Spatial mapping of human colorectal cancer cell states to identify co-targeting opportunities and determinants of drug response.

3140 Background: Colorectal cancer (CRC) progression relies on multiple mechanistically distinct proliferative and survival programs differing in lineage dependence, plasticity, and environmental sensitivity. LGR5-driven proliferation represents a stem-like program, whereas non-LGR5 modes—including progenitor, KRAS/MAPK, AP-1, and YAP-driven proliferation—reflect inflammatory and adaptive states that may confer therapeutic resistance. Single-cell studies revealed these programs, but technical barriers have restricted their quantitative and spatial analysis in human tumors, limiting translation to drug development and patient stratification. Methods: We integrated single-cell RNA sequencing (81 CRC patients) with spatial transcriptomics (Xenium; 16 matched primary CRC samples) to quantify epithelial cell states and tissue organization. Analyses focused on proliferative programs defined by LGR5, CDX2, ANXA1, and MKI67, stratified by KRAS genotype and intratumoral location, with extended spatial annotation of immune and stromal niches. Results: CRC tumors showed high compositional heterogeneity across subgroups (mean normalized Shannon entropy >0.6). On average, 36% of tumor epithelial cells were proliferative; only 50% expressed LGR5, with substantial inter-patient variability and genotype dependence. 14% of proliferating cells expressed ANXA1 alone, corresponding to inflammatory, YAP-high states with potential cytotoxic implications. LGR5-driven proliferation was enriched in KRAS-mutant tumors, contrasting with mouse models suggesting KRAS-mediated suppression and potentially indicating reduced sensitivity to KRAS inhibition. EGFR–LGR5 double-positive cells were abundant, identifying additional targetable niches. Spatial analyses revealed regions with high TGFβ signaling and co-localization with SPP1⁺ macrophages consistent with TGFβ-driven immunosuppressive niches described in complementary studies. These niches associate with reduced PD-1/PD-L1 blockade response, coincide with states responsive to combined TGFβ and PD-1 inhibition, and are not captured by genotype alone. Conclusions: Spatial transcriptomics provides quantitative, systems-level resolution of tissue architecture, revealing opportunities for rational co-targeting to disrupt tumor homeostasis and enhance immunotherapy efficacy. Ongoing analyses of neoadjuvant-treated tumors and paired primary–metastatic samples will extend this framework to capture therapy-induced and naturally evolving tumor states.