Predictive modeling of gene mutations for the survival outcomes of epithelial ovarian cancer patients

Abstract

Epithelial ovarian cancer (EOC) has a low overall survival rate, largely due to frequent recurrence and acquiring resistance to platinum-based chemotherapy. EOC with homologous recombination (HR) deficiency has increased sensitivity to platinum-based chemotherapy because platinum-induced DNA damage cannot be repaired. Mutations in genes involved in the HR pathway are thought to be strongly correlated with favorable response to treatment. Patients with these mutations have better prognosis and an improved survival rate. On the other hand, mutations in non-HR genes in EOC are associated with increased chemoresistance and poorer prognosis. For this reason, accurate predictions in response to treatment and overall survival remain challenging. Thus, analyses of 360 EOC cases on NCI's The Cancer Genome Atlas (TCGA) program were conducted to identify novel gene mutation signatures that were strongly correlated with overall survival. We found that a considerable portion of EOC cases exhibited multiple and overlapping mutations in a panel of 31 genes. Using logistical regression modeling on mutational profiles and patient survival data from TCGA, we determined whether specific sets of deleterious gene mutations in EOC patients had impacts on patient survival. Our results showed that six genes that were strongly correlated with an increased survival time are BRCA1, NBN, BRIP1, RAD50, PTEN, and PMS2. In addition, our analysis shows that six genes that were strongly correlated with a decreased survival time are FANCE, FOXM1, KRAS, FANCD2, TTN, and CSMD3. Furthermore, Kaplan-Meier survival analysis of 360 patients stratified by these positive and negative gene mutation signatures corroborated that our regression model outperformed the conventional HR genes-based classification and prediction of survival outcomes. Collectively, our findings suggest that EOC exhibits unique mutation signatures beyond HR gene mutations. Our approach can identify a novel panel of gene mutations that helps improve the prediction of treatment outcomes and overall survival for EOC patients.

Fig 1. Comparison of gene mutations between alive and deceased EOC patients.

A) The percentages of alive and deceased patients that contained each of 31 gene mutations were determined. 159 patients were alive, and 201 patients were deceased. B) The ratios of the percentages of alive to deceased patients were calculated. The bars indicate a fold change in the percentage of alive patients compared with deceased patients in each gene.

Fig 2. Kaplan-meier survival analysis of 360 EOC patients.

Kaplan-Meier survival analysis was performed on the basis of stratified patient populations containing mutations in positive genes (BRCA1, NBN, BRIP1, RAD50, PTEN, and PMS2) (A), negative genes (FANCE, FOXM1, KRAS, FANCD2, TTN, and CSMD3) (B), and HR-deficient genes (BRCA1, BRCA2, ATM, ATR, BRIP1, CDK12, CHEK2, FANCD2, FANCE, MRE11, NBN, PALB2, and RAD50) (C). The number of patients stratified (N) was also shown.

Fig 3. Mutation frequency in EOC populations.

A) Gene mutation spectrums of BRCA1- and BRCA2-mutated EOC. The percentages of 30 gene mutations are displayed in all, BRCA1-mutated, and BRCA2-mutated EOC cases. B) Gene mutation spectrums of EOC in alive and deceased patients. The percentages of 30 gene mutations are displayed for alive and deceased EOC patients. TP53 mutations are not included in the pie charts because these mutations are present in 96.4% of patients.

Fig 4. Heat map matrix and Eigenvectors of Pearson correlation analysis of 31 gene mutations and the alive outcome in 360 EOC patients.

Pearson correlation analysis was performed to make pair-wise comparisons to identify correlations among variables (alive outcome and 31 gene mutation). A) The correlation matrix is displayed as a heatmap. The correlations among variables are sorted to show hierarchical clustering. B) The Eigenvectors of Pearson correlations are shown to cluster variables and identify their relatedness on the PC1 versus PC2 scatter plot. The blue dash circle highlights the close relationship among the alive outcome and gene mutations.