NanoCMSer: A Clinically Applicable Tool for Stratification of Colorectal Cancer Samples [March2024]

Colorectal cancer (CRC) is a significant contributor to global cancer-related mortality, emphasizing the critical need for advanced predictive biomarkers to guide treatment decisions. In our pursuit of identifying novel biomarkers, as part of an international consortium, we have previously categorized CRCs into four transcriptome-based consensus molecular subtypes (CMS1-4). These subtypes show promise for predicting outcomes and facilitating biological investigations. To streamline the integration of CMS classification into clinical practice, where routinely formalin-fixed paraffin-embedded (FFPE) samples are used, we developed NanoCMSer. This NanoString-based CMS classifier reliably stratifies both fresh-frozen (FF) and FFPE samples using a concise panel of just 55 genes. Employing domain adaptation methods during training enhanced the generalizability of the classifier, resulting in high accuracy for both FF and FFPE samples. For FF NanoString-based datasets, an accuracy of 95% and for FFPE NanoString-based datasets an accuracy of 92% was achieved, which is the highest reported for FFPE tissues. Furthermore, using the NanoCMSer, a 96% accuracy across a comprehensive collection of 1,976 RNAseq-based samples from 23 datasets aggregated in this study was determined. Additionally, our findings highlight that even with a limited number of genes, the biological relevance of predictions at the individual sample level can be still identified, with distinct CMS biology readily recognizable upon conducting enrichment analysis. Furthermore, our observations revealed a substantial difference in 5-year recurrence free survival (RFS) for CMS2 and CMS3 patients in stage III compared to II (CMS2 21% more recurrences, CMS3 31% more recurrences), while this difference is less pronounced for CMS1 and CMS4 with 11% and 9%, respectively. We posit NanoCMSer as a robust tool with potential applications to assess the CMS status of CRCs in both tumor biology and clinical practice, accessible through the NanoCMSer R package.