Altered Expression of miR-181a-5p and miR-23a-3p Is Associated With Obesity and TNFα-Induced Insulin Resistance

Abstract

Context

The proinflammatory cytokine TNFα is a key player in insulin resistance (IR). The role of miRNAs in inflammation associated with IR is poorly understood.

Objective

To investigate miR-181a-5p and miR-23a-3p expression profiles in obesity and to study their role in TNFα-induced IR in adipocytes.

Design

Two separate cohorts were used. Cohort 1 was used in adipose tissue (AT) expression studies and included 28 subjects with body mass index (BMI) <30 kg/m² and 30 with BMI ≥30 kg/m². Cohort 2 was used in circulating serum miRNA studies and included 101 subjects with 4 years of follow-up (48 case subjects and 53 control subjects). miR-181a-5p and miR-23a-3p expression was assessed in subcutaneous and visceral AT. Functional analysis was performed in adipocytes, using miRNA mimics and inhibitors. Key molecules of the insulin pathway, AKT, PTEN, AS160, and S6K, were analyzed.

Results

Expression of miR-181a-5p and miR-23a-3p was reduced in adipose tissue from obese and diabetic subjects and was inversely correlated to adiposity and homeostasis model assessment of IR index. Overexpression of miR-181a-5p and miR-23a-3p in adipocytes upregulated insulin-stimulated AKT activation and reduced TNFα-induced IR, regulating PTEN and S6K expression. Serum levels of miR-181a-5p were reduced in case vs control subjects at baseline, suggesting a prognostic value. Variable importance in projection scores revealed miR-181a-5p had more effect on the model than insulin or glucose at 120 minutes.

Conclusion

miR-181a-5p and miR-23a-3p may prevent TNFα-induced IR in adipocytes through modulation of PTEN and S6K expression.

Obesity is associated with chronic low-grade inflammation of white adipose tissue (AT), which can subsequently lead to insulin resistance (IR), impaired glucose tolerance, and, ultimately, diabetes (1). TNFα is a proinflammatory cytokine whose expression in AT is elevated in obesity, where it can contribute to the modulation of lipid metabolism by altering insulin signaling (2, 3).

microRNAs (miRNAs) are small (17 to 24 nucleotides long) noncoding RNAs that bind to 3′-untranslated regions (3′-UTRs) of target mRNAs to regulate gene expression by translational repression or degradation. miRNAs regulate metabolic processes associated with type 2 diabetes mellitus (DM), including insulin signaling and glucose homeostasis (4), highlighting their potential as therapeutic targets for obesity and metabolic syndrome. Moreover, the finding of circulating miRNAs in biological fluids supports the potential utility of harnessing miRNAs as biomarkers in diseases ranging from cancer to DM (5). Along this line, differential expression of miRNAs in tissues has been reported in obese vs nonobese humans and in humans and animals with diabetes (6). However, the function of most miRNAs that are deregulated in obesity and IR is unknown.

miRNAs may provide a link between inflammation in obesity and IR. TNFα is a major initiator of inflammation and differentially regulates miRNA expression in several contexts (7, 8). Although various miRNAs modulated by TNFα have been described in AT, little is known about those directly involved in the regulation of the insulin pathway in mature adipocytes (6, 9, 10). Consequently, the underlying mechanisms linking miRNAs and TNFα-induced IR in adipocytes remain to be deciphered.

A recent microarray analysis identified miR-181a and miR-23a as being deregulated in blood from obese and nonobese subjects with and without DM (11). Furthermore, miR-181a expression has been found to be inversely related to adiponectin levels in AT (12), and its expression is known to alter hepatic insulin sensitivity (13). In relation to miR-23, diminished expression levels have been detected in the plasma of patients with cardiovascular disease (14) and a recent study has implicated miR-23 in glucose metabolism in the muscle of dogs (15).

Given the scarcity of data on miR-181a-5p and miR23a-3p within the context of obesity and TNFα-induced IR, we aimed in this study to examine the expression of these miRNAs in AT from obese and nonobese subjects, to investigate their role in TNFα-induced IR, and to test their possible functional mechanism in adipocytes. We also explored their potential utility as predictive serum markers of prediabetes in a prospective case-control study.

Methods

Subjects

Cohort 1: AT expression

We recruited 28 subjects with BMI <30 kg/m² and 30 subjects with BMI ≥30 kg/m², age and sex matched, at Hospital Joan XXIII, Tarragona, Spain (Table 1). All patients were white and reported that their body weight had been stable for at least 3 months before the study. They had no systemic disease other than obesity, and all had been free of any infections in the month before the study. Liver and renal diseases were specifically excluded by biochemical workup. Samples of visceral AT (VAT) and subcutaneous AT (SAT) were obtained from the same patient during elective abdominal surgical procedures. Samples and data from patients included in this study were provided by the BioBanc IISPV (B.0000853+B.0000854) integrated into the Spanish National Biobanks Network, and were processed following standard operating procedures with appropriate approval of the ethics and scientific committees.

Table 1.

Baseline Characteristics of the Participants in the Study

	BMI <30 (n = 28)	BMI ≥30 (n = 30)	P
Age, y	52.67 ± 14.84	45.8 ± 7.84	0.123
Sex, male/female	12/16	9/21	0.309
BMI, kg/m2	25.74 ± 2.44	43.7 ± 4.91	<0.001
Waist, cm	87.28 ± 11.77	127.24 ± 11.44	<0.001
Hip, cm	97.68 ± 11.53	142.16 ± 12.29	<0.001
Waist-to-hip ratio	0.89 ± 0.09	0.89 ± 0.09	0.515
SBP, mm Hg	129.57 ± 15	140.03 ± 21.84	0.044
DBP, mm Hg	72.89 ± 8.05	90.5 ± 10.85	<0.001
Glucose tolerance, normal /IGT/DM	22/0/6	12/11/7	0.001
Glucose, mM	5.7 ± 1.27	5.72 ± 0.99	0.413
Insulin, pmol/L	51.32 ± 66.67	116.95 ± 63.96	<0.001
HOMA-IR	1.47 ± 1.15	4.26 ± 2.31	<0.001
Cholesterol, mM	5.05 ± 1.16	4.89 ± 1.17	0.486
HDL cholesterol, mM	1.32 ± 0.36	1.16 ± 0.24	0.101
LDL cholesterol, mM	3.7 ± 1.03	3.69 ± 1.11	0.935
Triglycerides, mM	1.3 ± 0.36	1.7 ± 1.12	0.013
CRP, mg/L	0.99 ± 0.56	0.89 ± 0.58	0.564

Data given as mean ± standard deviation, unless otherwise indicated.

Abbreviations: CRP, C-reactive protein; DBP, diastolic blood pressure; HOMA-IR, homeostasis model assessment of insulin resistance index; HDL, high-density lipoprotein; IGT, impaired glucose tolerance; NGT, normal glucose tolerance; SBP, systolic blood pressure.

Cohort 2: miRNA circulating serum analysis

We selected 101 subjects from the Pizarra study (16) with 4 years of follow-up (48 incident cases of diabetes and 53 control subjects). The Pizarra study is a population-based prospective study undertaken in a population from southern Spain. The characteristics of the study have been reported elsewhere (16).

Analytical methods

Venous blood samples were collected between 9:00 am and 10:00 am or 2 hours after an oral glucose tolerance test. Samples were centrifuged at 4°C, and serum and plasma from each subject were stored at −80°C for later analysis. Plasma glucose, cholesterol, triglyceride, high-density lipoprotein cholesterol, and insulin levels were measured as described previously (17). IR was determined by the homeostasis model assessment of insulin resistance index (HOMA-IR). Cytokines (TNFα and its receptors R1 and R2, and IL-6) were measured by enzyme immunoassay, as described by Rubio-Martín et al. (18). Leptin, adiponectin, FABP4, REDOX, sex hormone–binding globulin levels were measured by enzyme-linked immunosorbent assay, as described previously (18).

Cell culture and treatments

Human preadipocytes were purchased from the European Collection of Cell Cultures (Salisbury, United Kingdom). The Simpson Golabi Behmel Syndrome (SGBS) cell line was kindly provided by Dr. M. Wabitsch (University of Ulm, Ulm, Germany). Cells were differentiated to adipocytes as described by Fischer-Posovszky et al. (19) and then incubated either with or without 50 ng/mL TNFα for 8 hours. Cell lysates were collected for RNA extraction.

miRNA mimics and inhibitors

Mimic miRNAs (miScript mimic-miR-181a-5p, miScript mimic-miR-23a-3p), nontarget control small interfering RNA (siRNA) and miRNA inhibitors (miScript anti-miR-181a-5p inhibitor and miScript antimiR-23a-3p inhibitor), and a negative control inhibitor were all purchased from Qiagen (Madrid, Spain). SGBS adipocytes were transfected at day 9 of differentiation with 50 nM of mimic or 50 nM of inhibitor in 0.66 µL/cm² Lipofectamine 2000 (Thermo Fisher Scientific, Madrid, Spain). See Supplemental Methods for dose-response assays. Twenty-four hours after transfection, adipocytes were left either unstimulated or were stimulated with 100 nM insulin (Actrapid; Novo Nordisk, Bagsvaerd, Denmark) for 10 minutes. In some experiments, posttransfected adipocytes were stimulated with 50 ng/mL TNFα for 8 hours, followed by a 10-minute stimulus with 100 nM insulin. Cells were then collected for protein analysis.

Luciferase reporter assays

Potential 3′-UTR specific binding sites for miRNAs were predicted by Microrna.org and microT-CDS (http://diana.imis.athena-innovation.gr), revealing potential sites for miR-181a-5p and miR-23a-3p in PTEN and S6K genes. LightSwitch 3′-UTR reporter GoClone RenSP (Active Motif, Carlsbad, CA) luciferase reporter constructs with the full-length 3′-UTR sequence of PTEN or S6K were cotransfected into HEK293 for 24 hours, as detailed in Supplemental Methods.

RNA isolation, complementary DNA synthesis, and real-time polymerase chain reaction

Frozen AT (400 to 500 mg) was homogenized with an Ultra-Turrax 8 (Ika, Staufen, Germany). Tissue total RNA was extracted with the RNeasy Lipid Tissue Midi Kit (Qiagen Science, Hilden, Germany). Adipocyte total RNA was extracted with the miRCURY RNA Isolation Kit–Cell & Plant (Exiqon, Vedbaek, Denmark). RNA quality control was assessed spectrophotometrically by Xpose (Trinean, Gentbrugge, Belgium).

The miRCURY RNA Isolation Kit–Biofluids (Exiqon) was used to extract total miRNA from serum. The Universal cDNA Synthesis Kit II (Exiqon) was used for total RNA retrotranscription. Quantitative reverse transcription polymerase chain reaction gene expression was performed using the ExiLENT SYBR Green master mix (Exiqon). miRNA expression levels were quantified on the 7900HT Fast Real-Time PCR platform (Applied Biosystems). Data were analyzed by RQ Manager software (Supplemental Methods).

Western blotting

Cellular proteins were subjected to sodium dodecyl sulfate–polyacrylamide gel electrophoresis, transferred to nitrocellulose membranes, and western blotting was performed using standard protocols. The following primary antibodies were used: pAKT, AKT, pAS160, AS160, S6K, and PTEN, all from Cell Signaling Technologies (Werfen, Barcelona, Spain). An antibody to β-actin was purchased from Sigma-Aldrich (Madrid, Spain). Detailed information is given in Supplemental Methods.

Statistical analysis

For clinical and anthropometrical variables, data are expressed as mean (standard deviation) or median (25th to 75th) quartiles, when appropriate. Differences in clinical or laboratory parameters or expression variables between groups were compared using the Kruskall-Wallis one-way analysis and Mann-Whitney U test for nonnormally distributed data or one-way analysis of variance with post hoc Bonferroni correction and Student t test for normally distributed data. The χ² test was used for categorical data to assess differences among groups. For paired plasma samples, Wilcoxon signed-rank test was performed. Spearman Rho test was used to assess the strength of association correlations between variables. Receiver operating characteristic (ROC) curve analysis was performed to evaluate the best predictive model. Variable importance in projection analysis and partial least squares discriminant analysis (PLS-DA) models were developed using the R programming platform. For in vitro studies, mean comparison was performed by one-way analysis of variance and Student t test. Statistical analysis was performed using Statistical Package for the Social Sciences, version 19 (IBM, Armonk, NY). Significance was considered at P < 0.05.

Results

miR-181a-5p and miR-23a-3p expression is reduced by TNFα in human adipocytes in vitro

TNFα is known to alter the expression of miR-181a-5p and miR-23a-3p in several cell types (20, 21). To determine whether this also occurred in human mature adipocytes, we treated differentiated human primary preadipocytes with 50 ng/mL TNFα for 8 hours and measured miRNA expression using quantitative reverse transcription polymerase chain reaction. The expression of both miR-181a-5p and miR-23a-3p was significantly lower in treated than in untreated adipocytes [Fig. 1(a)], whereas the expression of miR-155-5p, a validated TNFα-regulated miRNA (22), was elevated under the same conditions. Identical results were obtained in differentiated SGBS adipocytes [Fig. 1(a)], leading us to hypothesize that these microRNAs may be deregulated in inflammatory-activated obese AT.

Figure 1.

miR-181a-5p and miR-23a-3p expression is reduced in TNFα-treated adipocytes and in obese human AT. (a) miR-181a-5p and miR-23a-3p expression is reduced by TNFα in HAds. Mature HAds and SGBS adipocytes were stimulated at day 9 of differentiation with 50 ng/mL TNFα for 8 hours. Relative levels of miRNAs [mean ± standard error of the mean (SEM)] from four independent experiments are shown. *P < 0.05; **P < 0.01; ***P < 0.001. (b) Relative levels (mean ± SEM) of miR-181a-5p and miR-23a-3p expression from SAT and VAT samples stratified according to BMI. *P < 0.05; **P < 0.01; ***P < 0.001. (c) Relative levels (mean ± SEM) of miR-181a-5p expression from SAT and VAT samples stratified according to BMI and glucose tolerance status (i.e., NGT, impaired glucose tolerance, and DM). *P < 0.05; **P < 0.01; ***P < 0.001. (d) Relative levels (mean ± SEM) of miR-23a-3p expression from SAT and VAT samples stratified according to BMI and glucose tolerance status *P < 0.05, **P < 0.01, ***P < 0.001. HAd, human adipocyte.

miR-181a-5p and miR-23a-3p expression in AT from obese subjects and those with diabetes

To test the aforementioned idea, we compared miR-181a-5p and miR-23a-3p expression in paired VAT and SAT samples from cohort 1. As shown in Table 1, subjects with BMI ≥30 kg/m² had a worse metabolic profile than those with BMI <30 kg/m² concomitant with elevated mRNA levels of TNFα in VAT (Supplemental Fig. 1), as previously described (2). Analysis showed that miR-181a-5p and miR-23a-3p expression was significantly lower in VAT of patients with BMI ≥30 kg/m² than in those with BMI <30 kg/m² [Fig. 1(b)], whereas in SAT, this significant deregulation was observed only for miR-23a-3p. When subjects were reclassified according to glucose tolerance status, we found that miR-181a-5p expression in VAT was significantly lower in subjects with DM than in those with normal glucose tolerance (NGT), irrespective of adiposity [Fig. 1(c)]. A similar result was seen in SAT only in subjects with BMI <30 kg/m². Analysis of miR-23a-3p expression showed that it was significantly lower in SAT and VAT of patients with DM and with BMI <30 kg/m² than in subjects with NGT [Fig. 1(d)]. We also observed that only miR-23a-3p expression levels were significantly lower in patients with NGT whose BMI was ≥30 kg/m² than in those with BMI <30 kg/m² [Fig. 1(d)]. Finally, expression of both miRNAs was lower in VAT from patients with DM whose BMI was ≥30 kg/m² than in patients with DM and BMI <30 kg/m², pointing to a combined effect of the comorbidity.

Correlation analysis between miRNA expression and clinical and anthropometrical characteristics revealed that miR-181a-5p and miR-23a-3p expression inversely correlated with adiposity (measured by BMI and waist circumference) in VAT, whereas this correlation was found only for miR-23a-3p in SAT. The same correlation was observed with both miRNAs and HOMA-IR (Table 2). Moreover, the expression of both miRNAs was inversely related to TNFα expression in VAT, where the proinflammatory cytokine plays a prominent role in the development of IR (2). miR-155-5p expression levels were below detection in the majority of VAT samples; however, in those samples with detectable expression levels (n = 30), no significant correlation was observed with TNFα (r = 0.302; P = 0.316).

Table 2.

Significant Spearman Correlations for Compared Patient Characteristics and miR-181a-5p and miR-23a-3p Expression Levels From SAT and VAT Depots From Studied Patients

	SAT				VAT
	miR-181a-5p		miR-23a-3p		miR-181a-5p		miR-23a-3p
	R	P	R	P	R	P	R	P
BMI	—	—	−0.467	<0.001	−0.237	0.042	−0.686	<0.001
Waist, cm	—	—	−0.546	<0.001	−0.269	0.028	−0.651	<0.001
Hip, cm	—	—	−0.559	<0.001	—	—	−0.584	<0.001
Waist-to-hip ratio	—	—	—	—	−0.400	0.003	—	—
SBP, mm Hg	−0.277	0.040	−0.378	0.007	—	—	—	—
DBP, mm Hg	−0.278	0.039	−0.615	<0.001	—	—	−0.506	<0.001
Glucose, mM	—	—	–	–	−0.391	0.002	−0.315	0.011
Insulin, pmol/L	—	—	−0.497	<0.001	—	—	−0.545	<0.001
HOMA-IR	—	—	−0.492	<0.001	−0.289	0.033	−0.672	<0.001
Cholesterol, mM	0.256	0.038	—	—	—	—	—	—
HDL cholesterol, mM	0.240	0.048	—	—	0.207	0.071	0.282	0.020
LDL cholesterol, mM	—	—	—	—	—	—	—	—
Triglycerides, mM	−0.254	0.039	−0.314	0.014	—	—	−0.270	0.025
CRP, mg/L	—	—	—	—	—	—	—	—
Gene expression levels
TNFα mRNA	ND	ND	ND	ND	−0.404	0.031	−0.425	0.024
miRNA expression levels
miR-181a-5p SAT	1.000		0.748	<0.001	—	—	—	—
miR-23a-3p SAT	0.748	<0.001	1.000		—	—	0.537	<0.001
miR-181a-5p VAT	—	—	—	—	1.000		0.675	<0.001
miR-23a-3p VAT	—	—	0.537	<0.001	0.675	<0.001	1.000

Abbreviations: CRP, C-reactive protein; DBP, diastolic blood pressure; HOMA-IR, homeostasis model assessment of insulin resistance index; HDL, high-density lipoprotein; ND, not determined; SBP, systolic blood pressure.

miR-181a-5p and miR-23a-3p modulated insulin-stimulated AKT activation and reduced TNFα-induced IR in human adipocytes

Our findings suggest a possible link between miR-23a-3p and miR-181a-5p expression and TNFα-induced IR in obesity. To evaluate the contribution of both miRNAs to insulin signaling, we transiently overexpressed them using miRNA mimics in mature adipocytes, which we then stimulated with insulin or insulin plus TNFα. We then measured phosphorylated (p)AKT and AKT substrate of 160 kDa (pAS160) levels as a surrogate measure of insulin signaling. We observed that individual transfection of miR-181a-5p and miR-23a-3p significantly elevated the insulin-stimulated increase in pAKT levels by up to 28% and 32%, respectively, and in pAS160 levels up to 36% (by miR-181a-5p), with respect to nontarget control (NTC) siRNA [Fig. 2(a)]. Furthermore, overexpression of miR-181a-5p prevented, in part, TNFα-induced IR measured as a suppression of insulin-induced phosphorylation, by significantly increasing pAKT levels up to 32%, and pAS160 levels showed a clear tendency for upregulation [Fig. 2(b)]. By contrast, no preventive effect was detected when miR-23a-3p was overexpressed [Fig. 2(b)].

Figure 2.

Representative western blots of protein expression levels of pAKT, total AKT, pAS160, total AS160, and β-actin. (a) Single- and double-mimic transfection of SGBS adipocytes under 100 nM insulin treatment. (b) Single- and double-mimic transfection of SGBS adipocytes with 50 ng/mL TNFα plus 100 nM insulin. Graphs show relative expression values [mean ± standard error of the mean(SEM)] of five independent experiments. pAKT/AKT and pAS160/AS160 ratios were normalized to NTC for single transfection or NTC2× for double transfection. The 100% NTC mean ± SEM relative unit values are for five experiments: NTC+insulin pAKT/AKT, 0.68 ± 0.31; NTC2× +insulin pAKT/AKT, 0.45 ± 0.13; NTC+insulin pAS160/AS160, 0.71 ± 0.28; NTC2× +insulin pAS160/AS160, 0.53 ± 0.15; NTC+insulin+TNFα pAKT/AKT, 0.26 ± 0.08; NTC2× +insulin+TNFα pAKT/AKT, 0.25 ± 0.12; NTC+insulin+TNFα pAS160/AS160, 0.65 ± 0.31; and NTC2× +insulin+TNFα pAS160/AS160, 0.45 ± 0.12. *P < 0.05; **P < 0.01; ***P < 0.001. INS, insulin; M181, mimic miR-181a-5p; M181+M23, mimic miR-181a-5p + mimic miR-23a-3p; M23, mimic miR-23a-3p; NTC2×, double dose (100 nM) of siRNA NTC.

We next wondered whether combined overexpression of both miRNAs would have additive effect for improving insulin signaling. Cotransfection of both miRNAs in adipocytes significantly upregulated pAKT levels up to 20% after insulin stimulation and we observed a clear tendency for upregulation of pAS160 signaling; however, no improvement over single miRNA transfection was observed [Fig. 2(a)]. Nevertheless, miRNA cotransfection minimized the TNF-induced IR effect with greater efficiency than did individual miRNAs, by upregulating the level of AKT phosphorylation up to 135.8%, and up to 62% for AS160 phosphorylation, although the latter effect did not reach statistical significance [Fig. 2(b)]. These results suggest that both miRNAs may share targets in the insulin pathway regulated by the action of TNFα.

PTEN and S6K are potential targets of miR-181a-5p and miR-23a-3p

Key insulin signaling genes PTEN and S6K were identified as the best shared predicted targets for both miRNAs (Supplemental Fig. 2) (23, 24). To test these predictions, reporter constructs containing the Renilla luciferase gene fused to the PTEN 3′-UTR (luc-PTEN) or to the S6K 3′ UTR (luc-S6K) were transiently transfected into HEK293 cells jointly with miR-181a-5p or miR-23a-3p. As shown in Fig. 3(a), luc-PTEN 3′-UTR luciferase activity was significantly reduced by miR-181a-5p (−36.48%) and miR-23a-3p (−21.94%), and a similar reduction in luciferase activity was observed in luc-S6K 3′-UTR with miR-181a-5p (−45.44%) and miR-23a-3p (−46.52%).

Figure 3.

PTEN and S6K are targets for miR-181a-5p and miR-23a-3p in human adipocytes. (a) HEK293 cells were cotransfected with luc-PTEN 3′-UTR or luc-S6K 3′-UTR constructs and 50 nM miR-181a-5p mimic, miR-23a-3p mimic, or siRNA NTC for 24 h. Relative renSP luciferase activity is expressed as mean ± standard error of the mean (SEM) of three independent experiments normalized to NTC + luc-PTEN (or luc-S6K) 3′-UTR RLU/µg protein (100%): NTC + luc-PTEN 3′-UTR RLU/µg protein, 5,848 ± 848; NTC + luc-S6K 3′-UTR RLU/µg protein, 9,014 ± 1,577. *P < 0.05; **P < 0.01. (b) Representative western blots of protein expression levels of PTEN and S6K in SGBS adipocytes transfected for 24 hours with 50 nM M181 and M23, or in combination (M181+M23). (c) Representative western blots of protein expression levels of PTEN and S6K in SGBS adipocytes transfected for 24 hours with 50 nM iM181 or iM23, or combined (iM181+iM23). Relative values (mean ± SEM) of three independent experiments. PTEN/β-actin and S6K/β-actin ratios were normalized to NTC, NTC2×, iNC, or iNC2×, when appropriate. The 100% NTC mean ± SEM relative unit values for three experiments are given: NTC PTEN/β-actin, 0.37 ± 0.09; NTC2× PTEN/β-actin, 0.51 ± 0.13; NTC S6K/β-actin, 1.35 ± 0.12; and NTC 2× S6K/β-actin, 1.31 ± 0.28. The 100% iNC mean ± SEM relative unit values are as follows: iNC PTEN/β-actin, 0.41 ± 0.04; iNC2× PTEN/β-actin, 0.51 ± 0.06; iNC S6K/β-actin, 1.02 ±0.32; and iNC2× S6K/β-actin, 1.44 ± 0.49. *P < 0.05; **P < 0.01; ***P < 0.001. i23, miR-23a-3p inhibitor; i181, miR-181a-5p inhibitor; iNC, inhibitor negative control; iNC2×, double dose (100 nM) of inhibitor negative control; RLU, relative light unit. See Fig. 2 legend for expansion of other abbreviations.

Consistent with the results of the Renilla reporter assays, a marked reduction of PTEN and S6K protein expression was observed when miR-181a-5p (−32.33%) was overexpressed, but no significant changes where observed for miR-23a-3p [Fig. 3(b)]. However, cotransfection of both miRNAs significantly reduced PTEN expression [−21.33%; Fig. 3(b)]. Reduced S6K expression (−20.3%) was also observed with cotransfection [Fig. 3(b)], moderately affecting its phosphorylation status (Supplemental Fig. 3).

To bolster these findings, we transfected mature adipocytes with inhibitors for miR-181a-5p and miR-23a-3p, either individually or in combination. An increase in PTEN expression was observed when miR-181a-5p was inhibited alone (22.3%) or when both miRNAs were inhibited simultaneously (38.3%); this did not reach statistical significance [Fig. 3(c)]. By contrast, miR-23a-3p inhibition significantly elevated PTEN protein expression up to 78.7% above control levels. S6K protein expression was significantly increased by single-inhibitor miR-181a-5p (13.5%) and miR-23a-3p inhibition (18.4%). An increase in S6K expression was also observed when both miRNAs where inhibited simultaneously, reaching 16.5% when compared with the inhibitor negative control [Fig. 3(c)].

miR-181a-5p and miR-23a-3p were found circulating in serum

To assess the potential value of miR-181a-5p and miR-23a-3p as prognostic markers, we measured their levels in serum in cohort 2 (Supplemental Table 1). miR-181a-5p levels were significantly reduced in the case subjects vs control subjects at baseline, pointing to the prognostic value of miR-181a-5p [Fig. 4(a)]. We also found significant differences for miR-181a-5p and miR-23a-3p between case and control subjects after 4 years, when prediabetes was already diagnosed [Fig. 4(a)]. Interestingly, we observed that levels of miR-181a-5p increased in both groups after 4 years with respect to baseline levels.

Figure 4.

Circulating levels of miR-181a-5p may help identify the prediabetic phenotype. (a) Concentration of miR-181a-5p and miR-23a-3p in serum of control and case subjects at baseline and at 4 years of follow-up. *P < 0.05 between case and control subjects at baseline; ^¶P < 0.05 between case and control subjects at 4-year follow-up; ^##P < 0.01 for 4-year follow-up vs baseline. (b) Results of ROC for miR-181a-5p levels at baseline. (c) VIP of first component in the discriminating quadratic model. (d) ROC plot using miR-181a-5p, HDL cholesterol, CRP, and adiponectin for the analysis. AUC, area under the curve; CRP, C-reactive protein; FABP4, fatty acid binding protein 4; SHGB, sex hormone–binding globulin; VIP, variable importance in projection representation; WHR, waist-to-hip ratio.

To evaluate the usefulness of circulating miR-181a-5p as a potential prognosis biomarker of prediabetes, we performed an ROC analysis. The ROC curve of miR-181a-5p at baseline yielded an area under the curve (AUC) of 0.633 (95% confidence interval, 0.048 to 0.632; P = 0.028) with 82.7% sensitivity and 33.3% specificity [Fig. 4(b)]. We then applied a PLS-DA model to evaluate the potential of miR-181a-5p in the stratification of patients. The model was constructed using biochemical, anthropometrical, and clinical variables. Cross-validation analyses showed that a one-component model had an accuracy of 54.9% (Supplemental Fig. 4). With regard to the importance of individual variables, scores showed that high-density lipoprotein (HDL) cholesterol, C-reactive protein (CRP), and adiponectin had high importance in this model, with miR-181a-5p as the fourth most important variable and having more influence in the model than insulin or glucose at 120 minutes [Fig. 4(c)].

Finally, we performed a multivariate logistic regression analysis including miR-181a-3p, HDL cholesterol, CRP, and adiponectin—variables that had shown significantly different levels at baseline between groups (Supplemental Table 1). The resulting ROC curve yielded a larger AUC [AUC, 0.793; 95% confidence interval, 0.693 to 0.893; P < 0.001; Fig. 4(d)] and significantly higher specificity (72.9%), and the multivariate model correctly classified 72.3% of patients.

Discussion

The association between the expression levels of TNFα in AT and obesity and IR has long been recognized (2, 25). In the current study, we identified miR-181a-5p and miR-23a-3p as miRNAs downregulated by TNFα in human adipocytes in vitro. This was corroborated in human VAT biopsy specimens from obese subjects with BMI ≥30 kg/m² and was more pronounced in VAT of obese patients with diabetes. Although both SAT and VAT have been reported to correlate with IR, available data show that the VAT depot more strongly correlates with IR (26). We also show that expression of miR-181a-5p and miR-23a-3p is inversely related to adiposity irrespective of the fat depot and also to IR measured by HOMA-IR. In addition, both miRNAs are inversely related to TNFα expression in VAT, which argues in favor of their possible role in IR.

TNF-α alters the expression of many proteins that are required for insulin-stimulated glucose uptake in adipocytes, such as the insulin receptor, insulin receptor substrate-1, AKT, and AS160, overall affecting the translocation of the glucose transporter GLUT4 to the plasma membrane (27, 28). Other mechanisms implicated in TNFα-induced IR include the activation of the JNK signaling pathway and the regulation of the adipogenic master regulator, peroxisome proliferator-activated receptor γ (29).

We observed that transient overexpression of miR-181a-5p and miR-23a-3p increased insulin-stimulated pAKT and pAS160 expression in adipocytes to levels >25%, but only miR181a-5p blocked TNFα-induced suppression of pAKT and pAS160. Nevertheless, this rescue effect was augmented when both mimics were overexpressed simultaneously in adipocytes, clearly indicating that miR-181a-5p and miR-23a-3p can cooperate to target insulin pathway regulators.

PTEN regulates phosphatidylinositol 3-kinase–dependent insulin-signaling pathways in adipocytes (30) and S6K (p70S6K), a serine kinase, is involved in negative feedback regulation of insulin action (24). The presence of seed regions in PTEN and S6K for miR-181a-5p and miR-23a-3p, which could potentially regulate their expression, was validated experimentally in HEK293 cells, as shown by a downmodulation of the luciferase-reporter genes containing the wild-type PTEN and S6K 3′-UTRs. Analysis of the mechanisms through which these miRNAs interfered with insulin signaling revealed a clear downregulation of PTEN protein expression after combined overexpression of miR-181a-5p and miR-23a-3p, and also when miR-181a-5p alone was overexpressed. The effect of miR-23a-3p on PTEN expression was only evident when inhibition experiments were performed. A possible explanation for this finding is that because miR-23a-3p can target the deubiquitinase A20 (31), which is a negative regulator of NF-ĸB (32), its suppression may prevent PTEN inhibition by NF-ĸB (33). S6K expression was reduced when miR-181a-5p and miR-23a-3p were overexpressed simultaneously in adipocytes, but not individually. An upregulation of S6K protein after miRNA inhibition confirmed this target in adipocytes.

Despite their rather mild effect on AKT and AS160 phosphorylation targets, our results point to the participation of overexpressed miR-181a-5p and miR-23a-3p in insulin signaling in adipocytes, at least in vitro, because these findings were validated by loss of function and by miRNA/mRNA interaction experiments. Nevertheless, one has to be cautious about translating in vitro results to an in vivo context, because changes in individual miRNA levels in tissue or blood do not always induce observable physiologic effects (34).

Many studies have investigated the function of miR-181a in tissues where it is highly expressed, such as thymus (35) and brain (36). miR-181a has been described as an important negative regulator in hepatic insulin sensitivity (13) and belongs to the miR-181 family cluster, which has a crucial role as a positive regulator of phosphatidylinositol 3-kinase signaling, with PTEN as a target in lymphoid development (37). Accordingly, a robust metabolic phenotype might be expected after suppression of the entire miR-181 cluster in adipocytes, which might unveil new functions in the context of obesity and IR. Although a role for miR-23a-3p in cancer has been reported, by regulating invasion and migration of osteosarcoma cells via its targeting of PTEN (38), to the best of our knowledge, the potential participation of miR-23a-3p in PTEN regulation in adipocyte insulin signaling has not been reported before.

Circulating miRNAs are attractive candidates as biomarkers for disease diagnosis and monitoring (39). The potential involvement of miR-181a-5p and miR-23a-3p in insulin signaling led us to assess their usefulness as serum biomarkers of prediabetes before it is diagnosed. Interestingly, only miR-181a-5p levels were significantly reduced in prediabetic subjects at baseline when compared with control subjects, suggesting a possible prognostic role. Changes to miR-23a-3p levels were also observed with prediabetes, which is in concordance with other findings in a smaller cohort study (40). ROC curve analysis of miR-181a-5p showed that the sensitivity and the specificity were too low for a single diagnostic test; even so, miR-181a-5p could be used as a biomarker to support a positive diagnosis of prediabetes rather than the diagnosis being one of exclusion. Similar data were obtained when PLS-DA analysis was applied using all patient clinical and anthropometrical data and data for miR-181a-5p. In this case, patient stratification had an accuracy of 54.9%. Nevertheless, importance in projection of the first component of the PLS-DA model showed that miR-181a-5p was the fourth most important variable, having more effect than insulin or glucose at 120 minutes. Thus, although the diagnostic accuracy is moderate for all the indexes we examined, the four serum signatures (i.e., HDL cholesterol, CRP, adiponectin, and miR-181a-5p) could represent a potential biomarker panel that could enable early diagnosis of prediabetic patients.

In conclusion, the results presented here outline a potential regulatory role for miR-181a-5p and miR-23a-3p in obesity-related TNFα-associated IR in adipocytes. We also show that miR-181a-5p is deregulated before the onset of prediabetes. Additional studies will be needed to establish the molecular mechanisms through which miR-181a-5p and miR-23a-3p modulate adipocyte insulin signaling.

Abbreviations:

Abbreviations:
 
• AT

  adipose tissue

 
• AUC

  area under the curve

 
• BMI

  body mass index

 
• CRP

  C-reactive protein

 
• DM

  type 2 diabetes mellitus

 
• HDL

  high-density lipoprotein

 
• IR

  insulin resistance

 
• miRNA

  microRNA

 
• NGT

  normal glucose tolerance

 
• NTC

  nontarget control

 
• pAKT

  phosphorylated AKT

 
• PLS-DA

  partial least squares discriminant analysis

 
• ROC

  receiver operating characteristic

 
• SAT

  subcutaneous adipose tissue

 
• SGBS

  Simpson Golabi Behmel Syndrome

 
• siRNA

  small interfering RNA

 
• UTR

  untranslated region

 
• VAT

  visceral adipose tissue

Acknowledgments

We thank the patients enrolled in this study for their participation and the BioBanc IISPV (B.0000853+B.0000854) integrated in the Spanish National Biobanks Platform (PT13/0010/0029 and PT13/0010/0062) for its collaboration.

Financial Support: This study was supported by a project from the Fondo de Investigación Sanitaria (Grants PI14/00465 and PI17/00877 to M.R.C.), cofinanced by the European Regional Development Fund. M.R.C. is supported by the Research Stabilization Program of the Instituto de Salud Carlos III cofinanced by Institut Català de Salut in Catalonia.

Author Contributions: J.L.-B. and M.R.C. designed the experiments. J.L.-B. and A.A.-C. performed the experiments. G.L., M.P.-O., and E.R.-G. analyzed data. G.R.-M., J.V., and R.J. provided patients and supervised clinical information. M.R.C. drafted the paper and takes the full responsibility for the work as a whole.

Disclosure Summary: The authors have nothing to disclose.