Levels of circulating insulin cell-free DNA in women with polycystic ovary syndrome – a longitudinal cohort study

Polycystic ovary syndrome (PCOS) is the most common endocrine disorder of reproductive age women [1]. It is characterized by hyperandrogenism, polycystic ovary morphology, and anovulation, but also by metabolic disturbances; increased risk of hypertension, dyslipidemia, insulin resistance, impaired glucose tolerance (IGT) and an increased lifetime risk of Type 2 diabetes mellitus (T2D) [1, 2]. Preventive advice and monitoring for these future complications are important. However, it is controversial how to monitor women with PCOS according to the risk of T2D and the guidelines present with some inconsistencies. The latest recommendation is that measurements of HbA_1C, fasting glucose or oral glucose tolerance test (OGTT) should be repeated with 1 to 3 years interval, according to additional risk factors and an OGTT should be offered when planning pregnancy or seeking fertility treatment [1]. There is a general consensus that the risk of diabetes increases, corresponding to age and increasing BMI [2, 3], but to date, no definitive marker of these metabolic disturbances and T2D has been identified in women with PCOS. Further, the heterogeneous nature of PCOS with differences in metabolic risk profile [4, 5] clarifies the need for new biomarkers.

The abundance of insulin cfDNA that is unmethylated at the − 69 position of the Insulin gene has been suggested as a biomarker of β-cell death, as this site is unmethylated in insulin-producing islet β-cells but methylated in non-islet cells [6‐8]. Methylation of a certain genomic region is generally regarded as a deactivation mechanism, and is conducted by DNA methyltransferases (DMT) that catalyze the addition of a methyl group from S-adenosyl-methionine to the carbon in 5′ of cytosine residues in cytosine-phosphate-guanine (CpG) dinucleotides [9]. The active, unmethylated form of the insulin gene from β-cells is presumed to be released into the circulation from dying/dead islet β-cells [10]. Therefore, the amount of circulating unmethylated cfDNA would be a reflection of β-cell death, while methylated cfDNA is suggested to be a marker of cellular stress and death of cells/tissues that do not transcribe insulin [11]. An intense search for methods to quantify methylated and unmethylated insulin cfDNA is ongoing, as it would enable us to diagnose Type 1 Diabetes (T1D) and β-cell destruction before the loss of a significant number of insulin-producing cells. A crucial step in this process has been to identify CpG-sites within the insulin gene that are unique to islet β-cells. There are different approaches to this issue and no consensus of which sites and methods to apply [10, 12, 13].

Though insulin resistance, hyperglycemia and impaired glucose tolerance play a central role in T2D, these alone are not sufficient to lead to T2D without a β-cell defect [14]. In patients with early stages of T2D, β-cells secrete excessive insulin and expand their mass to compensate for the increased metabolic load and obesity-associated insulin resistance. After a period, the β-cell function then deteriorates along with loss of β-cell mass and apoptosis [15, 16]. It is suggested that epigenetic modulations could contribute to these alterations in β-cell function [17]. Currently, only a few groups have demonstrated the methylation pattern of the insulin gene in patients with T2D, and most of them only within the pancreatic tissue of donors, mouse models or pancreatic cell lines [6, 7, 18, 19]. Although one study demonstrated that at-risk subjects who progressed to T1D had higher levels of unmethylated insulin-coding cell-free DNA (INScfDNA) compared with healthy controls [20], there is no knowledge of whether an increased amount of unmethylated circulating INScfDNA could serve as a predictor of T2D, as there are no longitudinal studies of cohorts with an increased risk of T2D.

Women with PCOS are known to have a higher risk of IGT and IFG progressing to T2D [2]. Some studies indicate that the IGT is accompanied by a β-cell dysfunction in PCOS women especially in those with a family history of T2D [21] and might even occur prior to changes in stimulated insulin or glucose levels [20]. Since islet β-cell death is a common denominator to T1D as well as T2D, we aimed to assess if islet β-cell death, measured as circulating unmethylated INScfDNA could serve as a biomarker for risk stratification in PCOS women who progress to prediabetes or T2D.

Droplet digital (dd) PCR was used to measure the copy number of INScfDNA in women with PCOS compared with controls at baseline and at 6 years of follow-up using the method described by Fisher et al. [13]. We also investigated whether the abundance of circulating INScfDNA correlated with markers of insulin resistance and T2D in women with PCOS.

Islet β-cell death is known to be a common denominator in progression to type 1 or type 2 diabetes. Β-cell death in T2D is well known to be associated with exogenous insulin requirement. However, there are several lines of evidence [25] indicating that progression to Type 2 diabetes itself may result following loss of functional islet β-cells. Thus, our study aimed to measure biomarkers of islet β-cell death (i.e INS cfDNA) in PCOS/Control women who would develop T2D in future. At the moment, there is a consensus that elevated levels of unmethylated INScfDNA correlate with β-cell death and T1D in mice and humans [8, 10, 12], whereas methylated cfDNA is suggested as a marker of inflammation or sepsis [11]. With this study, we investigated whether this relatively new method could be used for detecting early signs of β-cell death and T2D in women with PCOS. Our results could not confirm this hypothesis, as we did not find any differences in methylated or unmethylated INScfDNA when we compared PCOS women with healthy controls. Likewise, we could not detect any increase in methylated or unmethylated INScfDNA over a period, even though the PCOS group presented with a more metabolically challenged profile at FU. Neither did we find a correlation between markers of insulin resistance and metabolic syndrome and insulin promotor DNA methylation status.

There are several issues that could contribute to these negative findings. First: the insulin gene and its promotor contain several methylation sites, but not all of them display a unique methylation pattern in β-cells compared with other tissues [26]. The tissue specificity of these methylation sites is essential for their usefulness as biomarkers of β-cell death. The evidence though is still sparse on this and there is no consensus of which sites that are the best determinants of islet β-cell stress or death and of future progression to T1D. Previous studies have also used the CpG site (− 69) upstream the TSS of the insulin gene, and reported increased levels of both unmethylated and methylated insulin cfDNA in patients with T1D, but not in patients with T2D [6]. Another study [27] primarily tested downstream CpG sites (located at + 255, + 273, + 303, + 329, + 364, + 370, + 396 and + 399) within the coding region of the insulin gene and confirmed increased levels of unmethylated cfDNA in patients with T1D as well as subjects with increased risk of T1D [28]. A third group identified a cluster of six other CpG sites, within the insulin promoter, with Illumina 450 K methylation arrays and next-generation sequencing, that was able to distinguish between healthy controls and patients with T1D [29].

Secondly, little research into INScfDNA has been done in patients with T2D, and none on women with PCOS. Likewise, the existing studies report results with some discrepancy. Studies indicate that the DNA methylation is not stable and not all CpG sites are consistently methylated during periods of cellular stress, which could be an explanation to the inconsistency in findings [7]. Further, it is most likely that the DNA methylation during T2D occurs at multiple loci with small effect sizes that contribute to an increased risk for disease [30]. Heterogeneity in the methylation sites affected would decrease the sensitivity of a cfDNA biomarker.

Fisher et al. [6] used the CpG site (− 69) and found that neither methylated nor unmethylated cfDNA levels were increased in patients with T2D compared with controls. Another group investigated the DNA methylation in 25 CpG sites within the insulin promoter and gene in pancreatic islets from patients with T2D and controls [7]. They reported that four sites (located at − 234, − 180, − 102 and + 63) showed increased methylation in pancreatic β-cells from patients with T2D compared with controls. Furthermore, they showed that the percentage of DNA-methylation was positively correlated to HbA_1C, but negatively correlated with insulin mRNA expression. [7] These findings indicate that increased DNA methylation of the insulin promotor in the pancreas could be a contributor to the downregulation of insulin expression in T2D as a pathological response to hyperglycemia. This is in line with the findings of Kenna et al. [31] who investigated the methylation pattern in women with gestational diabetes compared with pregnant women without gestational diabetes. They suggested that a decreased fraction of unmethylated insulin promotor DNA is a reflection of a decreased β-cell turnover and the decreased turnover could be a mechanism to compensate for the need for higher insulin levels.

Third, PCOS is a heterogeneous condition – there is variation within the group mainly explained by the syndrome definition of PCOS [1, 32], but also variation over time caused by e.g. pregnancy or interventions in lifestyle [33‐35]. Some studies show that the different PCOS phenotypes differs in cardiometabolic risk, with the highest risk in hyperandrogenic and anovulatory phenotypes [4, 5]. However, we did not detect any differences in unmethylated or methylated INScfDNA neither at baseline nor at follow-up in a subanalysis of PCOS phenotypes based on Rotterdam criteria (data not shown). Further, the cohort was recruited when referred to assisted reproductive treatment (ART), and as a standard procedure, all PCOS women with a BMI > 25 are asked to initiate lifestyle interventions. Looking at weight only, the control group increased theirs during FU, while the PCOS group did not. The steady weight in the PCOS group could be interpreted as an alteration in general lifestyle. The fact that our cohort is recruited prior to ART, could also have caused selection bias, as those women with need of ART possibly are challenged with a severe PCOS phenotype. Moreover, our participants were relatively young at the follow-up time point in terms of developing insulin resistance, prediabetes or T2D. The effect on INScfDNA methylation pattern could have been more pronounced if the participants were older or if the follow-up time had been longer. Another explanation to the negative results is a possible effect of PCOS status on liver function or the use of metformin in the PCOS women. PCOS women are known to have an increased risk of developing non-alcoholic fatty liver disease (NAFLD) [36], and although there is no studies on NAFLD and INScfCNA counts, one study have indicated that the presence of autoimmune hepatitis has a decreasing effect on levels of methylated and unmethylated INScfDNA compared with controls [6]. This could cause an underestimation of the differences between the groups and thereby our results. The use of metformin could also be a confounder in our study. There are no studies of the effect of metformin on INScfDNA counts, but insulin treatment in newly diagnosed T1D is shown to reduce both methylated and unmethylated INScfDNA significantly [6]. If the effect of metformin is similar, use of metformin in the PCOS group could also cause an underestimation of the differences between the groups. Unfortunately, we have no reliable records of the metformin use in our cohort between baseline and follow-up. Finally, one could argue that our sample size, and especially the control group, is too small. This study was designed with an assumption of that, in a PCOS cohort aged 30–40 and BMI > 25, the prevalence of prediabetes (IGT or IFG) would be approximately 40–45% [37] at follow up. The prevalence in our cohort did not fulfill this assumption, as only 17.5% of our PCOS cohort was prediabetic, causing very low power. If the prevalence had been as expected, the power would had been acceptable. The lower number of participating controls resulted due to a recruitment issue, which might cause an underestimation of the differences between PCOS and controls.

We were not able to detect any differences in unmethylated or methylated INScfDNA at any time in this study. However looking at samples with undetectable levels INScfDNA, there are a relatively larger number of samples without detectable levels of unmethylated INScfDNA, than of methylated INScfDNA. This advocates the theory of PCOS as an inflammatory condition with increased cell-turnover in general, as suggested by others [38, 39], even though we did not detect significant differences between controls and PCOS.

The cell-free DNA methylation pattern of the insulin gene promoter has been suggested as a marker of pancreatic islet β-cell destruction, and thereby as an early marker of diabetes. Since islet β-cell death is a common denominator in T1D as well as T2D, we explored if circulating insulin cell-free DNA can determine differences in BL and 6 year follow-up samples in a cohort of women with PCOS or without. We were not able to detect any differences in the levels of INScfDNA between PCOS and controls at BL as well as over a period of 6 years. Although these negative findings could be a result of the heterogeneous nature of PCOS or by the relatively short follow-up time, the publication of negative findings is an important contribution to scientific understanding of a topic, avoiding publication bias by preferentially publishing only papers with positive results increasing the risk that an incorrect prevailing view can persist. Further studies in larger and longer longitudinal cohorts are warranted.