Epigenetic plus cytokine therapy induces stem-like CD8+ T-cell reservoirs and intratumoral immune niches to overcome anti-PD-1 resistance

Abstract

Background

Checkpoint blockade (CPB) has limited efficacy in patients with colorectal cancer (CRC), warranting improved therapeutic strategies capable of remodeling the tumor microenvironment. Stem-like TCF1⁺CD8⁺ T cells, cytotoxic CD8⁺ T cells, and lymphoid aggregates have been associated with CPB responsiveness, either individually or as coordinated immune hubs.

Methods

We evaluated the combination of αPD-1, the class I histone deacetylase inhibitor entinostat, and the IL-15 superagonist N-803 in CPB-resistant murine CRC and breast cancer models with varying antigen presentation deficiencies. Proteomic, transcriptomic, spatial, and functional analyses were completed to identify treatment-specific immune remodeling. Translational relevance was assessed using publicly available CPB-treated cancer genomic datasets and functional assays with patient-derived peripheral blood mononuclear cells (PBMCs).

Results

Triple therapy increased stem-like TCF1⁺CD8⁺ T cells in tumor-draining lymph nodes and tumors, concomitant with increased intratumoral cytotoxic GZMB⁺CD8⁺ T cells. Therapeutic efficacy was associated with remodeling of the tumor microenvironment, uniquely marked by enrichment of intratumoral immune niches comprising TCF1⁺ and cytotoxic CD8⁺ T cells, type 1 conventional dendritic cells, and B cells, along with concerted cytokine and chemokine production. A transcriptional signature derived from triple therapy-induced immune niches predicted favorable clinical responses across multiple CPB-experienced cancer cohorts. These responses were functionally corroborated by increased markers of cytotoxicity in patient-derived PBMCs in vitro.

Conclusions

Treatment with entinostat, N-803, and αPD-1 boosts anti-tumor intratumoral immune niches comprising TCF1⁺CD8⁺ T cells. Collectively, these findings suggest that TCF1⁺CD8⁺ T cell immune niches previously associated with αPD-1 responsiveness and lacking in non-responders, including most CRC patients, may be restored by the addition of entinostat and N-803 to αPD-1 therapy to treat αPD-1-refractory patients.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Increasingly, checkpoint blockade (CPB)-responsive patients are seen with intratumoral increases in stem-like TCF1⁺CD8⁺ T cells, cytotoxic CD8⁺ T cells, and lymphoid aggregates either occurring individually or as coordinated stem-immune hubs. However, CPB therapies have limited efficacy in patients with colorectal cancer and other CPB-resistant tumors.

WHAT THIS STUDY ADDS

• Here, we report for the first time that the addition of histone deacetylase inhibition and an IL-15 superagonist to anti-PD-1 therapy significantly increases TCF1⁺CD8⁺ T cells in tumor-draining lymph nodes and tumors simultaneously with the formation of coordinated immune niches in αPD-1-refractory tumors. These findings further elucidate agent-specific cellular and genetic changes that may inform therapeutic designs.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• Together, these data introduce a novel translational strategy to reprogram the tumor microenvironment and restore effective antitumor immunity in anti-PD-1-refractory disease.

Introduction

Checkpoint blockade (CPB) therapies targeting programmed cell death protein-1 (PD-1) are effective in only a minor subset of patients with colorectal cancer (CRC) with microsatellite instability-high (MSI-H)/mismatch repair deficiency.¹ However, 70%–90% of CRC patients harbor CPB-refractory microsatellite stable (MSS)/mismatch repair proficient (MMRp) disease.¹ Hallmark mechanisms of CPB resistance in CRC include neoantigen depletion and impaired interferon gamma (IFNγ) and antigen presentation response pathways.² Patient tumors with MSS disease are further enriched for inflammatory yet tolerogenic CD8⁺ T cells, in contrast to the CD8⁺ T stem-like progenitor cell populations present in CPB-sensitive MSI-H tumors.³ KRAS mutations, associated with increased metastasis and poor prognosis, further promote primary CPB resistance by inhibiting IFN-related genes.^{1 2} Thus, strategies to enhance CPB responsiveness in CRC are warranted.

Accumulation of epigenetic alterations and histone deacetylase (HDAC) expression in CRC support the use of HDAC inhibitors (HDACi), with several HDACi approved by the Food and Drug Administration (FDA) only for hematologic malignancies.^{4 5} Entinostat is a class I HDACi that has shown clinical activity in solid tumors when used in combination regimens.^5–7 We and others have previously shown entinostat to synergize with CPB or immunocytokines to increase cytotoxic CD8⁺ T cells, immune memory, antigen presentation, and neoantigen abundance, while reducing immune suppression and CPB resistance in experimental solid tumors.^8–10 Clinically, HDACi combinations with αPD-1 have demonstrated favorable safety but limited effectiveness in solid tumors, including MSS CRC.^{7 11 12} Entinostat plus pembrolizumab demonstrated 9%‒14% overall response rate (ORR) in αPD-1-refractory metastatic non-small cell lung cancer (NSCLC) and uveal melanoma, with a favorable safety profile and promising clinical activity in a small cohort of CRC MSS patients.^{7 12 13} Recent studies incorporating a third agent, such as αCTLA4 or αVEGF, to HDACi plus CPB regimens yielded further improved outcomes, including in CRC MSS patients.^{6 14} However, treatment-related fatalities were reported in association with bevacizumab-containing combinations.¹⁴

The IL-15 superagonist N-803 is a potent activator of both CD8⁺ T and natural killer (NK) cells, and was recently FDA-approved to treat lymphopenia in cancer patients and in combination with Bacille Calmette-Guérin (BCG), for BCG-unresponsive non-muscle invasive bladder cancer.¹⁵ Preclinical studies combining entinostat with N-803 and vaccine, or N-803 with anti-programmed death-ligand 1 (αPD-L1), have demonstrated robust T cell and NK cell activation.^{9 16} Clinically, the N-803 and αPD-1 combination was well tolerated and achieved an objective response rate of 29% in NSCLC patients, with disease control achieved in 91% of patients with αPD-1-relapsed and refractory disease.¹⁷ Together, these findings support the clinical addition of N-803 as a third agent to entinostat plus αPD-1. Moreover, clinical data on HDACi plus cytokine immunotherapy for solid malignancies is lacking.

Mounting evidence supports the assertion that effective αPD-1 therapy requires multicellular immune interactions that sustain CD8⁺ T cell-mediated antitumor and survival benefits.^{18 19} CD8⁺ T cells encompass various differentiation states governed by T cell factor 1 (TCF1; Tcf7), including stem-like PD-1⁺TCF1⁺, cytotoxic granzyme B⁺ (GZMB⁺), and terminally exhausted TIM3⁺CD8⁺ T cells that can differentially influence therapeutic outcome.^{20 21} Importantly, CD8⁺ T cell differentiation and maintenance of stem-like states are tightly regulated by epigenetic programs controlling transcriptional accessibility and fate decisions.^{22 23} Recent studies have demonstrated that stem-like TCF1⁺CD8⁺ T cells are maintained as primed “reservoirs” within tumor-draining lymph nodes (tdLN), where they give rise to effector populations sustained in the tumor microenvironment (TME).²³ These cells interact with type 1 conventional dendritic cells (cDC1) within tertiary lymphoid structures (TLS) or similar multicellular ‘stem-immunity hubs’ correlating with enhanced tumor control, a process described largely in the context of αPD-1 therapy.^{19 21 24 25} Whether such stem-like CD8⁺ T cell reservoirs and coordinated immune architectures can be induced in CPB-refractory tumors remains unclear. Given that HDACi, IL-15 signaling, and αPD-1 have been implicated in multicellular activation, hematopoietic regulation, and expansion of stem-like T cells, we used CPB-resistant CRC models and interrogated tdLN and TME to examine how epigenetic and cytokine-based modulation with αPD-1 influences CD8⁺ T cell development and immune interactions.^{21 26 27}

Materials and methods

Mice

BALB/c and C57BL/6 female mice aged 6–8 weeks were acquired from the National Cancer Institute (NCI) Frederick mouse repository. Batf3 ^−/−(B6.129S(C)-Batf3^tm1Kmm/J; #013755) females were acquired from The Jackson Laboratory.²⁸ Mice were co-housed under specific pathogen-free conditions in an Association for Assessment and Accreditation of Laboratory Animal Care-accredited animal facility of the National Institutes of Health (NIH).

Murine tumor cell lines

Colon CT26 (CRL-2638) and breast EMT6 (CRL-2755) carcinoma cells obtained from American Type Culture Collection were cultured as recommended. MC38 colon carcinoma cells were cultured as previously described.^{8 16} Cells were used at low passage (<6) and routinely tested negative for Mycoplasma (MycoAlert Kit; Lonza). MC38, and EMT6 with Jak1 gene knocked out were generated as previously described.²⁹

Murine tumor studies

CT26 (1×10⁶) and MC38-Jak1 knockout (KO) (3–6×10⁵) cells were implanted subcutaneously (s.c.) into the right flank of BALB/c or C57Bl/6 mice, respectively. EMT6 Jak1 KO (5×10⁵) cells were implanted orthotopically into BALB/c mammary fat pad. When tumors reached 50–100 mm³, mice were randomized to entinostat (6 mg/kg/day) or control diet for 10 days with concurrent intraperitoneal (i.p.) administration of αPD-1 (200 µg) or phosphate-buffered saline (PBS) on alternating days (three doses total). Two days after treatment onset, N-803 (1 µg, s.c.) or PBS was administered ipsilaterally every 5 days (two doses). In selected studies, CT26 or EMT6 Jak1 KO tumor rechallenge was performed 45‒90 days after tumor resolution. For depletion studies, αCD8 or αCD4 antibodies (100 µg/100 µL, i.p.) were administered on days 7‒9 and weekly thereafter. Depletion efficiency was confirmed by flow cytometry. Tumor volumes were measured biweekly using digital calipers (length²×width/2). Experimental schematics were created with BioRender.com. All in vivo experiments were performed using independent cohorts of mice and repeated 2‒3 times with comparable results.

Flow cytometry

Cells (1–2×10⁶) stained with LIVE/DEAD (ThermoFisher) per the manufacturer’s instructions were incubated (10 min at 4°C) with Fc-block (Biolegend) prior to cell surface staining (30 min at 4°C) with designated antibodies (online supplemental table S1). Intracellular staining was executed using BD Biosciences cytofix/cytoperm or eBioscience transcription factor kits per the manufacturer’s instructions. Acquisition of 2.5×10⁵ cells for spleens, tdLN, or peripheral blood, 1–3×10⁵ cells for tumors, and 5×10⁵ cells for bone marrow was completed on the LSRFortessa (BD Biosciences). Data were analyzed in FlowJo Analysis Software V.10.10.0 (BD Biosciences) using designated gating strategies (online supplemental table S2, figure S8), and as previously described.^{10 30}

Multiplex immunofluorescence and spatial analysis

CT26 tumor sections preserved in Z-fix (Anatech) were paraffin-embedded and sectioned (5 µm) (Histoserv, Inc). Tumor sections were stained with validated multiplex immunofluorescence (MX-IF) using the Tyramine Signal Amplification Opal 6-plex kit (Akoya Biosciences) and the BOND RX Autostainer (Leica Biosystems). Antibodies, protocol, and Opal used are further described in online supplemental table S1. H&E and MX-IF whole slide scans were captured (PhenoImager HT, Akoya Biosciences) at 20× magnification. Image analysis and biomarker quantification (cells/mm²) were performed using Phenoplex analytical software (Visiopharm).

Healthy donor and cancer patient peripheral blood mononuclear cells

Deidentified healthy donor (HD) peripheral blood mononuclear cells (PBMCs) obtained from the NIH Clinical Center Blood Bank (NCT00001846) were cryopreserved. Pretherapy cryopreserved PBMCs were obtained from metastatic colorectal or breast cancer patients enrolled in trials (NCT00179309, NCT00088413) conducted at the NCI.

Human T cell stimulation assay

HD and cancer patient PBMCs were defrosted and cultured (1×10⁶ cells/well) overnight (37°C/5% CO₂) in Iscove’s Modified Dulbecco’s Medium (GIBCO) supplemented with 10% heat-inactivated human AB serum (Omega Scientific). Cells were then replated in 96-well plates and exposed to entinostat (500 nM) and/or Pembrolizumab Biosimilar (10 µg/mL; BioXCell) for 48 hours. Media was replaced with or without N-803 (25 ng/mL) for an additional 24 hours prior to being left unstimulated or stimulated with 0.5 µg/mL soluble αCD3 (OKT3, ThermoFisher) and αCD28 (28.2, ThermoFisher) for 24 hours. Cells were then stained and analyzed by flow cytometry (online supplemental tables S1, S2, figure S8).

Additional materials and methods are described in the online supplemental methods.

Statistics

Statistical analyses (two-tailed) were performed on GraphPad Prism V.10.2.3 (GraphPad) and listed in figure legends for bar and violin plots. For scRNA-Seq, a p value threshold of 0.05 and a fold-change threshold of 1.2 or 1.5 were used to delineate significant changes. Volcano plots utilized the Benjamini-Hochberg (BH) adjustment where genes are deemed differentially expressed with an adjusted p<0.05 following correction for multiple testing via the BH procedure. For gene set enrichment analysis (GSEA), the t-statistic of differential expression was used for gene-rank scores. The Fisher’s exact test was used for pathway analysis from gene ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG), and Hallmark sources with a p value cut-off of 0.05. Statistical significance was set at p<0.05. *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001, ns=not significant. Two or more biologically independent experimental replicates were performed and showed comparable results with representative results from one experiment shown. Samples from multiple experiments were pooled in some cases for statistical power and to demonstrate main points. The ROUT outlier test was performed for each bar or violin plot and outliers excluded at a false discovery rate (FDR) <5% depending on sample number.

Data availability

Data generated in this study are available within the article and its online supplemental data files. ScRNA-Seq data generated in this study are publicly available in Gene Expression Omnibus at GSE279544. Publicly available data generated by others and analyzed in this study are identified in the figure legends and referenced in online supplemental table S1. CRISPR/KO cell lines are available on request but will require completion of a Materials Transfer Agreement. All bioinformatics scripts have been deposited on GitHub (https://github.com/NIDAP-Community/Epigenetic-plus-Cytokine-Therapy-Evades-CPB-Resistance-by-Potentiating-CTL-Stem-Immunity-Clusters).

Results

Triple therapy outperforms single and dual agent therapies in murine CRC

CT26 is an experimental model often used as representative of MSS/MMRp human CRC due to lack of mutations in MMR genes Mlh1 and Mlh2, harboring mutated Kras and low response to CPB.³¹ To elucidate if the combination of entinostat (ENT), N-803, and αPD-1 could overcome CPB-resistant CRC, CT26 tumor-bearing mice were treated with one therapy cycle (figure 1A). Triple therapy rejected tumors in 54.5% (6/11) of mice, a 34.5% and 45.5% efficacy increase over entinostat combinations with αPD-1 and N-803, respectively (figure 1B). Within 21 days of tumor implant, significant tumor growth inhibition (TGI) was observed in cohorts treated with entinostat plus αPD-1 (TGI: 60%) and triple therapy (TGI: 77%) relative to PBS control. Although overall survival statistical significance was not reached between triple therapy and entinostat plus αPD-1, median overall survival was not reached with triple therapy versus 46 days with entinostat plus αPD-1, reflecting the increase in tumor rejection observed with triple therapy (54.5%) versus the doublet (20%) (figure 1C). Triple therapy-cured mice resisted tumor rechallenge (93%; 14/15), associated with CD8⁺ T cell responses specific to the endogenous CT26 immunodominant epitope AH1 (figure 1D; online supplemental figure S1A). To recapitulate clinical CPB resistance in CRC and additional malignancies, we developed CPB-resistant MC38 (CRC) and EMT6 (breast carcinoma) Jak1 KO cell lines.²⁹ MC38 has been characterized as analogous to MSI-H human CRC, sharing αPD-1 sensitivity with EMT6.^{31 32} These KO cell lines were confirmed deficient in IFNγ response and antigen presentation-associated genes (online supplemental figure S1B).²⁹ Only triple therapy demonstrated significant MC38 Jak1 KO TGI versus control (52%), doublets, and monotherapies, with 87.5% (7/8) of tumors displaying volumes <500 mm³ at day 18 post-tumor implant (figure 1E,F). Orthotopic EMT6 Jak1 KO also demonstrated improved TGI and survival with triple therapy, associated with increased cytotoxic (GZMB⁺) and proliferative (KI67⁺) CD8⁺ T cells in peripheral blood, with cured mice resisting tumor rechallenge (online supplemental figure S1C–E). This was concordant with findings in the CT26 model, where depletion of CD8⁺ but not CD4⁺ T cells negated triple therapy efficacy (figure 1G). Collectively, these data demonstrate that triple therapy synergizes for effective TGI in various CPB-refractory solid tumor models, including CRC. No significant weight loss or novel organ toxicities (on-treatment or long-term), or clinical signs of adverse events were observed with triple therapy relative to age-matched controls (online supplemental figure S1F, G).

Figure 1. Triple therapy outperforms single and dual agent therapies in murine colorectal cancer. (A) Treatment schedule for this study. (B) Individual CT26 tumor growth curves with percent cures indicated. (C) Mean tumor growth and survival for CT26 tumor-bearing mice (n=10‒11 mice/group). (D) Individual tumor growth curves on CT26 tumor rechallenge (n=12‒15 mice/group; data combined from three independent experiments). (E) Mean MC38 Jak1 KO tumor growth curves (n=7‒8 mice/group). (F) Individual MC38 Jak1 KO tumor volumes per treatment at day 18; one-way ANOVA with Tukey’s multiple comparisons test. (G) Tumor growth curves and survival for CT26 tumor-bearing mice treated with triple therapy and αCD8 or αCD4 cell depletion (n=12 mice/group). (B, C, E, F) Data from one experiment representative of 2‒3 independent studies. Tumor volume data show mean±SEM, two-way ANOVA with Tukey’s multiple comparisons test, and graph indicators as: entinostat feed, gray shade; αPD-1 treatment, dashed lines; N-803 treatment, notches on x-axis. Survival: Mantel-Cox. *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001, ns=non-significant. (A) Created in BioRender. Minnar, C. (2026) https://BioRender.com/o09dqz4. ANOVA, analysis of variance; ENT, entinostat; i.p, intraperitoneal; KO, knockout; mOS, median overall survival; PBS, phosphate-buffered saline; s.c., subcutaneous; TGI, tumor growth inhibition; UD, undefined.

Triple therapy boosts CD8⁺ T cell function and tumor homing

Since CD8⁺ T cells were determinant of triple therapy-mediated tumor control, we analyzed splenic and tumor-infiltrating lymphocytes (TILs) 3 days after the final dose of N-803 (day 17). All subsequent analyses were carried out at this timepoint unless specified otherwise. Triple therapy significantly increased CT26 CD8⁺ TILs relative to control, single agent, and N803-containing doublets, while displaying heightened proliferative capacity relative to all groups (figure 2A). Cytolytic TILs were seen significantly increased with entinostat, entinostat plus αPD-1, and triple therapy (figure 2B). Peripherally, proliferative (Ki67⁺) CD8⁺ T cells were significantly expanded with entinostat plus N803 and triplet groups, with only triple therapy showing significantly greater cytolytic CD8⁺ T cells relative to all single and double agents (figure 2C). Increased polyfunctionality was observed driven by N-803 (figure 2D). Augmented CD8⁺ T cell activation elicited by triple therapy was associated with elevated serum chemokines (CCL4, CXCL9, CCL22) and cytokines (IFNγ), including tumor necrosis factor-alpha (TNFα), CCL5, and CXCL13, indicating a proinflammatory milieu with coordinated crosstalk (figure 2E; online supplemental figure S2A). Next, we performed scRNA-Seq analysis of CD45⁺ cells isolated from CT26 tumors and paired tdLN to gain further insight into the mechanism underlying triple therapy antitumor effects (figure 2F; online supplemental figure S9). Differential expression of genes (DEGs) analysis of triple therapy CD3⁺CD8⁺ TIL clusters showed significant increases in Ccl5 and granzyme genes (figure 2G). Dissection of single agent effects on CD8⁺ TILs showed entinostat to significantly contribute to enrichment in granzyme and chemokine genes, such as Ccl5, whereas αPD-1 contributed to activation and maturation genes, including perforin and granzyme genes. Genes associated with naïve and stem-like CD8⁺ T cell states (Tcf7, Lef1, Sell, FoxP1, Satb1, Ccr7, Xcl1) were driven by N-803 (online supplemental figure S2B).^{21 33} Comparison of triple therapy with entinostat doublets showed αPD-1 contribution to granzymes, Ccl5, and TNFα inflammatory pathways, while N-803 increased stemness genes, cellular activation, and gene translation (online supplemental figure S2C). Interestingly, GSEA of triple-therapy CD8⁺ TILs revealed the hematopoietic cell lineage and cytokine-cytokine receptor pathways as two of the top enriched pathways (figure 2H). CD8⁺ T cell transcript frequency in tdLN was highest in N803-treated groups, indicating IL-15-mediated T cell reserve at this site (figure 2I). Altogether, these results indicate that triple therapy synergizes to heighten CD8⁺ T cell cytolytic and proliferative capacity, promote a lymphocyte-recruiting cytokine and chemokine milieu, and possibly engage stem-like precursors and a tdLN reservoir to sustain tumor control and resolution.

Figure 2. Triple therapy boosts CD8⁺ T cell function and tumor homing. All CT26 tumor-bearing mice treated as in figure 1A were sacrificed on day 17 unless specified otherwise. (A–D) Tumor and spleen CD8⁺ T cells were analyzed by flow cytometry (n=6 mice/group). (A) Total and Ki67⁺ CD8⁺ T cells/mg of tumor. (B) Granzyme B⁺ (GZMB⁺) CD8⁺ T cells/mg of tumor and representative histograms. (C) Frequency of Ki67-expressing and GZMB-expressing splenic CD44^hiCD8⁺ T cells. (D) Average frequency of IFNγ, TNFα, or (IFNγ/TNFα) double-positive (DP) splenic CD44^hiCD8⁺ T cells and representative zebra plot. (E) Heatmap illustrating normalized row z-scores for designated serum cytokines and chemokines (n=6 mice/group; IFNγ and TNFα analytes include pooled data from three independent experiments, n=16 mice/group). (F) Overview of CT26 tumors and matched draining lymph nodes (dLN) acquired for scRNA-Seq analyses (n=4‒6 mice/group). (G) Volcano plot shows differentially expressed genes (DEGs; Benjamini-Hochberg (BH)/false discovery rate (FDR) <0.05; fold change >±1.5 and p<0.05) in triple therapy-treated CD8⁺ TILs versus PBS control. (H) Top upregulated KEGG pathways elicited by triple therapy in CD8⁺ TILs versus PBS control. (I) Bar plots representing tumor and tdLN CD8⁺ frequencies out of total CD45⁺ filtered scRNA-Seq counts. (A–C) Mean±SEM, one-way ANOVA with Tukey’s multiple comparisons test; *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001, ns=non-significant. Data in (A‒E) are from one experiment representative of 2‒3 independent studies unless specified. (F) Created in BioRender. Minnar, C. (2026) https://BioRender.com/2iq8dew. ANOVA, analysis of variance; ENT, entinostat; FC, fold change; KEGG, Kyoto Encyclopedia of Genes and Genomes; PBS, phosphate-buffered saline; tdLN, tumor-draining lymph nodes; TILs, tumor-infiltrating lymphocytes; TNF, tumor-necrosis factor.

Triple therapy enriches activated stem-like TCF1⁺CD8⁺ T cells in tumors and tdLNs

Next, we investigated stem-like and memory components of paired CD8⁺ T cells in tdLN and TME. Pseudotime trajectory analysis demonstrated earlier CD8⁺ T cell phenotypes prevalent in tdLN with maturation occurring within the TME (figure 3A). Cluster 26, unique to triple therapy, revealed trajectory from naïve/stem-like genes enriched in tdLN toward effector programs in the TME (figure 3B). Uniform Manifold Approximation and Projection (UMAP) visualization further showed that stemness and self-renewal markers (Tcf7, Ly6a, Bach2, Slamf6) were also detectable within TILs, with Tcf7 preferentially expressed in the tdLN (figure 3C, online supplemental figure S2D). Flow cytometry analysis of tdLN CD8⁺ T cells showed primed (CD44⁺), central (CM), and effector (EM) memory phenotypes significantly increased in N803 treatment cohorts (figure 3D; online supplemental figure S2E). However, only triple therapy significantly increased TCF1⁺ and SCA1⁺ expression on EM and CM CD8⁺ T cells, respectively, relative to all treatment groups (figure 3E; online supplemental figure S2F). We extended this characterization to identify tdLN stem-like PD-1⁺TCF1⁺Tim3^neg CD8⁺ T cells previously shown as a determinant for αPD-1 efficacy.²¹ These cells were significantly expanded by entinostat plus N-803, and triple therapy, with no significant changes versus control noted in the more terminally differentiated, exhausted-like PD-1⁺TCF1^negTIM3⁺ CD8⁺ T cells (figure 3F; online supplemental figure S3A). Transcriptomic analysis of total and TCF1⁺TIM3^neg CD8⁺ T cells revealed that triple therapy augmented Ccl5 expression, with the latter also expressing genes of stemness, activation/memory, early effector function, and migration, corroborating that triple therapy induced a primed and migratory stem-like phenotype within the tdLN (figure 3G; online supplemental file 1).^{21 22 33} Triple therapy also significantly increased genes of migration and cDC1 recruitment (Ccl5, Xcl1, Cxcr3) in TCF1⁺Tim3^neg CD8⁺ T cell CM and EM populations. Notably, exhaustion-associated Tox gene was significantly decreased in triple therapy-treated total and EM TCF1⁺Tim3^neg populations (figure 3G; online supplemental figure S3C, arrow).³³ Strikingly, stem-like TCF1⁺ EM cells in tdLN correlated with serum TNFα levels, known to induce CCL5 (figure 3H).³⁴ This suggests a feedback loop in which a TNFα-rich milieu enhances Ccl5 expression, both previously reported to be expressed in conjunction with stem-like CD8⁺ T cells, further supporting triple therapy-induced migration of primed TCF1⁺CD8⁺ T cells into the TME.^{21 24 33}

Figure 3. Triple therapy enriches stem-like TCF1⁺CD8⁺ T cells in tumors and tdLN. (A–H) CD8⁺ T cells were analyzed in tdLN and (A–C, I–L) in matched tumor by flow cytometry (E, F, H, J) and scRNA-Seq (A–C, G, K, L). (A) Pseudotime trajectory of CD8⁺ T cell differentiation across paired tdLN and tumor with inset showing tdLN and tumor cell distributions. (B) Pseudotime trajectory heatmap illustrating increasing differentiation from tdLN cluster 5 to triple therapy exclusive cluster 26 in tumor as highlighted (red line). (C) Uniform Manifold Approximation and Projection (UMAP) plots of Tcf7, Ly6a, and Bach2 gene expression in CD8⁺ T tdLN and tumor cells. (D) Activated CD44^hiCD8⁺ T cells, and (E) TCF1 or SCA1 expression on EM and CM CD8⁺ T cells, as percent of CD45⁺ tdLN cells, respectively. (F) Radar plot shows average of PD-1⁺TCF1⁺TIM3^neg or PD-1⁺TCF1^negTIM3⁺ CD8⁺ T cells as percent of CD45⁺ tdLN cells. (G) Volcano plots show DEG (BH test/FDR<0.05) from triple combination‒treated mice versus PBS for total and TCF1⁺TIM3^neg CD8⁺ T cells in tdLN. (H) Serum TNFα correlation with percent tdLN TCF1⁺EM CD8⁺ T cells. (I) Heatmap illustrating normalized row z-scores for designated chemokines in tumor supernatants from treatments versus PBS (analytes include pooled data from two independent experiments, n=11 mice/group). (J) Number of PD-1⁺TCF1⁺TIM3^neg and PD-1⁺TCF1^negTIM3⁺ CD8⁺ T cells/mg of tumor. (K) Volcano plot showing DEG (BH test/FDR<0.05) in TCF⁺TIM3^neg CD8⁺ TILs from triple therapy-treated mice versus PBS. (L) BACH2 expression levels (gMFI) of TCF⁺TIM3^neg CD8⁺ TILs from triple therapy-treated versus PBS mice (one independent study; n=5 mice/group). (M) Chord diagrams show inferred cell-cell interactions for CCL, IFN-II, and MHC-I pathways (CellChat) between tumor CD8⁺ T, cDC1, and B cells from triple therapy-treated CD45⁺ cells. Red dots in volcano plots (G, K) represent individual genes (fold change >±1.2 and p<0.05). (D, E, J) Mean±SEM; one-way ANOVA with Tukey’s multiple comparisons test; *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001, ns=non-significant; (D–F, H, J) include n=6 mice/group representative of 2–3 independently repeated studies. SCA1 expression (E) is from one experiment. (F) Created in Flourish Studio. Minnar, C. (2026) https://public.flourish.studio/visualisation/17584892/. ANOVA, analysis of variance; BH, Benjamin-Hochberg; CCL, C-C motif chemokine ligand; cDC1, type one conventional dendritic cells; CM, central memory; DEG, differentially expressed genes; EM, effector memory; ENT, entinostat; FDR, false discovery rate; gMFI, geometric mean fluorescence intensity; IFN, interferon; MHC, major histocompatibility complex; PBS, phosphate-buffered saline; tdLN, tumor-draining lymph nodes; TIL, tumor-infiltrating lymphocyte; TNF, tumor-necrosis factor.

In tumors, scRNA-Seq analysis of triple-therapy CD8⁺ TILs showed EM cells enriched in cytolytic (Granzymes, Prf1) genes, while CM cells retained stemness genes (Tcf7, Bach2) with significantly decreased Tox expression (online supplemental figure S3D, E (arrow)). Both memory subsets retained high Ccl5 expression. Tumor supernatant analysis revealed that only triple therapy sustained elevated T cell‒homing chemokines, with the T and B cell–promoting chemokine CXCL13 showing the highest elevation (figure 3I; online supplemental figure S3F).³⁵ Consistent with T cell homing into the TME, triple therapy significantly increased PD-1⁺TCF1⁺TIM3^neg CD8⁺ TILs relative to all other groups while approaching significance compared with entinostat plus αPD-1 (p=0.0509); a concurrent decrease in the more terminally differentiated/exhausted-like PD-1⁺TCF1^negTIM3⁺ CD8⁺ TILs was observed with triple therapy relative to PBS control (figure 3J). Transcriptomic analysis of TCF1⁺TIM3^neg CD8⁺ TILs further revealed higher expression of stemness-associated genes when compared with PBS or entinostat doublets, with the highest expression noted for Bach2, a transcriptional repressor of T cell terminal exhaustion known to enforce long-term stem-like CD8⁺ T cell commitment (figure 3K; online supplemental figure S3G).^{21 22 33 36} Consistent with these findings, flow cytometric analysis demonstrated a significant increase in BACH2 protein expression with triple therapy (figure 3L).

IFNγ inducible SCA1, observed on CD8⁺ T memory cell subsets, marks murine bone marrow (BM) progenitor and T stem cell‒memory populations with self-renewal and rapid proliferation capacity.^{27 37} IL-15 cytokine also supports homeostatic expansion of lymphocytes.³⁸ Thus, we reasoned that triple therapy-driven expression of SCA1 may skew progenitors. Analysis of BM progenitors from CT26 tumor-bearing mice showed entinostat-driven increases in common lymphoid progenitors (CLP), early lymphoid committed progenitors (ELCP), and lineage^negSCA1⁺cKIT⁺ (LSK) populations with triple therapy inducing SCA1-high expression (online supplemental figure S4A–F).³⁹ No significant changes were noted in BM myeloid progenitors (online supplemental figure S4G). Collectively, these data show that triple therapy promotes lymphopoiesis in the BM and drives dynamic CD8⁺ T cell changes sustaining stem-like states. Primed stem-like TCF1⁺CD8⁺ T cells accumulate in the tdLN, recruiting additional cells via CCL5 and TNFα. In the TME, stem-like CD8⁺ T cells limit exhaustion, while memory cells upregulate Ccl5/Ccr5 in a CXCL13-high milieu, supporting interactions with cDC1 and B cells.^{24 35} Single-gene UMAPS of tumor CD45⁺ populations confirmed Ccl5 and Ifng gene expression by CD8⁺ T cells (online supplemental figure S5A). Consistently, signaling inference of triple therapy–treated tumors highlighted CCL5, IFN-II, and MHC-I pathways as key predictors of CD8⁺ T-, cDC1-, and B-cell crosstalk (figure 3M).⁴⁰

Triple therapy remodels the immune landscape via increased antigen-presenting cDC1s and B cells

In the tdLN, triple therapy increased the frequency of activated and resident cDC1s relative to control and significantly increased migratory CD86⁺ cDC1s relative to all other groups, consistent with enhanced antigen presentation and T cell priming (figure 4A–C). Within the tumor, scRNA-Seq analysis of migratory cDC1s (Clec9a⁺Itgax⁺Itgae⁺) revealed a significant upregulation of select activation (Cd86, H2-Eb2, Mki67, others) and migration (Batf3, Itgae, others) genes, along with a decrease in select inhibitory gene expression (Mif, Smad1), with triple therapy relative to PBS control and/or doublet treatments (figure 4D)⁴¹. These transcriptional changes were accompanied by enrichment of TNFα and regulation of T cell activation signaling pathways, mirroring CD8⁺ T cell responses (figure 4E). DEG and GSEA analyses further corroborated enhanced cDC1 function, with increases in activation, antigen presentation, and major histocompatibility complex (MHC)-II genes and pathways required for tumor rejection (figure 4F,G). Together, coordinated activation and accumulation of cDC1s in the tdLN and TME suggest enhanced CD8⁺ T cell-cDC1 crosstalk following triple therapy.

Figure 4. Triple therapy remodels the immune landscape via increased antigen-presenting cDC1s and B cells. (A–D) Flow cytometry of tdLN cDC1s, (E–G) scRNA-Seq of tumor cDC1s, and (H, I) tumor B cells from CT26 tumor-bearing mice. (A) Activated DCs (MHCII⁺CD11c⁺), (B) CD8⁺ resident cDC1 (CD24⁺ MHCII⁺CD11c⁺) with CD86⁺ resident cDC1, and (C) tdLN CD103⁺ migratory cDC1 with CD86⁺ migratory cDC1 as frequencies of CD45⁺ cells in tdLN (n=12 mice/group; combined from two independent experiments). (D–G) Downstream analysis of tumor cDC1 (Clec9a⁺Cd11c⁺Cd103⁺) scRNA-Seq clusters. (D) Tumor cDC1 gene expression representing activation, trafficking/migration, and inhibitory phenotypes for designated treatments; starred genes indicate DEG (BH test/FDR <0.05; fold change >±1.2) relative to PBS (black), ENT+αPD-1 (blue), or ENT+N803 (purple). (E, G) GSEA for triple therapy-treated tumor DCs versus PBS control. (E), NES by ranked genes for designated pathways. (F) DEG (BH/FDR <0.05) in triple therapy-treated tumor cDC1s versus PBS control. (G) Selected upregulated GO, KEGG, and Hallmark gene pathways with triple therapy in tumor cDC1s versus PBS. (H) CD19⁺Ki67⁺ B cells as percent CD45⁺ cells and cells/mg CT26 tumor (n=16 mice/group; combined from three independent experiments). (I) Serum CXCL13 correlation with CT26 tumor B cell infiltration from designated cohorts (n=6 mice/group). Red dots in (F) represent individual genes (fold change >±1.2 /p<0.05). (A–C, H) mean±SEM; one-way ANOVA/Tukey’s multiple comparisons test; *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001, ns=non-significant. ANOVA, analysis of variance; BH, Benjamini-Hochberg; cDC1, type 1 conventional dendritic cells; DEG, differentially expressed genes; ENT, entinostat; FC, fold change; FDR, false discovery rate; GO, gene ontology; GSEA, gene set enrichment analysis; KEGG, Kyoto Encyclopedia of Genes and Genomes; MHC, major histocompatibility complex; NES, normalized enrichment score; PBS, phosphate-buffered saline; tdLN, tumor-draining lymph nodes.

Given the marked increase in BM CLPs, elevated CXCL13, and inferred crosstalk, we examined B cells in the TME. Although proliferative B cells (Ki67⁺CD19⁺) were significantly increased with entinostat plus αPD-1 and triple therapy, only triple therapy strongly correlated with serum CXCL13 levels, consistent with enhanced homing and lymphoid organization in the TME (figure 4H,I). Transcriptomic analysis of tumor B cells (Cd3^negCd79b⁺) from triple therapy-treated mice revealed increased Cxcr4 and Mzb1 gene expression, consistent with antibody-secreting cells (ASCs). Other markers of activation and transition were also noted (online supplemental figure S5B).⁴² Transcriptomic analysis of B cells in tdLN (Cd79b⁺Cd19⁺) revealed increased expression of Ly6 family genes including Ly6a, and MHC-I genes relative to PBS control, corroborating enhanced activation and antigen presentation (online supplemental figure S5C).⁴³ Further comparison of triple therapy-treated proliferating B cells (Cd79b⁺Mki67⁺) versus entinostat plus αPD-1 revealed increased expression of immunoglobulin variable region and ASC genes (Xbp1, Mzb1) (online supplemental figure S5D).⁴² Proteomic quantification of serum and tumor lysate confirmed broad activation of humoral response on treatment with triple therapy (online supplemental figure S5E,F). These findings demonstrated that triple therapy induces activation of and antigen presentation by B cells and cDC1s.

Triple therapy-induced immune modulation gives rise to TCF1⁺CD8⁺ immune niches

The coordinated increases in tumor TCF1⁺CD8⁺ T cells, cDC1s, B cells, and chemokine milieu including CCL5 and CXCL13 with triple therapy are indicative of TLS induction, increasingly shown as a strong predictor of CPB response.^{19 25 35 44} Using multiplex immunofluorescence (mIF), we examined tumor sections for CD8⁺ T cells, migratory DCs, and B cells (Panel 1: CD8, CD11c, CD103, CD19, B220, CXCL13) to investigate the ability of triple therapy to induce TLS-like structures in CPB-resistant tumors (figure 5A, B). While clusters of B cells were identifiable in all groups, these cells were mostly dispersed. Cell clusters including CD8, CD11c, and CD19 markers were identifiable in all entinostat-treated groups, with larger and tighter structures observed only with triple therapy (figure 5B; online supplemental figure S6A,B). These less mature immune clusters devoid of a TLS-defining germinal center were observed only at tumor edges, as seen with “excluded” TLS.⁴⁴ In order to quantify these observations, we first measured whole section marker densities (cells/mm²) per treatment and observed that triple therapy was significantly increased for CD8 relative to all single agents and N803-double agents (figure 5C). Only triple therapy-treated tumors reached a significant increase from PBS control in the expression of CD11c and CD19 markers. TCF1⁺CD8⁺ T cell persistence in intratumoral TLS-like niches co-localized with differentiated CD8⁺ T cells has been associated with improved responses to CPB.^{18 19 25 45} To evaluate TCF1 expression in our aggregates, we examined the same tumors used for Panel 1 for stem-like TCF1⁺CD8⁺ T cells, cDC1s, and B cells (Panel 2: CD8, TCF1, PD-1, GZMB, CLEC9a, CD19) (figure 5D, E). TCF1⁺ and GZMB⁺ CD8⁺ T cells were found co-localized only within immune clusters in entinostat combinations, with tight aggregates co-located with CLEC9a exclusive to triple therapy, corroborating TCF1⁺ expression associated with therapeutic response (figure 5E; online supplemental figure S6C). Spatial analysis demonstrated triple and entinostat+N803 therapies to attain the highest B/DC/CD8 cell number thresholds in immune aggregates, with triple therapy eliciting an increased DC/CD8 ratio relative to entinostat+N803 (online supplemental figure S6D,E). Notably, spatial analysis confirmed that centrally located nuclei corresponded to cDC1s, which were significantly more abundant and positioned closer to CD8⁺ T and B cells in triple therapy-treated tumors (figure 5F). Analysis of TCF1⁺CD8⁺ T cells revealed no significant differences in proximity to cDC1 or B cells between triple and entinostat+N803 therapies (online supplemental figure S7A,B). However, immune niche composition of CD8⁺ T cells differed markedly from the remaining treatment groups, with entinostat-containing regimens exhibiting CD8-rich regions of interest (ROIs) composed of TCF1⁺ and/or GZMB⁺ CD8⁺ T cells, whereas other treatment groups contained fewer overall CD8⁺ T cells and little to no TCF1⁺ or GZMB⁺ CD8⁺ T cells (online supplemental figure S7C). Consistent with these findings, the abundance of CLEC9a⁺ cDC1s within immune niche ROIs strongly correlated with the presence of TCF1⁺CD8⁺ T cells (online supplemental figure S7D). Given the central role of cDC1s in aggregate formation and in orchestrating communication between T and B cells during triple therapy, we utilized Batf3^−/− mice that lacked resident CD8⁺ cDC1s required for cross-presentation.²⁸ Using the MC38 Jak1 KO model as described in figure 1, we observed significant loss in anti-tumor activity elicited by combination therapy relative to wild type (WT) mice (figure 5G–I). Peripheral blood analysis showed a marked reduction in activated GZMB⁺ and EM CD8⁺ T cells in response to triple therapy, accompanied by a relative increase in CM CD8⁺ T cells. These findings are consistent with impaired cDC1-dependent priming and skewing away from terminal effector differentiation, further supporting that cDC1s are required for mediating cytotoxic lymphocyte immunity induced by triple therapy (figure 5J,K; online supplemental figure S7E). No changes to proliferative or naïve compartments were observed relative to WT mice. Despite evidence of triple therapy-induced humoral activation, including increased proliferative B cells and immunoglobulin production, antitumor efficacy was unchanged in CT26-bearing Jh mice lacking mature B cells. This indicated that B cells were not required for therapeutic response (online supplemental figure S7F). Collectively, these findings indicate that triple therapy promotes the accumulation and spatial organization of B cells, cDC1s, and CD8 T cells within the TME characterized by higher densities and tighter spatial architecture, with anti-tumor responses heavily reliant on CD8⁺ T cells and to a lesser degree cDC1. Altogether, our data report that triple therapy increases primed stem-like TCF1⁺ and GZMB⁺ CD8⁺ T cells that co-localize with cDC1 and B cells within intratumoral immune niches, supported by cytokine and chemokine signaling and a tdLN reservoir, thereby prolonging anti-tumor responses in otherwise CPB-resistant tumors. These effects required the cooperative action of all three agents.

Figure 5. Triple therapy-induced immune modulation gives rise to TCF1⁺CD8⁺ immune niches. (A) Representative multiplex immunofluorescence (mIF) images show triple therapy-induced immunity niches (white square) with inset showing matched H&E, scale bar=400 µm. (B) Higher magnification of immunity niche shows tightly aggregated clusters of CD11c⁺ (cyan) and CD19⁺ (magenta) cells with overlay demonstrating presence of all markers in Panel 1, scale bar=50 µm. (C) Quantification of designated cells/mm² of tumor section. (D, E) Representative mIF shows panel 2 overlay of CD19⁺ (magenta), CLEC9a⁺ (cyan), TCF1⁺ (red), GZMB⁺ (white), PD-1⁺ (orange), and CD8⁺ T cells (green) within the same cluster as in (B), scale bar=10 µm, 63× magnification. (E) Representative mIF showing interactions between TCF1⁺ (red), GZMB⁺ (white), CD8⁺ T cells, and CLEC9A⁺ DCs (cyan), scale bar=10 µm, 63× magnification. (A–F) n=3‒6 mice/group; panels represent one independent experiment. (F) Histograms showing spatial imaging algorithm-based distances between DCs and B cells or CD8⁺ T cells in tumor ROIs that reached cell number thresholds. (G) Treatment schedule and study summary of MC38 Jak1 KO tumor-bearing C57BL/6 WT and Batf3^−/− mice for (H–K). (H) Aggregate and (I) individual tumor growth curves in WT versus Batf3^−/− mice. (J–K) Peripheral blood (day 15) flow cytometry analysis from triple therapy-treated mice versus PBS control (n=7 mice/group). (J) Activated (CD44⁺) CD8⁺ T cells as percent CD45⁺ and associated GZMB expression. (K) EM, CM, and naïve cells as frequency of CD8⁺ T cells. (C, F, H–K) Mean±SEM; (C) one-way ANOVA with Tukey’s multiple comparisons test; (F, H) two-way ANOVA with multiple comparisons test; (I–K) paired t-test; *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001, ns=non-significant. (G) Created in BioRender. Minnar, C. (2026) https://BioRender.com/93tuiql. ANOVA, analysis of variance; CM, central memory; DC, dendritic cells; EM, effector memory; ENT, entinostat; KO, knockout; mIF, multiplex immunofluorescence; PB, peripheral blood; ROI, region of interest; WT, wild type.

Triple therapy associates with clinical benefit and activates patient-derived CD8⁺ T cells in vitro

To interrogate whether triple therapy-induced transcriptional programs in mice are associated with clinical outcomes in human cancer datasets, we probed expression data from The Cancer Genome Atlas (TCGA). Analysis of the TCGA colorectal adenocarcinoma (COAD) dataset demonstrated CCL5 to be highly correlated with cytotoxic T cell (CTL) infiltration (figure 6A). Recent findings linked tumor immune cluster formation with stem-like T cells, highlighting cDC1 presence as ideal niches for differentiation of stem-like T cells.^{20 24} Using gene markers elevated with triple therapy in vivo, we observed high correlation between cDC1 (CLEC9A, ITGAE, CD86, HLA-DP) and stem-like T cell (CD8A, TCF7, TNF, CCL5) gene signatures in COAD datasets (figure 6B). We identified a gene signature representative of triple therapy-induced stem immunity niches (TCF7, CD8A, TNF, CCL5, ITGAE, CLEC9A, NCR1, CD19) in vivo to further probe patient survival and clinical response. Across multiple tumor types, triplet signature genes were broadly associated with improved survival (−log₁₀ HR), although the strength and consistency of this association varied by cancer type (figure 6C). Of note, XCR1 was used in place of CLEC9A for CRC dataset. We first interrogated if triple therapy-response gene signature was associated with clinical outcomes in patients with CRC, KRAS-mutated CRC, and BRCA cohorts, and found significant increases in survival (figure 6D,E). To interrogate if triple therapy-response gene signature was associated with clinical outcomes in patients who had received CPB, we applied it to αPD-1-treated pan-tumor (bladder, glioblastoma, and melanoma) and melanoma-only datasets and found significant survival benefit with median survival increases of 15.38 and 14.57 months, respectively (figure 6F). Another dataset from αPD-1–treated NSCLC, head and neck squamous cell carcinoma, and melanoma patients (GSE93157) also demonstrated significantly improved survival with high signature expression (figure 6G).⁴⁶ Within this dataset, triple therapy-associated CCL5 and TCF7 genes were highly correlated with increased CTL infiltration and patient overall survival, with significantly increased tumor TNF expression in responders versus non-responders (figure 6H,I; online supplemental figure S7G). Lastly, since CT26 exhibits an inflamed tumor phenotype, we analyzed a dataset from αPD-1-treated T cell-rich hepatocellular carcinoma patients (GSE206325).⁴⁷ CD8⁺ T cell subset analysis revealed CD8⁺ T EM and, to a lesser degree, CM cell subsets were associated with higher expression of GZMB, CCL5, IFNG, and TNF in responders versus non-responders (figure 6J; online supplemental figure S7H). Notably, IL7R and LY6E genes associated with CLP were also significantly increased in responders versus non-responders (online supplemental figure S7I). Our results capture a strong association with triple therapy transcriptional program and improved clinical outcomes. However, clinical validation in CRC-specific immunotherapy cohorts will be required to further validate our findings.

Figure 6. Triple therapy predicts clinical benefit and activates patient-derived CD8⁺ T cells in vitro. (A) Correlation of CCL5 expression to CTL infiltration in COAD (TIDE). (B) Stem-like T cell gene signature (CD8, TCF7, TNF, and CCL5) correlation with antigen-presenting cDC1 signature (CLEC9A, ITGAE, CD86, and HLA-DP) in COAD (GEPIA2). (C) OS log10 HR of triple therapy response genes (signature) across multiple cancer types (GEPIA2); log10(HR)<0 (blue squares) and log10(HR)>0 (red squares) denote improved and decreased risk of survival benefit, respectively. (D, E) Patient OS prognosis relative to triple therapy response gene signature for designated CRC cohorts and BRCA (KM plotter platform). (F) Patient OS prognosis relative to triple therapy response gene signature in designated αPD-1-treated patient cohorts (KM plotter platform). (G) OS (GSE93157), (F) CCL5 and TCF7 gene survival analyses, and (I) TNF gene expression in responders versus non-responders for αPD-1-treated NSCLC/HNSCC/melanoma patient cohorts (TIDE and GSE93157). (J) Gene expression analysis of CD8⁺ T cell subsets in responders versus non-responders in αPD-1-treated HCC patient cohort (GSE206325). (K) Summary of in vitro treatment schedule of PBMCs for panels (L–N),where previously cryopreserved healthy donor (HD; n=7) and CRC (n=5) patient PBMCs were treated as described and evaluated by flow cytometry. (L) GZMB expression (gMFI) on CD8⁺ T cells with representative histogram. (M) GZMB expression as frequency of CD8⁺ T cells and as cell surface expression (gMFI) with representative histogram from CRC patient PBMCs. (N) Ki67 and CCL5 expression as frequency of CD8⁺ T cells from CRC patient PBMCs. Data representative of 2 independent experiments (repeated on same donors) (HD) or one independent experiment (CRC). Bar and violin plots show mean±SEM; one-way ANOVA with Tukey’s multiple comparisons test; pairwise comparisons show paired t-test; *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001, ns=non-significant. TIDE, GEPIA2, KM plotter, and patient global gene expression datasets (GEO) were used for producing immune correlations, signature and gene-associated HRs, and gene expression profiles of αPD-1 responders and non-responders for human cancers. (I) Created in BioRender. Minnar, C. (2026) https://BioRender.com/wg9f1b5. ANOVA, analysis of variance; BRCA, breast invasive carcinoma; COAD, colon adenocarcinoma; CRC, colorectal carcinoma; CTL, cytotoxic T lymphocyte; EM, effector memory; ENT, entinostat; gMFI, geometric mean fluorescence intensity; HCC, hepatocellular carcinoma; HD, healthy donor; HNSCC, head and neck squamous cell carcinoma; HR, hazard ratio; KM, Kaplan-Meier; NSCLC, non-small cell lung cancer; OS, overall survival; PBMC, peripheral blood mononuclear cells; TIDE, tumor immune dysfunction and exclusion; TNF, tumor necrosis factor.

αPD-1 activity relies on a spectrum of direct and indirect immune interactions within the TME.^{21 23 35} To extend insight into the translational potential of triple therapy, we used HD and cancer patient PBMCs in vitro to isolate the direct effects of triple therapy on the activation and proliferation of CD8⁺ T cells in the absence or presence of CD3/CD28 stimulation (figure 6K). Triple combination led to significantly higher expression of GZMB on HD-stimulated CD8⁺ T cells (figure 6L). As expected, TCF1⁺ populations decreased in unstimulated CD8⁺ T cells treated with triple therapy in vitro, consistent with our observations of therapy-driven activation and differentiation towards effector T cells and indicating that TCF1 enrichment observed in vivo is beyond a direct effect on Tcf7 transcription (online supplemental figure S7J). Consistent with enhanced cytotoxicity, triple therapy increased H3K27ac expression, an effect driven by entinostat (online supplemental figure S7K). H3K27ac expression levels were strongly correlated with GZMB expression in triple therapy-treated cells, suggesting epigenetic remodeling associated with enhanced CD8⁺ T cell cytotoxicity (online supplemental figure S7L). Analysis of CRC patient PBMCs demonstrated GZMB⁺CD8⁺ T cell frequency to be significantly increased with triple therapy and N-803 doublets; however, surface expression of GZMB was highest with triple therapy alone relative to control (figure 6M). Additionally, triple therapy alone elicited significant increases in both Ki67⁺- and CCL5-producing CD8⁺ T cells relative to PBS control with variable increases relative to doublets (figure 6N). Similar results were observed in breast cancer patient PBMCs (online supplemental figure S7M and N). Collectively, these data show that triple therapy-induced immune niche signature observed preclinically is associated with improved clinical responses, suggesting potential benefit in CPB-refractory CRC.³ Supporting this, triple therapy increased CD8⁺ T cell cytolytic and proliferative activity in HD and patient PBMCs over mono-agent or double-agent therapy.

Discussion

Accumulating evidence indicates that αPD-1 efficacy requires multicellular coordination involving cytotoxic and stem-like TCF1⁺CD8⁺ T cells, effective antigen presentation, and a supportive cytokine-chemokine milieu. When most mature, such response-associated immune aggregates have been described as TLS, with less mature non-TLS aggregates designated niches or “hubs”.^{18 25 45} Strategies that promote this coordinated immune activation within the TME are increasingly recognized as critical for αPD-1 response. Here, we provide the first evidence that TCF1⁺CD8⁺ T cells in tdLN and intratumoral immune niches composed of cDC1s, TCF1⁺CD8⁺ T cells, GZMB⁺CD8⁺ T cells, and B cells can be induced in αPD-1-refractory tumors through the addition of entinostat and N803 to αPD-1 immunotherapy. Proteomic, single-cell transcriptomic, spatial, and depletion analyses show that triple therapy induces these features in the CT26 model, whereas single-agent and double-agent therapies lack anti-tumor robustness and the presence and/or fitness of all immune niche components. Depletion and Batf3 knockout studies confirmed that CD8⁺ T cells and cDC1s are required for triple therapy efficacy, in line with our findings demonstrating triple therapy to induce higher intratumoral CD8⁺ T cells, including proliferative and cytolytic subsets, in parallel with increased infiltration of migratory cDC1 with a transcriptomic enrichment supportive of augmented antigen presentation and functional fitness. These findings are consistent with recent reports identifying cDC1s as central to the development and maintenance of intratumoral immune aggregates in mice and patients.^{48 49} Analysis of human cancer genomic datasets, including from αPD-1-experienced patients, and human PBMCs further supports the potential translational relevance of this combination strategy.

Clinical studies combining entinostat with CPB have reported enhanced CD8⁺ T and B cell activation, consistent with our findings.^{6 13} Although entinostat plus αPD-1 elicited anti-tumor responses, treatment with entinostat plus N803 or triple therapy was associated with augmented peripheral CD8⁺ T cell activation and reservoir of stem-like TCF1⁺CD8⁺ T cells in tdLN, increasingly associated with robust anti-tumor responses with CPB.⁵⁰ Triple therapy-expanded TCF1⁺TIM3^neg CD8⁺ T cells within tdLN exhibited effector-associated features while lacking canonical exhaustion markers distinguishing them from previously described progenitor-exhausted (Tpex) phenotypes.^{23 33} Consistent with prior reports identifying stem-like TCF1⁺CD8⁺ T cells, we also observed expression of Ccl5 and Xcl1 supporting their role in cDC1 recruitment.^{20 21 25} TCF1⁺TIM3^neg CD8⁺ TILs expressed stem-like memory signature genes and proteins, notably Bach2, a transcriptional enforcer of long-term stemness, self-renewal, and exhaustion restraint.³⁶ These cells lacked Ccl5 or Xcl1 expression, suggesting a non-migratory role in the TME. Simultaneously, IFNγ-inducible SCA1 expression was elevated on T and B cells with concomitant increases in BM CLPs with triple therapy treatment. Given the central role of IFNγ/IL-15 signaling and HDAC-mediated epigenetic regulation of hematopoiesis, further investigation into their translational relevance in solid tumors is warranted.^{26 51 52} Notably, we recently reported similar expansion of stem-like T cell populations following IL-12-based cytokine therapy in cancer patients, underscoring the potential of cytokine-driven interventions to reinforce durable anti-tumor immunity.⁵³ Of note, triple therapy did not increase TCF1⁺ expression in unstimulated CD8⁺ T cells from HD PBMCs in vitro, further supporting in vivo findings showing expansion and maintenance of stem-like T cells is associated with remodeling of lymphoid compartments.

TNFα is a strong stimulator of chemokines driving leukocyte recruitment, including CCL5.³⁴ Here, CCL5 expression was significantly increased in tdLN and tumor CD8⁺ T cells, with serum TNFα strongly correlated with tdLN TCF1⁺CD8⁺ EM cells. This indicated an overall mechanism of action reliant on activation-induced TNFα. The TNFα signaling pathway was also enriched in cDC1s. Given that recruitment of tumor-infiltrating cDC1s has been attributed to T cell-derived XCL1 and CCL5 expression, and both CCL5 and TNFα were markedly increased with triple therapy, TNFα secretion likely contributed to coordinated crosstalk among these cell types.²⁴ Of note, cDC1s are increasingly implicated in the maintenance and activation of TCF1⁺ stem-like cells.^{20 24} Additional studies are needed to resolve the role of TNFα in crosstalk between triple therapy-driven TCF1⁺CD8⁺ T cells, cDC1s, and other TNFα-producing myeloid populations.

We have previously shown that entinostat synergizes with cytokine-based therapies to improve cure rates in CPB-sensitive and CPB-resistant murine tumor models.^8–10 In these studies, combining entinostat with N-803 and vaccine enhanced cytolytic CD8⁺ T cell responses and established a cytokine-chemokine milieu associated with effective anti-tumor immunity.⁹ Similar immune remodeling was observed with entinostat plus tumor-targeted IL-12 therapy in CPB resistant models, including simultaneous increases in tumor-infiltrating cDC1s and B cells.¹⁰ Together, these findings propose a model in which entinostat combination with cytokine activation reprograms the TME into a highly interactive immune network. Consistent with this framework, we observed pronounced induction of cytokines and chemokines with triple therapy in the highly inflammatory CPB-resistant CT26 model. In particular, increased IFNγ, CCL5 expression in serum, and a profound increase of CXCL13 in tumor as well as transcriptomic inference of cell signaling interactions prompted further interrogation of cDC1 and B cell interactions, leading to the identification of immune niches analogous to those reported in human CRC and lung cancer specimens.^{19 25 45} Here, spatial analysis further revealed that tightly organized immune niches comprising CD8 and B cells surrounding cDC1 nuclei specifically occurred in response to triple therapy. At the time point of tumor collection, distance analysis within these immune niches indicated similar proximity between TCF1⁺CD8⁺ T cells and cDC1s with both triple therapy and entinostat plus N-803. These findings support a model in which enhanced antitumor efficacy of triple therapy, including versus entinostat doublets, resulted from a convergence of coordinated mechanisms favorable to tumor suppression, including an enriched intratumoral chemokine milieu and increased cDC1s, CD8,⁺ and TCF1⁺CD8⁺ T cell densities. Consistent with mounting evidence in clinical samples, our observations noting the presence of tighter CD8/DC tumor immune niches in mice treated with triple therapy and inferring enhanced CD8⁺ T cell-cDC1 crosstalk supports a path of favorable TCF1⁺CD8⁺ T cell differentiation into the observed expanded pool of cytolytic CD8⁺ T cells via interaction with cDC1s displaying a transcriptomic profile consistent with increased antigen presentation and functional fitness.^54–56

Although triple therapy significantly induced B cell activation, accumulation, and increased immunoglobulin production, therapeutic efficacy was retained in B-cell-deficient Jh mice, supporting our findings that cDC1s are the determinant antigen-presenting cell in response to triple therapy. Similar findings were recently reported in studies of TLS, suggesting that the contribution of B cells to antitumor immunity may vary with tumor model and immunotherapeutic approach.⁴⁸ These findings are also consistent with mounting evidence indicating that B cells, including TLS-resident, can promote but also dampen antitumor immunity via IL-10 secretion, secretion of autoreactive antibodies in detriment of tumor antigen specificity, or antibody engagement with inhibitory Fc receptors.^{44 57} Additional studies are required to clarify the contribution and phenotypic diversity of B cell populations in immune aggregates. Collectively, these findings underlie the dynamic cellular interactions and CD8 differentiation within these clusters, with T cells differentiating to cytolytic effectors, and highlight the importance of cytokine-driven immune architecture in remodeling the TME. Going forward, it will be important to assess the role of high endothelial venules, innate immune cells also activated by N803, and myeloid cells that may further shape the formation and function of intratumoral immune niches. Collectively, these data suggest that the addition of N803 to entinostat and αPD-1 promotes the generation, recruitment, and sustenance of stem-like CD8⁺ TILs, enabling improved tumor control through coordinated action of epigenetic, cytokine, and checkpoint modulation. Planned clinical studies will evaluate safety and clinical outcomes with the addition of N-803 to HDACi/αPD-1 therapy.

Data availability statement

Data are available in a public, open access repository. Data are available on reasonable request. All data relevant to the study are included in the article or uploaded as supplementary information.