Identification and validation of the fibrosis-related molecular subtypes of hepatocellular carcinoma by bioinformatics

Abstract

We identified novel Molecular subtypes according to the expression of fibrosis-related genes (FRGs) and constructed a prognostic model using different expression genes (DEGs) for patients with Hepatocellular carcinoma (HCC). We downloaded the clinical data and transcriptome data of HCC from The Cancer Genome Atlas (TCGA) database, Gene Expression Omnibus (GEO) database, and International Cancer Genome Consortium (ICGC) database. We identified two fibrosis-related molecular subtypes of HCC by consensus unsupervised clustering analysis. Interestingly, these two molecular subtypes significantly differed in overall survival (OS) and clinical characteristics. Besides, the most minor absolute shrinkage and selection operator (Lasso) and multivariate Cox regression analysis were performed to develop a novel prognostic model by three genes (including KPNA2, LPCAT1, and AKR1D1). There was a statistically significant difference in OS between the high-risk and low-risk groups. The area under the ROC curve (AUC) of OS in 1-, 3-, and 5-year were satisfactory. Besides, the risk score was connected with critical clinical characteristics and could be an independent factor in predicting prognosis. Then, the nomogram was built by incorporating risk scores with clinical parameters. Additionally, the risk score was remarkedly correlated with TME and drug susceptibility. Finally, the results of H&E staining and immunohistochemistry of Ki67 showed that the tumor of higher-risk patients are more malignant. The FRGs-based subtype and signature explain the HCC heterogeneity, which might provide a new method to develop a more efficient treatment.

Keywords: Drug susceptibility; Fibrosis; HCC; Immunotherapy; Prognosis model.

Fig. 1

FRG subtypes and clinical characteristics, tumor immune cell microenvironments of two distinct subtypes of samples. a Consensus among clusters for each category number k. b Consensus matrix heat-map defining two clusters (k = 2). c Comparison of OS among two subtypes. d PCA analysis of transcriptomes in the two subtypes. e Comparison of clinical information between the fibrosis-related clusters. f Distribution of 22 types of immune cells among these two molecular subtypes. g Expression levels of PD-L1 in the two subtypes

Fig. 2

Development of a Novel Prognostic Signature. a The GO analysis. b The KEGG analysis. c Coefficients of DEGs shown by lambda parameter. d Partial likelihood deviance versus log (lambda) drawn by LASSO algorithm and cross-validation. e Interactions in HCC. Green and purple represent favorable and risk genes, respectively. f Expression differences of model genes. g The K-M curve of model genes. h Immunohistochemical staining protein level from the HPA database

Fig. 3

The overall survival analysis. a–d The K-M curve analysis in training, testing, all, validating cohort. e–h The AUC values of OS at 1-, 3- and 5- year. i–l The expression heatmap, the distribution of risk score and survival time

Fig. 4

The relationship between risk score and clinical characteristics. a The heatmap among this signature, clinical features and risk score in TCGA and GEO. b Associations of the risk score and clinical characters in TCGA and GEO. c K-M curves of the effect of clinical features on the prognosis of HCC in TCGA and GEO. d The heatmap in ICGC. e The association in ICGC. f K-M curves in ICGC

Fig. 5

Construction of a 1-, 3-, 5-year OS nomogram. a Nomogram for predicting the in the all set. b Calibration curves of the nomogram for predicting OS. c–f ROC curves in the all, training, testing, ICGC sets

Fig. 6

The evaluation of the immunocytes and N6-methyladenosine. a Correlations between risk score and immune. b Correlations between immune cells and three genes. c, d Expression of immune checkpoints and immune related pathways in the high and low-risk groups

Fig. 7

Comprehensive analysis in HCC. a The waterfall plot of somatic mutation features established of high-risk and low-risk patients. c TMB in different risk-score groups. d Relationships between risk score and CSC index. e Relationships between risk score and drug sensitivity

Fig. 8

Verification of the signature in our cohort. a The heatmap among clinical features and risk score. b Associations of the risk score and clinical characters (Stage, Grade, Fibrosis). c, d The immunohistochemistry of Ki67 and HE staining in HCC tissues in patient with lowest risk score and highest risk score