil R 1.2/1.3, 400), which had been 514 plasma-cleaned for 5 seconds using a mixture of N2/O2 (Gatan Solarus 950 Plasma system). The 515 grid was blotted and plunged into liquid Ethane using a Vitrobot Mark IV (ThermoFisher). 516

Cryo Electron Microscopy Data Collection and Processing 517
Samples were imaged on either FEI Titan Krios electron microscope (ThermoFisher) operating at 518 300 keV (RM20F dataset) or a FEI Talos Arctica electron microscope (ThermoFisher) operating at 519 200 keV (RM20J and RM20E1 datasets). Both microscopes were equipped with Gatan K2 Summit 520 direct electron directors operating in counting mode. Automated data collection was performed 521 using the Leginon software suite 59 . Micrograph movie frames were aligned and dose-weighted 522 using MotionCor2 71 , and CTF models were determined using Gctf 72 . Particle picking, 2D 523 classification, Ab-initio reconstruction, and 3D refinement were conducted using cryoSPARCv2 73 . 524 Data collection and processing parameters are reported in Table S4. 525 Initial molecular models of the BG505 SOSIP trimer/Fab complexes were built by docking the Env 526 portion of PDB: 5V8M 74 into the EM density maps along with the relevant Fab crystal structures 527 (PDB: 4JY5 was used for PGT122 75 ) using UCSF Chimera 62 . The Fab constant regions were 528 removed due to flexibility in the elbow region as commonly found in Fab structures 76 , the 529 appropriate stabilizing mutations (v4.1 or v5.2) were introduced into the Env sequence, and N-530 linked glycans were added using Coot 77 . The models were iteratively refined into the EM density 531 maps using RosettaRelax and Coot 70,77-80 . Glycan structures were validated using Privateer 81 . 532 Overall structures were evaluated using EMRinger 82 and MolProbity 83 . Protein interface 533 calculations were performed using jsPISA 84 . Final model statistics are summarized in Table S4. 534

Statistical analysis 535
Statistical models inherent to Relion 3.0 61 and cryoSPARC 73 were employed in image analysis to 536 derive 2D classes and 3D models. All ELISA and neutralization assays were conducted with at least 537 duplicate measurements.