Identification of a Cancer Stemness Driven Prognostic Model and Therapeutic Drug Targets by Machine Learning in Hepatocellular Carcinoma

Abstract

Background: Hepatocellular carcinoma (HCC) is one of the most common malignancies. Cancer stem cells (CSCs), characterized by self-renewal and drug-resistance, play an important role in the development and progression of diverse cancers, but the underlying association of HCC and CSCs is not fully researched.

Methods: Transcriptome and clinical data of 903 patients in four independent HCC cohorts were obtained from TCGA, ICGC, and GEO databases. We evaluated the stemlike index for each patient to reflect the cancer stemness by using one-class logistic regression (OCLR) algorithm. GISTIC 2.0, Maftools and GSVA were used to reveal the association between the stemness index and genomic variation and biological processes in HCC. The differential expression analysis, univariate Cox analysis and LASSO analysis were used to identify the prognostic stemness signatures. The HCC stemness-related risk score (HCSRS) was constructed to quantify stemness levels of individual tumors. Based on HCSRS, the nomogram was established for HCC prognosis in a quantitative approach. Additionally, single sample Gene Set Enrichment Analysis (ssGSEA) algorithm was used to evaluate the immune infiltration levels in HCC, and drug response analysis was adopted to identify potential agents with drug sensitivity in high-HCSRS score patients.

Results: The stemness index in HCC tissues was significantly higher than that in normal tissues, and there was a significant positive correlation with pathological grade. Patients with high stemness index showed higher somatic mutation frequency, tumor mutation load, and copy number variation frequency, and were significantly enriched in tumor-related signaling pathways. Meanwhile, the 7-gene based HCSRS model that was trained and validated in 4 independent cohorts exhibited high predictive significance for overall survival (OS). Further analysis revealed that patients with high HCSRS possessed higher immunosuppression status, characterized by significantly decreased infiltration of anti-tumor immune cells (CD8 T cells, cytotoxic T cells, DC cells, NK cells, etc.) and exhausted CYT responses. At last, a total of twelve agents were identified to have potential therapeutic effects in high-HCSRS patients.

Conclusion: In current study, we systematically analyzed the potential relationship of HCC stemness with genomic variation, tumor microenvironment and biological processes, provided a theoretical basis for individualized treatment of HCC patients.