Influence of perinatal and childhood exposure to tobacco and mercury in children's gut microbiota

Abstract

Background: Early life determinants of the development of gut microbiome composition in infants have been widely investigated; however, if early life pollutant exposures, such as tobacco or mercury, have a persistent influence on the gut microbial community, its stabilization at later childhood remains largely unknown.

Objective: In this exposome-wide study, we aimed at identifying the contribution of exposure to tobacco and mercury from the prenatal period to childhood, to individual differences in the fecal microbiome composition of 7-year-old children, considering co-exposure to a width of established lifestyle and clinical determinants.

Methods: Gut microbiome was studied by 16S rRNA amplicon sequencing in 151 children at the genus level. Exposure to tobacco was quantified during pregnancy through questionnaire (active tobacco consumption, second-hand smoking -SHS) and biomonitoring (urinary cotinine) at 4 years (urinary cotinine, SHS) and 7 years (SHS). Exposure to mercury was quantified during pregnancy (cord blood) and at 4 years (hair). Forty nine other potential environmental determinants (12 at pregnancy/birth/infancy, 15 at 4 years and 22 at 7 years, such as diet, demographics, quality of living/social environment, and clinical records) were registered. We used multiple models to determine microbiome associations with pollutants including multi-determinant multivariate analysis of variance and linear correlations (wUnifrac, Bray-Curtis and Aitchison ß-diversity distances), single-pollutant permutational multivariate analysis of variance adjusting for co-variates (Aitchison), and multivariable association model with single taxa (MaAsLin2; genus). Sensitivity analysis was performed including genetic data in a subset of 107 children.

Results: Active smoking in pregnancy was systematically associated with microbiome composition and ß-diversity (R² 2-4%, p < 0.05, Aitchison), independently of other co-determinants. However, in the adjusted single pollutant models (PERMANOVA), we did not find any significant association. An increased relative abundance of Dorea and decreased relative abundance of Akkermansia were associated with smoking during pregnancy (q < 0.05).

Discussion: Our findings suggest a long-term sustainable effect of prenatal tobacco exposure on the children's gut microbiota. This effect was not found for mercury exposure or tobacco exposure during childhood. Assessing the role of these exposures on the children's microbiota, considering multiple environmental factors, should be further investigated.

Keywords: 16S rRNA; birth cohort; children; diet; gut microbiota; mercury; second-hand smoke; smoking during pregnancy.

Figure 1

Graphic representation of the gut microbiome association analysis workflow in the Sabadell-INMA cohort (Sabadell, Catalonia, Spain). Data for gut microbiome at the genus level, exposure to pollutants (tobacco and mercury) and possible non-genetic determinants (clinical records, demographics, diet, and quality of living) were available for 151 children and were used in the main association analysis. For pollutant exposure and non-genetic determinants, missing data were imputed and included in the β-diversity association study, whereas the original data were included in the genus association study. When using imputations, an association was considered significant when a p-value <0.05 was observed in more than 17 out of 20 imputations. Sensitivity analysis was performed including genetic data of a subset of 107 children. A p-value or q-value (adjusted by confounders) <0.05 was considered, as appropriate. Full description of the statistical analysis is available in the materials and methods section.

Figure 2

Early life and childhood determinants and pollutant exposures investigated in this study. In the pie chart, the number of variables each period and type are indicated. Data for each pollutant exposure and non-genetic determinant, or the imputed value in the case of missing value, were available for 151 children (see statistical analysis section). Original data of fish intake at 32 weeks of pregnancy were also available for 148 children. Original genetic data were available for a subset of 107 children. Full description of the variables is presented in Supplementary File 1; Table S1.

Figure 3

Effect size (R²) of microbiome β-diversity multiple determinants identified in the school-aged children of the Sabadell-INMA cohort. The analysis was based on the envfit function in the vegan R package. Factors are sorted according to their effect size and colored based on metadata category. Mean R² was obtained for 20 imputed datasets considering, Aitchison distance (n = 151 children) Aitchison. Active smoking during pregnancy and SHS at 4 years based on urinary cotinine levels were associated with the β-diversity (Aitchison distance, p-value: 0.0045 and 0.0322, respectively). Having siblings at birth was also associated with the β-diversity (Aitchison distance, p-value: 0.0065).

Figure 4

Fecal microbiome β-diversity from 7-year-old children, categorized by the exposure to active smoking of their mothers at any time during pregnancy. Plots of principal component analysis (PCA) with the clr-transformed features from children’s feces were calculated for one imputation set. The 95% confidence interval prediction ellipses were calculated for each category. Amplicon sequence variants (ASVs) were collapsed at the genus level. Each children’s microbial community is represented by a dot. Length of the arrows shows the strength of the association between each genus and the differences in microbiota composition: the arrow length is proportional to the variance explained by each specific genus; the arrow angle is correlated with the distribution of this variance between PC1 and PC2. Akkermansia, Phascolarctobacterium, Dialister, Clostridium sensu-stricto_1, and Haemophilus were driving the differences between children’s gut microbiomes. No significant association was found (PERMANOVA, p-value > 0.05 in the crude or adjusted univariate model).

Figure 5

Fecal microbiome β-diversity from 7-year-old children, categorized by having siblings at birth. Plots of principal component analysis (PCA) with the clr-transformed features from children’s feces were calculated for one imputation set. The 95% confidence interval prediction ellipses were calculated for each category. Amplicon sequence variants (ASVs) were collapsed at the genus level. Each children’s microbial community is represented by a dot. Length of the arrows shows the strength of the association between each genus and the differences in microbiota composition: the arrow length is proportional to the variance explained by each specific genus; the arrow angle is correlated with the distribution of this variance between PC1 and PC2. Akkermansia, Phascolarctobacterium, Dialister, Clostridium sensu-stricto_1, and Haemophilus were driving the differences between children’s gut microbiomes. No significant association was found (PERMANOVA, p-value > 0.05 in the crude or adjusted univariate model).

Figure 6

Akkermansia and Dorea abundances were associated with tobacco exposure through active smoking during pregnancy. The results of genus association with the multivariable association model (MaAsLin2) analysis in a model adjusted by maternal education, maternal BMI, and having siblings at birth. Active smoking at any time during pregnancy was associated with a decrease in the relative abundance of Akkermansia (q-value 0.005, BH adjusted by genus). Sustained maternal smoking from the first to the third trimester of pregnancy was associated with an increased relative abundance of Dorea (q-value 0.01, BH adjusted by genus).