Introduction

The human gut microbiome is a complex ecosystem, comprising hundreds of microbial species with significant associations with human health and disease, including metabolic syndrome, cognitive disorders, and autoimmune diseases1. Microbial colonization is a key developmental process in early life. The timing and sequence of microbial exposures shape the development of the gut microbiome, which affects host metabolism and immunity later in life. In humans and other mammals, the first microbes encountered early in life, particularly upon vaginal delivery, are those from the maternal microbiome2,3. Subsequently, breastfeeding, skin-to-skin contact, and the living environment introduce new colonizing microbes into the neonatal gut4,5. With the addition of solid foods, the diversity and complexity of the infant gut microbiome rapidly increase, converging to the adult microbiome composition and reaching homeostasis around 3 years of age6,7. The initial colonization of the gut microbiome is extraordinarily sensitive to external stimuli8. Hence, infancy represents a critical window for the colonization and development of the gut microbiome, and early life modulation of the microbiome may have consequences for health later in life.

The initial colonization of the gut microbiome is influenced by several intrinsic and extrinsic factors, such as delivery mode and feeding patterns9,10,11,12,13. The maternal gut microbiome is an important reservoir of pioneering infant microbes. And the profile of species and strains shared between mother and infant is further influenced by mode of delivery and feeding patterns14. During birth, the neonate first encounters an immense quantity and diversity of maternal microbes from the maternal reproductive tract, gut, and skin15. Vaginally delivered neonates are exposed to maternal vaginal and fecal microbiome, resulting in neonatal gut colonization by vagina-associated microbes (e.g., Lactobacillus and Prevotella), which significantly differ from the microbiome that colonize C-section neonates16. A key factor shaping the neonatal microbiome development is the maternal breastfeeding status (exclusive or partial)17,18.

Modulation by the external environment also plays an important role in microbial colonization. Previous studies have confirmed that exposure to low doses or environmental concentrations of xenobiotics in early life can alter the structure and function of the initial neonatal gut microbiome, with profound implications for host health19,20,21. Iszatt et al. found that 2,4,4’-tribromodiphenyl ether (BDE 28) and perfluorooctanesulfonic acid (PFOS) in breast milk to be associated with a reduced relative abundance of Lactobacillus in 1 month-old infants and to alter the concentration of short-chain fatty acids (SCFAs) in feces19. A longitudinal birth cohort in the United States found that nutrient-toxic elements in toenails were associated with the infant gut microbiome. A negative association between arsenic and alpha diversity was observed in the whole population, and accentuated in infants breast-fed exclusively and exposed to peripartum antibiotics21. Urinary arsenic in infants was related to the composition of the gut microbiome at 6 weeks of age. Eight genera (six within the phylum Bacillota) were enriched under high arsenic exposure, and fifteen genera including Bacteroides and Bifidobacterium were negatively associated with urinary arsenic concentration22. Heavy metals are widely distributed in the environment, and long-term and low-dose exposures in early life can result in embryotoxicity and neurodevelopmental toxicity. At present, the mechanism of how heavy metals affect infant health is not fully understood, and the gut microbiome remains an underexplored target of heavy metals. In general, biological materials including maternal blood, amniotic fluid, placenta, and umbilical cord blood are used to assess prenatal exposure to elements. However, invasive sampling is not appropriate for vulnerable populations such as pregnant women and neonates. Compared to other biological materials, hair samples are collected non-invasively and provide an indication for the retrospective investigation of chronic exposure over several months to years23,24. Thus, maternal hair has the potential to indicate prenatal exposure of infants to trace elements.

Here, we hypothesize that prenatal exposure to trace elements impacts the development of the infant gut microbiome in early life. The present study provides a comprehensive analysis of trace elements and gut microbiome/metabolome/resistome of 146 mother-infant pairs at three postpartum visits (~3, 6, and 12 months). We show the development of the infant gut microbiome, metabolites and antibiotic resistance gene profiles, and associations with prenatal exposure to trace elements and maternal gut microbiome.

Results

Population characteristics

The 146 postpartum women enrolled in this study had a mean age of 32.8 ( ± 3.5) years old, with a relatively normal body mass index (mean 23.1 ± 2.9 kg/m2), more than 12 years of education (97.8%), and an average gestational week of 38.8 ( ± 1.1). The mean birth weight of the infants (N = 146, the sex ratio about 1:1) was 3257 ( ± 397) g. Approximately half of the infants were breast-fed, and 60.2% of the infants were born via vaginal delivery. We collected hair samples from 119 mothers in the postnatal 6 weeks outpatient service, and stool samples from 91 mother-infant pairs at the first postpartum visit (approximately at 3 months postpartum), 58 pairs at the second postpartum visit (approximately at 6 months postpartum), and 30 pairs at the third postpartum visit (approximately at 12 months postpartum). Hair samples and stool samples are matched in the 83 mother-infant pairs. Further characteristics of all participants are shown in Supplementary Data 1.

Gut microbiome profiles of paired mother-infant

During the three postpartum visits, a total of 20,551,471 valid sequences of 353 stool samples were acquired by 16S rRNA sequencing analysis, resulting in 11,355 amplicon sequence variants (ASVs) for the gut microbiome. Gut microbiome composition of mothers and infants was clearly distinct at the phylum level. The phyla Actinomycetota (43.3%) and Pseudomonadota (26.5%) were the dominant taxa within the microbial communities from infants (Fig. 1A), whereas mothers had a higher relative abundance of Bacillota (67.5%) and Actinomycetota (13.6%) (Fig. S1). Furthermore, microbial diversity significantly differed between infants and their paired mothers at different postpartum visits, and the gradual development of infant gut microbiome with increasing age eventually approached that of their mothers (p < 0.05, Fig. 1B and S2). Overlapping regions of the principal coordinate analysis (i.e., PC1 and PC2) showed that the microbiome structure at 12 months of age was closer to their mothers (Figs. 1C and S3). At 12 months, the predominant genera in infant gut were still Bifidobacterium (34.2%) and Escherichia-Shigella (9.55%), but compared to that of the 3 and 6 months of age, the relative abundance of Blautia and Collinsella was increased to 13.5% and 4.02%, respectively, while Klebsiella decreased from 9.96% − 0.40% as well as Enterococcus from 5.56% to 0.68%. Metagenomic data from maternal and infant stool samples also confirmed the trend of development in infant gut microbiome. The diversity of the gut microbiome became richer in older infants, and the dominant phyla shifted from Actinomycetota and Pseudomonadota to Actinomycetota and Bacillota (Fig. 1D, E). Across the infant age, changes in the community structure of bacteria with an overall effect size (Fig. S4, R2 = 0.281, p = 0.001), were similar to the 16S-based sequencing data (R2 = 0.289).

Fig. 1: Composition and diversity of the gut microbiome in paired mothers and infants.
figure 1

A Relative abundance of the top 4 most abundant phyla of the infant gut microbiome at different postpartum visits. Phyla with lower abundance are grouped into the “other” category. B The Shannon diversity in paired mother-infant stool samples, stratified by the visit time into six categories: Infant_First (n = 91), Infant_Second (n = 58), Infant_Third (n = 29), Mother_First (n = 89), Mother_Second (n = 56), and Mother_Third (n = 30). C Principal coordinate analysis of paired mother-infant samples based on the unweighted UniFrac distance of ASVs. The box plots below showed paired mother-infant samples projected on the first principal coordinate. Statistical test was permutational multivariate analysis of variance with FDR correction for multiple test comparisons, F-value = 28.3, R2 = 0.289. D Relative abundance (at genus and phylum level) and (E) Shannon diversity in metagenomic sequencing of 33 infants (First, n = 16; Second, n = 11; Third, n = 6) and 32 mothers (First, n = 13; Second, n = 6; Third, n = 13) at different postpartum visits. The vertical color bar represents the phylum level to which the genus is assigned. For (B), (C), and (E), the box ranges from the 25th to the 75th percentile, and the median value (middle line), while the whiskers extend from each quartile to the minimum and maximum values within 1.5 × interquartile range (IQR) of the box. Data points beyond this range are considered outliers (single points). For (B) and (E), statistical test was Kruskal–Wallis test with FDR correction for multiple test comparisons, p’ indicated the value before FDR correction, two-tailed. Source data are provided as a Source Data file.

Full size image

We found that delivery mode affected the infant alpha diversity (Shannon, Chao1, and Phylogenetic indices), with increased diversity in forceps delivery infants (Fig. 2A and S5). As shown in Fig. S6, the relative abundance of Bacteroides and Escherichia-Shigella (dominant in mothers as well) was markedly different between infants of forceps delivery, vaginal delivery, and cesarean delivery, and between infants of breast-fed and mix-fed. The covariate groups of infants were readily distinguished by principal coordinate analysis (Fig. 2B–D). There was no significant difference in microbial alpha and beta diversity between male and female infants. Notably, these variations in infant gut microbiome were only observed in infants at 3 and 6 months of age, not at 12 months of age. The occurrence in early infancy was related to the impact of delivery mode and feeding patterns on the colonization of the infant gut microbiome.

Fig. 2: Effects of delivery mode, feeding patterns, and sex on the infant gut microbiome at different postpartum visits.
figure 2

A The Shannon diversity of different stratified infants. The box ranges from the 25th to the 75th percentile, and the median value (middle line), while the whiskers extend from each quartile to the minimum and maximum values within 1.5 × interquartile range (IQR) of the box. Data points beyond this range are considered outliers (single points). Statistical test was Kruskal–Wallis test with FDR correction for multiple test comparisons, p’ indicated the value before FDR correction, two-tailed. B−D Principal coordinate analysis of different stratified infants based on the unweighted UniFrac distance of ASVs. Statistical test was permutational multivariate analysis of variance with FDR correction for multiple test comparisons. Number of subjects: Vaginal delivery, n = 56, 36, 21; Cesarean delivery, n = 18, 12, 5; Forceps delivery, n = 17, 10, 3; Breast-fed, n = 31, 17, 6; Mix-fed, n = 43, 30, 15; Male, n = 44, 28, 15; Female, n = 47, 30, 14 of three visits, respectively. Source data are provided as a Source Data file.

Full size image

Impact of trace elements exposure on infant gut microbiome

Hair samples provided by 119 mothers were used to assess prenatal exposure to trace elements in infants (Table 1). Compared to >90% of the other nine elements, the detection frequency of cadmium, arsenic, and nickel was slightly low at 67%, 78%, and 82%. The medium concentrations of mercury, cadmium, lead, arsenic, aluminum, iron, zinc, selenium, copper, manganese, nickel, and barium were 0.33, 0.01, 0.20, 0.03, 5.86, 10.8, 205, 0.484, 9.27, 0.26, 0.12 and 2.59 μg/g, respectively. Except for mercury and selenium, there were statistically significant correlations within trace elements (r = 0.23-0.48, p < 0.05, Fig. S7), such as arsenic and lead, aluminum, and manganese.

Table 1 Concentrations of trace elements in the maternal hairs (N = 119, μg/g)
Full size table

Linear regression model (Supplementary Data 2–4) showed that Shannon diversity of the infant microbiome at 3 months of age was positively correlated with selenium exposure [β = 0.017 per standard deviation increase in selenium concentration, 95% CI (0.002, 0.032), p = 0.023], and negatively correlated with copper exposure [β = -0.032 per standard deviation increase in copper concentration, 95% CI (−0.055, −0.008), p = 0.020]. Associations between most trace elements and alpha diversity were identified using stratified analysis by delivery mode, feeding patterns, and infant sex. For example, in mix-fed infants, prenatal exposure to iron was positively correlated with microbial Chao1 diversity [β = 3.40 per standard deviation increase in iron concentration, 95% CI (2.05, 4.75), p = 4.18e-05] and Phylogenetic diversity [β = 0.452 per standard deviation increase in iron concentration, 95% CI (0.283, 0.622), p = 1.73e-05]. Prenatal exposure to manganese increased microbial Chao1 diversity [β = 48.3 per standard deviation increase in manganese concentration, 95% CI (12.3, 84.4), p = 0.020] and Phylogenetic diversity [β = 5.89 per standard deviation increase in manganese concentration, 95% CI (1.14, 10.6), p = 0.024], and the correlations were stronger and more statistically significant in male infants. Prenatal exposure to mercury was negatively correlated with microbial Shannon diversity [β = −1.31 per standard deviation increase in mercury concentration, 95% CI (-2.50, −0.120), p = 0.048] in female infants. Although no association between arsenic exposure and alpha diversity was observed in the unstratified population, we found an association between arsenic exposure and increased Shannon diversity [β = 3.04 per standard deviation increase in arsenic concentration, 95% CI (1.02, 5.07), p = 0.014] in forceps delivery infants.

To further explore the impact of trace elements on the dynamics in infant gut microbiome, we tested the difference of microbial diversity among low (<P25), medium (≥ P25 – ≤ P75), and high (>P75) exposure groups. There was no significant difference in microbial alpha and beta diversity for all exposure groups (Fig. S8), except for the Shannon diversity among the copper exposure group at the first postpartum visit (Fig. 3). The Shannon diversity of the infant gut microbiome in the high copper exposure group (>11.6 μg/g) was a statistically significant reduction than that of the medium and low exposure groups (p = 0.046, Fig. 3A). Following two postpartum visits, the Shannon diversity of the high exposure group showed an apparent and continuing upward trend, especially from 6 to 12 months of age (Fig. 3B–D). As expected, by the third postpartum visit, microbial diversity in infants tended to a relatively stable plateau among the copper exposure group.

Fig. 3: The Shannon diversity of the infant gut microbiome among the copper exposure groups.
figure 3

A The differences in diversity between the copper exposure groups at the three postpartum visits. The box ranges from the 25th to the 75th percentile, and the median value (middle line), while the whiskers extend from each quartile to the minimum and maximum values within 1.5 × interquartile range (IQR) of the box. Data points beyond this range are considered outliers (single points). Statistical test was Kruskal–Wallis test with FDR correction for multiple test comparisons, two-tailed. B–D showed the trends of diversity in each group during the three postpartum visits, respectively. Number of subjects: High copper, n = 23, 16, 10; Medium copper, n = 41, 26, 13; Low copper, n = 17, 9, 3 of three visits, respectively. Source data are provided as a Source Data file.

Full size image

We also found differences in the relative abundance of specific bacterial taxa between low (<P25) and high (> P75) exposure groups. The infant gut may harbor the most differentially abundant genera classified as Bacillota and Actinomycetota (the genera with the relative abundance > 0.5%). As shown in Fig. 4, prenatal exposure to high level of aluminum significantly increased the relative abundance of Bifidobacterium, Erysipelatoclostridium, and Cutibacterium, and decreased the abundance of Hungatella at 3 months of infant age. High level of lead exposure was associated with increased relative abundance of Ruminococcus_gnavus_group, Erysipelatoclostridium, and Cutibacterium. For the high iron group, there was a significant depletion in the relative abundance of Enterococcus. Bifidobacterium and Erysipelatoclostridium were higher in relative abundance for infants exposed to high level of manganese. Due to the high variability of the gut microbiome during infancy, prenatal exposure to trace elements had varying effects on specific bacterial taxa at different postpartum visits (Fig. S9).

Fig. 4: Prenatal exposure to trace elements affected specific bacterial taxa.
figure 4

Significantly altered genera between high and low exposure groups (manganese, mercury, aluminum, barium, lead, iron, copper, and nickel) were identified by linear discriminant analysis effect size analysis (LEfSe, LDA > 2, p < 0.05) at the first postpartum visit. Bacterial taxa with Unclassified in their name were not displayed. Source data are provided as a Source Data file.

Full size image

Metabolic and functional profiles of the gut microbiome

A total of 3867 metabolites were identified in 198 stool samples. At the first, second, and third postpartum visit, metabolic profiles could be readily differentiated in infant samples, not in maternal samples (Fig. S10). The intersection of maternal and infant metabolome appeared at the third postpartum visit. Fig. 5A, B illustrated the variability in metabolic profiles of the infant gut microbiome at different ages (fold change >1.5, p < 0.05), with 56 metabolites (54 upregulated and 2 down-regulated) changed significantly between 3 and 6 months of age, and 515 varied metabolites (466 upregulated and 49 down-regulated) between 6 and 12 months of age. Linking specific metabolic products to gut microbiome was conducted by calculating the Spearman correlation coefficient (Fig. 5C, D). Clear correlations between most of the differential metabolites and dominant genera of infants were identified, including 43 metabolites between the first and second postpartum visit, and 49 metabolites between the second and third postpartum visit. These products of the gut microbial metabolism were diverse in Supplementary Data 5–6, such as amino acids, fatty acids, carbohydrates, flavones, and bile acids. For example, lacto-N-fucopentaose III as a polysaccharide (decreased by 1.99-fold from the first to second postpartum visit) was associated positively with Streptococcus, and negatively with the genera Bifidobacterium, Blautia, and Escherichia-Shigella. Similarly, violacein assigned as a hydroxyindole (increased by 2.98-fold) was positively related to Blautia. Compared to the infants at 6 months of age, 3-hydroxy-6,8-dimethoxy-7(11)-eremophilen-12,8-olide belonging to a member of terpene lactones was increased by 3.00-fold at 12 months of age, related positively to Blautia and negatively to Enterococcus. Prenatal exposure to trace elements affected a variety of metabolites in the infant gut (Fig. S11). For example, selenium showed positive associations with 4-trimethylammoniobutanoic acid, citramalic acid, N-methyl-2-pyrrolidinone, and 7-methylguanine, which belong to the fatty acid, pyrrolidone, and purine. Similarly, cadmium exhibited a positive association with lacto-N-difucohexaose, a polysaccharide. Heatmaps of the metagenomics-based KEGG pathways showed that the metabolome in the infant gut microbiome at 6 and 12 months was enriched in important pathways for infant growth and development, such as the metabolism of amino acid, lipid, carbohydrate, and vitamin (Fig. 6A). The infant gut metagenome was observably expanded and diversified from the age of 3 months to 12 months, while both taxonomic composition and metabolic capacity in infants remained distinct from the maternal gut microbiome (Fig. 6B, C).

Fig. 5: Infants showed different metabolic profiles at different postpartum visits.
figure 5

A 56 metabolites were significantly changed between the first and second postpartum visit, and (B) 515 metabolites were significantly changed between the second and third postpartum visit (fold change > 1.5, VIP > 1, p < 0.05). Statistical tests were Student’s t-test and orthogonal partial least squares discriminant analysis (OPLS-DA) with FDR correction for multiple test comparisons, two-tailed. C 43 metabolites (from the top 50 differential metabolites by fold change) were associated with the top 10 genera of the infant gut microbiome between the first and second postpartum visit, and (D) 49 metabolites (from the top 50 differential metabolites by fold change) were associated with the top 10 genera of the infant gut microbiome between the second and third postpartum visit. Statistical test was Spearman correlation analysis with FDR correction for multiple test comparisons, *p < 0.05, **p < 0.01, ***p < 0.001. Number of subjects: First postpartum visit, n = 41; Second postpartum visit, n = 41; Third postpartum visit, n = 17. The metabolites abbreviated in the figure have been given their full names in Supplementary Data 5–6. Source data are provided as a Source Data file.

Full size image
Fig. 6: Expansion and diversification of microbial community structure and function in paired mother-infant.
figure 6

A Distinct microbial gene profiles (according to KEGG pathway analysis) of the maternal and infant stool metagenomes at different postpartum visits. The relative abundance of pathways in the given sample was colored by its row Z-score (row mean/row SD). The vertical color bar represents the higher-order KEGG module belonging to the corresponding pathway. B Principal coordinate analysis of Bray-Curtis distances based on relative abundances of pathways demonstrates clustering by time point. Statistical test was permutational multivariate analysis of variance with FDR correction for multiple test comparisons, F-value = 1.73, R2 = 0.127. C Enumeration of the observed species within the maternal and infant stool samples. The box ranges from the 25th to the 75th percentile, and the median value (middle line), while the whiskers extend from each quartile to the minimum and maximum values within 1.5 × interquartile range (IQR) of the box. Data points beyond this range are considered outliers (single points). Statistical test was Kruskal–Wallis test with FDR correction for multiple test comparisons, p’ indicated the value before FDR correction, two-tailed. Number of subjects: Infant_First, n = 16; Infant_Second, n = 11; Infant_Third, n = 6; Mother_First, n = 13; Mother_Second, n = 6; Mother_Third, n = 13. Source data are provided as a Source Data file.

Full size image

Antibiotic resistance gene profiles in gut microbiome

We characterized antibiotic resistance genes (ARGs) profiles in the gut microbiome of 33 infants and 32 mothers. A total of 263 ARGs were detected and 150-160 ARGs were overlapping between infants and mothers (Fig. 7A, B). Specifically, ARGs conferred resistance to 33 drug classes, and the most abundant gene was tetracycline (Fig. 7C), followed by fluoroquinolone, rifamycin, and peptide in the infant gut, and by macrolide, lincosamide, and streptogramin in the maternal gut. The most prevalent mechanism of antibiotic resistance was antibiotic inactivation and antibiotic efflux pumps (Fig. 7D). Principal coordinate analysis highlighted a significant difference in relative abundance of ARG profiles between infants and mothers (Bray-Curtis, p = 0.001, Fig. 7E). The increased proportion of ARGs in infants at 12 months of age was macrolide, lincosamide and streptogramin, tending to resemble that of their mothers. Correlation analysis showed that the majority of ARGs in the infant gut may originate from members of family Enterobacteriaceae such as Escherichia and Klebsiella (Fig. S12).

Fig. 7: Characterization of ARG profiles in the stool samples of paired mother-infant.
figure 7

A Heatmap depicting the relative abundance of 263 ARGs across the samples. B Analyses of the unique and overlapping ARGs. “Unique” represents ARGs that were only present in mother or infant gut regardless of prevalence and abundance. C The relative abundance of ARGs associated with different drug classes at different postpartum visits. D Mechanism of resistance of each ARG, depicted as a proportion of all ARGs detected in the cohort. E Principal coordinate analysis based on Bray-Curtis distance of ARGs relative abundances in the maternal and infant stool samples. Statistical test was permutational multivariate analysis of variance with FDR correction for multiple test comparisons, F-value = 4.38, R2 = 0.270. Number of subjects: Infant_First, n = 16; Infant_Second, n = 11; Infant_Third, n = 6; Mother_First, n = 13; Mother_Second, n = 6; Mother_Third, n = 13. Source data are provided as a Source Data file.

Full size image

A slight difference in the ARGs richness between infants and their mothers was observed at the age of 3 months, where alpha diversity of multiple-drug resistance genes (MDR ARGs) was higher in the infant gut (p = 0.074, p = 0.005 before FDR correction, Fig. S13A). We discovered that 27.8% (73 types) of all ARGs were known as MDR ARGs, with the most common type conferred to rifamycin, peptide, and fluoroquinolone (Fig. S13B). Almost three-quarters of MDR ARGs overlapped, and unique MDR ARGs in the infant gut were higher at the age of 3 and 6 months (Fig. S13C). Prenatal exposure to trace elements affected the incidence of ARGs. For example, manganese exposure was negatively associated with tetW in the gut of 3 month-old infants. In 6 month-old infants, copper exposure was positively associated with AcrF and acrD, and arsenic exposure was positively associated with mdtB (Fig. S14). We further calculated the Jaccard distance to reflect the dissimilarity of DNA sequence profiles in the ARGs of paired mother-infant at 3 months postpartum (N = 8 pairs). In the high selenium exposure groups, the DNA sequence profiles of gene dissimilarity of paired mother-infant ARGs were higher, while the dissimilarity decreased in the high nickel exposure groups (Fig. S15).

Discussion

We recruited 146 mother-infant pairs and collected hair samples from 119 mothers in outpatient service. A total of 12 trace elements in maternal hairs were determined to evaluate prenatal exposure of the infants, and their association with the colonization of the gut microbiome. A longitudinal follow-up cohort, coupled with 16S rRNA gene amplicon and metagenomic sequencing, allowed us to characterize microbial composition and diversity in mothers and infants. Finally, the microbial metabolite profiles were tracked from 41 mother-infant pairs during three postpartum visits. Microbial alpha diversity in infants was associated with selenium, manganese, and copper. There were significant differences in the relative abundance of the gut microbiome between low (<P25, percentile 25 of element concentrations) and high (>P75, percentile 75 of element concentrations) exposure groups for various elements.

The levels of trace elements in maternal hair were within normal limits compared with previous studies25,26. Several external factors, including dietary intake and environmental exposure (e.g., air pollution and dust), may influence the bioaccumulation of trace elements in hair during pregnancy25. For nutrient elements, the association between copper and alpha diversity showed that infant gut exposed to high copper harbored a low abundance of multiple genera within Bacillota, such as Clostridium_innocuum_group, Coprobacillus, and Fusicatenibacter. Trends of this association showed that copper influence on alpha diversity did not completely disappear at 1 year old. Dietary copper has shown profound effects on microbial composition, and alterations caused by low or high copper exposure were found to increase the abundance of Bacillota and deplete the probiotic Akkermansia, respectively27. In addition, high manganese was associated with increased alpha diversity and specific bacteria in infants, especially for Bifidobacterium, there was a 23.9% difference in relative abundance between the high and low exposure groups (p = 0.025). Manganese is a redox active metal and responsible for the growth of bacteria like probiotics (e.g., Lactobacillus) or pathogens (e.g., Staphylococcus sp., Shigella, and Escherichia coli)28,29. Dietary exposure to extremely high levels of manganese (200 mg/L) has been reported to alter the microbial composition and further activate the NLRP3 inflammasome, triggering an inflammatory response in the central nervous system30; in contrast, no alterations were found in gut microbiome composition of mice from dietary manganese31. Given the importance of Bifidobacterium strains in the anti-inflammatory, this epidemiological study provided scientific support for the idea that gut microbiome weakens the manganese-induced inflammatory response. With respect to iron exposure, we found that the relative abundance of Enterococcus was lower in the high-iron exposure group. Iron fortification may enrich pathogenic E. coli strains and induce intestinal inflammation32. Iron supplements may disturb the composition and diversity of the gut microbiome and, a reduced abundance of Lactobacillus was found in pre-weaning and young adult animals fed iron in early life33.

For toxic elements, mercury was negatively associated with Shannon diversity in female infants. It was previously proposed that exclusive breastfeeding could confer some protection against mercury, or that formula supplementation imparted susceptibility. Correspondingly, in the subpopulation of non-exclusively breast-fed infants, mercury was related to decreased within-subject diversity21. The association between mercury and alpha diversity was not modulated by feeding patterns in our study. Both inorganic mercury (IHg) and methylmercury (MeHg) adversely affect the gastrointestinal tract in several animal studies34. An increase in methylating bacteria (e.g., sulfate-reducing bacteria) corresponded with the formation of MeHg28. With respect to lead, we found that prenatal exposure to high level of lead was associated with increased abundance of Ruminococcus_gnavus_group, Erysipelatoclostridium, and Cutibacterium. Higher lead level in tooth of infants corresponded with an increased abundance of Collinsella and Bilophila and a decreased abundance of Bacteroides20. For arsenic, it was negatively associated with microbial alpha diversity, and the association was stronger significant in infants exclusively breast-fed and exposed to peripartum antibiotics21. Furthermore, high arsenic exposure decreased the relative abundance of the family Clostridiaceae and the genera Bacteroides and Bifidobacterium22. We also found a positive association between arsenic exposure and Shannon diversity in forceps delivery infants. Disturbance of gut microbiome by aluminum may exacerbate the toxicity associated with these metals35. Our study found no association between high aluminum and infant microbial diversity but did find a significant decrease in the relative abundance of Hungatella and an increase in the relative abundance of Bifidobacterium, Cutibacterium, and Erysipelatoclostridium in the high-aluminum exposure group. Based on an in vitro colonic fermentation model, there was a significant differential clustering of microbial composition in the human gut, that dietary aluminum exposure significantly increased the abundance of Prevotella, Escherichia, and Collinsella, and decreased the abundance of Streptococcus, Roseburia, and Ruminococcus36. Accordingly, the gut microbiome has been evidenced as a crucial mediator of the bioavailability and toxicity of these metals. Aluminum decreased the microbial diversity and perturbed the overall community structure of the gut microbiome, such as Turicibacter, Lactobacillus murinus, Lactobacillus reuteri, and Bifidobacterium pseudolongum37. These findings indicated that high levels of nutrient elements (selenium, iron, and manganese) may improve gut microbial diversity in infants at 3 months of age, but not for high level of copper. Prenatal exposure to toxic elements aluminum, mercury, and lead may induce the enrichment of potentially detrimental bacteria (e.g., Erysipelatoclostridium and Ruminococcus_gnavus_group) in the infant gut. It is the first evidence of the association between element exposure and gut microbial colonization in a Chinese cohort of healthy mother-infant dyads. These findings are preliminary and further studies with larger cohorts are needed.

The complexity of gut microbiome development in infants was apparent12,38,39. Gut microbial community was usually dominated by the phyla Actinomycetota, Pseudomonadota, and Bacillota, corresponding with the genera Bifidobacterium, Escherichia-Shigella, Klebsiella, Blautia, and Streptococcus. These genera in the infant gut varied considerably at different postpartum visits. Especially at 12 months of age, a shift of microbial composition toward that of mothers attributed to the decreased Bifidobacterium and increased Blautia. The striking differences in both diversity and composition were found between mothers and infants at 3 months of age and subsequently diminished as the infant grew up to 12 months of age. The bacterial diversity increased with age in different populations, and the distance and variance in the gut microbiome among infants were greater than that of mothers40. Maternal and infant microbial communities with clear distinction were statistically stable but not similar41. Infants at 6 and 12 months had a low relative abundance of Bacteroides compared to the previous reports13,42,43. Bacteroides plays a crucial role in gut homeostasis and are susceptible to dietary patterns and drug use, information on which was not collected during the postpartum visit. Following transition from exclusive breastfeeding to partial and/or complete solid foods44, the similar observation was a downward trend of Bifidobacterium and Escherichia.

The colonization of the infant gut microbiome was influenced by delivery mode and feeding patterns, illustrated by a gradual decrease in their effects on the overall microbial composition as infants grew up. Gut bacteria of vaginal born infants during the first year of life were enriched in the genera Bacteroides, Escherichia, and Parabacteroides. Their abundances changed dynamically within the first 3 months, with the greatest difference in the microbial profiles between vaginal- and cesarean-born infants14. The relative abundance of Escherichia in vaginally born infants (10.1%) was much higher than that of cesarean born infants (2.07%) at the age of 3 months. By 12 months of age, the abundance of Escherichia in cesarean born infants increased to 8.43%, and differences in community structure caused by delivery mode disappeared. Compared to exclusively or partially breast-fed infants, microbial composition and diversity of 3 month-old infants weaned from breastmilk differed; When longer than 9 months, the impact of breastfeeding lasted until the infant at 1 year of age12. Microbial composition difference between mix- and breast-feeding infants was observed at 3 months of age, which did not persist for long. Sex differences are still inconclusive17,45, and this study found no differences in alpha and beta diversity between male and female infants. However, these findings were based on the relative abundance, and the absolute changes in these indicators remain unclear. In particular, the quest for differential biomarkers could be impacted by individual differences across studies and high microbial variability, due in part to their diet and health status.

Gut microbes make key contributions to the metabolism of ingested compounds that are linked to health benefits. However, our knowledge of how the gut microbiome–related metabolome develops within the first year of life remains unclear. We tracked the dynamics of stool metabolites in 41 mother-infant pairs at the ages of 3, 6, and 12 months. The metabolic profiles of the infants were differentiated over time, in contrast to the highly conserved profiles of mothers. A variety of differential metabolites upregulated with the infant age, mostly belonging to amino acids and fatty acids metabolic pathways. Interestingly, the consistent development of the microbiome and metabolome in the infant gut corroborated their mutual regulations. At 1 year of age, the relative abundance of Blautia increased, in association with 46 differential metabolites (e.g., amino acids, fatty acids, and flavones). Several longitudinal studies have also identified the effects of delivery mode, antibiotic use, and feeding patterns on the infant fecal metabolome and associated development46,47. Compared to the formula-fed infants, carbohydrate metabolic pathways were enriched in breast-fed infants48. Cesarean delivery was associated with lower fecal concentrations of tryptophan, bile acids, and phenylalanine, potentially attributed to the low abundance of the genera Bifidobacterium and Bacteroides46. We also found associations between prenatal exposure to trace elements and multiple metabolites in infant stools (e.g., 4-trimethylammoniobutanoic acid, citramalic acid and selenium, palmitic acid and cadmium, gluconic acid and iron). For example, gluconic acid is slowly fermented by Lactobacillus in the gut to produce lactate and acetate, which are essential substrates for butyrate production49. Fatty acids such as citramalic acid, palmitic acid, and 4-trimethylammoniobutanoic acid in the host and commensal bacteria regulate gut immune responses and diseases50,51,52. These findings suggest that exposure to trace elements not only perturbs the gut microbiome of infants but also substantially alters their metabolomic profiles, resulting in the disruption of metabolite homeostasis.

Human gut microbiome is a major reservoir for antibiotic resistance genes, including in infants53. We examined the abundance and distribution of ARGs in the gut of infants and their mothers during the first year after delivery. In this assemblage, a high abundance of tetracycline resistance genes was presented in the maternal and infant gut. This mirrors previous findings that ARGs are more prone to exist in the species of Pseudomonadota phylum. Similarly, the family Enterobacteriaceae contributed the largest abundant ARGs in the infant gut, such as Escherichia, Klebsiella, Citrobacter, and Enterobacter54,55. Evidences indicate a bimodal distribution of the adult and infant gut resistome, mainly driven by E. coli composition56,57. The presence of ARGs was predictive of neonatal sepsis and adverse birth outcomes in low- and middle-income countries58. Colonization with ARG-carrying bacteria may cause a delayed maturation of the infant microbiome, associated with an increased risk of asthma56,59. The infant gut resistome is significantly affected by environmental factors, such as antibiotic use and heavy metal exposure57,60. In this study, prenatal exposure to trace elements (e.g., copper, manganese, zinc, arsenic) showed strong correlations with several ARGs, such as tetW, mdtB, and acrD. Heavy metals as a selective agent in the proliferation of antibiotic resistance have become a compelling concern. Ding et al. found that high soil concentrations of copper, zinc, and cadmium increased the richness and abundance of ARGs in the collembolan gut through the co-resistance mechanism61. In particular, copper could increase conjugation and permeability of the cell membrane to facilitate the transfer of exogenous mobile genetic elements (MGEs) into bacterial cells while activating integron, and further induce co-selection of gut-associated ARGs through horizontal transfer of MGEs61,62. Most of previous studies were based on the natural environment or soil/aquatic animals, and our study is the first description of the association between metal exposure and the abundance of ARGs in infant gut. Despite a small cohort, it does provide a longitudinal view that prenatal exposure to trace elements may increase the abundance of antibiotic resistance genes in the infant gut. A large cohort is necessary to examine the robustness of the associations between xenobiotics, gut microbiome and ARGs in early life.

In conclusion, this study highlighted the emerging associations of trace element exposure (e.g., copper, selenium, and manganese) with the dynamics of infant gut microbial diversity, composition, and metabolic profile. We characterized the relationships between multiple elements and ARG abundance, this is the first description of the element exposure effect on the enrichment of ARGs in the infant gut. Yet more studies with a large birth cohort and careful considerations for the robustness of these novel findings are still needed. Future research is needed to understand the direct associations of xenobiotic exposure in early life with infant gut microbiome and long-term health effects.

Methods

Ethical statement

This study was approved by the Institutional Review Board of Peking University People’s Hospital (approval number: 2021PHB254-001) and complied with all relevant ethical regulations.

Study cohort and data collection

Hair and stool samples collected for this study were obtained from a nationally funded prospective clinical cohort study designed to assess the quality of postpartum recovery, initiated and conducted by Peking University People’s Hospital. All participants were recruited between November 2021 and March 2023, and enrolled after signing the informed consent documents, collecting demographic information and clinical records of the study population.

Hair and stool sample collection

Hair samples were collected by skilled professionals. Sanitary ceramic scissors pre-cleaned with 1% nitric acid and ultrapure water were used to cut the hair from 1–2 cm around the occipital area. The collected samples were placed into the sealed polyethylene plastic bags, transported to the laboratory in a medical transport box, and stored at −20 °C until analysis. At a regularly scheduled postpartum visit (approximately at 3, 6, and 12 months), mothers were asked to collect paired mother-infant stool samples (an infant stool sample in a provided diaper) at home. The samples were stored in the household freezer (-20 °C) immediately and then uninterrupted frozen storage (incubator with ice pack), chaperoned to the laboratory within 24 h, where stool samples were stored at -80 °C until analysis.

Trace elements analysis

119 hair samples were washed successively with acetone, ultrapure water for three times, and acetone, recommended by the International Atomic Energy Agency (IAEA)63. After freeze-drying to a constant weight without dust, preparations of the hair samples for the analysis of 11 trace elements were according to the report of Zheng et al.26. In brief, the first 10 cm of the hair sample was cut into small pieces, exactly 0.100 g of hair was added into the microwave digestion tube, mixed with 4 mL of nitric acid (69%, granted reagent) and 1 mL of 30% hydrogen peroxide (granted reagent). Further details of the digestion procedure can be consulted in the Supplementary Note 1. The concentrations of cadmium, lead, arsenic, aluminum, iron, zinc, selenium, copper, manganese, nickel, and barium were analyzed via inductively coupled plasma mass spectrometry (ICP-MS). The parameters of the ICP-MS analysis were given in the Supplementary Note 2. In addition, mercury determination was conducted following the EPA Method 747364. Approximately 0.020 g of precleaned and dry hair was weighed into a quartz boat and analyzed using an RA-915M Mercury Analyzer (Lumex, Canada) consisting of thermal decomposition, amalgamation, and atomic absorption spectrophotometry.

For the ICP-MS analysis, procedural blanks were prepared along with each batch of hair samples. The limit of detection (LOD) ranged from 0.001 μg/L (cadmium) to 0.567 μg/L (iron). The certified reference material (CRM, NCS ZC 81002b, China) was included every 16 samples, with recoveries ranging from 71.6% to 121% and a good precision of less than 20%. For the mercury analysis, blank samples (empty quartz boat) were below the LOD value (2.00 ng/g) with 115% of the CRM recovery.

16S rRNA gene sequencing and data processing

For 353 stool samples and negative control samples, genomic DNA was extracted using the E.Z.N.A.® Soil DNA Kit (Omega Bio-Tek) according to the manufacturer’s instructions. The hypervariable region V3-V4 of the bacterial 16S rRNA gene was amplified with primer pairs 338 F (ACTCCTACGGGAGGCAGCAG) and 806 R (GGACTACHVGGGTWTCTAAT) by an ABI GeneAmp® 9700 PCR thermocycler. Purified amplicons were pooled in equimolar and paired-end sequenced on an Illumina MiSeq PE300 platform (Illumina Inc., San Diego, CA, USA) according to the standard protocols by Majorbio (Shanghai, China). The raw 16S rRNA gene sequencing reads were demultiplexed, quality-filtered by fastp version 0.20.065 and merged by FLASH version 1.2.766. DADA2 was used for denoising amplicon sequence variants (ASVs) using a previously described standard protocol67,68. Taxonomic assignment of ASVs was performed using the naive bayes consensus taxonomy classifier implemented in the Qiime2 with a confidence threshold of 0.7 against the SILVA 138 database69. More details can be found in the Supplementary Information.

Metagenomic sequencing and data processing

We further performed the metagenomics on a random subset of 65 stool samples. This subset included 16 infant stool samples and 13 mothers collected at the first postpartum visit, and 11 infant stool samples and 6 mothers collected at the second, and 6 infant stool samples and 13 mothers collected at the third, respectively. All DNA extractions were used for paired-end library construction following NEXTFLEX™ Rapid DNA-Seq Kit (Bioo Scientific, Austin, TX, USA) protocol. Paired-end sequencing was performed on Illumina Novaseq Xplus (Illumina Inc., San Diego, CA, USA) at Majorbio (Shanghai, China) using Illumina NovaSeq Reagent Kits. Raw sequencing reads were trimmed of adapters, and low-quality reads (with a quality value <20 or having N bases) were removed by fastp (v0.20.0)65. Reads were aligned to the human genome by BWA (v0.7.17)70 and any hit associated with the reads and their mated reads were removed. The quality-filtered data were assembled using MEGAHIT (v1.1.2)71, and the contigs with length >300 bp were selected as final assembly result. Open reading frames (ORFs) from each assembled contig was predicted using Prodigal (v2.6.3)72, and ORFs >100 bp in length were retrieved. A non-redundant gene catalog was constructed using CD-HIT (v4.7)73 with 90% identity and 90% coverage of sequence. Gene abundance for a certain sample was estimated by SOAPaligner (v2.21)74 with 95% identity. Taxonomic assignments for non-redundant genes were obtained by aligning the integrated NCBI NR database (v20230830) using DIAMOND (v0.8.35)75 with an e-value cut off of 1e-05. The taxonomic level of each gene was determined based on the best hit algorithms from the significant matches. Similarly, functional assignments of non-redundant genes were obtained using DIAMOND with the KEGG (v94.2) and CARD (v3.0.9) databases, with an e-value cut off of 1e-05. A unigene was annotated as an ARG if the best BLASTP hit showed at least 80% identity over a query coverage of 80%76. Additionally, DNA sequence profiles based on kmers were produced from the metagenomic reads using sourmash (v4.4.0)77 with option -kmer 31. Subsets of reads mapping to ARGs were used to produce the profiles for ARGs.

Metabolomic analysis

Metabolomic profiles were recorded essentially using the ultra-high performance liquid chromatography tandem mass spectrometry (UHPLC-MS/MS). Due to the insufficient stool collection, 164 samples from 41 mother-infant pairs at the first and second postpartum visit, and 34 samples from 17 pairs at the third postpartum visit were included. Briefly, 50 mg stool were accurately weighed and extracted using a 400 µL methanol: water (4:1, v/v) solution with 0.020 mg/mL L-2-chlorophenylalanin as internal standard. The mixture was allowed to settle at -10 °C and treated by high throughput tissue crusher at 50 Hz for 6 min, then followed by ultrasound at 40 kHz for 30 min at 5 °C. The samples were placed at -20 °C for 30 min to precipitate proteins. After centrifugation at 13,000 g at 4 °C for 15 min, the supernatant was carefully transferred to sample vials for UHPLC-MS/MS analysis. Quality control (QC) sample was prepared by mixing equal volumes (20 μL) of all samples to be a pooled sample and then analyzed using the same method with the analytic samples. The QCs were injected at regular intervals (every 15 samples) throughout the analytical run to provide a set of data for repeated evaluation. For overall data, when relative standard deviation was <0.3, the cumulative proportion of peaks was >70%. Further details on UHPLC-MS/MS parameters and metabolite annotation were provided in the Supplementary Notes 5 and 6, respectively.

Statistics and reproducibility

Statistical analyses were performed using IBM SPSS Statistics (v27.0), R (v4.2.1), and GraphPad Prism (v10.1.2). Target elements with detection frequency (DF) >50% were included and undetectable concentrations were assigned as half of the LOD/2. Under the premise of a linear relationship between trace elements and alpha diversity (Shannon, Chao1, and Phylogenetic)19,21, linear regression model (backward) was conducted to investigate the relationship between each element and alpha diversity, adjusted by other elements and multiple covariates, including maternal alpha diversity, feeding patterns, delivery mode and infant sex. According to the postpartum visit, mothers and infants were divided into six categories as follows: Mother_First and Infant_First (at 3 months postpartum), Mother_Second and Infant_Second (at 6 months postpartum), Infant_Third, and Mother_Third (at 12 months postpartum). The infants were further divided into low (<P25), medium (≥P25 − ≤P75), and high (>P75) exposure groups according to the percentile of trace element concentrations, following to the methodology of the literature19,78. Group differences in microbial alpha diversity, observed species, and Jaccard distance were examined by Wilcoxon rank sum test or Kruskal–Wallis test, in beta diversity (generalized unweighted UniFrac or Bray-Curtis distances) by permutational multivariate analysis of variance (PERMANOVA/Adonis analysis), and in the relative abundance of bacteria by linear discriminant analysis effect size (LEfSe). Statistical significance was set at p < 0.05 after the Benjamini-Hochberg FDR correction. No statistical method was used to predetermine the sample size. No data were excluded from the analyses. The experiments were randomized. This was an observational study, participants received no interventions, and the investigators were not blinded to allocation during experiments and outcome assessment.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.