Redefining tumor classification and clinical stratification through a 1 colorectal cancer single-cell atlas 2 3

34 35 Colorectal cancer (CRC), a disease of high incidence and mortality, exhibits a large degree of 36 interand intra-tumoral heterogeneity. The cellular etiology of this heterogeneity is poorly 37 understood. Here, we generated and analyzed a single-cell transcriptome atlas of 49,859 CRC 38 cells from 16 patients, validated with an additional 31,383 cells from an independent CRC patient 39 cohort. We describe subclonal transcriptomic heterogeneity of CRC tumor epithelial cells, as well 40 as discrete stromal populations of cancer-associated fibroblasts (CAFs). Within CRC CAFs, we 41 identify the transcriptional signature of specific subtypes (CAF-S1 and CAF-S4) that significantly 42 stratifies overall survival in more than 1,500 CRC patients with bulk transcriptomic data. We also 43 uncovered two CAF-S1 subpopulations, ecm-myCAF and TGFß-myCAF, known to be associated 44 with primary resistance to immunotherapies. We demonstrate that scRNA analysis of malignant, 45 stromal, and immune cells exhibit a more complex picture than portrayed by bulk transcriptomic46 based Consensus Molecular Subtypes (CMS) classification. By demonstrating an abundant 47 degree of heterogeneity amongst these cell types, our work shows that CRC is best represented 48 in a transcriptomic continuum crossing traditional classification systems boundaries. Overall, this 49 CRC cell map provides a framework to re-evaluate CRC tumor biology with implications for clinical 50 trial design and therapeutic development. 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 Main 66 Colorectal cancer (CRC) is the third most commonly diagnosed cancer and a leading cause of 67 cancer-related mortality worldwide1,2. Approximately 50% of patients experience disease relapse 68 following curative-intent surgical resection and chemotherapy3,4. Despite the high incidence and 69 mortality of advanced CRC, few effective therapies have been approved in the past several 70 decades5. One barrier to the development of efficacious therapeutics is the biological 71 heterogeneity of CRC and its variable clinical course6. While landmark studies from The Cancer 72 Genome Atlas (TCGA) have defined the somatic mutational landscape within CRC, several 73 studies have shown that stromal signatures, including fibroblasts and cytotoxic T cells, are likely 74 the main drivers of clinical outcomes7–12. These findings suggest that the clinical phenotypes of 75 CRC and by extension, its tumor biology, is shaped by a complex niche of heterotypic cell 76 interactions within the tumor microenvironment (TME)8-12. 77 78 Bulk gene expression analyses by several independent groups have identified distinct CRC 79 subtypes13–15. Reflecting both the tumor and TME, an international consortium published the 80 Consensus Molecular Subtypes (CMS), which proposed four distinct subtypes of CRC15. 81 Unfortunately, the association between CMS and meaningful therapeutic response to specific 82 agents have shown inconsistent results across studies and CMS lacks a concordance between 83 primary and metastatic CRC tumors, limiting its utility thus far in clinical decision making16–23. As 84 a result, an improved CMS classification or an alternative classification system is required to 85 improve clinical utility. 86 87 To overcome the limitations of bulk-RNA sequence profiling, we utilized single-cell RNA 88 sequencing (scRNA-seq) to more thoroughly evaluate the CRC subtypes at the molecular level, 89 including within the context of the currently defined CMS classification. We dissected heterotropic 90 cell states of tumor epithelia and stromal cells, including a cancer-associated fibroblast 91 (CAF) population. The CAF population's clinical and prognostic significance became apparent 92 when CAF signatures were applied to large, independent CRC transcriptomic cohorts. 93