Nowadays, hepatocellular carcinoma still threatens millions of people’s life every year as the sixth prevalent malignant disease [16]. The current therapy on HCC mainly consists of surgery, ablation, interventional therapy and systematic treatment [17]. Unfortunately, these methods have their limitation because of the heterogeneity of liver cancer. Liver transplantation [18] is supposed to be most effective in countering HCC, however, the long waiting time and the contradiction between shortage of donors and huge population of patients restrict its application. Thus, early detection of the disease is still highlighted in HCC management, nevertheless effective biomarkers are lacked. The dysfunction of various RNA is pivotal in tumorigenesis, which exists in many cancer types [19–20]. RBPs are the main actor in regulating the performance of RNA at post transcription level [21] which accomplished by a series of modification including RNA alternative splicing, polyadenylation, localization and stabilization, etc. Existing literature have concluded that RBPs are crucial contributors in various phenotypes of malignant disease, like proliferation [22], apoptosis escape [23], angiogenesis [24], etc. However, the role of RBPs in HCC has not been systemically explored. So as to investigate the potential of RBPs as biomarkers in HCC, we analyzed the transcriptome and clinical data of LIHC patients from TCGA. According to our filtration standard, there were 330 DERBPs recognized between normal and cancerous samples. Then the biological processes and metabolic pathways involved these DERBPs were globally analyzed and the interaction between them was displayed by PPI network. After combining the sequencing and clinical data, survival-related RBPs were screened by univariate COX regression and KM analysis. These RBPs were next enrolled to LASSO and multivariate COX regression to get the core RBPs behind them. At last we build our six RBPs-based risk model, which had well performance in forecasting LIHC patients’ prognosis.
As what was shown in bar and bubble plot, GO functional analysis pointed out that a large proportion of DERBPs took part in non-coding RNA metabolic processing, non-coding RNA processing, RNA 3’-end processing, tRNA metabolic process, RNA splicing, ribonucleoprotein complex biogenesis, RNA catabolic process, regulation of translation, regulation of cellular amide metabolic process, RNA phosphodiester bond hydrolysis and defense response to virus, etc. In line with other studies, alternative splicing of RNA, RNA processing and translational regulation were critical in the occurrence and progression of malignant diseases [21, 25]. The elevated expression of RBP KHSRP can facilitate the proliferation of small cell lung cancer by promoting the maturation of miR-26a, which suppressed the function of PTEN subsequently [26]. Aberrant change of splicing target of RBP SRSF1 induces the proliferation of HCC via generating the anti-apoptotic isoforms of target genes [27]. By interacting with the internal ribosome entry site of target mRNA laminin B1 at translation level, RBP La protein increases the expression of laminin B1 and therefore promotes the invasiveness of HCC [28]. Ribonucleoprotein is the functional state of RBP, its formation relies on the specific domains on RBP. The dysfunction of ribonucleoprotein is reported in various cancer types [6]. In addition, the outcome of KEGG pathway enrichment indicated most of DERBPs were involved in pathways associated with RNA spliceosome, mRNA surveillance, RNA degradation, Ribosome biogenesis in eukaryotes and RIG − I−like receptor signaling pathway, etc.
We also investigated the co-expressional relations of these DERBPs and the STRING returned us a PPI network with 311 nodes. The top three important modules with MCODE score over 10 were further analyzed by GO and KEGG. According to the report, module 1 was mainly involved in ribonucleoprotein complex biogenesis, various kinds of RNA splicing and processing. Module 2 was associated with defense response to virus response, interferon production and the signaling of interferon, etc. In the meanwhile, module 3 majored in translational determination and elongation, process of mitochondrial translation, etc. In agreement with previous studies, some of these RBPs functioned in liver cancer. BOP1 was frequently expressed in HCC and promoted the invasiveness of cancer cells, which indicated poor prognosis in most cases [29]. High expression level of IFIT3 amplified the therapeutic effect of interferon-α in combating HCC, thus can be utilized to screen LIHC patients adapted to interferon treatment [30]. EIF5A2 was associated with poor survival and can advance the proliferation of HCC by accelerating the metabolic rate of cancer cell [31]. These researches corroborate the important role of RBPs in HCC.
In order to further investigate the relationship between these DERBPs and outcome of patients, we performed uni-, multi-COX regression and LASSO regression. There were six RBPs highlighted as core genes, including UPF3B, MRPL54, ZC3H13, DHX58, PPARGC1A and EIF2AK4. Previous study has demonstrated that ZC3H13 was associated with OS of LIHC patients [32] and PPARGC1A was related to people’s susceptibility of HCC [33], however the roles of the others in HCC are yet to be confirmed. Hereafter, these six RBPs together with their coefficient returned by multi-step COX regression were enrolled into our risk model, which was trained in TCGA datasets. As the result of time-dependent ROC test, our model showed well performance in distinguishing patients of different prognosis in both training and validating cohorts. Furthermore, a nomogram based on these hub RBPs were plotted for the purpose of approximately calculating the likelihood of patients’ survival. At last, the reliability of these six RBPs as prognostic genes were verified in Kaplan-Meier plotter and HPA database and the outcome agreed with our study.
Despite the encouraging results of our study, there were some unavoidable limitations. First, we got all the data in our study from public databases like TCGA and ICGC and we did not verify the prognostic model in other datasets or our own reliable prospective clinical trial. Second, some of the samples bonded with incomplete clinical information were abandoned during the analysis which may increase the bias error of our study.