Can oral microbiome predict low birth weight infant delivery?

This study aimed to identify the oral microbiota factors contributing to low birth weight (LBW) in Chinese pregnant women and develop a prediction model using machine learning.

A nested case-control study was conducted in a prospective cohort of 580 Chinese pregnant women, with 23 LBW cases and 23 healthy delivery controls matched for age and smoking habit. Saliva samples were collected at early and late pregnancy, and microbiome profiles were analyzed through 16S rRNA gene sequencing.

The relative abundance of Streptococcus was over-represented (median 0.259 vs. 0.116) and Saccharibacteria_TM7 was under-represented (median 0.033 vs. 0.068) in the LBW case group than in controls (p < 0.001, p = 0.015 respectively). Ten species were identified as microbiome biomarkers of LBW by LEfSe analysis, which included 7 species within the genus of Streptococcus or as part of 'nutritionally variant streptococci' (NVS), 2 species of opportunistic pathogen Leptotrichia buccalis and Gemella sanguinis (all LDA score>3.5) as risk biomarkers, and one species of Saccharibacteria TM7 as a beneficial biomarker (LDA= -4.5). The machine-learning model based on these 10 distinguished oral microbiota species could predict LBW, with an accuracy of 82 %, sensitivity of 91 %, and specificity of 73 % (AUC-ROC score 0.89, 95 % CI: 0.75-1.0). Results of α-diversity showed that mothers who delivered LBW infants had less stable salivary microbiota construction throughout pregnancy than the control group (measured by Shannon, p = 0.048; and Pielou's, p = 0.021), however the microbiome diversity did not improve the prediction accuracy of LBW.

A machine-learning oral microbiome model shows promise in predicting low-birth-weight delivery. Even in cases where oral health is not significantly compromised, opportunistic pathogens or rarer taxa associated with adverse pregnancy outcomes can still be identified in the oral cavity.

This study highlights the potential complexity of the relationship between oral microbiome and pregnancy outcomes, indicating that mechanisms underlying the association between oral microbiota and adverse pregnancy outcomes may involve complex interactions between host factors, microbiota, and systemic conditions. Using machine learning to develop a predictive model based on specific oral microbiota biomarkers provides a potential for personalized medicine approaches. Future prediction models should incorporate clinical metadata to be clinically useful for improving maternal and child health.