Short Report: Circulating microRNAs are associated with incident diabetes over 10 years in Japanese Americans

Epigenetic changes precede the development of diabetes by many years, providing clues to its pathogenesis. We explored whether the epigenetic markers, circulating microRNAs (miRNAs), were associated with incident diabetes in Japanese Americans. We conducted a pilot study (n = 10) using plasma from age- and sex-matched participants who did or did not develop diabetes in the Japanese American Community Diabetes Study, an observational study of diabetes risk factors. Extraction and high-throughput sequencing of miRNAs were performed using samples collected at baseline. Regression models were fit comparing circulating miRNAs (N = 1640) among individuals who did or did not develop incident diabetes at 10-year follow-up. Participants averaged 51.7 years of age at baseline; 60% were male. We identified 36 miRNAs present at different (10 higher and 26 lower) levels in individuals who developed diabetes compared to those who did not (log2fold change ≥1.25 and false discovery rate ≤5%). These included miRNAs with functions in skeletal muscle insulin metabolism (miR-106b and miR-20b-5p) and miRNAs with functions in both skeletal muscle insulin metabolism and cell cycle regulation in endocrine pancreas (miR-15a and miR-17). Circulating miRNAs were associated with subsequent development of diabetes among Japanese Americans over 10 years of follow-up. Results are preliminary. Large-scale miRNA sequencing studies could inform our understanding of diabetes pathogenesis and development of therapies, based on gene expression regulation, that target diabetes.

RPMs for five minutes to completely clear plasma of cells. Small RNAs were extracted from 500 μL plasma aliquots using the Exiqon (now Qiagen) miRCURY TM RNA Biofluids Isolation Kit (Exiqon, Woburn, MA). Integrity, purity and quantity of purified miRNA was assessed using spectrophotometry and an Agilent 2100 Bioanalyzer capillary electrophoresis system (Agilent Technologies Inc, Palo Alto, CA). The Qiagen QIAseq miRNA NGS Library Kit was used for library preparation. MiRNAs were sequenced using an Illumina sequencer. Lab personnel were blinded to participant outcomes. Additional technical details are provided in Supplemental Materials.
Statistical and bioinformatics analyses. Number (%) and mean (standard deviation) describe study population characteristics. Analyses were conducted in R version 3.4.0. Samples were normalized using a weighted trimmed mean of M-values (TMM), which calculates a normalization factor that is used to scale the library sizes. Because count data are not normally distributed and may have transcripts with zero counts 8 , a linear model based on the negative binomial distribution in the Bioconductor edgeR package 9 using quasi-likelihood F-tests 10 was used. Exploratory principal component analysis showed an apparent large batch effect captured by the first principal component; therefore, the first principal component was included as an adjustment variable, along with two surrogate variables detected using the Bioconductor sva package 11 . To protect against choosing miRNA transcripts that may be differentially expressed at a statistically significant but low-fold-change level that is not biologically meaningful, miRNAs with at least a 25% difference 12 in expression as well as a <5% false discovery rate (FDR) were identified as significantly different, as has been previously reported 13 . We used the Bioconductor sizepower package to estimate study power, post hoc. The average standard error was 0.999, and we tested for a 1.25-log2fold difference. We found 36 out of 1640 miRNAs significant. Accepting 10 false positives, we calculate we had 31% power to identify differences in circulating miRNAs between the groups. The Core Analysis feature of the Ingenuity Pathway Analysis (IPA) software program (Built version-486617 M; Content version-33559992; Ingenuity Systems, A Qiagen Company, Redwood City, CA) was used to identify transcriptional networks using microRNAs that were present at different levels in cases than in controls 14 . IPA's microRNA Target Filter to identify mRNA targets was also used, restricting the search to "experimentally validated" targets. IPA Core Analysis was performed on the lists of targets that were identified as described above by IPA's microRNA Target Filter feature. ethics approval and consent to participate. This study was approved by the institutional review board at the University of Washington as # STUDY00001823, and all participants provided written informed consent.

Discussion
In this exploratory analysis using archival samples from a community-based study of diabetes risk factors in Japanese Americans, we identified 36 miRNAs (FDR < 0.05, >1.25-fold change) present at different levels in plasma of individuals who did versus did not develop diabetes up to 10 years later. Although the sample size is small, this report establishes new findings and extends previous findings in this area in several important ways. First, we extended the study of circulating miRNAs with long-term follow-up for incident diabetes to a new population, Asian-Americans. Second, by using a comprehensive sequencing approach, we identified circulating miR-106b, miR-17 and miR-20b-5p not reported in previous long-term longitudinal studies. These miRNAs have putative roles in the pathogenesis of diabetes via actions in skeletal muscle and/or pancreatic islet cells. Lastly, we (2020) 10 www.nature.com/scientificreports www.nature.com/scientificreports/ replicated previous observations 4,5 that lower baseline levels of miR-15a were associated with incident diabetes, suggestingits potential as a predictive diabetes biomarker.
In the Spanish CORDIOPREV cohort (n = 462), higher levels of miR-150 and miR-30a-5p and lower levels of miR-15a and miR-375 were associated with higher diabetes risk at 60 months 4 . In the Bruneck (Italy) cohort, individuals with incident diabetes at 10 years (n = 19) had lower baseline plasma levels of miR-15a, miR-29b, miR-126 and miR-223 and higher levels of miR-28-3p than matched controls 5 . In our sample, except for miR-15a, the other associations were not replicated. A secondary analysis of the Practicing Restorative Yoga vs. Stretching for the Metabolic Syndrome study examined associations of circulating miRNAs with incident diabetes over 12 months of follow-up 15 . As in our study, baseline levels of miR-106b, miR-20b, miR-363, miR-486, miR-532 and miR-93 were associated with incident diabetes, as well as with response to a lifestyle intervention 15 .
In skeletal muscle from monozygotic twins discordant for type 2 diabetes, expression of miR-106b, miR-15b, miR-17, miR-20a, miR-20b, miR-25, miR-363, miR-451, miR-93 (identified in the current study), and others were downregulated in twins with type 2 diabetes compared to twins without diabetes 16 . In a community-based sample, miR-106b, miR-144, miR-15b, miR-451, and miR-93 were upregulated and miR-15a downregulated in skeletal muscle of individuals with type 2 diabetes compared to individuals with normal glucose tolerance 17 . In another study, higher levels of both skeletal muscle miR-15a and miR-15b were associated with higher fasting and 2-hour glucose in adults with and without diabetes 18 . Inconsistency of identified miRNAs in previous as well as the current study could be due to racial and ethnic differences in the study populations, heterogeneous mechanisms contributing to diabetes pathogenesis across the populations studied,or differences in methods (sample source, processing, and profiling). Carefully designed studies incorporating repeated measurement of miRNAs over time in cohorts at high diabetes risk will be needed to clarify these relationships.
Identified miRNAs including miR-17, miR-106b and miR-20b-5p have functions in skeletal muscle insulin metabolism. In skeletal muscle, miR-17 19 and miR-106b 20 downregulate components of the glucose transporter 4 pathway, while overexpression of miR-20b-5p impairs insulin signaling and suppresses genes in pathways related to immune function (AKTIP and STAT3) 21 . Some identified miRNAs have functions in islet cells as well. In pancreatic cell lines, miR-17 downregulates thioredoxin-interacting protein, a pro-apoptotic regulator of beta-cells in diabetes 22 . In mouse islets, miR-7 (miR-7a) targets multiple components of the mTOR signaling pathway, and its inhibition promotes adult beta-cell replication 23 . In rats, miR-25 negatively regulates expression of insulin I messenger transcripts 24 . Last, islet cell expression of miR-15a is upregulated in the presence of brief (1 h) exposure to hyperglycemia but depressed after a longer period of exposure 25 . At the same time, overexpression of miR-15a increases insulin levels and its repression inhibits insulin synthesis by inhibiting uncoupling protein-2 (UCP-2) gene expression 25 . Five miRNAs identified in this study are members of the highly conserved polycistronic miR-17~92 family (miR-17, miR-20, miR-25, miR-93, and miR-106b) 26 , which may be important in cell-cycle regulation, as suggested by our bioinformatics analysis results.
An important limitation of our pilot study is that it was underpowered to detect an exhaustive list of circulating miRNAs associated with incident diabetes. We used conservative correction for multiple testing to mitigate risk of type II error. Results are preliminary and need to be replicated in other populations. In addition, at baseline the groups differed in distribution of impaired fasting glucose (IFG) and impaired glucose tolerance (IGT). To further characterize baseline glycemic status, we used 30-, 60-, 120-, and 180-minute values of glucose and insulin from baseline oral glucose tolerance tests to calculate Matsuda and insulinogenic indices (IGI) for study