Adipose Tissue MicroRNAs as Regulators of CCL2 Production in Human Obesity

Cohort 1 comprised 30 obese (BMI >30 kg/m²) otherwise healthy and 26 nonobese (BMI <30 kg/m²) healthy women. They were investigated in the morning after an overnight fast in approximately the midst of their menstrual cycle. All were premenopausal and free of continuous medication. Height, weight, and waist circumference were determined. A venous blood sample was obtained for measurements of glucose and insulin. The values were used to construct an index of overall insulin sensitivity (homeostasis model assessment of insulin resistance [HOMA_IR]) as described (12). Thereafter, an abdominal subcutaneous WAT biopsy (∼1.5 g) was obtained by needle aspiration as described (13). One part (300 mg) of the tissue was used for measurement of release of CCL2 per number of fat cells, as described (14). Similar results were obtained when protein release was related to the WAT weight (values not shown). Methodological studies show that protein release was linear with time for at least 3 h, suggesting no important cell damage. Another part of the tissue (700 mg) was subjected to collagenase treatment, and mean adipocyte volume and weight were determined as described (15). Next, 200 µL packed fat cells as well as 400 mg intact WAT were frozen at −70°C for future mRNA and miRNA measurements. The remaining isolated fat cells were incubated for 2 h at 37°C in an albumin concentration buffer without or with increasing concentration of insulin, and lipogenesis (incorporation of [³H]glucose into lipids) was determined as described (15). From the insulin concentration-response curves, lipogenesis at maximum effective insulin concentration was determined and expressed per number of fat cells. Clinical characteristics of this cohort are given in Table 1. The obese subjects displayed in vivo (HOMA_IR) and in vitro (insulin-stimulated lipogenesis) insulin resistance as well as increased adipose tissue CCL2 secretion (in vitro). For experimental studies, subcutaneous WAT was obtained from healthy subjects undergoing cosmetic liposuction. In this group, there was no selection for age, sex, or BMI. In a subgroup of these tissue donors (n = 12), global gene expression profiling was determined during in vitro differentiation of primary human adipocytes. This study was approved by the ethical committee at the Karolinska University Hospital. All subjects were informed in detail about the studies and written informed consent was obtained.

From total RNA, we prepared and hybridized biotinylated complementary RNA to GeneChip Human Gene 1.0 ST Array (Affymetrix, Inc., Santa Clara, CA) and then washed, stained, and scanned the slides using standardized protocols (Affymetrix, Inc.). Preprocessing was performed using the Affymetrix Expression Console version 1.1 and the following settings: summarization, PLIER; background correction, PM-GCBG; and normalization, Global Median. Gene and miRNA expression have been deposited in the National Center for Biotechnology Information Gene Expression Omnibus (http://ncbi.nim.nih.gov/geo) and are accessible using Gene Expression Omnibus series accession number GSE25402.

Total adipose RNA (1 μg) from each subject in cohort 1 was labeled using the FlashTag biotin HSR labeling kit (Genisphere, Hatfield, PA) according to the supplier’s protocol. The labeled samples were placed in a hybridization cocktail mix containing 4% formamide and hybridized overnight to Affymetrix miRNA arrays following the indicated Genisphere protocol. The arrays were washed, stained, and scanned in an Affymetrix GCS 3000 scanner. Signal intensities and present calls were generated by using the miRNA QC tool by Affymetrix.

Isolation, culture, and in vitro differentiation of human adipocyte progenitor cells from subcutaneous WAT were performed exactly as described (16). 3T3-L1 and THP1 cells were handled as recommended in the protocols from American Type Culture Collection (Manassas, VA).

In vitro differentiated adipocytes were treated with 40 nmol/L miRIDIAN miRNA Mimics or 100 nmol/L miRIDIAN miRNA Hairpin Inhibitors (Thermo Fisher Scientific, Lafayette, CO) and HiPerFect Transfection Reagent (QIAGEN, Hilden, Germany) according to the manufacturers’ protocols. Optimal transfection conditions were determined in separate titration experiments, and transfection efficiency was assessed with quantitative real-time PCR (qRT-PCR) using miRNA probes and fluorescent transfection controls (Thermo Fisher Scientific). To rule out unspecific effects, control cells were transfected with miRIDIAN miRNA Mimic or Hairpin Inhibitor Negative Controls (Thermo Fisher Scientific). THP1 macrophages were transfected as described above for adipocytes.

Total RNA was extracted from in vitro differentiated adipocytes, 3T3-L1 cells, isolated adipocytes, intact WAT, and macrophages. RNA concentration as well as purity was measured spectrophotometrically using a Nanodrop ND-1000 Spectrophotometer (Thermo Fisher Scientific), and high quality (total and small) RNA was confirmed using the Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA).

Total RNA was reverse transcribed using Omniscript First-strand cDNA synthesis kit (QIAGEN) and random hexamer primers (Invitrogen, Carlsbad, CA). For confirmation and further studies of microarray results, expression levels of selected target genes were measured in 384-well TaqMan Low Density Array plates (Applied Biosystems, Foster City, CA). Equal amounts of cDNA were dispersed on the plates together with a TaqMan gene expression master mix (Applied Biosystems). Amplification of the PCR products was followed and registered in a 7900HT Fast Real-Time PCR System (Applied Biosystems). The resulting data were analyzed with the RQ Manager and Data Assist software (Applied Biosystems). Other human transcripts were detected using TaqMan Universal PCR Master Mix (Applied Biosystems) and TaqMan primers and probes (Applied Biosystems). qRT-PCR was performed on an iCycler IQ (Bio-Rad, Hercules, CA) using the amplification protocol provided by Applied Biosystems. All samples were run in duplicates, and arbitrary units were calculated using a comparative cycle threshold (Ct) method (i.e., 2^{ΔCt-target gene}/2^{ΔCt-reference gene}). 18S rRNA was used to normalize the expression of all analyzed human genes.

miRNA cDNA was synthesized using TaqMan miRNA Reverse Transcription Kit and human Megaplex primer pool A without preamplification (Applied Biosystems). For miRNAs not included in the primer pool, specific probes designed by Applied Biosystems were used. qRT-PCR was performed as described above on an iCycler IQ (Bio-Rad) using the amplification protocol provided by Applied Biosystems. Expression of miRNA was normalized to an internal reference gene RNU48, and arbitrary units were calculated as described above.

Cells were lysed in radioimmunoprecipitation assay buffer as described previously (17), and 15–20 µg total protein was separated by SDS-PAGE. Western blot was performed according to standard procedures, and secondary antibodies mouse/rabbit IgG–horseradish peroxidase were from Sigma-Aldrich. β-Actin (Sigma-Aldrich) was used as a reference for all Western blots. Primary antibodies detecting MYC-associated factor X (MAX) were unspecific and, therefore, could not be used.

Conditioned media from in vitro differentiated adipocytes, macrophages, and subcutaneous WAT pieces (300 mg in 3 mL medium) was saved at −70°C for determination of CCL2 and IL-6 levels by ELISA according to the manufacturer’s instructions (R&D Systems, Minneapolis, MN).

The luciferase reporter construct containing 3′-untranslated region (3′UTR) of v-ets erythroblastosis virus E26 oncogene homolog 1 (ETS1) (pGL3 ETS1 3′UTR/155up) was provided by Dr. Giovanna Marziali (Istituto Superiore di Sanitá, Rome, Italy) and has been described previously (18). The empty luciferase reporter vector and vectors containing 3′UTR of CCL2, ARNT, MAX, and RELB were obtained from GeneCopoeia (Rockville, MD). 3T3-L1 cells were plated in 24-well plates and transfected 48 h postplating with 150 ng luciferase-3′UTR reporter constructs or with the empty expression vector 2 ng Renilla luciferase reporter pRL-SV40 (Promega, Madison, WI) (for ETS1 only) together with miRIDIAN miRNAs (miRNA mimic or negative controls) at final concentration of 40 nmol/L using Lipofectamine and Plus Reagent (both from Invitrogen) according to the manufacturer’s protocol. After 4 h of incubation, the medium was refreshed with Dulbecco’s modified Eagle’s medium/F12 plus 10% FCS. The cells were incubated for 24 h and subsequently lysed, and firefly as well as Renilla luciferase activities was determined using the Dual-Glo Luciferase Assay System (Promega). All experiments were performed in triplicate, and the values of firefly luciferase activity were normalized by Renilla luciferase activities and empty vector, which served as internal controls.

Differences in WAT expression of mRNAs/miRNAs between obese/nonobese subjects and during adipocyte differentiation were analyzed using significance analysis of microarrays, false discovery rate 5%, as described previously (19).

Motif activity response analysis (MARA) is extensively described in the main text and supplement of our previous study (20). In brief, the method relies on the assumption that TFs regulate the expression of genes through binding to DNA sequence elements in promoter regions and that gene expression is a linear function of the number of TF binding sites in the promoter of the gene. Activities of known motifs recognized by human TFs were fitted to gene expression levels as determined by microarray. For interactions between TFs and their target genes, genes with a minimum number of the TF binding sites in their promoter region and with expression correlating with the corresponding motif activity over a threshold were identified as nodes in a transcriptional regulatory network together with the TFs. Finally, the subnetwork predicted to regulate CCL2 was extracted. Exact details on MARA and construction of networks used in this study are further described in results.

Unless otherwise stated, details on the reagents used in this study, including relevant catalog numbers, are listed in Supplementary Table 1.

Data shown in the figures and tables are mean ± SEM. When appropriate, the data were log transformed to become normally distributed. Results were analyzed with unpaired t test, linear regression (simple and multiple), sign test, or ANOVA (repeated-measures). For analysis of genome-wide data, see the respective subheading.

Using global miRNA expression profiling of WAT, we identified 20 miRNAs that were expressed in all subjects (cohort 1) and concomitantly altered in obesity (2 up- and 18 downregulated) (Table 2). The results were validated by qRT-PCR in intact WAT showing that 14 miRNAs were regulated in accordance with the array data (70% concordance). Adipose tissue contains many cell types. Since we were primarily interested in miRNAs expressed in human fat cells, we determined their expression in isolated adipocytes from cohort 1 and whether they were affected by obesity. All 20 miRNAs were expressed in mature fat cells. However, when overlapping the data from isolated fat cells and intact WAT, only 11 (miR-26a, -30c, -92a, -126, -143, -145, -193a, -193b, -652, let-7a, and let-7d) were found to be significantly altered (all downregulated) in obesity. These miRNAs were considered as candidate miRNAs.

To determine whether our candidate miRNAs could influence CCL2 secretion, we performed a transfection screen by overexpressing individual miRNAs in human in vitro differentiated adipocytes. Nine miRNAs (miR-26a, -92a, -126, -143, -193a, -193b, -652, let-7a, and let-7d) significantly reduced CCL2 secretion, one had no effect (miR-30c), and one led to increased release (miR-145) (Fig. 1A). We chose to focus our functional studies on the nine miRNAs that upon overexpression, downregulated CCL2 production since their expression in obesity (Table 2) and effects on adipocyte secretion (Fig. 1A) were concordant.

To assess whether the miRNAs could directly interact with the 3′UTR of CCL2, we performed an in silico analysis using information available in public databases (see research design and methods). According to this search, only miR-126 was predicted to bind to the 3′UTR of CCL2, which was confirmed in a luciferase reporter assay in 3T3-L1 cells (Fig. 1B). The specificity of this interaction was determined by transfecting a random subset of the other candidate miRNAs (miR-26a, -92a, and let-7d) in parallel experiments showing no effect on luciferase activity (Fig. 1B). That miR-126 regulates CCL2 production at the transcriptional level was supported by the fact that miR-126 overexpression downregulated CCL2 mRNA in human in vitro differentiated adipocytes (Fig. 1C).

Since only 1 out of 10 candidate miRNAs was predicted to directly interact with CCL2, we set out to construct a transcriptional regulatory network by which miRNAs could control CCL2 secretion. On the basis of global gene expression profiling in cohort 1, we performed MARA, a method previously developed by us to find key regulatory elements in monocyte differentiation (20,21). Using MARA, we characterized the activity of regulatory motifs by modeling gene expression patterns in terms of genome wide–predicted TF regulatory sites in proximal promoters (for details, see Supplementary Data). We studied 193 previously defined TF motifs, and a principal component analysis of the data revealed obesity as the second strongest factor affecting TF activity (Fig. 2A). While the first component showed no association with obesity-related anthropometric measures (data not shown), the second component correlated markedly and positively with BMI (Fig. 2B) and several other anthropometric measures, such as waist circumference and fat cell size (data not shown). Thus, there appears to be a clear association between the second component and obesity. Furthermore, we identified 96 TFs and corresponding motifs with altered activity in obesity (Supplementary Table 2). To assess whether these TFs are present in adipocytes, 47 were randomly chosen for qRT-PCR on paired samples of intact WAT and isolated fat cells from cohort 1 (Supplementary Table 3). Of the TFs, 45 were expressed in both adipose preparations and 36 (80%) were significantly more highly expressed in the fat cells. Altogether, these results allowed us to identify motifs and their cognate TFs with altered activity in adipocytes of obese subjects.

Since our transfection screen was performed in human in vitro differentiated adipocytes, we wanted to ensure that the motifs and cognate TFs identified above were also present in our cell model system. We therefore performed a MARA during in vitro differentiation of cultured human fat cells. The motifs identified in the two MARAs were overlapped, and 32 out of the 96 were regulated in both analyses (Supplementary Table 2). An adipocyte-specific transcriptional regulatory network controlling CCL2 production was constructed by linking TFs identified by MARA to CCL2 and our candidate miRNAs (for details, see Supplementary Data and Supplementary Table 4). Five (miR-92a, -126, -193b, -652, and let-7a) of the 10 candidate miRNAs were predicted to target TFs within the CCL2 network (Fig. 3A).

The proposed connections between miRNAs and their predicted TF targets were challenged by overexpressing each miRNA and assessing the effect on TF mRNA expression in human adipocytes. Only miR-92a and miR-193b altered the expression of their predicted targets (Table 3). However, the average effects of miR-92a were small (∼20%) and with some interexperimental variability. Furthermore, 3′UTR analyses (by reporter assays in 3T3-L1 cells) demonstrated that only miR-193b was able to interact with and directly regulate its proposed TFs (Table 3). We therefore focused further validation experiments on miR-193b and its role in the regulation of CCL2. As predicted, overexpression of miR-193b downregulated RELB, STAT6, MAX, ETS1, NFKB1, and CCL2 mRNA levels, while inhibition of miR-193b had reciprocal effects (Fig. 3B–D). The effects of miR-193b on RELB, STAT6, and ETS1 were confirmed at the protein level (Supplementary Fig. 1). The subnetwork comprising miRNAs, TFs, and CCL2 validated by these experiments is shown in Fig. 4A. To assess the functionality of this regulatory circuit, we expanded our network and included IL-6 as a target. According to our bioinformatic analysis, miR-193b is a predicted regulator of IL-6 via NFKB1/RELB. This was confirmed by the observation that overexpression of miR-193b in human adipocytes attenuated IL-6 secretion (Fig. 4B). In contrast, there was no predicted direct or indirect (through TFs) connection between miR-126 and IL-6. In accordance with this, miR-126 overexpression did not affect IL-6 release (Fig. 4B).

Our results identified a transcriptional framework in human adipocytes controlling CCL2 production. However, CCL2 is secreted by several cell types present within the adipose tissue among which macrophages play an important role. Therefore, we assessed the role of miR-193b and miR-126 in controlling CCL2 production in a human monocyte/macrophage cell line (THP1). Similar to our findings in adipocytes, overexpression of either miRNA decreased CCL2 mRNA levels (Fig. 4C) and secretion (Fig. 4D).

Associations between adipose expression of miR-193b, miR-126, CCL2 secretion, and a marker of macrophage infiltration were determined by multiple regression analysis in cohort 1 using BMI as a covariate. There was a significant relationship between miR-193b expression and CCL2 secretion independent of BMI (Table 4). Furthermore, the expression of both miR-193b and miR-126 correlated with integrin, α-X (ITGAX) mRNA levels (a gene specific for proinflammatory [M1] macrophages), regardless of body fat mass (Table 4).

In this study, we identified a set of miRNAs uniformly expressed in human WAT that are altered in obesity and that regulate CCL2 secretion from human fat cells and macrophages. CCL2 is a key factor involved in the development of obesity-induced inflammation. Our findings show that miRNAs regulate CCL2 production via both direct and indirect mechanisms, the latter involving a network of distinct TFs.

miRNAs have emerged as important transcriptional regulators in both physiological and pathophysiological conditions. A number of studies identify several miRNAs to be dysregulated in obese WAT (22–24). However, there has been only a modest overlap concerning the specific miRNAs that have been described in each study. Our present results defining a set of 11 adipocyte-specific miRNAs downregulated in obesity were found in the largest cohort of subjects published so far (n = 56) and performed in both WAT and isolated adipocytes. Admittedly, our candidates only partially overlap with the previous studies, possibly explained by differences in study designs, for example, the use of only limited numbers of lean subjects (24), the inclusion of patients with diabetes (22), and/or the use of different technical methodologies (22–24).

By use of an unbiased approach involving several genome-wide and computational studies, we were able to dissect pathways linking miRNAs, TFs, and CCL2. By overexpressing each miRNA, we found that 10 out of 11 candidate miRNAs affected CCL2 secretion from fat cells. Nine of these miRNAs attenuated CCL2 release, which is in line with the observation that they are downregulated in obesity. In contrast, overexpression of miR-145 resulted in increased CCL2 secretion. However, since miR-145 is downregulated in obesity, the role of this miRNA remains to be defined.

MARA allowed us to construct a transcriptional regulatory network present in adipocytes and altered by obesity that could mediate the effects of individual miRNAs on CCL2 secretion. The predicted network comprised several TFs with well-characterized roles in obesity-induced inflammation (1) and five miRNAs that were directly or indirectly linked to CCL2. For two of the miRNAs, we were able to confirm the regulatory pathways proposed by our model. In particular, we demonstrated a signal circuit from miR-126 directly to CCL2 and from miR-193b to CCL2 either via ETS1 as a single TF step or via a series of other TFs (MAX, NFKB1, RELB, and STAT6). These findings were corroborated by several independent validation experiments and demonstrate a hitherto unknown role for miRNAs in the regulation of adipocyte CCL2 production. miRNA-193b may be of particular importance since its expression was associated with CCL2 secretion from subcutaneous WAT, independent of BMI. To our knowledge, several miRNAs identified in this study have not been previously associated with adipocyte function (miR-26a, -92a, -126, -145, -652, and let-7a/d) or controlling inflammatory pathways (miR-92a, -193a/b, and -652). Moreover, in recent murine studies, let-7 and miR-193b were implicated in controlling muscle glucose metabolism (25) and brown fat differentiation (26), respectively. This further supports a role for miRNAs in regulating different aspects of metabolism.

In this study, we used primary cultures of in vitro differentiated human adipocytes and the human monocyte/macrophage cell line (THP1) in our functional experiments. It could be argued that freshly isolated mature fat cells and adipose-derived macrophages should have been investigated. However, recent publications show that purified mature fat cells are not ideal for studies on inflammation since the isolation procedure per se affects inflammatory pathways (27). In addition, for both practical and ethical reasons, it is not possible to obtain sufficient amounts of adipose-derived macrophages from human biopsies.

Using the present approach, we were not able to identify how 8 out of 10 candidate miRNAs affect CCL2 secretion. There could be several explanations for this. Most important, the network constructed using the present approach is not complete. For instance, TFs without known DNA-binding motifs and other regulatory proteins (i.e., transcriptional cofactors) are not included in the current model. Therefore, more complex transcriptional regulatory networks, which include these and other factors, need to be constructed in future studies. Furthermore, we examined only subcutaneous WAT obtained from female donors and although unlikely, we cannot exclude sex- and region-specific differences in the regulation of CCL2 production.

Although the focus of this study was on fat cells, our data in THP1 cells suggest that miR-193b and miR-126 may also be important for the regulation of CCL2 secretion in macrophages. However, whether the signaling circuit established in adipocytes is also present in macrophages remains to be determined. In any case, the clinical relevance of our findings is supported by the observation that there was a strong association between the expression of miR-193b or miR-126 and the mRNA levels of the M1 macrophage marker ITGAX, independent of BMI.

While CCL2 may induce insulin resistance in skeletal muscle cells via direct mechanisms (28), there are no reports of such effects in fat cells. This suggests that the link between CCL2, adipose tissue inflammation, and local insulin resistance is indirect. Hence, CCL2 may initiate and possibly sustain migration of proinflammatory cells into WAT, which in turn (together with the adipocytes) secrete a number of proteins (e.g., tumor necrosis factor-α and IL-6), which affect insulin signaling directly or indirectly, as discussed previously (29).

Taken together, our results suggest that specific miRNAs may be important regulators of inflammation in human adipose tissue through their effects on CCL2 secretion from adipocytes and macrophages. This may be mediated by direct interactions with CCL2 mRNA and/or via indirect effects on TF circuits. miRNAs could therefore constitute novel potential targets for treating insulin resistance and type 2 diabetes in obesity. The approach presented herein may form the basis for future studies focusing on complex interactions regulating other aspects of human adipose tissue function.

This work was supported by grants from the foundations of Magnus Bergvall, Gun and Bertil Stohne, and Åke Wiberg and Ageing Research at Karolinska Institutet (to E.A.); the Ministry of Education, Culture, Sports, Science, and Technology of Japan for RIKEN Omics Science Center and Innovative Technology (Cell Innovation Program) (to Y.H.); and the Swedish Research Council, the Swedish Heart and Lung Foundation, the Swedish Diabetes Foundation, the Strategic Research Programme in Diabetes at Karolinska Institutet, the Novo Nordisk Foundation, the European Foundation for the Study of Diabetes/Lilly Programme, the European Union (ADAPT, HEALTH-F2-2008-201100), and NordForsk (SYSDIET-070014) (to P.A.).

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

E.A. and C.O.D. designed the study; conducted microarray data, motif activity response, and network analyses; and wrote the manuscript. N.M. designed the study; conducted microarray data, motif activity response, and network analyses; collected tissue; performed the functional analyses; and wrote the manuscript. A.K., M.C., S.L.-C., A.E., J.L., P.H., and J.-F.T. collected tissue and performed the functional analyses. P.J.B., M.P., K.D.-W., Y.H., I.D., and E.v.N. conducted microarray data, motif activity response, and network analyses. M.R. collected tissue, performed the functional analyses, and wrote the manuscript. P.A. designed the study and wrote the manuscript. All authors contributed to data interpretation and reviewed and approved the final manuscript. C.O.D. and P.A. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

The authors thank Elisabeth Dungner, Marika Rönnholm, Eva Sjölin, Lena Lindberg, Kerstin Wåhlén, Gaby Åström, Britt-Marie Leijonhufvud, Katarina Hertel, and Yvonne Widlund (Department of Medicine, Huddinge, Lipid Laboratory, Karolinska Institutet) for excellent technical assistance.