Purpose: Tumor heterogeneity compromises the clinical applicability of transcriptomic subtypes of colorectal cancers (CRCs). We performed multiregional tumor sampling to map the intra-tumor heterogeneity of primary CRCs and to evaluate the prognostic relevance of features less vulnerable to heterogeneity. Methods: Gene expression profiling was performed of 708 primary tumor samples from 515 patients treated by surgery for stage I-IV CRC at Oslo University Hospital (Human Transcriptome 2.0 Arrays). This included 2-4 multiregional samples from each of 97 patients (n=290 samples; mean 2.9 per tumor), and single bulk tumor samples from 418 patients. Intra-tumor heterogeneity was evaluated according to consensus molecular subtype (CMS) classification, including computational assessment in the single samples by CMS enrichment scores. Gene-wise heterogeneity scores were calculated as intra-class correlation coefficients in the multiregional sample set, and genes with low heterogeneity were used as input for subtype discovery by non-negative matrix factorization. Results: Heterogeneous CMS classification of multiregional samples was found in 40% of tumors. Heterogeneous tumors were enriched with CMS3 (odds ratio 5.0) and CMS4 (odds ratio 12.5), and the most common combinations were CMS2/4 and CMS1/3. Microsatellite instability and BRAFV600E mutations were enriched among tumors with a major CMS1 component only. KRAS/NRAS mutations were most frequent in CMS3 tumors without CMS1. Intra-tumor heterogeneity was primarily driven by mesenchymal-like and stromal signals, but was independently associated with a poor 5-year relapse-free survival among patients with stage I-III CRC (multivariable hazard ratio 1.5 [1.0-2.2]). Heterogeneity explained a larger proportion of variation in survival (14%) than cancer-associated fibroblasts (6%). Genes with low intra-tumor heterogeneity were enriched in cancer cell intrinsic signaling pathways, including cell cycle progression and MYC targets. De novo subtyping based on these genes recapitulated the intrinsic iCMS groups identified from single-cell sequencing (Joanito et al., Nat Genet 2022;54:963-75) with high accuracy (Cohen’s κ=0.80). Further sub-stratification identified four subtypes with similar biological activity to the original CMS, but lower classification heterogeneity among multiregional samples (25%). The subtypes were called congruent CMS (cCMS) and provided stronger prognostic stratification than CMS in multivariable analysis with clinicopathological and molecular features (cCMS4: hazard ratio 5.9 [2.4-14.7]; cCMS2: 3.4 [1.5-7.6]; cCMS3: 2.0 [0.9-4.3] with cCMS1 as reference). Conclusion: Multiregional sampling of primary CRCs enabled identification of cancer-intrinsic transcriptomic features robust to intra-tumor heterogeneity and with prognostic relevance. Citation Format: Jonas Langerud, Ina A. Eilertsen, Hossein Moosavi, Ingeborg F. Backe, Merete Hektoen, Marine Jeanmougin, Ole Sjo, Arild Nesbakken, Ragnhild A. Lothe, Anita Sveen. Multiregional transcriptomics of colorectal cancers define prognostic features less vulnerable to tumor heterogeneity. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 4540.