Research ArticleMicrobiome

Antibiotics, birth mode, and diet shape microbiome maturation during early life

See allHide authors and affiliations

Science Translational Medicine  15 Jun 2016:
Vol. 8, Issue 343, pp. 343ra82
DOI: 10.1126/scitranslmed.aad7121

Snapshots of the developing infant gut microbiota

The intestinal “microbiota,” that is, the community of microbes inhabiting the human intestinal tract, undergoes many changes during the first 2 years of life. Bokulich et al. now show that this pattern of development is altered in children who are delivered by cesarean section, fed formula, or treated with antibiotics, compared to those babies who were born vaginally, breast-fed, or unexposed to antibiotics. Future studies will determine whether these disturbances influence the health of these babies.


Early childhood is a critical stage for the foundation and development of both the microbiome and host. Early-life antibiotic exposures, cesarean section, and formula feeding could disrupt microbiome establishment and adversely affect health later in life. We profiled microbial development during the first 2 years of life in a cohort of 43 U.S. infants and identified multiple disturbances associated with antibiotic exposures, cesarean section, and formula feeding. These exposures contributed to altered establishment of maternal bacteria, delayed microbiome development, and altered α-diversity. These findings illustrate the complexity of early-life microbiome development and its sensitivity to perturbation.


The establishment of stable microbial communities within the gastrointestinal tract closely parallels host growth and immune development during early life (1). The intestinal microbiome helps regulate host metabolism (2) and immune system development (3, 4) and thus could play an important role in directing host development. Delayed or altered establishment of the intestinal microbiota in childhood, termed microbiota immaturity, has been associated with diarrhea and malnutrition in Bangladeshi children (5). The causes of these microbiota disturbances and their consequences in other populations have not been established, but they may be linked to host development.

Antibiotic use during childhood is prevalent in most parts of the world, but the effects on maturation of microbiota and human health are poorly characterized (6). The average U.S. child receives about three courses of antibiotic by the age of 2 and 10 courses by the age of 10 (7). Antibiotics directly perturb the intestinal microbiota, leading to altered compositional states in children and adults (6, 8), but the consequences of these changes on host physiology are not well understood. Antibiotic exposure in children has been associated with increased risk of obesity (9), diabetes (10), inflammatory bowel disease (11), asthma (12), and allergies (13). We have shown previously that antibiotic exposure leads to increased adiposity in mice (14), that early-life exposures lead to prolonged effects on host metabolic characteristics (15, 16), and that the disturbed intestinal microbiota mediates these host effects (16).

Other disturbances, including birth mode and infant diet (17), also affect the intestinal microbiota during early life and are associated with later-in-life adiposity and other clinical effects. Cesarean delivery has been associated with asthma (18), allergies (19), type 1 diabetes (20), and obesity (21), possibly because of diminished exposure to maternal microbes during birth. Formula feeding similarly disrupts the intestinal microbiota (17) and may impair immune development (22) and normal metabolism (23).

Although the impacts of antibiotic exposures on intestinal dysbiosis in adults are well characterized, less attention has been given to their effects on microbiota development during early childhood (6). We hypothesized that antibiotics and other early disturbances may alter microbiome establishment during early life, potentially explaining associations with emerging health issues. We examined the intestinal microbiota to model its development over 2 years of life in a cohort of 43 healthy urban U.S. infants. We then assessed the effects of birth mode, infant nutrition, and antibiotic exposures on intestinal microbiota development in this population.


From the 53 mothers who initially enrolled in this study (table S1), a total of 43 infants were enrolled for follow-up until 2 years of age (table S2). Stool samples were collected from these infants; stool samples, vaginal swabs, and rectal swabs were collected from their mothers prepartum and postpartum. Birth mode, feeding, and systemic antibiotic exposures in the infants are shown in Table 1, and details of prenatal, perinatal, and postnatal antibiotic usage are shown in tables S3 to S5. The succession of bacterial taxa during the first 2 years of life followed a predictable pattern (Fig. 1, A to E), consistent with previous studies of the early-life microbiota (5, 24, 25). In the first month of life, stools were dominated by facultative aerobic Enterobacteriaceae before yielding to strict anaerobes—principally Bifidobacterium, Bacteroides, and Clostridium (Fig. 1A). These taxa were gradually displaced between months 6 and 24 by a diverse mixture of Clostridiales, roughly corresponding to the introduction and increased use of solid foods in these infants (Fig. 1F). However, even among infants who received no antibiotics in the first 6 months of life, those differing by birth mode and predominant diet showed substantial early differences (Fig. 1, B and E, fig. S1, and table S6). During the first 2 years of life, the microbiome was characterized by a period of gradual succession of different taxa. Although the infant microbiota began to resemble an adult’s microbiota at about 2 years of age (see below), it had not yet achieved an adult-like state, characterized by different alternative states that exist in quasi-equilibrium (26). Hence, we focused on the trajectory of microbiota development in children in the context of early disturbances.

Table 1. Characteristics of the 43 children in the study including systemic antibiotic exposure, delivery mode, and diet.
View this table:
Fig. 1. Microbial and dietary succession viewed over the first 2 years of life.

Mean relative abundance (RA) of fecal bacteria at the genus level at each month of life, for taxa with ≥1% mean RA across all samples. (A) All 43 infant subjects during the first 2 years of life. (B to E) The first 6 months of life for the 32 subjects who were not antibiotic-exposed, organized by delivery mode (vaginal or cesarean) and predominant feeding mode (breast or formula). Vaginal-breast, n = 15; cesarean-breast, n = 7; vaginal-formula, n = 3; cesarean-formula, n = 7. (F) Dietary trends in all infants across the study period.

Antibiotic exposures alter bacterial diversity

To assess the effects of disturbances on bacterial diversity, we measured α-diversity (that is, the species diversity in each individual sample), bacterial richness [total number of unique operational taxonomic units (OTUs)], phylogenetic diversity (the relative amount of diverse phylogenetic lineages detected), and species evenness in each sample (Fig. 2). Antibiotic use significantly diminished phylogenetic diversity immediately after birth (P < 0.0001), but diversity subsequently recovered during the first year of life to resemble that of infants not exposed to antibiotics (Fig. 2, A to C, and table S6). However, α-diversity was not significantly suppressed in individual children immediately after antibiotic administration (P > 0.05) (fig. S2). Thus, antibiotic exposure altered the trajectory of α-diversity changes during the first months of life, but transient effects were inconsistent.

Fig. 2. α-Diversity during the first 2 years of life in relation to antibiotic treatment, delivery mode, and predominant diet.

(A to I) Left: Mean phylogenetic diversity ± SEM. Middle: Mean observed OTUs ± SEM. Right: Mean Shannon equitability (evenness) ± SEM. α-Diversity is shown for antibiotic use (A to C), delivery mode (D to F), and diet (G to I). Asterisks indicate significant linear longitudinal model (LLM) (P < 0.05) group differences at baseline or rate-of-change differences across age ranges (dotted lines demarcate time periods tested).

β-Diversity, measuring similarities between samples as a function of microbial composition, allowed us to assess potential impacts on the composition and recovery of an entire microbial ecosystem. Infants who had not been exposed to antibiotics and those who had been exposed at any prior time differed significantly [weighted and unweighted UniFrac permutational multivariate analysis of variance (MANOVA), P < 0.001] (tables S7 and S8), although antibiotic exposure accounted for a small fraction of the overall variation among samples (R2 < 0.01). This observation provided evidence that antibiotics altered the intestinal microbiota in infants, but the effects were smaller than delivery mode (R2 = 0.02) and profoundly smaller than age (R2 = 0.14) (table S7). Even these factors accounted for a small fraction of community variation, and most of the variation could not be explained by age, delivery mode, diet, or antibiotics (R2 > 0.60) (tables S7 and S8), stressing the need for broader future studies that measure more factors that potentially contribute to microbiota development during early life.

Antibiotic exposure was associated with deficits in Clostridiales and Ruminococcus from 3 to 9 months of life, but with no consistent changes in other taxa (Fig. 3). Exposed subjects differed by age and by antibiotic class; thus, extensive variation in how the taxa were altered is not surprising. Drug type, timing, route, duration, underlying conditions, number of exposures, and differences between individual subjects all may be confounding factors.

Fig. 3. Antibiotic exposure alters bacterial abundance.

Antibiotic exposure significantly altered abundance of diverse bacterial taxa over the first 2 years of life. On the basis of LEfSe analysis, red-shaded taxa (rows) were significantly (P < 0.05) more abundant in antibiotic-exposed infants at the given time points (columns); blue shading indicates more abundant taxa in unexposed infants.

Antibiotic exposures delay microbiota maturation

During the first 2 to 3 years of life, intestinal microbiota undergoes a gradual succession, and although a large degree of interindividual variation occurs, these age-dependent succession patterns share many features in different human populations (24). Microbiota maturation, then, is defined as the rate at which a child’s microbiota develops, as measured by these age-dependent successional stages; a “mature” microbiota contains certain taxa that are markers for that child’s age group, whereas an “immature” or delayed microbiota resembles that of a younger child. Delayed microbiota maturation, as defined in healthy children, mirrors physiological disturbances in the host (5) and occurs in mice exposed to antibiotics (15). Thus, we examined whether antibiotic exposures and other disturbances similarly altered microbiota maturation in children, comparing relative maturation rates in a reference group (vaginally delivered, breast-fed, and unexposed to prenatal, perinatal, or postnatal antibiotics), and antibiotic-exposed subjects using a random forest (27) regression model to predict a child’s age as a function of their microbial composition, as reported previously (5). A defined microbiota maturation of the reference samples could be predicted using 22 key OTUs that explained the greatest degree of variation (pseudo-R2 = 82.1) in the model and, thus, are markers for normal development in this cohort (Fig. 4). Children exposed to antibiotics showed delayed microbiota maturation compared to those not exposed to antibiotics (Fig. 4A and table S9). These effects were most pronounced during months 6 to 12, and thereafter, no significant effect was observed. A delayed maturation pattern during early childhood was due to depletion of specific OTUs, including constituents of Lachnospiraceae and Erysipelotrichaceae (Fig. 4B). Within this cohort, intestinal bacterial communities followed a predictable pattern throughout the first 2 years (Fig. 1), but antimicrobials disturbed microbial succession at the OTU level, delaying microbial community development in the intestine relative to unexposed children (Fig. 4).

Fig. 4. Antibiotic exposure delays microbiota maturation during early life.

(A) Microbiota-by-age Z (MAZ) scores at each month of life between antibiotic-exposed and unexposed infants (infants never exposed to systemic pharmacologic antibiotic doses before the sampling time). MAZ scores indicate the number of SDs from the mean predicted age of age-matched control samples, as a function of microbiota maturation. Gray margins represent 95% confidence interval (CI). Asterisks indicate significant (LLM, P < 0.05) group differences at baseline or rate-of-change differences across age ranges (dotted lines demarcate time periods tested). The “unexposed” group contains both training set samples (from children who were never exposed to prenatal, perinatal, or postnatal systemic antibiotics; were vaginally delivered; and were dominantly breast-fed) and all other samples from children who had not been previously exposed to systemic postnatal antibiotics, regardless of delivery mode and diet. This accounts for the observed deviation from a 0 MAZ score in the unexposed group, as other factors influenced maturation. (B) OTU abundance heatmaps illustrate the RA Z scores of 22 maturation marker OTUs in the antibiotic-exposed and unexposed groups throughout the first 2 years of life. These OTUs were selected as those that best predict age of life in the control group and hence can be used as markers of normal maturation. Substantial departures from the normal maturation profile alter predicted age of other samples. The color scale represents RA Z scores for each OTU (that is, the number of SDs from the mean RA of that OTU) across all samples at that age. Red text indicates OTUs that appear most suppressed during months 6 to 12 after antibiotic exposure.

An important consequence of altered microbial patterns is changes to the functional gene repertoire present within the gut microbiome. We found substantial differences in the maturation of gene functions in the imputed metagenome (28), with relation to perturbations including antibiotics and predominant diet but not delivery mode (figs. S3 and S4 and table S10).

Delivery mode alters intestinal diversity

The first major microbial exposure for a vaginally born infant is in the birth canal, a potentially important event for establishing a healthy microbiome early in life. Cesarean section bypasses this exposure, altering the initial pool of microbes to which the neonate is exposed (29). We sought to investigate the impact of delivery mode on microbiota maturation and diversity during the first 2 years of life.

Compared to vaginally born infants, cesarean-delivered infants exhibited significantly greater (P < 0.05) phylogenetic diversity, richness, and evenness at baseline. However, these declined significantly in cesarean-born infants during the first month after birth, and cesarean-born children subsequently displayed lower diversity and richness up to 2 years of age, especially after 8 months of age (Fig. 2, D to F, and table S6). These effects were not due to differences in absolute bacterial abundance (fig. S5).

We also expected the comparative diversity between communities (β-diversity) to be altered in cesarean-delivered infants, reflecting both differing microbial exposures during birth and the altered α-diversity that was observed. For the babies’ first bowel movement (mean ± SD, 20.2 ± 18.3 hours of life), fecal β-diversity was not significantly different, suggesting that the microbiota colonizing infants was of similar complexity. Thereafter, cesarean section significantly altered microbial β-diversity compared to vaginal delivery (unweighted UniFrac permutational MANOVA, P < 0.001) but accounted for a small fraction of total variation among samples (R = 0.02) (Fig. 5A and table S7). For several common bacterial taxa, abundances were altered in cesarean-born infants compared to those vaginally born, underlying the differences in α- and β-diversity (tables S6 and S7) (30). Most prominently, Bacteroides abundance was significantly lower in cesarean-delivered infants (Fig. 5, B and C, and fig. S6), regardless of predominant feeding mode. By 12 months, the balance of Bacteroides, Bifidobacterium, and Enterobacteriaceae that dominated the first year of life in all infants was replaced by a mixture of Firmicutes, primarily Clostridiales (Fig. 1). Although particular Clostridiales and Enterobacteriaceae were significantly more abundant [linear discriminant analysis effect size (LEfSe), P < 0.05] in cesarean-born infants during the first year, filling the void left by Bacteroidales, few taxa were significantly different during the second year of life (Fig. 5C). This increasing similarity after 1 year in children born vaginally or by cesarean section indicates that both communities undergo gradual maturation, eventually resembling the adult fecal microbiota (Fig. 6). The taxa that dominated in the early months of life—whether or not disturbed by cesarean delivery or antibiotics—declined as later-life taxa replaced them.

Fig. 5. Delivery mode alters microbial diversity and composition.

(A) Unweighted UniFrac principal coordinates (PCs) analysis of the infant microbiome in relation to delivery mode over the first 2 years of life. Permutational MANOVA, P < 0.05 (table S7). (B) Bacteroides RA (means ± SEM) over time in relation to delivery mode. Asterisks and brackets indicate significant (LLM, P < 0.05) rate-of-change differences across age ranges (dotted lines demarcate time periods tested). (C) Cesarean section significantly altered abundance of diverse bacterial taxa over time. Red-shaded taxa (rows) were significantly more abundant (LEfSe, P < 0.05) in cesarean-delivered infants at the given time points (columns); blue shading indicates more abundant taxa in vaginally delivered infants.

Fig. 6. Bipartite network comparing the relationships among all samples (squares) and OTUs (circles).

(A) The distance between sample nodes and OTU nodes is a function of shared microbial composition. Samples with a large degree of OTU overlap (weighted by the number of observations of that OTU) form clusters. Edges connect a sample to each OTU detected in that sample, revealing shared OTUs between samples. Sample nodes and edges are colored by sample type; the border of sample nodes is a function of the age of the child, including prepartum (negative) values for maternal samples (key at top left). OTU nodes are colored by taxonomic family affiliation; the size of each OTU node is a function of that OTU’s overall abundance, registered as OTU count in all samples (key at middle left). See Fig. 7 for specific analyses. (B) Unweighted UniFrac distance between maternal vaginal, rectal, and stool microbiota and child stool microbiota as a function of child age. Shorter distance indicates greater similarity between microbial communities. (C) Unweighted UniFrac distance between stool microbiota from the same child (self) and other children (nonself) as a function of the difference in age between sampling (Δ months). (D) Unweighted UniFrac distance between maternal vaginal microbiota and stool microbiota of vaginally born dyads, unrelated children, or cesarean-delivered dyads as a function of child age. For (B) to (D), lines indicate rolling average mean values, and gray shading is 95% CI. ANOVA P values are shown.

We hypothesized that the disruption of cesarean section could also alter microbiota maturation patterns in infants, similar to antibiotic exposure (Fig. 4). Using the microbiota maturation model described in the Supplementary Materials, we found that cesarean- and vaginally delivered infants demonstrated similar degrees of microbiota maturation during the first 6 months of life. Subsequently, microbiota maturation stagnated in cesarean-delivered infants, with relative maturation dropping compared to vaginally born infants for the remainder of the study period (fig. S7).

Infant diet affects intestinal diversity

To assess the relationship between early diet and the microbiota, we routinely surveyed infant nutrition, documenting the extent of breast-feeding or formula feeding and the timing of solid food introduction (Fig. 1F). We compared two major dietary groups that best described dietary variation in this cohort: infants who were dominantly (>50% of feedings) breast-fed or dominantly formula-fed for the first 3 months of life. Phylogenetic diversity and bacterial richness growth rates were significantly decreased in formula-fed children during 12 to 24 months of life (P < 0.05) (Fig. 2, G to I, and table S6). Formula feeding also altered β-diversity (fig. S8A and table S7) and decreased microbiota maturation during 12 to 24 months of life (fig. S8B). During this period, Lactobacillus, Staphylococcus, Megasphaera, and Actinobacteria were more abundant in breast-dominant children, and various genera of Clostridiales and Proteobacteria were more abundant in formula-dominant children (fig. S8C).

Maternal bacteria populate the infant gut during early life

The earliest source of microbial colonizers in the infant gut is the mother’s own microbiota, from passage through the birth canal to breast-feeding and skin contact. However, the extent to which a mother’s microbiota successfully establishes in her child, its dynamics over time, and the relative contributions of bacteria from different body sites have not been well described. We characterized the microbiota of rectal swabs, vaginal swabs, and stool samples from mothers before and after birth (fig. S9) to explore the microbial relationships shared between mother-infant dyads, as well as unrelated individuals, in the context of birth mode.

Network analysis explores the relationships between samples and their constituent OTUs, revealing that the infant microbiota matured from a neonatal state, associated with maternal OTUs from both vagina and rectum, to a post-infancy state resembling maternal stools (Fig. 6). In this analysis, sample nodes are connected by edges to all OTUs that they contain, indicating explicit connections between samples via shared OTUs. Edges are weighted on the basis of OTU abundance, causing samples with similar OTU composition to cluster together along with their characteristic OTUs. These connections may be quantitatively measured directly as shared OTU counts between samples (Fig. 7 and fig. S10) and indirectly as UniFrac distance (Fig. 6, B to D). Infant stool samples clustered together, associated strongly with several key Bacteroidales, Clostridiales, Enterobacteriales, and Bifidobacteriales OTUs (Fig. 6A). Infants cluster away from maternal stool samples, instead having more connections to maternal vaginal samples via robust links with several Lactobacillales and Bifidobacteriales OTUs (Fig. 6A). As the children age beyond 12 months, their fecal microbiota came to resemble maternal stool and rectal samples more closely (Fig. 6B), indicating that their microbiota matured to a maternal-like configuration associated with numerous Clostridiales and Bacteroidales (Fig. 6A and fig. S10). Stool microbiotas from the same child (“self”) were more similar to each other than those from unrelated children (“nonself”) (unweighted UniFrac ANOVA, P < 0.001), and children’s stool microbiotas were most similar to those of the same age, highlighting that the succession of the intestinal microbiota occurs gradually (Fig. 6C). Conversely, at all ages, the children’s fecal microbiota was less similar to the mothers’ vaginal microbiota than to the mothers’ rectal and stool microbiotas (Fig. 6B). The dissimilarity between a mother’s vaginal microbiotas and her child’s fecal microbiotas was significantly more pronounced in cesarean-born children than in vaginally born mother-child pairs, whether dyads or unrelated (unweighted UniFrac ANOVA, P < 0.001) (Fig. 6D). Considering two important genera, Bacteroides and Bifidobacterium, there was a consistent diminution in the total OTU diversity acquired in cesarean-born infants, as well as shared OTUs; dyads shared significantly more of these taxa than did unrelated pairs (fig. S10).

Fig. 7. Shared OTUs reveal microbial relatedness among mothers and children.

(A) Shared OTU counts (median ± quartiles) between individual stool samples (top), rectal swabs and stool samples (middle), and vaginal swabs and stool samples (bottom), represented in Fig. 6. Distributions represent the total number of OTUs within a single sample (blue) or shared OTUs between samples from the same individual (self, yellow), another individual (nonself, white/black), or a mother-infant dyad (red). Lowercase letters indicate significantly different shared OTU count distributions [one-way ANOVA, P < 0.0001, followed by false discovery rate (FDR)–corrected Fisher’s protected least significant difference (PLSD) test]. Key indicates coloring for box plots in (A) or line plots in (B) to (I). (B to I) Shared OTU counts over time between mothers and unrelated children, mother-infant dyads, and total OTUs in child stool samples. (C) Samples from the same child or unrelated children at different times (Δ months). (D) Mothers’ rectal swabs and stool samples from their own children (dyad) or unrelated children. (E) Mothers’ vaginal swabs and stool samples from unrelated children or dyads of children delivered vaginally or by cesarean section. (F) Vaginal and rectal swabs from the same mother or other mothers. (G) Stool samples from the same mother or other mothers. (H) Rectal swabs from the same mother or other mothers. (I) Vaginal swabs from the same mother or other mothers. (B), (D), and (E) compare mothers versus children, and x axes indicate the child’s age (months). For (C) and (F) to (I), x axes indicate the differences in child age (Δ months) between the times when these samples were obtained. Lines indicate rolling average mean values, and gray shading is equal to 95% CI. *P < 0.0001, ANOVA, followed by FDR-corrected Fisher’s PLSD test.

Because microbial composition differs widely between adults and children, and between samples from different body sites, quantitation of shared OTUs is a metric to assess microbial transmission between these loci (Fig. 7). Samples with more shared OTUs are expected to have closer relationships, whether OTUs are transmitted directly between samples or both have a common external source. Maternal microbiota transmission implies that more maternal OTUs are expected in her child than in unrelated children, except when OTUs are highly dispersed among mothers. We examined the total number of OTUs in each sample, and the number shared between samples belonging to the same individual (self) or other individuals (nonself), or between mother-infant dyads (Fig. 7). Results support expectations that OTUs are more shared between self samples, and a mother’s rectal microbiota shares more OTUs with her own child’s stool than that of other children (Fig. 7A). Although vaginal swabs follow this trend, effects are not significant when examining samples from all time points and birth modes.

Child stool OTU counts gradually increased with time (Fig. 7B), approaching OTU counts in mothers after 2 years of age (compare Fig. 7, A and B). Unrelated mothers and children shared far fewer stool OTUs throughout this time, but, surprisingly, still shared more OTUs on average than mother-infant dyads (Fig. 7B). The shared OTU count increased as children grew older—for both dyads and unrelated mother-infant pairs—further evidence that children’s microbiotas gradually mature into an adult-like state. After 2 years of age, α-diversity (total OTU counts) approached adult levels (compare Fig. 7, A and B), and β-diversity became lower between children and adults (Fig. 6B), but OTUs shared with adults remained low, indicating that different strains colonized children and adults even when similar bacterial taxa were present.

A child’s stool shared more OTUs with their other stools that were collected <14 months apart, compared to nonself stool samples; as time between sampling increased, shared OTUs decreased until self and nonself stools had similar shared OTU counts (occurring for samples >14 months apart) (Fig. 7C), indicating the dynamism of the microbiota during the first 2 years of life. As expected, the number of OTUs shared between self-stool samples gradually decreased as the time between sampling increased, indicating gradual succession of OTUs. As children aged, they also shared more OTUs with maternal rectal swabs whether or not they were related (Fig. 7D). Dyads shared more OTUs across all times (Fig. 7A) but were not significant at individual time points (Fig. 7D). Mothers shared significantly more vaginal OTUs with their children if they had been vaginally delivered, compared to both cesarean-delivered and unrelated infants (Fig. 7E), an effect only significant after 1 year of life, peaking between 18 and 24 months of life. Vaginal OTUs, relevant during early infancy, were detected less frequently in fecal specimens in later childhood, leading to the decline in shared OTUs. As expected, peripartum maternal stool, vaginal, and rectal samples also showed fewer shared OTUs with samples from the same mother as the interval between sampling increased (Fig. 7, F to I). As sampling interval grew, maternal samples shared as many OTUs with samples from other mothers as they did with self. Stool samples were the exception, with more shared self-OTUs across sampling than from other mothers (Fig. 7G).


The purpose of this study was to characterize early-life microbial development in the context of antibiotic use, cesarean section, and formula feeding. Each of these has been associated with conditions emerging later in life, including obesity (9), diabetes (10, 20), and allergies (13, 19). The cause of these relationships is unclear, but altered patterns of microbiota assembly during early life are plausible. We profiled microbiota development during the first 2 years of life and documented disturbances related to antibiotic treatment, cesarean section, and diet.

Intestinal bacterial communities undergo a gradual succession during early life (Fig. 1A), after a predictable, age-dependent pattern that is conserved between disparate human populations and stabilizes after 3years of age (5, 24, 25). These events may reflect a coevolutionary relationship in which normal maturation of the gut microbiome during a critical window contributes to host development; disturbances may interrupt this delicate choreography, with consequences for long-term host health (16).

We observed three distinct phases in the childhood microbiome during the first years of life. During the first month of life, Enterobacteriaceae dominated the microbiota (Fig. 1), suggesting that these facultative anaerobes can exploit the naïve conditions of the neonatal intestine. The intermediate stages of development, from about 1 to 24months of life, were more dynamic and appeared to be sensitive to disturbances from birth mode, predominant nutrition, and antibiotic use. We speculated that even transient effects during this sensitive, developmental window could lead to long-lasting effects—as we have shown with short-term antibiotic exposures in mice (15, 16) and as has been observed in human children, associating antibiotic exposures to illness outcomes (913, 31). Finally, as children reached 2 years of age, the microbiota gradually stabilized toward an adult-like community (Fig. 6), characterized by both higher diversity and greater resilience to change. This period paralleled dietary transition from liquid to solid foods (Fig. 1F), an important catalyst for microbiota maturation in childhood (17) and a source of introduced microbes. Alternative hypotheses for the higher diversity include the diminution of the constraint imposed by cesarean section on microbial transmission between the mother and neonate (fig. S9) and between breast-feeding and its strong selective effects (32). Higher immunological tolerance during early life (33) may be ending, adding new constraints to acquiring nonfounding microbes. These conserved transitions may be important and suggest that disturbances during the first 2 years of life are likely to have strong effects on development of the microbiota and potentially for host health.

The particular microbial species and the specific gene pathways that best define “microbial age” can be used to track how an infant’s gut microbiome matures. In that sense, they can be markers, paralleling how weight-for-height tracks a child’s development, and are similarly vulnerable to disruptions affecting health (5, 15). Although the precise role that these bacteria play in development is unclear, identifying maturation markers actually linked to host processes at key junctures during an infant’s development has great potential. By suppressing early-life bacterial markers including Lachnospiraceae, Enterobacteriaceae, Erysipelotrichaceae, and numerous predicted gene pathways, antibiotic exposures effectively stall the development of the intestinal microbiome, causing the intestinal communities in these infants to appear less developed (younger) than unexposed counterparts (Fig. 4).

Lachnospiraceae sp. and other Clostridiales appeared particularly sensitive to antibiotic exposures, with significant depletion observed in antibiotic-treated infants often throughout early life (Fig. 3). These findings confirm previous studies of adult human and mouse microbiotas (8, 34). Lachnospiraceae live almost exclusively in the mammalian gastrointestinal tract, often producing butyrate and other short-chain fatty acids (35) that regulate host immunity via epithelial cell signaling, colonic T regulatory cells (36), and macrophages (37). Butyrate synthesis could explain how Lachnospiraceae induce T regulatory cells, suppressing colitis and allergic diarrhea in mice (38), and is consistent with protective associations against type 1 diabetes development in infants (39). Butyrate, serving as an energy source for host epithelial cells, also regulates the cell growth and differentiation-related AP-1 (activator protein 1) signaling pathway (40), potentially explaining links between butyrate producers (including Lachnospiraceae sp.) and body weight (41). Lachnospiraceae OTUs were also implicated as markers of intestinal microbiota maturation in Bangladeshi infants (5), suggesting that their blooms may be an important event in the developing infant gut across continents.

α-Diversity, an important ecosystem characteristic, was disturbed by delivery mode, antibiotic use, and diet (Fig. 2). During early life, overall diversity rapidly increases in the developing infant gut (Fig. 2) (24, 25) as children acquire bacterial strains encompassing greater phylogenetic diversity from diet, human contact, and environment. Because this process is occurring during a sensitive period when acquired intestinal bacteria train the nascent immune system (33), increasing diversity may be relevant to normal development (1). Disturbance from antibiotic use, formula feeding, or cesarean section, as well as the potential for cumulative damage from multiple disturbances, might affect intestinal homeostasis and long-term health. Decreased intestinal α-diversity during infancy precedes type 1 diabetes onset (39) and allergic manifestations (42) and has been reported in obese adults (41).

Whether intestinal α-diversity influences disease development or is only a marker, functional diversity differences could contribute to intestinal homeostasis (26). The substantial functional redundancy in the healthy human microbiome (26, 43) may provide insurance that the gut ecosystem can recover from temporary disturbances, maintaining productivity by minimizing risk of functional loss. In this model, a healthy microbiota with sufficiently high diversity can withstand normal environmental fluctuations, for example, due to diet change or illness, and maintain key functions. However, disturbances during development may reduce diversity below thresholds sufficient to maintain an essential functional repertoire. Such lack of resilience may delay ecosystem recovery or lead to a new stable state (26); either event may promote later disease development. One challenge to the translation of this theory will be identifying microbial ecosystem functions important for critical developmental steps; as targets for future investigation, we identify several gene pathways affected by early disturbances (see the Supplementary Materials).

In another study in this issue, Yassour et al. (44) have shown that antibiotic exposures also affect intestinal microbiota stability and diversity at the level of individual strains, highlighting the importance of strain-level dynamics to microbiome establishment during early life. If specific strains have unique functional niches in the developing intestine, their displacement could affect host-microbial interactions.

Another notable effect of cesarean section was a marked, persistent decrease in Bacteroides populations. Although several studies have noted decreased phylum Bacteroidetes in cesarean-delivered infants (45, 46), we now show that this effect persists throughout the first year of life and is associated with an altered metagenomic landscape in the gut, consistent with the findings of Yassour et al. (44). Because some Bacteroides sp. help regulate intestinal immunity (47), this long-lasting, large-scale disturbance could partially explain cesarean section–associated health consequences, including asthma, obesity, and allergies (18, 21). The deficiency of Bacteroides after cesarean section, with commensurate expansion of other taxa, may disrupt tolerogenic feedback loops (26), potentially contributing to development of inflammation and obesity. Disturbances to microbial metabolism and cell motility pathways in cesarean-born children suggest mechanisms by which microbial perturbation could influence the host; targets for future investigation include the altered production of short-chain fatty acids or other immunomodulatory metabolites (36).

The strong evidence for maternal transmission of early-life taxa in general, and specific dominant taxa in particular, provides support for the importance of the composition of the maternal microbiota for healthy development. Widespread practices that affect this reservoir include antibiotic therapies and prophylaxes but also cleansings (for example, maternal vaginal lavage, skin cleansing of neonate, and umbilical cord cleansing) that were aimed for neonates in developing countries but have recently spread to more developed populations (48).

Some limitations are inherent to our study. Our sample size lacks sufficient power to account for complex interactions between many potential microbial disturbances during early life. Larger and longer studies may assess whether early-life microbiota disruptions are indeed cumulative, counteractive, or independent. Longer studies can determine how community assembly during childhood transitions to climax (for example, adult) communities, characterized by diverse alternative stable states (49). Dissimilar trajectories of microbiota maturation during childhood may lead to different stable states in adulthood, without necessarily affecting the host. Microbial disruptions primarily delayed microbiota maturation during the first year, indicating transient effects rather than permanent alterations. Transient antibiotic disturbances during early life durably alter host development in mice (1416), but our results do not necessarily imply that this effect extends to humans. An alternative possibility is that disruptions to microbiota composition and maturation are nullified if they are replaced by other functionally similar taxa. The disparate microbiota states observed in adults have functional overlap despite compositional similarity, suggesting that community function may be more relevant for predicting host interactions (26, 49). Our results suggest that community function and functional maturation are altered by antibiotics, cesarean section, and formula feeding [using PICRUSt (Phylogenetic Investigation of Communities by Reconstruction of Unobserved States)–predicted metagenomes], but metagenomic and transcriptomic studies will be needed to assess whether early-life disruptions influence community function and behavior.

Our results identify multiple disturbances to microbial development during early life. These alterations were not linked to health outcomes, which we did not survey, but are potential targets for ongoing inquiry. The microbial populations and dynamics deserve further scrutiny for their role in host-microbial communication, host development, and health. Mounting evidence associates antibiotic use, cesarean section, and formula feeding with metabolic and immunological disorders later in life, possibly as a consequence of disturbed microbiome development. Although these interventions often strongly benefit recipients, the hidden costs imply the need for careful consideration in children with less severe illnesses. Our findings warrant studies to establish whether and how early-life microbial disturbances associated with antibiotic use, cesarean section, and formula feeding influence health outcomes. Recent studies of Canadian and Finnish children provide evidence that such early-life disturbances are important for asthma risk (12, 50, 51).


Study summary

An Institutional Review Board–approved study was conducted in healthy, pregnant mothers in New York City from 2011 to 2014. Maternal vaginal, rectal, and fecal specimens and fecal specimens from their infants were obtained from birth to the age of 3 years, chiefly in the first 2 years of life, and analyzed by microbiome sequencing. Information about home environment, delivery mode, infant diet, and antibiotic exposures was obtained for each mother-baby dyad. Complete details are provided in the Supplementary Materials.

Study design

The Early Childhood Antibiotics and the Microbiome (ECAM) study is a longitudinal study aiming to examine the development of the intestinal microbiota in early childhood. Enrollment of mothers, which was specific for this study, began in December 2011 and ended in December of 2014 at New York University (NYU) School of Medicine. Inclusion criteria included healthy pregnant mothers 18 to 45 years old. For better generalizability of the results and to reduce the likelihood of loss to follow-up, we excluded mothers who were at high risk for premature delivery (that is, incompetent cervix, premature labor, bed rest), had fetal anomalies detected in utero, and/or were unable to meet study obligations. We excluded infants who required treatment more than 7 days in a neonatal intensive care unit, premature infants of less than 38 weeks gestation, and those with genetic diseases and/or birth-related injuries requiring intensive monitoring. We sought to enroll at least 60 mothers, so that, taking dropouts into account, we would have a cohort of at least 40 babies who could be followed for at least 1 year. We assumed a cesarean section rate of ~50%, which would provide 20 babies in each group to examine the chief outcome variables. In total, 53 mothers were enrolled (table S1); after early dropouts, follow-up of their infants occurred in 43 cases. The mothers who continued with the study did not significantly differ from the initial enrollees (table S1). Their babies were followed up for up to the first 3 years of life (table S2). Study surveys were obtained at enrollment; 6 weeks postpartum; 4, 6, and 9 months during year 1; and every 6 months thereafter (Table 1). Maternal fecal and vaginal swab samples were collected at two separate times prepartum and once postpartum. Stool samples were collected from infants 12 to 24 hours after delivery, every month for the first year, and every other month for the second and third years of the study (see Sample collection). With 43 infants, we would have 80% power to detect a 0.82 standardized effect (standardized to sample SD of the outcome, such as diversity indices) in relation to a dichotomized factor, such as antibiotic use or delivery methods, with a prevalence of 0.50. This is a conservative estimate based on t test, because analyses based on longitudinal data have greater power. This study was approved by the Institutional Review Board at the NYU School of Medicine.

Sample collection

Maternal fecal and vaginal swab samples were collected at two separate times prepartum (4 to 110 days before parturition) and once postpartum (18 to 188 days postpartum, typically ~6 weeks; average, 51.4 days). Samples were collected by the obstetrician and/or through self-collection by the mothers using sterile cotton-tipped swabs; these were collected during routine prenatal visits and were kept chilled at 2° to 8°C immediately after collection and delivered within 2 hours to the laboratory, where they were immediately frozen at −80°C until processing. Stool samples were collected once from each mother, 1 to 63 days postpartum, typically within 1 to 3 weeks of parturition, with only one sample collected 16 days before parturition and an additional sample from one mother at 232 days postpartum. Stool samples were collected from infants 12 to 24 hours after delivery, every month for the first year, and every other month for the second and third years of study. Mothers, when possible, also collected stool samples from their children before, during, and after each course of antibiotic use. Mothers were provided with sterile supplies and instructions for sanitary collection and storage of feces from their own children. Mothers were also provided with ice packs and insulated cooler bags; samples were kept chilled at 2° to 8°C immediately after collection and delivered within 24 to 48 hours to the laboratory, where they were stored at −80°C until processing.

Summary of statistical methods

Microbiome sequencing data, bacterial composition, and analyses of α-diversity (observed OTUs, phylogenetic diversity, and species evenness) and β-diversity [UniFrac (30) phylogenetic distance between samples] were calculated in QIIME (Quantitative Insights into Microbial Ecology) (52). Random forest (27) regression models, using microbiota composition as a predictor of subject age, were used to examine relative rates of microbial maturation, as described previously (5). Significant differences in β-diversity were tested by permutational MANOVA (53). Significant longitudinal changes in α-diversity, microbiota maturation, and PICRUSt-predicted (28) metagenome maturation were tested by piecewise LLM (54). Significant differences in RA of microbial taxa were tested by LEfSe (55).


Materials and Methods

Fig. S1. LEfSe analysis of differentially abundant taxa between vaginally born, predominantly breast-fed (n = 15) and cesarean-delivered, predominantly formula-fed children (n = 7) from 1 to 6 months of life.

Fig. S2. Antibiotic exposures did not alter short-term α-diversity in individual subjects.

Fig. S3. Functional maturation of the microbiome is delayed by antibiotic exposure.

Fig. S4. Functional maturation of the microbiome is altered by formula feeding.

Fig. S5. Enumeration of 16S ribosomal RNA genes in fecal specimens in children at 1 and 12 months of age, by quantitative polymerase chain reaction.

Fig. S6. Average RA of Bacteroides in children in relation to feeding and delivery mode in the first 2 years of life.

Fig. S7. Cesarean section exerts a modest impact on microbiota maturation.

Fig. S8. Infant diet alters microbiota composition and maturation over the first 2 years of life.

Fig. S9. RA heatmap of major taxa in 296 maternal samples.

Fig. S10. Development of diversity of Bifidobacterium and Bacteroides OTUs in children in early life.

Table S1. Baseline characteristics of the 53 mothers in the study.

Table S2. Baseline characteristics of the 43 infants in the study.

Table S3. Prenatal antimicrobial use by class and purpose.

Table S4. Perinatal antibiotic use by class and indication.

Table S5. Postnatal antimicrobial use by class and age of child.

Table S6. LLM estimates of antibiotic treatment, diet, and delivery effects on α-diversity.

Table S7. Permutational MANOVA scores of antibiotic, diet, and delivery effects on unweighted UniFrac distance β-diversity.

Table S8. Permutational MANOVA scores of antibiotic, diet, and delivery effects on weighted UniFrac distance β-diversity.

Table S9. LLM estimates of antibiotic, diet, and delivery effects on microbiome maturation MAZ scores.

Table S10. LLM estimates of antibiotic, diet, and delivery effects on PICRUSt-predicted metagenome maturation MAZ scores.

References (5665)


  1. Acknowledgments: We thank S. Bedi and A. Radin for the clinical assistance. Funding: This study was supported by grant R01 DK090989 from NIH, Public Health Service Institutional Research Training Award T32 AI007180, the Diane Belfer Program for Human Microbial Ecology, the Daniel and Leslie Ziff Fund, and the C & D fund. Author contributions: M.J.B., M.J., Y.C., G.I.P.-P., and M.G.D.-B. planned and designed the longitudinal study design and the analytical experiments. N.H. and W.S. recruited study subjects. J.C. performed microbiome sequencing. X.Z. and M.C. performed quantitative polymerase chain reaction experiments. N.H., J.C., and T.B. managed the study data. N.A.B., T.B., A.D.L., F.W., J.C., and H.L. performed all analyses. N.A.B. and M.J.B. wrote the manuscript. All authors contributed to and reviewed the final manuscript. Competing interests: N.A.B. is a co-founder of MicroTrek Inc., a company focused on microbial analysis for food, beverage, health, and other industries. MicroTrek was not involved in funding, design, or interpretation of this study. M.G.D.-B., together with NYU Langone Medical Center (LMC), has intellectual property related to the restoration of the microbiome of newborns, which has been licensed to Commense Inc. M.G.D.-B. and M.J.B. are co-founders and equity holders in Commense Inc., which is focused on the restoration of depleted microbiota in early life. Commense was not involved in the funding, design, or interpretation of this study. M.J.B. serves as a consultant to Johnson & Johnson, which funds unrelated studies in his laboratory at NYU LMC. The other authors declare that they have no competing interests. Data and materials availability: All sequencing data are publicly available in QIITA database ( under study ID 10249.
View Abstract

Stay Connected to Science Translational Medicine

Navigate This Article