Advancing spatially resolved in vivo functional genomes will link complex genetic alterations prevalent in cancer to critical disease phenotypes within tumour ecosystems. To this end, we developed PERTURB-CAST, a method to streamline the identification of perturbations at the tissue level. By adapting RNA-templated ligation probes, PERTURB-CAST leverages commercial 10X Visium spatial transcriptomics to integrate perturbation mapping with transcriptome-wide phenotyping in the same tissue section using a widely available single-readout platform. In addition, we present CHOCOLAT-G2P, a scalable framework designed to study higher-order combinatorial perturbations that mimic tumour heterogeneity. We apply it to investigate tissue-level phenotypic effects of combinatorial perturbations that induce autochthonous mosaic liver tumours.