The colorectal cancer tissues obtained from surgical specimens were retrospectively collected from 55 patients diagnosed between 2000 and 2014. Patients were treated by neoadjuvant treatment (RT with 5-fluorouracil-based CT (nCRT; n=30) or without CT (nRT; n=13)) and radical surgery by total mesorectal excision. Among the patients that received nRT, 7 had long course (LC) and 6 had short course (SC) RT. All patients with nCRT received LC RT (TME). LC, that is, around 45 Gy in 25 fractions with delayed surgery and the nRT ‘short course’ that is, 25 Gy in 5 fractions with early surgery. Target volumes of RT were treated according to the International consensus guidelines on Clinical Target Volume delineation in rectal cancer.21 Response to neoadjuvant treatment was assessed by the Dworak tumor regression grading score22 from no regression/no response (Dworak 0) to a near tumor control/a near complete response (Dworak 3): Dworak 0 (n=8); Dworak 1 (n=14); Dworak 2 (n=13); Dworak 3 (n=8).

C57BL/6AnCrl and BALB/cAnCrl female mice were injected subcutaneously with 2.10⁵ TC1, CT26, MC38, B16F10, or B16OVA cells in 100 µL of phosphate buffered saline (PBS 1×, Gibco, France) into the left flank. Tumor growth was monitored every 3 days with a caliper and groups were formed when tumor reached a mean diameter of 50–60 mm². All CT procedures were conducted in the Pharmacy Department of the University Hospital of Besançon (France). Mice were administered a single dose of intraperitoneal cisplatin (5 mg/kg) and 5-fluorouracil (25 mg/kg). After 2 days, mice were anesthetized by injection of ketamine (100 mg/kg) and xylazine (10 mg/kg) and received single-fraction beam photon irradiation (8 Gy) using Clinac 2100 CD (Varian Medical System) with a maximal energy of 6 MV and in 2D with one beam. Mice from control group were injected with the solvent used to dilute the drug (the Pharmacy Department of the University Hospital of Besançon, France). Anti-mouse CTLA-4 antibody (9H10, Euromedex) and anti-PD-1 (RMP1-14, Euromedex) injections (200 µg/mouse two times a week for 2 weeks) started 2 days prior to CT and 3 days after RT, respectively. All experimental studies were approved by the local ethics committee in accordance with the European Union’s Directive 2010/63.

A unilateral linear-by-linear association test was applied to assess associations between PD-L1 score and patient treatment. The associations between clinicopathological parameters and cells densities or gene expression level were examined using T-test or Wilcoxon-Mann-Whitney as appropriate. Partitioning around medoids algorithm was used to cluster gene expression matrix. A logistic regression model was applied to test association between ICI PD-L1, PD1, LAG3 and CTLA4, related genes and TME inflammatory status (Hot/Cold). Correlation matrices were constructed by calculation of Pearson correlation coefficients for all marker combinations. Data visualizations and statistical analyses were achieved in R version 3.6.3 with ggpubr, ggplot2, coin, ComplexHeatmap and cluster packages. For preclinical models, the statistical analyses were performed with the Graphpad Prism 6 software (La Jolla, California, USA). Multiple groups comparison was performed using the one-way or two ways analysis of variance or Kruskal-Wallis tests and two groups comparisons were done with the Mann-Whitney test. P values lower than 0.05 were considered as statistically significant (*p<0.05, **p<0.01, ***p<0.001). Mice survival was estimated using the Kaplan-Meier method and the log-rank test. The event in the mice survival test was either when the size of the tumor reached 300 mm² or the death of the mouse. The log-rank test was corrected for multiples comparison using the Bonferroni method. P values lower than 0.01 were considered as significant.

More detailed information is available in the online supplemental material section.

To study the changes of adaptive T cell immunity in TME induced by CRT, we performed Nanostring-based transcriptomic analysis of tumor samples from patients with colorectal cancer (n=43) treated either by neoadjuvant radiation (nRT) or chemoradiation (nCRT) (figure 1A). The distribution of clinical parameters (age, gender, cancer location clinical (cTNM) and pathological (pTNM) tumor classifications) was well-balanced among treatment groups (online supplemental table 1). The expression level of T-cell-related genes was highly variable. Overall, 7% (n=3) exhibited a “hot signature” with high expression of genes related to Th1 immune orientation (eg, Tbx21, Stat1, Stat4, Irf1, Ifng,), cytotoxicity (Gzma/b/h/k/m, Gnly, Prf1), T-cell activation (eg, Icos, Tnfrsf9), migration (eg, Cx3cl1, Cxcl9, Cxcl10), while 33% (n=14) and 60% (n=26) exhibited the “lukewarm” and “cold” signatures, respectively, with low expression of such genes (figure 1B). All patients except one with the lukewarm/hot signature belonged to the nCRT group. In contrast, most patients treated with nRT presented non-inflamed immune signature in the TME related to a cold tumor. Notably, we found positive correlations (all p<0.01) between the immune signatures (cold, lukewarm, and hot) and the expression level of immune checkpoints such as CTLA-4, PD-L1, PD-1 and LAG3, suggesting the ability of nCRT to induce a higher inflamed TME than nRT alone (figure 1C). Accordingly, immunohistochemistry analysis showed a higher density of PD-1/PD-L1 or LAG-3 positive cells in the TME of nCRT group compared with patients who did not receive neoadjuvant treatment (figure 1D,E).

Next, we focused on PD-L1 expression (on immune cells or tumor cells). We found that most colorectal tumors treated by nCRT (82%) exhibited high expression of PD-L1 on immune cells (score 3) whereas only 33% of tumor treated by nRT highly expressed PD-L1 (p<0.0001). Similar trends were observed when considering PD-L1 expression on tumor cells (figure 1D,E). These results support the ability of CRT to create a strong adaptive immune resistance phenomenon in TME.

Next, we addressed the relationship between the immune response induced by CRT and the efficacy of the treatment evaluated by histological response. To this end, we used the tumor regression grading of rectal cancer after neoadjuvant treatment (Dworak).22 Best histological responses (Dworak 2 and 3) were statistically greater (p=0.003) in the nCRT group than in patients treated with nRT alone (figure 1F). Notably, high inflamed TME was significantly associated with a high rate (>60%) of Dworak 2 and 3 scores, compared with the cold TME (38%; p=0.02; figure 1G), suggesting that the strong local immune activation induced by CRT may contribute to its effectiveness. Collectively, CRT appears more effective to induce a proinflammatory TME suitable for ICI action.

To study more extensively the immunological changes of TME induced by CRT, we performed a comprehensive translational study in mouse tumor models exposed or not to CRT. These tumor models consist of the lung epithelial carcinoma HPV positive TC1 and CT26 tumors in C57BL/6 and BALB/c mice, respectively. Tumor-bearing mice were separated into four groups: vehicle untreated (CTRL), CT, (5-fluorouracil plus cisplatin), RT (8 Gy focal radiation) and CRT group that consists to a single dose of 5-fluorouracil plus cisplatin, two radiosensitive cytotoxic drugs combined with a single fraction of 8 Gy radiation. To reflect locally-advanced cancer stages, treatments were initiated when the tumor mass reached 50–60 mm² (figure 2A). As expected, a single CRT treatment was more effective than RT or CT alone to delay tumor growth (p<0.05) but failed to achieve complete remission of tumors in the majority of mice (online supplemental figure S1).

First, we performed RNA sequencing of tumor-infiltrating immune cells isolated from tumor-bearing mice 7 days after the respective treatments (figure 2A). In TC1 model, CRT selectively regulated 1344 differentially expressed genes (DEGs; n=571 (42.5%), induced; n=773 (57.5%), repressed) versus <330 DEGs (13%) regulated by CT or RT; with only 536 (22%) similarly regulated in ≥2 therapies (figure 2B). Notably, CRT-induced DEGs were highly enriched in inflammatory responses (eg, Il1b, Cxcr3, Tnfsf4), host defense (eg, Zbp1, Clec4a4), T cell-mediated immunity (eg, CD8α, Ifng, Prf1, Il12rb,) Th1 polarization (eg, Tbx21, Cxcr3, Eomes) and neutrophil/myeloid cells activation (online supplemental figure S2). CRT-repressed DEGs mainly included extracellular matrix organization (Mmp3, Col16a1), or vascular development (Angptl1, Angptl2), which can be involved in tumor infiltration and invasiveness. All treatments repressed G alpha signaling and positively regulated genes involved in lipoprotein metabolism, which is known to be potentially affected by RT or CT.23

To determine the immune cell composition in TME, we used PanglaoDB database.24 The correlation matrix analysis between 1258 highly variables genes within according to GSEA revealed the existence of 27 cell subtypes belonging to myeloid and lymphoid lineages. These cells were mainly distributed in three clusters (clusters 5, 7 and 8). Cluster 5 is a group of genes mainly associated with B-cells and plasma cells. Cluster 7 is related to macrophages, monocytes and natural killer (NK)-specific genes and the cluster 8 was associated with different T cell subsets (online supplemental figure S2). Unsupervised clustering analysis showed that CRT triggered stronger expression of genes associated with T-cell activation (eg, Cd28, Lck, Zap70, Egr2), exhaustion (eg, Ctla4, Havcr2, Pdcd1, Lag3, Tigit), memory (eg, Cd44, Id2, Klrg, Prdm1), and function (eg, Ifng, Tnf, Prf1, Gzma/b, Fasl) than RT and CT (decrease in the range: CRT>RT> CT) (figure 2C,D). Similar results were observed with transcriptomic data from CT26 tumor model in BALB/c mice (not shown). To estimate the abundance of immune cells, we used Microenvironment Cell Populations-counter (MCP-counter) method based on a set of mouse/human orthologous genes.25 As shown in figure 2E, CRT-treated tumors had an enrichment of T-cells, NK cells and monocytes, but also myeloid-derived suppressor cells (MDSC) and neutrophil markers. In contrast, B cell markers were downregulated after CRT as compared with CT or RT, suggesting a shift towards a strong T cell-mediated immunity after CRT.

To confirm transcriptomic data, the immune cell infiltrates isolated from tumor-bearing mice were analyzed ex vivo by flow cytometry after the respective treatments (figure 3A). Compared with non-CRT-exposed tumors, CRT improved the accumulation of CD4+ and CD8+ tumor-infiltrating lymphocytes (TILs) within the TME both in TC1 and CT26 tumors (figure 3B,C). To track tumor-reactive CD8⁺ TILs, we performed E7_49-57-K^b dextramer stainings in HPV-16 +TC1 model. We detected a 2–3-fold increase in the percentage of E7-specific CD8⁺TILs in CRT-treated mice compared with the CT and RT groups (30% vs 10%–15%) (figure 3D). Kinetic analysis showed that the E7-specific CD8⁺ TILs expansion reached its peak on day 7 after CRT and then returned to a basal level on days 15–21, corresponding to the contraction phase of a T-cell response (figure 3E). The high expansion of CD8⁺ TILs activated by CRT positively correlated with the increased expression of Cxcl9 and Cxcl10 and Cxcl16 transcripts, key chemokines ensuring effector T cells migration towards the tumor (figure 2D). Furthermore, we found that CD8⁺ TILs from CRT-treated mice produced higher level of effector cytokines such as, IFN-γ, TNF-α, and granzyme B (GzmB) as compared with each monotherapy (figure 3F). Similar observations were made in CT26 colorectal tumor model following CRT, suggesting the potent capacity of CRT to promote local activation of antitumor T cell immunity (online supplemental figure S3).

To determine whether the efficacy of CRT relies on effector T-cell, tumor-bearing mice were depleted in CD8 or CD4 T-cells before or after CRT. In contrast to T cell depletion prior treatment that did not impair CRT efficacy, we showed that the efficiency of CRT significantly decreased when T-cells were depleted from day 7, at the peak of T cell infiltration; this was also the case in CT26 model (figure 3G,H, and online supplemental figure S3). Furthermore, the efficacy of CRT was significantly impaired in IFN-γ knockout mice as compared with wild type mice (figure 4I).

To understand how this in situ T cell activation occurred, we focused our attention on CD103⁺ dendritic cells (DCs). CD103⁺ DCs are subset of cross-presenting DCs that are critical for antitumor T-cell stimulation.26 27 We showed that CRT triggered strong CD103⁺ expression as well as costimulation receptors such as CD40, CD80 on DC isolated from tumor draining lymph nodes (TDLN), in accordance with the transcriptomic signatures found in online supplemental figure S4. This suggests that DCs could be involved in antitumor CD8⁺T-cells priming after CRT. To test this assumption, we isolated CD11c⁺ DCs from TDLN of 3 days after CRT or from untreated B16-OVA-bearing mice. The DC were co-cultured in presence of anti-OVA T-cells from OTI or OTII mice and the specific production of IFN-γ was assessed by the ELISpot assay (online supplemental figure S4). In contrast to DCs isolated from untreated mice, those from CRT-treated mice triggered strong IFN-γ production by naïve OTI and OTII T-cells (online supplemental figure S4). Thus, CRT improved CD103⁺ DC activation, cross-presentation and the priming of antitumor T cells responses.

Collectively, CRT synergistically promotes robust CD8 T-cell expansion and activation in the TME and requires an active IFN-γ pathway in vivo.

Tissue resident memory T lymphocytes (T_RM) are a subpopulation of memory T cells, which ensure immune surveillance in non-lymphoid tissues. Increased T_RM infiltration in TME is associated to a better prognosis in several cancers.28 29 Furthermore, recent findings reported that the presence of T_RM in TME is associated to response to anti-PD-1/PD-L1 therapy.28 30 Thus, we wondered if the local inflammation promoted by CRT could also trigger T_RM activation (figure 4A).

As shown in TC1-bearing mice, the key genes related to T_RM signature such as retention and adhesion molecules ITGAe (CD103), ITGA4 (CD49a), RGS1 and RGS-2, immune checkpoints (eg, CTLA4, HAVCR2, Pdcd1) as well as transcription factors associated with T_RM (eg, Notch-RBPJ, Runx3) were highly upregulated after CRT as compared with RT and CT (figure 4B). Next, T_RM monitoring within TME by flow cytometry at day 7 post therapy showed that the rate of CD103⁺CD49a⁺CD8⁺T_RM was twofold to threefold increased after CRT as compared with RT, CT and untreated TC1-bearing mice thus, supporting the transcriptomic data (figure 4C,D). Interestingly, around 50% of tumor-specific CD8⁺ TILs displayed CD103⁺CD49a⁺ CD8⁺T_RM phenotype in the CRT-treated group (figure 4E). kinetic analysis revealed that a substantial proportion of E7-specific CD8⁺ T_RM was still detected 3 weeks after CRT and RT but not in CT-treated and untreated mice (figure 4F). Next, we investigated the functionality of the tumor-specific CD8⁺ T_RM and found that the frequencies of IFN-γ and TNF-α producing CD8⁺ T_RM in response to E7_49-57 peptide was higher within TILs from CRT-treated mice than in control groups (figure 4G). Of note, no obvious changes of CD103⁺CD49a⁺ CD8⁺T_RM were detected in the TDLN or in the spleen from tumor-bearing following CRT (not shown). In addition, the ability of CRT to activate tumor-specific CD103⁺ CD49a⁺ CD8⁺T_RM was also confirmed in CT26 tumor-bearing mice by tracking AH1 as model of tumor-associated antigen (figure 4H–J).

In patients with rectal cancer (online supplemental figure S5), we also found that the transcripts of T_RM-related genes, such as (ITGAE) CD103 and (ITGB4) CD69, together with CD8 gene were simultaneously upregulated in the TME of patients treated by nCRT as compared with nRT-treated patient. Furthermore, in CRT-exposed tumors, a strong positive correlation was found between T_RM markers, CD8 T cells, PD-1 and LAG-3. The strength of the T_RM signature and the expression of PD-1 and LAG-3 was significantly higher (around two times) in patients presenting the best histological response after neoadjuvant therapy (Dworak >2). This effect was especially observed in nCRT-exposed tumors as compared with colorectal patients treated with nRT alone (online supplemental figure S5).

Thus, our results demonstrate the ability of CRT to expand functional tumor reactive CD103⁺ T_RM in the TME.

Transcriptomic data from both patients with colorectal cancer and mouse TME indicates that CRT synergistically promotes the upregulation of genes related to immune checkpoint ligand/receptors. To confirm this statement, the expression of immune checkpoint exhaustion receptors was assessed by flow cytometry from TILs isolated after CRT or monotherapy.

We detected high percentages of exhausted phenotype PD1 +TIM-3+CD8⁺ TILs in CRT-treated mice as compared with the control groups (figure 5A,B). Similar phenotypic features were observed when focusing on tumor-specific CD8⁺ TILs. Indeed, a higher rate (>65%) of E7-specific CD8⁺ TILs displayed exhausted phenotype by coexpressing PD-1 and TIM-3 after CRT, as compared with RT and CT (50% and 30%, respectively) (figure 5C,D). The kinetic analysis showed that PD1⁺TIM-3⁺ anti-E7 CD8⁺ TILs reached a peak on day 7 and decreased 3 weeks later during the contraction phase of T cell response (figure 5E).

In parallel, we assessed the kinetic of PD-L1 expression in the TME by flow cytometry and found overexpression of PD-L1 in TC1 tumor tissues exposed to CRT. The induction of PD-L1 expression on TC1 cells started at 3 days after CRT, reached a peak at day 7 and decreased to a basal expression, similar to the kinetic of exhausted phenotype CD8⁺ TILs expansion (figure 5F). Similar expansion of exhausted phenotype CD8⁺TILs together with PD-L1 was also found in CT26-bearing mice (online supplemental figure S6). Moreover, transcriptomic data showed that expression of galectin 9, the ligand of TIM-3, was positively correlated with TIM-3⁺ CD8+TILs expansion (data not shown). All above suggest that CRT triggers strong adaptive immune resistance phenomenon in the TME.6

Previous evidence support the ability of RT to promote suppressive cells such as regulatory T cells in the TME which limit antitumor immunity.31–33 Here, we found that as RT, CRT induced high expansion of regulatory T-cells (Tregs) in the TME (figure 5G,H). By using DEREG mice, we demonstrated that in vivo depletion of Foxp3⁺Tregs with diphteria toxin (DTX) significantly increased CRT effectiveness (figure 5I–K). Furthermore, DTX treatment combined with CRT induced a high expansion of functional E7-specific CD8⁺TILs (figure 5L,M). Thus, CRT induces immune suppressive Treg in the TME.

All the above data suggest that CRT represents a potent approach to convert a cold TME toward a hot one suitable for ICI activity.

To this purpose, we first investigated the efficacy of CRT combined with a single agent anti-PD-1 or anti-CTLA-4 in HPV 16+TC1 tumor model. We found that these combinations, mainly with anti-CTLA-4 appeared more effective for delaying tumor growth and to increase mice survival compared with ICI alone or CRT (online supplemental figure 7). However, this approach failed to cure tumor in most mice, suggesting that this was not an optimal regimen.

Previous reports showed that concurrent blockade of PD-1 and CTLA-4 pathways may have complementary functions, leading to better tumor control.34

Thus, we sought out to evaluate the dual blockade of CTLA-4 and PD-1 pathways combined with CRT in order to sustain CRT-induced antitumor T cells.

To this end, TC1-bearing mice were treated with CT or RT, or CRT combined with dual anti-CTLA-4 and anti-PD-1 therapy (anti-CTLA-4/PD-1) and the control groups received the dual immunotherapy alone or the cytotoxic therapies (CRT, RT or CT) (figure 6A). The results showed a strong antitumor effect of the CRT plus anti-CTLA-4/PD-1 combination therapy. Indeed, more than 80% of mice treated with this combination achieved complete tumor response and experienced significant improvement of survival as compared with the control groups (figure 6B,C). Of note, the bitherapy anti-CTLA-4/PD-1 was inefficient to eradicate TC1 tumor reinforcing the capacity of CRT to sensitize this tumor to ICI. We observed several weeks after therapy, the occurrence of vitiligo in mice cured after the CRT plus anti-CTLA-4/PD-1 combo (figure 6D). This skin immune-related side effect occurs in patients with cancer during immune checkpoint therapy and is commonly associated with therapy effectiveness.35

To ensure that this combination strategy results to the establishment of a long-lasting protective antitumor immunity, we performed tumor rechallenge experiments (figure 6E). Indeed, all mice cured (>100 days after the combo CRT plus anti-CTLA-4/PD-1) rejected a second TC1 tumor graft, contrary to untreated control mice (figure 6F). Unexpectedly, these mice were also preserved from primary challenge with B16F10 melanoma, suggesting the establishment of a large diversity of tumor antigen specific T-cell responses (figure 6F). Accordingly, spleen T-cells from cured mice produced high amount of IFN-γ against a broad range of epitopes derived from E7 antigen expressed by TC1 cells as well as against MHC class I and II-restricted epitopes derived from mouse telomerase, as shared tumor-associated antigen (figure 6G,H). These results suggest that an epitope spreading phenomenon may also contribute to the efficacy of the CRT combined with anti-CTLA-4/PD-1 bitherapy.

In the CT26 model, which mimic microsatellite stable (MSS) colorectal cancer, we also found that CRT was able to sensitize this tumor to the dual anti-CTLA-4/PD-1 blockade in BALB/c mice. Notably, more than 75% of CT26 tumor-bearing mice treated with the combo CRT+anti-CTLA-4/PD-1 displayed complete tumor regression as compared with CRT combined with a single ICI (figure 7A).

Finally, we tested this combination approach in MC38 tumor, a second model of colorectal cancer, syngeneic in C57BL/6 mice. This tumor has additional advantage of being a microsatellite instable (MSI) CRC model known to be more sensitive to checkpoint inhibitors.36 Although tumor control was observed in all mice treated with anti-CTLA-4/PD-1, complete remissions were only achieved in MC38 tumor-bearing mice treated with the combo CRT+anti-CTLA-4/PD1 (60%) (figure 7B,C). Although the cured mice failed to reject primary TC1 graft (figure 7D,E), the growth of this tumor appeared to be slower in cured mice than in untreated control mice (figure 7E,F). In contrast, cured-mice successfully rejected a second challenge with MC38 cells and also exhibited functional CD8 T cells directed against A9M peptide, a neoepitope expressed by MC38 cells, thus demonstrating that the immune protection triggered by CRT was tumor specific (figure 7G,H).

Collectively, our results strongly support the capacity of CRT to establish a highly inflamed TME, suitable for ICI effectiveness.