Monocyte-derived transcriptome signature indicates antibody-dependent cellular phagocytosis as a potential mechanism of vaccine-induced protection against HIV-1

  1. Shida Shangguan
  2. Philip K Ehrenberg
  3. Aviva Geretz
  4. Lauren Yum
  5. Gautam Kundu
  6. Kelly May
  7. Slim Fourati
  8. Krystelle Nganou-Makamdop
  9. LaTonya D Williams
  10. Sheetal Sawant
  11. Eric Lewitus
  12. Punnee Pitisuttithum
  13. Sorachai Nitayaphan
  14. Suwat Chariyalertsak
  15. Supachai Rerks-Ngarm
  16. Morgane Rolland
  17. Daniel C Douek
  18. Peter Gilbert
  19. Georgia D Tomaras
  20. Nelson L Michael
  21. Sandhya Vasan
  22. Rasmi Thomas

  • Curated by eLife

    eLife logo

    Evaluation Summary:

    Attempts to produce effective vaccines against HIV have not yet been successful, with the lack of understanding correlates of protection being a significant limitation. This paper analyses gene expression in a number of human and non-human primate vaccine trials and identifies a profile that appears to correlate with protection from infection. This profile is linked primarily to monocytes and the ability of these cells to mediate antibody dependent cellular phagocytosis. The work has implications for ongoing attempts to generate effective vaccines against HIV and perhaps other viral diseases.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #2 agreed to share their name with the authors.)

Reviewed by

Read the full article

Listed in

Log in to save this article

Abstract

A gene signature was previously found to be correlated with mosaic adenovirus 26 vaccine protection in simian immunodeficiency virus and simian-human immunodeficiency virus challenge models in non-human primates. In this report, we investigated the presence of this signature as a correlate of reduced risk in human clinical trials and potential mechanisms of protection. The absence of this gene signature in the DNA/rAd5 human vaccine trial, which did not show efficacy, strengthens our hypothesis that this signature is only enriched in studies that demonstrated protection. This gene signature was enriched in the partially effective RV144 human trial that administered the ALVAC/protein vaccine, and we find that the signature associates with both decreased risk of HIV-1 acquisition and increased vaccine efficacy (VE). Total RNA-seq in a clinical trial that used the same vaccine regimen as the RV144 HIV vaccine implicated antibody-dependent cellular phagocytosis (ADCP) as a potential mechanism of vaccine protection. CITE-seq profiling of 53 surface markers and transcriptomes of 53,777 single cells from the same trial showed that genes in this signature were primarily expressed in cells belonging to the myeloid lineage, including monocytes, which are major effector cells for ADCP. The consistent association of this transcriptome signature with VE represents a tool both to identify potential mechanisms, as with ADCP here, and to screen novel approaches to accelerate the development of new vaccine candidates.

Article activity feed

  1. Version published to 10.7554/elife.69577 on eLife
  2. eLife

    Author Response:

    Reviewer #1 (Public Review):

    The authors carried out a post-hoc analyses of a protective gene expression signature previously observed in preclinical trials and clinical trials (RV144 and HVTN505) to identify a possible correlate of reduced risk of infection and whether able to provide a potential mechanism for protection. This monocyte signature they focus on was absent in the DNA/rAd5 human vaccine trial which did not show efficacy and was enriched in the partially effective RV144 human trial where the vaccine was and ALVAC/protein vaccine. Here they indicate that the signature is a correlate of reduced risk of infection.

    Identifying signatures of protection is an important issue in the development of a HIV vaccine, and signature analyses might be important to reveal a few markers that might be selected to evaluate vaccine trials. However, this analysis must be able to point to very few genes, as single cell analyses are not an option in a large clinical vaccine trial.

    We agree scRNA-seq might not be applicable to assessing large scale clinical trials. However it is useful for identifying the cellular lineage of the signal we previously were unable to identify from bulk gene expression datasets (see discussion section, line 324). The signature identified has 200 genes for which methods are increasingly available for economic screening of large sample numbers, for example as we did previously on the Fluidigm BioMark platform (Ehrenberg et al. 2019).

    It is unclear whether or how the conclusion of the previous publication by many of the same authors of this paper, including the senior author (Ehrenberg et al., 2019, identification of a gene signature in B cells that is associated with protection from SIV and HIV infection providing a new approach for evaluating future vaccine candidates) is compatible with this new one: signature primarily expressed in myeloid lineage being the one most consistently associated with vaccine efficacy. It is unclear which one of the two is correct or how they are reconciled. Was the single cell analysis done in monocytes only for this paper or simply not reported in the studies of Ehrenberg et al., 2019?

    The protective gene signature was identified initially in microarray data from total PBMCs in the RV144 study and so we did not know the cellular lineage of this signal. Although RV144 samples were depleted, we had the unique opportunity to investigate the cellular lineage of the gene signature in the RV306 trial, which is a vaccine trial that was performed in Thailand and used the same RV144 vaccine series, with additional boosts after the 4th vaccination. We performed scRNA-seq at the timepoints that were equivalent to the RV144 4th vaccination and concluded that the enriched genes in the signature were mostly expressed in monocytes (discussion section, lines 329-336). The current paper has 3 new datasets: HVTN 505 RNA-seq data, RV306 RNA-seq data and RV306 single cell CITE-seq data. The Ehrenberg et al. 2019 were primarily focused on the Ad26 vaccine preclinical trials in NHP. This formed the basis of our current findings that expanded to human studies such as RV144, HVTN 505 and RV306. The CITE-seq data from the RV306 study was performed in 2020, only after we confirmed using bulk-RNA-seq from blood that the gene signature associated with increased ADCP in the RV306 study. We have clarified the different studies in supplementary table 1.

    Figure 1: The gene expression score (GES) of this figure does not seem to be for a specific cell type. It is unclear how the GES reported here relates to the final GES of monocytes. What is the utility of this analysis? Can we observe here the same most significant genes that we observe in monocytes? This is important because if bulk analysis gives the same results as looking at monocytes an eventual marker identified in monocytes could be evaluated in luck analysis.

    The composite gene expression score in Figure 1 focuses on a GES of only the enriched genes which can be used as a continuous or categorical variable in a immune correlates analyses as shown in Figure 1 or 2 regardless of phenotype. We see enrichment of a geneset that associates with vaccine protection and ADCP across multiple studies and species irrespective of methods being used. We think this set of 200 genes have a coordinated expression and may not be specific to a cell type, but might mark a certain biological state, such as response to a cytokine, and may be picked up even in PBMC and blood samples. We clarify this further in the discussion section (line 348).

    Figure 2: it would be good to know whether the subset of the 63 genes can be restricted to the most significant and their GES can still retain the predictive value.

    As suggested by the reviewer, we made a GES of a subset of the 63 genes in the RV144 signature that had the most significant genes (32) that associated with HIV acquisition in Fig 5. (p <0.05, q <0.1). We see that the association is slightly stronger and the probability of acquiring HIV-1 is lower in individuals with high GES (OR = 0.35 and p value = 0.0001 compared to previous OR = 0.37 and p value = 0.0002). Vaccine efficacy in individuals with high GES has also increased to 81.4% from 75.1%. The Distribution of AUC and accuracy plotted after repeating the process 1000 times showed that GES of the significant subset of the genes is predictive of HIV-1 infection with AUC of 0.69 ± 0.08 and with accuracy of 0.81 ± 0.04 (compared to previous 0.67 ± 0.08 and 0.81 ± 0.04). We agree this smaller subset could potentially be useful and now include them in the results section (page 11) and as a new supplementary figure 2.

    Figure 3 deals with genes associated with antibody dependent cellular phagocytosis (ADCP). Can one derive a gene or a few genes that are predictive of significant ADCP?

    Thank you for this suggestion, we have been able to explore this and now include new panels in the main figures which identify genes predictive of ADCP. We made a GES of the 93 genes enriched at the day 3 timepoint associating with magnitude of ADCP. A prediction model was built using the 93 genes from Day 3 time point. Internal validation has an area under the curve (AUC) of 0.80, suggesting that this classifier was able to discriminate high ADCP from low ADCP measured 2 weeks after last vaccination. This model, consisting of 93 expressed genes, was then tested at the Week 2 time point and was also able to predict ADCP as a dichotomous variable at the week 2 time point (AUC = 0.73).

    We further examined 82 genes overlapping between the 118 enriched genes from week 2 and the 93 enriched genes from day 3 post the RV144 vaccine regimen that associated with ADCP. A GES was computed using the 82 overlapping genes for both time points. A prediction model was built using the 82 genes from the Day 3 time point. Internal validation has an AUC of 0.81, suggesting that this classifier was also able to discriminate high ADCP from low ADCP measured at week 2 after vaccination. This model, consisting of 82 expressed genes, was then tested at the week 2 time point and was also able to predict ADCP as a dichotomous variable at the week 2 time point (AUC = 0.75). We thank the reviewer for this suggestion and we have now included these data as Figures 3B and 4B.

    Reviewer #3 (Public Review):

    Strong points:

    1. This provides a novel mechanism into the RV144-mediated protection of HIV acquisition.

    2. The analyses are robust and statistically sound.

    3. The flow of the paper/figures is easy to follow.

    Weak points:

    1. the RV306 trial (Figure 3 A and B) RNA-SEQ analysis vs ADCP could benefit from a little more information:

    Are the 118 / 93 genes at Wk2 / Day 3 post-vaccination overlapping a lot?

    Per the reviewer’s suggestion we looked for overlapping genes between week 2 and day 3 post the RV144 immunization series in the RV306 study. There are 82 genes in common between the enriched genes at week 2 and day 3 ADCP data which are now detailed in Supplementary table 2. The number of enriched genes in the pathway at these two timepoints are summarized in the following Venn diagram and are now included in the manuscript as Figure 4A.

    What are those genes? Do they play a known direct role in ADCP or are they upstream regulators?

    All 82 genes are listed in supplementary table 2. There is not a lot of information about genes associated with ADCP specifically from previous publications, but when querying existing databases for genes associating with phagocytosis, we identified four of the 82 genes in the GO:0006909 phagocytosis pathway including SIRPA, SIRPB1, RAB20, and TYROBP. When using GeneMANIA a gene function prediction tool to investigate interaction networks of the 82 overlapping genes we identified 44 additional genes that were connected to the 4 genes previously implicated in phagocytosis. We have included this information as a new supplementary table 4.

    We also show other canonical pathways with gene membership including the Immune System (33 genes), Innate Immune System (23), Signaling by Interleukins (9), Hallmark Inflammatory Response (7), Hallmark TNFA Signaling Via NFKB (7), Cell-Cell Communication (5), Interleukin-10 Signaling (4), Signal Regulatory Protein Family Interactions (3), and Pentose Phosphate Pathway (3). We now include this information in the results and discussion section and hope that this information will clarify the field further (new Figure 4D, new Supplementary table 3).

    Perhaps a heatmap representation with the ADCP as an annotation track would help unfamiliar readers better understand.

    We have now included a heatmap that shows gene expression of the 82 genes at both timepoints, with ADCP group status annotated (Fig. 4C). The list of the 82 genes are also available in Supplementary Table 2 – (“Yes” for enrichment in the “RV306 ADCP day3” and “RV306 ADCP wk2” columns).

    1. I would nuance that ADCP is "A" primary mechanism, not "THE" (title). There could be more potent unidentified mechanisms, so the usage of "THE" in the title is in my opinion premature.

    The title has been updated accordingly.

    1. While I agree that it is possible that ADCP is a primary mechanism with the previously identified transcriptomic signature given the evidence, we cannot exclude that the signature in fact represents an upstream regulator of ADCP, inducing a myriad of cascades contributing to vaccine-induced protection. If that were the case, ADCP could be higher in individuals with higher protection without it being directly involved in that protection (more of a collateral effect). Showing an enrichment of ADCP-associated genes from external datasets with the tested gene signature would strengthen at least partly that this is a direct phenomenon. Otherwise, I would nuance the statement and say that ADCP is a likely/potential mechanism of vaccine-induced protection.

    We agree with these nuances and have updated the title and discussion accordingly (Lines 1-2, 314-316).

    1. Observations in Figure 4 are glanced too quickly in the Results section: this would require a more in-depth description.

    Based on the results from the current revisions we have updated the previous Figure 4 (now Figure 5) to provide an indepth description of gene function prediction based on networks. We have used GeneMANIA, which is an application that can find associated genes or pathways using its functional association data, to examine overlapping enriched genes from the different studies with either infection status or ADCP magnitude. Interestingly, TYROBP, which is associated with phagocytosis, is the gene with the most connections to other genes or pathways. We also do a clustering analysis to identify highly interconnected sets of genes and pathways from the enrichment results of different studies. We describe this now in the results section (lines 209-219), updated Figures 5A-B and discussion section (lines 299-319).

    1. It is not clear whether the expression level per monocyte for the subset of genes tested in the CITE-seq data is different in patients with higher ADCP vs those with lower ADCP, or is the differential enrichment the result of a different number of cells that express this signature? Or both?

    For the CITE-seq data we performed differential expression analyses transcriptome-wide and found that monocytes had a higher frequency of differentially expressed genes when comparing higher versus low ADCP (Fig 5E). This effect was independent of frequency of the different cell populations with a frequency of >1%. We now include a sentence to clarify our findings (results last paragraph) and show the frequency differences in the supplementary section (Supplementary Figure 3).

  3. eLife

    Evaluation Summary:

    Attempts to produce effective vaccines against HIV have not yet been successful, with the lack of understanding correlates of protection being a significant limitation. This paper analyses gene expression in a number of human and non-human primate vaccine trials and identifies a profile that appears to correlate with protection from infection. This profile is linked primarily to monocytes and the ability of these cells to mediate antibody dependent cellular phagocytosis. The work has implications for ongoing attempts to generate effective vaccines against HIV and perhaps other viral diseases.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #2 agreed to share their name with the authors.)

  4. eLife

    Reviewer #1 (Public Review):

    The authors carried out a post-hoc analyses of a protective gene expression signature previously observed in preclinical trials and clinical trials (RV144 and HVTN505) to identify a possible correlate of reduced risk of infection and whether able to provide a potential mechanism for protection. This monocyte signature they focus on was absent in the DNA/rAd5 human vaccine trial which did not show efficacy and was enriched in the partially effective RV144 human trial where the vaccine was and ALVAC/protein vaccine. Here they indicate that the signature is a correlate of reduced risk of infection.

    Identifying signatures of protection is an important issue in the development of a HIV vaccine, and signature analyses might be important to reveal a few markers that might be selected to evaluate vaccine trials. However, this analysis must be able to point to very few genes, as single cell analyses are not an option in a large clinical vaccine trial.

    It is unclear whether or how the conclusion of the previous publication by many of the same authors of this paper, including the senior author (Ehrenberg et al., 2019, identification of a gene signature in B cells that is associated with protection from SIV and HIV infection providing a new approach for evaluating future vaccine candidates) is compatible with this new one: signature primarily expressed in myeloid lineage being the one most consistently associated with vaccine efficacy. It is unclear which one of the two is correct or how they are reconciled. Was the single cell analysis done in monocytes only for this paper or simply not reported in the studies of Ehrenberg et al., 2019?

    Figure 1: The gene expression score (GES) of this figure does not seem to be for a specific cell type. It is unclear how the GES reported here relates to the final GES of monocytes. What is the utility of this analysis? Can we observe here the same most significant genes that we observe in monocytes? This is important because if bulk analysis gives the same results as looking at monocytes an eventual marker identified in monocytes could be evaluated in luck analysis.

    Figure 2: it would be good to know whether the subset of the 63 genes can be restricted to the most significant and their GES can still retain the predictive value.

    Figure 3 deals with genes associated with antibody dependent cellular phagocytosis (ADCP). Can one derive a gene or a few genes that are predictive of significant ADCP?

  5. eLife

    Reviewer #2 (Public Review):

    Shangguan, Shida et al report here that a "protective" gene set correlating with decreased risk in sorted B-cells from NHP immunized with Ad26 regimens, is also enriched in microarray of total in cultured PBMCs from human immunized with the moderately protective ALVAC-based vaccine in the RV144 HIV vaccine trial, but not in HVTN505, that afforded no protection against HIV.

    Strengths:

    The analyses of the efficacy of different HIV vaccine candidates across different species is challenging but is an important goal. Important findings of the current work stem from analysis of sorted B-cells and monocytes from the non-efficacious trial, HVTN505 and from the single cell analyses from peripheral blood in the RV306 immunogenicity human trial, where volunteers were immunized with the same regimen used in RV144 and received an additional vaccine boost. They found that the gene set that correlated with protection in RV144 was associated with monocytes, and not B-cells as previously thought and importantly, that the gene set correlating with protection in RV144, was not enriched in the HVTN505 failed trail. In addition, in the RV306 immunogenicity trial they found that this gene set correlated also with antibody-dependent cellular phagocytosis (ADCP). This finding supports the hypothesis that ADCP could be a mechanism of protection against HIV, a testable hypothesis in future efficacious HIV vaccine trial.

    Weaknesses:

    The analysis of data from additional failed trails in NHP could strengthen the authors' conclusions

  6. eLife

    Reviewer #3 (Public Review):

    Strong points:

    1. This provides a novel mechanism into the RV144-mediated protection of HIV acquisition.

    2. The analyses are robust and statistically sound.

    3. The flow of the paper/figures is easy to follow.

    Weak points:

    1. the RV306 trial (Figure 3 A and B) RNA-SEQ analysis vs ADCP could benefit from a little more information:

    Are the 118 / 93 genes at Wk2 / Day 3 post-vaccination overlapping a lot ?

    What are those genes ? Do they play a known direct role in ADCP or are they upstream regulators? Perhaps a heatmap representation with the ADCP as an annotation track would help unfamiliar readers better understand.

    2. I would nuance that ADCP is "A" primary mechanism, not "THE" (title). There could be more potent unidentified mechanisms, so the usage of "THE" in the title is in my opinion premature.

    3. While I agree that it is possible that ADCP is a primary mechanism with the previously identified transcriptomic signature given the evidence, we cannot exclude that the signature in fact represents an upstream regulator of ADCP, inducing a myriad of cascades contributing to vaccine-induced protection. If that were the case, ADCP could be higher in individuals with higher protection without it being directly involved in that protection (more of a collateral effect).

    Showing an enrichment of ADCP-associated genes from external datasets with the tested gene signature would strengthen at least partly that this is a direct phenomenon.

    Otherwise, I would nuance the statement and say that ADCP is a likely/potential mechanism of vaccine-induced protection.

    4. Observations in Figure 4 are glanced too quickly in the Results section: this would require a more in-depth description.

    5. It is not clear whether the expression level per monocyte for the subset of genes tested in the CITE-seq data is different in patients with higher ADCP vs those with lower ADCP, or is the differential enrichment the result of a different number of cells that express this signature ? Or both?

  7. Version published to 10.1101/2021.05.12.443737v1 on bioRxiv