β-Catenin Drives Butyrophilin-like Molecule Loss and γδ T-cell Exclusion in Colon Cancer

Intraepithelial lymphocytes (IEL) expressing the γδ T-cell receptor (TCR) account for nearly 50% of all T cells in the mouse gut and 10% to 30% of all T cells in the human intestinal tract. These cells actively migrate in the space between the enterocyte layer and the basement membrane, surveying for abnormalities, including cancer (1, 2). Although diverse, the TCRs of most mouse γδIELs include a Vγ7 chain that facilitates critical interactions with butyrophilin-like (BTNL) molecules, specifically, heterodimers consisting of BTNL1 with BTNL4 or BTNL6 (3–5). Vγ7⁺ cells ordinarily reside only in gut tissue, owing at least in part to the largely restricted expression of BTNL1, BTNL4, and BTNL6 to intestinal epithelial cells (IEC; refs. 3–6). The BTNL1/6 or BTNL1/4 interaction drives Vγ7⁺ IEL expansion and maturation during postnatal development and is thereafter required for maintaining the signature phenotype of Vγ7⁺ IEL (5, 7). The BTNL1/6–γδ T-cell axis in mice is also conserved in humans: human BTNL3 and BTNL8 dimers bind to and regulate Vγ4⁺ IELs (5, 6, 8). The localization of γδ IELs and of BTNL expressions aligns with a decreasing WNT signaling gradient that runs from crypt to villus. As such, Vγ7⁺ IELs are rarely found in the crypt regions where WNT signaling is high.

Most colorectal carcinomas exhibit mutations in members of the WNT pathway that drive tumor initiation and progression to malignancy. These mutations almost exclusively manifest in the form of truncating mutations in the APC tumor-suppressor gene, preventing the degradation of β-catenin, which leads to uncontrolled proliferation (9). Like intestinal stem cells residing in crypt regions, colon cancer cells require WNT signaling to maintain their stemness and de-differentiated phenotype (10, 11). However, the relationship between dysregulated WNT signaling in cancer and local, tissue-resident IELs remains wholly unexplored.

Here, we investigated Vγ7⁺ IEL function and the expression of BTNL molecules during tumor initiation and growth. We found that β-catenin signaling in IECs decreased expression of Btnl genes and the transcription factors that regulate them, HNF4A and HNF4G. This molecular rewiring also promoted γδ T-cell exclusion from tumors. Conversely, inhibition of β-catenin signaling restored HNF4 transcription factor expression, Btnl1 gene expression and intratumoral γδ T-cell infiltration. However, reexpression of BTNL1 and BTNL6 in cancer cells was not sufficient to increase Vγ7⁺ IEL infiltration into tumors. Collectively, our data suggest that aberrant WNT signaling in tumors elicits disarray in the tissue-resident γδ T-cell compartment, disrupting natural tissue immunosurveillance as cancer cells dedifferentiate and acquire stem cell–like characteristics.

Animal experiments were approved by the UK Home Office (70/8645 and PP6345023 to Karen Blyth, 70/8646, PP3908577 to Owen Sansom, P0B63BC4D to Simon Leedham) and Institutional Animal Care and Use Committees at the Universities of Glasgow and Oxford. Animal experiments were carried out in line with the Animals (Scientific Procedures) Act 1986 and the EU Directive 2010. All mice were maintained on the C57BL/6J background at the Cancer Research UK Beatson Institute, except Vil1-Grem1 mice and Lgr5-Cre^ERT2;Rspo3^INV mice, which were maintained at the Functional Genetics Facility, Wellcome Center for Human Genetics, University of Oxford. Mice were bred and housed in individually ventilated cages under specific pathogen-free conditions on a 12/12-hour light/dark cycle and fed and watered ad libitum. Both male and female mice at least 6 weeks old and ≥20 kg were used for experiments.

Mice with the following genetic modifications were as follows: Vil1-Cre^ERT2 (RRID:IMSR_JAX:020282), Apc⁵⁸⁰ (RRID:MGI:1857966), Kras^G12D (RRID:MGI:2429948), Bcl9^F/F (RRID:MGI:4398979), Bcl9l^F/F (RRID:MGI:4398981), Ctnnb1^ex3/+ (RRID: MGI:1858008), Lgr5-Cre^ERT2 (RRID:IMSR_JAX:008875). Rspo3^INV mice were obtained from John Hilkens (Netherlands Cancer Institute; ref. 12). Vil1-Grem1 mice were generated by Simon Leedham (University of Oxford; ref. 13). Btnl1^—/— mice were generated by Adrian Hayday (Francis Crick Institute; ref. 5). The generation of Vil1-Cre^ERT2;Apc^F/+ (VA) mice, Vil1-Cre^ERT2;Apc^F/+;Kras^G12D/+ (VAK) mice, Vil1-Cre^ERT2;Apc^F/F (VA^F/F) mice, Vil1-Cre^ERT2;Apc^F/F;Kras^G12D/+ (VA^F/FK) mice, VA;Bcl9^F/F;Bcl9l^F/F mice, V;Ctnnb1^ex3/+ mice, and V;Ctnnb1^ex3/+;Bcl9^F/F;Bcl9l^F/F mice has been described previously (14–17). Generation of Lgr5-Cre^ERT2;Rspo3^INV mice has been previously described (12). Cre-negative mice were used as controls. Recombination in these tumor models was induced by a single intraperitoneal injection of 80 mg/kg tamoxifen (Sigma-Aldrich T5648–1G), except for VA^F/F mice, which 2 injections of 80 mg/kg tamoxifen on 2 consecutive days. Mice were aged until they showed clinical signs (i.e., anemia, hunching and/or weight loss). Tumors were scored macroscopically after fixation of opened intestinal tissue. Tumor burden was calculated by summing the area of all tumors. VA^F/F mice were sacrificed 4 days after the first tamoxifen injection, and VA^F/FK mice were sacrificed 3 days after tamoxifen injection. Recombination in Lgr5-Cre^ERT2;Rspo3^INV mice was induced by intraperitoneal injection of 1-mg tamoxifen for 5 consecutive days. The porcupine inhibitor LGK-974 was administered by daily oral gavage at 1 mg/kg in 0.5% hydroxypropyl methylcellulose.

For CT26 experiments, female BALB/c mice aged 6 weeks were purchased from Charles River Laboratories. BALB/c mice were used as source for Vγ7⁺ cells for coculture experiments (below). At 8 weeks old, mice were injected with a total of 5×10⁵ CT26 (carrying empty pLIX401 vector) or CT26-B1/6 cells (carrying Btnl1-pLIX401-puromycin and Btnl6-pLIX401-blasticidin vectors as described below) in PBS subcutaneously into the left flank. Doxycycline (1 mg/mL DOX in 2% sucrose; Sigma-Aldrich D9891–25G) was added to drinking water on day of CT26 injection and provided until endpoint. Mice were monitored three times per week, and tumors were measured by calipers. Humane endpoint was defined as tumors reaching 15 mm in any direction. For orthotopic transplantation, a total of 1×10⁶ CT26 (carrying empty pLIX401 vector) or CT26-B1/6 cells in 70-μL PBS were injected into the submucosal layer of the colon of 10-week-old BALB/c mice, using a Karl Storz TELE PACK VET X LED endoscopic video unit as previously described (18). Doxycycline (1 mg/mL DOX in 2% sucrose) was added to drinking water 7 days after CT26 injection and provided until endpoint. Mice were monitored for weight loss and paling 3 days per week. Humane endpoint was defined as weight loss ≥20%.

CT26 cells were maintained in RPMI-1640 medium (Thermo Fisher Scientific 31870074) with 10% heat-inactivated FCS (Thermo Fisher Scientific 15808947), 100 U/mL penicillin/streptomycin (Thermo Fisher Scientific 15140148), 2 mmol/L l-glutamine (Thermo Fisher Scientific 25030081), 50 μmol/L β-mercaptoethanol (Thermo Fisher Scientific 21985023), 1 mmol/L sodium pyruvate (Thermo Fisher Scientific 11360088). MODE-K and HEK-293T cells were maintained in DMEM, 4.5 g/L D-glucose with GlutaMAX (Thermo Fisher Scientific 31331093), supplemented with 10% heat-inactivated FCS and 1% penicillin–streptomycin. Cells were kept at 37°C and 5% CO₂. CT26 cells were obtained from Stephen Tait (University of Glasgow, UK) in 2021. HEK-293FT cells (RRID:CVCL_HA71) were obtained from James Neil (University of Glasgow, UK) in 2017. MODE-K cells were obtained from Dominique Kaiserlain (University of Lyon, France) in 2014. No authentication was performed. Mycoplasma testing was performed monthly. After thawing, cells were not used past 15 passages.

The murine transcription factors Cdx1, Cdx2, Creb3l3, Gata5, Hnf4a, and Isx were cloned into pCSIGPW [lentiviral vector bearing an IRES-GFP reporter and resistance to puromycin, described in ref. (5)] by conventional RT-PCR, using cDNA derived from C57BL/6 IECs as template and the following primers (lowercase symbols correspond to spacers and restriction sites): Cdx1 forward 5′-atatgaattcCCCTGCGGTCACCATG-3′, reverse 5′-atatctcgagCAGGCTGCAAGGGGCTAG-3′; Cdx2 forward 5′-atatgaattcGTCCCTCGCCACCATG-3′, reverse 5′-atatctcgagCCACGGGAGGGGTCAC; Creb3l3 forward 5′-atatctcgagCAGGCACGGGACTCATG-3′, reverse 5′-atatgcggccgcAGGGCTGTCTGAGTCTGTCAC-3′; Gata5 forward 5′-atatctcgagCGCGGGGAAAAAAAATG-3′, reverse 5′-atatgcggccgcGTGACAGTTTCCTGAGCACCTAG-3′; Hnf4a forward 5′-atatgaattcCGTGGGTAGGGGAGAATG-3′; reverse 5′-atatgcggccgcCCCCAGCAGCTTGCTAG-3′; Isx forward 5′-atatgaattcAGCAGGGCTAGGCCATG-3′; reverse 5′-atatgcggccgcGCTGCCTGCCATCAC-3′. All plasmids used for lentiviral transduction were purified using a NucleoBond Xtra Midi EF kit (Macherey-Nagel 740420.50). Lentiviral particles were produced in HEK-293T cells by cotransfection of pCSIGPW either empty vector (EV) or encoding the indicated murine transcription factors, HIV-1 gag-pol pCR/V1 (19) and VSV-G env pHIT/G (20). Medium was replaced 16 hours after transfection and was collected at 48 hours, filtered through 0.45-μm cellulose acetate mesh, and then added to subconfluent MODE-K cells grown in 24-well plates. After 24 hours, MODE-K cells were transferred to 12-well plates in DMEM, 4.5 g/L D-glucose with GlutaMAX, 10% heat-inactivated FCS, 1% penicillin–streptomycin, supplemented with 1 μg/mL puromycin (Merck P8833–25MG), and selected for 2 weeks. Where necessary, cells were sorted on the basis of GFP expression to homogenize expression across lines. Expression of Btnl1, Btnl4, Btnl6, and Gapdh (positive control for the presence of cDNA) was evaluated in MODE-K lines transduced with the indicated transcription factors or the EV control (above) by conventional RT-PCR, using the following primers: Btnl1 forward 5′-GACTTGACCTTCCACTCTGATG-3′, reverse 5′-ATTCCCTGTGCACATCACTTAG-3′; Btnl4 forward 5′-ATGGAAAATCACCGCAAGCCC-3′, reverse 5′-CTTCTACATTCCCACAAGGAGC-3′; Btnl6 forward 5′-ATGGAAAATCACCGCAAGCCA-3′, reverse 5′-CTAACTTCTTCCCATTCTTCCC-3′; Gapdh forward 5′-GTAGACAAAATGGTGAAGGTCG-3′, reverse 5′-GACTCCACGACATACTCAGCAC-3′. cDNA derived from C57BL/6 IECs was used as a positive control of Btnl expression.

Mouse Btnl1 and Btnl6 cDNA was purchased from Invitrogen in pMA-T and pMK-RQ plasmids, respectively. DNA fragments of Btnl1 and Btnl6 with NheI and SbfI were inserted into the doxycycline-inducible plasmid pLIX401 with puromycin resistance (Addgene 41393-DNA.cg). To generate a blasticidin-resistant Btnl6-pLIX401 plasmid, a partial PGK promoter followed by a blasticidin-resistant gene in pMK-RQ, and the puromycin-resistant gene was replaced with the blasticidin-resistant gene at AscI and XbaI sites. Viral supernatants were prepared as before (21), following transient transfection of HEK-293FT cells with empty pLIX401, Btnl1-pLIX401-puromycin, or Btnl6-pLIX401-blasticidin together with pSPAX2 packaging vector (RRID:Addgene_12260) and pVSVG envelope vector (RRID:Addgene_85140) using Lipofectamine 2000 (Thermo Fisher Scientific, 11668019). CT26 cells were transduced with empty pLIX401, Btnl1-pLIX401-puromycin, or Btnl6-pLIX401-blasticidin viral supernatants, and then expanded in 6 μg/mL puromycin or 8 μg/mL blasticidin (Thermo Fisher Scientific A1113903), as appropriate. Stable CT26 cells containing Btnl1-pLIX401-puromycin vector were then transduced with Btnl6-pLIX401-blasticidin viral supernatants, and CT26 cells containing Btnl6-pLIX401-blasticidin vector were then transduced with Btnl1-pLIX401-puromycin. These cells were expanded in both 6 μg/mL puromycin and 8 μg/mL blasticidin to generate CT26 cells, expressing both Btnl1 and Btnl6. Confirmation of Btnl1 and Btnl6 mRNA expression was performed by qRT-PCR as described below after varying doses of doxycycline for 24 hours.

Tissues from Cre-negative, VA, VA;Btnl1^—/—, VAK, VA^F/F, VA^F/FK, VA;Bcl9^F/F;Bcl9l^F/F, V;Ctnnb1^ex3/+, V;Ctnnb1^ex3/+;Bcl9^F/F;Bcl9l^F/F, Vil1-Grem1, and Lgr5-Cre^ERT2;Rspo3^INV mice as well as CT26 tumors were fixed overnight in 10% neutral-buffered formalin, then embedded in paraffin. Staining was performed on 4-μm sections that had been heated at 60°C for 2 hours. Primary antibodies used for IHC were as follows: CDX1 (1:250; Invitrogen #PA5–23056), CDX2 (1:200; Abcam #ab76541), HNF4A (1:10,000; Perseus Proteomics #pph1414–00), HNF4G (1:1,000; Novus Biologicals #NBP1–82531), and SOX9 (1:500; Millipore #AB5535). HNF4A and SOX9 were detected by an Agilent AutostainerLink48 using high pH citrate buffer (Target Retrieval Solution, Aglient #K8004/K8005) and peroxidase blocking. CDX1, CDX2, and HNF4G were detected on a Leica Bond Rx autostainer, using ER2 antigen retrieval solution (Leica #AR9640). For RNAscope, the following probes were used from Advanced Cell Diagnostics: Btnl1 (436648) and Trdc (449358). Staining was performed on a Leica Bond Rx autostainer according to Advanced Cell Diagnostics instructions. Images were acquired with an Olympus BX51 or Zeiss Axio Imager.A2 microscope. For each antibody or RNAscope probe, staining was performed on tissue sections from at least three mice of each genotype, and representative images are shown for each staining. The average number of γδ T cells was determined by HALO image analysis software (Indica Labs, CytoNuclear v1.5 algorithm) in 10⁶ μm² tissue from 1 to 5 villi or tumors within each mouse.

RNA sequencing (RNA-seq) data from wild-type (WT), VA^F/F, and VA^F/FK mouse intestinal tissue were generated for previous studies (ArrayExpress E-MTAB-7546; refs. 14, 15, 22). In this study, we performed analysis of these data as previously described where raw counts per gene were determined using FeatureCounts (RRID:SCR_012919) version 1.6.4 (22). Differential expression analysis was performed using the R package DESeq2 (RRID:SCR_000154) version 1.22.2, and principal component analysis was performed using R base functions. RNA-seq data from WT, Hnf4a^Δ/Δ, Hnf4g^Δ/Δ and Hnf4a^Δ/Δ;Hnf4g^Δ/Δ mice were analyzed as previously described (23); these data were obtained from GSE112946.

Tumors and 1 cm² of jejunum from Cre-negative, VA, and VAK mice were cut into small pieces using the McIlwain Tissue Chopper and digested on the gentleMACS Octo Dissociator with Heaters (program, 37C_m_TDK_1) using the mouse Tumor Dissociation Kit (Miltenyi Biotec 130–096–730) according to the manufacturer's instructions, and prepared cells were resuspended in 0.5% BSA in PBS. Cells were stained in the brilliant stain buffer (BD Biosciences 566349) containing antibodies for 30 minutes at 4°C in the dark. The following antibodies were used: CD19-APC-eFluor780 (clone 1D3; 1:400; eBioscience 47–0193–82), CD3ε-BV650 (clone 17A2; 1:100; BioLegend 100229), CD8α-BUV805 (clone 53–6.7; 1:50; BD Biosciences 564920), EpCAM-APC-eFluor780 (clone G8.8; 1:100; eBioscience 47–5791–82), and TCRδ-FITC (clone GL3; 1:200; eBioscience 11–5711–85). Dead cells were identified with Zombie NIR Fixable Viability dye (BioLegend 423106). Cells were acquired using a 5-laser BD LSRFortessa flow cytometer with DIVA software (BD Biosciences). Data were analyzed using FlowJo Software (RRID:SCR_008520) version 9.9.6.

The Scotland cohort of paraffin-embedded colon tumors was assembled from 1,030 patients who had undergone a resection for Stage I to IV colon cancer between 1997 and 2007 at the Glasgow Royal Infirmary, Western Infirmary or Stobhill Hospital in Glasgow, UK. The West of Scotland Research Ethics Committee granted study approval (16/WS/0207). Patient information and tissue blocks were held within the Glasgow and Clyde Safe Haven (12/WS/0142). Tumors were staged using the 5th edition of AJCC/UICC-TNM staging system. A subcohort of 144 samples was selected for IHC, and tissue was available from 142 patients where both tumor and normal adjacent tissue was visible. The Norway cohort was assembled from 299 patients who had undergone a resection for Stage II to III colon cancer between 2000 and 2020 at the Southern Hospital Trust in Norway. The Sørlandet Hospital Ethics Committee granted study approval. Tumors were staged using the 5th edition of AJCC/UICC-TNM staging system from 2000 to 2009, the 7th edition from 2010 to 2017, and the 8th edition thereafter. A subcohort of 84 samples was selected for IHC, and tissue was available from 71 patients where both tumor and normal adjacent tissue was visible. The Thailand cohort was assembled from 411 patients who had undergone a resection for Stage I to IV colon cancer between 2009 and 2016 at hospitals in Thailand. These samples were approved by the Siriraj Institution Review Board (COA no.Si544/2015). Tumors were staged using the 6th or 7th editions of AJCC/UICC-TNM staging system. A subcohort of 136 samples was selected for IHC, and tissue was available from 122 patients where both tumor and normal adjacent tissue was visible. Across all cohorts, tissues were assembled retrospectively so no patient consent was obtained, and samples were excluded if patients had received neoadjuvant chemotherapy or died within 30 days of surgery. These studies were conducted in accordance with the ethical principles of the World Medical Association Declaration of Helsinki. Clinicopathological characteristics are listed in Supplementary Table S1.

IHC was performed on full tissue sections with citrate buffer (pH 6.0) antigen retrieval with standard protocols, using an anti-TCRδ antibody (1:300; clone H-41, Santa Cruz Biotechnology #sc-100289, lot K1318 or K2618) previously validated (24). Scoring of γδ T cells was conducted using VisioPharm software. The first level of tissue compartments (primary tumor, adjacent normal tissue) was manually annotated. A tissue classifier was built using RGB and hematoxylin features with the application of a K-means clustering algorithm and was trained using sections from all cohorts. A pan-lymphocyte detector was built using a five-pixel mean filter applied to the chromogenic DAB feature and a dual hematoxylin feature consisting of a polynomial smoothing filter and a polynomial Laplace filter at a field size of 15 pixels at an order of two. The output metric is defined as the percentage of total cells within an analyzed region that are positively identified as the target cell type.

TempO-Seq whole-transcriptome profiling was performed on 82 patients from the Scotland cohort, according to the manufacturer's instructions using whole formalin-fixed, paraffin-embedded (FFPE) tissue sections. 77 out of the 82 had matched γδ T-cell IHC data. FFPE tissue was deparaffinized before tissue digestion. Crude tissue lysates were used as input for whole-transcriptome analysis using the Human Whole Transcriptome v2.0 panel (Biospyder Technologies). Detector oligos, consisting of a sequence complementary to an mRNA target plus a universal primer landing site, were annealed in immediate juxtaposition to each other on the targeted RNA template and ligated (25). Amplification of ligated oligos was performed using a unique primer set for each sample, introducing a sample-specific barcode and Illumina adaptors. Barcoded samples were pooled into a single library and run on an Illumina HiSeq 2500 High Output v4 flowcell. Sequencing reads were demultiplexed using BCL2FASTQ software (Illumina). FASTQ files were aligned to the Human Whole Transcriptome v2.0 panel, which consist of 22,537 probes, using STAR (26). Up to two mismatches were allowed in the 50-nucleotide sequencing read. Deseq2 was used to normalize raw read counts. Linear regression analysis on paired samples was performed using Prism software (version 9.3.1).

Oncomine (oncomine.org) was used to query gene expression levels of BTNL3, BTNL8, HNF4A, HNF4G, CDX1, and CDX2 in microarray analyses from the TCGA (9) and Skrzypczak (27) cohorts. The TCGA dataset consisted of 19 normal and 101 tumor tissues. The Skrypczak dataset consisted of 24 normal and 45 tumor tissues. Median, minimum, maximum expression levels were extracted from Oncomine analysis and imported into Prism software (version 9.3.1) for visualization. Expression levels are presented as log₂ median-centered intensity.

The Marisa cohort consists of fresh-frozen primary tumor samples from patients with colon cancer collected and transcriptionally profiled as described previously (28). The normalized, batch corrected microarray data for the Marisa cohort were downloaded from Gene Expression Omnibus (GEO; RRID:SCR_005012) using the accession number GSE39582. This dataset had been processed using the Robust Multi-Array Analysis method and corrected for technical batch effects using ComBat as described previously (28). Probesets were collapsed to the gene level by selecting the probeset with the highest mean expression value across all samples for each gene using the collapseRows function (method = “MaxMean”) from the WGCNA package (29) using R (v3.3.2). Only tumor samples from patients with Stage II or III disease who did not receive adjuvant chemotherapy that had relapse-free survival data (n = 258) were used analysis.

Small intestine was harvested from mice of indicated genotypes. Organoids were generated as previously described (15, 22), cultured in Matrigel (Corning, 356231) with ENR medium [Advanced DMEM/F12 containing 2 mmol/L Glutamine, 10 mmol/L HEPES, 1× N2 supplement (Thermo Fisher Scientific 17502048), 1× B27 supplement (Thermo Fisher Scientific 12587010), 50 ng/mL EGF (PeproTech AF-100–150), 100 ng/mL Noggin (PeproTech 250–38), 1,000 ng/mL R-spondin 1 (PeproTech 120–38), 100 U/mL of penicillin, and 100 U/mL of streptomycin]. Organoids were split every 2 to 3 days. Where indicated, organoids from WT (Cre-negative mice), VA^F/F, VK, and VA^F/FK mice were treated with 1 μmol/L 4-Hydroxytamoxifen (4-OHT, Sigma T5648) in ENR medium for 48 hours. Organoids from Cre-negative mice were treated with 3 and 10 μmol/L CHIR-99021 (Sigma SML1046) or DMSO as a control (1:300 dilution) in ENR medium for 6 days. Medium containing CHIR-99021 (GSK3β inhibitor) was changed every day. After 6 days, organoids were cultured in ENR medium without CHIR-99021 or DMSO for 2 days. Organoids from Cre-negative mice were treated with 100 μmol/L BI-6015 (Cayman CAY12032) or DMSO as a control (1:100 dilution) in ENR medium for 3 days. Cells were collected for downstream analysis on indicated day after treatment. Biological replicates were generated from individual mouse organoid lines.

Short hairpin (sh)RNA target sequences designed against Cdx1, Cdx2, Hnf4a, Hnf4g, and Sox9 were selected from Merck Mission shRNAs (https://www.sigmaaldrich.com/GB/en/product/sigma/shclnd). 5 sequences per gene were subcloned into the pLKO.1-Puro lentiviral backbone (Addgene #8453), and inserts sequenced before use. Viral supernatants were prepared following transient transfection of 293FT cells with pLKO.1 encoding shRNAs, pSPAX2 packaging vector and pVSVG envelope vector using Lipofectamine 2000 as described previously (21). Two 24-hour supernatants were collected sequentially over a 48-hour period, pooled and filtered through a 0.45-μm syringe filter and then concentrated using the Lenti-X Concentrator solution (Clontech/Takara). Intestinal WT organoids were expanded 3 days before infection in normal growth medium supplemented with 1 μg/mL R-Spondin, 3 μmol/L CHIR-99021, 10 μmol/L Y27632 (ROCK inhibitor, Cambridge Bioscience SM02–1), and 1 μmol/L Jagged-1 (Notch Ligand 1, Cambridge Bioscience 188–204) to enrich stem and progenitor cells (30). VA^F/F organoids that received Sox9 shRNAs were similarly expanded but no supplements were added. Organoids were reseeded into the same medium 24 hours before infection. Freshly concentrated viral supernatants were added directly to harvested, manually disrupted organoids in the presence of 8 μg/mL polybrene (Merck TR-1003-G) and mixtures seeded into 12-well plates coated with a fine film of Matrigel. Organoid fragments were left to attach overnight and then drained before overlaying with a fine film of Matrigel. Organoids were expanded in culture medium as above, supplemented with 1 μg/mL R-Spondin (WT organoids only) and 3 μg/mL puromycin.

RNA was isolated from fresh intestinal organoids using the Qiagen's RNeasy kit (74106) with on-column DNA digestion. RNA concentration and purity (cutoff = 2.0–2.2 260/280 ratio) was determined using a Thermo Scientific NanoDrop spectrophotometer with NanoDrop 2000 software. cDNA was prepared from 0.5 to 1 μg RNA using a Quantitect Reverse Transcription Kit (Qiagen 205311) and diluted to 2.5 ng/mL in DEPC-treated water. For quantitative RT-PCR, 12.5 ng aliquots of cDNA were amplified in triplicate on an ABI 7500 real-time PCR machine using SyGreen Blue Mix Lo‐ROX PCR master mix (PCR Biosystems PB20.15–51) and primers (below), all at 2.0 μmol/L except for Btnl1 (1 μmol/L Fwd; 4 μmol/L Rv), with endogenous controls Hprt (Mm_Hprt_1_SG; Quantitect) and β-actin (Mm_Actn_1_SG; Quantitect). Relative expression was calculated by the ΔC_t method after averaging endogenous controls. Data are displayed as fold change (2^−ΔΔCt). The following primer sequences were used for each gene: Btnl1 forward 5′-CCGGGAACACGCTACTGTC-3′, reverse 5′-CAAACCAGGGCTACTTTCCAT-3′; Btnl2 forward 5′-TTTGCTATGGATGACCCTGC-3′, reverse 5′-TCCTGATTGCTGCTGTGTGT-3′; Btnl4 forward 5′-CATTCTCCTCAGAGACCCACACTA-3′, reverse 5′-GAGAGGCCTGAGGGAAGAA-3′; Btnl6 forward 5′-CGTGTGGAGGATAATAAGGCAGA-3′, reverse 5′-TCCTTGCGCCAATCTGCATAC-3′. The other primers were purchased from QIAGEN (Quantitect Primer): Hprt (QT00166768); Axin2 (QT00126539); Lgr5 (QT00123193); Sox9 (QT00163765); Cdx1 (QT00265139): Cdx2 (QT00116739); Cd44 (QT00173404); Hnf4a (QT00144739); Hnf4 g (QT00169799).

Promoter sequences for mouse Btnl1, Btnl2, Btnl4, and Btnl6 and human BTNL3 and BTNL8 (12 kb upstream of ATG start site) were extracted from the UCSC Genome Browser (31). These sequences were analyzed by The Open Regulatory Annotation database (ORegAnno; ref. 32) for putative transcription factor–binding sites using ApE software.

Chromatin immunoprecipitation (ChIP)-seq data were generated previously (23) on 3 mice per group using anti-HNF4A (6 μg, Santa Cruz Biotechnology #sc-6556 X, lot B1015) and anti-HNF4G (6 μg, Santa Cruz Biotechnology #sc-6558 X, lot F0310) antibodies. ChIP-seq data were obtained from GSE112946, analyzed as described previously (23), and visualized using IGV (33).

LEF1 and IgG-negative control cleavage under targets and release using nuclease (CUT&RUN) datasets and ATAC-seq data from three biological replicates of HEK-293T cells were analyzed as described previously (34, 35) and visualized in IGV (33). These data are available from ArrayExpress (E-MTAB-12077 and E-MTAB-12076). Conservation data were obtained from the UCSC genome browser 100 vertebrates’ track (31). ConTra (36) was used to determine JASPAR LEF1 core binding sites (MA0768.1). Genomic regions enrichment of annotations tool (GREAT) annotation analysis (37) was used on default settings to assign genes to genomic coordinates of the LEF1 signal enrichment.

Murine small intestine (SI) from naive Balb/c mice was isolated and lymph nodes and mesentery were removed. SI was washed with PBS and cut into 1×1-cm pieces and transferred into ice-cold PBS. SI pieces were dissociated in HBSS with 1 mmol/L DTT and 5 mmol/L EDTA at 37°C for 30 minutes with rocking at 75 rpm, after which tissue pieces were physically disrupted by pipetting. SI pieces were then transferred into fresh PBS for further physical disruption by pipetting again. SI pieces were removed and both dissociation solutions were combined, passed through 70-μm strainers and centrifuged to pellet cells. Cell suspension was centrifuged on 20/40/80% Percoll density gradient at 700 × g for 30 minutes. Lymphocytes were collected from the interface between 40/80% Percoll. Cells were stained for the following antibodies in Brilliant Stain buffer (BD Biosciences) followed by live dead cell staining using Zombie Violet Fixable Viability kit (BioLegend): NKp46-FITC (clone 29A1.4; 1:100; BioLegend 137606); EpCAM-FITC (clone G8.8, 1:50, eBioscience 11–5791–82); CD11b-BV786 (clone M1/70, 1:200, BioLegend 101203); CD3ε-BV650 (clone 17A2, 1:100, BioLegend 100229); TCRβ-BV510 (clone H57–597, 1:100, BioLegend 109234); CD8α-APC-eFluor780 (clone 53–6.7, 1:50, eBioscience 47–0081–82); Vγ4-PE-Cy7 (clone UC3–10A6, 1:100, BioLegend 137708); CD11c-PE-Cy5 (clone N418, 1:100, eBioscience 15–0114–82); Vγ1-PE (clone 2.11, 1:200, BioLegend 141106); and CD19-PE (clone 1D3/CD19, 1:100, BioLegend 152407). Cells were filtered and transferred into 2% FCS, 25 mmol/L HEPES, 2 mmol/L EDTA in PBS. Vγ7⁺ cells were sorted by negative selection into IMDM medium containing 8% FCS, 100U/mL penicillin streptomycin, 50 μmol/L β-mercaptoethanol using a FACSAria cell sorter with DIVA software (BD Biosciences).

Twenty-four hours before coculture, CT26 and CT26-B1/6 cells were plated into a 96-well plate at a concentration of a total of 2×10³ cells per well in RPMI medium containing doxycycline (1 μg/mL). Mouse Vγ7⁺ cells (2×10⁴ cells) were added to CT26 cells in medium containing IL2 (10 IU/mL, PreproTech 212–12), IL3 (100 IU/mL, PreproTech 213–13), IL4 (100 IU/mL, PreproTech 214–14), and IL15 (100 ng/mL, PreproTech 210–15) and doxycycline (1 μg/mL). Cells were cultured for 24 hours at 37°C. After 24 hours, medium was collected, and then CT26 cells were detached by trypsin. Detached CT26 cells and collected medium were combined, centrifuged to form a cell pellet and stained in Brilliant Stain buffer (BD Biosciences) with the following antibodies and analyzed using a BD Fortessa flow cytometer with DIVA software: EpCAM-eFluor780 (clone G8.8, 1:50, eBioscience 47–5791–82); CD3ε-FITC (clone 145–2C11, 1:100, eBioscience 11–0031–86); TCRδ-PE (clone GL3, 1:100, BioLegend 118108); Vγ7-Dylight650 (clone F2.67, 1:100, BioLegend 161702); CD44-PerCP-Cy5.5 (clone IM7, 1:50, BioLegend 103032); CD25-BV711 (clone PC61, 1:100, BioLegend 102049); CD45-BV605 (clone 30-F11, 1:50, eBioscience 83–0451–42); and CD122-PE/Dazzle594 (clone TM-β1, 1:100, BioLegend 123217).

An unpaired t test or the nonparametric Mann–Whitney test or paired t test was used to compare two groups. One-way ANOVA was used to compare groups of three or more followed by the Tukey or Dunnett post hoc test. The log-rank (Mantel–Cox) test was used to analyze Kaplan–Meier survival curves. Correlation between genes was determined using the Pearson correlation coefficient. P values less than 0.05 were considered statistically significant. Graphs were generated and statistical significance calculated using Prism software (version 9.3.1). The statistical tests used are indicated in figure legends. For all animal and organoid experiments, each data point represents an individual mouse or individual organoid line.

Data analyzed in this study were obtained from GEO at GSE39582 (RNA-seq and ChIP-seq data) and GSE112946 (microarray data), and ArrayExpress at E-MTAB-7546 (RNA-seq data), E-MTAB-12077 (CUT&RUN-LoV-U data), and E-MTAB-12076 (ATAC-seq data). The Tempo-Seq gene expression data generated in this study are publicly available in ArrayExpress at E-MTAB-13077. All other data generated in this study are available within the article and its Supplementary Data Files or from the corresponding author upon reasonable request.

To test the importance of gut-resident Vγ7⁺ cells in tumor initiation and progression, we crossed Villin-Cre^ERT2;Apc^F/+ (VA) mice with Btnl1^—/— mice, which harbor significantly diminished Vγ7⁺ cell compartments in the SI and colon (5). We confirmed that Btnl1 expression was absent from gut tissue of VA;Btnl1^—/— mice, whereas Btnl1 expression was maintained in VA mice (Fig. 1A). The number of γδ T cells in normal, tumor-adjacent regions was reduced in VA;Btnl1^—/— mice when compared with VA mice (Fig. 1B). Overall survival of tumor-bearing VA and VA;Btnl1^—/— mice was the same, and there was comparable tumor incidence and burden in the SI of tumor-bearing VA and VA;Btnl1^—/— mice (Fig. 1C and D). Conversely, tumor number and particularly tumor burden were increased in the colon of VA;Btnl1^—/— mice when compared with VA mice (Fig. 1D). The lack of phenotype in the SI may be explained by compensation from cytotoxic TCRαβ⁺ IELs and other γδ T-cell subsets (e.g., Vγ1⁺ cells), which partially offset Vγ7⁺ cell deficiencies in Btnl1-deficient mice (5). Because bacterial load is higher in the mouse distal colon than in the SI (38), the propensity for inflammation-driven tumors in this anatomical location may be more sensitive to the lack of Vγ7⁺ cells, which are crucial infection sensors, protectors from pathogens, and suppressors of inflammation (39, 40). However, we cannot rule out any cancer cell–intrinsic effect of Btnl1 deletion in the VA model. In sum, loss of Vγ7⁺ cells was associated with increased tumorigenesis, in agreement with the established role of Vγ7⁺ cells in cancer immunosurveillance (2).

We next asked whether the prevalence of Vγ7⁺ cells in normal gut tissue was maintained in tumors. Contrary to their abundance in gut tissue of WT (Cre-negative) mice, γδ T cells were sparse within adenomas from VA mice, as well as an additional model of colon cancer, Villin-Cre^ERT2;Apc^F/+;Kras^G12D/+ (VAK) mice (Fig. 2A). The cells’ frequency was estimated at 7- to 10-fold lower than in normal tissue (Fig. 2B). Because Vγ7⁺ IELs express CD8αα dimers, whereas most other intestinal γδ T cells do not (5), we used CD8α as a marker to specifically quantify the Vγ7⁺ cell representation in tumor-bearing VA and VAK mice. CD8α⁺ γδ T cells were apparent in the SI of WT and tumor-bearing VA and VAK mice; however, CD8α⁺ γδ T cells were almost absent from tumors in either model (Fig. 2C). These observations showed that tumor-infiltrating Vγ7⁺ cells and other γδ T-cell subsets were rare.

How quickly γδ T cells might be excluded from tumors was investigated using a short-term model, wherein both alleles of Apc were simultaneously deleted in gut tissue, thereby maximally activating β-catenin signaling. The mouse intestine cannot tolerate loss of Apc in this way, so mice are euthanized 3 or 4 days after CRE recombinase induction. γδ T cells were quantified in villi of the SI of Villin-Cre^ERT2;Apc^F/F (VA^F/F) mice and Villin-Cre^ERT2;Apc^F/F;Kras^G12D/+ (VA^F/FK) mice. The number of γδ T cells was reduced by about 3-fold in VA^F/F and VA^F/FK mice when compared with Cre-negative controls (Fig. 2D and E), indicating that deletion of Apc in epithelial cells had a rapid impact on γδ T-cell numbers, before the overt formation of a tumor.

To investigate whether our findings in mice might parallel human colon tumors, we examined samples from three human cohorts that were collected from Scotland, Norway, and Thailand. In all three cohorts, γδ T-cell densities were higher in normal adjacent tissue than tumor tissue (Fig. 2F and G), mirroring our observations in mouse models. Moreover, levels of γδ T cells were higher in the Scottish cohort when compared with the Norway and Thailand cohorts (Fig. 2G). We performed RNA-seq analysis on 82 human colon cancer samples from the Scotland cohort from which immunohistochemical γδ T-cell density data were available to glean information on the subtypes of γδ T cells present in these tumors. TRGV4 transcripts were more abundant than TRGV9 transcripts within the same tumor (Fig. 2H), indicating that Vγ4⁺Vδ1⁺ cells, which reflect colonic IEL, are on aggregate more abundant than Vγ9⁺Vδ2⁺ cells, which are typical of peripheral blood. These data corroborate but substantially extend findings by others (41, 42).

We investigated the expression of Btnl genes, which are essential to the phenotypic maintenance of Vγ7⁺ IEL in the adult gut (4, 5, 7). When tumor sections from VA and VAK mice were stained for Btnl1 mRNA, expression was apparent in epithelial cells surrounding adenomas but was absent from cancer cells (Fig. 3A). The kinetics of this loss of Btnl1 expression were examined in the short-term VA^F/F and VA^F/FK models. In these models, Btnl1 expression was slightly reduced when compared with WT SI (Fig. 3B). To verify this reduction, RNA-seq data from the SI of WT, VA^F/F and VA^F/FK mice were analyzed for Btnl1, Btnl2, Btnl4, and Btnl6 gene expression (14, 15). This analysis showed reduced RNA expression of all four Btnl family members following deletion of Apc in gut tissue (Fig. 3C).

We interrogated two human gene expression datasets, The Cancer Genome Atlas (TCGA) and Skrzypczak (9, 27), to determine whether the expression levels of BTNL3 or BTNL8—homologs of mouse Btnl1 and Btnl6—were different between normal gut and tumor tissues. BTNL3 expression levels were higher in normal tissue than tumor tissue in both the datasets, whereas BTNL8 expression was only higher in normal tissue in the Skrzypczak dataset (Fig. 3D). Together, our analyses demonstrate an evolutionarily conserved reduction of BTNL expression in tumors across species.

The relationship between expressions of BTNL3 and BTNL8 and γδ T-cell infiltration in human tumors was investigated in the Scotland cohort. Gene expression values were plotted with γδ T-cell density values from matched samples. Both BTNL3 and BTNL8 mRNAs were positively correlated with γδ T-cell density, with human tumors exhibiting high expression of BTNL3 and BTNL8 containing more γδ T cells than tumors with low levels of BTNL3 and BTNL8 (Fig. 3E). To more specifically address the relationship between Vγ4⁺Vδ1⁺ IELs, and BTNL3 and BTNL8 levels, we compared TRGV4 and TRDV1 expression levels with BTNL3/8 expression levels. TRGV4 mRNA was positively correlated with both BTNL3 and BTNL8 expression (Fig. 3F). TRDV1 mRNA was also positively correlated with both BTNL3 and BTNL8 expressions, although, TRDV1 mRNA was not detected in 33 of 82 samples (Fig. 3G). These data support the notion that loss of BTNL molecules in tumors is directly associated with the loss of Vγ4⁺Vδ1⁺ IELs in the tumor microenvironment of human tumors.

Prompted by the mouse data, we explored the relationship between WNT signaling, loss of BTNL molecules in cancer cells, and γδ T-cell exclusion in human tumors. Expression levels of CTNNB1 (β-catenin) and SOX9, which is a transcriptional target of the β-catenin transcription factor complex (43), were plotted together with γδ T-cell density values determined by IHC from the same, matched tumor samples from the Scotland cohort. This analysis revealed that higher expression levels of CTNNB1 and SOX9 were correlated with low numbers of γδ T cells in human colon cancer (Fig. 4A). Similarly, TRGV4 mRNA negatively correlated with both CTNNB1 and SOX9 expressions, although this correlation did not reach significance for the SOX9 comparison (Fig. 4A). We did not explore correlations with TRDV1 mRNA owing to the absence of detectable expression levels in many samples. In these human tumors, high CTNNB1 and SOX9 expression levels were correlated with low BTNL3 and BTNL8 expressions (Fig. 4B). To validate these findings, we analyzed the Marisa cohort, a publicly available microarray dataset containing 258 human colon cancer samples (28). Within this dataset, high CTNNB1 and SOX9 expression levels correlated with low BTNL3 expression levels, whereas SOX9 also negatively correlated with BTNL8 expression (Fig. 4C). A correlation between WNT surrogate genes and other butyrophilin family members, namely BTN3A1 and BTN2A1, which bind Vγ9Vδ2⁺ cells, was not observed in the Scotland cohort dataset, except for CTNNB1 and BTN2A1 expression levels (Supplementary Fig. S1A). These data support our hypothesis of a unique relationship between WNT signaling, Vγ4⁺Vδ1⁺ IEL exclusion from tumors, and a loss of BTNL genes in cancer cells.

To explore a mechanistic link between WNT signaling activation and downregulation of Btnl gene expression, we developed an ex vivo transformation assay using intestinal organoids derived from tamoxifen-naïve WT, VA^F/F, VK, and VA^F/FK mice. Cells were treated with tamoxifen in vitro to induce deletion of Apc or expression of mutant KRAS via Cre recombinase. Tamoxifen treatment failed to influence the shape or size of organoids derived from WT mice (Fig. 4D). By contrast, tamoxifen altered the morphology of organoids harboring Apc and Kras^G12D alleles, transforming their normal, budding shape into large spheres typical of tumor-derived organoids (Fig. 4D). Gene expression was measured in these four groups of organoids over the course of 1 week after tamoxifen treatment. We found that WNT pathway target genes, including Lgr5, Sox9, Axin2, and Cd44, were upregulated in VA^F/F organoids, but not in WT and VK organoids. Lgr5 and Cd44 were upregulated in VA^F/FK organoids (Supplementary Fig. S2A). These results show that the organoid system recapitulated cancer cell transformation in vivo by β-catenin signaling. Expression of Btnl1, Btnl2, Btnl4, and Btnl6 mRNA was measured in these four groups of organoids (Fig. 4E). Whereas expression of these genes remained constant in WT organoids, the deletion of Apc resulted by day 4 in reduced expression of all Btnl RNAs assayed. Activation of mutant KRAS (VK) had no effect on Btnl expression. However, the combination of Apc deletion and mutant KRAS expression in VA^F/FK organoids accelerated Btnl downregulation with reduced expression apparent by day 2 (Fig. 4E). These observations demonstrate that β-catenin activation via loss of Apc negatively regulates Btnl gene expression.

As an alternative approach to genetic manipulation of WNT signaling, organoids from WT mice were treated with the GSK3β inhibitor CHIR-99021 to activate β-catenin. Expression of Btnl1, Btnl2, Btnl4, and Btnl6 mRNA was measured after four days of treatment using two different concentrations of CHIR-99021. CHIR-99021 treatment reduced expression of all Btnl mRNAs assayed when compared with controls in a dose-dependent manner (Fig. 4F), thus supporting the hypothesis that activated β-catenin downregulates Btnl gene expression. The reversibility of this effect was tested by treating WT organoids with CHIR-99021 for 4 days, washing off drug, then culturing the treated organoids for another 4 days without drug. On day 8 after treatment began, expressions of Btnl1, Btnl2, Btnl4 and Btnl6 mRNA were measured by qPCR. Withdrawal of CHIR-99021 restored Btnl1, Btnl2, Btnl4, and Btnl6 mRNA expression to baseline or higher levels in these organoids (Fig. 4F).

To understand how WNT signaling negatively affects Btnl gene expression, we investigated how Btnl molecules are regulated in normal tissue. We searched for potential transcription factor–binding sites in the promoter regions of these genes. Using a publicly available database (OregAnno), we generated a list a putative transcription factor–binding sites, then narrowed down the list by focusing on gut-specific transcription factors. This analysis uncovered two sets of paralogs, CDX1 and CDX2, and HNF4A and HNF4G, with multiple binding sites found within 12 kb upstream of mouse and human BTNL gene start sites (Fig. 5A).

We determined whether CDX1, CDX2, HNF4A or HNF4G were localized specifically to the villus where BTNL molecules are expressed. CDX1 was expressed in crypt regions and lower villus, but expression decreased as enterocytes moved up the villus (Fig. 5B). CDX2 was expressed in both crypts and villi with higher expression in the crypt. HNF4A was also expressed in both crypts and villi; staining was also observed in cells residing within the lamina propria. HNF4G expression was specific to enterocytes in the villus, as no expression was observed in crypt regions (Fig. 5B). These data suggested HNF4G as the prime candidate for Btnl gene regulation, given their overlapping patterns of expression in the villus. However, all four transcription factors are expressed in the villus to some extent.

We investigated whether CDX1 and CDX2 mediated Btnl1, Btnl2, Btnl4, and Btnl6 transcription. Organoids from WT mice were transduced with 5 shRNA constructs targeting Cdx1 or Cdx2 mRNA. Two constructs achieved good knockdown efficiency for Cdx1, although organoid morphology and expression of Btnl molecules remained unchanged (Supplementary Fig. S3A and S3B). Attempts to knockdown Cdx2 proved difficult as organoids transduced with these constructs often died. In two replicate experiments where organoids survived antibiotic selection, knockdown of Cdx2 was sufficiently achieved with the shRNA_Cdx2–2 construct, but this failed to impact on organoid morphology or Btnl gene expression (Supplementary Fig. S3C and S3D). These data suggested that CDX2 was required for organoid survival. Indeed, conditional deletion of Cdx2 in adult intestine is lethal (44). We concluded from these experiments that CDX1 and CDX2 were not required specifically for Btnl1, Btnl2, Btnl4, and Btnl6 transcription.

HNF4A and HNF4G are paralogs that bind fatty acids and whose functions are somewhat redundant (23, 45, 46). These transcription factors recognize a nearly identical consensus motif on DNA, and they exhibit 98.7% commonality in DNA-binding profiles (23). It should be noted that HNF4A is expressed outside the gut at sites such as liver (47), where BTNL molecules are not expressed (4). To determine whether HNF4A and HNF4G bound the promoter region of Btnl1, Btnl2, Btnl4, and Btnl6 genes, we analyzed chromosome 17 in a HNF4 ChIP-seq dataset from mouse SI (23). This analysis confirmed that HNF4A/G bound all Btnl gene promoter regions (Fig. 5C).

We investigated whether HNF4A and HNF4G activity was causally linked to Btnl expression. Organoids from WT mice were transduced with 5 shRNA constructs targeting Hnf4a or Hnf4g mRNA. Organoid morphology was unaffected by Hnf4a constructs (Supplementary Fig. S3E). Knockdown of Hnf4a was not successful. Instead of reduced expression, we observed higher expressions of Hnf4a and Hnf4g in these cells, concomitant with higher expression of Btnl1, Btnl2, Btnl4, and Btnl6 genes (Supplementary Fig. S3F). These findings suggested that a feedback mechanism may be active, preventing Hnf4a knockdown, but provided indirect evidence that increased HNF4A and HNF4G expressions correlated with increased Btnl expression. To clarify this situation, we transduced MODE-K enterocytes that do not express Hnf4a with a series of cDNAs encoding gut-associated transcription factors, including Cdx1, Cdx2, and Hnf4a. Only in Hnf4-transduced cells was there overt upregulation of Btnl mRNAs, specifically those for Btnl4 and Btnl6 (Supplementary Fig. S3G), whereas Cdx1, Cdx2, Creb3l3, Gata5, and Isx failed to influence Btnl gene expression. For Hnf4g targeting in organoids from WT mice, two constructs achieved good knockdown efficiency with the shRNA_Hnf4g-5 construct exhibiting superior efficiency. However, only the shRNA_Hnf4g-5 construct reduced expression of Btnl1, Btnl2, and Btnl6 without affecting expression of Btnl4 (Fig. 5D). This was accompanied by the occasional appearance of sphere-shaped organoids (Supplementary Fig. S3H), as observed in other reports (23) and indicative of a stem cell–like state. Collectively, our data demonstrate that HNF4 transcription factors are regulators of Btnl gene expression, although there are seemingly differences in the degrees to which specific Btnl genes are dependent upon or influenced by HNF4A and HNF4G, respectively. These data further integrate Btnl expression with physiological enterocyte differentiation (23).

We analyzed Btnl gene expression in mouse models deficient for HNF4A or HNF4G or both. An RNA-seq dataset derived from intestine of WT, Villin-Cre^ERT2;Hnf4a^F/F (Hnf4a^Δ/Δ) mice, Hnf4g^—/— (Hnf4g^Δ/Δ) mice and Hnf4a^Δ/Δ;Hnf4g^Δ/Δ mice was used for this purpose (23). In these mice, deletion of Hnf4a failed to alter Btnl gene expression, whereas deletion of Hnf4 g reduced expression of all four Btnl genes (Fig. 5E). Simultaneous deletion of Hnf4a and Hnf4 g led to the most pronounced loss of Btnl expression when compared with WT tissue. Btnl1, Btnl2, and Btnl4 (but not Btnl6) mRNA was also lower in Hnf4a^Δ/Δ;Hnf4g^Δ/Δ intestine than in Hnf4a^Δ/Δ or Hnf4g^Δ/Δ intestine (Fig. 5E). We then used an inhibitor that targets both HNF4A and HNF4G, BI-6015 (HNF4i), in organoids from WT mice. This drug altered the morphology of the organoids, transforming them into spheres (Fig. 5F), similar to the morphology of Hnf4a^Δ/Δ;Hnf4g^Δ/Δ organoids previously described (23). Inhibition of these transcription factors by BI-6015 reduced expression of Hnf4a and Hnf4g, as well as Btnl1, Btnl2, Btnl4, and Btnl6 mRNA (Fig. 5F). Together, these data demonstrate that HNF4G is the main regulator of Btnl molecule expression with cooperation from HNF4A in enterocytes.

The relationship between transcription factor expression, γδ T-cell infiltration and BTNL gene expression was examined in human tumors. In the Scotland cohort, there was no correlation between CDX1, CDX2, and HNF4A expressions and γδ T-cell density (Supplementary Fig. S4A–S4C). There was a positive correlation between CDX1, BTNL3, and BTNL8 expressions in both the Scotland and Marisa cohorts (Supplementary Fig. S4D), but correlations between CDX2 or HNF4A and BTNL3 or BTNL8 were absent or inconsistent among both cohorts (Supplementary Fig. S4E and S4F). In contrast with the other transcription factors, increased HNF4G expression was correlated with higher γδ T-cell density in human tumors (Fig. 5G). Increased HNF4G expression also correlated with increased BTNL3 expression in both the Scotland and Marisa cohorts, whereas the relationship with BTNL8 expression was only observed in the Scotland cohort (Fig. 5H). These data establish an association between HNF4G, BTNL expressions, and tumor-infiltrating γδ T cells in human tumors and point to HNF4G regulation of BTNL gene expression as being conserved across species.

We hypothesized that disruption of the WNT gradient in the mouse intestine would interfere with enterocyte-specific HNF4G and Btnl gene expression and subsequently γδIEL abundance. To test this hypothesis, we used Vil1-Grem1 mice in which Gremlin1 is under the control of the Vil1 promoter. These mice develop ectopic crypts in the villi due to the antagonist actions of GREM1 on bone morphogenic proteins, a consequence of which is increased WNT signaling in the villi (13). Nuclear SOX9 was used to identify ectopic crypts in the villi of Vil1-Grem1 mice (Supplementary Fig. S5A). These SOX9-high, WNT-high ectopic crypts maintained HNF4A as in normal crypts, but lost expressions of HNF4G and Btnl1 mRNA (Supplementary Fig. S5A). Moreover, we quantified γδ T cells in villi of Vil1-Grem1 mice and found that these cells were reduced when compared with WT mice (Supplementary Fig. S5A and S5B). These results show that WNT signaling suppresses the HNF4G–Btnl–Vγ7⁺ cell axis.

Further testing of our hypothesis was carried out in an additional model that is WNT-ligand dependent in which R-spondin 3 (RSPO3) is expressed from LGR5⁺ stem cells: Lgr5-Cre^ERT2;Rspo3^INV mice (12). In this model, increased WNT signaling induces greater numbers of crypt regions, as demonstrated by increased SOX9⁺ cells at the base of the intestine, and reduced villus length (Supplementary Fig. S5C). We investigated HNF4A, HNF4G, and Btnl1 expressions in these mice. Staining patterns of these molecules were consistent with expression in intestine from WT mice where HNF4A was expressed in crypt regions and enterocytes, whereas HNF4G and Btnl1 expressions were restricted to enterocytes (Supplementary Fig. S5C). However, the expansion of WNT-high crypt regions and reduced villus length resulted in fewer γδ T cells in the villi of these in Lgr5-Cre^ERT2;Rspo3^INV mice, when compared with Cre-negative mice (Supplementary Fig. S5C and S5D). To determine whether reduced γδ T-cell numbers could be restored by interference with WNT signaling, Lgr5-Cre^ERT2;Rspo3^INV mice were treated with the porcupine inhibitor (PCPNi) LGK-974 to block the secretion of WNT ligands and prevent its activation of β-catenin. Expression patterns of HNF4A in crypt and villi regions as well as HNF4G and Btnl1 mRNA in enterocytes were unaltered by LGK-974 treatment. However, γδ T-cell numbers in the villi of these mice increased (Supplementary Fig. S5C and S5D). These data provide evidence that aberrant β-catenin activation in normal intestinal tissue disrupts γδIEL abundance.

We compared HNF4A and HNF4G expressions between normal human colon and tumor tissues in the TCGA and Skrzypczak datasets. Both HNF4A and HNF4G were reduced in tumor tissue from both datasets (Fig. 6A), mirroring reduced BTNL3 and BTNL8 expressions in human tumors (Fig. 3D).

We investigated whether expressions of Hnf4a and Hnf4 g mRNAs were affected by WNT signaling by examining mRNA levels in the SI of WT, VA^F/F and VA^F/FK mice. This analysis showed that Hnf4a levels were similar between normal Sl and Apc-deficient SI tissues, whereas Hnf4 g levels were reduced in Apc-deficient tissue (Fig. 6B). IHC on these models revealed that nuclear staining of both HNF4A and HNF4G was reduced or even absent from epithelial cells in the villus of VA^F/F and VA^F/FK tissue when compared with WT tissue (Fig. 6C). The addition of mutant KRAS to Apc loss had no influence over decreased expression of HNF4A and HNF4G. The discrepancy between Hnf4a mRNA and HNF4A protein levels may be explained by expression of HNF4A⁺ stromal cells in the lamina propria. These data show that expressions of HNF4A and HNF4G are rapidly reduced or lost completely in cells with high β-catenin activity.

End-stage tumors from VA mice were evaluated for the presence of HNF4A and HNF4G. Nuclear SOX9 staining was used to identify WNT-high tumors. We found that HNF4A and HNF4G were completely absent from cancer cells, whereas normal adjacent epithelial cells maintained nuclear HNF4A and HNF4G staining (Fig. 6D). This pattern of expression mimicked loss of Btnl1 staining in tumors from the same mouse model (Fig. 3A).

We used the organoid transformation assay to test the kinetics of Hnf4a and Hnf4g downregulation after β-catenin activation. After tamoxifen treatment, expression of these molecules remained constant in WT organoids (Fig. 6E). The deletion of Apc resulted in reduced expression of Hnf4a and Hnf4g by day 2, which was 2 days earlier than was observed for Btnl mRNA downregulation, as shown in Fig. 4E. Induction of oncogenic KRAS had no effect on Hnf4a and Hnf4g gene expressions, but the combination of Apc deletion and mutant KRAS expression in VA^F/FK organoids resulted in a downregulation of Hnf4a and Hnf4g by day 1 or 2 (Fig. 6E). These observations indicated that suppression of Hnf4a and Hnf4g RNAs by β-catenin preceded the downregulation of Btnl gene expression. Treatment of WT organoids with the GSK3β inhibitor CHIR-99021 reduced expression of Hnf4a and Hnf4g (Fig. 6F). As observed with Btnl1, Btnl2, Btnl4, and Btnl6 expressions (Fig. 4F), the inhibition of Hnf4a and Hnf4g mRNA was reversible after withdrawal of CHIR-99021 with expression levels returning to normal on day 8 (Fig. 6F).

Analysis of CUT&RUN datasets for the WNT/β-catenin mediator LEF1 in human HEK293T cells stimulated with 10 μmol/L CHIR-99021 (34) revealed a LEF1-binding site downstream of the HNF4G locus (orange track in Supplementary Fig. S6A). Using an ATAC-seq dataset (35), we found that this LEF1-bound genomic region had an open chromatin signal in response to CHIR-99021 treatment with high evolutionary conservation (Supplementary Fig. S6B), pointing to an active regulatory site. The presence of JASPAR-predicted LEF1 consensus sequences, which match the binding profile, further supported the notion that this genomic locus is a bona fide WNT Responsive Element (WRE; Supplementary Fig. S6C). Finally, GREAT unbiasedly annotated this WRE to the HNF4G promoter (Supplementary Fig. S6D), suggesting its direct regulation of HNF4G. By contrast, no obvious signal was detected for LEF1 binding in proximity to the HNF4A locus (Supplementary Fig. S6E).

Together, these data are consistent with the notion that β-catenin suppresses Btnl1/2/4/6 gene expression, via downregulation of HNF4A and HNF4G.

We investigated whether introduction of the BTNL1/BTNL6 heterodimer in BTNL1/BTNL6-deficient cells was sufficient to enhance Vγ7⁺ cell–mediated cancer cell death or to induce Vγ7⁺ cell infiltration into tumors and slow tumor growth. We cloned Btnl1 and Btnl6 into doxycycline-inducible vectors and transduced CT26 cells with them as CT26 cells naturally lack expression of Btnl1 and Btnl6 mRNA (Supplementary Fig. S7A). We confirmed increased expression of Btnl1 and Btnl6 mRNA by doxycycline in these cells (CT26-B1/6 cells) grown in vitro (Supplementary Fig. S7A). We tested whether the expression of BTNL1 and BTNL6 modulated the growth pattern of these cells. Upregulation of Btnl1 and Btnl6 mRNA failed to influence the proliferation of CT26 cells in vitro (Supplementary Fig. S7B). These cells were transplanted into the flank of syngeneic mice, but the growth of tumors remained unchanged when compared with control CT26 tumors despite achieving a high induction of Btnl1 and Btnl6 mRNA in CT26-B1/6 tumors (Supplementary Fig. S7C and S7D). These data indicate that ectopic expression of the BTNL1/BTNL6 heterodimer had no impact on cancer cell growth.

Vγ7⁺ cells were cocultured with CT26 vector control cells or CT26-B1/6 cells. We observed increased viability and expression of the activation marker CD25 by Vγ7⁺ cells when cocultured with CT26-B1/6 cells as compared with control cells (Fig. 7A and B; Supplementary Fig. S7E and S7F). These data showing increased activation of γδIELs after Vγ7-BTNL1/BTNL6 interaction accord well with previous studies (5, 7). We measured death of CT26 cells in these cocultures. We found that Vγ7⁺ cells increased CT26 cell death by approximately 5-fold; however, engagement with the BTNL1/BTNL6 heterodimer had no impact on cancer cell killing by Vγ7⁺ cells (Fig. 7C; Supplementary Fig. S7G). Therefore, the ability of Vγ7⁺ cells to induce cancer cell death is independent of the BTNL1/BTNL6 heterodimer, although the BTNL1/BTNL6 heterodimer can support Vγ7⁺ cell survival and activation.

We transplanted control and CT26-B1/6 cells into the colonic submucosa of syngeneic mice by colonoscopy-guided injection and treated tumor-bearing mice with doxycycline. We confirmed increased expression of Btnl1 mRNA by doxycycline in these tumors (Supplementary Fig. S7H). We found no difference in survival of tumor-bearing mice when Btnl1 and Btnl6 mRNA were induced (Fig. 7D). Moreover, the number of CD3⁺ T, CD8⁺ T, and γδ T was the same between groups (Fig. 7E; Supplementary Fig. S7I–S7K). Therefore, we concluded that the BTNL1/BTNL6 heterodimer may support Vγ7⁺ cell survival in tumors, but BTNL1/BTNL6 are not sufficient to induce Vγ7⁺ cell infiltration into the tumor microenvironment (or even other T cells), suggesting that other molecules are necessary to attract Vγ7⁺ cells into tumors.

Because ectopic expression of BTNL1/BTNL6 was insufficient to mediate recruitment of γδ T cells, we asked whether inhibition of β-catenin signaling could reverse γδ T-cell exclusion from tumors. The β-catenin transcription factor complex consists of several components, including B-cell lymphoma 9 (BCL9) and BCL9-like (BCL9L; refs. 48, 49), whose deletion in mouse tumor models abrogates β-catenin–mediated transcription (14, 50). Therefore, we investigated expression levels of Hnf4a, Hnf4g, Btnl1, Btnl2, Btnl4, and Btnl6 in tissue where Apc is deleted and Bcl9 and Bcl9L are absent. For this purpose, we examined RNA-seq data from intestinal tissue of VA^F/F mice and VA^F/F;Bcl9^F/F;Bcl9l^F/F mice that were treated with tamoxifen for 4 days to induce Cre recombinase. This analysis showed that Hnf4g, Btnl2, and Btnl4 levels were higher in VA^F/F;Bcl9^F/F;Bcl9l^F/F intestinal tissue, whereas Hnf4a and Btnl1 mRNA remained unchanged (Supplementary Fig. S7L). Btnl6 could not be detected in this dataset.

End-stage tumors from VA and VA;Bcl9^F/F;Bcl9l^F/F mice were assessed for expression of HNF4A, HNF4G and Btnl1 mRNA. SOX9 was used to detect WNT-high cancer cells. Expressions of HNF4A, HNF4G, and Btnl1 mRNA were absent from tumors in VA mice (Fig. 7F). By contrast, nuclear expressions of HNF4A and HNF4G as well as Btnl1 mRNA were apparent in some but not all areas of tumors from VA;Bcl9^F/F;Bcl9l^F/F mice (Fig. 7F). Previous reports indicate that recombination of Bcl9^F/F;Bcl9l^F/F alleles is inefficient in these mice (14), which provides an explanation for the sporadic expression pattern of HNF4A, HNF4G, and Btnl1 mRNA in these tumors. To determine whether the restoration of HNF4A, HNF4G, and Btnl1 expression in tumors from VA;Bcl9^F/F;Bcl9l^F/F mice affected tumor-infiltrating γδ T cells, we quantified these cells in tumor tissue. This analysis showed that γδ T cells were more abundant in tumors from VA;Bcl9^F/F;Bcl9l^F/F mice than VA mice (Fig. 7G).

Another colon cancer model, Villin1-Cre^ERT2;Ctnnb1^ex3/+ (V;Ctnnb1^ex3/+) mice, was used to validate these findings, where mutant β-catenin is used to drive tumor formation (14). Nuclear SOX9 expression was used to identify WNT-high cancer cells. In this model, nuclear HNF4A expression was evident in cancer cells (albeit weak expression), whereas HNF4G and Btnl1 expressions were lost from tumors (Fig. 7H). V;Ctnnb1^ex3/+ mice were crossed with Bcl9^F/F;Bcl9l^F/F mice. Tumors from V;Ctnnb1^ex3/+;Bcl9^F/F;Bcl9l^F/F mice retained expression of HNF4A. Nuclear HNF4G and Btnl1 mRNA expressions could be observed in overlapping regions of tumors (Fig. 7H), although staining was sporadic as in tumors from VA;Bcl9^F/F;Bcl9l^F/F mice (Fig. 7F). We quantified tumor-infiltrating γδ T cells in these mice and found that tumor-infiltrating γδ T cells are more abundant in tumors from V;Ctnnb1^ex3/+;Bcl9^F/F;Bcl9l^F/F mice than V;Ctnnb1^ex3/+ mice (Fig. 7I). These data indicate that inhibition of β-catenin signaling reverses suppression of HNF4A/HNF4G-driven Btnl gene expression and exclusion of Vγ7⁺ cells in tumors.

Our study shows how BTNL molecule expression is lost in mutated epithelial cells by dysfunctional WNT signaling. Our data in Btnl1-deficient mice suggest that cancer cells thrive in the absence of Vγ7⁺ cells, in line with recent evidence from several groups reporting on the pivotal role of human and mouse γδIELs in cancer immunosurveillance (2, 42, 51). In addition to direct cancer cell killing, γδIELs may also suppress protumorigenic inflammation, such as occurs during intestinal nutrient sensing (40), to impede cancer progression. We found that normal IECs use HNF4G (most likely dimerized with HNF4A) to induce expression of Btnl gene expression. These data are supported by a recent study focused on HNF4A-regulated gene expression throughout gut tissue (52), where HNF4A transcriptional activity seems to be more important in the mouse large intestine than small intestine. Although these paralogs bind the same consensus sites in promoter regions and exhibit redundant functions (23), our data suggest that HNF4G is the dominant regulator of Btnl molecules. Indeed, HNF4G expression, not HNF4A, correlated with BTNL3/BTNL8 expression and γδ T-cell infiltration into tumors in our human colon cancer dataset.

The biological basis for evasion from γδIEL immunosurveillance shown herein revolves around WNT-driven dedifferentiation of cancer cells toward a stem cell–like state. Dysregulated WNT signaling fosters the conversion of cancer cells toward a less differentiated phenotype reminiscent of LGR5⁺ stem cells that reside in intestinal crypts. LGR5⁺ stem cells, like cancer cells, fail to express HNF4G and BTNL molecules (5), making crypt regions and tumors immune privileged sites, devoid of γδIELs. Redifferentiation of colon cancer cells could reengage γδIEL immunosurveillance, and strategies to achieve redifferentiation could benefit γδIEL-based cancer immunotherapies. These approaches could be combined with anti–PD-1 therapy to boost γδIEL killing activity, particularly in patients lacking HLA expression (51). At the same time, redifferentiation would slow the proliferative signals induced by β-catenin in cancer cells. Our data suggest that redifferentiation may be possible given that inhibition of β-catenin transcriptional activity by deletion of BCL9 and BCL9 L results in reexpression of HNF4G and BTNL molecules and increased numbers of γδ T cells in tumors. This notion is supported by data from other disease settings. Individuals with celiac disease exhibit a loss of BTNL8 expression concomitant with a loss of Vγ4⁺Vδ1⁺ IELs, but elimination of dietary gluten can restore BTNL8 expression (53). Together, our two studies emphasize the reversibility of BTNL gene expression in different disease contexts. However, expression of other molecules, cytokines, and chemokines in addition to BTNLs will likely be necessary to attract γδIELs into the tumor microenvironment, as our data show that BTNLs only support survival and activation of γδIELs. Moreover, strategies to restore γδIEL immunosurveillance will be anatomical site–specific. BTNL-responsive mouse Vγ7⁺ cells and human Vγ4⁺ cells are restricted to the gut, so reengagement of endogenous γδIELs will not be possible for liver metastasis.

T. Suzuki reports grants from Naito Foundation and European Union during the conduct of the study. P. Vantourout reports a patent for WO2019053272A1 pending. C. Kersten reports personal fees from Astra Zeneca outside the submitted work. A. Hayday reports grants from Cancer Research UK core funding at the Francis Crick Institute (FC001093) during the conduct of the study; grants and personal fees from Gamma Delta Therapeutics, grants from Takeda Pharmaceuticals, grants and personal fees from Adaptate Biotherapeutics, personal fees from Prokarium, and grants and personal fees from ImmunoQure, AG outside the submitted work; as well as reports a patent for IP501/2117 EP (GB) pending, issued, licensed, and with royalties paid from Takeda, a patent for IP501/2488 US pending and with royalties paid from Takeda, a patent for IP501/2552 US3 pending and with royalties paid from Takeda, a patent for IP501/2616 US2 pending to Filed by Takeda, and a patent for IP501/3062 US4 pending assigned to Marengo. O.J. Sansom reports grants from Astra Zeneca, Boehringher Ingelheim, Cancer Research Horizons, and Novartis outside the submitted work. S.B. Coffelt reports grants from Wellcome Trust, Cancer Research UK, European Union, grants from Naito Foundation, and UK Medical Research Council during the conduct of the study. No disclosures were reported by the other authors.

T. Suzuki: Conceptualization, formal analysis, supervision, funding acquisition, investigation, visualization, methodology, writing–original draft, writing–review and editing. A. Kilbey: Conceptualization, data curation, formal analysis, supervision, investigation, visualization, methodology, writing–original draft, writing–review and editing. N. Casa Rodriguez: Formal analysis, investigation, writing–review and editing. A. Lawlor: Formal analysis, investigation, methodology. A. Georgakopoulou: Formal analysis, investigation, writing–review and editing. H. Hayman: Formal analysis, investigation, writing–original draft, writing–review and editing. K.L. Yin Swe: Formal analysis, investigation, methodology. A. Nordin: Formal analysis, investigation, methodology, writing–review and editing. C. Cantù: Resources, data curation, formal analysis, supervision, funding acquisition, investigation, writing–original draft, writing–review and editing. P. Vantourout: Formal analysis, investigation, methodology, writing–review and editing. R.A. Ridgway: Resources, data curation, supervision, writing–review and editing. R.M. Byrne: Formal analysis, investigation, visualization, methodology, writing–review and editing. L. Chen: Resources, formal analysis, investigation, writing–review and editing. M.P. Verzi: Resources, data curation, formal analysis, supervision, funding acquisition, investigation, visualization, methodology, writing–original draft, writing–review and editing. D.M. Gay: Resources, writing–review and editing. E. Gil Vazquez: Resources, writing–review and editing. H.L. Belnoue-Davis: Resources, writing–original draft, writing–review and editing. K. Gilroy: Formal analysis, investigation, visualization, methodology, writing–review and editing. A.H. Kostner: Resources, data curation, writing–review and editing. C. Kersten: Resources, data curation, supervision, methodology, writing–review and editing. C. Thuwajit: Resources, data curation, methodology, writing–review and editing. D.K. Andersen: Formal analysis, investigation, methodology, writing–review and editing. R. Wiesheu: Formal analysis, investigation, writing–review and editing. A. Jandke: Conceptualization, investigation, writing–review and editing. K. Blyth: Resources, supervision, writing–review and editing. A.K. Roseweir: Data curation, writing–review and editing. S.J. Leedham: Resources, supervision, funding acquisition, methodology, writing–original draft, writing–review and editing. P.D. Dunne: Formal analysis, supervision, funding acquisition, investigation, writing–review and editing. J. Edwards: Resources, data curation, supervision, funding acquisition, methodology, writing–original draft, writing–review and editing. A. Hayday: Conceptualization, resources, supervision, funding acquisition, writing–original draft, writing–review and editing. O.J. Sansom: Data curation, supervision, funding acquisition, writing–original draft, writing–review and editing. S.B. Coffelt: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

We are grateful to Daniel Murphy (University of Glasgow/CRUK Beatson Institute), Laura Machesky (University of Glasgow/CRUK Beatson Institute), Karen Edelblum (Rutgers University), and Catherine Winchester (CRUK Beatson Institute) for advice. We thank David Bryant (University of Glasgow/CRUK Beatson Institute) for lentiviral reagents and Natalie Church (Francis Crick Institute) for technical assistance. We thank the Core Services and Advanced Technologies at the Cancer Research UK Beatson Institute, with particular thanks to the Histology Core Facility and Biological Services Unit. This work was supported by grants from Wellcome Trust (Seed Award 208990/Z/17/Z; to S.B. Coffelt and Senior Clinical Research Fellowship 206314/Z/17/Z; to S.J. Leedham); Cancer Research UK Glasgow Center (A25142; to S.B. Coffelt); Marie Skłodowska Curie Actions European Fellowship (GDCOLCA 800112; to T. Suzuki); Naito Foundation Grant for Research Abroad (to T. Suzuki); Medical Research Council (MR/R502327/1; to S.B. Coffelt and J. Edwards and MR/R502327/1; to J. Edwards); Greater Glasgow and Clyde endowment (306620–01; to J. Edwards); Cancer Research UK (Early Detection Project Grant A29834; to P.D. Dunne and Career Establishment Award RCCCEA-Nov21\100003; to S.B. Coffelt); Cancerfonden (CAN 2018/542 and 21 1572 PJ), the Swedish Research Council, Vetenskapsrådet (2021–03075), Linköping University, and Knut and Alice Wallenberg Foundation (Wallenberg Molecular Medicine fellowship award; to C. Cantù); and the NIH (R01DK121915 and R01CA190558; to M.P. Verzi). A. Hayday was supported by Cancer Research UK core funding at the Francis Crick Institute (FC001093). O.J. Sansom and K. Blyth were supported by Cancer Research UK core funding at the Cancer Research UK Beatson Institute (A17196 and A31287).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.