USP19 and RPL23 as Candidate Prognostic Markers for Advanced-Stage High-Grade Serous Ovarian Carcinoma

Abstract

Ovarian cancer is one of the leading causes of deaths among patients with gynecological malignancies worldwide. In order to identify prognostic markers for ovarian cancer, we performed RNA-sequencing and analyzed the transcriptome data from 51 patients who received conventional therapies for high-grade serous ovarian carcinoma (HGSC). Patients with early-stage (I or II) HGSC exhibited higher immune gene expression than patients with advanced stage (III or IV) HGSC. In order to predict the prognosis of patients with HGSC, we created machine learning-based models and identified USP19 and RPL23 as candidate prognostic markers. Specifically, patients with lower USP19 mRNA levels and those with higher RPL23 mRNA levels had worse prognoses. This model was then used to analyze the data of patients with HGSC hosted on The Cancer Genome Atlas; this analysis validated the prognostic abilities of these two genes with respect to patient survival. Taken together, the transcriptome profiles of USP19 and RPL23 determined using a machine-learning model could serve as prognostic markers for patients with HGSC receiving conventional therapy.

Keywords: high-grade serous carcinoma; machine learning; next-generation sequencing; ovarian cancer; prognostic marker.

Figure 1

Transcriptome profiles of 51 patients with HGSC. (A) Heatmap of upregulated genes in patients with early-stage HGSC. The significant enrichment of individual pathways among the differentially expressed genes is depicted as a bar plot on the right. (B) Volcano plot depicting differentially expressed genes between different stage groups (I or II vs. III or IV).

Figure 2

Screening for the prognostic markers and patterns of USP19 and RPL23 expression. (A) Schematic depicting the workflow used for creating the random forest model and its validation. (B) Progression-free survival (PFS) in TCGA HGSC dataset (stage III or IV, n = 208) for validation of the predictive model. (C) The expression of USP19 mRNA in the recurrence and non-recurrence groups (t-test p = 0.015). (D) Graph depicting the correlation between PFS and USP19 expression in the training set. (E) Correlation between PFS and USP19 expression in the TCGA HGSC validation set. (F) The expression of RPL23 mRNA in the recurrence and no-recurrence groups (t-test p = 0.020). (G) Correlation between PFS and RPL23 expression in our training set. (H) RPL23-based evaluation of prognosis in the TCGA HGSC validation set.

Figure 3

Functional association between USP19 and DNA double-strand (DSB) repair genes such as BRCA1/2. (A,B) Co-expression pattern among USP19, 18 DSB repair genes, and two candidate genes in our patients with HGSC and TCGA datasets, respectively. (C) Violin plot depicting USP19 expression according to the BRCA1/2 mutation in our patients. (D) Sub-network of functional associations among USP19 and hereditary pathogenic genes responsible for causing breast and ovarian cancer syndrome. Node color represents the characteristics of genes; red: USP19, green: DSB repair genes, blue: gene connecting USP19 to DSB repair genes.