This study was approved by the Ethics. This study was registered at https://clinicaltrials.gov. All procedures were performed following relevant ethical regulations, Helsinki Declaration, and privacy legislation. All participants provided written informed consent.

A total of 25 obese adults with body mass index (BMI) ranging from 28 to 45 kg/m² were recruited from hospitals from April to November 2018. The recruitment process has shown in Supplementary Figure S1 . All patients had well-controlled hypertension, diabetes, hyperlipidemia, hypeluricemia, or metabolic syndrome. The individuals with serious cardiopathy, liver dysfunction, renal disease, hypoglycemia, anemia, malnutrition, hemopoietic system disease, systemic immune system disease, infectious disease, or neurological disorders were excluded. All participants provided written informed consent.

As shown in Figure 1A , in phase I, the participants were on normal diet without restriction on calories and food types for four days. The average daily energy intake of each subject was recorded according to 24-h dietary recalls for four days as their basic energy intake (Park et al., 2018). The IER meals were formulated by a clinical dietitian based on each participant’s basic energy intake and provided to each participant on the energy-restricted diet day. Each meal consisted of 55% carbohydrates, 15% protein and 30% fat. The phase II which is the high-controlled fasting phase included 32 days (8 d each stage, 4 stages in total).At stage 1, 2, 3 and 4, patients were provided with 2/3, 1/2, 1/3 and 1/4 of each participant’s basic energy intake respectively. Patients ate independently without restriction on every other unrestricted energy intake day at home. The Phase III was the low-controlled fasting phase which includes 30 days. During the Phase III, the meals were not provided to the participants. They were gave a list of food (consisted of 55% carbohydrates, 15% protein and 30% fat) to have and asked to have energy-restricted diet (600 call/day for men, 500 call/day for women) on the alternating day at home. The feces and blood samples were collected before phase I (baseline), at the midpoint (HC-W2) and endpoint (HC-W4) of phase II, and at the endpoint of phase III (LC-W4).

Body weight, waist circumference, body fat, systolic blood pressure, and diastolic blood pressure were measured at baseline, midpoint and endpoint of phase II, and endpoint of phase III. Blood samples were collected in the morning after overnight fasting at different timepoints as above mentioned. Serum was obtained by centrifugation at 3,000 rpm for 15 min at 23°C and stored at -80°C until use. Serum levels of fasting plasma glucose, glycosylated hemoglobin, total cholesterol, glutamyl transpeptidase, high-density lipoproteins, low-density lipoproteins, aspartate aminotransferase, alanine aminotransferase, alkaline phosphatase, creatinine, and uric acid were measured by the Laboratory of Hospital.

Brain activity was recorded by resting-state fMRI. The parameters were Repeat time (TR) = 2000 ms, Echo Time (TE) = 35 ms, Field of View (FOV) = 220 mm×220 mm, matrix size = 94×94, slices = 75, slice thickness = 2.2 mm, flip angle = 80°, and Simultaneous MultiSlice (SMS) factor = 3. The regional homogeneity (ReHo) was measured based on Kendall’s coefficient concordance as previously described (Zang et al., 2004). Data were analyzed using DPABI (http://rfmri.org/dpabi) and SPM12 (www.fil.ion.ucl.ac.uk/spm). Briefly, DICOM-format data were converted to NIFTI-format. After discarding the 10 initial scans, data were subjected to slice-timing and head motion correction. Data were then normalized to the Montreal Neurological Institute standard space by linearly registering using echo-planar imaging (EPI) template. After smoothing the data using a 6 mm full width at half maximum, data were detrended to eliminate the linear trend of time courses and filtered with low frequency fluctuations (0.01-0.08 Hz). Six head motion parameters, white matter mask, the whole-brain mask, and cerebrospinal fluid mask were regressed out of the EPI time series. Five participants with abnormal head motion parameters (displacement > 2 mm and/or rotation > 2 degrees) were excluded.

The resting-state ReHo maps were collected. The ReHo values were compared using paired t-test. The region of interest was selected as follows: first, the peak point of the paired t-test results was set as the center of the region of interest; second, the brain regions that were not included in the paired t-test but important for obesity research were selected. The regions of interest were stored as the mask of seed points by Anatomical Automatic Labeling (AAL) template, followed by loading into the results of the paired t-test. The activity of corresponding brain regions was obtained, and the peak coordinates and t values of the brain regions were found.

Fecal samples from 25 participants who successfully lost weight were subjected to metagenomic sequencing. Genomic DNA was isolated from fecal samples using Magen Magpure stool DNA KF kit B (Magen, China) following the manufacturer’s instructions. A paired-end DNA library with insert size of 250 bp was constructed. The DNA libraries were sequenced with paired-end reads of 2 × 100 bp on the BGISEQ-500 platform (BGISEQ, China).

Sequencing data were processed using KneadData v0.7.2 (https://bitbucket.org/biobakery/kneaddata/wiki/Home) as previously described (Liu et al., 2021). Briefly, the raw reads with 50% low quality bases (quality ≤ 20) or more than five ambiguous bases were excluded. The remaining reads were mapped to the human genome (hg19) using Short Oligonucleotide Analysis Package (SOAP) v2.22 (-m 100 -x 600 -v 7 -p 6 -l 30 -r 1 -M 4 -c 0.95), and the matching reads were removed. The high-quality non-human reads were considered clean reads. The clean reads were aligned against 11.4 M human gut microbial gene catalog using SOAP v2.22 (-m 100 -x 600 -v 7 -p 6 -l 30 -r 1 -M 4 -c 0.9) to generate the gene abundance profile. To obtain the taxonomic profiles, metaphlan2 (–input_type fastq –ignore_viruses –nproc 6) was used to generate species profile.

The raw sequence data reported in this paper have been deposited in the Genome Sequence Archive (Genomics, Proteomics & Bioinformatics 2021) in National Genomics Data Center (Nucleic Acids Res 2022), China National Center for Bioinformation / Beijing Institute of Genomics, Chinese Academy of Sciences (GSA: CRA013299) that are publicly accessible at https://ngdc.cncb.ac.cn/gsa (Chen et al., 2021; CNCB-NGDC Members and Partners, 2023).

Alpha diversity (within samples) was assessed by Shannon index. Beta diversity (between samples) was assessed by Bray-Curti’s dissimilarity based non-metric multidimensional scaling (NMDS). Data were analyzed using R 3.2.5, vegan package 2.4-4.

Differentially abundant Gut microbial species and pathways were identified by linear discriminant analysis (LDA). Logarithmic LDA scores threshold was 2.0. Kruskal-wallis test threshold was P < 0.05.

The measurement data were expressed as the mean ± standard deviation. Statistical analysis was performed using SPSS version 17.0 (SPSS, Chicago, IL, USA). Data with normal distributions were analyzed using ANOVA and two-tailed Student’s t-test. Nonnormal distributions were compared using Mann-Whitney, Kolmogorov-Smirnov, or Wald- Wolfowitz test. The ReHo values were compared using paired t-test and subjected to Gaussian random field correction. NMDS data was compared using Wilcoxon rank-sum test and Kruskal-wallis rank sum test. P < 0.05 was considered statistically significant. The correlations between gut microbiota and fMRI results were analyzed by Spearman’s rank correlation. Spearman`s correlation coefficient ≥ 0.2 and P < 0.05 were considered significant correlation.

As shown in Supplementary Figure S1 , a total of 41 participants were recruited, 35 completed the IER, and 25 successfully lost weight as evidenced by continuous, significant reductions in body weights during IER intervention (all P < 0.01 vs. baseline; Figure 1B ). In the 25 participants, by comparing with baseline parameters, we observed sustained, significant reductions in body mass index, body fat, waist circumference, percent of body fat, skeletal muscle (all P < 0.001), systolic blood pressure (all P < 0.05), and serum levels of glycosylated hemoglobin (all P < 0.05) and glutamyl transpeptidase (all P < 0.01) during IER intervention. The body weight of obese participants after IER intervention were significantly lower than those before intervention[(89.92 ± 14.98) vs. (97.53 ± 15.67) kg, P < 0.001]. In addition, diastolic blood pressure and serum levels of fasting plasma glucose, total cholesterol, high-density lipoproteins, low-density lipoproteins, alanine aminotransferase, aspartate aminotransferase, and alkaline phosphatase were significantly decreased on at least one time point during IER (all P < 0.05; Figures 1C–U ). These data suggest that IER not only reduces obesity but also alleviates obesity-related comorbidities such as hypertension, hyperlipidemia, and liver dysfunction.

Then, we performed resting-state fMRI to explore the effect of IER on brain activity. Representative fMRI images are shown in Figures 2A–F . Statistical analysis of ReHo values is shown in Table S1 . We found significant decreases in the Reho values of the left inferior frontal orbital gyrus ( Figure S2D ) and putamen ( Figure S2B ) at midpoint and endpoint of phase II, respectively, compared with those at baseline. We also observed significant decreases in the Reho values of the right inferior frontal orbital gyrus ( Figure S2A ), anterior cingulate cortex ( Figure S2C ), left dorsolateral prefrontal cortex ( Figure S2D ), and right putamen ( Figure S2F ) at endpoint of phase III, compared with those at baseline ( Figure S2 ). These brain regions are responsible for cognitive control, emotion and learning memory, as well as sensory (Volkow et al., 2003; Dagher, 2012; Ochner et al., 2013; Gettens and Gorin, 2017). No significant changes were observed in brain activity of the reward circuit or other brain regions ( Figure S2 ). These data suggest that IER reduces the activity of brain regions related to the regulation of food intake, thus contributing to successful weight loss and maintenance.

To explore the effect of IER on gut microbiota in obese patients, we performed metagenomic sequencing and bioinformatics analyses. Shannon diversity analysis at the species level showed that the gut microbial species at endpoint of phase II was significantly increased compared with that at baseline or midpoint of phase II. The gut microbial species at the end of phase III was significantly decreased compared with that at the end of phase II. No significant difference was observed in the gut microbial diversities between baseline and the end of phase III ( Figure 3A ). Bray-Curtis dissimilarity analysis showed that the gut microbiota of obese individuals at phase II and phase III clustered together ( Figures 3B , S3B ). These data suggest that IER intervention increases gut microbial richness and diversity in obese individuals.

Next, we sought to identify differentially abundant gut microbial species and pathways during IER. The results of LDA showed that food-borne pathogenic Escherichia coli was the most abundant species in the gut microbiota of obese individuals at baseline ( Figures 4A–C ). IER dramatically reduced E. coli abundance across multiple timepoints ( Figure 4E ). On the other hand, compared with those at baseline, obesity-related Faecalibacterium prausnitzii, Parabacteroides distasonis, and Bacterokles uniformis were remarkably increased in gut microbiota of obese individuals during phase II (all P < 0.05; Figure 4D ), peaking at midpoint of phase II and then gradually decreasing to near baseline levels. No significant differences were observed in the abundance of F. prausnitzii and P. distasonis between endpoints of phase II and phases III. Similar trends were observed in the abundance of Clostridium leptum, Flavonifractor plautii, Bacterokles intestinalls, Odoribacter splanchnicus, and Clostridium citroniae ( Figure 4D ).

Considering the important role of the BGM axis in weight maintenance, we investigated the dynamic correlation between differentially abundant gut bacteria and the brain regions responding to IER. The results showed that at baseline, the abundance of E. coli, C. comes, and E. hallii were negatively correlated with the activity of left orbital inferior frontal gyrus whereas the abundance of P. distasonis and F. plautii were positively correlated with the activity of right orbital inferior frontal gyrus and right putamen, respectively ( Figure 5A ). At midpoint of phase II, the abundance of B. intestinalis was negatively correlated with the activity of anterior cingulate gyrus whereas the abundance of C. leptum and O. splanchnicus were positively correlated with the activity of putamen and left orbital inferior frontal gyrus, respectively ( Figure 5B ). At endpoint of phase II, the abundance of B. caccae and B. thetaiotaomicron were negatively correlated with the activity of left orbital inferior frontal gyrus, the B. thetaiotaomicron abundance was negatively correlated with the activity of anterior cingulate gyrus, and the C. comes abundance was positively correlated with the activity of putamen ( Figure 5C ). At endpoint of phase III, the abundance of E. coli, P. distasonis, and F. plautiiwere were negatively correlated with the activity of right putamen, left dorsolateral prefrontal lobe, and right orbital inferior frontal gyrus, respectively, whereas the abundance of S. salivarius and F. plautii were positively correlated with the activity of right orbital inferior frontal gyrus and putamen, respectively ( Figure 5D ). The differential microbes were Coprococcus comes at baseline, Bacteroides uniformis and Parabacteroides distasonis at HC-W2, and Eubacterium Hallii and Agathobaculum butyriciproducens at HC-W4 ( Figure S3 ). These data suggest that IER induces a dynamic interaction between the brain and gut microbiota.

The last measurements were taken at 30 days after the IER. Further follow-up study is needed for revealing the BGM axis dynamical alteration which affect the long-term weight loss. In this study, it mainly aims at the specific BGM axis changes during the weight loss under the IER. The results provided primary insights for potential BGM axis changed during weight loss, and did not establish causation. The animal model is need to validate the BGM changes for more understanding about how these certain brain area activity work with specific gut flora in the weight loss.