Inflammation gene signature to predict survival outcomes in hepatocellular carcinoma (HCC): A machine-learning approach.

Abstract

551 Background: Inflammation is closely associated with HCC tumorigenesis. Understanding transcriptomic features of inflammatory response will elucidate potential therapeutic targets that could improve survival outcomes. Herein, we study the association between inflammation-based genes and survival outcomes in early-stage HCC. Methods: Transcriptomic and clinical data of 249 patients with resectable stage I and II HCC were obtained from The Cancer Genome Atlas. 132 genes associated with acute and chronic inflammatory response were identified from the molecular signature database. Those were then used in a principal component analysis (PCA) to cluster patients into high and low inflammation groups using the median of the first PCA. Inflammation related differentially expressed genes (DEGs) between the two groups were then identified & were fit in Least Absolute Shrinkage and Selection Operator model to reduce the multicollinearity. Uni- and multi-variate Cox proportional hazard model was used to find the independent predictors of survival. PCA was applied to significant genes to calculate the inflammation score (IS) and the optimal cut-off point was chosen based on survival. Log-rank test and Kaplan Meier curves were used to correlate the IS with overall survival (OS) & disease-free survival (DFS). Pearson’s Chi-squared, Wilcoxon rank sum, and Fisher’s exact tests were used to correlate the inflammation groups with clinical variables. Gene set variation analysis was performed to investigate enriched pathways between groups using the cancer hallmark gene sets. Results: We identified 6 inflammation-related DEGs associated with survival including: ADAM8, CCR7, FFAR3, S100A9, TNF, and TREM1. Two of those were independent predictors of survival: TREM1 & FFAR3. Based on IS cut off of 0.29 we have grouped patients into inflammation high (n=91) and inflammation low (n=158) groups. The low-inflammation group had significantly better OS & DFS (P=0.00039, P=0.039) respectively. No significant association was found between the inflammation groups and multiple variables including hepatitis B and C, alcohol and non-alcohol fatty liver disease. Stage I HCC patients had low-inflammation scores and better survival outcomes compared to stage II, indicative of an association between inflammation and HCC evolution. Pathway enrichment analysis notably showed significant enrichment in signal transduction pathways in the high inflammation group including Wnt/β-catenin, Notch, PI3K-AKT-mTOR, Hedgehog, TGFβ and upregulated KRAS signaling. Conclusions: Findings from this analysis highlight the close association between inflammation and survival outcomes in early-stage HCC and support the utility of an inflammation-based gene signature to predict survival in this disease setting. Our work also sheds light on potentially targetable pathways linked to IS that deserve further exploration.