Baseline tumor features and systemic immune dynamics underlying efficacy in MSS metastatic colorectal cancer treated with regorafenib, ipilimumab, and nivolumab

Abstract

Microsatellite-stable metastatic colorectal cancer (MSS mCRC) remains resistant to conventional immunotherapies. In a phase I trial, we observed encouraging efficacy of the combination of regorafenib, ipilimumab, and nivolumab (RIN), with a 27.6% overall response rate and a median overall survival of 20 months. The most pronounced benefits were observed in patients without liver metastases. To uncover immunological mechanisms underlying response and resistance, we performed correlative studies of the tumor microenvironment (TME) and systemic immune features. Tumor biopsies from eight patients and peripheral blood samples from 29 patients with MSS mCRC were collected and analyzed at baseline and during treatment. At baseline, tumors from good responders exhibited enhanced proliferation, DNA repair pathways, and STING expression, while poor responders showed enrichment of complement and metabolism pathways. In peripheral blood, good responders had a higher CD4/CD8 T cell ratio, increased dendritic cells, and intact type 1 cytokine responses. In contrast, poor responders exhibited more effector T cell differentiation, elevated immune checkpoint molecule expression, and increased DNA damage in lymphocytes. In good responders, RIN therapy increased tumor-infiltrating lymphocytes and upregulated immune activation genes, accompanied by heightened T cell proliferation and activation in peripheral blood, including expansion of low-frequency TCR clones in CD8⁺ T cells. Patients with liver metastases exhibited T cell senescence and metabolic hyperactivation, correlating with therapeutic resistance. These findings highlight that pre-existing tumor immunogenicity and T cell functional capacity are associated with response to RIN therapy, and that RIN treatment may facilitate both systemic T cell activation and local TME modulation.

Introduction

Colorectal cancer (CRC) is the third most common cancer and the second leading cause of cancer-related deaths globally. Most CRC cases are microsatellite-stable (MSS) and proficient in mismatch repair. MSS CRC, in particular, has demonstrated poor response to immune checkpoint inhibitors (ICIs), which have been successful in treating other malignancies (1–4). Such resistance to ICIs represents a major therapeutic challenge, underscoring the need for innovative strategies to sensitize MSS metastatic colorectal cancer (mCRC) to immunotherapy.

Emerging research highlights the potential of tumor microenvironment (TME) modulation as a strategy to overcome immune evasion in MSS mCRC. Regorafenib, a multi-kinase inhibitor with antiangiogenic properties, has demonstrated the ability to normalize aberrant tumor vasculature, relieve hypoxia, and reduce immunosuppressive elements within the TME (5–8). These effects create a more permissive environment for immune cell infiltration and activation. When combined with ICIs, regorafenib holds the potential to overcome immunotherapy resistance in MSS mCRC.

We previously conducted a phase I trial combining regorafenib with ipilimumab (anti-CTLA-4) and nivolumab (anti-PD-1), collectively termed the RIN regimen. This combination yielded an overall response rate (ORR) of 27.6%, with a median progression-free survival (PFS) of 4 months and a median overall survival (OS) of 20 months. Notably, patients without liver metastases experienced a higher ORR of 36.4% and durable benefits, including a 3-year PFS rate of 19.3% (9, 10). These results suggest that RIN therapy may enable antitumor immunity, contributing to durable responses in a subset of patients with MSS mCRC. However, the mechanisms underlying differential clinical responses remain poorly understood, particularly the poor outcomes observed in patients with liver metastases. A better understanding of these mechanisms is critical for optimizing treatment strategies and identifying predictive biomarkers.

Both the TME and systemic immunity play critical roles in shaping response to immunotherapy. Tumor-intrinsic features such as immune cell infiltration, checkpoint expression, and suppressive cytokines influence therapeutic efficacy (11, 12). Liver metastases, in particular, impose an additional barrier due to the liver’s immune-tolerant microenvironment, which promotes T cell dysfunction and immune escape (13–15). Moreover, systemic immune factors, including the composition and functional state of peripheral immune cells, may also alter treatment response (16, 17).

In this correlative study, we investigated the immunological and molecular determinants of response to RIN therapy in MSS mCRC. By analyzing paired tumor biopsies and peripheral blood samples collected before and during treatment, we performed RNA sequencing, multiplex immunofluorescence imaging, high-parameter flow cytometry, and profiling of serum cytokines. Our objectives were to (1) identify baseline tumor and peripheral immune features associated with response, (2) characterize how RIN therapy modulates the TME and systemic immunity, and (3) uncover mechanisms contributing to poor outcomes in patients with liver metastases. Through integrated analysis of clinical, molecular, and immune data, this study aims to advance our understanding of immunotherapy resistance in MSS mCRC and guide the development of personalized strategies to improve outcomes in this difficult-to-treat population.

Materials and Methods

Patient Enrollment and Sample Collection

A total of 29 patients with MSS mCRC were enrolled in this study under the City of Hope Clinical Trial IRB #19417 (NCT04362839). All patients provided written informed consent prior to participation. The study was conducted in accordance with the ethical standards outlined in the Belmont Report and the Declaration of Helsinki. Eligible patients had histologically confirmed MSS mCRC and were suitable candidates for the combination therapy with regorafenib, ipilimumab, and nivolumab.

Tumor biopsies were collected at baseline and at week 4 after treatment initiation. Samples were immediately snap-frozen in liquid nitrogen for RNA sequencing or formalin-fixed, paraffin-embedded (FFPE) for multiplexed immunofluorescence imaging (mIF). Peripheral blood was collected at baseline, week 4, and week 8. Plasma was stored at −80°C, and peripheral blood mononuclear cells (PBMCs) were isolated using Ficoll-Paque density centrifugation (Sigma-Aldrich, GE17-1440-02) and cryopreserved in liquid nitrogen for downstream analyses.

RNA Extraction and Sequencing

Total RNA was extracted from snap-frozen tumor biopsies and from fluorescence-activated cell sorting (FACS)-sorted non-naïve CD4⁺ and CD8⁺ T cells using the RNeasy Mini Kit (Qiagen, 74104) and RNeasy Micro Kit (Qiagen, 74004), respectively, following the manufacturer’s instructions. Each RNA-seq sample represents an independent biological replicate derived from an individual patient biopsy or sorted cell population.

Libraries and sequencing were performed by Novogene as a service project. Tumor-biopsy RNA libraries were generated using an rRNA-depletion workflow, whereas RNA from blood-derived sorted T cells was prepared as poly(A)-selected mRNA libraries. Libraries were sequenced on an Illumina NovaSeq 6000 (RRID:SCR_016387) to produce 150-bp paired-end reads, with a target depth of >30 million read pairs per sample.

Raw sequencing data underwent standard quality control by the sequencing provider. Reads passing a minimum Phred quality score of Q30 were retained for downstream analysis. Adapter trimming and low-quality base removal were performed as part of the provider’s standard preprocessing pipeline prior to alignment.

Bioinformatics Analysis

Raw RNA-seq data were processed using Partek Flow cloud-based platform. Reads were aligned to the GRCh38/hg38 human reference genome using STAR v2.7.8a. Gene expression quantification was performed using HTSeq v2.0.3, and differential gene expression was analyzed using DESeq2 v1.30.1. Genes with an adjusted p-value (Benjamini-Hochberg correction) of less than 0.05 were considered significant. Gene set enrichment analysis (GSEA) was conducted using Broad Institute’s GSEA software v4.1.0 (RRID:SCR_003199) with MSigDB hallmark and KEGG gene sets. Cell-type deconvolution from bulk RNA-seq was performed with xCell (RRID:SCR_026446) (18). Consensus molecular subtypes (CMS) were assigned using the CMSclassifier R package (v1.0.0) (19). T cell receptor (TCR) repertoire analysis was conducted with TRUST4 v1.0.12 (RRID:SCR_026162) (20), and clonality diversity metrics were calculated using the Immunarch R package v0.9.1 (RRID:SCR_023089).

Flow Cytometry

PBMCs were stained using four antibody panels (Supplementary Table 4). For surface staining, cells were incubated with fluorochrome-conjugated antibodies. Intracellular cytokine staining was conducted after stimulation with PMA (50 ng/mL, Sigma-Aldrich, P8139) and ionomycin (1 μg/mL, Sigma-Aldrich, I0634) in the presence of Brefeldin A (BioLegend, 420601) for 4 hours at 37°C. Cells were fixed, permeabilized using the Foxp3/Transcription Factor Staining Buffer Set (Invitrogen, 00-5523-00) and stained for intracellular markers.

Senescence-associated β-galactosidase (SA-β-gal) staining was performed using the SPiDER-βGal Detection Kit (Dojindo, SG04) following the manufacturer’s instructions. Briefly, cells were pretreated with Bafilomycin A1 in complete medium for 1.5 hours at 37°C, followed by incubation with SPiDER-βGal working solution for 30 minutes prior to fixation and surface antibody staining.

Samples were acquired on a Cytek Aurora spectral flow cytometer (RRID:SCR_019826) and analyzed using SpectroFlo v2.2 and FlowJo v10.9 (RRID:SCR_008520). Standardized gating was used to identify immune subsets and activation states (Supplementary Figure 2–5).

Multiplexed Immunofluorescence Imaging

Multiplex immunofluorescence (mIF) was performed on FFPE tumor sections using the Akoya PhenoCycler CO-Detection by indEXing (CODEX) platform (RRID:SCR_023773) (21). A custom panel of 50 antibodies targeting immune and stromal markers was applied (Supplementary Table 5). Staining was conducted per the manufacturer’s protocol.

Imaging was performed on the PhenoCycler-Fusion system (Akoya Biosciences), which integrates a widefield fluorescence microscope, automated stage, and sCMOS detector, using a 20× objective lens. Image acquisition was performed using the instrument’s proprietary software with standardized exposure settings across samples. QuPath v0.5 (RRID:SCR_018257) (22) was used for image visualization and manual annotation of regions of interest.

Cytokine Profiling

Plasma cytokine levels were quantified using a 30-plex Luminex assay (MilliporeSigma, HCYTMAG-60K-PX30) at the City of Hope Core Facility. Assays were performed according to the manufacturer’s instructions using xMAP technology and analyzed on a Luminex 200 system (RRID:SCR_018025).

Statistical Analysis

All statistical analyses were performed in R v4.4.3 (RRID:SCR_001905) and GraphPad Prism v10 (RRID:SCR_002798). Two-group comparisons were performed using the unpaired Student’s t tests. Data are presented as mean ± standard deviation, as noted. Significance thresholds were: *p < 0.05; **p < 0.01; ***p < 0.001.

Study Approval

All biospecimens were obtained from participants who provided written informed consent under an IRB-approved protocol at City of Hope (IRB #19417).

Results

Clinical Trial Outcomes Reveal Variability in Patient Responses to RIN Therapy

Clinical efficacy of the RIN regimen was evaluated in a cohort of 29 patients with MSS mCRC at the recommended phase II dose. Details of study treatment and clinical outcomes have been previously reported (9, 10). Over a three-year follow-up period, we observed a durable clinical benefit in patients without liver metastases, with a median PFS of 5 months and a median OS of 27.5 months. These patients generally had more favorable outcomes, including a 36.4% partial response (PR) rate and 31.8% stable disease (SD) rate, according to RECIST 1.1 criteria (Figure 1A). In contrast, no responses were observed in patients with liver metastases, among whom 57.1% experienced progressive disease (PD) and 42.9% achieved SD, highlighting an important association between metastatic site and therapeutic efficacy (Figure 1A).

Figure 1. Clinical outcomes and study overview of RIN therapy in MSS mCRC patients.

(A) Waterfall plot depicting the best percentage change in tumor size from baseline for individual patients (n = 29), stratified by the presence (red bars) or absence (blue bars) of liver metastases. Responses are classified per RECIST 1.1 criteria as progressive disease (PD), stable disease (SD), partial response (PR), or complete response (CR).

(B) Schematic representation of the study design. Tumor biopsies and peripheral blood samples were collected at baseline and on-treatment (week 4 and/or week 8), followed by integrated molecular and immunological analyses to assess treatment-associated changes. Created in BioRender. Ye, J. (2025) https://BioRender.com/4hyjvtj

Comprehensive correlative analyses were conducted using tumor biopsies and peripheral blood samples collected throughout treatment. Tumor biopsies from 8 patients were obtained at pre-treatment and/or at week 4, when available and of sufficient quality. These samples were analyzed by whole transcriptome RNA sequencing and multiplexed imaging (CODEX). Peripheral blood samples from all 29 patients were collected at baseline, week 4, and week 8, and were used for flow cytometry and Luminex cytokine assays. Additionally, sorted peripheral T cells from selected patients were subjected to transcriptional analysis and T cell receptor (TCR) profiling. Together, these multimodal analyses enabled characterization of both the TME and systemic immunity associated with clinical outcomes (Figure 1B).

To enable interpretable correlative analysis, patients were stratified into “good” and “poor” responders. Patients who achieved either progression-free survival (PFS) ≥ 6 months or a confirmed PR were categorized as good responders, while all others were classified as poor responders (Supplementary Table 1). This definition was chosen to reflect clinically relevant benefit, capturing both durable disease control and objective tumor regression.

Baseline Tumor Microenvironment Predicts Response to RIN Therapy

To investigate how the TME influences response to RIN, we performed RNA sequencing on tumor biopsies from eight patients with varying clinical outcomes. t-distributed stochastic neighbor embedding (t-SNE) analysis showed tight clustering of paired baseline and week 4 biopsies from the same individual, but distinct separation between good and poor responders, suggesting inter-patient TME heterogeneity even before treatment (Figure 2A).

Figure 2. Baseline tumor microenvironment profiles predict response to RIN therapy in MSS mCRC patients.

(A) t-SNE plot of tumor gene expression profiles from patients treated with RIN therapy.

(B) Bar plot of the top-ranked KEGG pathways enriched among genes upregulated in good (n=5, green bars) and poor (n=3, magenta bars) responders at baseline, as identified by gene set enrichment analysis (GSEA). Bar length indicates normalized enrichment score (NES). Shown pathways represent a subset of the most significantly enriched gene sets (FDR < 0.25).

(C) Volcano plot of differentially expressed genes (DEGs) at baseline between patients with good (n=5) and poor (n=3) response. Genes significantly upregulated in good responders are shown in green; those upregulated in poor responders are in magenta. Key DEGs are labeled.

(D) GSEA plots showing enrichment of cell cycle, mismatch repair, and cytosolic DNA sensing pathways in good responders.

(E) Heatmap of the top 20 differentially expressed genes contributing to enriched pathways in good responders at baseline. Samples are ordered by PFS duration. Expression values are z-score scaled; red indicates higher expression, blue indicates lower.

(F) Representative CODEX multiplex immunofluorescence images of baseline tumor sections from a good and a poor responder, stained for Ki-67 (white), STING (green), pan-cytokeratin (magenta), and CD45 (red). Scale bars, 100 μm.

Gene set enrichment analysis (GSEA) of KEGG pathways showed that tumors from good responders were enriched in pathways related to cell cycle regulation, DNA replication and repair, and cytosolic DNA sensing. These findings were supported by elevated expression of proliferation-associated genes such as MKI67, CDK1, CDC20, and CCNB1 (Figure 2B–E, Supplementary Table 2). Several immune-modulatory genes were also elevated in good responders. These included CXCL11, a type I/II IFN-induced chemokine involved in effector T cell recruitment (23); NOS2, which encodes inducible nitric oxide synthase (iNOS) and mediates IFN-driven cytotoxicity (24); and DMBT1, a scavenger receptor-like protein involved in epithelial barrier function and immune activation (25) (Figure 2C). In line with this, two patients with prolonged PFS had immune-permissive colorectal cancer consensus molecular subtypes (CMS): Pt_10 (CMS3, metabolic subtype) and Pt_7 (CMS2, canonical subtype) (19). Collectively, these data suggest that proliferative stress and cytosolic DNA sensing signaling contribute to a more immunogenic and treatment-responsive TME in good responders.

In contrast, poor responders, whose biopsies were all taken from liver metastases, exhibited significant enrichment of pathways related to complement and coagulation cascades, xenobiotic metabolism, amino acid metabolism, and lipid processing (Figure 2B; Supplementary Figure 1A). Genes such as C3, C4A/B, C5, SERPING1, SERPIND1, CYP2E1, FMO3, and ACSL1 were markedly elevated. These pathways are linked to complement-driven inflammation, metabolic detoxification, and adaptation to the nutrient-rich hepatic environment, all of which have been implicated in promoting immune evasion and therapy resistance (26, 27).

Tissue mIF analysis provided support for the transcriptomic findings. A baseline tumor from a representative good responder with the longest PFS demonstrated extensive nuclear Ki-67 staining colocalized with pan-cytokeratin, indicating active tumor cell proliferation (Figure 2F). This tumor also exhibited strong STING expression within tumor epithelial cells, suggesting activation of the cGAS-STING pathway. Elevated proliferation can result in replication stress and cytosolic DNA accumulation via DNA breakage or micronuclei formation. The presence of cytosolic DNA can activate cGAS-STING signaling, which promotes type I interferon production, enhances antigen presentation, and facilitates recruitment of cytotoxic lymphocytes (28). In contrast, a representative poor responder showed lower Ki-67 staining in tumor cells and STING expression restricted to CD45⁺ immune cells. Consistent with these observations, immune subset deconvolution and GSEA of transcriptomic data revealed that patients with prolonged PFS exhibited greater lymphoid infiltration and enrichment of lymphocyte activation pathways, whereas early progressors displayed myeloid-skewed immune profiles (Supplementary Figure 1B–E). Taken together, these integrated transcriptomic and histological findings suggest that a highly proliferative, interferon-activated TME, potentially driven by replication stress and cytosolic DNA sensing signaling, favors good responses to RIN therapy. In contrast, tumors from poor responders, particularly those in the liver, exhibited complement activation and upregulation of metabolic pathways associated with immune evasion and therapeutic resistance.

Peripheral Immune Biomarkers Predict Response to RIN Therapy

We comprehensively profiled pre-treatment peripheral blood mononuclear cells (PBMCs) from patients undergoing RIN therapy using multiple high-parameter flow cytometry panels quantifying immune cell subset composition and phenotype (Panels 1 and 2), cytokine production potential (Panel 3), and markers of immune exhaustion and senescence (Panel 4) (Figure 3A, Supplementary Figure 2–5, Supplementary Table 4). This analysis revealed distinct baseline immune signatures associated with clinical outcomes.

Figure 3. Peripheral immune biomarkers predict response to RIN therapy.

(A) Flowchart illustrating the blood collection and flow cytometry analysis comparing good responders (green) and poor responders (magenta) at baseline. Created in BioRender. Ye, J. (2025) https://BioRender.com/gq9ycl7

(B) Dot plots showing CD4/CD8 T cell ratio and the frequencies of CD4⁺ T cells, CD8⁺ T cells, Tc2 and Tc17 subsets, as well as dendritic cell populations (DC1 and DC2). Each dot represents an individual patient; horizontal lines indicate the median. Group comparisons were performed using unpaired two-sided Student’s t tests, with exact P values shown.

(C) Dot plot comparing immune cell subsets between good and poor responders. Data are shown as absolute cell counts, percentages of live cells, and percentages of parent populations. Color coding indicates higher levels in good responders (green) and poor responders (magenta).

(D) Representative flow cytometry plots and corresponding dot plots showing higher proportions of T_em cells in CD4⁺ and CD8⁺ compartments in poor responders.

(E) Dot plot of baseline cytokine production, immune activation markers, and checkpoint expression across lymphocyte populations.

(F) Dot plots showing the proportions of IFN-γ within CD8⁺ T_em and IL-4 within CD4⁺ T cells.

(G) Representative flow cytometry plots of Ki-67 staining in good and poor responders, with accompanying dot plots summarizing the results.

(H) Histograms showing γH2AX mean fluorescence intensity, along with corresponding dot plots highlighting group differences.

Good responders exhibited a higher CD4/CD8 T cell ratio and increased frequencies of both type 1 and type 2 dendritic cells (DC1 and DC2) (Figure 3B–C; Supplementary Figure 6A). In contrast, patients with poor outcomes had a greater total number of CD8⁺ T cells and higher proportions of effector memory (T_EM) subsets in both CD4⁺ and CD8⁺ compartments (Figure 3D), including higher frequencies of Tc2 and Tc17 cytotoxic subsets (Figure 3B–C). Monocyte subsets did not differ significantly between the two groups.

To assess functional capacity, we examined intracellular cytokine expression following ex vivo stimulation. CD8⁺ T cells from good responders had a higher frequency of IFN-γ⁺ cells, whereas those from poor responders showed increased proportions of type 2 (IL-4, IL-13) and type 3 (IL-22) cytokine-producing cells within both CD4⁺ and CD8⁺ compartments (Figure 3E–F; Supplementary Figure 6B). Poor responders also exhibited higher expression of immune checkpoint receptors (LAG-3, TIM-3) and the exhaustion-associated marker CD38, suggesting a more dysfunctional T cell phenotype (Figure 3E, Supplementary Figure 6B). This was accompanied by an increased percentage of Ki-67⁺ cells and γH2AX staining in lymphocytes (Figure 3G–H), particularly among patients with progressive disease (Supplementary Figure 6B–D), indicating increased proliferation and pre-existing DNA damage prior to treatment. Myeloid activation markers did not differ significantly between groups (Supplementary Figure 7A). However, plasma cytokine profiling revealed elevated soluble IL-2 receptor (sIL-2R) in poor responders, reflecting heightened systemic inflammation (Supplementary Figures 8).

To further dissect T cell phenotypes, we performed bulk RNA sequencing on sorted non-naïve CD4⁺ and CD8⁺ T cells (Supplementary Figure 9). CD4⁺ T cells from good responders were enriched for lipid metabolic pathways, consistent with a memory-like phenotype (29). In contrast, T cells from poor responders showed upregulation of pathogen-response pathways (e.g., bacterial, viral, and parasitic) and stress-related programs such as ferroptosis and necroptosis, indicative of chronic inflammation and disrupted immune homeostasis.

Together, these results indicate that patients who respond to RIN therapy have a baseline peripheral immune profile characterized by a higher CD4/CD8 T cell ratio, increased dendritic cells, and preserved type 1 cytokine responses. Conversely, poor responders exhibit skewed effector differentiation, type 2/3 cytokine polarization, and features of T cell exhaustion and DNA damage.

RIN Therapy Induces Immune Activation in the TME of Good Responders

To evaluate how RIN therapy modulates the TME, we compared the transcriptional profiles of paired pre-treatment and week 4 post-treatment tumor biopsies (Supplementary Table 3). Pre-treatment tumors exhibited elevated expression of genes involved in metabolic processes, including markers of xenobiotic metabolism (AKR1C2), glycolysis (ENO3) and insulin-like growth factor signaling (IGFBP3). Genes associated with tumor progression and invasion, such as MMP11, NCAM2, FN1, and NGFR, were also upregulated at baseline (Figure 4A–B).

Figure 4. RIN therapy induces immune activation in the tumor microenvironment of good responders.

(A) Bar plot of the top-ranked KEGG pathways significantly enriched among genes upregulated in tumor biopsies before (n=8, yellow) and after (n=6, purple) RIN treatment, identified by GSEA.

(B) Volcano plot of differential gene expression in tumor biopsies from all patients before and after RIN therapy. Genes upregulated pre-treatment (n=8) are shown in yellow, and those upregulated post-treatment (n=6) are in purple; selected key genes are labeled.

(C) GSEA plots illustrating significant enrichment of immune activation pathways post-treatment across all patients.

(D) Heatmaps of the top 15 genes contributing to enriched pathways. Paired pre- and post-treatment tumor samples are arranged side by side and ordered by best clinical response. Expression values are z-score scaled.

(E) Representative CODEX multiplex immunofluorescence images of tumor sections from a good responder (Pt_25) and a poor responder (Pt_13) at baseline and after 4 weeks of RIN therapy. Staining includes pan-cytokeratin (magenta), CD4⁺ T cells (cyan), CD8⁺ T cells (white), CD68⁺ macrophages (red), CD11c⁺ dendritic cells (green), and TCF1 (yellow); nuclei are counterstained with DAPI (blue). Scale bars, 100 μm.

In contrast, post-treatment biopsies showed significant enrichment of immune-related pathways, including antigen processing and presentation, T cell receptor signaling, natural killer cell-mediated cytotoxicity, and cytokine-cytokine receptor interactions (Figure 4A). Expression of T cell markers (CD3E, CD8A, TRAC), signaling molecules (ZAP70, LCK, ITK), and chemokines (CXCL9, CXCL13, CCL5) increased markedly, along with effector genes such as GZMA, indicating enhanced immune activation and recruitment (Figure 4B).

Paired sample analysis revealed that, although baseline immune profiles did not differ substantially between good and poor responders, the two patients with partial responses (Pt_25 and Pt_10) exhibited the most pronounced post-treatment increases in immune-related gene expression (Figure 4C–D). These transcriptomic changes were accompanied by increased infiltration of CD4⁺ and CD8⁺ T cells as well as CD68⁺ macrophages in the post-treatment tumors. In contrast, tumors from poor responders showed minimal changes in immune gene expression or immune cell infiltration (Figure 4E).

Although limited by sample size, these findings suggest that RIN therapy can enhance immune engagement within the TME of good responders. This shift was marked by transcriptional enrichment of antigen-processing/presentation pathways and increased expression of chemokines linked to immune cell recruitment, potentially contributing to therapeutic benefit.

Dynamic Modulation of Peripheral Immunity Correlates with Therapeutic Response

To delineate how RIN therapy shapes systemic immunity in MSS mCRC patients, we analyzed longitudinal changes in peripheral immune cells by comparing pre-treatment PBMCs with samples collected after the first (4 weeks) and second (8 weeks) treatment cycles. Good responders consistently maintained a higher CD4/CD8 T cell ratio, whereas poor responders exhibited persistently elevated CD8⁺ T cell levels across all time points (Figure 5A). When comparing percentage changes from baseline, poor responders showed a more pronounced expansion of lymphocytes, particularly CD8⁺ T cell and natural killer T (NKT) cell subsets, at week 4, whereas good responders displayed more moderate shifts in immune composition over time (Figure 5B–E, Supplementary Figure 10).

Figure 5. Dynamic modulation of peripheral immunity correlates with therapeutic response.

(A) Dot plots representing significant differences in immune cell subsets between good responders and poor responders after the first (4 weeks) and second (8 weeks) treatment. Data are presented in terms of absolute cell number, percentage of live cells, and percentage of parent population. Color coding indicates higher levels in good responders (green) and poor responders (magenta).

(B) Dot plots illustrating the relative percentage changes in immune cell populations of peripheral blood from good responders and poor responders. Changes are shown as percentages of live cells at 4 weeks minus baseline, 8 weeks minus 4 weeks, and 8 weeks minus baseline.

(C-E) Line graphs showing longitudinal changes in T cells (C), CD8⁺ T cells (D), and NKT cells (E). Each line represents an individual patient measured longitudinally. The relative changes between different time points are shown in the corresponding boxplots to the right.

(F) Dot plots illustrating the relative percentage changes in cytokine production and immune checkpoint expression of peripheral blood from good responders and poor responders. Dot color and size indicate direction and statistical significance, respectively.

(G-J) Line graphs depicting longitudinal changes in IFN-γ⁺ (G), IL-4⁺ (H), PD-1⁺ (I) and Ki-67⁺ (J) cells, comparing good responders and poor responders.

(K) TCR clonality analysis in CD8⁺ non-naïve T cells, highlighting significant clonal expansion in good responders during treatment compared to poor responders.

Analysis of cytokine-producing subsets and activation markers revealed divergent responses (Figure 5F–H, Supplementary Figures 11 and 12). Poor responders displayed a stronger Th1 profile, with increased frequencies of IFN-γ⁺ and TNF-α⁺ T cells (Figure 5G). By contrast, good responders showed milder Th1 induction but greater increases in IL-4⁺ and IL-13⁺ T cells (Figure 5H), and less reduction in exhaustion markers PD‐1, CTLA‐4, and TIM‐3 (Figure 5I). Myeloid activation markers, including CD83 and CD86, were upregulated from week 4 to week 8 in good responders (Supplementary Figure 13A–B). These contrasting dynamics likely reflect a pre-existing pro-inflammatory state in poor responders that favors rapid effector activation, while good responders may initiate a more attenuated immune response.

Despite their early lymphocyte activation and Th1 cytokine induction, poor responders showed limited expansion or restructuring of the TCR repertoire. In contrast, some good responders exhibited increased Ki-67 expression in naïve CD8⁺ T cells, suggesting new T cell recruitment or proliferation (Figure 5J). Consistent with this finding, these patients also displayed significant expansion of previously low‐frequency TCRs within non‐naïve CD8⁺ T cells, but not in CD4⁺ T cells (Figure 5K and Supplementary Figures 14 and 15). Together, these observations suggest a potential selective expansion of CD8⁺ T cell clones in good responders, which may contribute to enhanced antitumor immunity.

Transcriptomic profiling further emphasized these differences. Only good responders showed upregulation of immune activation pathways in both CD4⁺ and CD8⁺ T cells during treatment, including programs involved in host defense, antigen presentation, and T cell differentiation (Supplementary Figure 16). These signatures were absent in poor responders, underscoring their limited adaptive immune engagement and reduced clinical benefit.

Together, these findings suggest that RIN therapy elicits distinct peripheral immune responses in good versus poor responders. Poor responders display an early lymphocyte surge and Th1 cytokine production without sustained T cell expansion or transcriptional activation. In contrast, good responders show selective expansion of CD8⁺ T cell clones and activation of immune programs in both CD4⁺ and CD8⁺ compartments, which may underpin antitumor immunity in patients who benefit from RIN therapy.

Immune Senescence in Liver Metastases Correlates with Poor Response to RIN Therapy

We further investigated the immunological mechanisms underlying liver metastasis-associated poor response to RIN therapy in MSS mCRC patients by comparing peripheral immune profiles of patients with liver metastases to those of patients with lung metastases. Similar to the differences observed between good and poor responders, patients with liver metastases showed a lower CD4/CD8 T cell ratio that was attributable not to a loss of CD4⁺ T cells, but rather to an accumulation of CD8⁺ T cells, particularly effector memory (T_EM) and terminally differentiated effector memory re-expressing CD45RA (T_EMRA) subsets (Figure 6A–C). This imbalance in CD4/CD8 ratios persisted during RIN treatment, suggesting that the senescent immune environment associated with liver metastases remains refractory to short‐term RIN-mediated modulation (Figure 6A, Supplementary Figure 17A–B).

Figure 6. Immune senescence in liver metastasis correlates with poor response to RIN therapy.

(A) Dot plot comparing immune cell subset frequencies (normalized by live cells) between patients with liver metastases and those with lung metastases at baseline, week 4, and week 8. Color coding indicates higher levels in liver metastases (red) and lung metastases (blue).

(B, C) Representative flow cytometry and line graphs illustrating longitudinal changes in the CD4/CD8 T cell ratios and CD8⁺ T cell frequencies (B) and the T_EMRA subset within CD8⁺ T cells (C) in patients with liver versus lung metastases. Each line represents an individual patient followed longitudinally.

(D) Dot plot depicting immune senescence markers in patients with liver metastases compared to those with lung metastases at baseline, week 4, and week 8.

(E, F) Representative flow cytometry and line graphs showing the longitudinal expression changes of CD27 (E) and CD57 (F) in CD4⁺ and CD8⁺ T cells between the two groups.

(G) Schematic showing RNA sequencing analysis performed on sorted CD4⁺ and CD8⁺ T cells from three groups: good‐outcome patients without liver metastases (n = 7), poor‐outcome patients without liver metastases (n = 6), and patients with liver metastases (n = 7). Created in BioRender. Ye, J. (2025) https://BioRender.com/ktkfh4u

(H) KEGG pathway analysis of CD4⁺ and CD8⁺ T cells from patients with liver metastases. Comparisons are shown for patients with liver metastases versus poor outcome patients without liver metastases (left) and patient with liver metastases versus all patients without liver metastases (right).

(I) Line graphs showing PD-L1 and PD-1 checkpoint pathway signatures in CD4⁺ and CD8⁺ T cells over time.

(J) Gene set enrichment of post-treatment changes (4 weeks versus baseline) in CD4⁺ and CD8⁺ T cells across all three patient groups.

T_EMRA cells are known to demonstrate a senescent phenotype (30), suggesting a link between chronic immune activation and persistent hepatic tumor burden in patients with liver metastases. Further examination of senescence markers revealed that CD4⁺ T cells in patients with liver metastases exhibited more features of senescence as compared with patients with lung metastases (31–33). These included reduced expression of co-stimulatory molecules CD27 and CD28, increased levels of CD57 and senescence-associated β-galactosidase (SA-β-gal), and elevated granzyme K. Additionally, these cells showed a higher proportion of IFN-γ⁺ subsets, and upregulation of exhaustion markers such as PD-1 and LAG-3 across all time points (Figure 6D–F, Supplementary Figure 17C, D). CD8⁺ T cells displayed fewer liver metastasis-associated senescent features overall but did express higher granzyme K and less CD28 as compared to patients with lung metastases. Notably, levels of γH2AX, the marker of DNA damage, did not differ significantly between patients with liver and lung metastases (Figure 6D), in contrast to the elevated γH2AX levels found in poor responders (Figure 3E, Supplementary Figure 6B). This suggests that in the context of liver metastases, immune dysfunction is driven more by senescence-associated mechanisms than DNA-damage related mechanisms.

Transcriptional profiling supported this notion. Prior to the treatment, CD4⁺ T cells from patients with liver metastases exhibited increased expression of genes linked to apoptosis, senescence, autophagy, and chemokine signaling, reflecting a hyperactivated but dysfunctional immune state (Figure 6G–H). CD8⁺ T cells showed enrichment of metabolic and inflammatory pathways, consistent with an exhausted yet metabolically active phenotype (Figure 6H). Following RIN treatment, both CD4⁺ and CD8⁺ T cells from patients with liver metastases displayed persistent metabolic activation and biosynthetic pathway upregulation (Supplementary Figure 18A).

In contrast, patients without liver metastases showed a more favorable immune remodeling after RIN treatment. Their T cells upregulated pathways associated with immune regulation, including TCR signaling, T cell differentiation, and PD-1/PD-L1 checkpoint signaling (Supplementary Figure 18B), indicating a shift toward a more activated immune state. Notably, the KEGG pathway for PD-1/PD-L1 checkpoint signaling was upregulated by week 4 even in patients without liver metastases who experienced poor outcomes (Figure 6I–J), highlighting the liver’s unique capacity to dampen peripheral T cell responsiveness.

Together, these findings indicate that immune senescence and persistent metabolic stress characterize the T cell landscape in patients with liver metastases, contributing to therapeutic resistance. In contrast, patients without liver involvement exhibit greater immune plasticity and responsiveness. These observations underscore the immunological barrier posed by liver metastases and support the need for strategies that target immune dysfunction in this patient subset.

Discussion

Our study provides evidence that combining regorafenib, ipilimumab, and nivolumab elicits antitumor activity in patients with MSS mCRC, particularly those without liver metastases. Through integrated molecular and immunological analyses, we demonstrate that the RIN regimen engages both the TME and systemic immunity, contributing to immune activation in a subset of patients with otherwise immunotherapy-resistant disease.

When evaluating biomarkers predictive of response, distinct features emerged in both the TME and peripheral blood. Within the TME, transcriptional signatures linked to proliferative stress, cytosolic DNA sensing, and interferon signaling were enriched in good responders, suggesting that tumors with intrinsic immune activation may be more permissive to immune-based therapies. In peripheral blood, a higher CD4/CD8 T cell ratio, lower frequencies of effector memory T cells, and reduced levels of DNA damage in lymphocytes were associated with favorable outcomes. These findings support a model in which both a primed TME and a more intact, immunologically fit systemic immune state are associated with clinical benefit and may together serve as predictive biomarkers.

In contrast, poor responders, particularly those with liver metastases, exhibited features of immune dysfunction at both the local and systemic levels. These observations are consistent with features of immune exhaustion and senescence that may reflect chronic immune activation and the immunoregulatory hepatic microenvironment, thereby limiting effective antitumor responses. Addressing this resistance is a critical area for future investigation. Therapeutic strategies such as STING agonists (e.g., cyclic dinucleotide analogs or nanoparticle-based formulations) may enhance local interferon signaling and antigen presentation (34, 35). Meanwhile, systemic rejuvenation strategies, including senolytics (e.g., dasatinib plus quercetin; BCL-2 inhibitors) (36) or selective elimination of dysfunctional immune populations (e.g., antibody-mediated depletion of myeloid-biased hematopoietic stem cells) (37), may help restore immune competence in non-responders.

While our trial lacked an ICI-only comparator arm, the observed immune activation suggests that the addition of regorafenib may have enhanced responsiveness to checkpoint blockade. Regorafenib is known to normalize tumor vasculature and reduce suppressive myeloid populations, thereby facilitating T cell infiltration and activation (6, 7, 38). In our study, good responders exhibited increased post-treatment lymphocyte infiltration, immune activation gene expression, and clonal expansion of CD8⁺ T cells. These findings suggest that regorafenib may synergize with ICIs by remodeling the TME and improving the immune accessibility of tumors.

Several limitations should be acknowledged. The nonrandomized design and relatively small sample size may limit the generalizability of our findings. Larger, prospective studies with longitudinal profiling are needed to validate these results and clarify the mechanisms involved. While we defined “good responders” as patients with either PFS ≥ 6 months or a confirmed partial response to capture meaningful clinical benefit, this binary classification may group together patients with distinct underlying biology. Early tumor shrinkage may result from acute cytotoxic activity but may not necessarily translate into durable benefit. In fact, several cases of strong early regression in this trial were followed by rapid progression due to adverse events requiring treatment interruption and the administration of steroids. Consequently, early regression may be followed by disease progression driven by resistant clones, mixed responses (e.g., target lesion reduction with non-target lesion growth), or immunosuppression related to toxicity management, whereas prolonged PFS may instead reflect sustained immune pressure or effective control of resistant subclones. To account for this, stratified analyses are provided in Supplementary Figures to better delineate clinical and immunologic heterogeneity across patients.

In conclusion, our study highlights the therapeutic potential of the RIN regimen in MSS mCRC by demonstrating its ability to engage antitumor immunity and modulate both the TME and the systemic immune responses. The differential outcomes associated with metastatic sites underscore the need for personalized approaches, particularly in patients with liver metastases. Future efforts aimed at reversing immune dysfunction and enhancing T cell fitness will be critical to improving outcomes for this historically treatment-refractory population.

Synopsis:

Preexisting tumor immunogenicity and preserved T cell fitness are associated with responsiveness to Regorafenib plus checkpoint inhibitors in MSS colorectal cancer tumor and blood. T cell senescence is reported in liver metastases.

Data Availability

RNA-seq data are available via the NCBI Gene Expression Omnibus under accession numbers GSE289582 and GSE289583. Other data are available from the corresponding author upon reasonable request.