The interplay between PCOS pathology and diet on gut microbiota in a mouse model

To compare microbial composition between diet groups, Bray-Curtis resemblance matrices (similarities) were calculated from the relative abundance of bacteria at the OTU level (β-diversity) and ordinated and tested using PCoA and adonis. These analyses revealed that microbial composition was significantly different between each diet type, as mice consuming the same diet clustered together but away from those consuming another diet type (adonis: R² = 0.25, P < .001; Figure 1b).

We then investigated the influence of protein, carbohydrate, and fat on pooled control and PCOS-like samples. To elucidate if microbial diversity and composition, regardless of PCOS phenotype, are influenced by a specific macronutrient (protein, carbohydrate and/or fat), data were analyzed based on increasing energy availability from protein, carbohydrate, and fat across diets (Table S2). Analysis revealed that all three macronutrients (P, C and F) influenced microbial diversity as measured by Shannon’s diversity index (α-diversity) in control and PCOS-like mice pooled together (all P < .0001; Figure 1c), with the greatest diversity achieved on diets with medium amounts of energy from Pro (23%), Carb (38%) and Fat (38% and 48%). As peak α-diversity was observed in diets with medium percentage compositions of P, C and F (Figure 1c), a ternary plot was created to visualize if the individual macronutrient peaks coincided (Figure 1d). Results from the ternary plot confirm that in mice there is a maximal peak in Shannon’s diversity at a very specific ratio of P, C and F dietary content of 23% P, 38% C and 38% F (Figure 1d).

Bray-Curtis resemblance matrix (similarities) of the relative abundance of bacteria at the OTU level (β-diversity) was visualized using PCoA to compare individual macronutrient groupings (Table S2, Figure 1e). This analysis showed that microbial composition was significantly different across protein groups (R² = 0.15, P < .001), carbohydrate groups (R² = 0.12, P < .001), and fat groups (R² = 0.14, P < .001), demonstrating the strong impact of macronutrients on the composition of the gut microbiota in mice. In particular, the amount of dietary energy pertaining to P and F exhibited a specific pattern in which the shift of β-microbial diversity changed progressively as dietary P and F increased (shown in change from red to yellow groups in Figure 1e). This shift in β-microbial diversity corresponding to P and F intake was observed to occur in opposite directions, as shown by arrows (Figure 1e).

In contrast, analysis of microbial composition, β-diversity, of control and PCOS-like mice across the ten diet formulations (Figure 2b) was found to be significantly influenced by the hyperandrogenic environment as tested by PERMANOVA (Pseudo-F: 1.6, P = .017, df = 172) and confirmed by CCA (F = 1.6, P = .002) and RDA (F = 1.4, P = .001) constrained analyses.

The most abundant taxa identified to be significantly different between control and PCOS-like mice was Bacteroides OTU0003 (Table S3). The relative abundance of Bacteroides OTU0003 was decreased in PCOS-like mice compared to control mice (control: 7.88% and PCOS 5.38%; fold change = 1.464; P < .0001; Figure 3c). The consensus sequences of Bacteroides OTU0003 was assessed using blastn against the NCBI database, which showed a 99.21% similarity to Bacteroides acidifaciens.

Estrous cycle irregularity is a key diagnostic criterion for PCOS, and the development of this trait was confirmed by the detection of estrous acyclicity in all PCOS-like mice (Figure 5g, p < .01). Similar to PCOS-like mice, 0% of FMT-treated PCOS-like mice displayed estrous cyclicity (Figure 5g). Oligo/anovulation is a hallmark reproductive characteristic of PCOS, and the development of this trait was confirmed by the detection of anovulation in PCOS-like females. Analysis of the ovarian histological section revealed that PCOS-like females displayed zero corpora lutea (CL) which was significantly different from controls that displayed an average of three CL/ovary (Figure 5h,j; p < .05). Similarly, CL were absent in the ovaries of all FMT-treated PCOS-like mice (Figure 5h,j). Ovary weight did not differ significantly between the three groups, although there was a tendency for ovarian weight to be increased in PCOS-like mice compared to control and FMT-treated PCOS-like mice (Figure 5i).

LEfSe analysis identified six OTUs to be significantly different between control and PCOS-like mice and control and FMT-treated PCOS-like mice. FMT treatment resulted in a shift in the average relative abundances of the six key taxa in the PCOS gut microbiome toward that of the control (Figure 6e). Compared to control mice, in PCOS-like mice the relative abundance of Barnesiella OTU6 (P = .021), Parabacteroides OTU2 (P = .0023), Alistipes OTU56 (P = .0046), Deltaproteobacteria OTU25 (P = .016) and Aestuariispira OTU70 (P = .036) were significantly decreased, while the relative abundance of Acetatifactor OTU11 (P = .036) was significantly increased (Figure 6e). FMT treatment induced a shift in gut microbiome in PCOS-like mice, which was toward the gut microbiome of the control group. In PCOS-like mice, FMT treatment resulted in a significant increase in the relative abundance of Barnesiella OTU6 (P = .0012), Parabacteroides OTU2 (P = .011), Alistipes OTU56 (P = .0026), Deltaproteobacteria OTU25 (P = .0054) and Aestuariispira OTU70 (P = .015), but a significant decrease in the relative abundance of Acetatifactor OTU11 (P = .021) (Figure 6e).

PCOS-like mice were generated by postnatal exposure to DHT. Female mice were implanted subcutaneously with either a blank or a DHT-filled implant at 3–4 weeks of age. Implants were made using a 1-cm SILASTIC brand tubing with an inner diameter of 1.47 mm and an outer diameter of 1.95 mm made of medical grade–grade silicon (Dow Corning Corp, catalog no. 508–006). Implants were filled with ~10 mg DHT before being sealed. DHT-filled implants provide a continuous release of DHT over 90 days and induce a PCOS-like phenotype in mice.^29,30 This PCOS mouse model was chosen for this study to reproduce both metabolic and reproductive features of PCOS, which more closely resembles the ‘obese’ PCOS phenotype that affects most women with PCOS. Blank implants were empty 1-cm sized SILASTIC tubing.

Experiment 2. At 10 weeks of age, fecal samples were collected from all 8 control mice once a week for 6 weeks for FMT-processing. The fresh fecal pellets were resuspended in saline/Phosphate Buffered Solution in a 1.5 ml Eppendorf® tube, then vortexed until homogenized. It was then centrifuged at low speed (~200 x g). The resulting supernatant was collected and added to sterile 10% glycerol saline solution. All the samples were frozen immediately after collection and stored at – 80°C. Once all 6 weeks of sample supernatants were stored, they were all thawed once to pool all samples together, aliquoted into daily use tubes and frozen for later use. At 17-weeks of age, placebo gavage (10% glycerol saline solution) was administered to all control mice and 8 PCOS-like mice, while antibiotics were given to another 7 PCOS-like mice, 3 times/week for 1 week, to deplete their gut microbiota. At 18-weeks of age, 3 times/week for 2 weeks, placebo gavage was administered to all control (Control + saline group), and PCOS-like mice (DHT + saline group), while the PCOS-like mice previously treated with antibiotics were treated with pooled fecal supernatant (DHT + antibiotic + FMT group).

At the end of the experiment, the entire intestinal tract was removed from mice from both experiments 1 and 2, and fecal contents from the colon were collected aseptically. All samples were snap frozen in liquid nitrogen and stored at – 80°C immediately after collection until DNA extraction.

Fecal samples were homogenized, and DNA extracted using QIAmp ® PowerFecal ® DNA Kit (Hilden, Germany) following the manufacturer’s instructions. DNA concentration was measured using a NanoDrop 1000 Spectrophotometer (ThermoFisher).

Sequencing procedures were performed at the Ramaciotti Center for Genomics (UNSW Sydney). DNA was used in a PCR reaction to amplify the V4 region of the bacterial 16S rRNA gene using the following primers:

515 f (AATGATACGGCGACCACCGAGATCTACAC XXXXXXXX TATGGTAATT GT GTGYCAGCMGCCGCGGTAA) and

806 r (CAAGCAGAAGACGGCATACGAGAT XXXXXXXXXXXX AGTCAGCCAG CC GGACTACNVGGGTWTCTAAT) and the Platinum Hot Start PCR Master Mix (2x) from ThermoFisher (cat.no. 17000–11). The thermocycler conditions for amplification were activation at 94°C for 3 min followed by 35 cycles of amplification that included denaturation at 94°C for 45 seconds, annealing at 50°C for 1 minute, extension at 72°C for 1 minute and 30 seconds, and one final extension cycle at 72°C for 10 minutes; finally, the samples were cooled and held at 4°C. Amplified samples were then visualized on a 2% e-gel with an expected band size for 515 F/806 R of approximately 300–350 bp. PCR products were then cleaned and normalized using SequalPrep Normalization plates. Normalization was performed according to the manufacturer’s instructions. PCR products were then cleaned using beads on a portion of the pooled library, which therefore concentrated on it and allowed for the removal of any residual small fragments present. To ensure accurate sizing information, the clean amplicon pooled products were run on a 2200 TapeStation (or Agilent Bioanalyzer or LabChip GXII). Then, the concentration of the final clean pool was measured using a qubit. Finally, amplicon sequencing was performed with the Illumina MiSeq Reagent kit v3 (2x250 base pairs) .⁵⁴

Raw reads were analyzed using the MiSeq standard operating procedures within Mothur v1.39.1^54,55 with the SILVA database and Ribosomal Database Project reference employed for alignment and classification, respectively. Quality filtering of 16S rRNA sequences and the subsequent subsampling of the data resulted in n = 21,361 total clean reads/sample for experiment 1 and n = 16,976 total clean reads/sample for experiment 2.

Microbial communities were examined for α-diversity and β-diversity, and changes in individual taxa. All tests were performed in Primer-E v6, Calypso,⁵⁶ GraphPad Prism v7, and R. Before diversity comparisons, the operational taxonomic unit (OTU) counts were normalized by a total sum (% relative abundance) followed by square-root transformation. Chao1 species (OTU) richness, species evenness and Shannon diversity index were used as α-diversity measures, and linear mixed models were employed to examine the effects of DHT, FMT, diet and macronutrient composition. To assess the effect of each individual diet on microbial diversity measurements, diet groupings in Table S1 were used as depicted in Figure 1a-b and Figure 2a-b utilized. To assess the direct effect of each individual macronutrient (protein, carbohydrate, and fat), data were grouped by the percentage of P (5% to 60%), C (20% to 75%) and F (20% to 75%) present in each diet (Table S1), producing six different groups for each macronutrient (Table S2). Classification of diets into these six different groups, ranging from low to high concentrations of each macronutrient, allowed detailed analysis of the influence of each macronutrient based on its availability. Figure 1c-d and 1f and Figure 4a-c were analyzed using macronutrient classifications in Table S2. The ternary plot was generated using the R packages ggtern and ggplot2.⁵⁷ For β-diversity comparisons, the Bray-Curtis resemblance matrix was generated from the square-root transformed relative abundances of OTU counts and visualized through principal coordinate analysis (PCoA). Adonis, PERMANOVA, Canonical correspondence analysis (CCA), and redundancy analysis (RDA) were used to examine the effects of diet, macronutrient composition and DHT implant using Calypso or Primer-E v6. Identification of differentially abundant taxa was performed using DESeq2.⁵⁸

Estrous-cycle stage was analyzed using vaginal epithelial cell smears taken daily for 5 weeks prior to collection.^29,59,60 Smears were collected using 15 μL of 0.9% sterile saline and transferred to glass slides to air dry. Dry smears were stained with 0.5% toluidine blue before examination under a light microscope. Estrous-cycle stage was determined based on the presence or absence of leukocytes, cornified epithelial cells, and nucleated epithelial cells. Proestrus was characterized by the presence of mainly nucleated and some cornified epithelial cells; estrous by the presence of mostly cornified epithelial cells; metestrus by the presence of both cornified epithelial cells and leukocytes; and diestrus by the presence of predominantly leukocytes.

Ovaries were dissected, weighed, and fixed in 4% (weight/vol) paraformaldehyde overnight at 4°C and stored in 70% (vol/vol) ethanol before histological processing and analysis. Four Ovaries per group were processed through graded alcohol and embedded in glycol methacrylate resin (Technovit 7100; Heraeus Kulzer). Embedded ovaries were sectioned at 20 μm, stained with periodic acid-Schiff and counterstained with hematoxylin. The corpora lutea population was quantified as previously described,^29,59,60 where whole-section images of every third section were taken under a light microscope using a DP70 Olympus camera and corpora lutea was identified with morphological properties consistent with luteinized follicles and observed through several serial sections.

Parametrial fat pads were weighed, fixed in 4% paraformaldehyde, embedded in paraffin, and sectioned at 8 μm. Sections were stained with hematoxylin and eosin and imaged at 40x magnification under the Olympus BX60 light microscope for histomorphometric analysis. Five different pictures were taken and assessed from each of three sections of the fat pad, with a minimum of 200 μm separating these sections. Adipocyte area was quantified using ImageJ version 1.51 software (NIH), as previously described.^29,59

Livers were first weighed whole before excision of the right lateral lobe. The right lateral lobe was fixed in 10% neutral-buffered formalin overnight at 4°C and then soaked in 30% sucrose (weight/vol) for 48 h; 1-cm × 1-cm portion of the fixed right lateral lobe was excised and embedded in an OCT compound (Tissue-Tek). 10-μm sections were cryo-sectioned from the embedded liver portion and air-dried onto slides. To visualize lipid deposition, slides were stained with oil red O in 60% isopropanol,⁴⁸⁶⁰ before histo-morphometric analysis. ImageJ version software (NIH) was used to quantify oil red O-positive staining. Three separate images were analyzed from three different sections taken 200 μm apart.

Statistical analyses were performed using GraphPad Prism v9. The normal distribution of the data was calculated using the Shapiro–Wilk test. Data that were not normally distributed were transformed before analysis using square-root or log transformation. Data that were not normally distributed after transformation were analyzed using the non-parametric Kruskal–Wallis test followed by Dunn’s multiple comparison test as a post hoc test. Data that were normally distributed were analyzed using one-way ANOVA followed by Tukey’s multiple comparison test as a post hoc test. P values <.05 were considered statistically significant.