Relationships among the maternal and neonatal microbiomes
The studied dataset was collected from 164 mother-neonate dyads, including 16S rRNA taxonomic profiles from neonatal buccal (NB), rectal (NR), and stool (NS) sites, maternal buccal (MB), rectal (MR), and vaginal (MV) niches in the pregnant women, maternal and neonatal clinical and other metadata, and lipid and cytokine expression levels in the vaginal fluid of the pregnant women. The experimental designs and case numbers are shown in Fig. S1 and Supplementary Data 1 and 2.
Complexity (alpha diversity) quantified by the Shannon index, evenness of bacterial abundance distribution, and the number of observed taxa of the NB and NR microbiomes were lower than those of the MB and MR microbiomes, respectively (Fig. S2), illustrating that the neonatal microbiomes are less complex than the mature maternal microbiomes22.
Dissimilarity (beta diversity) of the neonatal and maternal microbiomes was visualized in a t-distributed stochastic neighbor embedding (t-SNE) plot (Fig. 1a and and Fig. S1a) and quantified in a heatmap showing median values of Bray-Curtis distance between each of the paired microbiomes (Fig. 1b). A significant difference in beta diversity between the NB and NR microbiomes was not observed on day 0, but appeared on days 1 and 2 (Fig. 1b and Fig. S3a). In contrast, the within-group dissimilarity of both the NB and NR microbiomes decreased on days 1 and 2 (Fig. 1d). These results are consistent with a previous report13 suggesting that the NB and NR microbiomes rapidly diversify from each other but that the microbiomes at each body site tend to converge among individuals after birth (Supplementary Movie 1).
Compared to the maternal microbiomes, the neonatal microbiomes generally clustered more closely on the t-SNE plot (Fig. 1a), suggesting a higher similarity among the neonatal microbiomes consistent with a shared source of taxa at the beginning of life. Recent reports have suggested that the NB microbiome is similar to the MV microbiome immediately after birth (≤ 5 minutes)11. Herein, on day 0 (≤ 24 hours), the NB microbiomes were more widely distributed on the t-SNE plot, and some of the NB microbiomes clustered among each of the three maternal microbiomes (Fig. 1a). These observations imply that microbes in the NB microbiome within 24 hours may derive from different maternal sources. The NB microbiome on day 0 was most similar to that of the MB microbiome (Fig. 1b). Consistent with previous observations23, the differences between the NB and MB microbiomes decreased with time, but the same phenomenon was not observed between the NR and MR microbiomes (Fig. S3b-d, and Supplementary Movie 1). These observations suggest that the composition of the NB microbiome tends to converge on the MB microbiome after childbirth, but the NR microbiome develops a composition that diverges from the MR microbiome at an early stage. However, the NB-MB distance in the paired mother-neonate dyads was not higher than in unpaired samples (Fig. S3e), indicating that general maternal factors shared by all the women were not primary factors modulating the composition of the NB microbiomes. The beta diversity of the NR microbiome on day 0 was most similar to that of the MV microbiome (Fig. 1b), suggesting that the main source of the initial NR microbiome may be the MV microbiome.
Composition of the neonatal microbiomes
Consistent with beta diversity analysis, the predominant taxa in the NB and MB microbiomes were the same Gram-positive and aerobic or facultative anaerobic bacterial taxa, e.g., Corynebacterium OTU 226, Streptococcus mitis, and Streptophyta OTU 179 (Fig. S4 and Supplementary Data 3). The predominant taxa in the NR microbiome on day 0 were similar to those abundant in the MV and MB microbiomes. However, the composition of the NR microbiome on days 1–2 differed from any of these three maternal microbiomes but was more similar to the NS microbiome; i.e., the dominant taxa were Escherichia coli and other Enterobacteriaceae spp.
Differential abundance analysis using ‘ALDEx2’28 showed that the relative abundance of the predominant taxa on days 1 and 2 and several closely related taxa, e.g., several Corynebacterium and Streptococcus spp. in the NB and Enterobacteriaceae spp. in the NR, were increased from day 0 to day 1 (Fig. 1d and Supplementary Data 4). However, no significant changes were observed from day 1 to day 2. These results suggested that the composition of the NB and NR microbiomes changed abruptly within 24–48 hours but were more stable by the third day postpartum. Furthermore, network analysis showed that predominant taxa, e.g., Streptococcus and Corynebacterium spp. in the NB microbiome (Fig. S5) and Enterobacteriaceae spp. in the NR microbiome (Fig. S6), had negative correlations with taxa abundant in other body sites in related microbiomes. Since the predominant taxa in the NB and NR microbiomes had increased relative abundance from day 0 to day 1 (Fig. 1d), it was not surprising that many of the taxa abundant at other body sites had reduced relative abundances from day 0 to day 1, although these abundance changes were not significant (Fig. 1d and Supplementary Data 4).
Maternal And Neonatal Factors Associated With The Neonatal Microbiome Structure
Maternal metadata, microbiomes, lipids, and cytokines on the last visit of pregnancy and paired neonatal metadata and microbiomes on the first visit after birth were selected to identify factors that are associated with the neonatal microbiomes (Fig. S1b and Supplementary Data 1).
The association between the alpha diversity of each neonatal microbiome and each factor in the metadata was measured by the Mann-Whitney U test or linear regression (see Methods). These maternal factors were mainly associated with the alpha diversity in the NB microbiome, i.e., yeast infection, histories of pelvic inflammatory disease and urinary tract infections, and other maternal diseases and disease histories; issues during pregnancy including contractions, vaginal bleeding, and progesterone administration (as previously reported29); behaviors including the age of the first sexual intercourse30 and birth control by Depo-Provera injections31; environmental factors involving education and annual household income; neonatal factors, e.g., admission to the NICU, baby’s sex, height and weight (Fig. 2a and S7). All the factors related to adverse maternal and neonatal health were associated with higher alpha diversity, whereas those associated with pregnancy and better environmental conditions were associated with a lower alpha diversity in the NB microbiome. These results suggested that a complex NB microbiome was associated with the mothers' and neonates' sub-optimal health state. Interestingly, in contrast to males, female neonates exhibited a higher alpha diversity in the NB microbiome. One possible explanation is that potential differences in the neonatal immune systems associated with sex could diversify the NB microbiomes32–34. The statistical powers of sample sizes in most of these associations are above 0.9 (Supplementary Data 2).
Alternatively, a non-linear regression analysis using a Leave-One-Out Cross-Validation strategy and the random forest algorithm was performed to predict the alpha diversity of the neonatal microbiomes with multiple variables, and the performance of the prediction was evaluated by linear regression between predicted and true values of the alpha diversity. When all the maternal factors in the metadata were applied as independent variables, the alpha diversities of the NB, NR, and NS microbiomes were accurately predicted, which implied the causal relationship between maternal health and the complexity of the neonatal microbiome (Fig. 2b and S8). Additionally, the factors in the metadata were grouped into twenty clusters based on the Spearman’s correlation among the factors (left panel in Fig. 2c, Fig. S9, and Supplementary Data 5), and twenty individual models were generated using factors in one cluster as independent variables in each model. The predicted values in the models built with factor clusters 9 and 19 are significantly linearly correlated with the true alpha diversity values of the NB, NR, and NS microbiomes (Fig. 2d), suggesting that some of the factors included in microbial infections and infection histories as well as abnormal vaginal odor and discharge on the last visit of pregnancy and presence of ovarian cysts are universal factors that impact the alpha diversity of all three neonatal microbiomes.
The association between the beta diversity of each of the neonatal microbiomes and the factors in the metadata was quantified by the Adonis test containing all the factors as independent variables (Supplementary Data 5, see details in Methods). Several factors influenced the beta diversity of one of the NB, NR, and NS microbiomes as follows: several maternal mental stresses; employment status, e.g., homemaker or student; environmental factors, including education; behaviors, e.g., vaginal douching, smoking, and moving to a new address; diseases and disease histories; e.g., abnormal Pap smear, bacterial vaginosis, and diabetes; C-section; the complexity of the MV microbiome; neonate factors involving baby’s pulse and health problems postpartum; and others (Fig. 2c). The time after birth was only significantly associated with the NB microbiome, implying a faster change of the NB microbiome than the NR and NS microbiomes. The impact of C-section3,11, 13–17, diabetes17,24, smoking36, mental health, and antibiotics24 on early-life microbiomes have been reported previously.
Another approach to quantify the influence of body site on the beta diversity of the neonatal microbiomes used twenty sets of paired NB, NR, NS microbiomes, metadata (Supplementary Data 1 sheet 7), and an additional ‘body site’ factor that indicated the niche of the neonatal microbiomes. According to the highest R-squared value, body site was the most important factor in the Adonis analysis, suggesting the importance of micro-environments on the maturation of the neonatal microbiomes (Fig. 2e).
Similar to a previous study9, more similar taxa were found in the paired mother-neonate dyads than in the unpaired dyads, but none of the observations were statistically significant (Fig. S10 and Supplementary Data 1 sheet 6), implying that a vertical mother-neonate microbe transmission had limited influence on the neonatal microbiomes within three days postpartum.
Mediation Of Maternal Lipids On The Association Between Maternal Factors And The Neonatal Microbiome
Pearson's correlation analysis showed that twenty-six lipids and four cytokines in the maternal vaginal fluid were associated with the alpha diversity of the neonatal microbiome, particularly the gastrointestinal microbiome (Fig. 3a and b and Supplementary Data 6). Interestingly, the alpha diversity of the NR microbiome was negatively correlated with six ceramides but positively correlated with ten sphingomyelins. Higher concentrations of sphingomyelins and lower levels of ceramides have been reported to be associated with decreased adaptive immune responses37. Thus, these observations are consistent with the hypothesis that a more complex NR microbiome is associated with an optimal maternal health condition and a less active adaptive immune system. Similarly, a more complex NS microbiome was correlated with lower concentrations of four cytokines that were associated with maternal inflammation (Fig. 3b). The statistical powers of sample sizes in all the significant associations are higher than 0.99 (Supplementary Data 6).
Our mediation analysis suggested that smoking, anxiety, and abnormal stress of the mothers increased the complexity of the NB microbiome, but meanwhile promoted the concentration of C20 ceramide and, as a result, attenuated the increase of the complexity of the NB microbiome (Fig. 3c and see details in Supplementary Data 6). The uptake of yeast infection medication and the HPV history of the mothers lowered the complexity of the NR microbiome. However, the influence of yeast infection medication and HPV history on the NR microbiome was attenuated by modulating the concentration of C18:0 ceramide and C14 sphingomyelin, respectively (Fig. 3d and see details in Supplementary Data 6). Since a complex NB microbiome has been associated with sub-optimal states of the mothers and neonates (Fig. 2) and high complexity of the gut microbiome is considered a hallmark of gut health48, the modulation of the maternal lipids is a protective mechanism by which the mothers limit the adverse impact of smoking, anxiety, abnormal stress, yeast infection medication, and HPV history on the neonatal microbiomes.
Maternal And Neonatal Factors Associated With The Risk For Nicu Admission
NICU admission rates have risen from 6.62% in 2008 to 9.07% in 2018 in the United States39 associated with increased incidence of very low birthweight neonates40. Not surprisingly, the NICU admission of the neonates was associated with lower birthweight, height, and BMI (Fig. 2c, S9, and Supplementary Data 5). The alpha diversity of the NB microbiome but not that of the NS or NR microbiome was significantly higher in the neonates being admitted to the NICU (Fig. 4a) and also in the paired mothers with microbial infectious diseases (Fig. 2a). Differential abundance analysis using the LEfSe41 showed that several taxa, including some potential pathogens; e.g., Neisseria spp., and Actinomyces spp.42, and several other microbes were enriched in the oral cavity of the neonates admitted to the NICU, but Streptococcus cristatus, a commensal oral microbe that inhibits the colonization of the oral pathogen Porphyromonas gingivalis43, was enriched in the controls (Fig. 4b and Supplementary Data 7). Hence, these results argue that a sub-optimal NB microbiome with higher complexity is associated with NICU admission.
The Mann–Whitney U test illustrated that several maternal factors, e.g., earlier gestational age at delivery, abnormal bed rest, yeast infection, hospitalization, higher frequency of vaginal douching, change of residence, smoking, abnormal stresses, and lower levels of education, were associated with an increased risk of NICU admission (Fig. 4c and Supplementary Data 7). This association was also examined by establishing a machine-learning model using the random forest algorithm and a cross-validation strategy as previously described44. The importance of variables in the model quantified by the mean decrease in Gini coefficient showed that gestational age at delivery had the strongest association with NICU admission, an unsurprising finding reflecting that most babies born prematurely are admitted to the NICU45. Both the Mann–Whitney U test and the machine learning method indicated that a sub-optimal maternal health condition, e.g., yeast infection and hospitalization, associated with higher alpha diversity of the NB (Fig. 2a), were also risk factors for NICU admission (Fig. 4c).
Interestingly, mothers whose babies would be admitted to the NICU had more complex oral microbiomes and less complex rectal microbiomes before childbirth (Fig. 4d). A higher alpha diversity of the oral microbiome has been associated with two most prevalent oral diseases, i.e., periodontitis and dental caries34,46, but a higher complexity of the gut microbiome is generally considered a hallmark of gut health38. Additionally, the LEfSe analysis showed that Veillonellaceae and a Saccharibacteria (TM7) sp. were enriched in the MB microbiome with matched neonates who would be admitted to the NICU (Fig. 4e and Supplementary Data 7) and several Veillonellaceae and TM7 spp. have been associated with periodontitis46 and preterm birth26. Thus, these data illustrated that sub-optimal MB and MR microbiomes were risk factors for NICU admission.