Initiation of HIV neutralizing B cell lineages with sequential envelope immunizations

A strategy for HIV-1 vaccine development is to define envelope (Env) evolution of broadly neutralizing antibodies (bnAbs) in infection and to recreate those events by vaccination. Here, we report host tolerance mechanisms that limit the development of CD4-binding site (CD4bs), HCDR3-binder bnAbs via sequential HIV-1 Env vaccination. Vaccine-induced macaque CD4bs antibodies neutralize 7% of HIV-1 strains, recognize open Env trimers, and accumulate relatively modest somatic mutations. In naive CD4bs, unmutated common ancestor knock-in mice Env+B cell clones develop anergy and partial deletion at the transitional to mature B cell stage, but become Env− upon receptor editing. In comparison with repetitive Env immunizations, sequential Env administration rescue anergic Env+ (non-edited) precursor B cells. Thus, stepwise immunization initiates CD4bs-bnAb responses, but immune tolerance mechanisms restrict their development, suggesting that sequential immunogen-based vaccine regimens will likely need to incorporate strategies to expand bnAb precursor pools.

UCA could be generated and tested for neutralization, to better investigate the contribution of the HC and LCs in neutralization.
Fig 7a. TF+w53 immunization results in approximately a log higher TF Env-specific B cells (and in neutralizing antibody responses) than immunization with TF. Why is that? Does the w53 envelope 'rescue' somehow envelope+ B cells? What happens if only wk53 envelope is used alone as an immunogen in these heterozygous mice?

Reviewer #2 (Remarks to the Author):
The manuscript by Williams and colleagues addresses the question of why it is so difficult to elicit broadly neutralizing antibodies against HIV ENV, focussing on the conserved CD3 binding site.
The study focuses on antibodies related to CH103, a broadly neutralizing antibody lineage that arose during chronic HIV infection of an African individual, where the evolution of ENV variants and antibody variants has previously been longitudinally tracked and analysed in remarkable detail.
One line of experimentation attempts to reproduce the elicitation of CH103-like antibodies by sequentially immunizing rhesus macaques with ENV proteins corresponding to different timepoints in the natural history of the CH103 lineage. This proved not to elicit broadly neutralizing antibodies, but did generate antibodies related to CH103 but lacking the corresponding lambda light chain -possibly because the macaques lacked the necessary Vlambda element present in humans.
The clearest conclusions come from the second line of experiments, where mice are engineered by gene targetting in their germline H and L chain genes with rearranged VDJ and VJ elements encoding the inferred unmutated common ancestor of the CH103 antibody. These experiments are elegant and conclusively demonstrate that the unmutated CH103 antibody undergoes strong negative selection.
There is lower surface IgM on nascent immature B cells in the bone marrow, editing of light chains, and further diminished surface IgM, calcium signalling and poor accumulation as transitional and mature B cells. These are all well established responses of developing B cells when their surface Ig binds too strongly to systemic self-antigens. The nature of the self-antigens responsible for negative selection of CH103 is not determined, but these findings alone provide the clearest evidence yet that CD4-binding site antibodies are subject to multiple mechanisms of self-tolerance and negative selection.
Env immunization experiments show that the CH103-expressing B cells can be induced to hypermutate and switch to IgG, albeit very inefficiently.
The results are for the most part compelling and will be of wide interest to readers interested in HIV, antibodies, vaccines and self-nonself discrimination.
As detailed below, there are some technical and presentation issues that should be addressed to further improve an already very interesting manuscript.
1. Excessive use of sub-field specific acronyms makes the study very hard to read. Avoid abbreviating rhesus macaque to RM, or NHP (as done in the figures); avoid shortening knock-in to KI; bnAbs is ok but nAbs better to spell out neutralizing antibodies; etc. Also, figure legends need to explain better how the results were obtained, and avoid interpretation of the results. For example legend to Figure 5D does not explain that the results are from sorting single B cells regardless of Env binding and sequencing H and L chain mRNA. Legend to Figure 2A needs to clarify % of what: all leukocytes, B cells, or memory B cells?
2. Nature of immunogen. In the macaque and mouse studies, it is not explained whether the immunization used ENV trimers, monomers, SOSIPs, were the glycans native complex carbohydrates or high mannose or EndoH treated? The form of antigen given should be explained in the results, figure legend and in more detail in the methods.
3. In Figure 2, need to provide representative flow cytometric plots showing the gating strategy to enumerate ENV-binding memory B cells. Full gating strategy should be provided in supplementary figure. The conclusion at line 156 of the results "sequentially-administered Envs also induced higher frequencies of differential-binding memory B-cells" is not supported by statistical evidence, and there appears little difference in Figure 2b. In Figure  2c, please use same Y axis scale for all three panels to improve comparison of the regimes. 4. Affinity for Env. In the results text and legend for Figure 3, it is important to clarify that deglycosylated Env was used for crystallography. When CD4bs antibodies have been crystallized with natively glycosylated Env, half the contact surface is between antibody and N-glycan's shrouding the binding site. What is known about the affinity of CH103 UCA for deglycosylated versus natively glycosylated Env?
Line 303 cites ref 6 where the Kd for (high mannose?) glycosylated Env was measured to be 36 nM. This was measured by SPR with divalent IgG, and may not reflect the monovalent affinity, which could be lower. Important to state the Kd, how it was measured, and whether it is the single-site affinity or bivalent avidity, in order to compare like with like vis other antibodies referred to here.
The Abstract states that the "non-edited precursors underwent limited affinity maturation"in the immunised mice. Results only show they underwent hypermutation, and the fraction of non-edited precursors among IgG+ cells appears comparable with repeated TF or sequential immunization. There is no convincing evidence that sequential immunization was better at recruiting unedited cells to hypermutate and switch. Figure 5g. Given that CH103 likely makes extensive contacts with N-glycans around the CD4binding site, it is likely that it or the unmutated common ancestor has substantial affinity for N-glycans on a variety of self proteins. It would be valuable to test this, since low affinity binding to cell surface N-glycans would explain the downregulation of surface IgM and negative selection observed. Figure 7 is the weakest part of the study. Need to show representative plots of IgG+ cells and gates used. Ideally would be good to show as IgG versus ENV TF binding. The text on line 339 states that non-differential Env binders "represent~35% of the total IgG+ memory pool" but the plots gated on this pool show less than 2% ENV binders, most of which appear to be differential binders to native and not CD4-binding site mutant ENV.

5.
The corresponding analysis of all B cells in Supp Fig 5 appears to have an error, because the saline injected mice should have ~8% ENV binders among all B cells (as shown in Figure  5A) yet there are none and the plots appear to have fewer events overall and be processed differently. Figure 7D should be omitted. It "overcooks" the data in 7D, and one can't do statistics on ratio since not normally distributed because numerator and denominator are reciprocals. Just leave at C, which is a more accurate yet convincing portrayal.
In Figure 7E and F it is surprising that only 9 of 81 cells sorted based on binding Env but not CD4bs-mutant Env actually express the targetted H and L chain, and indeed 70% of the presumed CH103-expressing memory B cells lack the CH103 H chain. One would expect nearly 100% to be expressing CH103 H and L chain. Is this a technical problem with ENV staining and sorting in the immunized mice? It casts doubt on the conclusion that the increased frequency of differential Env binding IgG+ cells upon sequential immunization in Figure 7A, and the increased serum IgG binding to wildtype relative to mutant Env, represents increased activation of unedited CH103 cells. Most of the cells appear not to employ the knocked in CH103 H chain, and only a minority of those that do employ the CH103 L chain. Does this mean the endogenous mouse repertoire is better equipped to mount IgG antibodies against the CD4 binding site? Needs discussion and clarification.

Reviewer #3 (Remarks to the Author):
The article by Williams et al. entitled 'Initiation of HIV Neutralizing B Cell Lineages with Sequential Envelope Immunizations' is a very thorough multi-center study describing the shaping of the immune response to HIV trimers in model systems with engineered B-cell lineages. The study represents a thorough exploration of a concept that many investigators in the field are interested and will therefore be of wide interest and importance.
Focusing on the crystallographic datasets corresponding to the reported structures of DH522.1, DH522UCA, DH522IA, DH522.2, and DH522.2-gp120 core. The data processing and refinement statistics are reported appropriately and all the datasets exceed expected minimum standards for the reported resolutions. The datasets are of suitable quality to reliably support the presented crystal structure models in Figure 3.
The cryo-EM data is of low resolution but the presented class averages of the DH522-CH505 SOSIP complex provide good supporting evidence of the inferred protomer neighbor clash of DH522 from the crystallographic studies. Given the low resolution and on-going variation in data processing methods adopted across the EM community, I would recommend including corresponding EM for all the antibodies studied crystallographically. This is particularly important given the low resolution and unusual conformation of the presented 'open' trimer.
We appreciate the reviewers' thorough and thoughtful comments and are glad they were so positive in their overall assessment of our study. We have answered all the questions from the reviewers, and where applicable, performed additional experiments requested. Our manuscript is now considerably stronger, and we now hope it is acceptable for publication in Nature Communications. For ease in review, the comments of the reviewers are numbered and listed below followed by our answers in blue font. The changes we have made are indicated in yellow in the manuscript.

Reviewer #1
The study discusses efforts to elicit CH103-like broadly neutralizing antibodies through a sequential immunization scheme with recombinant CH505-derived envelope immunogens. Immunizations were performed in non-human primates (NHP). The bottom line is that CH103-like antibodies were not elicited. However, the authors discuss in a very thoughtful manner, potential reasons for this and executed appropriate experiments to address this issue. This is a major strength of this study. Overall the study is outstanding and I do not have major concerns regarding the experiments or with the interpretation of the results.
1. The underlying mechanisms that prevent the rapid elicitation of CH103-like neutralizing antibodies in the KI mice are well discussed. However, the study does not provide direct evidence that similar tolerance-related blockages are operational in the NHP case. The only direct evidence provided for the lack of elicitation of CH103 antibodies is the lack (infrequent?) of expression of appropriate VL genes in NHP.  Fig 7G), thus resulting in a higher fraction of clones with affinity for, and thus further expansion by week 53 Env.

) diverting the non-lineage positive B-cell response, thus reducing competition with non-anergic clones. This would specifically occur via week 53 Env potentially cross-reacting more strongly than TF with either of two subsets: i) pre-immune knock-in light chain-edited (TF Env-) clones and/or ii) vaccine-induced knock-in heavy chain and light chain-edited (TF Env+ but non-differential) clones that start appearing after TF priming. b) preferential activation of the rare double knock-in positive clones initially activated by TF to switch and expand, because of their preferential (albeit limited) SHM induced (relative to heavy or light chain only knock-in clones;
These possibilities are now stated in the discussion on page 21 of the manuscript.

Reviewer #2
The manuscript by Williams and colleagues addresses the question of why it is so difficult to elicit broadly neutralizing antibodies against HIV ENV, focussing on the conserved CD3 binding site.
The study focuses on antibodies related to CH103, a broadly neutralizing antibody lineage that arose during chronic HIV infection of an African individual, where the evolution of ENV variants and antibody variants has previously been longitudinally tracked and analysed in remarkable detail.
One line of experimentation attempts to reproduce the elicitation of CH103-like antibodies by sequentially immunizing rhesus macaques with ENV proteins corresponding to different timepoints in the natural history of the CH103 lineage. This proved not to elicit broadly neutralizing antibodies, but did generate antibodies related to CH103 but lacking the corresponding lambda light chainpossibly because the macaques lacked the necessary Vlambda element present in humans.
The clearest conclusions come from the second line of experiments, where mice are engineered by gene targetting in their germline H and L chain genes with rearranged VDJ and VJ elements encoding the inferred unmutated common ancestor of the CH103 antibody. These experiments are elegant and conclusively demonstrate that the unmutated CH103 antibody undergoes strong negative selection.
There is lower surface IgM on nascent immature B cells in the bone marrow, editing of light chains, and further diminished surface IgM, calcium signalling and poor accumulation as transitional and mature B cells. These are all well established responses of developing B cells when their surface Ig binds too strongly to systemic self-antigens. The nature of the self-antigens responsible for negative selection of CH103 is not determined, but these findings alone provide the clearest evidence yet that CD4-binding site antibodies are subject to multiple mechanisms of self-tolerance and negative selection.
Env immunization experiments show that the CH103-expressing B cells can be induced to hypermutate and switch to IgG, albeit very inefficiently.
The results are for the most part compelling and will be of wide interest to readers interested in HIV, antibodies, vaccines and self-nonself discrimination.
As detailed below, there are some technical and presentation issues that should be addressed to further improve an already very interesting manuscript.

4.
Excessive use of sub-field specific acronyms makes the study very hard to read.
Avoid abbreviating rhesus macaque to RM, or NHP (as done in the figures); avoid shortening knock-in to KI; bnAbs is ok but nAbs better to spell out neutralizing antibodies; etc. Also, figure legends need to explain better how the results were obtained, and avoid interpretation of the results. For example legend to Figure 5D does not explain that the results are from sorting single B cells regardless of Env binding and sequencing H and L chain mRNA. Legend to Figure 2A needs to clarify % of what: all leukocytes, B cells, or memory B cells?

Answer:
We thank the reviewer for the suggestions. We have edited the manuscript text to refer to rhesus macaques as macaques or NHPs, and nAb abbreviation was replaced with neutralizing antibodies. In figures where these abbreviations are in the figure, we indicate the abbreviation key in the figure legend. We have also revised our figure legends for clarity in how data were generated.
With respect to Figure 5D in particular, we assume the reviewer means Figure 5F (since sorting/cloning was only performed for data related to that panel). As per the reviewer's suggestion, that figure legend has now been revised to state that single cell sorts of total (unselected for Env binding) mature IgG+ B-cells were performed and that the VDJ/VJ pairs from single cells were recovered by RT-PCR. Furthermore, the legend of Figure 2A is now also revised, to read as follows: "One million peripheral blood mononuclear cells (PBMCs) from immunized RMs were phenotyped by FACS analysis for memory B cells that had CH505 Env differential binding. Shown is frequency (%) of memory B cells in PBMC that demonstrated CH505 Env differential binding at timepoints throughout the immunization schedule per animal (each line on graph)" 5. Nature of immunogen. In the macaque and mouse studies, it is not explained whether the immunization used ENV trimers, monomers, SOSIPs, were the glycans native complex carbohydrates or high mannose or EndoH treated? The form of antigen given should be explained in the results, figure legend and in more detail in the methods.

Answer:
We thank the reviewer for the suggestions. As recommended, we outlined in more details the form of CH505 Envs used as immunogens in the methods of the supplemental materials, and were also more descriptive of the Env forms used in the results and figure legends by using the term CH505 gp120 Env monomer instead of CH505 Env. We included a recombinant Env gp120 expression section on page 3 of the supplemental materials. Previous work has demonstrated that the glycans on recombinant gp120 Envs are more complex than glycans on virus associated gp160 Envs (Doores et al. PNAS 2010;107(31):13800-5). We have studied CH505 Env gp120 proteins expressed in 293F cells by glycan site-specific mass spectrometry; the glycans were found to be a mixture of complex and high mannose residues (Desaire H, Alam SM, Haynes BF et al., unpublished). This is now stated on page 4 of the supplemental materials.
6. In Figure 2, need to provide representative flow cytometric plots showing the gating strategy to enumerate ENV-binding memory B cells. Full gating strategy should be provided in supplementary figure. The conclusion at line 156 of the results "sequentially-administered Envs also induced higher frequencies of differential-binding memory B-cells" is not supported by statistical evidence, and there appears little difference in Figure 2b. In Figure 2c, please use same Y axis scale for all three panels to improve comparison of the regimes.

Answer:
We agree with the reviewer that the conclusion on line 156 is not statistically supported. We have rephrased the sentences on page 7 of the manuscript to say "Among macaques immunized with TF gp120 Env, 11% of the CH505 Env-reactive antibodies were CH505 differential binders; however, among those immunized with sequential combinations of CH505 gp120 Envs, 16% were CH505 differential binders (Figure 2b). Thus, we observed a trend for animals immunized with sequential CH505 gp120 Envs to make more CH505 differential-binding antibodies, albeit not statistically significant." As recommended, Figure 2C has been edited to show the same Y axis scale for all three panels, and a new Supplementary Figure 2 shows representative flow cytometric plots with the gating strategy to enumerate Env-binding memory B cells. In the new Supplementary Figure 2, we showed the representative flow cytometry plots of CH505 differential binding memory B cells above background, at 2 weeks post 3 rd immunization for animals immunized with CH505 TF gp120 Env monomer alone (N=4) and sequential combinations of CH505 gp120 Env monomers (N=8). Most immunized animals had CH505 differential binding memory B cells at 2 weeks post 3 rd immunization.

7.
Affinity for Env. In the results text and legend for Figure 3, it is important to clarify that deglycosylated Env was used for crystallography. When CD4bs antibodies have been crystallized with natively glycosylated Env, half the contact surface is between antibody and N-glycan's shrouding the binding site. What is known about the affinity of CH103 UCA for deglycosylated versus natively glycosylated Env?
Answer: We agree with the reviewer that we should clarify that deglycosylated Envs were used for x-ray crystallographic studies. Negative stain electron microscopy studies were performed with fully glycosylated Envs. We have edited the legend of Figure 3E on page 31 to reflect these changes.
It was previously reported that glycan-deleted trimers, including CH505 DS SOSIP.∆Gly4, showed increased binding to CH103 (Zhou et al. Cell Reports 19, 2017;719-732), consistent with previous reports that glycans around the CD4 binding site occlude antibody access to the CD4 binding-site epitopes (Stewart-Jones et al. Cell 165, 2016;813-826 9. The Abstract states that the "non-edited precursors underwent limited affinity maturation"in the immunised mice. Results only show they underwent hypermutation, and the fraction of non-edited precursors among IgG+ cells appears comparable with repeated TF or sequential immunization. There is no convincing evidence that sequential immunization was better at recruiting unedited cells to hypermutate and switch.

Answer:
We agree with this comment by the reviewer. Thus, we have re-worded the abstract (see revised lines 33-34) in order to more clearly reflect this important distinction.
10. Figure 5g. Given that CH103 likely makes extensive contacts with N-glycans around the CD4binding site, it is likely that it or the unmutated common ancestor has substantial affinity for N-glycans on a variety of self-proteins. It would be valuable to test this, since low affinity binding to cell surface N-glycans would explain the downregulation of surface IgM and negative selection observed.  Figure 3c on page 29 of the supplemental materials. Figure 7 is the weakest part of the study. Need to show representative plots of IgG+ cells and gates used. Ideally would be good to show as IgG versus ENV TF binding. The text on line 339 states that non-differential Env binders "represent~35% of the total IgG+ memory pool" but the plots gated on this pool show less than 2% ENV binders, most of which appear to be differential binders to native and not CD4-binding site mutant ENV.