All animal protocols were reviewed and approved by the Pfizer Global Research and Development Institutional Animal Care and Use Committee (IACUC). C57BL/6 mice, as well as STING Golden ticket mice (C57BL/6J-Sting1^gt/J) were procured from The Jackson Laboratory (Bar Harbor, ME) and BALB/c mice were procured from Charles River Laboratories (San Diego, CA).

HPK1-deficient mice (HPK1^-/-) were generated by flanking the region encompassing exon 6 to exon 9 of the Map4K1 gene with LoxP sites in the same orientation and then crossing these mice with a Cre deleter mouse strain (C57BL/6N-Gt(ROSA)26Sor^tm1(cre)) and then bred to homozygosity sourced from The Jackson Laboratory. All studies were performed in animal rooms which were temperature (20 – 26 °C) and humidity (30 – 70 %) controlled. The animals were under a 12 h:12 h light-dark cycle and had ad libitum access to water and Laboratory Rodent Diet. Animals had at least 3 days of acclimation prior to study initiation.

All murine cancer cell lines were cultured at 37 °C and 5% CO₂. CT26 (colon carcinoma), B16F10 (melanoma), EMT6 (breast cancer) were obtained from the American Type Culture Collection (ATCC, Manassas, VA); MC38-Kerafast cells were obtained from Kerafast (Boston, MA) and all cell lines were cultured as per manufacturer’s instructions. MC38-Ribas cells were obtained from Antoni Ribas lab cultured in DMEM supplied with 2 mM l-glutamine, 10% FBS, penicillin (100 U/ml) and streptomycin (100 μg/ml) (Thermo Fisher Scientific, Waltham, MA). All cell lines were maintained by Pfizer Oncology Cell Bank and were validated using standard microbial and short tandem repeat (STR) profiling tests at IDEXX BioAnalytics (Columbia, MO).

Cells were detached using 0.25% Trypsin-EDTA (Thermo Fisher, Waltham, MA) and washed and resuspended in growth medium. After counting, cells were irradiated using X-Rad 225 (see below) and then seeded in 12-well plates. For EMT6 cells 1×10⁴ cells were seeded, for all other cell lines 2.5×10⁴ cells were seeded per well in a 12-well plate. Cell proliferation was monitored using the Incucyte^® live-cell analysis system (Sartorius AG, Göttingen, Germany). Phase-contrast images were acquired at regular intervals, and cell confluence was quantified as the percentage of the culture surface area occupied by cells using the manufacturer’s integrated image-analysis software.

For RT a X-Rad 225 biological irradiator (Precision X-Ray, Madison, CT) was used at a dose output of 225 kV, 10 mA. Cells were irradiated using a 2 mm Aluminum filter and targeted irradiation of tumor bearing mice was performed using a 0.3 mm Copper filter with a 5 × 5 mm collimator. In the study with two-fractionation arm, RT was delivered using the same setup and dose rate with 24 h of separation between each fraction.

For in vivo syngeneic tumor models, 5 × 10⁵ MC38 cells or 2 × 10⁵ B16F10 cells were subcutaneously implanted on the right flanks of 8- to 10-week-old female C57BL/6 mice. EMT6 cells (2.5 × 10⁵) or CT26 cells (2.5 × 10⁵) were implanted on the right flanks of 8- to 10-week-old female BALB/c mice. 5 × 10⁵ MC38 cells were implanted in STING-Golden ticket and HPK1-deficient mice.

Tumors were measured at least twice weekly using a caliper on the longest dimension (length) and the longest perpendicular dimension (width). Tumor volume was estimated with the formula: (L × W²)/2. Animals were randomized and enrolled into treatment arms when tumors were between 75–150 mm³. Mice with tumor volumes greater than 2000 mm³ were euthanized by CO₂ inhalation.

To obtain single-cell suspensions, tumors were processed using the Tumor Dissociation Kit, mouse (Miltenyi Biotec, Bergisch Gladbach, Germany) according to the manufacturer’s instructions. Following dissociation, cells were resuspended in autoMACS^® Rinsing Solution with MACS^® BSA Stock Solution diluted 1:20 and the cell suspensions were passed through a 70-µm cell strainer.

Single cell suspensions were obtained from mouse spleen and lymph node samples by mechanical dissociation, using a plunger to push tissues through a cell strainer (70-µm nylon mesh). Spleen, lymph node and when necessary, tumor samples were treated with eBioscience 1X RBC Lysis Buffer (Thermo Fisher Scientific, Waltham, MA). An aliquot of each tumor single-cell suspension was taken for cell count. Cells were stained for 20 min with BD Horizon™ BUV395 Rat Anti-Mouse CD45 antibody (BD Biosciences, San Jose, CA; Cat. No. 564279). Cells were resuspended in BD Pharmingen™ Stain Buffer (FBS) containing 1 μg/ml Propidium Iodide (Thermo Fisher Scientific; Cat. No. P3566) for dead cell exclusion and mixed at a 1:1 (v/v) ratio with 123count eBeads™ Counting Beads (Thermo Fisher Scientific; Cat. No. 01-1234-42), before measurement by flow cytometry (see below).

For dead cell exclusion, the resultant single-cell suspensions were treated with Zombie Aqua Fixable Viability kit (Biolegend, San Diego, CA) according to manufacturer’s instructions. Fc receptors were blocked using the TruStain FcX antibody (Biolegend, San Diego, CA) before staining for 20 min at 4-8 °C in the dark with fluorochrome-conjugated antibodies (Supplementary Table 1). Following staining, cells were fixed using FluoroFix™ Buffer (Biolegend, San Diego, CA). To stain for FoxP3 (Forkhead Box P3) and Ki67, intracellular staining was performed using the eBioscience™ Foxp3 Transcription Factor Staining Buffer Set (Thermo Fisher Scientific, Waltham, MA) following manufacturer’s instructions. The stained cells were analyzed using the BD LSRFortessa™ flow cytometer (BD Biosciences, San Jose, CA). Data was analyzed using FlowJo 10 (FlowJo, LLC, Ashland, OR). The gating strategy for the flow cytometric analysis is presented in Figure 1 (tumor), Supplementary Figure 1A (spleen), and Supplementary Figure 1B (tumor-draining lymph nodes, TDLN).

Tumors from treated mouse models were harvested on day 6 after RT and dissociated as described for flow cytometry (above) to obtain single-cell suspensions. Immune cells were isolated from cell suspensions using CD45 (TIL) MicroBeads, mouse (Miltenyi Biotec, San Diego, CA; Cat. No. 130-110-618) following manufacturer’s instructions. Isolated cells were counted using Cellaca MX Cell Counter (Nexcelom Bioscience, Lawrence, MA). Cells were processed for scRNAseq using the 10x Genomics chromium v3 3’ chemistry pipeline according to the manufacturer’s instructions. Cells were resuspended in 0.04% BSA in PBS at a concentration of 1 × 10⁶ cells/ml. Cells were loaded onto a Chromium Single Cell B Chip (10x Genomics, Pleasanton, CA) to aim for target cell recovery of 8 × 10³ cells. Library construction was performed according to Chromium Single Cell 3' GEM, Library & Gel Bead Kit v3 protocol (10x Genomics). Libraries were sequenced using the NovoSeq 6000 platform (Illumina; San Diego, CA) aiming for a minimum of 200 × 10⁶ reads per library (25,000 reads per cell).

The animals were euthanized at different time points, and tumors and tissues were harvested and fixed in 10% neutral buffered formalin. Single chromogenic IHC assay for CD8a was performed on 5-micron sections of formalin-fixed and paraffin embedded tumor tissue. Briefly, tissue sections were loaded onto a Leica Bond III instrument and deparaffinized, followed by pretreatment with epitope retrieval solution 2 (Leica Biosystems) for 20 minutes and then blocking buffer for 10 to 20 minutes. The anti-CD8a antibody (rat anti-CD8a, clone 4SM15, 1:2000; catalog number 14-0808-82; Thermo Fisher Scientific, Waltham, MA); or anti-FoxP3 antibody (rat anti-FoxP3, clone FJK-16s, 1:200; catalog number 14-5773-82; Thermo Fisher Scientific, Waltham, MA); was incubated for 20 minutes followed by biotinylated anti-rat linker antibody applied for 15 minutes prior to detection and color visualization by Refine diaminobenzidine polymer. A coverslip was applied on slides and scanned on Aperio AT2 whole-slide scanner (Leica Biosystems, Vista, CA) at 20X magnification.

TGI was assessed in non-clinical wildtype animal model studies using analysis of covariance (ANCOVA). Log-transformed tumor volumes, adjusted for baseline measurements, were compared between radiation therapy (RT) dose groups and the vehicle (0 Gy) control group within each model The null hypothesis stated that RT does not inhibit tumor growth, while the one-sided alternative hypothesis posited that RT results in tumor growth inhibition relative to the vehicle control. In TGI analyses comparing RT effects in the STING-deficient and wild-type MC38-K animal model, ANOVA tests with interaction between treatment and animal model were performed. To preserve growth differences between wildtype and engineered mice no baseline adjustments to the log-transformed tumor volumes were performed. In this analysis, the null hypothesis stated that RT effects are similar between wildtype and genetically modified mice, while the one-sided alternative hypothesis stated that RT tumor growth inhibition is higher in the wildtype animal model. ANOVA/ANCOVA analyses were conducted on the latest study day when all animals remained on study. This approach was taken to minimize potential bias associated with early removal of animals due to welfare concerns.

Time-to-event analyses were conducted to compare RT dose groups with the control group within each animal model. In these analyses, an event was defined as a tumor reaching a volume of 1500 mm³; observations not meeting this criterion were considered censored. Group comparisons were performed using log-rank tests with one-sided p-values to assess differences in time-to-event distributions. Kaplan–Meier curves were generated to summarize and visually represent the time-to-event data for C57BL/6 tumor-bearing mice treated with varying RT doses.

Tumor cell lines treated with varying doses of RT and control were evaluated for in vitro confluency. Four-parameter logistic regression models were fitted to replicate confluency curves. The estimated slope parameters from individual replicates were then compared between each treatment group and the control using ANOVA tests. The null hypothesis stated that RT does not inhibit cell growth, while the one-sided alternative hypothesis asserted that RT delays cell growth relative to the control.

Cell population frequencies obtained from flow cytometry assays and BioD scores were analyzed using a two-tiered statistical approach. The Shapiro–Wilk test was first applied to assess the normality of each cell population frequency distribution. When data significantly deviated from normality (p-value < 0.05), a log transformation was applied. If the transformed data met the normality assumption, group comparisons were performed using analysis of variance (ANOVA). If normality was not achieved, non-parametric Mann–Whitney tests were used for pairwise comparisons. The null hypothesis stated that RT does not alter cell population frequencies, while the two-sided alternative hypothesis posited that RT induces changes in cell population frequencies relative to the control. For TIL analysis experiments, tumors were not randomized on the day of treatment to avoid treatment-induced shrinkage that could preclude subsequent TIL isolation. All tumors were ≤5 mm in each dimension to accommodate the collimated irradiation field. Consequently, differences in tumor size between treatment groups were not tested.

Correlation analyses were conducted to evaluate the hypothesis that tumor hot/cold status may predict RT benefit. Spearman correlation coefficients were calculated between TGI estimates and baseline frequencies of CD8⁺ T cell populations or BioD scores, both within individual dose levels and across all dose levels combined. All p-values reported in this manuscript are unadjusted (raw). Statistical significance is denoted as follows: p < 0.05 (*), p < 0.01 (**), and p < 0.001 (***); “ns” indicates non-significant results. No corrections for multiple comparisons were applied in these analyses. The results are intended to support the generation of biological hypotheses which require further validation. We acknowledge the potential for false discoveries and imprecise effect size estimates. All statistical analyses were performed using R software, version 4.5.1. The data is representative of 1–3 experiments. The sample size for each experiment is mentioned in the Figure legend.

To test how immune cell infiltration relates to tumor-targeted RT-induced TGI and survival, we categorized tumors based on overall immune cell, CD8⁺ T cell infiltration and distribution. We chose syngeneic tumor models on two commonly used Mus musculus strains, namely BALB/c (CT26 and EMT6, T helper 2-prone) and C57BL/6 (MC38 and B16F10, T helper 1-prone). Because MC38 tumors are not available from ATCC and there may be different strains being used among the scientific community, we included these from two different sources: MC38-R tumors were obtained from Antoni Ribas and MC38-K tumors were obtained from Kerafast.

To compare syngeneic tumor models, we quantified TILs (Figure 2A; Supplementary Figure 2A) and determined the relative proportions of major immune cell types: macrophages, monocytes, granulocytes, CD8⁺ T cells, regulatory T cells (Tregs), B cells, DCs and NK cells (Figures 2B–I, as % of CD45 cells and Supplementary Figures 2B–I, as # cells per mg of tumor). We observed strong differences among the five models: B16F10 had the lowest number of immune cell infiltration, whereas EMT6 and CT26 showed the highest infiltration (Figure 2A; Supplementary Figure 2A). Additionally, the CT26 model had the largest fraction of CD8⁺ T cells (> 4% of CD45 cells; Figure 2E). The two MC38 models showed an intermediate immune cell infiltration (Figures 2A–I). Strikingly, EMT6 tumors contained the smallest fraction of CD8⁺ T cells (<0.5% of CD45 cells) (Figure 2E), despite having a large CD45⁺ immune cell population. Interestingly, B16F10 tumors had >4% of CD8⁺ T cells within the small pool of immune cells in the TME (Figure 2E) and was the tumor model with lowest number of CD8⁺ T cells per mg tumor (Supplementary Figure 2E). Monocytes and macrophages together accounted for >50% on immune cells present in all tumors (Figures 2B, C). Figure 2J summarizes inflammatory immune cell distributions across tumor models.

We utilized IHC to assess the intratumoral spatial location of cells expressing CD8α, the vast majority of which represents CD8⁺ T cells. Tumor models showed strong differences in intratumoral distribution, with EMT6 and CT26 tumors representing the extremes: While CD8α⁺ cells were rarely observed towards the center of EMT6 tumors (Figure 3A), they were readily observed in CT26 tumors (Figure 3B). To quantify this observation, we developed a biodistribution (BioD) score by dividing the tumor area into ten annular zones to compute the average CD8α cell density per zone (Supplementary Figure 3A). These zones were then grouped into three regions: core (C), intermediate (I), and periphery (P), with average densities calculated for each. The BioD score was derived from the pairwise ratios of these average densities across the regions (Supplementary Figure 3B). Because of the high melanin content of B16F10 melanoma tumors, BioD scores could not be reliably determined (data not shown). For all pairwise comparisons (C:P; C:I; I:P), CT26 showed a BioD score of >1, indicating an enrichment of CD8α⁺ cells towards the core of the tumor, relative to the periphery (Figure 3C). BioD scores for both MC38 tumor types were >0.5 and <1.0, representing a homogenous CD8α⁺ cell distribution, while EMT6 tumors showed consistently low BioD scores (<0.5), demonstrating an exclusion of CD8α⁺ cells from the tumor center and an enrichment towards the periphery. Based on the characterization of tumors by flow cytometry and IHC, we classified the tumor models from hot to cold phenotype with CT26 representing a hot phenotype and B16F10 a cold phenotype (Figure 2J).

Mice with established tumors (B16F10, CT26, EMT6, MC38-R and MC38-K) were treated with low (6 Gy), intermediate (12 Gy) and high (2 × 12 Gy, 24 Gy given in two fractions on two consecutive days) dose of tumor-targeted RT and evaluated for in vivo TGI (Figure 4A; Supplementary Table 2: TGI) and survival benefit (Figure 4B). Compared with untreated tumors, RT led to a dose-dependent TGI (Figure 4A) and a significant increase in survival (Figure 4B) for CT26 (hot) tumors at all doses evaluated, with TGI values of 64.8%, 83.9%, and 91.1% for low, intermediate, and high-dose RT, respectively. In the other tumor types, only intermediate- and high-dose RT led to significant TGI and survival benefit. The weakest RT response was seen in B16F10 tumors (cold), with TGI values of -0.4%, 47.1%, and 61.3% for low, intermediate, and high-dose RT, respectively; only high-dose RT led to a significant increase in survival. EMT6, as well as both MC38 tumors both responded strongly to intermediate and high-dose RT, showing similar TGI levels at both doses (EMT6: 86.6% vs. 83.0%; MC38-R: 72.9% vs. 84.1%; MC38-K: 76.7% vs. 86.4%), suggesting intermediate dose was sufficient to reach maximal benefit.

To understand the growth inhibitory effects of radiation on the cancer cells per se., in the absence of immune control, we treated monolayer cells in vitro with very low (3 Gy), low (6 Gy) and intermediate (12 Gy) irradiation (Figure 4C), using lower doses than in vivo, because cultured cells are more radiosensitive. Interestingly, B16F10 cells were most strongly affected by irradiation, a response comparable to CT26 cells. EMT6, MC38-K, and MC38-R cells tolerated low-dose irradiation but were sensitive to intermediate-dose treatment. For all cell lines, high-dose irradiation almost completely blocked tumor cell growth in vitro.

In summary, we found that intrinsic cell sensitivity to irradiation is not an accurate predictor for in vivo TGI by RT. After normalizing for dose, in vivo TGI demonstrated a reproducible positive association with CD8⁺ T-cell frequency and BioD scores (Supplementary Table 3). A striking example is provided by cell lines B16F10 and CT26, which exhibit similar sensitivity to in vitro irradiation. However, when observing in vivo tumor growth, tumors derived from cell line B16F10 display relatively lower sensitivity to RT compared to CT26-derived tumors suggesting that the TME and TILs could be contributing to in vivo TGI.

To better understand immune response dynamics and therapeutic implications, we analyzed TILs six days after intermediate and high-dose RT using multiparametric flow cytometry. Weights of tumors analyzed are shown in Supplementary Figure 4A. Treatment with RT induced complex tumor model-specific changes in immune cell infiltration (Figures 5A–H). Analysis of the relative contribution of each immune cell type (macrophages, monocytes, granulocytes, CD8⁺ T cells, NK cells, Tregs, B cells and DCs) to this immune cell infiltration by multi-parametric flow cytometry revealed a strong diversity between tumor types.

Macrophages and monocytes, which often adopt an immunosuppressive phenotype in tumors (33), were prominent in CT26, MC38, and EMT6 tumors. RT strongly reduced macrophages in CT26 and MC38-R tumors but not in EMT6 tumors (Figure 5A). Macrophages were much less frequent in MC38-K and B16F10 tumors, and only high-dose RT reduced their abundance in MC38-K tumors.

Monocytes, precursors to macrophages (34) and M-MDSCs, contribute to an immunosuppressive TME. Neutrophils, a type of granulocyte, can also acquire immunosuppressive function as PMN-MDSCs (35). Distinguishing MDSCs from neutrophils or monocytes by flow cytometry remains challenging due to substantial phenotypic overlap (36). Canonical surface markers are broadly expressed across mature myeloid cells and hematopoietic precursors. High-dimensional profiling revealed that cells within conventional MDSC gates exhibit extensive heterogeneity and lack a distinct transcriptional or functional signature (37). MC38-K tumors had a large monocyte fraction, which decreased after intermediate-dose RT (Figure 5B). Monocytes increased in CT26, MC38-R, and B16F10 after both RT doses, while in EMT6, only high-dose RT caused an increase. Granulocytes, abundant in EMT6 tumors, increased slightly after intermediate-dose RT (Figure 5C). In CT26, both intermediate- and high-dose RT led to their reduction. In B16F10, only high-dose RT caused a significant reduction, with intermediate-dose RT showing a similar trend.

CD8⁺ T cells are the key effector cells and play an important role for the abscopal effect (38). These were increased following intermediate-dose RT in CT26, MC38-R and MC38-K, but decreased in B16F10 tumors (Figure 5D; Supplementary Table 2: CD8T). No changes were observed in EMT6 tumors. Surprisingly, high-dose induced no changes in CD8⁺ T cells in any tumor type. The proportion of exhausted CD8⁺ T (T_EX) cells (PD-1⁺TOX⁺) did not increase following RT (Supplementary Figure 4B). NK cells, highly cytotoxic effector cells but often inhibited by an immunosuppressive TME (39), increased in CT26 and MC38 after RT (Figure 5E; Supplementary Table 2: NK). Only high-dose RT increased NK cells in B16F10, with no change in EMT6. No tumor type showed a reduction in NK cells post-RT.

Strikingly, FoxP3⁺ regulatory T (Treg) cells, CD4 T cells that suppress anti-tumor immune responses, were not reduced in any tumors following radiation. Instead, radiation increased the fraction of Tregs among immune cell infiltrates in CT26, MC38-R and MC38-K tumors (Figure 5F).

IHC analysis of the intermediate–dose RT cohort confirmed the flow cytometry findings: CD8α staining was sparse in B16F10 and EMT6 tumors, compared with MC38 and CT26 tumors. IHC of tumors 6 days after RT showed an increased frequency of CD8α⁺ cells (Supplementary Figure 5A). Similarly, FoxP3⁺ cells, which were less frequently observed than CD8α⁺ cells, were increased after RT (Supplementary Figure 5B).

Intra-tumoral B cells, which are commonly immunosuppressive were abundant in B16F10 tumors, whereas they were a smaller fraction of immune cells in MC38-R, MC38-K, and EMT6 tumors. There was a dose-dependent reduction in B cells following RT in all tumor types except for CT26 (Figure 5G). DCs were reduced after RT in CT26 and MC38-K tumors with either dose, but unchanged in the other tumor models (Figure 5H). In summary, RT induced complex and, in some cases, inverted dose-response effect and tumor-specific alterations in immune cell infiltration. Macrophages and DCs were reduced in select tumors, whereas monocytes and granulocytes increased in others, and key effector cells such as CD8⁺ T cells, NK cells, and Tregs were modulated in a dose- and context-dependent manner.

Changes in the cellular composition of distal immunological tissues indicate systemic effects of tumor-targeted RT. We therefore analyzed spleens of tumor bearing mice after treatment. RT induced distinct, but modest immune changes in the spleen, with the MC38 tumors showing the most pronounced effects (Supplementary Figures 6A–J). For example, NK cells increased significantly in MC38-R mice with both RT doses (Supplementary Figure 6D).

We also analyzed the TDLN (Supplementary Figures 7A–G). Lymph nodes of young healthy mice are dominated by two populations: 65 – 75% T cells and 10 – 15% B cells (40, 41). In the TDLNs of mice with MC38 tumors, there were notably fewer T cells at baseline. RT induced only marginal changes in T and NK cell populations in TDLNs (Supplementary Figures 7A, F). When analyzing overall B cells, we noted only a slight reduction after intermediate-dose RT in MC38-R and EMT6 tumors. Upon activation by their cognate antigen, B cells can initiate the formation of germinal centers (GCs), which are structures in secondary lymphoid organs where GC B cells proliferate and undergo selection to produce high-affinity antibodies (41). In tumors, GCs form within tertiary lymphoid structures, but since syngeneic tumor models usually lack these structures, we analyzed GC B cells in TDLNs instead (42). RT induced pronounced alterations in GC B cells (Supplementary Figure 7E). In mice bearing CT26 or EMT6 tumors, both doses prompted a decline in GC B cells, with high-dose RT resulting in a more marked reduction for EMT6 TDLNs. Although TDLNs of B16F10 tumors initially contained very few GC B cells, intermediate-dose RT further diminished this population, while high-dose RT exhibited no discernible effect. Intriguingly, intermediate-dose RT significantly increased GC B cells in mice bearing both types of MC38 tumors. Conversely, high-dose RT did not enhance, and notably diminished, this population in TDLN of MC38-R tumors.

Flow cytometric analysis revealed an increase in CD8⁺ T cells in the TME after intermediate-dose (12 Gy) RT across multiple tumor types (Figure 5D). Therefore, we chose intermediate-dose RT to compare and characterize the TILs from CT26, MC38-R, -K, EMT6, and B16F10) tumors. We isolated CD45⁺ TILs from syngeneic tumors 6 days after RT and analyzed their gene expression using single-cell RNA sequencing (scRNA-seq). We combined the datasets from the MC38 variants R and K for joint clustering (labeled as MC38) to identify shared transcriptional populations in a common space, while datasets from the other tumor models were clustered independently to optimize resolution of model-specific cell states. To visualize immune cell diversity across the tumor models we used t-distributed stochastic neighbor embedding (t-SNE) plots and differentially expressed genes for each cluster can be found in Supplementary Table 4. Most TIL populations were found in all tumors, with the exceptions of B cells and CD8⁺ T cells, which were absent from the sampled cells isolated from EMT6 tumors (Figure 6A). Their absence from the t-SNE map likely reflects their low abundance, causing them to cluster with transcriptionally similar populations rather than forming distinct groups. Consistent with flow cytometry, monocytes were the predominant population of cells by scRNA-seq.

To identify similar CD8⁺ T cell clusters across different tumor types, we performed hierarchical clustering and identified 4 superclusters with similar gene expression (Figure 6B). To illustrate the identity of the superclusters, we used MC38 TILs as an example and examined the expression of established marker genes (Figure 6C). Among the identified clusters, CT26-T2, CT26-T5, CT26-T6, and MC38-T3 (Figure 6C) expressed Klrg1, a marker gene associated with effector T cells (T_EFF). These cells are highly functional and induce tumor cell death through the secretion of cytotoxic molecules and cytokines (43, 44). In contrast, no T_EFF clusters were detected in EMT6 and B16F10 tumors among the cells sampled, consistent with their overall low CD8⁺ T cell infiltration.

Progenitor or precursor exhausted T cells (T_PEX), identified by the transcription factor Tcf7, constitute a self-renewing subset that gives rise to exhausted T cells (T_EX) upon stimulation (45–49). Clusters MC38-T2 (Figure 6C), CT26-T3, and B16-T1 contained T_PEX cells; however, no such cluster was detected among cells sampled from EMT6 tumors.

Clusters MC38-T4 (Figure 6C), CT26-T4, and B16-T3 expressed marker genes associated with proliferation. Proliferating T cells (T_PROL), characterized by Mki67, are actively dividing and represent a transitional phase between precursor and T_EFF or T_EX cells (50, 51). Chronic antigen stimulation in infections and cancer leads to T cell exhaustion, a process regulated by the transcription factor TOX (52–54). T_EX cells, marked by high Pdcd1 expression, progressively lose their effector functions and proliferative capacity. Notably, clusters MC38-T1, CT26-T1, and B16-T2 exhibited marker gene signatures of T_EX cells.

CT26 tumors contained the most CD8⁺ T cell clusters, three of which showed low similarity to other clusters. CT26-T4 expressed both Mki67 (a proliferation marker) and Pdcd1 (a marker of T_EX cells) (Figure 6D). CT26-T2 partially expressed Mki67, while CT26-T5 expressed high Pdcd1. CT26-T6 expressed neither Mki67 nor Pdcd1 (Figure 6D).

We noted that RT induced a strong increase in the fraction of clusters with proliferative gene signatures, except for B16F10, where only 2 of 4 tumors analyzed showed this change (Figure 6E; Supplementary Table 2: T_PROL). Although Ki-67 staining is widely used to assess proliferation, and its half-life in cancer cells has been reported to be short (1–1.5 hours), studies in T cells have shown that Ki-67 protein can remain stably expressed after cell division (55, 56). Consistent with this, we observed substantially higher Ki-67 expression by intracellular flow cytometry (Figure 6F). Notably, RT led to an increase in Ki-67⁺ CD8⁺ T cells in tumor CT26 and MC38-K, but not in MC38-R and B16F10, although a trend was observed in MC38-R.

Across the remaining T cell clusters, treatment was associated with a significant decrease in T_PEX cells in CT26 tumors (Supplementary Figure 8A), a significant increase in T_EX cells in MC38R tumors (Supplementary Figure 8B), and a significant increase in T_EFF cells in MC38R as well as in one effector T cell subset in CT26 tumors (Supplementary Figure 8C).

Therefore, despite a much different expression baseline expression level, we found that flow cytometry confirms the scRNA-seq finding that RT leads to an increase in (recently) dividing CD8⁺ T cells.

Hierarchical clustering of macrophage clusters from all tumor models resulted in a total of 65 clusters (Figure 7A), showcasing the complexity of the monocyte derived immune cells. We found three super clusters containing monocyte/macrophage populations with high expression of signature genes of proliferation (Figure 7A, top left corner of the heatmap). Similar to what was found for CD8⁺ T cells, these showed the highest inter-tumoral similarity characterized by hallmark genes of proliferation and could be categorized into three distinct groups. The t-SNE plot and clusters of MC38 model as an example is shown in Figure 7B. The first group of macrophage clusters, denoted by green box (Figure 7A), exhibited pronounced upregulation of Cdc20 and Birc5. The second cluster, marked by the blue box, exhibited heightened expression of Mki67 and Top2a. Lastly, the third cluster, indicated by the red box, displayed the highest expression levels of minichromosome maintenance genes, such as Mcm6 and Mcm5. Most of these populations were strongly reduced after RT (Figure 7C, boxed with matching colors with clusters depicted in Figure 7A; Supplementary Table 2: MPROLg, MPROLr and MPROLb). However, no reduction was observed for the two B16F10 populations (B16-M14 and B16-M7) belonging to the three proliferating macrophage groups.

In case of EMT6, a reduction was observed only for EMT6-M10, but not for the other two populations EMT6-M8 and EMT6-M7 (Figure 7C). Since CT26 tumors showed particularly strong T cell infiltration and TGI after RT, we were interested in pro-inflammatory myeloid populations (Cxcl9⁺ and Cd40⁺ cells) in these tumors (Figure 7D). In two of the four tumors analyzed monocytic dendritic cells (moDCs) increased. Strikingly, macrophage population M3, which is characterized by high expression of Cd40 and Cxcl9, has recently been shown to be a key population for the response to CPI by recruitment of protective CXCR3⁺ T cells (57). Among the diverse immune cell types, myeloid and CD8⁺ T cells and their subsets were significantly altered due to RT. Thus, we evaluated the STING and HPK1 pathways which are key pathways for myeloid/macrophage and CD8⁺ T cell function and activity.

We identified four clusters (macrophage/monocyte clusters 2 and 8, as well as neutrophils and moDCs) in the MC38 tumors that we define as “IFN-responders” based on higher expression of IFN-response genes (Ifit3) compared to all other immune cell clusters (Figure 8A). We noticed a strong difference (~2×) in the cluster sizes between MC38-R and MC38-K TILs, with MC38-K tumors showing a much larger fraction of these cells than MC38-R tumors (Figure 8B). As a whole, these were unchanged in MC38-R tumors following RT but further increased in MC38-K tumors (Figure 8B). Macrophage clusters 2 and 8 also expressed high levels of Ly6c, consistent with a monocyte phenotype. Following RT treatment, moDCs and neutrophils were significantly increased in MC38-K tumors, while macrophage/monocyte cluster 8 and neutrophil cluster were reduced in MC38-R tumors (Figure 8B). Thus, MC38-R and -K provided an interesting opportunity to compare two very similar cell lines and the STING pathway.

Liang et al. found that STING activation recruited MDSCs to the tumor following tumor-targeted RT of MC38 tumors and suppressed the anti-tumor immune response (58). In this study, MDSCs were identified as CD45⁺CD11b⁺Ly6C⁺ cells. Based on the low frequency of Ly6C⁺ cells among CD11b⁺ cells identified by Liang et al., we assumed that these tumors reflect our MC38-R tumors and not MC38-K tumors. To confirm the role of these MDSCs in RT-mediated TGI we utilized STING^GT mice. The STING^GT mouse carries a point mutation (T596A) in Sting, leading to an isoleucine-to-asparagine substitution (I199N) in the STING protein (59).

Interestingly, RT significantly increased monocytes in MC38-R tumors growing in STING^GT and wildtype animals, but this increase was significantly reduced in STING^GT animals and accompanied by an increased frequency of macrophages compared with wildtype animals (Figure 8C). Strikingly, the opposite was observed in MC38-K tumors, where RT decreased Ly6C⁺ monocytes and increased macrophages (Figure 8D). Additionally, MC38-R tumors grew faster in STING^GT mice than in wild-type mice, but RT-induced TGI was not different between the two strains of mice with MC38-R tumors (Figure 8C). Most importantly, MC38-K tumor growth was similar in STING^GT and wildtype mice, RT lead to a significantly stronger TGI in STING^WT than STING^GT mice, showing a synergistic effect (Figure 8D). The loss of TGI difference in MC38-R tumors, indicates that recruitment of Ly6C⁺ monocytes diminishes the RT-induced anti-tumor immune response in MC38-R tumors (60), but not MC38-K tumor model.

Since we observed an induction of CD8⁺ T cell expansion following RT, we were interested in testing whether HPK1, a kinase primarily expressed in hematopoietic cells limits RT-induced T cell response. HPK1 limits T cell activity by phosphorylating the adapter protein SLP76 in the TCR pathway (61, 62). Prostaglandin E2 (PGE2), an immunosuppressive molecule found in the TME has been shown to further activate HPK1 (63). Deficiency of HPK1 suppresses tumor growth by elevating the anti-tumor immune response, particularly in combination with anti-PD-L1 therapy (64, 65). Therefore, HPK1 is a promising target for cancer immunotherapy. While HPK1 inhibitors have been previously described, our study aimed to assess the biological consequences of HPK1 ablation. Since small-molecule inhibitors inherently introduce variables such as potency, selectivity, and off-target effects, we opted to use knockout mice to directly assess HPK1 function without confounding factors.

We tested the effects of RT on TILs in the absence of HPK1 in MC38-R tumors (Figures 9A–C). Without RT treatment, MC38-R tumors in HPK1^-/- mice contained a slightly reduced fraction of macrophages (Figure 9A). In HPK1^+/+ and HPK1^-/- mice, tumor-targeted RT led to a strong and significant reduction of macrophages, but there was no significant difference between treated tumors in both groups. NK cells on the other hand were slightly increased in tumors in HPK1^-/- mice, but after RT treatment there was again no difference for this cell population in tumors in HPK1^-/- and HPK1^+/+ animals (Figure 9B). Finally, CD8⁺ T cells were significantly increased in tumors growing in HPK1^-/- deficient mice (Figure 9C). RT led to an increase of the fraction of CD8⁺ T cells in HPK1^+/+ mice to about the same size as in HPK1^-/- mice without treatment, but this population was strongly increased in HPK1^-/- after RT, comprising about 12% of the total intra-tumoral immune cell fraction.

While HPK1-deficiency alone did not lead to an increased survival of MC38-R tumor bearing mice, RT synergized with HPK1-deficiency leading to a significantly increased survival (Figure 9D). In summary, our data demonstrate that HPK1 deficiency increases CD8⁺ T cell infiltration and synergizes with RT to improve survival, suggesting that the improved outcome may result from an augmented anti-tumor immune response.