Activation of the adipocyte CREB/CRTC pathway in obesity

Obesity is a major risk factor for the development of type II diabetes. Increases in adipose tissue mass trigger insulin resistance via the release of pro-inflammatory cytokines from adipocytes and macrophages. CREB and the CRTC coactivators have been found to promote insulin resistance in obesity, although the mechanism is unclear. Here we show that high fat diet feeding activates the CREB/CRTC pathway in adipocytes by decreasing the expression of SIK2, a Ser/Thr kinase that phosphorylates and inhibits CRTCs. SIK2 levels are regulated by the adipogenic factor C/EBPα, whose expression is reduced in obesity. Exposure to PPARγ agonist rescues C/EBPα expression and restores SIK2 levels. CRTC2/3 promote insulin resistance via induction of the chemokines CXCL1/2. Knockout of CRTC2/3 in adipocytes reduces CXCL1/2 expression and improves insulin sensitivity. As administration of CXCL1/2 reverses salutary effects of CRTC2/3 depletion, our results demonstrate the importance of the CREB/CRTC pathway in modulating adipose tissue function.

O besity is associated with inflammatory changes in white adipose tissue (WAT) that lead to systemic insulin resistance and type 2 diabetes 1 . Sustained low-grade inflammation in this setting impairs triglyceride and glucose metabolism. Following their migration to WAT depots in response to circulating cytokines and chemokines, neutrophils promote infiltration of macrophages 2,3 , which in turn release TNFα and other pro-inflammatory cytokines that enhance insulin resistance [4][5][6] , in part via the induction of NF-κB. Indeed, proinflammatory cytokines have also been found to disrupt catecholamine signaling in adipocytes by stimulating phosphodiesterase PDE3B activity and blocking PKA-induced lipolysis 7 .
The cAMP pathway promotes cellular gene expression via the PKA-mediated stimulatory phosphorylation of CREB [8][9][10] and inhibitory phosphorylation of the salt inducible kinases (SIKs). SIKs are active under basal conditions, where they phosphorylate and sequester the cAMP Responsive Transcriptional Coactivators (CRTCs) in the cytoplasm through an association with 14-3-3 proteins 11 . SIK2 is the most highly expressed of the three SIK family members in adipose tissue; and its downregulation in adipocytes of obese individuals is thought to contribute to insulin resistance 12,13 . Supporting this notion, mice with a knockout of SIK2 are glucose intolerant and insulin resistant 14 .
Dephosphorylation of the CRTCs in response to cAMP stimulates their translocation to the nucleus, where they bind to CREB over relevant promoters. Although cAMP signaling is itself associated with reduced inflammation in adipose tissue 15,16 , the adipocyte CREB/CRTC pathway has paradoxically been found to enhance insulin resistance in obesity [17][18][19] . Indeed, the loss of cAMP/PKA signaling in obese adipose tissue 20 would be expected to block CREB/CRTC activation.
Here we examine the mechanism by which HFD feeding stimulates CRTC2 and CRTC3 activities in adipose tissue, leading to the downstream induction of a subset of pro-inflammatory genes in conjunction with NF-κB. Our results demonstrate an unexpected link between CREB/CRTC and cytokine signaling pathways in modulating adipose tissue function.

Results
Disruption of cAMP/PKA signaling and adipogenic gene expression in obesity. We evaluated the effects of obesity on gene expression in WAT by feeding C57BL/6J mice a 60% HFD for 15 weeks (Fig. 1a, b). Body weight and fasting blood glucose levels were elevated in HFD-fed relative to NCD-fed mice (Supplementary Fig. 1a). Gene ontology analysis of RNA-seq data from epidydimal WAT (eWAT) mRNA, revealed upregulation of "inflammatory response" genes and downregulation of "glucose and lipid metabolism" genes in response to HFD (Fig. 1a). Metabolic genes with reduced expression under HFD conditions include the GLUT4 glucose transporter (SLC2A4), the Beta 3 Adrenergic receptor (ADRB3), salt inducible kinases (SIK2, SIK3) as well as CAAT/enhancer-binding protein alpha (C/EBPα). The expression of these HFD-downregulated genes is inversely correlated with insulin resistance and obesity in mice and humans 12,[21][22][23][24] .
In keeping with the induction of pro-inflammatory genes in response to HFD 1,25,26 , NF-κB target genes such as c-c motif chemokine ligand 2 (CCL2; also referred to as MCP1), IFN responsive regulator of G protein signaling 1 (RGS1), and prostaglandin E2 receptor 3 (PTGER3; also referred to as EP3) were all upregulated in response to HFD feeding (Fig. 1b, c).
Adipose tissue contains a number of cell types, including immune and endothelial cells as well as pre-adipocytes and mature adipocytes. To identify specific subsets of cells that display similar profiles of metabolic and inflammatory gene expression as adipose tissue in response to HFD, we performed fluorescenceactivated cell sorting (FACS) studies. Immune and endothelial cells from eWAT showed little change in either inflammatory or metabolic gene expression under HFD vs. NCD conditions; but pre-adipocytes and mature adipocytes showed profiles of both up-and downregulated genes that mimicked those of intact adipose tissue ( Supplementary Fig. 1e, f).
HFD feeding has been found to block cAMP signaling in adipocytes through the IκB Kinase (IKK)-mediated phosphorylation and activation of the phosphodiesterase PDE3B 7 . In line with this observation, PKA activity was decreased in HFD adipose tissue of WT mice, by western blot assay of both eWAT and iWAT using phospho-PKA substrate antibody as well as antisera against PKA-phosphorylated Hormone Sensitive Lipase (HSL) and phospho-AMPK 27 (Fig. 1d and Supplementary  Fig. 1b, c).
Effect of obesity on SIK2 expression in adipose tissue. The Ser/ Thr kinase SIK2 is the most highly expressed of the three SIK family members in adipocytes, where it promotes triglyceride homeostasis and whole-body insulin sensitivity 12,14,[28][29][30] . SIK activity is dependent on phosphorylation (at Thr175) by LKB1, a master kinase for AMPK family members. Triggering of the cAMP pathway blocks SIK2 activity via PKA-mediated phosphorylation 31 . Recognizing that SIK2 inhibits CREB/CRTC signaling, under basal conditions, by phosphorylating CRTCs at 14-3-3 binding sites 11 , we evaluated the phosphorylation status of these proteins in adipose tissue.
Consistent with the decrease in total as well as phospho (Thr175)-specific 28 SIK2 protein amounts (Fig. 1d), CRTC2 and CRTC3 are dephosphorylated to a greater extent in eWAT as well as iWAT from HFD-fed compared to NCD-fed mice ( Fig. 1d and Supplementary Fig. 1c). As a result, CRTC2/3 nuclear staining is increased in eWAT from HFD-fed mice, by immunohistochemical analysis ( Fig. 1e and Supplementary Fig. 1d). These results indicate that the loss of SIK2 expression in obese WAT is sufficient to promote increases in nuclear CRTC2/3.
Obesity triggers the expression and release of the proinflammatory cytokine TNFα from adipocytes and resident macrophages in adipose tissue, where it interferes with insulin signaling and triglyceride metabolism 4,32 . We explored the potential role of this cytokine in modulating SIK2 expression. Exposure of primary adipocytes or 3T3-L1 adipocytes to TNFα (6 h) decreased mRNA amounts for both SIK2 and C/EBPα (Fig. 2a).
Within the SIK2 promoter, we noted consensus binding sites for C/EBPα and CREB. Overexpression of CRTC2/3 had only modest effects on SIK2 promoter activity in transient assays using a SIK2 luciferase reporter, but C/EBPα overexpression strongly upregulated this reporter in HIB1b cells (Fig. 2b). In keeping with the ability for C/EBPα to auto-regulate its own promoter, C/EBPα overexpression also potentiated the activity of a C/EBPα reporter construct (Fig. 2c); these effects were further enhanced by cotreatment with the PPARγ agonist Rosiglitazone (Rosi). By contrast, overexpression of p65, the trans-activating subunit of the NF-κB hetero-dimeric complex 33 , inhibited C/EBPα reporter activity in HIB1b cells (Fig. 2c).
In keeping with the effects of C/EBPα on SIK2 reporter activity, exposure of 3T3-L1 cells to differentiation medium increased endogenous C/EBPα protein amounts, followed by increases in SIK2 expression (Fig. 2d). Modest C/EBPα overexpression in HIB1b cells rescued TNFα-induced decreases in SIK2 expression. As a result, phosphorylation of endogenous CRTC2/3 increased in cells over-expressing C/EBPα (Fig. 2e).
HFD feeding stimulated the expression and release of CXCL1/2 from adipose tissue into the circulation ( Fig. 1c and Supplementary Fig. 3a), leading to increases in neutrophil migration and infiltration of eWAT (Fig. 3a). Consistent with this scenario, exposure of 3T3-L1 adipocytes or primary adipocytes to FSK also induces CXCL1 expression (Fig. 3b). Indeed, exposure of 3T3-L1 adipocytes to a more physiologic cAMP stimulus (β3 adrenergic receptor agonist; CL316,243) also augments CXCL1 expression, albeit more modestly than FSK ( Supplementary Fig. 3b).
In keeping with the role of SIK2 in suppressing CRTC activity, exposure to SIK inhibitor HG upregulated CXCL1/2 expression comparably to FSK (Fig. 3c). SIK inhibition also increased the expression of other inflammatory genes (IL6) that have CREB and NF-κB binding sites on their promoters but not genes like MCP1, an NF-κB target lacking a CREB binding site.
Recognizing that the CXCL1 promoter contains two cAMP response elements (CREs) as well as an NF-κB binding site, we evaluated the extent to which these regulatory pathways modulate the expression of this gene. Exposure to either FSK or TNFα increased CXCL1 reporter activity 2-3 fold, while exposure to both FSK and TNFα increased reporter activity synergistically (10-fold) ( Fig. 3d and Supplementary Fig. 3c). We also observed cooperativity between NF-κB and CREB/CRTC pathways on a human CXCL8 (IL8) reporter in transient assays of HEK293T cells ( Supplementary  Fig. 3d). Indeed, co-expression of CRTC2/3 and p65 potentiated effects of FSK and TNFα on IL8 reporter activity. These results indicate that NF-κB and CREB/CRTC pathways co-regulate the expression of CXCL1/2 and perhaps other genes in adipocytes.
TNFα and other cytokines stimulate pro-inflammatory gene expression in part through induction of the inhibitor of nuclear factor κB kinase (IKK), which phosphorylates IκBs and thereby promotes their degradation 45,46 . Exposure to IKK inhibitor (IKK16) blocked the induction of the CXCL1 and CXCL2 genes in response to TNFα. Remarkably, IKK16 also interfered with induction of the CXCL1 and CXCL2 genes by SIK inhibitor (HG), suggesting that IKK activity is necessary for the cooperativity between NFκB and CREB pathways ( Supplementary Fig. 3e).
We wondered whether FSK and TNFα exert distinct or overlapping effects on p65 and CRTC2 recruitment to the CXCL1 promoter. In chromatin-immunoprecipitation sequencing (ChIPseq) studies of differentiated 3T3-L1 adipocytes ( Fig. 3e and Supplementary Fig. 4), CRTC2 occupancy over the CXCL1 promoter is low under basal conditions but increases following exposure (1 h) to FSK. In line with its ability to stimulate CRTC2/ 3 nuclear translocation, treatment with TNFα also promotes CRTC2 recruitment to the CXCL1 promoter ( Fig. 3e). Moreover, exposure to both FSK and TNFα further enhances CRTC2 as well as p65 occupancy over their respective binding sites. CRTC2 occupancy extends through the CXCL1 gene body, suggesting that it may contribute to both the transcriptional initiation and productive elongation of RNA polymerase II complexes along this gene.
In addition to their effects on CXCL1, FSK and TNFα increased the occupancy of both CRTC2 and p65 at 367 co-bound loci (Fig. 3f). 70 of these loci have CRE(s) (TGACG), and 86 have NFκB motif(s) (GGAAWTTCCC), while 12 have both consensus CRE and NFκB binding sites. Some co-bound promoters (e.g., CCL2, CXCL5) lack consensus CREs, but they contain Jun/AP1 binding sites, which also appear capable of mediating recruitment of CRTCs to the promoter, potentially via an association with jun/fos family members 47 ( Supplementary Fig. 4a-c). Consistent with this notion, CCL2 and CXCL5 promoters lack consensus CREs, yet they are bound by CRTC2; and these genes are downregulated in dAKO adipocytes relative to wild-type adipocytes exposed to TNFα ( Supplementary Fig. 4d).

Knockout of CRTC2/3 improves adipose function in obesity.
Of the three CRTC family members, CRTC2 and CRTC3 are expressed at the highest levels in adipose tissue relative to CRTC1, which is preferentially expressed in the brain. To determine the role of CRTCs in adipose tissue, we crossed mice with floxed alleles of both CRTC2/3 with Adipoq-Cre transgenic mice expressing Cre recombinase under the control of the adiposespecific adiponectin promoter. CRTC2/3 expression in WAT is reduced by more than 70% in CRTC2/3 double knockout (dAKO) mice by qPCR and western blotting analysis ( Fig. 4d and Supplementary Fig. 5a, b). Based on the proposed effects of CRTC2/3 on CXCL1/2 expression and insulin resistance, we evaluated whether depletion of CRTC2/3 in adipose tissue modulates the effects of HFD on insulin signaling and glucose metabolism.
Although they were indistinguishable from WT littermates under NCD conditions, dAKO mice gained less weight and had a lower fat mass in the context of HFD feeding, beginning after 4 weeks of HFD ( Fig. 4a and Supplementary Fig. 5c). HFD-fed dAKO mice have lower fasting blood glucose levels; glucose and TNFα activates CRTCs by decreasing c/EBPα and SIK2 expression in adipocytes. a Effect of TNFα on SIK2 and C/EBPα mRNA amounts in differentiated primary white adipocytes and 3T3-L1 adipocytes. TNFα (10 ng/ml) treatment for indicated times (**P < 0.01, *P < 0.05, one-way ANOVA, t-test; n = 3 per group). Data from experiments repeated three or more times. b Relative effect of CRTC2/3 (C2C3) and C/EBPα overexpression on mouse SIK2 promoter activity in HIB1b cells shown. Luciferase activity normalized to RSV-β gal activity (**P < 0.01, t-test; n = 3 per group). c Luciferase assay showing the effect of c/EBPα and NF-κB p65 overexpression on c/EBPα promoter activity in HIB1b cells. Luciferase activity normalized to RSV-β gal activity (**P < 0.01, one-way ANOVA; n = 3 per group). Treatment with PPARγ agonist Rosiglitazone (Rosi;1 µM, 16 h) indicated. d Immunoblot showing relative time course for SIK2 and c/EBPα expression in differentiating 3T3-L1 cells. e Immunoblots showing effects of C/EBPα overexpression on endogenous SIK2 protein levels in HIB1b cells exposed to TNFα. Cells were incubated with lentiviral C/EBPα or control vector for 2 days. Infected cells were then exposed to TNFα (T, 10 ng/ml) for 6 h. Densitometry analysis of SIK2 expression and CRTC phosphorylation shown. f Effect of PPARγ agonist (Rosi) on endogenous SIK2 protein amounts and CRTC phosphorylation in 3T3-L1 cells. Undifferentiated cells exposed to Rosi (5 µg/ml) for 24 h and HG9-91-01 (HG;10 µM) for the final 1 h. Immunoblots show C/EBPα and SIK2 protein amounts as well as phosphorylated forms of CRTC2/3. g Effect of TNFα (10 ng/ml, 6 h) on CRTC nuclear translocation in 3T3-L1 adipocytes. Immunoblots show CRTC2/3 protein amounts in nuclear and cytoplasmic fractions. p-CRTC and SIK2 amounts in whole-cell lysates indicated. Each lane represents one plate of cells. Right, bar graph showing densitometry analysis of fractionation study (**P < 0.01, t-test; n = 3 per group). Numbers below each lane in d-f indicate relative expression levels. Data in a-c represent mean ± SD and data in g represent mean ± SEM.
insulin tolerance is also improved in dAKO relative to WT (Fig. 4b, c and Supplementary Fig. 5d).
We performed indirect calorimetry studies to further characterize energy metabolism in dAKO mice ( Supplementary  Fig. 5e). Consistent with their lower body weight on HFD, dAKO mice have elevated oxygen consumption rate and energy expenditure relative to WT littermates. Despite these differences, physical activity and food intake appear comparable between CRTC2/3 mutant and WT littermates.
In keeping with the effects of CRTC2/3 on CREB target gene expression in adipose tissue, mRNA amounts for CXCL1 and CXCL2 in WAT are downregulated in dAKO mice relative to WT (Fig. 4d). Correspondingly, neutrophil numbers in adipose tissue from HFD-fed dAKO mice are 2-fold lower than in HFD-fed WT adipose tissue. mRNA amounts for neutrophil (CD11b) and M1 macrophage (F4/80) markers are decreased in dAKO WAT, suggesting that a reduction in CXCL1/2 mediated neutrophil recruitment also attenuates adipose tissue infiltration by other inflammatory cells.
In line with the improvements in adipose tissue inflammation, cAMP and insulin signaling pathways were more active in dAKO mice relative to wild-type ( Supplementary Fig. 5f, g). Metabolic (ADRB3, GLUT4) gene expression was also increased in eWAT from dAKO mice relative to WT (Fig. 4d). c/EBPα protein amounts are elevated in dAKO eWAT, and SIK2 expression is correspondingly higher.
CXCL1/2 mediate effects of CRTC1/2 on HFD-induced inflammation. Based on the effects of HFD in promoting cytokine gene expression, we wondered whether CRTC2/3 depletion would attenuate the reactivity of WAT tissue to an inflammatory stimulus. Intra-peritoneal administration of low-dose lipopolysaccharide (LPS; 500 ng/kg, 3 h) stimulated the expression of inflammatory genes that are targets of both CRTC and NFκB (CXCL1, CXCL2, IL6) or NFκB alone (TNFα, IL1β, MCP1) in WT eWAT. Depletion of CRTC2/3 reduced expression of NFκB/ CRTC coregulated targets and to a lesser extent NFκB selective targets in dAKO eWAT ( Supplementary Fig. 6). Collectively, these studies suggest that CRTC2/3 acts upstream of NFκB to promote the expression of cytokine genes in adipose tissue.
Having seen that CXCL1/2 expression is reduced in dAKO mice, we evaluated the extent to which these chemokines contribute to the effects of HFD on inflammation and lipid metabolism. Administration of CXCL1/2 neutralizing antiserum during the HFD feeding regimen reduces circulating concentrations of CXCL1/2 proteins to a similar extent in WT mice as those in dAKO mice ( Supplementary  Fig. 7a). Neutrophil and macrophage infiltration into adipose tissue is downregulated by immunohistochemical analysis of eWAT sections from immuno-neutralized mice; and glucose tolerance is correspondingly improved (Fig. 5a, b). Indeed, neutrophil and macrophage marker gene expression (CD11b, F4/80, MCP1) is also reduced in eWAT from anti-CXCL1/2 immuno-neutralized mice compared with control (Fig. 5c). In RNA-seq studies of eWAT, most of the gene clusters that are upregulated in HFD versus NCD feeding correspond to immune response and chemokine signaling ( Supplementary Fig. 7d). mRNA amounts for these genes are decreased in both dAKO and anti-CXCL1/2 antiserum injected mice.
Consistent with this decrease in pro-inflammatory gene expression, insulin signaling, evaluated by relative pSer(473) AKT and pSer(307)IRS1 protein amounts, is enhanced while stress signaling (pThr(183)JNK) is attenuated in eWAT from CXCL1/2-neutralized mice (Fig. 5d). As a result, circulating triglyceride and free fatty acid concentrations are decreased in HFD-fed dAKO and CXCL1/2 immuno-neutralized mice relative to WT. Indeed, lipid accumulation in the liver is reduced in dAKO mice (Fig. 5e and Supplementary Fig. 7b, c). Collectively, these results indicate that the induction of CXCL1/2 by CRTC2/3 in adipose tissue during HFD feeding, contributes in part to the development of insulin resistance.

Effect of CXCL1 knockout on diet-induced insulin resistance.
Based on the effects of CXCL1/2 immuno-neutralization on insulin signaling, we employed CXCL1 KO mice to evaluate whether this chemokine contributes to HFD-induced insulin resistance (Fig. 6). Under HFD conditions, CXCL1 KO mice had lower body weight and a tendency toward lower fat mass (P = 0.07, Fig. 6a). Glucose metabolism was improved in CXCL1 KO mice relative to WT littermates by intraperitoneal glucose (IPGTT) and insulin tolerance testing (ITT) (Fig. 6b and Supplementary Fig. 8a). Indeed, CXCL1 KO mice had lower circulating concentrations of triglyceride and free fatty acids relative to WT ( Fig. 6c and Supplementary Fig. 8b, c). Consistent with an improvement in lipid metabolism, triglyceride stores in BAT and liver are reduced in HFD fed CXCL1 KO mice relative to WT (Fig. 6c).
Inflammatory gene expression (F4/80, TNFα, CD11b, IL1β) is also downregulated in CXCL1 KO adipose tissue, while metabolic gene expression (ADRB3, Glut4, AdipoQ) is elevated. These salutary effects are associated with increases in insulin signaling (p-AKT) and adipogenic (c/EBPα) gene expression (Fig. 6e). In keeping with the stimulatory effects of c/EBPα on the SIK2 gene, protein levels for SIK2 are also increased in eWAT from HFD-fed CXCL1 KO relative to WT mice. Collectively, these results demonstrate that CXCL1 contributes to the inflammatory and metabolic changes that accompany HFD feeding ( Fig. 6d and Supplementary Fig. 8d).
To address the potential role of CXCL1/2 in promoting the migration of innate immune cells, we performed in vitro chemotaxis assays using the macrophage/monocyte cell line Raw 264.7. Exposure to conditioned medium from unstimulated WT adipocytes promoted macrophage migration (Fig. 7c); these effects were further potentiated by exposure to conditioned medium from WT adipocytes treated with TNFα. dAKO Relative mRNA levels WT dAKO Conversely, the addition of rCXCL1/2 increased macrophage migration even in conditioned medium from dAKO cells. Taken together, these results demonstrate that CXCL1/2 acts downstream of CREB/CRTC and NFκB pathways to promote insulin resistance in obesity.

Discussion
Obesity is a major risk factor for the development of insulin resistance and type II diabetes. High-fat diet feeding stimulates the release of cytokines from adipocytes, which in turn recruit innate immune cells that disrupt fat storage and block insulin signaling 4 . TNFα and other cytokines are thought to mediate the effects of HFD in promoting the expression of inflammatory genes and by inhibiting metabolic gene expression via the induction of NF-κB 21,45 . We found that the NF-κB pathway stimulates a subset of inflammatory genes cooperatively with CREB/ CRTC. Although CREB and CRTCs are typically activated by cAMP, HFD feeding appears to induce this pathway in adipocytes by decreasing the expression of SIK2. Indeed, SIK2 expression and function are inversely correlated with insulin resistance and obesity in humans 12,29 . While inhibition of SIK2 in macrophages appears to promote the release of anti-inflammatory cytokines 48 , adipocyte SIK2 has been found to regulate Glut4 expression and insulin signaling 28 . Correspondingly, SIK2 −/− mice have lower glucose tolerance and reduced insulin sensitivity, due to increases in macrophage infiltration 14 . The extent to which CRTCs contribute to this phenotype is unclear, however.
Although it disrupts adipocyte cAMP signaling, HFD feeding paradoxically activates the CREB/CRTC pathway, in part via cytokine-mediated decreases in SIK2 expression that lead to CRTC dephosphorylation, nuclear entry, and association with CREB over relevant promoters. Restoring SIK2 expression in adipocytes, by overexpression of C/EBPα or by administration of PPARγ agonist, rescued the inhibitory phosphorylation of the CRTCs. Future studies should reveal the extent to which PPARγ may also modulate CREB/CRTC activity in other tissues.
Following their induction in response to HFD feeding, adipocyte CRTC2/3 stimulates the expression of CXCL1 and CXCL2 as well as other pro-inflammatory genes (IL6, LIF) in cooperation with NFκB. In turn, CXCL1/2 recruit neutrophils and macrophages to adipose tissue, where they promote insulin resistance (Fig. 7d). Loss of CXCL1/2 expression, by immune-depletion or by knockout (of CXCL1), restored insulin sensitivity and glucose tolerance. Indeed, these chemokines have also been shown to function importantly in human obesity and insulin resistance 39,42,44 as well as in cancer metastasis 49,50 .
Although they are indistinguishable from wild-type littermates under NCD conditions, dAKO mice weigh less and have lower fat mass than wild-type littermates in response to HFD feeding. In this regard, CREB has been reported to promote the expression of the adipogenic program by stimulating the expression of the c/ EBPβ gene 51 . c/EBPβ mRNA levels tend to be decreased in dAKO WAT but adipogenic gene expression appears to be relatively unaffected, likely reflecting compensation by the closely related family member c/EBPδ 52 . c/EBPα and PPARγ protein levels are elevated in dAKO relative to wild-type WAT. And late adipogenic markers (e.g., FABP4, LPL, PEPCK) are either equally expressed or upregulated in dAKO WAT. Further studies should reveal how crosstalk between adipogenic and inflammatory pathways contributes to the development of insulin resistance.
In addition to their effects on triglyceride storage in WAT, proinflammatory cytokines have been shown to interfere with thermogenesis in beige and brown fat 53 . We imagine that the loss of pro-inflammatory (CXCL1, CXCL2) gene expression in dAKO mice could therefore enhance the expression of thermogenic genes, either in beige adipocytes or in brown fat. Indeed, oxygen consumption and energy expenditure are elevated in dAKO mice. Moreover, CXCL1 KO mice have reduced body weight and tend to have lower fat mass compared to wild-type littermates (Fig. 6a). Thermogenesis may be more active in dAKO mice, although the modulatory effects of CRTC3 alone in brown fat appear to be largely developmental 54 .
Knockout of CRTC2/3 not only reduces CXCL1/2 expression but also blunts the effects of NFκB activation on proinflammatory genes. These results suggest that CRTCs may prime certain promoters for NFκB recruitment through epigenetic mechanisms. CRTC2/NFκB co-bound peaks appear to be particularly enriched in AP1 binding sites, which are typically recognized by Jun/Fos but not CREB family members. Supporting this idea, CRTC1 has been reported to associate with AP1 and to mediate induction of target gene expression 47 . Future studies should reveal the extent to which CRTCs modulate adipose tissue function through other nuclear factors in addition to CREB.

Materials and methods
Animals. All procedures involving the use of animals were performed in accordance with the guidelines presented by Salk Institute's Animal Care and Use Committee. C57BL/6J mice were purchased from The Jackson Laboratory (strain #000664). CRTC2/3 double-floxed (dflox) mice were produced by crossing Crtc2floxed mice 55 with Crtc3-floxed mice 54 . Adipose-specific CRTC2/3 knockout (dAKO) mice were generated by crossing CRTC2/3 dflox mice with Adiponectin-Cre mice (Jackson Laboratory; strain # 010803). CXCL1 knockout mice, C57BL/   4 Mice with a knockout of both CRTC2 and CRTC3 in adipose tissue are resistant to HFD-induced obesity. a Body weight gain and fat mass in WT and double (CRTC2/3) adipose KO (dAKO) littermates after HFD feeding for 12 weeks (**P < 0.01, t-test; n = 9, 12 per group). HFD started at 4-5 weeks of age. b, c Glucose (1 g/kg glucose, IP) (b) and Insulin (1 U/kg insulin) tolerance testing (c) of dAKO and WT littermates. Mice were fed an HFD for 12 weeks (**P < 0.01, *P < 0.05, t-test; n = 9, 12 per group). d Q-PCR analysis of mRNA amounts for immune-related and metabolic genes in adipose from dAKO and WT littermates under HFD conditions for 12 weeks (*P < 0.05, t-test; n = 9, 12 per group). e Circulating concentrations of CXCL1 in dAKO mice and WT littermates after 6 weeks HFD or NCD feeding (**P < 0.01, t-test; n = 3 per group). f Flow cytometry analysis showing neutrophil (CD11b + ; Ly6g + ) numbers in eWAT from dAKO and WT littermates after 6 weeks HFD (*P < 0.05, t-test; n = 5, 4 per group). g H&E and macrophage (F4/80) staining in eWAT tissue sections from WT and dAKO mice after HFD feeding for 12 weeks. Representative images shown from more than 3 mice per group. h Relative mRNA amounts for CXCL1 and CXCL2 in differentiated primary adipocytes from dAKO or WT mice. Exposure to FSK (10 µM, 2 h) indicated. i Effect of FSK (10 µM, 6 h) on the secretion of CXCL1 from differentiated primary WT or dAKO adipocytes. j Effect of TNFα (10 ng/ml, 2 h) on the expression of CXCL1 and CXCL2 genes in differentiated primary adipocytes from dAKO or WT mice. k Effects of TNFα (10 ng/ml, 12 h) on the secretion of CXCL1 from differentiated primary adipocytes of dAKO and WT littermates. Significance determined by one-way ANOVA for h-k (**P < 0.01, *<0.05, n = 3 per group). Data in a-f and i, k represent the mean ± SEM, and h, j represent the mean ± SD.
For High Fat Diet (HFD) feeding, 4-6-week-old mice were fed 60% HFD (Research Diet # D12492) for indicated periods with regular day/night cycles. For glucose tolerance testing (GTT), mice fed with a 60% high-fat diet were fasted overnight and injected with 1 g/kg of glucose/body weight. Blood glucose levels were monitored at indicated time points for 2 h. For insulin tolerance testing (ITT), mice were fasted for 4 h and injected with 1 U/kg of insulin/body weight. Blood glucose levels were determined at indicated time points. Body fat composition was analyzed by EcoMRI-100H.
Histology. Mouse tissues were fixed in zinc-buffered formalin (Anatech) and paraffin-embedded. Sections (5-10 mm) were used for hematoxylin and eosin (H&E) staining or immunohistochemistry. For immunohistochemical staining of epididymal WAT (eWAT), sections were rehydrated and antigen-retrieved in sodium citrate. The sections were incubated with F4/80 (Abcam) or Ly6G (Biolegend) antiserum and visualized by the avidin-biotin-complex method using the chromogen diaminobenzidine (Vector Labs). Stained slides were imaged by lightfield microscopy (Nikon). Indirect calorimetry. Mice fed with a 60% high-fat diet for 3 months were individually housed for at least 3 days before experiments. Food intake, locomotor activity, oxygen consumption, and energy expenditure were simultaneously measured for individually housed mice with a LabMaster system (TSE Systems). Data were collected for 2-3 days and analyzed.
Primary adipocyte and 3T3-L1 cells culture. Primary adipocytes were generated and maintained as previously reported 54 . Interscapular BAT and eWAT were collected, minced, and digested with isolation buffer for a proper time at 37°C on a shaker. The isolation buffer contains 1.5 mg/ml Collagenase I (for eWAT) or 1 mg/ ml Collagenase IV (for BAT). After digestion and filtration, stromal vascular fraction (SVF) cells were cultured in a growth medium for 3 days, and then fresh media was changed every 2 days. Upon confluence, cells were exposed to an induction medium for 2 days and then a differentiation medium. After differentiation, cells were considered primary adipocytes. The induction medium contains DMEM (Mediatech), 10% FBS (Gemini), 1 µg/ml insulin (HumulinR; Lilly), 0.3 μM dexamethasone (Sigma) and 0.63 mM 3-isobutyl-1-methylxanthine (IBMX) (Sigma), 1 µM Rosiglitazone (Adipogen), 10 nM T3 (Sigma). The differentiation medium contains DMEM, 10% FBS, 1 µg/ml insulin, and 10 nM T3. For 3T3-L1 differentiation, we employed the same procedure as for primary adipocytes. Subcellular fractionation for TNFα treated 3T3-L1 adipocytes was performed using a NE-PER Nuc-Cyt extraction kit according to the manufacturer's protocol (Thermo Scientific).
Immunofluorescence staining. Immunofluorescence staining was performed as previously described 54 . eWAT from HFD (6 weeks) and age-matched NCD mice were fixed with ice-chilled 4% PFA for 3 h, stored overnight in 25% sucrose solution, and then embedded in OCT compound. After cutting at 15-30 micrometer using cryostat, slides were blocked for 1 h in blocking buffer containing PBS, 3% normal goat serum, 2% bovine serum albumin, and 0.2%Triton X-100. The sections were incubated with anti-CRTC3 antibody (Cell signaling technology) or anti-CRTC2 antibody (generated in our laboratory) or anti-F4/80 (Abcam) diluted in blocking buffer overnight at 4°C. Slides were washed three times with PBS containing 0.2% Triton X-100, and then incubated with secondary antibodies and DAPI (Vector Laboratories) for 1 h at room temperature, mounted and images were taken under Zeiss750 confocal microscope.
Flow cytometry. The stromal vascular fraction of epididymal adipose tissues were isolated as described above. Cells were stained for 1 h and washed three times with HBSS containing 1% FBS. The macrophages (F4/80 + CD11b + ) and neutrophils (F4/80-Ly6g + CD11b + ) were analyzed by FACS LSRII (BD Biosciences). The antibodies were purchased from Serotec (F4/80) and BD Biosciences (CD11b and Ly6g). For the collection of endothelial/immune cells, stained cells with CD31, CD45, Ter119 were collected by FACS Aria (BD Biosciences). The antibodies were obtained from eBioscience. The gating strategy is described in Supplementary Data.
Lentivirus production. Lenti-mouse c/EBPα was prepared by cloning c/EBPα cDNA into the pHRST vector. pHRST-control, pHRST-c/EBPα along with packaging plasmids were transfected in HEK293 cells. After 3 days of incubation, amplified virus particles were collected and concentrated.
RNA analysis. Total RNA was extracted from cells or tissue with TRIzol-based isolation Kits (Zymo Research) and 1 μg of RNA was converted to cDNA with Transcriptor first-strand cDNA synthesis kit according to the manufacturer's protocol (Roche). Quantitative PCR (qPCR) was performed with SYBR green master mix (Roche) by Light cycler 480 II (Roche) and the relative mRNA expression was calculated by 2 −ΔΔct method. L32 was used as a housekeeping gene for mRNA expression analysis. RNA-seq was performed as previously described 54 .
Neutralization and administration of CXCL1 and 2. For neutralization of CXCL1 and 2, 10 µg/mouse of anti-CXCL1 (KC), CXCL2 (MIP2), and control IgG antibody were introduced by retro-orbital injection. All antibodies were purchased from R&D systems. For recombinant protein administration, mice were injected 1 µg/mouse of recombinant murine CXCL1 and CXCL2 (Peprotech) or Saline intraperitoneally. IPGTT assay was conducted 1 week after each injection.
LPS injection study. Low dose (500 ng/kg; Sigma-Aldrich) of LPS or an equal volume of saline was administered by intra-peritoneal injection. After 3 h of incubation, tissues were collected and snap-frozen in liquid nitrogen.
Chemotaxis assay. Chemotaxis assay was performed with a slight modification of invasion assay 57,58 . Transwell inserts containing polycarbonate membrane (5 µm pore size; Corning costar) were preincubated in regular medium (DMEM supplemented with 10% FBS and 1% penicillin/streptomycin) for 30 min. Seeded LPSprimed (5 ng/ml, overnight) Raw264.7 murine monocyte cells (1 × 10 5 cells in 500 µl serum-free medium) in the transwell insert and filled outside of insert with conditioned medium from WT and dAKO adipocytes following overnight treatment with saline, 10 ng/ml TNFα, TNFα + α-CXCL1/2 antisera (0.5 µg/ml) and rCXCL1/2 (0.1 µg/ml). After overnight incubation, aspirated medium and removed non-migrated cells in the upper membrane of an insert with a cotton swab. Then, washed insert and fixed in 4% formaldehyde for 10 min and stained with 0.1% crystal violet for 30 min. Following three washes with PBS and dry membrane, filters were photographed and dye was eluted in 50% ethanol plus 1% acetic acid and quantitated by measuring absorbance at 600 nm (GloMax multi detection system; Promega Inc).
ELISA. Levels of CXCL1 and CXCL2 in serum and conditioned media were measured by the ELISA kit according to the manufacturer's protocol (R&D Systems). Circulating and hepatic triglyceride and free fatty acids were analyzed by Elisa kit according to the manufacture's protocol (BioVision).
Statistics and reproducibility. The data were analyzed using Microsoft Excel and Graphpad PRISM8. The data are presented with mean ± SEM or mean ± SD. Pvalues were calculated using unpaired two-tailed student's t-test and one-way ANOVA with post-hoc Tukey HSD test. Statistical significance was indicated as **P < 0.01 and *P < 0.05. A P-value of <0.05 was considered statistically significant.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The raw data for the main figure are available in Supplementary Data (graphs; Figs. 1-7) and Supplementary Fig. 10 (Uncropped Immunoblots; Figs. 1, 2, 5, 6, 7). GEO accession number for RNA-seq studies in Fig. 1a, b and Supplementary Fig. 7d is GSE160684. RNA-seq data for Supplementary Fig. 2a was taken from GEO accession number GSE109443 54 . GEO accession number for ChIP-seq studies in Fig. 3e and Supplementary  Fig. 4a-c is GSE160597. Additional details can be obtained from the corresponding author on reasonable request.