Connections Between the Gut Microbiome and Metabolic Hormones in Early Pregnancy in Overweight and Obese Women Free

The increasing prevalence of maternal obesity and its subsequent health outcomes are a significant public health concern and a major challenge for obstetrics practice. In early pregnancy, overweight and obese women are at an increased risk of metabolic complications that affect placental and embryonic development (1). Metabolic adjustments, such as a decline in insulin sensitivity and an increase in nutrient absorption, are necessary to support a healthy pregnancy (2,3); however, these metabolic changes occur on top of preexisting higher levels of insulin resistance in overweight and obese pregnant women, increasing the risk of overt hyperglycemia because of a lack of sufficient insulin secretion to compensate for the increased insulin resistance (3).

The potential role of the gut microbiome (the composite of the bacteria present in the gastrointestinal tract) in pregnancy has become the subject of considerable interest. In normal pregnancy, the maternal gut microbiota changes from first to third trimester with a decline in butyrate-producing bacteria and an increase in Bifidobacteria, Proteobacteria, and lactic acid–producing bacteria. Inflammation and weight gain that occurs during pregnancy might be the result of microbe-driven processes to increase energy supply for the fetus (4). These alterations might also be linked with the maternal metabolic profile and thereby contribute to the development of pregnancy complications (5,6) as well as affect the metabolic and immunological health of the offspring (7). In summation, modifications in the metabolic hormone milieu during gestation are proposed to be linked with changes in the maternal microbiota; however, no studies have reported a possible maternal hormonal-microbial interaction to date.

The aim of this study was to evaluate the relationships between gut microbiome composition and the metabolic hormonal milieu in overweight and obese pregnant women in early gestation (<16 weeks). To address this relationship, fecal microbiota profiles as assessed by 16S rRNA gene amplicon sequencing were correlated with the serum concentrations of nine metabolic hormones involved in glucose and energy metabolism.

This study included a subset of 29 overweight and 41 obese pregnant women from the SPRING (Study of Probiotics in the Prevention of Gestational Diabetes Mellitus) cohort (8). Women with known preexisting diabetes, impaired fasting glucose, or impaired glucose tolerance and/or early-onset gestational diabetes mellitus (GDM) were excluded. The women in this subset were selected based on completion of the pregnancy with two stool samples collected. Of the women meeting these criteria, all with GDM at 28 weeks' gestation were included and those with normoglycemia were matched for BMI, maternal age, and ethnicity to encompass a broad spectrum of metabolic health. Clinical characteristics and biological samples at <16 weeks' gestation are summarized in Table 1. Fasting blood glucose, HbA_1c, triglyceride, total cholesterol, HDL, LDL, and VLDL levels were analyzed immediately after collection, with further serum aliquots stored at −80°C. HOMA of insulin resistance (HOMA-IR) was calculated for each participant. Refrigerated fecal samples were self-collected by each participant at home and stored within 1 day after collection at −80°C until DNA extraction.

Stored serum samples were used to measure insulin, C-peptide, glucagon, gastrointestinal polypeptide (GIP), GLP-1, ghrelin, leptin, resistin, and visfatin by using the Bio-Plex Pro human diabetes immunoassay (Bio-Rad Laboratories, Hercules, CA) at room temperature according to the manufacturer’s instructions. Serum samples were allowed to thaw overnight at 4°C. Fifty microliters of undiluted serum were loaded in each well in duplicate. Standard curves were generated for all hormones. Hormone concentrations were collected and analyzed by using a Bio-Rad Bio-Plex 200 instrument equipped with Bio-Plex Manager 6.1 software (Bio-Rad Laboratories).

The repeated bead beating and column method (9) followed by QIAGEN AllPrep DNA extraction kit were used for microbial DNA extraction from 0.25 g of thawed stool samples. Mechanical disruption was achieved using sterile zirconia beads (0.1 and 0.5 mm diameter) with homogenization for 3 min at maximum speed (30 Hz) in the TissueLyser II (QIAGEN). DNA concentration was determined with a NanoDrop ND-1000 spectrophotometer (NanoDrop Technologies).

Fecal microbiome profiles were assessed by 16S rRNA gene amplicon sequencing by using the Illumina MiSeq system at The University of Queensland Australian Centre for Ecogenomics. Bacterial 16S rRNA gene sequences covering variable regions (V6–V8) were PCR amplified from purified genomic DNA by using the primers 926F (5′-TCG TCG GCA GCG TCA GAT GTG TAT AAG AGA CAG AAA CTY AAA KGA ATT GRC GG-3′) and 1392R (5′-GTC TCG TGG GCT CGG AGA TGT GTA TAA GAG ACA GAC GGG CGG TGW GTR C-3′) with overhang adapters, generating a 500-base pair product. In each PCR run, a positive control (Escherichia coli JM109) and negative control (sterile deionized water) were added. The V6–V8 PCR amplicons were cleaned using AMPure XP beads to remove unbound primers, primer dimer species, and nucleotides. Thirty-two self-correction bar codes were added to the primers using Nextera XT Index Kit followed by a second PCR cleanup step. Library quantification, normalization, and pooling were performed according to the manufacturer’s instructions.

Data files were processed using QIIME (Quantitative Insights Into Microbial Ecology) version 1.9.1 software (www.qiime.org). Forward and reverse sequence reads for each individual sample were demultiplexed, quality filtered, and joined (10). Sequences were assigned to operational taxonomic units (OTUs) with a pairwise identity threshold of 97% by using the GG (Greengenes) reference database (11). Sequences were mapped, yielding an OTU table, and a phylogenetic tree was constructed using FastTree (12), which was then used to generate unweighted and weighted UniFrac distance metrics. Any OTUs with overall relative abundance <0.0001 were excluded from further analysis. A total of 2,086,048 reads representing 787 OTUs were retained (median 26,049 reads/sample, range 3,180–88,127). To standardize sequence reads across samples, data were randomly rarefied to 3,000 sequences per sample. The relative abundances of OTUs were summarized across various taxonomic levels using default settings in QIIME.

Pregnant women were classified according to BMI cutoff values of 25.0–30.0 kg/m² (overweight) and >30.0 kg/m² (obese). Normally distributed variables are reported as mean ± SEM; otherwise, median with interquartile range is reported (Table 1). Comparisons between the two BMI categories were performed with the Mann-Whitney U test. Correlations between hormonal and anthropometric variables were performed with the Spearman correlation coefficient test. Statistical analysis of the demographic and clinical data were performed using GraphPad Prism 6 software. P < 0.05 was considered statistically significant.

Gut microbiota diversity within samples from each individual (α-diversity) was assessed by Chao1 index, which estimates the number of different taxa, and by Shannon diversity index, which evaluates the richness and evenness of taxa. High Shannon diversity values range from 2.5 to 3.0. Significant differences in microbial community composition between overweight and obese women or future GDM status (β-diversity) were calculated by permutational MANOVA (pMANOVA) of weighted UniFrac distances (Bray-Curtis distance method) with 999 permutations and illustrated by principal component analysis. The impact of maternal age and gestational age on microbiome composition was evaluated by redundancy analysis. For the hypothesis-driven analysis, canonical correspondence analysis (CCA) with permutation tests was computed and visualized by Calypso software (http://bioinfo.qimr.edu.au/calypso). Associations between the metadata (BMI category, hormonal measurements, glucose levels, HbA_1c, and clinical data) and taxa abundance were evaluated by bootstrapped Spearman rank correlation coefficient using 1,000 permutations. For all analyses, correction for multiple testing by false discovery rate (FDR) with the Benjamini-Hochberg procedure was done in QIIME, with values of <0.05 considered statistically significant.

The clinical characteristics of the pregnant women included in this substudy at <16 weeks' gestation are presented in Table 1. Seventy healthy pregnant women were grouped according to their BMI and matched for GDM status, maternal age, and ethnicity. Fasting serum levels of nine metabolic hormones (insulin, C-peptide, glucagon, GIP, GLP-1, ghrelin, leptin, resistin, and visfatin) were measured in all participants. Detectable hormone levels were present in all participants except glucagon and visfatin, which were detectable in 50% and 27%, respectively. Visfatin was not included in further analyses. There were significant differences in BMI, fasting glucose, HOMA-IR, insulin, C-peptide, leptin, GIP, and resistin between the two groups (Table 1), with obese women having a more disturbed metabolic profile. Fasting serum glucose was positively correlated with insulin and C-peptide concentrations (Table 2). HbA_1c was not correlated with fasting serum glucose or any of the nine metabolic hormones measured.

Stool samples were collected from all women to assess whether the composition of the gut microbiota was associated with maternal BMI. The ratio between the main phyla Firmicutes to Bacteroidetes was 3:1, with obese women tending to have a slightly higher abundance of Firmicutes (69.0% [59.0–81.4%]) than overweight women (64.2% [57.6–72.9%], P = 0.107). The relative abundances of two other dominant gut phyla tended to differ between overweight and obese women at 16 weeks' gestation (Fig. 1A) with a higher relative abundance of Tenericutes in the overweight group (P < 0.05) and higher abundance of Actinobacteria in the obese group (P = 0.084). At the family level, Lachnospiraceae (ρ = 0.42, P < 0.0001) and Rikenellaceae (ρ = 0.28, P < 0.0001) were positively correlated with maternal BMI. Microbiota composition was not significantly affected by either maternal age (P = 0.705) or gestational age (P = 0.176). Furthermore, the microbiome composition at 16 weeks' gestation was not different between women in whom GDM did or did not develop at 28 weeks' gestation (P = 0.200) (Supplementary Fig. 1).

We observed that overweight pregnant women tended to have a higher microbial richness (number of OTUs) and evenness (relative prevalence of the various OTUs within the gut) than obese pregnant women (Shannon index P = 0.098) (Fig. 1B); however, there was no difference in the total number of different taxa between the groups (Supplementary Fig. 2, Chao1 index). Examination of the bacterial community structure between the two groups as measured by β-diversity, principal component analysis (Supplementary Fig. 3), and pMANOVA showed no significant differences between overweight and obese women (pMANOVA P = 0.604); however, the hypothesis-based CCA showed that BMI was significantly associated with gut microbiota composition at the phylum level in early pregnancy (P = 0.009) (Fig. 1C).

Insulin, C-peptide, HOMA-IR, and fasting glucose were all significantly and positively correlated with increasing BMI (Table 2). To assess associations between maternal glucose metabolism and gut microbiota, insulin, C-peptide, HOMA-IR, and fasting glucose levels were correlated with intestinal bacteria abundance (Table 3). At the phylum level, insulin levels and HOMA-IR were positively correlated with the relative abundance of Actinobacteria and negatively correlated with Tenericutes, resembling the trend observed with increasing BMI. Here, Actinobacteria was primarily composed of two main families: Bifidobacteriaceae and Coriobacteriaceae. Women with increased serum levels of insulin and C-peptide exhibited a higher abundance of Coriobacteriaceae (P < 0.0001) but a similar abundance of Bifidobacteriaceae. The ratio of Bifidobacteriaceae to Coriobacteriaceae was 1:2. Further analysis within Coriobacteriaceae indicated that abundance of the genus Collinsella was positively associated with insulin (P = 0.003), HOMA-IR (P = 0.012), and C-peptide (P = 0.020) levels and different between overweight and obese pregnant women (P = 0.027) (Fig. 2); however, Collinsella abundance was not correlated with fasting glucose (P = 0.783) or BMI (P = 0.340) after correction for multiple testing. Furthermore, the abundance of the family Ruminococcaceae from the phylum Firmicutes was enriched in women with higher C-peptide concentrations (ρ = 0.41, P < 0.0001). A positive correlation between the Ruminococcaceae and insulin levels was observed from the bootstrapped Spearman rank correlation coefficient produced using 100 permutations (ρ = 0.28, P < 0.0001) but not with 1,000 permutations (ρ = 0.26, P > 0.05).

The relationship between the gut microbiome composition and incretin hormone levels was also assessed. Fasting serum GIP concentrations were negatively correlated with BMI in early pregnancy. Higher fasting GIP levels were correlated with a higher abundance of the genus Coprococcus (family Lachnospiraceae) (ρ = 0.47, P < 0.0001) but a lower abundance of Ruminococcaceae (ρ = −0.39, P < 0.0001), with both families belonging to phylum Firmicutes. There were no significant associations of GLP-1 with the gut microbiota and maternal BMI.

To further investigate the relationship between maternal gut microbiota and metabolism, correlations to serum adipokine levels were assessed (Table 3). Leptin was found to be positively associated with BMI (Table 2) and fasting glucose. Higher leptin levels were strongly correlated with Firmicutes family members Lachnospiraceae (ρ = 0.45, P < 0.0001) (Fig. 3) and Ruminococcaceae (ρ = 0.41, P < 0.0001). In contrast to leptin, resistin concentrations were negatively correlated with maternal BMI (Table 2) and Ruminococcaceae (ρ = −0.40, P < 0.0001) (Table 3). Ghrelin, the hunger hormone, exhibited a trend for negative correlation with maternal BMI in early pregnancy (Table 2). Of note, ghrelin levels were positively correlated with the family Bacteroidaceae (ρ = 0.33, P < 0.0001) and negatively with the family Prevotellaceae (ρ = −0.34, P < 0.0001) (Table 3), with both families belonging to the phylum Bacteroidetes.

The genus Collinsella, which was positively correlated to insulin levels, was also positively correlated with maternal triglycerides (ρ = 0.31, P = 0.009) and VLDL cholesterol levels (ρ = 0.33, P = 0.006) (Fig. 2C). In contrast, Collinsella exhibited a trend for negative correlation with maternal HDL cholesterol (ρ = −0.31, P = 0.09) (Supplementary Fig. 4). No associations were found with LDL cholesterol (P = 0.383).

This study is the first in our knowledge to report a relationship between maternal metabolic hormonal milieu and gut microbiome in early pregnancy. In obese pregnant women, the concentrations of fasting glucose, insulin, and leptin at 16 weeks' gestation resemble those in the third trimester of nonobese, nondiabetic pregnant women (13,14). Moreover, obesity is associated with a higher proportion of the major bacterial phylum Actinobacteria and a lower proportion of the phylum Tenericutes.

This study showed a positive correlation of the genus Collinsella (family Coriobacteriaceae, phylum Actinobacteria) with insulin, C-peptide, and HOMA-IR in pregnancy. Collinsella is well suited to colonize mucosal surfaces, metabolizes amino acids, and may directly interact with the host (15). Outside pregnancy, higher abundance of Collinsella has been reported in type 2 diabetes (16,17), which is reduced by weight loss (18). Whether limiting gestational weight gain in pregnancy lowers the abundance of Collinsella and whether this could contribute to lowering insulin resistance is not clear.

Incretins GIP and GLP-1 stimulate insulin secretion and are secreted by cells in the gastrointestinal tract. Levels of the incretin hormone GIP are positively associated with the abundance of the genus Coprococcus. Coprococci are considered to be butyrate producers (19). Butyrate is an important short-chain fatty acid (SCFA) that can serve as a second messenger as well as a source of energy (20). Oral administration of sodium butyrate in mice significantly increased GLP-1 and GIP levels (21), supporting the importance of butyrate in the stimulation of gut hormones.

In the current study, maternal triglycerides and VLDL cholesterol were positively correlated with the genus Collinsella. Outside pregnancy, Collinsella abundance has been positively correlated with serum cholesterol (22) and bile acid deconjugation (15). Additionally, in infants, a slower increase in Collinsella and Bifidobacterium abundance is associated with lower body fat at a later age (23). Thus, there is a possible link between Collinsella abundance and fat metabolism both inside and outside pregnancy. This link is further supported in rodent models, where Collinsella abundance is strongly associated with non–HDL cholesterol (24) and hepatic triglycerides (25). Therefore, manipulating the abundance of Collinsella may affect both lipid and glucose metabolism in pregnancy.

Leptin levels were positively correlated with the abundance of the families Lachnospiraceae and Ruminococcaceae. These families are dominant species in the human gut. Some members are reportedly responsible for breaking down indigestible polysaccharides (i.e., dietary fiber) into SCFAs such as butyrate; however, an abnormal increase in these taxa could lead to an excessive calorie uptake by the host through increased availability of SCFAs, contributing to obesity (26). Leptin synthesis and secretion are increased from early gestation mainly due to placental production (3). Leptin opposes ghrelin activity. Ghrelin is a fast-acting hormone that stimulates food intake (27) and was higher in overweight women than in obese women. Ghrelin is positively correlated with the abundance of Bacteroidaceae but negatively with Prevotellaceae. Bacteria belonging to family Bacteroidaceae are involved in carbohydrate fermentation and are underrepresented in nonpregnant obese subjects (28). Members of family Prevotellaceae are associated with a vegetarian diet and with chronic inflammatory conditions (29). When members of both families co-occur in the gut, one or the other predominates because they have an antagonistic relationship (29). In fact, an inverse association between Bacteroidetes and Prevotella has been reported with obesity and type 2 diabetes in nonpregnant individuals (30,31) and now in pregnancy.

In human studies outside pregnancy, progression of glucose intolerance is accompanied by changes in the proportion and diversity of the gut microbiota (16,32). An obese-type microbiome often is reported to be less diverse (33). Only limited information is available on the effect of maternal obesity on the shifts in fuel metabolism and changes in microbiota composition in pregnancy. Here, obese pregnant women tended to have a less diverse gut microbiome compared with the overweight group; however, no significant differences in bacterial richness and evenness between individuals (α-diversity) were detected. This is similar to the results reported by the only maternal microbiome sequencing study to date, where the α-diversity of first-trimester gut microbiota was independent of prepregnancy BMI (4); however, the differences in composition of the maternal microbiota in early pregnancy between groups of overweight and obese women (β-diversity) show that subtle differences may exist between these groups. To definitively establish whether the relationship between microbiome and BMI is linear, a cohort of normal-weight women should be investigated.

In early gestation, the maternal microbiome shows a predominance of Firmicutes over Bacteroidetes, similar to the pattern observed in obese nonpregnant subjects (34). Here, obese pregnant women have a predominance of Firmicutes over Bacteroidetes (ratio 3.1:1) compared with overweight pregnant women (ratio 2.7:1). Santacruz et al. (35) reported similar results in early gestation, with a higher abundance of Bacteroides in normal-weight compared with overweight pregnant women; however, Collado et al. (36) associated Bacteroides with excessive weight gain over the first trimester. Moreover, the enrichment in Firmicutes species has been correlated with an increased expression of key enzymes involved in polysaccharide digestion, through which more energy may be harvested from the same diet (37).

Here, the significantly decreased abundance of Tenericutes, which is a less-prominent phylum, in the obese group may be explained by a relative increase in phylum Actinobacteria and/or Firmicutes abundance. This pattern of increased abundance of Firmicutes and Actinobacteria with decreased abundance of Bacteroidetes has previously been reported in nonpregnant obese, insulin-resistant subjects (33,38,39). Moreover, consumption of a high-fat diet in rodents for 2 weeks reduces the relative abundance of Tenericutes (40). At the family level, the abundance of Lachnospiraceae, from phylum Firmicutes, and Rikenellaceae, from phylum Bacteroidetes, were positively correlated with BMI. Abundance of some members of the Lachnospiraceae family have been linked with obesity in both humans and mice (41). This may be due to their role in SCFA production (42). In addition, in obese and diabetic mice, the family Rikenellaceae is more abundant (43) and highly enriched in mice fed a high-fat diet (44), relating Rikenellaceae with the host’s phenotype and diet. Again, the patterns observed in early pregnancy mirror those outside pregnancy. Confirmation of these microbiome-hormonal associations in a larger cohort will be necessary.

No associations were found between GLP-1 and BMI or gut microbiota, even though these relationships have been well established outside pregnancy (45). The lack of association in this study could be due to samples being taken during the fasting state, whereas GLP-1 secretion is stimulated upon ingestion of glucose and rapidly cleared from the circulation. Also in the current study, total GLP-1 rather than active GLP-1 was measured.

The microbiome-hormonal correlations in the current study were assessed in overweight and obese pregnant women only. To definitively establish whether the relationship between microbiome and BMI is linear, a cohort of normal-weight women should be investigated. Confounding factors such as diet and antibiotic treatment may further affect both the metabolic profile and the gut microbiota composition (16). The current study excluded women treated with agents known to alter glucose metabolism, and antimicrobial treatment was carefully recorded.

It is not clear from the study whether the microbiome determines the hormone levels or vice versa or whether the regulation is in fact bidirectional. Animal models may help to establish this; however, modulation of the gut microbiota by the ingestion of prebiotics (as a food source for beneficial bacteria) and/or administration of antibiotics in nonpregnant subjects reportedly affects adiposity and metabolic hormone levels (45,46), indicating that gut microbiome composition affects hormonal levels. Proliferation of the probiotic bacteria Bifidobacterium species modulates inflammation in obese mice and increases incretin secretion, thereby decreasing insulin resistance (47). In pregnancy, probiotic supplementation in normal-weight pregnant women significantly reduces the incidence of GDM (48) and lowers maternal cholesterol (49) and triglyceride (50) levels. These data support the notion that gut microbiota modulation by selective prebiotics or probiotic bacteria constitutes an interesting target for modulating maternal metabolism. The current results suggest that suppression of Collinsella is an additional target for improving glucose homeostasis and lipid metabolism in early pregnancy; however, if probiotics belonging to the genus Bifidobacterium are administered during pregnancy, the ratio between Coriobacteriaceae and Bifidobacteriaceae may change in favor of Bifidobacterium species. This could possibly alleviate impairments in both glucose and lipid metabolism. The SPRING trial, in which overweight and obese pregnant women are supplemented with Lactobacillus rhamnosus GG and Bifidobacterium lactis BB12, may shed light on this (8).

This study brings clear insights into the relationships between microbiota and maternal metabolic regulatory hormones and their potential causal associations in overweight and obese pregnancy; however, as the gut microbiota composition changes throughout pregnancy, these microbial-hormonal associations need to be examined in the third trimester to provide insights into whether the changes at 16 weeks are predictive for the progression to diabetes. Pregnancy is also accompanied by immune changes allowing for fetal survival and immune cells to interact constantly in the gut. The host’s immune system can also contribute to gut microbiota composition and thereby affect maternal hormones; however, no studies of the relationship among the immune system, the gut microbiota, and metabolism in pregnancy currently exist.

In summary, a high abundance of family Ruminococcaceae in early pregnancy may be related to adverse metabolic health. In addition, a high abundance of the family Coriobacteriaceae and genus Collinsella could serve as markers of impaired glucose metabolism, and a higher abundance of the families Lachnospiraceae, Prevotellaceae, and Bacteroidaceae is associated with maternal energy metabolism.

This study shows that a relationship exists between the gut microbiome composition and the metabolic hormone milieu in early pregnancy, which could influence the metabolic health of the mother and infant. The study brings a better understanding of the composition of the gut microbiota in early obese pregnancy and suggests that proportion and diversity of certain microbiota members can serve as potential markers for a high risk in insulin resistance and abnormal lipid metabolism.

Acknowledgments. The authors thank the participants; the SPRING Trial Group (Katie Foxcroft, Anne Tremellen, Jacinta Tobin, Shelley Wilkinson, Chris McSweeney, Peter O’Rourke, and Barbara Lingwood); and Alicia Kang, Erin Shanahan, Naoki Fukuma (The University of Queensland Diamantina Institute), and Nicola Angel (Australian Centre for Ecogenomics, The University of Queensland) for expert technical advice.

Funding. L.F.G.-A. is supported by a Departamento Administrativo de Ciencia, Tecnología e Innovación scholarship and H.L.B. is supported by the Australian Diabetes Society Skip Martin Research Fellowship. The SPRING trial is supported by the National Health and Medical Research Council (grant 1028575) and a Royal Brisbane and Women’s Hospital Foundation grant.

Duality of Interest. The SPRING trial has received probiotics and placebo products from Chr. Hansen A/S. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. L.F.G.-A. contributed to the data research and analysis and writing and editing of the manuscript. H.L.B., H.D.M., and L.K.C. provided intellectual input and contributed to the editing of the manuscript. M.M. provided intellectual input and technical expertise and contributed to the editing of the manuscript. M.D.N. provided intellectual input and contributed to the research design and writing and editing of the manuscript. M.D.N. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.