Consensus molecular subtypes of morphological regions on colorectal tumors

Abstract

Background The consensus molecular subtypes (CMS) of colon cancer integrated the results of several different subtyping efforts and led to the definition of four distinct groups with clinical relevance (CMS1-4). The connection between tumor morphology and molecular subtypes has been previously described qualitatively and quantitatively. Moreover, it has been suggested that variability in tumor sampling strategies would lead to less reproducible subtype classifications. To better understand the sources of heterogeneity and for stabilizing the CMS classifier, we performed a morphology-directed tumor sampling for RNA extraction and compared the predicted CMS with whole-tumor results. We choose to base our sampling on well-established tumor morphologies to improve the interpretability and lower selection variability.
Methods From 100 colon tumors (hospital cohort), consecutive sections were used for whole-tumor and morphological-region RNA extraction. The following tumor morphological regions have been manually annotated in the virtual slides and macro-dissected: complex tubular, desmoplastic, mucinous, papillary, serrated, and solid trabecular. Additionally, a number of tumor-adjacent normal and stroma regions have been marked. We used the CMScaller package for subtype assignment to both whole tumor and regional profiles, with n=1000 permutations and FDR=25%.
Results In total, 100 whole tumor and 152 regional transcriptomics profiles were obtained, respectively. No subtype was assigned to 22% and 23% of whole tumor and regional profiles. For whole tumor CMS1-3, there was a good concordance with the subtype assigned to at least one of its corresponding morphological regions (73%, 62% and 78%, respectively). However, CMS4 appeared to be less stable, with only 44% of the regions being CMS4, other 21% being CMS2. In one case, the whole tumor subtype (CMS2) did not correspond to any of the associated regions profiled (complex tubular - CMS1, mucinous - CMS4 and tumor-adjacent normal - CMS3). When considering the subtypes assigned to each morphological region, we observed that: Complex tubular regions were mostly labeled as CMS1 (41%) and CMS2 (32%); Desmoplastic regions were labeled as CMS4 (54%) and CMS1 (31%). In addition, the subtype of the desmoplastic region, when in combination with other region types, determined the whole tumor subtype. Serrated regions were mostly labeled as CMS2 (42%) and CMS3 (32%); All tumor-adjacent stroma was labeled as CMS4. Surprisingly, most of the mucinous regions were assigned to CMS4 (60%). The analysis of subtype molecular markers indicated that canonical Wnt is mostly activated in mucinous, serrated and stroma regions. Also, the EMT signature had highest values in desmoplastic, mucinous and stroma regions.
Conclusions The morphological regions may have a different molecular subtype than the whole tumor, indicating the presence of additional subtypes. The current gene expression classifiers are sensitive to tumor sampling protocol. Clear specification of region(s) used for RNA profiling is necessary for improving the stability of the molecular subtype classification and multi-region sampling opens the possibility for refining the CMS classification. The CMS4 EMT characteristics are most probably due to the desmoplastic reaction.