Discussion
Main findings. This study demonstrates that there are measurable differences in patterns of DNA methylation at initial diagnosis of PCOS, between women who do, or do not, later develop diabetes. Genome-wide methylation array screening identified 273 differentially methylated sites in at least five of six case-control pairs. 19 CpG sites were differentially methylated in the same direction, separating best the women with PCOS and those that developed diabetes, and thus are the best targets for further investigation as potential at-diagnosis prognostic biomarkers.
Strengths. To identify potential clinically useful prognostic DNA methylation biomarkers, we used peripheral blood samples taken from women at their initial diagnosis of PCOS. These samples had accompanying long-term clinical follow-up data (up to 12 years), which enabled direct comparison of ‘at-diagnosis’ methylation patterns in women with PCOS who did or did not later develop diabetes. We reasoned that the best potential methylation biomarkers would be detectable at this early stage where dietary, exercise, and weight management interventions are most likely to be effective[5].
We used well-established methods of genome-wide methylation assessment (HumanMethylation450 arrays) to identify differences between women with PCOS, and women with PCOS who later developed diabetes. Similar to previous array-based prognostic marker screening studies, and to reduce the likelihood of false positive targets, we used a ≥0.2 β-value difference to define differential methylation[9].
Limitations . Given the number of samples evaluated and the variation in criteria used to define PCOS (and therefore variable population prevalence), we cannot estimate prognostic or predictive values. Furthermore, plotted Pyrosequencing™ data raise the possibility that single nucleotide polymorphisms (SNPs) may be implicated in methylation patterns observed in some of our candidates. Although known SNPs were excluded during array filtering, polymorphic CpG sequences (in all Infinium methylation arrays) necessitate caution in interpretation of results[11].
Interpretation. Although there are no striking visual differences between cases and controls in the heatmap of the differentially methylated CpG sites, clustering reassuringly separately grouped the women with PCOS who developed diabetes from the women that did not. The 273 differentially methylated CpG sites and the 19 putative biomarker candidates identified is broadly in keeping with numbers in previous methylation biomarker studies in benign tissue types[10], but is difficult to interpret in the context of lack of similar prior studies in PCOS. Notwithstanding, methylation differences were substantiated by Pyrosequencing™ in our candidates, confirming ‘proof of principle’ that DNA methylation holds the exciting potential as a prognostic biomarker in PCOS.
Whilst it cannot be implied from our data, it is possible that differential methylation at one or more of the sites identified may be contributory to diabetes development in women with PCOS. Further studies in larger patient cohorts are required to confirm our initial findings, evaluate the possible importance of SNPs in the differentially methylated sites identified, and to facilitate the mechanistic investigation of DNA methylation changes in women with PCOS who develop diabetes.