Evaluation and prediction of hepatocellular carcinoma prognosis based on molecular classification

Prediction of hepatocellular carcinoma (HCC) prognosis faced great difficulty due to tumor heterogeneity. We aimed to identify the prognosis-associated molecular subtypes existing in HCC patients and construct an evaluation model based on identified molecular classification.

Non-negative matrix factorization consensus clustering was performed using 371 HCC patients from The Cancer Genome Atlas (TCGA) to identify molecular subtypes, based on the expression profile of the survival-associated genes. Signature genes for different subtypes were identified by Significance Analysis of Microarray and Prediction Analysis for Microarrays. Model for subtype discrimination and prognosis evaluation was established using binary logistic regression. The model and its clinical implications were further validated in GSE5436 cohort and Fujian cohort.

Based on TCGA data, we observed two molecular subtypes with distinct clinical outcomes including significantly different overall survival, tumor differentiation, TNM stage, and vascular invasion (all P<0.05). The existence of these two molecular subtypes was further validated in five other Gene Expression Omnibus datasets. Furthermore, we constructed an evaluation model based on six subtype signature genes, which can discriminate different subtypes with the cutoff of 0.385. Meanwhile, both Cox regression analysis and stratification analysis showed that the calculated continuous prognostic value could also effectively indicate HCC prognosis, regardless of patients' clinical conditions. The prognostic evaluation model was successfully validated in GSE54236 cohort and Fujian cohort.

Two prognostic molecular subtypes existed among HCC patients, which provided promising strategies for overcoming HCC heterogeneity and could be utilized in future clinical application for predicting HCC prognosis.