Genomic Analysis Reveals Pre- and Postchallenge Differences in a Rhesus Macaque AIDS Vaccine Trial: Insights into Mechanisms of Vaccine Efficacy

Abstract

We have employed global transcriptional profiling of whole blood to identify biologically relevant changes in cellular gene expression in response to alternative AIDS vaccine strategies with subsequent viral challenge in a rhesus macaque vaccine model. Samples were taken at day 0 (prechallenge), day 14 (peak viremia), and week 12 (set point) from animals immunized with replicating adenovirus type 5 host range (Ad5hr) recombinant viruses expressing human immunodeficiency virus HIV_env89.6P, simian immunodeficiency virus SIV_gag239, or SIV_nef239 alone or in combination with two intramuscular boosts with HIV_89.6Pgp140ΔCFI protein (L. J. Patterson et al., Virology 374:322-337, 2008), and each treatment resulted in significant control of viremia following simian-human immunodeficiency virus SHIV_89.6P challenge (six animals per group plus six controls). At day 0, 8 weeks after the last treatment, the microarray profiles revealed significant prechallenge differences between treatment groups; data from the best-protected animals led to identification of a network of genes related to B cell development and lymphocyte survival. At peak viremia, expression profiles of the immunized groups were extremely similar, and comparisons to control animals reflected immunological differences other than effector T cell functions. Suggested protective mechanisms for vaccinated animals included upregulation of interleukin-27, a cytokine known to inhibit lentivirus replication, and increased expression of complement components, which may synergize with vaccine-induced antibodies. Divergent expression profiles at set point for the immunized groups implied distinct immunological responses despite phenotypic similarities in viral load and CD4⁺ T cell levels. Data for the gp140-boosted group provided evidence for antibody-dependent, cell-mediated viral control, whereas animals immunized with only the replicating Ad5hr recombinants exhibited a different evolution of the B cell compartment even at 3 months postchallenge. This study demonstrates the sensitivity and discrimination of gene expression profiling of whole blood as an analytical tool in AIDS vaccine trials, providing unique insights into in vivo mechanisms and potential correlates of protection.

A crucial aspect of vaccine design and evaluation is the ability to identify and measure correlates of protective immunity. This has proven particularly problematic in AIDS vaccine research (63) where, despite evidence for the role of virus-specific CD8⁺ T cells (14, 88) and the associated impact of type I major histocompatibility complex (MHC) haplotypes (38, 45) in mediating viral control, exceptions to such classifiers still confound the research community (4, 17, 79).

The need for such correlates is particularly highlighted by the outcome of the Merck STEP trial, in which vaccinees failed to exhibit superior viral control at set point despite the induction of high-frequency, virus-specific T cells as evidenced by enzyme-linked immunospot (ELISPOT) measurements against the immunogen sequences (22, 68). The recently completed ALVAC/AIDSVAX trial in Thailand offers a still more complex picture: the failure of the vaccine to achieve a significant impact on set point viremia may have been predicted on the basis of the poor cellular immune responses induced by the treatment. However, the reduced acquisition rate among vaccinees brought renewed attention to antibody-mediated immunity, suggesting benefits from such mechanisms even in the absence of appreciable levels of neutralizing antibodies (85).

Our laboratory has been applying global gene expression profiling and proteomics methods to various viral infection models, using these as systems-level views in exploring viral pathogenesis and host-pathogen interactions (57, 78). We have utilized these techniques in the context of nonhuman primate models with respiratory RNA viruses, such as influenza virus (16, 27, 60) and severe acute respiratory syndrome coronavirus (29), including our characterization of responses following influenza vaccination with either attenuated or inactivated viruses (15). For lentiviral systems, we have also employed high-throughput methods under the highly controlled circumstance of in vitro studies of human immunodeficiency virus type 1 (HIV-1) and simian immunodeficiency virus (SIV) (23, 24, 65, 100, 104). Recently, we reported a comprehensive investigation with global expression profiling in a nonhuman primate model, contrasting pathogenic versus nonpathogenic outcomes from SIV infection (37, 61).

In considering new tactics to characterize responses to candidate HIV vaccines, we believe global gene expression profiling offers an attractive approach for evaluating and defining the host response to vaccination and subsequent challenge. As such, gene expression profiling may reveal genomic markers of successful vaccination and protective immunity, which in turn could lead to potential natural targets for future vaccine development. Important proof-of-concept studies along these lines have been recently published in the characterization of a yellow fever vaccine (42, 83).

As an initial implementation of this approach, we performed microarray analyses on whole blood samples obtained from rhesus macaques at defined points during the course of an AIDS vaccine study (76). This study used replicating adenovirus type 5 host range (Ad5hr)-HIV/SIV recombinant virus priming in combination with a protein boost. Because adenovirus vectors preferentially infect cells that line the respiratory, gastrointestinal, and reproductive tracts, they have the particular merit of inducing immune responses at mucosal surfaces, the site where the majority of HIV infections are acquired (32, 71). In nonhuman primate models, replicating Ad-HIV/SIV recombinant viruses, in combination with a protein boost, have been demonstrated to protect chimpanzees from HIV challenge (66, 86) and rhesus macaques from challenge with SIV_mac251 (67, 77) or simian-human immunodeficiency virus SHIV_89.6P (30, 76). Correlates of protection have included Env-specific CD8⁺ T cell responses and Env-specific antibodies that mediate antibody-dependent cellular cytotoxicity (ADCC) and antibody-dependent cell-mediated viral inhibition (ADCVI) (39, 44, 48, 76).

In the study described by Patterson et al. (76), the goal was to characterize the protection afforded by immunization with replicating Ad5hr recombinant viruses expressing HIV_env89.6P, SIV_gag239, and SIV_nef239 alone or in combination with novel HIV envelope protein boosting, contrasting these results to earlier studies that employed only the corresponding SIV immunogens (77). This study revealed that animals boosted with HIV_89.6Pgp140ΔCFI protein, and to a lesser extent animals that did not receive a boost, showed significant reductions in acute and set point viremia when subsequently subjected to SHIV_89.6P challenge. The reduction in chronic viremia was long lasting and could be observed 32 weeks after the last priming immunization. While no prechallenge immune measurement provided a correlate of protection for viremia outcomes (76), a spectrum of Env-specific nonneutralizing antibody activities have now been correlated with control of viremia (107).

In the current study, we sought to expand upon these findings by performing global transcriptional profiling of whole blood samples taken prechallenge, at peak viremia, and at viral set point. Whole blood was chosen as a very tractable sample type with significant advantages when considering sampling in the context of large-scale clinical trials. Simple methods are now available that provide immediate stabilization of the RNA at the point of collection with robust storage thereafter, and sufficient RNA for array characterization can be obtained with as little as 2.5 ml of blood (36). Appreciating that whole blood is a complex mixture of cells, in terms of data quality this approach also avoids alterations in expression profiles that would attend ex vivo partitioning of cell populations, a process that also poses challenges in consistency, especially if implemented at multiple sites (11). Our objective was to determine whether we could identify gene expression “signatures” that would yield distinctions between vaccine groups and which would be predictive of protective immunity and to obtain new insights into the immune processes postchallenge. Our results show significant pre- and postchallenge gene expression differences between treatment groups and suggest the utility of gene expression profiling of whole blood as an analytical tool for application in AIDS vaccine research.

MATERIALS AND METHODS

Immunization and challenge.

Details of the immunization and challenge schedule were those described previously by Patterson et al. (76). Briefly, 24 juvenile male Indian rhesus macaques, all Mamu-A*01 negative, were divided equally into four groups. Groups I through III received identical priming immunizations of Ad5hr-HIV/SIV recombinant viruses at week 0 (intranasal and oral routes) and at week 12 (intratracheally). At weeks 24 and 36, group I received two intramuscular boosts of phosphate-buffered saline (PBS) alone, group II received two boosts with HIV_89.6Pgp140ΔCFI protein in PBS, and group III received two boosts with a HIV peptomer in monophosphoryl lipid A-stable emulsion (MPL-SE). The peptomer was composed of repeating 18-mers representing amino acids 419 to 436 of the viral envelope. Group IV, control animals, received empty Ad5hr vector followed by mock boosts of adjuvant (MPL-SE; three animals) or PBS (three animals), administered on the same schedule as the treatment groups. All macaques were challenged at week 44 with intravenous injection of a SHIV_89.6P challenge stock.

Sample collection and processing.

Samples of peripheral whole blood were collected from all animals as PAXgene (Qiagen) whole blood. The PAXgene collection system offers the advantage of immediately stabilizing the RNA in the sample at the time of the draw and avoids any possible perturbations that would accompany later manipulations. A common reference sample of RNA was derived from the whole blood of six naïve male rhesus macaques, independent of the immunized and control animals. These reference animals were screened by quantitative reverse transcription-PCR (qRT-PCR) on the whole blood samples to avoid inclusion of any animals that were subject to any heightened state of immune activation, using methods described previously (10). This was accomplished with an assay panel for expression from the following genes using primer/probe sequences based on the rhesus macaque genome: CCR5, CD2, CD4, CD8a, CD20, CXCR4, FCGRA, IFNG, MX1, and TNF (see Table S6 in the supplemental material for probe and primer sequences).

RNA isolation and synthesis of fluorescently labeled RNA probes was accomplished by standard linear amplification techniques (10). Per the manufacturer's protocols, linear amplification to generate dye-labeled cRNAs was performed without globin depletion; electropherograms (Agilent Bioanalyzer) of Cy5-labeled cRNA preparations from representative samples did not exhibit distinct and dominant peaks that could be ascribed to a globin product.

Microarray analysis.

Global gene expression profiling using rhesus macaque oligonucleotide microarrays (Agilent Technologies) (105) was performed using a two-color format as previously described by Baas et al. (10), measuring expression ratios for 17,221 genes. Experimental samples were measured as ratios against the common reference RNA (see above), with each measurement incorporating reverse dye labeling techniques, resulting in two measurements for each gene. Details of data processing have been described elsewhere (26). All data collected in this format were entered into a custom-designed relational database and subsequently uploaded into the Rosetta Resolver System 7.1 (Rosetta Biosoftware). The Resolver system combines replicates while applying error weighting to account for additive and multiplicative noise (93). A P value is generated, assessing the probability that a gene is differentially expressed relative to the comparator employed in the ratio [i.e., the probability of a null hypothesis, corresponding to the log(ratio) equal to 0].

A similar approach was used in combining ratio measurements that shared the same common reference by using the in silico reratio tool available in the Resolver suite, thereby yielding ratios of the test samples relative to each other, as well as P values for the resulting values, albeit with the penalty of increased variance that attends such indirect designs (25, 108). In this manner, microarray measurements against the common reference allowed the day 0 gene expression profile for each animal to serve as the genetically matched control for later time points; that is, the resulting gene expression values for an animal at day 14 or week 12 are given as the change relative to the animal's state on day 0. However, prevaccination samples were not available for the animals; consequently, results for day zero (posttreatment, prechallenge) could not be expressed relative to each animal's genetically matched, pretreatment state, and instead had to be expressed relative to a separate reference. To assist in visualizing the differences among the groups on day zero, the measured expression log(ratio) results of the individual animals were transformed to the log(ratio) relative to the average expression of control group IV on day zero. This was effected by first combining all the measured group IV day zero arrays to obtain the weighted average group IV day zero expression profile, with average values for each gene as the log(ratio) relative to the common reference. For each gene on day zero, subtracting this group IV average log(ratio) value from the measured log(ratio) value of each animal simply shifts the distribution of the individual animal values to center around the group IV average but does not alter the spread or variance of the values. As with indirect comparison strategies, the calculation yields for each gene a new log(ratio) relative to the group IV average of the gene. Primary microarray data are available at http://viromics.washington.edu, in accordance with proposed minimum information about a microarray experiment (MIAME) standards.

Statistical methods and functional analysis.

The day zero one-way analysis of variance (ANOVA) was performed with data filtered for expression ratios having the aforementioned P values of <0.01 in a minimum of four experiments (4,418 genes) and setting the Benjamini-Hochberg false-discovery rate (FDR) to a P level of <0.01. After testing a variety of data filters (including no filter), the chosen strategy provided a balanced approach to focus on genes showing larger changes in expression, returning a tractable number of statistically significant genes (1,059 genes) with few comparisons relying on ratios with low P values (i.e., marginal differential) for all the animals. Similar considerations were applied to the one-way ANOVA comparisons on day 14 and week 12, for which adding the 1.5-fold filtering threshold further focused the analyses on genes showing larger changes in the animals relative to their day zero state. The filtered inputs were 4,243 genes for the day 14 ANOVA and 3,336 genes for the week 12 ANOVA (results of these analyses are shown in the figure legends). No weighting was used in the ANOVA calculations. All post hoc tests used the Scheffé method with a P level of <0.1 for between-group comparisons.

For analysis of outlier animals for the day 14 expression comparisons, examination of the expression patterns for each treatment group by unsupervised clustering led to the observation of animal H696 as an outlier in control group IV and animal CJ95 as an outlier in the Env-boosted group II. The outlier status for an animal was further verified by comparing the anomalous animal to the rest of the group by a two-tailed t test (threshold P, ≤0.05), using genes showing significant expression differences at day 14 versus day 0. Neither of these animals appeared as clear outliers to their respective experimental groups on day 0 or week 12. Array data for control group IV at each time point were assessed by unsupervised clustering and by principal component analysis. With the exception of the outlier noted above, these methods failed to show significant differences among the animals based on the vehicle used during the mock boosts at weeks 24 and 36 in the treatment regimen, and they were treated as equivalent in all analyses.

The unsupervised two-dimensional hierarchal clustering of primary microarray data from day zero (see Fig. 2) was performed by the agglomerative technique for both the experiments and the expression values, using the Euclidean distance similarity measure with average linkage for joining subclusters; error weighting was applied only to the gene expression values. For the visualizations of the ANOVA results in Fig. 4 and 5, below, the experiments were fixed in order by the experimental groups used in the ANOVA and clustered only in the dimension of the gene expression data; the latter employed a self-organizing map algorithm with a three-by-three lattice, cosine correlation similarity metric, and application of error weighting to the gene expression values. For the results shown below in Fig. 4 and 5, clustering of the samples was not performed because the genes had already been selected via ANOVA to distinguish the treatment groups, and clustering under such biased input would not be an independent assessment of the relationships among the samples.

Functional analysis using the Ingenuity Pathways Analysis software (Ingenuity Systems) took advantage of the array annotation linking the macaque probe sequences to human Entrez GeneIDs; derivation of this annotation was described previously by Wallace and colleagues (105). Ingenuity Pathways Analysis employs a vendor-built database of finite functional classes, pathways, and biological interactions extracted from the literature. Given an input list of genes, it uses a Fisher's exact test to calculate probabilities of functional classes or pathways. Functional analysis using the open access tool DAVID was also performed using the human Entrez GeneIDs. This analysis was restricted to public gene ontologies and referenced to the human genome as the background (31, 50). Probability calculations in DAVID were performed using a metric very similar to Fisher's exact test.

RESULTS

Our gene expression profiling analyses were performed using whole blood samples obtained from rhesus macaques during the course of a vaccine study aimed at evaluating the ability of replicating (Ad5hr)-HIV/SIV recombinant virus priming alone or in combination with a HIV envelope protein boost to control viremia following SHIV_89.6P challenge. Details of the experimental design, and also a complete description of Env-specific antibody and CD8⁺ T cell responses, can be found in the report of Patterson et al. (76). Briefly, three groups of six macaques each were mucosally primed with replicating Ad5hr recombinant viruses expressing HIV_env89.6P, SIV_gag239, or SIV_nef239 (Fig. 1 A). Group I received no boost (adjuvant alone), group II received two intramuscular boosts with HIV_89.6Pgp140ΔCFI protein, and group III received two intramuscular boosts with a synthetic HIV polypeptide “peptomer” designed to mimic the critical CD4 binding site on HIV_env89.6P. A control group of six animals (group IV) received the Ad5hr vector alone. Three of the control animals received MPL-SE adjuvant alone in place of the protein boost, and three received PBS (76).

FIG. 1.

Immunization schedule and resultant viremia observed postchallenge for the experimental groups analyzed in this study. (A) Three groups of Mamu-A*01-negative macaques were primed at weeks 0 (orally and intranasally) and 12 (intratracheally) with three Ad5hr vectors each encoding the indicated immunogen. Group II received the indicated intramuscular boosts with purified Env protein homologous to the immunogen and challenge strain, while other groups received only vehicle. The intravenous challenge was 90 50% monkey infectious doses of SHIV_89.6P. (B) Plasma viremia following challenge. Geometric means are plotted with standard errors. Differences for group II versus control group IV reached statistical significance for both the acute phase (day 9 to week 3) and set point (weeks 8 to 24), whereas this difference was only significant for group I at the set point. Arrows shown above the viremia plot indicate the time points for the blood samples chosen for microarray analysis. The indicated SHIV_89.6P challenge at week 44 in panel A corresponds to day 0 in the viremia plot in panel B.

As described above, the treatment regimen of replicating adenoviral priming followed by boosting with soluble gp140 provided significant reductions in viremia relative to the nonimmunized control during both the acute phase of the infection as well as at set point (Fig. 1B). The group receiving only the recombinant Ad5hr recombinants (group I) trended to a lower peak viremia at day 14 versus group IV controls but only achieved statistically significant differences during the chronic phase of infection. Viral load data for the peptomer-boosted group in general paralleled those observed for group I, albeit just failing to achieve a statistically significant reduction in viremia at the early set point. We therefore targeted our microarray analyses to PAXgene blood collected from groups I, II, and IV, excluding group III, as it appeared unlikely to provide information distinct from that to be garnered from analysis of group I. We also believed a comparison of the vector immunogen with or without Env protein boosts to be the most relevant choice, given the significance of vector prime/protein boost strategies in clinical and preclinical research on HIV vaccine regimens (13, 21, 85). Blood samples were from day 0 immediately prior to challenge, day 14 at the recorded peak of viremia, and week 12 as a point representative of the set point (Fig. 1B).

The primary array measurements were performed by the two-color technique, with dye swap replication for each measurement. Due to limited quantities of RNA from the experimental samples, an indirect comparison strategy was employed in which all samples were measured against a common reference (25, 108). The common reference consisted of pooled RNA from PAXgene blood drawn from six rhesus macaques, chosen independently from the vaccine study. RNA from these reference samples was evaluated by qRT-PCR with a limited panel of assays for immune-related genes, to guard against inclusion in the reference pool of an animal having an active but undiagnosed infection. Measurements against the common reference were used for the analyses of day zero prechallenge samples. For samples from day 14 and week 12, the primary array data were also taken in the two-color format, relative to the common reference; however, these data were then transformed in silico by using each animal's corresponding prechallenge array measurement as a baseline. For each animal this resulted in expression changes at day 14 or week 12 relative to the animal's state on day 0.

Microarray profiles of whole blood showed prechallenge distinctions among treatment groups.

A visualization of the day zero prechallenge microarray data, presented as an unsupervised two-dimensional hierarchical cluster, is shown in Fig. 2. This analysis method was chosen to determine if patterns in the data were sufficient to result in spontaneous partitioning of the experimental groups. The filtering parameters for genes in the clustering diagram allowed for inclusion of genes that showed significant regulation in at least four out of six animals in a treatment group.

FIG. 2.

Unsupervised two-dimensional hierarchical clustering of microarray data from day zero PAXgene whole blood. The dendrogram on left shows the partitioning of the animals, labeled by experimental group and animal ID, color coded for ease of visualization. Gene expression values are ratiometric measures relative to a reference pool of blood from six rhesus macaques (see Materials and Methods) and are visualized on a log₁₀ scale using a color scheme with saturation at a 4-fold change. The 2,640 genes in the clustering diagram were selected as having a ≥2-fold change (P < 0.01) in at least four animals.

The dendrogram on the left side of Fig. 2 reveals the segregation of the expression profiles for group IV (unimmunized controls) as a distinct subtree, rooted high in the hierarchy and containing all six animals from this group. The coherence of group II (Ad5hr immunogens plus protein boosts) was also noteworthy, again with all animals in this group occurring in one subtree. The animals that received only the Ad5hr immunogens exhibited the least consistent behavior, with members displaying various extents of similarity to the other treatment groups. The behaviors of group IV and II animals to segregate separately and of group I animals associating with the other treatment groups were general observations of hierarchical clusters generated by a variety of criteria.

As manifest in the heat map of Fig. 2, all of the array measurements on the prechallenge PAXgene blood showed a broad extent of differential gene expression relative to the common reference derived from the nonstudy animals. Because all the treatment groups similarly showed many changes relative to the common reference, it would appear the study animals were all subject to a shared influence or condition that was distinct from the reference animals. Nevertheless, there are regions in the heat map where the treatment groups differed, thus driving the spontaneous organization of the animals in group IV and group I.

It is noteworthy that the prechallenge expression arrays for control group IV did not show any significant differences between the three animals that received mock boosting with MPL-SE (CK6V, CK7R, and M407) and those receiving mock boosts with PBS (H696, H700, and H701), whether assessed by statistical tests or by examining simple trends by clustering visualizations. Therefore, further analyses of prechallenge arrays considered all animals in group IV as equivalent, allowing for balanced group sizes for the ensuing ANOVAs.

gp140-boosted group II showed distinct upregulation of genes associated with humoral and cell-mediated immunity prior to challenge.

For statistical tests between the treatment groups on day zero (prechallenge), the measured log(ratio) expression data were transformed by subtracting the average value for the measurements of the six animals in control group IV. This resulted in recentering the log(ratio) data, giving expression changes relative to the group IV average, but did not alter the variance for any gene within or between treatment groups. One-way ANOVA based on the transformed log(ratio) values was used to find gene expression differences that distinguished the treatment groups prior to challenge, returning a total of 1,059 genes. A Scheffé post hoc comparison was then employed to parse these into subsets for pairwise differences between the treatment groups (see Materials and Methods).

A subset of 789 genes was thus identified as exhibiting significant differences between the protein-boosted group and the control. Functional characteristics for these genes were determined using the Ingenuity Pathways Analysis software (Table 1) (see Materials and Methods), providing the observation that the genes upregulated in group II relative to the expression levels in group IV were highly enriched for genes related to the immune response. As shown in Table 1, the infectious disorder of cells (under a broader category of response to infection) was the top-scoring function with the lowest P value and containing 33 genes. The broad categories of apoptosis and growth of cells contained within them the high-scoring, more-specific immunological functions for apoptosis of lymphocytes or for the proliferation of leukocytes. The representation of both these particular categories no doubt stems in part from the many genes these categories share but can also be rationalized in that apoptosis is a natural consequence following the proliferation of immune cells upon antigenic stimulation.

TABLE 1.

Summary of functional analysis for genes with increased expression at day zero prechallenge in gp140-boosted group II versus control group IV^a

Function	P value	No. of genes	Gene IDs
Response to infection		41
Infectious disorder of cellsb	7.07E-06	33	AKAP13, ANXA2, CTDP1, CTSZ, DLST, E2F4, ERCC3, ERCC5, GANAB, HNRNPF, HNRNPU, KIAA0922, KLF2, LAPTM5, NFKB1, NRBP1, NUP85, PDIA3, PI4KA, PRPF8, PURA, RAD23A, RANBP1, RBM10, RELA, RGPD5, SAP30BP, SDF4, SF3B2, SNRPD3, SPTAN1, TCF20, UQCRFS1
Transcription		56
Processing of mRNAb	1.37E-04	9	CTDP1, DHX38, HBB, HNRNPR, SFPQ, SFRS5, SFRS8, SFRS15, U2AF2
Apoptosis		45
Apoptosis of lymphocytesb	3.00E-03	11	ADORA2A, CD5, GNAS, GPR132, KIFAP3, KLF2, NFATC1, NFKB1, RAC2, RELA, ST3GAL1
Growth of cells		47
Proliferation of leukocytesb	2.71E-02	16	AKAP13, CD5, DIAPH1, HLA-DRB1, HNRNPA2B1, JUND, KLF2, NFATC1, NFATC3, NFKB1, PDCD1, POU2F2, PURA, RAC2, RELA, SH3BP2
Cell movement		28
Rho protein signalingb	8.809E-03	4	ARHGDIA, ARHGDIB, CFL1, KIFAP3
Inflammation		16
Hypereosinophiliab	5.94E-04	5	CD5, NFATC1, NFATC3, NFKB1, TNFSF8
Antibody response	2.29E-02	4	BST2, GPI, PDCD1, POU2F2
Modification of carbohydrate	4.79E-02	4	B4GALT1, GANAB, MGAT4B, PI4KA
Excision repair	1.53E-02	4	ERCC3, ERCC5, RAD23A, XRCC1
Targeting of protein	7.54E-03	5	KIF13B, NXT1, PACS1, XPO7, YWHAE
Synthesis of protein	2.78E-02	11	ABCF1, ALAS2, CKAP5, EIF2AK1, KLF2, NACA, NFKB1, PABPC4, PET112L, PPP1CA, RELA

^aGenes were selected by ANOVA of expression data for the three treatment groups followed by post hoc determination of those that distinguished group II from control. Functional analysis was performed with the Ingenuity Pathways Analysis program. The summary was restricted to categories with P values of <0.05 and represented by greater than three genes. General categories are summarized by the total number of associated genes.

^bWithin the indicated general category, the most specific functional entry with the lowest P value is indicated.

This set of upregulated genes also contained very particular markers relating to the antibody response in Env-boosted group II, including genes associated with early B cell development (BST2, GPI, and PDCD1) and those more indicative of mature B cells (POU2F2 and class II MHC proteins). The antibody response may also be reflected in the upregulation of the genes B4GALT1, GANAB, and MGAT5 (under the heading of modification of carbohydrate). These genes all belong to the KEGG pathway for N-linked glycan biosynthesis. Inasmuch as the protein boosting of group II resulted in Env-binding antibody titers 2 to 3 logs greater than the other groups, upregulation of this pathway may be a consequence of the increased production/secretion of antibodies glycosylated on the conserved N-linked sites.

At day zero prechallenge, group II animals (Env boosted) also had significantly greater expression of immune-associated genes than group I animals, who only received the Ad5hr recombinants expressing the Env, Gag, and Nef immunogens. Specifically, from the 789 genes described above, a subset of 276 genes was found that differentiated group II from both of the other groups (see Table S1 in the supplemental material). The functional attributes of this gene subset are very much in the same character as those described in Table 1, again reflecting such phenomena as proliferation and apoptosis of lymphocytes, and also including protein synthesis and N-linked protein glycosylation. Thus, at the prechallenge time point the expression data for the Env-boosted group exhibited features distinct from the other treatment groups and represented a breadth of immunological characteristics, including B cell and T cell functions.

Networked expression differences on day zero in group II highlight enhanced T cell and B cell functions and prosurvival signaling patterns.

A network was created with the immune function genes of Table 1, and membership was expanded in a limited manner by adding functionally associated genes which exhibited expression patterns similar to those in the ANOVA set, but at a more relaxed statistical threshold (Fig. 3). Lymphocyte function is represented by genes within downstream signaling pathways (e.g., LAT and BLNK, which are linkers in the T cell and B cell receptor pathways, respectively) rather than by differential levels of the receptors themselves. These culminated with upregulation of NFATC1 and NFATC3, whose transcription would also be favored by the enhancing transcription factor MEF2D. The upregulation of the MHC class II antigen presentation pathway may point to continuing development of the B cell compartment, an inference with support from other genes shown in Table 1 and Fig. 3.

FIG. 3.

Network diagram derived from day zero microarray measurements for T cell and B cell function and cell survival. Node colors show the average gene expression levels of vaccinated animals that received the HIV_89.6pgp140 protein boost (group II) relative to the average for unvaccinated control animals (group IV) (red, relative increase in expression; green, relative decrease in expression). The network was produced with the Ingenuity Pathway Analysis program, using a gene list derived from an ANOVA for the day zero microarray data, comparing all three treatment groups, followed by post hoc analysis to filter genes showing significant differences between groups II versus IV.

Genes associated with B cell development include transcription factors such as JUND (8) and BCOR (52) and other elements linked by the central node BCL6, which plays an important role in B cell development and establishment of immunological memory (99). While BCL6 failed to make the statistical cutoffs for genes distinguishing these groups on day zero, the placement of this central node was supported by direct interactions with other differentially regulated genes. The strong upregulation of POU2F2 (also known as OCT2) also distinguished group II from the other groups. This gene, which is ultimately downstream of the B cell receptor (34), also drives many aspects of B cell maturation, including driving expression of XRCC6. XRCC6 encodes an enzyme involved in V(D)J recombination (92), the process of genetic rearrangement in generating B cell receptor and T cell receptor diversity.

The network depicted in Fig. 3 also brings into consideration proliferative and prosurvival elements centering on NF-κB and AKT. Increased AKT1/2 activity would be enhanced further by increased levels of HSP90 and CD37, as well as by the downregulation of PPP2R4, an activator of phosphotyrosine phosphatases (9). The pivotal roles of AKT in cell growth and survival are well established (40). In addition, this segment of the network diagram includes integration of both upward and downward changes in expression levels, particularly around the apoptotic activator BAX. This critical gene did not show statistically significant differences among the treatment groups in these prechallenge data; however, the downregulation of other apoptosis facilitators in the BCL2 family (BCL2L11 and BAK1) would indicate antiapoptotic influences in group II. The products of YWHAE (74) and PYCARD (75) alter BAX localization, and the expression differences for these genes, indicated in Fig. 3, would result in diminished localization of BAX at the mitochondrial membrane, thereby also favoring cell survival. In Env-boosted group II, MYC also showed upregulation relative to levels in the other treatment groups, an observation that seems strongly supported by the very substantial increased expression of FUBP1, which is the activator of the MYC far upstream element (FUSE) (33). While increased levels of MYC are associated with proliferation and growth, expressed MYC protein can also activate the apoptotic action of BAX (49). However, the diversion from driving growth to inducing apoptosis appears dependent on thresholds of MYC expression and the integration of other proapoptotic signals (72).

In summary, differences in the expression profiles based on whole blood from day zero revealed gene expression features distinct to the Env-boosted animals, and these genes are associated with humoral and cellular immunity. These genes form a network encompassing T cell and B cell functions and B cell fate and include a subnetwork of prosurvival/antiapoptotic expression changes.

On day 14, expression comparisons were dominated by shared differences of vaccine groups versus the control.

As noted earlier, analysis of expression profiles for day 14 utilized transformed data, which showed changes in transcript levels relative to an animal's prechallenge state. As described in Materials and Methods, the expression data for each treatment group led to the exclusion of animal H696 as an outlier in control group IV and animal CJ95 as an outlier in Env-boosted group II.

As with the day 0 expression results, hierarchical clustering was used to assess the general trends of the day 14 profiles. The only self-organization evident in the data was the segregation of control group IV from the other animals; however, group I and group II were highly intermingled (data not shown). The three groups had a relatively limited number of genes gauged as commonly up- or downregulated (see Table S2 in the supplemental material). Among the 45 shared upregulated genes, there was clear evidence that all animals manifested a type I interferon response (IFIT2 and -3, IFI27 and -44, MX1, and OASL) as well as showed activation of pathogen-associated molecular pattern receptors, with the induction of IRF7 and ISG15. Activation of cytotoxic T lymphocytes or NK cells is implicated by the upregulation of granzymes A, B, and K, and activation of neutrophils is suggested by increased transcript levels for defensins and lysozyme. MHC class I-directed cell killing would also be stimulated by the consistent downregulation of the genes LILRB2, -3, and -5, leukocyte immunoglobulin-like receptors that attenuate T cell receptor signals from class I antigen-presenting genes. The other most consistent aspect for shared downregulated genes was diminished expression of genes associated with production of inflammatory lipids (CYP4F2, LENG4, forms of cytosolic PLA2, and PTGES).

Analysis of variance was applied to determine statistically significant expression differences between the three experimental groups, using the same filtering and thresholds applied in the analysis of the day zero data. Expression values for the resulting ANOVA gene set are depicted in Fig. 4 and are organized into two large blocks, in which either group IV showed increased expression relative to prechallenge conditions (494 genes) or group II showed such increased expression (1,209 genes). Group I (Ad5hr recombinants only) showed extensive similarity to group II except for a subset of genes trending toward upregulation, akin to those seen in the unimmunized animals. For functional analysis, the indicated blocks of genes were analyzed by functional annotation clustering of gene ontology terms using the DAVID Bioinformatics Resource (31, 50). This tool was chosen because it offered the option to restrict the functional analysis to the specific levels of the public gene ontologies; this proved more tractable when dealing with the very large gene sets under consideration on day 14. Trimming for redundancy and eliminating poorly descriptive, general categories (e.g., “cytoplasm”) returned the consolidated annotation shown in Fig. 4 (see also Tables S3 and S4 in the supplemental material).

FIG. 4.

ANOVA results for day 14 expression data, visualized as a self-organizing map. Changes in gene transcription levels are given as fold changes relative to the expression level on day zero, with each animal serving as its own matched comparator. The heat map color scheme is as described for Fig. 2. For inclusion in the statistical analysis, a gene had to manifest a ≥1.5-fold change (P ≤ 0.01) in a minimum of four experiments; the final FDR (P ≤ 0.01) yielded the display set of 1,703 genes. Treatment groups and animal IDs are indicated; outlier animals H696 in group IV and CJ95 in group II were not included in the analysis. Gene ontology annotations for bioprocess levels 3, 4, and 5 and cellular component levels 4 and 5 were performed for the indicated portions of the diagram using the DAVID bioinformatics resource to cluster similar functional categories.

On day 14, upregulated genes in control group IV implicated activation and expansion of B cells.

For the 494 genes that were upregulated in the control animals on day 14, there was a clear prevalence of fundamental processes in protein processing, including protein synthesis, protein folding, and finally protein trafficking. The protein biosynthesis set is a broad category and includes 26 ribosomal proteins, 9 proteins associated with translation initiation, 2 translation elongation factors, and 4 tRNA synthetases; it also includes proteins for posttranslational modifications, such as N-linked glycans. Increased levels of protein synthesis are reinforced by the increased expression of genes involved in mRNA processing, including mRNA splicing and nuclear export. Functionally downstream of ribosome-directed protein synthesis are the categories of protein folding and protein trafficking; associated genes encode chaperonins, proteins for signal sequence recognition, and components in the Golgi apparatus/vesicular transport. The proteasome constituents and antigen-processing category contains 11 proteins in the ubiquitination pathway along with other peptidases and factors that influence protein catabolism; processing for antigen presentation is implicated by the upregulation of interferon-induced immunoproteasome constituents PSME1, PSME2, and PSMB10 (89). These changes are all consistent with the functional response of B cells after antigen recognition that lead to increased numbers of ribosomes and elevated levels of protein synthesis and secretion (2). The rhesus microarray employed in the analysis contains very few probes for detecting immunoglobulin transcripts, leading to this indirect inference of a functional B cell response.

Genes related to apoptosis and mitochondrial dysfunction were also highly populous categories in this block, with limited overlap of only nine genes. Negative and positive regulators of programmed cell death had near-equal representation (12 and 8 genes, respectively). The upregulation of levels of CD74, BCL2L2, and FXR1 were specific for apoptosis/development of B cells and antigen-presenting cells, consistent with the above proposal of B cell activation (41, 110). Otherwise, there was not a strong association with inflammatory response pathways or with other particular lymphocyte signaling pathways within this category.

Genes under the classification of cell-mediated immunity included a small assortment associated with leukocyte trafficking and activation. For example, monocyte/macrophage recruitment was implicated by the very significant upregulation of CCR2 in groups IV and I (6-fold and 2-fold, respectively), whereas this gene was unchanged in the most protected group, group II. Granulysin (GNLY) was uniquely upregulated in control group IV, suggesting increased cell-mediated cytotoxicity by either cytotoxic T lymphocytes (CTLs) or NK cells. This set of genes also included members associated with B cell development (see Table S4 in the supplemental material). Particularly noteworthy among these are TNFSF13 (APRIL), which is the ligand to the B cell maturation antigen (19), and IGBP1, a phosphatase in the B cell receptor pathway (53). Likewise, in control group IV, BST-2 expression was increased over 6-fold from the prechallenge state; in addition to B cell activation/maturation, BST-2 is also an important host cell restriction factor for blocking the release of virions from infected cells (73). In contrast, BST-2 expression in group I showed only a modest increase (1.3-fold), and in group II the level was unchanged.

Class I and class II MHC genes showed increased expression in control group IV but the opposite behavior in gp140-boosted group II.

In the ontology analysis of the block of left-hand MHC genes, the set for antigen presentation was the highest-ranking DAVID annotation cluster with a quite-specific immunological character. This category was composed almost exclusively of genes in the major histocompatibility locus and included genes for both class I and class II presentation. A class I-related molecule that was also upregulated was MICB; when expressed on target cells, this molecule increases the cytolytic response of CTLs and NK cells (70). The indicated MHC class II genes in this set included the genes DMB, DOB, and CD74 for antigen loading. For the class II surface display of antigenic peptides, only the HLA-DR subset of genes exceeded the thresholds in this statistical analysis. However, in the full data set it was evident that the rhesus HLA-DP and HLA-DA genes followed a similar trend as the DP genes, being upregulated in control group IV, distinctly downregulated in protein-boosted group II, and with an intermediate character for the unboosted group I animals.

At day 14, immunized animals upregulated genes for immune homeostasis and for activated antigen-presenting cells and T cells.

For the block of 1,209 genes in Fig. 4, the functional annotation clustering tended to return the most general shared attributes of the genes; after trimming poorly descriptive categories, the remaining specific sets shown in the figure consisted of relatively few genes compared to the scale of the input. The set described here as “lymphocyte and leukocyte activation, differentiation, and proliferation” consists of genes with more distal associations to lymphocyte and leukocyte development and seems to reflect broad homeostasis of immune cells; these include such general growth factors with immunomodulatory character as prolactin (PRL), insulin (INS), and ghrelin (GHRL). Signaling by these growth factors, particularly insulin, may be associated with the limited number of upregulated genes in the category of protein biosynthesis, inasmuch as they influence protein synthesis by impinging on the mTOR pathway.

Beyond suggesting immune homeostasis, the expression values do indicate explicit processes for the increased activation and maturation of antigen-presenting cells, e.g., by either increased levels of colony-stimulating factor 2 (CSF2; also granulocyte-macrophage CSF) or via Toll-like receptor signaling through IRF3. This is supported by the increased expression of interleukin-27 (IL-27), an early product from antigen-presenting cells (80), and from the upregulation of CD86, which is a requisite costimulatory molecule for class I presentation to T helper cells and for costimulatory signals to CTLs. Increased levels of BCL10 and TNFRSF8 suggest increased lymphocyte activation in the immunized animals (58, 101, 109). Consistent with this, within the day 14 ANOVA set, the immunized animals were distinguished by increased expression of the component genes of the T cell and B cell receptor pathways (calmodulin, NFAT, phosphatidylinositol 3-kinase, RAS, and VAV). The overall changes under the category of “cell migration and motility” also implicate activated leukocytes inducing motility in response to various growth factors (PDGFB, VEGFA, SHH, EGFR, and NRD1) and adhesion stimuli (CNT4, ITGA3, PLXNB1, and PODXL). The control group provided a contrasting picture of motility, showing downregulation of the aforementioned genes as well as very high upregulation of transcripts for cofilin and thymosin beta 4, protein products that destabilize actin fibers, thereby implicating very little actin polymerization and thus little motility (56, 64).

The category of apoptosis also includes the aforementioned lymphocyte/leukocyte activation genes CSF2, BCL10, and TNFRSF8. The set includes both proapoptotic and antiapoptotic genes (e.g., ERN2, FOXL2, NLRP2, and AMIGO2; BAG4 and FAIM2; and TRAF4, respectively); unlike the prechallenge state, the expression values did not show a bias toward promotion of cell survival. However, of particular note in this set is the inclusion of CD38, for which increased expression in the immunized animals suggests B cell maturation to long-lived plasma cells (55).

Immunized animals showed increased expression of acute-phase response genes at peak viremia.

The acute-phase response complement pathway genes C3, C8b, and MASP were significantly increased at day 14 in immunized groups I and II relative to their prechallenge state and relative to the control group. Similarly upregulated were other acute-phase and inflammatory proteins that are typically induced by inflammatory cytokines and are generally associated with activated macrophages (e.g., AOX1, SFTPD, and SERPINA). Interleukins-17 and -27 are members of this annotation cluster. Proinflammatory IL-17B may originate from activated antigen-specific CD4 and CD8 cells and was uniquely observed in the immunized animals. IL-27, produced by macrophages and antigen-presenting cells, while not specifically proinflammatory has been reported to synergize with IL-12 or IL-2 to induce gamma interferon secretion by cells (80). Also induced were chemotaxis genes ATRN and CCL21, which are associated with early immune responses; notably, CCL21 also plays a role in homeostatic proliferation of CD4 T cells (82).

To summarize the findings for day 14, all groups manifested a strong interferon and inflammatory response; nonetheless, as determined by ANOVA we saw quite divergent expression profiles in blood for the vaccinated groups versus the control. Control group IV exhibited upregulation of many more genes involved in apoptosis and cell death than Env-boosted group II. Genes for antigen processing and presentation (MHC I and II) were distinctly upregulated in the controls, but otherwise the gene ontology analysis provided few indications of specific pathways for leukocyte activation. Control group IV also showed upregulation of a large proportion of genes for protein biosynthesis. Over 1,200 upregulated genes distinguished group II from the controls, and the ontology analysis identified many genes associated with immune homeostasis. There was upregulation of genes in the T cell and B cell receptor pathways and markers of later B cell development. Group II also showed increased expression levels of acute-phase proteins, such as complement components, IL-17B, and IL-23. In this block of ∼1,200 genes, group I, which received only the Ad5hr recombinant immunogens, exhibited expression patterns almost identical to those of group II. Group I diverged from group II with upregulation of some genes involved in antigen presentation and processing and in the category of apoptosis and cell death (i.e., bearing some resemblance to the controls); however, these differences did not reach statistical significance.

At week 12, array profiles for the immunized groups were significantly diverged.

Expression data at week 12 following infection were analyzed in a manner analogous to that applied for the day 14 data, exploiting the changes in transcript levels relative to the prechallenge state of an individual animal. While there was variability within the groups, unlike the day 14 results no single animals exhibited such anomalous results as to be considered outliers for exclusion. Employing hierarchical clustering as a general visualization tool with the data from all 18 animals suggested a segregation of the group I animals from groups II and IV. This inference is born out in the ANOVA results presented in Fig. 5: for the 551 genes composing the heat map, we saw a much more complex relationship between the study groups than was observed at day 14. Major blocks of genes evolved with diametrically opposite expression values for groups I and II, and they manifested differing extents of similarity to control group IV.

FIG. 5.

Self-organizing map of ANOVA results from week 12 array results. Changes in gene transcription levels are given as fold changes relative to the expression level on day zero. Genes for statistical comparison were filtered for those with a ≥1.5-fold change (P ≤ 0.01) in a minimum of four experiments, yielding the set of 551 genes deemed statistically significant at an FDR rate of ≤0.01. Descriptions are based on gene ontology terms (level, ≥3) with enrichment in the indicated map region.

To provide an overview of the functional attributes, the week 12 ANOVA gene set was parsed into the four blocks as depicted in the figure, representing the differing similarities between the groups, and each block was analyzed for gene ontology terms using the DAVID Bioinformatics Resource (Fig. 5; see also Table S5 in the supplemental material). Perhaps the most immediate observation to follow from this was the different prevalences of ribosomal and protein biosynthetic genes, occurring as an upregulated category in the immunized animals; this contrasts to the day 14 results, where such genes were a dominant feature of the controls. Likewise, mitochondrial genes and proteasome components were seen to be more prominently upregulated in one or both of the immunized groups, again representing a major shift from day 14 findings.

The tractable size of the gene set and the range of expression differences between the groups made the week 12 ANOVA results amenable to a more detailed differential assessment using the Ingenuity Pathways Analysis program (Fig. 6). In addition to providing more specific annotations, it highlighted distinctions between the groups not evident in the density of Fig. 5. For example, we immediately noted from this analysis that interferon signaling was the only pathway/biofunction where the expression differentials (and thus the P value) was correlated to the viral loads for the treatment groups. However, the other notable antiviral pathway (activation of IRF by cytosolic pattern recognition receptors) did not track with viral loads and instead showed the most significant upregulation in the unprotected group IV control and in the most-protected, group II animals.

FIG. 6.

Comparison of pathways and biofunctions for treatment groups at week 12. The bar graph indicates statistical significance for relevant pathways and biofunctions as assessed by gene set analysis using the Ingenuity Pathway Analysis program, displayed as the −log₁₀ values of the associated P values. The analysis employed error-weighted average expression ratio across all six animals in each group, calculated for the 551 genes from the week 12 ANOVA. The data were then constrained to genes in an experimental group with average expression ratios of ≥1.5, thereby emphasizing pathways and functions with the largest increases in expression relative to the prechallenge state.

Upregulated genes for group I at week 12 implicate an evolving B cell compartment.

The differential functional analysis clarifies that group I had the most significant upregulation of genes involved in protein ubiquitination and cell cycle control. The upregulation of genes in the mTOR signaling pathway would be associated with increased protein biosynthesis, a process already implicated for group I from the increased expression of ribosomal genes. Consistent with this emphasis on protein biosynthesis, group I is unique in upregulation of genes for N-glycan biosynthesis. The upregulation of genes for oxidative phosphorylation is often associated with mitochondrial dysfunction attendant to apoptosis. However, the differential analysis showed group I had the least significant upregulation of genes associated with apoptosis, and the increased expression of mitochondrial components for oxidative phosphorylation must serve an increased energy requirement for a cellular subset in the blood. Finally, the presence of genes for V(D)J recombination and double-stranded DNA repair may imply maturation of a lymphocyte compartment, for example, antibody isotype switching (see below). The importance of B cell maturation in group I is also inferred from the very increased expression of the B cell markers CD79B and MS4A1 (CD20) (see below).

Panel A of Fig. 7 integrates many of these observations and emphasizes a central role for the transcription regulator MYC, as it is associated with increased expression of the cell cycle genes cyclin D2 (CCND2) and proliferating cell nuclear antigen (PCNA) (46, 98). PCNA, in addition to being a cofactor for DNA polymerase δ, has numerous other functions during chromosome replication, including orchestrating chromatin remodeling, represented by the included histone deacetylase components RBBP7 and MTA (69). PCNA is also implicated in dsDNA repair (12), as occurs in V(D)J recombination, a biofunction previously noted for group I. The figure also shows Myc-driven expression of the nucleolin gene (NCL) (47); NCL is involved in the maturation of rRNAs, and so we see a relation to the block of upregulated ribosomal genes identified for group I in the heat map. The proteasome category of Fig. 5 also recurs in the network in upregulated CUL1; this is a component of the Scf ubiquitin ligase, which polices the cell cycle process via targeted destruction of cyclins and other components (7). Finally, the network suggests cellular specificity with the references to increased expression of B cell markers CD79B and MS4A1 (aka CD20). This is also tied to cell cycle control with contribution of the B cell receptor to the increased expression of cyclin D2 and progression of B cells through the G₁/S cell cycle stage (84).

FIG. 7.

Network diagrams illustrating contrasting outcomes for vaccine groups I and II at week 12 postinfection. Networks were seeded by automated generation in the Ingenuity Pathway Analysis program, using the genes from the differential functional analysis shown in Fig. 6, and then expanded with other genes in the ANOVA set. (A) Group I (Ad5hr immunogens only) upregulated genes associated with cell cycle regulation and the relationship to increased expression of MYC and genes in B cell function. (B) Network of upregulated genes in group II (Ad5 immunogens plus gp140 boost), illustrating disparities for genes involved in cytoskeletal processes impacting cell adhesion as well as the motility/phagocytic processes discussed in the text. Also shown is the role of SPI1 upregulation in driving expression of genes implicated in macrophage function (FCGR, LYL1, and AZU1). In panels A and B, colors of nodes represent relative expression values for groups I and II, respectively. Associated heat maps compare the ANOVA set genes contained in the network for all three groups, rendered at a 2-fold saturation.

Group II animals showed a distinct upregulation of processes implicating activated macrophages.

For the protein-boosted vaccine recipients, the differential functional analysis of Fig. 6 indicates a very specific role for phagocytic immune cells such as macrophages, or possibly neutrophils. This provides an organizing theme to the ontologies for the second block of genes in Fig. 5, where group I was upregulated, group I was downregulated, and the control group had a general similarity to group II. The identification of the phagocytosis pathways reflects the increased expression of associated signaling intermediaries, including enzymes such as phospholipase D (PLD4), protein kinase C (PRKCI and PRKCG), and more specific components, such as the kinase substrate ezrin (EZR) and G-proteins such as RAB11B and RALA (96). These pathways ultimately require actin and cytoskeletal components for their mechanical execution, and thus they are consistent with the large number of such actin-associated genes identified for this block in the heat map (see Table S5 in the supplemental material). However, very few actin genes are referenced in the indicated pathways, and the significance scores are not driven by the presence or absence of the cytoskeletal genes. Consistent with the suggested role of macrophages, we saw that group II also shared particular upregulation of genes promoting chemotaxis of macrophages and neutrophils, including the IL-8 receptor (IL8RB [CXCR2]), the receptor for complement fragment C5a (C5AR1), and the secreted factor azurocidin I (AZU1). The latter is a neutrophil-derived inflammatory mediator that has significant effects on macrophage/monocyte activation and leukocyte adhesion (20, 62).

The representation in Fig. 7B more clearly illustrates this contrast between group II and group I as it relates to the actin and cytoskeletal genes, here placed in the context of cell adhesion but again critical for those phagocytosis functions referenced previously. While upregulation of focal adhesion kinase (FAK) was not observed, we saw a pronounced increase in the interactor MAPK8IP3, which scaffolds FAK and JNK during their action in cytoskeletal remodeling (97). The protein-boosted animals at this time also showed very significant upregulation of fascin (FSCN2); fascins are actin-bundling proteins involved in membrane structures important for cell motility and adhesion, for phagocytosis, and for antigen presentation on mature dendritic cells (3, 6). Growth factors (exemplified here by upregulated fibroblast growth factor [FGF]) also influence cytoskeletal processes via the downstream action of ERK1/2.

The right-hand side of the figure shows a convergence of cytokine signaling with the ERK1/2 signaling pathway via the STAT3-induced expression of myeloid transcription factor SPI1. SPI1 plays a significant role in the production of IL-12 by macrophages, thereby influencing the functions of T cells and NK cells (81). In support of the aforementioned phagocytic functions of macrophages, it also interacts with the transcriptional program downstream of ERK1/2, resulting in increased expression of the Fcγ receptor (as well as the inflammatory mediator azurocidin [see above]) (5). Also downstream of SPI1 is another very significantly upregulated gene, LYL1, a transcription factor generally found expressed in macrophages (95). LYL1 provides a connection to the last functional category highlighted in the figure: various genes implicating p53 signaling and regulated cell death in this experimental group (87).

To summarize, between day 14 and week 12, the expression profiles from whole blood for the two protected groups diverged considerably. Animals in Env-boosted group II showed upregulation of genes implicating activated macrophages, characterized by chemotaxis, adhesion, and phagocytosis. Upregulated genes for the unboosted group I emphasized protein biosynthesis and cell growth, with indications of B cell signaling and development. Genes for interferon signaling and inflammation trended with viral load and were most highly expressed in the control group.

DISCUSSION

In the context of the current vaccine study, we have demonstrated that gene expression profiling of whole blood is a method of sufficient detail and sensitivity to distinguish study groups months after the last treatment and prior to viral challenge and to give a view into gene expression programs that are particular to the best-protected treatment group in the study. The method also illuminates postchallenge differences between the treatment groups: at peak viremia, showing the extraordinarily different responses of controls versus immunized animals, and at early set point, suggesting distinct mechanisms of viral control for immunized groups that appear quite similar phenotypically.

Prechallenge expression profiles reveal gene expression networks for establishing B cell and T cell memory.

The unsupervised hierarchical clustering of prechallenge array data points to a distinction of Env-boosted group II compared to either the group IV controls or to group I animals. We believe we are observing continued dynamics in the peripheral lymphocytes in group II animals that ensued after the administration of the second protein boost 8 weeks earlier, and global expression profiling is sufficiently sensitive to detect this signature even after such a long interval. Genes that distinguish this group were enriched in immune functions and reflected specific changes in the T cell and B cell compartment, a result consistent with the induction by the boosting regimen of increased levels of Env-specific T cells and increased Env-binding antibody titers observed for this group at this prechallenge point (76, 107). Moreover, the changes in NFAT, NF-κB, and MYC expression suggest a functional and proliferative response of lymphocytes, but in association with an integrated network of expression changes that foster a prosurvival outcome. One hypothesis is that these are changes attendant to establishing the memory cohort of T and/or B cells elicited by the replicating Ad5hr-recombinant/protein boost vaccine approach (30, 76, 111). We note in this network the kinase gene AKT1; beyond a general attribution of promoting cell survival, AKT1 also mediates the phosphorylation of transcription factor FOXO3a, a process that drives the survival of memory T cells and B cells (102, 103).

At peak viremia, expression differences in whole blood reflect immunological distinctions other than effector T cell functions.

This is perhaps an unanticipated result given the results of Patterson et al. that showed prechallenge the clear induction of virus-specific memory T cell responses for groups I and II, which received the replicating Ad5hr recombinants, as well as the enhanced Env-specific T cell responses in the group II animals that also received the Env-protein boosts (76). For example, the expression levels of granzymes A, B, and K, associated with CTL effector function, trended with viral load and were highest in the control animals (see Table S2 in the supplemental material). Likewise, the ANOVA results did not clearly differentiate the vaccine groups by the extent of type II interferon signaling. It is important to consider that other cells also produce granzymes and gamma interferon; nonetheless, at peak viremia their expression did not appear enhanced in the immunized animals. It is possible that this level of detail is obscured when looking in the complex milieu of whole blood. But in light of recent publications showing that CD8⁺ T cell suppression of viremia is not affected by targeted killing of infected cells, our result also begs the question of other mechanisms of T-cell-mediated viral control (59, 106).

The expectation for control group IV would be mobilization of a de novo immune response, driven by very strong engagement of innate signaling pathways, and a type I interferon response. While these functions were not uniquely associated with group IV from the day 14 ANOVA gene set, many such genes showed trends with the greatest levels of expression in the controls (see Table S2 in the supplemental material). We posit that a consequence of the stronger signaling in the control group is the distinct upregulation of the antigen presentation molecules. A puzzling aspect is the contrasting behavior of Env-boosted group II, which showed lower expression levels of MHC genes on day 14, when the animals were exhibiting superior viral control. One possible explanation is the uncompensated downregulation of these genes that attends HIV and SIV infection (90, 94). An alternative is that the cell type(s) that was the primary source of these transcripts on day zero in group II have migrated from the blood to secondary lymphoid organs following activation or in response to homing signals such as CCL21, which was noted as upregulated for group II.

At peak viremia, control group IV exhibited extensive upregulation of a set of genes for protein biosynthesis, protein folding and trafficking, and RNA processing. This may reflect a proliferative burst of leukocytes, as the immune systems of the naive animals respond to the viral antigens or an attempt to repopulate after the apoptosis and cell death occasioned by unrestrained viral growth and attendant inflammation. A more specific explanation could be the functional response of B cells after initial antigen recognition leading to increased numbers of ribosomes and elevated levels of protein synthesis (2). Protein-boosted group II already has an established cohort of Env-specific B cells and does not manifest this generative process for newly activated B cells; rather, it appears to be further maturing the humoral response by generating antibody-producing plasma cells. Such preexisting B-cell-mediated immunity is not as strong in unboosted group I but expands considerably in the first 2 weeks postchallenge; in line with this reasoning, at day 14 for this group we saw partial overlap of upregulated protein biosynthetic genes to those identified for group IV.

The day 14 arrays suggest protective roles for IL-27 and complement pathways in immunized animals.

For the vaccine groups, increased expression levels of several genes were ascribed to the increased activity and maturation of antigen-presenting cells. Among the most intriguing of these was IL-27, a cytokine produced from activated dendritic cells that impacts responsiveness of naive cells, rather than memory cells. While the presence of this cytokine would not directly stimulate the T cell anamnestic response in the vaccine groups, IL-27 is a known inhibitor of HIV-1 replication in CD4⁺ T cells and macrophages and therefore would confer considerable benefit to the vaccinated animals (35). Immunized animals of groups I and I also exhibited increased expression of complement pathway genes, whose production is ascribed to activated macrophages. This may present a particular advantage to the protein-boosted animals in group II: the higher levels of Env-binding antibodies generated through this vaccine regimen may have exerted a protective effect through the classical complement pathway. Such complement-mediated viral lysis has already been reported for human clinical specimens, in which subject nonneutralizing antibodies mediate such lysis against both autologous and heterologous strains of HIV-1 (1, 51). This mechanism would synergize with ADCC and ADCVI, mechanisms already demonstrated as operative in animals receiving such prime-boost regimens and correlating with reduced acute viremia (39, 48, 107).

Despite the phenotypic similarities of the immunized groups at week 12, the microarray profiles indicate differing mechanisms of viral control.

The extreme similarity in expression patterns for groups I and II at day 14 stands in contradistinction to their virological and immunological differences through the early part of the infection, specifically, the ∼10-fold-higher viral load for the group I animals versus group II, as well as the stronger anamnestic responses in the group II animals as gauged by gamma interferon ELISPOT and Env-binding antibody titers. In contrast, when approaching set point at week 12, the expression profiles for the vaccinated groups diverged considerably, an outcome not entirely anticipated on the basis of phenotypic similarities at this stage of the infection. Both groups are entering set points of comparable viral loads, comparable CD4⁺ counts, and similar levels of antigen-specific T cells at week 12, and yet the expression profiles showed substantial divergence from each other. When we assessed the relationships between the groups by post hoc treatment of the week 12 ANOVA results, we saw the greatest number of differences between the two immunized groups (453 genes), suggesting that groups I and II employing different immunological mechanisms for virus control at this early set point.

It is striking that at 12 weeks postinfection, expression signatures for B cell development are different between the protected groups, while in contrast, titers of anti-Env antibodies have plateaued and are not statistically different between the groups (107). Maturation (or possibly replenishment) of this subset postchallenge appears as a strong feature for group I, as evidenced by the upregulation for this group of the B cell receptor subunit CD79B, MS4A1 (also known as CD20, suggesting plasma cell formation), and genes for V(D)J recombination. Absence of these features for the control animals could be attributed to the lack of CD4⁺ T helper cells, which are required for full induction of the B cell response, including memory development. Quite different considerations apply to Env-boosted group II. Measurements from related studies have demonstrated that prime-boost regimens such as the one employed here induce abundant IgG and IgA virus-specific memory B cells in the peripheral blood prechallenge, and these memory cells persist through the chronic phase (E. Brocca-Cofano et al., unpublished data). Concordant with this, the array data suggest that the postchallenge evolution of the B cell compartment is quite different for group II compared to those animals immunized with only the replicating Ad5hr recombinants and persists on a different course long after viral infection.

From the differential functional analysis based on the week 12 expression data (Fig. 6), we propose a very specific role for phagocytic immune cells such as macrophages in the immune response for Env-boosted group II, a process that would be quite strongly linked to the antibody response of this group. This proposition parallels recent ex vivo studies demonstrating that group II has significantly greater ADCC and ADCVl activities, which are established prechallenge and persist through the chronic phase (107). The higher degree of ADCC and ADCVI for group II correlates to the greater avidity of the Env-binding antibodies during the immunization regimen. Our conclusion from the expression features of group II for the distinct operation of activated macrophages, including Fcγ receptor signaling, provides evidence of ADCC and ADCVI as an in vivo mechanism of virus control in these best-protected animals.

Recent studies have reported on longitudinal expression profiling of peripheral leukocytes from SIV infections in rhesus or pig-tailed macaques, in which aspects of pathogenesis were assessed and contrasted to observations for natural hosts (18, 54, 61). While the primary objective of our analysis was to determine differences among vaccine treatment groups, our observation of strong and persistent upregulation in control group IV of interferon and innate signaling pathways is consistent with these and other reports (43). At peak viremia, the expression data did not show statistically significant differences among the study groups in the regulation of these pathways. However, at set point, the expression levels of these immune activation genes in the two immunized groups were now significantly lower than the controls. This difference, as revealed by the whole-blood expression profiles, may represent a significant therapeutic benefit from the vaccine regimen, inasmuch as chronic immune activation is a superior predictor compared to viral load of poor clinical outcome for HIV-infected individuals (28).

We present these results with the caveat that the expression changes we observed can arise from two sources: modulation of transcripts in a fixed population, or changes in representation of cell types within a sampled population. We nonetheless believe we have demonstrated that expression profiling of whole blood does allow us to draw relevant and unanticipated conclusions about the host responses to vaccination and infection. Other functional genomics studies of vaccine response have utilized isolated subsets from peripheral blood, and these methods may have advantages for specific mechanistic questions (83). However, the use of whole blood offers a physiological integration of the various aspects of the host response to infection and may provide truer in vivo measurements for exploring hypotheses. It may also reveal aspects that originate from many small differences that would be overlooked in analyses of isolated cellular subsets and yet are amplified in the context of the host. In the future, we hope to apply this technique to analysis of earlier time points in nonhuman primate vaccine trials in which more distinct immunization regimens are compared. We have the expectation that treatment-induced gene expression changes will be more robust in shorter time frames, offering the prospect of identifying immediate innate immune responses and progressive adaptive immune responses. In addition, we plan to investigate statistically significant correlations of expression features to virological and immunological data from this and other studies, and similarly, to exploit new techniques for deconvoluting expression features of specific cell types within complex mixtures, such as whole blood (91). A key goal in this regard would be to ultimately enable prospective evaluation of vaccine regimens by simply sampling whole blood, rather than particular cell types or tissues. Particularly advantageous would be identification of surrogate markers of mucosal immune responses in the blood, precluding the necessity of sampling mucosal tissues in large clinical trials.