Abstract

Brown fat dissipates energy as heat and protects against obesity. Here, we identified nuclear factor I-A (NFIA) as a transcriptional regulator of brown fat by a genome-wide open chromatin analysis of murine brown and white fat followed by motif analysis of brown-fat-specific open chromatin regions. NFIA and the master transcriptional regulator of adipogenesis, PPARγ, co-localize at the brown-fat-specific enhancers. Moreover, the binding of NFIA precedes and facilitates the binding of PPARγ, leading to increased chromatin accessibility and active transcription. Introduction of NFIA into myoblasts results in brown adipocyte differentiation. Conversely, the brown fat of NFIA-knockout mice displays impaired expression of the brown-fat-specific genes and reciprocal elevation of muscle genes. Finally, expression of NFIA and the brown-fat-specific genes is positively correlated in human brown fat. These results indicate that NFIA activates the cell-type-specific enhancers and facilitates the binding of PPARγ to control the brown fat gene program.

Main

Obesity and its complications including diabetes amount to a world-wide epidemic. While white adipose tissue (WAT) stores energy as lipids and expands in obesity, brown adipose tissue (BAT) is specialized to dissipate energy through the uncoupling protein-1 (UCP1) on the mitochondrial inner membrane. When activated, UCP1 dissipates the electrochemical gradient that is normally used for adenosine triphosphate (ATP) synthesis, resulting in energy expenditure in the form of heat1. Although the existence of human BAT was controversial until recently, since the re-discovery of BAT in human adults2,3,4,5,6, it has been considered a potential target in the treatment of obesity. Human BAT activity is inversely correlated with body mass index3, and studies have shown that chronic cold exposure7,8 and β3 adrenergic receptor agonist administration9 successfully recruit human BAT and increase systemic energy expenditure.

Lineage tracing has demonstrated that brown fat and skeletal muscle share a common progenitor, but brown fat and white fat do not10. Both brown fat and skeletal muscle derive from a Myf5-positive precursor, and a transcriptional cofactor PRD1-BF1-RIZ1 homologous domain containing 16 (PRDM16) works as a cell-fate switch10,11,12. The master transcriptional regulator of adipogenesis—peroxisome proliferator-activated receptor γ (PPARγ)—and its agonist were also shown to activate the brown fat gene program13,14. Motif analysis of PPARγ-binding sites in BAT identified early B cell factor 2 (EBF2) as a transcriptional regulator of brown fat15. However, much remains elusive in the genome-wide landscape of brown fat development.

To gain insight into the underlying mechanism of brown fat development in a global and unbiased manner, we performed formaldehyde-assisted isolation of regulatory elements (FAIRE) coupled with high-throughput sequencing16 on murine BAT and WAT to profile the tissue-specific accessible chromatin regions. Through motif analysis of BAT-specific open chromatin regions, we identified NFIA as a transcriptional regulator of brown fat. NFIA exerts its effects by co-localizing with PPARγ at cell-type-specific enhancers.

RESULTS

The NFI motif within BAT open chromatin

BAT and WAT share a common transcriptional program regulated by PPARγ and CCAAT/enhancer binding proteins (C/EBPs). However, these tissues also have depot-selective gene programs that are responsible for their specific functions11,15,17. Regulatory elements controlling gene expression are characterized by open chromatin structures accessible to transcription factors and cofactors. We performed FAIRE-seq analyses of murine interscapular BAT, inguinal WAT (iWAT) and epididymal WAT (eWAT) to map open chromatin regions genome-wide, and we identified 24,322 FAIRE peaks for BAT, 10,012 for iWAT and 12,656 for eWAT (Fig. 1a, b). Genes near BAT-specific FAIRE peaks were associated with gene ontology (GO) terms such as brown fat cell differentiation (Supplementary Fig. 1a, b), suggesting that the FAIRE-seq experiments unbiasedly identified functionally active, depot-specific accessible chromatin regions. Through motif analysis, in addition to known regulators such as C/EBPβ, EBF2 and PPARγ (refs 13,15,18), we found that the binding motif for the NFI transcription factor was the most highly enriched within BAT-specific open chromatin regions (Fig. 1c).

Figure 1: The NFI-binding motif is highly enriched in brown-fat-specific open chromatin regions.
Figure 1

(a) Representative FAIRE-seq tracks of murine BAT, iWAT and eWAT. (b) Venn diagram showing overlap of BAT, iWAT and eWAT FAIRE peaks. (c) Motifs enriched in BAT-specific FAIRE peaks and eWAT- or iWAT-specific FAIRE peaks. The motif analysis was performed once on the basis of the FAIRE-seq data set. (d) mRNA expression levels of the NFI family in C2C12 myoblasts, immortalized brown adipocytes and 3T3-L1 white adipocytes before (pre) and after (diff) differentiation (mean  ±  s.e.m.; n = 3 independent samples). (e) mRNA expression levels of the NFI family in soleus muscle, quadriceps muscle, BAT, iWAT and eWAT (mean ± s.e.m.; n = 7 independent samples; P < 0.05, P < 0.01). (f) Western blot analysis of NFIA in representative samples from soleus muscle, quadriceps muscle, BAT, iWAT and eWAT. β-actin was used as a loading control. Representative images of two independent experiments are shown. Unprocessed original scans of blots are shown in Supplementary Fig. 8.

Of the four isoforms of the NFI family19, we found that Nfia was highly expressed in brown adipocytes compared with its expression in 3T3-L1 white adipocytes or C2C12 myoblasts. Moreover, the gene expression level of Nfia was robustly induced during brown adipocyte differentiation (Fig. 1d). NFIA was also highly expressed in BAT compared with WAT or skeletal muscle at both the RNA and protein levels (Fig. 1e, f). Furthermore, the expression level of Nfia was increased when mice are challenged by exposure to cold or β3-agonist CL316,243 (Supplementary Fig. 1c–e). These findings indicate that NFIA is a candidate transcriptional regulator that defines brown adipocyte identity.

NFIA induces brown adipogenesis

To examine whether NFIA can induce adipocyte differentiation from myoblasts, we introduced NFIA into C2C12 myoblast cell lines using retroviral vectors (Fig. 2a). The cells were grown to confluence and treated with an adipogenic cocktail. Strikingly, NFIA-expressing cells differentiated into lipid-filled adipocytes (Fig. 2b). Consistent with cell morphology, NFIA induced expression of the master regulator Pparg and the general adipocyte marker Fabp4 (Fig. 2d). In contrast, myogenic genes such as Myod1 and Myog were suppressed by NFIA (Fig. 2c). NFIA also induced brown-fat-specific genes including Cidea and Ppargc1a as well as the thermogenic gene Ucp1 in response to elevated cyclic AMP through forskolin treatment (Fig. 2e). Functionally, NFIA-expressing cells showed induced uncoupled respiration (Fig. 2f).

Figure 2: NFIA is capable of—and required for—driving brown adipocyte differentiation.
Figure 2

(a) Western blot analysis of NFIA in control and NFIA-expressing cells. β-actin was used as a loading control. (b) Control and NFIA-expressing C2C12 myoblasts were stained with Oil Red O seven days after inducing adipocyte differentiation. Scale bar, 50 μm. (ce) Myogenic genes (c), common adipocyte genes (d) and brown-fat-specific genes (e) were quantified by RT-qPCR at the indicated times (mean ± s.e.m.; n = 3 independent samples; P < 0.05, P < 0.01). fsk, forskolin. (f) Oxygen consumption of control and NFIA-expressing C2C12 myoblasts (mean ± s.e.m.; n = 10 independent samples; P < 0.01). (g) Control shRNA or shRNA for NFIA was introduced into immortalized brown adipocytes and the cells were stained with Oil Red O six days after inducing adipocyte differentiation. Scale bar, 50 μm. (hjNfia (h), Pparg and Fabp4 (i), Ppargc1a and Ucp1 (j) were quantified by RT-qPCR at the indicated times (mean ± s.e.m.; n = 3 independent samples; P < 0.05,P < 0.01; NS, not significant.). (k) Western blot analysis of NFIA, UCP1 and PPARγ in the indicated cells. β-actin was used as a loading control. Representative images of two independent experiments are shown. Unprocessed original scans of blots are shown in Supplementary Fig. 8.

Hierarchical clustering of genes quantified by RNA-seq showed the global changes in gene expression caused by the introduction of NFIA (Supplementary Fig. 2a). When we defined genes selective for BAT and skeletal muscle (SKM) by fold changes of expression levels between these tissues, BAT-selective genes were enriched in the cluster of genes upregulated by NFIA (P = 9.9 × 10−28, chi-square test), while SKM-selective genes were enriched in the cluster of genes downregulated by NFIA (P = 2.3 × 10−32). And GO analysis independently supported this observation (Supplementary Fig. 2b). Taken together, these data indicate that introduction of NFIA into myoblasts drives brown adipocyte differentiation while inhibiting myogenic differentiation.

We also tried introduction of NFIA into 3T3-F442A white preadipocyte cells (Supplementary Fig. 2d). The effect of NFIA on lipid accumulation and on common adipocyte gene expression after differentiation was modest (Supplementary Fig. 2c, e). However, NFIA very strongly increased the expression levels of the brown-fat-specific genes (Supplementary Fig. 2f), suggesting that NFIA can drive the brown fat gene program also in white preadipocytes.

To test the endogenous role of NFIA, we next performed loss-of-function experiments. We introduced a short hairpin (sh) RNA for NFIA into brown adipocytes and achieved significant knockdown throughout the differentiation (Fig. 2h, k). While the effects of NFIA knockdown on cell morphology and on common adipocyte gene expression were not significant (Fig. 2g, i), expression levels of the brown-fat-specific genes such as Ppargc1a and Ucp1 were significantly reduced (Fig. 2j). Expression of UCP1 protein was also highly reduced (Fig. 2k). Similarly, when we introduced a small interfering (si) RNA for NFIA by electroporation into fully differentiated brown adipocytes, expression levels of the brown-fat-specific genes were significantly reduced (Supplementary Fig. 2g–j). These results suggest that NFIA is required for both activation and maintenance of the brown-fat-specific gene expression. Altogether, our gain- and loss-of-function experiments show that NFIA is capable of and required for controlling the brown fat gene program.

NFIA and PRDM16 work in parallel with each other

The effect of NFIA on the brown fat gene program prompted us to examine the relationship with and requirement for PRDM16, which perform a similar function in this context. Although mass-spectrometric analysis suggested that NFIA is included in the PRDM16 protein complex18, our co-immunoprecipitation experiments showed that NFIA does not bind physically to PRDM16 (Fig. 3a). Introduction of NFIA into myoblasts did not induce Prdm16 expression. However, introduction of PRDM16 did induce Nfia expression while introduction of PPARγ did not (Fig. 3b, c). Importantly, PRDM16 was dispensable for the effect of NFIA, because NFIA was capable of stimulating adipocyte differentiation and stimulating the brown-fat-specific gene expression even when PRDM16 was knocked down (Fig. 3d–g). Interestingly, the opposite was also true (Fig. 3h–k). Overall, these results suggest that NFIA and PRDM16 work in parallel with each other.

Figure 3: PRDM16 is dispensable for the effect of NFIA.
Figure 3

(a) Left, immunoprecipitation of V5-tagged PRDM16 from HEK293 cells expressing V5-tagged PRDM16 and/or 3×FLAG-tagged PPARγ followed by western blot analysis to detect 3×FLAG-tagged PPARγ, as a positive control10. Right, immunoprecipitation of V5-tagged PRDM16 from HEK293 cells expressing V5-tagged PRDM16 and/or NFIA followed by western blot analysis to detect NFIA. Representative images of two independent experiments are shown. (b) RT-qPCR analysis of Prdm16 in control or NFIA-expressing C2C12 cells during the induction of adipocyte differentiation (mean ± s.e.m.; n = 3 independent samples; NS, not significant). (c) RT-qPCR analysis of Nfia in control, PPARγ- or PRDM16-expressing C2C12 cells (mean ± s.e.m.; n = 3 independent samples; P < 0.05, P < 0.01). (d) Control shRNA or shRNA for PRDM16 was introduced into control or NFIA-expressing C2C12 myoblasts, and the cells were stained with Oil Red O seven days after inducing adipocyte differentiation. Scale bar, 50 μm. (eg) Prdm16 and Nfia (e), the general adipocyte marker Fabp4 (f) and the brown-fat-specific genes (g) were quantified by RT-qPCR at the indicated times (mean ± s.e.m.; n = 3 independent samples; P < 0.05, P < 0.01; NS, not significant). (h) Control shRNA or shRNA for NFIA was introduced into control or PRDM16-expressing C2C12 myoblasts, and the cells were stained with Oil Red O seven days after inducing adipocyte differentiation. Scale bar, 50 μm. (ik) Nfia and Prdm16 (i), the general adipocyte marker Fabp4 (j) and the brown-fat-specific genes (k) were quantified by RT-qPCR at the indicated times (mean ± s.e.m.; n = 3 independent samples; P < 0.05, P < 0.01; NS, not significant). Unprocessed original scans of blots are shown in Supplementary Fig. 8.

NFIA binds to the brown fat enhancers

To dissect the genome-wide binding landscape of NFIA in brown adipocyte differentiation, we performed chromatin immunoprecipitation coupled with high-throughput sequencing (ChIP-seq) analysis using the NFI antibody, which reacts primarily with NFIA, but also with NFIC and NFIX. We also performed ChIP-seq for PPARγ, C/EBPα, C/EBPβ, EBF2 and H3K27 acetylation (H3K27Ac). We additionally performed assays for transposase-accessible chromatin coupled with high-throughput sequencing (ATAC-seq) for accessible chromatin regions20. We further performed ChIP-seq for NFIA with the FLAG M2 antibody in C2C12 myoblasts that expressed FLAG-tagged NFIA. We identified 12,486 and 12,748 NFI-binding sites, respectively, on day 0 and day 6 of differentiation and the majority of NFI-binding sites were located distal to genes, as is the case with PPARγ (Fig. 4a and Supplementary Fig. 3a). Motif analysis showed that NFI-binding sites in brown adipocytes and NFIA-binding sites in NFIA-expressing C2C12 myoblasts were strongly enriched with NFI motif (Fig. 4b and Supplementary Fig. 3b, c), in agreement with direct DNA binding. We observed that NFI binds to the enhancers of the master regulator Pparg and the brown-fat-specific genes such as Cidea and Ucp1 (Fig. 4c). Importantly, most of the NFI-binding sites in brown adipocytes and NFIA-binding sites in NFIA-expressing C2C12 myoblasts overlapped each other at BAT FAIRE peaks (Supplementary Fig. 3e). Furthermore, when we defined BAT- and WAT-selective genes by fold changes of expression levels between these tissues, BAT-selective genes were closer to NFI-binding sites than were WAT-selective genes (Fig. 4d), and BAT-selective genes harboured more NFI-binding sites than did WAT-selective genes or all genes within ±50 kilobases (kb) of the transcription start site (TSS). (Fig. 4e). And the NFI binding signal was highly enriched near BAT-selective genes compared with that signal near WAT-selective genes or all genes (Fig. 4f). These results indicate that NFI binding is enriched at brown-fat-specific enhancers.

Figure 4: NFI binding is enriched at brown-fat-specific enhancers.
Figure 4

(a) Genomic location of NFI-binding sites in brown adipocytes at day 0 and day 6 of differentiation. (b) Enriched known motifs within NFI-binding sites in brown adipocytes at day 6 of differentiation. (c) Representative tracks of ChIP-seq for NFI, PPARγ, C/EBPα, C/EBPβ, EBF2 and H3K27Ac in brown adipocytes, ATAC-seq of brown adipocytes, FAIRE-seq of in vivo BAT, and ChIP-seq for NFIA in NFIA-expressing C2C12 myoblasts at Pparg, Cidea and Ucp1 loci. (d) The distance from the TSS to the nearest NFI-binding site for BAT- and WAT-selective genes. The definitions of BAT- and WAT-selective genes (n = 549 and n = 849, respectively) are shown in the Methods. (e) Number of NFI-binding sites within ±50 kb of the TSS for BAT- and WAT-selective genes (n = 549 and n = 849, respectively). (f) Box plot showing the strength of the NFI binding signal (MACS score) near all genes (n = 21,258), BAT- and WAT-selective genes (n = 549 and n = 849, respectively). The box shows the median, and the first and third quartiles. The whisker shows the value still within one-and-a-half times the interquartile range. The genome-wide analyses were performed once on the basis of the ChIP-seq data set.

Co-localization of NFIA and PPARγ

We also found that the binding sites of NFI and PPARγ often overlapped each other. Since co-localization of transcription factors at the chromatin is crucial for both cell-fate decision21 and cell-type-specific signalling22, we investigated co-localization of NFI and PPARγ genome-wide. We found that the binding sites of NFI overlapped those of PPARγ at 63% of all binding sites in differentiated brown adipocytes (8,001 of 12,748 sites, Fig. 5a). Most strikingly, the majority of the co-localizing peaks were pre-occupied by NFI but not occupied by PPARγ, C/EBPα, C/EBPβ nor EBF2 before differentiation, and these sites exhibited a high level of H3K27 acetylation and chromatin accessibility, markers of active enhancers even before differentiation (Fig. 5b, c). And the binding sites of NFI were substantially concordant between day 0 and day 6 of differentiation, unlike other transcription factors examined (Supplementary Fig. 3f–j).

Figure 5: Co-localization of NFI and PPARγ at the brown-fat-specific enhancers.
Figure 5

(a) Venn diagram showing the overlap of NFI and PPARγ ChIP-seq peaks in brown adipocytes at day 6 of differentiation. (b) Venn diagram showing the overlap of the co-localizing peaks at day 6 of differentiation, NFI ChIP-seq peaks at day 0 and PPARγ ChIP-seq peaks at day 0. (c) Heat map showing ChIP-seq for NFI, PPARγ, C/EBPα, C/EBPβ, H3K27Ac and ATAC-seq tag densities at the co-localizing peaks of NFI and PPARγ at day 6. (d) Bar graph showing the number of co-localizing sites per gene within ±50 kb of BAT- and WAT-selective genes stratified by the fold changes of gene expression. The genome-wide analyses were performed once on the basis of the ChIP-seq data set.

To test whether the co-localization of NFI and PPARγ is associated with gene expression, we counted the number of co-localizing sites per gene within ±50 kb regions around the BAT- and WAT-selective genes stratified by the fold change of expression. The results showed that the more the genes were expressed in brown fat compared with expression in white fat, the higher the number of co-localizing sites per gene (Fig. 5d). We observed that the co-localization was enriched near BAT-selective genes also in white adipocytes (Supplementary Fig. 3k, l). NFI-binding sites near BAT-selective genes were closer to DR-1 motifs (the consensus motif for PPARγ) compared with those near WAT-selective genes (Supplementary Fig. 3m), suggesting that the co-localization near BAT-selective genes is, at least in part, determined by the DNA sequence itself.

To reveal the functional consequences of NFIA and PPARγco-localization, we utilized a model system in which we introduced into C2C12 myoblasts either PPARγ alone or both PPARγ and NFIA. As previously reported, introduction of PPARγ alone is sufficient to promote adipocyte differentiation10. Co-expression of NFIA with PPARγ did not alter the degree of differentiation—as judged by both Oil Red O staining and general adipocyte marker Fabp4 expression (Fig. 6a, b). We confirmed that NFIA binds to the brown-fat-specific enhancers in this model system (Fig. 6d). Note especially that the binding of PPARγ to those enhancers near Cidea, Ppara, Ppargc1a and Ucp1 was significantly facilitated when NFIA co-localized with PPARγ (Fig. 6e), even though PPARγ protein levels were similar in both cells (Fig. 6c). We observed that the binding of PPARγ to some of those enhancers was already facilitated before differentiation (Supplementary Fig. 4a, b). Conversely, the binding of NFIA to those enhancers was independent of the co-localization of PPARγ, since NFIA in cells without PPARγ was able to bind to these loci as strongly as NFIA in cells with PPARγ (Supplementary Fig. 4c, d and Fig. 6d). Moreover, chromatin accessibility of those enhancers was dramatically increased when NFIA and PPARγ were co-localized (Fig. 6f). Finally, the co-localization robustly activated transcription of those genes (Fig. 6g). Together, these results demonstrate that co-localization of NFIA facilitates PPARγ binding to the brown-fat-specific enhancers for controlling the brown-fat-specific gene expression.

Figure 6: Co-localization of NFIA facilitates PPARγ binding to the brown-fat-specific enhancers and drives active transcription.
Figure 6

(a) C2C12 myoblasts expressing only PPARγ —and those expressing both PPARγ and 3×FLAG-NFIA—were stained with Oil Red O seven days after inducing adipocyte differentiation. Scale bar, 50 μm. (b) Indicated genes were quantified by RT-qPCR at day 7 after adipocyte differentiation (mean ± s.e.m.; n = 3 independent samples; P < 0.01; NS, not significant). (c) Western blot analysis of the PPARγ protein in PPARγ − or PPARγ + 3×FLAG-NFIA-expressing C2C12 myoblasts. β-actin was used as a loading control. (d) ChIP-qPCR analysis of NFIA. Cidea 29 kb, Ppara 21 kb, Ppargc1a − 97 kb and Ucp1 9.5 kb are background sites. The representative result of three independent experiments is shown (n = 2 independent samples; mean ± s.e.m.). (e) ChIP-qPCR analysis of PPARγ. Cidea 29 kb, Ppara 21 kb, Ppargc1a − 97 kb and Ucp1 9.5 kb are background sites. The representative result of three independent experiments is shown (n = 2 independent samples; mean ± s.e.m.). Source data for d,e are provided in Supplementary Table 2. (f) FAIRE-qPCR analysis. Cidea 29 kb, Ppara 21 kb, Ppargc1a − 97 kb and Ucp1 9.5 kb are background sites (mean ± s.e.m.; n = 3 independent samples; P < 0.01). (g) RT-qPCR analysis of the indicated genes (mean ± s.e.m.; n = 3 independent samples; P < 0.05, P < 0.01). Unprocessed original scans of blots are shown in Supplementary Fig. 8.

Role of NFIA in BAT in vivo

To evaluate the physiological relevance of NFIA in BAT in vivo, we analysed BAT in NFIA-knockout (KO) mice. NFIA-KO mice are born in Mendelian ratios but die within a week of birth due to neurological deficits including agenesis of corpus callosum and hydrocephalus23. We therefore analysed the BAT of neonates soon after birth. BAT masses and morphology were comparable among three genotypes (Fig. 7a). However, we observed significantly decreased expression of Ucp1 messenger RNA (Fig. 7b) and UCP1 protein (Fig. 7c) in NFIA-KO tissues. By ChIP-qPCR analysis of those tissues, we also observed severely impaired PPARγ binding to the Ucp1 − 4.5 kb enhancer (Fig. 7d). Notably, we observed co-localization of NFIA and PPARγ at the Ucp1 − 4.5 kb enhancer in our model system (Fig. 6d, e) and expression levels of Pparg were comparable between wild-type (WT) and NFIA-KO tissues (Supplementary Fig. 5a). Consistently, the binding of PPARγ to the Ucp1 − 4.5 kb and Ucp1 − 11.7 kb enhancer was highly decreased when NFIA was knocked down in brown adipocytes (Supplementary Fig. 5b). These results suggest that co-localization of NFIA and PPARγ is necessary for optimal expression of the Ucp1 gene in vivo.

Figure 7: Deficiency of NFIA causes an impaired brown fat gene signature and reciprocal elevation of skeletal muscle gene expression in vivo.
Figure 7

(a) Macroscopic pictures (scale bar, 2.5 mm) and haematoxylin and eosin (HE) staining (scale bar, 100 μm) of BAT sections from neonates. (b) RT-qPCR analysis of the Ucp1 gene (mean ± s.e.m.; n = 11 mice for WT, 24 mice for Nfia+/−, and 15 mice for Nfia−/−, respectively; P < 0.05). (c) Western blot analysis of UCP1 protein. β-actin was used as a loading control. Representative images of two independent experiments are shown. (d) ChIP-qPCR analysis of in vivo BAT. Ucp1 9.5 kb is a background site (mean ± s.e.m.; n = 3 independent samples; P < 0.05; NS, not significant). Representative results of two independent experiments are shown. (e) Volcano plot of RNA-seq analysis. BAT- and SKM-selective genes are depicted in red and blue, respectively. The definitions of BAT- and SKM-selective genes (n = 254 and n = 312, respectively) are shown in the Methods. (f) Top GO terms of genes down- or upregulated by NFIA-KO. (g) Scatter plot showing fold changes of gene expression by NFIA introduction into C2C12 myoblasts and NFIA-KO in BAT. BAT- and SKM-selective genes are depicted in red and blue, respectively. The genome-wide analyses were performed once on the basis of the RNA-seq data set. Unprocessed original scans of blots are shown in Supplementary Fig. 8.

To evaluate genome-wide changes in gene expression by NFIA-KO, we performed RNA-seq analysis (Fig. 7e). The expression levels of a battery of brown-fat-specific genes and mitochondrial genes were significantly decreased, while common fat genes were relatively maintained. In contrast, muscle-specific genes were reciprocally elevated by NFIA-KO (Supplementary Fig. 5c–f). Genes downregulated by NFIA-KO were associated with GO terms such as triglyceride biosynthetic process and brown fat differentiation, and genes upregulated by NFIA-KO were associated with GO terms such as skeletal myofibril assembly (Fig. 7f). A scatter plot of gene expression changes that were due to introduction of NFIA into myoblasts and NFIA-KO in BAT (Fig. 7g) demonstrated that BAT-selective genes (p = 5.1 × 10−29, chi-square test) and SKM-selective genes (p = 1.7 × 10−14) are reciprocally regulated by both gain- and loss-of-function studies. Finally, in db/db mice—a mouse model of obesity and diabetes—we found that expression levels of both Nfia and Ucp1 in BAT were suppressed compared with levels in C57BL6/J mice (Supplementary Fig. 5g), suggesting that downregulation of NFIA may play a pathophysiological role in the development of obesity and diabetes. Together, these results demonstrate that NFIA is required for the optimal activation of the brown fat gene program and repression of the muscle gene program in vivo.

NFIA in human brown fat

To explore the possible role of NFIA in human BAT, we analysed the perirenal BAT of human patients with pheochromocytoma and with non-functional adrenal tumours. Human perirenal BAT is considered relatively close to murine classical BAT in terms of developmental origin and gene signature24. Perirenal BAT is activated in patients with pheochromocytoma, and the gene expression pattern is similar to that of classical BAT in mice25. We observed that NFIA expression was higher in patients with pheochromocytoma compared with those who had non-functional adrenal tumours (Fig. 8a). Furthermore, expression levels of the brown-fat-specific genes including UCP1 and PPARGC1A were positively and significantly correlated with NFIA expression (Fig. 8b). We also analysed human brown and white adipocytes, obtained from supraclavicular and subcutaneous regions, respectively26. Expression levels of NFIA were higher in human brown adipocytes compared with white adipocytes throughout the differentiation (Fig. 8c). Finally, we measured the expression levels of NFIA in BAT and WAT of human necks27. The expression of NFIA was numerically higher in BAT than expression in WAT, although the difference did not reach statistical significance, possibly due to limited sample size (Fig. 8d). These results indicate that NFIA may also control the brown fat gene program in humans.

Figure 8: Expression of NFIA and the brown-fat-specific genes is positively correlated in perirenal brown fat of human patients with pheochromocytoma.
Figure 8

(a) Expression levels of NFIA in perirenal brown fat of human patients with pheochromocytoma or non-functional adrenal tumours (mean  ±  s.e.m.; n = 7 independent samples for non-functional adrenal tumours and n = 11 independent samples for pheochromocytoma; P < 0.01). (b) Correlation of expression of the brown-fat-specific genes and NFIA expression. (c) Expression levels of NFIA in human brown (supraclavicular) and white (abdominal subcutaneous) adipocytes before and after differentiation (mean ± s.e.m.; n = 5 independent samples for white adipocytes and n = 6 independent samples for brown adipocytes; P < 0.01). (d) RT-qPCR analysis of NFIA in human neck WAT (subcutaneous) and BAT (carotid sheath) (n = 5 independent samples). The analyses were performed once because human samples were highly limited.

DISCUSSION

PPARγ is the master transcriptional regulator of adipocyte differentiation28. Here, we show that NFIA co-localizes with PPARγ at the brown-fat-specific enhancers to control the brown fat gene program. The binding of NFIA precedes and facilitates the binding of PPARγ. NFIA may recruit chromatin remodelling complexes such as Swi/Snf, as reported in the case of human adrenal cells29. Alternatively, NFIA may work as a pioneer factor30 by facilitating structural changes of chromatin that increase accessibility and by recruiting PPARγ. Consistently, knockdown of PPARγ results in almost complete loss of the effect of NFIA (Supplementary Fig. 6a–d). In support of this, genetic variations that alter NF-1 motifs were reported to affect not only the binding of NFI, but also the binding of PPARγ (ref. 31), chromatin accessibility32 and the enhancer activity33. Proximally co-occupied transcription factors often compete with a nucleosome to access DNA34, and co-localization of NFIA and PPARγ is more likely to result in nucleosome displacement than binding of NFIA or PPARγ alone, probably leading to increased chromatin accessibility, enhancer activity and gene expression. Nevertheless, further studies are needed to clarify the whole picture of chromatin remodelling during brown fat development caused by NFIA and other factors.

Although NFIA and PRDM16 perform a similar function in brown adipocyte differentiation, we show that these two factors work in parallel with each other. And both NFIA and PRDM16 are indispensable for the full activation of the brown fat gene program. Indeed, expression levels of Prdm16 were relatively unaffected in NFIA-KO mice (Supplementary Fig. 5c), and the opposite was also the case35 (fold change of Nfia expression in KO/WT was 0.99). Additionally, PGC1α is also dispensable for the effect of NFIA, because the effect of NFIA is totally maintained even when PGC1α is knocked down (Supplementary Fig. 6e–h). Although expression levels of Adrb3 were severely reduced by NFIA-KO, Adrb3 was also dispensable for the effect of NFIA (Supplementary Fig. 6i–l).

The BAT of NFIA-KO mice showed impaired expression of the brown fat gene program. However, the appearance and mass of the BAT was relatively preserved, suggesting that co-localization of NFIA and PPARγ alone may not be sufficient for determining brown fat cell fate. Cell-fate determination may be achieved through the contribution of multiple factors, and deficiency of NFIA alone could be compensated for. The investigation at the prenatal stage36 and lineage tracing experiments10 will help characterize the role of NFIA more definitively. Additionally, tissue-specific deletion of NFIA in mice will be required for investigating the role of NFIA in adult BAT in systemic metabolism.

Since brown fat and skeletal muscle share a common progenitor, repressing the muscle gene program is required in addition to inducing the brown fat gene program to direct the cell-fate determination toward brown fat. However, many of the myogenic genes do not have NFIA-binding sites at their enhancers, and the repressive effect of NFIA on muscle genes may be indirect. Note that PPARγ is reported to suppress MyoD—a master regulator of myogenesis—through enhanced ubiquitylation and degradation of the MyoD protein37. NFIA may suppress myogenesis by inducing PPARγ; uncharacterized, PPARγ-independent mechanisms may also be involved.

Originally, NFI was found to bind to an enhancer of the Fabp4 gene38, and we previously reported that the NF-1 motif is enriched within open chromatin regions of differentiated white adipocytes compared with undifferentiated cells39. In this study, we found that the NF-1 motif is enriched in BAT-specific—not WAT-specific—open chromatin regions in vivo. NFIA is important for general adipogenesis—probably by inducing PPARγ—and NFIA is more important for the BAT-selective gene program than the WAT-selective gene program. Indeed, the overlap of NFI-binding sites with BAT FAIRE peaks is higher than that of NFI-binding sites with eWAT FAIRE peaks (Supplementary Fig. 7a, b). In support of this, a transcriptome analysis independently suggested that NFIA would be a positive regulator of brown adipocyte differentiation40. Interestingly, it is reported that NFIC negatively regulates adipocyte differentiation41. Future studies will be needed to understand the mechanisms underlying differential regulation of brown adipocyte differentiation by different isoforms of NFI family.

Developing a therapy for obesity through activating BAT is highly anticipated. Recently, an epigenome-wide association study showed that DNA methylation of a CpG site at the intron of NFIA in human adipose tissue shows a positive and significant association (P = 4.0 × 10−20) with obesity42, suggesting that downregulation of NFIA may contribute to pathophysiology of obesity in humans. Identifying an upstream regulator of NFIA may open a door for BAT-targeted anti-obesity therapy.□

Methods

Retroviral expression system.

For gain-of-function experiments, we used the pMXs retroviral expression system as previously described39. The sequences used for shRNA-mediated loss-of-function experiments were: shNFIA no. 1, 5′-CCUUCUCAACUCUGUAACA-3′; shNFIA no. 2, 5′-GUCAGCAGUUACAUACAUA-3′; shPPARγ, 5′-CAAGAGAUCACAGAGUAUG-3′; shPRDM16, 5′-GAAGAGCGUGAGUACAAAU-3′. The corresponding double-stranded DNA sequences were subcloned into the pLMP retroviral vector (Open Biosystems) using XhoI and EcoRI restriction enzyme sites. For retrovirus production, Platinum E packaging cells (Cell Biolabs) were transfected with the vector using Lipofectamine 2000 (Invitrogen). Two days later, conditioned medium was centrifuged at 2,000 r.p.m. (780g) for 5 min, and the supernatant was supplemented with 10 mg ml−1 Polybrene and used for overnight infection. Subsequently, infected cells were selected using appropriate antibiotics.

Cell culture.

C2C12 myoblasts and 3T3-L1 adipocytes were purchased from the American Type Culture Cell Collection (ATCC). 3T3-F442A adipocytes were a gift from S. Kajimura11 (UCSF Diabetes Center, University of California, San Francisco, USA). Immortalized brown adipocytes were a gift from K. Ueki43 (Diabetes Research Center, National Center for Global Health and Medicine, Japan) and S. Kajimura44. Immortalized brown adipocytes44 and C2C12 myoblasts were negative for mycoplasma contamination. Other cell lines were low passage and not tested for mycoplasma. None of the cell lines mentioned above are listed in the database of commonly misidentified cell lines maintained by International Cell Line Authentication Committe (ICLAC). For adipocyte differentiation of C2C12 transfected with NFIA and/or PPARγ expression vector, cells were treated for 48 h in medium containing 10% FBS, 0.5 mM isobutylmethylxanthine, 125 nM indomethacin, 1 mM dexamethasone, 850 nM insulin, 1 nM T3 and 1 mM rosiglitazone. After 48 h, cells were switched to medium containing 10% FBS, 850 nM insulin, 1 nM T3 and 1 mM rosiglitazone. To stimulate thermogenic gene expression, cells were incubated with 10 μM forskolin (fsk) for 4 h.

siRNA-mediated gene knockdown.

For NFIA knockdown experiments by electroporation, a control siRNA and a siRNA for NFIA were purchased from Sigma (Mission_SIC-001 and Mm_Nfia_9630). Differentiated brown adipocytes at day 6 of differentiation were washed, trypsinized, centrifuged and transfected by the Neon transfection system (Invitrogen). The cells were harvested 2 days after the transfection. For PGC1α or Adrb3 knockdown experiments by lipofection, a control siRNA and a siRNA for PGC1α or Adrb3 were purchased from Santa Cruz Biotechnology (sc-37007, sc-38885 and sc-39869). The siRNAs were transfected using Lipofectamine RNAiMAX (Invitrogen) 2 days before confluence, according to the manufacturer’s instructions.

ChIP.

Samples were treated by nuclear extraction buffer (10 mM Tris-HCl, pH 7.4, 10 mM NaCl, 3 mM MgCl2 and 0.1% IGEPAL CA-630) for 10 min and immediately crosslinked with 1% formaldehyde for 7.5 min at room temperature. Crosslinking was quenched using 125 mM glycine for 5 min. The chromatin was sheared by a probe sonicator (Branson) and was spun at 15,000 r.p.m. (20,400g) for 5 min. Antibodies were added for overnight incubation at 4 °C. Mixes of Protein A and Protein G Sepharose (GE) were added to samples for 4 h at 4 °C. Subsequent procedures were performed as described previously39. The antibodies used were FLAG M2 (1 μg per immunoprecipitation (IP), Sigma, F3165), NFI (12 μg per IP, Santa Cruz Biotechnology, sc-30198), PPARγ (4 μg per IP, mix of Santa Cruz Biotechnology, sc-7273, and Perseus Proteomics, A3409A), C/EBPα (4 μg per IP, Santa Cruz Biotechnology, sc-61), C/EBPβ (4 μg per IP, Santa Cruz Biotechnology, sc-150), EBF2 (6 μg per IP, R&D Systems, AF7006) and H3K27Ac (4 μg per IP, Abcam, ab4729). ChIP-seq libraries were prepared using the KAPA Hyper Prep Kit (KAPA Biosystems) according to the manufacturer’s instructions.

FAIRE.

FAIRE was performed as described previously39, with optimization for experiment using in vivo tissues. Briefly, freshly collected adipose tissues obtained from 8-week-old male C57BL/6J mice were minced with scissors and crosslinked with 1% formaldehyde for 7 min at room temperature, followed by quenching with 125 mM glycine for 5 min. A Pasteur pipette was used to carefully wash the floating minced samples with cold PBS. The chromatin was sheared by using a homogenizer and then a probe sonicator (Branson). Subsequent procedures were performed as described previously39. FAIRE-seq libraries were prepared using ChIP-Seq Sample Prep Kit (Illumina) according to the manufacturer’s instructions.

ATAC.

Fifty thousand nuclei of brown adipocytes before and after differentiation were transposed using Tn5 transposase as previously described20. Briefly, cells were lysed using ice-cold lysis buffer (10 mM Tris-HCl pH 7.4, 10 mM NaCl, 3 mM MgCl2 and 0.1% IGEPAL CA-630) and were spun at 2,400 r.p.m. (500g) for 10 min. The pellet was resuspended in the transposase reaction mix and incubated at 37 °C for 30 min. The sample was column-purified and amplified by 15 cycles of PCR before high-throughput sequencing.

RNA expression analysis.

Total RNA from cultured cells or tissues was isolated using TRIzol reagent (Invitrogen) and RNeasy minicolumns (QIAGEN). Isolated RNA was reverse-transcribed using ReverTra Ace qPCR RT Master Mix kit (Takara). Real-time quantitative PCR (SYBR green) analysis was performed on a 7900HT Fast Real-Time PCR System or QuantStudio 7 Flex Real-Time PCR System (Applied Biosystems). Rplp0 was used as an internal normalization control of murine samples. TBP (Fig. 8a, b), RN18S (Fig. 8c) and PPIA (Fig. 8d) were used as an internal normalization control of human samples. For RNA-seq, libraries were prepared using TruSeq Stranded mRNA Library Prep Kit (Illumina) according to the manufacturer’s instructions.

High-throughput sequencing.

High-throughput sequencing was performed by using the Illumina Genome Analyzer System or HiSeq 2500. Demultiplex and base calling were performed with the CASAVA 1.8.2 software (Illumina).

ChIP-seq, FAIRE-seq and ATAC-seq data processing.

The sequence reads were mapped to UCSC build mm9 (NCBI Build 37) assembly of the mouse genome using the ELAND mapping program. Peak calling was performed using the MACS version 1.4 with default parameters45. Peaks of ChIP-seq, FAIRE-seq and ATAC-seq were visualized by a GenomeJack browser (version 3.1, Mitsubishi Space Software). Galaxy cistrome46 was used for genomic region handling. For the Venn diagrams, note that the sum of the number of peaks in each component may not equal the number of overall peaks, because a single peak in one sample could overlap with multiple peaks in another sample. Motif analysis was performed using CentriMo47 version 4.10.2 with default parameters. We used the licensed version of the TRANSFAC database48. Gene ontology annotation analysis was performed using DAVID49. Biological process terms ‘GO_BP_FAT’ were used and GO terms with FDR less than 0.05 were shown in descending order of the fold enrichment. A heat map representation was generated using in-house software.

RNA-seq data processing.

The sequence reads were mapped to the mm9 mouse genome using TopHat. Fragments per kilobase of exon per million fragment mapped (FPKM) values were calculated for each gene using CuffLinks. Differentially expressed genes were analysed using DeSeq2 (ref. 50). Genes with FPKM value <1 in all of the samples were excluded for the differential expression analysis. A heat map representation was generated using GenePattern online software51.

Definition of tissue-selective genes for genome-wide analysis.

We defined BAT- and WAT-selective genes (n = 549 and n = 849, respectively) using the publicly available microarray data set GSE28440. BAT-selective genes were defined as genes expressed twofold or more in BAT than in WAT with statistical significance (P < 0.05). We likewise defined WAT-selective genes. For BAT- and SKM-selective genes (n = 254 and n = 312, respectively), we used microarray data set GSE70857 and defined the tissue-selective genes as above.

Western blotting.

Tissues were lysed in radioimmunoprecipitation assay (RIPA) buffer containing 0.1% SDS, 1% NP-40, 0.5% Na deoxycholate, 150 mM NaCl, 50 mM Tris-Cl (pH 8.0), 1 mM EDTA supplemented with protease inhibitor (Roche). Proteins were separated by SDS–PAGE, transferred to nitrocellulose membrane, and detected with the antibodies anti-NFIA (1:500 dilution, Santa Cruz Biotechnology, sc-133816), anti-UCP1 (1:2,000 dilution, Abcam, ab10983), anti-PPARγ (1:500 dilution, Santa Cruz Biotechnology, sc-7196), anti-β actin (1:1,000 dilution, Santa Cruz Biotechnology, sc-1616), FLAG M2 (1:2,000 dilution, Sigma, F3165) and V5 (1:5,000 dilution, Invitrogen, R960-25).

Co-immunoprecipitation.

HEK293 cells were transfected with pcDNA 3.1-V5-PRDM16, pcDNA 3.1-FLAG-PPARγ or pcDNA 3.1-NFIA expression vector as indicated in the figures and figure legends. pcDNA 3.1-V5-PRDM16 and -FLAG-PPARγ vectors were gifts from C. Villanueva17 (Department of Biochemistry, University of Utah School of Medicine, USA). Two days after transfection, cells were lysed using RIPA buffer supplemented with protease inhibitor. Dynabeads protein A (Thermo Fisher) were incubated with V5 antibody (2 μg per IP, Invitrogen) at 4 °C for 2 h, and then the lysate and antibody–beads complex were incubated at 4 °C overnight. The beads were washed with RIPA buffer five times. Eluted proteins were analysed by western blotting as described above.

Oxygen consumption assay.

Control or NFIA-expressing C2C12 myoblasts were plated on gelatin-coated XF24 culture microplates (Seahorse Bioscience), grown to confluence and treated with an adipogenic cocktail. At day 7 of differentiation the medium was replaced with XF24 assay medium supplemented with 1 mM sodium pyruvate and 25 mM glucose. The oxygen consumption rate was measured using an XF24 flux analyser (Seahorse Bioscience). The cells were treated with 1 μM oligomycin, 0.5 μM FCCP and 1 μM antimycin/rotenone in succession.

Animal studies.

All animal work was conducted according to the institutional guidelines at The University of Tokyo. NFIA-KO mice (stock number: 010318-UNC)23 were purchased from MMRRC (Mutant Mouse Regional Resource Center). The founder of this stock was 129S6 and subsequently was backcrossed to C57BL6/J for more than 20 generations. All of the experiments were performed using male mice. The age of the mice for each experiment is indicated in the main text or figure legends.

Human studies.

Perirenal BAT samples were obtained from eleven patients with pheochromocytoma and seven with non-functioning adrenal tumours, as previously described25. All procedures were approved by the Hiroshima University Ethics Committee. Supraclavicular brown adipocytes and abdominal subcutaneous white adipocytes were obtained by head or neck tumour surgery and gall bladder surgery, respectively, as previously described26. All procedures were approved by the Scientific-Ethics Committees of the Capital Region and of Copenhagen and Frederiksberg Municipalities Denmark, journal numbers H-A-2009-020, H-A-2008-081 and (KF) 01-141/04, respectively. BAT and WAT samples from necks were obtained during thyroidectomy or anterior cervical spine surgery, as previously described27. All procedures were approved by the Human Studies Institutional Review Boards of Beth Israel Deaconess Medical Center, Joslin Diabetes Center and Massachusetts General Hospital. The entire study was approved by the research ethics committee of the Graduate School of Medicine, the University of Tokyo. All of the procedures described above were conducted according to the Declaration of Helsinki, and all of the patients gave written informed consent before taking part in the study.

Human adipocyte culture and differentiation.

At the confluence, preadipocytes were induced to undergo adipocyte differentiation and cultured for 12 days, as previously described26. Samples were collected when the cells were 45–65% confluent and the cells were fully differentiated (12 days after inducing differentiation).

Statistics and reproducibility.

At least five mice were used for all of the animal studies. This group size was based on our previous studies. Randomization was not performed, and the investigators were not blinded to mouse genotype or type of the human samples. No samples were excluded for analysis. Two-tailed Student’s t-test was performed to determine the statistical significance between two groups unless otherwise specified, with a P value of less than 0.05 considered significant. We checked that the data met the assumption of the statistic tests, and variances were similar between the groups being tested. Experiments independently repeated three or more times were Figs 1d, e, 2a–e, g–j, 3b–k, 6a–g and 7a, b and Supplementary Figs 2c–j, 5a, b and 6a–d. Experiments independently performed two times were Figs 1f, 2f, k, 3a and 7c, d and Supplementary Figs 1c–e, 4a–d, 5g and 6e–l. Experiments or analyses performed once were Fig. 1a–c (FAIRE-seq), Fig. 4a–f (ChIP-seq and ATAC-seq), Fig. 5a–c (ChIP-seq and ATAC-seq), Fig. 7e–g(RNA-seq) and Fig. 8a–d (experiments using human samples) and Supplementary Fig. 1a, b (RNA-seq), Supplementary Fig. 2a, b (RNA-seq), Supplementary Fig. 3a–m (ChIP-seq), Supplementary Fig. 5c–f (RNA-seq) and Supplementary Fig. 7a, b (FAIRE-seq and ChIP-seq).

Code availability.

The in-house software for a heat map representation of ChIP-seq is available from the corresponding authors on request.

Data availability.

High-throughput sequencing data have been deposited at the Gene Expression Omnibus (GEO) under the accession number GSE83764. Unprocessed original scans of western blot analysis are shown in Supplementary Fig. 8. Source data for Fig. 6d, e are provided in Supplementary Table 2. All other data supporting the findings of this work are available from the corresponding authors on reasonable request.

Additional Information

Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Accessions

Primary accessions

Gene Expression Omnibus

References

  1. 1.

    & Brown adipose tissue: function and physiological significance. Physiol. Rev. 84, 277–359 (2004).

  2. 2.

    , & Unexpected evidence for active brown adipose tissue in adult humans. Am. J. Physiol. Endocrinol. Metab. 293, E444–E452 (2007).

  3. 3.

    et al. Cold-activated brown adipose tissue in healthy men. N. Engl. J. Med. 360, 1500–1508 (2009).

  4. 4.

    et al. Identification and importance of brown adipose tissue in adult humans. N. Engl. J. Med. 360, 1509–1517 (2009).

  5. 5.

    et al. Functional brown adipose tissue in healthy adults. N. Engl. J. Med. 360, 1518–1525 (2009).

  6. 6.

    et al. High incidence of metabolically active brown adipose effects of cold exposure and adiposity. Diabetes 58, 1526–1531 (2009).

  7. 7.

    et al. Cold acclimation recruits human brown fat and increases nonshivering thermogenesis. J. Clin. Invest. 123, 3395–3403 (2013).

  8. 8.

    et al. Short-term cold acclimation improves insulin sensitivity in patients with type 2 diabetes mellitus. Nat. Med. 21, 6–10 (2015).

  9. 9.

    et al. Activation of human brown adipose tissue by a β3-adrenergic receptor agonist. Cell Metab. 21, 33–38 (2015).

  10. 10.

    et al. PRDM16 controls a brown fat/skeletal muscle switch. Nature 454, 961–967 (2008).

  11. 11.

    et al. Transcriptional control of brown fat determination by PRDM16. Cell Metab. 6, 38–54 (2007).

  12. 12.

    et al. PRDM16 binds MED1 and controls chromatin architecture to determine a brown fat transcriptional program. Genes Dev. 29, 298–307 (2015).

  13. 13.

    , , & PPARγ agonists induce a white-to-brown fat conversion through stabilization of PRDM16 protein. Cell Metab. 15, 395–404 (2012).

  14. 14.

    et al. Genome-wide profiling of peroxisome proliferator-activated receptor γ in primary epididymal, inguinal, and brown adipocytes reveals depot-selective binding correlated with gene expression. Mol. Cell. Biol. 32, 3452–3463 (2012).

  15. 15.

    et al. EBF2 determines and maintains brown adipocyte identity. Cell Metab. 17, 562–574 (2013).

  16. 16.

    , , , & FAIRE (Formaldehyde-Assisted Isolation of Regulatory Elements) isolates active regulatory elements from human chromatin. Genome Res. 17, 877–885 (2007).

  17. 17.

    et al. Adipose subtype-selective recruitment of TLE3 or Prdm16 by PPARγ specifies lipid storage versus thermogenic gene programs. Cell Metab. 17, 423–435 (2013).

  18. 18.

    et al. Initiation of myoblast to brown fat switch by a PRDM16-C/EBP-β transcriptional complex. Nature 460, 1154–1158 (2009).

  19. 19.

    Roles of the NFI/CTF gene family in transcription and development. Development 249, 31–45 (2000).

  20. 20.

    , , , & Transposition of native chromatin for fast and sensitive epigenomic profiling of open chromatin, DNA-binding proteins and nucleosome position. Nat. Methods 10, 1213–1218 (2013).

  21. 21.

    et al. Simple combinations of lineage-determining transcription factors prime cis-regulatory elements required for macrophage and B cell identities. Mol. Cell 38, 576–589 (2010).

  22. 22.

    et al. Master transcription factors determine cell-type-specific responses to TGF-β signaling. Cell 147, 565–576 (2011).

  23. 23.

    et al. Disruption of the murine nuclear factor I-A gene (Nfia) results in perinatal lethality, hydrocephalus, and agenesis of the corpus callosum. Proc. Natl Acad. Sci. USA 96, 11946–11951 (1999).

  24. 24.

    & Brown and beige fat: development, function and therapeutic potential. Nat. Med. 19, 1252–1263 (2013).

  25. 25.

    et al. Activation of classical brown adipocytes in the adult human perirenal depot is highly correlated with PRDM16-EHMT1 complex expression. PLoS ONE 10, 1–13 (2015).

  26. 26.

    et al. A classical brown adipose tissue mRNA signature partly overlaps with brite in the supraclavicular region of adult humans. Cell Metab. 17, 798–805 (2013).

  27. 27.

    et al. Anatomical localization, gene expression profiling and functional characterization of adult human neck brown fat. Nat. Med. 19, 635–639 (2013).

  28. 28.

    , & Stimulation of adipogenesis in fibroblasts by PPAR γ2, a lipid-activated transcription factor. Cell 79, 1147–1156 (1994).

  29. 29.

    et al. Regulation of CSF1 promoter by the SWI/SNF-like BAF complex. Cell 106, 309–318 (2001).

  30. 30.

    & Pioneer transcription factors: establishing competence for gene expression. Genes Dev. 25, 2227–2241 (2011).

  31. 31.

    et al. Genetic variation determines PPARγ function and anti-diabetic drug response in vivo. Cell 162, 33–44 (2015).

  32. 32.

    et al. Large-scale identification of sequence variants influencing human transcription factor occupancy in vivo. Nat. Genet. 47, 1393–1401 (2015).

  33. 33.

    et al. Systematic dissection of genomic features determining transcription factor binding and enhancer function. Proc. Natl Acad. Sci. USA 114, E1291–E1300 (2017).

  34. 34.

    , & The genetics of transcription factor DNA binding variation. Cell 166, 538–554 (2016).

  35. 35.

    et al. Prdm16 is required for the maintenance of brown adipocyte identity and function in adult mice. Cell Metab. 19, 593–604 (2014).

  36. 36.

    et al. Ebf2 is a selective marker of brown and beige adipogenic precursor cells. Proc. Natl Acad. Sci. USA 111, 14466–14471 (2014).

  37. 37.

    , , , & Antagonism between the master regulators of differentiation ensures the discreteness and robustness of cell fates. Mol. Cell 54, 526–535 (2014).

  38. 38.

    , , & Identification of a potent adipocyte-specific enhancer: involvement of an NF-1-like factor. Genes Dev. 5, 428–437 (1991).

  39. 39.

    et al. Global mapping of cell type-specific open chromatin by FAIRE-seq reveals the regulatory role of the NFI family in adipocyte differentiation. PLoS Genet. 7, e1002311 (2011).

  40. 40.

    et al. Dissecting the brown adipogenic regulatory network using integrative genomics. Sci. Rep. 7, 42130 (2017).

  41. 41.

    et al. Nuclear factor I-C reciprocally regulates adipocyte and osteoblast differentiation via control of canonical Wnt signaling. FASEB J. 31, 1939–1952 (2017).

  42. 42.

    et al. Epigenome-wide association study of body mass index, and the adverse outcomes of adiposity. Nature 541, 81–86 (2017).

  43. 43.

    , , & Differential roles of insulin receptor substrates in the anti-apoptotic function of insulin-like growth factor-1 and insulin. J. Biol. Chem. 277, 31601–31611 (2002).

  44. 44.

    , , , & EHMT1 controls brown adipose cell fate and thermogenesis through the PRDM16 complex. Nature 504, 163–167 (2013).

  45. 45.

    et al. Model-based analysis of ChIP-Seq (MACS). Genome Biol. 9, R137 (2008).

  46. 46.

    et al. Cistrome: an integrative platform for transcriptional regulation studies. Genome Biol. 12, R83 (2011).

  47. 47.

    & Inferring direct DNA binding from ChIP-seq. Nucleic Acids Res. 40, 1–10 (2012).

  48. 48.

    et al. TRANSFAC: an integrated system for gene expression regulation. Nucleic Acids Res. 28, 316–319 (2000).

  49. 49.

    , & Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat. Protoc. 4, 44–57 (2009).

  50. 50.

    , & Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 15, 550 (2014).

  51. 51.

    et al. GenePattern 2.0. Nat. Genet. 38, 500–501 (2006).

Download references

Acknowledgements

We are grateful to S. Mandrup for her suggestion about our study. We thank K. Shiina and K. Tatsuno for their help in library preparation for high-throughput sequencing, S. Fukuda and T. Umehara for their help in computational analysis, K. Ueki, S. Kajimura and C. Villanueva for providing cells and plasmids, K. Nakashima for his suggestion regarding NFIA-KO mice, and T. Sugiyama, T. Kubota and N. Kubota for their help in animal experiments. We also thank T. Wada for his technical assistance. This work is funded by an AMED-CREST research grant from the Japan Agency for Medical Research and Development (AMED) to T.Y.; by a grant-in-aid for scientific research (B) from the Japan Society for the Promotion of Science (JSPS), grant number 25293209 to H.W.; by a grant-in-aid for JSPS fellows from JSPS, grant number 15J02835 to Y.Hiraike; and by a junior scientist development grant from the Japan Diabetes Society to Y.Hiraike. Y.Hiraike has been supported by a research fellowship from JSPS. The Centre of Inflammation and Metabolism (CIM) and the Centre for Physical Activity Research (CFAS), Department of Infectious Diseases, Rigshospitalet is supported by a grant from TrygFonden. CIM/CFAS is a member of DD2—the Danish Center for Strategic Research in Type 2 Diabetes (the Danish Council for Strategic Research, grant no. 09-067009 and 09-075724). T.J.L. has been supported by a research grant from the Danish Diabetes Academy supported by the Novo Nordisk Foundation. The Novo Nordisk Foundation Center for Basic Metabolic Research is supported by an unconditional grant from the Novo Nordisk Foundation to University of Copenhagen.

Author information

Author notes

    • Yuta Hiraike
    •  & Hironori Waki

    These authors contributed equally to this work.

Affiliations

  1. Department of Diabetes and Metabolic Diseases, Graduate School of Medicine, The University of Tokyo, Tokyo 113-8655, Japan

    • Yuta Hiraike
    • , Hironori Waki
    • , Jing Yu
    • , Masahiro Nakamura
    • , Kana Miyake
    • , Ken Suzuki
    • , Hirofumi Kobayashi
    • , Wei Sun
    • , Tomohisa Aoyama
    • , Yusuke Hirota
    • , Toshimasa Yamauchi
    •  & Takashi Kadowaki
  2. Department of Molecular Science on Diabetes, Graduate School of Medicine, The University of Tokyo, Tokyo 113-8655, Japan

    • Hironori Waki
  3. Department of Molecular and Internal Medicine, Graduate School of Biomedical and Health Sciences, Hiroshima University, Hiroshima 734-8551, Japan

    • Gaku Nagano
    • , Haruya Ohno
    • , Kenji Oki
    •  & Masayasu Yoneda
  4. Genome Science Division, Research Center for Advanced Science and Technology, The University of Tokyo, Tokyo 153-8904, Japan

    • Ryo Nakaki
    • , Shogo Yamamoto
    • , Shuichi Tsutsumi
    •  & Hiroyuki Aburatani
  5. Department of Orthopedic Surgery, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, Massachusetts 02215, USA

    • Andrew P. White
  6. Section on Integrative Physiology and Metabolism, Joslin Diabetes Center, Harvard Medical School, Boston, Massachusetts 02215, USA

    • Yu-Hua Tseng
  7. Translational Physiology Section, Diabetes, Endocrinology, and Obesity Branch, National Institute of Diabetes and Digestive and Kidney Diseases, NIH, Bethesda, Maryland 20892, USA

    • Aaron M. Cypess
  8. The Centre of Inflammation and Metabolism and the Centre for Physical Activity Research, Department of Infectious Diseases, Rigshospitalet, University of Copenhagen, DK-2100 Copenhagen, Denmark

    • Therese J. Larsen
    • , Naja Z. Jespersen
    •  & Camilla Scheele
  9. Danish Diabetes Academy, Odense University Hospital, DK-5000 Odense C, Denmark

    • Therese J. Larsen
  10. Novo Nordisk Foundation Center for Basic Metabolic Research, Section for Metabolic Receptology, University of Copenhagen, 2200 Copenhagen, Denmark

    • Naja Z. Jespersen
  11. Novo Nordisk Foundation Center for Basic Metabolic Research, University of Copenhagen, 2200 Copenhagen, Denmark

    • Camilla Scheele

Authors

  1. Search for Yuta Hiraike in:

  2. Search for Hironori Waki in:

  3. Search for Jing Yu in:

  4. Search for Masahiro Nakamura in:

  5. Search for Kana Miyake in:

  6. Search for Gaku Nagano in:

  7. Search for Ryo Nakaki in:

  8. Search for Ken Suzuki in:

  9. Search for Hirofumi Kobayashi in:

  10. Search for Shogo Yamamoto in:

  11. Search for Wei Sun in:

  12. Search for Tomohisa Aoyama in:

  13. Search for Yusuke Hirota in:

  14. Search for Haruya Ohno in:

  15. Search for Kenji Oki in:

  16. Search for Masayasu Yoneda in:

  17. Search for Andrew P. White in:

  18. Search for Yu-Hua Tseng in:

  19. Search for Aaron M. Cypess in:

  20. Search for Therese J. Larsen in:

  21. Search for Naja Z. Jespersen in:

  22. Search for Camilla Scheele in:

  23. Search for Shuichi Tsutsumi in:

  24. Search for Hiroyuki Aburatani in:

  25. Search for Toshimasa Yamauchi in:

  26. Search for Takashi Kadowaki in:

Contributions

Y.Hiraike and H.W. designed the study. Y.Hiraike., H.W., J.Y., K.M., H.K., W.S., Y.Hirota and T.A. performed experiments. M.N., R.N., K.S., S.Y., S.T., Y.Hiraike and H.W. performed computational analysis. G.N., H.O., K.O. and M.Y. performed analysis of human perirenal BAT. A.P.W., Y.-H.T. and A.M.C. performed analysis of human neck BAT and WAT. T.J.L., N.Z.J. and C.S. performed analysis of human brown and white adipocytes. Y.Hiraike and H.W. wrote the manuscript. H.A., T.Y. and T.K. supervised all aspects of this work.

Competing interests

The authors declare no competing financial interests.

Corresponding authors

Correspondence to Hironori Waki or Hiroyuki Aburatani or Toshimasa Yamauchi or Takashi Kadowaki.

Integrated supplementary information

Supplementary information

PDF files

  1. 1.

    Supplementary Information

    Supplementary Information

  2. 2.

    Supplementary Information

    Life Sciences Reporting Summary

Excel files

  1. 1.

    Supplementary Table 1

    Supplementary Information

  2. 2.

    Supplementary Table 2

    Supplementary Information