Reduced stress-associated FKBP5 DNA methylation together with gut microbiota dysbiosis is linked with the progression of obese PCOS patients

Polycystic ovary syndrome (PCOS) is a common endocrine disease in females that is characterized by hyperandrogenemia, chronic anovulation, and polycystic ovaries. However, the exact etiology and pathogenesis of PCOS are still unknown. The aim of this study was to clarify the bacterial, stress status, and metabolic differences in the gut microbiomes of healthy individuals and patients with high body mass index (BMI) PCOS (PCOS-HB) and normal BMI PCOS (PCOS-LB), respectively. Here, we compared the gut microbiota characteristics of PCOS-HB, PCOS-LB, and healthy controls by 16S rRNA gene sequencing, FK506-binding protein 5 (FKBP5) DNA methylation and plasma metabolite determination. Clinical parameter comparisons indicated that PCOS patients had higher concentrations of total testosterone, androstenedione, dehydroepiandrosterone sulfate, luteinizing hormone, and HOMA-IR while lower FKBP5 DNA methylation. Significant differences in bacterial diversity and community were observed between the PCOS and healthy groups but not between the PCOS-HB and PCOS-LB groups. Bacterial species number was negatively correlated with insulin concentrations (both under fasting status and 120 min after glucose load) and HOMA-IR but positively related to FKBP5 DNA methylation. Compared to the healthy group, both PCOS groups had significant changes in bacterial genera, including Prevotella_9, Dorea, Maihella, and Slackia, and plasma metabolites, including estrone sulfate, lysophosphatidyl choline 18:2, and phosphatidylcholine (22:6e/19:1). The correlation network revealed the complicated interaction of the clinical index, bacterial genus, stress indices, and metabolites. Our work links the stress responses and gut microbiota characteristics of PCOS disease, which might afford perspectives to understand the progression of PCOS.


INTRODUCTION
Polycystic ovary syndrome (PCOS) is characterized by hyperandrogenemia, chronic anovulation, and polycystic ovaries and has a prevalence ranging from 6 to 20% among females of reproductive age; in addition, PCOS is the most common endocrine disorder and a cause of infertility in reproductive-aged women 1,2 . Many reports indicate that PCOS is linked to a higher risk of metabolic disorders, such as insulin resistance (IR), type 2 diabetes mellitus (T2DM), dyslipidemia, and cardiovascular diseases 3,4 . To date, the etiology of PCOS is unknown but multifactorial elements, including inherent genetics, intrauterine environment, lifestyle, and potential alteration in the gut microbiota, are thought to be involved in its development 5 .
Accumulating evidence indicates that gut inhabitants play a significant role in the development of obesity, obesity-associated inflammation, and insulin resistance [6][7][8] . For instance, Zhao et al. 9 revealed that short-chain fatty acid (SCFA) producers may be beneficial in the improvement of hemoglobin A1c levels in T2DM partly by increasing glucagon-like peptide-1(GLP-1) production. The glutamate-fermenting commensal, Bacteroides thetaiotaomicron, reduced plasma glutamate concentration and alleviated dietinduced body weight gain and adiposity in mice, and its abundance was related to bariatric surgery efficiency 10 . A recent randomized controlled trial showed that fecal microbiota transplantation (FMT) halted the decline in endogenous insulin production in recently diagnosed patients with T1DM in 12 months 11 . PCOS has obvious heterogeneity in which a large proportion of patients have IR and metabolic abnormalities; however, there are still some patients with normal body mass index (BMI) and metabolic normalities. In addition to race, lifestyle, and diet habits, different subtypes caused by different etiologies may be an important reason for the difference in gut microbiota. Limited research has shown that PCOS patients have decreased alpha diversity [12][13][14] and featured bacteria from Bacteroidaceae, Clostridiaceae, Erysipelotrichidae, Lachnospiraceae, Lactobacillaceae, Porphyromonadaceae, Prevotellaceae, Ruminococcaceae 15 , and Actinobacteria 5,12,16,17 . Qi et al. 14 revealed that Bacteroides vulgatus was markedly elevated in the gut microbiota of PCOS individuals, accompanied by reduced glycodeoxycholic acid and tauroursodeoxycholic acid levels. Hyperandrogenism is associated with gut microbial dysbiosis, indicating that androgens may modulate the gut microbial community and that modulation of the gut microbiota may be a potential treatment target for PCOS 18 . For example, FMT from PCOS women or exposure to certain bacteria resulted in a PCOS-like phenotype in mice, while exposure to a healthy gut microbiome resulted in protection from developing PCOS-like traits in mice 19 . The PCOS-associated lipidomic analysis revealed that PCOS patients had distinguished lipid characteristics that made potential lipid biomarkers for PCOS diagnosis possible 20,21 . For example, Jiang et al. 22 revealed that phosphatidylcholine (PC) was higher, whereas lysophosphatidylcholine was lower in PCOS women than in healthy controls. In addition, metabolomics analysis between the healthy and PCOS groups also revealed metabolic disorders associated with lipid and amino acid metabolism 23 .
Depression and anxiety are more common in patients with PCOS 24,25 , and these unhealthy psychological conditions may play a role in the development of PCOS. The gut microbiota composition has changed as a response to stressful conditions 26,27 . The brain-gut-microbiota axis plays a role in the pathogenesis of stress-related psychiatric disorders 28,29 . FK506binding protein 5 (FKBP5) is an important modulator of the stress response, and Zannas et al. suggested that methylation measurement of FKBP5 CpGs may be associated with stress-related disease 8 . Interestingly, with respect to hormone-binding function, FKBP5 serves to increase androgen receptor (AR) function, suggesting that FKBP5 can directly or indirectly target the ligand-binding domains of AR 30 . However, the methylation level of FKBP5 CpGs in PCOS patients and its relationship with gut microbiota have not been reported.
To elucidate the relationship among gut microbiota, metabolites, and PCOS clinical features, we characterized gut microbiota from a large number of PCOS patients with high BMI (BMI ≥ 24) and normal BMI (BMI < 24) using 16S rRNA gene sequencing and untargeted metabolomics, and their association with FKBP5 CpG methylation was also investigated. Our findings can help to illuminate the microbiome-based process underpinned in PCOS disease and in the development of advanced approaches to develop biomarkers for diagnose of PCOS.
Comparison of alpha-and beta-diversities in the gut microbiome among the three participant groups An average of 69,379 raw reads, 68,424 clean reads, 59,441 OTU sequences/samples, and 227 OTUs were obtained from 136 samples (Supplementary Table 2). Permutational multivariate analysis of variance (PERMANOVA) (Df = 1, F = 1.73, R 2 = 0.01, p < 0.05) showed significant differences in the overall bacterial community between the PCOS and healthy groups (Fig. 1c). However, there were no significant differences in beta diversity among the healthy, PCOS-HB and PCOS-LB groups (PERMANOVA, Df = 2, F = 1.21, R 2 = 0.02, p > 0.05) (Fig. 1d), and there was no significant difference (Df = 1, F = 0.63, p > 0.05) in Bray-Curtis distance between the PCOS-HB and PCOS-LB groups relative to the healthy group (Fig. 1e), indicating that the bacterial community was homogenized within PCOS patients. However, the bacterial community difference between healthy individuals and PCOS-LB is significantly smaller (Df = 1, F = 78.48, p < 0.05) than that of the difference between PCOS-HB and PCOS-LB (Fig. 1f).
Shared "universal" OTUs (found in samples from all PCOS and healthy individuals) accounted for 78.9% of the total OTUs. Individuals in the PCOS-HB group had more exclusive OTUs (3.45%) than patients who were in the PCOS-LB group (1.87%) and individuals in the healthy group of (2.44%) (Fig. 2a). Compared with healthy participants, patients with PCOS-LB displayed a lower Chao1 index but higher J indices. No significant difference (p > 0.05) was observed between PCOS-HB and PCOS-LB groups for either Chao1 or J indices, while reduced Shannon indices were observed in the PCOS-HB group relative to the PCOS-LB group (Supplementary Table 3 and Fig. 2b). Further correlation between the bacterial diversity indices and clinical parameters showed that observed species, Chao1 and ACE values were significantly negatively related to I0 and I120 but were significantly positively related to FKBP5-Met1 and FKBP5-Met2 (p < 0.05) (Supplementary Table 4 Table 5 and Supplementary Figure 1a). Firmicutes and Actinobacteria were abundant in the healthy group, while Bacteroidetes and Proteobacteria were lower in the PCOS group. PCOS-HB group was featured as higher abundance of Proteobacteria and Fusobacteria (Fig. 2d). The abundant bacterial genus (relative abundance > 1%) added up to 76.89%; therein, the average abundance of Bacteroides (25.49%), Faecalibacterium (9.34%), Prevotella_9 (8.63%), Roseburia (6.23%), un_f_Lachnospiraceae (5.41%), Blautia (2.86%), and Megamonas (2.06%) reached >2% (Supplementary Table 6). Healthy individuals were featured as higher Faecalibacterium and Prevotella_9 while lower Bacteroides, and the PCOS-HB group had a higher abundance of Bacteroides and Megamonas than the healthy group (Fig. 2e).
To screen out the differentiated microbiota taxa, the Wilcox test between groups was conducted. The comparison between the PCOS and healthy groups identified a total of 29 distinct bacterial genera. Compared with the healthy group, the highly increased taxon in the PCOS group was Escherichia. Shigella   The data are shown as the mean ± SD. n = 38 in the control group, n = 48 in the PCOS-LB, and n = 50 in the PCOS-HB group for all outcomes. Letters indicate the ANOVA grouping among groups. BMI body mass index, LDH lactate dehydrogenase, AST aspartate aminotransferase, ALT alanine aminotransferase, GGT glutamyltransferase, ALP alkaline phosphatase, CHE cholinesterase, MAO monoamine oxidase, AFU α-L-fucosidase, TP total protein, ALB albumin, GLB globulin, TBIL total bilirubin, DBIL direct bilirubin, IBIL indirect bilirubin, BUN urea nitrogen, Cr creatinine, UA uric acid, TG triglycerides, TC total cholesterol, HDL-C high-density lipoprotein cholesterol, LDL-C low-density lipoprotein cholesterol, FT3 free triiodothyronine, FT4 free thyroxine, TSH thyroid-stimulating hormone, LH luteinizing hormone, FSH folliclestimulating hormone, PRL prolactin, E2 estrogen, PROG progesterone, AD androstenedione, TT total testosterone, DHEA dehydroepiandrosterone, DHEA-S dehydroepiandrosterone sulfate, FAI free androgen index, SHBG sex hormone-binding globulin, G0 glucose level at fasting status, G120 glucose level at 120 min after glucose load, I0 insulin level at fasting status, I120 insulin level at 120 min after glucose load, HOMA-IR homeostasis model assessment for IR, IL-22  Table 7). There were 26 differential genera in the comparison between healthy and PCOS-HB groups and Prevotella_9, Faecalibacterium, Lachnoclostridium, Subdoligranulum, and Escherichia. Shigella ranked as the fifth most-abundant genera (Fig. 2f). Prevotella_9, Megamonas, Alistipes, Romboutsia, Dorea, and Alloprevotella were the most abundant taxa, which differentiated the healthy and PCOS-LB groups ( Supplementary Fig. 1b), and among the selected distinguishing taxa, Lactococcus, Romboutsia, and Slackia were increased in the PCOS-LB group, while a low amount of Mailhella only existed in the group of healthy individuals. Further comparisons between PCOS-HB and PCOS-LB were made, and 18 distinguished genus taxa were screened out ( Supplementary Fig. 1c). For example, the relative abundances of Erysipelotrichaceae_UCG.003, Holdemania, Sellimonas, Terrisporobacter, Turicibacter, and un_f_Flavobacteriaceae in PCOS-LB were twice higher than those in PCOS-HB (Supplementary Table 8).
Functional profiling of the gut microbiota by PICRUSt analysis To predict bacterial functions coded by the gut microbiome, PICRUSt analysis was performed to compare the difference between the PCOS and healthy groups. The mean nearest sequenced taxon index (NSTI) value was 0.078 ± 0.025 for all samples (Supplementary Table 9). The Wilcox test was performed to compare the significantly different functions in KEGG level 3, and those featuring |logFC | >1 (the abundance in one group was twice more than that in the other group) were determined. Compared with PCOS patients, the p53 signaling and cardiac muscle contraction pathways were more highly expressed in the healthy group ( Supplementary Fig. 3a), the two of which were also distinguished pathways between healthy and PCOS-LB patients ( Supplementary Fig. 3b). However, cardiac muscle contraction was the only function that occurred between the healthy and PCOS-HB groups. The relative abundance of alpha-linolenic acid metabolism in PCOS-HB patients was twice more than that in PCOS-LB patients ( Supplementary Fig. 3c).

Metabolite profile related with PCOS disease
To assess whether the profiles of plasma metabolites were associated with PCOS, 20 healthy individuals, 20 PCOS-HB and 20 PCOS-LB patients were included for the untargeted metabolome analysis. Significant differences in the composition of plasma metabolites were observed between the healthy group and the PCOS group (Fig. 3a, b, Fig. 4, and Supplementary Fig. 4a Table 11). It is intriguing that estrone sulfate existed in the difference comparison between the healthy and PCOS groups, indicating its potential link with the progression of PCOS. We observed great similarity between the PCOS-HB and PCOS-LB groups, and the metabolite comparison results between the PCOS-HB and PCOS-LB groups showed that bilirubin, 4-chlorophenol, hydrocinnamic acid, caffeine, quinoline, prostaglandin E2, and PC(2:0/16:1) were the main distinguished metabolite types ( Supplementary Fig. 5).

Association between microbial taxa, metabolites and clinical parameters
The significantly distinguished clinical properties and gut bacterial taxa were screened out between patients with PCOS and healthy individuals. The relationships between different clinical properties and FKBP5-Met were determined, and the results showed that methylation was positively associated with urea nitrogen (BUN), prolactin (PRL) and sex hormone-binding globulin (SHBG) but negatively related to BMI, cholinesterase (CHE), LH, free androgen index (FAI), and interleukin-22 (IL-22) and I120. In addition, FKBP5-Met displayed a significantly positive correlation with Faecalibacterium, Lachnospiraceae, Prevotella_9, and Romboutsia but was negatively correlated with Granulicatella (Fig. 5a). Correlation of

DISCUSSION
PCOS is a common endocrine disease in females of reproductive age. To date, the exact etiology and pathogenesis of PCOS are unknown. Genetic factors, unhealthy lifestyle, and neuroendocrine, immune and metabolic dysfunctions 31 are believed to be involved in the pathogenesis of PCOS. Recently, increasing evidence has indicated the gut microbiota characteristics of PCOS patients and linked the gut dysbiosis with the progression of the disease, which helps researchers understand the pathogenesis of PCOS from a new perspective 5,12,13,32,33 . Nevertheless, there have been no reports associating FKBP5 DNA methylation with PCOS and gut microbiota differentiation among PCOS patients with different BMI conditions. α diversity indices reflect thespecies type and abundance and productivity of the local ecosystem of the intestine. PCOS patients have been shown to have decreased bacterial α diversity compared with that of healthy controls 5,12,13,32-34 , with which our results were consistent. Our results further showed that the Chao1 index is negatively correlated with HOMA-IR, I0 and I120 while positively correlated with level of FKBP5 DNA methylation . Most studies have demonstrated a reduction in gut microbiota diversity and richness in obese subjects 35 . Exposure to social stress also leads to a decrease in diversity in the gut microbiota of mice and Syrian hamsters 36,37 . Our results showed that PCOS patients, especially obese individuals, often have IR and a reduced level of stress-associated FKBP5 DNA methylation. Therefore, these two pathophysiological status may partly explain the decline in gut microbiota diversity.
Some researchers reported that β diversity was significantly different between patients with PCOS and healthy controls 32,38,39 , which was also the case in this study, suggesting that PCOS patients have a specific gut microbiota composition. PCOS patients have been shown to have a higher abundance of Catenibacterium, Kandleria, Ruminococcaceae, Bacteroidaceae, Parabacteroides, Clostridium, Prevotella, and Alistipes 40 while a lower abundance of Prevotellaceae 39 . The differences in predominant bacteria in the gut microbiota of PCOS patients have been shown to be significantly affected by race, lifestyle disease severity and sample size. Qi et al. 14 reported that Bacteroides vulgatus content was markedly elevated in the gut microbiota of individuals with PCOS, with reduced IL-22 secretion, which was different from our comparison that was featured as Prevotella_9. Several reasons may account for the different results. First, the patients and the control subjects in Qi's study were located in northern China, but the subjects in our study were located in southeastern China. The differences in lifestyle expecailly diet and anthropometrics between northern and southern China are very substantial. It has been reported that diet 7,41,42 , drug useand anthropometrics together explain 20% of gut microbiota variability 43 44 . Second, the characteristics, including BMI and HOMA-IR, of the recruited PCOS patients and the control individuals were somewhat different from those in Qi's and our study. For individuals without PCOS, a higher abundance of Firmicutes in those with obesity was observed, and the Firmicutes/Bacteroidetes ratio seemed to be higher in women with a high BMI [45][46][47] . This ratio was reduced with a decrease in Firmicutes and an increase in Bacteroidetes abundance after bariatric surgery and weight loss 48 . However, some studies have pointed out that the decrease of Bacteroidetes abundance is accompanied by an increase in Actinobacteria rather than Firmicutes abundance 49 , or a synchronous increase in Firmicutes and Actinobacteria abundance. A meta-analysis found that no significant difference in the Bacteroidetes-to-Firmicutes ratio between obese and lean rodents 50 . Furthermore, we paid attention to gut microbiota compositions in PCOS patients with different BMI levels. Liu et al. 17 found that Bacteroides, Escherichia/Shigella, and Streptococcus were positively correlated with BMI while Akkermansia and Ruminococcaceae were negatively correlated with BMI. Adolescents who had PCOS and obesity were found to have a higher abundance of Actinobacteria, lower abundance of Bacteroidetes, and similar abundances of Firmicutes and Proteobacteria compared with those of controls who had obesity 5 . Zeng et al. 39 found that Prevotellaceae abundance was dramatically lower in PCOS patients, especially in the IR-PCOS group. Lachnoclostridium, Fusobacterium, Coprococcus_2, and Tyzzerela 4 were found to be the characteristic genera of scores were used to rank the discriminating power of different taxa between the PCOS and control groups. A taxon with a VIP score of >1 was considered important in the discrimination. Red, green, and blue color represent Healthy, PCOS-HB and PCOS-LB group separately. Fig. 4 The distribution of distinguished metabolites in PCOS patients. The metabolites were screened out based on metabolites with VIP > 1, p < 0.05, and FC ≥ 2 or FC ≤ 0.5. Red and green color represent Healthy and PCOS group separately. *p < 0.05 denotes significant difference. The data are shown as the mean ± SD and error bar was used. patients with PCOS and obesity 34 . Our results showed that there were obvious differences in α and β diversities between normal subjects and PCOS patients, but the difference was not obvious between PCOS patients with elevated BMI and normal BMI. These results suggest that the effect of PCOS on the gut microbiota is greater than that of BMI.
PCOS women tend to experience mildly elevated anxiety and depression significantly more often than women without PCOS 51-53 and to have a higher level of perceived stress 25 . Whether these unhealthy psychological states are the results of PCOS or one of the reasons that cause or deteriorate PCOS is still unknown. Patients with PCOS and obesity have been found to have higher depression scores than women with PCOS who do not have obesity, and depression scores were found to be significantly correlated with IR and lipid parameters and with the number of metabolic syndrome components 24 . Livadas et al. 54 showed that the degree of anxiety, state, and trait (STAI-S, STAI-T) appeared to vary in a pattern similar to that of hyperandrogenemia and IR independent of age and BMI in 130 PCOS patients. FKBP5 is an important modulator of stress responses. Aging/stress-driven FKBP5-NF-κB signaling mediates inflammation, potentially contributing to cardiovascular risk, and may thus point to novel biomarkers and treatment possibilities [55][56][57][58][59] . Furthermore, FKBP5 is also a positive regulator of the androgen receptor 30 , and this mechanism may be related to the incidence and severity of PCOS. Our study showed that patients in the PCOS-HB group had the lowest levels of FKBP5-Met compared with the PCOS-LB and control groups. The level of FKBP5 DNA methylation was proved to be associated with age and stress. In our study, the patients with PCOS had a narrow range of ages and there was no difference in age among groups. Therefore, the level of FKBP5 DNA methylation was more obvious in stress-associated individuals than in age-associated individuals. In our study, the results of FKBP5 DNA methylation analysis suggested that patients with a high BMI had more stress than the patients with a normal BMI. Stress may be related at least in part to certain clinical features of PCOS, including obesity and hirsutism 60 or an awareness of PCOS 61 . Furthermore, a low level of FKBP5-Met would lead to higher expression of FKBP5 protein and increased androgenic activity, which might be involved in the occurrence and development of PCOS in patients who have a high BMI.
The gut microbiota composition changes as a response to stressful situations and interventions that can modulate microbiota stress in human and animal models 26,27 . Stressor exposure was shown to decrease the relative abundance of bacteria in the genus Bacteroides while increasing the relative abundance of bacteria in the genus Clostridium in mice 62 . Mice colonized with gut microbiota from stressed mice with a lower relative abundance of Lactobacillus and a higher relative abundance of Akkermansia showed similar behaviors 63 . Probiotics as interventions can improve anxiety symptoms 64 . Overall, stress may be involved in PCOS in multiple ways. Among them, an important factor to target may be the composition of the gut microbiota, while enhancing androgen receptor activity through FKBP5 may be another option.
In the nontargeted metabolomics in our study, we found that metabolic products, especially lipid profiles, were different among the PCOS-LB, PCOS-HB, and control groups. Li et al. 65 reported that polyunsaturated fatty acids levels were reduced and that longchain saturated fatty acids levels were increased in patients with PCOS and obesity compared with that in lean controls. We screened the differentially abundant metabolites that could be linked with gut microbiota to explore the possible pathogenesis. On the other hand, our findings can be used as a potential basis for disease diagnosis. For example, Daan et al. 66 showed that retinol-binding protein 4,(RBP-4), dipeptidyl peptidase IV(DPP-IV), and adiponectin, as potential discriminative markers for PCOS with obvious hyperandrogenemia, had a specifically strong correlation in cases with a higher BMI. Our results showed that estrone sulfate, which is the most abundant estrogen precursor found in the bloodstream of women and men 67,68 , was notable in the difference comparison between the healthy and PCOS groups, indicating its potential role in the progression of PCOS.
Our study had limitations. First, although it included healthy individuals with normal BMI, additional healthy participants with a high BMI are also be needed to reveal the background difference between healthy subjects with a normal BMI and those with a high BMI. Second, our trial needs to be repeated in other geographical locations since microbiome composition is affected by ethnicity and diet. Third, it is difficult to conclude whether the change in FKBP5 gene methylation and gut microbiota composition is the cause or the result of PCOS, as most data were not functionally validated, the characteristic biomarkers should be verified in the future to reveal their clinical potential in disease diagnose. Lastly, we mainly used amplicon-based metagenomics and nontargeted metabolomics tools, and shotgun metagenome sequencing and lipidomics strategies could be used further to supply better characteristic resolution.
In conclusion, based on multi-omics data from patients with PCOS and healthy controls, our results suggest that PCOS patients with different BMI levels have differentially altered stress responses, gut microbiota compositions, and metabolites. The association between the top contributing genus, Prevotella_9, and the metabolite, estrone sulfate, in the comparison between patients with PCOS and healthy controls revealed their potential involvement in PCOS which needs further causal verification. In addition, the close connection among the stress-associated index, FKBP5-Met, and the PCOS-related phenotype may provide a therapeutic target.

MATERIALS AND METHODS Human participants
The study and all experimental procedures were approved by the Ethics Committee of the First Affiliated Hospital Shantou University Medical College according to the Council for International Organizations of Medical Science (ChiCTR2000041108). All participants were recruited from the Department of Endocrinology at the First Affiliated Hospital Shantou University Medical College between June 2019 and September 2020. Written informed consent was obtained from all participants.
The healthy volunteers who had regular menstrual cycles and normal ovarian morphology were from the general community. Their hormone levels, BMI, glucose tolerance status, blood pressure and serum lipids were all in the normal ranges. Women who were breastfeeding or pregnant within the past year were excluded from the study. Women with PCOS were diagnosed according to the 2003 Rotterdam criteria, which require the presence of at least two of the following: (1) oligo-ovulation and/or anovulation; (2) clinical and/or biochemical signs of hyperandrogenism; and (3) ultrasound findings of polycystic ovaries in 1 or 2 ovaries, ≥12 follicles measuring 2-9 mm in diameter, and/or ovarian volume ≥10 mL. Diagnoses of PCOS were made after the exclusion of other etiologies for hyperandrogenemia or ovulatory dysfunction 8 . All PCOS patients were first-visit patients and had not received PCOS-related treatment.

Clinical parameters determination
Ninety-eight PCOS patients with a normal BMI (PCOS-LB, BMI < 24), 50 PCOS patients with high BMI (PCOS-HB, BMI ≥ 24), and 38 healthy individuals with a normal BMI were recruited from the First Affiliated Hospital Shantou University Medical College Hospital between September 2018 and July 2020. All the participants were asked to come to our department during days 2-4 of spontaneous cycles after an overnight fast. Height, body weight, and BMI were calculated. Blood pressure was measured after at least 15 min of rest. Peripheral blood samples were collected from all subjects for TT, AD, DHEA,DHEA-S, E2, SHBG, LH, FSH, PRL, progesterone (PROG), FT3, free thyroxine (FT4), thyrotropin (TSH), G0, I0, and biochemical indexes included liver function, kidney function, and blood lipid measurement. Blood was also collected for plasma metabolomics and FKBP5 DNA methylation assays. The oral glucose tolerance test (OGTT) and insulin-releasing test were performed on the same day. After the procurement of the first blood sample, within 5-10 min, all the subjects ingested a solution containing 75 g glucose diluted in 300 mL of water. Subsequently, one additional blood sample was obtained at 120 min after the ingestion of the solution to estimate the glucose level (G120: glucose level after 120 min) and insulin concentrations (I120: insulin level after 120 min). SHBG was measured using a luminescence immunoassay (Siemens, New York, USA). The levels of serum FSH, LH, PRL, E2, PROG, FT3, FT4, and TSH were tested by radioimmunoassay (Beckman, MN, USA). Serum insulin concentration was estimated using a direct chemiluminometric assay (Siemens, New York, USA). A liquid chromatography (LC) (ACQUITY UPLC I Class, Water, MA, USA) coupled to tandem mass spectrometry (MS) (Triple Quad™5500, AB SCIEX, MA, USA) system was used to quantitate TT, AD, DHEA, and DHEA-S with the multiple reaction monitoring (MRM) model in BGI (Shenzhen, China). IL-22 and IL-8 was measured by ELISA (R&D, Minneapolis, MN, USA). The levels of plasma glucose and biochemical indexes included liver function, kidney function, and blood lipid were measured using an autoanalyzer (Beckman Coulter AU5800, MN, USA). HOMA-IR was defined as I0 (mIU/L) × G0 (mmol/L)/22.5. The free androgen index (FAI) was calculated with the formula FAI = TT (ng/mL) × 100 × 3.467/SHBG (nmol/L).

16S rRNA gene sequencing and data analysis
Fecal samples from healthy individuals and PCOS patients were collected on the day of the medical examination and immediately frozen at −80°C. Fecal microbial DNA was extracted from approximately 200 mg of the fecal samples using HiPure Stool DNA Kits B (D3141-03B, Guangzhou Meiji Biotechnology Co., Ltd., China) according to the manufacturer's protocols. The purity and concentration of the isolated DNAs were assessed by Qubit3.0 Fluorometer (Invitrogen, Carlsbad, CA, USA). DNA samples were stored at −20°C before being used as templates for 16S rRNA gene sequencing library construction. The 16S rRNA gene was amplified from the DNA samples with barcoded forward primers (5′-CCTACGGRRBGCAS-CAGKVRVGAAT-3′) and reverse primers (5′-GGACTACNVGGGTWTCTAATCC-3′), which were designed by GENEWIZ (Suzhou, China) (Supplementary Table 1), and aimed at relatively conserved regions bordering the V3 and V4 hypervariable regions of bacteria and Archaea 16S rDNA 69,70 . DNA library concentrations were validated by Qubit3.0 Fluorometer and further multiplexed and loaded on an Illumina MiSeq instrument according to the manufacturer's instructions (Illumina, San Diego, CA, USA). The bioinformatics analysis procedure for raw reads was performed according to previously described methods 71,72 . The sequence data will be available at NIH Sequence Read Archive under SUB8691740.

Untargeted metabolomics measured by LC-MS/MS
Plasma samples (100 μL) and prechilled methanol (400 μL) were mixed by vortexing. LC-MS/MS analyses were performed using a Vanquish UHPLC system (Thermo Fisher) coupled with an Orbitrap Q Exactive series mass spectrometer (Thermo Fisher) 73 . Identification and quantification of metabolites were performed using the mzCloud database by the search engine Compound Discoverer 3.0 (Thermo Fisher Scientific). Partial least squares discriminant analysis (PLS-DA) and featured metabolites based on variable importance in projection (VIP) scores were performed at