Differential DNA Methylation and Expression of miRNAs in Adipose Tissue From Twin Pairs Discordant for Type 2 Diabetes

In subjects with type 2 diabetes (T2D), the adipose tissue function is frequently disturbed by insulin resistance, leading to impaired glucose uptake, dysregulation of adipokines, and impaired suppression of lipolysis. This results in elevated lipid levels in circulation and storage of fat in other tissues such as muscle, liver, and pancreas (1). Even though insulin-stimulated glucose disposal in adipose tissue accounts for <5% of an oral glucose load (2), selective GLUT4 knockdown (KD) in adipose tissue causes insulin resistance in muscle and liver (3).

miRNAs are small noncoding RNA molecules that regulate expression and, consequently, many physiological processes. In adipose tissue, miRNAs regulate adipogenesis and specific endocrine/metabolic functions (4,5). For example, some miRNAs control differentiation of adipocytes by regulating the Wnt/β-catenin signaling pathway (5), regulate insulin sensitivity of adipocytes via the PTEN/PI3K/AKT pathway (6), and regulate oxidative catabolism of fatty acids via PPARα (7). However, it is less clear which miRNAs might be involved in T2D pathology in human adipose tissue in vivo.

DNA methylation of miRNA promoters can inhibit transcription of miRNAs (8). Indeed, epigenetic regulation of miRNA levels have been described in several pathophysiological processes (9,10). For instance, transcriptional silencing by hypermethylation causes the loss of tumor-suppressor miRNAs in cancer (11,12). Programmed epigenetic changes in miRNAs may also contribute to the link between early-life nutrition and long-term risk of metabolic disease (13). Nevertheless, it is not well established whether such epigenetic changes are responsible for impaired adipose tissue function in T2D.

miRNAs may provide attractive therapeutic targets for T2D because of their involvement in insulin secretion (14) and insulin sensitivity (15). Some studies have investigated the role of miRNAs in human adipose tissue in relation to insulin resistance (16). For instance, eleven miRNAs displayed differential expression in adipose tissue from obese insulin-resistant compared with obese insulin-sensitive women (17), and another study suggested that miR-103 affects the development of adiposity and glucose metabolism (18). However, there is still a lack of detailed knowledge of miRNA regulation of T2D pathologies in human adipose tissue.

In the current study, we aimed to identify differentially expressed miRNAs and epigenetic regulation of their expression in T2D using adipose tissue from a unique human cohort of monozygotic (MZ) twin pairs discordant for T2D. We further explored whether identified miRNAs affect phenotypes related to T2D in adipocytes cultured in vitro. The genomic identity of MZ twins provides a powerful design for detection of differentially expressed miRNAs (19) and acquired DNA methylation changes in miRNA genes related to T2D. The fact that twins provide naturally matched pairs reduces the effects of possible confounders such as genetics, age, and sex. We successfully identified differentially expressed miRNAs in adipose tissue from the twins and inferred that the differences in expression are associated with differential DNA methylation of miRNA promoters. We validated these findings in an independent T2D case-control cohort. Silencing of the most promising candidate miRNAs in adipocytes demonstrated their effects on glucose uptake, insulin signaling, and regulation of T2D pathology–related genes.

Fourteen MZ twin pairs discordant for T2D recruited through Scandinavian twin registries were included in the study (nine Swedish and five Danish twin pairs). Zygosity was confirmed by analysis of 730,525 genetic markers using HumanOmniExpress arrays (Illumina, San Diego, CA). Further, 28 normal glucose tolerant (NGT) and 28 T2D unrelated subjects, pairwise matched for age and sex, and with no difference in BMI between the groups, were selected from a larger twin cohort (20) and included for biological validation as a case-control cohort. After an overnight fast, subcutaneous adipose tissue biopsies were obtained under local anesthesia, frozen in liquid nitrogen, and stored at –80°C. Clinical characteristics of both cohorts are described in Table 1 and have previously been published (21). Glucose tolerance was measured with a 75-g oral glucose tolerance test, with T2D determined according to the 1999 World Health Organization criteria. All study participants gave informed consent. The study was carried out in accordance with the Declaration of Helsinki.

Total RNA was extracted with the miRNeasy kit (QIAGEN, Hilden, Germany). RNA quantity and purity were determined spectrophotometrically (NanoDrop Technologies, Wilmington, DE). RNA integrity was determined with the Experion system (Bio-Rad Laboratories, Hercules, CA).

miRNA expression was analyzed in adipose tissue from 12 of the 14 discordant twin pairs with miRNA 3.0 arrays (Affymetrix, Santa Clara, CA) according to the manufacturer’s recommendations. Robust multichip average (RMA) expression measures were computed with the oligo package in Bioconductor (release 3.10) (22). Since not all of 1,733 miRNAs included in the array are expressed in adipose tissue, the analysis was restricted to 408 miRNAs known to be expressed in adipose tissue, as determined by either of two studies (23,24). Comparisons between discordant twins were based on paired two-tailed Wilcoxon statistics. Intra–twin pair correlations of within–twin pair differences in MZ twins were analyzed using Spearman statistics.

RNA was amplified and reverse transcribed to cDNA with TaqMan PreAmp Master Mix and MicroRNA Reverse Transcription Kit (Thermo Fisher Scientific, Rockford, IL). miRNA expression was quantified with specific TaqMan miRNA assays (hsa-miR-30a-5p, assay 000417; hsa-miR-30b, assay 000602; hsa-miR-30c, assay 000419; hsa-miR-30d, assay 000420; hsa-let-7a, assay 000377; hsa-let-7b, assay 002619; hsa-let-7f, assay 000382; and hsa-let-7g, assay 002282; all Thermo Fisher Scientific) and normalized to RNU48 expression (human, assay 001006; Thermo Fisher Scientific) or snoRNA202 (mouse, assay 001232; Thermo Fisher Scientific). miRNA levels were quantified with the QuantStudio 7 Flex Real-Time PCR System (Applied Biosystems, Foster City, CA). Validation of miRNA expression differences between subjects with and without T2D were based on one-tailed Mann-Whitney statistics.

3T3-L1 murine fibroblasts were cultured in DMEM (25 mmol/L glucose; Thermo Fisher Scientific) supplemented with 10% FBS (Thermo Fisher Scientific) and 100 units/mL penicillin/streptomycin (Thermo Fisher Scientific) in 5% CO₂ humidified atmosphere at 37°C. For differentiation to adipocyte-like cells, 2-day postconfluent cells were incubated with 1 μmol/L dexamethasone (Sigma-Aldrich, St Louis, MO), 0.5 mmol/L 3-isobutyl-1-methylxantine (Sigma-Aldrich), and 1.74 μmol/L insulin (Sigma-Aldrich) for 72 h and then cultivated in DMEM media. The differentiation rate was typically 80–90%.

Nine days postdifferentiation, cells were separately transfected with 100 nmol miRCURY LNA inhibitors (mmu-miR-30a-5p, mmu-miR-30b-5p, and mmu-miR-30c-5p or mmu-miR-30d-5p, respectively; QIAGEN) or with 100 nmol of negative control (NC) siRNA (Negative control A, miRCURY LNA miRNA Inhibitor Control; QIAGEN) using Lipofectamine 2000 transfection reagent (Thermo Fisher Scientific), according to the manufacturer’s instructions. Cells were then incubated for 48 h and RNA was isolated using miRNeasy kit (QIAGEN). Unspecific downregulation of miR-30d by silencing of miR-30a and unspecific downregulation of miR-30b by silencing of miR-30c, and vice versa, were observed (Supplementary Fig. 1). Sequence analysis of the miR-30 family revealed 100% identity between human and mouse miR-30a, b, c, d, and e and a great similarity between miR-30a and miR-30d (1 nucleotide [nt] substitution), miR-30a and miR-30e (1 nt substitution) and miR-30c and miR-30b (2 nt deletions) that could explain the unspecific silencing (Supplementary Fig. 2). Therefore, 3T3-L1 adipocytes were cotransfected with miR-30a and miR-30c inhibitors (100 nmol in total) for complete miR-30a, b, c, and d silencing (n = 6).

RNA (six independent passages) was used for mRNA microarray analysis with Clariom S Assay (902930; Thermo Fisher Scientific), according to the manufacturer’s instructions. Preprocessing was performed with RMA using the Expression Console v1.4.1.46 software (Thermo Fisher Scientific). Data were log2 transformed and analyzed with the limma package in Bioconductor (release 3.10) (25). Data are presented as log2 fold change (log2 FC) between mmu-miR-30a+c silencing and NC. Data were analyzed by paired t test, and P values were adjusted with the use of Benjamini-Hochberg false discovery rate (FDR) analysis. FDR <5% (q < 0.05) was considered significant.

Microarray data were analyzed by gene set enrichment analysis (GSEA) using Reactome pathways (26) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways (www.kegg.jp) available from fgsea package and reactome.db (Bioconductor, release 3.10).

A total of 1,500 base pairs (bp) DNA fragments of hsa-miR-30a (GRCh38.p13, chromosome 6 [chr6]: 71403621–71405121) and hsa-let-7a-3 (GRCh38.p13, chr22: 46110248–46112748) promoter regions were synthetized and inserted into a CpG-free firefly luciferase reporter vector (pCpGL-basic; GenScript, Piscataway, NJ). The construct was then methylated using SssI methyltransferase with S-adenosylmethionine as a methyl donor (B9003S; New England Biolabs, Cambridge, MA). The hsa-miR-30a and hsa-let-7a-3 promoter sequences contain 10 and 59 SssI target sites, respectively. Nonmethylated constructs were used as a background control. Nondifferentiated 3T3-L1 cells were seeded into a 96-well plate and cotransfected with the methylated or nonmethylated construct, and the pRL Renilla luciferase control-reporter vector (pRL-CMV vector; Promega, Madison, WI), using FuGENE HD Transfection Reagent (E2311; Promega), according to the manufacturer’s instructions. Cells were lysed after 48 h, and the firefly and Renilla luciferase activity was measured with the Dual-Luciferase Reporter Assay system (E1910; Promega) in six independent passages. Firefly luciferase activity was normalized to Renilla luciferase activity and corrected for background value. Paired t test was used to determine differences in the transcriptional activity of unmethylated and methylated vectors.

3T3-L1 adipocytes were cotransfected with miR-30a and miR-30c inhibitors or NC siRNA for 45 h, starved in Krebs-Ringer bicarbonate HEPES buffer (0.75 mmol/L CaCl₂ * 2 H₂O, 120 mmol/L NaCl, 4 mmol/L KH₂PO₄, 1 mmol/L MgSO₄ * 7 H₂O, 10 mmol/L NaHCO₃, and 30 mmol/L HEPES [pH 7.4]) for 2 h, and incubated with or without insulin (0 nmol/L [basal], 1.7 nmol/L, and 100 nmol/L insulin; Novo Nordisk, Bagsværd, Denmark) or cytochalasin B (10 μmol/L, Sigma-Aldrich) for 30 min. Glucose uptake was determined as previously described (27). Experiments were performed on five independent passages and in triplicates. Cytochalasin B values were subtracted, and results were normalized to protein concentrations measured with a Bradford protein assay and NC basal values (28). Paired t tests were used to determine differences between glucose uptake of mmu-miR-30a+c KD and NC cells.

Differentiated 3T3-L1 adipocytes (day 9) were cotransfected with miR-30a and miR-30c for 48 h as previously described and stimulated with or without 1 nmol/L insulin (Novo Nordisk) for 15 min. Protein samples were collected by lysis in cOmplete Protease Inhibitor Cocktail (1 tablet/50 mL), 1 mmol/L dithiothreitol, 1 mmol/L EDTA, 1 mmol/L EGTA, 5 mmol/L Na-pyrophosphate, 0.27 mol/L sucrose, 50 mmol/L NaF, 50 mmol/L Tris-Base, 1 mmol/L Na-orthovanadate, and 1% NP40, and their concentrations were measured with Bradford assay. Paired t tests were used to determine differences between KD and NC.

Western blot was performed using the following primary antibodies: AGPAT9 (GPAT3, HPA029414; Atlas Antibodies, Bromma, Sweden), AS160 (TBC1D4, 07-741; Sigma- Aldrich), phosphorylated AS160 T642 (TBC1D4, 44-1071G; MyBioSource, San Diego, CA), PKB (AKT1, 9272; Cell Signaling Technology, Danvers, MA), phosphorylated PKB S473 (AKT1, 44621G; Thermo Fisher Scientific), ELOVL6 (PA5-13455; Thermo Fisher Scientific), and HSP90 (610418; BD Biosciences, Franklin Lakes, NJ). GLUT1 antibody was kindly provided by Samuel W. Cushman (National Institutes of Health).

For Western blot analysis in cultured adipocytes, cell lysates containing LDS sample buffer (Thermo Fisher Scientific) were heated for 5 min in 95°C and loaded on precast Novex Bis-Tris 4–12% polyacrylamide gels (Thermo Fisher Scientific). Proteins were transferred to nitrocellulose membrane (Amersham Protran Western blotting membrane; Sigma-Aldrich) and blocked for 1 h in 10% milk in TBS-T (Tris-buffered saline, pH 7.6, and 0.1% Tween 20; Sigma-Aldrich). Membranes were incubated overnight in primary antibodies at 4°C diluted in TBS-T with 5% BSA (Sigma-Aldrich). After incubation in mouse or rabbit secondary antibody (NA931, Sigma-Aldrich, and A16096, Thermo Fisher Scientific), signals were visualized with ECL (SuperSignal West Pico and Femto Chemiluminescent Substrates; Thermo Fisher Scientific) and detected with a Bio-Rad ChemiDoc CCD camera and the Image Lab software (Bio-Rad Laboratories). Signals were quantified and normalized to HSP90.

Human adipose tissue biopsies were lysed in ice-cold radioimmunoprecipitation assay buffer (50 mmol/L Tris-HCl, pH 7.5; 150 mmol/L NaCl; 2 mmol/L EDTA; 1% Triton X-100; 0.5% Na-deoxycholate; and 0.1% SDS) containing cOmplete Protease Inhibitor Cocktail (1 tablet per 50 mL; Roche). Biopsies were homogenized in a TissueLyser II (QIAGEN). Protein concentrations were determined with the BCA method (Pierce BCA Protein Assay Kit; Thermo Fisher Scientific).

Total protein (5 µg) was mixed with sample buffer (60 mmol/L Tris-HCl, pH 6.8; 2% [w/v] SDS; 10% [v/v] glycerol; 2% [v/v] 2-mercaptoethanol), separated with gel electrophoresis (Criterion TGX Stain-Free Precast Gradient Gels; Bio-Rad Laboratories) and transferred to Low Fluorescence PVDF Membranes (0.45 µm; Bio-Rad Laboratories). Membranes were blocked in 5% (w/v) milk in TBS-T (0.05% [w/v] Tween-20), incubated with primary antibody diluted in 5% (w/v) BSA (HyClone; Cytiva) in TBS-T overnight at +4°C and horseradish peroxidaseconjugated secondary antibody (goat anti-rabbit IgG, RRID:AB_2099233; Cell Signaling Technology) for 1 h in room temperature. Proteins were detected with enhanced chemiluminescence (SuperSignal West Femto) in a ChemiDoc MP (Bio-Rad Laboratories) and quantified based on total protein normalization with stain-free gels in Image Lab version 6.1.0 (Bio-Rad V3 Western Workflow; Bio-Rad Laboratories).

Simpson-Golabi-Behemil syndrome (SGBS) cells were cultured and differentiated into mature adipocytes as previously described (29). Cells were used at day 9 with the differentiation rate 70–80% (Supplementary Fig. 3). Silencing was performed under the same experimental conditions and with the use of the same inhibitors for miR-30a, miR-30c, and NC as in 3T3-L1, due to 100% similarity between these human and mouse miRNAs. Cells were collected after 48 h, RNA was extracted and reverse transcribed, and efficiency of KD was determined (Supplementary Fig. 4).

RNA was extracted from miR-30a+c KD and NC SGBS adipocytes (five independent passages) and used for mRNA microarray analysis with Clariom S human arrays (Thermo Fisher Scientific) according to the manufacturer’s instructions. Array quality control, data normalization (Signal Space Transformation (SST)-RMA), and log2 transformation were performed with Transcriptome Analysis Console (TAC) Software (Thermo Fisher Scientific). Data were analyzed with eBayes ANOVA between KD and NC, calculated as condition (KD vs. NC) + repeated measures (passages). FDR <5% (q < 0.05) was considered significant.

We applied GSEA (30) to expression array data using KEGG pathways. Probes corresponding to transcripts were used and ranked according to the fold change.

The data sets generated in the course of the current study are available from the corresponding author upon reasonable request. No applicable resources were generated or analyzed during the current study.

We first analyzed expression of 408 miRNAs in adipose tissue from 12 MZ twin pairs discordant for T2D (Table 1). Thirty miRNAs (7.4% of those analyzed) were differentially expressed in the twin pairs (P < 0.05) (Table 2), which is more than expected by chance (P < 0.05, χ² test). The expression of the vast majority of the identified miRNAs (25 [83%]) was lower in the T2D than in non-T2D adipose tissue. Since miRNAs within the same family often target the same mRNAs, we asked whether the differentially expressed miRNAs in discordant twins represented the same family (or families), as defined by miRBase (release 22.1) (31). Indeed, three miRNA families were represented by at least two differentially expressed miRNAs (Table 2). The hsa-miR-30 and hsa-let-7 families, with 5 and 13 family members, respectively, were overrepresented among the dysregulated miRNAs, with 4 family members each downregulated in T2D twins (Fig. 1A). Further, intra–twin pair correlations revealed that the differences in log2 expression levels of 4 of the 30 identified dysregulated miRNAs (miR-128, miR-151b, miR-26b, and miR-30b) correlated with differences in fasting glucose levels (ρ = 0.61–0.67, P < 0.05) (Supplementary Table 1).

To test the robustness of the above observations, we next investigated the expression of hsa-miR-30a–d and hsa-let-7a, b, f, and g in adipose tissue from NGT and T2D unrelated subjects (Table 1). All miRNAs tested followed the same pattern as in T2D discordant twins, with reduced levels in adipose tissue of T2D subjects compared with that of control subjects (Fig. 1B) (all P ≤ 0.05, except for miR-30d [P = 0.1]).

To evaluate the possible functional involvement of these miRNAs in T2D, we next used TargetScan (release 7.2) (32) to predict conserved targets of the differentially expressed miRNAs. We then identified the conserved targets with differential expression in discordant twins, using previously published mRNA data (21) (Supplementary Table 2).

TargetScan predicted 1,576 conserved targets for the human miR-30 family; 285 of these exhibited differential expression in adipose tissue from the T2D discordant twins (P < 0.05) (Supplementary Table 2). The identified targets participate in pathways related to, e.g., membrane transport, amino acid metabolism, and development and regeneration (Supplementary Fig. 5). The targets included ELOVL6 and B4GALT6, two genes with the largest expression difference in adipose tissue of T2D twins versus twins without diabetes (21). Since ELOVL6 and B4GALT6 are important for lipid metabolic processes, we next analyzed the relationship between expression of hsa-miR-30a–d and ELOVL6 and B4GALT6, focusing on intra–twin pair correlations in the MZ twins. We observed significant intra–twin pair correlations between differences in ELOVL6 mRNA and differences in hsa-miR-30a levels, miR-30b levels, and miR-30c levels (Fig. 1C–E). Moreover, differences in B4GALT6 mRNA and differences in hsa-miR-30d levels were also significantly correlated (Fig. 1F).

Further TargetScan analysis predicted 1,207 conserved targets for the human let-7 family; 210 of these were differentially expressed in adipose tissue from T2D discordant twins (Supplementary Table 2). The predicted differentially expressed let-7 target genes are involved in pathways associated with, e.g., metabolism of terpenoids and polyketides, aging, and cancer (Supplementary Fig. 6). One target was AGPAT9, one of the genes with the largest expression difference in adipose tissue from T2D twins versus twins without diabetes (21). The difference in AGPAT9 expression between co-twins was significantly correlated in intra–twin pair analysis with the difference in hsa-let-7a levels (Fig. 1G). We further analyzed adipose tissue AGPAT9 protein levels in 8 T2D subjects and 13 control subjects without diabetes (Supplementary Table 3). There was no difference in protein levels between T2D versus control subjects but a significant negative correlation between AGPAT9 and BMI (r = −0.51, P = 0.02) (Supplementary Fig. 7A), suggesting that AGPAT9 might be related to obesity rather than T2D. Indeed, also let7a (r = −0.65, P = 0.02), let7b (r = −0.68, P = 0.015), and AGPAT9 mRNA (r = −0.58, P = 0.046) levels correlate negatively with BMI in analysis of the 12 twins without diabetes in the discordant twin cohort (Supplementary Fig. 7B–D). AGPAT9 is a glycerol-3-phosphate acyltransferase, which catalyzes the initial step of de novo triacylglycerol synthesis.

Since miRNA expression could be regulated by methylation, we analyzed DNA methylation of CpG sites annotated to miRNA genes (based on Illumina manifest and genome build 37) in adipose tissue from 14 twin pairs discordant for T2D. In the current study we only assessed DNA methylation of sites annotated to miRNA genes; results from the analysis of DNA methylation data not focusing on sites annotated to miRNAs have previously been presented (21). A total of 147 sites, which are included on and could be analyzed with the Illumina Infinium HumanMethylation450 BeadChip, were annotated to the 30 miRNAs with differential expression in discordant twins. Ten of these sites, annotated to nine different miRNAs, exhibited differential methylation between T2D twins and twins without diabetes (P < 0.05, 6.8% of analyzed sites) (Table 3). Methylation of the majority of these sites (8 sites [80%]) was increased in T2D twins compared with twins without diabetes.

Next, to test the validity of the methylation differences identified in the discordant twins, we analyzed DNA methylation in adipose tissue from the cohort of unrelated subjects. We analyzed methylation of sites annotated to the 30 differentially expressed miRNAs in the discordant twin cohort, in 28 T2D and 28 non-T2D subjects. We identified 26 differentially methylated sites annotated to 14 miRNAs (P < 0.05, 18.1% of 144 analyzed sites) (Table 3), which is more than expected by chance (P < 0.05, χ² test). Methylation of the majority of the identified sites (23 [88%]) was higher in T2D subjects than in subjects without diabetes. Five of 10 sites with differential methylation in the discordant twins were replicated in this case-control cohort (Table 3), including cg16506910 in the region within 1,500 bp of the transcription start site (TSS1500) of hsa-miR-30a and cg26371705 and cg15702185 in TSS1500 of hsa-let-7a-3 (Fig. 2A–C).

Based on the DNA methylation data in the two cohorts, supporting epigenetic regulation of miRNA expression in adipose tissue, we proceeded to test the effects of promoter region methylation of selected miRNA genes on their transcriptional activity. We analyzed methylation of the promoters of hsa-miR-30a and hsa-let-7a-3, since their methylation was increased and expression decreased in adipose tissue of T2D subjects (Figs. 1A and B and 2A–C). Luciferase reporter gene assays revealed that methylation of the proximal promoter region (1,500 bp upstream) of hsa-miR-30a significantly suppressed its transcriptional activity compared with unmethylated promoter (Fig. 2D). A similar effect was observed for hsa-let-7a-3, although it did not reach statistical significance (Fig. 2E).

To better understand the biological relevance of the identified miRNAs in adipocytes, we tested the effect of silencing specific miRNAs in vitro. We selected the miR-30 family for in vitro functional analysis based on the downregulation of four of five family members in adipose tissue of T2D subjects (Fig. 1A and B) and the identified epigenetic regulation of miR-30a (Fig. 2A and D). We silenced the expression of mmu-miR-30a, b, c, and d in mature 3T3-L1 adipocytes (Fig. 3A). We then analyzed the transcriptome of the silenced cells using microarrays and found 732 significantly downregulated genes (q < 0.05) and 246 significantly upregulated genes (q < 0.05) in 3T3-L1 adipocytes where mmu-miR-30a, b, c and d had been silenced compared with NC (Supplementary Tables 4 and 5). The 50 genes with the largest expression differences between mmu-miR-30–silenced adipocytes and NC (25 upregulated and 25 downregulated) are shown in Table 4. According to a literature search (articles deposited in the PubMed database prior to 2020) on the relationship of these genes with T2D, diabetes complications, or obesity in human, animal models, or cells, approximately one-third of these top 50 genes are linked with one or more of these search terms, i.e., the gene encoding AMP-activated protein kinase α2 (Prkaa2/Ampka2) (Table 4).

We then compared the list of genes with significantly altered expression after mmu-miR-30 silencing (Supplementary Tables 4 and 5) with the differentially expressed genes in adipose tissue from MZ twins discordant for T2D (21). This revealed 17 common genes regulated in the same direction, after conversion of mouse genes to human orthologs, via the biomaRt package (Bioconductor, release 3.10). Two genes (Ctsz and Evc2) were upregulated and 15 genes were downregulated in both data sets (human T2D adipose tissue and 3T3-L1 miR-30 KD) (Fig. 3B). Ctsz, encoding cathepsin Z, has been associated with inflammatory conditions in different tissues and has been proposed to be used as a clinical marker for systemic inflammation in humans (33). The downregulated genes were associated with signal transduction (Braf) or lipid, amino acid, and carbohydrate metabolism (Hadh, Acat1, Aldh6a1, and Hibch). Interestingly, Tet1, encoding CpG demethylase, an enzyme involved in regulation of DNA methylation, was downregulated in both human T2D adipose tissue and 3T3-L1 miR-30 KD adipocytes and is a conserved predicted target of the miR-30 family.

We next used GSEA to identify sets of biologically related genes that are altered in the miR-30–silenced adipocytes. Pathway analysis revealed 41 significantly upregulated gene sets and 15 significantly downregulated gene sets (q < 0.05) (Supplementary Table 6). The upregulated pathways included gene sets involved in regulation of translation, metabolism of RNA, and the immune system (Fig. 3C). Downregulated pathways included gene sets related to the metabolism of carbohydrates and lipids and to vesicle-mediated transport (Fig. 3D).

We analyzed the overlap between differentially expressed genes and in silico predicted miR-30 targets. TargetScan analysis predicted 1,244 conserved targets for the mouse miR-30 family (92.5% matched the predicted targets of the miR-30 family in human). Eight of the predicted target genes were upregulated and 84 were downregulated in miR-30–silenced 3T3-L1 adipocytes (q < 0.05) (Supplementary Table 7). Of these, the target gene Rhebl1 exhibited the largest increase in expression in response to miRNA silencing (log2 FC = 0.67).

Furthermore, we detected 81 downregulated and 14 upregulated transcription factor (TF) genes in silenced adipocytes (q < 0.05) (Supplementary Table 8). Expression of Irf7 encoding the interferon regulatory factor 7 was significantly upregulated (log2 FC = 1.33). The expression of 13 TF genes recognized as conserved predicted targets in mouse by TargetScan was downregulated after silencing, including PGC1α (log2 FC = –0.32).

Finally, we used the RcisTarget package (34) (Bioconductor, release 3.10) to search for DNA-binding motifs 5,000 bp upstream and 5,000 bp downstream of the TSS regions of significantly regulated genes in the silenced adipocytes. Indeed, we identified enriched motifs annotated to the IRF family (IRF1, IRF2, IRF4, IRF7, and IRF9) in promoter regions of significantly upregulated genes (Supplementary Table 9). Promoters of the significantly downregulated genes contained enriched motifs associated with ELK4, TAF1, and PHF8 TFs (Supplementary Table 9). Collectively, the data presented above indicate that miR-30 regulates key genes involved in T2D pathology, overlapping with genes differentially expressed in human adipose tissue from T2D versus control subjects.

The increased fasting plasma and 2-h glucose levels (Table 1), the decreased expression of hsa-miR-30a–d (Table 2) in subjects with T2D, and the positive intra–twin pair correlation between hsa-miR-30b and fasting glucose levels (Supplementary Table 1) suggested that the miR-30 family is involved in regulation of glucose metabolism or is regulated by hyperglycemia. We therefore investigated the possible effect of mmu-miR-30a–d silencing on glucose uptake and insulin signaling in 3T3-L1 adipocytes and the effect of 24-h exposure to high glucose concentration on miR-30a, b, c and d levels in these cells. mmu-miR-30 silencing in 3T3-L1 adipocytes resulted in a significantly reduced glucose uptake both in the basal state and in presence of insulin compared with NC (P < 0.05) (Fig. 3E). There was either no effect (1.7 nmol/L) or an increase (100 nmol/L) in the induction of glucose uptake by insulin, i.e., the fold increase compared with the basal, when silencing miR-30 (Supplementary Fig. 8). In search for mechanisms underlying the reduction in glucose uptake, we investigated phosphorylation/activity status of Akt and its substrate TBC1D4—a Rab-GAP whose Akt-mediated phosphorylation is important for GLUT4 translocation to the membrane (35). There was no change in protein levels or phosphorylation of Akt at the activity-controlling site Ser473 after silencing of miR-30 (Supplementary Fig. 9A and B). However, we found a consistent and marked reduction in total TBC1D4 phosphorylation at the Akt site Thr642 in miR-30–silenced 3T3-L1 adipocytes in the basal state (Fig. 3F) and a nominal reduction in the presence of insulin (Fig. 3G). We also detected reductions in protein levels and specific phosphorylation (phosphorylation/protein level) of TBC1D4, following miR-30 silencing, although these differences were smaller than for total phosphorylation (Fig. 3F). Also, in line with the reduced glucose uptake, mmu-miR-30 silencing resulted in reduced expression of the glucose transporter Glut1 (Slc2a1) (q < 0.05) (Supplementary Table 5). However, GLUT1 protein expression was not changed in miR-30–silenced 3T3-L1 adipocytes (Supplementary Fig. 10). Expression of Glut2, 3, and 4 was not affected by miR-30 silencing (Supplementary Table 5).

Direct exposure of 3T3-L1 adipocytes to high glucose (50 vs. 25 mmol/L glucose) for 24 h had no effect on miR-30a, b, c, or d expression (Fig. 3H).

We finally tried to validate some of our findings from human T2D adipose tissue and 3T3-L1 miR-30 KD adipocytes in human adipocytes where miR-30 was silenced. Here, we analyzed the transcriptome of miR-30–silenced human SGBS adipocytes using microarray and found 668 significantly downregulated genes (q < 0.05) and 249 significantly upregulated genes (q < 0.05) compared with NC (Supplementary Table 10). We then studied the overlap of differentially expressed genes (q < 0.05) in miR-30–silenced 3T3-L1 versus SGBS adipocytes after conversion of mouse genes to human orthologs (biomaRt package, Bioconductor, release 3.10) and identified 111 common genes regulated in the same direction, which is more than expected by chance (P < 0.0001, χ² test [marked in Supplementary Table 10]).

We next compared the list of genes with significantly altered expression after miR-30 silencing in SGBS adipocytes (Supplementary Table 10) with the differentially expressed genes in adipose tissue from MZ twins discordant for T2D, where also we discovered reduced miR-30 expression (21) (Fig. 1A). This revealed 10 common genes regulated in the same direction including S100A4, which was upregulated, as well as ELOVL6 and TET1, which were downregulated, in both data sets (human T2D adipose tissue and SGBS miR-30 KD) (Fig. 3I). Notably, S100A4 was recently identified as a novel adipokine associated with insulin resistance and inflammation/adipocyte hypertrophy (36). We then used GSEA to identify sets of biologically related genes that are altered in both the miR-30–silenced human adipocytes and in human T2D adipose tissue. GSEA revealed that the pathway “biosynthesis of unsaturated fatty acids” is downregulated in both SGBS miR-30 KD and twins with T2D (21) (Fig. 3J and Supplementary Table 11).

Next, we analyzed the overlap between differentially expressed genes in miR-30–silenced SGBS adipocytes and in silico predicted miR-30 targets. TargetScan analysis predicted 1,576 conserved targets for the human miR-30 family. Fifteen of the predicted target genes were upregulated and 117 were downregulated in miR-30–silenced SGBS adipocytes (q < 0.05) (Supplementary Table 12). Twenty-four target genes were differentially expressed in both miR-30–silenced 3T3-L1 adipocytes and SGBS adipocytes (marked in Supplementary Table 12). Interestingly, TET1 and ELOVL6 were downregulated in SGBS miR-30 KD adipocytes and human T2D adipose tissue and are conserved predicted targets of the human miR-30 family. In line with this, there was a trend toward lower ELOVL6 protein levels in SGBS adipocytes after miR-30 silencing (Supplementary Fig. 11).

The role of miRNAs in the pathology of T2D and their potential as therapeutic targets remain to be explored adequately. In this study, we identified a set of miRNAs that were differentially expressed in T2D by using adipose tissue from two human cohorts. We were able to demonstrate the potential regulation of miRNA expression by DNA methylation and the physiological consequences of miRNA level dysregulation, reminiscent of the T2D pathology, in vitro (Fig. 4).

We identified 25 miRNAs as downregulated and 5 as upregulated in twins with T2D compared with their corresponding co-twins without T2D. The hsa-miR-30 and hsa-let-7 families were overrepresented among the dysregulated miRNAs in discordant twins, with four members of each family downregulated in T2D individuals, a finding confirmed in the case-control cohort. The coordinated behavior of miRNAs belonging to the same family and, thus, targeting the same mRNAs is thought to strengthen their concerted effect on gene regulation. The miR-30 and let-7 family members have previously been shown to be associated with T2D pathology, as they affect adipogenesis, glucose metabolism, and inflammatory processes (37–40). For example, hsa-miR-30e and hsa-let-7i were shown to be downregulated in muscle from MZ twins with T2D compared with their co-twins without diabetes (19). Additionally, mir-30a, miR-30c, and miR-30e were downregulated in adipose tissue macrophages of mouse fed a high-fat diet, affecting Notch1 signaling and pro-inflammatory cytokine production (41).

Our previous analysis of mRNA expression in adipose tissue from MZ twins discordant for T2D (21) revealed decreased expression of energy metabolism genes and increased expression of inflammatory genes in subjects with T2D. Based on the findings of the current study, we speculate that altered miRNA levels in T2D subjects could underpin this expression pattern. To investigate this further we therefore performed experiments where we silenced miRNAs in cultured adipocytes. The main part of these experiments was performed in the well-characterized mouse 3T3-L1 cell line, in which protocols for gene silencing and the study of insulin action were already established. However, we also used the SGBS cell strain to validate our results in cells of human origin. Indeed, silencing of the miR-30 family in cultured adipoctes affected expression of almost 1,000 genes, with dysregulated genes related to glucose and lipid metabolism, and the immune system, similar to the pattern seen in adipose tissue from T2D subjects. This suggests that the downregulation of miR-30 in T2D subjects can, at least partially, be responsible for changes observed in human T2D adipose tissue. miRNA regulatory networks are enormously complex because hundreds of miRNAs might target one mRNA, either directly or indirectly, e.g., through TFs. In fact, we detected nearly 100 up- or downregulated genes encoding TFs in miR-30–silenced 3T3-L1 adipocytes. Irf7 is of particular interest among the upregulated TF genes: it is a predicted miR-30 target, and the promoter regions of significantly upregulated genes in miR-30–silenced adipocytes are enriched with TF-binding motifs annotated with the IRF family. Moreover, IRFs are recognized transcriptional regulators of adipogenesis (42).

Analysis of overlapping genes in the adipose tissue mRNA data set from T2D discordant twins and 3T3-L1 adipocytes lacking miR-30a–d revealed genes related to fatty acid or amino acid metabolism with links to T2D and/or adipocytes, including HADH, ACAT1, ALDH6A1, and HIBCH. Notably, downregulation of the genes encoding acetyl-CoA acetyltransferase 1 (ACAT1) and aldehyde dehydrogenase 6 family, member A1 (ALDH6A1), has previously been observed in adipose tissue from obese T2D patients (43). Analysis of overlapping genes in the adipose tissue mRNA data set from T2D discordant twins and human SGBS adipocytes lacking miR-30a–d also revealed genes related to fatty acid metabolism with links to T2D and/or adipocytes. In particular, ELOVL6, encoding a key enzyme in the elongation of long-chain fatty acids, was downregulated in both T2D subjects and miR-30–silenced SGBS adipocytes.

We detected only a small overlap between the in silico predicted targets and significantly upregulated genes in miR-30–silenced adipocytes. This could indicate a primary involvement of miR-30 in posttranscriptional regulation, with a minor effect on mRNA levels, or the fact that not all mRNA targets are expressed in adipocytes. This could also be a matter of timing, since some mRNAs might be degraded only after an initial inhibition of translation (44). Interestingly, more genes were significantly downregulated than upregulated in miR-30–silenced adipocytes. According to previous studies, silencing of a miRNA leads to both up- and downregulation of mRNA and protein levels, suggesting that the ability of miRNAs to upregulate mRNA and protein levels may be more common than currently appreciated. Upregulation may, for example, be a result of the miRNA blocking the binding of repressing factors to mRNA (45), or it could be caused by competition of multiple miRNAs for the same mRNA target (46).

Although changes in miRNA expression have been demonstrated in various diseases, knowledge of the underlying mechanisms is limited. miRNA biogenesis is regulated on multiple levels, including transcription, processing, and editing. A link between epigenetic changes and miRNAs has been described for several physiological processes, and miRNA expression can be downregulated by methylation of the promoter (8–12). In the current study we demonstrated, for the first time, the association of differential DNA methylation with altered miRNA expression in adipose tissue from subjects with T2D. Methylation of five sites annotated to differentially expressed miRNAs was increased in adipose tissue from T2D subjects in the two cohorts analyzed, including sites in the promoters of hsa-miR-30a and hsa- let-7a-3. In this study we analyzed DNA methylation using the Illumina Infinium HumanMethylation450 array. Importantly, we have previously both technically and biologically validated DNA methylation data generated using this array in several human cohorts via other methods (9,47). Further, the luciferase reporter experiments confirmed that methylation of the hsa-miR-30a promoter suppresses expression of that gene. This is in line with another study where miR-30a promoter hypermethylation correlated with reduced expression (48). DNA hypermethylation in the promoter and decreased hsa-let-7a-3 expression have been observed in blood from patients with diabetic nephropathy (49). Methylation of the hsa-let-7a-3 promoter in the luciferase experiments in our study nominally suppressed the miRNA expression, although the effect was not statistically significant. This suggests that DNA methylation might be one, but not the sole, mechanism regulating let-7 levels. For instance, the long noncoding RNA H19 is an important regulator of the let-7 family (50). Additional epigenetic mechanisms, such as histone modifications, may also play a role in miRNA expression regulation, which should be investigated in future studies.

Differentially expressed miRNAs could be an effect and/or a cause of metabolic imbalance. It is important to distinguish the primary mechanisms relevant to development of T2D from secondary changes caused by the disease. The observations of dysregulated miRNAs in T2D adipose tissue and intra–twin pair correlations between differences in fasting glucose and miRNA levels suggest that these miRNAs are either involved in regulation of glucose metabolism or are regulated by hyperglycemia as an effect secondary to T2D. Accordingly, we focused on the miR-30 family to investigate its potential role in adipocyte glucose metabolism in detail. Downregulation of miR-30 in adipocytes resulted in decreased glucose uptake in the basal state and to a lesser extent in the presence of insulin. In line with this, miR-30 silencing reduced phosphorylation of TBC1D4, which in its unphosphorylated form contributes to anchoring of GLUT4-containing vesicles in the cytosol (35). Notably, another study found lower TBC1D4 phosphorylation in adipose tissue from T2D subjects versus subjects without diabetes (51). As for glucose uptake, phosphorylation of TBC1D4 was in our study markedly reduced in the basal state and slightly less so in the presence of insulin. The mechanism underlying reduced TBC1D4 phosphorylation does not appear to involve changes in Akt activity, as Akt expression and phosphorylation were unaffected by miR-30 silencing. Direct exposure of 3T3-L1 adipocytes to high glucose did not affect miR-30 expression, suggesting that the reduced miR-30 expression in T2D adipose tissue is not caused by hyperglycemia.

The current study has some limitations. First, although the concept of MZ twin pairs discordant for T2D makes for a powerful study design, these twins are rare and the relatively small sample size increases the risk of type II errors. Additional dysregulated miRNAs might have been identified in a larger cohort. However, the performed validation in an independent case-control cohort supports the biological relevance of the presented findings in relation to T2D. Further, it is well established that the methylome is heritable (21,52–54), and this may explain why we identified more DNA methylation differences in unrelated subjects from the case-control cohort than in the MZ discordant twins.

Understanding the miRNA-guided network of gene expression regulation can potentially provide new tools for the diagnosis and therapy of many human diseases. Several years ago, the miR-34a mimic MRX34, a master regulator of tumor suppression, was the first miRNA mimic to reach phase 1 studies (55). Recent data suggest that miRNAs may constitute attractive therapeutic targets for T2D too by improving β-cell function (14) or insulin sensitivity (15), and our findings help substantiate this statement. In the future, in vivo delivery of miRNA mimics or anti-miRNAs specifically in insulin target tissues could permit correcting the level of key miRNAs under diabetes conditions and lead to new strategies for treating the disease.

Acknowledgments. The authors thank Swegene Center for Integrative Biology at Lund University (SCIBLU) genomics facility for help with global DNA methylation and miRNA and mRNA expression analyses as well as the Bioinformatics and Expression Analysis core facility at Karolinska Institute for help with mRNA expression analyses. The authors thank Jette Bork-Jensen (University of Copenhagen, Copenhagen, Denmark) for valuable miRNA discussions, Martin Wabitsch (University of Ulm, Ulm, Germany) for providing the SGBS cell line, and Kasper Pilgaard for excision of human adipose tissue from three individuals with T2D and three control subjects at Steno Diabetes Center Copenhagen. The authors thank the Nordic Network for Clinical Islet Transplantation (JDRF award 31-2008-413) and the tissue isolation teams and Human Tissue Laboratory within EXODIAB/Lund University Diabetes Centre for providing human adipose tissue from 5 individuals with T2D and 10 control subjects.

Funding. This work was supported by the Swedish Research Council (grants Dnr 2016-02486, 2018-02567, and 2019-01406 and Strategic Research Area Exodiab Dnr 2009-1039), Region Skåne and ALF, the Novo Nordisk Foundation, the Swedish Foundation for Strategic Research (Dnr IRC15-0067), the Syskonen Svensson Foundation, the Diabetes Foundation, Kungliga Fysiografiska Sällskapet i Lund, Magnus Bergvall Foundation, Åke Wiberg Foundation, the European Foundation for the Study of Diabetes/Lilly Programme, the Söderberg Foundation, and the Påhlsson Foundation. The Swedish twins were recruited from the Swedish Twin Registry, which is supported by grants from the Swedish Department of Higher Education and the Swedish Research Council.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. E.N. and C.L. designed the study, researched data, and wrote the manuscript. M.V. performed in vitro experiments, analyzed data, and wrote the manuscript. J.S. performed in vitro experiments and analyzed data. A.P. analyzed data. P.-A.J. and P.P. collected the clinical material and data. J.L.S.E., L.E., and O.G. supervised and contributed to the experimental design. A.V. contributed to study design and collected the clinical material and data. All authors reviewed and approved the final version of the manuscript. E.N. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.