|Home | About | Journals | Submit | Contact Us | Français|
Many of the immune and metabolic changes occurring during normal pregnancy also describe metabolic syndrome. Gut microbiota can cause symptoms of metabolic syndrome in non-pregnant hosts: To explore their role in pregnancy, here we characterized fecal bacteria of 91 pregnant women of varying pre-pregnancy BMIs and gestational diabetes status, and their infants. Similarities between infant-mother microbiotas increased with children’s age, and the infant microbiota was unaffected by mother health status. Gut microbiota changed dramatically from first (T1) to third (T3) trimesters, with vast expansion of diversity between mothers, an overall increase in Proteobacteria and Actinobacteria, and reduced richness. T3 stool showed strongest signs of inflammation and energy loss, however microbiome gene repertoires were constant between trimesters. When transferred to germ-free mice, T3 microbiota induced greater adiposity and insulin insensitivity compared to T1. Our findings indicate that host-microbial interactions impacting host metabolism can occur, and may be beneficial, in pregnancy.
Over the course of a normal, healthy pregnancy, the body undergoes substantial hormonal, immunological, and metabolic changes (Mor and Cardenas, 2010; Newbern and Freemark, 2011). Body fat increases early in pregnancy followed by reduced insulin sensitivity later in gestation (Barbour et al., 2007). Reduced insulin sensitivity has been correlated with changes in immune status in pregnancy, including elevated levels of circulating cytokines (e.g., TNF α and IL-6; Kirwan et al., 2002) that are thought to drive obesity-associated metabolic inflammation (Gregor and Hotamisligil, 2011). In contrast to the obese state where they are detrimental to long-term health, excess adiposity and loss of insulin sensitivity are beneficial in the context of a normal pregnancy as they support growth of the fetus and prepare the body for the energetic demands of lactation (Di Cianni et al., 2003; Lain and Catalano, 2007; Nelson et al., 2010).
The cause of reduced insulin sensitivity in pregnancy remains unclear. In the context of non-pregnant obesity, recent work suggests a role for gut microbiota in driving metabolic disease, including inflammation, weight gain and reduced insulin sensitivity (Cani et al., 2007; Vijay-Kumar et al., 2010). The gut microbiota is shaped by environmental factors, such as diet (Wu et al., 2011), host genetics (Spor et al., 2011), and the immune system, which in particular, can have profound effects on the composition of the gut microbiota (Salzman et al., 2010; Slack et al., 2009; Vijay-Kumar et al., 2010). In pregnancy, immunological changes occur at the placental interface to inhibit rejection of the fetus, while at the mother’s mucosal surfaces, elevated inflammatory responses often result in exacerbated bacterially-mediated diseases, such as vaginosis and gingivitis (Beigi et al., 2007; Straka, 2011). In the gut, bacterial load is reported to increase over the course of gestation (Collado et al., 2008), but a comprehensive view of how microbial diversity changes over the course of normal pregnancy is lacking. The contribution of intestinal host-microbial interactions in promoting weight gain and other metabolic changes in the context of pregnancy remains to be evaluated.
In the present study, we have characterized the changes in the gut microbiota that occur from the first (T1) to the third (T3) trimester of pregnancy, and assessed the potential of T1 and T3 microbiota to induce metabolic changes using germ-free (GF) mouse transfers. We provide evidence that the gut microbial community composition and structure are profoundly altered over the course of pregnancy. Furthermore, the T3 microbiota induces metabolic changes in GF recipient mice that are similar to aspects of metabolic syndrome. These changes are associated with metabolic disease in non-pregnant women and men, but may be beneficial in the context of a normal pregnancy.
To address how pregnancy alters the gut microbiome, we obtained stool samples, diet information, and clinical data for 91 pregnant women, previously recruited for a prospective, randomized mother-infant nutrition study in Finland (see Supplemental Information for details; Collado et al., 2008, 2010; Laitinen et al., 2009). Each pregnant woman donated stool during T1 (13.84±0.16 weeks) and T3 (33.72±0.12 weeks) of pregnancy, and a subset donated stool 1-month post-partum; Additionally, a stool sample was obtained from the womens’ infants at 1-month of age, and a subset was resampled at 6 months and 4 years of age. Prior to pregnancy, the majority of the women in the study had normal body weights, although a subset was either overweight, or obese (Table S1), and 15 women were diagnosed with gestational diabetes mellitus (GDM, Table S1). The womens’ diets at T1 and T3 were evaluated by nutritionists using 3-day food records; 16 of the women took probiotic supplements over the course of pregnancy, and 7 used antibiotics at either T1 or T2 (see Supplemental Information; Laitinen et al., 2009). Health markers (i.e., HOMA, GHbA1C1, insulin, and 4 others) and anthropometric measurement indicators of adiposity gains were obtained during clinical visits (Table 1). Overall, the diets of the women, including total energy intake, were unchanged between sampling times. From T1 to T3, the women gained adiposity, and had higher integrated levels of circulating glucose (i.e., higher GHbA1c1), greater circulating levels of leptin, insulin and cholesterol, and increased insulin resistance (i.e., significant changes in HOMA and QUICKI values, Table 1).
We employed a culture-independent approach to compare the gut microbial communities of women during pregnancy (T1, T3), postpartum, and of their children at the different ages. PCR was used to amplify the V1–V2 variable region of the 16S rRNA gene, and samples were multiplexed and pyrosequenced, followed by quality filtering and chimera checking (see Experimental Procedures), yielding 925,048 high quality 16S rRNA gene sequences (average per sample: 2,873 ± 156). We then clustered sequences into operational taxonomic units (OTUs, clustered at 97% pairwise sequence identity) and assigned taxonomies. We applied the UniFrac distance metric (Lozupone et al., 2005), which provides a measure of the evolutionary distance between microbiota (β-diversity), to assess pregnancy effects on between-individual variation in community composition. The weighted UniFrac analysis (sensitive to abundances of taxa) revealed a dramatic expansion of β-diversity with gestational age (Figure 1A), and the unweighted UniFrac analysis (sensitive to rarer taxa) showed a global shift in microbial community composition from T1 to T3 (Figure S1A). The magnitude of the change in β-diversity (weighted and unweighted UniFrac) from T1 to T3 was unrelated to pre-pregnancy BMI, GDM development or previous number of births (Figure 1B–D and S1B,C). Within individual women, we could not relate changes in β-diversity to their health status before or during pregnancy, nor to their use of probiotics or antibiotics during pregnancy (Table 1; Supplemental Information for additional analyses). Additionally, although we used the same techniques that have previously shown relationships between OTU abundances and components of the diet (Wu et al., 2011), we did not detect any significant relationships between aspects of the microbiota and our diet records either within or between trimesters (See Supplemental Information for details), which may reflect other methodological differences between these two studies. The lack of any correlations between covariates studied here and changes in β-diversity between trimesters raises the possibility that they may be related to immune or hormonal changes.
From T1 to T3, the relative abundances of Proteobacteria increased on average (T1: 0.73%±0.08, T3: 3.2%±0.68, p=0.0004) as did Actinobacteria (T1: 5.1%±0.47, T3: 9.3%±1.32, p=0.003, paired t-tests; Figure 2A; see Supplemental Folder for full taxonomic information by sample), and although these changes did not occur in all subjects, they occurred in 69.5% and 57% of women, respectively. Figure 1 indicates that the greatest component of the variation between samples (PC1, 33%) relates to the gradient of Bacteroidetes and Firmicutes abundances across samples (Figure 1E,F), and that the separation of T3 samples from T1 along PC2 reflects enrichment of Proteobacteria in many of the T3 samples (Figure 1G).
The number of OTUs was significantly reduced as individual women progressed from T1 to T3 (T1: 219±4.1, T3: 160±5.7, paired t-test p≤0.0001, note that enterotypes were not present within trimesters, see Supplemental Information). Similarly, T1 microbial communities had greater within-sample (α) phylogenetic diversity than T3 microbiota, regardless of pre-pregnancy BMI and health state (Fig 1H, Table S2). T1 samples also had significantly more even taxonomic distributions than T3 samples (Gini coefficients, Table S2). Together with β-diversity patterns these findings indicate that by T3, microbiota were depleted of bacterial phylogenetic diversity in ways that differed between individuals.
We used machine learning techniques to identify 29 OTUs whose relative abundance reliably discriminated T1 and T3 samples (clustering confidence > 80%, Figure 2B). Eighteen of these discriminatory OTUs were over-represented in T1 and belonged mostly to the Clostridiales order of the Firmicutes (e.g., butyrate producers, such as Faecalibacterium and Eubacterium; Figure 2B). OTUs that were over-represented in the T3 samples included members of the Enterobacteriaceae family and Streptococcus genus (Figure S2). No correlations were found between the abundances of specific OTUs (at any level of taxonomy) and the use of probiotics, antibiotics, number of previous births, health markers or the diet data (See Supplemental Information for details on the statistical tests employed to search for associations; Table 1). These results indicate changes in immunity and/or hormonal levels may also induce changes in phylogenetic content of the microbiota.
The large differences in β-diversity for T1 and T3 samples raised the question of which of these two sets of pregnancy samples was most similar to the non-pregnant state. To answer this question, we placed our data in the context of the Human Microbiome Project’s (HMP) recently generated healthy reference dataset of microbial diversity across the human body (Consortium, Accepted), which includes 16S rRNA gene sequences for 191 stool samples obtained from 98 men and 93 non-pregnant women. The HMP 16S rRNA gene sequence data consisted of 2 different region of the 16S rRNA gene (both V1–V2 and V3–V5), therefore we compared these sets to our data by picking OTUs against a common full-length reference set (Greengenes, see Experimental Procedures). A combined weighted UniFrac analysis showed clearly that the β-diversity of T1 is similar to HMP normal controls (Figure 3, S3A). In contrast, the T3 β-diversity is far higher than for T1 and HMP samples (Figure 3). The combined PCoA analysis of the UniFrac matrix shows a separation of T1 samples from HMP controls along PC1, reflecting differences in Bacteroidetes and Firmicutes content (Figure 1E,F). This indicates that the between-individual variation is similar for T1 and HMP samples, even though the community structure for these samples differs somewhat. Various factors may account for the compositional differences between the sample sets, but the difference between the pregnancy samples and the HMP samples is much larger than the difference between the HMP stool assayed with two different primer regions (Figure S3A), such that primer region is not an explanatory factor for this shift. However, in contrast to the HMP protocol, we lyophilized our samples prior to homogenization and DNA extraction, and our comparison of handling methods on a small subset of samples indicates handling may also account for part of the shift (Figure S3B,C). It is also highly likely that either the onset of pregnancy (i.e., hormonal or behavioral changes) induces a shift in composition reflected in PC1, and/or that the provenance of the samples (Finland versus USA) may be important, as geographical/cultural factors have been shown to impact gut microbial diversity (De Filippo et al., 2010; Yatsunenko et al., 2012).
We tested if the change in β-diversity from T1 to T3 was driven by samples obtained from women that had above-normal pre-pregnancy BMI’s or who developed GDM. Results showed that women who were overweight or obese prior to pregnancy, and women who developed GDM, also had a significant shift in β-diversity from T1 to T3 (weighted and unweighted UniFrac, Figure S3D,E). Removal of these subjects from the whole dataset showed that the shift from T1 to T3 also occurred in the healthy women alone (Figure S3D,E). These results strongly suggest that the expansion of β-diversity between women is a widely shared phenomenon driven by pregnancy regardless of health status.
We further observed that women who were obese prior to pregnancy had the lowest within-subject (α) diversity at both T1 and T3, although this was not significantly different from normal-weight women. In addition, GDM+ women tended to have the most depleted microbial richness at T1 (Table S2), although their microbiota did not differ in composition from those of matched controls (Figure 1C and S1C; no significant differences for OTU abundances, FDR of 0.05). Importantly, GDM did not impact negatively the microbiota of the children. Children of GDM+ mothers did not differ from children of GDM- mothers in terms of their microbiota’s α-diversity, Gini coefficients, or OTU abundances (FDR of 0.05). These results suggest that although a low phylogenetic diversity may be a biomarker for GDM, this condition does not appear to negatively impact the microbiota of infants born to GDM+ mothers.
The high levels of between-individual variation in community composition observed in T3 persisted for women 1-month postpartum (Figures 1A,,4).4). We found that the relative abundance of the genus Streptococcus, which is significantly enriched in T3 and 1-month postpartum samples compared to T1, is in highest abundance in the 1-month olds (ANOVA for children’s data, p≤0.05, Figure S2). Additionally, infants aged 1 month and 6 months also had elevated levels of β-diversity, but by 4 years of age, children had levels of β-diversity similar to mothers at T1 (Figure 4). These results indicate that differences in gut microbiota between infants are higher than what is observed in non-pregnant adults, as previously reported (Koenig et al., 2011; Palmer et al., 2007). (It is important to note that the V1–V2 region primers used in this study are biased against Bifidobacteria (Kuczynski et al., 2012), an important component of the developing infant microbiota (Koenig et al., 2011), although this bias has been shown not to impact the diversity of other taxa (Sim et al., 2012).) Using UniFrac to measure microbiota distances between mother-infant pairs, we found that the T1 microbiota was more similar to the childrens’ microbiota at all ages than the T3 (Figure S4). Although infant/child microbiotas (at all ages) were not more similar to their own mothers’ microbiota compared to unrelated mothers’ microbiota (at T1), the similarity to their own mother was greatest for the 4-yr olds (p value=0.003, paired t-test). These patterns are consistent with observations that within-family similarities in microbiomes are observed for older children but not for infants (Turnbaugh et al., 2009a; Yatsunenko et al., 2012).
Gut microbial community composition has been linked to how efficiently energy can be extracted from components of the diet reaching the colon and undergoing bacterial fermentation (Jumpertz et al., 2011; Turnbaugh et al., 2006, 2009a,b). Thus, we asked if changes in community structure could be related to energy loss in stool. Using bomb calorimetry, we measured a significant increase in stool energy content between trimesters within individual women (4.4 ± 0.6 vs. 4.7 ± 0.6 Kcal/gdw, p=0.002, paired t-test). This difference in stool energy content (i.e., ~10%) has been considered relevant to host adiposity in studies of obese and lean mice (Turnbaugh et al., 2006) and for altered microbiomes associated with excess nutrient load (Jumpertz et al., 2011). Here, however, these changes in stool energy content may not be related to diet or levels of food energy intake, because these remained constant from T1 to T3 (Table 1), but may be related to changes in host energy uptake or gut microbiota.
Previous studies have shown that a microbiome’s energy extraction efficiency from the diet is correlated with an enrichment of specific metabolic pathways, particularly those for carbohydrate transport and utilization (Turnbaugh et al., 2009a; Turnbaugh et al., 2006). To assess if this was the case for the T1 vs. T3 microbiomes, we performed a shotgun metagenomic analysis of T1 and T3 samples obtained from 10 mothers selected at random (Figure S5A) using the Illumina HiSeq 2000 (4.1e+07 ± 5.9e+06 sequences/sample; Table S3). The metagenome-based community composition matched the 16S rRNA-based phylogenetic profile (Figure S5B). Unlike patterns observed in obesity-associated microbiomes, this analysis did not reveal differences in the mean relative abundance of gene categories (clusters of orthologous groups, COGs) or metabolic pathways (KEGG) between trimesters (Figure 5A). This finding may reflect the similar abundances of the major phyla across trimesters: Levels of Bacteroidetes and Firmicutes, which can impact microbiome gene content (Turnbaugh et al., 2006), were not significantly different between trimesters (Figure 5B). It was interesting to note, however, that a network analysis of correlations between COG abundances across samples (using the maximal information-based nonparametric exploration (MINE) statistics; Reshef et al., 2011), indicated that the T1 functional network had a lower degree of random connectivity between functionally unrelated genes and a greater degree of modularity than the T3 network (modularity of 0.69 vs. 0.64, Figure S5C–H). Disease-associated microbiomes have recently been shown to consist of less modular metabolic networks compared to health-associated microbiomes (Greenblum et al., 2012). The loss of network modularity in T3 is likely related to reduced phylogenetic diversity and the more uneven distribution of taxa. Overall, the metagenomic analysis indicated that the shifts in microbiome during pregnancy are not associated with the functional changes previously observed in the context of obesity (Ley et al., 2005; Turnbaugh et al., 2006), and may not be linked directly to the energy content of the stool.
The higher average proportion of Proteobacteria in T3 microbiota (Figure 2A), including elevated levels of Enterobacteriaceae, raised the question of whether the T3 microbiota can induce a greater inflammatory response in the host compared to T1 microbiota, as Proteobacteria are often associated with inflammatory conditions (Mukhopadhya et al, 2012). To address this question we first measured levels of cytokines in T1 and T3 stool (stool cytokine levels can be biomarkers for inflammation in the gut, Saiki et al., 1998). Levels of the pro-inflammatory cytokines IFNγ IL2, IL6, and TNFα were significantly higher in T3 than T1 (Tukey HSD, p≤0.05, p≤0.001, p≤0.001 and p≤0.005 respectively, Tables 1 and S4). Although pregnancy is associated with anti-inflammatory conditions at the placental interface (Mor and Cardenas, 2010), our data suggest that the T3 mucosal surfaces of the gastrointestinal tract present low-grade inflammation.
A powerful approach to investigate whether changes in the microbiota are a cause or a consequence of greater levels of inflammation is to transfer microbiota to GF wildtype recipient mice, which can be colonized with human microbiota in a manner that maintains the complex communities of the original donor samples (Turnbaugh et al., 2009b). To investigate the potential of the pregnancy-associated microbiota to promote inflammation, we transferred T1 and T3 microbiota into female GF wildtype Swiss-Webster mice. T1 and T3 inocula were created from pooled samples derived from T1 and T3 samples of 5 healthy-weight women chosen at random without a-priori knowledge of their microbial diversity profiles (these 5 were also used in the metagenomic analysis, Figure S5A). Post-hoc 16S rRNA gene sequence analysis of the donor samples and pooled inocula revealed that the donors all exhibited a consistent shift in diversity that was also captured by the pooled inocula (Figure S5A). To verify that differences in T1 and T3 microbiota observed in the donors were maintained in the recipient mice, we also sequenced 16S rRNA genes derived from mouse stool obtained 7 and 14 days post-transfer, and of cecal samples obtained day 15. This analysis showed that the shift between T1 and T3 microbiota observed in the donors (Figure S5A) was maintained in mice over the 2-week course of the experiment (Figure 6A, S6A,B).
The transfer of specific gut microbiota to otherwise healthy germ-free wildtype mice is sufficient to induce symptoms of metabolic syndrome, which, in addition to inflammation, include reduced insulin sensitivity and excess weight gain (Vijay-Kumar et al., 2010). Likewise, after 2 weeks, levels of inflammation markers were significantly higher overall in the stool and cecal samples from the T3 sample recipients compared to those of T1 recipients (ANOVA p≤0.001; Figures 6B, S6C–K, Table S5). Levels of lipocalin, which has recently been described as a sensitive marker of inflammation in mice (Carvalho et al., In Review), were also significantly higher in the T3 than T1 recipients (Table S5). Furthermore, we found that mouse recipients of the women’s T1 microbiota gained less adiposity compared to T3 recipients (37.9% ± 5.9 and 49.9% ± 4.4, for T1 and T3 respectively; p=0.06, one tailed t-test, Figure 6C), despite similar food consumption. Levels of insulin were slightly lower in T1 than T3 recipients after 2 weeks (0.266±0.017 vs. 0.281±0.066 ng/mL, n.s., respectively). Levels of blood glucose were slightly but significantly higher in T3 recipients after 30 minutes in an oral glucose tolerance test (Figure 6D). The observations that T3 recipients have reduced oral glucose tolerance, as well as greater inflammation and adiposity gains, than T1 recipients, together indicate that the T3 microbiota in particular has the capacity to induce metabolic changes in the host that resemble those occurring in both metabolic syndrome and pregnancy.
We describe a dramatic remodeling of the gut microbiota over the course of pregnancy. The first trimester gut microbiota are similar to one another and comparable to those of normal healthy controls, but shift substantially in phylogenetic composition and structure over the course of pregnancy. By the third trimester, the between-subject diversity has greatly expanded, even though within-subject diversity is reduced, and an enrichment of Proteobacteria and Actinobacteria is observed a majority of T3 samples. Furthermore, the abundances of health-related bacteria is impacted. For instance, Faecalibacterium, which is a butyrate-producer with anti-inflammatory effects that is depleted in inflammatory bowel disease (Sokol et al., 2008), is less abundant on average in T3. By the third trimester, each woman’s microbiota has diverged in ways that could not be predicted from the T1 composition, and were not associated with health status or our diet records. Nonetheless, in the majority of women, the shift from T1 to T3 includes an increase in the abundance of Proteobacteria, which has been observed repeatedly for inflammation-associated dysbioses (Mukhopadhya et al., 2012).
One of the questions raised by the observation of greater inter individual bacterial diversity and the decrease in bacterial richness in T3 and 1-month postpartum is that an aberrant microbiota might colonize the baby and contribute negatively to the shaping of the immune system from birth, with long-term consequences for health problems, such as allergy development (van Nimwegen et al., 2011). Nevertheless, we found that regardless of their age, the children’s microbiota were most similar to their mothers’ microbiota at T1, which may indicate that the taxa prevalent in T3 are at a selective disadvantage in the developing infant gut. Furthermore, we did not detect any differences between the microbiotas of GDM+ and GDM− mothers. We did observe an enrichment of Streptococcus in T3 and in post-partum samples on average, and highest levels for children were in the gut microbiomes of the 1-month olds (although it should be noted that many members of the Streptococcus are commensal). Such enrichments may serve to educate the developing immune system to important members of the microbiota. As was recently reported for children on 3 continents (Yatsunenko et al., 2012), similarities between the child and mother microbiota increased with the age of the children, which underscores the importance of shared diet and environment on shaping the microbiota (Koenig et al., 2011).
Metabolic syndrome is a range of phenotypes that increase an individual’s risk of developing type 2 diabetes, including hyperglycemia, insulin resistance, excess adiposity and low-grade inflammation (Tilg and Moschen, 2006; Vijay-Kumar et al., 2010). Similarly, the latter stages of pregnancy have been described as a diabetogenic state that maintains hyperglycemia in the mother and a continuous supply of nutrients to the fetus. Gains in adiposity also prepare the female body for the energetic demands of lactation. Elevated levels of circulating pro-inflammatory cytokines have been reported for late pregnancy and correlated with levels of insulin resistance, suggesting a possible mechanistic link (Mor and Cardenas, 2010). The women in our study had reduced insulin sensitivity, circulating blood glucose levels and adiposity over gestation, and in addition, we observed an increase in levels of inflammation markers in stool from T1 to T3. We suggest that a low-grade inflammation develops during pregnancy at the intestinal mucosal epithelium, and this inflammation may drive the microbial dysbiosis into a positive feedback loop with the altered host response (Lupp et al., 2007).
Two principal mechanisms have been proposed for how the gut microbiota can contribute to host adiposity: (i) increased energy extraction efficiency from the diet, and (ii) altered host-microbial interactions that promote metabolic inflammation. The results of our microbiota transfer experiments suggest that pregnancy is most similar to the second mechanism, in which a dysbiosis drives changes in metabolism. Our results are very similar to the recently described mouse model for metabolic syndrome in which the microbiota are sufficient and required to transfer aspects of metabolic syndrome to otherwise healthy germ-free wildtype recipient mice, including inflammation, excessive weight gain, hyperglycemia, and reduced insulin sensitivity (Vijay-Kumar et al., 2010).
The dysbiosis observed in T3 and reported for the mouse model of metabolic syndrome (Carvalho et al., In Review; Vijay-Kumar et al., 2010) are also strikingly similar: Both scenarios are characterized by elevated levels of Proteobacteria, greater between-individual variation, and excess bacterial load (described by Collado et al., 2008). Proteobacteria are active participants in inflammatory bowel disease (Mukhopadhya et al., 2012), and indeed colonization with just one member of this group (Escherichia coli) is sufficient to induce macrophage infiltration into white adipose tissue and impaired glucose and insulin tolerance in GF mice (Caesar et al., 2012). Not all women showed elevated levels of Proteobacteria in T3, however, indicating that other factors, such as other members of the microbiota, and potentially gene expression profiles, are also likely to be important for promoting inflammation. Although in the present study we pooled randomly selected donor microbiomes, comparison of individual donor effects on mouse phenotype will help identify the specific components of the microbiota driving metabolic inflammation. If the microbiota are not only sufficient but also required for metabolic changes in pregnancy, these components should be widely shared amongst women with normal pregnancies, and might share features with microbiomes of non-pregnant individuals of both sexes with metabolic syndrome.
It is interesting to note that some of the features of the T3 microbiota are similar to those of the obesity-associated microbiome shown to have enhanced energy extraction efficiency. For instance, both the low taxonomic richness and reduced metabolic network modularity that we observed in T3 have previously been reported for obese microbiomes (Greenblum et al., 2012; Qin et al., 2010; Turnbaugh et al., 2009a). In the T3 microbiome, the drivers of these traits are quite different from aspects of the obesity-associated microbiome. In the studies of obesity mentioned above, the microbiome is depleted in Bacteroidetes, such that gene categories related to simple sugar uptake, for instance, are over-represented in obese compared to lean microbiomes. Furthermore, excess energy intake has been shown to favor Firmicutes over Bacteroidetes (Jumpertz et al., 2011), and in obesity the microbiota have been exposed long term to excess energy intake. In T3 versus T1, the relative abundances of Bacteroidetes and Firmicutes are largely unchanged, and we see no shift in the abundances of specific gene functional categories or metabolic pathways. Additionally, in stark contrast to the obese microbiome, the T3 microbiome is associated with a greater amount of energy lost in stool compared to T1. Thus, although some of the features of the microbiome are shared between the obese and T3 microbiota, the underlying mechanisms by which they impact host adiposity can differ.
In summary, we have shown pregnancy to be associated with a profound alteration of the gut microbiota. The first trimester gut microbiota is similar in many aspects to that of healthy non-pregnant male and female controls, but by the third trimester, the structure and composition of the community resembles a disease-associated dysbiosis that differs among women. The underlying mechanisms resulting in the alteration of the microbiota remain to be clarified, but we speculate that the changes in the immune system at the mucosal surfaces in particular precipitate changes in the microbiota, although hormonal changes may also be important.
Dysbiosis, inflammation, and weight gain are features of metabolic syndrome, which increases the risk of type 2 diabetes in non-pregnant individuals. These same changes are central to normal pregnancy, where they may be highly beneficial, as they promote energy storage in fat tissue and provide for the growth of the fetus. Our work supports the emerging view that the gut microbiota affect host metabolism, however, the context (pregnant or not) defines how the outcome is interpreted (healthy or not). Metabolic changes are necessary to support a healthy pregnancy, which in itself is central to the fitness of a mammalian species. We hypothesize that in mammalian reproductive biology, the host can manipulate the gut microbiota to promote metabolic changes. Thus, the origins of host-microbial interactions that skew host metabolism towards greater insulin resistance, and which underlie much of the present-day obesity epidemic, may lie in reproductive biology.
Enrollment of human subjects, collection of samples, and clinical and biometric data were described previously (Laitinen et al., 2009). Samples were collected as previously described (Collado et al., 2008, 2010).
Bacterial 16S rRNA gene sequences (V1–V2 region) were generated from PCR amplicons that were multiplexed and pyrosequenced, and data analyzed using the QIIME software package (Caporaso et al., 2010a) as described in Supplemental Information.
We combined our data with the recently released HMP 16S rRNA (V3–V5 region) gene sequence data (Consortium, Accepted), and used a reference-based approach to pick OTUs at 97%ID using the Greengenes latest release (McDonald et al., 2011). We compared β-diversity using weighted UniFrac distances (Lozupone and Knight, 2005) calculated from the phylogenetic tree (Greengenes), after applying a rarefaction of 500 sequences/sample to standardize sequence counts.
Gross energy content of paired T1 and T3 samples (20 mothers chosen at random) was determined by bomb calorimetry using an IKA C2000 basic calorimeter system (Dairy One, Ithaca, NY).
Samples from 5 mothers chosen randomly, and the samples used as donors in the mouse transfer experiments, were selected for shotgun metagenomic sequencing using the Illumina HiSeq 2000. Sequence data were quality filtered and uploaded to MG-RAST. Taxonomy assignments (LCA), COG and KEGG relative abundance data for protein-coding reads were summarized using MG-RAST. Maximal information coefficient (MIC; (Reshef et al., 2011)) values were used to mine for between-COG ecological relationships within the two groups T1 and T3, accounting for linear as well as non-linear relationships. A conservative cutoff of MIC=1 was used to define between-COG edges in a network analysis of both T1 and T3 samples (MIC scores of 1 were well below p=0.05 based on a Bonferroni correction). See Extended Eperimental Procedures for details.
T1 and T3 stool samples from five women (age 24–30 years, normal pre-pregnancy BMIs) were used to colonize GF mice (n=6 for T1 and n=6 for T3). Adiposity was determined by DEXA as previously described (Backhed et al., 2004). Body weight and chow consumption were monitored weekly. Fecal pellets were collected at days 7 and 14. Oral glucose tolerance tests were performed by gavage with glucose (2 g/kg body weight) after a 4 h fast. At day 15, mice were sacrificed after measurements of total body fat content by DEXA, plasma insulin was measured and cecal content was removed. Body, gonadal white adipose tissue and cecum weights were recorded for each mouse.
Data are expressed as mean ± SEM. For complete statistical analysis methods see Supplemental Information.
We thank Mary-Claire King, Andrew Clark and Andrew Gewirtz for their contributions to the manuscript, and Daniel McDonlad and Nick Scalfone for technical assistance. This research was supported by The Hartwell Foundation, the NIH Human Microbiome Project DACC, the David and Lucile Packard Foundation, the Arnold and Mabel Beckman Foundation, the Cornell Center for Comparative Population Genomics, the Ragnar Söderberg Foundation and the Päivikki and Sakari Sohlberg Foundation.
Supplemental Information includes Extended Experimental Procedures and Analyses, six figures, and five tables.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.