Deep Transcriptional Sequencing of Mucosal Challenge Compartment from Rhesus Macaques Acutely Infected with Simian Immunodeficiency Virus Implicates Loss of Cell Adhesion Preceding Immune Activation

ABSTRACT

Pathology resulting from human immunodeficiency virus (HIV) infection is driven by protracted inflammation; the primary loss of CD4⁺ T cells is caused by activation-driven apoptosis. Recent studies of nonhuman primates (NHPs) have suggested that during the acute phase of infection, antiviral mucosal immunity restricts viral replication in the primary infection compartment. These studies imply that HIV achieves systemic infection as a consequence of a failure in host antiviral immunity. Here, we used high-dose intrarectal inoculation of rhesus macaques with simian immunodeficiency virus (SIV) SIV_mac251 to examine how the mucosal immune system is overcome by SIV during acute infection. The host response in rectal mucosa was characterized by deep mRNA sequencing (mRNA-seq) at 3 and 12 days postinoculation (dpi) in 4 animals for each time point. While we observed a strong host transcriptional response at 3 dpi, functions relating to antiviral immunity were absent. Instead, we observed a significant number of differentially expressed genes relating to cell adhesion and reorganization of the cytoskeleton. We also observed downregulation of genes encoding members of the claudin family of cell adhesion molecules, which are coexpressed with genes associated with pathology in the colorectal mucosa, and a large number of noncoding transcripts. In contrast, at 12 dpi the differentially expressed genes were enriched in those involved with immune system functions, in particular, functions relating to T cells, B cells, and NK cells. Our findings indicate that host responses that negatively affect mucosal integrity occur before inflammation. Consequently, when inflammation is activated at peak viremia, mucosal integrity is already compromised, potentially enabling rapid tissue damage, driving further inflammation.

IMPORTANCE The HIV pandemic is one of the major threats to human health, causing over a million deaths per year. Recent studies have suggested that mucosal antiviral immune responses play an important role in preventing systemic infection after exposure to the virus. Yet, despite their potential role in decreasing transmission rates between individuals, these antiviral mechanisms are poorly understood. Here, we carried out the first deep mRNA sequencing analysis of mucosal host responses in the primary infection compartment during acute SIV infection. We found that during acute infection, a significant host response was mounted in the mucosa before inflammation was triggered. Our analysis indicated that the response has a detrimental effect on tissue integrity, causing increased permeability, tissue damage, and recruitment of SIV target cells. These results emphasize the importance of mucosal host responses preceding immune activation in preventing systemic SIV infection.

INTRODUCTION

Human immunodeficiency virus (HIV)-induced immune cell depletion is primarily caused by protracted inflammation. However, recent studies indicate that the earliest time period following HIV infection, the acute phase, is critical in HIV pathogenesis (1). Nonhuman primate (NHP) models have revealed that during this phase, mucosal immunity can inhibit viral replication and prevent systemic infection. The low rate of sexual transmission among humans (≤0.5% occurrences per sexual contact) also implies that the virus runs a high risk of dying in the primary infection compartment or adjacent tissues before it can infect a sufficient number of CD4⁺ cells to spread systemically. Even after exposure to the virus, early administration of antiretroviral medication can clear the virus from the host (2, 3). This makes the acute phase a critical time point in the transmission process.

Currently, simian immunodeficiency virus (SIV) infection of NHPs is the most accurate model to study the early events following HIV infection in humans. In this study, we used deep RNA sequencing (RNA-seq) to study the host transcriptional response at the site of inoculation during early SIV infection of rhesus macaques (RMs). The aim was to identify aspects of the host response that contribute to early viral control, or the loss thereof, leading to subsequent systemic HIV/SIV infection. Understanding these processes could greatly contribute to the development of therapies to decrease transmission rates between individuals. To our knowledge, this is the first in-depth look at the mucosal host response to SIV infection during the acute phase.

The acute phase of HIV/SIV infection follows a characteristic time course, which provides an opportunity to identify key events that could be modulated by vaccines or other therapeutics to limit subsequent pathogenesis. Sexual transmission of HIV/SIV is followed by a time period when the virus is undetectable in the circulation, termed the “eclipse phase” (4). NHP studies have led to the discovery of several mechanisms by which the host controls SIV, including viral entry blockage (e.g., SDF-1, MIP1a/b), alpha/beta interferon (IFN-α/β) expression by mucosal dendritic cells, and expression of host restriction factors (e.g., BST2, CD317) (1). If these mechanisms fail, a small number of virions eventually escape host restriction and infect mucosal CD4⁺ T cells, macrophages, and dendritic cells (DCs). Carried by DCs, the virus eventually reaches the draining lymph nodes, which brings it into contact with a large number of target CD4⁺ CCR5⁺ T cells (5). Viral replication then increases rapidly as the infection spreads first to other lymph nodes and then to the bloodstream. This process takes about 1 week in NHPs. The rising viremia is accompanied by an adaptive immune response involving specific B cells and CD8⁺ and CD4⁺ T cells (6). At about day 12, viremia reaches its peak and then decreases to the viral set point. At the viral set point, although the viral load is reduced, the immune system is gradually depleted through inflammation-driven apoptosis. The protracted inflammation at the viral set point is a key driver in the development of AIDS, as shown by three main findings: (i) most of the immune cells that are lost in HIV infection are themselves not infected by the virus but, rather, are bystander cells (7), (ii) NHP species that are natural carriers of SIV (e.g., African green monkeys) resolve the inflammatory response after peak viremia and do not develop immunodeficiency, despite a high viral load (8, 9), and (iii) levels of inflammatory markers predict disease progression more accurately than viral load (10, 11).

In the interest of learning how viral spread from the primary infection compartment can be prevented, we focused this study on the eclipse phase. Building upon in vivo studies of NHPs that found mucosal antiviral mechanisms capable of restricting SIV replication (12), we investigated the critical question of how this control is maintained and lost. The fact that the sexual transmission rate of HIV is low implies that the physical barriers and immune systems in the mucosal tissues are normally successful at preventing infection (13, 14). In apparent contradiction, activation of the innate immune system during the acute phase can also be harmful to the host; early responses by NK cells, macrophages, and dendritic cells attract CD4⁺ T lymphocytes and other target cells to the site of infection (15). Mucosal immune activation has additional harmful effects, as it can result in compromise to the integrity of the mucosal barrier, thereby leading to bacterial translocation into the lamina propria, enhancing protracted inflammation (16). It has not been determined whether the loss of epithelial integrity is merely a consequence of ongoing inflammation or if the virus itself can cause epithelial damage. While in vitro studies have found that HIV/SIV can affect the integrity of epithelial explants without the influx of immune cells (17), our use of NHPs provides a unique opportunity to explore this issue in vivo.

In this study, we carried out high-dose intrarectal inoculation of Indian-origin RMs with SIV_mac251 and sacrificed animals at 3 or 12 days postinoculation (dpi), corresponding to the eclipse phase and the time of peak viremia, respectively. Rectal tissue specimens were obtained at the site of inoculum deposition and analyzed by deep mRNA sequencing (mRNA-seq) to capture the host response at the infection site. During the eclipse phase (3 dpi), we observed the activation of genes encoding cytoskeletal remodeling and cell adhesion proteins. This response was followed by a strong inflammatory response during peak viremia (12 dpi).

A notable of advantage of RNA-seq is that it can quantify both coding gene and noncoding RNA (ncRNA) expression. The role of ncRNA in SIV or HIV pathogenesis has not been extensively explored, but it has been shown that the suppression of the enzymes required for microRNA (miRNA) biogenesis leads to enhanced HIV replication in peripheral blood mononuclear cells (18). Likewise, knockdown of particular long ncRNA can affect HIV replication (19), similar to the knockdown of coding genes known as HIV dependency factors. In addition, RNA sequencing of HIV-infected SUP-T1 cells has shown the differential regulation of several classes of ncRNA (20). Whereas these previous studies were carried out in vitro, the design of the present study allowed us to examine ncRNA expression in vivo on a genomic scale.

This study suggests that even during the eclipse phase, the rectal mucosa is mounting a response that compromises epithelial integrity. The fact that these changes preceded both local inflammation and viremia raises the possibility that these changes are independent of inflammation and play an important role in the loss of viral containment within the challenge compartment. This is the first study to evaluate mucosal host responses associated with epithelial compromise as early as 3 days after SIV inoculation.

MATERIALS AND METHODS

Animals and ethics statement.

All animal procedures were performed by standard protocols and according to guidelines approved by the University of Washington Environmental Health and Safety Committee, the Occupational Health Administration, the Primate Center Research Review Committee, and the Institutional Animal Care and Use Committee. The eight male RMs that underwent intrarectal SIV challenge were housed at the Washington National Primate Research Center. Control tissues from three uninfected animals were obtained from the tissue distribution program run by the National Primate Research Center. All challenged animals were specific pathogen free (SPF) and were negative for the protective major histocompatibility complex class I alleles Mamu A01, B01, and B17.

Intrarectal SIV challenge.

The eight RMs were intrarectally challenged with SIV_mac251 using 1 ml of a high-dose inoculum (6,000 50% tissue culture infective doses [TCID₅₀s]/ml). SIV inocula were deposited at a rectal depth of 25 mm from the anus. Baseline rectal tissue samples were obtained 14 days prior to viral challenge by pinch biopsy.

Tissue preparation and RNA extraction.

At necropsy, all rectal tissue samples were taken at a depth of 25 mm from the anus, corresponding to the depth at which the inocula were deposited. Rectal tissues were immediately perfused in RNAlater and stored at −80°C until further processing. Tissues were homogenized in 20 volumes of RLT reagent (Qiagen) using an Omni TH tissue homogenizer (Omni International, Kennesaw, GA). RNA was extracted from the rectal tissue homogenate using an AllPrep DNA/RNA/protein kit (Qiagen). RNA concentrations were quantified using an ND-2000c UV-visible spectrophotometer (NanoDrop; Thermo Scientific, Wilmington, DE) and controlled for integrity and purity on a capillary electrophoresis system (Agilent 2100 bioanalyzer; Agilent Technologies, Santa Clara, CA).

Blood samples for viral load measurement were taken (i) at the time of baseline sampling 14 days prior to inoculation, (ii) at the necropsy at 3 dpi, (iii) at 6 dpi from the animals to be sacrificed at 12 dpi, and (iv) at the necropsy at 12 dpi. Whole blood was collected into EDTA tubes (Becton, Dickinson, Franklin Lakes, NJ) for use in plasma isolation. The contents of the tubes were mixed by inversion, and the tubes were subsequently centrifuged at 1,300 × g for 10 min. The upper (plasma) layer was carefully removed and stored at −80°C for later analysis. Viral RNA was prepared from EDTA-anticoagulated, cell-free plasma using a Gentra Puregene RNA isolation kit according to the manufacturer's instructions (Gentra Systems, Minneapolis, MN). RNA was precipitated in the presence of glycogen, resuspended in 50 μl of nuclease-free water, and immediately analyzed.

Viral load measurement.

Plasma viral load was determined by real-time reverse transcription (RT)-quantitative PCR (qPCR) based on published methods (21). The intracellular viral RNA load in the rectal mucosa was quantified as previously described (22).

mRNA library preparation.

mRNA libraries were constructed using an Illumina TruSeq RNA preparation kit (Illumina, San Diego, CA) according to the manufacturer's guide. Libraries were quality controlled and quantitated using a BioAnalzyer 2100 system and qPCR (Kapa Biosystems, Woburn, MA). The libraries were clonally amplified on a cluster generation station using Illumina (version 4) cluster generation reagents to achieve a target density of approximately 700,000/mm² in a single channel of a flow cell.

Next-generation sequencing and read mapping.

The resulting libraries were sequenced on a Genome Analyzer IIx apparatus (Illumina, San Diego, CA) using Illumina (version 5.0) sequencing reagents, which generated paired-end reads of 75 nucleotides (nt). Image analysis, base calling, and error estimation were performed using Illumina Analysis Pipeline (version 2.8) software. Raw reads were trimmed to 50 bp, and adapter sequences were removed. The 50-bp reads were mapped to known ribosomal sequences (human, mouse, rat) using the short-read aligner software Bowtie to remove potential rRNA sequences to maximize the coverage of reads mapped to our annotation (23). Viral reads were then determined by mapping to the SIV_mac251 genome (GenBank accession no. M19499.1) using the gapped aligner software TopHat, which predicts splicing junctions and maps intron-spanning reads to known splicing junctions (24). We then mapped all the remaining reads to the rhesus macaque reference genome (source, Ensembl; build, Mmul_1) from Illumina's igenomes using TopHat. After mapping, we assigned aligned read counts from BAM files to exons and genes using the python package HT-Seq (25). HT-Seq provided the most accurate way of aligning read counts to overlapping exons. Reads that mapped to multiple positions were removed. Annotations for human large intergenic noncoding RNA (lincRNA) were obtained from a previously published catalogue (26). Annotations for novel macaque ncRNA were obtained from the Nonhuman Primate Reference Transcriptome Resource (27, 28).

For visualization, BAM files were generated using TopHat and SAMtools (29) and displayed using the Integrative Genomics Viewer (IGV) genome browser. Read count refers to the number of sequenced cDNA fragments that map to a particular genomic feature. Normalization and differential expression analysis were carried out using R (version 2.14.1) and software package edgeR. Normalization comprised calculation of a size factor for each sample (as the median ratio of the read counts for each feature and sample to the geometric mean of the read counts for each feature across samples) and dividing all of the read counts in a particular sample by the sample size factor (30).

Differential expression analysis.

Differentially expressed mRNA and noncoding RNA were determined using a generalized linear model implemented in the Bioconductor package edgeR (R, version 2.15.3; edgeR, version 1.8.3). To avoid bias between samples obtained by necropsy and baseline samples obtained by pinch biopsy, differential expression was determined by two tests. First, each gene underwent a paired test between necropsy samples and baseline samples from the same animal. Second, each gene underwent a group-wise unpaired test for the time point of interest (3 dpi or 12 dpi) versus all eight pinch biopsy baseline samples and uninfected rectal necropsy samples. P values were adjusted for multiple testing by use of the false discovery rate (FDR) (see Fig. S1 in the supplemental material). Differentially expressed (DE) coding genes and noncoding RNA were defined as those with an adjusted P value of <0.05 and an absolute fold change of >1.5 in both tests.

Functional enrichment analysis.

Functional enrichment of differentially expressed genes was carried out using Ingenuity Pathway Analysis (Ingenuity Systems, Inc.). Predefined, manually curated functional categories containing given genes were tested for statistically significant enrichment with differentially expressed genes using Fisher's exact test. The functional categories have a hierarchical organization, with more specific subcategories (e.g., activation of lymphocytes) being aggregated into more generic categories (e.g., cellular signaling and interaction). Since the specific subcategories can be highly overlapping, we present only the most enriched subcategory in each broad category. In addition to determining the enrichment of DE genes in functional categories, the analysis includes information on genes that inhibit or activate each function and uses a regularization Z-score to predict whether an enriched functional category is inhibited or activated (31). The categories shown in Fig. 2B were chosen by a manual survey of each generalized category and selection of specific subcategories that included a large number of DE genes and exhibited a strong enrichment P value.

FIG 2.

(A) Differential expression analysis at 3 dpi and 12 dpi. Differentially expressed genes were defined as having an adjusted P value of <0.05 and an absolute log₂ fold change of >1.5. Each column corresponds to an individual animal. Colors represent the fold changes of expression of each gene. The upper section includes genes that were DE at 3 dpi only, the middle section includes genes that were DE at both time points, and the lower section includes genes that were DE at 12 dpi only. The middle section includes 888 genes. (B) Functional enrichment analysis, showing representative biological functions from the most enriched functional categories, shown on the right side. Circle sizes indicate the total number of differentially expressed genes in each function; color intensities indicate enrichment significance. The functional enrichment also distinguished between genes that activate and inhibit each function. Where the functional enrichment predicts activation or inhibition, this is indicated by an upward or downward pointer, respectively. -Log₁₀P, −log₁₀ P value. (C) Canonical pathway enrichment. Unlike the biological functions, canonical pathways describe signaling cascades that are activated by extracellular signals. Diff., differentiation; Cell., cellular.

Functional enrichment analysis of coexpressed gene sets was carried out using the GOSim library.

Coexpression analysis.

All differentially expressed coding RNA and ncRNA from both time points were binned together, amounting to 4,015 transcripts. Coexpression between all pairs of transcripts was determined using biweighted midcorrelation, a measure which has shown good performance compared to alternative methods (32). Coexpressed transcripts were organized into modules by hierarchical clustering using the Ward method (33) and adaptive branch pruning (34).

To avoid the risk that a small set of outlier samples plays a dominating role in generating coexpressed gene sets, we evaluated the hierarchical clustering by a bootstrap test, wherein the hierarchical clustering was repeated on randomized subsets of the data (35).

The coexpression network was constructed by connecting each transcript to the two other transcripts with which it shared the highest biweighted midcorrelation. This method does not require a correlation cutoff, and it also avoids the construction of networks consisting of large completely connected groups of genes or groups of genes that are completely unconnected (36).

Quantitative PCR.

RNA from rectal tissue samples was reverse transcribed using a QuantiTect reverse transcription kit (Qiagen, Valencia, CA). The resulting cDNA samples were diluted 50 times. SYBR green qPCR assays were run for each sample in triplicate. Relative expression was calculated using the ΔΔC_T method with averaged ΔC_T values (where C_T stands for threshold cycle) for the rhesus macaque 18S rRNA (FJ436026.1) and ACTB1 (NM_001033084.1) genes as a calibrator, as the expression of neither of these genes changed significantly over time in the mRNA sequencing data.

Analysis of public data sets related to UC.

Four data sets contrasting colon biopsy specimens from patients with ulcerative colitis (UC) to those from healthy controls were obtained from the Gene Expression Omnibus database (accession numbers GSE9686, GSE10191, GSE22619, and GSE38713). All data sets were individually normalized using quantile normalization; DE genes were determined using the Bioconductor package limma. Genes were classified as up- or downregulated if their P values were below 0.05 and their fold changes showed the same direction (up or down) in all four analyses.

Accession number.

The mRNA sequencing data from this study are available in the Gene Expression Omnibus database (accession number GSE56845) and the Sequence Read Archive.

RESULTS

Viral reads detectable in rectal mucosa at 3 dpi.

Eight Indian-origin rhesus macaques (RMs) were infected intrarectally with SIV_mac251, using an inoculum containing 6,000 TCID₅₀s. The inoculation protocol was designed to avoid abrasions in the rectal mucosa. Four animals were sacrificed at 3 dpi, and four additional animals were sacrificed at 12 dpi (Fig. 1A). Rectal tissues were obtained at necropsy from all animals at the site of inoculum deposition. Rectal tissues were examined to ensure that no visible damage had been caused by the inoculation procedure. Uninfected baseline samples were obtained by pinch biopsy 14 days prior to inoculation (referred to as −14 dpi). To avoid bias arising from the different sampling techniques, we also included rectal mucosal tissue obtained from three uninfected RMs at necropsy. Host responses in rectal mucosa were examined at the transcriptomic level using deep mRNA sequencing.

FIG 1.

(A) Experimental design. Eight animals were intrarectally inoculated with SIV_mac251. Baseline samples were obtained by rectal pinch biopsy and blood draw 14 days prior to inoculation. Four animals were sacrificed at 3 dpi (group D3), and 4 animals were sacrificed at 12 dpi (group D4). (B) Quantification of viral load by quantitative RT-PCR (qRT-PCR) in plasma (log₁₀ SIV RNA copies/ml; left) or rectal mucosa (log₁₀ SIV RNA copies/μg total RNA; center) and of viral reads in rectal mucosa by deep RNA sequencing (log₁₀ SIV reads; right). The x axis represents the time point (days postinoculation); viral load is represented on the y axis. The result for each animal is shown as an individual bar. Viremia was detectable after 6 days. In contrast, viral reads were detected in the rectal mucosa by day 3 in all animals by deep RNA sequencing.

To examine viral levels at these time points, virus was quantified in (i) peripheral blood and rectal mucosa by quantitative RT-PCR and (ii) rectal mucosa by mapping sequenced mRNA reads to the SIV genome (Fig. 1). Viral DNA was quantified by quantitative RT-PCR (see Fig. S2 in the supplemental material). Viral RNA was undetectable in blood samples taken at the baseline (data not shown) and at 3 dpi. At 6 dpi, low levels of viremia ranging from 355 to 8,730 RNA copies/ml of plasma (geometric mean = 3,020 RNA copies/ml) were detected. At 12 dpi, viremia had risen to between 1.41 × 10⁷ and 2.59 × 10⁷ RNA copies/ml of plasma (geometric mean = 2.10 × 10⁷ RNA copies/ml; Fig. 1B, left). Similarly, at 3 dpi, no virus was detected in the rectal mucosa of three out of four animals, while a high viral load was found at 12 dpi (range, 3.00 × 10⁵ to 6.99 × 10⁵ RNA copies/μg total RNA; Fig. 1B, center). In contrast, using deep mRNA-seq, low numbers of SIV reads ranging from 1 to 18 reads per sample (geometric mean = 7.54 reads per sample) were detected at 3 dpi in mucosal samples from all four animals. Although these numbers were very low, no reads mapping to the SIV genome were found in any of the baseline mucosal samples or in mucosa from uninfected control animals. At 12 dpi, the number of reads in rectal mucosa had almost risen above 7,000 in all animals (geometric mean number of reads = 15,882; Fig. 1B, right).

These analyses imply that at 3 dpi, the virus was present in rectal mucosa but had not reached the bloodstream, corresponding to the eclipse phase. Virus in blood was detectable at 6 dpi; by 12 dpi, the viral load had increased by several orders of magnitude both in the challenge compartment and in blood.

A strong mucosal transcriptional response was detected both during the eclipse phase and at peak viremia.

To characterize the mucosal host response at 3 and 12 dpi, mRNA-seq data were mapped to the RM genome, enabling the quantification of coding gene and noncoding RNA (ncRNA) expression. To limit technical noise, we first summed the number of mapped reads for each annotated gene and ncRNA across all 19 samples and removed those with less than 20 detected reads; 18,926 of 30,246 annotated coding genes were expressed in the rectal samples (i.e., genes with >20 total reads). This cutoff was chosen to enable detection of transcripts expressed in only a small subset of cells in the mucosal samples (e.g., immune cells).

Coding genes and ncRNA that were differentially expressed at 3 and 12 dpi were identified by contrasting postinoculation necropsy samples to (i) matched baseline pinch biopsy specimens from each animal using a paired test and (ii) a pool of all baseline pinch biopsy specimens and necropsy samples from uninfected animals by an unpaired test (see Table S1 in the supplemental material). Differential expression was defined as having an adjusted P value (FDR) of <0.05 and an absolute fold change of ≥1.5 (|log₂ fold change| ≥ 0.58) in both of these tests. Despite these stringent criteria, we identified a large number of differentially expressed coding genes as well as ncRNA at both time points. The number of differentially expressed coding genes at 3 and 12 dpi was 1,507 and 2,905, respectively (Fig. 2A; see the supplemental material). Notably, over 85% of all DE genes at each time point showed upregulation.

DE genes at 3 dpi showed a strong tendency to remain DE at 12 dpi, particularly in two of the animals. Specifically, there were 888 DE genes common to both time points, which amounted to a 3.84-fold enrichment compared to that expected to occur at random (P < 10⁻¹⁵, Fisher's exact test based on the 18,926 expressed genes). All of these genes showed the same direction of change at both time points, meaning that a quarter of the genes that were differentially expressed at peak viremia were already differentially expressed during the eclipse phase.

The mucosal immune response is preceded by the differential expression of genes associated with rearrangement of the cytoskeleton and cell adhesion.

For a comprehensive biological overview of the DE genes, we utilized Ingenuity Pathway Analysis to determine the functional gene categories that were enriched among the DE genes, as well as what categories showed activation or repression (Fig. 2B; see Tables S2 and S3 in the supplemental material). At 3 dpi, the functional enrichment primarily implicated genes associated with increased rearrangement of the cytoskeleton and the formation of cellular protrusions. These were associated with several aspects of HIV infection in humans, including endocytosis, exocytosis, and recruitment of coreceptors to an HIV-bound CD4 receptor. However, at 3 dpi a significant number of DE genes were also involved in differentiation of several types of connective tissue cells, including stromal cells, smooth muscle cells, and adipocytes (Fig. 2B; see Table S2 in the supplemental material). We also found that a significant number of DE genes were involved in maintenance of epithelial tissue integrity (e.g., cell-to-cell adhesion, formation of focal adhesions, gap junction signaling). This suggests that the reorganization of the cytoskeleton is not limited to intracellular structures but also indicates early effects on epithelial cell adhesion. Notably, while our analysis showed significant transcriptional perturbation at the inoculation site early after SIV infection, we also observed a marked absence of inflammatory functions. At 3 dpi, we did observe a small number of DE genes with a documented role in the host response to HIV infection, including the IFNA21 (37), TRIM22 (38), and ISG20 (39) genes. These genes could represent the first signs of an antiviral immune response.

In contrast, the analyses at 12 dpi showed a strong enrichment of DE genes in immune functions, predominantly T cell activation (Fig. 2B; see Table S3 in the supplemental material). Whereas many of these functions related to T cells in general, two T cell subsets were specifically implicated: Th1 cells and cytotoxic T lymphocytes. These specialized functions are subsets of generic T cell functions. Activation of B cells was also evident, including earlier phases of development (e.g., development of pro-B lymphocytes) and IgG production. The enriched functions also implicated the activation and recruitment of several innate immune cell types, including NK cells, dendritic cells, eosinophils, and macrophages. This suggests a well-developed antiviral immune response at 12 dpi involving the recruitment of innate and adaptive cell types. Many of the functions that were enriched at 3 dpi included the same number of DE genes at 12 dpi, but with the total number of DE genes at 12 dpi having increased by a factor of ∼2, these functions were not statistically significantly enriched at the later time point. If this study had focused exclusively on peak viremia, the statistically significant enrichment of DE genes associated with the activation of cytoskeletal functions would not have been detected.

We also more closely analyzed the underlying pathways corresponding to intracellular signaling cascades that were enriched at the two time points (Fig. 2C). This analysis confirmed the clear difference between the cytoskeletal and cell adhesion pathways that were activated at 3 dpi and the immune system pathways that were activated at 12 dpi. Among the many immune cell-related pathways, interferon signaling ranked among the highest at 12 dpi, with DE genes including the genes for six type I interferons (IFNA2, IFN6, IFN8, IFN10, IFN14, and IFNB1) and the type II interferon IFNG. Consistent with the presence of interferon signaling, we identified 133 DE interferon-stimulated genes (ISGs) at 12 dpi, which constituted a 2.95-fold enrichment (P < 10⁻¹⁵). In fact, the 100 most significantly DE genes at 12 dpi included 54 ISGs, amounting to a 26.1-fold enrichment. Several of the most activated pathways also involved the strong upregulation of genes encoding pattern recognition receptors, including Toll-like receptor 2 (TLR2; 9.64-fold), TLR3 (5.03-fold), and CLEC7A (7.41-fold). Whereas TLR3 activates IRF3/7 upon recognition of viral antigens (double-stranded RNA), TLR2 and CLEC7A both activate NF-κB after recognizing bacterial and fungal antigens, respectively.

Coexpression analysis implicates coding genes and ncRNA in the loss of cell adhesion.

While the analyses presented above provided an overview of the functions associated with differentially expressed coding genes, RNA-seq can also be used to quantify ncRNA. To predict the functions of ncRNA in acute SIV infection, we carried out a coexpression analysis between differentially regulated ncRNA and coding genes. This also assisted with the functional characterization of transcripts with specific expression patterns.

For a comprehensive overview of rhesus macaque ncRNA, sequenced reads were mapped to three annotations that explored different categories of ncRNA (Fig. 3A). First, we used previously characterized ncRNAs from the Ensembl rhesus macaque reference genome (build, MMUL_1), which provided an overview of classes of ncRNA (e.g., rRNA or miRNA). We term these “known ncRNA.” The second annotation consisted of sequences orthologous to human large noncoding intergenic RNA (26), which we term “lincRNA.” The third annotation consisted of transcripts from the nonhuman primate reference transcriptome resource (27) that did not correspond to any characterized transcript in Ensembl and that showed low protein-coding potential. This was the most comprehensive annotation, containing 6,027 transcripts in total. We term these “unannotated ncRNA.” Given that these transcripts were not previously annotated, we carried out RT-qPCR validation of five strongly upregulated unannotated ncRNAs, which showed a high reproducibility of the RNA sequencing results (see Fig. S2 in the supplemental material).

FIG 3.

(A) Heat maps giving an overview of differentially expressed ncRNA by three different mappings: ncRNA annotated in Ensembl, macaque homologues of human long ncRNA, and previously unannotated ncRNA from the nonhuman primate reference transcriptome project. (B) Coexpression analysis of differentially expressed coding RNA and ncRNA across all 19 biological samples. Differentially expressed transcripts formed eight coexpressed clusters. The heat map shows the average log₂ fold changes (FC) at the two time points compared to the level of expression at the baseline. The squares show enrichment of coexpressed gene sets among up- or downregulated genes in ulcerative colitis. Cluster 4, which was uniquely downregulated at both time points, was organized into a coexpression network for a more detailed view of its internal correlation structure. Each gene was connected to the gene with which it shared the strongest and second strongest correlation. The line thickness is dependent on the correlation coefficient. Nodes are color coded according to RNA class and sized according to their number of interactions; color intensity is dependent on the fold change at 3 dpi.

Due to the lack of a functional annotation for ncRNA, we carried out a coexpression analysis between coding genes and ncRNAs to associate ncRNAs with the functions of their coexpressed coding genes. We pooled the 4,015 differentially expressed transcripts (both coding and noncoding) and organized them into coexpressed groups by hierarchical clustering, which produced eight coexpressed gene sets (referred to as clusters 1 to 8; Fig. 3B; see Table S4 in the supplemental material). The reproducibility of this analysis was tested by a bootstrapping test, which showed a highly reproducible hierarchical structure (P < 0.0001 for all eight clusters). Most of these clusters showed predominant upregulation at either 3 or 12 dpi. The most notable exception was cluster 4, which showed downregulation at both 3 and 12 dpi. For each cluster, we performed a functional enrichment analysis using the gene ontology biological processes of their coding genes (see Table S5 in the supplemental material). Clusters 1, 2, 3, and 6, which were predominantly activated at 12 dpi, were associated with antiviral, innate immunity. Cluster 8 was primarily activated at 3 dpi and was associated with microtubule organization, cell spreading, and cell adhesion.

The downregulated cluster 4 is associated with wound healing, cell-cell adhesion, and tissue formation, similar to the functions enriched in the full DE list at 3 dpi. The coexpression analyses showed that a significant number of genes associated with these functions were downregulated at both time points. Notably, this coexpressed transcript set contained 17% ncRNA, constituting the highest proportion of all clusters. In cluster 4, the largest family of genes associated with cell adhesion proteins encoded claudin tight junction proteins of the epithelium (CLDN3, CLDN4, CLDN5, CLDN23). The rest encoded non-voltage-sensitive sodium channels (SCNN1B, SCNN1G), extracellular matrix proteins (COL5A1, LAMC2), and a gap junction protein (GJB2). The expression pattern and functional associations of these genes suggest their involvement in the loss of epithelial integrity.

The early downregulation of genes encoding claudins and other proteins involved in cell adhesion (cluster 4) was accompanied by the upregulation of genes associated with similar functions (e.g., microtubule organization, cell spreading, and cell adhesion) in cluster 8. However, at 12 dpi, the levels of expression of most of these genes had returned to nearly baseline levels, while the downregulation of cluster 4 persisted. Cluster 8 contained genes encoding the integrins ITFG2 and ITGA1 and genes encoding ligands of integrin, collagen (COL11A1, COL12A1, COL24A1), and laminin (LAMA2, LAMA4, LAMB1, LAMC1). Several other genes in cluster 8 can affect the structure of the cytoskeleton, including caveolin (CAV1, CAV2) and cofilin (CFL2).

Coexpression network analysis implicates pathological mechanisms of other gastrointestinal diseases.

To view the internal correlation structure of cluster 4, we organized the transcripts into a coexpression network. This allowed us to identify particularly strong correlations between coding genes and ncRNA to make functional inferences. This also allowed us to identify hub transcripts that shared high correlations with a large number of other transcripts, which would imply that they play a central role in the function of the cluster. The network was constructed by connecting each transcript to the two other transcripts with which it shared the strongest correlations. To identify genes with roles in the loss of cell adhesion, we examined the hubs in the resulting network. Several of the hub genes had known associations with the pathology of colorectal mucosa. The most notable example was ABCB1, a transporter protein that interacts with several drugs and which has been genetically associated with the failure of first-line protease inhibitors in HIV-infected patients (40). Other important hub genes included RPS5, a ribosomal protein associated with colorectal cancer, and RNF186, a ring finger protein that has been genetically associated with ulcerative colitis (UC). The most connected gene was GPRC5A, a G-protein-coupled receptor protein which has been associated with epithelial cell differentiation, followed by EMP1, a tight junction protein (41, 42). The most connected ncRNA was an unannotated transcript, XLOC_045516, connected to four other coding genes and two ncRNAs. The coding genes included a transporter protein (SLC5A10), an actin cytoskeleton reorganizer (VAV1), a signaling protein (PRKAR1), and one uncharacterized protein.

Because a gene (RNF186) associated with UC was a hub gene in cluster 4, we compared the eight coexpressed gene clusters to genes affected by this disease. UC is an inflammatory disorder of the gastrointestinal tract associated with compromised epithelial integrity and bacterial translocation (43). The purpose of this analysis was to examine what coexpressed gene clusters were shared with other pathological transcriptional perturbations of the gastrointestinal tract. We used four data sets comparing patients with UC with healthy controls. A meta-analysis of these data sets produced one set of genes that was upregulated (n = 483) in all four data sets and one set of genes that was downregulated (n = 521) in all data sets. We then tested these two gene sets for enrichment among the coexpressed gene clusters. We found that the upregulated genes were associated with cluster 1 (P = 1.12 × 10⁻¹⁹), cluster 2 (P = 0.0459), cluster 3 (P = 8.59 × 10⁻¹⁶), cluster 6 (P = 8.30 × 10⁻¹⁰), and cluster 8 (P = 3.48 × 10⁻⁵). In contrast, the downregulated genes were enriched only in cluster 4 (P = 0.0384; Fig. 3B).

In short, the coexpression analysis showed that genes encoding cell adhesion proteins were highly overrepresented among downregulated genes at 3 dpi, and many of these genes are also downregulated in other pathological states associated with intestinal epithelial damage.

Immunohistochemistry and quantitative RT-PCR confirm the downregulation of tight junction genes.

As a validation of the downregulated coexpression network, we examined the differential expression of 10 transcripts, protein coding as well as noncoding, from the network using quantitative RT-PCR. Candidates were selected from among the genes that showed the strongest downregulation and occupied important positions in the coexpression network (determined by their number of interactions). The candidates included tight junction genes CLDN3, CLDN4, and EMP1, the extracellular matrix gene LAMC2, and the network hub gene GPRC5A. We also included four known noncoding RNAs and one unannotated noncoding RNA. In most cases, the selected RNA showed downregulation in all animals at both time points (Fig. 4A).

FIG 4.

Quantitative RT-PCR and immunohistochemistry analysis of downregulated cell adhesion genes in rectal mucosa. (A) Validation of mRNA-seq results using quantitative RT-PCR. Ten RNAs, five coding RNAs, four known noncoding RNAs, and one unannotated noncoding RNA were analyzed. (B) (First panel) Quantification of CLDN3 expression, measured in number of positive pixels/μm². CLDN3 showed strong downregulation at 3 dpi in two animals compared to its expression in the uninfected controls. At 12 dpi, CLDN3 was strongly suppressed in all four animals. (Second to fourth panels) Representative slides showing CLDN3 (stained brown) in rectal tissue samples obtained at necropsy from unchallenged animals, an animal that showed downregulation of CLDN3 at 3 dpi, and an animal that showed downregulation of CLDN3 at 12 dpi, respectively.

In the coexpression network, CLDN3 shared interactions with other cell adhesion proteins, including CLDN4 and PDLIM2. These genes were significantly downregulated compared to the level of expression at the baseline at 12 dpi and, to a lesser degree, at 3 dpi as well. To examine whether the trends in the mRNA levels were indicative of protein levels, we performed immunohistochemistry analysis of the tight junction protein CLDN3. This is a major tight junction protein (44) that has been implicated in the loss of mucosal epithelial integrity in several gastrointestinal diseases, including colorectal cancer (45), celiac disease (46), and chronic SIV infection (21).

The quantification of the CLDN3 protein showed a strong decrease in two animals at 3 dpi compared to the amount in tissue from uninfected animals. By 12 dpi, the level of CLDN3 protein had significantly decreased in all animals (P = 0.00327; Fig. 4B, first panel). The variation in CLDN3 protein expression between individual animals could indicate that the 3-dpi time point represents a transition period, during which the tissue is undergoing the first changes that lead to subsequent loss of tissue integrity. During this time, differences in response kinetics could cause a high degree of heterogeneity between individuals, while the strong antiviral immune response at peak viremia caused a consistent downregulation.

In uninfected animals, CLDN3 was primarily expressed in the mucosal surface facing the intestinal lumen and surrounding the crypts. In the two animals that showed a loss of CLDN3 protein expression at 3 dpi, as well as at 12 dpi, CLDN3 expression was diminished in all these locations (Fig. 4B, second to fourth panels). Thus, the contribution of CLDN3 and possibly other downregulated tight junction proteins to tissue integrity can be lost mere days after SIV infection.

Taken together, these results confirm that the earliest mucosal response to SIV infection includes downregulation of several genes with important roles in cell adhesion and tissue integrity. This downregulation is observable in both RNA levels and protein levels at as early as 3 dpi.

DISCUSSION

Here, we describe the first whole-genome transcriptional profiling of rectal mucosa from SIV-infected nonhuman primates. By focusing on early time points after mucosal challenge, we aimed to characterize mechanisms that contribute to viral spread from the inoculation site. In short, we found that at 3 dpi, corresponding to the eclipse phase, there was a strong transcriptional response at the site of inoculation. This response did not show a significant association with immune activation. Instead, it was predominantly associated with cytoskeleton reorganization, cell morphology, and cell adhesion. In particular, we observed a downregulation of genes encoding claudins, a class of tight junction proteins. At peak viremia, we observed a strong inflammatory response involving both innate and adaptive immune functions and antibacterial functions, in addition to antiviral functions.

The first challenge presented to the virus is crossing the mucosal epithelium into the lamina propria, where it gains access to CD4⁺ target cells. Any circumstance that causes damage to the epithelium can elevate the risk of SIV/HIV transmission, including minor wounds caused by sexual intercourse and preexisting inflammation (47). The rate of sexual transmission is normally low, which implies that undamaged mucosal epithelium is an effective barrier to HIV infection. HIV-positive patients and SIV-infected NHPs show significant damage to the mucosal epithelium, which leads to the translocation of bacterial antigens from the gastrointestinal tract to the lamina propria, further enhancing protracted inflammation and immune cell depletion and driving increased epithelial damage (48). Yet, it has not been determined whether this degenerative cycle is first triggered by the antiviral inflammatory response or by an alternative pathway. Here, we found differential expression of cytoskeletal rearrangement, cell adhesion, and, in particular, downregulation of tight junction proteins before immune activation. This study supports the hypothesis that the virus affects epithelial integrity directly within days of mucosal challenge.

Taken together, the findings of this study suggest a scenario whereby a low number of virions cross the thin rectal epithelium, giving them access to CD4⁺ target cells in the mucosa. This triggers morphological and structural changes in the epithelium within days of infection, which includes the loss of cell adhesion by the downregulation of genes encoding tight junction proteins, in particular, claudins. The coexpression network describing the downregulated genes associated with cell adhesion and tight junctions showed that the central genes in this process are associated with other pathologies of the intestinal mucosa. Specifically, the most interconnected hub genes in the network were associated with the failure of protease inhibitors in HIV, colorectal cancer, and ulcerative colitis.

Ulcerative colitis shares a number of disease mechanisms with SIV infection: inflammation in the intestinal mucosa leading to compromised epithelial integrity and bacterial translocation (43). This prompted us to compare genes that are affected by UC with genes that are affected by acute SIV infection. By this analysis, we found that expression changes in UC were primarily correlated with the downregulation of cell adhesion and tight junction proteins at 3 and 12 dpi, in addition to immune activation at 12 dpi.

While a separate set of genes associated with tissue integrity is activated at 3 dpi, even these cell adhesion molecules are downregulated at peak viremia, and when immune cells migrate to the site of infection, the tissue has already developed increased permeability. Thus, tissue inflammation facilitates viral spread by driving additional inflammatory activation and the recruitment of additional target cells. This scenario suggests that if the downregulated expression of claudins and other tight junction proteins could be prevented, the host would stand a better chance of restricting the virus to the primary infection compartment (and adjacent tissues) for a longer period of time. The potential benefit of delaying SIV infection is supported by previous studies that implicated slow CD8⁺ lymphocyte (6) and type I interferon (20, 49) responses in the loss of HIV/SIV control by failing to manifest itself when the virus is the most vulnerable. Maintenance of mucosal integrity would provide additional opportunity for these antiviral mechanisms to come into play and increase the chance that the virus would be eliminated before it could infect enough host cells to spread systemically (Fig. 5).

FIG 5.

Hypothetical description of events leading up to systemic SIV infection. The events described in this study (red line) begin with the downregulation of genes maintaining epithelial integrity in mucosal tissues within days of exposure to SIV. This response is accompanied by a low rate of viral replication in rectal mucosa. When immune activation takes place, the rectal mucosa is already compromised and tissue integrity is quickly lost, leading to microbial translocation, increased tissue damage, recruitment of SIV target cells, and accelerated SIV replication. This process allows the virus to infect a large number of CD4⁺ cells, leading to systemic infection. This suggests that if expression of cell adhesion proteins (and mucosal integrity) could be maintained before inflammation is triggered (blue line), tissue damage, microbial translocation, and recruitment of SIV target cells could be limited. Thereby, viral replication would be slower, delaying and perhaps allowing the prevention of systemic infection.

Aside from its role in epithelial tissue structure, the cytoskeleton plays an important role in intracellular events required for immunodeficiency virus infection; studies of HIV have shown that during engagement of HIV gp120 with the CD4 receptor, actin filaments participate in recruitment of coreceptors CCR5 and CXCR4, used by HIV for cellular entry (50–52). However, this local assimilation of polymerized actin can stabilize the plasma membrane and inhibit viral entry by endocytosis (53). To overcome this, HIV gp120 engagement with CXCR4 can induce cofilin to sever actin polymers, enabling invagination of the plasma membrane and endocytosis (53). The fact that alterations to the cytoskeleton can be both beneficial and detrimental to the virus could explain why disruption of the cytoskeleton during HIV infection in vitro has been shown to promote (54) or inhibit (50) viral infection. While these previous studies were carried out primarily in CD4⁺ cells, the strong enrichment in cytoskeletal functions in our study suggests that similar processes are activated by SIV in other cell types in the rectal mucosa, such as stromal cells, smooth muscle cells, and adipocytes. Indeed, the transcriptional regulation of genes with cytoskeletal functions by HIV proteins has also been reported in DCs (55) and macrophages (56).

Together, our results emphasize the importance of mucosal integrity to control HIV/SIV infection. Through the downregulation of adhesion proteins, events leading to systemic infection take place within days of infection. The lack of immune activation at this time point suggests that future studies of vaccines and interventions should also focus on noninflammatory mechanisms that could reduce viral replication rates in mucosal tissues. While this study highlights the roles of cell morphology and cell adhesion mechanisms in the loss of viral restriction, other mechanisms could be discovered by examining additional and still earlier time points. Furthermore, the insights gained from these studies of hosts in which SIV is pathogenic should be compared to findings for natural SIV hosts, to specifically explore the pathogenic aspects of this process. A previous study of sooty mangabeys chronically infected with SIV did not find signs of damaged epithelial mucosa (21), suggesting that the expression patterns of tight junction and cell adhesion genes would be different during the acute phase.

These questions will be further investigated in future studies, where hosts in which SIV is pathogenic will be compared to hosts in which SIV is not pathogenic. These studies will also include other tissues, including lymph nodes, in addition to earlier time points, starting at day 1.