Male C57BL/6J mice (23.0 ± 0.9 g, 6 weeks of age) were obtained from Shanghai SLAC Laboratory Animal Co., Ltd (Shanghai, China). All animals were maintained under standardized conditions (23 ± 1°C; 12-h light-dark cycle). All animal procedures were performed in full accordance with the “Regulation for the Use of Experimental Animals” of Zhejiang Province, China. This study was specifically approved by the Animal Care and Use Committee of Zhejiang University (ETHICS CODE Permit no. ZJU20170529). Animals were fed commercial feed (#P1101F, Slacom, Shanghai, China) composed of fish meal, wheat, corn, soybean meal, wheat bran, vitamin, mineral and amino acids that contained at least 20.5% crude protein, 4% crude fat, 1.32% lysine, and 0.78% methionine + cystine, and ≤ 5% crude fiber, and ≤ 8% crude ash. After acclimatization for 10 days, mice were randomly assigned into four groups (12 mice/group): control (C), intermittent fasting (F), melatonin (M), and intermittent fasting plus melatonin (MF). The C and M groups mice were provided with ad libitum access to food and water, while the F and MF groups underwent alternative-day feed deprivation (15 cycles total). Melatonin was administered in the drinking water of the M and MF groups. Due to the experimental design, we are not able to determine the melatonin intake for each mouse; instead, melatonin was titrated to an averaged dose of 10 mg/kg body weight and provided in the drinking water. The melatonin water was prepared daily and kept in a normal bottle with an aluminum foil cover to prevent light-induced melatonin degradation.

Body weight and food/water intake were measured daily. The mice were euthanized at the end of the trial with an intraperitoneal injection of pentobarbital sodium (50 mg/kg body weight), and blood and tissue samples were collected.

Blood samples were centrifuged at 3,000 × g for 10 min at 4°C to produce serum samples. The levels of serum glucose, triglyceride (TG), total cholesterol (TC), low density lipoproteins cholesterol (LDL-C), and high density lipoproteins cholesterol (HDL-C) were quantified using corresponding ELISA kits (no. F006, no. A110-1, no. A111-1, no. A113-1, and no. A112-1, respectively; Nanjing Jiancheng Bioengineering Institution, Nanjing, China). Serum levels of alanine aminotransferase (ALT) (no. C009) and aspartate aminotransferase (AST) (no. C010) were measured using kinetics-based assays with commercially available kits (Nanjing Jiancheng Bioengineering Institution, Nanjing, China) and an automatic biochemistry analyzer (SELECTA XL; Vital Scientific, Newton, MA, USA) according to protocols provided by the manufacturers. Serum melatonin and insulin levels were measured using ELISA kits (no. H256-1-2 and no. H203-1-2, respectively; Nanjing Jiancheng Bioengineering Institution, Nanjing, China).

H&E staining was performed as previously described (24). Epididymal adipose, liver, ileal, and colonic samples were fixed, dehydrated, and paraffin embedded. The sections were prepared and subsequently stained with H&E. Photomicrographs were obtained using an optical microscopy system (Olympus Corporation, Tokyo, Japan). Quantitative measurement of adipocyte count, ileal villi height, ileal crypt depth, and colonic fold height were conducted with ImageJ (National Institutes of Health, Bethesda, MD, USA). Adipocytes area was calculated using at least three histological sections and a total of 300 adipocytes per mouse. A pathologist evaluated the liver slides.

TEM of liver tissues was performed as previously described (25, 26). The specimen was sectioned using a LEICA EM UC7 ultratome, and sections were stained with uranyl acetate and alkaline lead citrate. Section were visualized with a Hitachi H-7650 TEM.

The entire intestinal contents were collected and used for bacterial 16S rDNA sequencing. High-resolution 16S rDNA genes of the intestinal bacterial flora were analyzed using a previously described method with modifications (27). DNA was extracted from the intestinal contents using the Qiagen DNA Kit (51640, Germany) according to the manufacturer's instructions. The selected region of 16S rDNA amplification was the V3–V4 region PCR products were quantified using Qubit (Invitrogen, USA). An Illumina NovaSeq PE250 (Illumina, San Diego, USA) was used for on-board sequencing, followed by bioinformatics analysis. Chimeric sequence detection and de novo operational taxonomic units (OTU) picked up with 0.97 identities were performed using Usearch (version 7.0) and UPARSE (http://drive5.com/uparse/), respectively (28). Alpha/beta diversity and the relative abundance of bacteria at the phylum and genus level were analyzed with QIIME1 (V1.9.1). Principal coordinates analysis (PCoA) of Weighted-Unifrac distance was used to visualize the bacteria communities of each of the four groups. PICRUSt was used to predict the functional composition of the metagenome, and the Kyoto Encyclopedia of Genes and Genomes (KEGG) Orthology functional database was used as reference for this prediction (29). Linear discriminant analysis (LDA) effect size (LEfSe) was used to estimate the effect of each component metabolic pathway on the difference, and to find out the metabolic pathway that had a significant difference on the sample division (The default screening condition is LDA >2).

About 50 mg sample with 500 μL pre-cold extraction mixture [methanol/chloroform (v:v) = 3:1] and 10 μL internal standard (L-2-Chlorophenylalanine, 1 mg/mL stock) were added into a 2 mL tube. Then the sample was vortexed for 30 s and homogenized with ball mill for 4 min at 35 Hz, followed by ultrasonication for 5 min on ice water. The mixture was extracted and centrifuged at 13,300 × g for 15 min at 4°C. Supernatant was collected and analyzed using gas chromatography coupled with a time-of-flight mass spectrometer (GC-TOF-MS) according to a previous method (30), which did not distinguish between positive and negative, and the results did not distinguish between polarity and non-polarity. The LECO-Fiehn Rtx5 database was used to identify metabolites by matching the mass spectrum and the retention index (31). Peaks detected in less than half of the QC samples or with a RSD >30% in the QC samples was removed (32). Follow-up analysis of the obtained data was performed using existing methods (33). The final dataset containing the information of peak number, sample name and normalized peak area was imported into MetaboAnalyst 4.0 (https://www.metaboanalyst.ca/) for multivariate analysis. Partial least squares discrimination analysis (PLS-DA) and t-test were performed between the two groups, with a false discovery rate (FDR) adjusted P < 0.05 and variable importance in projection (VIP) > 1.5 used to identify a significant different in metabolites. Commercial databases including KEGG (http://www.genome.jp/kegg/) were used for pathway enrichment analysis.

Statistical analysis was performed using SAS software (SAS9.04, Cary NC, USA). All data are presented as mean ± standard error (SE). Significant differences were identified using 2-factor ANOVAs with a general linear model for this 2 x 2 factorial experiment and IF x melatonin interaction was tested, followed by Duncan's multiple range tests. A P-value of <0.05 was considered statistically significant.

Compared with the C group, IF and melatonin did not affect cumulative food intake (Figure 1A). However, IF (F and MF) led to significantly lower BW (Figure 1B) and significantly lower BW gain (50 and 64% lower than C, respectively, Figure 1C). Food intake after fasting for 24 h sharply increased (Supplementary Figure 1A). When the mice fasted, their water intake was also lower than normal (Supplementary Figure 1B). The average dose of melatonin over the study period was shown in Supplementary Figure 1C (one cycle means a non-fasting day and a fasting day), and confirms that the average daily melatonin intake was close to 10 mg/kg. In addition, epididymal fat mass in the F and MF groups was significantly lower than in the C group (25 and 30%, respectively; Figure 1D). Melatonin alone, and the use of melatonin with fasting did not have a significant impact on BW and epididymal fat mass (P > 0.05).

The F and MF groups had significantly lower serum glucose (66 and 61% lower, respectively, P < 0.05, Figure 2A), TC (29 and 26% lower, respectively, P < 0.05, Figure 2B) and TG (31 and 34% lower, respectively, P < 0.05, Figure 2C) levels than the C group. There was no significant difference in serum glucose, TC, or TG between the M and C groups, or between the MF and F groups. However, the MF group had significantly lower serum glucose (61% lower, P < 0.05, Figure 2A), TC (19% lower, P < 0.05, Figure 2B) and TG (27% lower, P < 0.05, Figure 2C) than the M group. Compared with the C group, MF group mice had significantly lower serum LDL-C (25% lower, P < 0.05, Figure 2D). No significant difference was observed in serum HDL-C (Figure 2E).

Serum insulin level was significantly lower in the F (36% lower) and MF (35% lower) groups than the M group (P < 0.05, Figure 2F). Serum melatonin level was significantly higher in the M and MF group than in the C (17 and 21% higher, respectively, P < 0.05) and F (9 and 13% higher, respectively, P < 0.05) groups (Figure 2G). Compared with the C group, the F group had significantly lower serum ALT (37% lower, P < 0.05, Figure 2H) and AST (32% lower, P < 0.05, Figure 2I), whereas the MF group had significantly lower serum AST (41% lower, P < 0.05) and an equivalent serum ALT. There was little interaction between IF and melatonin in serum parameters (P > 0.05).

The morphology of ileal and colonic tissues was visualized with H&E staining and shown in Supplementary Figures 2A,B, respectively. Compared with the C group, the F group mice had significantly higher villus height (13% increase, P < 0.05) and villus height/crypt depth ratio (20% increase, P < 0.05, Supplementary Figure 2C). The villus height/crypt depth of the MF group was also significantly higher than that of the C group (15% increase, P < 0.05). There were no significant differences in ileal crypt depth identified. There were no significant differences in colonic fold length between groups (P > 0.05, Supplementary Figure 2D). No significant interaction between IF and melatonin was found on villus height, crypt depth and colonic fold length (P > 0.05).

Histologic analysis of the epididymal white adipose tissue (eWAT) revealed an increase in the number of multilobular adipocytes in IF mice (Figure 3A), a typical characteristic of beige adipocytes. F and MF group mice had a significantly smaller adipocyte size (37 and 20% decrease, respectively, P < 0.05, Figure 3B), but a significant increase in adipocyte number per square millimeter (78 and 32% increase, respectively, P < 0.05, Figure 3C), compared with the corresponding non-fasting group (C and M, respectively). M group mice had a significantly higher number of adipocytes per square millimeter than C group mice (30% increase, P < 0.05). The interaction between fasting and melatonin had significant effects on adipocytes size and number (P < 0.05, Figures 3B,C).

Liver tissue morphology was normal in all groups, without obvious signs of inflammation or lesions (Figure 4). However, sections of the liver assessed with TEM showed a greater number of lipid droplets in the fasting groups (F and MF) compared with the C and M groups (Figure 5).

The rarefaction curve of observed species (Supplementary Figure 3A) and Chao 1 index (Supplementary Figure 3B) of gut microbiota plateaued when the read increased to a certain level. The F and M groups had significantly lower alpha diversity (7.7 and 8.7% lower, respectively, P < 0.05) than the MF group (Figure 6A). Adonis analysis showed no noticeable separation between the M and C groups (Figure 6B). However, the fasting groups (F and MF) were clearly separated from the C and M groups (Figure 6B).

LEfSe analysis showed the following metabolic pathway (LDA >2) between each group: secretion system, biosynthesis of unsaturated fatty acid for the F group; phosphotransferase system and other ion coupled transporters for the MF group; membrane and intracellular structural molecules and the citrate cycle TCA cycle for the M group (Figure 6C).

There were 32 genera of intestinal flora that were significantly different between the groups. The 6 genera with the most significant difference were shown in Figures 7A–F. Compared with the C group, the F group exhibited a significant increase in the abundance of Lactobacillus (fold change, FC = 28.87, P < 0.05, Figure 7A), Ruminococcus (FC = 12.21, P < 0.05, Figure 7B) and Akkermansia (FC = 24.19, P < 0.05, Figure 7C) and significantly lower abundance of Helicobacter (FC = 0.34, P < 0.05, Figure 7D), Prevotella (FC = 0.05, P < 0.05, Figure 7E) and Parasutterella (FC = 0.31, P < 0.05, Figure 7F). Compared with the C group, the M group had a significantly lower abundance of Prevotella (FC = 0.67, P < 0.05, Figure 7E). In addition, the interaction between fasting and melatonin had a significant effect on the abundance of Akkermansia (P < 0.01). Relative abundance by phylum is shown in Supplementary Figure 3C. IF but not melatonin had significant effects on intestinal Firmicute, Bacteroides and the radio of Firmicutes/Bacteroides (P < 0.05, Supplementary Figures 3D–F). In addition, the interaction between fasting and melatonin had a significant effect on the relative abundance of intestinal Firmicute and Bacteroides (P < 0.05).

The partial least-squares discriminate analysis (PLS-DA) scores plot showed a clear separation between all treatments (Figure 8A). And the permutation test was evaluated based on the corresponding PLS-DA model (Figure 8B). A total of 482 metabolites were detected by metabolomics (Supplementary Table 1) and the top 19 metabolites with the highest abundance were shown in Figure 8C. Differences in metabolites between groups were shown in Supplementary Figure 4. Fasting significantly decreased glucose abundance (P < 0.05, Supplementary Figure 4A) but increased ribose, alanine, glycine, valine, isoleucine, tyrosine and 2-ketoadipate abundance (P < 0.05, Supplementary Figures 4B–H). Melatonin alone had a significant effect on 2-ketoadipate abundance (P < 0.05, Supplementary Figure 4H). The interaction between fasting and melatonin had significant effects on alanine, valine and isoleucine abundances (P < 0.05, Supplementary Figures 4C,E,F).

The KEGG enrichment pathways of different metabolites were further analyzed (Figure 8D). The main enrichment pathways were: galactose metabolism; starch and sucrose metabolism, alanine, aspartate and glutamate metabolism, glycolysis or gluconeogenesis, aminoacyl-tRNA biosynthesis and the synthesis and degradation of ketone bodies. Metabolites related to butyric acid metabolism including 4-hydroxybutyrate and 3-hydroxybutyric acid are shown in Figures 9A,B. Fasting had a significant effect on intestinal 4-hydroxybutyrate (Figure 9A) and 3-hydroxybutyric acid (Figure 9B) (P < 0.05).

We analyzed correlations between the phenotype and the differences of gut microbita or metabolome (Supplementary Figure 5A). Serum glucose and TC levels were positively correlated with the Prevotella abundance. The epididymal fat mass and serum glucose level were negatively correlated with the metabolites alanine, isoleucine and methionine.

We also analyzed correlations between differential bacteria (genus) and metabolites (Supplementary Figure 5B). Bacteria from genera Lactobacillus, Olsenella, Desulfovibrio, and Parvibacter were positively correlated with the metabolites ribose, 3-hydroxybutyric acid, and the amino acids methionine, alanine, isoleucine and valine. However, these metabolites were negatively correlated with the genera Prevotella, Parasutterella, Parabacteroides, Helicobacter, Paraprevotella, Barnesiella, and Vampirovibrio.