CMS-PDX: A 20-gene genomic panel to predict consensus molecular subtypes in patient-derived xenografts (PDX) of colorectal cancer

• Published in: Journal of clinical oncology Vol. 37; no. 4_suppl; p. 598
• Main Authors: Buechler, Steven Allen, Gokmen-Polar, Yesim, Badve, Sunil S.
• Format: Journal Article
• Language: English
• Published: 01.02.2019
• ISSN: 0732-183X; 1527-7755
• DOI: 10.1200/JCO.2019.37.4_suppl.598

Abstract only 598 Background: The consensus molecular subtypes (CMS1-4) partition colorectal cancer (CRC) into subgroups with varying prognoses and treatment responses. We previously reported the ColotypeR-CMS genomic scores (4 scores defined with 5 genes each) that accurately define the CMS subtypes of primary CRC tumors. There is also a critical need to develop an epithelial cell based algorithm for CMS subtyping of PDX models. Methods: To define scores predicting CMS subtypes for PDX models, genes that both predict CMS subtypes in primary CRC tissue and that have linearly correlated expression in primary tumor and tissue-matched PDX models, were identified in GSE35144 (N = 64, Affymetrix microarray). Using the identified genes, a 20-gene CMS-PDX score was developed using multistate gene methodology. ROC curve (AUC) analysis was used to measure the ability of CMS-PDX scores to predict subtypes obtained with CMScaller (473 genes), a preclinical CMS classifier. The ability of CMS subtypes to determine cetuximab response in PDX cohorts was analyzed in GSE78806 (N = 121, Affymetrix microarray) and GSE76402 (N = 529, Illumina beadarray). Results: In GSE78806 (N = 121), CMS1-4 subtypes obtained with CMScaller were predicted by CMS-PDX scores with AUC 0.88, 0.94, 0.81, 0.86, respectively; in GSE78806 CMScaller failed to classify 32% of samples. In CRC-derived PDX models, higher values of the CMS-PDX CMS2 score were predictive of reduced tumor volume following cetuximab treatment (p = 1.7 x 10 -5 in GSE78806, p < 2.2 x 10 -16 in GSE76402, by linear regression.) In contrast, higher values of the CMS1 score were predictive of increased tumor volume (p = 3.8 x 10 -5 in GSE78806, p = 2.9 x 10 -10 in GSE76402.) Change in tumor volume was independent of the CMS-PDX CMS4 score in both PDX cohorts. Conclusions: We showed that the 20-genes CMS-PDX scores are predictive of CMS subtypes obtained with CMScaller. Moreover, CMS2, by CMS-PDX scores, was highly predictive of sensitivity to cetuximab treatment in CRC PDX models.