All animal experiments were performed after approval by the Institutional Animal Care and Use Committee (IACUC-14030182) of Nanjing Medical University. Institute of Cancer Research (ICR) mice and Sprague-Dawley rats were purchased from Shanghai SLAC Laboratory Animal Co. Ltd. (Shanghai, China), while Lep^−/+ rats were a kind gift from Dr. Guoping Fan's lab [20]. The animals were acclimated for 1 week before the study began. The Lep^−/+ rats were crossed to obtain wild-type (WT) and Lep^−/− rats. The animals were housed in a temperature- and humidity-controlled (23℃±1℃, 53%±2%) room and maintained on a 12-hour/12-hour light/dark cycle with food and water available ad libitum. For generation of diet-induced obese (DIO) rats, 9-week-old rats were fed a 60% kcal HFD (Cooperative Medical Biological Engineering Co. Ltd., Nanjing, China) until the age of 20 weeks.

For the oral glucose tolerance test, animals were fasted from the beginning of the dark cycle. After 15 hours of fasting, blood glucose levels were measured from the tail vein. Subsequently, dextrose solution (2 g/kg BW) was administered by gavage. Blood glucose levels were measured from the tail vein 15, 30, 60, 90, and 120 minutes following glucose treatment with an automated reader (Roche Ltd., Welwyn Garden City, UK).

For the insulin tolerance test, the mice were fasted as mentioned above. Insulin (1 IU/kg BW) was intraperitoneally injected. Blood glucose levels were measured 15, 30, 60, 90, and 120 minutes following insulin injection from the tail vein with an automated reader.

To monitor the effects of celastrol treatment on metabolic parameters, 20-week-old DIO rats were housed individually and given free access to food and water (n=3 for each group). They were orally administered celastrol for 1 week for adaptation, and BW, oxygen consumption, food intake, the level of physical activity, etc., were measured for the next 3 days with a TSE Labmaster Caging System (TSE Systems, Bad Homburg, Germany). The first recorded cycle following the first day of measurement in the metabolic chambers was named the dark 1 cycle, which was followed by the light 1, dark 2, light 2, dark 3, and light 3 cycles.

Blood from the tail was collected in nonheparinized tubes once a week and transferred to ice-cold Eppendorf tubes. The tubes were centrifuged at 2,000 ×g for 30 minutes at 4℃. We collected the serum layer and stored it at −80℃ for long-term storage. Serum leptin levels were measured with enzyme-linked immunosorbent assay (ELISA) kits (Merck-Millipore, Darmstadt, Germany). Triglyceride and cholesterol levels were assayed by the Department of Animal Core Facility of Nanjing Medical University.

Dissected livers and abdominal white adipose tissue (WAT) from DIO rats that were treated with vehicle or celastrol for 3 weeks were fixed in 10% neutral buffered formalin and embedded in paraffin according to standard procedures. Tissue sections (5-mm thick) were stained with hematoxylin and eosin (H&E) kits following the manufacturer's instructions (Beyotime, Shanghai, China). Histopathological slides were scanned using Pannoramic Scan (3DHISTECH Ltd., Budapest, Hungary), which were taken at 40× magnification. Three photomicrographs per rat were compiled and analyzed using Adobe Photoshop version CS6 (Adobe, San Jose, CA, USA); ImageJ, version 2 (National Institutes of Health, Bethesda, MD, USA).

To prepare total lysates from the hypothalamus, samples were lysed in ice-cold tissue lysis buffer (25 mM Tris-HCl, pH 7.4; 100 mM NaF; 50 mM Na₄P₂O₇; 10 mM Na₃VO₄; 10 mM EGTA; 10 mM EDTA; and 1% NP-40; supplemented with phosphatase and protease inhibitors) using Misonix Sonicator 3000 (Misonix, Nassau County, NY, USA). Following 2 minutes of homogenization, we centrifuged the samples at 13,400 ×g for 20 minutes at 4℃. Supernatants from the hypothalamus were used as protein extracts. We measured the protein concentration of each sample using a BCA protein assay kit (Beyotime), and an equal amount of protein from each sample was denatured in sample loading buffer by boiling at 95℃ for 5 minutes. The proteins were separated on sodium dodecyl sulfate polyacrylamide (SDS-PAGE) gels and transferred onto polyvinylidene fluoride membranes at 4℃ at 100 V for 2 hours. The membranes were blocked for 1 hour in tris-buffered saline (TBS) with 5% bovine serum albumin. The membranes were then incubated with primary antibody in TBS with 0.1% Tween-20 (TBST) at 4℃. After overnight incubation, the membranes were washed three times in TBST for 10 minutes and then incubated with secondary antibody in TBST for 1 hour at room temperature. Following three cycles of 15-minute washes with TBST, the membranes were developed using a chemiluminescence assay. Intensity analysis was performed on scanned Western blot images using ImageJ software. The primary antibodies used in this study included phospho-protein kinase-like endoplasmic reticulum kinase (PERK) (Thr980; Cell Signaling Technologies, Beverly, CA, USA), phosphorylated-signal transducer and activator of transcription 3 (p-STAT3, Tyr705; Cell Signaling Technologies), β-actin (Cell Signaling Technologies), PERK (T-19; Santa Cruz Biotechnology Inc., Santa Cruz, CA, USA) and STAT3 (F-2; Santa Cruz Biotechnology Inc.).

Total RNA was extracted from the hypothalamus and WAT and brown adipose tissue (BAT) tissues with TRIzol (Invitrogen, Carlsbad, CA, USA), and 1 µg of total RNA was reverse transcribed into cDNA with a PrimeScript Kit (Takara Bio Inc., Kusatsu, Japan) according to the manufacturer's instructions. Quantitative real-time polymerase chain reaction was performed in triplicate for each sample and analyzed with the ABI 7900 system (Applied Biosystems, Waltham, MA, USA). The data were expressed as arbitrary units after normalization to the levels of expression of the internal control glyceraldehyde 3-phosphate dehydrogenase (Gapdh) for each sample. The sequences of the primers used in this study are listed in Supplementary Table 1.

Stool samples from random rats from the vehicle-, celastrol-, and celastrol+antibiotics-treated groups (n=6 for each group) were freshly collected and stored at −80℃ until DNA isolation. The bacteria in the feces were extracted using a QIAamp DNA Stool Mini Kit (Qiagen, Hilden, Germany) according to the manufacturer's instructions. Then, the DNA was sent to Majorbio Bio-Pharm Technology Co. Ltd. (Shang, China) for 16S rRNA gene sequencing using primers for the V3 region (forward primer: CCAGACTCCTACGGGAGGCAG; reverse primer: CGTATTACCGCGGCTGCTG). After quality control and trimming of the adaptors, paired-end reads were joined and mapped to the Greengenes 13_8 release database using Qiime software version 1.8 (http://qiime.org/). Operational taxonomic units were picked against the Greengenes database using a 97% similarity threshold. To adjust for differences in sequencing depth, all samples were subsampled to the same number in the subsequent analysis. The workflow for microbiome data analysis [22] was used to filter unidentified and rarely prevalent phyla and compare the differential bacterial abundance among the groups with default settings. Plots were made by the Phyloseq package using R version 3.5.1 (R Foundation for Statistical Computing, Vienna, Austria).

For transcriptome profiling, three hypothalamic tissues from vehicle- and celastrol-treated rats were randomly collected and sent to Novogene Co. Ltd. (Beijing, China) for RNA extraction, library preparation and mRNA sequencing on an Illumina Hiseq 4000 platform (Illumina Inc., San Diego, CA, USA). The RNA-seq data were filtered and processed with STAR to generate read alignments with rn6. Raw read counts for annotated genes were obtained with HTSeq, normalized and analyzed using DEseq2 in accordance with the workflow for gene-level exploratory analysis [23]. A false discovery rate-adjusted P value <0.05 was used to identify the differentially expressed genes. We performed pathway enrichment analysis using Ingenuity Pathway Analysis (IPA; Qiagen).

For statistical analyses, Student's t-test or the rank sum test were used for comparisons between two groups. One-way analysis of variance (ANOVA) followed by Dunnett's post hoc test was used for multiple comparisons versus the vehicle-treated group. The results are shown as the mean±standard error using GraphPad Prism version 7.03 software (GraphPad, San Diego, CA, USA). P<0.05, P<0.01, and P<0.001 were considered as the threshold of statistical significance for all analyses.

The RNA-seq data from this publication have been submitted to the Sequence Read Archive (SRA) database under the SRA accession number PRJNA494595.

According to the effective dose identified in a previous study [1416], we first treated DIO rats with celastrol at doses ranging from 15 to 250 µg/kg BW by oral gavage in the pilot test. As shown in Supplementary Fig. 1, the BW of the DIO rats did not decrease until the dose increased to 250 µg/kg BW To ensure the anti-obesity effect of celastrol, a dose of 500 µg/kg BW, which equals approximately 1,000 µg/kg BW in mice, was chosen for subsequent experiments.

The administration of celastrol significantly reduced the BW of the DIO rats from 813.28±12.20 g on day 0 to 787.69±12.13 g over the 3-week period, representing 3.11%±1.02% weight loss compared to the initial weights (Fig. 1A and B); however, the BW of the vehicle-treated rats showed a slight increase of 0.48%±0.50%. Consistent with this reduction, we observed that the livers of the vehicle-treated rats were filled with large lipid droplets and that lipid accumulation in the celastrol-treated rats decreased. The size and number of adipocytes in abdominal WAT and triglyceride and cholesterol levels, however, were not altered (Fig. 1C, F, and I). These results indicated improved lipid metabolism in the liver, but not in white fat, after treatment with celastrol.

Before treatment with celastrol, the fasting blood glucose levels of the rats from the vehicle- and celastrol-treated groups were 7.28±0.46 and 7.03±0.12 mmol/L, respectively, representing a phenotype of prediabetes. After 3 weeks of treatment, the impaired fasting glycemia was slightly improved (6.29± 0.16 and 6.23±0.26 mmol/L, respectively), but the rats still showed impaired tolerance to a glucose load (Fig. 1D and E, Supplementary Fig. 2A and B). Regarding response to insulin, the glucose levels of the celastrol-treated rats were slightly diminished and restored to the level of the vehicle-treated group after intraperitoneal insulin injection (Fig. 1G and H), although no significant difference in the area under curve (AUC) was observed (Supplementary Fig. 2C and D). Collectively, these results demonstrated that 3 weeks of treatment with celastrol ameliorated obesity in DIO rats.

The gut microbiota, an intermediate between the host and the environment, comprises trillions of bacteria and contributes to energy harvesting and metabolism regulation [24]. Considering celastrol was orally administered, an interaction between celastrol and the gut microbiota may have occurred, but there was no evidence of this association. To assess this association, we collected the feces of the rats from the two groups after the 3 weeks of treatment on the second morning after cage change, and the microbiota composition in the feces was analyzed using 16S rRNA sequencing. Weighted UniFrac-based principal coordinate analysis (PCoA) revealed a scattered distribution of microbiota in the vehicle-treated group, reflecting a state of HFD feeding, while the microbiota of the celastrol-treated rats was concentrated (Fig. 2A). Within-habitat diversity, i.e., α-diversity, showed that the microbiota composition of the celastrol-treated rats was more abundant than that of the vehicle-treated group (Fig. 2B). Based on the negative correlation between α-diversity and the onset of obesity, the promotion of the gut microbiota accompanied by weight loss in the celastrol-treated rats was consistent with previous findings in humans [25]. As shown in Fig. 2C, there was an obvious difference in the relative abundance between the two groups at the phylum level, with Bacteroidetes, Firmicutes, and Proteobacteria as the major constituent phyla. After 3 weeks of treatment with celastrol, the relative abundance of Bacteroidetes in the feces was significantly increased by approximately 85% compared to that in the vehicle-treated group, while the relative abundance of Firmicutes was diminished by approximately 26%; this result is compatible with the aforementioned studies in murine models of obesity [24]. This finding indicated that celastrol improved the homeostasis of the microbiota composition in DIO rats, but there was still a lack of understanding of the dynamic changes.

To determine whether there was an association between the duration of treatment and the microbiota composition, we collected the feces of the celastrol-treated rats after 1, 2, and 3 weeks of treatment. After treatment with celastrol, the relative abundance of the microbiota in the DIO rats gradually changed and stabilized after 2 weeks (Fig. 2D). Linearized regression revealed a weekly 5.7% elevation in Bacteroidetes and a 6.0% reduction in Firmicutes, suggesting a synchronous change of the two phyla (Fig. 2E). Taken together, our data provided evidence that there was a close relationship between celastrol and the gut microbiota.

Although we demonstrated the association between celastrol and the gut microbiota, it was still unclear whether the microbiota mediated the anti-obesity effect of celastrol. Therefore, we added a mixture of antibiotics, including neomycin, penicillin, streptomycin, vancomycin and metronidazole [21], to the drinking water beginning the second week of treatment with celastrol (Fig. 2F) so that we could disturb the composition of the microbiota, observe its influence on the BW of the DIO rats and confirm the weight loss that was observed during the first week of treatment with celastrol. A separation between celastrol alone and the combination of celastrol and antibiotics is shown in the PCoA plot, and there was a weakened α-diversity in the rats after the addition of antibiotics, which suggested that 2 weeks of antibiotics administration disrupted the microbiota homeostasis promoted by celastrol (Fig. 2G and H). Consistent with the previous results, the BW of the rats in the celastrol+antibiotic-treated group decreased at a rate equivalent as that of the rats in the celastrol only-treated group during the first week (from day 1 to day 7) before the addition of antibiotics, which validated the anti-obesity effect of celastrol (Fig. 1B). Nevertheless, the rate slowed after the addition of antibiotics, and the slope of the broken line graphs indicates two totally different statuses of weight loss. By generating another group of rats that received only antibiotics, we aimed to control for the effects of the antibiotics themselves. As shown in Fig. 1B, the BW of the rats in this group diminished immediately after drinking water with antibiotics, and this was supposed to be induced by the disturbance of the gut microbiota [26]. Weight gain was observed in these rats 5 days later. These results implied that the gut microbiota but not antibiotics mediated the anti-obesity effect of celastrol.

There is growing evidence that the bidirectional communication system between the gut and the brain, in other words, the gut-brain axis, is associated with metabolic diseases [27]. This is also a route through which the gut microbiota may impact neurodevelopmental processes and brain functions. To profile the changes in the hypothalamus after 3 weeks of treatment, we extracted RNA and performed transcriptome sequencing. The vehicle- and celastrol-treated rats were clustered separately (Fig. 3A and Supplementary Fig. 3A) in the principal component analysis plot, and there were 13 downregulated genes and 153 strongly upregulated genes in the celastrol-treated group. We analyzed enriched pathways with IPA, and it was revealed that G-protein coupled receptor signaling, dopamine-DARPP32 feedback in cyclic adenosine monophosphate (cAMP) signaling, protein kinase A signaling, etc., were enriched (Fig. 3B). Notably, we found that genes that participate in leptin signaling, including protein kinase cAMP-activated catalytic subunit beta (Prkacb), forkhead box Ow 1 (Foxo1), and adenylate cyclase 5 (Adcy5), were differentially expressed, indicating the potential involvement of leptin signaling in the anti-obesity effect of celastrol. However, the mRNA expression levels of Pomc, Agrp, Npy, etc., were expressed at comparable levels in the two groups (Fig. 3F and Supplementary Fig. 3B).

To determine the connection between leptin and celastrol-induced weight loss, we tested successive serum leptin levels, but there was no significant difference between the vehicle- and celastrol-treated rats after 1, 2, or 3 weeks of treatment (Fig. 3E). Nevertheless, the level of leptin in the celastrol-treated group showed a tendency to decrease, reflecting the association of a negative feedback loop of leptin with a reduction in body fat, which has been reported by others [28]. For leptin signaling in the hypothalamus, we found that the phosphorylation of STAT3, a downstream effector of leptin, at Tyr705 was significantly increased in the celastrol-treated group, which implied the activation of leptin signaling (Fig. 3C and D). In contrast to that in the celastrol-treated group, the ratio of p-STAT3 to total STAT3 protein in the celastrol+antibiotic-treated group was slightly reduced, which was in accordance with the slowing of weight loss (Fig. 1B). Together, our data demonstrated that the hypothalamic leptin signaling pathway was activated after treatment with celastrol even though the serum leptin level did not change; this may have been due to the improved leptin sensitivity in the hypothalamic center for energy homeostasis.

It is well established that leptin and insulin resistance in obesity results from endoplasmic reticulum (ER) stress and that improvements in ER homeostasis induced by chaperones such as 4-phenylbutyric acid (4-PBA) contributes to the improved sensitivity to leptin [29]. We thus analyzed the level of PERK protein as well as its phosphorylation at Thr980, but the reduction in PERK was not obvious (Supplementary Fig. 3C–F), implying that the involvement of ER homeostasis still needs to be validated by further studies.

Given the activation of leptin signaling in celastrol-treated rats, heterozygous (Lep^−/+) and homozygous leptin knockout (Lep^−/−) rats were used to verify the role of leptin in celastrol-induced weight loss. We orally administered different doses of celastrol to the Lep^−/+ rats. As shown in Fig. 4A and B, the rats administered 500 µg/kg BW celastrol maintained a stable BW, and the difference between the BWs of the 500 µg/kg BW celastrol- and vehicle-treated groups was not statistically significant. When the dosage increased to 1,000 µg/kg BW, the percent decrease in BW was markedly altered beginning the 18th day of treatment. As the number of genomic copies of leptin was decreased in the Lep^−/+ rats compared to that in the WT rats, the elevation of the celastrol dose helped to restore its function and decrease BW. For Lep^−/− rats, however, even 1,000 µg/kg BW celastrol did not decrease the BW (Fig. 4C and D), suggesting that the effectiveness of celastrol is partially dependent on leptin.

Obesity is the result of an interrupted balance between the intake and consumption of energy. To lose weight, an organism must consume more calories than it takes in; however, whether celastrol affects intake or consumption remains unresolved. We first recorded the daily food intake of the rats, which was approximately 20 g, and no difference was observed between the WT, Lep^−/+ or Lep^−/− rats treated with celastrol and those treated with vehicle (Fig. 5A–C and Supplementary Fig. 4). This implied that the oral administration of celastrol at the present dose had no impact on food intake, although the signaling pathway of leptin, the appetite hormone, was activated.

Therefore, we conducted a metabolic chamber experiment to examine whether EE, the respiratory exchange ratio (RER) or the level of physical activity were altered after 1 week of treatment, when the gut microbiota had improved. Regarding RER, the rats from the two groups maintained relatively stable respiratory function during the dark and light cycles (Fig. 5D and G), although the RER value was slightly but not significantly increased in the celastrol-treated rats. The RER reflected the relative use of carbohydrates versus lipids as a source of energy, and the results implied a potential switch of fuel to carbohy drates in the celastrol-treated group. The level of physical activity was the total distance that the rats moved in the chamber. The celastrol-treated rats exhibited higher activity than that of the vehicle-treated rats in the light phase, but a statistically significant difference was not observed (Fig. 5F and I). Regarding EE, the red broken lines were obviously higher than the black lines in the dynamic longitudinal chart, indicating the enhancement of thermogenesis (Fig. 5E). By calculating the average levels in the dark and light phases, we found that the rats treated with celastrol did expend more energy than those in the vehicle-treated group and that the difference was more obvious in the light cycle, which may have been due to hyperactivity, which was revealed by increased physical activity (Fig. 5H). Our data demonstrated that celastrol-induced weight loss resulted from the negative regulation of energy homeostasis and that the EE was enhanced while there was no alteration in food intake.

To examine increased EE at the level of transcription, we assessed the expression of some genes involved in fatty acid metabolism in both abdominal WAT and interscapular BAT. In WAT, markers of browning, including uncoupling protein 1 (Ucp1), cell death inducing DFFA like effector a (Cidea), and PPARG coactivator 1 alpha (Ppargc1a), were not differentially expressed (Fig. 5J), and a slight but not marked induction of carnitine palmitoyltransferase 1A (Cpt1a), adiponectin, C1Q and collagen domain containing (Adipoq), and fatty acid synthase (Fasn) was observed. In BAT, we found strong activation of the entire gene set, especially Cpt1a, lipoprotein lipase (Lpl), Adipoq, Fasn, and fatty acid binding protein 4 (Fabp4), were expressed at a high level, in rats treated with celastrol (Fig. 5K). Taken together, the results suggested that the upregulation of genes associated with fatty acid metabolism in BAT supported the enhancement of EE in DIO rats treated with celastrol.

Now that we knew that celastrol reduced BW by EE expenditure and that the hypothalamic leptin signaling pathway was activated without an influence on feeding behavior, we next tested whether leptin signaling played a role in such enhancement (Fig. 6A). We fed WT and Lep^−/− rats with the same volume of HFD to test metabolism based on equivalent caloric intake. However, under normal conditions, the weight of the WT and Lep^−/− rats was significantly different compared to that of the Lep^−/− rats, which were extremely obese, and this may have affected metabolism. To avoid the effects of weight, we restricted the food intake of Lep^−/− rats after weaning and acquired a baseline BW that was the same as that of WT rats (Fig. 6B).

As shown in Fig. 6C-H, no alteration was observed in the RER of Lep^−/− rats. Additionally, compared to that of the WT rats, the level of physical activity of the Lep^−/− rats was increased in the light cycle but reduced in the dark cycle. However, the EE of the Lep^−/− rats was significantly diminished upon treatment with the same dose of celastrol, especially in the light cycle. This finding indicated the participation of leptin in the enhancement of EE by celastrol, which helps to explain the ineffectiveness of celastrol when leptin is homozygously deleted.

For the potential clinical application of celastrol, we aimed to assess the acute toxicity of different doses of celastrol using a mouse model based on the dose of celastrol used in the rats. Mice were orally administered 0, 0.1, 0.5, 2.5, 12.5, or 62.5 mg/kg BW celastrol and then kept in cages for surveillance over a 5-day period (Supplementary Fig. 5A). Consistent with the results of the long-term chronic toxicity test [14], no death occurred even when the dose increased to 62.5 mg/kg BW, which is equivalent to approximately 30 mg/kg BW in rats, which was a high dose in usual tests (Supplementary Fig. 5B). No significant differences in organ coefficients or blood biochemistry were revealed by blinded observation (Supplementary Figs. 6 and 7). This finding suggested the relative safety of celastrol when administered through an oral route. In contrast, the intraperitoneal injection of celastrol induced an adverse reaction that severely inhibited the appetite of the rats. During the pilot test, we dissected the rat abdomen and observed precipitate, which we thought may have been due to the different absorption rates of the solute and solvents, i.e., celastrol, ethanol, and DMSO (Supplementary Fig. 5C and D).