Development and validation of a novel pseudogene pair-based prognostic signature for prediction of overall survival in patients with hepatocellular carcinoma

Hepatocellular carcinoma (HCC) is the most prevalent subtype of hepatic malignancies worldwide, accounting for 90% of primary liver cancers [1]. HCC is particularly prevalent in developing countries, particularly in East Asia and sub-Saharan Africa when compared with developed countries [2, 3]. Previous epidemiological studies have reported there to be approximately 250,000 new subjects and approximately 500,000 to 600,000 deaths due to HCC annually [1]. Despite the rapid advances in imaging techniques, surgical resection, and comprehensive therapy to treat HCC in recent years, the 5-year survival rate of HCC patients remains poor [4]. Therefore, it is necessary to uncover novel prognostic signatures that may identify groups of patients with a high risk of poor survival.

Pseudogenes are non-coding genes similar to their corresponding homologous protein-coding genes and long been considered ‘gene fossils’ or ‘junk genes’ because they do not encode functional proteins due to different kinds of mutations in the coding sequences [5]. In recent years, accumulating evidence has overwhelmingly revealed that individual pseudogenes involve in multiple human diseases including malignancy [6]. Multiple tumor-related pseudogenes have been confirmed as predictors for both diagnosis and prognosis. For example, the pseudogene DUXAP10 was found to be upregulated in several kinds of malignancies and could serve as a novel biomarker with high diagnostic and prognostic value for many cancers [7]. In HCC, high expression of the pseudogene ANXA2P2 has been found to be related to a worse prognosis. ANXA2P2 could be a novel predictive factor for evaluating the risk of recurrence or metastasis in HCC patients [8]. However, the molecular characteristics of pseudogene interactions and the prognostic value of pseudogenes in HCC have not been comprehensively explored.

Numerous studies have established mRNA expression profile-based signatures for outcome prediction in HCC patients [9‐14]. However, these models have been failed to utilize clinically due to the diversity of data types, batch effects, and subsequent normalization of expression data, which poses a daunting obstacles for data processing given the possible biological heterogeneity among various data series and technical differences across different platforms [15]. Recently, a novel algorithm according to the relative orders of gene expression levels was established to remove the disadvantages of mRNA/miRNA expression normalization and scaling and has demonstrated robust results in previous studies [16, 17].

In this study, we identified 19 pseudogene-pairs based on univariate and LASSO regression analyses, and established a risk score model to predict the outcome of patients with HCC. Time-dependent receiver operating characteristic (ROC) curves were used to investigate the model’s performance in predicting the 1-, 3-, and 5-year overall survival (OS) of patients with HCC in two cohorts. Further, subgroup analysis was implemented to explore the prognostic performance of the signature in different stages, grades, sexes, and age groups. The C-index of the prognostic signature was compared with several established risk models. Pearson correlation analyses were done between the included pseudogenes and protein-coding genes based on their expression levels. Subsequently, we explored the biological functions and possible signaling pathways associated with the identified pseudogenes in the risk signature.

HCC remains a major and growing global public health challenge. However, the molecular pathogenesis of HCC is not fully understood. Given the extensive heterogeneity of HCC, there is a need for more accurate individualized prognostic signatures. Recently, increasing evidence has demonstrated that abnormal expression of pseudogenes is involved in multiple diseases, including malignancy [6]. For example, in HCC, upregulation of the pseudogene RP11-564D11.3 has been found to be associated with adverse survival [20]. Numerous researches have built gene expression profile-based signatures for survival prediction in patients with HCC [9‐14]. However, previous reports aiming to build a prognostic model have focused on mRNAs, lncRNAs, and miRNAs, neglecting pseudogenes as potential biomarkers in HCC. Therefore, the development of a robust pseudogene pair signature contributes to clinical decision-making for individualized treatment of HCC patients.

In this study, we established a novel 19-pseudogene pair signature that could successfully classify patients into two groups with different OS. We found that patients in high-risk group had a worse survival rate than patients in the low-risk group in both cohorts. Subgroup analysis by age, family history, sex, grade, vascular invasion, and stage yielded the same conclusion. We found the signature to be a stable prognostic predictor for patients with HCC. Multivariate analyses demonstrated that the risk score may be a clinically independent prognostic predictor for HCC. The AUC values of the prognostic model for OS prediction also present excellent predictive performance in both cohorts. The signature was reproducible and robust in the independent validation cohort, demonstrating its value and effectiveness. These conclusions confirmed that the novel model could offer an accurate survival prediction for patients with HCC. Moreover, the C-index of our signature was larger than that of established signatures. We employed a more comprehensive and novel approach to develop a robust prognostic signature for HCC and successfully validated it in the ICGC cohort. Therefore, this novel prognostic model is accurate, robust, and interpretable.

Although numerous prognostic models have been established for the prediction of HCC survival [9‐14, 21, 22], these prognostic models have seldom been widely utilized clinically due to their need for proper data standardization across various expression profiles for further analysis [16, 17]. In this study, based on the relative orders of the mRNA expression, the signature was generated only by weight-pairwise comparison within a given sample without requiring for data normalization and can remove the batch effects between different platforms. Furthermore, the cutoff value derived from the risk score formula used in this study could be employed across multiple datasets, showing a great advantage when compared with previous models, and may be easily translated into clinical application. This novel algorithm has been validated to be accurate and robust in previous cancer-related reports [16, 17, 23, 24].

The identified pseudogene-related protein-coding genes were primarily involved in cytokine and chemokine receptor activity, and cancer-related pathways, such as T cell receptor signaling, NF-κB signaling, PD-L1 expression, and PD-1 checkpoint pathway in cancer. It has also been reported that IL-8 and IL-6, important chemokines, are involved in tumor angiogenesis, growth and metastasis, and can therefore act as vital chemokines for blood vessel formation in HCC [25‐27]. Previous studies have confirmed that T cells gather in the blood of HCC patients, and tumor necrosis factor (TNF), which is involve in the T cell receptor signaling pathway, was downregulated in sorafenib-treated HCC patients, demonstrating that the T cell receptor signaling pathway may also be involved in HCC [28, 29]. A previous study has provided evidence for the inhibitory effect of PPARα on HCC via the NF-κB signaling pathway [30]. IGFBP2 can serve as a new therapeutic target that activates the NF-κB-ZEB1 signaling axis and contributes to HCC tumorigenesis [31]. Expression of immune checkpoint molecules, such as PD-1/PD-L1, has been confirmed in HCC [32]. Furthermore, in September 2017, the FDA has granted accelerated approval to PD-1 checkpoint inhibitors for the treatment of HCC patients [33]. Therefore, the novel established pseudogene pair signature could be associated with HCC-related biological pathways and the functional dysregulation could be well associated with the survival of HCC patients.

The signature based on the relative expression ordering exhibited no difficulty in clinical transformation and application. For future study, if the expression matrix of 21 pseudogenes was obtained, a 19-pseudogene pair signature was therefore constructed. Furthermore, the signature only involves pairwise comparison within a given sample without the requiring for data normalization and batch effects from different laboratories and platforms. Based on the cutoff value derived from the risk score formula, patients can be grouped into different groups, and their prognosis can be predicted. Thus, the prognostic model can act as an individualized, single-sample prediction of outcome of HCC and can be easily translated to clinical application.

This is the first research to establish a pseudogene pair-based prognostic signature in HCC. However, this study also has limitations. First, prospective research is suggested to validate the prognostic function of the pseudogene pair signature. Moreover, to better understand the functional role of these pseudogene pairs in HCC, experimental studies investigating these pseudogenes should be carried out.

We developed and validated an accurate and novel robust pseudogene pair signature capable of accurately predicting the prognosis of HCC patients, with higher risk scores demonstrating adverse prognosis. The signature is reproducible and robust in an independent external cohort and outperforms other established signatures, demonstrating its value and effectiveness. Additionally, this signature could act as an encouraging independent prognostic predictor for HCC.