Age-dependent loss of adipose Rubicon promotes metabolic disorders via excess autophagy

The systemic decline in autophagic activity with age impairs homeostasis in several tissues, leading to age-related diseases. A mechanistic understanding of adipocyte dysfunction with age could help to prevent age-related metabolic disorders, but the role of autophagy in aged adipocytes remains unclear. Here we show that, in contrast to other tissues, aged adipocytes upregulate autophagy due to a decline in the levels of Rubicon, a negative regulator of autophagy. Rubicon knockout in adipocytes causes fat atrophy and hepatic lipid accumulation due to reductions in the expression of adipogenic genes, which can be recovered by activation of PPARγ. SRC-1 and TIF2, coactivators of PPARγ, are degraded by autophagy in a manner that depends on their binding to GABARAP family proteins, and are significantly downregulated in Rubicon-ablated or aged adipocytes. Hence, we propose that age-dependent decline in adipose Rubicon exacerbates metabolic disorders by promoting excess autophagic degradation of SRC-1 and TIF2.

M etabolic disorders, and the insulin resistance associated with them, represent a worldwide health concern due to their increasing incidence in developed countries and their contribution to life-threatening cardiovascular diseases 1,2 . Adipose tissue, especially white adipose tissue (WAT), regulates systemic metabolism via two functions 3,4 . Adipocytes store surplus energy as triglyceride in a single monolayer organelle, the lipid droplet (LD) 5,6 , and they also orchestrate other metabolic organ functions through the secretion of several hormones or cytokines (called adipocytokines), such as leptin and adiponectin 7 . Because diabetes, fatty liver and dyslipidaemia are closely associated with aging, a mechanistic understanding of ageassociated metabolic complications is of particular interest. The connection between adipose tissue and aging is illustrated by ageassociated systemic changes in adipose tissue such as ectopic lipid deposition and adipose-derived hormone changes 8 . However, the direct cause of adipocyte dysfunction in aged organisms, which in turn leads to age-associated metabolic disorders, remains unknown.
Autophagy is an intracellular bulk degradation system that plays a pivotal role in the maintenance of cellular homoeostasis and the prevention of a wide range of age-associated diseases, such as neurodegeneration 9,10 , hepatic disorders 11,12 and heart failure 13 . Consistent with this, recent advances in the field have shown that autophagy is crucial for both adipose tissue homoeostasis and longevity 14 . Autophagy, especially mitophagy, regulates adipocyte differentiation 15,16 or beige adipocyte maintenance 17,18 , whereas a form of selective autophagy, lipophagy, degrades LDs in brown adipose tissue (BAT) and liver [19][20][21] . The age-associated decline in autophagic activity 22,23 , along with the extension of longevity by upregulation of basal autophagy 24,25 , indicates that autophagy is involved in the aging process. Rubicon negatively regulates autophagy by binding to the PI3K complex [26][27][28] , which is essential for autophagy. Our recent studies have revealed that Rubicon levels increase with obesity or aging, and that accumulation of this factor suppresses autophagy and disrupts cellular homoeostasis in several tissues 29,30 . However, the role of Rubicon in adipocytes, especially in aging, remains largely unknown. In addition, it is important to determine whether, and if so how, the age-associated change in adipose autophagy causes metabolic diseases.
Here, we report that an age-dependent loss of Rubicon and an increase in autophagic activity are key feature of aged adipocytes. We found that genetic ablation of Rubicon in adipocytes causes inadequate upregulation of autophagy, which in turn disrupts proper adipocyte functions, leading to fat atrophy, glucose intolerance, dyslipidaemia and hepatic fat accumulation. These changes arise from a reduction in adipogenic gene expressions, which can be restored by activation of PPARγ. SRC-1 and TIF2, coactivators of PPARγ, are degraded by autophagy and thus significantly downregulated in Rubicon-ablated or aged adipocytes. We propose that age-dependent loss of adipose Rubicon results in excess autophagy, which decreases the levels of SRC-1 and TIF2, contributing to the development of age-associated metabolic disorders.

Results
Rubicon deletion in adipocytes causes systemic fat atrophy. In previous studies, we showed that upregulation of Rubicon, a negative regulator of autophagy 26 , is a signature of aging 30 . We hypothesised that Rubicon also accumulates in aged adipocytes, and that its accumulation could decrease autophagic activity in adipose tissue. Surprisingly, we found that adipose tissue in aged mice exhibited a significant decline in levels of Rubicon and the autophagic substrate p62 12 (Fig. 1a), suggesting that autophagic activity in adipocytes increases with age. Next, to determine the fundamental role of Rubicon in adipose tissue, we genetically deleted Rubicon specifically in adipocytes. To this end, we crossed Adipoq-Cre mice 31 with Rubicon-floxed mice 29 (Supplementary Fig. 1a) to generate Rubicon flox/flox ; Adipoq-Cre mice (Rubicon ad−/− mice) and confirmed that Cre-mediated recombination ( Supplementary  Fig. 1b) and the absence of Rubicon occurred specifically in both the WAT (Supplementary Fig. 1c) and BAT ( Supplementary  Fig. 1d) of Rubicon ad−/− mice. Rubicon knockout had no significant effect on other ATG protein levels in both adipose tissues (Supplementary Fig. 2a, b). Rubicon ad−/− mice were born in a Mendelian ratio, exhibited no visible abnormalities at birth, and had normal body weight at 5 weeks of age. These observations suggested that Rubicon deletion by Adipoq-Cre does not cause significant abnormality during development. Notably, after 7 weeks of age, the body weight of Rubicon ad−/− mice increased at a lower rate than their control counterparts, and failed to catch up even at 21 weeks of age on a normal chow diet (NCD) (Fig. 1b, c). Consistent with this, we observed a significant decrease in the organ weights of epididymal (eWAT), mesenteric (mWAT) and inguinal WAT (iWAT), as well as interscapular BAT (iBAT), but not liver, in Rubicon ad−/− mice (Fig. 1d, e). Haematoxylin and eosin (H&E) staining of the tissue sections revealed that Rubicon deficiency caused a reduction in adipose cell size in eWAT (Fig. 1f, g) and iBAT (Fig. 1h). These facts suggest that Rubicon maintains the size of adipocytes. Remarkably, these characteristics of Rubicon knockout were not observed in mice also harbouring a knockout of Atg5, a gene essential for autophagy (Fig. 1b, d, f-h). We confirmed that the reduction of Rubicon and ATG12-ATG5 complex in the double knockout mice was comparable with that in the single knockout mice ( Supplementary Fig. 1c, d). We also found that Rubicon ad−/− mice exhibited a significant decrease in the levels of two autophagic substrates, LC3-II and p62, but not LC3-II/I ratio, in eWAT ( Supplementary Fig. 2c) and the iBAT ( Supplementary Fig. 2d). These decreases were completely abolished by additional Atg5 deletion ( Supplementary Fig. 2c, d). Importantly, An ex vivo autophagic flux assay 20,21 revealed an increase in autophagic activity in the eWAT ( Supplementary Fig. 2e), the iWAT (Supplementary Fig. 2f) and the iBAT ( Supplementary Fig. 2g) in Rubicon ad−/− mice. These data collectively indicate that systemic fat atrophy in Rubicon ad−/− mice is caused by inadequate upregulation of autophagy. Rubicon participates positively in LC3associated phagocytosis (LAP); 32 however, the phenotypes of Rubicon ad−/− mice must be independent of LAP because their phenotypes were abolished by additional knockout of Atg5, which is also required for this process 32 .
Notably, neither deletion of Rubicon nor Atg5 had any effect on dietary intake ( Supplementary Fig. 3a), oxygen consumption ( Supplementary Fig. 3b), or carbon dioxide production (Supplementary Fig. 3c), indicating that the observed phenotypes in Rubicon ad−/− mice did not arise from changes in food intake or energy consumption. Moreover, neither Rubicon nor Atg5 knockout had an effect on the respiratory quotient (RQ) ( Supplementary Fig. 3d), excluding the possibility that the fat atrophy in Rubicon ad−/− mice was caused by switching between glucose and fat utilisation in vivo. Taken together, our findings show that the loss of Rubicon in adipocytes causes systemic atrophy without affecting energy balance.
Rubicon deletion in adipocytes leads to metabolic disorders. To determine whether systemic fat atrophy in Rubicon ad−/− mice is associated with systemic changes in metabolism, we measured a series of indicators of metabolic disorder. Strikingly, we found that plasma triglyceride (Fig. 2a) Fig. 1 Loss of Rubicon in adipocytes causes a reduction in body weight and adipose tissue mass. a Immunoblotting to detect Rubicon and p62 in the eWAT depots of 3-or 25-month-old wild-type mice on an NCD. n = 3 mice. b Body weight chart for mice of the indicated genotypes on an NCD. Intraperitoneal glucose tolerance test (IPGTT) (Fig. 2c) and intraperitoneal insulin tolerance test (IPITT) were performed at 17 and 19 weeks, respectively. Control, n = 35; Rubicon ad−/− , n = 21; Atg5 ad−/− , n = 27; Rubicon ad−/− ; Atg5 ad−/− , n = 20. c Representative image of a 21-week-old control and Rubicon ad−/− mouse on an NCD. d Organ weight of eWAT, mWAT, iWAT, iBAT and liver from 21-week-old mice of the indicated genotypes, described in (b). Control, n = 35; Rubicon ad−/− , n = 21; Atg5 ad−/− , n = 27; Rubicon ad−/− ; Atg5 ad−/− , n = 20. e Representative images of fat pads from a 21-week-old control and Rubicon ad−/− mouse on an NCD. f, h Representative images of H&E staining of eWAT (f) and iBAT (h) sections from mice of the indicated genotypes on an NCD. Scale bars, 50 μm. n = 5 mice with similar results. g Distribution of adipocyte area in the eWAT sections in f. n = 5 mice. Quantification of mean adipocyte area is shown in the graphs at right. Error bars indicate means ± SEM. Data were analysed by two-tailed Student's t test (d, g) or two-way repeatedmeasures ANOVA followed by Fisher's LSD test (b). P value from left to right: 0.0003, 0.7257 (b), 0.0258, 0.0067, 0.0494, 0.2337, 0.0005, 0.8217, 0.8418, 0.9407, 0.5023, 0.5205 (d), 0.0008 (g). *P < 0.05; ** or † P < 0.01; *** or ‡ P < 0.001; § P < 0.0001. N.S. not significant. NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-17985-w ARTICLE NATURE COMMUNICATIONS | (2020) 11:4150 | https://doi.org/10.1038/s41467-020-17985-w | www.nature.com/naturecommunications mice. In addition to lipid metabolic disorders, Rubicon ad−/− mice exhibited glucose intolerance, which was less significant in an autophagy-deficient background (Fig. 2c), strongly suggesting that Rubicon in adipocytes plays a crucial role in systemic metabolism by regulating proper autophagic activity. Consistent with this idea, we found that Rubicon ad−/− mice exhibited hepatic lipid accumulation. Rubicon knockout in adipocytes increased hepatic triglyceride (Fig. 2d) and cholesterol content (Fig. 2e). Histological analysis revealed that Rubicon deficiency in adipocytes caused abnormal LD formation in liver (Fig. 2f). We assumed that metabolic disorders in Rubicon ad−/− mice stemmed from a decline not only in the lipid storage function, but also in Liver Control ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-17985-w the endocrine function. Indeed, we found that the Rubicon ad−/− mice had significantly lower levels of plasma adiponectin (Fig. 2g) and leptin (Fig. 2h), which are adipocyte-derived hormones maintaining glucose and lipid metabolism in vivo [33][34][35] , whereas Rubicon deletion in the Atg5-knockout background caused no detectable change in these protein levels (Fig. 2g, h). Western blot of eWAT lysates revealed that adiponectin was markedly reduced in Rubicon ad−/− mice (Fig. 2i). This reduction was cancelled in Rubicon ad−/− ; Atg5 ad−/− mice (Fig. 2i), indicating that Rubicon deletion in adipocytes decreases adiponectin production in an autophagy-dependent fashion. We also found that Atg5 ad−/− mice showed a reduction in plasma cholesterol (Fig. 2b) and leptin (Fig. 2h), and an increase in hepatic cholesterol (Fig. 2e). These results suggest that an autophagy-dependent or -independent role of ATG5 in adipocytes could be important for leptin secretion or cholesterol metabolism.
Our results described above indicated that the loss of Rubicon in adipocytes inappropriately upregulates autophagy, resulting in metabolic disorders. To investigate the role of adipose Rubicon in obesity, we fed Rubicon ad−/− mice a high-fat diet (HFD). Notably, Rubicon ad−/− mice weighed less than controls fed an HFD ( Supplementary Fig. 4a), and adipose-specific Rubiconknockout mice fed an HFD exhibited reduced adipose weight  Fig. 4i), but not leptin ( Supplementary Fig. 4j). These phenotypes of obese Rubicon ad−/− mice were not seen in the Atg5-deleted background (Supplementary Fig. 4a-j). Together, these results indicate that Rubicon in adipocytes maintains glucose and lipid metabolism by modulating autophagy independently of diet. Interestingly, HFD-fed Atg5 ad−/− mice exhibited a reduction in visceral eWAT and an increase in subcutaneous iWAT and iBAT, compared with the control counterparts ( Supplementary Fig. 4b). This fat maldistribution was associated with a reduction in plasma adiponectin (Supplementary Fig. 4i), and an increase in obesity-induced glucose intolerance ( Supplementary Fig. 4f) and hepatic triglyceride accumulation ( Supplementary Fig. 4h). The Atg5 deficiency in adipocytes also promoted formation of the HFD-induced crownlike structure (CLS) in the eWAT (Supplementary Figs. 4c and 8d), consistent with a previous report demonstrating the role of ATG3 and ATG16L1 in adipose tissue inflammation during obesity 36 . These results suggest that ATG conjugation system, which includes ATG3, ATG5 and ATG16L1, in adipocytes could prevent obesity-related metabolic disorders.
Autophagic activity in adipose tissue increases with age. To further investigate the role of autophagy in adipocytes during the aging process, we examined Rubicon protein levels in adipose tissue from 12-or 18-month-old mice, and found that the levels of Rubicon and p62 were significantly reduced at the age of 12 months (Fig. 3a). Rubicon mRNA levels also decreased with age, whereas p62 mRNA levels did not (Fig. 3b), suggesting that Rubicon is downregulated with age at the transcriptional level, followed by upregulation of autophagy that is marked by a decrease in an autophagic substrate p62 at a protein level. Importantly, an ex vivo autophagic flux assay using LC3-II 20,21 revealed that autophagic activity in adipose tissue increased with age ( Fig. 3c). At the age of 25 months, a decline in transcripts of p62 in eWAT became evident as well ( Supplementary Fig. 5a), suggesting a possible involvement of transcriptional regulation in age-associated changes of adipose p62.
Next, to clarify the role of adipose Rubicon in the aging process, we allowed Rubicon ad−/− mice to age until they were 12 or 18 months old. Remarkably, the aged Rubicon ad−/− mice exhibited reductions in body weight ( Collectively, these data indicate that Rubicon in adipocytes prevents excess autophagy, thereby maintaining metabolic homoeostasis over the aging process. More importantly, we found that hepatic triglyceride ( Fig. 3i) and cholesterol levels ( Fig. 3j) increased with age, and that these increases were significantly attenuated in aged Atg5 ad−/− mice. Histological analysis also revealed that the loss of autophagy in adipocytes ameliorated age-associated hepatic steatosis (Fig. 3k). Taken together, our data strongly suggest that excess autophagy due to the loss of Rubicon promotes ageassociated metabolic disorders, including liver steatosis.
Rubicon is upregulated during adipogenesis. To determine the molecular mechanism by which Rubicon maintains adipocyte homoeostasis, we monitored changes in autophagic activity and Rubicon expression along with adipogenesis. We found that during adipogenesis, 3T3-L1 cells exhibited a gradual decrease in autophagic flux, as determined based on the degradation of autophagic substrates LC3-II and p62 (Fig. 4a). This was associated with an increase in Rubicon protein (Fig. 4b) and mRNA levels (Fig. 4c). Because Rubicon knockdown, which results in upregulation of autophagy (Fig. 4d), dramatically decreased Adiponectin protein level in differentiated 3T3-L1 cells (Fig. 4e), the increase in Rubicon must be crucial for adipogenesis. Indeed, we found that Rubicon knockdown decreased the expression of a wide range of adipogenic genes, including Adipoq, Lep, Fabp4, i Immunoblotting to detect adiponectin in the eWAT depots of 21-week-old mice of the indicated genotypes on an NCD. n = 5 mice. Quantification of adiponectin levels is shown in the graphs at right. Error bars indicate means ± SEM. Data were analysed by two-tailed Student's t test (a, b, d, e, g, h), oneway ANOVA followed by Tukey's test (i), or two-way repeated-measures ANOVA followed by Fisher's LSD test (c). Cd36, Glut4, Acaca and Fasn (Fig. 4f). Consistent with this, we found that Rubicon deletion in adipocytes in vivo caused a decrease in adipogenic gene expression that was again reverted by additional Atg5 knockout (Fig. 4g).
To further investigate lipid metabolism pathways in Rubicon ad−/− mice, we examined the expression of Perilipin1 and Perilipin2, which participate in lipolysis or LD stabilisation 37 Fig. 6a, b), possibly causing the reduction in LD size. We also found that gene expression of lipoprotein lipase (LPL), which hydrolyses triglycerides in chylomicrons or VLDL 38 , was significantly decreased in the adipose tissue of Rubicon ad−/− mice ( Supplementary Fig. 6c). In addition, expression of ABCA1 and ABCG1, which participate in cholesterol efflux 38 , was elevated in Rubicon ad−/− mice ( Supplementary Fig. 6c), whereas levels of the cholesterol receptor LDL-R and SR-B1 38 were not changed ( Supplementary Fig. 6c). Previous reports showed that knockout of Lpl increases both plasma triglyceride and cholesterol 39 , and that adipose-specific knockout of Abca1 decreases plasma cholesterol 40 . Therefore, the reduction of LPL and the increases in ABC proteins could mediate the lipid metabolic disorders in Rubicon ad−/− mice. Rubicon deletion in adipocytes did not increase the expression of lipogenic genes in the liver ( Supplementary  Fig. 6d), supporting the idea that the lipids in adipocytes could be transferred into the liver in Rubicon ad−/− mice. Moreover, we investigated whether Rubicon knockout induces lipolysis in adipocytes, and found that Rubicon deletion had no impact on characteristic features of BAT ( Supplementary Fig. 6e). The lipolytic genes Atgl and Hsl 41 in WAT tended to be downregulated or at least not changed (Supplementary Fig. 6f). Rubicon deletion in adipocytes did not increase the plasma levels of non-esterified fatty acid (NEFA), a product of lipolysis, under either fed or fasted conditions ( Supplementary Fig. 6g). Notably, ex vivo lipolysis assay showed that catecholamine-induced release of glycerol was not significantly increased in eWAT ( Supplementary Fig. 6h), iWAT ( Supplementary  Fig. 6i) or iBAT ( Supplementary Fig. 6j) from Rubicon ad−/− mice. That of NEFA was not also increased in Rubicon ad−/− mice ( Supplementary Fig. 6k-m). These data collectively suggest that the loss of Rubicon does not enhance lipolysis in adipocytes. Notably in this regard, the reduction in adipogenic gene expressions was also seen in younger Rubicon ad−/− mice ( Supplementary Fig. 6n), suggesting that Rubicon is needed for early adipocyte development. Considering a role of mitophagy 42 , mitochondria-targeted autophagy, in adipogenesis 15,16 , we tested whether any possible changes in mitophagy due to the loss of Rubicon in adipocytes contributes to the phenotype shown above. We assessed the mitochondrial respiration in Rubicon-knockdown adipocytes using tetramethylrhodamine methyl ester (TMRM), an indicator of the mitochondrial membrane potential. We found that the loss of Rubicon did not cause any change in the mitochondrial membrane potential while an ionophore valinomycin remarkably decreased the membrane potential as a positive control ( Supplementary Fig. 7a). For assessment of Parkin-dependent mitophagy, we treated the cells with valinomycin to see the recruitment of Parkin onto the mitochondria, and found that Parkin was recruited to the damaged mitochrondria in Rubiconknockdown cells at a level comparable to control ( Supplementary  Fig. 7b). Consistently, Rubicon knockdown in adipocytes did not affect the interaction between LC3-II and p62 that works as a potential receptor for mitophagy ( Supplementary Fig. 7c). The degradation rate of outer and inner mitochondrial proteins was not significantly changed in Rubicon-knockdown cells (Supplementary Fig. 7d). In addition, the protein levels of Parkin and PINK1 were not affected in the Rubicon-knockdown cells ( Supplementary Fig. 7d) or the WAT (Supplementary Fig. 2a) and the BAT (Supplementary Fig. 2b) of Rubicon ad−/− mice. These data collectively indicate that the downregulation of Rubicon in adipocytes does not cause any significant change in mitophagy. Furthermore, ex vivo culture of adipose tissue with a lysosomal inhibitor did not increase the protein level of LDresident Perilipin1, Perilipin2 and mitochondrial Complex III subunit Core 1, but did increase LC3-II ( Supplementary Fig. 7e), suggesting that lipophagy 42 or mitophagy in adipocytes works only at an undetectable level and may play a relatively minor role in the absence of exogenous stimuli.
Cell death or inflammation often leads to a reduction in adipogenic gene expression in adipose tissue. Moreover, excess autophagy is closely involved with cell death 43,44 ; thus, upregulation of autophagy by Rubicon knockout could affect cell proliferation or cell death. We performed immunohistochemistry for PCNA to assess cell proliferation in Rubicon ad−/− mice. Although PCNA-positive cells were observed in the epididymis of control mice, PCNA-positive adipocytes were not detected in eWAT in either control or Rubicon ad−/− mice ( Supplementary  Fig. 8a). This is consistent with the idea that adipocytes are nonproliferative, and that Rubicon knockout does not affect adipocyte proliferation. Moreover, Rubicon knockout did not increase the level of Cleaved Caspase-3 in adipose tissue ( Supplementary  Fig. 8b). TUNEL-positive adipocytes were observed in a DNase Itreated sample, used as a positive control, but not in untreated samples from control or Rubicon ad−/− mice ( Supplementary  Fig. 8c). CLSs, dead adipocytes surrounded by macrophages 45 , were not detected in the eWAT in either control or Rubicon ad−/− mice (Fig. 1f, Supplementary Fig. 8d), whereas they were present in HFD-fed mice ( Supplementary Figs. 4c and 8d). These data indicate that adipocyte death is not induced in Rubicon ad−/− mice. To determine whether systemic inflammation is induced in Rubicon ad−/− mice, we measured the plasma IL-6 or TNFα level, and found that those in Rubicon ad−/− mice were comparable to those in control mice ( Supplementary Fig. 9a, b). In addition, the gene expression of pro-inflammatory cytokines or macrophage markers was not significantly elevated in eWAT ( Supplementary  Fig. 9c, d) or iBAT ( Supplementary Fig. 9e, f) of Rubicon ad−/− mice. Consistently, flow cytometry analysis showed that the rate of CD45 + hematopoietic cells was not increased in the stromal vascular fraction (SVF) from the eWAT of Rubicon ad−/− mice ( Supplementary Fig. 9g). The rate of CD11b + F4/80 + macrophage in the CD45 + hematopoietic cells was not increased in Fig. 3 Excess autophagy in adipocytes promotes age-related hepatic steatosis. a Immunoblotting to detect Rubicon and p62 in eWAT depots of 3-, 12and 18-month-old control mice. n = 4 mice. b Relative mRNA expression of Rubicon and p62 in eWAT depots of 3-, 12-and 18-month-old control mice. n = 6 mice. c Ex vivo autophagic flux assay based on LC3-II degradation in the eWAT depots of mice of the indicated ages and genotypes. n = 4 mice. Body weight (d) and eWAT weight (e) of mice of the indicated ages and genotypes. 12-month control, n = 21; 12-month Rubicon ad−/− , n = 16; 12-month Atg5 ad−/− , n = 19; 12-month Rubicon ad−/− ;Atg5 ad−/− , n = 20; 18-month control; n = 18; 18-month Rubicon ad−/− , n = 23; 18-month Atg5 ad−/− , n = 18; 18month Rubicon ad−/− ;Atg5 ad−/− , n = 13. Representative images of H&E staining of eWAT (f) and liver (k) sections from mice of the indicated ages and genotypes. Scale bars, 50 μm. n = 5 mice with similar results. Distribution of adipocyte area in eWAT sections from 12-(g) and 18-(h) month-old mice in (f). n = 5 mice. Quantification of mean adipocyte area is shown in the graphs at right. Liver triglyceride (i) and cholesterol (j) levels in mice of the indicated ages and genotypes. n = 14 mice, except for 3-month-old mice (n = 5). Aged mice were maintained on an NCD. Quantification data are shown in the graphs at the right of each blot. Error bars indicate means ± SEM. Data were analysed by two-tailed Student's t test (d, e, g, h), one-way ANOVA followed by Tukey's test (a-c) or Dunnett's test (i, j).  Rubicon ad−/− mice ( Supplementary Fig. 9h). Histological analysis also revealed that inflammatory cells were not seen in the eWAT (Fig. 1f) or the iBAT (Fig. 1h) in Rubicon ad−/− mice. These results suggest that inflammatory cell infiltration does not occur in adipose tissue of Rubicon ad−/− mice. Based on these findings, we conclude that cell death and inflammation are not major contributors to adipocyte dysfunction in Rubicon ad−/− mice.
Our data suggest that downregulation of Rubicon causes adipocyte dysfunction, which could stem from a reduction in a wide range of adipogenic gene expressions. Because PPARγ is a master transcription factor involved in adipogenesis 46 , and Adipoq-Cremediated knockout of Pparg causes a phenotype that is similar to, but much stronger than, that of Rubicon knockout 47 , we hypothesised that PPARγ activity is reduced in Rubicon ad−/− mice. To test this hypothesis, we treated Rubicon-depleted cells with thiazolidinedione (TZD), an exogenous ligand of PPARγ 48 . Strikingly, TZD treatment restored the reduction in Adiponectin protein level (Fig. 5a), adipogenic gene expression (Fig. 5b) and triglyceride content (Fig. 5c) caused by Rubicon depletion. Consistent with this, Oil Red O (ORO) staining revealed that Rubicon depletion decreased LDs (Fig. 5d), whereas no such reduction was seen under TZD treatment (Fig. 5d). Furthermore, we treated Rubicon ad−/− mice with TZD, and found that TZD treatment in vivo almost completely rescued the reduction in plasma adiponectin (Fig. 5e) and leptin (Fig. 5f) levels, as well as adipose mass (Fig. 5g). Our findings suggest that activation of PPARγ reverses adipocyte dysfunction caused by loss of Rubicon.
SRC-1 and TIF2 are degraded by autophagy in adipocytes. The results described above indicate that upregulation of autophagy by loss of Rubicon causes adipocyte dysfunction, which could cause PPARγ activity to be suppressed. Therefore, we assumed that autophagy degrades the PPARγ coactivator in adipocytes in Rubicon ad−/− mice. Previous reports showed that both SRC-1/NCoA1 and TIF2/SRC-2/NCoA2 work as coactivators of PPARγ 49 , and that SRC-1 and TIF2 are localised both in the cytosol and nucleus 50,51 . Importantly, a protein in the same family, NCoA4, is an autophagic substrate 52 . These reports prompted us to investigate whether SRC-1 and/or TIF2 are degraded by autophagy in Rubicon-ablated adipocytes.
Remarkably, Rubicon depletion decreased the levels of both SRC-1 and TIF2 proteins in differentiated 3T3-L1 cells, whereas the lysosomal inhibitor increased the levels of both proteins (Fig. 6a). Also, in mice, SRC-1 and TIF2 levels were increased in an ex vivo culture with lysosomal inhibitor (Fig. 6b). Rubicon ad−/− mice exhibited decreases in SRC-1 and TIF2 (Fig. 6c), but these reductions were abolished in the Atg5-knockout background (Fig. 6d). Collectively, these data indicate that SRC-1 and TIF2 are degraded by autophagy, and thus downregulated, in adipocytes of Rubicon ad−/− mice. We also found that SRC-1 and TIF2 were dramatically decreased in aged adipose tissue, along with the reduction in adiponectin (Fig. 6e,  f), suggesting that the reductions in SRC-1 and TIF2 could contribute to age-dependent adipocyte dysfunction. Notably, in addition to aged mice, Rubicon was significantly reduced in fasted mice, concomitant with the reductions in SRC-1 and TIF2 (Fig. 6g). 3T3-L1 adipocytes also showed a starvationinduced decline in Rubicon, SRC-1 and TIF2 (Fig. 6h), followed by a reduction in most of adipogenic gene transcripts (Fig. 6i). These data implicate the physiological system that a reduction of Rubicon in adipocytes actively causes a decline in adipocyte function during fasting.
SRC-1 and TIF2 interact with GABARAP family proteins. The LC3-interacting region (LIR)/GABARAP-interacting motif (GIM) is necessary for autophagic degradation of various substrates 53 . Using the iLIR database 54 , we found that SRC-1 and TIF2 have evolutionarily conserved LIR/GIM-like motifs (Fig. 7a), which are comparable to the ATG8-interacting motif in yeast ATG3 55 , suggesting that their autophagic degradation is mediated by binding to LC3 and/or GABARAP. To test this possibility, we generated mutant constructs in which the tryptophan in the potential LIR/GIM was replaced with alanine, and performed immunoprecipitation of FLAG-tagged SRC-1 or TIF2. Importantly, GABARAP and GATE16, but not LC3, were coimmunoprecipitated with SRC-1 (Fig. 7b), as well as TIF2 (Fig. 7c). The LIR/GIM mutant SRC-1 and TIF2 exhibited a reduction in the interaction with endogenous GABARAP family proteins (Fig. 7b, c). Our findings suggest that the tryptophan in the LIR/GIM of SRC-1 or TIF2 is essential for the interaction with GABARAP or GATE16. Moreover, the LIR/GIM mutant SRC-1 and TIF2 failed to exhibit an increase in the presence of a lysosomal inhibitor (Fig. 7d, e), as well as a reduction upon Rubicon depletion (Fig. 7f, g), suggesting that autophagic degradation of SRC-1 and TIF2 by Rubicon depletion is mediated by the interaction with GABARAP family proteins. To clarify their contribution to the phenotype in Rubicon-knockdown adipocytes, we overexpressed the LIR/GIM mutant SRC-1 and/or TIF2 in Rubicon-knockdown cells, and found that both overexpression of the mutant SRC-1 and TIF2 rescued the reduction in adipogenic gene expressions in Rubicon-knockdown cells (Fig. 7h). Taken together, these findings indicate that the age-dependent loss of Rubicon causes adipocyte dysfunction, which is mediated by autophagic degradation of both SRC-1 and TIF2, leading to ageassociated metabolic disorders.

Discussion
Here, we showed that downregulation of autophagy by Rubicon is achieved during adipogenesis and is crucial for adipocyte function, and that excess autophagy with the loss of Rubicon in adipocytes occurs in aging, leading to age-associated metabolic disorders. Consistent with these findings, previous studies showed that the insulin receptor (IR)-AKT-mTOR complex 1 (mTORC1) axis 56 , which negatively regulates autophagy via phosphorylation of the ULK1 complex 57 and TFEB 58 , also maintains adipocyte function. Indeed, recent studies using Adipoq-driven Cre have shown that the development or maintenance of adipocyte functions requires each component of the IR-AKT-mTORC1 axis [59][60][61][62][63] . Because this axis promotes lipogenesis by enhancing glucose incorporation, upregulating lipogenic gene sets and increasing glycolysis 64 , the strong phenotypes caused by genetic disruption of this axis must arise from multiple factors, including autophagy. However, our results highlight the significance of downregulation of autophagy in the maintenance of adipose tissue homoeostasis. Notably in this regard, the two systems for negatively regulating autophagy are functionally connected: Rubicon is regulated by the mTORC1 pathway via phosphorylation of its binding partner UVRAG 28 . In addition, a previous study showed that mTOR activity in adipose tissue is downregulated with age, but maintained by calorie restriction 65 , which extends lifespan in several organisms. Based on these reports, along with the findings in this study, we propose that the age-dependent decline in Rubicon and mTORC1 activity collectively causes adipocyte dysfunction via excess autophagy, leading to age-associated metabolic disorders. Importantly, as in the liver and kidney 30 , Rubicon in adipose tissue is also transcriptionally regulated in aging (Fig. 3b). The transcriptional regulator of Rubicon in aged adipocytes remains to be elucidated. Mechanistically, our findings suggest that excess autophagy by Rubicon suppression in adipocytes leads to an LIR/GIM-dependent reduction in SRC-1 and TIF2, which work as coactivators of PPARγ. It remains to be determined why SRC-1 and TIF2 are degraded by autophagy. One clue regarding the answer to this question is provided by the observation that SRC-1 and TIF2 are significantly decreased in adipose tissue from fasted mice (Fig. 6g), or starved 3T3-L1 adipocytes (Fig. 6h). Therefore, we speculate that SRC-1 and TIF2 in adipocytes are transported to the cytosol and degraded by autophagy in the response to outside energy demand, such as a reduction in insulin signalling, and that these reductions cause a decline in PPARγ activity and adipocyte function, leading to energy supply for other tissues. Furthermore, because overexpression of LIR/GIM mutant SRC-1 and TIF2 did not completely rescue the reduction in adipogenic gene expressions by Rubicon knockdown (Fig. 7h) 66 , was recently shown to be degraded in an autophagy-dependent manner in some cell types 67,68 .
In summary, we have revealed that Rubicon plays an essential role in proper maintenance of adipocyte function and systemic metabolic homoeostasis by preventing excess basal autophagy. The absence of Rubicon in adipocytes occurs in aging and leads to metabolic disorders, which are recovered by autophagy deficiency. Thus, we anticipate that inhibition of autophagy in adipocytes could be highly beneficial for human health in an aging society. . The following antibody was used for immunohistochemistry at the indicated dilution: mouse monoclonal anti-PCNA (Santa Cruz Biotechnology, sc-56, 1:3000). The following antibody was used for immunocytochemistry at the indicated dilution: rabbit polyclonal anti-Tomm20 (Abcam, ab78547, 1:500), Alexa Fluor 488 goat antirabbit IgG (Abcam, ab150085, 1:1000). Bafilomycin A1 was purchased from Cayman Chemical.

Methods
Animals. C57BL/6J mice were obtained from the Nihon-CLEA. Rubicon-floxed mice 29 were previously generated in our laboratory. Adipoq-Cre mice 31 and Atg5floxed mice 10 were obtained from Dr. Evan Rosen (Beth Israel Deaconess Medical Center) and Dr. Noboru Mizushima (The University of Tokyo), respectively. Adipoq-Cre mice were crossed with Rubicon-floxed mice or Atg5-floxed mice to generate mice harbouring homozygous deletion of Rubicon or Atg5 specifically in adipocytes, respectively. Rubicon-floxed mice were crossed with Atg5 ad−/− mice to generate mice with homozygous deletion of both Rubicon and Atg5 specifically in adipocytes. All mice used in this study were not littermates because we used Adipoq-Cre mice as controls to exclude the possibility that the phenotypes of the knockout mice arise from Cre-recombinase-mediated cellular dysfunction 69 . All kinds of transgenic mice were initially backcrossed into the C57BL/6J wild-type strain at least six times, and each mouse was generated from the same mouse colony maintained in the C57BL/6J background. The controls were maintained in the same mouse colony and same space. The control colony was used for the generation of the knockout strains. Just after weaning, age-matched mice were randomly co-housed independently from genotype, for the aging and HFD studies. No fighting was observed in the mice. The following primer sets were used for genotyping by PCR: 5′-ACAACGACAATCACACAGAC-3′ and 5′-TGAC-GAGGGGTAATGGATAG-3′ for Rubicon WT and floxed allele; 5′-ACAACGAC AATCACACAGAC-3′ and 5′-AATCCTTCGCCCCTTTTACC-3′ for Rubicon deleted allele; 5′-GAATATGAAGGCACACCCCTGAAATG-3′, 5′-GACAGGTCG GTCTTGACAAAAAGAAC-3′, and 5′-GTACTGCATAATGGTTTAACTCTT GC-3′ for Atg5 WT and floxed allele; 5′-GCTCTTAGTCCAGAACCTAAACC-3′ and 5′-GTACTGCATAATGGTTTAACTCTTGC-3′ for Atg5-deleted allele; 5′-GCATTACCGGTCGATGCAACGAGTGATGAG-3′ and 5′-GAGTGAAC-GAACCTGGTCGAAATCAGTGCG-3′ for Cre. Unless otherwise specified, all mice used for the experiments were 21-week-old males. These mice were maintained on an NCD in 12-h light/12-h dark cycles. The ambient temperature and humidity was 23 ± 1.5°C and 45 ± 15%, respectively. Food and water were provided ad libitum. Sequence information about primers used for genotyping is available upon request. Five-week-old mice were fed with HFD for 16 weeks. Pioglitazone (Tokyo Chemical Industry) was administered at 0.02% (w/w) for 3 weeks from 21week-old by admixing it with food. Experimental procedures using mice were approved by the Institutional Committee of Osaka University.
Plasmids. pmCherry-Parkin was purchased from Addgene. pENTR1A and pcDNA3.1 were purchased from Invitrogen. DNA fragments containing attR1, CmR, ccdB and attR2 were amplified from pAd/CMV/V5-DEST (Invitrogen) and subcloned into pcDNA3.1 (pcDNA3.1-DEST2). pcDNA3.1-DEST2-Nt-3xFLAG was generated from pcDNA3.1-DEST2 using In-Fusion (Takara). Full-length mouse SRC-1 and TIF2 sequences were amplified from mouse liver cDNA and cloned into pENTR1A (pENTR1A-WT-SRC-1 and pENTR1A-WT-TIF2). Mutant SRC-1 and TIF2 vectors were generated from wild-type vectors using In-Fusion (pENTR1A-W288A-SRC-1 and pENTR1A-W296A-TIF2). The sequences inserted into the pENTR1A plasmids were transferred into pcDNA3.1-DEST2-Nt-3xFLAG using Gateway LR Clonase (Invitrogen).  6 SRC-1 and TIF2 are degraded by autophagy, and thus are significantly reduced in Rubicon-ablated adipocytes or aged adipocytes. a Immunoblotting to detect SRC-1 and TIF2 in Luciferaseor Rubicon-knockdown 3T3-L1 cells treated with or without 125 nM Baf A1 for 8 h. Knockdown was performed for 48 h starting on day 8. n = 3 independent experiments. b Immunoblotting to detect SRC-1 and TIF2 in wild-type eWAT depots explanted in DMEM, treated with or without 20 mM ammonium chloride and 200 μM leupeptin for 2 h. n = 3 independent experiments. c, d Immunoblotting to detect SRC-1 and TIF2 in eWAT depots of 21-week-old mice of the indicated genotypes on an NCD. n = 5 mice. e Immunoblotting to detect the indicated proteins in eWAT depots of 3-or 25-month-old wild-type mice on an NCD. n = 3 mice. f Immunoblotting to detect SRC-1 and TIF2 in the eWAT depots of 3-, 12-, 18-month-old control mice on an NCD. n = 4 mice. g Immunoblotting to detect the indicated proteins in eWAT depots of fed or 48-h-fasted 3-month-old wild-type mice. n = 3 mice. h Immunoblotting to detect the indicated proteins in 3T3-L1 cells. The cells on day 10 were subjected to starvation for the indicated times. n = 3 independent experiments. i Relative mRNA expression of adipogenic genes in 3T3-L1 cells. The cells on day 10 were subjected to starvation for the indicated times. n = 4 independent experiments. Quantification data are shown in the graphs at the right of each blot. Error bars indicate means ± SEM. Data were analysed by two-tailed Student's t test (b, c, i), one-way ANOVA followed by Tukey's test (d, f) or Dunnett's test (a).