Single-cell analysis of immune cell transcriptome during HIV-1 infection and therapy

To determine the host immune cell transcriptional changes during HIV-1 infection and ART, we performed scRNA-seq of peripheral blood mononuclear cells (PBMCs) from six untreated chronically HIV-1-infected (untreated), three ART-treated HIV-1-infected (treated) and three HIV-1 seronegative control (seronegative) participants using the 10 × Genomics scRNA-seq platform (Table 1). The HIV-1-infected individuals on ART had undetectable viral load, but were only on ART for less than one year which could allow for residual viral replication compared to individuals on ART for longer periods. In order to normalize the data and to filter out low-quality cells, we utilized the CellRanger aggregation pipeline which normalized for effective sequencing depth of each sample by subsampling the reads to the sample with the lowest depth. Next, we excluded cells that had greater than 20% of the genes that were mitochondrial genes (Additional file 1: Fig. S1A). Finally, to correct for any batch effects we used the Seurat analysis pipeline Multi CCA method to regress out cell–cell variation in gene expression in order to control for technical variation. This resulted in the elimination of 994 cells from the final dataset. Among all individuals, we determined the transcriptomes of 81,297 single cells detecting a median of 951 genes per cell (Fig. 1A; Additional file 1: Table S1).

We first used an unsupervised analysis approach and performed dimensionality reduction using Uniform Manifold Approximation and Projection (UMAP) and graph-based clustering to visualize and identify 23 transcriptionally distinct cell clusters in the total single cell dataset (Fig. 1B). To define broad cell types contained in each cluster we used the SingleR reference-based mapping to assign cell type identities to each cluster using both the Immune cell expression and Monaco Immune databases (Additional file 1: Table S2). In parallel, we determined the pattern of expression of transcripts that define immune cell identity to further classify the cell clusters. We determined the expression of CD14 expressed in monocytes; CD3D, CD4, CD8A expressed in T cell populations; KLRC1 expressed in natural killer (NK) cell populations; and CD79A expressed in B cell populations (Additional file 1: Fig. S2A). We then identified the genes that were uniquely upregulated in each cell cluster, and used the top three upregulated genes in each cluster to generate a heatmap of gene expression across the clusters (Fig. 1D; Additional file 2: Table S3).

Next, we used a supervised analysis approach to identify cell types but also more specific cell states using the Azimuth reference-based mapping pipeline using refence PBMC datasets. This identified naïve, memory, cytotoxic and regulatory populations of both CD4 and CD8 T cells, natural killer (NK) cell, B cells, monocyte, dendritic cells and small populations or platelet and erythroid cells (Fig. 1C). This analysis recapitulated many of the expected immune cell lineages present in PBMCs.

We identified cells by the HIV group and found that cells from HIV-1 infection (treated and untreated) had overlapping populations with the HIV-1 seronegative samples indicating overall similar transcriptomes of certain cellular populations, as well as clusters of cells that displayed a unique transcriptome (Fig. 1D; Additional file 1: Fig. S2). We then determined the fraction of cells in each Azimuth-identified cell type that was had a frequency above 1% of the total cells (Fig. 1E). The most abundant cell types in all groups were CD4 T central memory, CD8 T effector memory and CD14 monocytes. Individuals with HIV-1 infection (treated and untreated) had increased frequency of CD8 T effector memory cells (29.2%, ART HIV-1; 28.6%, Untreated HIV-1) compared to seronegative individuals (14.2%; Fig. 1E). Conversely, HIV-infected individuals had reduced frequencies of CD4 T central memory cells (20.0%, ART HIV-1; 14.0%, Untreated HIV-1) compared to seronegative individuals (27.2%; Fig. 1E). Individuals with treated or untreated HIV-1 infection also had increased frequencies of CD14 (13.2% ART HIV-1; 19.3%, Untreated HIV-1) and CD16 (5.5%, ART HIV-1; 7.2%, Untreated HIV-1) monocyte populations compared to the CD14 (11.5%) and CD16 (3.6%) monocyte populations in seronegative individuals (Fig. 1E). The reduction in CD4 T cell and increase in CD8 T and monocyte populations were also evident when the fraction of cells within each unsupervised defined cluster (Additional file 1: Fig. S2C). Despite ART treatment and no detectable HIV-1 viral load, transcriptionally unique populations of immune cells were still present in the treated group especially among monocytes and T cells (clusters 4, 6, 12, 16), indicating that ART did not fully restore the immune cell transcriptional landscape after infection (Figs. 1E; Additional file 1: Fig. S2). Cluster 6 represented over 20% of the cells in the treated group and contained a population of regulatory T cells. Clusters 12 and 16 were also nearly exclusively made up of cells in the treated group and represented populations of innate like cells (ILC) and CD16 monocytes, respectively (Additional file 1: Fig. S2C; Additional file 2: Table S3). These data revealed that HIV infection perturbed immune cell identity and that short-term treatment to undetectable HIV-1 viral load did not fully restore the cellular phenotypes.

We next determined the genes that were differentially expressed in cells from the untreated HIV-1-infected individuals compared with the seronegative individuals. We found 96 genes that were upregulated and 45 genes that were downregulated in the cells from untreated HIV-1-infected individuals (Fig. 2A; Additional file 3: Table S4). The top seven upregulated genes are all associated with interferon response (S100A8/9, MT2A, IFITM3, IFI6, ISG15) [17], while the top downregulated gene ZNF90 is a transcription factor with unknown roles in immune cells followed by the genes CXCR4, IL7R and BTG1, which are key regulators of lymphocyte proliferation and motility.

Next, we identified transcripts that were differentially expressed in cells from HIV-1-treated individuals compared with the HIV-1 seronegative individuals to determine genes that are changed during HIV-1 infection despite ART control of viral load. We found 83 genes upregulated and 31 genes downregulated in cells from the HIV-1-treated individuals compared to seronegative individuals (Fig. 2B; Additional file 4: Table S5). CCL4 was the highest upregulated transcript in cells from treated individuals. The CCL4 gene encodes the protein MIP-1b that can bind HIV-1 co-receptor CCR5 and inhibit infection [18]. The next three upregulated genes in treated individuals were DNAJB1, which is a part of the HSP40 complex that has been shown to play a role in anti-inflammatory processes in autoimmunity [19]; and RGS1 and SOD1, which are important for antiviral immune responses [20]. The top downregulated transcripts in cells from treated individuals were the transcription factor ZNF90; NHSL2, which has unknown function; and IL7R, which has been shown to play a role in CD4 T cell loss during HIV-1 infection [21].

We performed Gene Set Enrichment Analysis (GSEA) using the differentially expressed transcripts in untreated HIV-1 or ART-treated HIV-1 compared to seronegative individuals and plotted the hallmark pathways that were significantly enriched (Fig. 2C, D). There were 10 hallmarks that had significant FDR-corrected q-values for the untreated HIV-1 compared to seronegative controls and 15 hallmarks that were significant in the ART-treated individuals (Additional file 1: Table S6 and S7). Both gene sets were enriched for pathways involved in inflammatory responses, complement immune responses and interferon responses (Fig. 2C, D). There were also several enriched pathways involved in metabolism and cell death. The Hallmark Inflammatory Response was significantly enriched in both HIV-1-infected groups (Fig. 2C, D). These data demonstrated that genes associated with the interferon response are upregulated in cells during HIV-1-infection, and key immune genes important for inflammatory processes and controlling virus replication, such as CCL4, remain changed even during ART therapy.

We selected and performed graph-based clustering of cells that identified as T cells and NK cells from the larger dataset to determine higher resolution differences in cellular clusters (Clusters 0, 1, 2, 3, 4, 6, 10, 12, 13, 14, 19, 21; Fig. 1). We identified 17 new distinct transcriptional clusters within this subset of cells (Fig. 3A). We next determined the genes that were upregulated within each cluster of cells, and generated a heatmap of the top five genes that defined each cluster to further classify the identity of each cell cluster (Additional file 1: Fig. S3A; Additional file 5: Table S8). We also determined the expression of nine specific lineage-defining transcripts in order to further classify the cell clusters (Additional file 1: Fig. S3B). CD3D encodes for CD3 which identified T cell populations and the T cells were further subdivided into CD4+ and CD8+ T cells using CD4 and CD8A expression, respectively. This analysis identified a single major cluster of CD4 T cells, but multiple CD8 T cell clusters that were transcriptionally distinct (Additional file 1: Fig. S3B). We found that the T cells were segregated by naïve T cell marker CCR7 as well as a marker of T cell activation MKI67. There were also smaller populations of T cells with the regulatory marker FOXP3 and exhaustion marker PD-1 (PDCD1; Additional file 1: Fig. S3B). We also identified two major clusters of cells that lacked T cell markers (cluster 11 and 13). Cluster 11 was further classified as NK cells due to expression of granzyme B and CD16 transcripts (GZMB and FCER1G) and cluster 13 had markers of monocyte populations (CD16, FCER1G; lysozyme, LYZ; Additional file 1: Fig. S3). In parallel, we utilized the reference-based cell identify mapping approach Azimuth to identify cell populations. This revealed similar cellular populations of CD4 and CD8 naïve and memory populations, T regulatory cells, and NK cells on the major group of single-cells on the UMAP. There were also smaller segregated populations of proliferating cell types, and platelets, B cells, monocytes and dendritic cells (Fig. 3B).

Cells from untreated HIV-1 and seronegative individuals had many overlapping clusters, while the cells from the ART-treated individuals had more unique clusters of T cells (Fig. 3C). Similar to what was observed when analyzing the entire PBMC dataset, there were smaller frequencies of CD4 TCM and increased frequencies of CD8 TEM cell populations in the cells from the untreated or treated HIV-1 infected individuals compared to the cells from the seronegative individuals (Fig. 3D). There was a shift in the T cell populations for ART-treated HIV-1 individuals that were made up of clusters 4, 5 and 10 and represented mostly CD4 cytotoxic T lymphocytes (CTL; Fig. 3A–D). These cells represented 6.14% of the NK/T cells from ART-treated individuals compared to 2.46% and 1.67% of the untreated HIV-1 and seronegative groups, respectively (Fig. 3D). Notably, these clusters had upregulation of two genes that have been shown to play a role in the development of regulatory T cells NFKBIA and FTH1 [22] (Additional file 1: Fig. S3A).

Next we performed differential gene expression comparisons of the CD4+ and CD8+ T cell populations. CD4+ and CD8+ T cells (CD4 and CD8A-expressing cells, respectively) from untreated HIV-1 individuals had upregulation of interferon response genes (IFI6, MT2A, S100A8, S100A9) and downregulation of the transcription factor ZNF90 and IL7R that encodes CD127, important for T cell differentiation [23], when compared to cells from seronegative donors (Additional file 6: Table S9 and Additional file 7: Table S10). When comparing differentially expressed genes in cells from ART individuals and seronegative, we found that CD4+ T cells had upregulation of chemokines and IL1B, which contribute to the proinflammatory state (Fig. 3E; Additional file 8: Table S11). In CD8+ T cells from untreated HIV-1-infected compared to seronegative individuals, there is upregulation of LGALS1 that encodes galectin-1 that has been shown to induce apoptosis in effector T cells and increase immunoregulatory roles of T cells [24] (Fig. 3F; Additional file 9: Table S12). CD8+ T cells had upregulation of DUSP2 and RGS1 that are important for T cell inflammation and cell migration [25] (Fig. 3F; Additional file 9: Table S12). We observed a similar downregulation of ZNF90 and IL7R when we compared ART to seronegative cells as when we compared untreated HIV cells were compared to seronegative, indicating these transcripts are changed in T cells during untreated or treated HIV-1 infection (Fig. 3E, F). These data demonstrated that CD4+ and CD8+ T cells remain in a proinflammatory state despite ART.

To identify genes that are critical for regulating CD4+ T cell resistance or sensitivity to initial HIV-1 infection, we used scRNA-seq in a primary CD4+ T cell model of HIV-1 infection. CD4+ T cells were enriched from the PBMCs of two healthy HIV-1 uninfected donors and stimulated for 72 h with recombinant human IL-2 and antibodies directed against CD3 and CD28. After stimulation, the cells were infected with a full-length primary HIV-1 infectious molecular clone (CH058; [26]) for 24 h. Cells were harvested and subjected to scRNA-seq (Fig. 4A). We analyzed 7359 single T cells from the two donors after filtering for poor quality cells (Additional file 1: Fig. S1B). The cells from both donors clustered similarly using UMAP (Fig. 4B). Annotation of reference-based cell identities using Azimuth revealed that most cells were CD4 proliferating cells, with small populations of cells with CD4 TCM identity (Fig. 4C). There were also some cells that had gene expression profiles that matched CD8 naïve, CD8 and NK proliferating cells (Fig. 4C). Expression of HIV-1 gag transcripts was detected in 2005 of the cells, indicating that after 24 h of virus exposure approximately 27% of the cells had active viral transcription (Fig. 4D). In addition to HIV-1 gag transcript, we identified 12 genes differentially expressed in cells with HIV-1 transcripts compared with HIV-1 negative cells (Fig. 4E, F). Two long-noncoding RNAs (NEAT1 and MALAT1) were upregulated as were other markers of T cell activation and interferon signaling (CD69, CCL3, TNFRSF18, GZMB, GZMA, PRDM1, IFNG; Fig. 4E, F). The three downregulated transcripts had variable function in cell metabolism, transcription and immune signaling (HSP90AA1, EIF5A, CYP1B1; Fig. 4F). Notably, the top three genes that correlated (Pearson) with HIV-1 gag expression levels were the two long-noncoding RNAs NEAT1 and MALAT1, and PRDM1, which encodes the transcriptional regulator BLIMP1 (Fig. 4G; Additional file 10: Table S13). All three of these genes have been implicated in controlling HIV-1 virus transcription and latency [27‐29]. Among the genes that anti-correlated with gag expression were members of the SR protein splicing factor family (SRSF2, SRSF3, SRSF7) that have been shown to be important for HIV-1 RNA splicing [30] (Fig. 4G; Additional file 10: Table S13). These data show that the two long-noncoding RNAs NEAT1 and MALAT1, SR protein splicing factors, and T cell activation status are critical for acute HIV-1 infection and viral gene transcription. Future studies will be required to determine if manipulation of these genes affects sensitivity to HIV-1 infection will be required.

Deidentified peripheral blood mononuclear cell (PBMC) samples were used from existing cohorts of HIV-1-infected and seronegative individuals stored at the Duke Human Vaccine Institute that were approved by the Duke Medicine Institutional Review Boards as well as the ethics boards of the local sites. Individuals were mostly male and of African descent (Table 1). These individuals were a part of a large HIV-1 clinical study that received longitudinal tissue sample collection. Whole venous blood was drawn and then fractionated into PBMCs and cryopreserved. This was an exploratory study, and the selected participants were not statistically matched for external factors such as age, sex, viral load or ethnicity. Future studies should be performed to confirm these cofounders. Cells from six untreated HIV-infected, three ART-treated HIV-1-infected (HIV-1 viral load below the limit of detection), and three HIV-1 seronegative (control) individuals were used to perform single-cell RNA sequencing studies (10× Genomics; Table 1).

PBMCs from two different HIV-1 seronegative individuals (not included in other analysis) were thawed and T cells were activated by stimulation with CD3 (clone OKT3; eBioscience, 150 ng/mL) and CD28 (clone CD28.2; BD Biosciences, 150 ng/mL) antibodies in RPMI media supplemented with 20% FBS and recombinant human IL-2 (NIH AIDS Reagent Program, Division of AIDS, NIAID, National Institutes of Health, 30 U/mL). Activation was allowed to proceed for 72 h at 37 °C 5% CO₂. CD4+ T cells were then enriched by removal of CD8+ T cells using magnetic CD8 microbeads (Miltenyi Biotec, Germany). Cells were then infected with full-length HIV-1 infectious molecular clone virus derived from an infected individual CH058 [26] by spinoculation at 1200×g for 2 h at room temperature [42]. Infected cells were incubated in RPMI media supplemented with 20% FBS and 30 U/mL IL-2 for 24 h. Cells were then harvested and washed with PBS for scRNA-seq.

PBMCs were thawed, washed and placed in single-cell suspensions with PBS + 0.04% BSA. Single-cells after activation and 24 h infection from the primary cell model of HIV-1 infection were also washed and also placed in single-cell suspension with PBS + 0.04% BSA. Cellular suspensions were loaded on a GemCode Single-Cell instrument (10× Genomics) to generate single-cell beads in emulsion. Single-cell RNA-seq libraries were then prepared using a GemCode Single Cell 3’ Gel bead and library kit version 2 (10× Genomics). Single-cell barcoded cDNA libraries were quantified by quantitative PCR (Kappa Biosystems) and sequenced on an Illumina NextSeq 500 [9, 14, 43, 44]. Read lengths were 26 bp for read 1, 8 bp i7 index, and 98 bp read 2. Cells were sequenced to greater than 50,000 reads per cell.

After sequencing, the Cell Ranger Single Cell Software Suite (version 2.1.1) was used to generate sequencing fastq files and to perform sample de-multiplexing, barcode processing, reference alignment and single-cell 3’ gene counting [45]. Reads were aligned to the human genome (GRCH38) combined with the consensus sequence of HIV-1 clade C gag derived from viruses in the HIV Sequence Database (https://​www.​hiv.​lanl.​gov). Clade C was selected due to it being the predominant circulating HIV-1 strain in southern Africa where most of the individuals originate. Samples were aggregated using the CellRanger Aggr function to create a single matrix of cell barcodes and gene counts for the groups. During the process each library is normalized for mapped sequencing depth. Reads are subsampled from higher-depth libraries until they all have an equal number of reads per cell that are confidently mapped to the transcriptome in order to control for variation in the number of reads per sample (sequencing depth). Finally, to correct for any batch effects we used the Seurat analysis pipeline Multi CCA method to regress out cell–cell variation in gene expression in order to control for technical variation. The union of variable genes across all individual samples are utilized to renormalize the data.

Matrices of cell barcodes and gene counts generated by Cell Ranger were loaded into Seurat R package (v3.2.3) for graph-based cell clustering, dimensionality reduction and data visualization [46‐48]. We filtered low quality cells that had lower than 200 expressed transcripts and percentage of mitochondrial genes expressed greater than 20% and for the primary cell model we reduced this threshold to greater than 10% mitochondrial genes. We included up to 45 PCA dimensions for the PBMCs and 48 PCA for the primary cell model for downstream graph-based clustering and UMAP visualization. All other parameters we followed the default Seurat recommendations. SingleR (v. 1.0.5) was utilized to assist with immune cell type identification [49]. Supervised identification of cell identity and states were performed using Azimuth [50]. Cells were annotated using the RunAzimuth, a Seurat object was provided as a query, function available on azimuth package in R using default settings. The Human-PBMC dataset was used as the reference dataset for Azimuth [48]. The reported cell type annotations are based on “celltype.I2” Azimuth output. Differentially expressed genes between cell clusters or groups were determined using Seurat by the Wilcoxon rank sum test. Genes that correlated with HIV-1 transcript expression were calculated by Pearson correlation and corrected for multiple comparisons using Bonferroni. Graphs and plots were generated using the Seurat and ggplot2 (v3.3.3) R packages and Graphpad Prism version 8.

GSEA v4.1.0 was used for gene set enrichment analysis. For the top 2000 variable genes, raw counts were linearly transformed (using ScaleData function available on Seurat Package) and averaged to generate expression dataset. Next, phenotype file was prepared based on group information for the samples. Hallmark gene sets (v7.2) and Human_HGNV_ID_MiSigDB (v7.2) chip hosted on the GSEA-MSigDB file servers were used as Gene sets and Chip annotation file respectively. Genes weren’t collapsed and “gene_set” parameter was used for permutation. While rest of the options were run with default settings.