Prognostic value analysis of cholesterol and cholesterol homeostasis related genes in breast cancer by Mendelian randomization and multi-omics machine learning

Abstract

Introduction: The high incidence of breast cancer (BC) prompted us to explore more factors that might affect its occurrence, development, treatment, and also recurrence. Dysregulation of cholesterol metabolism has been widely observed in BC; however, the detailed role of how cholesterol metabolism affects chemo-sensitivity, and immune response, as well as the clinical outcome of BC is unknown.

Methods: With Mendelian randomization (MR) analysis, the potential causal relationship between genetic variants of cholesterol and BC risk was assessed first. Then we analyzed 73 cholesterol homeostasis-related genes (CHGs) in BC samples and their expression patterns in the TCGA cohort with consensus clustering analysis, aiming to figure out the relationship between cholesterol homeostasis and BC prognosis. Based on the CHG analysis, we established a CAG_score used for predicting therapeutic response and overall survival (OS) of BC patients. Furthermore, a machine learning method was adopted to accurately predict the prognosis of BC patients by comparing multi-omics differences of different risk groups.

Results: We observed that the alterations in plasma cholesterol appear to be correlative with the venture of BC (MR Egger, OR: 0.54, 95% CI: 0.35-0.84, p<0.006). The expression patterns of CHGs were classified into two distinct groups(C1 and C2). Notably, the C1 group exhibited a favorable prognosis characterized by a suppressed immune response and enhanced cholesterol metabolism in comparison to the C2 group. In addition, high CHG score were accompanied by high performance of tumor angiogenesis genes. Interestingly, the expression of vascular genes (CDH5, CLDN5, TIE1, JAM2, TEK) is lower in patients with high expression of CHGs, which means that these patients have poorer vascular stability. The CAG_score exhibits robust predictive capability for the immune microenvironment characteristics and prognosis of patients(AUC=0.79). It can also optimize the administration of various first-line drugs, including AKT inhibitors VIII Imatinib, Crizotinib, Saracatinib, Erlotinib, Dasatinib, Rapamycin, Roscovitine and Shikonin in BC patients. Finally, we employed machine learning techniques to construct a multi-omics prediction model(Risklight),with an area under the feature curve (AUC) of up to 0.89.

Conclusion: With the help of CAG_score and Risklight, we reveal the signature of cholesterol homeostasis-related genes for angiogenesis, immune responses, and the therapeutic response in breast cancer, which contributes to precision medicine and improved prognosis of BC.

Keywords: Mendelian randomization; breast cancer; cholesterol homeostasis; immune microenvironment; machine learning method; prognosis prediction.