A central question in colorectal cancer (CRC) biology is why specific oncogenic driver mutations preferentially arise in defined anatomical regions of the intestine. Although left- and right-sided CRC display marked molecular and clinical differences, the role of regional epithelial zonation in constraining tumour initiation and evolution remains largely unexplored.

We assembled a longitudinal cohort of 463 tissue samples spanning normal epithelium, precancerous lesions, and established tumours from 19 genetically engineered mouse model (GEMM) combinations. Samples were collected across the proximal–distal intestinal axis (small intestine, caecum, and colon). Integrating bulk RNA-seq, single-cell transcriptomics, and spatial profiling, we mapped transcriptional phenotypes across anatomical location and disease stage.

Multi-omic interrogation revealed reproducible, genotype-specific disease trajectories emerging from zonation-restricted normal epithelial states. Driver mutation selection was fundamentally constrained by regional physiological signalling gradients. Apc-driven tumours exhibited expansion of canonical Lgr5+/Notum+ stem populations, consistent with a conventional adenoma bottom-up mode of initiation. In contrast, right-sided Braf/Alk5-driven tumours exploit a terminally differentiated EMP1+ enterocytes that in normal tissue is restricted to the proximal colon in both mouse and human, via a serrated polyp top-down trajectory. These findings demonstrate that tumour phenotypes reflect genotype–zonation interactions embedded within normal tissue architecture rather than purely emergent evolutionary states. Importantly, GEMMs faithfully recapitulate human transcriptional subtypes and mirror spatially restricted disease trajectories observed in human CRC.

Our data establish intestinal zonation as a critical determinant of oncogenic competence, constraining which driver mutations can successfully initiate transformation. This framework provides a mechanistic basis for anatomical tumour bias in CRC and offers a strategy for transcriptome-guided positioning of preclinical models.

This work was supported by CRUK CRC-STARS, Early Detection programme and ACRCelerate, alongside an all-Ireland AICRIstart programme. We thank the GEMMs Working Group for generating the diverse mouse models essential to this work.