Additive Effects of Micrornas and Transcription Factors on CCL2 Production in Human White Adipose Tissue

1248 Diabetes Volume 63, April 2014

Agné Kulyté,1 Yasmina Belarbi,1 Silvia Lorente-Cebrián,1 Clara Bambace,1 Erik Arner,1,2 Carsten O. Daub,3 Per Hedén,4 Mikael Rydén,1 Niklas Mejhert,1 and Peter Arner1

Additive Effects of MicroRNAs and Transcription Factors on CCL2 Production in Human White Adipose Tissue

Adipose tissue inflammation is present in insulin- converged on the nuclear factor-kB pathway. In resistant conditions. We recently proposed conclusion, TF and miRNA-mediated regulation of a network of microRNAs (miRNAs) and transcription CCL2 production is additive and partly relayed by factors (TFs) regulating the production of the cell-specific networks in human adipose tissue that proinflammatory chemokine (C-C motif) ligand-2 may be important for the development of insulin (CCL2) in adipose tissue. We presently extended and resistance/type 2 diabetes. further validated this network and investigated if the Diabetes 2014;63:1248–1258 | DOI: 10.2337/db13-0702

METABOLISM circuits controlling CCL2 can interact in human adipocytes and macrophages. The updated subnetwork predicted that miR-126/-193b/-92a White adipose tissue (WAT) function plays an important control CCL2 production by several TFs, including role in the development of insulin resistance/type 2 di- v-ets erythroblastosis virus E26 oncogene homolog 1 abetes. Fat cells present in WAT secrete a number of (avian) (ETS1), MYC-associated factor X (MAX), molecules, collectively termed adipokines, which affect and specificity protein 12 (SP1). This was confirmed insulin sensitivity by autocrine and/or paracrine mecha- in human adipocytes by the observation that gene nisms (1,2). In insulin-resistant obese subjects, WAT silencing of ETS1, MAX, or SP1 attenuated CCL2 displays a chronic low-grade inflammation, which is production. Combined gene silencing of ETS1 and proposed to alter insulin signaling by increased fatty acid MAX resulted in an additive reduction in CCL2 release and altered adipokine production (3). production. Moreover, overexpression of miR-126/ Chemokine (C-C motif) ligand-2 (CCL2), also referred -193b/-92a in different pairwise combinations as monocyte chemoattractant protein-1, is an early reduced CCL2 secretion more efficiently than either proinflammatory signal in WAT that acts by promoting miRNA alone. However, although effects on CCL2 adipose infiltration of macrophages (4,5). After diet- secretion by co-overexpression of miR-92a/-193b induced obesity, Ccl2-/- mice are less insulin-resistant and and miR-92a/-126 were additive in adipocytes, the display a less pronounced inflammatory reaction in WAT combination of miR-126/-193b was primarily additive (6). Adipocytes and macrophages both contribute sig- in macrophages. Signals for miR-92a and -193b nificantly to CCL2 secretion from WAT.

1Department of Medicine, Huddinge, Lipid Laboratory, Karolinska Institutet, Corresponding authors: Peter Arner, [email protected], and Agné Kulyté, agne Stockholm, Sweden [email protected]. 2 RIKEN Center for Life Science Technologies, Division of Genomic Technologies, Received 3 May 2013 and accepted 15 December 2013. Yokohama, Japan This article contains Supplementary Data online at http://diabetes 3Department of Biosciences and Nutrition, Huddinge, Karolinska Institutet, .diabetesjournals.org/lookup/suppl/doi:10.2337/db13-0702/-/DC1. Stockholm, Sweden 4Akademikliniken, Stockholm, Sweden © 2014 by the American Diabetes Association. See http://creativecommons .org/licenses/by-nc-nd/3.0/ for details. diabetes.diabetesjournals.org Kulyté and Associates 1249

Little is known about the mechanisms regulating WAT Another part of the tissue (at least 500 mg) was sub- inflammation. Results in recent years have demonstrated jected to collagenase treatment, and mean adipocyte that microRNAs (miRNAs) play important roles in con- volume and weight were determined as described (16). trolling inflammation in other conditions and non- Packed fat cells (200 mL) and intact WAT (400 mg) were adipose tissues, as reviewed (7). In mammals, miRNAs frozen at 270°C for future mRNA and miRNA mea- act by inhibiting the expression of target genes after surements as described (11). The Karolinska University binding to defined complementary sites in target 39- Hospital Ethical Committee (Stockholm, Sweden) ap- untranslated regions (UTRs) (8). proved this study. All subjects were informed in detail Cellular function is controlled by transcriptional about the studies, and written informed consent was (transcription factors [TFs]) and posttranscriptional obtained. mechanisms (miRNAs). Using global expression profiling For in vitro studies, subcutaneous WAT was obtained data and mathematical models, the transcriptome can be from healthy men and women undergoing cosmetic li- mapped into transcriptional regulatory networks (TRNs), posuction. This experimental group was not selected for with regulatory elements (e.g., TFs and miRNAs) as age, sex, or BMI. Isolation of human adipocyte progenitor nodes and their interactions as edges (9,10). We recently cells from subcutaneous WAT was performed as de- described and partly experimentally verified a miRNA-TF scribed (17). The progenitor cells obtained from separate network in the subcutaneous WAT of insulin-resistant individuals were not mixed. obese women that controls the expression of CCL2 (11). Affymetrix GeneChip Human Gene 1.0 ST and miRNA This TRN suggests that several miRNAs and TFs interact Array Protocols in different combinations to regulate CCL2 production. Although the reason for this complex regulation and re- Data obtained from gene and miRNA arrays from human dundancy is not entirely clear, it may allow for more exact subcutaneous WAT (Gene Expression Omnibus number fine-tuning of gene expression and/or signal enhancement GSE25402) and the protocols have been described (11). fi in response to speci c environmental cues. In line with Cell Culture this, recent studies in cancer cells have proposed that Culture and in vitro differentiation of human adipocyte partly overlapping regulatory circuits allow signal ampli- progenitor cells were performed as described (17). 3T3-L1 fication in an additive or synergistic manner (12,13). To andTHP1cellswerehandledasrecommendedinthepro- the best of our knowledge, it is not known whether these tocols from American Type Culture Collection (Manassas, types of mechanisms are present in WAT or whether they VA). For the induction of monocyte–macrophage differen- differ between specific cell types within the tissue. tiation, THP1 cells were stimulatedwith50ng/mLphorbol Through combinatorial gain-of-function/loss-of- 12-myristate 13-acetate (PMA). function studies focusing on TFs and miRNAs, we investigated how transcriptional regulators (TFs and miRNA and Small Interfering RNA Transfection miRNAs) can integrate effects on the expression of In vitro differentiated adipocytes (day 10–12 post- inflammatory factors in WAT. First, we extended and induction) were treated for 24–48 h with various con- further validated the proposed TRN (11). We sub- centrations (5–40 nmol/L) of miRIDIAN miRNA mimics sequently silenced potentially important TFs in the net- for overexpression of miRNA activity or 60 nmol/L of work, alone or in combination, and investigated how this miRIDIAN miRNA Hairpin Inhibitors for native miRNA altered CCL2 production in human adipocytes. Finally, inhibition (Thermo Fisher Scientific, Lafayette, CO) and we determined whether concomitant overexpression of HiPerFect Transfection Reagent (Qiagen, Hilden, miRNAs in human adipocytes or macrophages resulted in Germany), according to the manufacturers’ protocols. increased effects when compared with overexpressing Optimal transfection conditions were determined in each miRNA alone. separate titration experiments. Transfection efficiency was assessed with real-time quantitative PCR (qPCR) using RESEARCH DESIGN AND METHODS miRNA probes/assays (Applied Biosystems, Foster City, Clinical Material CA; and Qiagen, respectively). To rule out unspecific Cohort 1 comprised 30 obese (BMI .30 kg/m2) and effects, control cells were transfected with miRIDIAN 26 nonobese (BMI ,30 kg/m2) women who had no miRNA Mimic or Hairpin Inhibitor Non-Targeting Nega- chronic disease and were free of continuous medication. tive Controls (Thermo Fisher Scientific). This cohort has been described in detail previously (11). In some experiments, in vitro differentiated adipo- The subjects came to the laboratory in the morning after cytes were cotransfected with miScript Target Protector an overnight fast. Height, weight, and waist circumfer- for the miR-92a binding site on specificity protein 12 ence were determined. An abdominal subcutaneous WAT (SP1; 500 nmol/L; Qiagen) and miR-92a mimics (40 biopsy specimen (;1.0–1.5 g) was obtained by needle nmol/L) or appropriate control reagents (miScript Target aspiration as described (14). One part (300 mg) of the Protector Negative Control; Qiagen) and miRIDIAN tissue was used for measurement of CCL2 release and miRNA Mimic Non-Targeting Negative Control (Thermo expressed per number of fat cells as described (15). Fisher Scientific). THP1 cells were transfected with 1250 Additive Regulation of CCL2 by miRNAs and TFs Diabetes Volume 63, April 2014 miRNA mimics, as described above, for adipocytes 48 h Global Gene Expression Analysis, Motif Activity after incubation with PMA. Response Analysis, and Network Construction For RNA interference experiments, adipocytes were Motif activity response analysis and construction of the treated with various concentrations (10–40 nmol/L), as adipocyte CCL2 TRN used in this study has been de- described above, using ON-TARGETplus SMARTpool scribed in detail (11,19). Predicted targets of miRNAs small interfering RNA (siRNA) for v-ets erythroblastosis were updated according to the latest release of TargetScan virus E26 oncogene homolog 1 (avian) (ETS1), SP1, or (March 2013). MYC-associated factor X (MAX) and appropriate concentrations of siRNA Non-Targeting Negative Control Statistical Analyses (Thermo Fisher Scientific) instead of miRNA reagents. Data presented are mean 6 SEM. When appropriate, the data were log-transformed to become normally distrib- RNA Isolation, cDNA Synthesis, and Real-Time PCR uted. Results were analyzed with unpaired t test, linear Total RNA was extracted from in vitro differentiated regression (simple and multiple), or ANOVA (repeated- adipocytes, 3T3-L1 cells, and THP1 macrophages, and measures). purity control was performed as described (11). Syn- thesis of coding-gene cDNA and real-time qPCR using RESULTS Taqman probes (Applied Biosystems) was performed as Strategy for Investigating Additive Effects of TFs and described (11). Synthesis of miRNA cDNA and following miRNAs real-time qPCR was performed using Applied Bio- To dissect the regulation of adipocyte CCL2 pro- systems reagents as described (11) or using the miScript duction by miRNAs and TFs, we extended and further II RT Kit and miScript Primer Assays (Qiagen) according validated a recently described TRN present in human to the manufacturer’s instructions. Relative gene ex- fat cells and perturbed in obese insulin-resistant sub- pression was calculated using the comparative Ct D D jects (11). In brief, this in silico–generated TRN links method (i.e., 2 Ct-target gene/2 Ct-reference gene)with18S as for several miRNAs (miR-92a/-126/-193b/-652 and the internal control. Expression of miRNA was normalized let-7a), directly or indirectly, through one or two in- to a reference gene RNU48 or SNORD68. Levels of 18S, termediate TF steps, to CCL2 production. For the RNU48, and SNORD68 did not differ between groups. current study, we focused on a subpart of the TRN Western Blot comprising two partly validated circuits, from miR-126 Cells were lysed in radioimmunoprecipitation assay and -193b to CCL2, and a third possible candidate, buffer as described (18). SDS-PAGE was used to separate miR-92a, which was not fully characterized in our A 15–20 mg total protein, and Western blot was previous study (Fig. 1 ). performed according to standard protocols. The miR-92a Regulates CCL2 Production by SP1 in membranes were blocked in 3% enhanced chemilu- Adipocytes minescence Advance Blocking Agent (GE Healthcare, Because the regulatory edges linking miR-92a to CCL2 Buckinghamshire, U.K.). Primary antibodies against have not been explored in detail, we first tested if miR- ETS1 were from Novocastra Leica Biosystems AB (Wet- 92a affected its predicted target genes SP1 and ETS1 in zlar, Germany) and against SP1 were from Santa Cruz human adipocytes. We overexpressed or inhibited miR- Biotechnology (Santa Cruz, CA). b-Actin (Sigma-Aldrich, 92a using miRNA mimics/inhibitors and evaluated ex- St. Louis, MO) was used as a loading control. Secondary pression of SP1 as well as mRNA expression of ETS1. antibodies mouse/rabbit IgG-horseradish peroxidase were Modulation of miR-92a expression levels affected the from Sigma-Aldrich. All tested primary antibodies detecting mRNA and protein of SP1, whereas ETS1 expression MAX were unspecific and results therefore could not be remained unaltered (Fig. 1B–D and Table 1), suggesting demonstrated. Antibody-antigen complexes were detected that SP1, but not ETS1, is a direct target of miR-92a. by chemiluminescence using ECL Select Western Blotting The levels of CCL2 in conditioned media mirrored the DetectionKit(GEHealthcare). changes in SP1 mRNA/protein expression (Fig. 1B and ELISA D). Luciferase reporter assays confirmed that miR-92a 9 CCL2 levels in conditioned media from in vitro differ- interacted directly with the 3 UTR region of SP1 E entiated adipocytes and macrophages were analyzed using (Fig. 1 ). an ELISA assay from R&D Systems (Minneapolis, MN). To verify that SP1 could regulate CCL2 production in human adipocytes independently of miR-92a, the Luciferase Reporter Assay expression of SP1 was silenced using siRNA at two Empty luciferase reporter vector and vector containing different concentrations (10 and 20 nmol/L). SP1 gene a part of 39UTR of SP1 (enclosing predicted binding site knockdown significantly attenuated CCL2 secretion for hsa-miR-92a) were obtained from GeneCopoeia (Fig. 2A) and reduced SP1 mRNA/protein levels (Fig. (Rockville, MD). The luciferase reporter assay in 3T3-L1 2B and C). In addition, SP1 silencing affected mRNA cells was performed as described (11). expression of two predicted direct targets in the TRN: diabetes.diabetesjournals.org Kulyté and Associates 1251

Figure 1—miR-92a regulates SP1 and CCL2 in human adipocytes. A: The proposed detailed CCL2 regulatory subnetwork in human adipocytes depicting miR-193b, -126, and -92a (□) as well as motifs and cognate TFs (○) altered by obesity and targeting CCL2 (◇). T-bars represent inhibition; arrows represent stimulation. Bold edges represent physical interactions between nodes described previously in adipocytes, and thin lines represent predictions by network analyses and/or interactions in other cell types. Under the NF-kB complex, the following TFs are merged: NF-kB1, RELA (v-rel reticuloendotheliosis viral oncogene homolog A [avian]), RELB, and REL. B: Predicted interaction between miR-92a and SP1 was investigated by overexpressing miR-92a and assessing the effects on SP1 mRNA expression in human in vitro differentiated adipocytes. Changes of CCL2 secretion in the same experiments are shown in 1252 Additive Regulation of CCL2 by miRNAs and TFs Diabetes Volume 63, April 2014

Table 1—Interactions between miRNAs or TFs and ﬁrst predicted neighbors miRNA/TF Predicted ﬁrst Fold change t test siRNA neighbor treatment/NegC P value Reference miR-92a SP1 0.78 0.031 This study (Fig. 1B) ETS1 1.09 0.3108 This study miR-193b ETS1 0.54 0.0001 (11) MAX 0.58 0.0001 (11) SP1 1.04 0.681 This study miR-126 CCL2 0.67 0.0112 (11) REL 1.16 0.3714 This study siSP1 CCL2 0.76 0.007 This study REL 0.77 0.0365 This study and (28) (not adipocytes) NFKB1 0.85 0.3563 This study ETS2 NE — This study and (29) (not adipocytes) siETS1 MAX 0.82 0.0178 This study NFKB1 0.80 0.0242 This study and (27) (not adipocytes) STAT6 0.67 0.0023 This study CCL2 0.56 0.0055 This study and (23) (not adipocytes) siMAX RELB 0.84 0.0106 This study

NE, not evaluated due to low expression. Predicted interactions between miRNAs and first neighbors were investigated by overexpression of each miRNA and assessing mRNA expression of TFs in human in vitro differentiated adipocytes. Predicted TF interactions with first neighbors were investigated by transfecting human in vitro differentiated adipocytes with siRNA and assessing the effects on mRNA expression of first neighbors. NegC, negative control.

CCL2 and v-rel reticuloendotheliosis viral oncogene levels, suggesting that they are not true targets for these homolog [avian] (REL), a part of the nuclear factor miRNAs in human adipocytes (Table 1). (NF)-kB network (Table 1). To further investigate the role of miR-92a in CCL2 production by SP1, we con- ETS1 and MAX Regulate CCL2 Production in Human comitantly inhibited expression of miR-92a and SP1. Adipocytes by TRN in an Additive Fashion The combination of miR-92a inhibition and silencing According to Fig. 1A, ETS1 and MAX are entry TFs for of SP1 abolished their individual effects on CCL2 miR-193b; moreover, they have a predicted interaction production (Fig. 2D). Attenuation of miR-92a levels and have been validated as direct targets of miR-193b increased CCL2 secretion to the same extent as pre- (11). To determine whether ETS1 and MAX could affect viously shown in Fig. 1D. Moreover, simultaneous CCL2 secretion independently of miR-193b, we knocked down ETS1 and MAX in adipocytes using various concen- transfection of miR-92a mimics and a target protector trations of siRNAs, followed by measurements of mRNA (corresponding to the miR-92a target sequence on SP1 levels of their first predicted neighbors in the TRN (Fig. 39UTR) abolished the effect of miR-92a on CCL2 pro- 1A). We could confirm that silencing of the ETS1 gene E duction (Fig. 2 ). decreased the mRNA levels of MAX, signal transducer and Previous validations of the interactions in the CCL2 activator of transcription 6, interleukin-4 induced (STAT6), fi TRN identi ed that miR-126 targeted CCL2 directly, nuclear factor of k light polypeptide gene enhancer in whereas miR-193b affected its production indirectly B-cells 1 (NFKB1), and CCL2,whereassilencingofMAX through several TFs, including ETS1 and MAX (11). In decreased the mRNA levels of v-rel reticuloendotheliosis the current study, we identified two additional TFs as viral oncogene homolog B (RELB)(Table1). possible targets for miR-126 and miR-193b, REL and Silencing of ETS1 and MAX with 10 nmol/L of siRNA SP1, respectively (Fig. 1A). However, overexpression of had modest effects on CCL2 secretion, but higher con- miR-126 or miR-193b did not alter REL and SP1 mRNA centrations (i.e., 20 nmol/L) resulted in significantly

parallel (n = 8). C: Representative Western blots of SP1 protein after overexpression and inhibition of miR-92a. D: Quantiﬁcation of SP1 protein levels after overexpression and inhibition of miR-92a. Changes of CCL2 secretion are shown in parallel (n =6–9). E: miR-92a was overexpressed together with SP1 39UTR luciferase reporter in 3T3-L1 cells, and luciferase activity was measured (n = 8). NegC, negative control. ***P < 0.001; **P < 0.01; *P < 0.05. diabetes.diabetesjournals.org Kulyté and Associates 1253

Figure 2—SP1 regulates CCL2 in human adipocytes. A: TF SP1 was silenced with siRNA at 10 and 20 nmol/L final concentrations in in vitro differentiated adipocytes, and secreted levels of CCL2 were assessed by ELISA (n =12–18). B: TF SP1 was silenced with siRNA at 10 and 20 nmol/L final concentration in in vitro differentiated adipocytes, and mRNA levels of SP1 were assessed (n =6–9). C: Protein levels of SP1 were measured after silencing of SP1 (n = 6). Representative blot is shown. D: Expression of SP1 and levels of miR-92a were attenuated using siRNA and inhibitor of miRNA-92 individually or in pair with each other, and CCL2 secretion was assessed using ELISA (n =6–14). E: Interaction of miR-92a-SP1 was studied by cotransfecting human in vitro differentiated adipocytes with miR-92a mimics and target protector (corresponds to the binding site of miR-92a on the 39UTR of SP1) after evaluation of CCL2 secretion by ELISA (n = 7). To rule out unspecific effects, cells in parallel were transfected with appropriate negative control (nontargeting negative control for miRNA mimics or inhibitor, nontargeting negative control for siRNA, or negative control for target protector). Levels of mRNA were calculated using a comparative Ct-method (i.e., 2DCt-target gene/2DCt-reference gene) with 18S as reference gene. SP1 protein levels were evaluated by Western blotting. Levels of b-actin were used to control loading. Data were analyzed using the t test and are presented as fold change 6 SEM relative to the negative control. NegC, negative control. ***P < 0.001; **P < 0.01; *P < 0.05.

decreased CCL2 production (Fig. 3A). Because CCL2 when silenced alone at 10/20 nmol/L or in combination protein levels were affected by both TFs, and MAX mRNA (10+10 nmol/L). ETS1-siRNA treatment reduced protein expression was affected by silencing of ETS1, we assessed levels of ETS1 (Fig. 3D). Owing to the poor specificity of whether ETS1 and MAX could interact with each other in tested MAX antibodies, we were not able to determine controlling CCL2 secretion. Indeed, concomitant silenc- protein levels of this TF. ing of ETS1 and MAX using low concentrations (10 As mentioned above, we observed that 20 nmol/L nmol/L) of siRNA resulted in a more pronounced re- siRNA (ETS1 or MAX) caused a more pronounced de- duction of CCL2 secretion (10+10 nmol/L of each siRNA) crease in CCL2 production than 10 nmol/L. Un- compared with single knockdown of either TF (Fig. 3A). fortunately, we did not detect significant quantitative Expression of CCL2 mRNA followed the same trend as differences in the mRNA levels of ETS1 and MAX be- CCL2 secretion (Supplementary Fig. 1A). tween 10 and 20 nmol/L of siRNA at a given time point. Knockdown efficiency was determined by quantifying To control for off-target effects, we treated in vitro adi- ETS1 and MAX mRNA levels (Fig. 3B and C). There was pocytes with transfection agent alone or transfected with a marked decrease in mRNA levels of ETS1 and MAX various concentrations of siRNA nontargeting negative 1254 Additive Regulation of CCL2 by miRNAs and TFs Diabetes Volume 63, April 2014

Figure 3—ETS1 and MAX regulate CCL2 production in human adipocytes. A: TFs ETS1 and MAX were silenced with siRNA alone at 10 or 20 nmol/L ﬁnal concentration in in vitro differentiated adipocytes, and secreted levels of CCL2 were assessed by ELISA. ETS1 and MAX were also cosilenced in pair with each other (10+10 nmol/L each miRNA), and secreted levels of CCL2 were assessed (n =11–17). B: ETS1 and MAX were silenced alone (10 and 20 nmol/L) or cosilenced in pair with each other (10+10 nmol/L each miRNA), and mRNA of ETS1 was assessed (n =11–17). C: ETS1 and MAX were silenced alone (10 and 20 nmol/L) or cosilenced in pair with each other (10+10 nmol/L each miRNA), and mRNA of MAX was assessed (n =11–17). D: Protein levels of ETS1 were measured after silencing (n = 6). Repre- sentative blot is shown. To rule out unspeciﬁc effects, cells in parallel were transfected with siRNA nontargeting negative control. Levels of mRNA were calculated using a comparative Ct-method (i.e., 2DCt-target gene/2DCt-reference gene) with 18S as the reference gene. ETS1 protein levels were evaluated by Western blotting. Levels of b-actin were used to control loading. Data were analyzed using the t test and are presented as fold change 6 SEM relative to the negative control. NegC, negative control. ***P < 0.001; **P < 0.01; *P < 0.05.

control, followed by mRNA measurements. Indeed, basal overexpression of miR-193b, -126, or -92a at high con- expression levels of CCL2, ETS1,orMAX remained stable centrations (i.e., 40 nmol/L) downregulate CCL2 pro- (Supplementary Fig. 2). duction to a substantial degree (30–50%) (11). Here we Taken together, experiments in Figs. 1–3 demonstrate assessed how co-overexpression of miRNAs affected that miR-193b and -92a attenuate CCL2 production in CCL2 secretion. A caveat is that overexpression of mul- human adipocytes independently by TFs nodes com- tiple miRNA mimics, particularly at high concentrations, prising ETS1/MAX and SP1, respectively. increases the risk for off-target effects. Therefore, lower concentrations of miRNA mimics were used in the miRNA-193b, -126, and -92a Cooperatively Affect combined transfections. At 5 nmol/L, single over- CCL2 Production in Human Adipocytes expression of miR-193b, -126, and -92a had signiﬁcantly Because our data presented above indicate that TFs are weaker effects on CCL2 secretion compared with con- able to additively control CCL2 secretion, we assessed if comitant overexpression of miR-92a with miR-193b or this mode of action is also present when combining -126 (Fig. 4A). However, combining miR-193b and -126 miRNAs. We have previously shown that single did not cause a more pronounced effect than miR-126 diabetes.diabetesjournals.org Kulyté and Associates 1255

Figure 4—miRNAs regulate CCL2 in an additive manner in human adipose tissue. A: miR-193b, -126, and -92a were overexpressed alone at 5 nmol/L final concentration or co-overexpressed in pairs (5+5 nmol/L of each miRNA) in in vitro differentiated adipocytes, and secreted levels of CCL2 were assessed by ELISA (n =16–24). B: miR-193b, -126, and -92a were overexpressed alone at 5 nmol/L final concentration or co-overexpressed in pairs (5+5 nmol/L each miRNA) in in vitro differentiated adipocytes, and mRNA levels of NFKB1 were assessed (n =14–19). C: miR-193b, -126, and -92a were overexpressed alone at 5 or 10 nmol/L final concentration or co-overexpressed in pairs (5+5 nmol/L of each miRNA) in in vitro differentiated adipocytes, and mRNA levels of RELB were assessed (n =14–19). D: miR-193b, -126, and -92a were overexpressed alone at 10 or 20 nmol/L final concentration or co-overexpressed in pairs (10+10 nmol/L each miRNA) in in vitro differentiated THP1 cells, and secreted levels of CCL2 were assessed by ELISA (n = 9). To rule out unspecific effects, cells in parallel were transfected with an equal concentration of nontargeting negative control for miRNA mimics. Levels of mRNA were calculated using a comparative Ct-method (i.e., 2DCt-target gene/2DCt-reference gene) with 18S as the reference gene. Results were analyzed using the t test and are presented as fold change 6 SEM relative to the negative control. NegC, negative control. ***P < 0.001; **P < 0.01; *P < 0.05.

alone. The effects on CCL2 secretion were paralleled by RELB in the miRNA co-overexpression experiments. Al- similar changes in mRNA expression (Supplementary though no changes occurred in the expression levels of Fig. 1B). NFKB1 (Fig. 4B), RELB mRNA expression was attenuated To rule out the possibility that the additive effects on by the combined overexpression of miR-92a with miR- CCL2 production by combined overexpression of miRNAs 193b or miR-126 (Fig. 4C). were not due to alterations in transfection efficiency, we quantified miRNA levels after transfection. Indeed, there Cooperative Effects of miRNA 193b, -126, and -92a on were no differences in miRNA abundance irrespective of CCL2 Production in Human Macrophages whether they were overexpressed alone or in pairs That CCL2 is secreted by several other cell types present (Supplementary Fig. 3). within WAT, among which macrophages are probably the As suggested in Fig. 1A, the signaling circuits for miR- most important (20), is well known. Therefore, we 193b and -92a converge on TFs in the NF-kB pathway. assessed the effects of miR-193b, -126, and -92a on CCL2 We therefore quantified mRNA levels of NFKB1 and secretion in the THP1 human monocyte/macrophage cell 1256 Additive Regulation of CCL2 by miRNAs and TFs Diabetes Volume 63, April 2014

line. We previously used high concentrations of mimic Table 2—Correlation between CCL2 secretion, expression reagents (40 nmol/L) to show that miR-193b and -126 of miRNAs in adipose tissue and activity of ETS1 and MAX affected CCL2 production in these cells (40 nmol/L) determined by motif activity response analysis (11). To observe possible additive effects of miRNAs Single regression Stepwise regression on CCL2 production, we overexpressed miRNAs at r P F P lower concentrations (10 nmol/L). In this experimen- Regressor value value value value tal setting, overexpression of miR-92a alone signifi- miR-193b alone 20.39 0.007 —— cantly decreased CCL2 secretion, whereas miR-126 alone 20.21 0.15 —— overexpressing miR-193b caused somewhat increased miR-92a alone 20.28 0.053 —— (albeit not statistically significant) CCL2 secretion. —— In contrast to the results obtained in adipocytes, miR-193b + -92a For miR-193b 10.0 co-overexpression miR-193b and -126 decreased the For miR-92a —— 5.8 0.002 fi secretion of CCL2 signi cantly, whereas the combina- miR-193b + -126 —— tions of miR-193b/-92a and miR-126/-92a showed less For miR-193b 13.8 prominent effects (Fig. 4D). For miR-126 —— 7.2 0.0009 Similar to the experiments in adipocytes, we quanti- ETS1 alone 0.57 ,0.0001 —— fi ed the expression of miRNAs after transfection in THP1 MAX alone 0.41 0.004 —— macrophages. There were no differences in the abundance ETS1 + MAX —— of miR-126 or -92a irrespective of whether they were For ETS1 19.8 overexpressed alone or in pairs (Supplementary Fig. 4). For MAX —— 7.7 ,0.001 Owing technical problems of tested probes for miR-193b, we were not able to determine levels of this miRNA in An F value .4 indicates that the regressors have significantly THP1 cells. entered the model. CCL2 secretion was expressed per number of fat cells. Association Between Expression of Adipose Tissue miR-193b, -126, and -92a and CCL2 Secretion Thepossibleinvivorelevance was assessed by corre- lating expression levels of miRNAs (miR-193b, -126, DISCUSSION and -92a) with CCL2 secretion in adipose tissue of In this study, we tested whether combined silencing of cohort 1 (results in Table 2). In simple regression, only TFs (ETS1 and MAX) and overexpression of miRNAs miR-193b correlated significantly (and negatively) (miR-193b/-126/-92a) resulted in augmented effects in with CCL2 secretion. However, when miRNA levels CCL2 production. Our main conclusion is that the three were combined in a stepwise regression, miR-193b/ investigated miRNAs (miR-193b, -126, and -92a) regu- -92a and miR-193b/-126 interacted significantly in lated CCL2 production by distinct pathways involving the negative correlation with CCL2. The combinations specific TFs. Concomitant overexpression of miRNA pairs explained 21–24% of the interindividual variations in or paired silencing of their cognate target TFs (ETS1 and CCL2 secretion (adjusted r2). There was no significant MAX) resulted in additive effects on CCL2 production. correlation with CCL2 release when miR-92a and -126 The enhanced effects of miRNAs pairs were qualitatively were combined as regressors (values not shown). different in human adipocytes compared with human miRNA-193b always entered as the first step in the macrophages. This suggests that TRNs regulating CCL2 models. production are cell-specific and that the combined effects As shown in Fig. 1A, miR-193b targets two TFs in the of miRNAs influence cellular phenotypes. TRN: ETS1 and MAX. We applied regression analysis to Several biological processes, including diseases such as further investigate if the activity of these TFs covaried in cancer, have been causatively associated with dis- their relationship with adipose CCL2 secretion (Table 2). turbances in the interplay between miRNAs and TF in In simple regression, the motif activity of either TF was vitro and in vivo, as reviewed (21). MiRNAs and TFs are significantly and positively correlated with adipose CCL2 transactivating factors that interact with cis-regulatory secretion (Table 2). In a stepwise regression, their motif elements, potentially generating complex combinatorial activities acted together in a significant manner with effects. Here, we extensively characterized the first regard to their correlation with CCL2 secretion. Together downstream targets of miR-193b and -92a in the pro- they explained 40% of the CCL2 variation (i.e., adjusted posed CCL2 TRN. By knocking down ETS1, MAX, and r2). ETS1 entered as the first step in the relationship. We SP1, we demonstrate that these TFs downregulate CCL2 also investigated by regression analysis if the activity of production in fat cells independently of the targeting SP1 (target of miR-92a) covaried with adipose CCL2 se- miRNAs. This suggests that these three TFs are upstream cretion. A simple regression showed the SP1 motif ac- regulators of CCL2 in human adipose tissue. The results tivity was significantly and positively correlated with with ETS1 and SP1 are in accordance with findings in adipose CCL2 secretion (r = 0.42, P = 0.003). nonadipose tissue demonstrating that both TFs control diabetes.diabetesjournals.org Kulyté and Associates 1257

CCL2 (22,23). Although the MYC/MAX/MAD family are and -126, the effects are most probably obtained due to well-established regulators of key processes in basic cell a combination of the TF network (miR-92a acts through physiology, including cell metabolism (24), their in- SP1) and the direct interaction of miR-126 with CCL2 volvement in the regulation of CCL2 has not been de- (interaction A2). scribed before. Our study therefore suggests a novel role We also investigated if combinations of miRNAs could for MAX in the regulation of adipose inflammation. Be- alter signaling further downstream in the TF network. cause ETS1 and MAX were both direct targets for miR- Co-overexpression of miR-92a with miR-193b or miR-126 193b, we explored the effects on CCL2 production of downregulated RELB mRNA expression. This suggests that combined knockdown of these two genes. This resulted in adipocytes, an additive effect of miRNAs on CCL2 in more pronounced inhibition of CCL2 production production may be due to signals converging onto the NF- compared with single knockdown of either TF. Thus, kB/REL pathway. These results are in concordance with miR-193b may amplify its signal to CCL2 through previously published data on other cell types demon- interactions within the TF network. strating that NF-kB/REL are downstream of SP1, ETS1, To study the combined effects of miRNAs in human and MYC (an interaction partner with MAX) (25–27). adipocytes, we overexpressed miR-193b/-126/-92a in- To study cell-specific miRNA effects on CCL2 regula- dividually and in pairs. In fat cells, the combination of tion, we performed similar experiments in the human miR-92a with miR-193b or -126 caused a more marked macrophage THP1 cell line. In contrast to fat cells, only downregulation of CCL2 production compared with ei- the combination of miR-193b and miR-126 had an ad- ther miRNA alone. This demonstrates two types of signal ditive effect on CCL2 production (Fig. 5; interaction M1). amplifications, which are summarized in Fig. 5. For miR- Whether the miRNA-TF regulatory circuits established in 92a and -193b, there are interactions through the TF adipocytes is also acting in macrophages remains to be network that in an additive manner amplify the effects determined. Nevertheless, our data demonstrate that on CCL2 secretion (interaction A1 and A3). For miR-92a miRNAs regulate CCL2 differentially in human macrophages compared with fat cells. Are additive effects of miRNA signaling of clinical relevance? Unfortunately, determining this in vivo is not possible. To shed some light on this relevant issue, we performed extensive correlation analyses between miRNA expression, TF activities, and CCL2 secretion in human subcutaneous adipose tissue. The combined expression of miR-193b/-92a or miR-193b/-126 explained interindividual variations in CCL2 secretion to a larger extent than either miRNA alone. The same was true for the combined expression of MAX and ETS1. This sup- ports the hypothesis that additive effects of miRNA and TFs may be clinically relevant although, admittedly, caution should be taken when extrapolating these statistical correlations into an in vivo situation. It is also important to stress that the TRN constructed using the present approach is not complete and most likely includes other factors not evaluated in the current and previous analysis (11). Most of the work characterizing miRNAs associated with obesity and/or insulin resistance concerns single miRNAs rather than combinations or clusters of miRNAs. We propose the following model for how miRNAs affect CCL2 production in human adipose tissue, where the sig- Figure 5—Suggested model of integrated regulation of CCL2 by miRNAs and TFs in human adipocytes and macrophages. The vali- nal may be relayed in at least three different ways (Fig. 5): dated transcriptional regulatory subnetwork including three miRNAs 1) as the combined effects of direct interactions with CCL2 (miR-126, -193b, and -92a; □), three TFs (MAX, ETS1, and SP1; ○), and indirect interactions with specificTFs,2) as the in- k and an NF- B subnetwork all directly or indirectly regulating the ex- teraction of several miRNAs in a TF network, and 3)asthe pression and secretion of CCL2 (◆)fromtheadiposetissueinobesity (upper part). In the lower part of the model, miRNAs (miR-126 and interaction of a single miRNA with several TFs. These -193b; □) and possible TF network (○) regulating CCL2 production in combined effects are specific for different cell types within macrophages are presented. T-bars represent inhibition, and arrows adipose tissue (e.g., fat cells and macrophages). In addition, represent stimulation. Bold edges represent interactions between there are probably other signaling pathways that could be nodes described in this and previous studies in adipocytes, and thin lines are predicted by network analyses or shown in other cells types. important. For example, the expression of other cytoche- Dashed edges/lines in the lower part represent possible interactions. mokines is probably under the control of other TRNs, and 1258 Additive Regulation of CCL2 by miRNAs and TFs Diabetes Volume 63, April 2014 other WAT regions besides the subcutaneous fat could be 9. Martinez NJ, Walhout AJ. The interplay between transcription factors and controlled by miRNAs in a different way. microRNAs in genome-scale regulatory networks. Bioessays 2009;31: In summary, human adipose CCL2 production is reg- 435–445 ulated by a local miRNA-TF network that allows diverse 10. Schadt EE. Molecular networks as sensors and drivers of common human signal amplification and is specific for different cell types diseases. Nature 2009;461:218–223 present in the tissue. This may be an important factor 11. Arner E, Mejhert N, Kulyté A, et al. Adipose tissue microRNAs as regulators linking adipose tissue inflammation, insulin resistance, of CCL2 production in human obesity. Diabetes 2012;61:1986–1993 and type 2 diabetes. 12. Hobert O. Common logic of transcription factor and microRNA action. Trends Biochem Sci 2004;29:462–468 13. Ivanovska I, Cleary MA. Combinatorial microRNAs: working together to – Acknowledgments. The authors thank Eva Sjölin, Gaby Åström, make a difference. Cell Cycle 2008;7:3137 3142 Elisabeth Dungner, Kerstin Wåhlén, Britt-Marie Leijonhufvud, Katarina Hertel, 14. Kolaczynski JW, Morales LM, Moore JH Jr, et al. A new technique for and Yvonne Widlund (Department of Medicine, Lipid Laboratory, Karolinska biopsy of human abdominal fat under local anaesthesia with Lidocaine. Int Institutet) for excellent technical assistance. J Obes Relat Metab Disord 1994;18:161–166 Funding. This work was supported by several grants from the Swedish 15. Arvidsson E, Viguerie N, Andersson I, Verdich C, Langin D, Arner P. Effects Research Council, the Swedish Diabetes Foundation, Diabetes Wellness, the of different hypocaloric diets on protein secretion from adipose tissue of Diabetes Program at Karolinska Institutet, the Swedish Society of Medicine, obese women. Diabetes 2004;53:1966–1971 Tore Nilsson Foundation, Foundation for Gamla Tjänarinnor, Åke Wiberg 16. Löfgren P, Hoffstedt J, Näslund E, Wirén M, Arner P. Prospective and Foundation, European Association for the Study of Diabetes/Lilly program, Novo controlled studies of the actions of insulin and catecholamine in fat cells of Nordisk Foundation, and a research grant from the Ministry of Education, obese women following weight reduction. Diabetologia 2005;48:2334– Culture, Sports, and Technology in Japan to the RIKEN Center for Life Science 2342 Technologies. 17. van Harmelen V, Skurk T, Hauner H. Primary culture and differentiation of Duality of Interest. No potential conflicts of interests relevant to this human adipocyte precursor cells. Methods Mol Med 2005;107:125–135 article were reported. 18. Stenson BM, Rydén M, Venteclef N, et al. Liver X receptor (LXR) regulates Author Contributions. A.K. and Y.B. designed the study, performed the human adipocyte lipolysis. J Biol Chem 2011;286:370–379 functional analyses, and wrote the manuscript. S.L.-C. and C.B. performed the functional analyses and wrote the manuscript. E.A. and C.O.D. conducted 19. FANTOM Consortium; Suzuki H, Forrest AR, van Nimwegen E, et al. The microarray data, motif activity response, and network analyses, and wrote the transcriptional network that controls growth arrest and differentiation in manuscript. P.H. collected tissue and wrote the manuscript. M.R. designed the a human myeloid leukemia cell line. Nat Genet 2009;41:553–562 study, collected tissue, and wrote the manuscript. N.M. designed the study; 20. Fain JN. Release of interleukins and other inflammatory cytokines by conducted microarray data, motif activity response, and network analyses; per- human adipose tissue is enhanced in obesity and primarily due to the formed the functional analyses; and wrote the manuscript. P.A. designed the nonfat cells. Vitam Horm 2006;74:443–477 study and wrote the manuscript. All authors contributed to data interpretation and 21. Pitto L, Ripoli A, Cremisi F, Simili M, Rainaldi G. microRNA(interference) reviewed and approved the final manuscript. A.K. and P.A. are the guarantors of networks are embedded in the gene regulatory networks. Cell Cycle 2008; this work and, as such, had full access to all the data in the study and take 7:2458–2461 responsibility for the integrity of the data and the accuracy of the data analysis. 22. Ping D, Boekhoudt G, Zhang F, et al. Sp1 binding is critical for promoter References assembly and activation of the MCP-1 gene by tumor necrosis factor. J Biol – 1. Fain JN. Release of inflammatory mediators by human adipose tissue is Chem 2000;275:1708 1714 enhanced in obesity and primarily by the nonfat cells: a review. Mediators 23. Zhan Y, Brown C, Maynard E, et al. Ets-1 is a critical regulator of Ang II- Inflam 2010;2010:513948. mediated vascular inflammation and remodeling. J Clin Invest 2005;115: 2. Ouchi N, Parker JL, Lugus JJ, Walsh K. Adipokines in inflammation and 2508–2516 metabolic disease. Nat Rev Immunol 2011;11:85–97 24. Dang CV, Resar LM, Emison E, et al. Function of the c-Myc oncogenic 3. Cancello R, Clément K. Is obesity an inflammatory illness? Role of low- transcription factor. Exp Cell Res 1999;253:63–77 grade inflammation and macrophage infiltration in human white adipose 25. Hirano F, Tanaka H, Hirano Y, et al. Functional interference of Sp1 and tissue. BJOG 2006;113:1141–1147 NF-kappaB through the same DNA binding site. Mol Cell Biol 1998;18: 4. Anderson EK, Gutierrez DA, Hasty AH. Adipose tissue recruitment of leu- 1266–1274 kocytes. Curr Opin Lipidol 2010;21:172–177 26. Keller U, Nilsson JA, Maclean KH, Old JB, Cleveland JL. Nfkb 1 is dispensable 5. Deshmane SL, Kremlev S, Amini S, Sawaya BE. Monocyte chemoattractant for Myc-induced lymphomagenesis. Oncogene 2005;24:6231–6240 protein-1 (MCP-1): an overview. J Interferon Cytokine Res 2009;29:313– 27. Lambert PF, Ludford-Menting MJ, Deacon NJ, Kola I, Doherty RR. The 326. nfkb1 promoter is controlled by proteins of the Ets family. Mol Biol Cell 6. Kanda H, Tateya S, Tamori Y, et al. MCP-1 contributes to macrophage 1997;8:313–323 infiltration into adipose tissue, insulin resistance, and hepatic steatosis in 28. Viswanathan M, Yu M, Mendoza L, Yunis JJ. Cloning and transcription obesity. J Clin Invest 2006;116:1494–1505 factor-binding sites of the human c-rel proto-oncogene promoter. Gene 7. Sayed D, Abdellatif M. MicroRNAs in development and disease. Physiol Rev 1996;170:271–276 2011;91:827–887 29. Carbone GM, McGuffie EM, Collier A, Catapano CV. Selective inhibition of 8. Bartel DP. MicroRNAs: genomics, biogenesis, mechanism, and function. transcription of the Ets2 gene in prostate cancer cells by a triplex-forming Cell 2004;116:281–297 oligonucleotide. Nucleic Acids Res 2003;31:833–843