TGF-β mediates epigenetic control of innate antiviral responses and SIV reservoir size

Abstract

Immunotherapeutic approaches to eliminate latently HIV-infected cells are focused on the adaptive immune system. Herein we provide mechanistic evidence for a molecular cascade characterized by epigenetic reprogramming of innate myeloid cells and CD4 T cells. The coordinate regulation and gene expression mediated by transcription factors (TFs) IRF3, IRF7, STAT1 and C/EBPβ versus AP-1, promoted the development of innate antiviral immunity in these cells which was associated with control of viral load and decay of cell associated viral DNA (CA-vDNA) following analytical treatment interruption (ATI) in SIV-infected rhesus macaques (RMs) treated with anti-IL-10 and anti-PD-1. The prevalence of TGF-β/SMAD signaling in a subset of combo-treated RMs with high CA-vDNA (CA-vDNA^hi) suppressed this antiviral activity through histone deacetylases, including HDAC11, as the latter reduced chromatin accessibility of IRFs and STATs and impeded their antiviral functions. The addition of HDAC inhibitors in vitro restored antiviral response in the presence of TGF-β. Induction of IL-6, a target gene of C/EBPβ, in CA-vDNA^lo RMs, amplified the antiviral network through IRF9, a transcription factor upstream of IRF7. We identified a similar molecular cascade in HIV elite controllers, who maintain low to undetectable viremia and small viral reservoirs without treatment. These data highlight the importance of epigenetic regulation of the host in shaping innate antiviral immune responses that control viral rebound following ATI and reduce the viral reservoir, providing insight into potential strategies for HIV cure interventions.

Fig 1.. IFN signaling is the major pathway induced in PBMCs and LNMCs of combo-treated RMs pre-ATI and is inversely correlated with virological readouts 24 weeks post-ATI.

A. GSEA was used to identify pathways (MSigDB, Hallmark) that significantly (p ≤ 0.05) differentiated combo-treated RMs from control RMs or RMs treated only with aIL-10 pre-ATI in PBMCs and LNMCs. Heatmap illustrates NES ranging from high (red) to low (blue). Rows represent gene-sets and columns represent contrasts observed in PBMCs and LNMCs for pathways of the IFN module. B. Network inference (Cytoscape, Genemania) was used to plot the common IFN significant LEGs (p < 0.05) of pathways in Fig 1A that were up-regulated in PBMCs and LNMCs of combo-treated RMs (red circular nodes). Connecting lines highlight the type of gene-gene interactions, color-coded by network type. C. Heatmap illustrates SLEA scores for significant (p < 0.05) Reactome/Hallmark IFN signaling pathways and RFs pre-ATI in LNMCs from control (black), aIL-10 only (red) and combo-treated RMs (blue). Rows represent pathways and columns represent individual samples. A red-white-blue gradient was used to depict the relative SLEA scores of the pathways, where blue represents low and red represents high relative levels of expression. The levels of CA-vRNA, CA-vDNA and 2-LTR circles (white to blue, green, purple, respectively) 24 weeks post-ATI were plotted as annotations at the top of the heatmap. Note the higher enrichment of IFN pathways in a subset of combo-treated RMs with low SIV CA-vDNA levels (boxed). D. Left panel: heatmap illustrating results of regression analysis of gene expression (top 50 genes) in LNMCs as a function of viral outcomes, i.e. log SIV CA-vDNA (green), log CA-vRNA (blue), and log 2-LTR circles (purple) 24 weeks post-ATI (p < 0.05). Rows represent genes and columns represent individual RMs: controls (black), aIL-10 only (red), and combo-treated (blue). Gene expression levels were centered on zero and ranged to a standard deviation of 1 (z-score). A blue-white-red gradient was used to depict relative gene expression, where blue represents low expression and red high expression. Right panel: network representing the Gene Ontology functional annotation of the top IFN-stimulated genes with antiviral capacity that were negatively correlated (NES = −1.44, p = 0.021) to outcomes at week 24 post-ATI. Blue nodes represent inverse correlation with virological outcomes. E. IFN and antiviral pathways (p ≤ 0.05) persistently upregulated in CA-vDNA^lo RMs pre- and postATI. Heatmap illustrates NES for IFN related pathways. Rows represent gene sets and columns represent comparisons at both pre and post ATI timepoints in LMNCs. F. LEGs up-regulated at weeks 0 and 24 in CA-vDNA^lo RMs are represented in the network. G. Network inference (Cytoscape, Genemania) was used to plot the common IFN-associated genes that were up-regulated in PBMCs and LNMCs (significant LEGs). Red circular nodes represent up-regulated genes in combo-treated RMs. Lines highlight gene-gene interactions. H. Experimental design. Cells were pre-treated or not with IFN-β, then divided into 2 pools. Unlabeled CD4⁺ T cells were infected with HIV by spinoculation. Uninfected cells were labeled with CTV. On day 2, unlabeled HIV-infected cells and CTV-labeled HIV-uninfected cells were mixed (1:1 ratio) and cocultured for 4 days in the presence of saquinavir. I. Frequencies of cells positive for HIV p24 were measured by flow cytometry on day 6 in CTV⁻ and CTV⁺ cells. Gray bars: untreated conditions; blue bars: IFN-β-treated conditions. Circles: HIV-infected CD4⁺ T cells; triangles: CTV-labeled bystander CD4⁺ T cells. *p < 0.05, **p < 0.005, paired t-test. 2-LTR circles: 2 long terminal repeat circles; ATI: analytical treatment interruption; CA-vDNA: cell associated viral DNA; CA-vRNA: cell associated viral RNA; CTV: cell trace violet.; IFN: interferon; GSEA: gene set enrichment analysis; LEGs: leading-edge genes; LNMCs: lymph node mononuclear cells; NES: normalized enrichment score; PBMCs: peripheral blood mononuclear cells; RFs: HIV restriction factors; RMs: Rhesus macaques; SIV: simian immunodeficiency virus; SLEA: sample level enrichment analysis

Fig 2.. Transcriptomic analysis of PBMCs and LNMCs of combo-treated RMs with lower CA-vDNA levels reveals higher IFN-associated restriction factor signaling and lower SMAD signaling and epigenetic signatures.

A. Quantification of TGF-β isoforms in plasma from combo-treated RMs pre-ATI who became CA-vDNA^lo or CA-vDNA^hi 24 weeks post-ATI. Left panel: PCA highlighting significantly different (Wilcoxon test, p < 0.05) levels of plasma cytokines pre-ATI between CA-vDNA^hi and CA-vDNA^lo RMs (24 weeks post-ATI). Right panel: Jitter plots highlighting the different levels of all TGF-β isoforms (i.e., TGF-β1, TGF-β2, TGF-β3), IL-6 and MIP-1α pre-ATI in CA-vDNA^hi and CA-vDNA^lo RMs 24 weeks post-ATI. B. GSEA was used to test the enrichment of pathways (MSigDB c2, c7, Hallmark) that distinguished CA-vDNA^lo from CA-vDNA^hi RMs. The heatmap illustrates NES of pathways related to epigenetic modifications, SMAD2/3 targets, innate antiviral signatures, and inflammation (p < 0.05). The NES scale ranges from low (blue) to high (red). Columns represent relative enrichment in CA-vDNA^hi (left) and CA-vDNA^lo RMs (right) whereas rows represent pathways. C. Network inference (Cytoscape, Genemania) was used to plot the significant LEGs of SMAD targets and epigenetic signatures (both downregulated, blue nodes), and RFs and C/EBPβ target genes (both upregulated, red nodes) pre-ATI in LNMCs from CA-vDNA^hi versus CA-vDNA^lo, respectively. Edges highlight gene-gene interactions. The table highlights the Spearman correlation coefficients and p values between the RF pathways and the other pathways represented in the gene network. D. An integrated model was generated in the R Mixomics package (http://mixomics.org/) using plasma cytokine and bulk RNA-Seq data. This integrated network highlights the Spearman correlation between gene expression (blue circular nodes) and plasma cytokines (blue square nodes). The thickness of connecting lines represents the Spearman correlation coefficient associated with marker pairs. E. Experimental design. CD4⁺ T cells were pre-treated with TGF-β, with or without an anti-TGF-β antibody for 24h, followed by stimulation with IFN-β (24h). CD4⁺ T cells were then infected with HIV by spinoculation. HIV p24 levels were evaluated by flow cytometry 4 days post-infection. F. Bar graph depicting MFI of phosphorylated SMAD2/3 in live CD3⁺ cells from healthy human donors (n = 6) following stimulation with TGF-β and stimulation with TGF-β in presence of anti-TGF-β antibody. G. Bar graph depicting MFI of phosphorylated STAT1 in CD3⁺CD4⁺CD45RA⁻ cells from healthy human donors (n = 6) under unstimulated conditions, stimulation with IFN-β, and pre-stimulation with TGF-β followed by stimulation with IFN-β. H. Bar graph depicting frequencies of p24⁺ cells in cultures of CD3⁺CD4^lo cells under different stimulation conditions (UNS vs TGF-β vs TGF-β + anti-TGF-β antibody). I. Bar graph depicting frequencies of p24⁺ cells in cultures of CD3⁺CD4^lo cells under different stimulation conditions (IFN-β vs TGF-β + IFN-β vs TGF-β + IFN-β + anti-TGF-β antibody). *p < 0.05, **p < 0.005, ***p < 0.0005, ****p < 0.0001, paired t-test. CA-vDNA: cell associated viral DNA; FMO: fluorescence minus one; GSEA: gene set enrichment analysis; IFN: interferon; LEGs: leading-edge genes; MFI: median fluorescence intensity; NES: normalized enrichment score; PCA: principal component analysis; RFs: restriction factors; RMs: Rhesus macaques; UNS: unstimulated.

Fig 3.. HDAC inhibition leads to enhanced interferon signaling.

A. Co-expression network (cytoscape) illustrates a set of ISGs up-regulated 2h post the first dose of vorinostat (day 0 + 2h) and 2h after the seventh dose of vorinostat (day 7 + 2h) compared to baseline (no administration). Red circular nodes indicate significantly up-regulated genes (LIMMA, p < 0.05). Colored edges represent gene-gene interaction. B. Heatmap illustrates results of regression analysis between gene expression measured 2h post-vorinostat administration and viral outcome cell associated HIV RNA (white to black gradient) measured on day 84 (70 days after discontinuation of vorinostat). Rows represent genes (p <0.05) and columns represent individual samples. Gene expression levels were centered at zero (z-score). A blue-white-red gradient was used to depict relative gene expression, where blue represents low relative expression and red high relative expression. C. Experimental design. PBMCs from healthy donors were treated with panobinostat, romidepsin (FK228) or SIS17 (HDAC11 inhibitor) for 2h, and then treated with TGF-β for 24h prior to stimulation with IFN-β (24h). pSMAD2/3, pSTAT-1, and IRF-1 were evaluated by flow cytometry 24h later. D-F. Bar plots with individual data points showing the induction or suppression of IRF1 under various TGF-β, panobinostat, romidepsin (FK228) or SIS17 (HDAC11 inhibitor), and IFN-β, stimulation conditions. * p < 0.05, ** p < 0.005, *** p < 0.0005, **** p < 0.0001, One-way ANOVA, Tukey post-test. IFN: interferon; ISGs: IFN-stimulated genes; LIMMA: linear models for microarray and RNA-Seq data; PBMCs: peripheral blood mononuclear cells.

Fig 4.. Upregulation of IFNs and antiviral ISGs and downregulation of inflammatory pathways in innate and adaptative cells isolated from LNMCs pre-ATI is associated with lower levels of CA-vDNA post-ATI.

A-B. UMAP showing the normalized per cell expression of IFN-γ and antiviral ISGs module scores in lymphoid and myeloid cells sourced from LNMCs, respectively. The antiviral ISG module was calculated using the AddModuleScore function from Seurat and a list of 50 antivirals ISGs sourced from Schoggins JW, Rice CM, Curr Opin Virol 2011. C. UMAP plot with CD4⁺ and CD8⁺ T cell subsets. CD4⁺ and CD8⁺ T cells from LNMCs were re-clustered and manually annotated based on expression of marker genes. D. Overrepresentation analysis of DEGs in T cell subsets between CA-vDNA^lo and CA-vDNA^hi RMs. Red dots represent pathways upregulated while blue dots represent pathways that are downregulated in CA-vDNA^lo RMs. The size of the dots reflects the number of DEGs in each pathway. « EC_vs_nonHIC » pathways include differentially expressed genes from CD4⁺ and CD8⁺ T cells and monocytes of ECs obtained from dos Santos et al. (personal communication). Tirosh et al., pathway is composed of genes from Tirosh et al, 2016. E. Heatmap showing the fold change in levels of expression of DEGs belonging to antiviral pathways (left) and TNF-α signaling pathways (right) in selected CD4⁺ and CD8⁺ T cell subsets upregulated (red) or downregulated (blue) in CA-vDNA^lo RMs. Genes that did not reach the significance threshold (FDR adjusted p value < 0.1) are represented by grey boxes. F. UMAP plot illustrating the expression of the upregulated (red) antiviral ISGs OAS2, DDX60, EIF2AK2, and the overall antiviral ISG module in T cell subsets as described in B. G. UMAP plot depicting the clustering of myeloid cells and DCs sourced from LNMCs obtained from combo-treated RMs pre-ATI, including DCs, FDCs, macrophages, and pDCs. Manual annotation of the clusters was performed based on expression of marker genes and compared to a reference-based annotation with the Azimuth package. H. Overrepresentation analysis of DEGs in myeloid cell subsets and DCs between CA-vDNA^lo and CA-vDNA^hi RMs. Red dots represent pathways upregulated while blue dots represent pathways that are downregulated in combo-treated CA-vDNA^lo RMs. The size of the dots reflects the number of DEGs in each pathway. I. Heatmap showing the fold change in expression of DEGs from antiviral pathways (left) and TNF-α signaling pathways (right) in subsets of myeloid cells and DCs between combo-treated CA-vDNA^lo and CA-vDNA^hi RMs. J. Gene expression of the upregulated (red) antiviral ISGs DDX60, IFI44L, IFI6 and the overall antiviral ISG module in myeloid cells and DCs. The antiviral ISG module was assembled as described under B. ATI: analytical treatment interruption; CA-vDNA: cell-associated viral DNA; CA-vRNA: cell-associated viral RNA; DCs: dendritic cells; DEGs: differentially expressed genes; ECs: elite controllers; FDCs: follicular dendritic cells; FDR: false discovery rate; IFN: interferon; ISG: IFN-stimulated genes; LNMCs: lymph node mononuclear cells; NES: normalized enrichment scores; PBMCs: peripheral blood mononuclear cells; pDCs: plasmacytoid dendritic cells; UMAP: uniform manifold approximation and projection.

Fig 5.. Greater chromatin accessibility of IFN-associated transcription factors (STAT/IRF) binding sites and closed chromatin state of AP-1 family transcription factors binding sites pre-ATI is associated with lower CA-vDNA levels post-ATI.

A. Average difference of TFs motif chromatin accessibility, i.e., motif activity scores, in CD4⁺ and CD8⁺ T cell subsets from LNMCs isolated from CA-vDNA^lo RMs. TFs from the STAT, IRF and AP-1 families are indicated by arrows. Per-cell TFs motif activity was computed with ChromVar. Differentially active motifs were identified by using the Wilcoxon rank sum test on the average differences of scaled TFs motif activity by cell subset and group, with an adjusted p value (FDR) < 0.05. Only the TFs that showed statistical significance in at least two T cell subsets are shown. B. UMAP representation of chromatin accessibility of motif binding sites for SMAD2/3 and FOS/JUN and IRF7 and STAT1 in T cells subsets from CA-vDNA^lo RMs pre-ATI. Lower levels of chromatin accessibility (i.e., SMAD2, FOS/JUN) are in blue and higher levels of accessibility (i.e., IRF7, STAT1) are in red. C. Average differences in TFs activity scores in myeloid cells and FDCs from LNMCs from CA-vDNA^lo RMs pre-ATI. TFs motif activity was calculated and compared between groups as described for the T cell subsets (panel A). The top 5 TFs per subset based on highest absolute fold-change are indicated by arrows. D. UMAP representation of chromatin accessibility of motif-binding sites for IRF4, IRF7, FIS and RFX2 in myeloid cells and FDCs from combo-treated CA-vDNA^lo RMs pre-ATI. Lower levels of chromatin accessibility are in blue and higher levels of accessibility are in red. ATI: analytical treatment interruption; CA-vDNA: cell-associated viral DNA; FDC: follicular dendritic cells; FDR: false discovery rate; LNMCs: lymph node mononuclear cells; RMs: rhesus macaques; UMAP: uniform manifold approximation and projection; TFs: transcription factors.

Fig. 6.. Overlap of gene expression signatures between HIV elite controllers and CA-vDNA^lo RMs.

A. Upregulated pathways in T cell subsets from CA-vDNA^lo RMs and ECs. Larger nodes represent pathways, and the size of the nodes reflects the number of DEGs in each pathway. Shared leading edge genes are also represented (small nodes) and bold labeled nodes indicate genes from ECs that overlap with the DEGs from the CA-vDNA^lo RMs. Targets of C/EBPβ are labeled in red. Colors inside the nodes indicate the cell subset in which the genes or pathways in question are statistically significant. B. Downregulated pathways in T cell subsets from CA-vDNA^lo RMs and ECs. Targets of C/EBPβ are labeled in blue. C. Upregulated pathways in myeloid subsets and FDCs from CA-vDNA^lo RMs and ECs. D. Downregulated pathways in myeloid subsets and FDCs from CA-vDNA^lo RMs and ECs. CA-vDNA: cell-associated viral DNA; DEGs: differentially expressed genes; ECs: elite controllers; FDC: follicular dendritic cells; RMs: rhesus macaques.