Androgen-induced gut dysbiosis disrupts glucolipid metabolism and endocrinal functions in polycystic ovary syndrome

To induce a PCOS-like model, SD rats were injected subcutaneously with DHEA (6 mg/100 g bodyweight) daily for 35 days in the DHEA group, while control rats were injected with phosphate buffer saline (PBS). During the 5-week treatment, the weight of the DHEA-treated rats was similar to their PBS-treated counterparts (Fig. 1a) and the levels of fasting blood glucose (FBG) were similar between groups (Fig. 1b). Compared with PBS-treated rats, the DHEA-treated rats displayed higher glucose intolerance (Fig. 1c, d) and disrupted oestrous cycles (Fig. 1e). Ovaries from the PBS group contained follicles at different stages of development and several corpora lutea (CL). In contrast, those from the DHEA group showed multiple cyst-like follicles with collapsed walls and fewer CL (Fig. 1f). In addition, the levels of total testosterone (TT), sex hormone-binding globulin (SHBG), free androgen index (FAI), and LH were elevated in the DHEA-treated rats (Fig. 1g–j). The increased TT and FAI verified the efficiency of DHEA administration and the successful development of hyperandrogenic rats. However, no differences were observed in follicle-stimulating hormone (FSH) levels or the LH/FSH ratio (Fig. 1k, l). These results validate the successful establishment of a PCOS-like rat model with disorders in metabolism, ovarian morphology, and endocrine function.

Faecal samples of PBS- and DHEA-treated rats were sequenced using the Illumina-MiSeq platform. Regarding alpha diversity, the phylogenetic diversity (PD) whole tree index significantly decreased in the gut microbiota of DHEA-treated rats compared with those of PBS-treated ones (Additional file 1: Figure S1a). Principal coordinate analysis (PCoA) based on the unweighted UniFrac distance did not show significant differences in clustering in faecal samples between the DHEA and PBS groups (Additional file 1: Figure S1b). At the genus level, the DHEA group was characterised by the decreased relative abundance of Turicibacter, Anaerofustis and Clostridium sensu stricto (Additional file 1: Figure S1c). From linear discriminant analysis effect size (LEfSe) analysis, the genus Turicibacter, family Erysipelotrichaceae, order Erysipelotrichales, class Erysipelotrichia was top-ranking by linear discriminant analysis (LDA) score in contributing to sample separation between groups (Additional file 1: Figure S1d, e).

Rats in both the DHEA group and DHEA + antibiotics group were injected subcutaneously with 6 mg/100 g bodyweight DHEA daily for 35 days. During this period, to establish pseudo germ-free rats, rats in the DHEA + antibiotics group received a freshly prepared antibiotic cocktail (ABX) containing 1 mg mL⁻¹ ampicillin sodium, 1 mg mL⁻¹ neomycin sulfate, 1 mg mL⁻¹ metronidazole, and 0.5 mg mL⁻¹ vancomycin hydrochloride in drinking water, while rats in the DHEA group were provided with non-acidified drinking water during the establishment of the PCOS model. 16S rRNA gene sequencing was performed to confirm the effects of ABX administration. There was a substantial decrease in alpha diversity in the DHEA + antibiotics group (Additional file 2: Figure S2a, b), indicating that ABX significantly suppressed the gut microbiota. The variation, confirmed by PCoA and analysis of similarities (ANOSIM) analyses, was higher between groups than within groups (Additional file 2: Figure S2c, d). These findings demonstrated that the ABX treatment provided a suitable environment for FMT. The body weight, FBG and glucose tolerance were not affected by the clearance of gut microbiota (Fig. 1a–d). Regarding the oestrous cycle, both groups were acyclic (Fig. 1e). Ovarian morphological anomalies, such as cyst-like follicles and thinner granulosa cell layers, were also observed in the DHEA + antibiotics group (Fig. 1f). These data suggest that the lack of a gut microbiome does not affect disorders in glucose metabolism and ovarian function in hyperandrogenic rats. Serum TT, FAI, LH, FSH, and the LH/FSH ratio decreased in the DHEA + antibiotics group compared with the DHEA group (Fig. 1g, i–l). However, serum SHBG remained unchanged between groups (Fig. 1h). The serum levels of reproductive hormones were significantly affected by gut microbiota clearance. In general, the presence of a gut microbial community may not be necessary for the development of the PCOS phenotype in DHEA-treated rats.

The timeline for the animal experiments with FMT recipients is shown in Fig. 2a. A three-course antibiotic cocktail was used for 27 days to reduce gut microbiota and to establish a pseudo germ-free rat model [21]. After ABX treatment, the gut microbial loads were significantly decreased (Additional file 3: Figure S3a), providing a favourable environment for the FMT. Subsequently, the rats were divided into 2 groups, an FMT recipient from the PBS group (t-PBS, n = 6), and an FMT recipient from the DHEA group (t-DHEA, n = 6). For the following 14 consecutive days, the pseudo germ-free rats in the t-PBS and the t-DHEA groups received a daily gavage of gut microbiota (0.1 g fecal pellet in 1 ml mixture of PBS and skimmed milk) from the PBS- and DHEA-treated rats, respectively. To ensure that recipients received similar volumes of microbes from the FMT donors, microbial loads in both groups of donors were measured and shown in combination in Additional file 3: Figure S3a. During the 27 days of washout and the 14 days of transplantation, the body weight was measured weekly and was found similar between groups (Fig. 2b). No differences in FBG were observed between groups (Fig. 2c). Compared with rats in the t-PBS group, the t-DHEA group displayed increased glucose levels and area under the curve in glucose tolerance tests (GTT) (Fig. 2d, e). These results show that FMT from PCOS-like rats could lead to glucose intolerance in pseudo germ-free rats. Rats receiving faecal microbiota from PBS-treated donors showed normal oestrous cycles, while their counterparts receiving microbiota from DHEA-treated donors exhibited disrupted oestrous cycles (Fig. 2f). The ovaries of t-PBS rats showed follicles at different stages of development and CL. Rats in the t-DHEA group exhibited several cyst-like follicles with attenuated granulosa cell layers and thickened follicle walls (Fig. 2g). Therefore, faecal microbiota from PCOS-like rats disrupts the oestrous cycle and leads to ovarian morphological anomalies in recipients. The TT, SHBG and FAI levels were higher in the t-DHEA group than in the t-PBS group (Fig. 2h–j). The LH levels decreased insignificantly, while the FSH levels were lower in the t-DHEA group than in t-PBS group (Fig. 2k, l). The LH/FSH ratio was unchanged (Fig. 2m). Interestingly, gut dysbiosis in the DHEA-induced PCOS-like rats led to hyperandrogenism in FMT-recipient rats. In general, FMT from the DHEA group induced impaired glucose tolerance, disrupted oestrous cycle, polycystic ovaries and unbalanced reproductive hormones in recipient pseudo germ-free rats, suggesting an important role of gut dysbiosis in the development of PCOS.

In the DHEA-treated donors, free circulating DHEA was transformed into dehydroepiandrosterone sulfate (DHEA-S) by sulfation in the liver and excreted into the gastrointestinal tract. Untargeted metabolomic analysis showed that DHEA was not detected in the faecal samples of either group; however, the DHEA-S levels were elevated in the DHEA-treated donors (Additional file 3: Figure S3b). Elevated faecal DHEA-S levels were also found in PCOS patients [22]. The DHEA-S levels in faecal samples of both groups were quantified (Additional file 3: Figure S3c). Although there was a significant increase in the DHEA-S level in the gut content of DHEA compared with PBS-treated donors, absolute quantification showed that the levels were insignificant in both groups. The DHEA-S levels were 665 ± 63.21 ng/gram of feces in DHEA-treated donors and 109.8 ± 10.11 ng in PBS-treated donors, which are far below the minimal effecting concentration for elicit a PCOS phenotype [23].

Faecal samples of the t-PBS and t-DHEA groups were sequenced to confirm the efficiency of FMT and further investigate the underlying mechanisms of FMT induction of the PCOS-like phenotype. Unsurprisingly, the microbial characteristics of DHEA-treated donors reappeared in recipients of FMT from the t-DHEA group. Recipient rats showed a similar alpha diversity index to donor rats (Fig. 3a). Beta diversity analysis revealed that samples from the t-DHEA group formed a distinct cluster from the t-PBS group (Fig. 3b). At the phylum level, Firmicutes, Bacteroidetes, Verrucomicrobia and Proteobacteria were most abundant in both the donor and recipient groups (Additional file 4: Figure S4a, d). Clostridia, Bacteroidia, Verrucomicrobiae and Bacilli dominated the gut microbiota at the class level in the PBS- and DHEA-treated groups and their recipients (Additional file 4: Figure S4b, e). Akkermansia, Bacteroides, Lactobacillus and Parabacteroides ranked most abundant at the genus level before and after FMT in donors and recipients, respectively (Additional file 4: Figure S4c, f). These differentially abundant taxa were overrepresented in the t-PBS group (Fig. 3c), including the genera Turicibacter and Clostridium sensu stricto, families Erysipelotrichaceae and Clostridiaceae 1, order Erysipelotrichales and class Erysipelotrichia which were also increased in PBS-treated donors. Among these differentially abundant microbes, the genus Turicibacter, family Erysipelotrichaceae, order Erysipelotrichales and class Erysipelotrichia were top-ranking by LDA score in the LEfSe analysis, which coincided with the gut microbiota profiling of donor rats (Fig. 3d, e). SourceTracker analysis (SourceTracker software version 1.0.1) showed that the donors were the principal origin of t-PBS and t-DHEA rat-gut microbiota (Fig. 3f). Thus, the DHEA-shaped gut microbiota was successfully transferred to recipient rats.

Kyoto Encyclopaedia of Genes and Genomes (KEGG) analysis was performed using the predicted gene abundance to understand the functional contribution of gut dysbiosis. For module enrichment, the synthesis and degradation of ketone bodies and G protein-coupled receptors were enriched in the t-PBS group, while beta-alanine metabolism, biotin metabolism and mineral absorption were enriched in the t-DHEA group (Fig. 3g). These data suggest that aberrant functions in environment perception, energy, vitamin, amino acid and mineral metabolism may occur in the gut microbiota after DHEA-shaped microbiome transplantation. Then, the correlations of genus-level taxa with clinical parameters were assessed (Fig. 3h). The testosterone level was negatively correlated with the relative abundances of Clostridium sensu stricto, Turicibacter, Olsenella and Anaeroplasma. The FAI was negatively associated with Olsenella levels only. The FSH level was positively associated with Clostridium sensu stricto, Olsenella and other taxa. In addition, the area under the curve of the GTT, and FBG levels were linked with levels of Clostridium sensu stricto, Anaeroplasma, and other taxa.

The faecal metabolome was investigated to further understand the functional alterations following FMT. The top 30 abundant metabolites are displayed in Fig. 4a. The faecal metabolites with significantly different levels between the t-PBS and t-DHEA groups included amino acid metabolites, fatty acids and bile acids (Fig. 4b, c). The levels of 2-hydroxystearic acid, argininosuccinic acid, 1-octen-3-yl primeveroside, daidzein and 4-acetamidobutanoic acid were significantly associated with the relative abundances of Anaeroplasma, Turicibacter and Clostridium sensu stricto (Fig. 4d). Figure 4e displays differentially expressed metabolites which were correlated with at least one clinical parameter. The correlation analyses between each 2 of these metabolites are shown in Fig. 4f. Functional enrichment analysis showed that arginine biosynthesis and alanine, aspartate and glutamate metabolism were affected by the DHEA-shaped gut microbiota transplantation (Fig. 4g).

FMT revealed that gut dysbiosis significantly affected metabolic function. The liver serves as the major site for physiological metabolism. Therefore, hepatic genetic expression and morphology were evaluated.

Haematoxylin & eosin (H&E) and oil red O (ORO) staining revealed that liver tissue sections in the t-DHEA group contained more and larger lipid droplets than the t-PBS group (Fig. 5a, b). The serum lipid profile was also measured (Fig. 5c). NEFA (non-esterified fatty acids) levels were significantly increased in the t-DHEA group, while high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), total cholesterol (TCHO) and triglycerides (TG) were similar between groups. These results suggest abnormal lipid metabolism; thus, hepatic gene expression profiling was performed to gain more insight into the regulation of glucose and lipid metabolism in the liver. Next-generation sequencing of the liver transcriptome identified differentially expressed genes (DEGs) according to the criteria p < 0.05 and fold-change > 1.2. A total of 725 DEGs were selected for subsequent analysis, including 274 upregulated and 451 downregulated genes (Fig. 5d). Canonical pathway analysis using Ingenuity Pathway Analysis (IPA) software revealed that the cholesterol biosynthesis superpathway, LXR/RXR activation and Gαi signalling were upregulated. The downregulated pathways included GP6 signalling, MIF regulation of innate immunity, LPS/IL-1 mediated inhibition of RXR function, MIF-mediated glucocorticoid regulation, endothelin-1 signalling, LPS-stimulated MAPK signalling and 14-3-3-mediated signalling (Fig. 5e). The changes in these pathways indicated that hepatic glucolipid metabolism and immune response were affected by DHEA-shaped gut microbiota transplantation. The interactions between DEGs regarding carbohydrate and lipid metabolism were visualised using IPA (Fig. 5f). Tumour necrosis factor and prostaglandin-endoperoxide synthase-2 (PTGS2) occupied the centre of the working network. Regulator effects analysis was used to combine upstream regulator analysis and downstream function analysis (Fig. 5g). It indicated that alterations in hepatic gene expression regulated fatty acid concentration, which coincides with dyslipidaemia. These data indicated that the dysregulation of hepatic gene expression might be involved in metabolic derangement triggered by the DHEA-shaped microbiome transplantation.

Thirty SD rats (female, 3-week old) were purchased from Vital River Laboratory Animal Technology Co Ltd., Beijing, China. Two rats per cage were housed in 15 individually ventilated cages in a specific pathogen-free environment and maintained under a 12-h light/dark cycle, controlled temperature and stable humidity, with 24 h free access to standard irradiated rodent feed and non-acidified drinking water. The rats’ body weights were measured weekly.

Eighteen rats were randomly allocated to 3 groups: PBS, DHEA and DHEA + antibiotics groups (n = 6 each group). To induce the PCOS phenotype, rats in the DHEA group were subcutaneously injected with 6 mg/100 g bodyweight DHEA (cat. no. H10940064, Yangzhou Pharmaceutical Co., Ltd) daily for 35 consecutive days. Rats in the PBS group were injected with PBS. Rats in the DHEA + antibiotics group were injected with DHEA in the same manner as the DHEA group and received a freshly prepared antibiotic cocktail daily in their drinking water. The antibiotic cocktail, consisting of ampicillin sodium (1 mg mL⁻¹), neomycin sulfate (1 mg mL⁻¹), metronidazole (1 mg mL⁻¹), and vancomycin hydrochloride (0.5 mg mL⁻¹) (all purchased from Sangon Biotech, China) [54], was administered to rats every day immediately after preparation. The rats in all 3 groups were sacrificed after 5 weeks of treatment.

Twelve other rats were randomly allocated into 2 groups, a t-PBS and a t-DHEA group (n = 6 each group). First, a three-course antibiotic treatment was administered to both groups of FMT recipient rats to establish the pseudo germ-free rat model as previously described [31, 54]. They were given an antibiotic cocktail in drinking water for 7 days, non-acidified drinking water for 2 days, the antibiotic cocktail in drinking water for 7 days, non-acidified drinking water for 2 days, the antibiotic cocktail in drinking water for 7 days, and non-acidified drinking water for 2 days before FMT. The faecal samples were collected and processed within 30 min to ensure bacterial viability in the transplant. The faecal pellets were pooled and homogenised with 1 mL PBS and 1 mL skimmed milk per pellet (0.2 g) [21]. Then, the mixture was allowed to settle by gravity for 5 min, and 1 mL supernatant was transferred to recipients by oral gavage every day for 14 days. After the 21-day PCOS model establishment, the donors continued to receive DHEA for another 14 days during FMT. The microbial loads in donors after DHEA treatment and in recipients after ABX treatment were assessed using MAKLER counting chamber. The recipients received approximately 1 × 10⁹ microbes by oral gavage daily and were sacrificed after 2 weeks of FMT.

All rats were fasted overnight for 16 h before GTT. The blood glucose levels were measured using tail vein blood before and 15, 30, 45, 60, 90 and 120 min after intraperitoneal glucose administration (50% glucose, 2 g/kg body weight). The total area under the GTT curve was calculated using GraphPad Prism 7.00.

Vaginal smears were taken daily between 9:00 am and 11:00 am for 8 consecutive days before sacrifice. The oestrous cycle stage was determined by observation of the vaginal smears using a light microscopic (Zeiss, Germany). Briefly, samples dominated by nucleated epithelial cells indicated the proestrus stage, and those with primarily cornified squamous epithelial cells indicated the estrus stage. The metestrus stage was indicated by comparative quantities of cornified squamous epithelial cells and leukocytes, while the diestrus stage was dominated by leukocytes.

The rats were euthanised by cervical dislocation. The ovaries were carefully dissected and fixed in 4% paraformaldehyde overnight at room temperature before being embedded in paraffin. The livers were also carefully separated. A sample of liver tissue was snap-frozen in liquid nitrogen and kept at – 80 °C for future use. Others were fixed and embedded using the same procedure as the ovaries. Eight-micrometre-thick tissue sections of ovaries and livers were processed by H&E and ORO staining according to standard histological procedures and examined using a light microscope (Zeiss, Germany).

At the end of the treatments, blood samples were collected from the abdominal aorta of anaesthetised rats using a pro-coagulation tube and centrifuged at 2500 rpm for 15 min at 4 °C. Then the upper supernatants were carefully transferred to cryogenic vials and immediately stored at – 80 °C for further evaluation. The levels of TT (582701, Cayman, USA), SHBG (MBS014745, Mybiosource, USA), LH (MBS2018978, MyBioSource, USA) and FSH (EKU04249↗, BIOMATIK, Canada) were analysed using enzyme-linked immunosorbent assay kits. To assess the biologically active amount of testosterone, the FAI was calculated using the following formula: FAI = (TT × 100)/SHBG [55]. Separate serum samples were sent to the Shanghai Model Organisms Center, Inc., to evaluate serum lipids.

The faecal pellets were collected in cryogenic vials immediately after being discharged before sacrifice. The faecal samples were snap-frozen and stored at – 80 °C before analysis. Microbial DNA was extracted from rat faecal samples using QIAamp Fast DNA Stool Mini Kit (51604, Qiagen, Hilden, Germany) according to the manufacturer’s instructions. A Class II biosafety cabinet was used for the preparation of the extracted DNA. The studies were carried out attentively by the same technician to reduce the risks of contamination or operational mistakes. The quantity and purity of the extracted DNA were controlled using NanoDrop 2000 (Thermo Scientific, USA). The integrity and size of the isolated DNA were evaluated using agarose gel electrophoresis.

The V3–V4 hypervariable region of the 16S rRNA gene was amplified by PCR under the following conditions: 95 °C for 3 min, followed by 30 cycles at 98 °C for 20 s, 58 °C for 15 s, 72 °C for 20 s and a final extension at 72 °C for 5 min. The primers used were 341F 5′-CCTACGGGRSGCAGCAG)-3′ and 806R 5′-GGACTACVVGGGTATCTAATC-3′ [56]. The 30-μL reaction system consisted of 15-μL 2 × KAPA Library Amplification ReadyMix, 20-μM primers, 50-ng template DNA and double-distilled water. The amplification products were isolated using 2% agarose gel, purified using a AxyPrep DNA Gel Extraction Kit (Axygen Biosciences, USA), and quantitatively measured by Qubit®2.0 (Invitrogen, USA). After library establishment, the tags were sequenced using the MiSeq platform (Illumina, Inc., USA) for 250-bp paired-end reads.

After trimming the primers and barcodes, the lengths and average base quality of the tags were evaluated. The 16S tags were limited to the region between 220 bp and 500 bp. The average Phred score of the bases were at least 20 (Q20), and at most 3 ambiguous N. The enumeration of the tags was followed by the removal of redundancy. Only the reliable tags (frequency > 1) were allocated into operational taxonomic units (OTUs) labelled by a representative tag. Next, Userach (version 7.0) was used to cluster the OTUs at 97% similarity and filter out the chimeric sequences. Every representative tag was aligned against the RDP database (http://rdp.cme.msu.edu↗) and assigned to a taxon (confidence threshold: 0.8). Tags which could not be distributed to any known taxonomy level were defined as “unclassified.” Python scripts of Qiime (V1.9.1) were included for OTU standardisation, OTU profiling and alpha/beta diversity assessment. Random levelling of samples was performed to reduce the deviation resulting from size differences. The microbial richness and evenness were evaluated by α-diversity by the PD whole tree index, which is based on the phylogenetic tree and evaluates diversity by summing the length of all branches. PCoA, adonis analysis and ANOSIM were employed to differentiate the microbial composition between groups [57]. The non-parametric Wilcoxon test was used to determine the statistical significance of taxa and KEGG orthologues (KOs). The differences in the relative abundance of bacterial taxa between groups were identified using the LEfSe method (version 1.0). The significance level was set at p < 0.05 and the effect size at LDA score > 2.0. STAMP (version 2.1.3) was used to identify differentially enriched KO modules between groups. Correlation analysis between gut microbiota and clinical parameters, gut microbiota and faecal metabolome, and faecal metabolome and clinical parameters was conducted using Spearman’s rho correlation test. The degree of transfer of gut microbiota from donor rats to pseudo germ-free rats was evaluated using SourceTracker software version 1.0.1.

Faecal metabolome analysis was conducted using Liquid Chromatography-Mass Spectrometry (LC-MS, Agilent 1290 UHPLC). The data were collected using AB 5600 Triple TOF under the control of Analyst TF 1.7 (AB Sciex). The raw data were converted using ProteoWizard and processed using XCMS, followed by metabolite identification. The metabolites were annotated using m/z, retention time, and MS/MS fragmentation patterns. The differential metabolites between groups were identified using the student’s t test and orthogonal projections to latent structures-discriminant analysis (OPLS-DA). Volcano and bubble plots were produced using the R package (version 3.5.1). The corrplot package was used for correlation analysis, the pheatmap package was used for hierarchical clustering, and the ropls package for OPLS-DA. DHEA-S was quantified in each pellet of the donor animals’ faeces using LCMSMS.

A RNeasy mini kit (Qiagen, Germany) was used to extract total RNA from the rat livers. The TruSeq® RNA Sample Preparation Kit (Illumina, USA) was employed to construct paired-end libraries based on the manufacturer’s guidelines. An Agilent 2100 bioanalyser was used to evaluate the concentration and size distribution of the cDNA library (Agilent Technologies, USA). Finally, the libraries were sequenced on the HiSeq Xten platform (Illumina, USA).

First, the raw reads were filtered by Seqtk (https://github.com/lh3/seqtk↗). Clean reads were obtained after adapter removal, quality control and host contaminant filtering of raw reads using Seqtk. Quality control was based on the Phred score of each base being more than 20, and the read length greater than 25 bp. Hisat2 (version 2.0.4) was employed to map the cleaned reads to Rattus norvegicus Rnor_6.0 [58]. The genes fragments were counted using Stringtie (version:1.3.0) followed by FPKM (Fragments Per Kilobase of exon model per million mapped reads) normalisation [59]. EdgeR was used to identify the DEGs based on FPKM values with the thresholds of p < 0.05 and fold-change > 1.2. The heat map was produced by R package pheatmap. IPA software was used for bioinformatics analysis of DEGs.

GraphPad Prism 7.00 was used to perform statistical analyses and data visualisation. The sample sizes were calculated using GraphPad StatMate 2.0 with power analysis. A Kolmogorov–Smirnov test was performed to examine the quantitative data for Gaussian distribution. A two-tailed unpaired student’s t test, Mann–Whitney test and one-way analysis of variance coupled with Turkey’s multiple comparisons test were performed to assess the differences between and among groups. The quantitative data were shown as means ± standard error of the mean (SEM). A difference of p < 0.05 was considered statistically significant.

All animal experimental procedures were approved by the Shanghai Research Center (company) for Model Organisms Institutional Animal Care and Use Committee (SRCMO IACUC, 2019-0031-06).

Not applicable

The authors declare no competing interests.

The 16S rRNA gene sequencing data have been deposited in the NCBI database under accession code PRJNA635286. Hepatic mRNA sequencing data is available at NCBI Gene Expression Omnibus (GEO), series accession number GSE151220↗. All data can be obtained in this manuscript or from the authors upon request.