Impact of Common Modifications on the Antigenic Profile and Glycosylation of Membrane-Expressed HIV-1 Envelope Glycoprotein

Recent HIV-1 vaccine development has centered on “near native” soluble envelope glycoprotein (Env) trimers. These trimers are artificially stabilized laterally (between protomers) and apically (between gp120 and gp41). These same stabilizing mutations have been leveraged for use in membrane-expressed Env mRNA vaccines, although their precise effects in this context are unclear. To address this question, we investigated the effects of Env mutations expressed on virus-like particle (VLP) in 293T cells. Uncleaved (UNC) trimers were laterally unstable upon gentle lysis from membranes. However, gp120/gp41 processing improved lateral stability. Due to inefficient gp120/gp41 processing, UNC is incorporated into VLPs. A linker between gp120 and gp41 (NFL) neither improved trimer stability nor its antigenic profile. An artificially introduced enterokinase cleavage site allowed processing post-expression, resulting in increased trimer stability. Gp41 N-helix mutations I559P and NT1-5 both imparted lateral trimer stability, but concomitantly reduced gp120/gp41 processing and/or impacted V2 apex and interface NAb binding. I559P consistently reduced recognition by HIV+ donor plasmas, further supporting antigenic differences. Mutations in the gp120 bridging sheet failed to stabilize membrane trimers in a pre-fusion conformation, reduced gp120/gp41 processing and exposed non-neutralizing epitopes. Reduced glycan maturation and increased sequon skipping were common effect of mutations. In some cases, this may be due to increased rigidity which limits access to glycan processing enzymes. In contrast, viral gp120 did not show glycan skipping. We observed a minor species of high mannose glycan only gp160 in particle preparations. This was unaffected by any mutations and instead bypasses normal folding and glycan maturation processes. Including the full gp41 cytoplasmic tail led to markedly reduced gp120/gp41 processing and increased the proportion of high mannose gp160. Remarkably, NAbs were unable to bind to full-length Env trimers. Overall, our findings suggest caution in leveraging mutations to ensure they impart valuable membrane trimer phenotypes for vaccine use. AUTHOR SUMMARY A vaccine that induces virus-fighting antibodies to block HIV-1 infection remains elusive. To ablate HIV-1 infection, antibodies must bind to authentic envelope (Env) glycoprotein on the virus surface. However, Env can exist in various forms, many of which are relatively easy targets for non-effective antibodies. Therefore, a key challenge of vaccine design is to create pure authentic Env that is unfettered by these other forms of Env. Vaccine research to date has focused largely on stabilizing soluble Env trimers, as this format simplifies purification and translational studies. However, incomplete Env authenticity may blunt the efficacy of this approach. By comparison, the manufacture of particle-based vaccines that express Env in situ on membranes is cumbersome. In an alternative approach, lipid particles can deliver mRNA vaccines encoding membrane trimers, bypassing the manufacturing challenges of previous methods. Stabilizing mutations derived from soluble trimers are now being leveraged for membrane trimers. Here, we evaluated the effects of these mutations. Our results show that some mutations alter Env conformation, and therefore might best be omitted from membrane Env vaccines.

10 274 in the context of JR-FL SOS gp160CT. To enable us to better judge the effects of NFL, we also 275 made a double mutant (K510S+R511S) [65], to create a simple uncleaved (UNC) mutant.

283
Previously, we used the globally sensitive A328G mutant to measure non-neutralizing 284 antibodies in vaccine sera [26]. A328G was sensitive to non-NAbs 39F and 15e (S2C Fig, left 285 panels, open symbols). Interestingly, D197N was somewhat b12-resistant compared to the parent 286 but became more b12-sensitive when combined with A328G mutation. This overt change was 287 mirrored by higher sensitivity to 39F and 15e compared to the other A328G mutant. In addition, 288 PG16 sensitivity was lost, consistent with a "globally sensitive" conformation. In BN-PAGE, 289 A328G showed reduced trimer, suggesting lower stability (S2D Fig, bottom panel, lanes 2 and 4).

295
Although SOS trimers are stabilized in the apical plane by a disulfide bond, they may 296 dissociate laterally following lysis and BN-PAGE [30]. To check this point, we used BS 3 to 297 crosslink selected clones. Crosslinking significantly reduced monomer in all cases, particularly for 298 I559P or DS (Fig. 6, compare lanes 6, 8, 10, 12 to lanes 5, 7, 9, 11). Notably, SOS NFL I559P 299 behaved similarly to SOS I559P (Fig. 6, lanes 5, 6, 9 and 10). Since I559P prevents gp120/gp41 300 cleavage, these clones differ only by the length of the peptide linking gp120 and gp41, which 301 evidently has little effect here. For non-I559P mutants, some monomer shifted upwards slightly 302 with BS 3 (Fig. 6, lanes 2 and 4, black asterisks). The survival of some trimer in the face of boiling 303 in SDS and DTT verifies that BS 3 crosslinking was effective (Fig. 6, lanes 13 and 14). Overall, this 304 is consistent with the idea that Env is mostly if not all trimeric but different forms can exhibit 305 differing lateral stabilities.

308
The DS mutant was designed based on the SOSIP gp140 trimer structure to insert a 309 disulfide between positions 201 and 433 to restrict soluble CD4 (sCD4) binding and downstream 310 conformational changes [11]. Soluble DS gp140 trimers bind only one molecule of sCD4, 311 preventing full exposure of CD4-inducible (CD4i) and V3 epitopes. To check this with membrane 312 trimers, excess 4 domain sCD4 was incubated with parent or DS VLPs, followed by a wash and 313 BN-PAGE to evaluate shifts. DS trimers were shifted like the parent trimers, suggesting the 314 binding of 3 sCD4 molecules (S3 Fig, compare lanes 2 and 4) [30]. Overall, the similar sCD4 315 binding patterns suggest that DS does not limit sCD4 binding to membrane trimers.    16 475 bond that must be broken by exposure to reducing agent, providing more time for these MAbs to 476 neutralize.

477
Since the 2G12 epitope is constitutively exposed [10,37,40], relative 2G12 capture can be 478 used to standardize capture for each Env clone, regardless of any expression differences. There 479 was a ~2-fold range of 2G12 capture efficiency, with SOS NFL A433P being most sensitively 480 captured, while both WT and SOS FL were the least well captured (S10A Fig).

481
We next evaluated JR-FL mutant MAb capture, expressed as % relative to 2G12 capture 482 (Fig. 9). Capture of each VLP clone was compared to SOS-VLPs, testing for significant 483 differences by Kruskal-Wallis non-parametric rank test. MAb  491 However, here we observed that D197N+NT1-5 was captured poorly by all V2 MAbs (Fig. 9), 492 while D197N mutant alone did not significantly affect capture, although it trended to weaker 493 capture. This reflects the poor V2 binding to NT1-5 by BN-PAGE (Fig. 8)

498
Next, we considered UNC and NFL. PGT145 only modestly captured both clones, 499 suggesting that I559P is important for PGT145 binding to SOS NFL I559P (Fig. 9). PG16 and 500 CH01 capture was also poor, like SOS NFL I559P. However, VRC38 capture was unaffected 501 gp120/gp41 cleavage. VRC03 capture was weak, as its quaternary epitope is evidently poorly 502 presented in uncleaved mutants. Conversely, 15e and V3 capture was effective. Overall, we 503 conclude that I559P differentially affects V2 epitopes, with NFL having little impact.

504
PGT145, CH01 and VRC03 failed to capture SOS NFL A433P and SOS D197N+A328G 505 mutants, consistent with their "globally sensitive" phenotypes. PG16 and VRC38 captured A433P 506 moderately, but not D197N+A328G. Both mutants were well-captured by 15e, 39F and 447-52D, 507 albeit to differing extents. Overall, we infer that both mutants adopt tier 1 conformations, but with   , we used this method to analyze unfractionated "total 529 VLP" samples, in which the resulting data derives from the sum of various forms of Env that we 530 described above in Western blots, including gp120/gp41 trimers, gp160m, gp160i and gp41 531 stumps. To try to generate "cleaner" samples for analysis and to enable us to discriminate 532 between the glycan profiles of different Env isoforms, here, we ran VLP samples on reducing 533 SDS-PAGE and cut out bands for analysis. In the case of the parent sample, we cut out 3 bands 534 corresponding to gp160 (including gp160m and gp160i, which could not be separated sufficiently 535 by this method), gp120 and gp41. For the other samples (NFL, NFL I559P and NFL A433P), we 536 cut out gp160 bands (gp160m and gp160i), as there are no gp120 and gp41 bands when the NFL 537 is used.

538
Detailed glycopeptide data is provided in S1 Raw glycopeptide folder, containing a file for 539 each sample. Data was obtained for all samples except for the gp41 band of the parent VLP, 540 which was insufficient for detection. Data is plotted in bar charts (S12 Fig) that reveal the relative 541 percentages of oligomannose, hybrid, complex and core glycans at each site, as well as 18 542 unoccupied sequons due to skipping. These charts also show fucosylated and sialylated glycans 543 (NeuAc/NeuGc) at each site.

544
To summarize the data and to facilitate comparisons, each glycan type was given a score 545 from 1 to 19, depending on the average maturation state (S1 Main glycan analysis). Specifically, 546 the high mannose glycan, M9Glc, has a score of 1, while the most highly branched and 547 fucosylated complex glycan HexNAc(6+)(F)(x) has a score of 19. Glycans at each site were then 548 given an overall score based on the percentage of each glycoform at each site multiplied by its 549 percent prevalence and rounded to the nearest whole number. Overall glycan scores were 550 computed in S1 Main glycan analysis for each clone and are color coded for clarity. Glycan data 551 were modeled for the gp160 bands (S13A Fig) and the gp120 band (S13B Fig) of

554
To determine the effects of mutations on glycan profiles, we first need to select a reference 555 profile. The profiles of the 3 independent parent "total VLPs" (2020, 2021 and 2023) (S12, S13C-556 E, S1 Main glycan analysis score sheet) were largely similar. However, data for glycans at some 557 positions were missing in some preparation but present in others. Therefore, we decided to 558 combine and average the parent data as a reference to the mutants (S12 Fig, S13F Fig, S1 Main 559 glycan analysis score sheet).

560
We next compared the glycan profiles of gp160 bands (containing gp160m and gp160i) and 561 gp120 band with each other and to the average parent as well as the previous gp120 monomer 562 (S13G Fig). Glycan score changes at each position are determined by subtraction. Score changes 563 are color coded according to glycan processing: decreased processing is shown in shades of red 564 and increased processing is shown in shades of blue (Fig. 11). For clarity, we refer to VLP gp120 565 as viral gp120 hereafter to distinguish it from gp120 monomer. There were stark differences 566 between parent viral gp120 and gp160 bands, particularly at N356, that was completely flipped to 567 complex in viral gp120. Increased viral gp120 glycan maturation was also found at N88, N188, 568 N392 and N463. These are modeled in Fig. 11A, summarized in S1 Main glycan analysis and 569 shown in detail in S1 Sheet.

570
Notably, viral gp120 did not show any sequon skipping. This contrasts with gp160 bands, 571 where there was partial skipping at N188 and N339 (gray shading in S12A Fig, S1 Main glycan 572 analysis). Total VLP samples also exhibited detectable skipping (S12A  575 glycan maturation at the same positions (S1 Sheet). In contrast, the glycan profiles of the parent 19 576 average and the parent gp120 band (S13B Fig) were more consistent (S1 Main glycan analysis, 577 S1 Sheet).

578
If viral gp120 and gp160m express equivalently, one might expect increased diversity at the 579 variable glycan sites in total VLPs, as they would reflect the combined diversity of both gp120 and 580 gp160m components (S1 Sheet). To some extent this is the case at N188 and N463 (especially 581 the 2023 parent sample), but not so much at N88 and N356 that was biased more towards viral 582 gp120 profile. In the 2023 total VLP sample, N392 exhibited intermediate processing. Overall, this 583 suggests that functional gp120/gp41 trimers are the predominant form of Env in total VLPs and 584 gp160m is the weaker form. Furthermore, the near complete complex glycans at some positions 585 suggests that gp160m expression is dominant over gp160i, which exhibits high mannose glycans 586 at all positions [46].

587
The gp120 monomer profile at variable positions N88, N160, N356 and N463 was also 588 more akin to viral gp120 than the gp160 bands (S1 Sheet). However, the glycan at N392 was far 589 more processed on gp120 monomer than on viral gp120 or total VLPs (S1 Sheet). We infer that 590 N392 is uniquely buried in native trimers.

591
592 Mutant glycan profiles compared to parent.

593
The key information we sought here is whether mutants affect glycan compositions, as an 594 indicator of conformational differences. We therefore compared the parent average to NFL, NFL 595 I559P and NFL A433P (Fig. 11B-D, S1 Main glycan analysis). SDS-PAGE data above indicated 596 that I559P bands were fast moving and slightly more endo H sensitive, whereas A433P bands 597 were slightly slower moving. We hoped that glycan analysis would help us to reconcile these 598 effects.