Epithelial TGFβ engages growth-factor signalling to circumvent apoptosis and drive intestinal tumourigenesis with aggressive features

Abstract

The pro-tumourigenic role of epithelial TGFβ signalling in colorectal cancer (CRC) is controversial. Here, we identify a cohort of born to be bad early-stage (T1) colorectal tumours, with aggressive features and a propensity to disseminate early, that are characterised by high epithelial cell-intrinsic TGFβ signalling. In the presence of concurrent Apc and Kras mutations, activation of epithelial TGFβ signalling rampantly accelerates tumourigenesis and share transcriptional signatures with those of the born to be bad T1 human tumours and predicts recurrence in stage II CRC. Mechanistically, epithelial TGFβ signalling induces a growth-promoting EGFR-signalling module that synergises with mutant APC and KRAS to drive MAPK signalling that re-sensitise tumour cells to MEK and/or EGFR inhibitors. Together, we identify epithelial TGFβ signalling both as a determinant of early dissemination and a potential therapeutic vulnerability of CRC's with born to be bad traits.

Fig. 1. Stromal content has no prognostic value in pT1 CRC.

a Comparison of transcriptome-based Microenvironment Cell Populations (MCP) scores for fibroblasts (left) and cytotoxic T lymphocytes (right) in relapse (n = 10; red) and non-relapse (n = 17; blue) pT1 cohort. Statistical significance was determined by two-sided wilcoxon rank-sum exact test using wilcox.test function in stats R package. Horizontal line represents median values, boxes indicate the inter-quartile range and bars denote the maximum and minimum values. b Left, representative H&E of pT1 tumour scored for stromal content (red) using QuPath software. Right, correlation between transcriptome-based fibroblast scores and digital pathology stromal scores. Two-sided P = 0.00009, rho = Spearman correlation coefficient. Grey region shows the 95% confidence interval (CI) for predictions from a linear model (lm). Scale bar, 250 µm. c Proportion of the main mutations in CRCs in relapse (red) and non-relapse (blue) pT1 cohort. d Gene set enrichment analysis (GSEA) showing positive enrichment of Hallmark curated gene sets for KRAS and mTORC1 signalling in relapse cases as compared to non-relapse using fast gene set enrichment analysis (fgsea) method. e Group-wise GSEA performed using fgsea R package on the log expression ratio of relapse vs non-relapse cases to assess enrichment of CMS/CRIS subtypes. X-axis shows normalised enrichment score (NES), and the adjusted p value is indicated on the colour bar, from red (significant) to blue (not significant). padj, adjusted p value (computed and corrected for multiple testing using the Benjamini–Hochberg procedure). f Hallmark gene set enrichment analysis of TGFβ signalling in relapse cases as compared to non-relapse using fast gene set enrichment analysis (fgsea) method. TGFβ_Up gene set from. g Correlation analysis between transcriptome-based fibroblast scores (x-axis) and sample-level gsea hallmark TGFβ signalling scores (y-axis) in relapse (red) and non-relapse (blue) pT1 cohort. Two-sided P = 0.6, rho = Spearman correlation coefficient. Grey region shows the 95% CI for predictions from (a). h Enrichment of cancer-associated fibroblast (CAF), left, and stromal (right) signature gene sets in relapse cases compared to non-relapse samples using fgsea method; Benjamini–Hochberg FDR <0.2 was considered as significant. NES normalised enrichment score, FDR false discovery rate, padj adjusted p value (computed and corrected for multiple testing using the Benjamini–Hochberg procedure).

Fig. 2. Epithelial cell-intrinsic activation of TGFβ signalling is tolerated in vivo, but not in vitro.

a Representative images of small intestinal organoids derived from uninduced VilCre^ER;Alk5^CA (Alk5^CA) and VilCre^ER;Alk5^+/+ (Alk5^+/+) mice, following 5 days of in vitro treatment with vehicle control (EtOH for 4-hydroxytamoxifen; PBS/BSA for TGFβ1), 4-hydroxytamoxifen (Tmx, top right panel) or TGFβ1 (bottom right panel). n = 3 organoid lines per group. Treatments were repeated twice. Scale bar, 50 μm. b Relative viability (dead; grey vs alive; blue) of Alk5^CA and Alk5^+/+ organoids measured 5 days after the indicated treatments. Alk5^CA organoids were treated with EtOH, Tmx, or Tmx plus ALK5i, whereas Alk5^+/+ organoids were treated with PBS/BSA or TGFβ1. Representative images of treated organoids, with or without ALK5i, are shown in Supplementary Fig. 2a. n = 3 biological replicates per group, each in technical triplicate. Data were ±s.e.m; *P = 0.05, one-tail Mann–Whitney U-test. c qPCR for TGFβ-target genes expressed by organoids described in a. n = 3 biological replicates per group; EtOH (green), tamoxifen (yellow), PBS/BSA (grey), TGFβ1 (red) each in technical triplicate. Data were ±s.e.m; *P = 0.05, one-tail Mann–Whitney U-test. d Representative IHC (p-SMAD3, cleaved caspase-3 (CC3), BrdU and lysozyme) and ISH (Olfm4) in the small intestine of VilCre^ER;Alk5^+/+ (Alk5^+/+) and VilCre^ER;Alk5^CA (Alk5^CA) mice 4 days post-tamoxifen induction. Arrows indicate CC3-positive apoptotic cells. Scale bar, 100 μm. e–h Quantification of small intestinal epithelial cells stained for cleaved caspase-3 (CC3) (e), BrdU (f), Olfm4 (g), and lysozyme (h) cells from mice of the indicated genotypes 4 days post-tamoxifen induction. n = 3 mice per group and *P = 0.05, except for BrdU scoring; n = 5 Alk5^+/+;Smad4^+/+, n = 6 Alk5^CA;Smad4^+/+ and n = 4 Alk5^CA;Smad4^fl/fl mice, *P = 0.03. All data were ±s.e.m; two-tail Mann–Whitney U-test.

Fig. 3. Apc-mutant cells are sensitive to cell-intrinsic TGFβ signalling.

a Quantification of BrdU-positive (left) and apoptotic cells (right) in VilCre^ER;Apc^fl/fl;Alk5^+/+ (blue) and VilCre^ER;Apc^fl/fl;Alk5^CA (orange) small intestinal crypts 3 and 4 days post-tamoxifen induction. n = 4 mice per group, except n = 3 mice per group for d3 BrdU data. Data were ±s.e.m; *P = 0.05, one-tail Mann–Whitney U-test. b Representative BrdU and H&E staining of small intestinal tissue from VilCre^ER;Apc^fl/fl;Alk5^+/+ and VilCre^ER;Apc^fl/fl;Alk5^CA mice 4 days post-tamoxifen induction. Scale bar, 100 μm. c Schematic illustrating the treatment timeline and tissue analysis from VilCre^ER;Apc^fl/fl;Alk5^CA mice. Tmx tamoxifen, ALK5i ALK5-inhibitor. d Representative IHC (p-SMAD3) and ISH (Alk5^CA) staining on VilCre^ER;Apc^fl/fl;Alk5^CA mice following vehicle or ALK5i treatment. Scale bar, 200 μm. e Quantification of BrdU-positive (left) and apoptotic cells (right) from VilCre^ER;Apc^fl/fl;Alk5^CA mice following vehicle or ALK5i treatment. n = 3 mice per group, except for scoring of apoptotic bodies where n = 5 mice per group. Data were ±s.e.m. *P = 0.05, **P = 0.008; one-tail Mann–Whitney U-test for BrdU; two-tail Mann–Whitney U-test for apoptosis.

Fig. 4. Epithelial TGFβ/ALK5 signalling synergises with Wnt and MAPK signalling to accelerate tumourigenesis.

a Survival plot for VilCre^ER;Apc^fl/+;Kras^G12D/+;Alk5^+/+ (n = 12, grey) and VilCre^ER;Apc^fl/+;Kras^G12D/+;Alk5^CA (n = 22, red) mice aged until clinical endpoint following tamoxifen induction. P = 1.0 × 10⁻⁴, log-rank test. b, c Small intestinal (SI) tumour number (b) and burden (c) per mouse from mice described in a. n = 12 Alk5^+/+ mice (grey), n = 16 Alk5^CA mice (red) for tumour number dataset; n = 7 Alk5^+/+ mice (grey), n = 15 Alk5^CA mice (red) for tumour burden dataset. Data were ±s.e.m; P = 0.064 (b), P = 0.063 (c), two-tail Mann–Whitney U-test. d Relative percentage of small intestinal tumours from VilCre^ER;Apc^fl/+;Kras^G12D/+;Alk5^CA (Alk5^CA) mice positive (grey) or negative (blue) for Alk5^CA expression (ISH). n = 4 mice. Data were ±s.e.m. e Representative IHC for p-ERK, β-catenin and p-SMAD3 on tumour tissue from mice described in a. Scale bar, 200 μm.

Fig. 5. Oncogenic KRAS-driven MAPK signalling confers tolerance to epithelial TGFβ activation.

a Representative H&E and Alk5 ISH of small intestinal tissue from VilCre^ER;Apc^fl/fl;Kras^G12D/+;Alk5^CA mice 3 days post-tamoxifen and daily treatment with vehicle or ALK5-inhibitor (ALK5i). Scale bar, 100 μm. b Quantification of BrdU-positive (left) and apoptotic (right) cells in mice described in a. n = 4 mice per group (BrdU scoring); n = 5 mice per group (apoptosis scoring). Data were ±s.e.m; P = 0.68 (BrdU), P = 0.15 (apoptosis); two-tail Mann–Whitney U-test. c Representative Itgβ6 ISH and p-ERK IHC on mice of the indicated genotypes 3 days post-tamoxifen. Scale bars, 100 μm. d Quantification of p-ERK-positive cells per crypt-villus unit in mice described in c. n = 5 per group. Data were ±s.e.m; P = 0.003. Two-tail Mann–Whitney U-test. e H&E staining of small intestinal tissue from VilCre^ER;Apc^fl/fl;Kras^G12D/+;Alk5^CA mice 3 days post-tamoxifen and daily treatment with vehicle or MEK1/2 inhibitor (MEKi). Right panels show magnification of boxed areas in left panels, with apoptotic cells indicated by black arrowheads. Scale bar, 100 μm. f Quantification of BrdU-positive (left) and apoptotic (right) cells in mice described in e. n = 3 mice per group. Data were ± s.e.m. *P = 0.05; one-tail Mann–Whitney U-test.

Fig. 6. Mice with intestinal epithelial TGFβ signalling mimic transcriptional features of human born to be bad pT1 tumours.

a Heatmap of 60 genes with predictive value obtained from prediction analysis for microarrays (PAMR) using data from VilCre^ER;Alk5^CA (ALK5^CA-like) and non-TGFβ/Alk5-activated (WT-like) mice 4 days post-tamoxifen. n = 3 mice per genotype. b Proportion of relapse and non-relapse pT1 patients enriched with ALK5^CA-like (red) or WT-like (blue) signature. Two-sided Fisher’s exact test using fisher.test function in stats R package. c Pseudocolour overlay images of representative patient tumours from each cluster stained for p-S6, p-mTOR, 4EBP1 and TGFβ1. Marker expression was assessed using multiplexed immunofluorescence and single-cell staining of epithelial tumour cells within a cohort of stage II CRC (n = 282). Scale bar, 200 µm. d Proteins (p-S6, 4EBP1, TGFβ1 and p-mTOR) associated with relapse in the pT1 cohort were used to stratify stage II CRC patients into two groups, high (red) and low (blue) expression (P = 0.003; log-rank test). e, f GSEA of TGFβ (e) and KRAS and mTORC1 (f) signatures on shrunken log expression ratio of VilCre^ER;Apc^fl/fl;Kras^G12D/+;Alk5^CA (AKA^CA; n = 7) vs wild type (WT; n = 3) intestinal tissue 3 days post-tamoxifen using fgsea method. g CRIS-B subtype enrichment analysis on mice described in e. h Group-wise GSEA performed using fgsea R package on the log expression ratio of AKA^CA vs WT mouse intestinal tissue to assess enrichment of CMS/CRIS subtypes. X-axis shows normalised enrichment score (NES), and the adjusted p value is indicated on the colour bar, from red (significant) to blue (not significant). padj adjusted p value (computed and corrected for multiple testing using the Benjamini–Hochberg procedure). i CRIS-B signature single-sample GSEA analysis on VilCre^ER;Apc^fl/fl;Kras^G12D/+ (AK, green), VilCre^ER;Apc^fl/fl;Kras^G12D/+;Alk5^CA (AKA^CA, grey) and VilCre^ER;Apc^fl/fl;Kras^G12D/+;Alk5^fl/fl (AKAKO, purple) tissue 3 days post-tamoxifen. AK, AKAKO n = 3 mice, AKA^CA n = 7 mice. Statistical significance determined by two-sided t-test using t.test function in stats R package (4.2.0). Horizontal line represents median values, boxes indicate the inter-quartile range and bars denote the maximum and minimum values. ***p ≤ 0.001; **p ≤ 0.01; *p ≤ 0.05. j Heatmap of genes associated with growth-factor and TGFβ signalling differentially expressed in intestinal tissue harvested 3 days post-tamoxifen injection from mice of the indicated genotypes. Wild-type (WT, n = 3, mustard), VilCre^ER;Apc^fl/fl (A, n = 6, lavender), VilCre^ER;Apc^fl/fl;Kras^G12D/+ (AK, n = 3, green), VilCre^ER;Alk5^CA (Alk5^CA, n = 7, aqua), VilCre^ER;Apc^fl/fl;Alk5^CA (AA^CA, n = 6, blue), VilCre^ER;Apc^fl/fl;Kras^G12D/+;Alk5^CA (AKA^CA, n = 7, grey), VilCre^ER;Apc^fl/fl;Kras^G12D/+;Alk5^CA;Smad4^fl/fl (AKA^CAS, n = 3, brown) and VilCre^ER;Apc^fl/fl;Kras^G12D/+;Alk5^fl/fl (AKAKO, n = 3, purple).

Fig. 7. Epithelial TGFβ/ALK5 signalling exposes therapeutic vulnerability to MAPK-targeted therapies.

a Survival plot for VilCre^ER;Apc^fl/+;Kras^G12D/+;Alk5^CA mice treated daily with vehicle or vandetanib/ZD6474 and aged until clinical endpoint following tamoxifen induction. n = 6 mice per group. P = 0.02, log-rank test. b Representative CD31 and p-ERK staining on small intestinal tumour tissue from mice described in a. Scale bar, 100 μm. c Small intestinal (SI) tumour number (left) and burden (right) per mouse from mice described in a. n = 5 vehicle (pink), n = 6 ZD6474 (aqua) mice. Data were ±s.e.m; P = 0.98 (tumour number), *P = 0.02 (tumour burden). Two-tail Mann–Whitney U-test. d Survival plot for VilCre^ER;Apc^fl/+;Kras^G12D/+;Alk5^CA mice treated daily with MEK1/2 inhibitor (MEKi) or EGFR inhibitor (EGFRi) and aged until clinical endpoint following tamoxifen induction. n = 24 untreated (grey), n = 14 MEKi (red), n = 10 EGFRi (blue) mice. P = 1.0 × 10⁻⁴ (MEKi), P = 3.0 × 10⁻² (EGFRi), log-rank test. e Representative H&E staining of small intestinal tissue from VilCre^ER;Apc^fl/+;Kras^G12D/+;Alk5^CA mice following indicated treatments. The bottom panels are a magnification of the boxed areas in the corresponding top panels. Scale bar, 100 μm. f Total tumour number (left) and burden (right) per mouse from mice described in d. n = 15 untreated (grey), n = 8 MEKi (red), n = 13 EGFRi mice (blue). Data were ±s.e.m; Tumour number: P = 0.09 (MEKi), **P = 0.005 (EGFRi), Tumour burden: *P = 0.01 (MEKi), *P = 0.03 (EGFRi). Two-tail Mann–Whitney U-test.