MAPK-driven epithelial cell plasticity drives colorectal cancer therapeutic resistance

Abstract

The colorectal epithelium is rapidly renewing, with remarkable capacity to regenerate following injury. In colorectal cancer (CRC), this regenerative capacity can be co-opted to drive epithelial plasticity. Although oncogenic MAPK signalling in CRC is common, with frequent mutations of both KRAS (40-50%) and BRAF (10%)¹, inhibition of this pathway typically drives resistance clinically. Here, given the development of KRAS inhibitors and licensing of BRAF inhibitor combinations^2-4, we have interrogated key mechanisms of resistance to these agents in advanced preclinical CRC models. We show that oncogenic MAPK signalling induces epithelial-state changes in vivo, driving adoption of a regenerative/revival stem-like population, whereas inhibition leads to rapid transcriptional remodelling of both Kras-mutant and Braf-mutant tumours, favouring a WNT-associated, canonical stem phenotype. This drives acute therapeutic resistance in Kras-driven and delayed resistance in Braf-driven models. Where plasticity is restrained, such as in early metastatic disease, or through targeting ligand-dependent WNT pathway Rnf43 mutations, marked therapeutic responses are observed. This explains the super response to BRAF + EGFR-targeted therapies previously observed in a BRAF-RNF43 co-mutant patient population, highlighting the criticality of cellular plasticity in therapeutic response. Together, our data provide clear insight into the mechanisms underpinning resistance to MAPK-targeted therapies in CRC. Moreover, strategies that aim to corral stem cell fate, restrict epithelial plasticity or intervene when tumours lack heterogeneity may improve therapeutic efficacy of these agents.

Fig. 1. KRAS activation drives an RSC phenotype.

a, Schema representing CRC cell states, associated key markers and driver mutations. Cells can adopt a CBC state, a RSC state or exist on a continuum between these states. The SCI is a composite measure of these, describing the phenotypic continuum between states. Data adapted from ref. . The diagram was created in BioRender. White, M. (2025) https://BioRender.com/3s904bg. b, Representative images of Anxa1, Ly6a (RSC markers), Lgr5, Axin2 (CBC markers) fluorescence in situ hybridization (FISH) and pERK IHC in colons of VillinCre^ERT2 (WT), KG12D and KG12C. The representative images of mouse colonic tissue were sampled 30 days post-intraperitoneal tamoxifen induction (DPI) (n = 5 for WT, n = 3 for KG12D and n = 6 for KG12C). c, Time course of APC and AKG12D following intracolonic tamoxifen injection 4, 7, 14 and 21 DPI. Representative images of serial sections of β-catenin and BrdU IHC and with Anxa1 (red) and Lgr5 (green) FISH. For BrdU IHC, n = 3 (days 14 and 21) and n = 4 (days 4 and 7) for APC, n = 3 (days 7, 14 and 21) and n = 4 (day 4) for AKG12D. For β-catenin IHC and Anxa1/Lgr5 FISH analyses, n = 3 (day 21) and n = 4 (days 4, 7 and 14) for APC, n = 4 (days 4 and 7) and n = 3 (days 14 and 21) for AKG12D. d,e, Proportion of Anxa1⁺DAPI⁺ cells (d) and Anxa1⁺:Lgr5⁺ ratio (e) in the time course of APC and AKG12D at 4, 7, 14 and 21 DPI. Significance was determined by two-tailed Student’s t-tests, no multiple comparison correction. Data are mean ± s.e.m. f, Scaled heatmap of normalized gene expression associated with the CBC or RSC phenotype by RNA-seq of intracolonic tumours from APC (n = 13) and AKG12D (n = 9) mice. Statistical testing was performed by two-sided Wald test; the gene symbols in red have a Benjamini–Hochberg adjusted P < 0.05. g, SCI determined from RNA-seq of intracolonic tumours from APC (n = 13) or AKG12D (n = 9) mice. Boxes are median and interquartile range (IQR), and whiskers extend to the minimum and maximum values reaching up to 1.5× the lower and upper IQR. Significance was determined using a two-tailed Student’s t-test.

Fig. 2. KRAS inhibition shifts epithelial phenotype.

a, Uniform manifold approximation and projection (UMAP) visualization of epithelial cells and major cell-fate clusters with density overlay mapping from scRNA-seq of colonic tumours from APC (2,442 cells, n = 4), AKG12D (4,439 cells, n = 3), AKG12D mice + vehicle (5,429 cells, n = 3) and AKG12D mice + MRTX1133 (5,823 cells, n = 5). b, Alluvial plot showing the proportion of cell states across experimental conditions in panel a. c, Schema summary of key epithelial cell states in CRC derived from panel a based on oncogenic signalling pathways and key marker genes for each state. The x axis represents increasing expression of genes relating to MAPK activation and regenerative programmes, and the y axis represents expression of WNT pathway targets. The schematic was created in BioRender. White, M. (2025) https://BioRender.com/lxn2xfr. d, Representative images of spatial distribution annotated cell states in Xenium in situ datasets for APC or AKG12D tumours treated with vehicle or MRTX1133 for 4 days (n = 3 tumours per condition). Non-tumour epithelial cells are in grey. e, Alluvial plot showing the proportion of cell states across experimental conditions from spatial transcriptomic data of panel d. f, PCA of cell-type probability from spatial transcriptomic data of panel d for the six tumour cell populations generated from scRNA-seq. PC1 describes 49.8% of variance and the top three loadings are high proliferative, WntMapk^hi and Wnt^hi. PC2 describes 36.4% of variance and the top three loadings are Wnt^hi, high proliferative and Mapk^hi. The cell colour denotes the highest probability cell-type call. g, Histogram of gene set scores for canonical MAPK target genes, WntMapk^hi and Wnt^hi in tumour epithelial cells of spatial transcriptomic data of panel d. The x axis is gene score, and the y axis is cell density (total cells per condition normalized to an area under the curve of 1). Blue denotes APC (76,475 cells), yellow indicates AKG12D + vehicle (153,013 cells) and green shows AKG12D + MRTX1133 (150,894 cells). h, Tumour epithelial cells projected on the PCA space from panel f. Blue denotes APC, yellow indicates AKG12D + vehicle and green shows AKG12D + MRTX1133.

Fig. 3. Early CRC liver metastases are WntMapk^hi enriched and vulnerable to MAPK inhibition.

a, UMAP of scRNA-seq and cell-fate clusters of the tumour-derived spheroid lines AKPT small intestinal tumour (SIT) line (6,099 cells, n = 3) and KPN liver metastasis (LMet) line (1,501 cells, n = 3). b, Alluvial plot showing the proportion of cell fates from spheroids in panel a. c, Representative Anxa1 and Lgr5 in situ hybridization stains of AKPT metastases sampled 4, 7, 14 and 28 days post-intrasplenic engraftment. n = 3 per time point. d, Representative spatial profiling image of AKPT-derived liver metastasis 28 days post-intrasplenic transplantation (n = 3). Tumour epithelial cells are coloured by the highest probability cell-type call; non-tumour cells are in grey. e, Composite image of cells by scaled Axin2 (cyan) and Anxa1 (magenta) expression in AKPT-derived liver metastasis. f, PCA of cell-type probability in tumour epithelial cells. PC1 (x axis) describes 44.8% of variance; the top three loadings are Wnt^hi, WntMapk^hi and Mapk^hi. PC2 (y axis) describes 37.2% of variance; the top three loadings are high proliferative, Wnt^hi and WntMapk^hi. Cells are coloured by their highest probability cell-type call, KPN intrasplenic (n = 4, blue) and AKPT intrasplenic (n = 3, orange). g, Liver metastatic burden in AKPT treated from 7 days post-transplantation and continued for 21 days. AZD6244 versus vehicle (n = 7 per group, one-tailed Mann–Whitney test); MRTX1133 versus vehicle (n = 6 per group, one-tailed Mann–Whitney test); and RMC9805 versus RMC6236 versus combined RMC9805–RMC6236 versus vehicle (n = 6 per group except n = 5 in vehicle, Kruskal–Wallis test) are shown. Data are mean ± s.e.m. h, Liver metastatic burden in KPN or KcPN treated from 7 days post-transplantation for 21 days (KPN AZD6244), 28 days (KPN MRTX1133) and 35 days (KcPN AZD4625). AZD6244 versus vehicle (n = 5 per group, one-tailed Mann–Whitney test); MRTX1133 versus vehicle (n = 5 per group, one-tailed Mann–Whitney test); and AZD4625 versus vehicle (n = 8 per group, two-tailed Mann–Whitney test) are shown. Data are mean ± s.e.m. i, Percentage of KPN with liver metastasis when treated from 85 DPI with AZD6244 (n = 6) or vehicle (n = 9; n (%), two-sided Fisher’s exact test).

Fig. 4. BRAF + EGFR inhibition causes adoption of a Wnt^hi epithelial state in Braf-mutant tumours.

a, Boxplot of SCI in BRAF-driven colonic tumours compared with APC tumours. Scores are derived from RNA-seq of APC (n = 13) BA (n = 4), VillinCre^ERT2 Braf^V600E/+ Trp53^fl/fl Alk5^fl/fl (BPA, n = 10) and BPNA (n = 12) mice. Boxes are median and IQR, and whiskers extend to the minimum and maximum values reaching up to 1.5× the lower and upper IQR. Significance was determined using two-tailed Student’s t-tests. b, Kaplan–Meier survival curves with log-rank tests of intestinal tumour-free survival of BPNA mice treated with vehicle (n = 20), dabrafenib (n = 13), erlotinib (n = 9) or a combination of dabrafenib + erlotinib (n = 15) continuously from 20 DPI. Data are in days treated. c, Representative images of dual in situ hybridization Anxa1 (red) and Lgr5 (green) in BPNA tumours after 3 days of treatment with vehicle or dabrafenib + erlotinib from 27 DPI. d,e, Quantification of the percentage of Lgr5⁺ (d) and Anxa⁺ (e) cells in BPNA tumours after 3 days of treatment with vehicle or dabrafenib + erlotinib (D+E; n = 3 per group). Data are mean ± s.e.m. Significance was determined using a two-tailed Student’s t-test. f, UMAP of epithelial cells and cell-fate clusters with density overlay mapping from scRNA-seq of colonic tumours from APC (4,625 cells, n = 6 biological replicates), BPNA (3,169 cells, n = 5 biological replicates), BPNA mice treated with vehicle (1,305 cells, n = 5) and BPNA mice treated with dabrafenib + erlotinib (2,383 cells, n = 4). g, Alluvial plot showing the proportion of cell fates in panel f. h, Boxplots of single sample gene set enrichment analysis scores of CBC and RSC gene sets and SCI of RNA-seq of end point BPNA tumours in treatment groups (n = 16 for vehicle, n = 8 for dabrafenib, n = 5 for erlotinib and n = 3 for dabrafenib + erlotinib combination therapy). Boxes are median and IQR, and whiskers extend to the minimum and maximum values reaching up to 1.5× the lower and upper IQR. Significance was determined using two-tailed Student’s t-tests. i, Heatmap of scaled expression of normalized counts associated with WNT target genes in RNA-seq of BPNA colonic tumours post-treatment (for biological replicates: n = 16 for vehicle, n = 8 for dabrafenib, n = 5 for erlotinib and n = 3 for dabrafenib + erlotinib combination therapy). Statistical testing was by a two-sided Wald test; the gene symbols in red have a Benjamini–Hochberg adjusted P < 0.05 of vehicle versus dabrafenib + erlotinib.

Fig. 5. RNF43 loss sensitizes to BRAF + EGFR inhibition via restricting tumour epithelial plasticity.

a, Boxplot of SCI in APC or BRAF-driven small intestinal tumours. Scores are from RNA-seq of APC (n = 13), BP (n = 6), BPN (n = 30) and BPRNF (n = 5) mice. The boxes are median and IQR, and whiskers extend to the minimum and maximum values reaching up to 1.5× the lower and upper IQR. Significance was determined using two-tailed Student’s t-tests. b, Heatmap of scaled expression of normalized counts associated with WNT target genes from RNA-seq of BP (n = 6) and BPRNF (n = 5) tumours. The gene symbols in red have a Benjamini–Hochberg adjusted P < 0.05. c, Kaplan–Meier curves with log-rank tests of intestinal tumour-free survival of BP mice treated with vehicle (n = 7) or dabrafenib + erlotinib (n = 11) and BPRNF mice treated with vehicle (n = 15) or dabrafenib + erlotinib (n = 18) continuously from 130 DPI. Plotted are days treated. d, Boxplot of SCI of end point BPRNF small intestinal tumours treated with vehicle (n = 10) or dabrafenib + erlotinib (n = 10). The boxes are median and IQR, and the whiskers extend to the minimum and maximum values reaching up to 1.5× the lower and upper IQR. Significance was determined using a two-tailed Student’s t-test. e, Representative dual in situ hybridization Anxa1 (red) and Lgr5 (green) images in BPRNF end point tumours treated with vehicle (n = 5) or dabrafenib + erlotinib (n = 2). f, Representative haematoxylin and eosin (H&E), Anxa1, Lgr5 and Axin2 in situ hybridization images in BPRNF tumours following 3 days of vehicle (n = 5) or dabrafenib + erlotinib (n = 5) from 130 DPI. g, Primary tumour burden in BPRNF mice following 3 days of vehicle (n = 5) or dabrafenib + erlotinib (n = 5) treatment from 130 DPI. Data are mean ± s.e.m. Significance was determined using a two-tailed Mann–Whitney test. h, UMAP of epithelial cells and cell-fate clusters with density overlay mapping from scRNA-seq of BPRNF small intestinal tumours treated with vehicle (8,869 cells, n = 3) and dabrafenib + erlotinib (11,379 cells, n = 4) for 1 day or vehicle (20,957 cells, n = 4) and dabrafenib + erlotinib (13,140 cells, n = 5) for 3 days. i, Alluvial bar graph showing the proportion of cell states between BPRNF mice treated with vehicle or the dabrafenib + erlotinib combination for 3 days in panel h. j, Graphical summary representing the key epithelial cell states in CRC and response to therapy. The x axis represents increasing expression of MAPK and regenerative-associated genes, and the y axis represents expression of WNT pathway targets. The red arrows indicate the ability of GEMMs to shift epithelial cell state under treatment pressure. The schematic was created in BioRender. White, M. (2025) https://BioRender.com/z7mw8xa.

Extended Data Fig. 1. KRAS activation drives RSC marker expression in colonic tissue and tumours.

A-E Quantification of indicated stains of colonic tissue from Fig. 1b across VillinCre^ERT2 (WT), KG12D and KG12C 30 days post tamoxifen induction. Mean ± SEM, One-way ANOVA with Dunnett’s correction for multiple comparisons. WT n = 5, KG12D n = 3, KG12C n = 6. F- Schematic description of genetic crossing strategies to generate APC, AKG12D and AKG12C mice. The schematic was created in BioRender. White, M. (2025) https://BioRender.com/c0vov8d. G-H- Quantification of percentage of Lgr5⁺ DAPI⁺ cells and BrdU positive cells in recombined tissue of time-course sampling of APC and AKG12D at intracolonic tamoxifen injection site 4,7,14 and 21 days post induction. For Lgr5+ DAPI+ stain, APC n = 3 (day 21) and n = 4 (days 4, 7 and 14). AKG12D n = 4 (days 4 and 7) n = 3 (days 14 and 21). For BrdU+ stain, APC n = 3 (days 14 and 21) and n = 4 (days 4 and 7). AKG12D n = 3 (days 7, 14 and 21) and n = 4 (day 4). Mean ± SEM. Two-tailed t-tests, no correction for multiple comparisons. I- Representative images for annotated stains of colonic tumours from APC, AKG12D, AKG12C mice, induced with intracolonic 4-hydroxytamoxifen (4-OHT). J-M- Quantification of indicated stains of APC (n = 4 and n = 5), AKG12D (n = 5) and AKG12C (n = 4) colonic tumours. Mean ± SEM, One-way ANOVA with Dunnett’s correction for multiple comparisons.

Extended Data Fig. 2. Epithelial cell phenotyping in colon tumours from APC and AKG12D with or without KRAS G12D inhibition.

A-B- Regenerative stem cells (RSC) and crypt base columnar (CBC) scores from bulk RNA sequencing of intracolonic tumours from APC (n = 13) or AKG12D (n = 9) mice. Boxes are median and IQR, whiskers extend to minimum and maximum values reaching up to 1.5*lower and upper IQR. Two-tailed t-test. C- Schematic description of single cell RNA sequencing series experiment of APC, AKG12D, AKG12D mice treated with vehicle and AKG12D mice treated with MRTX1133 (KRAS-G12D inhibitor). The schematic was created in BioRender. White, M. (2025) https://BioRender.com/c02uw1d. D-E- UMAP visualisation of epithelial cells from scRNAseq of colonic tumours from APC mice (2,442 cells from 4 biological replicates), AKG12D mice (4439 cells from 3 biological replicates), AKG12D mice treated with vehicle (5,429 cells from 3 biological replicates) and AKG12D mice treated with MRTX1133 (5,823 cells from 5 biological replicates). Original 12 clusters (D) and the 7 key cell states (E) alongside the cluster numbers that make up the 7 key states. F-G- Violin plots of module scores of MAPK pathway activation (“Dry_MAPK” and “Dry_MEK_Inhib”), WNT pathway activation (“Apc_adenoma” and “Sansom_Wnt_signature”), YAP/Foetal (“Gregorieff_YAP” and “Foetal_Mustata”), crypt base columnar (CBC) scores, regenerative stem cell (RSC) scores and stem cell index (SCI) across 12 clusters (C) and the 7 key cell states (D). Horizontal line represents median of all the module scores. H- Key individual marker genes relating to epithelial cell state across cell populations. Dot size represents percentage of expressing cells and colour indicates average expression.

Extended Data Fig. 3. Epithelial cellular plasticity in response to KRAS inhibition.

A- Bar graph showing major epithelial populations across experimental conditions form scRNAseq associated with Fig. 2a,b at an individual sample level B- Boxplots comparing the cell type proportion of four experimental conditions. Each dot represents the cell type proportion of each sample. Colours indicated each experimental condition. Two-tailed t-test. Boxes are median and interquartile ranges (IQR) with whiskers extending from IQR to minimum and maximum values up to 1.5 * lower and upper IQR respectively. C- Individual marker genes relating to apoptosis, proliferation and MAPK-pathway signalling across each experimental condition from scRNAseq. Dot size represents percentage of expressing cells and colour indicates average expression. D- Representative images of AKG12D colonic tumours following 4-day treatment with vehicle or MRTX1133. IHC images of the apoptotic marker cleaved PARP (cPARP) or proliferation (Ki67) are presented. E- Quantification of percentage cPARP positive cells in colonic tumours of AKG12D mice treated with vehicle (n = 4) or MRTX1133 (n = 5) for 4 days. Mean ± SEM. Two-tailed t-test. F- Quantification of percentage Ki67 positive cells in colonic tumours of AKG12D mice treated with vehicle (n = 4) or MRTX1133 (n = 5) for 4 days. Mean ± SEM. Two-tailed t-test. G- UMAP visualisation of epithelial cells and major cell fate clusters from single cell RNA sequencing of colonic tumours from APC mice (30980 cells from 4 biological replicates), AKG12C mice (11821 cells from 2 biological replicates), AKG12C mice treated with vehicle (41963 cells from 4 biological replicates) or AZD4625 (33527 cells from 5 biological replicates). H- Alluvial bar graph showing percentage of cells of major epithelial populations across experimental conditions from scRNAseq in Extended Data Fig. 3g. I-J Individual marker genes relating to apoptosis, proliferation and MAPK-pathway signalling across each experimental condition in scRNAseq data from Extended Data Fig. 3g. Dot size represents percentage of expressing cells and colour indicates average expression.

Extended Data Fig. 4. Spatial transcriptomics show epithelial plasticity with KRAS G12D inhibition.

A- UMAP projection of APC, AKG12D mice treated with vehicle or MRTX1133 (n = 3 for each condition) cells after 10X Xenium spatial transcriptional profiling. Cells are coloured based on Leiden clustering. Tumour epithelial clusters were identified based on their spatial distribution in the tissue and key marker gene expression. B- Stacked bar graph of Leiden cluster assignment of tumour epithelial cells grouped by their experimental condition (n = 3 for each). C- Representative images of spatial distribution of all cells coloured by their Leiden cluster association (n = 3 for each). Scale 500um. D- Heatmap showing alignment of Leiden clusters presented in Extended Data Fig. 4a to key marker genes of cell populations from scRNAseq data. Colour bar is gene expression column min/max. E- PCA from Fig. 2f showing highest cell type call in this space. Cells are coloured by their highest probability for cell type assignment. F- Gene set scores for the individual gene sets overlayed on PCA presented in Fig. 2f.

Extended Data Fig. 5. Murine tumour derived spheroids are heterogeneous and are in a Mapk^hi state in early liver metastasis development.

A- UMAP visualisation of scRNAseq and major cell fate clusters of murine tumour derived tumour organoids. Lines are AKPT small intestinal primary tumour (AKPT SIT, 6099 cells from 3 biological replicates), KPN liver metastases (LMet) (1501 cells from 3 biological replicates) and KPN small intestinal tumour (SIT) (5251 cells from 3 biological replicates. B- Alluvial bar graph showing percentage of cells of major epithelial populations across experimental conditions form scRNAseq from Extended Data Fig. 5a. C- UMAPs of annotated gene sets in the three experimental conditions. D- Representative images of H&E and Anxa1, Lgr5 and Axin2 ISH stains in time-course sampling of AKPT and KPN liver metastases sampled 4, 7, 14 and 28 days post intrasplenic transplantation of spheroids. n = 3 at each time point. The Anxa1 and Lgr5 stains of the AKPT are the same image as the main Fig. 3c. E-H- Liver tumour number and burden in AKPT and KPN intrasplenic transplantation mice across different time points. Mean ± SEM, Kruskal-Wallis test. 3 biological replicates for each time points in each model. ns= not significant *=p < 0.05.

Extended Data Fig. 6. Liver metastasis are transcriptionally heterogeneous and correlates with size.

A- UMAP visualisation of AKPT intracolonic, AKPT intrasplenic, KPN intracolonic, and KPN intrasplenic cells after spatial transcriptional profiling. Cell colours denote Leiden clustering. Tumour epithelial and normal adjacent colon clusters were identified based on their spatial distribution in the tissue and key markers. B- Stacked bar graph of Leiden cluster assignment of tumour epithelial cells from KPN intracolonic (n = 4) and intrasplenic (n = 4) transplanted mice. C- Stacked bar graph of Leiden cluster assignment of tumour epithelial cells from AKPT intracolonic (n = 4) and intrasplenic (n = 3) transplants. D- Representative image of spatial distribution of cells coloured by their Leiden cluster association and cell type assignment of tumour epithelial cells in AKPT spheroid derived intracolonic tumours (n = 4). Scale bar 1 mm. E- Representative image of spatial distribution of cells coloured by their Leiden cluster association and cell type assignment of tumour epithelial cells in AKPT spheroid-derived liver metastasis, 28 days post intrasplenic injection (n = 3). Scale bar 1 mm. F- Representative image of spatial distribution of cells coloured by their Leiden cluster association and cell type assignment of tumour epithelial cells in KPN organoid derived intracolonic tumours (n = 4). Scale bar 1 mm. G- Representative image of spatial distribution of cells coloured by their Leiden cluster association and cell type assignment of tumour epithelial cells in KPN organoid-derived liver metastasis, 28 days post intrasplenic injection (n = 4). Scale bar 1 mm. H- Scatter plot of individual metastasis area vs percentage largest cluster present in AKPT derived liver metastasis, pooled from three biological replicates. Representative images of large transcriptionally heterogeneous and small transcriptionally homogenous liver metastasis. Tumour epithelial cells were coloured in their respective Leiden cluster colour. I- PCA from Fig. 3f showing highest cell type call in this space. Cells are coloured by their highest probability for cell type assignment. J- Gene set scores for the individual gene sets overlayed on PCA presented in Fig. 3f. K- Cells coloured by experimental condition (AKPT intrasplenic n = 3, AKPT intracolonic n = 4, KPN intrasplenic n = 4, KPN intracolonic n = 4) overlayed on PCA presented in Fig. 3f.

Extended Data Fig. 7. Limited impact of MEK inhibition in AKPT and KPN intracolonic transplants.

A- Experimental schematic of AKPT murine derived organoid intracolonic transplantation treated with AZD6244 or vehicle control when tumour was visible on endoscope. The schematic was created in BioRender. White, M. (2025) https://BioRender.com/zz3wsxv. B- Kaplan-Meier survival curves with Log Rank test of intestinal tumour free survival of AKPT transplants treated with vehicle (n = 7) or AZD6244 (n = 5). C- Colonic tumour burden for each treatment group (vehicle, n = 7 or AZD6244, n = 4). Mean ± SEM. Two tailed t-test. D- Representative images and associated quantification of Anxa1 ISH in vehicle (n = 7) and AZD6244 (n = 5) treated tumours. Mean ± SEM. Two tailed t-test. E- Experimental schematic of KPN murine derived organoid intracolonic transplantation treated with AZD6244 or vehicle control when tumour was visible on endoscope. The schematic was created in BioRender. White, M. (2025) https://BioRender.com/mhszqez. F- Kaplan-Meier survival curves with Log Rank test of intestinal tumour free survival of KPN transplants treated with vehicle (n = 11) or AZD6244 (n = 7). G- Colonic tumour burden each treatment group (vehicle, n = 11 or AZD6244, n = 7). Mean ± SEM. Two tailed t-test. H- Representative images and associated quantification of Anxa1 ISH in vehicle (n = 6) and AZD6244 (n = 5) treated tumours. Mean ± SEM. Two tailed t-test.

Extended Data Fig. 8. Characterisation of advanced BRAF mutant GEMMs.

A- Schematic description of genetic crossing strategies to generate VillinCre^ERT2 Braf^V600E/+ Trp53^fl/fl, VillinCre^ERT2 Braf^V600E/+ Alk5^fl/fl VillinCre^ERT2 Braf^V600E/+ Trp53^fl/fl Alk5^fl/fl, VillinCre^ERT2 Braf^V600E/+Trp53^fl/fl Rosa26^N1icd/+, and VillinCre^ERT2 Braf^−V600E Trp53^fl/fl Alk5^fl/fl Rosa26^N1icd/+ mice (BP, BA, BPA, BPN, and BPNA respectively). Cre, Cre-recombinase; ER, Oestrogen receptor; LoxP, Cre-Lox recombination site. The schematic was created in BioRender. White, M. (2025) https://BioRender.com/7yy6mmn. B- Kaplan-Meier survival curves of intestinal tumour free survival. Median tumour free survival days post induction (DPI); BA 192 days (n = 18), BPA 60 days (n = 37) and BPNA 37 days (n = 33). P-value calculated by log-rank test. C- Bar graph of primary tumour location across BA, BPA and BPNA models. D- Representative images of BPNA colon and colonic tumour from cohort (n = 33) described in (B), scale bar 2 mm low power image, 500μm high power image. E- Representative images of Anxa1 and Lgr5 ISH in tumours of each genotype (n = 5 each genotype). Scale bar 100μm. F-G- Boxplots of single sample gene set enrichment analysis (ssGSEA) scores of crypt base columnar (CBC) and regenerative stem cell (RSC) gene sets from bulk RNA sequencing of VillinCre^ERT2 Apc^fl/fl (APC, n = 13), BA (n = 4), BPA (n = 10), BPNA (n = 12) mice. Boxes are median and IQR, whiskers extend to minimum and maximum values reaching up to 1.5*lower and upper IQR. Two-tailed t-tests. H- Representative images of a BPNA colonic tumour stained with β-Catenin IHC staining showing membranous rather than nuclear staining. n = 3. Scale bar 500μm. I- Experimental schematic of VillinCre^ERT2 Braf^−V600E Trp53^fl/fl Alk5^fl/fl Rosa26^N1icd/+ (BPNA) mice treated from 20 days post induction. The schematic was created in BioRender. White, M. (2025) https://BioRender.com/a9j6uli. J- Representative images of H&E and Dusp6 ISH in BPNA colonic tumours under each treatment condition. K- Quantification of Dusp6 positive cells in tumour area across each group (n = 5 in each group). Mean ± SEM, One-way ANOVA with Tukey correction for multiple comparisons L- Heatmap of scaled expression of normalised counts associated with MAPK activation markers identified by bulk RNAseq in each treatment condition of BPNA tumours (vehicle n = 16, dabrafenib n = 8, erlotinib n = 5, dabrafenib + erlotinib combination therapy n = 3). Statistical testing by two-sided Wald test, red gene have a Benjamini-Hochberg adjusted p-value < 0.05 for dabrafenib + erlotinib vs vehicle. M- Log2 fold change of MAPK activation target genes from bulk RNA sequencing of BPNA tumours treated with dabrafenib + erlotinib vs vehicle control. Statistical testing by two-sided Wald test, red bars have a Benjamini-Hochberg adjusted p-value < 0.05. ns= not significant. N- Representative images of dual FISH Anxa1 (red) and Lgr5 (green) in endpoint BPNA tumours after prolonged treatment from Day 20 post induction until clinical endpoint with vehicle (n = 10) or dabrafenib and erlotinib (n = 6). Scale bar 250μm O- Associated quantification of Lgr5+ and Anxa1+ cells in endpoint BPNA tumours after prolonged treatment from Day 20 post induction ongoing until clinical endpoint with vehicle (n = 10) or dabrafenib and erlotinib (n = 6). Mean ± SEM two-tailed t-test.

Extended Data Fig. 9. Epithelial cellular plasticity in response to BRAF+EGFR inhibition in BPNA tumours and induces apoptosis with Lgr5 targeting radiotherapy.

A- Boxplots comparing the cell type proportion of four experimental conditions. Each dot represents the cell type proportion of each sample. Colours indicated each experimental condition. Boxes are median and IQR, whiskers extend to minimum and maximum values reaching up to 1.5*lower and upper IQR. Two-tailed t-test. B- Individual marker genes relating to apoptosis, proliferation and MAPK-pathway signalling across each experimental condition from sc RNAseq. Dot size represents percentage of expressing cells and colour indicates average expression. C- Heatmap of scaled expression of normalised counts of genes associated with the CBC or RSC phenotype by bulk RNAseq of intracolonic tumours from BPNA tumours in annotated treatment conditions (vehicle n = 16, dabrafenib n = 8, erlotinib n = 5, dabrafenib + erlotinib combination therapy n = 3). Statistical testing by two-sided Wald test, red gene symbols have a Benjamini-Hochberg adjusted p-value < 0.05. D- Volcano plot of showing up and down regulated genes post dabrafenib + erlotinib compared to vehicle controls in BPNA mice. Statistical testing by two-sided Wald test. E- Tumour burden (mm²) in VillinCre^ERT2 Braf^−V600E Trp53^fl/fl Alk5^fl/fl Rosa26^N1icd/+ (BPNA) mice following 3-days of vehicle (n = 6) or 3-days of dabrafenib + erlotinib (n = 3) combination treatment from day 27 post-induction; or 3-days of vehicle + a single dose of Radiotherapy (RT) (4 Gy) (n = 3) or 3-days of dabrafenib + erlotinib (n = 3) combination treatment + a single dose of RT (4 Gy). Mean ± SEM. Kruskal-Wallis test with Dunn’s correction for multiple comparisons. F- Representative images of H&E, Lgr5, Anxa1 and Axin2 ISH, p-ERK, Ki67 and c-PARP IHC in BPNA mice following 3-days of vehicle (n = 3), dabrafenib + erlotinib (n = 3), vehicle + RT (n = 3), dabrafenib + erlotinib + RT (n = 3) from 27 DPI. Scale bars 500μm and 100μm as indicated. G-L- Quantification of indicated ISH (Lgr5, Anxa1 and Axin2) and IHC (Dusp6, Ki67 and c-PARP) staining in BPNA mice following 3-days of vehicle (n = 3), dabrafenib + erlotinib (n = 3), vehicle + RT (n = 3), dabrafenib + erlotinib + RT (n = 3) from 27 DPI. Mean ± SEM. Kruskal-Wallis test with Dunn’s correction for multiple comparisons.

Extended Data Fig. 10. RNF43 does not drive a Wnt active phenotype in the context of Braf mutation and unaffected by Wnt ligand reduction.

A- Schematic description of genetic crossing strategies to generate VillinCre^ERT2 Rnf43^fl/fl (RNF43), VillinCre^ERT2 Braf^V600E/+ Trp53^fl/fl (BP), VillinCre^ERT2 Trp53^fl/fl Rnf43^fl/fl (PRNF), VillinCre^ERT2 Braf^V600E/+ Rnf43^fl/f (BRNF), VillinCre^ERT2 Braf^V600E/+ Trp53^fl/fl Rnf43^fl/fl mice (BPRNF). Cre, Cre-recombinase; ER, Oestrogen receptor; LoxP, Cre-Lox recombination site. The schematic was created in BioRender. White, M. (2025) https://BioRender.com/k5ttyup. B-C- Boxplots of single sample gene set enrichment analysis (ssGSEA) scores of crypt base columnar (CBC) and regenerative stem cell (RSC) gene sets from bulk RNA sequencing of endpoint Braf-driven small intestinal tumours (APC, n = 13, BP, n = 6, BPN, n = 30, BPRNF, n = 5). Boxes are median and IQR, whiskers extend to minimum and maximum values reaching up to 1.5*lower and upper IQR. Two-tailed t-test. D- Kaplan-Meier survival curves with Log Rank test of intestinal tumour free survival of RNF43 (n = 13), PRNF (n = 26), BRNF (n = 18), BP (n = 9) and BPRNF (n = 55) mice aged until clinical endpoint. Data presented as days post induction (DPI). Median tumour free survival: RNF43 undefined, PRNF 449 days, BRNF 454 days, BP 252 days, BPRNF 162 days. E-F- Tumour number and tumour burden (mm²) in PRNF (n = 21), BRNF (n = 18), BP (n = 7) and BPRNF (n = 52) mice aged until clinical endpoint. Mean ± SEM. Kruskal-Wallis test with Dunn’s correction for multiple comparisons. Total mouse number (n) for tumour number (E) and tumour burden (F) is lower as mice sampled for reasons other than intestinal tumorigenesis have been removed. G- Representative image of H&E, Anxa1 and Lgr5 ISH, a-SMA and Muc5ac IHC staining in BPRNF intestinal tumours at clinical endpoint. Scale 100μm. H-J- Quantification of indicated ISH (Anxa1, Lgr5 and Axin2) in BP (n = 3) and BPRNF (n = 6) intestinal tumours at clinical endpoint. Mean ± SEM. Two-tailed Mann-Whitney. K- Heatmap of scaled expression of normalised counts associated with key RSC and CBC genes identified by bulkRNAseq of tumours from BP (n = 6) and BPRNF (n = 5) mice. Statistical testing by two-sided Wald test, gene symbols in red have a Benjamini-Hochberg adjusted p-value < 0.05. L- UMAP visualisation of epithelial cells and major cell fate clusters with density overlay mapping from scRNAseq of colonic tumours from APC (4625 cells from 6 biological replicates), BP (1616 cells from 6 biological replicates) and BPRNF (5676 cells from 8 biological replicates) mice. M- Alluvial bar graph showing percentage of cells of major epithelial populations across experimental conditions from scRNAseq from Extended Data Fig. 10l. N- Kaplan-Meier survival curves with Log Rank tests of intestinal tumour free survival of BPRNF mice treated continuously with vehicle (n = 15) or LGK974 (PORCN inhibitor, n = 16) from 130 DPI. Data presented as days on treatment. O- Boxplot of single sample gene set enrichment analysis (ssGSEA) scores of stem cell index (SCI) from bulk RNA sequence (bulk RNAseq) of BPRNF tumours treated with either vehicle (n = 6) or LGK974 (n = 5). Boxes are median and IQR, whiskers extend to minimum and maximum values reaching up to 1.5*lower and upper IQR. Two-tailed t-test. P- Heatmap of scaled expression of normalised counts associated with Wnt target genes as identified by bulk RNAseq of tumours from BPRNF mice treated with vehicle (n = 6) or LGK974 (n = 5). None of the displayed genes have a Benjamini-Hochberg adjusted p-value < 0.05. Q- Representative images of Anxa1, Lgr5 and Axin2 ISH in BPRNF mice treated with vehicle (n = 3) or LGK974 (n = 3) from 130 DPI. Scale 1 mm. R-T- Quantification of indicated ISH (Anxa1, Lgr5 and Axin2) in BPRNF mice treated with vehicle (n = 3) or LGK974 (n = 3). Mean ± SEM. Two-tailed Mann-Whitney. U- Representative images of β-Catenin staining in BPRNF mice treated with vehicle (n = 3) or LGK974 (n = 3). Black arrows indicate clones of β-Catenin positivity. Scale 500μm.

Extended Data Fig. 11. RNF43 loss prevents cell state plasticity resistance in Braf mutant tumours.

A- Experimental schematic of BP and BPRNF mice treated from 130 DPI. The schematic was created in BioRender. White, M. (2025) https://BioRender.com/effw414. B- Kaplan-Meier survival curves with Log Rank tests of intestinal tumour free survival of BPRNF mice treated continuously with vehicle (n = 15), dabrafenib (n = 16), erlotinib (n = 8) or dabrafenib + erlotinib (n = 18) combination from 130 DPI. Data presented as days on treatment. Median tumour free survival: vehicle 63 days, dabrafenib 72 days, erlotinib 54 days, dabrafenib + erlotinib 191 days. C- Kaplan-Meier survival curves with Log Rank tests of intestinal tumour free survival of BP mice treated continuously with vehicle (n = 7), dabrafenib (n = 3), erlotinib (n = 5) or dabrafenib + erlotinib (n = 11) combination from 130 DPI. Data presented as days on treatment. Median tumour free survival: vehicle 37 days, dabrafenib 78 days, erlotinib 63 days, dabrafenib + erlotinib 46 days. D-E- Tumour number and tumour burden (mm²) in BP vehicle (n = 6), BP dabrafenib + erlotinib (n = 10), BPRNF vehicle (n = 14), BPRNF dabrafenib + erlotinib (n = 16) mice. Mean ± SEM. Kruskal-Wallis test with Dunn’s correction for multiple comparisons. Total mouse number (n) for tumour number (D) and tumour burden (E) is lower as mice sampled for reasons other than intestinal tumorigenesis have been removed. F-G- Boxplot of bulk crypt base columnar (CBC) and regenerative stem cell (RSC) score of BPRNF mice treated with either vehicle or dabrafenib + erlotinib (n = 10 both conditions). Boxes are median and IQR, whiskers extend to minimum and maximum values reaching up to 1.5*lower and upper IQR. Two-tailed t-test. H-I- Heatmap of scaled expression of normalised counts associated with Wnt target genes and stem cell genes respectively as identified by bulkRNAseq of tumours from BPRNF mice treated with vehicle (n = 10) or dabrafenib + erlotinib (n = 10). Statistical testing by two-sided Wald test, no displayed genes have a Benjamini-Hochberg adjusted p-value < 0.05. J- Representative images of Anxa1, Lgr5 and Axin2 ISH, p-ERK, Ki67 and c-PARP IHC in BP vehicle (n = 3), BP dabrafenib + erlotinib (n = 5), BPRNF vehicle (n = 6) and BPRNF dabrafenib + erlotinib (n = 4) mice treated continuously from 130 DPI. Scale 500μm as indicated. K-P- Associated quantification of indicated ISH (Anxa1, Lgr5 and Axin2) and IHC (p-ERK, Ki67 and c-PARP) in BP vehicle (n = 3), BP dabrafenib + erlotinib (n = 5), BPRNF vehicle (n = 6) and BPRNF dabrafenib + erlotinib (n = 4) mice treated continuously from 130 DPI. Mean ± SEM. Kruskal-Wallis test with Dunn’s correction for multiple comparisons.

Extended Data Fig. 12. Short treatment exposure does not impact BPRNF epithelial phenotype and RNF43 loss prevents epithelial cellular plasticity in response to BRAF+EGFR inhibition.

A- Tumour number in VillinCre^ERT2 Braf^V600E/+ Trp53^fl/fl Rnf43^fl/fl (BPRNF) mice treated with vehicle (n = 5) or dabrafenib + erlotinib (n = 5) for 3 days from 130 DPI. Mean ± SEM. Two-tailed Mann-Whitney. B- Representative images of p-ERK, Ki67 and c-PARP IHC in BPRNF mice treated with vehicle (n = 5) or dabrafenib + erlotinib (n = 5) for 3 days from day 130 post-induction. Scale bars 500μm and 100μm as indicated. C-H- Associated quantification of indicated ISH (Anxa1, Lgr5 and Axin2) and IHC (p-ERK, Ki67 and c-PARP) in BPRNF mice treated with vehicle (n = 5) or dabrafenib + erlotinib (n = 5) for 3 days from day 130 post-induction. Mean ± SEM. Two-tailed Mann-Whitney test. I- Dot plot showing the expression of genes associated with apoptosis, proliferation and MAPK pathway targets between experimental conditions in BPRNF mice treated with vehicle or dabrafenib + erlotinib combination for 1 day and 3 days from scRNAseq. J- Alluvial bar graph showing percentage of cells of major epithelial populations between experimental conditions in BPRNF mice treated with vehicle or dabrafenib + erlotinib combination for 1 day from scRNAseq from Fig. 5h. K- Boxplots comparing the cell type proportion of four experimental conditions. Each dot represents the cell type proportion of each sample. Colours indicated each experimental condition. Boxes are median and IQR, whiskers extend to minimum and maximum values reaching up to 1.5*lower and upper IQR. Two-tailed t-test.