Environmental and Maternal Imprints on Infant Gut Metabolic Programming

Abstract

Early life is a critical period for immune and metabolic programming, but developmental patterns remain underexplored in populations from low- and middle-income countries. Here, we profiled the microbiome and metabolome of 55 Bangladeshi mother-infant dyads over the first six months of life. Importantly, we observed an increase in microbially-derived bile amidates and N-acyl lipids with age in conjunction with reads matching the bile salt hydrolase/transferase (bsh) gene. While microbial source tracking confirmed maternal fecal seeding, a substantial environmental contribution was also highlighted. Differences in infant fecal metabolic profiles were associated with delivery mode, maternal milk composition, household assets, and household-level water treatment. C-section delivery and untreated drinking water were linked to transient metabolic differences, including increases in bile amidates, N-acyl lipids, and other host-microbe co-metabolic products, including acylcarnitines. Multi-omics analysis revealed specific microbial-metabolite relationships, highlighting how early environmental and maternal living circumstances shape metabolic gut programming through the microbiome.

Fig. 1:. Rapid shifts in key signalling metabolites during infant fecal metabolic gut programming.

a, Experimental overview. Longitudinal samples were collected from 55 mother-infant dyads in Dhaka, Bangladesh. Infant samples included stool, oral (tongue), skin (elbow), and blood; maternal samples included stool, milk, and a vaginal swab (see Methods for details). Open circles indicate random collection time within the first 60 days. n reflects the initial number of samples per biospecimen, prior to quality control. b, UpSet plot of metabolic features across infant fecal (n=377) and plasma (n=34) samples, and maternal milk (n=119) samples. The left panel shows features unique to or shared between sample types, while the right panel shows chemical classes and pathways predictions based on NPC classification from CANOPUS (probability > 0.5). c, Principal Component Analysis (PCA) of maternal milk showing significant development over time (Permutational Multivariate Analysis of Variance (PERMANOVA): p < 0.001). d, Boxplot showing the natural log peak areas of key discriminant features across lactation stages: colostrum (< 5 days), transitional (5–15 days), and mature milk (> 15 days). Statistical significance was determined by the Wilcoxon rank-sum test, with p values adjusted for multiple comparisons using the Benjamini-Hochberg (BH) method. Repeated measures were avoided by retaining only the highest value per subject within each lactation stage. e, PCA of fecal metabolic profiles across early infancy (0–6 months). Differences in the metabolic profiles were associated with infant age, delivery mode, maternal secretor status, and drinking water treatment (PERMANOVA: all p < 0.05). f, Longitudinal abundances of free di- and tri-hydroxylated bile acids in infant feces over the first 6 months of life. g, Abundance trajectories of taurine (tau), glycine (gly), and amino acid (AA) conjugated bile acids, as well as oxo bile acids in the same period. h, Scatter plot showing the association between the natural log of bile salt hydrolase/transferase (bsh) gene read count matches and the ratio of amino acid to taurine conjugated bile acids using Pearson correlation for significance (linear mixed effect model (LME): β = 0.25, p < 0.001). A linear regression line with 95% CI is shown, with points colored by infant age. i, Longitudinal abundance trajectories of N-acyl lipids with a C2:0 fatty acid chain (left panel) and a C3:0 fatty acid chain (right panel) amidated to different head groups. j, Longitudinal abundances of MS/MS spectral matches to various short-chain acylcarnitines (left panel) and long-chain acylcarnitines (right panel) in infant feces. For panels f, g, i, j, β and p values were derived from LME with subject as a random effect, and scatter plot smoothing (‘LOESS’) was used for visualization purposes. All boxplots show the first (lower), median, and third (upper) quartiles, with whiskers 1.5 times the interquartile range.

Fig. 2:. Sources driving the developing gut microbiome.

a, phylo-RPCA EMPeror plot of fecal samples with more than 100,000 reads (n=219) colored by infant age bin. b, Top and bottom 12 differentially abundant infant fecal genera identified by ANCOM-BC2 (p_adj < 0.05, LFC > 0.5). Full plot shown in Extended Data Fig. 3a. Enriched (yellow) or depleted (purple) after 30 days (later timepoints) are shown, black error bars indicate standard error. c, Scatterplot with lineplot of the natural log-ratio of the summed counts of enriched (numerator) and depleted (denominator) fecal genera found in b over time (LME: β = 0.026, p < 0.001); lineplot used binned timepoints. FEAST: Infant fecal samples as sinks. d, Stacked barplot of the mean percentage contribution of each maternal body site on infant fecal samples for matched (left) or all available (right) maternal samples. e, Stripplot/pointplot of the FEAST estimated percentage contribution of maternal fecal samples to matched (left) or all available (right) infant fecal samples. f, Stripplot/pointplot of the FEAST estimated percentage contribution of unknown elements to matched (left) or all available (right) infant fecal samples. Notation: ** = p < 0.01.

Fig. 3:. Delivery mode and water treatment are associated with infant fecal metabolic trajectories in early-life.

a, PCA of infant fecal samples showing separation by delivery mode (PERMANOVA: p < 0.001). b, Longitudinal natural log ratios of the summed peak areas of the top 100 discriminating features for delivery mode in infant feces. LME was used to determine statistical significance between groups. Scatter plot smoothing (‘LOESS’) and removal of three outliers for visualization. c, Barplot of metabolic features significantly differing between C-section and vaginal birth (p_adj < 0.10) in infant feces collected during days 10–30 after birth. For each infant, the sample with the highest log ratio within this interval was selected. d, Boxplot of the natural log ratios of amino acid to taurine conjugated bile acids by delivery mode at selected timepoints (Wilcoxon rank-sum test, day 14–28: p = 0.042, p_adj = 0.13). e, Boxplot of bsh gene reads in infant feces by delivery mode (Wilcoxon rank-sum test, day 14–28: p = 0.07, p_adj = 0.08). f, Trajectories for valine-C2:0 and 2-aminovaleric acid-C2:0 by delivery mode in infant feces, with smoothing for visualization. The difference between C-section and vaginal birth was assessed for significance using LME. g, PCA of infant fecal samples showing separation by household drinking water treatment (PERMANOVA: p = 0.020). h, Longitudinal natural log ratios of the summed peak areas of the top 100 discriminating features for treated vs. untreated drinking water. Smoothing added for visualization, and statistical significant difference between groups was assessed using LME. i, Molecular network of acylcarnitines associated with household drinking water treatment (untreated=brown, treated=dark green). Node size represents mean VIP score from three different PLS-DA models accounting for different purification methods, with larger nodes indicating higher VIP score. Only acylcarnitines that were significant in all three models (VIP > 1) and with the same directionality were included in the network. j, Boxplot showing differences in the natural log peak areas of various acylcarnitines enriched in the untreated water group at day 0–7. Statistical significance was determined by the Wilcoxon rank-sum test with BH adjusted p values. k, Differentially abundant infant fecal genera by water treatment identified by ANCOMB-BC2 (q < 0.05, log fold change (LFC) > 1). Enriched (grey) or depleted (brown) in treated water are shown, black errors indicate standard error. l, Scatterplot/lineplot of the natural log-ratio of the summed counts of enriched (numerator) and depleted (denominator) fecal genera found in k (LME: −0.058, p = 0.13); lineplot used binned timepoints. For b, f, and h, β and p values were derived from LME models with subject as random effect. All boxplots show the first (lower), median, and third (upper) quartiles, with whiskers 1.5 times the interquartile range. For boxplots, repeated measures were avoided by retaining only the highest value per individual within each time window.

Fig. 4:. Maternal secretor status shapes HMO profiles in human milk and infant feces.

a, Longitudinal natural log ratios of the summed peak areas of the top 70 discriminating features for maternal secretor status in human milk. Statistical significant difference between groups was assessed with LME. Scatter plot smoothing (‘LOESS’) added for visualization. b, Boxplot of metabolic feature (m/z 471.1709) predicted to be 2’FL, targeted quantification of 2’FL, metabolic feature (m/z 657.2211) predicted to be DFLac, and targeted quantification of DFLac, respectively, grouped by secretor status (Wilcoxon rank-sum test, with BH adjusted p values). For subjects with multiple samples within a stage, only the sample with the highest 2′FL concentration was retained. c, PCA of infant fecal samples showing separation by maternal secretor status (PERMANOVA: p = 0.006). d, Longitudinal natural log ratios of the summed peak areas of the top 100 discriminating features for maternal secretor status in infant feces, with the significance of the difference assessed by LME and smoothing for visualization. e, Piechart showing the distribution of NPC predicted chemical classes or pathways (probability > 0.7) for the discriminant infant fecal features by maternal secretor status. Only features with VIP > 2.5 were included. f, Cross-biofluid molecular network of amino sugars, aminoglycosides, and saccharides, as predicted by CANOPUS, colored by biofluid and secretor status. Node size represents the mean VIP score of significant features across the two PLS-DA models for secretor status in human milk and infant feces, with larger nodes indicating higher VIP score. Selected features enriched in both maternal milk and infant feces are shown in boxplots. g, Boxplot of the natural log-transformed peak areas of a candidate trihydroxy bile acid (Δ 163.0841) grouped by secretor status (Wilcoxon rank-sum test with BH adjusted p values). h, Boxplot of the natural log-transformed peak areas of putative 3α,7α-(OH)₂-12-oxo based on secretor status at select timepoints (Wilcoxon rank-sum test with BH adjusted p values). All boxplots show the first (lower), median, and third (upper) quartiles, with whiskers 1.5 times the interquartile range. For boxplots in f, g, and h, repeated measures were avoided by retaining only the highest value per individual within each time window. For a and d, β and p values were derived from LME models with subject as random effect.

Fig. 5:. Multi-omics integration identifies microbial mediators of metabolic trajectories.

a, Heatmap of correlating metabolites and microbial species from infant fecal samples collected at day ~30 identified by joint-RPCA. Metabolic features shown were selected based on previous analyses. Asterisk indicate MS/MS spectral match against microbeMASST, with (*) indicating a general match and (**) indicating a match to a monoculture of the same genus as taxa shown in the heatmap. Microbiome features shown were selected based on the overlap between 3 previous analyses, including needing to belong to the differentially abundant genera found in the ANCOM-BC2 analysis from Figure 2b, the top and bottom 20 most differentially ranked features for joint-RPCA Axis 1, and the top and bottom 10% differentially ranked features from TEMPTED Axis 1 (overlapping features=23).