Merck Ad5/HIV induces broad innate immune activation that predicts CD8⁺ T-cell responses but is attenuated by preexisting Ad5 immunity

Abstract

To better understand how innate immune responses to vaccination can lead to lasting protective immunity, we used a systems approach to define immune signatures in humans over 1 wk following MRKAd5/HIV vaccination that predicted subsequent HIV-specific T-cell responses. Within 24 h, striking increases in peripheral blood mononuclear cell gene expression associated with inflammation, IFN response, and myeloid cell trafficking occurred, and lymphocyte-specific transcripts decreased. These alterations were corroborated by marked serum inflammatory cytokine elevations and egress of circulating lymphocytes. Responses of vaccinees with preexisting adenovirus serotype 5 (Ad5) neutralizing antibodies were strongly attenuated, suggesting that enhanced HIV acquisition in Ad5-seropositive subgroups in the Step Study may relate to the lack of appropriate innate activation rather than to increased systemic immune activation. Importantly, patterns of chemoattractant cytokine responses at 24 h and alterations in 209 peripheral blood mononuclear cell transcripts at 72 h were predictive of subsequent induction and magnitude of HIV-specific CD8⁺ T-cell responses. This systems approach provides a framework to compare innate responses induced by vectors, as shown here by contrasting the more rapid, robust response to MRKAd5/HIV with that to yellow fever vaccine. When applied iteratively, the findings may permit selection of HIV vaccine candidates eliciting innate immune response profiles more likely to drive HIV protective immunity.

A highly efficacious HIV vaccine offers the greatest promise to halt the HIV pandemic. Results of the RV144 study conducted in Thailand, where a canarypox vector prime and subunit protein boost regimen showed 31% efficacy for reducing HIV-1 acquisition (1), have given hope that development of a successful HIV vaccine is possible, and suggest that the vector prime is important for shaping a protective response. Innate immune responses direct the adaptive immune response and thus influence the potential for inducing long-lived protective immunity (2). A comprehensive understanding of the molecular programs underlying optimal innate responses would therefore facilitate enhanced vaccine design. Little is known at present about the innate immune responses induced by candidate HIV vaccines, how these responses drive adaptive immunity, and how these innate responses compare with those induced by licensed efficacious vaccines against other pathogens.

To begin to fill these gaps in our knowledge, we conducted a phase Ib clinical trial (HVTN 071) to analyze, at the systems level, human innate immune responses to the replication-incompetent Merck adenovirus serotype 5 vaccine vector containing HIV-1 inserts gag/pol/nef (MRKAd5/HIV), in parallel with two phase IIb efficacy trials being conducted using the same vaccine. Although this vaccine did not offer protection from HIV acquisition or lower viral loads in the phase IIb Step or Phambili studies (HVTN 502 and 503), it elicited high CD8⁺ T-cell response rates to the HIV-1 inserts (3–5), and recent sieve analyses provide evidence that vaccine responses exerted selective pressure on infecting HIV-1 strains (6). The MRKAd5/HIV vaccine received particular attention when the Step Study analysis revealed that certain vaccine subgroups with baseline Ad5 seropositivity exhibited increased HIV-1 acquisition rates, halting its further use in all HIV-1 vaccine trials involving Ad5 seropositive subjects. Although hypotheses have been generated that may explain vaccine-induced increased HIV-1 infection rates (3, 7, 8) and enhanced acquisition was recently recapitulated in the simian immunovirus (SIV) challenge model (9), no clear mechanisms have been identified to date. These findings, coupled with the importance of the Ad5 and other adenovirus serotype vectors to vaccine development against many other pathogens (10, 11), reinforced our motivation to use an unbiased systems biology approach to better understand the innate immune response triggered by MRKAd5/HIV.

Systems biology integrates global molecular measurements and computational analysis with prior knowledge to generate holistic biological insights. This approach therefore provides a framework to address complex vaccine-induced immunological responses (12, 13). Crosstalk and feedback can be elucidated between immune signaling pathways and gene regulatory networks operating on multiple spatial and temporal scales. We have previously applied systems analysis to identify gene and signaling networks that coordinately amplify and attenuate Toll-like receptor (TLR)-mediated responses underlying innate immune cell activation (14–17). Recent systems analyses of responses to vaccination with the highly efficacious YF-17D yellow fever vaccine (18, 19) and seasonal influenza vaccine (20) have yielded novel insights about their mechanisms of action. Building on this systems-level approach, we describe here the innate immune responses induced by MRKAd5/HIV, how they are impacted by preexisting Ad5 neutralizing antibodies (nAb), how they relate to induction of T-cell responses, and how they differ from those induced by live-attenuated YF-17D.

Results

MRKAd5/HIV Dramatically Remodels Peripheral Blood Mononuclear Cell Transcriptomes by Triggering Robust Innate Immune and Cell Trafficking Responses.

We assessed the innate immune response to MRKAd5/HIV by profiling transcriptomes of peripheral blood mononuclear cells (PBMC) isolated from seven Ad5 nAb seronegative individuals (Ad5 nAb titer ≤18; Ad5^Neg) during the first week after vaccination, by gene-level analysis of Affymetrix exon microarrays. Responses to MRKAd5/HIV peaked at 24 h, with 1,026 genes exhibiting enhanced and 1,048 genes exhibiting repressed expression levels compared with prevaccination (Fig. 1A and Dataset S1, tab 1). At 72 h postvaccination, the differentially expressed genes were a small subset of those detected at 24 h (Dataset S1, tab 2). No significantly differentially expressed genes were detected at 168 h.

Fig. 1.

Systems analysis identifies widespread innate immune activation and cellular trafficking responses response to MRKAd5/HIV vaccination in humans. (A) In vivo PBMC transcriptional responses to vaccination with MRKAd5/HIV [n = 7 Ad5 seronegative individuals, false-discovery rate (FDR) ≤ 10%, absolute average log₂ fold-change ≥ 0.5]. Genes significantly differentially expressed in response to MRKAd5/HIV vaccination at any time point are annotated and grouped according to membership in functional gene modules (21, 64). Each column represents subject-specific log₂(fold-changes) compared with prevaccination. To emphasize regulation patterns, expression fold-changes for each gene are scaled by the maximum observed expression response. Pink intensity indicates up-regulation compared with prevaccination, cyan indicates down-regulation. (B) Functional gene modules significantly differentially expressed in PBMC in response to vaccination. Each black line represents the average log₂(fold change) of all genes in the module for a single subject. Up-regulated modules have values >0, down-regulated modules have values <0. FDR < 10%, |average log₂(fold-change)|>0.5. (C) Quantitative RT-PCR validation of the differential expression of module-associated genes identified by microarray analysis. Each line represents the response of one individual (n = 7). (D) Protein-level validation of vaccine-induced cytokine and chemokine differential expression by multiplex analyte analysis. (E) Validation of vaccine-induced cellular fluxes predicted by module-level analysis of the PBMC transcriptional responses. *P < 0.05 with Hochberg adjustment from the statistical model assessing change in concentration over time.

We used a modular analysis framework (21) to interpret the transcriptional response. This approach deconvolutes complex transcriptional profiles into functionally interpretable patterns through the evaluation of combined expression responses of predefined disease, cell type, and stimulus-specific coexpressed gene groups. We used versions of the functional modules defined by Chaussabel et al. (21, 22) that were updated through meta-analysis of a much larger transcriptional dataset encompassing many more disease states (23), to annotate the differentially expressed gene lists and to examine the differential expression of the overall modules themselves. We confirmed the functional annotations of the gene modules themselves by performing canonical pathway enrichment analysis (Dataset S1, tab 3). Mirroring the gene-level results, the modular response peaked at 24 h (13 up-regulated and 11 down-regulated modules), waned by 72 h (two up-regulated modules), and returned to baseline by 168 h (Fig. 1B). Modules induced by MRKAd5/HIV were associated with cell intrinsic innate immune responses (“Inflammation” and “Interferon response” modules) and influx of inflammatory cells (“Myeloid lineage” module). Concomitantly, the “Lymphoid lineage,” “T cells,” and “Cytotoxicity” modules were suppressed, leading to the hypothesis that the vaccine was stimulating an influx of myeloid cells and an efflux of lymphoid cells from the circulation. This hypothesis was further supported by comparing the lists of up- and down-regulated genes with published cell-type enriched gene lists generated from meta-analysis of a compendium of sorted cell transcriptomes (20). Thirty-two percent of the genes we detected as up-regulated at 24 h were identified as preferentially expressed in monocytes in that study, whereas 28% of the down-regulated genes we detected were preferentially expressed in lymphocytes (Dataset S1, tab 1). Rapid lymphocyte trafficking in response to MRKAd5/HIV is consistent with similar observations made in previous studies with an adenoviral-vectored vaccine (24). Direct canonical pathway enrichment analysis of the regulated gene sets provided additional support for the module analysis results, indicating that innate immune pathways and cell types were up-regulated in response to vaccination, and lymphocyte cell types and pathways were down-regulated (enrichment results and pathway figures in Dataset S1, tabs 4 and 5).

We validated the microarray results at the transcript, protein, and cellular levels. First, we quantified and confirmed the differential expression of mRNAs associated with several vaccine-regulated modules, including “Interferon response” [C-X-C motif chemokine 10 (CXCL10), ISG-15, and STAT1] (Fig. 1C). Next, we corroborated the differential expression of many cytokines and chemokines at the protein level using multiplex serum analyte analysis (Fig. 1D and Dataset S1, tab 6), detecting robust changes in serum levels of IP-10, I-TAC, monocyte chemoattractant protein-1 (MCP-1), and MCP-2, as well as immunoregulatory IL-10 and IL-1Ra. Finally, we validated the cellular trafficking responses predicted from the modular analysis by directly assessing circulating peripheral blood leukocyte concentrations (Fig. 1E and Dataset S1, tab 7), confirming vaccine-induced influx of monocytes and pronounced efflux of lymphocyte populations (T, B, and NK cells). Monocyte increases are likely a result of recruitment from the bone marrow in response to MCP-1 and other chemokines (25). Taken together, these results validate the robust systemic innate immune response to MRKAd5/HIV revealed by the transcriptional profiling.

The In Vivo Innate Immune Response to MRKAd5/HIV Is Recapitulated in Vitro and Engages a Coordinately Regulated Interacting Network Involving Unique Gene Isoforms.

To decouple the in vivo innate responses intrinsic to the circulating cells from those associated with cells trafficking into and out of the circulation, we extended our transcriptional profiling to PBMC stimulated with the vaccine vector in vitro. We profiled RNA from unstimulated PBMC and PBMC incubated for 24 h with MRKAd5 at a dose sufficient to induce robust cytokine responses (Fig. S1). We found that 8 of 13 (62%) modules induced in vivo were also induced in vitro and these consisted of the three “Interferon response modules” as well as unannotated modules largely comprised of innate immune response genes (Fig. 2A). Remarkably, 92% concordance between the in vivo and in vitro induction of IFN response genes was observed (Dataset S1, tab 8). Many of the modules discordant between the in vitro and in vivo responses were associated with particular cellular lineages (myeloid, lymphoid, T cell, B cell) or cell-type specific attributes (cytotoxicity) (Fig. 2B), suggesting that the much of the discrepancy between the in vivo and in vitro responses arose from an absence of cell trafficking in vitro. Comparison with cell-type specific genes lists (20) indicated 35% of the genes up-regulated in vivo but not in vitro are preferentially expressed in monocytes (Dataset S1, tab 8), supporting this hypothesis.

Fig. 2.

Responses to MRKAd5/HIV are recapitulated in vitro and involve coordinate regulation of an innate immune network involving unique gene isoforms. (A and B) In vitro responses of modules regulated by MRKAd5/HIV in vivo. Each black line represents the average log₂ fold-change of all genes in the module for PBMC derived from separate donors and stimulated in vitro (n = 4); red line indicates average module-level response in vivo. *P ≤ 0.01 comparing module fold-changes observed in vitro those observed in vivo. (A) Gene modules consistently regulated by MRKAd5 in vivo and in vitro. (B) Gene modules regulated in vivo but not in vitro. (C) Single gene heatmaps showing average exon-level expression responses to MRKAd5/HIV in vivo (n = 7) and in vitro (n = 4). Pink intensity indicates up-regulation, cyan indicates down-regulation. (D) Innate immune response interaction network of genes consistently induced in response to MRKAd5 in vivo and in vitro, arranged by representative subcellular localization. Node colors indicate functional module associations; node shapes indicate the mode of differential expression. Red lines indicate protein-protein interactions, gray lines indicate mutual membership in common complexes, and blue edges indicate protein–DNA interactions.

Our exon-level transcriptional analyses from previous studies demonstrated that defective alternative mRNA splicing results in profound phenotypic differences in memory T cells (26), and that alternative exon use occurs in the innate response. We therefore extended our analysis of the MRKAd5-induced innate response to the exon-level to further enrich our understanding of the action of the vaccine, with the primary focus of identifying vaccine-regulated genes not already detected by the gene-level analysis, particularly those behaving concordantly in vitro and in vivo. Exon-level analysis led to the identification of 94 additional vaccine-induced genes in vivo and in vitro (Dataset S1, tab 9), including critical innate immune pathway genes (TLR3, RIPK1, and NLRC5) and several genes with important roles in HIV infection (APOBEC3G, APOBEC3F, CCR5, and CD74). Additionally, alternative transcription analysis identified 16 genes with vaccine-induced responses that varied strongly from exon to exon, but were nevertheless consistent in vitro and in vivo, including FANCA, FARP2, RERE, GBP6, and GBP7 (Fig. 2C and Dataset S1, tab 10). Although the IFN-γ–induced antimicrobial GTPases GBP6 and GBP7 have been associated with immune responses, most of the other 16 genes have not been, suggesting additional leads that could be investigated to further understand vaccine-induced immunological memory. Induction of the unique short isoform of FANCA as part of the MRKAd5-induced innate immune response, for example, provides a compelling link between DNA damage pathways and the immunogenicity of adenoviral vectors.

We performed interaction network analysis to determine whether the genes commonly regulated by MRKAd5 in vitro and in vivo constituted established pathways or represented isolated nodes. This analysis revealed a densely interconnected network involving multiple modules and included genes detected by gene-level analysis, exon-level analysis, and alternative transcription analysis [visualized using Cytoscape (27) in Fig. 2D]. These findings indicate coordinate regulation of large functional subnetworks, including viral nucleic acid sensors, innate immune adaptors, inflammasome components, and antiviral effectors.

Preexisting Neutralizing Antibodies to Ad5 Attenuate the Innate Immune Response to MRKAd5/HIV.

An important observation in the Step Study was that the presence of Ad5 nAb before vaccination resulted in increased postvaccination risk of HIV acquisition (4), and thus far, no clear mechanism for this has been elucidated (3, 7, 8). We therefore analyzed the in vivo innate immune responses of vaccinated Ad5 seropositive subjects to determine if we could identify alternate programs of innate activation in these individuals. The early termination of the Step and Phambili studies resulted in the cessation of HVTN 071, limiting the number of subjects we could analyze. Given the possibility of threshold effects of Ad5 nAb (3, 4), we compared the responses between subjects with Ad5 nAb titers ≤ 200 and >200, and thereby identified 306 seropositivity effect genes (302 at 24 h, six at 72 h) for which the vaccine-induced responses were markedly attenuated (Dataset S1, tab 11). Canonical pathway enrichment analysis of these genes revealed that induction of complement pathways, innate immune sensors, and G-protein coupled receptor signaling was significantly attenuated (Dataset S1, tab 12). This attenuation extended to all modules regulated by the MRKAd5 vaccine (Fig. 3A), including impaired down-regulation of lymphocyte modules and genes preferentially expressed in lymphocytes (Dataset S1, tab 11), suggesting suppression of the acute lymphopenia observed in the Ad5 seronegative subjects (Fig. 1E). Direct comparison between regulation of innate immune networks in seronegative and Ad5 nAb >200 subjects revealed coordinate dysregulation that included the RIG-I, NLR/inflammasome, and TLR pathways (visualized using Cytoscape in Fig. 3B). Finally, we validated these transcriptional results at the protein level by analyzing serum analytes from a larger set of vaccinated subjects (Fig. S2). Consistent with the transcriptional results, cytokine responses were markedly attenuated in Ad5 nAb >200 subjects compared with Ad5 nAb ≤ 200 subjects (Fig. 3C and Dataset S1, tab 13). Taken together, these results suggest that the predominant effect of preexisting Ad5 nAb on the innate immune response is global attenuation. These data do not support the hypothesis that preexisting immunity leads to enhanced systemic innate immune activation.

Fig. 3.

Preexisting Ad5 neutralizing antibodies attenuate the MRKAd5/HIV-induced innate immune response in vivo. (A) Effect of Ad5 nAbs on gene modules regulated by MRKAd5/HIV. Black line represents the average fold-change of all genes in the modules, averaged over 8 Ad5 nAb ≤ 200 subjects. Green lines represent the average fold-change of all genes in the module for the two Ad5 nAb > 200 subjects. (B) Attenuated induction of critical innate immunity pathways in Ad5 nAb > 200 compared with Ad5 nAb ≤ 200 subjects. Nodes are colored according to average extent of induction in Ad5 nAb ≤ 200 subjects at 24 h (Left) and average extent of response attenuation in Ad5 nAb > 200 subjects compared with Ad5 nAb ≤ 200 subjects (Right). Edges represent protein–protein interactions between nodes obtained from InnateDB (65). (C) Representative attenuation of MRKAd5/HIV serum cytokine induction in Ad5 nAb > 200 subjects. Serum IP-10/CXCL10 concentrations are contrasted between Ad5 nAb ≤ 200 (n = 27) and Ad5 nAb > 200 (n = 6) individuals. Shading represents the interquartile ranges with the 75th percentile shown on top and the 25th percentile shown below the median line. *P < 0.05 after Hochberg adjustment.

MRKAd5/HIV Innate Immune Responses Predict Immunogenicity.

We next identified MRKAd5/HIV-induced innate immune signatures that predict subsequent HIV-specific adaptive immune responses. Based on the frequency of Gag-specific CD8⁺ T-cell responses detected at day 28 after one immunization (Fig. S3), we categorized vaccine recipients (n = 31) into high, moderate, or low responders. We determined whether fold-changes in serum cytokine concentrations, measured 24 h postvaccination (Fig. 1D and Dataset S1, tab 6), could predict the Gag-specific CD8⁺ T-cell response magnitudes. We performed two analyses: (i) discrimination between subjects with detectible (CD8^pos = CD8^mod and CD8^high) and undetectable (CD8^neg) responses; and (ii) discrimination between subjects with high (CD8^high) and moderate or undetectable (CD8^mod, CD8^neg) responses. The predictive potential of individual cytokines and all cytokine pairs was evaluated by 60 iterations of eightfold cross-validation of linear discriminant analysis (LDA) classifiers.

Two chemokines, MCP-1 and MCP-2 (Fig. S4 A and B), discriminated between CD8^mod/CD8^high subjects and CD8^neg subjects with high accuracy (81% and 88%, respectively), and thus were qualitatively predictive of the vaccine CD8⁺ T-cell immunogenicity. In both cases, higher chemokine induction predicted increased likelihood of positive CD8⁺ T-cell responses. Combining MCP-1 and MCP-2 into a single classifier did not increase predictive accuracy. However, the accuracy was increased by combination with other cytokines that were not individually predictive (Fig. 4A and Fig. S4A). For example, the growth factor PDGF-AA was 71% predictive individually but 85% predictive in combination with MCP-1. The network of predictive pairwise signatures for CD8⁺ T-cell responses is shown in Fig. 4A, and the receiver operating characteristic (ROC) for predicting positive CD8⁺ T-cell responses based on GRO and MCP-2 is shown in Fig. 4B. Classifiers performing as well as MCP-2 individually (or top pairs involving MCP-2) were generated only 13% of the time when the analysis was repeated on randomized datasets, indicating that the result is moderately robust. Nevertheless, repeat analyses using similar vaccines are required to confirm the association between these chemokines and CD8⁺ responses. In the second analysis, the combination of RANTES (regulated upon activation, normal T cell expressed and secreted) and IL-28A was predictive of CD8⁺ response magnitudes with high accuracy (87%), even though neither cytokine was strongly predictive individually (Fig. S4 C and D). Strong down-regulation of RANTES or up-regulation of IL-28A was associated with induction of high magnitude CD8⁺ T-cell responses (Fig. S4D). The ROC for predicting high magnitude CD8⁺ T-cell responses based on IL-28A and RANTES is shown in Fig. 4C. Repeating the analysis on randomized datasets generated classifiers performing as well as IL-28A and RANTES 25% of the time, indicating that this particular result should be regarded as a hypothesis until additional studies have validated the role of these cytokines in the fine tuning of adenoviral vector CD8⁺ T-cell immunogenicity.

Fig. 4.

Vaccine-induced serum cytokine and chemokine responses predict the magnitudes of CD8⁺ T-cell responses induced by MRKAd5/HIV. (A) Network depiction of serum cytokines and chemokines with 24-h vaccine-induced fold-changes that discriminate subjects who develop CD8⁺ T-cell responses (28 d after vaccination) from those who do not. Node sizes indicate predictive accuracy of the factors individually (percentages inside nodes), node color, and intensity indicate positive (pink) or negative (blue) correlations between induction levels of the factor and CD8⁺ T-cell response magnitudes. The presence of an edge between nodes indicates that the two cytokines in combination have increased predictive accuracy compared with either cytokine alone. Edge width is proportional to predictive accuracy of the pair-wise signature of the connected nodes (red percentages). (B) ROC for predicting which subjects will develop CD8⁺ T-cell responses, based on 24-h vaccine induced fold-changes of GRO and MCP-2 in a LDA model. The red line indicates the tradeoff in false positive rate required as the true positive rate is increased. Prediction accuracies were estimated from over 60 rounds of eightfold cross-validation. AUC, area under the curve. (C) ROC for predicting which subjects will develop high magnitude CD8⁺ T-cell responses, based on 24-h vaccine induced fold-changes of IL-28A and RANTES in a LDA model. Notation is as in B.

Using Systems Biology to Generate Additional Hypotheses Regarding the Immunogenicity of MRKAd5/HIV.

To identify additional potential mechanisms controlling MRKAd5/HIV-induced T-cell responses, we extended our signatures analysis to the transcriptional level. Given our small data sample sizes, it was not possible to implement approaches described above or in previous studies (19, 20), and the results must be regarded as hypothesis-generating until clinical studies using similar vaccines make validation of the results possible. We defined groups of subjects with high-, moderate-, or low-magnitude CD8⁺ T-cell responses by 1D clustering of the gag-specific responses. Genes with statistically significant differences in vaccine-induced transcriptional responses between the high and low CD8⁺ groups were then identified through direct comparison. Surprisingly, significant differences between these groups of subjects were only identified from the transcriptional responses measured 72 h postvaccination (88 genes were positively associated and 121 genes were negatively associated) (Fig. 5 A and B and Dataset S1, tab 14), and none of the MRKAd5-responsive modules differed significantly between the groups. Several of the implicated genes have clear functional relationships to cytotoxic responses, including the inhibitory killer cell Ig-like receptor KIR2DL1, the NK-cell activating receptor CLEC2D (28), and the NK-cell signaling adaptor EWS-FLI1–activated transcript 2 (EAT-2) (29) (Fig. 5 A and B). Consistent with this association between EAT-2 expression and CD8⁺ T-cell responses, it was recently reported that adenoviral expression of EAT-2 as part of a vaccine strategy enhanced vaccine-induced T-cell responses (30). Because interaction network analysis of the overall CD8⁺ T-cell response gene set itself was found to be uninformative, we investigated whether any the gene set members are protein–protein interaction neighbors of genes belonging to MRKAd5 regulated functional modules (Fig. 1B). We found that many of the CD8⁺ T-cell response associated genes are nearest neighbors of members of the “Cytotoxicity,” “T cells,” and “Lymphoid lineage” modules (Fig. 5C), providing additional support for the association between these genes and CD8⁺ T-cell immunogenicity.

Fig. 5.

Systems analysis identifies innate immune response genes that are associated with the immunogenicity of MRKAd5/HIV. (A) The 209 genes with 72-h MRKAd5/HIV-induced expression responses that are positively (Upper) or negatively (Lower) associated with the magnitude of MRKAd5/HIV-induced CD8⁺ T-cell responses. Subjects with low and high magnitude CD8⁺ T-cell responses are on the Left and Right halves of the heatmap, respectively. Gene selection criteria: FDR ≤ 20% and |average log₂(fold-change)| ≥ 0.5. Rows are scaled as in Fig. 1A. Pink intensity indicates up-regulation compared with prevaccination, cyan indicates down-regulation. (B) Seventy-two hour MRKAd5/HIV-induced expression responses of two representative genes, stratified by the magnitude of the CD8⁺ T-cell responses observed in the same subjects. (C) Protein–protein interaction network emphasizing links between CD8⁺ T-cell response-associated genes identified in Fig. 5A (triangles) and constituents of functional gene modules differentially regulated by MRKAd5/HIV (circles), colored according to module associations.

Replication-Incompetent MRKAd5 Induces a Greater Number of Innate Immune Genes than Does Live-Attenuated YF-17D, but the Response Is More Transient.

Recent studies have suggested that the efficacy of live-attenuated YF-17D, a yellow fever vaccine (31), may result from robust innate immune activation (18, 19, 32). We therefore contrasted the transcriptional responses induced MRKAd5 and published in vivo profiles for YF-17D (19). Although the MRKAd5/HIV vaccine induced more than 1,000 genes and repressed a similar number, the YF-17D vaccine only induced 181 genes and repressed 10 genes (Fig. 6A). However, the response to MRKAd5/HIV was rapid and transient, but the response to YF-17D lagged and was persistent (Fig. 6A). Modular analysis further illuminated differences between the two vaccines (Fig. 6B). Whereas MRKAd5/HIV induced the “Inflammation,” “Interferon response,” and “Myeloid lineage” modules and inhibited the “Lymphoid lineage,” “T cells,” and “Cytotoxicity” modules (Figs. 1B and 4B), YF-17D vaccination induced only a subset of the “Interferon response” modules (M1.2 and M3.4) (Fig. 6B).

Fig. 6.

MRKAd5 induces more extensive innate immune activation than the gold-standard yellow fever vaccine, YF-17D. (A) MRKAd5/HIV and YF-17D trigger innate immune responses in PBMC with distinct kinetics in vivo. The number of genes significantly up-regulated (upper half) or down-regulated (lower half) in response to vaccination with MRKAd5/HIV (red, present study) or YF-17D [blue (19)] are plotted. Significant differential expression defined as FDR ≤ 10%, |average[log₂(fold-change)]| > 0.5; n = 7 (MRKAd5/HIV), n = 25 (YF-17D). (B) MRKAd5/HIV regulates a broader spectrum of gene modules in vivo than does YF-17D. Each line represents the average fold-change of all genes in a module for a given individual (red lines, MRKAd5/HIV vaccines; blue lines, YF-17D vaccinees). Modules labeled in red differ significantly between MRKAd5/HIV and YF-17D in vivo (FDR ≤ 10%), modules labeled with asterisks differ significantly between MRKAd5 and YF-17D in vivo and in vitro (FDR ≤ 10%). (C) Average expression responses in vivo and in vitro of 43 genes preferentially induced by MRKAd5. Genes associated with functional gene modules are indicated. Rows are scaled as in Fig. 1A. Pink intensity indicates up-regulation compared with controls, cyan indicates down-regulation. FDR < 10%, |average log₂(fold-change)|>0.5 for induction in response to MRKAd5 and comparing MRKAd5 to YF-17D, in vivo and in vitro. (D) Putative transcriptional regulatory network controlling innate immune responses preferentially induced by MRKAd5. Squares indicate transcription factors preferentially induced by MRKAd5 in vivo and in vitro; circles indicate target genes preferentially induced (pink) or repressed (light blue) by MRKAd5 in vivo and in vitro. Lines indicate protein-DNA transcription factor–target gene interactions identified from published ChIP-seq datasets [blue, KLF4 targets (38); purple, IRF1 targets (37); orange, STAT5A targets (39)]. Pink edges indicate IRF1 target genes identified by conventional methods (66–68). (E) Expression responses of CRIP3 and NPB (72–168 h postvaccination) are negatively associated with CD8⁺ T-cell response magnitudes induced by MRKAd5/HIV and YF-17D. Shown are log₂(fold-changes) compared with prevaccination of the two genes for both vaccines, stratified by the magnitudes of vaccine-induced CD8⁺ T-cell responses observed in the same subjects. Lines indicate mean values.

Given that dosage and replication kinetics could likely account for the gross differences in innate immune activation between replication defective MRKAd5 and live-attenuated YF-17D, we performed a new set of comparative in vitro experiments to directly contrast responses to the two vaccines, focusing first on differences identified in vivo that recapitulated in vitro. We identified 43 genes preferentially induced by MRKAd5/HIV in vivo that confirmed in vitro (Fig. 6C), including several associated with innate immune responses (IRF1), the complement pathway (C1QB), pathogen recognition (TLR8), the inflammasome (CASP10, P2RX7), and NK-cell activation [SLAMF7 (33)]. This group also included several immunosuppressive factors, including the T-cell inhibiting IDO1 (34), M2-macrophage skewing TF KLF4 (35), macrophage inhibitor PSTPIP2 (36), and the PD-1 death receptor ligand PDCD1LG2. The preferential induction of three transcription factors by MRKAd5, IRF1, KLF4, and STAT5A suggested that these factors may partly account for the PBMC transcriptome induced by MRKAd5. By mining published ChIP-Seq datasets (37–39), we confirmed that several MRKAd5-specific genes are direct targets of these transcription factors (Fig. 6D), and that these transcription factors potentially coregulate each other. Although there were no genes preferentially induced by YF-17D in vivo that validated in vitro, we identified a robust gene set, predominantly consisting of members of the “Interferon response” modules (including STAT1, STAT2, IRF7, and IFI27), that was induced by both vaccines in vitro and in vivo (Dataset S1, tab 15).

To generate hypotheses about differences in innate immune activation that may result at the actual sites of MRKAd5 or YF-17D vaccination, we also performed a direct comparison between the in vitro responses to the two vaccines, without constraining them by the in vivo results. Unexpectedly, the innate activation profiles of MRKAd5 and YF-17D differed more strongly in vitro than we had originally observed in vivo, with 349 and 313 genes being preferentially induced by MRKAd5 and YF-17D, respectively, compared with 226 genes being robustly induced in common (Dataset S1, tab 16). Similar differences in the down-regulated gene sets were observed, with 190 and 229 genes being preferentially down-regulated by MRKAd5 and YF-17D, respectively, compared with 137 genes being down-regulated in common. MRKAd5 preferentially induced T-cell chemoattractants (CXCL9/10/11), MHC genes, and T-cell–associated cytokines (IFNG, IL-2, and IL-7) but YF-17D preferentially induced the IFNA family of antiviral cytokines and several neutrophil chemokines (IL-8, CXCL2, -3, -5, and -6). Canonical pathway enrichment analysis reinforced the differences between the two vaccines, with MRKAd5-specific gene enrichments including “Antigen Presentation Pathway” and “T Helper Cell Differentiation” and YF-17D–specific gene enrichments, including “Systemic Lupus Erythematosus Signaling” and several IL-17 associated pathways (Dataset S1, tab 17). These results indicate that greater specificity in vaccine-induced innate immune responses may be revealed by profiling local, rather than systemic responses.

Finally, we evaluated whether the transcriptional signatures associated with enhanced CD8⁺ T-cell responses induced by MRKAd5/HIV (Fig. 5A) were also associated with enhanced CD8⁺ T-cell responses to YF-17D, despite the numerous difference between the vaccines. By reanalyzing published YF-17D transcriptome and longitudinal CD8⁺ T-cell response data (19) using the approach implemented above, we identified two genes, CRIP3 and NPB, with vaccine-induced expression responses that were consistently associated with impaired CD8⁺ T-cell responses to both vaccines (Fig. 6E). Strengthening the associations, the average fold-changes for these genes for moderate CD8⁺ response subjects was always between that of high and low CD8⁺ response subjects, even though the data for the moderate subjects was not used in the gene selection.

Taking these data together, we have defined the early innate immune response to the MRKAd5/HIV vaccine, identified an attenuated innate response in individuals with Ad5 nAb, and defined innate response signatures that predict CD8+ T-cell responses to Gag. These data suggest previously unexplored targets for enhancing the immunogenicity of next-generation HIV vaccines.

Discussion

Systems biology analysis can contribute to rational vaccine design in four major ways. First, it can enable the identification of correlates of immunogenicity and protection; second, it can reveal the regulatory networks within cells that lead to the desired host immune response; third, it can guide the reengineering of the vaccine regimen to favor desirable responses; and finally, it can supply tools to glean insight from failed candidate vaccines. We believe this study provides fresh understanding in each of these ways to the highly immunogenic but nonefficacious MRKAd5/HIV vaccine.

Despite inducing T-cell responses at a high frequency, MRKAd5/HIV neither reduced HIV-1 acquisition nor lowered viral loads postacquisition in two independent clinical trials (3, 5). Furthermore, Ad5 seropositive male vaccine recipients in the Step study showed an increased rate of HIV-1 acquisition, making the influence of Ad5 nAbs on vaccine responses an area of intense research (3, 4, 7, 8, 40–42). Although additional factors likely played a role in acquisition (43–45), the effect of Ad5 nAbs was significant and has been supported in the nonhuman primate SIV challenge model. One current hypothesis is that antibody-mediated internalization of Ad5 results in increased dendritic cell activation, which may lead to enhanced HIV-1 infectivity (40). In our study, we found no evidence for enhanced or prolonged systemic innate immune responses in volunteers with preexisting Ad5 nAb. Instead, we observed attenuation of the overall transcriptional response (Fig. 3 A–C), which we confirmed at the protein level by multiplex serum cytokine analysis (Fig. 3D and Dataset S1, tab 13). Our results are compatible with the suggestion that Ad5 nAb may effectively lower the dose of the vaccine detected by the innate immune system (8, 46) and are consistent with the reduced vaccine immunogenicity seen in vaccine recipients with nAb titers >200 (3). Our results are also compatible with a model in which nAbs negatively regulate innate signaling pathways. The latter hypothesis is of interest given the possibility that the opsonized vector could interact with Fc receptors on antigen presenting cells; an event that might result in an inappropriate context for presentation of the vaccine-encoded antigens. Regardless of the precise mechanism, our observations highlight the impact of preexisting type-specific immunity to the vector on vaccine responses and open new avenues for mechanistic studies into the effects of this important variable.

Few vaccine vectors in development match Ad5 in terms of the magnitude and frequency of vaccine insert-specific CD8⁺ T cells they induce (Fig. S3) (3). Ad5-induced CD8⁺ T cells are functional in some settings, because they are essential to the efficacy of Ad5-vectored Ebola vaccines in the nonhuman primate model (10) and also appear to have exerted selective pressure on infecting HIV-1 in the Step Study (6). Results in the murine model, however, show that Ad5 induced CD8⁺ T cells may not properly differentiate into memory cells required for protective responses.† These contrasting results suggest that although the strong Ad5-induced CD8⁺ T-cell response may be sufficient for vaccine efficacy in some systems, increases in the efficacy of Ad5-based vaccines may be achieved if the quality of the induced CD8⁺ T cells is optimized.

To determine how the magnitude and quality of vaccine-induced T-cell responses are shaped and may ultimately be optimized by activation of innate pathways, we performed two hypothesis-generating analyses. First, we evaluated innate immune response signatures that were associated with vaccine-induced CD8⁺ T-cell magnitude, and second, we compared its innate activation profile with that of the highly efficacious yellow fever vaccine YF-17D.

In the signature analyses, we found that serum induction of the two chemokines, MCP-2 and MCP-1, 24 h postvaccination, predicted whether or not a subject would develop a measureable CD8⁺ T-cell response 4 wk postvaccination (Fig. 4 A and B, and Fig. S4 A and B). Predictive accuracy was increased to nearly 90% by coupling these chemokines with proinflammatory cytokines (Fig. 4 A and B, and Fig. S4 A and B). Although additional studies are required to confirm this result, a role for these chemokines in CD8⁺ T-cell responses is supported by the strong T-cell chemoattractant function they exhibit (47) and the reported CD8⁺ T-cell adjuvant activity of MCP-1 (48). Furthermore, there was an indication in our data that subjects with the highest CD8⁺ T-cell responses were those who either strongly down-regulated RANTES or up-regulated IL-28A (Fig. 4C,and Fig. S4 C and D). One hypothesis is that strong down-regulation of serum RANTES postvaccination may indicate increased uptake by migrating inflammatory cells, but increased serum IL-28A may indicate an immunogenic role for the antiviral activities of this cytokine (49, 50). The transcriptional CD8⁺ signature analysis also revealed many genes exhibiting responses to MRKAd5/HIV at the 72-h timepoint that were significantly associated with CD8⁺ T-cell response magnitudes (Fig. 5 A and B), including several that are nearest neighbors of CD8⁺ T-cell response-associated module genes (Fig. 5C). One compelling component of the gene signature was EAT-2 (Fig. 5B), which was recently reported to enhance the frequency of vaccine induced T cells when encoded in an Ad5 vector (30). Despite the striking differences in innate response kinetics between the vaccines, we also tested whether the CD8⁺ signature for MRKAd5/HIV was associated with CD8⁺ T-cell response magnitudes for YF-17D. Reanalysis of the published YF-17D dataset (19) identified two genes, CRIP3 and NPB, with induction patterns that were associated with the CD8⁺ T-cell responses of both MRKAd5/HIV and YF-17D (Fig. 6E). Roles for both of these genes in vaccine mechanisms are plausible, given the function of CRIP3 in thymic cellularity (51) and the high expression levels of NPB in lymphoid tissues (52).

In the comparative analysis, we found a striking difference in the temporal innate immune activation profile of MRKAd5/HIV and YF-17D (Fig. 6A) that is consistent with, but not completely explained by, the dosage and pharmacokinetics of the vaccines: although replication-incompetent MRKAd5/HIV is present at the highest levels immediately after injection (53), live-attenuated YF-17D takes 5–7 d to reach maximal titers in the host (31). Unexpectedly, the innate immune response to MRKAd5/HIV was much more extensive than that induced by YF-17D (Fig. 6 A and B). Although the peak response to MRKAd5/HIV involved induction of over a dozen functional modules, the peak response to YF-17D involved induction of only two. In vitro stimulation experiments with the two vaccines identified which of the innate immune response differences observed in vivo are because of cell-intrinsic differences in innate immune signaling (Fig. 6 C and D). MRKAd5 preferentially induced a transcriptional regulatory network involving three transcription factors, IRF1, KLF4, and STAT5A, in vivo and in vitro (Fig. 6 C and D). Activation of the IRF1 network may play a part in the strong CD8⁺ T-cell immunogenicity of MRKAd5/HIV, given the role of this transcription factor in regulating MHC class I presentation and CD8⁺ T-cell responses (54). Interestingly, a number of the MRKAd5-specific genes are also associated with immunoregulatory functions. Among these, KLF4 promotes anti-inflammatory “M2” and inhibits proinflammatory “M1” macrophage polarization (35), and PDCD1LG2 and IDO1 suppress T-cell activation through a variety of mechanisms (55, 56). Further study will determine whether pharmacological inhibition of these molecules will lead to enhanced Ad5-induced T-cell functionality. Finally, direct comparison between MRKAd5- and YF-17D–induced innate immune responses in vitro revealed additional functionally relevant differences between the vaccines, suggesting that profiling of local responses may complement measurements of systemic responses obtained by from blood cell transcriptomes.

Our comprehensive analysis of the immediate systemic response following vaccination with MRKAd5 provides fresh understanding of vaccine-induced innate immune activation, how it is modulated by preexisting immunity, and how it relates to the subsequent adaptive immune responses. Such understanding will play an important role in the development of a highly efficacious HIV vaccine.

Materials and Methods

Subjects.

We enrolled 35 healthy HIV-1-uninfected adults [median age 37 y (range 20–50); 21 female; 28 Caucasian, 7 African American]. Eleven subjects were Ad5^Positive and 24 were Ad5^Neg (Fig. S2). Microarrays were run on five males and five females [median age 33 y (range 22–43); nine Caucasians]. Female participants were counseled to use birth control and avoid pregnancy during the study. All participants provided written informed consent, and each of the four United States trial sites obtained approval for the study through their institutional review boards.

Study Design.

HVTN 071 was a Phase 1b multicenter, open-label trial (ClinicalTrials.gov #NCT00486408). At the start of the trial (day 0), all volunteers were intramuscularly vaccinated with 1.5 × 10¹⁰ genomes of the previously-described MRKAd5/HIV vaccine (3); 24 received a second vaccination at day 28 before all MRKAd5/HIV vaccinations were suspended (4). Blood was collected immediately before vaccination and at 4–6, 24, 72, and 168 h postvaccination for 11 individuals. Serum was obtained from an additional 24 individuals at the prevaccination and 24-h time points.

Microarrays and Quantitative Real-Time PCR.

PBMC were isolated from blood as previously described (57). RNA was extracted from PBMC using the RNeasy Protect Cell protocol (Qiagen). Before labeling, the integrity of samples was checked using an Agilent 2100 Bioanalyzer.

Affymetrix Exon Arrays.

RNA expression for the in vivo study and one in vitro study was analyzed using the Human Exon ST 1.0 microarray platform (Affymetrix) essentially as described in ref. 17. For in vivo profiling, 50 samples were analyzed: five time points (prevaccination and 6, 24, 72 and 168 h postvaccination) for 10 subjects (three Ad5-seropositive and seven Ad5-seronegative). For in vitro profiling, eight samples were analyzed: PBMC obtained from four Ad5 seronegative donors stimulated for 24 h with MRKAd5 empty vector at 20,000 particles per cell and GTS buffer mock control.

Agilent 3′ Arrays.

RNA from the MRKAd5 vs. YF-17D comparative in vitro study was analyzed using the Agilent SurePrint G3 Human GE 8 × 60K microarray platform (Agilent Technologies), essentially as described in ref. 17. For comparative in vitro profiling, 16 samples were analyzed: PBMC obtained from four Ad5 seronegative donors stimulated for 24 h with MRKAd5 empty vector at 60,000 particles per cell and GTS buffer mock control or YF-17D at 30 particles per cell and DMEM 2% (vol/vol) FCS buffer mock control. Microarray data analysis procedures are described in the SI Materials and Methods. Quantitative real-time PCR was performed as described in ref. 17.

Multiplex Cytokine Analysis.

Serum cytokine analysis was performed using the Lincoplex High Sensitivity kit (Millipore Cat# HSCYTO-60SPMX13) and regular sensitivity kits (Millipore Cat# MPXHCYTO60KPMX42, MPXHCYP2:00 PMX23, and MPXHCYP3-PMX9), according to the manufacturer’s instructions (Linco/Millipore), and samples were analyzed on a Luminex 200 (Luminex). PBMC supernatants were assayed similarly for a subset of the analytes. Data were analyzed using a custom in-house export and quality control program in conjunction with the Ruminex program (58).

Enumeration and Phenotyping of Fresh Blood Cell Populations.

Trucount tubes (BD) were stained with CD45-ΑPC, CD14-PE, CD3-PerCp, and CD8-FITC (all from BD). For further phenotyping, whole blood was diluted 1:10 in Pharmlyse RBC lysis buffer (BD), incubated 10 min at room temperature and centrifuged at 750 × g for 5 min. RBC lysis was repeated and cells were resuspended in cold PBS; 2 × 10⁶ to 6 × 10⁶ cells were stained with Aqua Viability Dye (Invitrogen), followed by one of three antibody mixtures (details provided upon request). Cells were fixed with PBS containing 1% paraformaldehyde and stored at 4 °C until analysis by flow cytometry. All samples from one volunteer were analyzed together within 7 d of staining.

Statistical Analysis of Cell Concentration and Multiplex Cytokine Data.

Analysis of analytes altered after vaccination was performed by running a mixed model with normally distributed errors and an unstructured covariance matrix, with cytokine or cell concentration as the dependent variable and categorical sampling time as the independent variable, allowing random intercepts for each participant for each immunization series and each cytokine or cell type. Sex and age were included as possible confounders. P values for time were adjusted within a given vaccine series using the Hochberg method (59). Methods for identification of serum analyte profiles that were predictive of CD8⁺ T-cell responses are provided in the SI Materials and Methods.

In Vitro PBMC Stimulations.

One-million PBMC from healthy Ad5-seronegative individuals were stimulated with MRKAd5 empty vector or GTS buffer mock control (60) or YF-17D or DMEM with 2% (vol/vol) FCS mock control [courtesy of Charles Rice, Rockefeller University, New York (61–63)] in RPMI containing 10% (vol/vol) FCS, penicillin and streptomycin at a range of multiplicities of infection. After 24 h, cell-culture supernatants were harvested for multiplex cytokine analysis and cells were frozen in RLT buffer (Qiagen) containing β-mercaptoethanol, for RNA purification and microarray analysis.