Practical and robust identification of molecular subtypes in colorectal cancer by immunohistochemistry

Abstract

URPOSE: Recent transcriptomic analyses have identified four distinct molecular subtypes of colorectal cancer (CRC) with evident clinical relevance. However, the requirement for sufficient quantities of bulk tumor and difficulties in obtaining high quality genome-wide transcriptome data from formalin-fixed paraffin-embedded tissue are obstacles towards widespread adoption of this taxonomy. Here, we develop an immunohistochemistry-based classifier to validate the prognostic and predictive value of molecular CRC subtyping in a multi-center study. EXPERIMENTAL DESIGN: Tissue microarrays from 1076 CRC patients from four different cohorts were stained for five markers (CDX2, FRMD6, HTR2B, ZEB1 and KER) by immunohistochemistry and assessed for microsatellite instability. An automated classification system was trained on one cohort using quantitative image analysis or semi-quantitative pathologist scoring of the cores as input, and applied to three independent clinical cohorts. RESULTS: This classifier demonstrated 87% concordance with the gold-standard transcriptome-based classification. Application to three validation datasets confirmed the poor prognosis of the mesenchymal-like molecular CRC subtype. In addition, retrospective analysis demonstrated the benefit of adding cetuximab to bevacizumab and chemotherapy in patients with RAS wild type metastatic cancers of the canonical epithelial-like subtypes. CONCLUSION: This study shows that a practical and robust immunohistochemical-assay can be employed to identify molecular CRC subtypes and uncover subtype-specific therapeutic benefit. Finally, the described tool is available online for rapid classification of CRC samples, both in the format of an automated image analysis pipeline to score tumour core staining, and as a classifier based on semi-quantitative pathology scoring.