Peroxisome Proliferator-Activated Receptor γ and C/EBPα Synergistically Activate Key Metabolic Adipocyte Genes by Assisted Loading

Abstract

Peroxisome proliferator-activated receptor γ (PPARγ) and CCAAT/enhancer binding protein α (C/EBPα) are key activators of adipogenesis. They mutually induce the expression of each other and have been reported to cooperate in activation of a few adipocyte genes. Recently, genome-wide profiling revealed a high degree of overlap between PPARγ and C/EBPα binding in adipocytes, suggesting that cooperativeness could be mediated through common binding sites. To directly investigate the interplay between PPARγ and C/EBPα at shared binding sites, we established a fibroblastic model system in which PPARγ and C/EBPα can be independently expressed. Using RNA sequencing, we demonstrate that coexpression of PPARγ and C/EBPα leads to synergistic activation of many key metabolic adipocyte genes. This is associated with extensive C/EBPα-mediated reprogramming of PPARγ binding and vice versa in the vicinity of these genes, as determined by chromatin immunoprecipitation combined with deep sequencing. Our results indicate that this is at least partly mediated by assisted loading involving chromatin remodeling directed by the leading factor. In conclusion, we report a novel mechanism by which the key adipogenic transcription factors, PPARγ and C/EBPα, cooperate in activation of the adipocyte gene program.

INTRODUCTION

The conversion of preadipocytes to mature fat-laden adipocytes involves a complex network of transcription factors that have been shown to act in two waves (1–4). The adipogenic cocktail used to stimulate in vitro differentiation of preadipocytes induces the first wave of factors that then subsequently induce the second wave of adipogenic transcription factors. Key components of this second wave are peroxisome proliferator-activated receptor γ (PPARγ) and CCAAT/enhancer binding protein α (C/EBPα), each of which has been shown to directly promote expression of the adipocyte gene program. In particular, several ex vivo and in vivo studies have demonstrated that PPARγ is obligate for adipogenesis (5–7) as well as maintenance of the adipocyte phenotype (8). Similarly, several in vivo models have demonstrated the importance of C/EBPα for differentiation and maintenance of white adipose tissue (9–11). C/EBPα is also important for ex vivo differentiation of adipocytes, although it is not strictly required for adipocyte differentiation in the presence of ectopic PPARγ expression (12). Furthermore, C/EBPα has been reported to be specifically involved in activating the gene program driving insulin-stimulated glucose uptake (13, 14).

It is well established that PPARγ and C/EBPα cooperate to accelerate adipogenesis (12, 15) through their mutual induction of each other (14, 16–18). In addition, at least some target genes have been shown to be activated by both transcription factors (19–25). More recent genome-wide profiling of C/EBPα and PPARγ binding sites using chromatin immunoprecipitation (ChIP) combined with microarray analysis (ChIP-chip) or deep sequencing (ChIP-seq) have shown that most adipocyte genes in murine adipocytes are associated with binding of both factors (26–28), indicating that a hitherto unrecognized high number of genes are regulated by both PPARγ and C/EBPα. Interestingly, there is also a high degree of overlap between the binding sites of the two factors in murine (26, 27) as well as human (29) adipocytes, suggesting that an additional layer of cooperativeness through joint binding to common regions in chromatin may exist. However, the mechanism as well as the functional importance of this overlap in binding is unknown. In support of a functional importance of cobinding of PPARγ and C/EBPα, shared binding sites are more highly retained between mouse and human adipocytes than sites occupied by only one of the factors (29).

Investigation of the molecular mechanism of cooperation between PPARγ and C/EBPα on adipocyte chromatin is complicated by the fact that the two factors induce the expression of each other and the fact that they are both intricately involved in the induction of adipogenesis. The aim of this study was to determine the molecular cross talk between PPARγ and C/EBPα at the chromatin level. To do this, we established a model system in which expression of PPARγ2 and C/EBPα could be independently controlled in PPARγ^−/− mouse embryonic fibroblasts (MEFs). Using this system, we show that many genes in the metabolic gene program in adipocytes are synergistically activated by PPARγ and C/EBPα and that this is at least partially mediated through assisted loading (30) to common binding sites. Thus, C/EBPα dramatically reprograms PPARγ binding and PPARγ reprograms C/EBPα binding, although to a lesser extent, near key metabolic genes. Our results provide the first demonstration that the main adipogenic transcription factors cooperate in gaining access to chromatin and that this cooperativeness is important for activation of key metabolic genes.

MATERIALS AND METHODS

Cell culture and differentiation.

PPARγ^−/− MEF (12), PPARγ^−/− MEF-coxsackievirus-adenovirus receptor (CAR), and Phoenix E cells were cultured in Dulbecco's modified Eagle's medium (DMEM; Gibco) supplemented with 10% fetal bovine serum (Gibco). 3T3-L1 fibroblasts were grown in DMEM supplemented with 10% bovine serum (Gibco) and differentiated to adipocytes in DMEM supplemented with 10% fetal bovine serum (Biochrom AG) by stimulation with dexamethasone, 3-isobutyl-1-methylxanthine, and insulin, as described previously (31).

Generation of PPARγ^−/− MEFs stably transduced with CAR.

Phoenix E cells at 90% confluence were transfected with the LXSN-hCARΔ1 vector (32) expressing the truncated coxsackievirus-adenovirus receptor (CARΔ1) using polyethyleneimine (PEI). At 2 days posttransfection, virus supernatant was harvested and filtered through a 0.45-μm-pore-size filter (Durapore polyvinylidene difluoride membrane; Millex-HV). PPARγ^−/− MEF cells at 50% confluence were transduced with a 1:1 dilution of virus supernatant and fresh growth medium in the presence of 6 μg/ml Polybrene (Sigma-Aldrich). One day following transduction, cells were subjected to puromycin selection for 4 days.

Expression of C/EBPα and PPARγ in PPARγ^−/− MEF-CAR cells.

Retroviruses were produced by transfection of Phoenix E cells with pBABEpuro-C/EBPα, pBABEpuro-C/EBPα ΔTE-III (33), or empty pBABEpuro vector as a control, as described above. PPARγ^−/− MEF-CAR cells were seeded in medium containing a 1:1 dilution of virus supernatant and fresh growth medium in the presence of 6 μg/ml Polybrene (Sigma-Aldrich). Twenty-four hours after retroviral transduction, PPARγ^−/− MEF-CAR cells were incubated in medium containing AdHA-PPARγ2 (an adenoviral [Ad] vector carrying hemagglutinin [HA] and PPARγ2) (34), AdHA-PPARγCDE (an adenoviral vector carrying an HA tag and an A/B domain-truncated version of PPARγ2), or AdEmpty (empty adenoviral vector) (35) for 2 h. Here, the medium was exchanged with medium supplemented with the vehicle (dimethyl sulfoxide [DMSO]; Sigma) or 1 μM rosiglitazone (Rosi; Novo Nordisk A/S, Denmark). Cells were harvested after another 6 h. The experimental procedure is outlined in Fig. 2A.

FIG 2.

Independent expression of PPARγ and C/EBPα in PPARγ^−/− MEF-CAR cells synergistically activates adipocyte genes through facilitated binding to shared sites. (A) Experimental outline showing the timing of virus transduction and harvest of PPARγ^−/− MEF-CAR cells. (B) Western blot demonstrating the independent expression of HA-PPARγ2 and C/EBPα. TFIIB was used as a loading control. (C and D) Timing of expression of C/EBPα, PPARγ, and their target genes following viral transduction. (Top) Western blots displaying protein expression of C/EBPα and PPARγ, respectively, and the loading control, TFIIB; (bottom) mRNA expression of the C/EBPα target gene Cd36 and of the PPARγ target gene Fabp4 relative to the corresponding expression of Gtf2b mRNA. (E) Endogenous C/EBPα expression is low and not affected by ectopic PPARγ expression. The Western blot shows the expression of C/EBPα in control PPARγ^−/− MEF-CAR cells and cells transduced with AdHA-PPARγ2 and/or pBabe-C/EBPα. TFIIB was used as a loading control. (F) Endogenous C/EBPβ expression in PPARγ^−/− MEF-CAR cells is not affected by ectopic PPARγ and/or C/EBPα expression and is comparable to that in undifferentiated 3T3-L1 cells. The Western blot shows the expression of the isoforms of C/EBPβ in PPARγ^−/− MEF-CAR cells and 3T3-L1 cells at the indicated time points of differentiation. TFIIB was used as a loading control. (G) Key metabolic adipocyte genes are synergistically activated by PPARγ-C/EBPα coexpression. The mRNA levels of selected adipocyte genes were measured by qPCR and normalized to the corresponding level of Gtf2b mRNA. Synergy factors, defined as the target gene level upon coexpression of PPARγ and C/EBPα relative to the sum of the target gene levels upon individual expression of PPARγ and C/EBPα with the background subtracted, are indicated. Error bars indicate the SEMs of triplicates. Results are representative of four independent experiments. ND, not detectable. (H and I) Facilitated transcription factor recruitment is common at shared sites. C/EBPα (H) and PPARγ (I) ChIP-qPCR for selected shared sites in the vicinity of synergistically regulated genes. Binding to a negative-control region, denoted “No gene,” was used as a control. p.p., proximal promoter. Bars represent the means of three independent experiments + SEMs. FDR-adjusted P values were <0.05 (*), <0.01 (**), and <0.001 (***) (one-way analysis of variance followed by Tukey's honestly significant difference test).

Western blotting and ECL detection.

Cells were harvested in hypotonic SDS buffer (36) and subjected to Western blotting as previously described (34). Membranes were probed with the primary antibodies anti-HA (12CA5 [37]), anti-C/EBPα (catalog no. sc-61; Santa Cruz), anti-C/EBPβ (catalog no. sc-150; Santa Cruz), and anti-TFIIB (catalog no. sc-225; Santa Cruz). The secondary antibodies used for enhanced chemiluminescence (ECL; Amersham Pharmacia Biotech) detection were horseradish peroxidase-conjugated goat anti-mouse immunoglobulin G (IgG; catalog no. P0447; Dako) and swine anti-rabbit IgG (catalog no. P0339; Dako).

ChIP and ChIP-seq.

ChIP was performed as described previously (38), except that di-(N-succinimidyl) glutarate (DSG) cross-linking was performed for 20 min and chromatin samples were sonicated at the maximum level using a Diagenode Biorupter twin apparatus for 3 sets of 16 cycles of 30 s on/30 s off. Antibodies directed against PPARγ (catalog no. sc-7196; Santa Cruz), C/EBPα (catalog no. sc-61; Santa Cruz), mediator subunit 1 (MED1; catalog no. sc-8998; Santa Cruz), CREB-binding protein (CBP; catalog no. sc-369; Santa Cruz), and Brahma-related gene 1 (BRG1; catalog no. 2822-1; Epitomics) were used for immunoprecipitation. For ChIP-seq, the ChIP reactions were scaled to gain a minimum of 10 ng DNA. DNA that had undergone ChIP was analyzed by quantitative real-time PCR (qPCR) or prepared for sequencing according to the manufacturer's protocol (Illumina).

Sequential ChIP (ChIP-reChIP).

Chromatin from mature 3T3-L1 adipocytes was prepared as previously described (38), except that DSG cross-linking was omitted and samples were sonicated at the maximum level using the Diagnode Biorupter twin apparatus for 5 sets of 8 cycles of 30 s on/30 s off. The first ChIP reaction was performed as previously described (38) using anti-PPARγ (catalog no. sc-7196; Santa Cruz) antibody, and the chromatin was eluted in 1% SDS and 10 mM dithiothreitol for 30 min at 37°C with shaking at 800 rpm. The supernatant was diluted 60 times total in reChIP buffer (1% Triton X-100, 0.15 M NaCl, 1 mM EDTA, 20 mM Tris pH 8.0, 1× Complete protease inhibitor cocktail [Boehringer Mannheim]), and the final SDS concentration was adjusted to 0.1%. Ten percent of the supernatant was saved as input for the second ChIP. The remaining supernatant was used for the second ChIP reaction using anti-C/EBPα (catalog no. sc-61; Santa Cruz) or normal rabbit IgG antibody (catalog no. sc-2027; Santa Cruz) using the standard ChIP protocol, as described previously (38). DNA that had undergone ChIP was analyzed by qPCR.

FAIRE and FAIRE-seq.

Chromatin from PPARγ^−/− MEF-CAR cells was prepared as described above for ChIP, except that chromatin was cross-linked using only 1% formaldehyde. Open DNA regions were isolated by three successive extractions using 1 volume of cold phenol-chloroform-isoamyl alcohol (PCI; 25:24:1) saturated with 10 mM Tris, pH 8.0, 1 mM EDTA (Sigma), mixing by inverting the tube 15 times, and separating the phases by spinning at 15,000 × g for 5 min at 4°C. Extracted DNA was purified as chromatin immunoprecipitated DNA. DNA obtained by formaldehyde-assisted isolation of regulatory elements (FAIRE) analysis was analyzed by qPCR or sequenced (FAIRE-seq) according to the instructions of the manufacturer (Illumina).

ChIP-seq data analysis.

Sequence reads from each ChIP-seq library were trimmed for bar code sequences, resulting in sequences of 36 bases. Each library was aligned to the genome (mm9) using the Bowtie program (39) with the following parameters: –best, –strata, and –m 3. The HOMER program (40) was used to identify binding sites using a false discovery rate (FDR) of ≤0.01%. Only peaks with a ≥4-fold tag-count enrichment over the input and local background (a 20-kb region centered at the putative peak) were used for further analysis. HOMER was used to generate a single peak file containing all peaks, where peaks within 200 bp were merged. The number of ChIP-seq reads in a 250-bp window at each peak was determined for each ChIP-seq data set using HOMER. Peaks with ≥20 tags and ≥4-fold tag count enrichment over input were called transcription factor binding sites. We defined gained and lost sites as those having a ≥2.5-fold increase or decrease in tag count, respectively, in cells coexpressing the two transcription factors relative to that in cells expressing only one factor. The counts for constitutive sites changed less than 2.5-fold.

Motif discovery analysis.

A 250-bp region at the top 1,000 (by tag density) constitutive PPARγ and C/EBPα binding sites was used for motif analysis using the MEME algorithm (41). The distribution of motifs in different groups of binding sites was determined by HOMER using log₂ odds detection thresholds of 5 and 7 for the PPARγ motif and C/EBPα motif, respectively.

RNA extraction, cDNA synthesis, and qPCR.

RNA extraction, cDNA synthesis, and qPCR were performed as previously described (42). The primer sequences used for qPCR are available upon request.

RNA-seq.

RNA from two independent experiments was prepared in triplicate. Technical triplicates were pooled after qPCR validation. RNA was prepared for sequencing according to the manufacturer's instructions (Illumina). Forty-mer RNA sequencing (RNA-seq) libraries were mapped to the mm9 genome and a pseudogenome containing all exon-exon junctions using Bowtie with the parameters –best, –strata, and –m 1. Tags in exons were counted using HOMER, and significantly regulated genes (FDR-adjusted P value, ≤0.05) were identified using the DESeq program (43). Induced gene groups were defined as described below: (i) synergistically activated genes had values in both biological replicates of [(PPARγ + C/EBPα)/control] ≥ 1.5 × {[(PPARγ/control) + (C/EBPα/control)] − 1}, where PPARγ indicates gene activation in cells expressing PPARγ, C/EBPα indicates gene activation in cells expressing C/EBPα, PPARγ + C/EBPα indicates gene expression in cells coexpressing the two transcription factors, and control indicates control gene expression; (ii) dually activated genes were significantly but not synergistically upregulated relative to control under all three conditions; (iii) genes upregulated by PPARγ only were significantly upregulated under conditions where PPARγ2 was expressed but not synergistically upregulated or significantly upregulated by C/EBPα alone; and (iv) genes upregulated by C/EBPα only were defined by criteria similar to those for genes upregulated by PPARγ only.

PPARγ- and C/EBPα-downregulated genes were defined as those genes that were significantly downregulated under both conditions where the factor was expressed. Constitutively expressed genes were defined as genes where the tag count was within ±5% of that for control cells under each condition.

Enrichment analysis of binding sites.

The number of gained, lost, and constitutive binding sites within 50 kb of the transcription start site (TSS) of regulated genes relative to the number for constitutive genes was determined using the BEDtools program (44).

RESULTS

C/EBPα and PPARγ cooccupy chromatin in mature 3T3-L1 adipocytes.

From the previously reported overlap between PPARγ and C/EBPα binding sites in adipocytes (26, 27, 29), it was not possible to determine whether the two factors bind simultaneously or in an alternating manner to the shared sites in the genome. To investigate this, we performed sequential ChIP analysis (ChIP-reChIP) on chromatin from mature 3T3-L1 adipocytes (day 6 of differentiation), analyzing five shared binding sites and two control sites to which only PPARγ or C/EBPα binds (Fig. 1). These data show that PPARγ and C/EBPα bind simultaneously to the selected shared sites in mature 3T3-L1 adipocytes but not to the control sites. This prompted us to further investigate the interplay between the two transcription factors at common binding sites.

FIG 1.

PPARγ and C/EBPα are simultaneously recruited to shared PPARγ-C/EBPα sites in mature 3T3-L1 adipocytes. 3T3-L1 cells were differentiated to mature adipocytes (day 6), and chromatin was subjected to PPARγ-C/EBPα and PPARγ-IgG ChIP-reChIP–PCR for selected shared PPARγ-C/EBPα sites, as well as for one PPARγ-only and one C/EBPα-only site (A), as determined by PPARγ ChIP-PCR (B) and C/EBPα ChIP-PCR (C). No gene, a negative-control region not occupied by any of the two transcription factors. p.p., proximal promoter. Bars represent the means plus SEMs of three independent experiments. FDR-adjusted P values were <0.05 (*), <0.01 (**), and <0.001 (***) for PPARγ-C/EBPα reChIP versus PPARγ-IgG reChIP (one-way analysis of variance followed by Tukey's honestly significant difference test). ns, not significant.

PPARγ2 and C/EBPα synergistically activate adipocyte genes.

Since PPARγ and C/EBPα mutually induce the expression of each other during adipogenesis, it is intrinsically difficult to study their cooperativeness on chromatin in adipocytes. To overcome this problem, we generated a model system based on PPARγ^−/− mouse embryonic fibroblasts (MEFs) (12), where we could independently control the expression of the two transcription factors (Fig. 2A and B). MEFs were first stably transduced with the truncated coxsackievirus-adenovirus receptor (CARΔ1) to facilitate adenoviral uptake (32, 34). C/EBPα was then introduced using a retroviral expression construct, leading to robust induction of C/EBPα and the C/EBPα target gene Cd36 at 24 h following transduction (Fig. 2C). PPARγ2 was introduced using an adenoviral vector leading to robust expression of PPARγ2 and its target gene, Fabp4, after 4 h (Fig. 2D). Based on these time scales of C/EBPα and PPARγ2 expression, we chose the experimental setup outlined in Fig. 2A, which results in short exposure to PPARγ2 and C/EBPα expression (4 and 8 to 12 h, respectively), thereby ensuring that we primarily investigated the direct transcriptional effects of these factors. Importantly, the PPARγ^−/− MEFs had extremely low endogenous expression of the C/EBPα protein that was unaffected by ectopic PPARγ2 expression (Fig. 2E). Furthermore, PPARγ2 and C/EBPα expression did not affect the levels of the other major adipocyte C/EBP protein, C/EBPβ, which was expressed at a level comparable to that found in undifferentiated 3T3-L1 cells (Fig. 2F).

We used this cell system to investigate the individual and combined transactivation potential of the two transcription factors at selected adipocyte genes. Interestingly, our data show that coexpression of PPARγ and C/EBPα leads to a highly synergistic activation of several adipocyte genes (Fig. 2G). The synergy factor, defined as the target gene level upon coexpression of PPARγ and C/EBPα relative to the sum of the target gene levels upon individual expression of PPARγ and C/EBPα with the background subtracted, is indicated. The investigated genes displayed different patterns of synergy between PPARγ2 and C/EBPα. Expression of the genes Cidec (encoding cell death activator CIDE-3) and Plin1 (encoding perilipin1) was highly PPARγ dependent, and expression of Lpl mRNA (encoding lipoprotein lipase) could be induced by both C/EBPα and PPARγ, whereas expression of Cd36 (encoding cluster of differentiation 36) and Acsl1 (encoding acyl coenzyme A [acyl-CoA] synthetase long-chain family member 1) in the absence of the PPARγ agonist rosiglitazone was C/EBPα dependent. For all five genes, transcription factor coexpression led to highly synergistic gene activation (synergy factor > 1.5).

PPARγ and C/EBPα mutually potentiate the binding of each other to shared sites.

To investigate if synergistic gene activation is associated with cooperative binding of PPARγ and C/EBPα to chromatin, we examined PPARγ and C/EBPα binding to shared sites in the vicinity of the synergistically activated target genes. This showed that whereas C/EBPα did not occupy the proximal promoter of Cidec or its upstream enhancer when expressed alone, coexpression with PPARγ led to significant recruitment of C/EBPα to these sites (Fig. 2H). This correlates well with the finding that C/EBPα contributes to transactivation of the Cidec gene only in the presence of PPARγ (Fig. 2G). In contrast, PPARγ was able to bind to both the proximal promoter and enhancer of Cidec in the absence of C/EBPα (Fig. 2I), consistent with its ability to transactivate Cidec alone. Other sites investigated were characterized by C/EBPα-facilitated PPARγ recruitment (6.2 kb downstream from the Lpl TSS [Lpl + 6.2 kb], Acsl1 intron1, and Acsl1 − 450 bp) (Fig. 2I). These data suggest that cooperative binding of PPARγ and C/EBPα contributes to synergistic gene activation and that at some sites one transcription factor assists the loading of the other transcription factor.

To further investigate the cooperative binding of PPARγ and C/EBPα to chromatin, we determined the genome-wide binding profiles of these factors using ChIP-seq. We identified a total of 33,660 PPARγ binding sites (Fig. 3A, top) and 25,525 C/EBPα binding sites (Fig. 3A, bottom). Binding to the majority of binding sites for both factors was independent of coexpression of the other factor (i.e., 21,724 and 22,695 constitutive sites for PPARγ and C/EBPα, respectively). Interestingly however, we also identified two groups of sites for each factor where the binding intensity increased or decreased by ≥2.5-fold upon coexpression of the other factor (i.e., 7,997 gained and 3,939 lost sites for PPARγ and 1,818 gained and 1,012 lost sites for C/EBPα). This demonstrates that expression of C/EBPα induces extensive reprogramming of PPARγ binding, whereas the effect of PPARγ expression on C/EBPα binding is more modest. We next asked to what extent constitutive (C/EBPα-independent), gained (C/EBPα-dependent), and lost PPARγ binding sites overlap with C/EBPα binding (Fig. 3B, top). Notably, gained PPARγ binding was highly associated with C/EBPα binding (79% overlap), whereas C/EBPα binding was much less pronounced at constitutive and lost PPARγ binding sites (15% and 1% overlap, respectively). This indicates that C/EBPα is directly involved in the recruitment of PPARγ to gained sites. Interestingly, C/EBPα binding to most of the gained PPARγ sites (73%) was constitutive (independent of PPARγ), whereas only a smaller fraction of the sites gained both PPARγ and C/EBPα binding upon coexpression (6%), indicating that C/EBPα acts as a pioneer factor for PPARγ at most gained PPARγ sites.

FIG 3.

C/EBPα extensively reprograms PPARγ recruitment by priming for binding at shared sites. (A) C/EBPα induces extensive reprogramming of PPARγ binding upon coexpression. Heat maps illustrate PPARγ (top) and C/EBPα (bottom) binding intensity in a 2-kb window around the peak centers in control cells (top and bottom) and cells expressing PPARγ2 and treated with rosiglitazone (top), C/EBPα (bottom), and PPARγ2 and C/EBPα plus Rosi (top and bottom). Groups of gained and lost sites were defined by having a binding intensity increasing or decreasing by ≥2.5-fold upon coexpression, respectively. Constitutive sites had a binding intensity increasing or decreasing <2.5-fold upon transcription factor coexpression. (B) The majority of gained sites bind both transcription factors. Pie charts indicating the percentage of constitutive (Const.), gained, and lost sites of one transcription factor overlapping the constitutive, gained, lost, and no binding of the other transcription factor.

The number of gained (PPARγ-dependent) C/EBPα binding sites (1,818) is much lower than the number of gained (C/EBPα-dependent) PPARγ binding sites (7,997). However, similar to what was found for PPARγ, a large fraction (62%) of the gained C/EBPα binding sites overlapped with PPARγ binding sites, indicating that PPARγ is directly involved in the recruitment of C/EBPα to gained sites. Intriguingly, however, this overlap contained a large fraction of gained PPARγ binding sites (Fig. 3B, bottom), showing that many sites (490) are sites where both PPARγ and C/EBPα are highly dependent on each other for binding. Cumulatively, the data presented here show that C/EBPα directs PPARγ binding to new sites in the genome by directly targeting these sites itself, and PPARγ also reprograms C/EBPα binding in a similar way, although to a lesser extent.

Gained PPARγ and C/EBPα binding sites are enriched for C/EBP and PPARγ motifs, respectively.

Facilitated recruitment can be the result of direct binding to DNA via assisted loading (30) to an adjacent element or of indirect recruitment mediated by protein-protein interactions (Fig. 4A). To explore which of these possibilities is the most prevailing, we used MEME (41) to identify transcription factor motifs at the 1,000 most intense constitutive PPARγ and C/EBPα binding sites. Reassuringly, we found the consensus motifs previously reported for PPARγ:RXR (i.e., a direct repeat 1 [DR-1]) (26, 27, 45) and C/EBP (46) (Fig. 4B and C). Interestingly, the total fraction of PPARγ binding sites having a PPARγ motif was comparable between constitutive, gained, and lost PPARγ binding sites; however, gained PPARγ sites were highly enriched for the C/EBP consensus motif and many sites showed cooccurrence of both motifs (36%) (Fig. 4D, left). These data are consistent with the results from ChIP-seq and indicate that C/EBPα-facilitated binding of PPARγ is primarily associated with bona fide C/EBP binding sites and also, in many cases, an adjacent PPARγ motif. This suggests that at several sites, C/EBPα facilitates PPARγ binding through a mechanism that involves direct contact of both factors with DNA. Notably, however, other gained PPARγ sites displayed only the C/EBP motif and not a PPARγ motif (42%), suggesting that either PPARγ binds indirectly to these sites or C/EBPα or other transcription factors are able to relax the sequence specificity of PPARγ.

FIG 4.

Gained binding sites for PPARγ are enriched for the C/EBP motif and vice versa. (A) Model illustrating C/EBPα-facilitated PPARγ binding through either direct binding by assisted loading (top) or indirect recruitment (bottom). (B and C) De novo motif analysis using the MEME algorithm on the 1,000 most intense PPARγ (B) and C/EBPα (C) binding sites identified the PPARγ and C/EBP consensus motifs, respectively. (D) Distribution of C/EBP and PPARγ consensus motifs at constitutive, gained, and lost PPARγ and C/EBPα binding sites.

Similar to what we observed for PPARγ, gained C/EBPα sites were enriched for the PPARγ motif (Fig. 4D, right), and we identified sites that contained both motifs as well as sites with only a PPARγ motif. This indicates that PPARγ can also facilitate binding of C/EBPα through an assisted loading mechanism involving direct binding of both factors to DNA.

Several key metabolic adipocyte genes are synergistically activated by PPARγ-C/EBPα coexpression.

To correlate the cooperative binding of PPARγ and C/EBPα with global gene expression, we performed RNA-seq in two independent experiments. Significantly induced genes were identified using DESeq (43) and grouped into four groups: genes activated only by C/EBPα (212 genes), genes activated only by PPARγ (685 genes), genes activated by both transcription factors in a nonsynergistic manner (84 genes, synergy factor < 1.5), and genes synergistically activated by coexpression (41 genes, synergy factor ≥ 1.5) (Fig. 5A) (for detailed definitions of the gene groups, see Materials and Methods). Examples of genes from each category are shown in Fig. 5B. Interestingly, the largest fold change in gene expression was seen when the two transcription factors cooperated in gene activation rather than when genes were activated only by one transcription factor, and synergistically activated genes were more highly induced than dually activated genes (Fig. 5A). Notably, the synergistically activated genes are also among the most highly induced genes during adipogenesis (Fig. 6A), indicating that cooperation between the two master regulators PPARγ and C/EBPα has an important impact on the adipocyte gene program. In addition to the induced genes described above, we also identified two groups of genes repressed by PPARγ and C/EBPα, respectively, and these genes were also significantly downregulated during adipogenesis (Fig. 6A).

FIG 5.

Synergistically activated target genes are key metabolic adipocyte genes enriched for nearby gained binding sites. (A) RNA-seq data from two independent replicates were analyzed for the tag count within exons. Significantly (FDR-adjusted P value, ≤0.05) induced genes were identified using DESeq (43) and grouped into four groups: genes activated only by C/EBPα, genes activated only by PPARγ, dually activated genes (synergy factor < 1.5), and genes synergistically activated by transcription factor coexpression (synergy factor ≥ 1.5). Box plots illustrate the fold change in tag count relative to that for control cells for each group of upregulated genes. (B) Examples of RNA-seq data and PPARγ- and C/EBPα ChIP-seq data for genes from each group of upregulated genes viewed in the UCSC genome browser (68). chr, chromosome. (C) Enrichment of binding sites within 50 kb of the TSS of genes in each gene group is viewed relative to the number of binding sites in the vicinity of constitutively expressed genes.

FIG 6.

PPARγ and C/EBPα activate adipocyte genes and bind to adipocyte regulatory sites in the vicinity of these genes in MEFs. (A) Transcriptional changes during 3T3-L1 adipogenesis, as measured by changes in RNA polymerase II (RNAPII) occupancy between day 0 and day 4 of differentiation (27), were determined for each group of target genes. *, genes significantly up- or downregulated during adipogenesis, determined using the Wilcoxon rank sum test (for target genes activated by PPARγ only, P = 2.2 × 10⁻¹¹; for dually activated target genes, P = 6.8 × 10⁻³; for synergistically activated [act.] target genes, P = 8.9 × 10⁻⁴; for PPARγ-downregulated (PPARγ down) target genes, P = 2.2 × 10⁻⁹; for C/EBPα-downregulated (C/EBPα down) target genes, P = 2.9 × 10⁻⁵). (B) Overlap between binding sites for PPARγ and C/EBPα in MEF and 3T3-L1 cells. The percentage of PPARγ and C/EBPα binding sites in MEF cells found within 50 kb of the TSS of constitutive genes, genes activated only by C/EBPα or PPARγ, dually activated genes, and synergistically activated genes that overlap a binding site in 3T3-L1 cells (47) is shown. (C) Model illustrating key metabolic pathways within the adipocyte. Proteins encoded by the synergistically activated genes and the pathway that they are involved in are highlighted in red. BCAA, branched-chain amino acid; FA, fatty acid; FFA, free fatty acid; GSV, GLUT4 storage vesicle; OAA, oxaloacetate; oxLDL, oxidized low-density lipoprotein; PPP, pentose phosphate pathway; TAG, triacylglycerol; TCA, tricarboxylic acid (citric acid) cycle; VLDL, very-low-density lipoprotein.

To investigate the extent to which the PPARγ and C/EBPα binding sites identified by ectopic expression in MEFs recapitulate binding sites in mature adipocytes, we compared binding sites from this study with our previously published binding sites from mature 3T3-L1 adipocytes (47). This showed that whereas only approximately 18% of all PPARγ and 34% of all C/EBPα binding sites identified in MEFs overlapped a binding site in 3T3-L1 adipocytes, the overlap near synergistically activated genes was much higher (approximately 42% and 54%, respectively) (Fig. 6B). Thus, many PPARγ and C/EBPα binding sites near synergistically activated genes are conserved between the fibroblastic model system and 3T3-L1 adipocytes.

Cumulatively, these findings demonstrate that although we used a fibroblastic model system to study the interplay between PPARγ and C/EBPα, the binding sites and target genes resembled those that can be found in in vitro-differentiated adipocytes.

The group of synergistically activated genes (see Table S1 in the supplemental material) comprises many metabolic genes not previously shown to be cooperatively induced by C/EBPα and PPARγ (Fig. 6C). This included genes encoding proteins known to play a role in fatty acid uptake (CD36 and LPL), activation of free fatty acids (ACSL1), triglyceride synthesis (glycerol kinase [GyK], glycerol-3-phosphate acetyltransferase 3 [GPAT3], and diacylglycerol O-acyltransferase 1 [DGAT1]), fatty acid oxidation (acyl-CoA thioesterase 2 [ACOT2]), cholesterol uptake (oxidized low-density lipoprotein [LDL] receptor 1 [OLR1]), branched-chain amino acid catabolism (branched-chain α-keto acid dehydrogenase subunit β [BCKDHβ]), the pentose phosphate pathway (transaldolase 1 [Taldo1]), transport of pyruvate across the mitochondrial membrane (mitochondrial pyruvate carrier 2 [MPC2]), carboxylation of pyruvate to enter lipogenesis or the tricarboxylic acid (TCA) cycle by pyruvate carboxylase (PCX), and regulation of glucose uptake (sortilin1 [SORT1]). Moreover, several lipid droplet-coating proteins regulating lipid droplet formation and lipolysis (i.e., PLIN1, CIDEC, and comparative gene identification 58 [CGI-58]) were represented in this gene group. Finally, expression of the nuclear receptor liver X receptor α (LXRα), which is known to induce expression of several lipogenic genes (48), was synergistically upregulated. These data demonstrate that synergy between PPARγ and C/EBPα is of great importance for activation of the adipocyte gene program.

Synergistically activated genes are associated with gained binding sites.

To investigate whether synergistic gene activation is correlated with cooperative binding of PPARγ and C/EBPα to chromatin, we determined the enrichment of constitutive, gained, and lost PPARγ and C/EBPα binding sites in the vicinity (±50 kb of the TSS) of the four groups of upregulated genes relative to constitutively expressed genes (constitutively expressed genes are defined as genes where the change relative to control cells is within ±5% for all three conditions). Interestingly, both gained PPARγ binding sites and gained C/EBPα binding sites were highly enriched near synergistically activated genes (Fig. 5C), indicating that synergistic gene activation is mediated in part by mutually facilitated recruitment of PPARγ and C/EBPα to target sites. In contrast, genes that were activated by both transcription factors but in a nonsynergistic manner had lower levels of enrichment of gained binding sites, indicating that these genes, although they are activated by both PPARγ and C/EBPα, may not be similarly dependent on facilitated recruitment of these transcription factors. It is possible that recruitment of PPARγ and C/EBPα to sites near these genes is facilitated by other factors. Unexpectedly, genes targeted by C/EBPα only were also enriched for gained PPARγ binding, and genes targeted by PPARγ only were also enriched for gained C/EBPα binding, suggesting that coexpression in some cases facilitates binding without affecting transactivation of nearby genes. It is possible that synergistic activation of these genes requires binding of additional factors not expressed in our fibroblast model system to the nearby enhancers. However, we cannot exclude the possibility that these enhancers are more important for the regulation of distant genes, which would not be picked up by this type of analysis based exclusively on proximity.

PPARγ and C/EBPα cooperate in cofactor recruitment and chromatin remodeling.

To investigate how cofactor recruitment is affected at facilitated binding sites, we performed ChIP-qPCR on three cofactors, mediator subunit 1 (MED1), the histone acetyltransferase CREB-binding protein (CBP), and the catalytic subunit of the SWI/SNF remodeling complex Brahma-related gene 1 (BRG1). This demonstrated a highly cooperative recruitment of MED1 and CBP at C/EBPα-facilitated PPARγ binding sites (Fig. 7A). In contrast, cooperative recruitment of MED1 and CBP was seen at only one of three tested PPARγ-facilitated C/EBPα binding sites (Lpl + 73 kb), whereas recruitment to the two other sites (Nr1h3 + 86 kb; Cidec − 3.1 kb) was independent of C/EBPα. The finding that recruitment of MED1 to these sites primarily depends on PPARγ is consistent with our recent report demonstrating that changes in MED1 occupancy correlate better with changes in PPARγ occupancy than with changes in C/EBPα occupancy (47).

FIG 7.

PPARγ and C/EBPα cooperate in cofactor recruitment and chromatin remodeling. (A) Chromatin from PPARγ^−/− MEF-CAR cells was subjected to MED1 and CBP ChIP-PCR for three PPARγ-facilitated C/EBPα binding sites and three C/EBPα-facilitated PPARγ binding sites, all displaying both the PPARγ and C/EBP consensus motifs. Binding to a negative-control (NC) region, denoted “No gene,” was used as a control. Results are represented as the means of three independent experiments + SEMs. FDR-adjusted P values were <0.05 (*), <0.01 (**), and <0.001 (***) versus the control and <0.05 (^), < 0.01(^^), and <0.001 (^^^) (one-way analysis of variance followed by Tukey's honestly significant difference test). ns, not significant. (B) Recruitment of BRG1 to shared sites follows the leading transcription factor. BRG1 ChIP-PCR was performed as described for panel A. Bars represent the means of three independent experiments + SEMs. FDR-adjusted P values were <0.05 (*), <0.01 (**), and <0.001 (***) versus the control (one-way analysis of variance followed by Tukey's test). ns, not significant. (C) Coexpression of PPARγ and C/EBPα leads to cooperative remodeling of chromatin at shared sites. Chromatin from PPARγ^−/− MEF-CAR cells was subjected to FAIRE analysis, and DNA recovery was measured using qPCR for a few selected sites, as described for panel A. The β-globin promoter was used as a negative-control region. Error bars indicate the SEMs of triplicates. Results are representative of four independent experiments.

Contrary to the findings for MED1 and CBP, recruitment of the remodeling factor BRG1 was not cooperative but followed that of the leading transcription factor, i.e., C/EBPα at C/EBPα-facilitated PPARγ binding sites and PPARγ at PPARγ-facilitated C/EBPα binding sites (Fig. 7B). This indicates that recruitment of BRG1 is primarily dependent on the leading transcription factor.

To investigate the role of chromatin remodeling in facilitated binding, we performed formaldehyde-assisted isolation of regulatory elements (FAIRE) analysis (49, 50) of shared sites having both the PPARγ and the C/EBPα motifs. Interestingly, we found that at sites where PPARγ facilitated C/EBPα binding, an open chromatin structure was induced by PPARγ expression and not C/EBPα expression (Nr1h3 + 86 kb, Cidec − 3.1 kb, and, to a lesser extent, Lpl + 73 kb) (Fig. 7C). This suggests that PPARγ is required for chromatin remodeling at these sites, explaining the dependency of PPARγ for C/EBPα binding. At sites where C/EBPα facilitated PPARγ binding (Acsl1 intron 1, Dgat1 + 3 kb, and Cdo1 − 17.5 kb), chromatin openness was induced only by C/EBPα expression, explaining the C/EBPα dependency for PPARγ binding. This is consistent with the assisted loading model (30) and suggests that C/EBPα facilitates PPARγ binding to gained sites where both motifs are present by increasing the accessibility of the PPARγ motif and vice versa.

Notably, although BRG1 recruitment was determined only by the leading transcription factor at the binding sites investigated (Fig. 7B), coexpression cooperatively enhanced chromatin openness at all shared sites. This might be an effect of recruitment of other remodeling factors or of the fact that facilitated recruitment of transcription factors (Fig. 2 and 3) and other cofactors (e.g., MED1 and CBP) (Fig. 7B) forces the chromatin to remain open due to increased activity at the genomic location.

C/EBPα-PPARγ cooperativeness is determined by chromatin structure, motif strength, and distance between motifs.

One intriguing question is why PPARγ and C/EBPα facilitate the binding of each other at some joint binding sites but not at others. To investigate whether chromatin structure is an important predictor of facilitated binding events, we used FAIRE-seq and FAIRE-qPCR to determine the baseline chromatin accessibility in nontreated (control) cells at gained and constitutive C/EBPα and PPARγ binding sites, all containing consensus motifs for both factors. This revealed that gained binding sites are characterized by a closed chromatin structure prior to expression of C/EBPα or PPARγ, whereas constitutive sites, where C/EBPα and PPARγ can bind independently of each other, are significantly more open (Fig. 8A and B). This indicates that cooperativeness becomes important for gaining access to DNA if the chromatin structure is more closed. Furthermore, the distance between the C/EBPα and PPARγ motifs is significantly shorter (on average, about 10 bp shorter) for sites where both factors facilitate binding of each other (Fig. 8C), indicating that the short distance between DNA binding at these sites facilitates cooperation between the two factors. Interestingly, motif analysis showed that compared to constitutive sites, gained sites generally have stronger motifs for the leading factor and weaker motifs for the other factor (Fig. 8D). For example, the motif score of PPARγ motifs at gained C/EBPα sites, where PPARγ facilitates C/EBPα binding, was higher than that at constitutive and gained PPARγ sites.

FIG 8.

C/EBPα-PPARγ cooperativeness is determined by chromatin structure, motif strength, and distance between motifs. (A and B) Gained C/EBPα and PPARγ binding sites are characterized by closed chromatin in untreated cells. (A) Average tag count per bp of FAIRE-seq data from untreated PPARγ^−/− MEF-CAR cells in a 4-kb window centered on constitutive and gained C/EBPα and PPARγ binding sites with both C/EBPα and PPARγ motifs. (B) FAIRE-qPCR on randomly selected constitutive shared sites as well as on gained PPARγ and/or C/EBPα sites with both motifs. Recovery at gained sites was significantly different from that at constitutive sites (P = 0.007, Wilcoxon rank sum test, n = 10). (C) Reduced distance between C/EBPα and PPARγ motifs at sites with both gained C/EBPα and PPARγ binding. →, constitutive binding; ↑ gained binding; *, P = 2.044 ×10⁻⁵ versus constitutive C/EBPα and PPARγ (Wilcoxon rank sum test). (D) Gained C/EBPα and PPARγ binding sites are enriched for high-confidence motifs of the leading factor. →, constitutive binding; ↑, gained binding.

Taken together, these data indicate that when the binding sequences are in a closed chromatin conformation and closely spaced, PPARγ and C/EBPα are more likely to bind in a cooperative manner. Generally, C/EBPα is more likely to act as the leading factor than PPARγ; however, for both PPARγ and C/EBPα, the stronger the motif score, the more likely it is that the factor will act as the leading factor at closed chromatin regions.

Synergy between PPARγ and C/EBPα is independent of the PPARγ A/B domain.

We have previously highlighted the importance of the N-terminal A/B domain of PPARγ2 for transcriptional activity and lipogenesis (35, 51), and we therefore compared the ability of full-length PPARγ2 and an A/B domain-truncated version of PPARγ (PPARγCDE) to synergistically activate target genes together with C/EBPα (Fig. 9). As previously reported (35), deletion of the A/B domain reduces the transactivation capacity, but interestingly, the synergistic activation of target genes is maintained (Fig. 9C). This demonstrates that although the A/B domain is important for transactivation of many adipocyte genes, it is not critical for the synergy between PPARγ2 and C/EBPα.

FIG 9.

The A/B domain of PPARγ2 is not required for synergistic gene activation by PPARγ and C/EBPα. (A) Schematic representation of PPARγ2 with the indicated domains and PPARγ from which the A/B domain was deleted (PPARγCDE). (B) Western blot displaying equal expression of HA-PPARγ2 and HA-PPARγCDE. TFIIB was used as a loading control. (C) Synergistic expression of adipocyte genes is not impaired by A/B domain deletion. The mRNA levels of selected adipocyte genes were measured by real-time qPCR and normalized to the corresponding level of Gtf2b mRNA. Synergy factors for PPARγ2-C/EBPα and PPARγCDE-C/EBPα are indicated. Error bars indicate the SEMs of triplicates. Results are representative of three independent experiments. ND, not detectable.

The TE-III domain is required for the ability of C/EBPα to act as a pioneer factor for PPARγ.

The ability of C/EBPα to recruit the SWI/SNF complex through direct interaction of transactivation element III (TE-III) of C/EBPα (Fig. 10A) with the hBRM and BAF155 subunits has been shown to be important for the ability of C/EBPα to promote adipogenesis (33). We therefore asked to what extent the TE-III domain is important for synergy with PPARγ and chromatin remodeling. Our results showed that deletion of the TE-III domain compromises the ability of C/EBPα to activate adipocyte genes in the absence of PPARγ (Fig. 10B and C) and markedly reduces the ability to synergize with PPARγ for activation of most adipocyte genes investigated. Importantly, binding of both C/EBPα and PPARγ to sites where C/EBPα facilitates PPARγ binding was significantly reduced by deletion of the TE-III domain, whereas binding of these factors to sites where PPARγ facilitates C/EBPα binding was not affected (Fig. 10D). This is consistent with the assisted loading model, in which the leading transcription factor facilitates binding of the other by making the chromatin template more accessible through recruitment of remodeling factors.

FIG 10.

Deletion of transactivation element III (TE-III) impairs the ability of C/EBPα to act as a leading factor. (A) Schematic representation of C/EBPα with the indicated transactivation elements (TE-I through TE-III) and the basic leucine zipper (bZIP) domain. (B) Western blot demonstrating equal expression of full-length C/EBPα and C/EBPα with TE-III deleted. TFIIB was used as a loading control. (C) Synergistic activation of adipocyte genes requires the TE-III domain of C/EBPα. The mRNA levels of selected adipocyte genes were measured by real-time qPCR and normalized to the corresponding level of Gtf2b mRNA. Synergy factors for PPARγ-C/EBPα and PPARγ-C/EBPα ΔTE-III coexpression are indicated. Error bars indicate the SEMs of triplicates. Results are representative of four independent experiments. (D) Deletion of TE-III impairs recruitment of C/EBPα and PPARγ to shared sites where C/EBPα facilitates binding. C/EBPα and PPARγ ChIP-PCR was performed for selected shared sites displaying both consensus motifs in the vicinity of synergistically regulated genes. Binding to a negative-control (NC) region, denoted “No gene,” was used as a control. The results of C/EBPα ChIP are represented as the means of three independent experiments + SEMs. The results of PPARγ ChIP are represented as the means of two independent experiments plus range. (E, G, and H) Recruitment of BRG1, MED1, and CBP to C/EBPα-facilitated binding sites is reduced by TE-III deletion. Results for BRG1 (E), MED1 (G), and CBP (H) ChIP-PCR at selected shared sites, all displaying both consensus motifs, three PPARγ-facilitated C/EBPα binding sites, and three C/EBPα-facilitated PPARγ binding sites are shown. A no-gene region was used as a negative control. Results are the means of two independent experiments + range. (F) Cooperative chromatin remodeling is impaired at C/EBPα-facilitated binding sites by deletion of TE-III. FAIRE-qPCR was performed at selected shared sites, as described for panel E. The β-globin promoter was used as a negative-control region. Error bars indicate the SEMs of triplicates. Results are representative of three independent experiments.

In further support of the assisted loading model, deletion of TE-III led to a marked reduction in recruitment of BRG1 and chromatin openness (as determined by FAIRE analysis) at C/EBPα-facilitated PPARγ binding sites but not at PPARγ-facilitated C/EBPα binding sites (Fig. 10E and F). The exception was remodeling at the Lpl + 73 kb site (a PPARγ-facilitated C/EBPα binding site), which was compromised in cells coexpressing PPARγ and C/EBPα ΔTE-III. Remodeling at this site was highly cooperative and might therefore also depend on C/EBPα's ability to recruit chromatin remodeling factors, even though PPARγ is the leading transcription factor. As expected, deletion of TE-III also primarily affected recruitment of MED1 and CBP to C/EBPα-facilitated binding sites, signifying the role of remodeling for establishment of functional enhancers (Fig. 10G and H). Taken together, these data indicate that assisted loading is an important mechanism through which C/EBPα and PPARγ facilitate binding of each other to chromatin.

The dependence on TE-III for synergistic activation of the Acsl1, Dgat1, and Cdo1 genes corresponds well with the TE-III-dependent recruitment of C/EBPα and PPARγ to C/EBPα-facilitated PPARγ binding sites in the vicinity of these genes (Fig. 10C and D), indicating an important functional role of C/EBPα-facilitated binding of PPARγ to these sites for the regulation of the corresponding genes. In contrast, although TE-III deletion impaired the synergistic activation of the Cidec, Lpl, and Nr1h3 genes by C/EBPα and PPARγ, this deletion did not interfere with the recruitment of the two transcription factors or the three cofactors investigated to the nearby PPARγ-facilitated C/EBPα binding sites (Fig. 10D, E, G, and H). This most likely reflects the fact that these genes are not exclusively regulated by these enhancers but are also regulated by other putative enhancers where C/EBPα facilitates PPARγ binding in a TE-III-dependent manner (see Fig. S1 in the supplemental material).

Collectively, the data presented here provide clear evidence for a model where synergistic gene activation of key metabolic adipocyte genes occurs at least in part through an assisted loading mechanism where C/EBPα, as the leading factor, primes for the recruitment of PPARγ and subsequent cooperative cofactor recruitment, as illustrated in Fig. 11.

FIG 11.

Model for synergistic gene activation by PPARγ and C/EBPα through assisted loading to adjacent binding motifs. An enhancer/promoter, displaying consensus motifs for both transcription factors, in the vicinity of a target gene is found in closed chromatin in PPARγ^−/− MEF-CAR cells. Ectopically expressed PPARγ is unable to bind and remodel the chromatin, preventing transcriptional activation. Ectopically expressed C/EBPα binds to DNA at strong C/EBPα binding motifs and recruits BRG1 and low levels of CBP and MED1 in a TE-III-dependent manner. This leads to remodeling of the chromatin and low levels of expression of the target gene. If PPARγ is coexpressed, C/EBPα assists binding of PPARγ by making its binding motif more accessible, and the two transcription factors cooperatively enhance chromatin remodeling as well as MED1 and CBP recruitment, leading to synergistic gene activation.

DISCUSSION

It is well established that PPARγ, in particular, and also C/EBPα play important roles in adipogenesis and that these transcription factors cooperate in activation of the adipocyte gene program, partly by mutually inducing the expression of each other. Genome-wide analyses have recently suggested that a hitherto unrecognized high number of adipocyte genes may be coregulated by PPARγ and C/EBPα through common regulatory regions (26, 27, 29). This suggests that direct cross talk between PPARγ and C/EBPα may exist at the chromatin level.

In this study, we investigated, for the first time, the cross talk between these two key regulators of adipocyte differentiation at the chromatin level. We circumvented their interdependent expression by ectopically expressing the two transcription factors in PPARγ^−/− MEFs engineered to stably express CARΔ1. Using RNA-seq to analyze gene expression in this cell system, we demonstrated that many genes involved in adipocyte metabolism are synergistically activated by PPARγ and C/EBPα. Notably, genes synergistically regulated in PPARγ^−/− MEF-CAR cells are the genes most highly upregulated during adipogenesis of 3T3-L1 preadipocytes, indicating that this synergistic activation is important for activation of the adipocyte gene program. Thus, in addition to C/EBPα's role as an activator of PPARγ expression (14, 16, 18) and an activator of the gene program governing insulin sensitivity (13, 14), it appears to directly synergize with PPARγ in gene activation.

Our genome-wide profiling of PPARγ and C/EBPα binding under conditions where they were expressed alone or in combination revealed that the two factors mutually facilitate the binding of each other at several thousand genomic binding sites. In particular, there are many sites (7,997) where C/EBPα is required for the binding of PPARγ, and C/EBPα binds to the majority (79%) of these sites. However, there are also sites (1,818) where PPARγ is required for C/EBPα binding, with PPARγ binding to 62% of these sites. Hence, the two key regulators of adipogenesis function as pioneer factors for each other at many sites in the genome. Importantly, we show that sites with such interdependent binding of C/EBPα and PPARγ are highly correlated with the synergistic activation of target genes, indicating that interdependent binding of C/EBPα and PPARγ is functionally important.

Interdependent binding of transcription factors to chromatin was initially described at the single-site level (52, 69). More recent genome-wide profiling of transcription factor binding combined with loss-of-function analyses has revealed that some transcription factors act as pioneer factors for others (53, 54). Thus, Forkhead box A1 (FoxA1) has been shown to function as a pioneer factor for the estrogen receptor (53, 55, 56) and the androgen receptor (55, 57–59), PU.1 has been shown to facilitate C/EBPβ and LXR binding during macrophage development (40), and C/EBPs and AP-1 have been shown to facilitate glucocorticoid receptor (GR) binding in several different model systems (60–63). The current study extends cooperative binding of transcription factors to the major regulators of adipocyte differentiation and demonstrates that they can act as mutual pioneer factors for each other.

Interestingly, our results demonstrated that the main feature that distinguishes sites with interdependent binding of PPARγ and C/EBPα from sites with independent binding of the two factors is the chromatin conformation in the parent MEF cell line. Thus, binding sites with both motifs that are in an open chromatin configuration in nontransduced MEFs tend to bind the factors independently, whereas sites in a closed chromatin configuration tend to display facilitated binding of the factors. This clearly indicates that cooperation between transcription factors becomes more important for gaining access to DNA if the chromatin structure is closed and that chromatin remodeling is involved in the facilitated binding events. The distance between motifs also appears to play a role, since this is generally shorter for interdependent binding sites. Another interesting finding is that the identity of the leading factor at facilitated binding sites appears to be determined by motif strength; i.e., at sites where C/EBPα is the leading factor, the C/EBPα motif is generally strong and the PPARγ motif is weak and vice versa. This is in line with the finding that GR has pioneering potential at a subset of GR binding sites with a high incidence of the GR binding motif (61).

The exact molecular mechanisms underlying facilitated binding of transcription factors are not clear but are believed to involve indirect binding by tethering (64–66) or direct binding by assisted loading, where the leading transcription factor recruits the chromatin remodeling complex, thereby increasing DNA accessibility, which facilitates binding of the secondary factor to its motif (30). The model system used in this study provides a unique possibility to investigate the modes of cross talk between PPARγ and C/EBPα at the chromatin level, because the expression levels of the two factors can be independently controlled. Our data indicate that assisted loading is an important mechanism for facilitated binding of PPARγ and C/EBPα. First, analysis of the DNA sequences at sites with interdependent binding shows that in a large percentage of the cases both motifs are present. Furthermore, the PPARγ motif is almost as abundant at sites with C/EBPα-facilitated PPARγ binding as at sites with constitutive PPARγ binding. Similarly, the C/EBP binding motif is almost equally abundant at sites with PPARγ-dependent and -independent C/EBPα binding. This indicates direct binding of both factors to DNA at many interdependent binding sites. Second, the ability of C/EBPα to facilitate PPARγ binding is dependent on its ability to recruit the SWI/SNF nucleosome remodeling complex and increase chromatin accessibility. In contrast, the ability of C/EBPα to recruit the SWI/SNF complex is dispensable at sites where PPARγ is the leading transcription factor. Third, as mentioned above, PPARγ and C/EBPα binding sites in a closed chromatin context are more likely to display facilitated binding than binding sites in an open chromatin configuration, indicating that chromatin remodeling is important for the facilitated binding events. Thus, our data provide strong support for assisted loading as an important mechanism for interdependent PPARγ and C/EBPα binding. It should be noted that some gained PPARγ binding sites do not have a PPARγ motif, suggesting that facilitated binding can also occur by indirect binding. Alternatively, C/EBPα and other cooperating factors are able to relax the sequence specificity so much that PPARγ can interact with nonconsensus sequences, e.g., direct repeat half sites. Consistent with this, previous work from our group indicated that both PPARα and liver X receptor can associate directly with sequences that have relatively little resemblance to their consensus sequences (67).

In conclusion, the results presented here have important implications for our understanding of the transcriptional network of adipocytes. We show for the first time that the key transcriptional activators of adipogenesis, PPARγ and C/EBPα, synergize in activation of the adipocyte metabolic gene program in part by facilitating the binding of each other to chromatin, followed by cooperative cofactor recruitment. This facilitated binding of PPARγ and C/EBPα occurs at least in part by assisted loading, where the leading transcription factor plays the major role in recruitment of the remodeling complex. By analogy, it is likely that transcriptional regulators of cell identity in other cell types synergize with each other by cooperative binding to chromatin and cooperative recruitment of coactivators.