In-depth Clinical and Biological Exploration of DNA Damage Immune Response as a Biomarker for Oxaliplatin Use in Colorectal Cancer

Colorectal cancer is the fourth most common cancer and the second most common cause of cancer-related death in the United Kingdom (1). Colorectal cancer diagnostic classification relies on the World Health Organization (WHO) classification and the tumor—node–metastasis staging system. While histologic assessment provides valuable prognostic information, it cannot identify specific patient subgroups within tumor type, grade, or clinical stage that respond best to chemotherapy. Despite advances in treatment regimens, 5-year overall survival (OS) rates in the unresectable metastatic setting remain at 10% (2). In patients with stage III or histologically high-risk stage II tumors, recurrence is seen in 45% and 16% of patients, respectively, following surgery and adjuvant 5-fluorouracil (5-FU)-based chemotherapy (2). The addition of oxaliplatin to 5-FU–based regimens has led to a 20% risk reduction in OS following surgery for patients with stage III colorectal cancer (3–5). However chronic peripheral neuropathy occurs in approximately 50% of patients exposed to oxaliplatin (6), and there is no clinically validated test available to predict oxaliplatin response. Therefore, a significant proportion of patients may endure distressing side effects from this treatment with no clinical benefit (7). This highlights the need for the development of improved predictive tools to guide treatment decision-making and ultimately improve patient outcomes (8).

Numerous models suggest that conventional chemotherapy elicits high levels of DNA damage and DNA strand breaks in highly proliferative cancer cells that can either prime them for cell death, or tip already primed cells into apoptosis (9). The efficacy of chemotherapy in cancer cells is often compromised because of dysfunctional damage detection or cell death mechanisms, allowing cell survival (9). Certain chemotherapeutic agents target vulnerabilities inherent in tumors with defective DNA damage repair machinery, leading to neoplastic cell death. In colorectal cancer, the most common defective DNA damage repair mechanism occurs in tumors with microsatellite instability (MSI), characterized by defects in DNA mismatch repair (MMR). MSI tumors account for approximately 15% of stage II/III colorectal cancer and approximately 4% of stage IV patients, and are largely characterized by hypermutation, an increase in cancer-specific neoantigen production, high immune infiltration, and a favorable prognosis in earlier stages (10, 11). Interestingly, in the recent FOxTROT (Fluorouracil and Oxaliplatin With or Without Panitumumab In Treating Patients With High-Risk Colon Cancer That Can Be Removed by Surgery) neoadjuvant colon cancer chemotherapy clinical trial, this immune-rich MSI subgroup, defined by loss of MMR, specifically failed to gain a clear significant benefit from oxaliplatin-based neoadjuvant therapy (7). The DNA damage immune response (DDIR) signature, which comprises a 44-gene transcriptional signature based on loss of the Fanconi anemia/BRCA (FA/BRCA) DNA damage response pathway, was previously developed in breast cancer, where it demonstrated clinical utility for the identification of patients with a good response to anthracycline and/or cyclophosphamide-based neoadjuvant chemotherapy (12, 13). DDIR-positive tumors (exhibiting defective DNA damage repair) are characterized by an inflammatory tumor microenvironment (TME), upregulation of IFN signaling genes, and high lymphocytic infiltration. Additional studies in breast cancer indicated that DDIR-positive tumors have increased levels of CXCL10 and enhanced signaling through the cGAS/STING pathway (14).

Given these predictive findings, the Stratification in COloRecTal cancer (S:CORT) consortium (15) hypothesized that the DDIR signature would be predictive of oxaliplatin benefit in colorectal cancer, based on its ability to predict benefit from DNA-damaging therapy in breast cancer. In this study, we tested the ability of the DDIR signature to identify patients that may respond to oxaliplatin-based chemotherapy in both metastatic and neoadjuvant colorectal cancer settings, employing transcriptional profiling and bioinformatic analysis of subsets of samples from the FOCUS (Fluorouracil, Oxaliplatin, irinoteCan, Use and Sequencing; first-line metastatic, n = 391) and FOxTROT (first-line neoadjuvant, n = 97 randomized controlled trials. We ascertained whether DDIR positivity was associated with improved outcomes in patients with metastatic colorectal cancer treated with FOLFOX (5FUFA plus oxaliplatin) compared with 5FUFA alone (bolus and infusion 5-FU and folinic acid on the modified de Gramont schedule), and in patients with localized disease treated with FOLFOX in the neoadjuvant setting. We also performed a series of analyses to comprehensively characterize the underlying biology of DDIR subtypes in colorectal cancer compared with breast cancer.

As part of the UK Medical Research Council (MRC) S:CORT (15), tumor biospecimens with associated clinical trial data were identified for exploration of potential stratifiers for oxaliplatin treatment. The randomized MRC FOCUS trial was selected for exploration in the metastatic setting and the FOxTROT trial was selected for exploration of short-course FOLFOX in the neoadjuvant setting. The studies were performed in accordance with the Declaration of Helsinki. All subjects provided written informed consent for further research on their samples at the time of consent to the clinical trials. Both the original clinical trials (FOCUS Ref: 79877428; FOxTROT 07/SO703/57) and the studies reported here (S:CORT ref 15/EE/0241) were approved by the National Research Ethics Service in the United Kingdom.

FOCUS was a large UK-based randomized controlled trial comparing different strategies of sequential or combination therapies of 5FUFA with or without oxaliplatin or irinotecan as first- or second-line therapies in patients with newly diagnosed advanced colorectal cancer (16). A total of 2,135 patients were recruited between 2000 and 2003 and randomized between three strategies of first- or second-line combination therapy. Control strategy: first-line 5FUFA alone, followed by single-agent irinotecan; second strategy: first-line 5FUFA alone, followed by second-line combination chemotherapy; third strategy: combination chemotherapy in first-line treatment. Within the two research strategies, the combination regimen was an additional randomization: either 5FUFA plus oxaliplatin (FOLFOX), or 5FUFA plus irinotecan (FOLFIRI). For the DDIR analysis, samples from patients with colonic primaries from a biobank of archival diagnostic tissue were selected from consenting patients in the relevant arms where a randomized comparison could be made between first-line 5FUFA alone or in combination with oxaliplatin (85 mg/m² two-weekly; Supplementary Fig. S1A). A total of 385 samples were obtained from 371 primary resections, 8 primary biopsies, and 6 metastatic samples (3 liver, 2 nodal, and 1 lung). The primary outcome for FOCUS was OS, but data were also available for progression-free survival (PFS) and objective response rate (ORR).

FOxTROT was an international randomized trial (1,052 patients) which has reported its main finding (7). Patients were eligible if they had been diagnosed with locally advanced colon cancer without evidence of distant metastasis and with surgical resection of the primary tumor planned. Patients were randomized into one of three chemotherapy groups:

• Group A: Patients had 6-week presurgery chemotherapy with FOLFOX or oxaliplatin plus capecitabine (OxCap) and 18-week chemotherapy that commenced 4–8 weeks after surgical resection of the tumor.

• Group B: Patients had no presurgery chemotherapy but had 24-week chemotherapy (FOLFOX or OxCap) after their surgical resection.

• Group C: For patients who were RAS wild-type on baseline biopsy and randomized to neoadjuvant chemotherapy, the option of a secondary randomization between panitumumab or not, for the 6 weeks prior to surgery.

For patients randomized into Group A, FOxTROT provided an opportunity to measure DDIR in the tissue biopsy in a subset at baseline and determine whether DDIR was predictive of response to neoadjuvant FOLFOX therapy prior to resection surgery, excluding patients in Group C and those with complete response (Supplementary Fig. S1B).

All the archival formalin-fixed paraffin-embedded (FFPE) tumor tissue samples were tested at Almac's Diagnostic CLIA Laboratories. Samples were reviewed and tumor material identified on an adjacent hematoxylin and eosin–stained slide for microdissection. Total RNA was extracted from two sequential 5-μm sections using the Roche High Pure FFPE Extraction Kit (Roche Life Sciences) and amplified using the NuGen Ovation FFPE Amplification System v3 (NuGen San Carlos). The amplified product was hybridized to the Almac Diagnostics XCEL array (Almac), a cDNA microarray-based technology optimized for archival FFPE tissue, and analyzed using the Affymetrix Genechip 3000 7G scanner (Affymetrix) as described previously (12). Microarray data were quality checked (see Supplementary Methods) then preprocessed where raw CEL files underwent the robust multiarray average normalization for the Almac Diagnostic XCEL array with the affy package (v1.56.0; ref. 17). Gene expression profiles from a total of 391 samples from FOCUS and 97 samples from FOxTROT were made available.

For the biological analysis, a subset of gene expression profiles from n = 361 primary tumor resection samples from FOCUS were used (exclusions detailed in Supplementary Fig. S1A) and n = 97 pretreatment biopsy samples from FOxTROT (exclusions detailed in Supplementary Fig. S1B). Probes were annotated using annotation file “Xcel Annotations, CSV format, Release 36” available for download from Affymetrix (http://www.affymetrix.com/support/technical/byproduct.affx?product = xcel), and then collapsed to their corresponding genes using WGCNA package (version 1.68), based on the probe with highest average value for each gene (18). For comparative analysis between breast cancer and colorectal cancer, TRANSBIG breast cancer cohort (19) containing gene expression profiles for 198 fresh-frozen samples from patients with node-negative T1-T2 (≤5 cm) breast performed on Affymetrix Human Genome U133A array was downloaded from Gene Omnibus Expression (GEO; www.ncbi.nlm.nih.gov/geo/; accession number “GSE7390”).

A total of 484 clinical samples (391 from FOCUS and 97 from FOxTROT) had DDIR signature scores calculated and predefined cut-points applied. The predefined threshold of 0.1094 was optimized in an independent technical study of 260 colorectal cancer samples whereby the optimal threshold was detected at the score where the sensitivity and specificity meant a joint maximum to accurately detect the DDIR-positive subgroup as defined in hierarchical clustering (personal communication Almac Diagnostics). The threshold was then applied independently to the validation cohorts, dichotomizing patients as DDIR positive (>0.1094) or DDIR negative (≤0.1094).

TRANSBIG breast cancer cohort (19) used in the original study had information available on predetermined DDIR threshold of 0.37 along with DDIR continuous score (12), that was used on our analysis.

To obtain consensus molecular subtype (CMS) calls, genes with multiple probesets were collapsed by mean and the CMSclassifier package was used (20). Classification by random forest with the default posterior probability of 0.5 showed a higher frequency of unclassified samples compared to the original publication (20). To derive calls with comparable frequencies, single sample predictor calls were computed after row-centering the expression data. Final CMS calls were generated when there was a match between both methods without applying any cut-off. To obtain colorectal cancer intrinsic subtype (CRIS) calls, probesets with the highest average levels for each gene were selected and the CRISclassifier package was used (21). Samples with a Benjamini-Hochberg–corrected FDR > 0.2 were left unclassified as originally reported (21).

Mutation data were generated by DNA target capture (SureSelect, Agilent) spanning all coding exons of 80 colorectal cancer driver genes (listed in Supplementary Methods) followed by next-generation sequencing (Illumina). Variant calling was performed with Caveman for point mutations and Pindel for indel mutations. Driver mutations in KRAS, NRAS, PIK3CA, and TP53 were considered for binary classification (e.g., depending on whether genes are dominant/recessive, mutations reported as recurrent or an internal curated list) based on frequency and relevance. BRAF was classified as mutated only with a V600E mutation. Tumors showing more than two mutations in n = 123 MSI markers within the panel were classified as MSI, otherwise as microsatellite stable (MSS). The FOxTROT cohort showed a high failure rate (55/97 missing data, 57%) due to lack of enough tissue in small biopsies after RNA profiling. Therefore, MSI classification from additional FOxTROT tumors were derived with an RNA signature (22). Two borderline tumors were not classified.

Gene set enrichment analysis (GSEA) was performed in the three cohorts to investigate biological pathways associated with DDIR (23, 24), using Hallmarks gene set collection [h.all.v6.2.symbols.gmt (Hallmarks)] from Molecular Signature Database (MSigDB; refs. 25, 26). GSEA version 19.0.26 was accessed from the GenePattern cloud server web interface: https://cloud.genepattern.org. All default parameters were utilized, with the exception of “collapse dataset” which was set to “FALSE,” as the probes were collapsed to their genes a priori, and the random seed was stated to be “40218336.” Normal enrichment score (NES) and FDR values were noted for each gene set within the two phenotypic (DDIR) groups, where FDR q value below 25% was justified to be a significant gene set.

The MCPcounter (version MCPcounter_1.1.0) R package was downloaded from GitHub (https://github.com/ebecht/MCPcounter), and was used to generate microenvironment cell population (MCP) estimation scores for ten stromal and immune cell infiltrates from the transcriptomic data of the three cohorts (27). Estimates were compared between DDIR positive and DDIR negative to determine their stromal/immune content, and the differences in cellular composition between the cancer types.

Partek Genomics Suite version 6.6 was utilized to perform ANOVA analysis to identify differentially expressed genes with FDR of < 0.05, and fold change (FC) adjusted to 1.5 for FOCUS and FOxTROT cohorts; for TRANSBIG due to the large number of differentially expressed genes, FC value was increased to 2.5. Differentially expressed genes were assessed using Ingenuity Pathway Analysis (IPA - 49932394) to examine any significant biological pathways associated with DDIR subtypes. All parameters were set to default.

Statistical analyses were conducted according to prespecified statistical analysis plans that were agreed prior to inspection of any DDIR-stratified outcome data. All clinical-related analyses for ORR, PFS, and OS were performed using Stat version 15.0 (Stata Corporation) or R (version 3.4.1). Further detailed statistical analysis on FOCUS and FOxTROT cohort is available in Supplementary Methods.

All statistical analyses undertaken for further biological exploration, including Pearson correlation coefficient, Fisher exact test, Student t test, Wilcoxon rank-sum test, Kruskal–Wallis rank-sum test, and one-way ANOVA followed by Tukey honest significance difference test were performed to generate P values for statistical significance using R stats package in R (version 3.4.0) and RStudio (version 1.1383). In addition to base R packages, ggplot2 R package (version 3.2.1) with other supporting packages, including cowplot (version 0.9.4), ggpubr (version 0.2.3), and grid (version 3.4.0) were used for graphical visualization.

FOCUS and FOxTROT gene expression dataset and clinicopathologic information are provided from S:CORT (https://www.s-cort.org/), with transcriptional data, mutation data (for KRAS, NRAS, PIK3CA, BRAF, and TP53), and MSI call available on GEO under reference GSE156915. All scripts required to reproduce figures in this article are available from corresponding author on request or from www.dunne-lab.com.

A total of n = 391 patients were available for DDIR analysis from the FOCUS trial. Following exclusion of rectal cancer cases and prioritization of resected tissue to ensure there was sufficient tumor tissue for molecular analyses, n = 310 from the 5FUFA alone group and n = 81 in the FOLFOX group were used for outcome analyses (Supplementary Table S1). Assessment of baseline characteristics of patients excluded from the DDIR analysis compared with those included in the DDIR analysis revealed that there were no other obvious selection biases between the groups (Supplementary Table S1; Supplementary Fig. S1). A total of 76/391 patients were classified as DDIR positive (Supplementary Fig. S2), generating a prevalence of 19% [95% confidence interval (CI), 16–24] overall, with a reasonable balance between the randomized groups of 63 (20%) versus 13 (16%) in the 5FUFA and FOLFOX groups, respectively, (χ² P value for difference = 0.39; Supplementary Table S1).

The overall prevalence of DDIR was lower than anticipated when compared with data from other cohorts of patients with colorectal cancer (28) and other disease indications (12, 13, 29), but was similar to the technical study of 260 metastatic colorectal cancer used to set the threshold for DDIR positivity (Personal communication Almacgroup).

During the course of follow-up between May 16, 2000 and October 18, 2006, there were a total of 383 PFS events (357 during the first 15 months) and 342 OS events. During the first 12 weeks of first-line chemotherapy, there were 157 (40%) complete or partial responders and 234 (60%) stable or progressive disease nonresponders. A comparison between randomized groups, without stratification for DDIR, confirmed the anticipated treatment effect of oxaliplatin; PFS adjusted HR = 0.63 (95% CI, 0.48–0.81), P = 0.001 and ORR adjusted OR = 4.07 (95% CI, 2.37–7.01), P < 0.001 (Supplementary Fig. S3).

In the FOCUS control arm, we identified no prognostic effect of DDIR status for patients with metastatic colon cancer treated with first line 5FUFA alone, either on OS [unadjusted HR = 0.95 (95% CI, 0.71–1.28), P = 0.73, test of proportional hazards: χ² = 1.42 on 1 d.f. (degree of freedom), P = 0.20; Supplementary Fig. S2B], or on PFS [adjusted HR = 1.11 (95% CI, 0.79–1.54), P = 0.55]. This result remained nonsignificant when adjusted for clinical variables, CMS status, and other molecular variables.

Using fully adjusted models, we next explored the predictive effects of DDIR for all outcomes, with PFS at 15 months as the primary outcome (Fig. 1A). Contrary to the expectation that DDIR-positive patients would derive the most benefit from oxaliplatin, DDIR-negative patients appeared to respond more frequently to FOLFOX [ratio of ORs for ORR = 0.15 (95% CI, 0.04–0.65], test for P_interaction = 0.011; Table 1; Fig. 1B). Although this inverted direction of effect was the same for the survival outcomes, the tests for interaction were nonsignificant (Table 1).

Following these analyses in the metastatic setting, we next assessed the clinical utility of the DDIR in the colorectal cancer neoadjuvant setting. A total of 97 patients who received neoadjuvant FOLFOX were selected from Group A of the FOxTROT dataset. Patients were excluded if they withdrew from the trial if they did not receive neoadjuvant chemotherapy or if they received OxCap prior to surgery. In addition, no patients with complete pathologic response were forwarded to S:CORT for analysis. These selections led to a somewhat biased subset compared with the main study with less responders, less MSI, and more KRAS wild-type tumors (Supplementary Table S2). Of these 97 patients, 4 had no associated response data, leaving a total of 93 patients who were included in the final analysis. There were a total of 40 nonresponders, 29 mild-responders, 17 moderate responders, and 7 marked responders. The DDIR threshold was set at the same value defined in the FOCUS cohort, resulting in 57% DDIR-positive patients, which was considerably higher than the 19% seen in the metastatic FOCUS dataset (Supplementary Fig. S2C). Using ordinal regression across the 4 response groups, there were marginally better responses in the DDIR-negative group (Fig. 1C), but this was not statistically significant using unadjusted ordinal regression OR = 0.62 (95% CI, 0.29–1.33), P = 0.218 (Table 1). After adjustment for age, sex, pT stage, pN stage, primary tumor location, MSI, and RAS status, the coefficient reduced slightly to 0.55 (95% CI, 0.21–1.39), P = 0.205. Employing DDIR as a continuous variable, the unadjusted OR for response was 0.19 (95% CI, 0.02–1.79), P = 0.148. When adjusted for age, sex, T-stage, N-stage, left/right, MSI, and RAS status, the OR reduced to 0.11 (95% CI, 0.01–1.66), P = 0.110 (Supplementary Table S2).

Given these counterintuitive findings, we next set out to investigate whether there was a biological explanation for this potentially inverted and inconsistent effect between previous breast cohorts and our colorectal cancer trial cohorts.

Investigation into the biological relevance of DDIR signature led to the comparison against colorectal cancer CMS which is largely based on histologic (stroma and immune) features (20). In the FOCUS cohort, immune-rich CMS1 tumors are significantly associated with increased DDIR scores when compared with all other CMS subtypes (Fig. 2A; Kruskal–Wallis, P < 0.0001). Despite CMS1 tumors having a significantly higher proportion of DDIR-positive tumors compared with the other subtypes (Supplementary Fig. S6A; Fisher exact test, P = 0.0002), given the low prevalence of DDIR positivity across the whole cohort, 68% of CMS1 subtypes are below the DDIR threshold (Fig. 2A). Of note, there are proportionally more CMS4 tumors within DDIR-negative classification in the FOCUS cohort (Supplementary Fig. S6A). In pretreatment biopsies from the smaller FOxTROT cohort, CMS1 tumors show a nonsignificant trend toward DDIR positivity (Fig. 2B; Kruskal–Wallis, P = 0.4695, and Supplementary Fig. S6B; Fisher exact test, P = 0.4879). In addition, we also examined DDIR CRISs that represents colorectal cancer tumor-intrinsic (epithelial) biology (21). Contrary to CMS, no significant association between the CRIS subtypes and DDIR-positive or DDIR-negative tumors in both the FOCUS and FOxTROT cohorts was found (Supplementary Fig. S6C–S6F). These findings suggest that, in colorectal cancer, DDIR positivity is primarily associated with (and potentially influenced by) CMS-related TME factors, such as differences in stromal/immune infiltrates, rather than epithelial-derived intrinsic factors.

Originally, DDIR signature was developed on the basis of defective DNA damage response and repair machinery of homologous recombination (HR) and FA in breast cancer (12). However, there is limited evidence on their role in colorectal cancer tumorigenesis (30). Thus, we explored the relationship between HR/FA and DDIR in colorectal cancer cohorts and compared against the TRANSBIG breast cancer cohort which was used in the development of the DDIR signature. Our investigation suggested that within colorectal cancer, these pathways do not show any association with DDIR, contrary to that in breast cancer (see Supplementary Results; Supplementary Fig. S4). MSI, a result of defective DNA MMR mechanisms, defines a proportion of patients with colorectal cancer associated with high tumor mutational burden, leading to development of immune-responsive TME. Despite the limited number of MSI tumors in the metastatic FOCUS colorectal cancer cohort (n = 13), we observe that MSI tumors contain a significantly higher proportion of DDIR positives (Fig. 2C; Fisher exact test, P = 0.0211). However, DDIR positivity is not a biomarker of MSI status, as only 46% of MSI tumors are DDIR positive (6 out of 13) while the majority of DDIR-positive tumors overall are MSS (Fig. 2D; MSI/DDIR+ n = 6, MSS/DDIR+ n = 59). In the FOxTROT cohort, MSI trends observed are in line with the larger FOCUS cohort (Fig. 2E; Fisher exact test, P = 0.2522, and Fig. 2F; Student t test, P = 0.0737), but this result cannot be used to confirm the FOCUS findings due to small (n = 3) MSI sample size (Fig. 2F). Furthermore, while MSI tumors collectively contain higher mutational burden than MSS as expected, mutational burden is not associated with DDIR positivity in either of the colorectal cancer cohorts (Supplementary Fig. S6G; Student t test, P = 0.1279 and Supplementary Fig. S6H; Student t test, P = 0.4534).

To further characterize the biological functions and pathways associated with DDIR, we performed GSEA, using the “Hallmark” collection, to compare DDIR-positive and DDIR-negative tumors in FOCUS and FOxTROT colorectal cancer cohorts, compared with the same analyses in the TRANSBIG breast cancer cohort. GSEA between DDIR-positive and DDIR-negative tumors generated different numbers of significant Hallmarks gene sets in each cohorts (Supplementary Fig. S7). However, in general, between the three cohorts five common significantly enriched gene sets in DDIR-positive colorectal cancer and breast cancer tumors were identified, namely, allograft rejection, IL6/JAK/STAT3 signaling, inflammatory response, IFNα response, and IFNγ response (Fig. 3A; FDR q-value < 0.25), suggesting that a common immune and/or inflammatory-like signaling defines DDIR positivity, regardless of the cancer type. Interestingly, we also observe eight unique gene sets that are only associated with DDIR in breast cancer and not in colorectal cancer (Fig. 3A).

Previous studies of DDIR signaling in breast cancer have highlighted increased levels of the IFNγ-induced chemokine CXCL10 gene/protein expression in DDIR-positive tumor cells, leading to lymphocytic trafficking into the tumor (14). Here, we showed that CXCL10 expression has a strong positive (>6) correlation with DDIR scores in both breast cancer and colorectal cancer cohorts (Fig. 3B,–D). In addition, it was previously demonstrated that DDIR positivity in breast cancer was specifically associated with activation of cGAS/STING/TBK1 innate immune-response axis (14). This, however, was not found to be the case in colorectal cancer (see Supplementary Results).

We tested the association between immune/stromal composition, based on gene expression profiles using MCP analysis, where we identified consistent correlations between DDIR scores and T cell, B cell, and monocytic immune lineages, confirming an increase in lymphocytic infiltration in DDIR-positive breast cancer (Fig. 4A; Pearson r; T cells = 0.7167, B lineage = 0.5075, monocytic lineage = 0.7042). While we also observe correlative trends in both colorectal cancer cohorts (Fig. 4B; Pearson r; T cells = 0.3509, B lineage = 0.2774, monocytic lineage = 0.2358 and Fig. 4C; Pearson r; T cells = 0.4038, monocytic lineage = 0.5152, B lineage = 0.3666), these correlations were not as strong as those observed in breast cancer. Moreover, cytotoxic lymphocyte scores also demonstrate a positive correlation with DDIR using both a positive versus negative categorical (Fig. 4D; Student t test, P < 0.0001) or DDIR continuous score (Fig. 4D; Pearson r = 0.6106) in the TRANSBIG breast cancer cohort. Similar, albeit weaker, correlations were observed in both FOCUS (Fig. 4E; Student t test, P < 0.0001; Pearson r = 0.436) and FOxTROT (Fig. 4F; Student t test, P = 0.0004; Pearson r = 0.5251) colorectal cancer cohorts using the MCP-derived cytotoxic lymphocyte scores. Incorporation of CMS in the colorectal cancer analyses demonstrated the association between CMS1, lymphocytic infiltration, and increased DDIR score. Levels of cytotoxic CD8⁺ T-lymphocytic infiltration were further assessed in situ in the FOCUS cohort by IHC (Fig. 4G), where a significant association between CD8 IHC scores and DDIR score was observed, in line with MCP assessments in these tumors (Fig. 4H; Student t test, P < 0.0001; Pearson r = 0.4388). Conversely, fibroblast levels and CMS4 subtypes were negatively correlated with DDIR score in the FOCUS cohort (Supplementary Fig. S8A and S8B; t test, P = 0.0109; Pearson r = −0.1597), while no association was noted in the FOxTROT cohort (Supplementary Figs. S8C and S4D; t test, P = 0.9984; Pearson r = 0.0291).

Next, we set out to identify overlapping individual differentially expressed genes between DDIR subtypes in both breast cancer and colorectal cancer. Differential gene expression analysis comparing DDIR-positive and DDIR-negative tumors identified 66 and 60 differentially expressed genes in FOCUS and FOxTROT cohorts, respectively (FDR < 0.05, FC = 1.5; Fig. 5A). We observed 975 differential genes between DDIR-positive and DDIR-negative tumors in the breast cancer cohort compared with colorectal cancer; thus, to limit these analyses to a similar sized gene list for the TRANSBIG cohort, we increased the FC for analysis, identifying 110 differentially expressed genes (FDR < 0.05, FC = 2.5; Fig. 5A). Comparison of gene lists from the three cohorts identified nine genes that are consistently upregulated in DDIR-positive tumors in both cancer types (Fig. 5A). This list contained members of chemokines family, including two genes (CXCL10 and IDO1) that are part of the 44-gene DDIR signature. Using these nine differentially expressed genes common in all three cohorts, pathway analysis was performed, which revealed 18 potential upstream regulators of conserved biology contributing to DDIR positivity across colorectal cancer and breast cancer, including key regulators of inflammatory and IFN-related signaling, such as IFNα, IFNγ, STAT1, and the NFkB complex (Fig. 5B; Supplementary Fig. S9A).

Using these nine consensus DDIR-related genes to generate an unweighted cumulative score, we observed a strong positive correlation between this new overlapping ranked sum score and the original DDIR score (Fig. 5C; Pearson r = 0.6291, P < 0.0001). In line with this overlap, we also observed similar correlative trends for both CMS and MSI (Supplementary Fig. S9B and S9C), with the 9-gene score as observed with the original DDIR score (Fig. 2). Finally, a Cox regression model (for PFS) and a logistic regression model (for response) were fitted with main effects for oxaliplatin and for each of three quartiles of Almac DDIR or 9-gene score relative to Q1 (reference), and interactions between oxaliplatin and the three quartiles (Fig. 5D). As with the response and outcomes analyses using the original DDIR score, this overlapping 9-gene score fails to predict a benefit for the addition of oxaliplatin to 5FUFA in the FOCUS trial. Importantly, however, this new refined colorectal cancer DDIR signature removes the trend for increased response to oxaliplatin observed in the DDIR-negative group in the original DDIR.

The original characterization of the DDIR signature demonstrated its predictive value as a biomarker for platinum-based chemotherapy treatment in breast cancer, and subsequently esophageal adenocarcinoma (12, 29). In the initial breast cancer study, the biology underpinning DDIR was based on dysfunctional DNA damage response and repair machinery regulated via the HR and FA/BRCA pathways, which is targeted by some chemotherapies as a mode of action (31). The multidisciplinary S:CORT consortium (15) was established to identify and test new molecular stratification methods to predict colorectal cancer response to treatments, through the discovery of new and/or validation of existing molecular biomarker-based assays. In this study, we tested the clinical utility of the 44-gene DDIR signature from archival FFPE tumor tissue profiled at Almac's Diagnostic CLIA Laboratories as previously described, to predict response to the addition of oxaliplatin to 5-FU–based chemotherapy in both metastatic colorectal cancer (FOCUS cohort) and neoadjuvant colorectal cancer (FOxTROT) clinical trial settings. Accompanying this clinical assessment, we utilized the molecular and histologic data generated to further interrogate the biological signaling associated with colorectal cancer–specific DDIR positivity in contrast to breast cancer.

DDIR positivity was observed in 19% of primary tumors from stage IV FOCUS cohort and 57% of primary tumor biopsy material from stage II/III FOxTROT cohort. A previous study of DDIR positivity in colorectal cancer reported a 35% incidence in a predominantly (94%) nonmetastatic population (28). This was comparable with findings in breast cancer (34%; ref. 12) and esophageal adenocarcinoma (24%; ref. 29). Differing DDIR rates in our study could be credited to the cancer stage or other (molecular) criteria used for patient selection in the original trials. Patients with localized disease, as in the neoadjuvant FOxTROT study, have a higher proportion of tumors with immune infiltration (32), a factor associated with DDIR positivity in breast cancer and esophageal adenocarcinoma, and also with MSI and CMS1 tumors in colorectal cancer. Similarly, the reduction in DDIR positivity to approximately 20% in metastatic disease is consistent with a lower relative proportion of patients with MSI in metastatic disease, which falls from approximately 20% in localized colon cancer to approximately 4% in metastatic colorectal cancer, as in the FOCUS cohort.

MSI is the most notable feature in colorectal cancer displaying defective DNA damage response and repair via MMR system (30). MSI and CMS1 are closely linked together with high tumor mutation burden, overproduction of tumor-specific neoantigens, increased immune infiltration, and show favorable clinical outcome in early-stage disease (20). Given their high levels of immune infiltration and mutation burden, these tumors have responded well to checkpoint blockade immune-oncology (IO) treatments (33). There is a strong association of DDIR status with CMS1, MSI status (ref. 28; Fig. 2) in FOCUS cohort, and a similar trend is observed in FOxTROT cohort, given its small sample size (Fig. 2), reflecting the observed clinical utility of immunotherapeutic interventions in this molecular subtype (34, 35). However, our findings do not validate the correlation between DDIR and mutational burden in the FOCUS cohort observed in the colorectal cancer threshold development abstract (28), likely due to the difference in disease stage (FOCUS as metastatic colorectal cancer) and mutational panel sequencing methods used with S:CORT.

Contrary to our primary hypothesis, it was noted that response to the addition of oxaliplatin to 5FUFA was more likely to benefit DDIR-negative patients in both FOCUS and FOxTROT cohorts rather than DDIR-positive patients. While this was only statistically significant in terms of response in the metastatic FOCUS trial setting (ratio of ORs for ORR = 0.15, test for P_interaction = 0.011), the trend was consistent across all endpoints in both cohorts examined. However, the refinement of DDIR gene signature to only 9-gene signature through our analysis showed no additional benefit from oxaliplatin for either DDIR-positive or DDIR-negative patients (Fig. 5). The original and subsequent DDIR study in breast cancer with the South Western Oncology Group (13) demonstrated improved response to anthracycline and/or cyclophosphamide-based neoadjuvant and adjuvant chemotherapy in DDIR-positive patients. Similarly, in esophageal adenocarcinoma, DDIR positivity was predictive of improved response to cisplatin-containing chemotherapy (29). Oxaliplatin is known to differ in its mechanism of cytotoxicity compared with cisplatin and may have more complex mode of action in colorectal cancer (36).

Although we show no additional interaction between DDIR positivity and oxaliplatin treatment, biologically, our study highlights promising immunotherapeutic opportunities among DDIR-positive colorectal cancer patients, beyond the use of general immune infiltration or MSI status. DDIR positivity may have value in identifying additional subsets of patients with MSS colorectal cancer who exhibit high tumor mutational burden and/or high TME activity, who have the potential to respond to immune checkpoint blockade such as PD-L1 inhibition (35, 37, 38). The search for biomarkers to distinguish immune “cold” tumors (that display limited response to IO) from immune “hot” tumors (that respond to IO) has gained traction in recent years. Our findings indicate that in colorectal cancer, although DDIR positivity is associated with increased levels of both innate and cytotoxic infiltration, likely to be driven by IFN-related signaling, the immune system is in an “exhausted” state and unable to efficiently clear these tumors, due to the concurrent expression of checkpoints such as IDO1 and PD-L1 (CD274). These findings may also provide an explanation for the noncorrelation of DDIR with oxaliplatin-based chemotherapy response, as induction of immune tolerance is a common response pattern to inflammation in the gut and tumor-associated inflammation (as seen in DDIR-positive tumors) that leads to a predominantly immune-suppressive milieu, which is further reinforced by additional chemotherapy-related inflammatory signaling. Indeed, MSI tumors are largely nonresponsive to chemotherapy, as has been demonstrated recently in the neoadjuvant FOxTROT trial (7), as are immune-rich/MSI tumors when assessed in other nontrial adjuvant cohorts (39). Very recent trial data reported 100% response rate in early-stage MSI colon cancer, including 60% pathologic complete response, to neoadjuvant IO treatment (combined CTLA-4 and PD1 blockade; ref. 40). Results from that study also indicate that only 27% of MSS tumors displayed any response. Importantly, however, these data confirmed the predictive nature of CD8⁺ T-cell infiltration for IO response in MSS tumors, a phenotype associated with the biology underpinning DDIR positivity in MSS colorectal cancer presented in this study, supporting clinical testing of DDIR as a predictive assay to select patients with MSS in this setting.

The approach adopted in our study highlights the clinical utility and high success rates associated with molecular profiling of FFPE material (Supplementary Table S1), even in tissue-limited pretreatment diagnostic biopsy material used to guide treatment decisions in the neoadjuvant setting, as in FOxTROT. The TRANSBIG data used in the original DDIR study pose a potential limitation on our breast cancer analysis due to the platform employed in the original analysis (Affymetrix Human Genome U133A Array) not being identical to the one used for the transcriptional profiling in the colorectal cancer cohorts, which was the Almac XCEL array. To ensure cross-platform comparison for DDIR was not confounding our study, Almac have classified DDIR according to their diagnostic assay on all cohorts tested.

In summary, our study shows that, in contrast to breast cancer and esophageal adenocarcinoma, DDIR does not predict improved response or survival to oxaliplatin treatment. We have identified the underlying biology of the signaling associated with DDIR in colorectal cancer that could affect the outcome. While we identify significant overlap in DDIR signaling across breast cancer and colorectal cancer, particularly immune-related TME signaling, we also highlight that signaling associated with both HR/BRCA and STING pathways is not significantly associated with DDIR in colorectal cancer. Overall, our data support further testing of the utility of the DDIR signature in selecting patients who may respond to IO-based therapy.

S.B. Malla reports grants from MRC CRUK during the conduct of the study. K.L. Redmond reports grants from MRC/CRUK during the conduct of the study. S. Walker reports personal fees from Almac Diagnostics (employee) during the conduct of the study; in addition, S. Walker has a patent for gene signature for immune therapies in cancer issued. G.E. Logan reports employment at Almac Diagnostic Services. R. Amirkhah reports grants from MRC & CRUK during the conduct of the study. D.G. Morton reports grants from MRC and CRUK during the conduct of the study. P. Quirke reports grants from Medical Research Council UK during the conduct of the study. P. Quirke also reports personal fees from Nordlai Adlyte; grants, personal fees, and nonfinancial support from Roche; and grants from Halio, Amgen, GeneFirst, ONI, Yorkshire Cancer Research, National Institute of Health Research UK, and Cancer Research UK outside the submitted work. N.P. West reports grants from Yorkshire Cancer Research during the conduct of the study; in addition, Dr. West reports personal fees from Eisai Ltd outside the submitted work. R.D. Kennedy reports personal fees from Almac Diagnostic Services (employee of Almac Diagnostic Services and responsible for developing the DDIR gene expression signature) during the conduct of the study. R.D. Kennedy also reports personal fees from Almac Diagnostic Services (employed as the medical director) outside the submitted work, and has a patent for WO2017013436A1 issued (gene signature for immune therapies in cancer). V.H. Koelzer reports grants from Swiss National Fond (P2SKP3_168322/1 and P2SKP3_168322/2) during the conduct of the study; in addition, V.H. Koelzer reports nonfinancial support from Indica Labs (has served as an invited speaker on behalf of Indica labs) outside the submitted work. R.S. Kaplan reports grants from Medical Research Council during the conduct of the study. M. Lawler reports grants and personal fees from Pfizer (honoraria and unrestricted grant unrelated to the submitted work), as well as personal fees from EMD Serono (honorarium unrelated to the submitted work), Roche (honorarium unrelated to the submitted work), and BMS (honorarium unrelated to the submitted work) outside the submitted work. T.S. Maughan reports grants from Medical Research Council, Cancer Research UK, and nonfinancial support from Almac diagnostics (performed transcriptomic analysis) during the conduct of the study. T.S. Maughan also reports grants and personal fees from AstraZeneca (advisory board), personal fees from Pierre Fabre (IDMC services), and grants from Merck Serono outside the submitted work. P.D. Dunne reports grants from MRC and CRUK during the conduct of the study. No disclosures were reported by the other authors.

S.B. Malla: Formal analysis, investigation, writing-review and editing. D.J. Fisher: Formal analysis, visualization, writing-review and editing, detailed clinical statistics. E. Domingo: Data curation, formal analysis, supervision, investigation, methodology, writing-original draft, writing-review and editing, led the multi-omic analyses of the clinical samples. A. Blake: Data curation, software, methodology, writing-review and editing, curated and quality assured all the clinical and biological data. S. Hassanieh: Investigation, visualization, writing-review and editing. K.L. Redmond: Formal analysis, gene expression analysis. S.D. Richman: Resources, data curation, investigation, methodology, writing-review and editing. M. Youdell: Conceptualization, resources, project administration, project manager for S:CORT consortium. S.M. Walker: Resources, formal analysis, validation, writing-review and editing, led Almac scoring of the DDIR signature in samples. G.E. Logan: Resources, data curation, project administration. A. Chatzipli: Data curation, writing-review and editing, calling mutations in genotyping. R. Amirkhah: Formal analysis, writing-review and editing. M.P. Humphries: Formal analysis, writing-review and editing. S.G. Craig: Investigation, visualization, writing-review and editing, immunohistochemistry. U. McDermott: Conceptualization, supervision, designed Sanger panel. M.T. Seymour: Resources, validation, writing-review and editing, Chief Investigator of FOCUS trial, approval for use of samples and data. D.G. Morton: Resources, validation, writing-review and editing, Chief Investigator of FOxTROT trial, approval for use of samples and data. P. Quirke: Resources, supervision, methodology, writing-review and editing. N.P. West: Resources, formal analysis, validation, writing-review and editing, pathology lead for FOxTROT trial. M. Salto-Tellez: Resources, supervision, investigation, visualization, writing-review and editing, oversaw immunohistochemistry. R.D. Kennedy: Conceptualization, supervision, writing-review and editing, original discovery of DDIR signature. P.G. Johnston: Conceptualization, resources. I. Tomlinson: Conceptualization, resources, supervision, writing-review and editing. V.H. Koelzer: Formal analysis, supervision, investigation. L. Campo: Investigation, methodology. R.S. Kaplan: Supervision, funding acquisition, writing-review and editing. D.B. Longley: Investigation, writing-review and editing. M. Lawler: Conceptualization, resources, supervision, writing-review and editing. T.S. Maughan: Conceptualization, resources, supervision, funding acquisition, writing-review and editing. L.C. Brown: Conceptualization, formal analysis, writing-review and editing. P.D. Dunne: Conceptualization, resources, formal analysis, supervision, writing-original draft, writing-review and editing.

We are grateful to all the patients and their families who participated in the FOCUS and FOxTROT clinical trials and gave consent to further research on their samples. We are also grateful to the Trial Management Groups and Trial Steering Committees for FOCUS and FOxTROT trials who allowed this work to proceed. This work was originally led by Paddy Johnston from Queen's University Belfast. Sadly, soon after the project commenced Paddy passed away and we would like to dedicate this work to him.

The stratification in colorectal cancer consortium (S:CORT) is funded by a UK Medical Research Council (MRC) Stratified Medicine Consortium programme grant (grant ref MR/M016587/1) and co-funded by Cancer Research-UK. L.C. Brown, D.J. Fisher, and R.S. Kaplan are partially funded by an MRC Core funding grant for the MRC Clinical Trials Unit at UCL (grant code 12023/20). This work was supported by a Cancer Research UK programme grant (to P.D. Dunne, D.B. Longley, P.G. Johnston; C212/A13721). Sample collection for FOxTROT was funded by Yorkshire Cancer Research.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.