Associations of plasma miRNAs with waist circumference and insulin resistance among women with polycystic ovary syndrome – Pilot Study

Abstract

Background:

Insulin resistance (IR) and central obesity are common in polycystic ovary syndrome (PCOS), but pathomechanisms for IR in PCOS are not established. Circulating microRNAs (miRNAs) are non-invasive biomarkers of epigenetic regulation that may contribute to the pathogenesis of IR and central adiposity in PCOS.

Methods:

We conducted a pilot study to examine associations of circulating miRNAs with IR and central adiposity among women with PCOS (n=11) using high-throughput miRNA sequencing. We fit generalized linear models examining associations of waist circumference and HOMA-IR with plasma miRNAs. We used false discovery rate (FDR)-adjusted cutoff p<0.1 to correct for multiple testing. We used miRDB’s Gene Ontology (GO) tool to identify predicted pathways for top hits.

Results:

Mean age and BMI of participants were 27.9 years and 32.5 kg/m², respectively. Lower levels of miR-1294 were associated with higher waist circumference (β = −0.10, FDR=0.095). While no miRNAs were associated with HOMA-IR at our FDR cut off <0.1, 11 miRNAs were associated with waist circumference and 14 miRNAs with HOMA-IR at unadjusted p<0.01, including members of the highly conserved miR-17/92 cluster and miR-1294 (β = −0.10, p<0.001). The GO analysis of miR-1294 identified 54 overrepresented pathways, including “negative regulation of insulin receptor signaling” (FDR=0.019), and 6 underrepresented pathways.

Conclusions

Plasma miR-1294 along with members of the miR-17/92 cluster and miRNAs involved in insulin signaling may be associated with central obesity and insulin resistance in PCOS. Larger studies among women with and without PCOS are needed to validate these findings.

Introduction

Polycystic ovary syndrome (PCOS) is the most common endocrinopathy in reproductive-age women¹. Insulin resistance (IR) affects roughly 60% of women with PCOS² and frequently co-occurs with central obesity³. Diabetic female mice exposed to high levels of androgens have decreased skeletal muscle glucose uptake due to decreased activation of serine/threonine-specific protein kinase AKT and decreased GLUT4 expression⁴, suggesting that skeletal muscle IR in PCOS may be mediated at least in part by androgens. A definitive study of IR-related changes in gene expression in women with PCOS would require the collection of relevant tissue samples such as skeletal muscle, adipose tissue, and ovary, which is not feasible in large-scale clinical studies. Therefore, identification of circulating factors that contribute to IR in PCOS is an important next step in understanding its pathogenesis. Circulating microRNAs (miRNAs), a class of post-transcriptional epigenetic regulators of gene expression, may have value in this setting because they participate in inter-organ crosstalk, have potential as therapeutic targets, and can be readily measured in the cell-free circulation¹. Ubiquitous in tissues and in cell-free body fluids, miRNAs are short non-coding RNAs that pair to the 3′ untranslated region of messenger RNAs leading to translational repression or a decrease in transcript stability^5,6. We^7,8 and many others^9–11 have identified alterations in circulating miRNAs that precede development of diabetes and may contribute to skeletal muscle insulin resistance^12,13. For example, lower levels of miRNAs such as miR-17 and miR-20a/b have been detected in men and women who subsequently develop type 2 diabetes compared to controls⁷. Although women with PCOS may have unique mechanisms contributing to IR, previous studies examining IR-related miRNAs using plasma or serum from women with PCOS have used PCR or TaqMan low-density arrays for miRNA profiling^14,15, limiting discovery of novel miRNA-IR associations in this population. In this preliminary study, we hypothesized that plasma miRNAs, including novel miRNAs identified by next-generation miRNA sequencing, would be related to IR and central obesity in a clinic population of women with PCOS.

Methods

Study setting and population

Participants were recruited from the University of Washington Endocrinology Clinic and the Diabetes Institute, Seattle, WA. Eligible participants were women ≥18 years old without diabetes who fulfilled the Rotterdam Criteria for PCOS^16–18 as determined by an endocrinologist (ST) (n=11). The study was approved by the institutional review board at the University of Washington, and all participants provided written informed consent.

Data collection

Data and sample collection were done at the University of Washington Endocrinology Clinic and the Diabetes Institute, Seattle, WA. Waist circumference (inches) was measured at least twice at the smallest circumference of the torso by trained examiners. Blood specimens were drawn at enrollment by trained phlebotomists after participants were asked to fast overnight. Plasma or serum glucose and serum insulin were assayed as part of clinical care using the facilities’ standard protocols. The homeostatic model assessment for insulin resistance (HOMA-IR) was calculated as (fasting plasma insulin [mU/l] * fasting plasma glucose [mmol/l])/22.5¹⁹. Plasma for miRNA was isolated and stored at −80 degrees C.

Pre-processing, extraction, and profiling of plasma miRNAs

Thawed samples were spun at 3000g for 5 minutes to completely clear plasma of cells. Small RNAs were extracted from 400 μL plasma aliquots using the Qiagen miRCURY^™ RNA Biofluids Isolation Kit (Qiagen, Woburn, MA). Integrity, purity and quantity of purified miRNA was assessed using an Agilent 2100 Bioanalyzer capillary electrophoresis system (Agilent Technologies Inc, Palo Alto, CA) and a Qubit microRNA assay kit (Thermo Fisher Scientific, Waltham, MA). The Qiagen QIAseq miRNA NGS Library Kit was used for library preparation. MiRNAs were sequenced using an Illumina sequencer.

Statistical analyses

Selected participant characteristics were summarized using mean/SD for continuous variables and n/percent for categorical variables. We excluded two samples with very low unique molecular index (UMI) depth (73K and 115K, respectively), resulting in a final sample size of n=11 participants. One sample had an unexpectedly high UMI count (14 million UMI reads), but the remainder were between 1 million and 3 million reads. Prior to fitting the linear models, we excluded miRNA transcripts with very low counts because these may represent unexpressed miRNA transcripts and are unreliable due to the low signal to noise ratio. We used a simple heuristic, selecting those miRNA transcripts with a mean log counts/million (logCPM) > 2.5, based on our experience that the distribution of the mean logCPM is bimodal and the assumption that the first mode represents primarily unexpressed transcripts. There were 546 transcripts with a mean logCPM > 2.5 that were then carried forward in our analysis. We used Qiagen’s Gene Globe software to process the FASTQ files, which identifies the UMI barcodes and then summarizes UMI counts/transcript for 2084 small non-coding mRNA transcripts (1970 miRNA and 114 piRNA transcripts). All quality control measures indicated that the 10 samples used were of comparable quality.

We fit generalized linear models and tested for differences using a quasi-likelihood F-test. We used the Bioconductor sva package to estimate surrogate variables intended to account for unobserved variability. The surrogate variables did not appreciably affect the model fit, so we chose to continue with unadjusted models. For the primary analysis, miRNAs with a Benjamini-Hochberg false discovery rate (FDR)-adjusted p-value <0.1 were considered significant. We also examined miRNA associations that were significant at α<0.01. Analyses were conducted in R version 4.1.0. Because the explanatory variables are continuous, the reported coefficient (β) is the log₂ slope of the fitted regression line and can be interpreted as the log₂ change in expression for every 1-inch increase in waist circumference (or unit increase in HOMA-IR). As an example, a β of −0.1 indicates a log₂ reduction in miR level of 0.1 for every 1-inch increase in waist circumference or, alternatively, a −1 log₂ change in miR level for every 10-inch increase in waist circumference, indicating that level of the miR is reduced by two-fold for every 10-inch increase in waist circumference. To identify predicted gene targets in silico, we used the miRDB tool (http://mirdb.org/²⁰. We analyzed all miRs with an FDR<0.1 and also the top 10 most statistically significant miRs (α<0.01) (Table 2). miRDB calculates a target score for each target (50–100). The higher the score, the higher the confidence that the prediction is real. A predicted target with a score > 80 is most likely to be real. Therefore, we used a prediction score > 80 to identify targets. We also used miRDB’s Target Ontology (GO) tool to identify predicted pathways for top miR hits (FDR<0.1).

Table 2.

Plasma microRNAs associated with waist circumference and/or HOMA-IR in women with PCOS (p<0.01), n=11

miRNA	β	logCPM	F	PValue	FDR
WAIST CIRCUMFERENCE
hsa-miR-1294	−0.1	5.323	26.228	<0.001	0.095*
hsa-miR-4732-5p	−0.098	6.472	12.899	0.003	0.266
hsa-miR-451a	−0.145	14.439	12.709	0.003	0.266
hsa-miR-486-5p	−0.08	15.531	12.409	0.004	0.266
hsa-miR-16-5p	−0.082	18.619	12.335	0.004	0.266
hsa-miR-15a-3p	−0.252	3.548	12.195	0.004	0.266
hsa-miR-451b	−0.146	8.121	11.083	0.005	0.266
hsa-miR-326	0.121	6.865	11.061	0.005	0.266
hsa-miR-4732-3p	−0.094	6.474	10.605	0.006	0.266
hsa-miR-1180-3p	−0.077	6.079	9.487	0.008	0.266
hsa-miR-339-5p	0.111	8.241	9.219	0.009	0.266
HOMA-IR
hsa-miR-486-5p	−0.57	15.531	17.481	0.001	0.338
hsa-miR-503-5p	−0.413	5.94	12.114	0.004	0.338
hsa-miR-16-5p	−0.543	18.619	11.963	0.004	0.338
hsa-miR-132-3p	0.327	5.939	11.909	0.004	0.338
hsa-miR-4732-5p	−0.638	6.472	11.662	0.005	0.338
hsa-miR-363-3p	−0.584	9.45	11.506	0.005	0.338
hsa-miR-25-3p	−0.378	12.89	10.925	0.006	0.338
hsa-miR-451a	−0.809	14.439	10.483	0.006	0.338
hsa-miR-20b-5p	−0.43	8.156	10.469	0.006	0.338
hsa-miR-15a-5p	−0.455	10.22	10.092	0.007	0.338
hsa-miR-4732-3p	−0.606	6.474	9.806	0.008	0.338
hsa-miR-92a-3p	−0.313	14.765	9.731	0.008	0.338
hsa-miR-144-3p	−0.852	10.199	9.681	0.008	0.338
hsa-miR-7-5p	−0.371	8.986	9.518	0.009	0.338

Results

Selected cohort characteristics are shown in Table 1. Table 2 shows log₂ slope (β) values, log CPM, unadjusted p-values, and FDR-adjusted p-values for plasma miRNAs that are associated (FDR<0.1) with waist circumference and/or HOMA-IR. Lower levels of miR-1294 were associated with higher waist circumference (β = −0.10, FDR=0.095). No miRNAs were associated with HOMA-IR at FDR<0.1. At α<0.01, higher levels of 2 miRNAs were associated with higher waist circumference: miR-326 (β = 0.12, p=0.005) and miR-339–5p (β = 0.11, p=0.009). Lower levels of 9 miRNAs were associated with higher waist circumference, including miR-1294 (β = −0.10, p<0.001), miR-4732–5p (β = −0.10, p=0.003), miR-451a (β = −0.15, p=0.003), miR-486–5p (β = −0.08, p=0.004), miR-16–5p (β = −0.08, p=0.004), and miR-15a-3p (β = −0.25, p=0.004). Higher levels of one miRNA (miR-132–3p) were associated with higher HOMA-IR (β = 0.33, p=0.004). Lower levels of 13 miRNAs were associated with higher HOMA-IR, including miR-486–5p (β = −0.57, p=0.001), miR-503–5p (β = −0.41, p=0.004), miR16–5p (β = −0.54, p=0.004), miR-4732–5p (β = −0.64, p=0.005), miR-363–3p (β = −0.58, p=0.005), and miR-25–3p (β = −0.38, p=0.006). Scatter plots of the top miRNAs associated with waist circumference and HOMA-IR are shown in Fig. 1. Pathways identified in the GO analysis of miR-1294 included “negative regulation of insulin receptor signaling” (FDR-adjusted p=0.019, Supplementary Tables 1–4) and “protein localization to phagophore assembly site” (FDR-adjusted p=0.045).

Table 1.

Characteristics of women presenting with polycystic ovary syndrome, February 2019–November 2020, n=11

Age, years (SD)	27.9	(4.4)
Reports White race (vs. non-White race), %y (n)	45%	5
Reports Hispanic/Latinx ethnicity (vs. no) Hispanic/Latinx ethnicity), %y (n)	18%	2
Oligo- or anovulation, %y (n)	91%	10
Clinical signs of hyperandrogenism, %y (n)	82%	9
Polycystic ovaries by ultrasound, %y (n)	64%	7
Mean body mass index, kg/m2 (SD)	32.5	(15.3)
Mean waist circumference, inches (SD)	36.4	(10.1)
Mean fasting glucose, mg/dL (SD)	90	(7.5)
Mean fasting insulin, uIU/mL (SD)	9.8	(6.9)
Mean HOMA-IR (SD)	2.2	(1.5)
Mean OGTT 120-minute glucose, mg/dL (SD)	107.2	(32.6)
Mean OGTT 120-minute insulin, uIU/mL (SD)	76.1	(76.7)
Mean total testosterone, ng/dL (SD)	0.6	(0.3)
Mean free testosterone, ng/dL (SD)	12.0	(7.5)
Mean sex hormone binding globulin, nmol/L (SD)	40.1	(27.4)

OGTT: oral glucose tolerance test; SD: standard deviation

Fig. 1.

Scatter plots of top six miRNAs associated with (A) waist circumference and (B) HOMA-IR

Conclusions

In this preliminary study, lower levels of miR-1294 were associated with higher waist circumference among women with PCOS (FDR<0.1). Further, 11 miRNAs were potentially associated with waist circumference and 14 miRNAs were potentially associated with HOMA-IR (unadjusted p<0.01). To our knowledge this is the first analysis to identify an association of miR-1294 with central adiposity (and possibly HOMA-IR) among women with PCOS.

Two previous studies have examined associations of HOMA-IR and/or a measure of central adiposity with circulating miRNAs in PCOS. In an analysis by Sorenson et al, serum miR-20a-5p, miR-361–5p, and miR-1225–3p were associated with HOMA-IR among women with PCOS, although only the association with miR-1225–3p remained after adjustment for age and BMI¹⁴. There was no overlap with the individual miRNAs identified in the current study, but both analyses identified miRNAs that are members of the miR-17/92 cluster (e.g., miR-20a-5p, miR-92a) or one of its paralogues (e.g., miR-20b-5p, miR-25)²¹. MiRNAs in these clusters are highly conserved and regulate genes critical to cell death (e.g., PTEN, BCL2L11) and proliferation (e.g., TGFBRII)²¹, among other critical processes. Murri et al. identified a profile of serum miRNAs that were associated with HOMA-IR or waist-to-hip ratio in women with PCOS¹⁵, but there was no overlap with the findings in the current study. Reasons for these differences may reflect differences in race/ethnicity, in clinical characteristics of the participants, or in experimental methods, or limited statistical power in the current analysis.

One factor that may have contributed to the differences in miRNAs identified between this analysis and previous reports is the use of next-generation miRNA sequencing rather than TaqMan arrays or PCR. Each of these methods has relative merits²². The major benefit conferred from the use of miR-seq in comparison to these other methods in this setting is the potential for discovery of novel miRNAs²³. This is especially important when examining disease-miR associations in a population in which unique mechanisms may contribute to the development of disease, such as insulin resistance in women with PCOS. Used in this way, miR-seq has facilitated discovery of novel miR-disease associations in neurodegenerative disease²⁴, lymphoma²⁵, and other conditions. On the other hand, low power is a recognized limitation that may contribute to bias in small miR-seq studies²⁶, therefore adequately powered miRNA discovery and validation studies are needed to follow up the current findings.

Eleven of the miRNAs potentially associated with waist circumference and/or HOMA-IR in the current study (miR-4732–5p, miR-451a, miR-486–5p, miR-15a, miR-451b, miR-1180–3p, miR-503–5p, miR-363–3p, miR-25–3p, miR-144–3p, and miR-7–5p) were also associated with incident diabetes after 5–10 years in a study we previously conducted among middle-aged Japanese Americans²⁷. In that analysis, mean age was 51 years, compared to 28 years in the current study. About half the participants were male. The overlap in findings suggests that these miRNAs may contribute to IR in a variety of populations, not just women with PCOS. In vitro studies are needed to characterize the roles of these miRs in target tissues including skeletal muscle and beta cell.

We found that lower levels of circulating miR-1294 were associated with higher waist circumference and may be associated with IR as well. This miRNA is downregulated as part of a competing endogenous RNA network in follicular fluid from women with PCOS and may help suppress estradiol production²⁸. In addition, overrepresented pathways in the gene ontology analysis for miR-1294 include “negative regulation of insulin receptor signaling.” Insulin activates two major signaling pathways, the RAS-MAPK pathway and the phosphatidylinositol-3-OH kinase (PI(3)K)-AKT pathway, which is thought to be the key pathway by which insulin controls metabolic processes²⁹. It is possible that this miRNA may contribute to the pathogenesis of insulin resistance among women with PCOS; however, replication studies with larger sample sizes and functional analyses are needed to improve understanding of the pathophysiologic role of this miRNA in IR among women with PCOS.

Strengths of the study include the use of fasting values for insulin, glucose, and miRNAs; as well as the anthropometric measurements performed by trained examiners. Limitations include the small sample size, lack of a comparison population without PCOS, cross-sectional design, as well as lack of validation of identified miRNAs.

In conclusion, miR-1294 along with members of the miR-17/92 cluster and miRNAs involved in insulin signaling are associated with central obesity and insulin resistance in PCOS. Better understanding of mechanisms underlying the pathogenesis of IR in PCOS may facilitate development of novel therapies. Larger studies among women with and without PCOS are needed to validate the findings.