Integrative mRNA-microRNA analyses reveal novel interactions related to insulin sensitivity in human adipose tissue

Abstract

Adipose tissue has profound effects on whole-body insulin sensitivity. However, the underlying biological processes are quite complex and likely multifactorial. For instance, the adipose transcriptome is posttranscriptionally modulated by microRNAs, but the relationship between microRNAs and insulin sensitivity in humans remains to be determined. To this end, we utilized an integrative mRNA-microRNA microarray approach to identify putative molecular interactions that regulate the transcriptome in subcutaneous adipose tissue of insulin-sensitive (IS) and insulin-resistant (IR) individuals. Using the NanoString nCounter Human v1 microRNA Expression Assay, we show that 17 microRNAs are differentially expressed in IR vs. IS. Of these, 16 microRNAs (94%) are downregulated in IR vs. IS, including miR-26b, miR-30b, and miR-145. Using Agilent Human Whole Genome arrays, we identified genes that were predicted targets of miR-26b, miR-30b, and miR-145 and were upregulated in IR subjects. This analysis produced ADAM22, MYO5A, LOX, and GM2A as predicted gene targets of these microRNAs. We then validated that miR-145 and miR-30b regulate these mRNAs in differentiated human adipose stem cells. We suggest that use of bioinformatic integration of mRNA and microRNA arrays yields verifiable mRNA-microRNA pairs that are associated with insulin resistance and can be validated in vitro.

insulin resistance increases the risk of developing Type 2 diabetes, which is associated with multiple adverse outcomes, including cardiovascular disease, neuropathy, blindness, and renal failure. Adipose tissue dysfunction inhibits insulin action through multiple mechanisms, such as increasing circulating levels of free fatty acids (13), promoting a proinflammatory environment (17, 34), or altering the secretion of various endocrine factors (35). These actions are thought to promote insulin resistance in other tissues, particularly skeletal muscle and liver. Moreover, adipose tissue itself can become resistant to the effects of insulin, as characterized by increased basal lipolysis rates and decreased insulin-stimulated glucose transport (6). Therefore, impaired glucose utilization may be partially mediated by dysfunctional biological process within adipose tissue.

Although obesity and insulin resistance are correlated (1, 27), each can occur independently of the other, suggesting that they may have both shared and distinct etiological factors. Previous studies have employed high-throughput microarray analyses to assess transcriptional changes in insulin resistant and/or obese humans vs. healthy control subjects (11, 28). These studies have identified various pathways that may be dysregulated in insulin-resistant (IR) humans, including fatty acid and carbohydrate metabolism, inflammation, angiogenesis, and insulin signaling (11, 28). However, microarray technologies only measure steady-state transcript levels, without discriminating between changes in transcriptional rates vs. altered mRNA stability. Therefore, determining which mechanisms contribute to transcript levels may have important implications for interventions that target these pathways.

It is now well recognized that microRNAs are important posttranscriptional regulators of mRNA transcript levels by modulating mRNA stability. This has led many to study the role of microRNAs in adipogenesis (19, 32), obesity, and insulin resistance (8, 29, 41, 44). Insulin resistance is associated with altered expression of several microRNAs whose predicted target genes are involved in insulin signaling (INSR, PI3K, GLUT4) (8, 44), metabolism (ADIPOR1) (29), and inflammation (ETS1) (37). However, these studies were designed to focus on obesity; obesity and insulin resistance are only moderately correlated, and thus, previous studies were not able to identify potential microRNAs that specifically regulate insulin resistance.

Despite the well-defined role of microRNAs as posttranscriptional regulators, no studies have utilized high-throughput mRNA and microRNA arrays on the same adipose tissue samples to identify molecular mechanisms that may contribute to insulin resistance in humans. Using human subcutaneous adipose tissue, we sought to identify putative mRNA-microRNA pairs that were strongly correlated with insulin sensitivity (S_I) and displayed reciprocal expression patterns in insulin-sensitive (IS) vs. IR subjects. We found that microRNA (miR)-26b, miR-30b, and miR-145 had strong positive correlations with S_I. Furthermore, these microRNAs showed a reciprocal expression pattern with various predicted mRNA targets with potential relevance to insulin resistance, including a disintegrin and metalloproteinase domain-22 (ADAM22), myosin VA (MYO5A), lysyl oxidase (LOX), and GM2 ganglioside activator (GM2A). Finally, in differentiated human adipocytes in vitro, we demonstrate that miR-145 decreases the expression of MYO5A, LOX, and GM2A, while ADAM22 expression is decreased by both miR-30b and miR-145.

MATERIALS AND METHODS

Human subjects.

In accordance with the guidelines set by the Declaration of Helsinki (modified in 2008), all protocols were approved by the Institutional Review Boards of the University of Arkansas for Medical Sciences and/or the University of Kentucky. All subjects were made aware of the design and purpose of the study, and all signed consent forms. Subjects were excluded for: diabetes, coronary disease, congestive heart failure, chronic inflammatory diseases, or body mass index (BMI) >43, and the recruitment and initial analysis of these subjects were described previously (11). Participants included both normal-weight and obese subjects. Some subjects demonstrated impaired fasting glucose or impaired glucose tolerance.

Measurement of S_I.

To measure S_I, the frequently sampled intravenous glucose tolerance test (FSIVGTT) was performed and analyzed by the MINMOD method (2, 33), and S_I (min × μU⁻¹ × ml⁻¹ × 10⁻⁴) was determined. Subjects were considered to be IR if S_I was <2.8.

Tissue collection and RNA isolation.

All biopsies were obtained in the fasting state, either before the FSIVGTT or on a different day. Adipose biopsies were obtained under local anesthesia from abdominal subcutaneous fat by incision, rinsed in saline, and frozen in liquid nitrogen. Total RNA was isolated from adipose using the RNeasy Lipid Tissue Mini kit (cat. no. 78404; Qiagen, Valencia, CA). The quantity and quality of the isolated RNA were determined by ultraviolet spectrophotometry and electrophoresis using an Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA). High-quality RNA with an average RNA integrity number of 8.5 was used for genome-wide transcriptome analysis.

Gene microarray and data processing.

Figure 1 shows a schematic of the bioinformatic analyses workflow. For Agilent whole genome microarrays, 62 subjects (n = 31 IS, 31 IR) were recruited. Subjects used for the Agilent microarray displayed a wide range of S_I and were matched for BMI. These subjects have been described in previous reports generated from this microarray data set (Gene Expression Omnibus accession GSE40234) (11). Clinical characteristics of these subjects are given in Table 1A.

Fig. 1.

Bioinformatic analyses schematic. Differentially expressed mRNA and microRNAs were identified in adipose tissue of insulin-sensitive (IS) and insulin-resistant (IR) individuals. Potential biological interaction were identified from reciprocal expression patterns, followed by inverse correlations. Gene regulation by specific microRNAs was confirmed in vitro.

Table 1.

Clinical characteristics of all study participants

	Insulin Sensitive	Insulin Resistant	P Value
	n = 31	n = 31
Female	25 (80.6%)	23 (74.2%)	NS
Age	40.8 ± 1.4 (22–53)	39.5 ± 1.7 (22–54)	NS
BMI	30.5 ± 1.0 (19.6–42.5)	30.3 ± 0.75 (21.5–40)	NS
SI	6.5 ± 0.44 (2.9–11.3)	1.5 ± 0.09 (0.9–2.7)	<0.0001
WHR1	0.86 ± 0.01 (0.71–1.03)	0.89 ± 0.01 (0.76–1.02)	NS
B: NanoString microRNA Array
	n = 14	n = 12
Female	12 (86%)	11 (92%)	NS
Age	37.3 ± 2.3 (22–51)	41.4 ± 2.4 (25–57)	NS
BMI	27.3 ± 1.4 (19.2–36.9)	33.2 ± 0.54 (30.0–36.6)	<0.01
SI	7.4 ± 0.78 (3.7–13.6)	1.5 ± 0.14 (0.91–2.4)	<0.0001
WHR2	0.81 ± 0.02 (0.70–0.95)	0.91 ± 0.02 (0.80–1.02)	<0.01
	n = 9	n = 7
Female	7 (78%)	6 (86%)	NS
Age	37.6 ± 3.0 (22–46)	41.0 ± 3.1 (25–49)	NS
BMI	30.0 ± 1.5 (23.0–36.9)	32.5 ± 0.7 (30–35)	NS
SI	6.7 ± 0.87 (3.7–11.3)	1.2 ± 0.07 (0.9–1.5)	0.0002
WHR2	0.84 ± 0.03 (0.71–0.95)	0.92 ± 0.02 (0.83–1.02)	<0.05

Means ± SE (Range). ¹Waist/hip ratio (WHR) data were not available for 2 insulin-sensitive (IS) subjects. ²WHR data were not available for 1 IS subject. IR, insulin resistant; BMI, body mass index; SI, insulin sensitivity; NS, not significant.

Genome-wide transcriptome analysis and initial processing was performed by GenUs Biosystems (Northbrook, IL) using Human Whole Genome 4 × 44 k arrays (Agilent Technologies), according to the vendor-recommended protocol. The previous report generated from this data set (11) utilized Statistical Analysis for Microarray software for data processing, while the analysis presented in this report employed Partek Genomics Suite (St. Louis, MO) for quantile normalization, background correction, and data processing.

microRNA expression profiling and data processing.

For microRNA arrays, 26 subjects (n = 14 IS, 12 IR) were chosen based on RNA quality and availability. These subjects displayed a wide range of S_I values, but they were not matched for BMI. Clinical characteristics of these subjects are given in Table 1B. A subset of these subjects overlapped with subjects included in the mRNA microarray analyses (see Table 1C, described below).

MicroRNA gene expression was measured using the nCounter Human v1 microRNA Expression Assay (NanoString Technologies, Seattle, WA) (14, 31, 40). RNA (100 ng) from each biopsy was hybridized with the nCounter probe set. According to manufacturer recommendations, positive and negative control corrections were applied to the raw data, and then gene expression was normalized to the mean of 120 microRNAs with the highest expression in our samples.

mRNA-microRNA target identification.

Differentially expressed mRNA and microRNA data sets were imported into Ingenuity Pathway Analysis for target identification. mRNA-microRNA interactions were identified using the microRNA Target Filter function which utilized the TarBase database for experimentally observed interactions and the Human TargetScan database for predicted interactions. Only interactions that demonstrated reciprocal expression patterns, based on fold-change differences between groups, were investigated further. Finally, correlations between mRNA expression and microRNA expression were run on samples for which there were data for both types of transcript (n = 9 IS, 7 IR). Clinical characteristics of these subjects are given in Table 1C.

In vitro microRNA target validation.

Adult-derived human adipose stem cells were obtained from subcutaneous fat following liposuction in healthy nonobese women from Lexington, KY. Cells lines from three different women were grown in triplicate and induced to differentiate with 33 μM Biotin, 17 μM panthothenate, 100 nM human insulin, 1 μM dexamethasone, 0.25 mM IBMX, and 1 μM rosiglitazone (5). When cells were 90% differentiated, they were transfected with 50 mM of the miRIDIAN microRNA mimics hsa-miR-145-5p, hsa-miR-30b-5p or hsa-miR-26b-5p (cat. no. C-300613-05-0005, C-300590-03-0005, or C-300501-07-0005; GE Dharmacon, Lafayette, CO) or a combination of both miR-145 and miR-30b microRNA mimics using Lipofectamine RNAiMAX (Life Technologies) for 36 h and then harvested for RNA.

Gene expression.

RNA was extracted using the RNeasy Lipid Tissue Mini Kit (cat. no. 74804, Qiagen). RNA quality and integrity were assessed using the Agilent 2100 Bioanalyzer. For messenger RNA gene expression, reverse transcription was performed using the miScript II RT Kit (cat. no. 218160, Qiagen). Quantitative real-time RT-PCR was performed using KiCqStart qPCR ReadyMix (cat. no. KCQS07; Sigma-Aldrich, St. Louis MO), and gene expression was normalized to the geometric mean of four housekeeping genes: β2 microglobulin, cyclophilin A, phosphoglycerate kinase, and 18S RNA. Primer pairs are available upon request.

Statistical analyses.

For mRNA microarray analyses, gene expression data (AU) were log base 2 transformed prior to analysis, and differentially expressed genes were identified by t-tests with an false discovery rate-corrected P value < 0.05 and a minimal fold-difference of ±1.4. For the microRNA NanoString analyses, differentially expressed genes were identified using a P value < 0.01 with minimal fold-difference of ±1.4. Because clinical S_I data were nonnormally distributed, values were log transformed prior to analyses. Linear regressions were performed using the Pearson product-moment correlation coefficient. Linear regressions shown in Fig. 3 had adjusted power >0.71, while those shown in Fig. 5 had adjusted power >0.92. For transfection experiments, paired t-tests were run between expression values from control and microRNA-transfected cell lines. For all paired t-tests, statistical power was >0.80. For all analyses significance was set at P < 0.05.

Fig. 3.

microRNA (miR)-145 (A), -26b (B), and -30b (C) demonstrate strong positive relationships with insulin sensitivity. MicroRNA expression was determined by NanoString technology and correlated against individuals Log S_I. P < 0.01.

Fig. 5.

MYO5A, LOX, GM2A, and ADAM22 expression levels are negatively associated with insulin sensitivity. Gene expression was determined by Agilent microarray and correlated against individuals Log S_I. A: MYO5A; B: GM2A; C: LOX; D: ADAM22. P < 0.0001

RESULTS

Differential expression of adipose mRNA and microRNAs in IS and IR subjects.

Table 1A provides clinical characteristics of the subjects used for Agilent mRNA gene expression analysis. For Agilent gene expression arrays, 31 IR and 31 IS subjects were matched for BMI. Overall, 644 mRNA probe sets demonstrated significant differences in expression between IS and IR individuals (Supplementary Table S1).¹ Of these probe sets, 539 mapped to known mRNA transcripts in Ingenuity Pathway Analysis software, representing 498 unique transcripts.

Table 1B provides clinical characteristics of the subjects used for NanoString microRNA gene expression analysis. This analysis was performed on adipose tissue from 14 IS and 12 IR subjects who were not matched for BMI. For microRNA analysis, counts were normalized using a factor based on the geometric mean of the top 120 microRNAs with the highest expression. The normalization factor did not differ between IR and IS (IR = 1.27 ± 0.24, IS = 1.57 ± 0.21, P = not significant). Seventeen microRNAs were differentially expressed. Of these, 16 microRNAs were lower in IR than IS subjects (Table 2). Hierarchical clustering revealed a strong contrast between the profiles of IS and IR individuals (Fig. 2). Of the differentially expressed microRNAs, miR-1, -574-3p, -1246, 30c, -145, and -23b all showed greater than threefold higher expression in IS individuals, while only miR-629 was higher in IR vs. IS.

Table 2.

microRNAs that are differentially expressed in subcutaneous adipose tissue of IR vs. IS individuals

microRNA	Mean Expression (IS)	Mean Expression (IR)	Fold Difference (IR vs. IS)	ANOVA P Value	BMI Correlation Coefficient	BMI P Value
hsa-miR-1	146.5	21.5	−6.8	0.0019	−0.22	0.290
hsa-miR-574-3p	58.7	12.7	−4.61	0.0004	−0.25	0.226
hsa-miR-1246	82.5	21.2	−3.89	0.009	−0.27	0.180
hsa-miR-30c	297.1	85.1	−3.49	0.0028	−0.30	0.142
hsa-miR-145	5034.7	1559	−3.23	0.0019	−0.33	0.096
hsa-miR-23b	811.3	262.5	−3.09	0.0041	−0.18	0.380
hsa-miR-99b	119.1	40.1	−2.97	0.0051	−0.18	0.384
hsa-let-7f	2671.7	1014.1	−2.63	0.0049	0.08	0.700
hsa-miR-30b	2517.4	1077.8	−2.34	0.0023	−0.38	0.053
hsa-miR-195	441.1	193.5	−2.28	0.002	−0.39	0.051
hsa-miR-374b	145.3	73.1	−1.99	0.0024	−0.18	0.370
hsa-miR-551b	65.3	33.6	−1.94	0.0011	−0.37	0.061
hsa-miR-103	1360.7	738.6	−1.84	0.0052	−0.27	0.189
hsa-let-7a	13249.1	7299.9	−1.81	0.0097	−0.28	0.164
hsa-miR-26b	795.1	496.4	−1.6	0.0061	−0.25	0.210
hsa-let-7d	2868.7	2132.8	−1.35	0.0098	−0.18	0.378
hsa-miR-629	20.7	41.4	2	0.0018	0.22	0.286

Fig. 2.

Hierarchical clustering of differentially expressed microRNAs based on insulin sensitivity (S_I). Hierarchical clustering reveals discrete clustering of microRNA profiles based on an individual's insulin sensitivity. Sixteen out of 17 differentially expressed microRNAs are lower in IR vs. IS subjects.

Identification of microRNA target genes.

Using the MicroRNA Target Filter function within Ingenuity Pathway Analysis, we identified which differentially expressed microRNAs potentially targeted any of the 498 differentially expressed mRNA transcripts. Since one of the primary mechanisms of action for microRNAs is regulation of mRNA stability, we focused on microRNA-mRNA interactions that demonstrated a reciprocal relationship (e.g., microRNA upregulated and mRNA target downregulated). MicroRNAs that possess the same sequence (e.g., miR-30b and miR-30c) and, therefore, target the same mRNA, were combined into microRNA families. Using these criteria, we identified 102 predicted or validated interactions between 12 microRNA families and 53 mRNAs (Supplementary Table S2).

miR-145, miR-26b, and miR-30b demonstrate a positive relationship with S_I.

To further refine our list of microRNAs, we correlated S_I against the 12 microRNA family members that had predicted targets (Table 3). From this list, miR-145, -26b, -30b, -103, -1, and -551b all demonstrated significant positive relationships with S_I (P < 0.01, R > 0.5). Because miR-1 and miR-551b had relatively low expression values (Table 2), we narrowed our focus to miR-145, -26b, -30b, and -103. In addition, since miR-103 has previously been described as a potential regulator of S_I in both murine and human models (4, 38, 43), we focused the remainder of our analyses on miR-26b, -145, and -30b (Fig. 3, A–C). Neither miR-26b nor miR-145 was significantly correlated with BMI (Table 2). However, miR-30b displayed a trend toward an inverse correlation with BMI (P = 0.052, R = −0.38).

Table 3.

Correlation between microRNA expression and S_I value

microRNA	Correlation Coefficient	P Value
hsa-miR-145	0.58	0.0021
hsa-miR-26b	0.56	0.0028
hsa-miR-30b	0.54	0.0047
hsa-miR-103	0.52	0.0065
hsa-miR-1	0.52	0.0069
hsa-miR-551b	0.51	0.0081
hsa-miR-1246	0.48	0.0137
hsa-miR-30c	0.47	0.0145
hsa-miR-374b	0.47	0.0156
hsa-miR-195	0.46	0.0169
hsa-let-7f	0.40	0.0411
hsa-miR-23b	0.40	0.0451

Identification of reciprocal relationships between microRNAs and predicted mRNA targets.

For 16 subjects (n = 9 IS, 7 IR), we had both mRNA and microRNA expression data. Clinical characteristics of these subjects are provided in Table 1C. To identify putative biological relationships within our microRNA-mRNA interaction list, we correlated the expression level of the microRNA and mRNA target genes against one another to identify those with significant inverse relationships (Table 4). Of these, miR-30b showed the strongest reciprocal relationship with its predicted targets. Gene targets that demonstrated significant inverse relationships were TNIP1, PLXNA4A4, LOX, MYO5A, TSPAN33, CDCP1, ADAM22, LIFR, and TAOK1 (Table 4). There was redundancy in the mRNA targets for miR-30b and miR-145 (PLXNA4A4, MYO5A) and for miR-30b and miR-26b (TAOK1), perhaps suggesting synergist relationships between these microRNAs (Supplementary Table S2). Furthermore, miR-145 showed a significant reciprocal relationship with GM2A, and miR-26b demonstrated reciprocal relationships with PARP14, RAB31, and LAMA1 (Table 4).

Table 4.

Relationship between microRNAs, mRNA targets, and S_I

microRNA	mRNA Target	Correlation Coefficient (microRNA vs. mRNA Target)	P Value	Correlation Coefficient (mRNA vs. SI)	Mean mRNA Expression (log2 AU)
miR-30b	TNIP1	−0.87	0.000	−0.59	13.02
	PLXNA4A	−0.75	0.001	−0.44	9.68
	LOX	−0.74	0.001	−0.48	9.90
	MYO5A	−0.70	0.002	−0.65	9.03
	TSPAN33	−0.62	0.011	−0.60	9.07
	CDCP1	−0.58	0.018	−0.56	4.68
	ADAM22	−0.57	0.023	−0.41	9.46
	LIFR	−0.54	0.030	−0.40	8.97
	TAOK1	−0.50	0.048	−0.51	8.83
miR-145	PLXNA4A	−0.68	0.004	−0.44	9.68
	GM2A	−0.53	0.034	−0.52	8.78
	MYO5A	−0.50	0.047	−0.65	9.03
miR-26b	PARP14	−0.61	0.011	−0.36	8.20
	RAB31	−0.58	0.019	−0.64	11.57
	LAMA1	−0.53	0.037	−0.36	5.68
	TAOK1	−0.50	0.050	−0.51	8.83

In vitro validation of microRNA targets.

For in vitro validation, we selected mRNAs that were inversely correlated to S_I (R < −0.40) and were relatively abundant in adipose tissue (expression > 8.5 log2 AU). Thus, CDCP1, LAMA1, and PARP14 were excluded from in vitro analyses. To validate microRNA regulation of mRNA targets, primary adult-derived human adipose stem cells were transfected with miR-30b, miR-145, miR-26b, or a combination of miR-30b and miR-145. For all conditions, we performed real-time RT-PCR for all genes of interest. Relative to scrambled control, transfection with miR-145 resulted in decreased mRNA abundance of MYO5A and GM2A (Fig. 4, A and B). Despite demonstrating a weak reciprocal relationship in vivo, our in vitro results confirm LOX as a target of miR-145 (Fig. 4C and Supplementary Table S2). Although ADAM22 was not a predicted target to miR-145 in the TargetScan database, miR-145 transfection decreased ADAM22 expression in vitro (Fig. 4D). However, miR-145 had no effect on PLXN4A4 mRNA levels (data not shown). Compared with scrambled control, miR-30b had no effect on the expression level of its predicted targets TNIP1, PLXN4A4, LOX, MYO5A, TSPAN33, LIFR, and TAOK1 (data not shown). However, miR-30b transfection induced a significant reduction in ADAM22 gene expression (Fig. 4E). Lastly, miR-26b had no effect on RAB31 or TAOK1 mRNA abundance (data not shown).There was no synergistic effect of cotransfecting miR-145 and miR-30b on the expression of shared gene targets. These results suggest that miR-145 regulates the expression of multiple predicted gene targets, which all demonstrated strong relationships with S_I (Fig. 5 and Table 4).

Fig. 4.

miR-145 and miR-30b regulate the expression of their predicted mRNA targets. ADHASC cells from 3 different individuals were transfected with miR-145, miR-30b, or scrambled control. miR-145: MYO5A (A), GM2A (B), LOX (C), ADAM22 (D); miR-30b: ADAM22 (E). Data are presented as percent reduction in gene expression within each different cell line. *P < 0.05.

DISCUSSION

Accumulating evidence suggests that dysregulated expression of microRNAs has significant effects on the cellular transcriptome, ultimately leading to defective biological functions. In the current study, we utilized an integrative mRNA-microRNA array approach to identify potential interactions in subcutaneous adipose tissue that may contribute to insulin resistance. Using this approach, we identified several microRNAs that were differentially expressed in IR vs. IS individuals. Intriguingly, of 17 microRNAs that were differentially expressed between IR and IS, 16 were lower in IR individuals. These findings are in keeping with recent evidence indicating that adipocyte-specific reductions in microRNA levels result in decreased adipose mass and severe insulin resistance in mice (30). This loss of adipose tissue following reduced microRNA levels may be due to impaired adipogenesis, as there is a global upregulation in microRNA levels during adipocyte differentiation (12). The current evidence suggests that impaired adipogenesis contributes to insulin resistance (9, 21). These findings provide evidence that adequate expression of microRNAs in adipose tissue is essential for adipose health, therefore contributing to maintenance of insulin sensitivity.

Although we identified 17 microRNAs that differed in IR vs. IS, we focused our attention on miR-26b, miR-30b, and miR-145 because they were novel and displayed the strongest correlations with S_I. We next identified putative gene targets of these microRNAs on the basis of reciprocal expression patterns. Since a single microRNA can target multiple transcripts, our identified microRNAs shared several predicted gene targets. These results suggest that these genes may be regulated by multiple microRNAs in a complex, combinatorial manner. These findings highlight the importance of utilizing integrative high-throughput approaches that allow for the identification of mRNA-microRNA regulatory networks.

miR-26b and -30b.

The higher miR-26b and -30b levels that we observed in IS subjects may stimulate adipogenesis and terminal adipocyte differentiation; in primary human adipocyte culture, miR-26b is upregulated during differentiation, miR-26b knock-down inhibits differentiation, and miR-26b overexpression enhances brite adipogenesis (23). Similarly, miR-30 family members contribute to thermogenesis (20), are upregulated during adipocyte differentiation (12), and their expression can directly impact adipogenesis (22, 46).

miR-145.

Recent evidence implicates miR-145 in adipocyte lipolysis (25, 26). In mice, miR-145 is reported to suppress lipolysis in adipose tissue by inhibiting FOXO1 and CGI58 (25). Conversely, other evidence indicates that miR-145 increases lipolysis in human adipocytes, possibly through increased TNF-α (26). In the current study, we show that miR-145 expression in subcutaneous adipose tissue is higher in IS subjects and is positively associated with S_I. In addition, we demonstrate that treatment with miR-145 mimic destabilizes predicted gene targets in cultured adipocytes, resulting in decreased expression levels of ADAM22, MYO5A, LOX, and GM2A. Relative to miR-30b, it appears that miR-145 may be more biologically active in vitro despite demonstrating weaker negative relationships with its predicted targets in whole human adipose tissue.

miR-145 and MYO5A.

Previously, miR-145 has been shown to directly target the 3′-untranslated region of MYO5A (10), a motor protein involved in cytoplasmic vesicle transport and insulin-stimulated GLUT4 translocation in adipocytes (7, 45). It would be reasonable to hypothesize decreased levels of MYO5A in IR vs. IS. But we observed increased MYO5A expression in IR vs. IS subjects and decreased MYO5A expression in miR-145 transfected adipocytes (Fig. 4). Thus, the role of MYO5A in adipose tissue insulin resistance is likely complex.

miR-145 and LOX.

We also demonstrated LOX as a target of miR-145. LOX acts by cross-linking collagen I and III to form fibrillar collagen fibers (16), and inhibition of LOX expression leads to decreased fibrosis in adipose tissue and an improved metabolic phenotype (16). Congruent with this mechanism, LOX expression was inversely correlated with S_I in our human cohort. miR-145 transfection induced a modest decrease in LOX expression (∼18%), which may help to explain why we did not see a significant inverse correlation between miR-145 and LOX in whole adipose tissue. Nonetheless, inhibition of LOX expression by miR-145 in adipose tissue may enhance extracellular matrix function and insulin signaling.

miR-145 and GM2A.

Finally, we identified a novel interaction between miR-145 and GM2A, a lysosomal glycoprotein that is involved with GM2 ganglioside processing (42). In adipose tissue, both mRNA and protein levels of GM2A are elevated in obese individuals, and GM2A may directly impair adipose insulin receptor signaling by disrupting caveolar architecture (18, 36, 39). Therefore, decreased levels of GM2A via miR-145 would act to promote insulin sensitivity, consistent with our findings that miR-145 is elevated in IS individuals.

ADAM22.

In vitro, ADAM22 gene expression was reduced by both miR-30b and miR-145 transfection. ADAM22 is involved in cell adhesion and migration (3), and it has been reported to be a miR-145 target (24). However, its role in adipose tissue and insulin sensitivity is currently unknown.

Although numerous reports indicate that insulin resistance is more highly correlated with visceral than subcutaneous adipose tissue mass, some have shown that subcutaneous fat mass better predicts S_I (15), and the vast majority of body fat is subcutaneous_. Visceral fat can only be obtained through surgery, and hence our study utilized subcutaneous fat from well-characterized subjects. Since adipose depots differ in molecular and physiological processes, it is likely that the microRNA-mRNA pairs identified here relate specifically to abdominal subcutaneous fat.

In our human adipose tissue samples, most of the differentially expressed microRNAs were positively associated with S_I. The underlying mechanism resulting in this increase is unclear, as we did not detect any significant differences in total microRNA expression or transcript levels for genes involved in microRNA biogenesis. Because we identified genes that were differentially expressed in IR vs. IS subjects matched for BMI, we hoped to identify microRNA-mRNA interactions that may modulate S_I, regardless of obesity. Neither miR-145 nor its gene targets were correlated with BMI, strengthening the potential relevance of miR-145 to the pathology of insulin resistance. With advances in high-throughput screening, bioinformatic identification of microRNA-mRNA targets like these may prove valuable for identifying pathways that are dysregulated in insulin resistance.

GRANTS

This work was supported by National Institutes of Health Grants DK-071349 (P. A. Kern and C. A. Peterson) and UL1 TR000117, a VA merit grant (N. Rasouli), and a grant from the Sturgis Foundation to the University of Arkansas for Medical Sciences.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author(s).

AUTHOR CONTRIBUTIONS

Author contributions: T.J.K., R.G.W., B.Z., R.U., and P.A.K. performed experiments; T.J.K., R.G.W., B.S.F., B.Z., and R.U. analyzed data; T.J.K., B.S.F., R.U., N.R., C.A.P., and P.A.K. interpreted results of experiments; T.J.K. and R.G.W. prepared figures; T.J.K. and R.G.W. drafted manuscript; T.J.K., R.G.W., B.S.F., B.Z., R.U., N.R., C.A.P., and P.A.K. approved final version of manuscript; B.S.F., N.R., C.A.P., and P.A.K. conception and design of research; B.S.F., N.R., C.A.P., and P.A.K. edited and revised manuscript.