Biostatistical Analysis of DNA Methylation Profiling in Ovarian Cancer

Abstract

Ovarian cancer is the most lethal gynaecological cancer. Although having good response to chemotherapy, the majority of the patients with advanced disease will eventually relapse. Aberrant DNA methylation in tumours has been proposed as biomarkers to predict patients’ clinical outcome and response to chemotherapy. An algorithm, Methylation Linear Discriminant Analysis (MLDA), was developed for large-scale methylation analysis using differential methylation hybridsation (DMH). MLDA identified loci differentially methylated between cisplatin sensitive and resistant derivatives of an ovarian tumour cell line with 89% accuracy and showed hypermethylation, rather than hypomethylation, predominantly occurred during the acquisition of cisplatin resistance. Customised microarrays targeting promoter CpG islands in 10 key signaling pathways were designed for DMH analysis. Based on the power analysis epithelial ovarian tumours (screening study n=120, validation study n=61) prospectively collected through a cohort study, were firstly analysed by DMH at 302 loci spanning 189 promoter CGIs at 137 genes in the Wnt pathways for the association with progression free survival (PFS). Increased methylation of 6 loci, at FZD4, FZD9, DVL1, NFATC3, ROCK1 and NKD1 genes, were associated with shorter PFS independent from clinical parameters. A multivariate Cox model incorporates only NKD1 and DVL1, identifying two groups differing in PFS (HR=2.72; permutation test p = 4x10-3). Consistent with epigenetic regulation, reduced expression of FZD4 and DVL1 is associated with poor relapse free survival in an independent cohort (p<0.05,n=321). Analysis in further 9 pathways/families found 6 more independent biomarkers relevant to PFS at PIK3R5, AKT1 and VEGFB from AKT/mTOR pathway, PRDX2 and TR2IT2 from Redox pathway and MLH3 from MMR system. The study shows DNA methylation changes are involved in acquired drug resistance, and demonstrates the importance of methylation at multiple promoter CGIs in key signaling pathways, especially in the Wnt pathway, for predicting clinical outcome in ovarian cancer and their potential as stratification biomarkers in future clinical studies for personalised treatment.