The Interleukin 22 Pathway Interacts with Mutant KRAS to Promote Poor Prognosis in Colon Cancer

Colorectal cancer is a complex disease driven by the interplay between tumor mutations, environmental factors, and aberrant immunity (1, 2). Chronic intestinal inflammation (e.g., inflammatory bowel disease) is a well-known risk factor for colorectal cancer, but colitis-associated cancers account for only a small fraction of total tumor incidence (3). However, colorectal cancers that are not associated with colitis also elicit inflammatory responses, which are thought to be key regulators of tumor progression and therapeutic resistance (4). Indeed, inflammatory cytokines control several key features of malignancy, including cancer cell proliferation, survival, and dissemination (5).

Evidence from preclinical murine models of both sporadic and colitis-associated colorectal cancer has implicated IL23 and its downstream effector IL22 in the initiation and progression of colon tumorigenesis (6–9). Furthermore, IL22 has been associated with human gastric cancer progression (10) and has been described to promote colorectal cancer stemness (11). IL23-responsive CD4⁺ T cells and innate lymphoid cells secrete IL22 in the tumor microenvironment (7, 8), where it signals through a heterodimeric receptor comprised of IL10 receptor beta (IL10RB) and IL22 receptor alpha 1 (IL22RA1; ref. 12). In the intestine, IL22RA1 expression is restricted to the epithelium (13) and activates diverse signal transduction pathways in response to IL22, including JAK/STAT3, MAPK, PI3K, and NF-κB, which collectively induce expression of genes involved in cell-cycle progression and inhibition of apoptosis (14–16).

Oncogenic Ras isoforms modulate inflammatory pathways in cancer by directly influencing inflammatory cytokine expression and signal transduction (5). IL23, IL17, and IL22 are elevated in human colorectal cancer, and their expression is modulated by the presence of mutant Ras (17). Activating mutations in KRAS, a major Ras isoform, occur in 40% to 45% of colorectal cancers and drive constitutive activation of the Ras-Raf-MEK-ERK pathway (18). Although KRAS mutations are strongly associated with resistance to EGFR-targeted therapy, they do not correlate with resistance to chemotherapy (19–21). However, given the cytokine-modulating capacity of the Ras family, KRAS mutations may be clinically significant in tumors with active cytokine signaling. Oncogenic KRAS could influence virtually all signal transduction pathways downstream of IL22, but a specific interaction between IL22 signaling and KRAS mutation has not been described.

In light of the strong evidence pointing toward a protumorigenic role for IL22 in murine models of colorectal cancer, we sought to determine the clinical importance of this cytokine in human disease and to identify subtypes of colorectal cancer in which IL22 signaling may be particularly influential. Our results identify a previously unrecognized interaction between IL22 signaling and oncogenic KRAS that contributes to tumor cell proliferation and poor patient prognosis.

Four colorectal cancer transcriptomic datasets were analyzed in this study (Supplementary Fig. S1A). The French National Cartes d'Identité des Tumeurs (CIT) cohort (GSE39582; ref. 22) contains 469 patients with stage II/III colorectal cancer who underwent surgery between 1987 and 2007 at 7 different centers. Patients who received preoperative chemotherapy/radiotherapy were excluded from the study. RNA from fresh-frozen primary tumors was assessed using the HG-U133A Affymetrix platform, which contains one probe that detects IL22RA1 (220056_at).

The Pan-European Trials in Alimentary Tract Cancers (PETACC3) cohort (NCT00026273) was used as a verification set (ArrayExpress:E-MTAB-990). Data from stage II (n = 108) and stage III (n = 644) colorectal cancer were analyzed (23, 24). Gene expression data were obtained using the ALMAC Colorectal Cancer DSA platform, which is a customized Affymetrix chip that includes 61,528 probe sets mapping to 15,920 unique Entrez Gene IDs. The PETACC3 dataset contains three different probes for IL22RA1. The probe used in our analysis was that which displayed the greatest variation in the dataset (ADXCRAD_BX089163_s_at). Of note, the second probe was not well mapped, and the third probe had a very narrow dynamic range.

Finally, a merged dataset comprising patients with stage II/III colorectal cancer of GSE39582 (22), PETACC3 (23, 24), The Cancer Genome Atlas (TCGA; ref. 25), and ALMAC (26), representing 1,820 patients, was analyzed (referred to in tables as the “Combined” dataset). The ALMAC dataset was obtained from ArrayExpress (www.ebi.ac.uk/arrayexpress) on the A-AFFY-101 platform (customized Affymetrix chip) and is a merge of E-MTAB-863 and E-MTAB-864 (26). Clinical information on overall survival (OS) was available for 1,734 patients and on relapse-free survival (RFS) for 1,499 patients. Gene expression profiles of the patients in the combined dataset were merged at the gene level and then normalized and corrected for batch effects using the Combat R package. For the individual analyses of the GSE39582 and PETACC3 datasets, gene expression profiles from each of the datasets were used independently (not normalized to the others). Tumors proximal to the splenic flexure were defined as proximal, and tumors distal to the splenic flexure were classified as distal.

All stage II/III patients determined to have a KRAS-mutant tumor were included in the analysis. The KRAS mutation sites by codon (where available) are detailed for each of the datasets in Supplementary Fig. S1B.

The interaction between mutant KRAS and all available cytokine/cytokine receptor genes was assessed by Cox proportional hazards tests for interaction using the combined dataset. The top tertile was considered high expression, and genes which yielded an uncorrected P value of less than 0.05 were considered for further analysis.

Tissue composition of the TCGA and an independent cohort, Stratification in COloRecTal cancer (S:CORT), was assessed by visual pathology review. The S:CORT cohort is composed of 385 formalin-fixed, paraffin-embedded (FFPE) tumors from the FOCUS-randomized clinical trial which assessed response to different chemotherapy regimens in patients with advanced colorectal cancer (27). Hematoxylin and eosin slides were scanned at high resolution with an Aperio scanner at a total magnification of 200×. Tissue areas were annotated using the Indica Labs HALO digital pathology platform. Small image tiles representing tumor epithelium, desmoplastic stroma, inflamed stroma, mucin hypocellular stromal areas, necrosis, and glass background were generated from >1,500 pathologist-annotated tissue areas. A deep neural network algorithm (Simoyan and Zisserman VGG https://arxiv.org/abs/1409.1556) was trained to segment and quantify individual tissue areas as described in ref. 28. Area measurements were normalized by total stromal content. Tumors were stratified into molecular subgroups based on IL22RA1 mRNA expression (as described above) and KRAS mutation status.

Differential expression analysis between IL22RA1-high, KRAS MUT and IL22RA1-high, KRAS-WT tumors in the combined dataset was performed using the limma R package (version 3.24.4). Using the t statistics from the differential expression analysis as a weighting factor, a preranked gene set enrichment analysis (GSEA) using the GenePattern interface from the Broad Institute (genepattern.broadinstitute.org/) was performed. GSEA was run on MYC_TARGETS_V2 signature from the MSigDB portal (http://www.broadinstitute.org/gsea/msigdb).

Tumors from the combined dataset were classified according to the “consensus molecular subtypes” (CMS) of colorectal cancer defined by Guinney and colleagues (29). Tumors were then stratified according to IL22RA1 and KRAS mutation status (as described above). The proportion of tumors in each CMS in the molecular subtypes of interest is displayed.

The DLD-1 isogenic pair was purchased from Horizon Discovery Group [Parental Line (KRAS G13D): HD PAR-111 Passage 4; Wild-type reverted line (KRAS±): HD 105-040 Passage 4]. Experiments with the isogenic pair were performed between passages 4–9. KRAS mutation status in the isogenic pair was confirmed in-house using the qBiomarker Somatic Mutation PCR Array Human KRAS Gene (Qiagen, Plate E Format, SMH-806ARE-12 337021). Colo205, SW480, and DLD-1 cells were a generous gift from Dr. Simon Leedham (Wellcome Trust Centre for Human Genetics, Oxford, UK). Colo205 and DLD-1 cells were maintained in RPMI medium (Sigma) with 10% FBS and 1% penicillin/streptomycin (P/S; Sigma). SW480 cells were cultured in DMEM (Sigma) with 10% FBS and 1% P/S. Cultures were maintained in 37°C and 5% CO₂. All cell lines were tested for mycoplasma, and determined to be mycoplasma free.

Cells, 3 × 10⁴ per well in 48-well plates, were cultured overnight. Cytokines [recombinant human IL22 (R&D Systems) and recombinant human IL6 (Peprotech)] were prepared in cell line–specific media at the concentrations indicated. All cytokine stimulation experiments in the DLD-1 isogenic pair were performed in serum-free conditions with 1X ITS (Insulin, Transferrin, Selenium; Sigma). Survival assays confirmed equal viability of both KRAS G13D and KRAS± DLD-1 cells in serum-free media with 1X ITS.

RNA was extracted using the Qiagen RNeasy Mini Kit (Qiagen) according to the manufacturer's instructions. cDNA was synthesized using the High Capacity cDNA Kit (Applied Biosystems) according to the manufacturer's protocol using a C1000 Touch Thermal Cycler (Bio-Rad). Real-time qPCR analysis was performed using TaqMan Real-Time PCR assays (Life Technologies) and Precision Fast qPCR Mastermix with ROX at a lower level, with inert blue dye (PrimerDesign) and run on a ViiA7 Real-Time PCR System (Applied Biosystems). Data were analyzed using the ΔC_T method [2^−ΔCT where ΔC_T = C_{T(target gene)} – C_{T(endogenous control)}]. RPLPO was used as the endogenous reference gene for all qPCR analyses.

Cells were lysed using a solution of 50 mmol/L Tris, pH 6.8, 20 mmol/L EDTA, 5% SDS, 1 mmol/L DTT, and 10% glycerol. Equal protein amounts (15–20 μg) were loaded into precast NuPAGE Novex 4%–12% Bis-Tris Gels (Life Technologies), separated by SDS-PAGE, and transferred onto Immobilon-FL PVDF membranes (Millipore). Membranes were blocked (Odyssey Blocking Buffer; Li-COR), incubated with primary antibody [total c-Myc (1:1,000; Cell Signaling Technology; Clone: D84C12), β actin (1:4,000; Sigma; Clone: Ac-74)] overnight at 4°C, and fluorescently conjugated secondary antibodies (goat anti-rabbit-800CW; 1:15,000; Li-COR; Clone: 925-32211) for 1 hour at room temperature before detection using the Odyssey CLx detection system (Li-COR).

For analysis of IL22RA1 expression, adherent cells were dissociated using TrypLE (Life Technologies), filtered through 70 μm filters, and stained with anti-IL22RA1 (1:10; PE; R&D; Clone 305405) or matched isotype control (1:10; anti-mouse IgG1κ; PE; R&D; Clone 11711), along with anti-EpCAM (1:100; CD326; FITC; Biolegend; Clone 9C4) and a fixable viability dye (1:1,000; APC Cy7; eBioscience) for 30 minutes at room temperature before fixation in Fix/Lyse Solution (BD).

For Phosflow experiments, 1 × 10⁶ cells were stimulated in technical duplicates with 0 to 100 ng/mL IL22 (R&D Systems) for 30 minutes at 37°C with 5% CO₂. Cells were fixed [Cytofix (BD)] and permeabilized [Perm III Buffer (BD)] according to the manufacturer's protocol and stained for 1 hour at room temperature with fluorescently conjugated antibodies to phosphorylated residues on intracellular signaling proteins of interest [anti-pSTAT3-Y705 (1:10; AF-647; BD; Clone: 4) and anti-ERK1/2 (pT202/pY204; 1:10; AF-647; BD; Clone: 20A (RUO))].

Flow cytometric analysis was performed on an LSR or Fortessa X-20 (BD). Data analysis was performed using FlowJo 10 software (Tree Star).

Cells, 1 × 10⁴ per well, were seeded in 48-well plates and allowed to adhere overnight. Cells were stimulated with 0 to 200 ng/mL IL22 or IL6 for 24, 48, and 72 hours. Cellular viability/proliferation was measured by adding 50 μg of MTT to each well 2 hours prior to the end of the IL22 incubation. At the end of the incubation, supernatants were aspirated, formazan particles were solubilized with DMSO, and samples transferred to a flat-bottom 96-well plate (Costar). Absorbance was measured at 540 nm on a BMG SPECTROstar NANO Microplate Reader (BMG Labtech).

Cells, 3 × 10⁴ per well in 48-well plates, were seeded and allowed to adhere overnight. siRNA was incubated with DharmaFect 2 (Dharmacon) in Opti-MEM (Thermo Fisher) for 20 minutes prior to transfection, and cells were transfected using 0.8 μL DharmaFect 2 and 40 nmol/L ON-TARGET Plus Human MYC SMARTpool (Dharmacon) or ON-TARGET Plus control pool (Dharmacon). Media were replaced with serum-free RPMI with 1X ITS (Sigma) after 48 hours.

Patients diagnosed with colorectal cancer scheduled for colectomy at the Churchill Hospital (Oxford, UK), were consented for research use of their resected cancer and matched normal mucosa during a preoperative appointment. Ethical approval for the study was provided by the National Research Ethics Committees of the UK National Health Service (NHS) in accordance with the Declaration of Helsinki (Reference numbers: 11/YH/0020, 16/YH/0247) under the HTA license 12217. The human samples used in this publication were ethically sourced, and their research use was in agreement with the written informed consent. Tumor mutation status was obtained from the Oxford Molecular Diagnostics Centre. Detailed protocols for human organoid culture were generously provided by Marc van de Wetering (Hubrecht Institute, Netherlands). Fresh colorectal cancer biopsies (2 mm³) were homogenized using Soft Tissue Homogenizing CK14 Kit vials (Stretton Scientific) in a Precellys 24 (Bertin Instruments) homogenizer (30 seconds at 3,500 RPM). RNA extraction, cDNA synthesis, and qPCR analysis were performed as described above.

Five to 7 mm fresh punch biopsies of resected tumor were used to establish organoid cultures as described (30). Genomic DNA was extracted from organoids using the QiaAmp DNA MiniKit (Qiagen) and used for KRAS mutation typing. KRAS mutation status was determined using the qBiomarker Somatic Mutation PCR Array Human KRAS Gene (Qiagen, Plate E Format, SMH-806ARE-12 337021) assay according to the manufacturer's protocol. KRAS mutation status clinically determined from whole tissue sections was compared with organoid KRAS mutation status to ensure concordance.

For proliferation assays, organoids were dissociated to single cells with TrypLE, resuspended in reduced growth factor basement membrane extract 2 (BME 2; Cultrex, Amsbio), and seeded in 5 μL BME in 96-well plates. Cells were stimulated with 1 ng/mL IL22 in tumor organoid culture media (30). Media were replaced after 48 hours. Ninety-one hours after initial seeding, the media were refreshed again with 1 ng/mL IL22 where appropriate and 10% Alamar Blue (BioRad) and incubated for an additional 5 hours. Absorbance was measured at 570 and 600 nm as a reference on a BMG SPECTROstar NANO Microplate Reader (BMG Labtech). The percent difference in Alamar Blue reduction for each condition compared with the no treatment condition was determined.

For confocal microscopy, organoids were seeded onto autoclaved 9-mm diameter glass coverslips (VWR) in 24-well plates. Note that 400 μL of organoid media with 1 ng/mL IL22, as appropriate, were added, and organoids were incubated for 30 minutes. Organoids were then fixed on the slides with 4% paraformaldehyde and permeabilized with methanol at −20°C for 10 minutes. Nonspecific antibody binding was blocked by incubation with 10% goat serum in TBS, and organoids were stained with primary antibodies [Phospho-Stat3 (Tyr705) XP Rabbit mAb (1:1,000; 9145; clone D3A7; Cell Signaling Technology) and anti–E-cadherin (1:300; 610181; clone 36/E-cadherin; BD)] overnight at 4°C, followed by staining with secondary antibodies [goat anti-rabbit IgG Alexa Fluor 555 (1:400; Life Technologies; A32732) and goat anti-mouse IgG Alexa Fluor 488 (1:400; Life Technologies; A11001)] for 1 hour at room temperature. Nuclei were stained with Hoechst 33258 (1:20,000; Life Technologies) for 10 minutes at room temperature. Coverslips were inverted and mounted onto Superfrost Plus slides (VWR). Images were acquired on an Olympus FV1200 IX83 Confocal System.

Three independent experiments were performed with DLD-1 isogenic cells at passages 6–8. RNA was extracted following the procedures described above. Library preparation, RNA-sequencing, and quality control were performed by the High-Throughput Genomics Group (Wellcome Trust Centre for Human Genetics, Oxford, UK). Bioanalysis of RNA was performed, and all samples had RNA integrity number (RIN) = 10. Samples were normalized to 1 μg total RNA, and the mRNA fraction was isolated by polyA selection. Libraries were prepared using the TruSeq Stranded mRNA Kit (Illumina) and paired-end sequenced using the Illumina HiSeq4000 platform (Illumina). “Strandedness” was tested as part of internal quality control, and the proportion of reads mapping to the correct strand was found to be very high. Adapter trimming was performed using cutadapt, alignment was performed using hisat2, indexing was performed with SAMBAMBA, and mapped reads were quantified using featurecounts. Principal component analysis (R package prcomp version 3.3.1) was performed on all samples. Pathway analyses were performed by single-sample GSEA (ssGSEA) using the GenePattern web interface from the Broad Institute (genepattern.broadinstitute.org/). The Hallmark collection of well-annotated gene sets from the MSigDB portal (http://www.broadinstitute.org/gsea/msigdb) was used. Enrichment scores from ssGSEA were used for differential analysis using the generalized linear model as implemented in the limma package (version 3.30.0). Differentially expressed pathways with an adjusted P value <0.01 were considered significant. Heat mapping was performed using the pheatmap package (version 1.0.8). Sequencing data were deposited in NCBI's Gene Expression Omnibus and are available through the GEO Series accession number GSE149262.

For prognostic studies in tumor transcriptomic datasets, all analyses were performed using R software (version 3.03). ROC analysis was performed to determine an optimal IL22RA1 cutpoint based on log₂ expression values in the discovery cohort (GSE39582). The presence or absence of disease relapse at last follow-up was used as the binary variable in this analysis. The IL22RA1 cutoff value associated with the maximum Youden index (J = sensitivity + specificity −1) was found to be the 67th percentile. This value was used for all subsequent analyses of IL22RA1 in the discovery and verification datasets, and was also used for analyses of additional cytokine/cytokine receptor genes. Contingency analysis (Fisher exact test with Bonferroni multiple comparisons correction) was used to assess association of clinicopathologic features with IL22RA1 expression status. Univariate, multivariate, and interaction analyses of RFS and OS were performed using Cox's proportional hazard regression models fitted with the survival R package. Interaction analyses were used to assess interactions between KRAS mutation status and the expression of cytokine and cytokine receptor genes. HRs were estimated with model coefficients and 95% confidence intervals (CI), and P values were computed with Wald tests. Time-to-event curves were prepared using Kaplan–Meier methods in GraphPad Prism 7.

For the analysis of in vitro experiments in cell lines and ex vivo experiments in primary organoids, all statistical tests were two-sided and are specified in figure legends with exact P values displayed. The number of independent experiments performed for each in vitro assay is indicated in figure legends. A data point for a given independent experiment is the average of technical replicates (for which 2–3 were performed for each independent experiment). Differences were considered significant when P < 0.05. All line graphs display mean ± SEM. For organoid experiments, single dots represent a value from an individual patient (often the average of technical triplicates). Parametric tests were used when n < 8. These analyses were performed using GraphPad Prism 7.

To assess the role of the IL22 pathway in colorectal cancer, we first examined expression of IL22RA1 in stage II/II colorectal cancer specimens using transcriptomic data from a publicly available population-based cohort (GSE39582, N = 566; ref. 22; Supplementary Fig. S1A). Findings were validated using gene expression data from patients with stage II (N = 108) and stage III (N = 644) colorectal cancer enrolled in the PETACC3 (NCT00026273) phase III clinical trial (N = 752; refs. 23, 24). Finally, we analyzed a merged dataset comprised of patients with stage II/III colorectal cancer from GSE39582, PETACC3, and two additional independent cohorts, TCGA and ALMAC (referred to in tables as the “Combined” dataset; Supplementary Fig. S1A). After excluding patients with missing data or nonstage II/III tumors, 1,083 patients remained for analysis in the combined dataset (see CONSORT diagram; Supplementary Fig. S1A). IL22RA1 was expressed as a continuous variable with a roughly normal distribution (Supplementary Fig. S1C). To determine a cutoff value for classification of IL22RA1-high versus low status, we employed ROC statistics and derived the 67th percentile as the cutoff for all subsequent analyses. High IL22RA1 expression was not significantly associated with patient gender, tumor–node–metastasis stage, KRAS mutation, or BRAF mutation, but was negatively associated with microsatellite instability (MSI) and proximal tumor location (Supplementary Fig. S1D).

In the GSE39582 discovery cohort, IL22RA1 expression had no impact on RFS or OS (Fig. 1A; Supplementary Fig. S2A). Given that Ras is a mediator of IL22 signaling, we next stratified patients based on both IL22RA1 expression and KRAS mutation status for further survival analysis. Consistent with prior reports (21), activating KRAS mutations were modestly associated with poor clinical outcome (HR = 1.57; 95% CI, 1.11–2.23; P = 0.0103; Fig. 1B; Supplementary Fig. S2B). However, among cases with high IL22RA1 expression, KRAS mutations were strongly associated with poor RFS (HR = 2.93; 95% CI, 1.59–5.43; P = 0.0006; Fig. 1C) and OS (HR = 2.45; 95% CI, 1.38–4.36; P = 0.0023; Supplementary Fig. S2C). By contrast, KRAS mutations had no prognostic impact in patients with IL22RA1-low tumors (RFS HR = 1.16; 95% CI, 0.76–1.78; P = 0.4840; OS HR = 1.05; 95% CI, 0.69–1.61; P = 0.813; Fig. 1D; Supplementary Fig. S2D). These findings were validated in the PETACC3 clinical trial cohort (Table 1) and in the large combined cohort (RFS HR = 2.05; 95% CI, 1.45–2.89; P < 0.0001; OS HR = 1.65; 95% CI, 1.09–2.50; P = 0.018; Table 1). Multivariate interaction analysis revealed that the IL22RA1–KRAS interaction was independent of tumor site, MSI status, BRAF mutation status, and sex (Table 2).

Notably, the prognostic effect of the IL22RA1–KRAS interaction was most profound in proximal (right-sided) tumors (RFS HR = 4.23; 95% CI, 1.38–13.01; P = 0.012; OS HR = 9.41; 95% CI, 2.13–41.60; P = 0.003; Fig. 1E and F; Supplementary Table S1). This observation was independent of MSI and BRAF mutation, both of which are common features of proximal tumors (Supplementary Table S2). Previous studies have demonstrated poor prognosis and response to palliative chemotherapy in advanced colorectal cancers with mucinous histology (31, 32). Given that mucinous adenocarcinoma is associated with proximal tumor location (33), we evaluated differences in the mucinous composition of IL22RA1-high proximal versus distal tumors. A histologic analysis was performed on colorectal cancer samples from the TCGA and an independent cohort, S:CORT, composed of 385 FFPE tumors from the FOCUS-randomized clinical trial which assessed chemotherapy response in patients with advanced colorectal cancer (27). There were no differences in the mucinous composition of IL22RA1-high distal versus proximal tumors (Supplementary Fig. S3C).

Low immune cell infiltration is a poor prognostic factor for many tumor types, including colorectal cancer (34). Because IL22RA1 is not expressed on hematopoietic cells, high IL22RA1 expression in specimens analyzed by bulk tumor transcriptomics could be a surrogate for tumors with a paucity of immune infiltrate. In the GSE39582 discovery cohort, IL22RA1 expression was not correlated with markers of antitumor immunity (GZMB, IFNG, and CD8a; Supplementary Fig. S3A). As a control, IFNG was compared with GZMB and CD8a in the cohort. Both comparisons revealed significant correlations, by Pearson correlation analysis (Supplementary Fig. S3A). To better characterize the cellular composition of tumors according to IL22RA1 and KRASstatus, a histologic analysis was performed on colorectal cancer samples from the TCGA and S:CORT cohorts. The tumor cell density, inflamed stromal cell density, and nonneoplastic cell count did not differ between the molecular subtypes of interest (Supplementary Fig. S3B). Taken together, these data suggest that IL22RA1-high tumors do not simply represent a subset of tumors with low immune infiltrate.

Clinical and preclinical studies have suggested that several cytokines, including IL6 and IL17A, may influence colorectal cancer progression (6, 35–40). However, we detected no interaction between IL6R (IL6 receptor) or IL17RA (IL17 receptor) expression (classifying the highest expression tertile for each gene as “high”) and KRAS mutation status in the combined cohort (Table 2). To determine whether other cytokines and/or their cognate receptors interact with KRAS mutation in terms of survival, a Cox proportional hazard interaction analysis was performed on all cytokine and cytokine receptor genes (classifying the highest expression tertile for each gene as “high”) and KRAS mutation status in the combined cohort. Although several cytokine/cytokine receptor genes interacted with KRAS mutation, IL22RA1 displayed the strongest poor-prognosis interaction (Fig. 2A). Similarly, IL10RB, which encodes the second subunit of the heterodimeric IL22 receptor, was also a strong interactor (Fig. 2A). Detailed survival analysis in the discovery cohort (GSE39582) revealed that like IL22RA1-high tumors, KRAS mutation was selectively associated with poor prognosis in IL10RB-high tumors (RFS HR = 3.62; 95% CI, 1.95–6.70; P < 0.0001; OS HR = 2.43; 95% CI, 1.33–4.45; P = 0.0039; Supplementary Fig. S4); this observation was confirmed in the combined cohort (Supplementary Table S3).

To determine the effect of high expression of both IL22RA1 and IL10RB, patients were classified into six groups based on KRAS mutation status, IL22RA1, and IL10RB. KRAS-mutant tumors with high expression of both cytokine receptor subunits displayed the lowest 5-year RFS rates (Fig. 2B). In the cohorts analyzed in this study, 18.9% of stage II/III colorectal cancers were KRAS mutant and expressed high amounts of IL22RA1 and/or IL10RB compared with 18.8% of colorectal cancers that were KRAS mutant and expressed low amounts of both receptors (Fig. 2C).

IL22 can promote epithelial proliferation in response to various stimuli (41–43). To investigate a possible functional interaction between IL22 and mutant KRAS, we tested the proliferative response of a panel of KRAS-mutant and wild-type colorectal cancer cell lines to IL22. All three cell lines [Colo205 (KRAS-WT), DLD-1 (KRAS-MUT), and SW480 (KRAS-MUT)] were functionally responsive to IL22, as demonstrated by induction of SOCS3 (encoding suppressor of cytokine signaling 3), a well-established direct transcriptional target of STAT3 (ref. 44; Fig. 3A). However, there was no proliferative response to IL22 in the KRAS wild-type Colo205 line (Fig. 3B). To determine whether proliferative differences were due specifically to the presence or absence of mutant KRAS, we employed a DLD-1 isogenic cell line system in which the parental line carries a heterozygous KRAS G13D mutation (KRAS-MUT) that was removed to produce paired “wild-type” cells (KRAS-WT). IL22RA1 expression at both the protein (Fig. 3C) and mRNA (Fig. 3D) level was equivalent in the KRAS-MUT and KRAS-WT isogenic cells, thus providing a pair of cell lines with matched IL22RA1 expression, differing only in the presence or absence of mutant KRAS. As expected, phosphorylation of ERK1/2 (pERK1/2) was elevated at baseline in the KRAS-MUT cells compared with their KRAS-WT counterparts based on flow cytometry analysis; but pERK1/2 was not induced by IL22 (Fig. 3E). In contrast, IL22 induced activation of STAT3 (based on flow cytometry analysis of phosphorylation at Y705) in a dose-dependent fashion, but no differences were observed between KRAS-MUT and KRAS-WT cells (Fig. 3F). In accordance with the equivalent STAT3 activation in both lines, SOCS3 induction after IL22 stimulation was similar in the KRAS-MUT and KRAS-WT cells across a range of IL22 doses (Fig. 3G). Next, we explored whether KRAS mutation alters the proliferative response to IL22. Notably, although IL22 enhanced proliferation of both KRAS-MUT and KRAS-WT cells, this effect was significantly greater in KRAS-MUT cells (Fig. 3H). Like IL22, IL6 is a well-known activator of STAT3. However, there was no difference in the proliferative response to IL6 between KRAS-MUT and WT cells (Fig. 3I).

To validate our observations using cells derived from primary tumors, we established organoid cultures from eight freshly resected human colorectal cancer samples. Three of these tumors bore oncogenic KRAS mutations. IL22 stimulation induced robust activation of STAT3 in these organoids (Fig. 3J). KRAS-mutant and wild-type organoids were dissociated to single cells and seeded into organoid culture conditions in the presence or absence of 1 ng/mL IL22 for 96 hours. Consistent with data derived from cell lines, IL22 significantly increased the viable cell number in cultures of KRAS-MUT organoids, but had no consistent effect on KRAS-WT organoids (Fig. 3K).

Together, these data indicate cooperativity between mutant KRAS and IL22 signaling that promotes the proliferation and accumulation of cancer cells.

To identify potential modes of interaction between IL22 and mutant KRAS, we next performed unbiased transcriptomic analysis of DLD-1 isogenic cells following IL22 stimulation. RNA-sequencing was performed on cells stimulated with 10 ng/mL IL22 for 2 hours to identify early transcriptional changes and after 24 hours to explore late changes (Fig. 4A). Given that IL6R expression does not interact with mutant KRAS in a prognostically significant manner (Table 2), nor does it enhance proliferation in KRAS-mutant versus wild-type cells (Fig. 3I), we also stimulated cells with 10 ng/mL IL6 for 2 and 24 hours to identify pathways that are uniquely regulated by IL22 and not IL6 (Fig. 4A).

Principal component analysis revealed that genotype (KRAS-MUT or WT), cytokine stimulation [no treatment (NT), IL22, IL6], and time (2, 24 hours) were the major sources of transcriptomic variation in the dataset (Fig. 4B). GSEA analysis of 24-hour RNA-sequencing data revealed that among the Hallmark gene sets (MSigDB), only the Myc pathway (“MYC_TARGETS_V2”) was uniquely induced by IL22 (and not IL6) in KRAS-MUT but not WT cells (Fig. 4C). MYC expression was significantly increased in KRAS-mutant versus wild-type cells after 2-hour IL22 stimulation (Fig. 4D). Although Myc target gene expression was modestly increased by IL22 in KRAS-MUT cells after 2 hours (Supplementary Fig. S5A), clear preferential induction of Myc targets in KRAS-MUT cells was observed after 24 hours (Fig. 4E).

Based on the results of our transcriptomic analyses, we predicted that c-Myc is preferentially induced in KRAS-MUT cells following IL22 stimulation. Indeed, Western blot analyses demonstrated a marked time-dependent induction of c-Myc protein in KRAS-MUT cells following IL22 stimulation, whereas c-Myc did not accumulate in KRAS-WT cells (Fig. 4F; Supplementary Fig. S5B). To determine whether the IL22-induced proliferation observed in KRAS-MUT cells was dependent on c-Myc, we used siRNA to suppress MYC expression in DLD-1 KRAS-MUT cells, which reduced c-Myc protein expression by approximately half (Supplementary Fig. S5C). MYC knockdown reduced IL22-induced cell proliferation in DLD-1 KRAS-MUT cells (Fig. 4G).

Multiple cell types that are not responsive to IL22 in the tumor microenvironment express MYC, complicating the interpretation of MYC pathway analysis in whole tumor tissue. Nonetheless, to determine whether the functional differences in MYC pathway induction in IL22RA1-high, KRAS-MUT versus WT cells are reflected in the colorectal cancer cohorts assessed in this study, we performed GSEA analysis on transcriptomic data from the combined cohort. The MYC_TARGETS_V2 pathway was not enriched in IL22RA1-high, KRAS MUT tumors compared with IL22RA1-high, KRAS WT tumors (Supplementary Fig. S6A). However, when tumors in the combined cohort were classified into the CMS subtypes defined by Guinney and colleagues (29), IL22RA1 expression was associated with CMS2 status regardless of KRAS mutation (Supplementary Fig. S6B). CMS2 is characterized by a molecular signature of epithelial MYC and WNT activation. Moreover, qPCR analysis of IL22RA1 and MYC in prospectively collected whole tumor biopsy tissue from patients undergoing colectomies at the Churchill Hospital (Oxford, UK), revealed that IL22RA1 positively correlates with MYC expression (Supplementary Fig. S6C). IL22RA1 is not a reported direct transcriptional target of c-Myc. Indeed, MYC knockdown in DLD-1 cells did not alter IL22RA1 expression compared with mock-transfected controls (Supplementary Fig. S6D). Therefore, elevated MYC expression in IL22RA1-high tumors could be a functional consequence of more active IL22 signaling.

Inflammatory responses are observed in both colitis-associated and sporadic colorectal cancer, but whether specific cytokines play beneficial or deleterious roles in cancer is largely context dependent. Moreover, tumors develop adaptations to exploit the growth and survival benefits conferred through cytokine signaling (5). Although data from preclinical studies have suggested that IL22, or its upstream driver cytokine IL23, may be useful therapeutic targets for colorectal cancer, there has been a lack of compelling clinical evidence to support this concept (6–9, 37, 39, 45, 46). Our study extends these previous efforts by demonstrating that IL22 signaling may be critical for a specific subset of patients with colorectal cancer, defined by tumors with high IL22 receptor expression and oncogenic KRAS mutations. Furthermore, our data suggest that IL22 and oncogenic KRAS collaborate to promote cell proliferation via enhanced c-Myc activity. These findings emphasize the important and often unpredictable effects that oncogenic mutations can exert on signals derived from the tumor microenvironment.

Prospective analysis of the PETACC3 cohort has previously demonstrated that KRAS mutations alone have no prognostic value for patients with stage II/III colorectal cancer who receive adjuvant chemotherapy (21). Therefore, the negative prognostic effect of KRAS mutation in patients with IL22RA1-high tumors may be due to a unique interaction between IL22 signaling and a constitutively active KRAS pathway. Notably, in our screen for cytokines and cytokine receptors that interact prognostically with mutant KRAS, IL22RA1 and IL10RB (which encode the IL22 receptor subunits) were the only KRAS-interacting genes that showed a significant negative prognostic association. Indeed, despite the well-described tumor-promoting role of IL6 (35), which is similar to IL22 in its ability to activate STAT3, neither IL6 nor IL6R expression interacted prognostically with KRAS. Intriguingly, high interferon gamma receptor 1 (IFNGR1) expression interacted with KRAS and was associated with improved prognosis, which is consistent with the known antitumor properties of IFNγ (47).

Much of the current understanding of the function and signaling downstream of IL22 has come from studying its physiologic role in microbial defense and tissue regeneration at barrier surfaces. Due to its antiapoptotic and mitogenic effects, IL22 supports intestinal stem cell integrity and tissue regeneration in models of colitis (48–50) and graft versus host disease (43). Recent evidence highlights its critical role in regulating DNA damage response pathways in intestinal epithelial stem cells to protect from mutation acquisition and tumor development (51). However, IL22 can exacerbate immune pathology when produced chronically (52, 53). Based on our data, we propose that IL22 can also be pathogenic in colorectal cancers with adaptations that alter the functional outcome of IL22 signaling. In agreement with previously reported findings from Kryczek and colleagues (11), we found that IL22 promotes expansion of DLD-1 cells, but only if they expressed constitutively active KRAS. This mutant KRAS-dependent proliferative response to IL22 was similarly observed in primary colorectal cancer organoids.

To date, the only evidence of a potential link between IL22 signaling and oncogenic KRAS is derived from Kras-induced lung cancer models in which genetic ablation of Il22 (54) or Il22ra1 (55) reduced tumor burden. Moreover, in a small cohort of patients with KRAS-mutant lung cancer (N = 39), high IL22RA1 expression conferred poor recurrence-free survival (54). Intriguingly, evidence from a preinvasive pancreatic neoplasia (PanIN) model revealed that oncogenic KRAS and chronic pancreatitis synergistically induce IL22 and IL17A production by CD4⁺ and γδ T cells and that oncogenic KRAS augments IL17 receptor expression on PanIN epithelial cells (56). Based on contingency analyses in the tumor transcriptomic cohorts we studied, mutant KRAS and IL22RA1 expressions appear to be independent variables in colon cancer. Consistent with these findings, DLD-1 isogenic cells with or without mutant KRAS had nearly identical IL22RA1 expression. This suggests that the interaction between IL22RA1 and mutant KRAS is at the level of downstream signaling.

Previous studies of IL22-driven colorectal cancer attributed the protumorigenic activity of IL22 to its STAT3-activating effect (7, 8, 11). Interestingly, we observed no difference in the amount of IL22-induced STAT3 activation between KRAS-mutant and wild-type DLD-1 cells. Moreover, the well-described STAT3-activator IL6 did not augment proliferation in KRAS-mutant versus wild-type cells, nor did expression of IL6 or IL6R interact with KRAS to impact prognosis. Therefore, the interaction between IL22 and KRAS is not due to activation of STAT3 alone. Interestingly, the presence of mutant KRAS enhanced IL22-induced expression of c-Myc and its downstream target genes. c-Myc is one of the earliest described proto-oncogenes and is a master transcriptional regulator of genes involved in cellular growth, proliferation, metabolism, and biosynthetic processes (57). Notably, the oncogenic activity of c-Myc is largely a consequence of overexpression, and, in colorectal cancer, mutations within its coding sequence are very rare (57). Myc expression and stability is regulated at both the transcriptional and posttranslational level. STAT3 has been found to augment MYC transcription (58, 59), whereas ERK1/2 and AKT activity downstream of constitutively active Ras signaling enhances c-Myc protein stability and extends its half-life (60, 61); both mechanisms could be at play in IL22-induced c-Myc in KRAS-mutant cells.

Based on the evidence presented here, we propose a stratification strategy in which tumors from patients with colorectal cancer are subjected to standard KRAS mutation typing paired with quantification of IL22 receptor expression. Although both IL22RA1 and IL10RB contribute prognostic information, IL22RA1 expression is restricted to epithelial cells in the colon and may thus be a more suitable biomarker. Colorectal cancers harboring KRAS mutations and high expression of IL22RA1 and/or IL10RB comprise approximately 19% of the total colorectal cancer population. Such patients are predicted to have lower responsiveness to conventional chemotherapy, increased frequency of disease relapse, and reduced OS time. Beyond prognostic stratification, our data suggest that pharmacologic blockade of IL22 signaling may be therapeutically beneficial for KRAS-mutant tumors. Notably, monoclonal antibodies targeting IL22 and its upstream driver IL23 have been developed for inflammatory indications and were well tolerated in phase I and II studies (NCT00563524, NCT00883896, NCT01866007, NCT02203032, and NCT01225731). Furthermore, antibodies specific to the p19 subunit of IL23 are now being investigated in phase II/III trials in inflammatory bowel disease (NCT03926130, NCT03466411, NCT03759288, and NCT03398135) and will provide gut-specific safety and efficacy data.

Although we have revealed a novel interaction between KRAS and IL22 in colorectal cancer, several questions remain. For example, our analyses focused on the direct effect of IL22 on cancer cell–intrinsic pathways, but whether IL22 differentially affects the tumor microenvironment based on KRAS mutation status is unknown. In addition, our cohorts were not sufficiently powered to study whether IL22 signaling interacts with constitutively active BRAF, an immediate downstream mediator of Ras signaling that is mutated in approximately 10% of colorectal cancers. Because IL22RA1 expression manifests as a continuous, nonbiphasic variable, further prospective studies assessing IL22RA1 expression are necessary (and ongoing in our laboratory) to establish a clinically relevant cutpoint for identifying patients with high expression.

Collectively, the evidence presented here suggests a method to both identify a subset of poor-prognosis colorectal cancer patients with KRAS-mutant tumors and to target these tumors using therapeutic blockade of the IL22 pathway. Interactions between cytokines and oncogenic mutations are not limited to IL22 and KRAS. Exploration and exploitation of cytokine–oncogene interactions more broadly may be beneficial for precise patient stratification and the application of immunomodulatory therapies in cancer.

S. McCuaig, N.R. West, and F. Powrie are inventors on an unlicensed patent owned by Oxford University Innovation (Technology Transfer University of Oxford) which describes a method for the treatment and prognosis of colorectal cancer, especially proximal colorectal cancer, and also relates to identifying patients with colorectal cancer who are likely to respond to therapy with an inhibitor of interleukin 22 signaling. N.R. West is an employee/paid consultant for Genentech. F. Powrie is an employee/paid consultant for GSK and Genentech, and reports receiving commercial research grants from Roche and Janssen. No potential conflicts of interest were disclosed by the other authors.

Conception and design: S. McCuaig, N.R. West, F. Powrie

Development of methodology: S. McCuaig, D. Barras, M. Friedrich, V.H. Koelzer, N.R. West

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. McCuaig, D. Barras, E.H. Mann, S.J. Bullers, A. Janney, L.C. Garner, E. Domingo, V.H. Koelzer, M. Delorenzi, S. Tejpar, T.S. Maughan, N.R. West

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. McCuaig, D. Barras, E.H. Mann, V.H. Koelzer, M. Delorenzi, S. Tejpar, T.S. Maughan, N.R. West, F. Powrie

Writing, review, and/or revision of the manuscript: S. McCuaig, E.H. Mann, E. Domingo, V.H. Koelzer, S. Tejpar, N.R. West, F. Powrie

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. McCuaig, D. Barras, S.J. Bullers, V.H. Koelzer, M. Delorenzi

Study supervision: T.S. Maughan, N.R. West, F. Powrie

Other (support with data acquisition via cell line experiments): A. Janney

We thank the Oxford IBD Cohort Investigators (Carolina V. Arancibia-Cárcamo; Adam Bailey; Ellie Barnes; Elizabeth Bird-Lieberman; Oliver Brain; Barbara Braden; Jane Collier; James East; Alessandra Geremia; Lucy Howarth; Satish Keshav; Paul Klenerman; Simon Leedham; Rebecca Palmer; Fiona Powrie; Astor Rodrigues; Jack Satsangi; Alison Simmons; Peter Sullivan; Simon Travis; Holm Uhlig) for helping to facilitate and coordinate GI specimen collection. Thank you to Cloe Vassart, James Chivenga, Ngonidzashe Charumbira, and David Maldonado-Perez who consented patients and collected samples from the operating theatres for this study. We thank the team of pathologists at the JR Hospital for biopsying specimens and S. Page in the Oxford Molecular Diagnostics Centre for providing tumor mutation status data. We thank Dr. Simon Leedham for generously providing cell lines and Dr. Marc Van de Wetering of the Hubrecht Institute, Utrecht, for providing human organoid protocols. We thank the High-Throughput Genomics Group at the Wellcome Trust Centre for Human Genetics (funded by Wellcome Trust grant reference 090532/Z/09/Z) for generation of the RNA-sequencing data and Nicholas Illott for assisting with NCBI GEO data deposition. We thank the patients and their families for contributing to this study.

This work was funded by an Oxford/CRUK Development Fund Grant (C25255/A18085), an MRC Experimental Medicine Grant (MR/N02690X/1), and CRUK OPTIMISTICC Grant (C10674/A27140). S. McCuaig was supported by the Rhodes Trust. N.R. West was supported by an Irivington Institute Postdoctoral Fellowship (Cancer Research Institute). L.C. Garner is supported by a Wellcome Trust PhD Studentship (109028/Z/15/Z). E. Domingo is supported by the S:CORT Consortium which is funded by a grant from the Medical Research Council and Cancer Research UK. This work was also supported by the NIHR Oxford Biomedical Research Centre. The views expressed are those of the authors and not necessarily those of the NHS, the NIHR, or the Department of Health.

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.