HDAC inhibitors directly modulate T cell gene expression and signaling and promote development of effector-exhausted T cells in murine tumors

Abstract

Epigenetic regulation plays a crucial role in the development and progression of cancer, including the regulation of antitumor immunity. The reversible nature of epigenetic modifications offers potential therapeutic avenues for cancer treatment. In particular, Histone deacetylase (HDAC) inhibitors (HDACi) have been shown to promote antitumor T cell immunity by regulating myeloid cell types, enhancing tumor antigen presentation, and increasing expression of chemokines. HDACi are currently being evaluated to determine whether they can increase the response rate of immune checkpoint inhibitors (ICIs) in cancer patients. While the potential direct effect of HDACi on T cells likely impacts antitumor immunity, little is known about how HDAC inhibition alters the transcriptomic profile of T cells. Here, we show that two clinical-stage HDACi profoundly impact gene expression and signaling networks in CD8⁺ and CD4⁺ T cells. Specifically, HDACi promoted T cell effector function by enhancing expression of TNFα and IFNγ and increasing CD8⁺ T cell cytotoxicity. Consistently, in a murine tumor model, HDACi led to enrichment of CD8⁺ T cell subsets with high expression of effector molecules (Prf1, Ifng, Gzmk, Grmb) but also molecules associated with T cell exhaustion (Tox, Pdcd1, Lag3, Havcr2). HDACi further generated a TME dominated by myeloid cells with immune suppressive signatures. These results indicate that HDACi directly and favorably augment T cell effector function but also increase their exhaustion signal in the TME, which may add a layer of complexity for achieving clinical benefit in combination with ICIs.

Introduction

While immune checkpoint inhibitors (ICI) have revolutionized treatment paradigms for many cancer types, the response rates are still relatively low, e.g., 20–25% for unselected non-small cell lung cancer (NSCLC) with single agent ICI anti-PD-1 ^1–4. Multiple preclinical studies have demonstrated that epigenome modulatory agents have the potential to overcome resistance to ICIs ^5–19. Histone deacetylases (HDACs) remove acetyl groups from histones leading to gene silencing, including genes critically involved in immune function. In previous work, we and others have found that HDAC inhibitors (HDACi) increase MHC and T cell chemokine expression in tumor cells leading to enhanced T cell infiltration into tumors and response to ICIs ^{17, 20}. Additionally, HDACi mediated mechanisms that increase tumor immunogenicity include apoptosis of suppressive myeloid cells (MDSCs), direct stimulatory effects on antigen-presenting cells (APCs), and de novo induction of expression of “silenced” tumor antigens ^5–16. Furthermore, epigenetic agents which promote genomic accessibility in a mechanistically distinct manner from HDACi were also shown to promote T cell chemokine expression ^{21, 22}. Specifically, it was shown that enhancer of zeste homologue 2 (EZH2)-mediated histone H3 trimethylation (H3K27me3) and DNA methyltransferase 1 (DNMT1)-mediated DNA methylation repress ovarian tumor cell expression of CXCL9 and CXCL10 ^{21, 22}. Importantly, inhibitors of EZH2, such as DZNep or GSK126, and DNMT1 inhibitor 5-AZA-dC enhance chemokine expression to promote effector T cell trafficking to the tumor and boost the response to PD-L1 blockade immunotherapy ^{21, 22}. Therefore, enhancement of T cell chemokine expression may be a common feature of different epigenome targeting agents. While the above studies indicate that HDACi increase tumor cell susceptibility to T cell killing, the direct effects of HDACi on T cells are not well understood.

Based on promising preclinical findings, multiple clinical trials have been initiated to determine whether epigenetic modulators, in particular HDACi, can increase the response to ICI therapies. While some promising signals of clinical activity have been seen ^23–26, further studies including randomized trials will be needed to determine whether HDACi represent a promising treatment option to combine with ICIs. In our recent studies, the combination of HDACi vorinostat and ICI pembrolizumab in ICI refractory NSCLC patients led to a disease control rate of 67% ²³. With the continuing interest in combining HDACi with T cell activating therapies, it is imperative to investigate the direct effect of HDACi on T cells.

Here, using exome-wide approaches, we determined whether two clinical-stage HDACi (romidepsin and vorinostat) can directly impact T cell gene expression responses. Our results indicate that a short-term treatment with either HDACi profoundly impacts gene expression responses in activated murine CD8⁺ and CD4⁺ T cells. A notable effect was the activation in TNFα and IFNγ signaling pathway activity that was mediated in part by direct upregulation of expression of TNFα and IFNγ. Moreover, our findings revealed that HDAC inhibition downregulated inhibitory markers (Lag3, Tim3) and enhanced CD8⁺ T cell cytotoxicity. Single cell RNA sequencing analysis in a murine tumor model found that HDACi profoundly modulate the TME, often in a manner similar to ICIs. Notably, both HDACi and ICI treated tumors exhibited an enrichment of CD8⁺ T cell populations with a transcriptomic profile of effector and exhausted cells, accompanied by a substantial increase in myeloid subsets with immune suppressive signatures. While these studies indicate that HDACi enhance T cell effector functionality, the increase in exhaustion-associated receptors on T cells and increase in suppressive myeloid cells in the TME suggest that potential immune-suppressive mechanisms may also be induced by HDACi.

Materials and Methods

Mice

All mice were housed in the animal facility at the Moffitt Cancer Center. Wild-type C57BL/6 mice were purchased from Charles River. OT-1 mice were bred within the animal facility. All experiments were approved by the Institutional Animal Care and Use Committee (IACUC).

RNA analysis and Microarray studies

Microarray was performed as recently described ²⁷. Splenic CD4⁺ and CD8⁺ T cells were isolated by negative selection using the MojoSort Mouse CD4⁺ and CD8⁺ T Cell Isolation Kit (Biolegend) following the manufacturer’s instructions. Isolated cells were plated at 2×10⁶ per mL in RPMI 1640 containing 2mM L-glutamine and 25 mM HEPES, supplemented with 10 mM Sodium Pyruvate, nonessential amino acids, 100 U/ml penicillin/streptomycin, 55 μM β-ME, and 10% fetal calf serum (complete media). T cells were stimulated with plate-bound 3ug/ml anti-CD3 and solubilized 5ug/ml anti-CD28 antibodies (eBioscience) for 2 days. Activated T cells were treated with 0.5, 2, 10uM Vorinostat or 2, 10, 30nM Romidepsin for 1 day and harvested for RNA extraction using RNeasy Plant Mini Kit (QIAGEN) following the manufacturer’s instructions.

For gene specific qPCR analysis, activated T cells were treated with Romidepsin (2, 10nM) for either 1, 2, or 3 days before RNA extraction. RNA was reverse transcribed into cDNA using TaqMan Reverse Transcription kit (Applied Bioscience, Cat. No. N8080234) following manufacture’s instructions. cDNA was then utilized to determine the expression of different effector and inhibitory genes (Ifng, Prf1, Gzmk, Tnf, Pdcd1, Tox, Havcr2, Lag3) by qPCR using SYBR Green PCR Master Mix (ThermoFisher, Cat. no. 4309155) and analyzed by QuantStudio3 Real-Time PCR System (Applied Biosystems). Gene expression was normalized to the internal control (S18) and plotted as fold change relative to untreated sample. Gene specific primers were purchased from RealTimePrimers (Elkins Park, PA). Microarray analysis was performed using the Mouse Genome 430 2.0 Arrays. Microarray analysis was initiated with 100 ng of total cellular RNA from each treated condition. RNA was converted to cDNA and then amplified and labeled with biotin using the Ambion Message Amp Premier RNA Amplification Kit (Life Technologies, Grand Island, NY) following the manufacturer’s protocol. The labeled RNA was hybridized to Affymetrix U133 Plus 2.0 microarrays and the staining and scanning of the chips followed the prescribed procedure outlined in the Affymetrix technical manual. Results were analyzed and annotated using the Bioconductor R packages: affy, gcrma and hgu133plus2.db. All microarray data will be deposited in a public database upon acceptance of this manuscript.

Microarray gene set enrichment analysis:

GCRMA-normalized probe intensities were summarized to gene level by taking the summation of multiple probes represent a gene. Mouse hallmark gene sets were downloaded from Mouse MsigDB Collections. Single-sample gene set enrichment analysis was performed using gsva function (R package GSVA²⁸) with parameter method=”ssgsea” and default settings. The scaled enrichment scores were compared across different dosage and time points.

DC-T co-culture studies

Bone marrow from C57BL/6 mice were used to generate DCs as previously reported ²⁹. In subsequent T cell coculture, same sex mice were used to generate DCs and isolate T cells. 2×10⁶ live cells were plated in 10ml complete RPMI media supplemented with 10% fetal bovine serum, penicillin-streptomycin-glutamine, 1M HEPES (Gibco), 1X non-essential amino acids (Corning), 1mM sodium pyruvate (Corning), and 0.11uM 2-mercaptanoethanol (Gibco), and 20ng/ml of rmGM-CSF (Fisher). After 3 days, 10ml of fresh RPMI media was added along with 10ng/ml of rmGM-CSF. At day 6,100ng/ml of LPS (Sigma-Aldrich) was added to the culture to activate DCs overnight. Activated DCs were washed, then resuspended at 1×10⁶ cells/ml with 100ng/ml SIINFEKL peptide (Invivogen) and plated at 1×10⁵ cells/well in a 96-well round-bottom plates. DCs were incubated with SIINFEKL peptide for 3–4 hours and then resuspended in 100ul T cell media. CD8⁺ T cells were isolated from OT-1 mouse spleens by negative selection, using the Mojo Sort Mouse CD8⁺ T cell isolation kit (Biolegend) following manufacturer’s instructions. 1×10⁵ CD8⁺ T cells were plated in the DC wells with either 2nM or 10nM doses of romidepsin (Selleckchem) or DMSO (Fisher) vehicle control, in a total final volume of 200ul per well. Each experimental condition was plated in 5 replicate wells. After 3 days, aliquots of supernatant saved and stored at −80°C for ELISA testing.

DC Fixation:

After 2 hours of SIINFEKL peptide pulsing, DCs were washed with 100ul of cold 0.1% PBS-BSA. Plates were then centrifuged at 800g for 2 minutes at 4°C. This washing step was repeated twice before resuspending DCs in 50ul/well of freshly made 0.008% (vol/vol) PBS-glutaraldehyde (Sigma). DCs and PBS-glutaraldehyde were mixed by pipetting. Plates were then incubated on ice for 5 minutes. Immediately after this, 50ul/well of 0.4M PBS-glycine (Sigma) was added to the wells. Plates were centrifuged at 800g for 2 minutes at 4°C. Wells were then resuspended in 100ul 0.2M PBS-glycine, centrifuged again, then washed twice with 200ul T cell media. Wells were centrifuged one final time before fixed DCs were resuspended in 100ul T cell media.

ELISA Assays

Freshly thawed DC-T cell coculture supernatant was tested using mouse TNFα and mouse IFNγ ELISA kits (Biolegend). 0.5% PBS-Tween20 (Sigma) was used as washing buffer. Supernatants were tested following kit manufacturer instructions. Wells were measured using a Gen5 Microplate reader (BioTek Instruments) within 15 minutes of completing the assay.

KRAS-G12D TP53-null (KP) lung tumor (KPN1) cell-line (KP-OVA) and OT1 T cell coculture for T cell functional assays

CD8⁺ T cells were isolated from OT1 splenocytes by negative selection, using the Mojo Sort Mouse CD8⁺ T cell isolation kit (Biolegend) following manufacturer’s instructions. Isolated T cells were stimulated with plate-bound 3ug/ml anti-CD3 and solubilized 5ug/ml anti-CD28 antibodies (eBioscience) with mouse IL2 (Peprotech, cat no. 212–12) (100U/ml). KP mouse lung tumor cells were pulsed overnight with OVA peptide (0.1ug/ml) and treated with mouse IFNγ (5ng/ml) to upregulate antigen presentation. Pulsed KP tumor cells were then irradiated (10,000 rad) and cocultured with activated OT1 T cells at density 2*10⁵ each/2ml media/well in 24-well plate and treated for 1, 2 or 3 days with Romidepsin (Selleckchem, cat. No. S3020) (10nM), and/or anti-mouse PD-1 (Clone RMP1–14, cat no. BE0146, Bio-X-Cell) (10ug/mL), or its matched isotype IgG (cat no. BE0089, Bio-X-Cell). At the indicated end points, cells were pelleted, and supernatants were collected for ELISA of IFNγ. Next, cells were stained with anti-mouse CD45.2 -BUV395 (BD Bioscience, clone 104, cat no. 454616), CD8-PE (Biolegend, Clone 53–6.7, Cat# 100708), LAG3-BV711 (BD Bioscience, Clone C9B7W, Cat no. 563179), PD1-APC (eBioscience, Clone RMP1–30, cat no. 17–9981-82), TIM3-BV421 (Biolegend, Clone RMT3–23, cat no. 119723) and Live dead Near IR (Invitrogen, cat no. L34976). Flow cytometric analysis was performed on BD FACS Symphony TM and analyzed using FlowJo software (Tree Star). Each experimental condition was plated in 3 replicate wells, and the data shown are representative of 2 independent experimental runs. T cell cytotoxicity was determined after coculture of OVA-peptide pulsed KP target cells with OT1 CD8⁺ T cells by Real Time Cell assay (RTCA) using xCELLigence^® E-plate 96 PET (Agilent, cat no. 00300600910) following manufacture’s instructions. Briefly, 5,000 KPN (unpulsed, or pulsed overnight with OVA peptide) were seeded in E plate overnight, before OT1 CD8⁺ T cells being added at the following ratio (E:T= 1:1 and 1:5) and treated with the indicated conditions. T cell cytotoxicity was assessed using the xCELLigence Real-Time Cell Analyzer (ACEA Biosciences), as measured by the kinetics of electrical impedance over time due to change in cell adherence of KP tumor target cells every 15 minutes until the end of the experiment. The data analysis was performed by the xCELLigence RTCA software package and plotted as normalized cell index.

KP subcutaneous tumor implantation:

KRAS-G12D TP53-null (KP) lung tumor model (KPN1) cell-line was generously provided by Dr. Nikhil Joshi (Yale School of Medicine) ³⁰. KPN1 cells were maintained in complete DMEM media (10% FBS, 1X pen/strep). On the day of injection, cells were washed briefly with 1X PBS and harvested by trypsinization. Cells were then resuspended in phenol red free 1X DMEM media with 2% FBS and counted. 1×10⁶ cells were subcutaneously implanted in the right flank of C57BL/6 mice.

Mouse tumor growth studies:

When tumors reached a measurable size (200mm³), mice were randomized into 4 groups (6 mice each): CTRL, romidepsin, anti-PD1/CTLA4, romidepsin + anti-PD1/CTLA4 groups. Each group received their indicated treatment for 2 weeks: romidepsin (Selleckchem, cat. No. S3020) (intraperitoneal, every 2 days, 2mg/kg), anti-mouse CTLA-4 (Clone UC10–4F10–11, cat no. BE0032, Bio-X-Cell) + anti-mouse PD-1 (Clone RMP1–14, cat no. BE0146, Bio-X-Cell) (intraperitoneal, every 3 days, 200ug each/dose), or their matched isotype IgG (cat no. BE0089 and cat no. BE0091, Bio-X-Cell). Tumor measurements were taken every 3 days with digital caliber and the tumor volume was calculated using the formula length X width² /2.

Single cell RNA sequencing

Tumor processing:

KPN1 tumor-bearing mice were randomized and treated as previously described. Data represents cells from 3 tumors/group pooled together. Before tumor collection, mice were cardiac perfused with 10 ml of PBS/Heparin (10 Unit/mL) to clear peripheral blood. Tumors were minced and digested in digestion buffer (10 ml DMEM media with 1mg/ml Collagenase A (cat no. 11088793001, Roche) + DNase I (50 U/ml) (cat no. 10104159001, Roche), for 20 minutes at 37°C with agitation. Digested tumors were then filtered through a 70-μm cell strainer and reconstituted in 10ml ACK lysis buffer for 2 minutes at room temperature to lyse the red blood cells. Subsequently, single cell suspensions were prepared and stained with anti-CD45.2-FITC (1:100) (Biolegend, clone no. 104, cat. No. 109805). DAPI (1ug/ml) was added prior to sorting. DAPI⁻ (live cells) or DAPI⁻CD45⁺ cells were sorted by Aria SORP sorter in the flow cytometry core at Moffitt Cancer Center. Sorted cells were then washed in 0.04% BSA in 1X PBS buffer, counted, and resuspended in the same buffer at density 1×10⁶ cells/ml as recommended by the cell preparation guide from 10X Genomics. Poly-adenylated mRNA was reverse transcribed inside each droplet at 53°C. cDNA libraries were completed in a single bulk reaction, to generate 50,000 sequencing reads per cell on the Illumina NovaSeq 6000 instrument, following the 10X Genomics Chromium NextGEM Single-Cell 3′ Reagent Kit v3.1 user guide. Single cell RNA seq data were then demultiplexed by 10X Genomics CellRanger v6.1.2 software and analyzed using 10X Genomics Loupe browser v6.0.0.

Single-cell RNA-seq data processing, batch effect correction, and clustering:

Samples were then processed by the Molecular Genomics Core at the Moffitt Cancer Center. Cell viability and concentration was assessed using AO/PI dual fluorescent staining and visualization on the Nexcelom Cellometer K2 (Nexcelom Bioscience LLC). Up to 10,000 cells/sample were encapsulated by the 10X Genomics Chromium Single-Cell Controller. Sequencing reads were mapped against GRCh38 human transcriptome and processed for UMI counting and barcodes filtering using Cell Ranger (v3.0, 10X Genomics), following the procedure outlined in a recent study ³¹. Principal component analysis was performed on the integrated data and a shared nearest neighbor (SNN) graph was constructed based on the first 40 principal components. A total of 28 clusters were identified using Louvain clustering ³² implemented in FindClusters function at resolution=1. Differential expression analysis for each cluster was performed using FindAllMarkers function in Seurat with default settings. Genes with Bonferroni-corrected p-value <0.05 and an average log-fold change > 0.25 were considered differentially expressed. Clusters were further annotated by comparing differential genes with markers previously associated with B cells (Cd79a,Cd19), T cells (Cd3e, Cd3d, Cd3g), NK cells (Klrd1, Nkg7), plasma cells (Jchain, Igkc), epithelial cells (Clu, Krt7, Krt19), Macrophages (Cd68, C1qb, C1qc, Marc1), Monocytes (Lyz2, Vcan, Chil3, Fn1), cDC1 (Xcr1, Clec9a, Itgae, Batf3), cDC2 (Lilrb4a, Itgax, Csf1r, Mgl2), mregDC (Fscn1, Ccl22, Cacnb3, Ccr7, Fabp3), pDC (Siglech, Ccr9, Bst2, Pacsin1, Tcf4), and Neutrophils (Cxcr2, Mmp9, S100a8, S100a9). Uniform manifold approximation and projection (UMAP) was used to visualize gene expression and clusters.

All clusters annotated as myeloid populations were isolated from the complete dataset for further analysis. Cells were reclustered into 16 clusters using SNN-based clustering based on the first 40 principal components with resolution = 1. Differentially expressed genes for each cluster were generated using FindAllmarkers with default setting. Based on DE genes and the expression of known canonical markers, clusters were annotated as 5 Macrophage subpopulations, 5 Monocytes subpopulations, 2 Neutrophil subpopulations, cDC1, cDC2, mregDC, and pDC.

Similarly, T and NK cells were extracted for further subpopulation identification. T cell receptor and immunoglobulin genes were removed from the variable genes to prevent clustering based on V(D)J transcripts. Normalization, data scaling, PCA analysis were re-performed on T and NK cells. A total of 11 clusters were identified using SNN-based clustering based on the first 40 principal components with resolution = 1. Differentially expressed genes for each cluster were generated using FindAllmarkers with default setting. Clusters were annotated by comparing DE genes to markers previously associated with various T cell functions subpopulations, such as T cell stemness/memory (Sell, Tcf7, Il7r, Cd28, Ccr7, Cd27), Treg (Cd4, Foxp3), Tfh (Izumo1r, Slamf6), Th2 (Il1rl1, Gata3), gama-delta T cells (Trdc, Tcrg), CD8⁺ T effector/cytotoxic (Cd8a, Gzmk, Gzmb, Prf1, Ifng), T cell exhaustion/dysfunction (Tox, Lag3, Pdcd1, Havcr2), and proliferation (Stmn1, Cdk1, Mki67). To validate the annotation, enrichment scores of marker genes reported in previous scRNA-seq studies^33–36, including T cells stemness, effector, precursors of exhaustion, exhaustion in CD4⁺T and CD8⁺ T cells, were calculated using AUCell algorithm implemented in SCENIC ³⁷. Marker gene expression was visualized using UMAP projection, violin plots, and bubble plots to display the z-score normalized average expression and percentage of expressing cells per cluster.

Trajectory analysis:

CD8⁺ T cells were extracted for trajectory analysis using Monocle 3 ³⁸. Briefly, a cell data set project was constructed from raw count matrix and cell type information of CD8⁺ T cells obtained above. Trajectory was learned using default setting in learn_graph function. CD8⁺ stem-like T cells were manually selected as the root node in the graphical interface. Pseudotime was measured by ordering cells according to their progress through the developmental trajectory.

Statistical Analysis

All statistical analyses were done using graph-Pad prism software. Student’s t tests were used to determine statistical significance between two experimental groups. Significance was set at p < 0.05. One way ANOVA was used to determine statistical significance between 3 or more experimental groups. Repeated-measures 2-way ANOVA followed by Tukey’s or Dunnett’s multiple-comparison test was used to determine statistical difference in tumor growth studies and OT1/KP-OVA experiments, which involved multiple conditions with different time points. Dunnett’s comparison was employed to compare each group with its untreated control, while Tukey’s method was adopted for comparison among all conditions.

Datasets

Microarray and scRNA-seq data are available in GEO (accession numbers GSE248990 and GSE249002) https://www.ncbi.nlm.nih.gov/geo/.

Results

HDACi exert broad effects on gene expression and signaling pathway activity in CD8⁺ and CD4⁺ T cells

Since T cells in the TME have in general been previously primed, we sought to determine HDACi impact on activated CD8⁺ and CD4⁺ T cells. To this end, we first determined gene expression changes in T cells stimulated with anti-CD3/CD28. We determined dose-dependent effects of 2 structurally distinct clinically used HDACi, romidepsin and vorinostat, on genome-wide mRNA expression in activated CD8⁺ and CD4⁺ T cells. Purified mouse spleen CD4⁺ and CD8⁺ T cells were first activated for 48 hours with anti-CD3/CD28 and then the two HDACi at 3 different concentrations each were added for an additional 24 hours. RNA isolated from unstimulated, anti-CD3/CD28 and anti-CD3/CD28 + HDACi treated cells was subjected to RNA microarray analysis in order to provide a broad assessment of gene expression changes (Fig. 1A). In 2 independent experiments, a total of 265 genes were up-regulated ≥2-fold in CD8⁺ T cells at the median concentration of romidepsin of 10nM and vorinostat of 2uM (common genes) in comparison to anti-CD3/CD28 stimulation alone, while 36 genes were downregulated ≥2-fold. Similarly, in CD4⁺ T cells after romidepsin and vorinostat treatment, a total of 268 genes were up-regulated ≥2-fold, while 254 were downregulated (common genes). These results imply that HDACi directly impact expression of a broad range of genes in CD8⁺ and CD4⁺ T cells.

Figure 1. HDACi exert broad effects on gene expression and signaling pathway activity in CD8⁺ and CD4⁺ T cells.

A. Schematic representation of HDACi treated CD8⁺ and CD4⁺ T cells for microarray. Purified mouse spleen CD4⁺ and CD8⁺ T cells were activated for 48 hours with anti-CD3/CD28 antibodies and HDACi at 3 different concentrations were added for an additional 24 hours. RNA isolated from unstimulated, anti-CD3/CD28 and anti-CD3/CD28 + HDACi treated cells was subjected to RNA microarray analysis to assess gene expression changes in 2 independent experiments. B-F. Gene Set Enrichment Analysis (GSEA) Hallmark pathways analysis of microarray showing differential pathway activation in different doses of HDACi vs. anti-CD3/CD28 stimulated T cells control treatments. Normalized enrichment scores (NES) results are indicated.

We next performed Gene Set Enrichment Analysis (GSEA) to define pathways impacted by HDACi. Amongst pathways relevant to CD8⁺ and CD4⁺ T cell function, one of the top increased pathways by HDACi was the TNFA-NFkB pathway (Fig. 1B). The median dose of both HDACi resulted in the highest activity, which was decreased at the highest dose, potentially due to adverse effects of high HDACi concentrations (Fig. 1B). A similar effect of HDACi was seen in IFNG pathway but primarily in CD8⁺ T cells (Fig. 1C). Another pathway that showed increased activity specifically in CD8⁺ T cells was the Protein Secretion pathway, indicative of greater secretion activity after HDACi treatment (Fig. 1D). Also increased in CD8⁺ T cells was the APOPTOSIS pathway (Fig. 1E), while proliferation-associated pathways, such as G2M checkpoint, were decreased by HDACi in both CD8⁺ and CD4⁺ T cells (Fig. 1F). Notably, both romidepsin and vorinostat had similar impact on these pathways confirming these effects are indeed being mediated by HDAC inhibition. In summary, these results suggest that HDACi directly impact multiple aspects of T cell function that could lead to an increase in effector function but also to a decrease in proliferative capacity.

HDACi promote expression of TNFα and IFNγ effector cytokines in T cells

TNFα and IFNγ secretion comprises a major T cell effector mechanism. We determined whether HDACi-induced increase in TNFA-NFKB and IFNG pathways is mediated by increased expression of TNFα and IFNγ. Indeed, Tnf mRNA expression was elevated after HDACi treatment in CD8⁺ and CD4⁺ T cells, suggesting increase in pathway activity could be mediated by increased expression of TNFα (Supplemental Figure 1A,B). An overall similar effect of HDACi treatment was seen with Ifng mRNA (Supplemental Figure 1C,D). Of note, expression of multiple TNFα and IFNγ regulated chemokines including Ccl4, Ccl5, Cxcl9, and Cxcl10 was increased by HDACi treatment in both CD8⁺ and CD4⁺ T cells (Supplemental Figure 2A,B). These results raise the possibility that HDACi-induced increase in TNFα and IFNγ could modulate the expression of multiple functionally important genes in T cells.

We next determined whether HDACi could increase TNFα and IFNγ secretion in the setting of antigen-induced T cell activation. To test this, we used ovalbumin SIINFEKL peptide-pulsed dendritic cells (DCs) and OT-1 CD8⁺ T cells to specifically test the effect of HDACi on antigen-induced secretion (Fig. 2A). DCs were pulsed with SIINFEKL peptide and plated in a 1:1 ratio with OT-1 CD8⁺ T cells with either 2nM or 10nM romidepsin or DMSO vehicle control for 3 days. ELISA of co-culture supernatant showed that both TNFα and IFNγ secretion was increased with increasing romidepsin concentrations (Fig. 2B–C). This experimental setup utilizes live DCs, which may also be impacted by HDACi. To investigate romidepsin effect specifically on CD8⁺ T cells, we used a DC model wherein peptide-pulsed DCs were first fixed with glutaraldehyde (Fig. 2D). Notably, romidepsin also increased both TNFα and IFNγ secretion by OT-1 T cells in this model (Fig. 2E–F), suggesting that HDACi can directly increase antigen-induced T cell cytokine release.

Figure 2. TNFα and IFNγ secretion by antigen activated CD8+ T cells.

A. Schema of TNFα and IFNγ secretion experimental setup shown in B-C. B. TNFα secretion at indicated doses of Romidepsin (Rom). C. IFNγ secretion at indicated doses of Romidepsin (Rom). D. Schema of TNFα and IFNγ secretion experimental setup using fixed DCs shown in E-F. E. TNFα secretion at indicated doses of Romidepsin (Rom). F. IFNγ secretion at indicated doses of Romidepsin (Rom). Each column represents the mean of three independent replicates +/− SD. Representative results of 1 out of 2 independent experiments are shown. Statistical significance is indicated by p-values or as *p<0.05, **p<0.01, ***p<0.001. NS: not significant.

Broad modulation of the TME by HDACi includes enrichment of effector-exhausted CD8⁺ T cell subsets and multiple myeloid cell populations

A major goal of recent HDACi clinical studies has been to improve benefit from ICIs ^5–20. To provide an in-depth assessment of effect of HDACi and ICI alone and in combination, we used a KRAS-G12D TP53-null (KP) lung tumor model ³⁰. As in previous tumor models used by others and us ^5–20, the combination ICIs and HDACi resulted in a more pronounced anti-tumor effect than the individual treatments (Supplemental Figure 3). We performed single cell RNA sequencing (scRNA-seq) after a 2-week treatment to determine effect on T cells, myeloid cells as well as the overall effect on all detectable CD45⁺ immune cell populations (Fig. 4). Within the CD45⁺ cell population, a total of 2990 T and NK cells were subjected to sc-RNA-seq analysis (Fig. 3A). Compared to untreated controls, T/NK cells from anti-PD-1/anti-CTLA4 treated mice showed several marked changes including a decrease in stem-like CD4⁺ and CD8⁺ T cells and increase in effector-exhausted and early-exhausted CD8⁺ T cells (Fig. 3B–C). A substantial population of Treg were present in controls, which were unchanged after anti-PD-1/anti-CTLA4 treatment but increased in Romidepsin treated tumors (Fig. 3B–C). Populations of γ/δ T cells and NK cells were very small in all treatment groups. Interestingly, the most substantial change after romidepsin treatment in comparison to controls was an increase in effector-exhausted CD8⁺ T cells (Fig. 3B–C). This population had high expression of effector genes (Prf1, Ifng, Gzmk, Grmb) as well as genes associated with a terminal exhaustion state (Tox, Pdcd1, Lag3, Havcr2) (Fig. 3B). Moreover, trajectory analysis of CD8⁺ T cells revealed a transition from stem-like to an effector/exhausted phenotype, suggesting that HDACi likely promote this transition (Fig. 3D, E). Both treatments also increased the proportion of early-exhausted CD8⁺ T cells. As the combination of anti-PD-1/anti-CTLA4 and romidepsin led to a pronounced anti-tumor response, we obtained very few cells from tumors after this treatment (79 T/NK cells). With this limitation in mind, we did observe that changes after the combination treatment were comparable to those after single agent treatments.

Figure 4. Major cell types impacted after Romidepsin treatment or/and CTLA4/PD1 blockade.

A. UMAP projections displaying 12 major cell lineages constituting CD45⁺ cells in KPN1 tumors, colored and labeled by their cell type. Each cluster is determined based on its predefined gene signature. B. Marker gene bubble plot showing the relative abundance of key cell type-associated genes, Z score normalized, and colored from red (high) to blue (low), based on their expression in each cell lineage. Size of the dot represents the percentage of positive cells in each cell lineage. C. Bar graph displaying the percentage composition of each cell lineage in the indicated treatments, showing the total cell count processed for analysis in each treatment condition.

Figure 3. Impact of Romidepsin and CTLA4/PD1 blockade in reshaping the transcriptomic profile of infiltrating T cells and NK cells in KPN1 tumors.

A. UMAP projections displaying 10 different T cell clusters and 1 NK cell cluster infiltrating KPN1 tumors, colored and labeled by their cell type. T cell clusters are determined based on their predefined gene signature. B. Heatmap analysis showing relative expression of different phenotype-related gene signatures. Relative gene expression was determined as Z score normalized and depicted for each cluster from red (high) to blue (low). C. Bar graph displaying the percentage composition of colored T cell and NK cell clusters in the indicated treatments, showing the total T/NK cell count processed for analysis in each treatment condition. (D) UMAP showing trajectory and (E) pseudotime of CD8⁺ T cells estimated by Moncole 3. CD8⁺ stem-like T cells were selected as root node of the trajectory.

In assessing the overall presence of immune cell types, we found that while control tumors were dominated by B and T lymphocytes (~75%), both anti-PD-1/anti-CTLA4 and romidepsin led to a dramatic increase in monocytes and macrophages which together accounted for ~80% of total CD45⁺ cells (Fig. 4A–C). Multiple sub-types of transcriptionally distinct monocytes and macrophages were detected, indicating they may be playing functionally distinct roles in the TME (Supplemental Figure 4A–C). Delving into the transcriptomic signature of these monocytes and macrophages subsets, we found that Romidepsin promoted the expansion of myeloid subsets with immune suppressive signatures (Ccr2⁺, Arg⁺ and Chil3⁺ monocyte clusters) ^39–44 (Supplemental Figure 4A–C). Similarly, two of these immune suppressive myeloid clusters (Ccr2⁺ and Arg⁺ monocyte clusters) were shown to be expanded by anti-PD1/anti-CTLA4 ICI, substantiating our notion that HDAC inhibition by Romidepsin reshapes TME in a pattern resembling that induced by ICI. While the combination also led to increase in monocytes and macrophages, another notable change was the dramatic increase in neutrophils (Fig. 4A–C). This increase in neutrophils may be due, at least in part, to the strong anti-tumor effect of the combination following which neutrophils traffic to clear cell debris. Overall, the increases in myeloid populations observed could be due to direct effects of HDACi on these cell-types, adaptive changes resulting from increase in T cell effector functions (e.g., feedback increase in suppressive myeloid cells), as well as secondary effects of strong anti-tumor activity of the combination. In conclusion, these results indicate HDACi induce dramatic changes in the immune microenvironment of tumors, which in many instances, are similar to those induced by ICIs.

HDACi impact on T cell activating and inhibitory gene expression

Our scRNA-seq findings indicate that in the TME, HDACi increase CD8⁺ T cells with high expression of effector molecules but also inhibitory receptors (Fig. 3A–C). Therefore, we asked the question whether direct HDAC inhibition on activated T cells would result in a similar increase in T cell effector and inhibitory molecule expression. We treated CD8⁺ T cells (pre-stimulated with anti-CD3/anti-CD28) with Romidepsin at either 2 and 10 nM doses for 1, 2, and 3 days before determining impact on gene expression. Consistent with our in vivo observations, HDACi significantly increased the expression of effector genes (Ifng, Prf1, Gzmk, Tnf) with a more pronounced effect after 2 or 3 days of treatment (Fig. 5A). HDACi significantly downregulated key inhibitory genes (Lag3, Havcr2) after 2 or 3 day treatment regimens (Fig. 5B). On the other hand, HDACi increased expression of Pdcd1 and importantly also that of Tox, a master transcriptional regulator of the T cell exhaustion program^{45, 46}. These findings indicate that HDAC inhibition, at least over the short-term, promotes a T cell activation program but with mixed effects on expression of genes involved in T cell exhaustion.

Figure 5. HDACi impact on T cell activating and inhibitory gene expression.

Transcriptional expression of effector genes (A), or inhibitory genes (B) determined by gene specific qPCR analysis after treating CD8⁺ T cells (pre-stimulated with anti-CD3/anti-CD28 Abs) with either 2 or 10 nM Romidepsin for 1, 2, or 3 days. The expression level of each gene was normalized to internal control (S18) and plotted as fold change relative to untreated (UT) sample. Each column represents the mean of three independent replicate +/− SD. Statistical significance is shown as compared to UT control. Data shown here represent one of two independent experiments.

We next used a different experimental system to investigate the impact of Romidepsin and ICI anti-PD1 on the effector function and inhibitory marker expression on CD8⁺T cells. We utilized T cell-tumor cell coculture system comprising of Ovalbumin peptide (SIINFEKL) specific OT1 CD8⁺ T cells and KP lung tumor cells pulsed with OVA peptide. This mixed system allowed us to dissect the effect of HDAC inhibition and immune checkpoint blockade on the bimodal T cell/tumor cell cross-talk through both antigen-induced T cell activation and PD1-PDL1 mediated T cell inhibition. OT1 T cells were stimulated with anti-CD3 and anti-CD28 and then cocultured with KP tumor cells pulsed overnight with OVA peptide at ratio Effector:Target (E:T = 1:1) in the presence of Romidepsin (10nM) and/or anti-PD1 (10ug/ml) for 1, 2 and 3 days. We first assessed the direct effect of HDACi on CD8⁺ T cell cytotoxicity (Fig. 6A). While anti-PD1 ICI did not enhance OT1 T cell-mediated killing, Romidepsin strongly promoted OT1 T cell cytotoxicity at both E:T ratios (1:1, 1:5). Furthermore, as in above studies, we observed that Romidepsin promoted increase in IFNγ secretion at the three different timepoints (Fig. 6B). While PD1 blockade alone exerted modest effect on IFNγ secretion, potentially because of insufficient PD-L1 expression as in the TME PDL1 is provided through both cancer and immune cells ^{47, 48}, the combination displayed the highest secretion of IFNγ, especially after 3 days (Fig. 6B). Finally, our findings revealed again that Romidepsin significantly and steadily downregulated exhaustion markers (especially LAG3 and TIM3) after 1,2 and 3 days of treatment (Fig. 6C). PD-1 was also decreased but to lesser extent then LAG3 and TIM3. Interestingly, while anti-PD1 ICI have resulted in a compensatory slight upregulation of TIM3 and LAG3 inhibitory markers, the combination treatment was generally associated with the most significant downregulation of T cells exhaustion signal (Fig. 6C). Collectively, these results indicate that HDAC inhibition promotes CD8⁺ T cell cytotoxicity and decreased inhibitory receptor expression.

Figure 6. Impact of HDAC inhibition by Romidepsin and immune check point blockade by anti-PD1 on T cell cytotoxicity and expression of activation and inhibitory receptors.

A. OT1 T cell cytotoxicity assessed by xCELLigence^® Real-Time Cell Analysis for the Cellular impedance kinetics of KP mouse lung tumor cells pulsed with OVA peptide then cocultured with OT1 T cells at a ratio (effector: target =1:1, and 1:5) and treated with Romidepsin (ROMI) (10nM) and/or anti-mouse PD1 (10ug/ml) over 70 hours period. Data is shown as mean of 3 independent replicates ± SD and compared to unpulsed tumor cells cultured alone or with OT1 T cells (E:T ratio = 1:1). B. IFNγ secretion measured by ELISA in the culture media collected after 1,2 and 3 days of OT1 T cells coculture with OVA-peptide pulsed KP tumor cells and treated with Romidepsin (10nM) and/or anti-mouse PD1 (10ug/ml). C. quantification of PD1, TIM3 and LAG3 inhibitory marker surface expression on CD45⁺ CD8⁺ live gated cells measured as mean fluorescence intensity by flow cytometry after 1,2 and 3 days of coculture with OVA-peptide pulsed KP tumor cells and treated as indicated in B. Data is shown as mean of 3 independent replicates ± SD and compared to OT1 T cells cultured alone (untreated or treated with Romidepsin) or cocultured with unpulsed tumor cells. Significance is indicated as compared to untreated KP-OVA + OT1 T cells coculture and analyzed using repeated-measures 2-way ANOVA followed by Dunnett’s multiple-comparison test. Data are representative of 2 independent experimental runs.

Discussion

In the preclinical setting, multiple studies have demonstrated that HDACi can reverse immune suppression leading to a more immunostimulatory TME and greater responses to ICIs ^5–20. Some of the proposed mechanisms of action of HDACi include direct effects on tumor cells and myeloid cells. Here, we show that HDACi also profoundly impact T cell gene expression and drive T cells towards an effector phenotype. A key effect was an increase in the expression of Th1 response-associated cytokines (TNFα and IFNγ). As these cytokines themselves modulate the expression of multiple immune function genes, some of the HDACi-induced changes in gene expression in T cells may be secondary to the upregulation of TNFα and IFNγ, e.g., the marked increase in expression of multiple chemokines. Furthermore, we found that activity of TNFA-NFKB and IFNG pathways were increased after HDACi treatment. Supportive of our findings, a recent clinical study of histone deacetylase inhibitor mocetinostat with ipilimumab (anti-CTLA4) and nivolumab (anti-PD-1) in melanoma showed increase in systemic levels of TNFα and IFNγ ²⁶.

Within the TME, HDACi increased the proportion of CD8⁺ T cells with high expression of effector genes (Prf1, Ifng, Gzmk, Grmb). These findings complement our in vitro studies showing increase in effector markers and cytokine expression in two different systems of T cells activation. Multiple preclinical studies have demonstrated that HDACi enhance T cell therapy response ^5–20. Our results indicate that the increase in the effector CD8⁺ T cell response is likely involved in mediating this benefit. However, these HDACi-enhanced CD8⁺ T cell populations also had the highest expression of terminal exhaustion-associated genes (Tox, Pdcd1, Lag3, Havcr2). Interestingly, both HDACi and ICIs triggered an increase in the effector-exhausted CD8⁺ T cell population, underscoring the immune modulatory effect of HDACi on T cells. IFNγ mediates anti-tumor immunity but also plays a key role in regulating mechanisms that enhance T cell exhaustion ^{49, 50}. In this respect, IFNγ and TNFα could function as a double-edged sword where, in the short-term, they drive anti-tumor responses but eventually enhance T cell exhaustion ⁴⁹. Furthermore, HDACi can increase the cellular response to IFNγ ²⁰, including increase in PD-L1 expression, which may additionally promote T cell exhaustion.

Both our in vitro and in vivo results indicate an increase in effector markers, including cytokines, and an increase in CD8⁺ T cell cytotoxicity. In vitro inhibition of HDACs in activated CD8⁺ T cells showed a decrease in late inhibitory markers like Lag3 and Tim3. On the other hand, PD-1 was increased and interestingly, expression of Tox, a master key regulator of the T cell exhaustion machinery was also increased ^{45, 46}. These results suggested that while HDACi augment T cell effector function, the increase in exhaustion regulators such as Tox could lay the groundwork for a T cell dysfunctional state. Furthermore, the enrichment of Tregs and immune suppressive myeloid subsets in the TME of HDACi-treated tumors as shown by our scRNA-seq results could also contribute to T cell dysfunction. Indeed, the recruitment of these immunosuppressive cells is possibly an immune-regulatory mechanism as a consequence of the early increase in activation of T cells by HDAC inhibition. Collectively, our in vitro and in vivo results indicate that Romidepsin induces a marked increase in T cell effector function but in the long-term may also promote an exhausted T cell phenotype.

In the TME, we also found that HDACi (and ICI) dramatically increase the population of monocytes and macrophages. Among them, three distinct monocyte subsets (Ccr2⁺, Arg⁺ and Chil3⁺ monocyte clusters) were shown to be expanded in Romidepsin treated tumors. The transcriptomic profile of these clusters revealed that they are likely of immunosuppressive nature. Chil3-high monocytes have been reported to have M2 macrophage phenotype with anti-inflammatory activity ^{39, 40}. CCR2 and its ligand CCL2 are commonly associated with recruitment of immune suppressive myeloid subsets that limit T-cell immunity and facilitate immune evasion ^{41, 42}. Moreover, a plethora of evidence has linked Arg⁺ monocytes, which likely have monocytic myeloid-derived suppressor cells (mo-MDSCs) characteristics, with tumor promotion and T cell suppression in TME ^{43, 44}. Together, our scRNA-seq data demonstrate that HDACi favorably enhance the effector function of T cells in the TME. However, it also enhances a suppressive TME dominated by Treg and immunosuppressive monocytes that could ultimately accelerate T cell exhaustion and diminish antitumor T cell immunity. Indeed, these findings substantially propose pairing immune check point blockade with HDACi as a potential therapeutic strategy to delay T cell exhaustion and improve T cell cytotoxicity mediated by HDAC inhibition.

Many clinical trials are currently determining whether the combination of HDACi with ICIs can increase patient benefit. In lung cancer, this combination has not led to the high response rates that were anticipated ^{23, 24}. While further clinical and correlative studies are needed, it is possible that the direct effects of HDACi on T cells impact the overall clinical responses to ICIs. In this respect, we show that HDACi have a major impact on T cell gene expression pathways, with both in vitro and in vivo studies showing that HDACi can promote a T cell effector response. However, our findings also indicate that HDACi could eventually promote T cell exhaustion, increase Treg and immune suppressive monocytes to diminish T cell functionality, which could potentially dampen the anti-tumor activity of ICIs. Findings from ongoing clinical studies will help determine the potential impact of HDACi on patient response and survival in the setting of ICI treatment.

Key points:

1. HDACi directly promote expression of activation-associated genes in T cells.

2. In the TME, HDACi promote both T cell activation and exhaustion.

3. HDACi mediate expansion of immunosuppressive myeloid populations.