The salivary microbiome as a diagnostic biomarker of health and periodontitis: a large-scale meta-omics analysis before and after the removal of batch effects

Abstract

Abstract Background Methodological differences in 16S rRNA sequencing studies have significant effects on the diversity of the results obtained, making comparisons in the form of traditional reviews controversial. This meta-omics study applies the best practices based on the available evidence and employs sequences from different Illumina V3-V4 bioprojects. The goal is to evaluate the salivary microbiota at the amplicon sequence variant (ASV) level in terms of differential proportionality and predictive models. This is done in periodontally healthy and untreated periodontitis patients, both before and after the removal of batch effects (BEs). Results Before the removal of BEs, 265 ASVs from 115 species (2.69% and 20.07% of the totals detected, respectively) revealed significant differences in their centred log-ratio abundance values for healthy and diseased patients. After their removal, this number fell to 190 ASVs from 94 species (1.93% and 16.40%), with 148 ASVs from 75 species (1.50% and 13.09%) common to those obtained before removal. In the predictivity analysis, models constructed before BEs removal using all the samples (796) consisted of 16 ASVs (0.16%) and had an area under the curve (AUC) of 0.944; models built using two-thirds of the specimens (training = 531) comprised 35 ASVs (0.36%) and had an AUC of 0.955 after being validated on one-third of the samples (test = 265). After BEs removal, these figures deteriorated - the models required more ASVs (all = 200 − 2.03%; training = 100 − 1.01%) and had slightly lower AUC (all = 0.935; test = 0.947). Conclusions Nearly half of the differential proportionality relationships before the exclusion of BEs were spurious. Although removing them reduced the number of ASVs with differential proportionality for health and periodontitis by approximately one-third, almost twelve (all samples) and three (training/test) times as many predictive ASVs were required to distinguish between clinical conditions than before the BEs exclusion. Nevertheless, both before and after their eradication, all models suggested that saliva has relevant value as a biomarker for diagnosing health and periodontitis, with a small percentage of salivary taxa having an outstanding capacity to discriminate between these conditions. The main health-predictor ASV was Streptococcus oralis dentisani-AV1042; for periodontitis, these were Fusobacterium nucleatum vincentii-AV10, Mycoplasma faucium-AV213, Parvimonas HMT110-AV21, Treponema denticola-AV38, and Tannerella forsythia-AV15.