All experiments were approved by the ethics committee of the Fujian Academy of Medical Sciences (#DL-2021-07). Twenty-one-day-old female Sprague-Dawley rats were purchased from Shanghai Laboratory Animals Center Co., Ltd. (Shanghai, China). All rats were housed individually and maintained under a 12 h light:dark cycle at 22 ± 2°C and 50%-60% humidity and were provided free access to food and water. Animal feed was obtained from Research Diets, Inc. (New Brunswick, NJ, USA). A standard rodent diet (D12450J, Research Diets; 3.85 kcal/g) is composed of 20% protein, 70% carbohydrate, and 10% fat. A high-fat diet (HFD, D12492, Research Diets; 5.24 kcal/g) is composed of 20% protein, 20% carbohydrate, and 60% fat. Dehydroandrosterone (HEA) was obtained from Cayman Chemical (Michigan, MI, USA).

Fifty-nine rats were randomly divided into three groups: (a) in the control group (CON, n = 15), rats were fed a normal rodent diet composed of 10% fat, and rats administered 0.2 mL tea oil were used as negative controls; (b) in the DHEA group (DHEA, n = 22), rats were fed a 10% fat diet and subcutaneously injected with DHEA (6 mg/100 g of body weight dissolved in 0.2 mL of tea oil for 21 consecutive days, administered daily (12, 13)); (c) in the DHEA+HFD group (DHF, n = 22), rats were fed a 60% HFD and treated according to the same schedule with DHEA subcutaneous injection. After 3 weeks of modeling, eight rats from each group were selected randomly for analyzing PCOS-related parameters. The remaining rats in the DHEA and DHF groups were randomly divided into the SG and sham surgery groups, whereas rats from CON underwent sham surgery. There were five groups for surgery: (1) CON+Sham group (CONSh, n = 7), (2) DHEA+Sham group (DHEASh, n = 7), (3) DHEA+SG group (DHEASg, n = 7), (4) DHF+Sham group (DHFSh, n = 7), (5) DHF+SG group (DHFSg, n = 7). DHEA was administered continuously via subcutaneous injection until the rats were euthanatized.

Surgery was performed on the 21^st day after DHEA injection. All rats were fasted for 12 h before surgery. Water was available ad libitum until 4 h before the intervention. The rats were anesthetized by administering 10% chloral hydrate (0.3 mL/100 g injected intraperitoneally).

SG was performed according to a previously published method (13) (Online Resource 1). To prevent potential interference with the gut flora, we avoided using antibiotics in this study.

In the first 24 h post-surgery, the rats were fasted without water and food to promote the healing of the newly stitched tresis vulnus. They were administered a liquid diet 24 h later, with gradual transition to a normal diet within 4-5 days post-surgery.

The plasma levels of insulin, C-peptide, and the inflammatory factors tumor necrosis factor alpha (TNFα) and interleukin 6 (IL-6) were measured using the MILLIPLEX MAP RAT METABOLIC MAGNETIC BEAD PANEL KIT according to the manufacturer’s instructions (Merck Millipore, Billerica, MA, USA). The serum levels of testosterone, progesterone, and estradiol (E2) were measured using enzyme-linked immunosorbent assay kits (R&D Systems, Inc, Minneapolis, MN, USA). The serum levels of luteinizing hormone (LH) and follicle-stimulating hormone (FSH) were measured using a LUMINEX 200 (Luminex, Austin, TX, USA).

The vaginal smears of the rats were monitored daily to determine the stage of cyclicity from 6 weeks of age to the day before surgery and till 1 week post-surgery. The regular estrus cycle comprises four stages: proestrus, estrus, metestrus, and diestrus. After the rats were euthanized, the ovaries were harvested and fixed immediately with 10% formalin for 24 h. When the rats were anesthetized, the ovaries were rapidly removed from the animals. Adipose tissue was separated and bilateral ovaries were weighed. The left ovarian vesicle was fixed in 4% paraformaldehyde for later ovarian histomorphological examination. The right ovary was placed in a cryopreservation tube into liquid nitrogen for cryopreservation, and then stored at -80°C for later use. Hematoxylin and eosin (HE) staining was performed. The number of cystic follicles and corpora lutea (CL) was counted by two pathologists blinded to grouping.

Each rat was weighed daily, starting at 28 days of age, until they were sacrificed. Glucose tolerance tests (GTTs) and insulin tolerance tests (ITTs) were conducted on the 18^th and 22^nd days post-surgery, respectively. The rats were fasted for 12 h. Blood from orbital veins was collected to measure the blood glucose level. For GTT, after the intraperitoneal injection of dextrose (2 g/kg body weight), the blood glucose levels were measured at 0, 15, 30, 45, and 60 min. For ITT, after the intraperitoneal injection of insulin (1 IU/kg body weight), the blood glucose levels were measured at 0, 30, 60, 90, and 120 min. GraphPad Prism 8.4 (GraphPad Software Inc., CA, USA) was used to calculate the total area under the glucose response curve (AUC). The serum levels of adiponectin and leptin were determined using enzyme linked immunosorbent assay (ELISA) kits (R&D Systems, Inc, Minneapolis, MI, USA).

Fresh fecal samples were collected in 1.5-mL sterile Eppendorf tubes, snap-frozen in liquid nitrogen and stored at -80 °C for subsequent analyses. The sequences of the 16S rDNA hypervariable regions V3-V4 were amplified using the barcoded primers 341F 5’-CCTACGGGRSGCAGCAG-3’ and 806R 5’-GGACTACVVGGGTATCTAATC-3’ using the Illumina MiSeq platform (Online Resource 1). The composition of the gut microbial community was analyzed using Unweighted UniFrac of ANOSIM and principal coordinates analysis (PCoa). Next, to determine the differences in the community composition among the groups, we performed principal component analysis (PCA) at the genus level. To investigate the differences in the microbial composition between the SG and sham groups, we used the linear discriminant analysis effect size (LEfSe) method by coupling standard tests for statistical significance with additional analyses that evaluated the biological consistency and effect relevance.

Methanol was purchased from Sigma-Aldrich (St. Louis, MO, USA). Methyl chloroformate, acetonitrile, and isopropyl alcohol were obtained from Thermo Fisher Scientific (Fairlawn, NJ, USA). All standard chemicals required for the measurement of other microbial metabolites were obtained from Sigma-Aldrich, Steraloids Inc. (Newport, RI, USA), and TRC Chemical (Toronto, ON, Canada). Ultrapure water was obtained using a Mill-Q Reference system equipped with an LC-MS Pak filter (Millipore, Billerica, MA, USA) (Online Resource 1). In order to analyze changes in the composition of metabolites related to the gut microbiota composition, we conducted ultra-performance liquid chromatography-tandem mass spectrometry analysis to measure the metabolite content.

Statistical analysis was performed using GraphPad Prism 8.4, SPSS 24.0 (IBM Corp., NY, USA), and R (version 3.6.3). Experimental data are expressed in terms of mean ± standard deviation from at least three independent experiments. Data were analyzed using one-way analysis of variance to compare the mean values of variables from different groups. Bonferroni’s test or Tukey’s post-hoc test was performed for the pairwise comparison of mean values among the groups. Spearman’s correlation analysis was performed to identify the correlations among microbiotas. Pearson’s correlation analysis was performed to determine the correlation among clinical phenotypes, gut microbiota, and fecal metabolites. P < 0.05 was considered to represent statistical significance.

A value of 0.99 was obtained for the Good’s coverage index, which indicated that the data was sufficient for analyzing the intestinal microbial community using the study samples ( Supplementary Figure 1A ). The Chao1 index indicated the richness of the gut microbial community. There were no significant differences between the Chao1 indices of CON and PCOS rats pre-surgery or between those of DHFSg and DHFSh rats post-surgery ( Figure 1A ). However, the Chao1 index was lower in the DHEASg group than in the DHEASh group ( Figure 1A ). The Shannon and Simpson indices indicated the diversity and evenness of intestinal flora. There were no significant differences between the parameters obtained for the CON and PCOS groups ( Figures 1B, C ) or those obtained for the corresponding SG and sham surgery groups ( Figures 1B, C ). The results showed that SG did not affect the α diversity of the gut microbiota. Interestingly, SG decreased the abundance of the gut microbiota in the DHEASg group but not in the DHFSg group ( Figures 1A–C ).

The PCoA results indicated a significant difference among species present in fecal samples from the CON, DHEA, and DHF groups ( Figure 1D and Supplementary Figure 1B ). The intestinal microbiota in the CON and PCOS groups were distinct with respect to the organismal structure. Notably, the PCoA plots revealed a distinct separation between the SG and sham surgery groups, suggesting that SG altered the gut microbiota composition in rats with PCOS ( Figure 1D ). Besides, ANOSIM analysis yielded similar results ( Supplementary Figure 1B ).

At the phylum level ( Figures 1E, F ), Bacteroidetes, Firmicutes, and Proteobacteria were the major microbial taxa detected in the groups. The abundance of Firmicutes decreased in both DHEASg and DHFSg rats compared to that in the corresponding sham groups; this change could be attributed to SG. Furthermore, the abundance of Bacteroidetes increased in DHEASg and DHFSg rats. At the genus level ( Figures 1G, H ), we analyzed the top twenty bacteria. The abundances of Bacteroides and Blautia increased more in the SG groups than in the sham groups, whereas the relative abundances of Ruminococcus, Alloprevotella, Clostridium, and Alistipes were low.

Furthermore, the abundances of Bacteroides, Escherichia_Shigella, Collinsella, Actinomyces, Blautia, Gemella, Peptostreptococcus, Flavonifractor, Streptococcus, and Fusobacterium increased in the PCOS-SG group. In contrast, the abundances of Ruminococcus, Alistipes, Anaeroplasma, Lachnospiracea_incertae_sedis, and Elusimicrobium increased significantly in the DHEASh and DHFSh groups ( Figure 2A and Supplementary Figure 1C ). Compared to the rats from the DHEASh group, those from the DHEASg group had a greater abundance of Bacteroides, Blautia, and Parabacteroides at the genus level ( Figure 2A and Supplementary Figure 1D ). However, the abundances of Alistipes, Ruminococcus, Paraprevotella, and Alloprevotella were notably lower in the DHEASg group ( Figure 2A and Supplementary Figure 1D ). Additionally, the abundances of Bacteroides and Bifidobacterium increased and those of Lactococcus, Peptococcus, and Enterococcus decreased considerably in the DHFSg group than in the DHFSh group ( Figure 2B and Supplementary Figure 1D ). Therefore, the abundances of microbial species in the SG and sham groups differed considerably.

PCA is primarily applied to analyze the chief source of alterations among samples. An observable distance between the PCOS (DHEA and DHF) and CON groups suggested that PCOS disturbed the fecal metabolite composition in rats ( Supplementary Figure 2A ). Based on the PCA results, the SG group was found to be metabolically distant from the corresponding groups, indicating that SG modulated the metabolic disturbance in rats with PCOS ( Supplementary Figure 2A ). Next, we conducted orthogonal partial least squares discriminant analysis (OPLS-DA) to maximize differences and filter core metabolites reliable for categories among groups. The generated models rendered a reliable predictive ability (DHEASg-DHEASh:R2Y = (0.0,0.9533), Q2 = (0.0, −0.1709) ( Supplementary Figure 2B ), and DHFSg-DHFSh:R2Y = (0.0,0.9391), Q2 = (0.0, −0.058) ( Supplementary Figure 2B )). Based on the values of variable influence on projection > 1.0, both the DHEASg and DHFSg groups were found to exhibit a significant shift in the levels of butyric and valeric acids compared to those in the corresponding sham surgery groups ( Figure 2C ). However, the levels of acetic acid and propionic acid were reduced significantly only in the DHEASg group. These findings indicated that SG-induced alterations in gut microbiota may markedly increase the fecal levels of SCFAs, particularly butyric and valeric acids, in rats with PCOS ( Figure 2 and Supplementary Figure 2 ).

Analysis of the estrus cycle revealed that both PCOS models experienced a significantly longer diestrus and shorter proestrus and estrus than CON rats ( Figures 3A–C ). Intriguingly, SG intervention alleviated the estrus cycle disorder in rats with PCOS ( Figures 3D, E ). In the DHEA and DHF rats, which were acyclic prior to surgery, the regular estrus cycle was restored from 12 days after SG ( Figure 3F ). However, the DHEASh and DHFSh rats continued to exhibit an irregular cycle throughout the intervention period ( Figure 3F ). DHEA and DHF rats had polycystic ovaries, whereas CON and PCOS rats did not. Multiple ovarian follicles from DHEA and DHF rats showed the presence of a tangled granulosa cell compartment with abnormal granulosa layer thickness, a typical characteristic of atretic antral follicles ( Figures 4A–C ). Notably, the ovaries of DHEA and DHF rats lacked CL, whereas the ovaries of CON rats had multiple CL, indicative of ovulation ( Figures 4I–L ). HE staining revealed that the granulosa layers of ovarian tissue from both PCOS-Sham groups showed a reduced number of cystic degenerating follicles. In contrast, SG increased the formation of granulosa layers and CL, indicating that SG improved the pathological changes observed in PCOS ( Figures 4D–H, K, L ). After modeling, the left ovaries of rats in each group were weighed, as shown in Figures 4M–O . After DHEA injection, the ovarian weight was lower than that of the control group, regardless of whether the diet was high-fat or not. Compared with the DHFSh group, the ovarian weight of PCOS rats in DHFSg group was lower. While there were no differences in ovarian weights between the DHEASh and DHEAsg groups.

Hyperandrogenism is the most prominent feature of PCOS. The serum testosterone levels increased considerably in DHEA and DHF rats than in CON rats ( Figure 5A ), indicating hyperandrogenism in the former. In addition, DHEA injection, with or without HFD, increased the serum levels of LH but not of FSH and progesterone ( Figures 5B–D ). Notably, both testosterone and LH levels were reduced, and almost restored to the normal levels, in DHEASg and DHFSg rats ( Figures 5F, G ). In addition, the circulating levels of TNFα and IL-6 were significantly elevated in both DHEA and DHF rats but were normalized upon SG intervention ( Figures 5K–N ). The above results confirmed that SG treatment considerably improved the levels of sex steroid hormones and inflammatory factors in PCOS ( Figure 5 ).

We detected the expression level of adiponectin, the results showed adiponectin was remarkably decreased in PCOS models compared with control rats ( Figure 6A ). The serum level of adiponectin was restored in DHEASg group and in DHFSg group compared to their corresponding sham operation group ( Figure 6A ). Inversely, leptin was significantly elevated both in DHEA and DHF rats, but were normalized by SG ( Figure 6B ). Moreover, DHF-treated rats showed a remarkably higher levels of leptin than DHEA-treated rats ( Figure 6B ). The DHEA and DHF rats gained more weight than CON rats during the 3-week treatment period ( Figures 6C–E and Supplementary Figure 3A ). Post-surgery, both DHEASg and DHFSg rats lost significantly more weight than DHEASh and DHFSh rats, starting 1 week post-surgery, and maintained a lower body weight till the end of the experiment ( Figures 6C–E and Supplementary Figure 3B ). Glucose homeostasis is typically impaired in patients with obesity and metabolic syndrome, including patients with PCOS. We found that the basal glucose levels were elevated in both DHEA and DHF rats ( Figure 6F ). Moreover, both DHEA and DHF rats exhibited impaired insulin sensitivity ( Figure 6G ). Post-surgery, the AUC for the SG excision group was significantly lower than that for the sham surgery and CON groups. As expected, both insulin sensitivity and glucose tolerance were significantly improved in DHFSg and DHEASg rats compared to that in the sham controls ( Figures 6H, I ). These results indicated the significant benefits of SG intervention on insulin resistance ( Figure 6 ).

In order to assess the association among gut microbiota, inflammation, sex steroid hormones, and fecal SCFAs, we performed a correlation analysis. The abundances of Bacteroides, Blautia, and Bifidobacterium were negatively correlated with the fecal SCFA content, whereas those of Alistipes and Ruminococcus were positively correlated with the fecal SCFA content ( Figure 7A ). The abundance of Bacteroides exhibited a significant negative correlation with the butyric acid, propionic acid, and total SCFA content. Conversely, the levels of LH and testosterone were positively associated with the levels of SCFAs, including butyric acid, propionic acid, acetic acid, valeric acid, and total SCFAs ( Figure 7B ). The levels of butyric acid were positively associated with those of insulin and C-peptide. Moreover, the abundances of Bacteroides and Blautia were negatively correlated with the LH, testosterone, insulin, and c-peptide levels ( Figure 7B ). In contrast, the abundances of Alistipes and Ruminococcus were positively correlated with the LH and testosterone levels. The abundance of Bacteroides was negatively associated with the IL-6 levels, the relative abundance of Clostridium was positively correlated with the TNFα and IL-6 levels, and the abundance of Alistipes was only positively associated with the TNFα level ( Figure 7 ).