Fig. 4

Weighted co-occurrence network graphs identifies two clusters of co-occurring taxa which were associated with age and early childhood caries case status among 855 longitudinal, 16S rRNA gene sequenced saliva samples from 189 children from Appalachia (191 records) in an incidence-sampled case-control study (Center for Oral Health Research in Appalachia 2 (COHRA2) study) and were reproducible in an independent longitudinal cohort of similarly aged children with a 10% prevalence of early childhood caries. A Spaghetti plots showing the summed module relative abundance of two of the five identified network modules from weighted co-occurrence networks. Networks were named using the two most abundant amplicon sequence variants in the network and the most central amplicon sequence variant. Summed module relative abundance calculated by summing the relative abundance of all amplicon sequence variants assigned to the same cluster. Thin, transparent lines are individuals over time, thick lines represent smoothed means, dots and bars are mean and bootstrapped 95% confidence intervals at each visit, including both pre-incident and incident visits. B Network graphs of the two network modules shown in A. On the left, the protective network module was dominated by H. parainfluenzae (turquoise) and a Neisseria taxon (gold), with central taxon Fusobacterium periodonticum (green). On the right, the cariogenic network module was dominated by Streptococcus (red) and Veillonella (brown), with additional Actinomyces (pink) and Prevotella (purple) members. Amplicon sequence variants (nodes) that were more abundant in cases are shown as triangles, those more abundant in controls are shown as squares. Larger nodes represent more abundant amplicon sequence variants, and nodes are colored by genus. Amplicon sequence variants which were among the top 10 most important features in the supervised random forests are annotated with an asterisk, *. Thicker edges represent stronger correlations between amplicon sequence variants. C Repeating the co-occurrence graph analysis in an independent longitudinal cohort of similarly aged children (Holgerson et al.) identified a Haemophilus (turquoise) and Neisseria (gold) dominated network with central taxon Fusobacterium periodonticum, similar in composition and structure to the protective network module identified in COHRA2. A Veillonella dispar (brown), Prevotella (purple), and Streptococcus (red) network module was similar in composition to the cariogenic network module identified in COHRA2. Larger nodes represent more abundant amplicon sequence variants, and nodes are colored by genus. Amplicon sequence variants which were shared between the corresponding modules in COHRA2 and the Holgerson et al. cohort are annotated. For visualization purposes, edges with weights < 0.03 were not included