Modulating the tumor immune phenotypes by radiotherapy: formulating and validating the combination therapy of radiation, PD-L1, and TIM-3 blockade in colorectal cancer

Abstract

Background

Most colorectal cancers (CRCs) are mismatch repair-proficient (pMMR) and microsatellite stable (MSS), and they respond poorly to immune checkpoint inhibitors (ICIs). Radiotherapy (RT) can promote antitumor immunity but may also trigger adaptive immune suppression through checkpoint upregulation, providing a rationale for combination therapies.

Methods

We integrated transcriptomic analyses from The Cancer Genome Atlas and Gene Expression Omnibus cohorts—annotated for microsatellite instability (MSI)/MMR status—with single-cell RNA-sequencing data from the Human Colorectal Cancer Atlas. Mechanistic and translational experiments were conducted using CT26 (MSS/pMMR) and MC38 (microsatellite instability (MSI)-high/mismatch repair-deficient (dMMR)) mouse models, patient-derived organoid (PDO)-immune co-cultures, and clinical CRC specimens. Assessments included multicolor flow cytometry, immunohistochemistry/immunofluorescence, bulk RNA-sequencing, and immune profiling.

Results

HAVCR2 (T cell immunoglobulin and mucin-domain containing-3 (TIM-3)) and LGALS9 (galectin-9) were broadly expressed in CRC, remaining relatively high in MSS/pMMR tumors compared with most other checkpoints. In CT26 tumors, RT preferentially increased programmed cell death protein 1 (PD-1) and TIM-3 co-expression on intratumoral CD8⁺ T cells and natural killer (NK) cells, whereas these changes were weaker and less consistent in MC38 tumors. The addition of TIM-3 blockade to RT plus programmed death-ligand 1 (PD-L1) blockade produced the most durable antitumor activity in CT26 tumors, improving primary tumor control, abscopal effects, and protection against tumor rechallenge. In PDO-immune co-cultures, pMMR PDOs showed a consistent incremental benefit when TIM-3 blockade was added to RT+PD-L1 blockade, whereas this added benefit was less consistent in dMMR PDOs. In clinical datasets and specimens, RT-containing treatment was associated with increased T-cell infiltration and higher TIM-3/PD-1 signals; stereotactic body radiotherapy was accompanied by systemic immune alterations, including TIM-3 induction on circulating immune subsets. Across pan-cancer cohorts treated with ICIs, concomitantly high expression of HAVCR2 and PDCD1 correlated with an immune-activated transcriptional profile and improved clinical outcomes.

Conclusions

These findings identify TIM-3/PD-L1 as a context-dependent adaptive resistance axis following RT. The greatest incremental value of TIM-3 blockade was observed in MSS/pMMR settings, where RT more consistently increased PD-1 and TIM-3 co-expression on intratumoral CD8⁺ T cells and NK cells. Combining RT with dual TIM-3/PD-L1 blockade warrants further clinical evaluation for immunotherapy-refractory CRC.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Most microsatellite stable (MSS)/mismatch repair-proficient (pMMR) colorectal cancers (CRCs) derive limited benefit from immune checkpoint inhibitor monotherapy.

• Radiotherapy (RT) can prime antitumor immunity but may also trigger adaptive immune resistance, including checkpoint upregulation.

WHAT THIS STUDY ADDS

• RT preferentially enriches programmed cell death protein 1 (PD-1)⁺ T cell immunoglobulin and mucin-domain containing-3 (TIM-3)⁺ CD8⁺ T cells and natural killer (NK) cells in the pMMR CT26 model, supporting TIM-3/programmed death-ligand 1 (PD-L1) as a key inhibitory axis following RT in this context.

• The addition of TIM-3 blockade to RT plus PD-L1 blockade enhances tumor control, systemic (abscopal) activity, and protective memory in CT26 tumors.

• Corresponding immune changes were also observed in pMMR patient-derived organoid-immune co-cultures and in clinical CRC specimens after RT-containing treatment.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• Provides a rationale to test RT combined with dual TIM-3/PD-L1 blockade in MSS/pMMR CRC, a population with few effective immunotherapy options.

• Supports incorporating RT-induced checkpoint states (eg, PD-1/TIM-3 co-expression on CD8⁺ T/NK cells) into biomarker and correlative plans for future RT-immunotherapy trials.

Background

Colorectal cancer (CRC) remains a major cause of cancer-related morbidity and mortality worldwide.¹ The clinical management of locally advanced rectal cancer (LARC) is particularly challenging due to anatomic constraints, treatment-related toxicities, and the persistent risk of local recurrence and distant metastasis. For metastatic CRC (mCRC), durable disease control is still limited by therapeutic resistance, cumulative toxicity, and substantial interpatient heterogeneity,² underscoring the need for more effective, mechanism-guided combination strategies.

Immune checkpoint inhibitors (ICIs), particularly those targeting cytotoxic T-lymphocyte-associated protein 4 (CTLA-4) and the programmed cell death protein 1 (PD-1)/programmed death-ligand 1 (PD-L1) axis, have transformed outcomes in multiple solid tumors.^{3 4} In CRC, however, meaningful and durable responses are largely restricted to tumors that are mismatch repair-deficient (dMMR) or microsatellite instability-high (MSI-H). By contrast, the majority of mismatch repair-proficient (pMMR)/microsatellite-stable (MSS) tumors remain poorly responsive, representing a major unmet clinical need.^{5 6} Therefore, identifying key immunoregulatory pathways in pMMR CRC and rationally combining ICIs with immune-priming approaches remain important priorities.

T cell immunoglobulin and mucin-domain containing-3 (TIM-3, encoded by HAVCR2) is an inhibitory receptor frequently upregulated on chronically stimulated T cells and expressed on various other immune cell subsets.7,10 By engaging ligands such as galectin-9 (Gal-9; LGALS9) and phosphatidylserine, TIM-3 restrains effector functions and has been implicated in adaptive immune resistance, including in the context of PD-(L)1 pathway blockade.78 11,15 These properties position the TIM-3/Gal-9 axis as a plausible therapeutic target in immune-refractory CRC, especially if TIM-3 induction presents a mechanism of adaptive resistance to treatment.

Radiotherapy (RT) is a cornerstone of rectal cancer treatment and can reshape the tumor microenvironment by inducing immunogenic cell death, enhancing antigen release and antigen presentation, thus providing a mechanistic rationale for combining RT with ICIs.16,18 However, RT can also trigger compensatory inhibitory pathways in both tumors and infiltrating immune cells, including upregulation of checkpoint molecules that may undermine sustained antitumor immunity.^{19 20} Although preclinical studies support combining RT with checkpoint blockade, the most relevant inhibitory axes to target—especially in pMMR CRC—and the clinical scenarios in which multi-checkpoint regimens provide added benefit remain poorly defined.21,24

Here, we integrated transcriptomic data from public CRC cohorts and single-cell atlases to profile the TIM-3/Gal-9 axis across molecular subtypes. We then employed complementary murine models, patient-derived organoid (PDO)-immune co-culture systems, and clinical specimens to dissect RT-associated immune remodeling and to evaluate the therapeutic rationale for combining RT with PD-L1 and TIM-3 blockade. Furthermore, we explored whether TIM-3/PD-1 co-expression correlates with immune activation states that could inform patient stratification for RT-based combinatorial immunotherapy.

Methods

Experimental mice

C57BL/6J (RRID:IMSR_JAX:000664) and BALB/c (RRID:IMSR_APB:4790) mice (aged 6–8 weeks) were obtained from Shanghai Experimental Animal Center and housed under specific pathogen-free conditions at the Sir Run Run Shaw Hospital, School of Medicine, Zhejiang University.

Human samples

Patients with liver or lung metastases scheduled for three to five fractions of stereotactic body radiotherapy (SBRT) between 2020 and 2021 were recruited. Blood samples were collected before RT and 7–14 days post-RT. Patients with LARC received neoadjuvant short-course RT (SCRT; 5×5 Gy) or long-course RT (LCRT; 50 Gy in 25 fractions) with concurrent capecitabine and oxaliplatin, followed by surgery between 2018 and 2020. Pre-treatment colonoscopic biopsies and post-treatment surgical specimens were obtained from the Sir Run Run Shaw Hospital, School of Medicine, Zhejiang University. Detailed clinical information for LARC-treated and SBRT-treated patients is provided in online supplemental tables S1 and S2, respectively. For PDO and immune co-culture construction, fresh tumors and peripheral blood were collected from newly diagnosed patients with CRC undergoing surgery between 2023 and 2025 (online supplemental table S3). Patients with an Eastern Cooperative Oncology Group performance status >2 were excluded. All participants completed the study without attrition.

Cell lines

The CT26 (RRID:CVCL_7254), MC38 (RRID:CVCL_B288), and 4T1 (RRID:CVCL_0125) cell lines were purchased from the cell bank of the Chinese Academy of Sciences (Shanghai, China) and confirmed mycoplasma-negative. All cells were cultured in RPMI-1640 medium (Hyclone, Cat# SH30027.01) supplemented with 10% fetal bovine serum (FBS) at 37°C in a humidified incubator with 5% CO₂.

Transplant tumor models and treatments

BALB/c mice were inoculated subcutaneously with 5×10⁴ CT26 cells, and C57BL/6J mice were inoculated subcutaneously with 5×10⁴ MC38 cells. Mice were randomized to treatment groups on day 3 after tumor inoculation and received intraperitoneal injections of anti-TIM-3 (Bio X Cell, Cat# BE0115, RRID:AB_10949464), anti-PD-L1 (Bio X Cell, Cat# BE0101, RRID:AB_10949073), or isotype control (Bio X Cell, Cat# BE0089, RRID:AB_1107769) at 10 mg/kg every 3 days for a total of five doses. Tumors were irradiated on day 7 after tumor inoculation using 6 MV X-rays from a linear accelerator (Siemens, Erlangen, Germany) after two antibody administrations. Mice were placed in a jig shielding the body except for tumor-bearing regions, which received a single 12 Gy dose. Tumor dimensions were measured every 2 days using calipers, and volumes were calculated as 0.5×length×width². Mice were euthanized when tumor volume exceeded 1000 mm³ or when humane end points were reached.

For memory and metastasis experiments, mice achieving complete responses were re-challenged according to the experimental scheme indicated in the corresponding figure legends (including subcutaneous and/or intravenous CT26 re-challenge followed by a subsequent subcutaneous 4T1 challenge, as applicable). For experimental metastasis assays, CT26 cells (1.5×10⁵) were injected via the tail vein to establish lung metastases. For liver metastasis assays, CT26 cells (2×10⁵) were injected into the spleen. Animals were euthanized at the indicated time points, and metastatic burden was assessed as described.

Isolation of immune cells

Lymph nodes and spleens were mechanically homogenized in RPMI medium containing 2.5% FBS and 1% penicillin-streptomycin, then filtered through a 40 μm cell strainer (BD Falcon). Red blood cells were lysed, and samples were washed twice with phosphate-buffered saline (PBS). Tumor-infiltrating lymphocytes (TILs) were isolated by enzymatic digestion of tumor tissues with collagenase I (0.1% w/v; Sigma, Cat# SCR103) and DNase I (0.005% w/v; Sigma, Cat# 10104159001) for 1 hour at 37°C, followed by enrichment on a discontinuous Percoll gradient (Cytiva, Cat# 17089101). Peripheral blood mononuclear cells (PBMCs) were isolated by Ficoll density-gradient centrifugation, and the mononuclear cell layer was collected and washed with PBS.

PDO establishment, culture, and apoptosis assay

Tumor tissues were processed as previously described.^{25 26} Briefly, specimens were washed in PBS containing 10% antimicrobials (Gibco, Cat# 15240-096), trimmed, minced (~1 mm), pelleted (300×g, 5 min), and digested for 30 min at 37°C in PBS with 2% FBS, antimicrobials, collagenase II (Sigma, Cat# C6885), and dispase II (Sigma, Cat# D4693). Cells were embedded in Matrigel (Corning, Cat# 356231) and maintained in IntestiCult medium (STEMCELL, Cat# 06010_C). Medium was refreshed every 2–3 days and organoids were passaged every 7–10 days using Cell Recovery Solution (Corning, Cat# 354253) with mechanical dissociation and, when needed, TrypLE Express (Gibco, Cat# 12604013). PDOs were cryopreserved in Recovery Freezing Medium (Gibco, Cat# 12648010) and stored in liquid nitrogen.

PDO-immune co-culture was performed based on established tumor organoid-T cell co-culture protocols with minor modifications.^{25 26} Organoids were recovered in medium containing Y-27632 (10 μM; Sigma, Cat# Y-0503) and primed with interferon (IFN)-γ (200 ng/mL; PeproTech, Cat# 300-02) for 24 hours. PBMCs were thawed, treated with benzonase (Merck, Cat# 70746-3; 1:1000; 15 min, 37°C), and rested overnight in T-cell medium with interleukin (IL)-2 (150 U/mL; PeproTech, Cat# 200-02). PBMCs were then resuspended at 1×10⁶/mL in T-cell medium containing IL-2 (300 U/mL) and co-cultured with PDOs at defined effector-to-target ratios in 96-well U-bottom plates precoated with anti-CD28 (5 μg/mL; 50 μL/well; 4°C overnight; Thermo, Cat# 11161D, RRID:AB_2916088). Co-cultures were maintained as previously described.^{25 26}

For apoptosis assays, PDOs were irradiated (6 Gy) and co-cultured with autologous PBMCs (1:5 ratio) pre-activated as described above. Co-cultures were seeded on Matrigel-coated 96-well plates and treated with control, anti-PD-L1 (Durvalumab, AstraZeneca), anti-TIM-3 (SHR1702, Hengrui), or the combination. Caspase-3/7 Green Reagent (Invitrogen, Cat#R37111) and Hoechst 33342 (Invitrogen, Cat#135102) were added at initiation. Fluorescence imaging was performed at 48 hours using a Zeiss microscope and analyzed with Zen Blue software.

Antibodies and flow cytometry

Single-cell suspensions were stained with Fixable Viability Dye eFluor 520 (eBioscience, Cat# 65-0867-14) or Zombie UV (BioLegend, Cat# 423107), followed by surface and/or intracellular staining. For degranulation and intracellular cytokine assays, cells were stimulated for 4 hours with Cell Stimulation Cocktail (eBioscience, Cat# 00-4970-93) in the presence of brefeldin A (BioLegend, Cat# 420601), with anti-CD107a added during stimulation. Cells were then stained for additional surface markers, fixed/permeabilized using the FoxP3 Fixation/Permeabilization kit (eBioscience, Cat# 00-5523-00), and stained for intracellular cytokines. Data were acquired on a BD FACSFortessa and analyzed in FlowJo; antibody details are provided in online supplemental table S4.

Transcriptomic data analysis

Total RNA from fresh-frozen tumors was extracted using RNeasy Plus (Qiagen, Cat# 74134) and quality-controlled by NanoPhotometer (Implen) and Bioanalyzer 2100 (Agilent). Libraries were prepared (NEBNext Ultra RNA Library Prep Kit) and sequenced on Illumina NovaSeq 6000 (PE150). Expression was quantified as FPKM using HTSeq (V.0.6.1). Immune cell composition was inferred using TIMER V.2.0 (RRID:SCR_018737). Immune function scores were computed by single-sample gene set enrichment analysis (Gene Set Variation Analysis (GSVA); RRID:SCR_021058). The IFN-TBX21-TIM-3/PD-1 module score was calculated using the gene set: Havcr2, Pdcd1, Tbx21, Eomes, Tox, Irf1, Irf7, Cxcl10, Stat1, Stat3, Il27ra.

Single-cell RNA-sequencing data (Human Colon Cancer Atlas, c295) were analyzed with Seurat (V.4.0) and annotated with SingleR (V.1.8.1) using the Human Primary Cell Atlas reference (celldex V.1.4.0).

Bulk cohorts (GSE15781, GSE13294, GSE39582) were obtained from Gene Expression Omnibus (GEO). GSE13294 and GSE39582 were stratified by annotated MSI/MMR status; GSE39582 was additionally stratified by chemotherapy status. Microarray probes were collapsed to gene symbols by the median. Comparisons used Wilcoxon tests with multiple-testing correction where applicable. StromalScore/ImmuneScore was calculated using ESTIMATE (RRID:SCR_026090). The Cancer Genome Atlas (TCGA)-COAD/READ RNA-sequencing data were obtained via the GDC Data Portal and UCSC Xena (RRID:SCR_018938).²⁷

Pan-cancer immunotherapy survival analyses were performed using Kaplan-Meier Plotter (RRID:SCR_018753)²⁸ in PD-(L)1 and/or cytotoxic T-lymphocyte-associated protein 4 (CTLA-4) blockade-treated cohorts (see figure legend for cohort sizes).

Bulk T cell receptor sequencing

T cell receptor beta repertoire profiling was performed by Lc-Bio (Hangzhou, China) using a 5′ rapid amplification of complementary DNA ends-based unbiased protocol with unique molecular barcodes to reduce amplification bias and correct PCR/sequencing errors. Sequencing was conducted on Illumina NovaSeq/NextSeq platforms (paired-end). V(D)J segments were assigned and CDR3 sequences were assembled into clonotypes; out-of-frame and stop-codon clonotypes were excluded. Clonotype abundance was quantified by counts of identical CDR3 nucleotide sequences.

Immunohistochemistry

Paraffin-embedded sections were dewaxed, rehydrated, and subjected to antigen retrieval, followed by staining for PD-1 (Cell Signaling Technology, Cat# 86163, RRID:AB_2728833) and TIM-3 (Abcam, Cat# ab47997, RRID:AB_883007). Immune-cell positivity was scored on a four-tier scale (0: 0%–1%; 1: 1%–5%; 2: 5%–10%; 3: >10%) across 10 randomly selected ×40 fields per slide, normalized to tumor area.

Immunofluorescence staining

Paraffin-embedded sections were deparaffinized, rehydrated, antigen-retrieved in citrate buffer (pH 6.0), quenched with 3% H₂O₂, and blocked with 3% bovine serum albumin. Primary antibodies were incubated overnight at 4°C, followed by fluorophore-conjugated secondary antibodies. Multiplex staining was performed sequentially with signal amplification as described; nuclei were counterstained with 4’,6-diamidino-2-phenylindole. Images were acquired on a Nikon fluorescence microscope. Reagents and antibodies are listed in online supplemental table S4.

Statistical analysis

Analyses were performed in Prism V.8.0.1 or R V.4.0.2. Statistical tests are specified in figure legends. P values are denoted as ns (p>0.05), ^*p<0.05, ^**p<0.01, ^***p<0.001, ^****p<0.0001.

Results

The TIM-3/Gal-9 axis is broadly expressed in CRC and remains prominent in pMMR tumors

Transcriptomic profiling of 530 CRC samples from TCGA revealed heterogeneous expression of immune checkpoint genes, with HAVCR2 (TIM-3) and LGALS9 (Gal-9) showing consistently high baseline expression across tumors (figure 1a). To assess generalizability, we analyzed two independent bulk datasets with annotated MSI/MMR status (GSE13294 and GSE39582). In both cohorts, HAVCR2 and LGALS9 expression was higher in MSI-H/dMMR tumors, consistent with an immune-inflamed phenotype (online supplemental figure S1a,b). Notably, even within MSS/pMMR tumors, HAVCR2 and LGALS9 remained comparatively high relative to most other checkpoint molecules (figure 1a; online supplemental figure S1a,b), indicating that the TIM-3/Gal-9 axis remains prominent in pMMR disease.

Figure 1. TIM-3 expression landscape in CRC and progressive TIM-3 accumulation in CT26 tumors. (a) TCGA CRC (n=530): relative mRNA expression of HAVCR2 (TIM-3), LGALS9 (Gal-9), PDCD1 (PD-1), CD274 (PD-L1), CTLA4, LAG3, TIGIT, and NRP1. (b) Single-cell RNA-sequencing atlas (Human Colon Cancer Atlas, c295): UMAP of annotated cell types (SingleR with HPCA reference); top 20 cell types labeled. (c, d) Checkpoint expression across cell types in c295 shown as UMAP feature plots (c) and dot plot (d) (dot size, fraction of expressing cells; color, scaled mean expression). (e) Cell-type composition across normal, pMMR, and dMMR samples in c295. (f) Checkpoint expression within TILs across normal, pMMR, and dMMR samples in c295. (g) Frequencies of PD-1 and TIM-3 on CD8⁺ T cells and NK cells in Ln, Sp, and TILs from CT26 tumor-bearing mice (2×10⁵ cells, subcutaneous injection; analyzed at ~300 mm³; n=6). (h) PD-1 and TIM-3 on splenic and intratumoral CD8⁺ T and NK cells at days 14 and 21 after CT26 implantation (n=6/time point); baseline values from tumor-free Ctrl (n=5). (i) Frequencies of cells expressing PD-1, LAG3, CTLA4, TIGIT, GZMB, CD226, CD107a, IFN-γ, or TNF-α among tumor-infiltrating TIM-3^– CD8⁺ T cells, TIM-3⁺ CD8⁺ T cells (key), TIM-3^– NK cells, or TIM-3⁺ NK cells (key) from CT26 tumor-bearing mice (n=6). Data represent ≥3 independent experiments. *P<0.05, **p<0.01, ***p<0.001, and ****p<0.0001. Statistical analysis: unpaired two-tailed t-test (g, h); paired two-tailed t-test (i). CRC, colorectal cancer; Ctrl, control; DC, dendritic cell; dMMR, mismatch repair-deficient; CTLA-4, cytotoxic T-lymphocyte-associated protein 4; CMP, common myeloid progenitor; GZMB, granzyme B; GMP, granulocyte-macrophage progenitor; HPCA, Human Primary Cell Atlas; IFN, interferon; iPS, induced pluripotent stem; LAG-3, lymphocyte-activation gene 3; Ln, lymph node; mRNA, messenger RNA; NK, natural killer; ns, not significant; PD-1, programmed cell death protein 1; PD-L1, programmed death-ligand 1; pMMR, mismatch repair-proficient; Sp/Spl, spleen; TCGA, The Cancer Genome Atlas; TIGIT, T cell immunoreceptor with Ig and ITIM domains; TIL, tumor-infiltrating lymphocyte; TIM-3, T cell immunoglobulin and mucin-domain containing-3; TNF, tumor necrosis factor; UMAP, uniform manifold approximation and projection.

To define cell-type specificity, we analyzed single-cell RNA-sequencing data from the Human Colon Cancer Atlas (c295). Unsupervised clustering followed by SingleR-based annotation identified major immune and non-immune populations (figure 1b). HAVCR2 was enriched in macrophages, T cells, NK cells, and dendritic cells (DCs), whereas LGALS9 was broadly expressed across myeloid and epithelial compartments (figure 1c,d). In contrast, PDCD1 (PD-1) was largely restricted to T cells, whereas CD274 (PD-L1) was enriched in antigen-presenting cells. Across normal tissue, pMMR, and dMMR tumors, dMMR samples showed enrichment of T and NK cells, while macrophages/monocytes were abundant in both subtypes (figure 1e). Feature plots further confirmed higher PDCD1 and CD274 in dMMR tumors, whereas HAVCR2 and LGALS9 were elevated in both pMMR and dMMR tumors (figure 1f).

We next validated these findings in the pMMR CRC mouse model CT26. Flow cytometry demonstrated higher frequencies of TIM-3⁺ and PD-1⁺ cells among tumor-infiltrating CD8⁺ T cells and NK cells than in spleen or tumor-draining lymph nodes (figure 1g). From day 14 to day 21 postimplantation, TIM-3 increased further, reaching 55.28%±12.4% in CD8⁺ T cells and 42.30%±7.8% in NK cells by day 21 (p<0.05) (figure 1h). Compared with TIM-3⁻ counterparts, TIM-3⁺ CD8⁺ T and NK cells exhibited reduced IFN-γ, TNF-α, and lower surface CD107a, despite higher intracellular granzyme B (GZMB), and co-expressed elevated levels of PD-1, lymphocyte-activation gene 3, CTLA-4, and T cell immunoreceptor with Ig and ITIM domains (figure 1i), consistent with an impaired effector phenotype with preserved cytotoxic granule content.

Together, these data identify the TIM-3/Gal-9 axis as a broadly expressed immunoregulatory pathway in CRC that remains prominent in pMMR tumors and is associated with dysfunctional effector lymphocyte states.

Radiotherapy induces a transient immune activation and checkpoint induction program in pMMR CT26 tumors

To compare RT-associated immune remodeling across models, we performed bulk RNA-sequencing analysis on CT26 (MSS/pMMR)²⁹ and MC38 (MSI-H/dMMR)³⁰ tumors collected 2 weeks after a single 12 Gy irradiation. This dose was selected based on prior studies showing that single fractions within the 5–12 Gy range can enhance antitumor immunity while preserving cyclic GMP-AMP synthase-STING Stimulator of interferon genes (STING) signaling.^{31 32} Cell-type Identification By Estimating Relative Subsets Of RNA Transcripts (CIBERSORT) deconvolution suggested increased CD8⁺ T cells, memory B cells, and M1/M2 macrophages in irradiated CT26 tumors (figure 2a,b). Consistently, multiple immune regulatory and costimulatory genes—including Havcr2 (TIM-3), Pdcd1 (PD-1), Cd274 (PD-L1), and Lgals9—were upregulated in CT26, whereas changes in MC38 were less pronounced (figure 2c). GSVA further revealed enrichment of pathways related to CD8⁺ T-cell responses, antigen presentation, IFN-γ signaling, and cytotoxic lymphocyte-mediated killing in CT26 (figure 2d; online supplemental figure S2a), supported by Gene Ontology and Kyoto Encyclopedia of Genes and Genomes analyses (online supplemental figure S2b,c).

Figure 2. RT triggers model-dependent immune remodeling and checkpoint induction in CT26 vs MC38. (a–d) Transcriptomic analysis of CT26 and MC38 tumors 2 weeks after tumor-directed 12 Gy X-ray irradiation or no treatment. Groups: MC38_Ctrl (n=3), MC38_IR (n=3), CT26_Ctrl (n=4), CT26_IR (n=6). (a) CIBERSORT-inferred immune composition (mean per group). (b) Estimated fractions of CD8⁺ T cells, memory B cells, and M1/M2 macrophages. (c) Heatmap of checkpoint/co-stimulatory genes (row-wise Z-score). (d) GSVA heatmap of top variable immune pathways (MSigDB C7; Z-score). See online supplemental table S5 for details. (e) Flow cytometry histograms of PD-1 (left) and TIM-3 (right) expression on tumor-infiltrating CD8⁺ T cells (top) and NK cells (bottom) in CT26 or MC38 tumor-bearing mice 2 weeks post-IR (IR, CT26 n=7, MC38 n=6) or untreated (Ctrl, CT26 n=7, MC38 n=5). (f) Frequencies of CD8⁺ T and NK cells, including PD-1⁺ and TIM-3⁺ subsets, and the PD-1⁺ TIM-3⁺/PD-1⁻ TIM-3⁻ ratio, in CT26 (yellow dots) and MC38 (green squares) tumors from panel e. (g) Frequency of PD-1 and TIM-3 and the ratio of PD-1⁺ TIM-3⁺ to PD-1⁻ TIM-3⁻ on CD8⁺ T cells (above) and NK cells (bottom) in PBMCs from the same mice as in panel e. (h–l) Transcriptomic analysis of CT26 tumors at days 5, 14, and 21 after tumor-directed 12 Gy X-ray irradiation (n=4 per time point). (h) CIBERSORT-estimated immune cell proportions across time points. (i) CD8⁺ T cell infiltration inferred by five deconvolution methods (CIBERSORT, CIBERSORT-ABS, EPIC, QUANTISEQ, TIMER). (j) Heatmap of selected immune-related gene expression (log2-transformed). (k) Box plots of expression levels for 10 immune-related genes across time points. (l) GSVA heatmap of immune-related pathways (MSigDB C7; Z-score). Detailed pathway and enrichment score information is provided in online supplemental file 2. *P<0.05, **p<0.01, ***p<0.001, and ****p<0.0001. Statistical analysis: Kruskal-Wallis test (b, k), unpaired two-tailed t test (f, g); Wilcoxon rank-sum test (i). CIBERSORT, Cell-type Identification By Estimating Relative Subsets Of RNA Transcripts; CIBERSORT-ABS, CIBERSORT absolute mode; Ctrl, control; EPIC, Estimating the Proportions of Immune and Cancer cells; GSVA, Gene Set Variation Analysis; IR, irradiation; NK, natural killer; ns, not significant; PBMC, peripheral blood mononuclear cell; PD-1, programmed cell death protein 1; QUANTISEQ, quanTIseq immune deconvolution method; RT, radiotherapy; TIM-3, T cell immunoglobulin and mucin-domain containing-3; TIMER, Tool for Immune Cell Estimation.

At day 14 post-RT, flow cytometry confirmed increased intratumoral frequency of CD8⁺ T cells (19.4%±3.1%) and NK cells (14.5%±2.9%) in CT26 tumors, with higher TIM-3 and PD-1 expression on both populations (p<0.05) (figure 2e,f). Similar trends were observed in circulating lymphocytes (figure 2g). Compared with CT26, MC38 tumors showed higher baseline infiltration of CD8⁺ T cells and PD-1 expression; however, intratumoral TIM-3 remained low and was minimally affected by RT, and only a modest increase in peripheral PD-1⁺ CD8⁺ T cells was observed after RT (figure 2f,g). Accordingly, the PD-1⁺ TIM-3⁺ to PD-1⁻ TIM-3⁻ ratio increased in CT26 but not in MC38 tumors.

Longitudinal RNA-sequencing analysis of CT26 tumors at days 5, 14, and 21 post-RT showed that inferred abundances of CD8⁺ T cells, activated NK cells, and M1 macrophages peaked at day 14 (figure 2h,i). Expression of checkpoint and costimulatory genes—including Havcr2, Pdcd1, Cd274, Ctla4, Nrp1, Icos, Cd86, and Cd28—also peaked at day 14 (figure 2j,k), consistent with a transient window of immune activation following RT (figure 2l; online supplemental figure S2d).

To explore potential drivers of TIM-3 induction, we analyzed pre-RT and post-RT bulk RNA-sequencing from CT26 and MC38 tumors, focusing on transcription factors linked to HAVCR2 and IFN-response genes. CT26 tumors showed coordinated upregulation of Tbx21, Eomes, Tox, Irf1/Irf7, Cxcl10, and Stat1/Stat3, together with Havcr2 and Pdcd1, whereas corresponding changes in MC38 were weaker (online supplemental figure S2e,f). Havcr2 expression also correlated with prioritized transcription factor (TF) and IFN genes, with stronger associations observed in CT26 tumors (online supplemental figure S2g). A composite module score reflecting IFN-TBX21-TIM-3/PD-1 signaling (Eomes, Tox, Irf1, Irf7, Cxcl10, Tbx21, Il27ra, Stat1, Stat3, Havcr2, Pdcd1) increased more markedly after RT in CT26 than in MC38 tumors (online supplemental figure S2h), supporting model-dependent induction of an IFN-linked adaptive resistance program in CT26 tumors.

TIM-3 blockade augments PD-L1 blockade and RT efficacy in CT26 but not in MC38

To evaluate therapeutic combinations, CT26 and MC38 tumor-bearing mice were treated with anti-PD-L1, anti-TIM-3, dual blockade, or isotype, with or without RT (12 Gy) (figure 3a). Anti-TIM-3 was administered according to a validated early delivery schedule.¹⁵ Compared with isotype controls, dual blockade delayed tumor growth and prolonged survival in both models but provided limited incremental benefit over anti-PD-L1 monotherapy (figure 3b–e). Interestingly, RT exhibited diverse therapeutic outcomes when combined with anti-PD-L1 and/or anti-TIM-3 therapy. In MC38 tumors, even though the expression of PD-1 and TIM-3 was not significantly changed (figure 2c,e), RT sensitized anti-PD-L1 therapy (figure 3b–e). This is pointing to a PD-L1-dominant adaptive resistance after irradiation: PD-L1 protein levels increased after RT in MC38 tumor cells in vitro (online supplemental figure S2i), which is consistent with another report.³³ However, across treatment regimens, anti-TIM-3 did not provide measurable additional benefit in this model: anti-TIM-3+anti-PD-L1 performed similarly to anti-PD-L1 alone, RT+anti-TIM-3 was comparable to RT alone, and triple therapy showed outcomes similar to RT+anti-PD-L1 (figure 3c,e). All treatments were well tolerated, as indicated by stable body weight (figure 3f).

Figure 3. RT combined with dual TIM-3 and PD-L1 blockade enhances antitumor efficacy in CT26 CRC models.(a) Experimental protocol for the mouse models used in panels b–g: mice received i.p. injections of TIM-3 mAb, PD-L1 mAb, or isotype IgG at various times, combined with a single dose of 12 Gy X-ray irradiation directed to the tumors, after a s.c. injection of 5×10⁴ CT26 (b, d, f, g) or MC38 (c, e) tumor cells on day 0. Isotype IgG (CT26 n=8; MC38 n=6), anti-TIM-3 (CT26 n=9; MC38 n=6), anti-PD-L1 (CT26 n=10; MC38 n=6), anti-TIM-3+anti-PD-L1 (CT26 n=10; MC38 n=6), isotype IgG+RT (CT26 n=11; MC38 n=6), anti-TIM-3+RT (CT26 n=9; MC38 n=6), anti-PD-L1+RT (CT26 n=9; MC38 n=6), anti-TIM-3+anti-PD-L1+RT (CT26 n=9; MC38 n=6). (b, c) Tumor growth curves for CT26 (b) and MC38 (c). Tumor volumes were compared at day 25 (no RT groups) and day 61 (RT groups) for CT26, and at day 37 (no RT groups) and day 61 (RT groups) for MC38. (d, e) Survival of CT26 (d) and MC38 (e) cohorts. (f) Body weight of CT26-bearing mice at day 19. (g) Frequencies of GZMB⁺, IFN-γ⁺, and CD107a⁺ cells among tumor-infiltrating CD8⁺ T cells and NK cells in CT26-bearing mice on day 19. *P<0.05, **p<0.01, ***p<0.001, and ****p<0.0001. Statistical analysis: one-way analysis of variance test (b, c, f, g), Mantel-Cox test (d, e). CRC, colorectal cancer; GZMB, granzyme B; IFN, interferon; i.p., intraperitoneal; mAb, monoclonal antibody; NK, natural killer; ns, not significant; PD-L1, programmed death-ligand 1; RT, radiotherapy; s.c., subcutaneous; TIM-3, T cell immunoglobulin and mucin-domain containing-3.

By contrast, in CT26 tumors, triple therapy (RT+anti-PD-L1+anti-TIM-3) provided the strongest tumor control and survival benefit (figure 3b,d). On day 34, mean tumor volumes were 6.05 mm³ in the triple-therapy group vs 243.0 mm³ with RT+anti-PD-L1, 602.1 mm³ with RT+anti-TIM-3, and 514.0 mm³ with anti-TIM-3+anti-PD-L1 (figure 3b), which might be due to increased frequencies of intratumoral CD8⁺ T cells and NK cells expressing GZMB, surface CD107a, or IFN-γ after dual PD-L1 and TIM-3 signaling blockade (figure 3g). Similarly, body weights remained stable across all groups (figure 3f).

To further examine whether an MMR-deficient-like context in the same host background alters response patterns, we generated a stable Mlh1-knockdown CT26 variant (online supplemental figure S3a–c) and assessed tumor control and immune phenotypes under the indicated regimens (online supplemental figure S3d,e). In this Mlh1-knockdown CT26 model, the frequencies of tumor-infiltrating CD8⁺ T cells and NK cells, as well as the PD-1⁺/TIM-3⁺ fractions within these populations, did not show marked differences across the treatment conditions (online supplemental figure S3e). Consistent with these immune readouts, the addition of anti-TIM-3 did not further improve tumor control beyond anti-PD-L1+RT in Mlh1-knockdown CT26 tumors (online supplemental figure S3d).

Together, these data indicate that TIM-3 blockade contributes meaningful additional benefit in the MSS/pMMR setting—most notably when added to RT plus PD-L1 blockade—whereas in dMMR tumors, the therapeutic efficacy is largely driven by PD-L1 blockade, with minimal added value from TIM-3 inhibition.

TIM-3 and PD-L1 blockade enhances RT-induced immune memory and abscopal activity in CT26 tumors

Building on the antitumor activity observed in figure 3, complete tumor regressions occurred most frequently with triple therapy (anti-PD-L1, 20%; anti-TIM-3+anti-PD-L1, 20%; anti-TIM-3+RT, 11.1%; anti-PD-L1+RT, 33.3%; RT+anti-PD-L1+anti-TIM-3, 88.9%) (figure 4a). Mice achieving complete responses were re-challenged subcutaneously with CT26 cells on day 100 without further treatment (figure 4b). All mice previously treated with RT-containing regimens rejected the re-challenge, whereas tumors grew in one out of two mice from the antibody-only groups and in all age-matched-naïve controls (figure 4c). Tumor-free survivors remained protected against a subsequent intravenous CT26 challenge but failed to reject a later subcutaneous challenge with unrelated 4T1 cells, indicating durable and antigen-specific immunity (figure 4c,d).

Figure 4. TIM-3/PD-L1 blockade enhances RT-induced immune memory and abscopal effects in the CT26 model. (a) Complete response rates in CT26-bearing BALB/c mice treated with anti-TIM-3, anti-PD-L1, or isotype control (rat IgG), with or without tumor-directed RT (12 Gy; n=6–11/group). ‘O’ denotes mice without tumor rejection. (b) Experimental protocol for panels c and d: mice with complete response from panel a were re-challenged s.c. with 2×10⁵ CT26 cells in the right hind limb. Tumor-free mice were subsequently injected i.v. with 1.5×10⁵ CT26 cells to generate lung metastases. (c) Rejection rates were compared with age-matched naïve controls. (d) Survivors were challenged with 2×10⁵ 4T1 cells s.c. in the flank, and the rejection rates were compared with those for the CT26 tumor they initially rejected. (e–g) In an independent cohort (n=5/group), mice received CT26 (5×10⁴, s.c.) followed by RT (12 Gy) and indicated antibody regimens; cured mice were rechallenged i.v. with CT26 (1.5×10⁵) (e). Representative lungs (f) and H&E staining (g) are shown 2 weeks later. (h–j) Cured mice received splenic injection of CT26 (2×10⁵) (h). Liver weight and macrometastatic nodules (>2 mm) were quantified 2 weeks later (i) with H&E confirmation (j). (k, l) Frequencies of intratumoral CD8⁺ T cells, NK cells, and MDSCs (k), and CD107a⁺ and GZMB⁺ cells within CD8⁺ T and NK compartments (l), measured 20 days after treatment (as in e). (m–p) Bilateral CT26 tumors were established (left: 5×10⁴; right: 2.5×10⁴), with RT (12 Gy) delivered to the left tumor only, plus indicated antibody regimens (n=8–12/group) (m). Growth of irradiated (n) and non-irradiated tumors (o) and overall survival (p) are shown. *P<0.05, **p<0.01, ***p<0.001, and ****p<0.0001. Statistical analysis: one-way analysis of variance test (i, k, l, n, and o), Mantel-Cox test (p). Ctrl, control; GZMB, granzyme B; i.p., intraperitoneal; i.v., intravenous; mAb, monoclonal antibody; MDSC, myeloid-derived suppressor cell; NK, natural killer; ns, not significant; PD-L1, programmed death-ligand 1; RT, radiotherapy; s.c., subcutaneous; TIM-3, T cell immunoglobulin and mucin-domain containing-3.

To evaluate protection in metastatic settings, complete responders were inoculated intravenously with CT26 cells and lungs were assessed 2 weeks later (RT, 2/5; anti-PD-L1+RT, 4/5; triple therapy, 5/5) (figure 4e,f). H&E staining revealed extensive metastases in naïve mice, reduced lesions after RT or RT+anti-PD-L1, and no detectable lung metastases after triple therapy (figure 4g). In a splenic-injection liver metastasis model (figure 4h), triple therapy reduced liver weight and the number of metastatic nodules >2 mm by day 14, as confirmed by H&E (figure 4i,j), indicating a systemic immune memory response was initiated by triple therapy.

Mechanistically, triple therapy increased intratumoral CD8⁺ T-cell and NK-cell infiltration (figure 4k) and enhanced effector function, elevating CD107a and GZMB in both populations (figure 4l). RT+anti-PD-L1 increased the proportion of myeloid-derived suppressor cells, an effect attenuated by adding anti-TIM-3 (figure 4k). In a bilateral tumor model in which only the left tumor was irradiated (figure 4m), triple therapy delayed growth of both irradiated and non-irradiated tumors and extended survival compared with the other regimens (figure 4n–p).

Consistent with enhanced antigen presentation and clonal focusing after RT, intratumoral DC numbers increased (online supplemental figure S4a), along with elevated CCR7 and intracellular pSTAT1 following RT; pSTAT1 was further increased by triple therapy, whereas CD86 showed a non-significant upward trend (online supplemental figure S4b–d). Central memory (Tcm) and effector memory (Tem) populations also expanded after RT-containing regimens (online supplemental figure S4e,f). Bulk T cell receptor (TCR)-β sequencing demonstrated increased clonality, reduced diversity/evenness indices after RT, and a more focused oligoclonal expansion pattern (online supplemental figure S4g,h), indicating RT induces a strong antitumor T cell response by expanding several tumor antigen-specific clones in the tumor microenvironment. These expanded dominant oligoclonal T cells might be more readily able to form memory T cells. However, there are no further changes for the TCR landscape by the addition of TIM-3/PD-L1 blockade (online supplemental figure S4g,h), suggesting that, instead of triggering clonal expansion, the immune checkpoint inhibition might affect the other development programs of memory T cell in these expanded dominant clones post-RT.

Collectively, RT combined with TIM-3/PD-L1 blockade elicited durable, antigen-specific protection, limited metastatic outgrowth, and enhanced abscopal activity in the CT26 model.

The combination of RT with PD-L1 and TIM-3 blockade enhances tumor killing in PDO-immune co-cultures, most prominently in pMMR models

To extend these findings to human CRC, we established PDO-immune co-cultures from 11 pMMR and two dMMR tumors (online supplemental figure S5a and table S3). In pMMR PDO co-cultures, triple therapy increased the proportions of CD8⁺ T cells and NK cells and reduced PDO counts (figure 5a). RT increased PD-1 and TIM-3 expression on CD8⁺ T cells, whereas checkpoint blockade attenuated this induction (figure 5b). Triple therapy increased Ki-67 mean fluorescence intensity in CD8⁺ T cells (figure 5c) and enhanced cytotoxic features, including higher expression of GZMB and CD16 (figure 5d,e). Given the role of TCF-1 (encoded by TCF7) in sustaining CD8⁺ T-cell memory and responsiveness to PD-1 pathway blockade,^{34 35} we assessed this compartment and observed an increased TIM-3⁻ TCF1⁺ fraction within PD-1⁺ CD8⁺ T cells following triple therapy (figure 5f). Multiplex immunofluorescence showed increased CD3, PD-1, PD-L1, and TIM-3 signals after RT; triple therapy reduced checkpoint signals while maintaining elevated CD3 infiltration (figure 5g). Apoptosis and oxidative DNA damage markers in the PDO compartment (cleaved-caspase 3 and 8-hydroxy-2′-deoxyguanosine) were strongest with triple therapy, consistent with smaller organoid morphology and higher caspase-3/7 activity (figure 5h–k).

Figure 5. Antitumor efficacy of RT combined with TIM-3/PD-L1 blockade in PDO immune co-culture models. PDOs received a single 6 Gy X-ray dose, were co-cultured with autologous immune cells, and treated with anti-PD-L1 with or without anti-TIM-3 for 48 hours. (a) Flow cytometry analysis of CD8⁺ T, NK, and PDO (tumor) cells. (b) PD-1 and TIM-3 expression on CD8⁺ T cells. (c–e) Quantification of Ki-67 MFI (c), GZMB⁺ frequency (d), and CD16⁺ frequency (e) within CD8⁺ T cells. (f) TIM-3⁻ TCF1⁺ fraction within PD-1⁺CD8⁺ T cells. (g, h) Multiplex immunofluorescence of co-culture systems showing CD3, PD-1, TIM-3, and PD-L1 (g), and apoptosis/oxidative DNA damage (c-CAS3, 8-OHdG) (h). (i) Bright-field images at days 1, 3, and 6 after RT under indicated conditions. (j, k) Caspase-3/7-based apoptosis imaging at 48 hours (j) and quantification (k). *P<0.05, **p<0.01, and ****p<0.0001. Statistical analysis: one-way analysis of variance test (c–f and k). 8-OHdG, 8-hydroxy-2′-deoxyguanosine; c-CAS3, cleaved-caspase 3; Ctrl, control; DAPI, 4′,6-diamidino-2-phenylindole; MFI, mean fluorescence intensity; NK, natural killer; ns, not significant; PD-1, programmed cell death protein 1; PD-L1, programmed death-ligand 1; PDO, patient-derived organoid; RT, radiotherapy; TIM-3, T cell immunoglobulin and mucin-domain containing-3.

In the dMMR PDO cohort (n=2), neither RT nor triple therapy increased CD8⁺ T-cell or NK-cell proportions, although PDO counts showed a downward trend (online supplemental figure S5b). Frequencies of PD-1⁺ CD8⁺ T cells changed minimally, whereas TIM-3⁺ CD8⁺ T cells decreased under triple therapy (online supplemental figure S5c). RT increased Ki-67^hi and GZMB⁺ CD8⁺ T cells; anti-PD-L1 further increased these proportions, with no clear additional effect from anti-TIM-3 (online supplemental figure S5d,e). CD16⁺ CD8⁺ T cells and TIM-3⁻ TCF1⁺ PD-1⁺ CD8⁺ T-cell subsets showed no consistent differences across conditions (online supplemental figure S5f,g). Intracellular staining confirmed that cleaved-caspase 3 signal was minimal in CD45⁺ immune cells and largely confined to PDO cells; RT increased cleaved-caspase 3⁺ PDO cells, with a further upward trend under RT plus anti-PD-L1, while adding anti-TIM-3 did not increase this signal (online supplemental figure S5h,i), consistent with multiplex immunofluorescence and bright-field imaging (online supplemental figure S5j,k).

Overall, RT combined with PD-L1/TIM-3 blockade enhanced immune activation and PDO killing, with robust effects in pMMR PDOs and limited incremental benefit from TIM-3 blockade in the dMMR subset.

RT-associated immunophenotypic changes in patients with CRC support the clinical relevance of TIM-3/PD-1 co-expression

To assess RT-associated immune modulation, we first analyzed GSE15781 rectal cancer samples annotated as pre-treatment versus post-neoadjuvant chemoradiotherapy (nCRT, 50 Gy in 25 fractions with capecitabine). Post-treatment samples showed increased inferred T cells and macrophages/monocytes, along with higher immune scores estimated by TIMER V.2.0 and ESTIMATE (figure 6a). Upregulated genes included costimulatory molecules (CD28, CD86), immune regulatory genes (including HAVCR2 and CD200), and HLA genes (figure 6b,c). This was accompanied by enhanced cytolytic and IFN-response signatures (figure 6d), indicative of concurrent activation and adaptive regulation following nCRT.

Figure 6. Immunophenotypic changes in patients with CRC following conventional and ablative RT. (a–d) Bulk transcriptomic profiling of rectal cancer samples pre-nCRT versus post-nCRT in GSE15781. Immune cell estimates and ImmuneScore/StromalScore (a), checkpoint genes (b), HLA genes (c), and ssGSEA immune-function scores (d) are shown. (e–i) Paired colonoscopy biopsies (before RT) and surgical tissues (after RT) from patients with LARC (n=56) post-nCRT/SCRT were analyzed: study schematic (e), representative IHC for TIM-3 and PD-1 (f), paired scores for TIM-3/PD-1 in ICs and TIL percentage (g), proportion of patients with or without increased TILs and/or TIM-3/PD-1 after treatment (h), and TRG distribution (i). (j–n) Peripheral blood was collected from patients with liver or lung metastases before and 7–14 days after SBRT for flow cytometry analysis (n=32): study schematic (j), frequency of the circulating CD8⁺ T cells, NK cells, M/N, and Treg (k). PD-1, TIM-3, CD16 expression on CD8⁺ T cells and TIM-3⁻ TCF1⁺ percentage among PD-1⁺ CD8⁺ T cells (l). Frequency of PD-1 and TIM-3 on NK cells (m). PD-L1 and TIM-3 expression on M/N cells (n). Box plots: median with IQR; whiskers=1.5× IQR; points denote outliers. Schematics in (e, j) were created with BioRender.com. *P<0.05, **p<0.01, ***p<0.001, and ****p<0.0001. Statistical analysis: Wilcoxon test (b–d), Paired t-test (g, k–n). CRC, colorectal cancer; GSE, Gene Expression Omnibus Series; GSVA, gene set variation analysis; HLA, human leukocyte antigen; IC, immune cell; IHC, immunohistochemistry; LARC, locally advanced rectal cancer; M/N, monocytes/neutrophils; nCRT, neoadjuvant chemoradiotherapy; NK, natural killer; ns, not significant; PD-1, programmed cell death protein 1; PD-L1, programmed death-ligand 1; RT, radiation therapy; SBRT, stereotactic ablative radiotherapy; SCRT, short-course radiotherapy; ssGSEA, single-sample gene set enrichment analysis; TIM-3, T cell immunoglobulin and mucin-domain containing-3; TRG, tumor regression grade; Tregs, regulatory T cells.

We next analyzed paired pre-treatment biopsies and post-treatment surgical specimens from 56 patients with LARC treated with either SCRT (5×5 Gy; n=13) or LCRT-based regimens (median 50 Gy; n=43) (figure 6e; online supplemental table S1). Most tumors were pMMR (55/56, 98.2%). Matched IHC showed increased TILs and higher TIM-3 and PD-1 scores in immune cells after RT (figure 6f,g), with increases observed in 60.7% (34/56) of cases for TILs, 51.8% (29/56) for TIM-3, and 60.7% (34/56) for PD-1 (figure 6h). These changes were associated with an improved distribution of tumor regression grades (figure 6i). When stratified by RT regimen, SCRT alone significantly increased TILs and TIM-3, while PD-1 showed a non-significant upward trend (online supplemental figure S6a). In the GSE39582 dataset, checkpoint gene expression did not differ between chemotherapy-treated and chemotherapy-naïve CRC cases (online supplemental figure S6b), suggesting that the RT-associated induction of TIM-3-related immune programs is not merely a consequence of chemotherapy.

Consistent with heightened effector activity, GZMB transcripts were higher in tumor than in normal tissues and were further increased in post-treatment rectal cancer samples from GSE15781 (online supplemental figure S6c). Multiplex immunofluorescence analysis of paired LARC specimens corroborated these findings, showing increased CD3 infiltration together with higher GZMB, PD-1, and TIM-3 signals post-RT (online supplemental figure S6d,e).

We also analyzed peripheral blood from 32 patients with advanced malignancies who underwent SBRT for liver or lung metastases, with peripheral blood collected before and 7–14 days after SBRT (figure 6j; online supplemental table S2). Post-SBRT, NK cell proportions increased, while CD8⁺ T cell, monocyte/granulocyte, and Treg levels remained stable (figure 6k). Within CD8⁺ T cells, PD-1⁺, CD16⁺, and TIM-3⁻ TCF1⁺PD-1⁺ subsets expanded, although TIM-3 expression itself remained low (figure 6l). In contrast, TIM-3 was highly expressed on NK cells and further upregulated after SBRT, whereas PD-1 remained low on this subset (figure 6m). Increased expression of PD-L1 and TIM-3 was also observed on monocytes/granulocytes post-SBRT (figure 6n). Collectively, these findings demonstrate that both conventional and ablative RT substantially remodel the immune landscape, promoting local and systemic immune activation alongside checkpoint expression.

To evaluate the biomarker relevance of these observations, we analyzed transcriptomes from patients with colon cancer from TCGA (n=512). Patients with concomitantly high HAVCR2 and PDCD1 expression exhibited significant enrichment of gene signatures related to cytokine signaling, T-cell activation, IFN-response, and T helper 1/T helper 17 differentiation programs (online supplemental figure S6f–i), as well as higher expression of immune activation genes (GZMB, TNF, TGFB1, PRF1, IL18) (online supplemental figure S6j). Finally, in a pan-cancer cohort of nine tumor types treated with PD-(L)1 and/or CTLA-4 blockade, the HAVCR2^highPDCD1^high subgroup demonstrated improved overall survival and progression-free survival compared with HAVCR2^lowPDCD1^low (online supplemental figure S6k,l), supporting the potential of TIM-3/PD-1 co-expression as a candidate biomarker for predicting benefit from immunotherapy.

Discussion

Integrated analyses across TCGA, two independent GEO cohorts with annotated MSI/MMR status (GSE13294 and GSE39582), and single-cell data support the TIM-3/Gal-9 axis as a broadly represented immunoregulatory pathway in CRC. Although HAVCR2 and LGALS9 are enriched in MSI-H/dMMR tumors, their expression remains comparatively elevated in MSS/pMMR tumors relative to most other immune checkpoints. This suggests that this axis is not restricted to ‘inflamed’ dMMR disease and may be relevant in immunotherapy-refractory pMMR CRC. In the CT26 model, TIM-3 progressively accumulated on intratumoral CD8⁺ T cells and NK cells during tumor progression, consistent with sustained antigen exposure and dysfunction/exhaustion-like programs.

A key finding from the CT26-MC38 comparison is that the same RT regimen drove distinct immune remodeling in the two models. In CT26 tumors, RT triggered a coordinated checkpoint-associated program, marked by increased Havcr2/Pdcd1 expression together with IFN-response and TF modules. In contrast, the induction of these signatures was weaker and less consistent in MC38 tumors. We do not attribute this contrast solely to MMR status. Notably, a strain-matched, dMMR-like CT26 variant generated by Mlh1 knockdown showed a pattern similar to MC38, with no clear incremental benefit from adding anti-TIM-3. Given that this is a single-gene perturbation, we interpret this finding cautiously and do not consider it a definitive surrogate for MSI-H/dMMR biology. CT26 and MC38 tumors differ in their tumor genetics, baseline immune contexture, and host background (BALB/c vs C57BL/6). Moreover, strain-dependent immune programming can shape T-helper cell polarization and IFN-γ-associated responses, thereby influencing the interaction between RT and ICI therapy.³⁶

These biological differences aligned with a clear therapeutic hierarchy. In the MSS/pMMR CT26 model, dual PD-L1/TIM-3 blockade enhanced cytotoxic lymphocyte activity and, when combined with RT, produced the most durable antitumor effects among the tested regimens, evidenced by prolonged survival, immune memory, suppression of metastatic outgrowth, and abscopal activity. In the MC38 model, RT did not measurably upregulate PD-1 or TIM-3 on intratumoral CD8⁺ T cells or NK cells, yet RT plus anti-PD-L1 remained superior to RT plus isotype control. Notably, adding anti-TIM-3 provided no additional benefit across treatment regimens (anti-TIM-3+anti-PD-L1 vs anti-PD-L1 alone; RT+anti-TIM-3 vs RT alone; triple therapy vs RT+anti-PD-L1). Together with the observed RT-induced upregulation of PD-L1 on tumor cells in vitro, these results support a PD-L1-dominant adaptive resistance axis in the MSI-H/dMMR MC38 context, consistent with clinical and preclinical evidence that RT/CRT can increase PD-L1 expression in CRC tissues and models.³³

The durability of immune memory after RT-containing regimens in CT26 tumors warrants focused mechanistic investigation. Prior work indicates that radiation can engage cytosolic DNA sensing and type I IFN pathways that promote cross-priming, with dose/fractionation schedule and nucleases such as TREX1 influencing whether these programs are productively activated.³⁷ In our system, increased DC infiltration/activation markers and RT-associated TCR focusing are consistent with enhanced antigen presentation and clonal selection. Establishing causality will require targeted tests of cross-presenting DC dependence, type I IFN/STING signaling, and CD8⁺ memory T cell differentiation. Regarding memory T cell differentiation, antigen-specific T cells typically undergo clonal expansion, followed by a contraction phase.³⁸ Our bulk TCR-β sequencing data from tumor tissues revealed that dominant T cell clones expanded significantly after RT in the tumor microenvironment, which are potentially important sources for tumor antigen-specific memory T cells. However, the addition of TIM-3/PD-L1 blockade did not induce further significant changes in the TCR repertoire. Given that ICIs can influence memory T-cell differentiation through multiple mechanisms,³⁹ TIM-3/PD-L1 blockade may modulate memory T cell development via pathways independent of clonal expansion. Therefore, additional work is needed to clarify the impact of dual TIM-3/PD-L1 signaling blockade on the formation and maintenance of T cell memory in the context of RT.

Human systems exhibited a similar hierarchical pattern. In pMMR PDO-immune co-cultures, the addition of TIM-3 blockade to RT plus PD-L1 blockade consistently enhanced immune activation and tumor killing, whereas the dMMR PDO subset (n=2) showed less reproducible incremental benefit, a finding that should be interpreted cautiously given small sample size and inter-PDO heterogeneity. In patient specimens, RT-treated rectal cancer tissues displayed increased T-cell infiltration along with elevated TIM-3/PD-1 signals. Within the SCRT subgroup, TILs and TIM-3 increased significantly, whereas PD-1 showed a numerical but statistically non-significant rise, possibly reflecting limited statistical power. This is consistent with reports that RT and/or cytotoxic therapy can induce TIM-3 expression,^{40 41} whereas opposite findings⁴² may reflect differences in sampling and incomplete patient-level pairing. Notably, in an external CRC cohort, checkpoint expression did not differ between chemotherapy-treated and chemotherapy-naïve cases, arguing against chemotherapy exposure alone as a sufficient explanation for the TIM-3-related immune signature in RT-based datasets. We also investigated whether RT enhances effector functions of immune cells in human CRC. GZMB expression increased after treatment at both transcript and protein levels. Any discrepancy between the post-RT transcript rise and more modest protein changes in some flow cytometry analyses may be attributable to differences in timing, anatomical compartment (bulk tumor vs gated lymphocyte subsets), and layered regulation of granule-stored effectors such as GZMB.⁴³ The additional human analyses—including higher tumor-versus-normal GZMB transcripts with further post-treatment elevation, plus multiplex immunofluorescence showing increased CD3 infiltration and elevated GZMB/PD-1/TIM-3 signals after RT—help link pathway-level inferences with compartment-resolved protein data. Peripheral immune profiling after SBRT further supports systemic remodeling, characterized by TIM-3 induction across multiple immune subsets.

Clinically, while nCRT followed by surgery remains standard for LARC and TNT or organ-preserving strategies have improved outcomes in selected patients, recurrence and distant failure continue to pose clinically relevant challenges.⁴⁴ For mCRC, durable disease control is limited by resistance, systemic toxicity, and interpatient heterogeneity.^{45 46} RT is known to promote immunogenic tumor cell death, antigen release, and antigen presentation, providing a rationale for its combination with ICIs.⁴⁷ However, clinical studies integrating PD-1/PD-L1 blockade with nCRT in pMMR LARC have yielded mixed outcomes,48,50 underscoring that ‘RT+PD-(L)1’ is not uniformly sufficient and that the dominant adaptive resistance pathways may differ by tumor context. Our data support a model of transient RT-driven immune activation followed by adaptive resistance, in which TIM-3 blockade adds greatest value when RT more consistently induces a TIM-3/PD-1-linked immunosuppressive program in pMMR disease.

Several limitations of this study should be acknowledged. First, the comparison between CT26-MC38 models is confounded by multiple variables beyond MMR status, including host genetic background and baseline immune contexture. Second, certain public datasets and clinical cohorts analyze pre-treatment vs post-treatment samples without strict patient-level pairing, and the SCRT subgroup sizes are modest, limiting statistical power to detect PD-1 changes. Third, the number of dMMR PDOs is small, leaving the incremental benefit of TIM-3 blockade in dMMR settings unclear.

In summary, our findings demonstrate that RT combined with TIM-3 and PD-L1 blockade synergistically augments antitumor immunity and promotes durable, systemic immune memory in pMMR CRC. The concordance of results across mouse models, PDO co-cultures, and clinical samples underscores the translational potential of the approach, while also highlighting the context-dependent nature of the response that should inform future mechanistic studies and clinical trial design.

Data availability statement

Data are available on reasonable request.