Poor-prognosis young-onset colorectal cancer is defined by the mesenchymal subtype and can be predicted by integrating molecular and histopathological characteristics

Abstract

Background: Young-onset colorectal cancer (CRC), affecting individuals <50 years of age, presents a significant health threat worldwide. The molecular and clinical characteristics of young-onset CRC are poorly understood, complicating the development of effective biomarkers for precision oncology. This study aimed to dissect age-dependent molecular heterogeneity of CRC and establish a model for identifying high-risk young-onset patients. Methods: We analyzed clinical data for 564 439 patient samples across three large cohorts. For molecular characterizations, a subset of 1874 patient samples was used. A deep learning framework was used to analyze hematoxylin-eosin-stained whole-slide images to quantify Shannon diversity indices (SDIs). Subsequently, a multivariate model, integrating SDI, microsatellite status and promoter methylation of miR-200s, was developed for predicting the consensus molecular subtype (CMS)4-mesenchymal subtype, followed by internal and external clinical validations. Results: Young-onset CRC patients exhibited better overall survival but worse relapse-free survival and higher metastasis rates compared with late-onset cases. Molecular subtyping analysis found that young-onset CRC also comprises the same four subtypes (CMS1-4), but the prevalence differs from late-onset CRC. Stratified analysis suggested that the poor outcomes in young-onset CRC were due to higher prevalence of the CMS4-mesenchymal subtype. To predict CMS4, we established an effective risk-scoring model (area under the curve = 0.87) combining molecular and histological markers, with multiple independent validations. Conclusions: CRC shows age-dependent molecular heterogeneity, with young-onset cases more frequently presenting the CMS4 subtype. To predict CMS4, we developed and validated a robust risk-scoring model integrating molecular and histological markers, offering a new translatable tool for more optimized management of young-onset patients.

Feng GAO

Professor

My research leverages AI and big data to improve diagnostics, prognostics, and ultimately, outcomes in cancer and other biomedical fields.