Systems toxicogenomics of prenatal low-dose BPA exposure on liver metabolic pathways, gut microbiota, and metabolic health in mice

Abstract

Bisphenol A (BPA) is an industrial plasticizer widely found in consumer products, and exposure to BPA during early development has been associated with the prevalence of various cardiometabolic diseases including obesity, metabolic syndrome, type 2 diabetes, and cardiovascular diseases. To elucidate the molecular perturbations underlying the connection of low-dose prenatal BPA exposure to cardiometabolic diseases, we conducted a multi-dimensional systems biology study assessing the liver transcriptome, gut microbial community, and diverse metabolic phenotypes in both male and female mouse offspring exposed to 5 μg/kg/day BPA during gestation. Prenatal exposure to low-dose BPA not only significantly affected liver genes involved in oxidative phosphorylation, PPAR signaling and fatty acid metabolism, but also affected the gut microbial composition in an age- and sex-dependent manner. Bacteria such as those belonging to the S24–7 and Lachnospiraceae families were correlated with offspring phenotypes, differentially expressed liver metabolic genes such as Acadl and Dgat1, and key drivers identified in our gene network modeling such as Malat1 and Apoa2. This multiomics study provides insight into the relationship between gut bacteria and host liver genes that could contribute to cardiometabolic disease risks upon low-dose BPA exposure.

1. Introduction

The prevalence of cardiometabolic disorders, such as obesity, metabolic syndrome, type 2 diabetes (T2D), and cardiovascular diseases (CVD) has been increasing at a high rate. For instance, a study evaluating the prevalence of diabetes in the US showed that around 40% of the participants were classified as diabetic or pre-diabetic (Cowie et al., 2009). There has also been an alarming increase in the prevalence of obesity and metabolic syndrome in both adults and children (Dehghan et al., 2005; Smith, 2007; Sahoo et al., 2015). Among the risk factors for cardiometabolic disorders is exposure to environmental contaminants such as endocrine disrupting compounds (EDCs) (National Academies of Sciences, 2018). EDCs are a heterogenous group of chemicals that include pesticides, pharmaceutical agents, and industrial chemicals present in all compartments of the environment (Diamanti-Kandarakis et al., 2009). Among environmental EDCs, bisphenol A (BPA) is one of the most extensively studied due to its broad, global exposure. Humans are consistently exposed to BPA as a result of its ability to leach from consumer goods in plastic and lined metal containers (Sun et al., 2006; Tsai, 2006; Vandenberg et al., 2007; Calafat et al., 2008). According to several studies, the adult human BPA intake can range from 30 ng to 1 μg/kg/day (Kang et al., 2006; Lakind and Naiman, 2008, 2011; Huang et al., 2017; Huang et al., 2018). Currently, the no observed adverse effect level (NOAEL) is 5 mg/kg/day and tolerable daily intake level (TDI) in the U.S. is 50 μg/kg/day according to the EPA (EPA, 2010; Almeida et al., 2018). Moreover, animal studies using exposure doses below TDI (e.g. 0.5 μg-10 μg/kg/day) have shown significant effects on various health endpoints (Richter et al., 2007; Marmugi et al., 2012; Ziv-Gal et al., 2015; Lejonklou et al., 2017).

BPA exposure has been linked to increased body weight, obesity, insulin resistance, diabetes, and cardiovascular diseases in both human epidemiological and animal studies, although not without controversy (Teppala et al., 2012; Beydoun et al., 2014; Ranciere et al., 2015; Han and Hong, 2016; Hao et al., 2017; Mouneimne et al., 2017; Rubin et al., 2017; Wassenaar et al., 2017; Le Magueresse-Battistoni et al., 2018). The controversies in BPA findings are likely due to its complex effects. First, BPA effects are non-monotonic and vary significantly by exposure window (developmental vs adulthood) and dosages. In terms of exposure window, previous studies have suggested that the in utero period is a critical time point for BPA vulnerability (Liu et al., 2013). BPA has the capacity to cross the placenta and has been observed in both maternal and fetal plasma (Takahashi and Oishi, 2000; Schonfelder et al., 2002; Nishikawa et al., 2010). However, the effects and mechanisms of prenatal exposure remain poorly understood, particularly at low doses mimicking human exposure. Second, males and females have been reported to exhibit different BPA responses; however, sex-specific differences have yet to be fully elucidated. Lastly, previous molecular studies of BPA exposure have implicated the involvement of numerous tissues and molecular pathways. In particular, the liver and gut microbiome communities, which are important for metabolic health, have been found to play a role in BPA-induced adverse effects (Malaise et al., 2017; Meng et al., 2019; Shu et al., 2019; Feng et al., 2020). The establishment of the microbiome during childhood is not only important for normal development, but also plays a critical role in health status later in life (Ihekweazu and Versalovic, 2018). The microbial community is critical in maintaining homeostasis of the gut-liver axis, and the liver also plays a role in shaping the intestinal microbial communities (Tripathi et al., 2018; Albillos et al., 2020). Modulation of the intricate relationship between the host and microbiome can contribute to the development of many diseases such as cancer, metabolic, cardiovascular, immune and neurobehavioral disorders (Wallace et al., 2011; Sanz et al., 2015; Mayer and Hsiao, 2017; Tang et al., 2017; Vuong et al., 2017; Vivarelli et al., 2019). The role of gut microbiome-liver interactions has not been investigated for prenatal exposure to BPA at a human relevant dose.

To address the remaining gaps in our understanding of BPA activities as outlined above, we conducted a multi-dimensional systems biology study to examine the influence of prenatal low-dose BPA exposure on offspring across a broad range of metabolic parameters, liver transcriptome, and the gut microbiome in both sexes. We chose to use 5 μg/kg/day since it is a dose below TDI but has been shown to cause developmental effects. In addition, when considering the human to mouse dose conversion factor, this dosage is close to the range of estimated daily human intake (Nair and Jacob, 2016). Compared to most previous studies that have examined different molecular aspects separately, our study design allows for systematic investigation of the multidimensional connections across biological domains such as identifying bacteria that are associated with liver genes, pathways and metabolic phenotypes. Our study provides valuable insight into how BPA, particularly at a low dose, disrupts the gut microbiome-liver axis and elucidates the sex-specific molecular mechanisms involved that contribute to metabolic disorders.

2. Materials and methods

2.1. Chemicals

Due to the poor solubility of BPA, we first dissolved 1 g of BPA (Milipore Sigma, Cat. No.239658–50G, USA) into 3 mL of pure ethyl alcohol (Sigma-Aldrich, Cat. No. E7023, USA). This mixture was then added to 97 mL of corn oil (Fisher Scientific, Cat. No S25271, USA) to make a stock solution at 10 μg/μL concentration of BPA and 3% ethanol. The stock solution was further diluted as follows: 5 mL of the stock was added to 95 mL of corn oil to make Dilution One at a BPA concentration of 0.5 μg/μL and 0.15% of ethanol. Then, 2.5 mL of Dilution One was added to 247.5 mL of corn oil to make Dilution Two (working solution), which was at a BPA concentration of 0.005 μg/μL and 0.0015% ethanol. The appropriate amount of Dilution Two was added into a 1/2 Honey Nut Cheerio (treat) (Honey Nut Cheerio, General Mills) to have a final exposure dose of 5 μg/kg/day of BPA. The control group also received a treat with corn oil and the same amount of ethanol (0.0015%) as the BPA-treated group. Both control and treatment groups were administered with the appropriate amount of BPA and corn oil in the treat, according to body weight, for 19–21 consecutive days.

2.2. Animal care and BPA exposure

Eight-week-old female and male C57BL/6J mice were purchased from JAX Laboratory (Bar Harbor, ME, USA). Eight breeding pairs, each consisting of one female and one male mouse, were housed in BPA-free opaque cages with glass bottle water dispensers. Mice were maintained on a 12-hour light/dark cycle environment and were given ad libitum access to food and water. Food and water intake were measured throughout the breeding period. Mice were fed “Select Rodent Diet 50 IF/6F 5V5R” chow (Newco Distributors, Inc, Rancho Cucamonga, CA, USA) consisting of 50 ppm total isoflavones (genistein, daidzein and glycitein) to regulate dietary estrogenic activity. Due to low survival rate of litters from first-time breeders, the first litter of each breeding pair was removed from the experiment. Male breeders were removed at the time of litter birth to ensure that new pregnancies did not occur until after a repository period. Female mice did not receive any treatment during the first-time breeding phase of the experiment. A week after giving birth to the first litter, females were paired up with the corresponding males from the first breeding round. One or two days after breeding pairs were set up, a vaginal sperm plug was observed to confirm mating. After mating was confirmed, treatment of 5 μg/kg/day BPA or corn oil was administered to female mice for 19–21 consecutive days during the gestation period. Male mice did not receive any BPA or corn oil administration at any point during the experiment. BPA or corn oil treatment was administered every evening between 8 pm and 10 pm. BPA or corn oil was loaded into the treat with a pipette with caution to ensure no loss or leaking of the chemical or control solution. During the treatment period, female breeders were removed from their home cage and weighed in order to prepare the proper dosage of BPA or corn oil daily. Mice were individually placed into empty BPA-free opaque cages with only the treat containing BPA or corn oil and returned to home cages after full consumption. Treatment consumption time took up to two hours for the first few days until acclimation to the new environment was established. Within a few days, female mice consumed the treat within 10 min and were quickly returned to their home cage. BPA and corn oil treatments were terminated after 19–21 days (gestation period) in preparation for pup delivery. After delivery of pups, male breeders were removed from breeding cages. Pups were checked for milk spots regularly to ensure that dams were nursing. At three weeks of age, pups were weaned from each breeding cage and separated into new BPA-free opaque cages with glass bottle water dispensers according to sex with 3–4 mice per cage. Randomization of pups from different breeding cages was done by distributing pups from the same breeding cage into different cages to reduce the risk of cage effects. Each breeding pair had litter sizes ranging from 3 to 8 pups. Litter size was normalized by distributing small (3–5 pups) and large (6–8) litter sizes equally into new cages. Pup cages were maintained on a 12-hour light/dark cycle environment and were given ad libitum access to food and water. Metabolic phenotypes such as body weight was monitored weekly starting at two weeks of age until 24 weeks by weighing mice on a scale. Plasma levels of lipids and glucose tolerance were also measured, as detailed below. Fecal samples were taken at 11, 14, and 20 weeks of age and liver samples were collected at 24 weeks of age, as described below. At the end of the experiment, liver and white adipose tissues were collected and weighed. Liver tissues were subject to lipid assays, as detailed below. All experimental procedures were approved by the UCLA Animal Research Committee.

2.3. Intraperitoneal glucose tolerance test (IPGTT)

Male and female offspring mice were tested for disturbances in glucose metabolism using IPGTT at 10 and 21 weeks of age. Mice were fasted overnight for 14 hours prior to testing and had access to drinking water at all times. A 20% glucose solution was prepared for the IP injection (2 g of glucose/ kg body mass). A small incision was made on the mouse tail using a sterile scalpel blade to obtain a blood sample for the glucometer (<5 μL). Baseline glucose for mice was obtained prior to 20% glucose IP injection. Blood glucose was measured at 15, 30, 60 and 120 min post glucose injection to test clearance of glucose.

2.4. Plasma lipid and insulin quantification

To investigate possible perturbations in lipid and glucose metabolism in mice exposed to BPA in utero, plasma lipid, and insulin quantifications were performed following blood collection by retro-orbital bleeding. Briefly, mice were fasted overnight for 14 hours prior to being anesthetized with isoflurane for blood collection via retro-orbital bleeding using a microcapillary tube at seven weeks of age. Approximately 1% of total body weight of blood was collected into K2 EDTA tubes and placed on ice. To obtain plasma, blood was spun at 1500 g for 10 min at 4 °C and then stored at −80 °C. Plasma triglyceride (pTG), total cholesterol (pTC), high-density lipoprotein (pHDL), unesterified cholesterol (pUC), free fatty acid (pFFA), and insulin (pInsulin) were analyzed using enzymatic colorimetric assays at UCLA GTM Mouse Transfer Core as previously described (Meng et al., 2016). Plasma very low-density lipoprotein (pVLDL) was calculated using the formula: VLDL= (TG/5) and low-density lipoprotein (pLDL) was calculated using the formula: LDL = TC - HDL – VLDL.

2.5. Liver lipid assays

Lipids were extracted from liver samples using the Folch method as previously described (Folch et al., 1957). Briefly, approximately 60 mg of liver tissue was homogenized in methanol then incubated overnight in a 2:1 mixture of chloroform and methanol. After filtration, magnesium chloride was added to the mixture and centrifuged. The resulting aqueous phase and soluble proteins were aspirated. The remaining phase was evaporated using nitrogen gas then the dried lipids were dissolved in 1% Triton X-100 in chloroform. The samples were again dried and then resuspended in 1.8% Triton X-100 in water. Samples were then stored in −80 °C prior to lipid analysis. Liver triglyceride (livTG), total cholesterol (livTC), unesterified cholesterol (livUC), phospholipids (livPC) levels were measured using a colorimetric assay at the UCLA GTM Mouse Transfer Core.

2.6. Fecal collection

Fecal samples were collected at weeks 11, 14, and 20 from male and female offspring mice. At the time of fecal collection, mice were removed from their home cages and individually placed in an empty autoclaved BPA-free opaque cage without bedding or access to food or water. Mice were allowed to defecate undisturbed. The time of fecal collection was at the beginning of the 12-hour dark cycle. Using sterilized tweezers, 3–5 pieces of fecal samples were collected into sterile 2 mL Eppendorf tubes and flash frozen in liquid nitrogen. Tweezers were thoroughly cleaned in between sample collections with 80% ethanol. Samples were then stored at −80 °C until analysis.

2.7. 16S rRNA sequencing and data analysis

DNA was isolated from fecal samples using the PowerSoil DNA Isolation Kit (Mo Bio, Carlsbad, CA) and the bacterial 16S rRNA gene was amplified for amplicons spanning the variable region 4 (V4) using barcoded primers. The amplicons from different samples were pooled, purified, and sequenced using the Illumina HiSeq4000 instrument. Microbial sequence data was analyzed using QIIME 2 (version 2018.6). Sequencing reads were denoised and clustered into amplicon sequence variants using DADA2 (Callahan et al., 2016). Using DADA2, the reads were truncated at 150 bases to obtain only sequences with quality control measure above 33. Resulting sequences were taxonomically classified against Greengenes reference sequences (McDonald et al., 2012). To compare the control and BPA groups, principal coordinates analysis (PCoA) was performed using the Bray-Curtis dissimilarity indices, and principal component analysis (PCA) was performed using Hellinger-transformed amplicon sequence variant (ASV) abundances. For each ASV, DESeq2 was used to compare the abundances between BPA and control groups at each time point within each sex to obtain p values for differential abundance between groups. The p-values were adjusted using the Benjamini-Hochberg (BH) method to estimate false discovery rate (FDR). ASVs that resulted in an FDR < 0.05 were considered significant. The ASVs altered by BPA exposure were then compared across time points and sexes to identify consistent microbes affected by BPA.

2.8. Liver RNA sequencing (RNAseq) and data analysis

The whole liver tissue was dissected out upon terminal sacrifice at 24 weeks of age. A portion of the right lobe was then collected, flash frozen in liquid nitrogen, and stored in −80 °C until RNAseq analysis. RNA was extracted from the liver tissues using the MiRNeasy Mini Kit purchased from Qiagen following the manufacturer’s instruction (Hilden, Germany). Liver samples used were from mice randomly selected from litters of different dams from independent cages (n = 4/group/sex). The RNA concentration and quality of all 16 liver samples were measured using the Thermo Scientific NanoDrop instrument (Waltham, MA) and the Agilent High Sensitivity RNA ScreenTape System (Agilent, Santa Clara, CA). Libraries were prepared using the Total RNA Library Prep Kit with Ribo-Zero Gold (Illumina, San Diego, CA). Single-end read sequencing was performed on an Illumina Hiseq4000 system. After sequencing, the read quality was assessed using FastQC (http://www.bioinformatics.babraham.ac.uk/projects/fastqc). After quality control, sequence alignment and transcript assembly were done using the HISAT-StringTie pipeline (Pertea et al., 2016). Reads were mapped to the GRCm38/mm10 reference genome (https://genome.ucsc.edu/index.html). Identification of differentially expressed genes (DEGs) were conducted using DEseq2 (Love et al., 2014) and FDR was estimated using the BH method. Genes with an FDR < 0.01 were considered significant. To investigate the pathways enriched in the BPA-associated DEGs, we used Gene Set Enrichment Analysis (GSEA) (Mootha et al., 2003; Subramanian et al., 2005) to annotate the DEGs with known biological pathways curated from the Kyoto Encyclopedia of Genes and Genomes (KEGG) (Kanehisa and Goto, 2000). KEGG pathways at FDR < 0.05 were considered significant. Fisher’s exact test was applied to evaluate the significance of overlapping genes and pathways between DEG sets, such as between male and female liver DEGs. The RNAseq data has been submitted to the Gene Expression Omnibus (GEO) under accession number GSE155845.

2.9. Identification of transcription factor (TF) and non-TF regulators using network analyses

To perform network analyses on the liver DEGs identified, we used Mergeomics (https://bioconductor.org/packages/devel/bioc/html/Mergeomics.html) to identify key regulators (Shu et al., 2016). To identify potential TFs important for the BPA-induced effects, we first assessed whether BPA-associated DEGs were downstream targets of specific liver TFs using Marker Set Enrichment Analysis (MSEA). Liver tissue TF regulatory networks were obtained from the FANTOM5 database (Marbach et al., 2016). Using MSEA, we evaluated the enrichment of BPA-induced DEGs within the target genes for each TF. We then used Weighted Key Driver Analysis (wKDA) to identify potential non-TF network regulators using previously constructed liver Bayesian networks (Zhu et al., 2007; Zhu et al., 2008; Shu et al., 2019). In brief, these networks were built using multiple large mouse and human liver transcriptome and genetic datasets. To identify key drivers (KDs), we searched for nodes/genes whose neighboring subnetworks were enriched for BPA-induced DEGs in the liver. TFs or non-TF network regulators with FDR < 5% were considered statistically significant. The cytoscape software was used for network visualization (Shannon et al., 2003).

2.10. Correlation analysis between gut microbiota and metabolic phenotypes or liver signature genes

To investigate the relationships among the liver transcriptome, gut microbiome, and metabolic phenotypes affected by low-dose BPA exposure, correlation analysis was performed. We first correlated the abundance of significantly altered microbial ASVs with metabolic phenotypes, including body weight, area under the curve (AUC) for glucose tolerance, plasma insulin, plasma lipids, liver lipids and tissue weight (liver and adipose). The bacteria significantly correlated with phenotypes were then correlated to liver DEGs affected by BPA. Correlation was assessed using Biweight midcorrelation (bicor), which is robust to outliers. P-values were adjusted using the BH approach and FDR < 0.05 was considered significant. The BPA-affected genes demonstrating significant bacterial correlations were evaluated for enriched KEGG pathways using the GSEA tool. Pathways and biological functions at FDR < 0.05 were considered significant.

3. Results

3.1. Low-dose BPA exposure in utero exhibits sex-specific effects on metabolic phenotypes in offspring

After in utero BPA exposure, various metabolic phenotypes of the offspring were monitored. In female offspring, significantly increased body weight in BPA-treated mice was observed at 13, 14, 20 and 22 weeks of age compared to controls (Fig. 1A). IPGTT results used to evaluate glucose tolerance was not altered in females at both week 10 and week 21 timepoints (Fig. 1B–1D). There was also no effect observed for plasma insulin (pInsulin) levels (Fig. 1E). In contrast, BPA-treated male offspring showed a significant decrease in body weight at week 4 (Fig. 1F) and faster glucose clearance based on the IPGTT results at week 10 and 21 week (Fig. 1G–1I) in the absence of significant changes in fasting plasma insulin levels (Fig. 1J). These results suggest enhanced insulin sensitivity in BPA-treated males. We also measured circulating plasma lipids in female and male offspring (pTG, pFFA, pTC, pUC, pHDL, pLDL and pVLDL). BPA-treated female mice did not show significant changes in any of the plasma lipid traits (Fig. 1K) compared to control mice. BPA-treated male mice showed significantly decreased pTG, pFFA, pUC, and pVLDL, but not pTC, pHDL, or pLDL (Fig. 1L). Lipids in the liver (livTG, livTC, livUC, livPC) were also measured. In BPA-treated females, there was a significant decrease in livTG levels (Fig. 1M). However, in males there was an increase in livTC and livUC (Fig. 1N). In both females and males, no difference was observed in the liver and adipose weight (Fig. 1O–1P). There was a significant difference in tissue weight to body weight ratio for liver but not for adipose in BPA-treated females (Fig. 1Q). No effect in tissue weight to body weight ratio in males was observed (Fig. 1R). Overall, our findings suggest sex-specific differences in BPA-induced metabolic phenotypes.

Fig. 1.

Metabolic phenotypes of offspring exposed to BPA in utero. Female offspring average body weight (A) and glucose tolerance test at 10 weeks (B) and 21 weeks (C) of age after in utero BPA exposure. Bar plot of glucose tolerance AUC in female offspring (D). Female fasting insulin (pInsulin) levels in plasma at 7 weeks (E). Male offspring average body weight (F) and glucose tolerance test at 10 weeks (G) and 21 weeks (H) of age after in utero BPA exposure. Bar plot of glucose tolerance AUC in male offspring (I). Male fasting insulin (pInsulin) levels in plasma at 7 weeks (J). Bar plot of female (K) and male (L) plasma lipid profiles at 7 weeks. Plasma triglycerides (pTG), free-fatty acids (pFFA), total cholesterol (pTC), unesterified cholesterol (pUC), high-density lipoprotein (pHDL), low-density lipoprotein (pLDL), very low-density lipoprotein (pVLDL). Bar plot of female (M) and male (N) liver lipid profiles: liver triglycerides (livTG), total cholesterol (livTC), unesterified cholesterol (livUC), phospholipids (livPC). Bar plot of the liver and white adipose weight in females (O) and males (P). Bar plot of the tissue weight to body weight ratio of liver and white adipose in females (Q) and males (R). Data are shown as means ± S.E; n = 6–10 per group/sex. Statistical analysis was performed using 2-way ANOVA followed by Bonferroni’s multiple comparisons test (A-C, F-H). Statistical analysis was performed using a two-tailed Student’s t-test for unpaired comparisons (D, E, I, J, K-R). Asterisk (*) represents p-value ≤ 0.05.

3.2. Low-dose BPA exposure in utero affects liver transcriptome and pathways in both male and female offspring

To investigate the host tissue responses to low-dose BPA exposure, RNAseq was performed on offspring liver tissue. As a result, 1958 DEGs (1001 down-regulated and 957 up-regulated) were observed in the females and 1429 DEGs (788 down-regulated and 641 up-regulated) were observed in the males (FDR < 0.01) (Fig. 2A and B; full DEG list in Supplementary Table 1). There were 817 common DEGs between males and females (454 down-regulated and 363 up-regulated), which represents a significant overlap in DEGs between sexes (fold enrichment = 7.026, p < 1.0e−100 by Fisher’s exact test) (Fig. 2C–E) despite the differences in the phenotypic manifestations and certain unique gene expression changes in each sex.

Fig. 2.

Liver transcriptomic alterations and pathway enrichment analysis of differentially expressed genes (DEGs) in the liver of male and female offspring exposed to BPA in utero. Volcano plots of differentially expressed genes (DEGs) identified in the female (A) and male (B) offspring determined by Deseq2 analysis. Red dots represent significant genes FDR < 0.01. Venn diagram comparing all DEGs (C), down-regulated DEGs (D), and up-regulated DEGs (E) in female and male liver transcriptome analysis. Significant overlap was determined by Fisher’s exact test. Bar plot of top 20 enriched Kegg pathways identified in the female (F) and male (G) offspring. Enrichment FDR q-value (shown beside the bars) was determined by GSEA. Venn diagram comparing the significant Kegg pathways identified using all DEGs (H), down-regulated DEGs (I), and up-regulated DEGs (J) in female and male liver transcriptome analysis. Significant overlap was determined by Fisher’s exact test. Sample size n = 4 per group/sex. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Pathway enrichment analysis of all DEGs in female showed 99 overrepresented KEGG pathways (Fig. 2F; full pathway list in Supplementary Table 2) and male all DEGs were enriched for 89 KEGG pathways (Fig. 2G; full pathway list in Supplementary Table 2). When comparing the pathways between sexes, 73 were common (fold enrichment = 1.541, p = 1.1e−15) (Fig. 2H), including oxidative phosphorylation, neurodegenerative diseases, PPAR signaling, fatty acid metabolism, and insulin signaling. Female-specific pathways included thyroid cancer and gluconeogenesis, while T cell receptor, regulation of autophagy and mTOR signaling were unique for males (Fig. 2H).

When considering only the down-regulated DEGs, 30 pathways were overlapping between females (41 pathways) and males (54 pathways) (fold enrichment = 2.520, p = 1.3e-12), including oxidative phosphorylation, neurodegenerative diseases, and WNT signaling. Pathways such as thyroid cancer and cardiac muscle disease were unique for the females while xenobiotic metabolism and T and B cell receptor signaling were unique for the males among the downregulated DEGs (Fig. 2I; full pathway list in Supplementary Table 2).

When only considering the up-regulated DEGs, 26 pathways were common between females (61 pathways) and males (33 pathways) (fold enrichment = 2.402; p = 1.8e−10), including oxidative phosphorylation, neurodegenerative diseases and glutathione metabolism. Drug metabolism and ABC transporter are examples of pathways that were unique for the females while ECM receptor interaction and fatty acid metabolism were unique for the males among the up-regulated DEGs (Fig. 2J; full pathway list in Supplementary Table 2).

The above results indicate that numerous metabolic pathways were perturbed in both males and females in addition to certain sex-specific pathways, which could underlie the observed sex-specific differences in phenotypic manifestation.

3.3. Identification of TF and non-TF regulators potentially governing BPA induced molecular perturbations

To explore the TFs that likely regulate the liver DEGs affected by BPA exposure, we leveraged liver-specific TF regulatory networks from the FANTOM5 project and identified TFs whose target genes were enriched in BPA DEGs. At FDR < 5%, there were 13 and 18 TFs identified in males and females, respectively (Full TF list in Supplementary Table 3). There were six (e.g. Tfeb, Nr5a1, Egr4) unique TFs identified in males and 11 TFs (e.g. Rora, Srebf2, Hnf4) in females (Fig. 3A and B). The male-specific TFs identified play roles in lysosomal biogenesis, development, and mitogenesis. Female-specific TFs play roles in cholesterol homeostasis, lipid metabolism, and liver development. There were also seven TFs (Zfp143, Yy1, Elf5, Etv5, Elk4, Etv1 and Arnt) shared between sexes (Fig. 3C). The common TFs are involved in drug metabolism, cell growth and differentiation, and circadian rhythm.

Fig. 3.

Transcription factor and key driver analysis of BPA-induced liver gene expression changes. Liver transcription factor (TF) regulatory networks unique to male (A), female (B), and shared among male and female (C). Networks are of significant TFs (FDR < 5%) identified based on enrichment of liver DEGs among TF downstream targets. TFs are represented as diamond-shaped nodes while known DEG targets are represented as circles. Topology of the network was based on FANTOM5. Only network edges that are also DEGs are presented. For the male (A) transcription factor network, DEGs shown are those with a FDR < 1e-3. For the female (B) and shared transcription factor network, DEGs shown are those with a FDR < 1e-5. Gene-gene regulatory subnetworks of top 10 key drivers (KD) in male (D), female (E) and shared among male and female offspring (F). Networks are of the top 10 KDs (FDR < 5%) identified based on enrichment of liver DEGs. Topology of the network was based on Bayesian modeling of liver datasets from mouse and human. KDs are represented as larger circular nodes. DEGs shown had a FDR < 0.01.

To identify regulatory genes that mediate the action of BPA on downstream targets through non-TF mechanisms, we leveraged data-driven liver-specific Bayesian networks (BNs) to predict gene-gene regulatory relationships and identify key drivers (KDs) of the BPA DEGs. At FDR < 0.05, we identified 35 and 36 KDs in male and female offspring, respectively (Full KD list in Supplementary Table 4). The top five unique KDs identified in males were Capza2, Apoa2, Ttr, Sepp1 and Akr1c12 (Fig. 3D). The top male KDs have roles in lipid metabolism, protein transport, and hormone metabolism. In females, the top five unique KDs were Ktn1, Tmed5, Golga4, F2 and Sult1a1 (Fig. 3E). The female KDs are involved in hemostasis, xenobiotic clearance, and protein transport. There were four shared KDs identified between male and female offspring (Dcxr, Erbb2ip, Tlk1 and Malat1) (Fig. 3F). The common KDs have functions in lipid pathways, xenobiotic metabolism, and chromatin assembly. These KDs, along with the identified TFs from the above analysis, may represent the regulatory targets which transmit the in vivo biological effects of BPA.

3.4. Low-dose prenatal BPA exposure affects the abundance and diversity of gut microbial species in both male and female offspring

Next, to investigate the effects of low-dose BPA exposure on the gut microbiota, 16S rRNA sequencing was performed on feces collected from both male and female offspring at 11, 14, and 20 weeks of age. Diversity metrics were used to evaluate community composition within a sample. As shown in the rarefaction curves based on observed ASVs, the number of ASVs detected increased and plateaued as the number of extracted sequences increased, verifying the sequencing depth was appropriate to observe all sample species (Fig. 4A). Alpha diversity was calculated using the Shannon diversity index, which indicated no significant difference between treatments (Fig. 4B). Next, differences in the gut communities between BPA- treated animals and controls were evaluated. Both PCoA based on Bray-Curtis distance and PCA using Hellinger-transformed abundance of the gut microbial community showed that the BPA groups clustered separately from the control offspring (Fig. 4C and D) in an age-dependent manner, with the most pronounced effect observed at week 20. We observed fluctuations in the abundance of bacterial families such as Bifidobacteriaceae, Bacteriodaceae and S24–7 at different stages (Fig. 4E).

Fig. 4.

Gut microbiome diversity in mice offspring exposed to BPA in utero. Rarefaction plot using observed amplicon sequence variants (ASVs) comparing fecal profiles of offspring exposed to BPA in utero (A). Box-and-whiskers plots of Shannon diversity index of fecal profiles from offspring exposed to BPA in utero (B). PCoA using Bray-Curtis distance metrics of offspring fecal profiles exposed to BPA in utero (C). PCA using Hellinger-transformed abundances of fecal samples offspring exposed to BPA in utero (D). Each dot represents the fecal microbiota of each mouse at week 11 (W11), week 14 (W14), and week 20 (W20). Taxa bar plot of most abundant bacterial families in offspring fecal samples at different time points (W11, W14 and W20) after in utero exposure to BPA (E). Sample size n = 6–10 per group/time/sex.

We then analyzed the abundance of specific ASVs between BPA-treated and control samples. In females, we identified 55, 63, and 71 differentially abundant ASVs in week 11, 14 and 20, respectively (Fig. 5A; full ASV list in Supplementary Table 5). Among these, seven ASVs (Oscillospira, Prevotella, YS2, members of the Rikenellaceae, Lachnospiraceae, and Ruminococcaceae families with unknown species, and member of the order Clostridiales unknown species) were consistently significant in females at all time points (Fig. 5A–C). Prevotella was found in all time points in control samples while its abundance was low or not observed in BPA-treated samples in the female offspring (Fig. 5B). In the male offspring, we identified 112, 137, and 108 differentially abundant ASVs in week 11, 14, and 20, respectively (Fig. 5D; full ASV list in Supplementary Table 6), among which 30 ASVs (e.g. Lactobacillus, Coprococcus, Dehalobacterium, Ruminococcus, and Streptococcus) were consistently significant at all time points (Fig. 5D). Lactobacillus had decreased abundance in the BPA-treated samples at all time points in males (Fig. 5E). Interestingly, among the ASVs consistently altered in all time points between female and male offspring there were four ASVs in common, one of which belongs to the Rikenellaceae family (Fig. 5C and F). This ASV was increased in the BPA-treated mice regardless of sex.

Fig. 5.

Taxonomic composition at different time points of offspring exposed to low-dose BPA in utero. Venn diagram representing the number of unique and shared differentially abundant amplicon sequence variants (ASVs) at each time point in female fecal samples (A). Bar plots of the abundance of Prevotella (B) and Rikenellaceae unknown species (C) at all time points in females. Venn diagram representing the number of unique and shared differentially abundant ASVs at each time point in male (D) fecal samples. Bar plots of the abundance of Lactobacillus (E) and Rikenellaceae unknown species (F) at all time points in males. Bar plot data are presented as mean abundance ± SEM. Bar plot of the number of ASVs significantly correlated to female and male phenotypes (G). Correlation plot between S24–7 unknown species abundance and female body weight (H). Correlation plot between S24–7 unknown species abundance and male glucose tolerance AUC (I). Bicor score = Biweight midcorrelation (bicor) coefficient, Padj = Benjamini-Hochberg adjusted P-values. Sample size n = 6–10 per group/time/sex.

3.5. Relationship among host metabolic phenotypes, gut microbiota, and liver genes in response to low-dose BPA

To evaluate the relationship between the gut microbiota, host liver genes, and metabolic phenotypes, we first correlated microbial ASVs that showed alterations by BPA at any given time point with metabolic phenotypes collected from the same sets of mice. There was a total of 21 and three different ASVs correlated to metabolic phenotypes in female and male offspring, respectively (Fig. 5G; full correlation list in Supplementary Table 7). In females, many of the ASVs correlating with phenotypes belonged to the S24–7 (Fig. 5H), and several belonged to the Erysipelotrichaceae and Lachnospiraceae families. Of the 21 ASVs correlated to female phenotypes, two (YS2 and Prevotella) were consistently altered across all three time points. The remaining were only significant in one (nine ASVs) or two time points (ten ASVs). Out of the 21 phenotype-correlated ASVs, 18 were significantly correlated to body weight and 5 were correlated to pTG and pVLDL levels. There were two ASVs (Lactobacillus and S24–7 unknown species) that were correlated to body weight, pTG and pVLDL. In the male offspring, none of the three correlated ASVs were consistently altered across time points; each ASV was unique to a specific time point. There were no significantly correlated bacteria to insulin or any of the lipids quantified in either plasma or liver in males. Only one ASV was correlated to body weight. This bacterium was identified as part of the Allobaculum genus which belongs to the Erysipelotrichaceae family. The other two ASVs were correlated to glucose tolerance AUC. One bacterium belonged to the S24–7 family (Fig. 5I) and the other to the order Clostridiales.

Next, we examined the correlation between the phenotype-correlated ASVs and liver DEGs. Among the 21 ASVs correlated with female phenotypes, only seven ASVs were correlated to BPA-responsive liver DEGs (Table 1; full correlation list in Supplementary Table 8). Among these seven ASVs, four belong to the Clostridiales order and two belonged to the S24–7 family, and the DEGs correlated to bacterial abundance were enriched for pathways such as oxidative phosphorylation, RNA metabolism, glycan biosynthesis and insulin signaling (Table 1; full pathway list in Supplementary Table 9). Among these correlated DEGs, several were KDs, such as F2, Malat1 and Ktn1 (Table 1). For the three male ASVs correlating with body weight or glucose tolerance, two ASVs (Clostridiales and S24–7 both unknown species) showed correlations with liver DEGs involved in oxidative phosphorylation, drug metabolism, mTOR signaling and fatty acid metabolism (Table 1; full correlation list in Supplementary Table 8; full pathway list in Supplementary Table 10;). Among these correlated DEGs, several were also KDs in our network analysis, such as Apoa2, Akr1c12 and Malat1 (Table 1). These results suggest connections between the liver-gut axis and metabolic phenotypes.

Table 1.

Correlation between BPA altered microbiota and liver DEGs in offspring.

	Taxa of correlated ASVs	No. of Correlated DEGs BH padj < 0.05	Enriched KEGG Pathways	Correlated Key Drivers
Female	S24–7 unknown species	1678	Oxidative Phosphorylation, Neurodegenerative diseases, RNA metabolism, Citrate cycle, Insulin signaling pathway, PPAR signaling pathway, Glycan biosynthesis	Ktn1, Golga4, F2, Gcc2, Gls2, Aasdhppt, Qdpr, Plod3, Tlk1, Hmgcl, Hspa1b, Cryl1, Gnmt, Malat1, Apoc3, Dnajb1, Esf1
	Clostridiales unknown species	1660	Oxidative Phosphorylation, Neurodegenerative diseases, Lysosome, PPAR signaling pathway, Insulin signaling pathway	Ktn1, Golga4, F2, Gcc2, Gls2, Aasdhppt, Qdpr, Plod3, Tlk1, Hmgcl, Hspa1b, Cryl1, Gnmt, Apoc3, Dnajb1, Esf1, Dcxr
	Coprococcus	1140	Oxidative Phosphorylation, Neurodegenerative diseases, Citrate cycle, RNA metabolism, Insulin signaling pathway	Ktn1, Golga4, F2, Gls2, Qdpr, Plod3, Hspa1b, Cryl1, Malat1, Dnajb1
	YS2 unknown species	867	Oxidative Phosphorylation, Neurodegenerative diseases, Citrate cycle, Endocytosis, Amino acid metabolism	F2, Gls2, Qdpr, Plod3, Tlk1, Cryl1, Atp2a2, Dnajb1
	Clostridiales unknown species	11	none	none
	Lachnospiraceae unknown species	9	none	none
	S24–7 unknown species	1	none	none
Male	Clostridiales unknown species	1394	Neurodegenerative disease, Oxidative phosphorylation, Amino acid metabolism, PPAR signaling pathway, Fatty acid metabolism, Drug metabolism, mTOR signaling	Apoa2, Sepp1, Tlk1, Akr1c12, Erbb2ip, Angptl3, Maob, Hrsp12, Dynlt3, Pebp1, Fabp1, Fbxo31, Acadm, Malat1
	S24–7 unknown species	522	Amino acid metabolism, Propanoate, metabolism, Neurodegenerative disease, Oxidative phosphorylation, Fatty acid metabolism, Insulin signaling pathway, mTOR signaling	Apoa2, Akr1c12, Angptl3, Maob, Hrsp12, Dynlt3, Acadm

Amplicon sequence variants (ASVs); Benjamini-Hochberg adjusted p value (BH padj).

4. Discussion

Previous studies that examined developmental exposure to BPA at higher concentrations have reported effects on liver and gut microbiome separately. In our current study, we uniquely looked at the effect of low-dose BPA exposure (5 μg/kg/day) on the gut-liver axis from the same animals in both male and female offspring exposed in utero. This systematic multiomics investigation (tens of metabolic phenotypes, gut microbiome, and liver transcriptome) allowed us to investigate the relationship across molecular and phenotypic domains. Our study is the first to show that in samples obtained from the same animals, abundances of certain bacteria are correlated with liver gene expression and diverse metabolic phenotypes, supporting a relationship between the gut microbiome, the host liver, and metabolic perturbations induced by low-dose, prenatal BPA exposure (Fig. 6). Importantly, we found sex-specific differences in metabolic phenotypes, liver transcriptome, and gut bacteria, although consistencies were found primarily in various metabolic pathways such as oxidative phosphorylation and PPAR signaling. Different BPA-altered bacteria in male and female offspring were also correlated with liver DEGs which are involved in different aspects of metabolism. In this study, the specific microbial species and the associated pathways in host liver tissues revealed insights into the molecular mechanisms through which BPA exposure contributes to metabolic risk, opening unique opportunities to modulate gut bacteria to neutralize EDC adverse effects.

Fig. 6.

Summary of the effects of in utero BPA exposure on liver metabolic pathways, gut microbiota, and metabolic phenotypes. Underline denotes shared between male and female offspring. Fatty acid metabolism (FA metabolism); Oxidative phosphorylation (Oxidative phos).

In the present study, we observed differences between the male and female offspring in BPA-induced metabolic phenotypes (Fig. 6). Agreeing with the distinct metabolic phenotypes between sexes, our study of the liver transcriptome also revealed unique pathways perturbed by low-dose prenatal BPA exposure between male and female offspring that potentially play a role in determining phenotypic differences (Fig. 6). Specifically, in utero BPA-exposed females had a significant increase in body weight in adulthood compared to the BPA-exposed males. This increase in body weight in females occurred despite no alteration in white adipose tissue weight, suggesting that other tissues not evaluated are potentially affected by BPA. The female DEGs were uniquely enriched for glycan biosynthesis pathway, which has been shown to be important in shaping the gut microbiota and is important for metabolic health (Koropatkin et al., 2012). Agreeing with this, the glycan pathway was also correlated to BPA affected bacteria such as S24–7, which has been linked to diverse metabolic phenotypes including body weight. Therefore, the glycan pathway may underlie the weight phenotype seen in females. In contrast to females, males had a significant decrease in weight at an early timepoint. This was accompanied with lower plasma lipid levels such as TG, which is consistent with our previous prenatal study at 5 mg/kg/day of BPA (Shu et al., 2019). However, this is in contrast to other studies that have reported an increase in plasma TG (Lejonklou et al., 2017). When evaluating all the liver DEGs a pathway identified for males was mTOR signaling which has been linked to restricted fetal growth (Hart et al., 2019). Our study at both the phenotypic and liver transcriptome levels in general confirms significant sex-specific differences in metabolic related processes such as energy metabolism noted in BPA studies and reviews (Mackay et al., 2013; Strakovsky et al., 2015; Nadal et al., 2017; Marraudino et al., 2018), but uniquely point to sex-specific pathways such as glycan biosynthesis for females and mTOR for males. In addition, our study found that xenobiotic metabolism pathways were enriched in downregulated DEGs in males but were enriched in upregulated DEGs in females, which could also contribute to the sex-specific phenotypes observed. Enzymes involved in xenobiotic metabolism have been shown to play a role in metabolic diseases (Xu et al., 2016). Altered expression of important enzymes in xenobiotic metabolism such as the cytochrome p450 family have been observed in different diabetic and obesity mouse models and can affect disease susceptibility (Roe et al., 1999; Yoshinari et al., 2006; Kudo et al., 2009; Kumar et al., 2018; Heintz et al., 2019). Along with the sex-specific pathways, our study showed shared liver pathways between male and female offspring such as oxidative phosphorylation, PPAR signaling pathways, and fatty acid metabolism (Fig. 6). Despite these shared pathways, we did not identify consistencies between sexes at the phenotypic level. As liver is not the only tissue involved in metabolism, it is possible that interactions of these liver pathways with sex-specific pathways in other tissues not examined here contribute to unique metabolic phenotypes between sexes.

To put our results from low-dose prenatal exposure in the broader context of previous findings and to explore molecular insights into the non-monotomic effects of BPA, we compared the male liver gene expression changes from our low-dose prenatal exposure study with previous liver transcriptomic studies of different doses and exposure windows. We previously examined the liver transcriptome of male offspring exposed to a high-dose of BPA (5 mg/kg/day) in utero (Shu et al., 2019). There were 114 genes in common between the two prenatal studies of different dosages (Supplementary Fig. 1). Pathways such as drug metabolism, PPAR signaling, and fatty acid metabolism were some of the common pathways identified between studies. Pathways related to oxidative phosphorylation and VEGF signaling were unique for the current low-dose study, while steroid hormone biosynthesis and glycolysis/gluconeogenesis were unique for the high-dose study. We also compared the male DEGs in the present study to a previous adult BPA exposure study done at 50 μg/kg/day dose in male mice (Marmugi et al., 2012). When comparing all the DEGs regardless of the direction, there were several genes shared between the two studies such as Sc5d, Acacb, Slc22a5 and Erbb2ip. Moreover, two DEGs (Slc22a5 and Sc5d) were shared across all three studies across dosages and exposure windows. Slc22a5 plays a role in carnitine transport, which is important for fatty acid metabolism (Longo et al., 2016). Sc5d is gene that encodes for an enzyme involved in cholesterol synthesis (Krakowiak et al., 2003). These pathways and genes consistently affected by BPA exposure at different dosages and exposure windows point to the core processes underlying the cardiometabolic effects of BPA exposure. Overall, the phenotypic and molecular alterations induced by BPA are complex and are driven by differences in experimental design, dose and sex.

To provide insights into potential regulators of liver transcriptomic and pathway alterations, we took several complementary approaches. We first used different network modeling approaches, including TF networks and Bayesian gene regulatory networks, to predict key regulators. The TF-based analysis revealed unique TFs in male (e.g. Tfeb and Egr4) and female (e.g. Srebf2 and Rora) offspring as well as shared TFs (e. g. Arnt and Zfp143). The BN approach also identified unique KDs in male (e.g. Apoa2 and Ttr) and females (e.g. Sult1a1 and F2) as well as common KDs (e.g. Malat1 and Dcxr) (Fig. 6). These network regulators could mediate the BPA effects on downstream genes and pathways in the liver. We then examined the gut microbiome composition of the offspring at multiple ages. We found certain bacteria to be consistently altered at all time points, whereas, others were only altered at one time point measured (Fig. 6). This shows the dynamic nature of the gut microbiome and the importance of evaluating microbial changes at multiple time points. BPA-exposed offspring had marked changes in the relative abundances of different groups within the Firmicutes phylum, the order Clostridiales, and families Lachnospiraceae, Ruminococcaceae, Erysipelotrichaceae, and Lactobaciliaceae. Previous BPA studies done on various doses have also revealed a change in the relative abundance of bacteria in the Firmicutes families, such as Ruminococcaceae, and Lachnospiraceae, which produce short-chain fatty acids (SCFAs) (Lai et al., 2016; Xu et al., 2019). A shift in bacterial composition from Bacteroidetes to Firmicutes has been previously identified as a marker for cardiometabolic risk (Rosenfeld, 2017). A fall in Bifidobacteria and certain Clostridia groups have also been observed in response to both perinatal and direct BPA exposure and decreases in these bacteria are linked to obesity and diabetes (Larsen et al., 2010; Lai et al., 2016; Malaise et al., 2017). In our current study, we also observed a decrease in the Bifidobacteriaceae and Clostridiaceae families at the early time points (weeks 11 and 14) but an increase at the latest time point (week 20). Given that the rise of these families occurred at the latest time point, it may be of interest to examine how their dynamic fluctuations play a role in occurrence of metabolic phenotypes in future studies.

By evaluating multiomics data together from the same animals after in utero low-dose BPA exposure, our study allows direct integration and assessment of the relationships between the multidimensional data, a significant departure from previous studies that investigate each aspect separately. In our study, we identified specific bacterial classes that are significantly influenced by prenatal low-dose BPA exposure and correlated with metabolic phenotypes (Fig. 6). In females, we identified several bacteria belonging to the clostridia class and S24–7 family to be correlated to body weight. In contrast, in males, the bacterial abundance of a member of the S24–7 was correlated with glucose tolerance. These correlated bacteria are important for maintenance of gut homeostasis. Interestingly, abundances of bacteria from the S24–7 family have been reported to be altered in obese models, during liver regeneration, disease remission, and in a myriad of other gut-microbiome responses (Liu et al., 2016; Ormerod et al., 2016; Lagkouvardos et al., 2019). Additionally, the S24–7 family have also been implicated in providing oxidative stress protection (Ormerod et al., 2016). Due to the challenges in characterizing and culturing the S24–7 family, it is difficult to directly test their role in cardiometabolic disease pathogenesis.

Our data also uniquely identified several BPA-responsive bacteria to be correlated to liver DEGs. Pathway enrichment analysis of the DEGs correlated to the BPA-responsive bacteria were enriched for core metabolic pathways such as oxidative phosphorylation, PPAR signaling pathways and fatty acid metabolism (Fig. 6). These results support potential interactions between gut microbiome and liver upon BPA exposure. Several of the important metabolic DEGs were correlated to specific bacteria. In females, the abundance of Coprococcus was correlated to the expression of Acadl and KD Malat1. Coprococcus belongs to the family Lachnospiraceae and is involved in the production of SCFAs (Louis et al., 2004; Reichardt et al., 2014; Koh et al., 2016). Acadl is a gene that encodes for a member of the acyl-CoA dehydrogenase family, which are enzymes that play a role in energy metabolism (Kurtz et al., 1998). Malat1 is a long noncoding RNA that regulates genes involved in cancer progression and diabetes (Zhang et al., 2017). In males, the abundance of a member of the S24–7 family was correlated to two KDs Acadm and Apoa2 (Fig. 6). Acadm is a gene that encodes an enzyme that metabolizes fatty acids for energy production. Loss of function in the enzyme produced by Acadm results in hypoglycemia (Grosse et al., 2006; Lim et al., 2018). Apoa2 is gene that encodes for an abundant HDL protein and has a regulatory role in lipid metabolism and atherosclerosis development (Warden et al., 1993). The exact role these bacteria play in regulating the host liver genes and KDs induced by BPA needs to be further studied.

The data presented here in our multiomics study represents the discovery phase of a comprehensive investigation of in vivo BPA activities, which opens new opportunities for future testing of the specific findings. However, we acknowledge the following limitations. First, we did not conduct qPCR validation of the RNAseq-based findings. Although using low-throughput qPCRs to validate RNAseq data was a common practice during the early years of RNAseq studies, over the past decades sufficient evidence has been raised to support the accuracy of RNAseq measurements. In a systematic technical comparison study, the agreement between RNAseq and qRT-PCR was found to be >80% (SEQC/MAQC-III Consortium, 2014), which is consistent with our own findings in several previous studies (Chen et al., 2016; Meng et al., 2016; Meng et al., 2017; Shu et al., 2019). In addition, many DEGs that were significant in our data have also been observed in previous work which provides additional validation to the consistency across completely independent studies. Moreover, our study mainly focuses on identifying the pathways and networks perturbed by BPA instead of the accuracy of individual DEGs. Additional limitations to our study include PCR biases that exist in 16S sequencing, food and water intake measurements were not evaluated which can be affected by BPA exposure (Angle et al., 2013; Laursen et al., 2017; Nadal et al., 2017), and no BPA residue was measured. We also acknowledge that our study mainly identifies correlated molecular factors, and the causality of the altered microbiome and liver genes require future functional studies.

5. Conclusion

In summary, our systematic investigation of low-dose BPA prenatal exposure on a range of metabolic phenotypes, the liver transcriptome, and gut microbiota help further elucidate the gut-liver axis involved in metabolic reprogramming. Our data highlights sex-specific and shared pathways affected by BPA in the liver of male and female mice exposed in utero. In the same set of mice, we also evaluated the microbiome at multiple time points and identified several bacteria that potentially play a key role in interacting with liver metabolic pathways to regulate metabolic phenotypes induced by BPA. However, future studies are needed to further evaluate how dysbiosis of the gut microbial community can affect the host genome, causing the reprogramming of gene networks in critical metabolic tissues to promote metabolic disease in offspring. Such information may help guide public health policies and help devise novel preventative and therapeutic avenues that engage gut microbiome modulation to counteract environmentally induced metabolic diseases.