What induces Treg cells to express Il23r remains largely unknown. (11) Given that pro-inflammatory mediators such as IL-6 are involved in the polarization of naïve T cells towards IL-17 producing Th17 cells, which also upregulate Il23r for Th17 cell maintenance (12), we hypothesized that chronic inflammation may increase Il23r expression in colonic Treg cells. To experimentally test this, Il23r ^flox/+ Foxp3 ^YFP-iCre mice were subjected to three cycles of dextran sulfate sodium (DSS) for 5 days that were interrupted by a 10-day rest period each cycle to mimic the relapsing and remitting inflammation experienced by patients with IBD ( Figures 1A, B ). Reverse transcription-quantitative polymerase chain reaction (RT-qPCR) was performed on flow-sorted colonic Treg cells form DSS treated mice based on YFP expression. Il23r and Il12rb1, which forms a heterodimer with IL-23R and required for IL-23R signaling, were both increased following DSS treatment indicating that a functional IL-23R could be formed ( Figure 1C ). We then sought to determine if IL-23R signaling in Treg cells was required for chronic intestinal inflammation induced by DSS. To this end, we generated Il23r ^flox/flox Foxp3 ^YFP-iCre mice (Il23r ^ΔTreg) that lack IL-23R specifically in FOXP3⁺ Treg cells and compared these to IL-23R competent littermate Foxp3 ^YFP-iCre mice (hereafter referred to as WT). At the experimental endpoint no changes in weight loss or colonic inflammation were observed by histology indicating that IL-23R signaling in colonic Treg cells is not required for chronic intestinal inflammation induced by chronic DSS ( Figures 1D, E ).

For patients with IBD, chronic inflammation contributes to an increased risk for developing intestinal carcinogenesis. Recent therapeutic strategies have targeted the IL-23R pathway to induce remission in patients with IBD. Although there is no evidence suggesting increased risk of CRC in these patients, IL-23 blockade is relatively new. Our recent work showed that colonic Treg cells unable to signal through IL-23R have increased capacity to restrict effector T cell responses and exhibit reduced cell turnover. (10) Thus, we hypothesized that within the tumor microenvironment, disrupting IL-23R signaling in Treg cells may interfere with anti-tumor immunity. To test this, we employed a model of inflammation-associated carcinogenesis (IAC) utilizing the mutagen azoxymethane (AOM) followed by repeated cycles of DSS ( Figure 1F ). We fist examined Il23a expression and found elevated expression in AOM/DSS tumors compared to adjacent normal tissue of WT mice ( Figure 1G ), consistent with a previous report (13) and likely the product of antigen presenting cells. (14) Following tumor induction with AOM/DSS in Il23r ^ΔTreg and WT mice, no differences in weight loss or survival were observed between groups ( Figures 1H, I ) while endoscopic examination confirmed tumor development in both groups ( Figure 1J ). At the experimental endpoint, tumor number was similar between groups while tumor volume of individual tumors was increased in Il23r ^ΔTreg mice ( Figure 1K ). Furthermore, a larger proportion of tumors exhibited high-grade dysplasia in Il23r ^ΔTreg compared to WT mice ( Figures 1L, M ). These findings suggest that the enhanced suppressive capacity and reduced turnover of IL-23R-deficient colonic Treg cells previously reported by our group (10) may restrict anti-tumor immunity in IAC.

To gain insight into the potential mechanism by which IL-23R-deficient Treg cells restrict anti-tumor immunity in IAC, we examined whether a numerical difference in Treg cells might drive the observed increase in tumor volume in Il23r ^ΔTreg mice. Our previous work established that the frequency of Treg cells in mLN and spleen was similar between Il23r ^ΔTreg and WT mice at baseline and only differed in the colon. (10) In contrast, following tumor induction with AOM/DSS, the frequency of Treg cells in the spleen and mLN was higher in Il23r ^ΔTreg mice, suggesting that intestinal inflammation can alter Treg cell frequency at other lymphoid compartments in an IL-23R-dependent manner ( Figure 1N ). Surprisingly, the frequency of Treg cells in AOM DSS-induced colonic tumors did not differ between Il23r ^ΔTreg and WT mice ( Figure 1O ). Finally, carcinogenesis in the AOM/DSS model is accompanied by the formation of tertiary lymphoid structures. (15) We assessed number of lymphoid aggregates and found no correlation with tumor number or genotype ( Supplementary Figure S1A ), nor did a survey of lymphoid aggregates indicate a difference in Treg cell frequency within the aggregates themselves ( Supplementary Figure S1B ). Collectively, these data suggest that a difference in Treg cell function or impact on another cell type rather than Treg cell frequency contributes to the difference in tumor volume between Il23r ^ΔTreg and WT mice.

To explore the potential impact of IL-23R-deficient Treg cells on other immune cell subsets within the tumor microenvironment, we defined transcriptional alterations within the tumor microenvironment of WT and Il23r ^ΔTreg mice by performing RNA-sequencing (RNA-seq) of tumor and tumor-adjacent colon tissue (hereafter adjacent normal). Differentially regulated genes were used to identify common and unique signatures of WT and Il23r ^ΔTreg tumors. Numerous cytokines and chemokines were upregulated in tumors of WT mice compared to adjacent normal tissue that were not altered when comparing Il23r ^ΔTreg tumors versus adjacent normal tissue ( Figure 2A ). Gene set enrichment analysis (GSEA) was employed to identify pathways differentially regulated in tumors of WT vs. Il23r ^ΔTreg mice ( Figure 2B ). Interestingly, pathways related to Th1, Th2, and Th17 cell differentiation as well as cytokine-receptor interactions were enriched in tumors of WT mice compared to Il23r ^ΔTreg mice. Conversely, a pathway uniquely upregulated in the tumors of Il23r ^ΔTreg mice was both canonical and non-canonical Wnt pathway genes. Considering the important role of Wnt signaling in epithelial stem cell biology and colorectal carcinogenesis, we quantified the number of OLMF4⁺ cells per high power field (hpf) as a proxy to enumerate intestinal stem cells. OLMF4 is a stem cell marker that has been reported to inhibit AOM/DSS-induced inflammation and crypt proliferation. (16, 17) No differences were observed in the number of OLMF4⁺ cells between WT and Il23r ^ΔTreg tumors ( Supplementary Figure S2A ). Collectively, these data suggest that differences in immune cell composition may contribute to the differences in tumor burden between WT and Il23r ^ΔTreg mice.

To delineate the cellular composition of AOM/DSS tumors, cell type analysis was performed computationally to extrapolate cell type identity from the RNA-seq data. After exclusion of cell types predicted to be absent from the tissue ( Supplementary Figure S2B ), 49 cell types remained ( Figure 2C ). Analysis using PERMANOVA showed that genotype contributed to the differences in cell composition between tumors. Post-hoc analysis indicated that dendritic cells and macrophages were decreased in tumors recovered from Il23r ^ΔTreg mice compared to tumors of WT mice. To corroborate the findings from the deconvolution analysis, flow cytometry of intratumoral dendritic cells (DCs) and macrophages was performed in independent validation experiments. This analysis showed that the frequency of intratumoral DCs (CD11c⁺CD45⁺CD3^negCD19^negNK1.1^neg) was increased in tumors from Il23r ^ΔTreg mice, while the frequency of intratumoral macrophages (CD11c^negLy6G^negCD45⁺CD3^negCD19^negNK1.1^neg) trended lower in the tumors Il23r ^ΔTreg when compared to WT ( Figure 2D ). Thus, these results were partially concordant with the deconvolution analysis and suggest that the absence of IL-23R signaling in Treg cells decreases intratumoral macrophages during AOM/DSS-mediated carcinogenesis. Finally, we examined the regulation of the intratumoral effector CD4⁺ T cell response in absence of IL-23R-signaling in Treg cells and found that effector T cells produce more IFNγ in absence of IL-23R signaling in Treg cells ( Figure 2E ). Altogether, these data show that IL-23R signaling in Treg cells restricts AOM/DSS-mediated carcinogenesis, potentially via regulation of intratumoral macrophages and/or cytokine production by CD4⁺ effector T cells.

Since the majority of CRC cases are sporadic, we also sought to examine the role of IL-23R signaling in Treg cells in sporadic CRC. Prior work has utilized MC-38 cells injected subcutaneously (SC).(9) However, to employ a more relevant in vivo setting, we performed endoscope-guided orthotopic injections of MC-38 cells into the colonic submucosa ( Figure 3A ). We then generated Foxp3 ^RFP Il23r ^GFP/+ mice (IL-23R Treg reporter) and validated IL-23R protein expression using in vitro induced Th17 cells (not shown). We then determined expression of IL-23R protein in Treg cells in colon tissue and MC-38 tumors injected SC or orthotopically ( Figure 3B ). We then compared tumor development and burden of submucosally injected MC-38 tumors cells in WT and Il23r ^ΔTreg mice two weeks following injection of MC-38 cells. The survival of WT mice was lower compared to the survival of Il23r ^ΔTreg mice ( Figure 3C ) with tumor volume in Il23r ^ΔTreg mice lower compared to tumor volume in WT mice ( Figures 3D, E ). Interestingly, this was in contrast to our findings in the inflammation-associated carcinogenesis model. Since transcriptional analysis and flow cytometry implicated antigen presenting cells as being differentially regulated in Il23r ^ΔTreg mice following AOM/DSS, we assessed pro-inflammatory macrophages (Ly6C⁺MHCII⁺) by flow cytometry and found the frequency of intratumoral pro-inflammatory macrophages to be increased in Il23r ^ΔTreg mice ( Figure 3F ). To determine whether MC-38 tumors are larger in Il23r ^ΔTreg mice due to macrophages, we depleted macrophages via intraperitoneal clodronate injection and orthotopically injected these mice with MC-38 tumors cells. However, due to clodronate-attributed mortality no conclusion could be drawn regarding the macrophage dependency of the orthotopic MC-38 model ( Figure 3G ). Given that Il23r ^ΔTreg mice have an increased frequency of colonic Treg cells that are also more suppressive than colonic Treg cells from WT mice (10), we anticipated that intratumoral Treg cells may be increased in the tumor MC-38 model. However, the frequency of intratumoral Treg cells did not differ between Il23r ^ΔTreg and WT mice ( Supplementary Figure S3A ). Collectively, these data suggest that IL-23R-deficient Treg cells protect mice from MC-38-mediated sporadic colorectal carcinogenesis, likely via increased intratumoral pro-inflammatory macrophages that promote anti-tumor immunity.

Previous work shows that IL-23R signaling in Treg cells in SC-injected MC-38 cells inhibit IFNγ production by CD8⁺ T cells. (9) Using a similar approach, we quantified cytokine production following orthotopic injection of MC-38 cells and observed a decreased IFNγ production by CD4⁺ cells that approached statistical significance, but no increase was observed in CD8⁺ T cells ( Supplementary Figures S3B–C ). Similarly, an increase in the frequency of IL-17⁺ CD4⁺ T cells also approached statistical significance in Il23r ^ΔTreg mice compared to WT mice ( Supplementary Figure S3C ). The frequency of intratumoral dendritic cells and neutrophils was not altered between WT and Il23r ^ΔTreg mice ( Supplementary Figure S3D ).

Since the injection medium utilized with the orthotopic model differs from traditional SC injections, we mirrored the orthotopic model by injecting MC-38 cells SC into Il23r ^ΔTreg and WT mice. Here we observed a decrease in tumor volume from Il23r ^ΔTreg mice compared to WT mice similar to the orthotopic model ( Supplementary Figure S4A ). However, in contrast to the lower CD4⁺ T cell IFNγ production observed in Il23r ^ΔTreg mice orthotopically injected with MC-38 cells, in SC MC-38 cell tumors, there was no significant difference in IFNγ production by CD4⁺ or CD8⁺ T cells between genotypes ( Supplementary Figures S4B–C ). Similarly, no differences in the frequency of intratumoral Treg cells were observed ( Supplementary Figure S4D ). To further define the immune cell tumor microenvironment, we quantified the frequency of myeloid cells including dendritic cells, neutrophils, and macrophages. While the frequency of intratumoral dendritic cells and neutrophils between WT and Il23r ^ΔTreg mice was similar ( Supplementary Figure S4E–F ), the frequency of intratumoral pro-inflammatory macrophages was increased in Il23r ^ΔTreg mice compared to WT mice ( Supplementary Figure S4G ). Finally, to determine the role of macrophages in driving enhanced anti-tumor immunity in Il23r ^ΔTreg mice, we depleted macrophages in the SC model using paratumoral injections as previously described. (18) Interestingly, depletion of macrophages increased MC-38 tumor size in Il23r ^ΔTreg mice whereas the tumor size in WT mice treated with clodronate was not statistically significant different from WT mice treated with control liposomes ( Supplementary Figure S4A ). Thus, IL-23R-deficient Treg cells inhibit tumorigenesis in the subcutaneous sporadic MC-38 model of carcinogenesis in a macrophage-dependent manner.

Our prior investigations in mice show that IL-23R signaling in Treg cells impairs their suppressive function of effector T cells and leads to increased Treg cell turnover. (10) Given the poor prognosis associated with increased tumor Treg cells in many solid tumors in humans, we sought to investigate whether IL-23R signaling in Treg cells may be active in this setting. To explore this possibility, we surveyed data obtained from The Cancer Genome Atlas (TCGA) to examine IL23R expression across various cancer types and found that IL23R expression was increased specifically in colon and rectal cancers when compared to other cancer types ( Figure 4A ). We subsequently focused on sporadic CRC as this form of CRC is more prevalent than IAC in humans. By comparing paired tumor and adjacent normal tissue samples from 24 individuals with colon cancer, we found transcript levels for the IL-23-specific subunit IL23A and FOXP3 were both elevated in tumor tissue compared to adjacent normal tissue ( Figure 4B ). The increase in tumor IL23A was corroborated in a validation cohort via qPCR using patient-matched tumor and non-tumor biopsies from 9 CRC patients at Vanderbilt University Medical Center ( Figure 4C ). Since whole tissue expression of IL23R does not provide the cell type-specific resolution, we analyzed a CRC single-cell RNA-seq dataset (19) that included cells from 12 patients. In this cohort we found IL23R expression to be readily detectable in tumor Treg cells but not in circulating Treg cells ( Figure 4D ). Taken together, this suggest that both Treg cells and IL-23 are increased within colon tumors and highlights the importance of evaluating tumor resident cells.

Given the increase in IL23A transcripts in tumors, it is reasonable to consider the possibility that IL-23 may act directly on tumor epithelial cells to promote tumor development as intestinal epithelial cells also express IL23R. To explore this possibility, we treated generated human organoids derived from healthy colon or CRC tumors and treated with exogenous IL-23 and measured the impact on cell growth. In this ex vivo model, IL-23 did not potentiate organoid growth after 3 days in culture ( Supplementary Figure S5A ). Since this may have been an artifact of the ex vivo culture system, we examined activation the downstream target of IL-23R (i.e., STAT3) in patient tumors. Using a tumor array containing intestinal tissue biopsies from CRC patients and adjacent normal tissue, we performed immunohistochemical staining for phospho-STAT3 (pTyr705). Activated STAT3 was readily detected in non-epithelial cells of tumor sections, but not adjacent normal tissue ( Supplementary Figure S5B ). While these data would be consistent with IL-23R pathway activation in tumor tissue, STAT3 activation can also be mediated by IL-6 or IL-10. It is noteworthy, however, that expression of IL6 and IL10 was similar between tumor and adjacent normal tissue (data not shown). Collectively, these data support the idea that IL-23R signalling in Treg cells may antagonize the normal suppressive function of Treg cells to promote anti-tumor immunity in CRC.

C57BL/6 Foxp3 ^YFP-iCre (The Jackson Laboratory, Bar Harbor, ME, stock #016959, designated WT in the text), ll23r ^ΔTreg, lacking IL-23R in FOXP3⁺ Treg cells, Foxp3 ^RFP (Jackson, stock #008374) and Il23r ^GFP (Jackson stock #035863) mice were used. Il23r ^ΔTreg mice have been previously described. (10) Mice were housed in a specific-pathogen free vivarium at Vanderbilt University Medical Center (VUMC). All experiments were performed using 6–15-week-old female or male mice unless otherwise indicated. Littermates were used where possible. Experiments were approved by the VUMC Institutional Animal Care and Use Committee.

7-8-week-old mice were co-housed, bedding was mixed 14 days prior to the start of the protocol, and 3 days prior to start of the protocol mice were placed on deprivation caps to acclimate mice to drinking exclusively from water bottles. Mice were intraperitoneally injected with 10 mgkg^-1 azoxymethane (Sigma-Aldrich) and exposed to three 5-day cycles of 3.5% dextran sodium sulphate (TdB consultancy, Uppsala, Sweden). Each DSS cycle was followed by a 16-day recovery period of regular autoclaved water.

MC-38 cells were grown in DMEM with high glucose and pyruvate (Gibco, NY, #11995065) supplemented with 10% FBS and 100 U/mL pen/strep. MC-38 cells were confirmed to be pathogen-free by the Vanderbilt Translational Pathology Shared Resource prior to in vivo use. All experiments were performed with cells from the same passage (passage 34). For injections, cells were suspended in 40% PBS, 50% DMEM, and 10% matrigel (Corning, NY, #356231). 5x10⁴ cells in 50 uL were injected SC into the right flank or colonoscope-guided orthotopically injected into the mucosal lining of the distal colon (31).

At euthanasia, images were taken using a SMZ1270 dissecting scope (Nikon, Melville, NY) in concert with NIS Elements version 5 (Nikon). Tumors were located and identified by vascular, non-dimpled appearance from these images in conjunction with direct visualization. Tumors were measured macroscopically in two perpendicular dimensions using digital calipers, and tumor volume was calculated using the previously validated formula W2*L/2, where W (width) was the shorter caliper measurement, and L (length) was the longer caliper measurement (32).

Macrophage depletion was done with clodronate (brand name clodrosome) and compared to mice injected with control liposome (both from Encapsula NanoSciences, TN). For the orthotopic MC-38 model, mice were maintained on enrofloxacin water (0.8 mgmL^-1) and injected with 100 uL clodronate or control liposomes one day prior or one the same day as MC-38 injection followed by a second injection three days after the first injection. For the subcutaneous MC-38 model, mice were injected with 50 uL clodronate or control liposomes every other day with the first injection on the same day as the MC-38 injection.

Hematoxylin and eosin (H&E) staining was performed by the Vanderbilt Translational Pathology Shared Resource on formalin-fixed paraffin-embedded (FFPE) colon sections or a custom tumor micro-array. For IHC, slides were placed on the Leica Bond IHC stainer. All steps besides dehydration, clearing, and coverslipping were performed on the Bond stainer. Slides were deparaffinized and hydrated. IHC for OLFM4 was done by FR with the following protocol: (1) Antigen Retrieval: Citrate buffer pH 6.0, decloaking chamber 105°C for 15 minutes, 10 minute bench cool down; (2) Peroxidase Block: 0.03% H₂0₂ with sodium azide, 5 minute incubation; (3) Primary Antibody (Cell Signaling, 39141, Massachusetts): 1:500 dilution, overnight incubation; (4) Detection: Dako (K4003) Envision + HRP Labelled Polymer, 30 minute incubation; (5) Chromogen: DAB⁺, 5 minute incubation. All incubations are at room temperature. Scoring of histology was performed by pathologist (MKW) who was blinded to experimental conditions.

Tumors were isolated, minced, and placed into 25 mL prewarmed RPMI 1640 media containing 0.1 mgml^-1 Liberase TL (Roche, Basel, Switzerland), 0.05% DNAse I (Sigma-Aldrich, D5025) and 20 mM HEPES and shaken at 37°C for 60 minutes in a non-CO₂ MaxQ4450 horizontal shaker (Thermo Fisher Scientific, Waltham MA). Cells were pulled through a 10 mL syringe 20 times and filtered through a 70-μm cell strainer into an equal volume of cold RPMI 1640 media containing 5% FBS, 0.05% DNAse I, 20 mM HEPES on ice. Cells were spun for 10 minutes at 4°C and 475 g and resuspended in 40% Percoll (Sigma-Aldrich) solution and underlaid using 90% Percoll. The 40/90 gradient was spun for 25 minutes at 20°C at 475 g with no brake or acceleration. The interphase layer was recovered and washed in fluorescence cytometry (FC) buffer (PBS w/o Ca2⁺/Mg2⁺ containing 2% FBS and 2 mm EDTA) and spun again for 10 minutes at 20°C and 475 g prior to downstream application(s).

Images from whole slides were captured using a high-throughput Leica SCN400 Slide Scanner automated digital image system from Leica Microsystems. Whole slides were imaged at 40X magnification to a resolution of 0.25 µm/pixel. Images were also taken with a EVOS XL Core followed by adjustment of brightness and contrast using Adobe Photoshop (equal for all images).

Extraction of RNA from whole tissue or organoids was performed on tissue stored in RNA-later (Sigma-Aldrich) until homogenization using a Tissue-Tearor (Dremel, Racine, WI), followed by phenol/chloroform extraction as described (33) and clean-up with the Rneasy Mini Kit with on-column DNAse digestion according to manufacturers’ instructions.

Quality control for RNA was performed by the Vanderbilt Technologies for Advanced Genomics core using RNA 6000 Pico (Agilent, Santa Clara CA), and cDNA library preparation using a NEB library preparation kit. Paired end 150bp sequencing was performed on a NovaSeq 6000 (Illumina, San Diego, CA).

Samples were trimmed with fastp (version 0.20.0) using default parameters. Quantification was performed using Salmon (version 1.4.0) against a decoy transcriptome (Mus musculus Gencode version 21). Further analysis was performed in R (version 4.1.2) in R studio (version 2021.09.2 + 382). For differential expression analysis, limma (version 3.50.3) was used on log-CPM transformed counts with prior count set to 3 or DESEq2 (version 1.34.0) was used on non-normalized counts. Unless otherwise indicated, genes with an adjusted p value <0.05 (-log₁₀ of 0.05 is 1.3) and log₂ fold change >|1| were considered differentially regulated. For heatmaps with gene counts, a Z-score, i.e. the number of standard deviations above or below the mean, of normalized counts was calculated using scale function in R.

For GSEA using Webgestalt, the web-based version accessible at webgestalt.org was utilized with their default parameters. The input was a list of all genes sorted on fold change in increasing order generated using Limma. Resulting normalized enrichment scores (NES), which reflect the degree to which a gene set is overrepresented at the extremes of the entire ranked list (34) and size of the gene sets were plotted. Annotation was done with AnnotationDbi (version 1.56.2) using org.Mm.eg.db (version 3.14.0). Images were generated with pheatmap (version 1.0.12), complexheatmap, RColorBrewer (version 1.1-3), ggplot2 (version 3.3.5), and EnhancedVolcano (version 1.12.0).

For cell type analysis (also known as deconvolution) xCell was used following the instructions from the authors of the package. (35) Cell types with a p-value cutoff of 0.20 in a beta distribution were excluded from the analysis, as this suggested that these cell types were not present in the tissue. Then, cell type scores were recalculated. An arbitrary cutoff of 4 samples was used to determine whether gene expression related to a certain cell type was present, i.e. if the gene expression associated with a cell type was present in less than 4 samples (in any group), the cell type was excluded from analysis to increase the specificity of the cell types that were present.

Since the data were not normally distributed as determined by the multivariate normality test (using mvnormtest), a permutational analysis of variance (PERMANOVA, Bray-Curtis, 999 permutations) was used to examine contribution of genotype and tissue (tumor or normal). To this end, genotype and tissue were combined into a composite variable with 4 groups. Post-hoc tests were done with a multilevel pairwise comparison (pairwise adonis, version 0.4.1) and corrected for multiple comparisons according to Benjamin-Hochberg. Adjusted p<0.05 was considered statistically significant. Finally, for the multiple comparisons that differed, Similarity Percentages (SIMPER) analysis was used to examine cell types contributing to dissimilarity. Again, adjusted p<0.05 was considered statistically significant.

To stimulate T cells, single cell suspensions were incubated for 5 hours at 37°C (5% CO₂) with 50 ngmL^-1 phorbol myrisate acetate (PMA) and 1 μgmL^-1 ionomycin (Biolegend, California) in the presence of Golgi-Stop (BD) for the last 4 hours. For cell surface staining, samples were blocked using 30 mL normal rat serum (StemCell Technologies) and incubated in the antibody cocktail for 20 minutes at 4°C in the dark. Intracellular cytokine staining was performed using Cytofix/Cytoperm (BD, New Jersey) according to manufacturer’s instructions. Flow cytometric analysis was performed using a 4-Laser Fortessa, 5-laser LSRII (BD) with FACSDiva software (BD), or a Cytek Aurora (Cytek, CA). Fluorescence-activated cell sorting (FACS) was performed on a FACS Aria III (BD). Analyses were performed using FlowJo (BD Biosciences). For all flow experiments, a live/dead stain (ThermoFisher) was used to only assess live cells. Antibodies used for flow cytometry or cell sorting are listed in Supplementary Table S1.

Deidentified human colorectal tumor and adjacent normal colon tissues were collected at Vanderbilt University Medical Center and provided by the NCI Cooperative Human Tissue Network in accordance with the Vanderbilt Institutional Review Board. Organoids were established in accordance with published methods. (36) Briefly, for tumor organoids, minced tumor tissue was incubated in digestion buffer (Advanced DMEM F12, Gibco; 10% FBS; 100 U/mL penicillin; 100 μgmL^-1 streptomycin; 0.125 mgmL^-1 Dispase II, Roche; 0.1 mgmL^-1 collagenase XI; Sigma-Aldrich) for 1 hour at 37°C. Following digestion, the supernatant containing digested tumor fragments was collected, washed, and suspended in growth factor reduced Matrigel (356231, Corning). For normal colon organoids, intestinal mucosa was minced, and tissue fragments were then incubated in chelation buffer (2 mM EDTA in phosphate-buffered saline) for 30 minutes at 4°C before 2 minutes of gentle shaking to free intestinal crypts. Crypts were collected and suspended in growth factor reduced Matrigel (Corning). Human organoids were overlaid in Advanced DMEM F12 media containing 50% LWRN-cell (ATCC) conditioned media, B27 (ThermoFisher), N2 (ThermoFisher), 100 U/mL pen/strep (Gibco), 1% HEPES (Corning), 1% Glutamax (Gibco), 1 mM N-acetylcysteine (A9165, Sigma), 50 ngml^-1 EGF (2028EG200, R&D), 500 nM A-83-01 (2939, Tocris), 10 μM SB202190 (S7067, Sigma), 10 mM nicotinamide (Sigma), 10 nM [Leu15]-Gastrin I (G9145, Sigma), 100 μgmL^-1 Primocin (Invivogen), and 3 μM CHIR99021 (4423, Tocris, for colon organoids only). To assess changes in size, human tumor and colonic organoids were split by ezymatic digestion (TrypLE, Gibco) or manual dissociation by vigorous pipetting, respectively, and replated. Fields of organoids were imaged, and the same fields were imaged again after 3 days. Size for all images was quantified in ImageJ by an investigator blinded to treatment status.

cDNA was synthesized from 1 µg of RNA using qScript XLT cDNA SuperMix (Quantabio, Massachusetts). RT-qPCR was performed in technical duplicates with TaqMan™ probes (ThermoFisher, Massachusetts) and TaqMan™ Universal PCR Master Mix (ThermoFisher, Massachusetts). RT-qPCR results were analyzed by the delta-delta Ct method and normalized to Gapdh. For RT-qPCR of mouse sorted Treg cells the methods previously described were used (10).

Data for Figure 4A was retrieved using the web interface from UALCAN (37). For Figure 4B , expression levels were queried from Illumina HiSeq and Illumina GA RNASeqV2 data in The Cancer Genome Atlas [TCGA(38)] colon adenocarcinoma (COAD) data set (N=264 colorectal cancer) and normalized to expression levels observed in adjacent normal colon (N=39 normal colon). Only patients that had both tumor and adjacent normal were included in the latter analysis. No statistics were done.

Details regarding the statistical analyses are indicated in the figure legends. N represents biological replicates, unless otherwise indicated. Data shown are representative or pooled data from at least two independent experiments with similar results. RNA-sequencing (RNA-seq) was performed on multiple samples from one experiment. The sequence of sample processing was counterbalanced. Age-, sex- and, where feasible, littermate-matched mice were used. Cage-effects were examined. Apart from bulk RNA-seq, statistical analyses were performed using GraphPad Prism (9.1.2). Other than RNA-seq, graphs were made in GraphPad Prism and edited with Adobe Illustrator. Sample size was determined empirically. Outliers were not removed except for Figure S4A , in which one outlier was removed from the WT control mice and one outlier was removed from the ll23r ^ΔTreg control mice. Positive cells in IHC images were detected using QuPath (0.4.3) using the “positive cell detection” function, except for OLMF4, which was counted manually by two blinded investigators.