Differentially expressed RBP identification
In total, we evaluated the expression of 1542 different RBPs in 374 HCC tumors and 50 normal tissue samples [6]. Of these, we identified 81 differentially expressed RBPs, including 54 and 27 that were upregulated and downregulated, respectively (|log2FC| > 1.0 and P < 0.05) (Figure 1).
Functional enrichment analyses
GO and KEGG analyses were next used to assess the potential functional roles of up- and down-regulated RBPs in HCC patients. GO analyses revealed upregulated RBPs to be enriched for roles in mRNA metabolic processes, RNA catabolic processes, DNA methylation or demethylation, DNA modification, and mRNA catabolic processes (Figure 2A). In contrast, downregulated RBPs were enriched for roles in RNA catabolic processes, intracellular mRNA localization, translational regulation, 3'−UTR−mediated mRNA destabilization, and RNA phosphodiester bond hydrolysis (Figure 2B). With respect to molecular functions, upregulated RBPs were enriched in mRNA 3'−UTR binding, catalytic activity, acting on RNA, translation regulator activity, poly(U) RNA binding, and poly−pyrimidine tract binding (Figure 2A), whereas downregulated RBPs were enriched in mRNA 3'−UTR AU−rich region binding, AU−rich element binding, mRNA 3'−UTR binding, ribonuclease activity and double−stranded RNA binding (Figure 2B). Upregulated RBPs were additionally enriched in the cytoplasmic ribonucleoprotein granule, ribonucleoprotein granule, cytoplasmic stress granule, telomerase holoenzyme complex, and cytosolic large ribosomal subunit compartments (Figure 2A), while downregulated RBPs were primarily enriched in mRNA cap-binding complex, RNA cap-binding complex, endolysosome membrane, and apical dendrite compartments (Figure 2B). Upregulated RBPs were additionally enriched in the mRNA surveillance pathway, microRNAs in cancer, RNA transport, RNA degradation, DNA replication, and cysteine and methionine metabolism KEGG pathways (Table 1), whereas downregulated RBPs were enriched in the influenza A, mRNA surveillance, and Hepatitis C pathways (Table 1).
PPI network construction and analysis
We next utilized Cytoscape (3.7.1) to construct a PPI network based on the STRING database. The resultant network incorporated 66 nodes and 127 edges (Figure 3A). Key co-expressed modules within this network were then identified using the MCODE plugin (Figure 3B). Functional enrichment analyses revealed that hub RBPs within this network were enriched in mRNA catabolic processes, RNA catabolic processes, mRNA surveillance pathways, and ribosome pathways.
Identification of hub RBPs associated with HCC patient prognosis
We next randomly separated 343 total HCC patients in the TCGA-LIHC dataset that had survived for a minimum of 30 days into a training cohort (n = 172) and a test cohort (n = 171). These two patient cohorts were then used to conduct survival analyses, leading us to identify 22 hub RBPs that were associated with patient OS (Figure 4A). A further multivariate Cox regression analysis determined that seven of these hub RBPs (SMG5, BOP1, LIN28B, RNF17, ANG, LARP1B, NR0B1) were independently associated with HCC patient OS (Figure 4B).
Construction and validation of a hub RBP-based prognostic model
We next utilized these seven independent prognostic hub RBPs to construct a prognostic risk score model as follows: risk score =0.7291*ExpressionSMG5+0.4424*ExpressionBOP1+0.0610*ExpressionLIN28B+0.0936*ExpressionRNF17+(0.2779)*ExpressionANG+0.6005*ExpressionLARP1B+0.0731*ExpressionNR0B1. Risk scores for each patient in the training set were then calculated, and the Survminer R package was used to calculate the median risk score in this patient cohort. This median value was used to stratify patients into low- and high-risk groups, and survival outcomes between these groups were then compared via Kaplan-Meier survival and time-dependent ROC analyses. This analysis confirmed that the OS of HCC patients in the high-risk group was significantly reduced relative to that of patients in the low-risk group (Figure 5A), with an area under the ROC curve value of 0.801 for this seven RBP risk score model (Figure 5B), consistent with its moderate diagnostic performance. In Figure 5C, mRNA expression levels, survival status, and risk score values for patients in the low- and high-risk groups are shown. We then utilized this same risk score formula to analyze patients in the test cohort (n=171) (Figure 6A-C). Consistent with the above results, HCC patients in the low-risk group exhibited an OS that was significantly longer than that of patients in the high-risk group, with an area under the ROC curve of 0.676. This thus indicates that our prognostic model was able to successfully predict HCC patient survival outcomes.
Construction of a hub RBP-based prognostic nomogram
A nomogram incorporating the results of the above multivariate Cox regression analysis pertaining to the seven hub RBPs was next constructed and used to predict 1-, 3-, and 5-year HCC patient OS (Figure 7) in our training dataset. This analysis revealed that patient 1-, 3-, and 5-year OS declined as risk scores increased, consistent with our above results, confirming the prognostic value of this risk nomogram.
The relationship between risk scores and clinical parameters
Logistic regression analyses were used to assess the relationship between risk scores and HCC clinical characteristics, revealing that high risk scores were associated with low histological grade (G3-4 vs G1-2, OR=2.060) and high AFP levels (>20 ng/mL vs <=20 ng/mL, OR=1.986) (P<0.05). In contrast, these scores were unrelated to hepatitis status, vascular invasion, or alcohol intake (Table 2).
RBP risk scores independently predict HCC patient prognosis
We next conducted univariate Cox analyses or factors associated with prognosis in 226 patients that survived for a minimum of 30 days and for whom complete clinical data were available. These analyses revealed that cancer tissue stage, T stage, and risk scores were all associated with HCC patient OS (P < 0.001) (Figure 8A). Subsequent multivariate Cox analysis confirmed that the RBP risk score was an independent predictor of HCC patient OS, with a hazard ratio (HR) of 1.160 and a 95% confidence interval of 1.095-1.229 (P=4.305E-07)(Figure 8B).
Validation of hub RBP prognostic value
Lastly, the relationship between identified hub RBPs and HCC patient OS was evaluated using the Kaplan-Meier plotter database. This analysis confirmed that 4/7 hub RBPs (ANG, LIN28B, SMG5, and NR0B1) were significantly associated with HCC patient OS, with respective P-values of 0.017, 0.013, 0.002, and 0.003 (Figure 9A-D).
SMG5-PCG co-expression network analysis
A correlation analysis of SMG5 and PCGs revealed that there were 3756 total PCGs correlated with SMG5, of which 7 were negatively correlated and 3749 were positively correlated. The top five PCGs positively correlated with SMG5 expression were ISG20L2, DENND4B, UBQLN4, PI4KB, and SLC39A1 (Table 3), while the top five PCGs negatively correlated with SMG5 expression were TTC36, CLEC4M, FCN2, MFSD2, and MT1X (Table 3). GO and KEGG analyses of these SMG5-related PCGs revealed them to be primarily enriched in the mTOR, AMPK, VEGF, and hepatitis B signaling pathways, indicating that they are closely related to tumor development.