Genetic association of tertiary lymphoid structure-related gene signatures with HCC based on Mendelian randomization and machine learning and construction of prognosis model

Tertiary lymphoid structures (TLS) are formed in numerous cancer types. However, their value and significance in hepatocellular carcinoma (HCC) is unclear. We performed differential genes expression analysis of TLS-related Genes (TLSG) based on The Cancer Genome Atlas (TCGA) database, and performed Mendelian randomization (MR) analysis using expression quantitative trait loci, and then took their intersecting genes. A TLSG prognostic signature (TLSGPS)-based risk score was constructed using Least Absolute Shrinkage and Selection Operator (LASSO), univariate and multivariate COX regression analysis, and survival analysis was then performed. We used the International Cancer Genome Consortium for outside validation. We also performed biological function, tumor mutational burden, immune infiltration, single-cell analysis, CeRNA and drug sensitivity analysis based on TLSGPS. Three TLSGs (HM13, CSTB, CDCA7L) were identified to construct the TLSGPS, which showed good predictive ability and outperformed most prognostic signatures. MR suggested that HM13 (OR = 0.9997, 95 %CI: 0.9994–0.9999, P = 0.014) and CSTB (OR = 0.9997, 95 %CI: 0.9995–0.9999, P = 0.048) were negatively correlated with the risk of HCC onset, while CDCA7L (OR = 1.0004, 1.0001–1.0007, P = 0.0161) was the opposite. The differences in biological functions between the TLSGPS-based high-risk group (HRG) and low-risk group (LRG) involved cell proliferation, differentiation, and drug metabolism. HRG plus high mutations exhibited extremely poor survival. HRG had higher abundance of immune cell-oncogenic phenotypes, higher immune escape ability, and greater sensitivity to Afatinib, Dasatinib, and Gefitinib. 3 TLSGs identified by machine learning and MR can predict the onset, prognosis and clinical treatment of HCC patients, and had significant genetic association with HCC.