Sequencing and data preprocessing statistics
A total of 26,208,640 paired-end reads were obtained (31,804 to 1,23,233 reads/sample), of which 95.24% reads merged successfully. Around 17% (4,310,661) of the sequences passed our stringent quality filtration, resulting in a final sequencing depth of 4,681 to 20,450 reads/sample. Raw sequences were submitted to Sequence Read Archive (SRA) under submission no. SUB11981103 (in process).
General microbial profiles of the inocula and in vitro microbiomes
A total of 119 and 137 species belonging to 39 and 40 genera and 8 phyla were detected in the two health inocula, respectively; in the two periodontitis inocula, 178 and 172 species, 55 and 65 genera, and 9 phyla were identified, respectively. In the in vitro-grown microbiomes, the comparable profiles were 58–103 species, 20–33 genera, 4–7 phyla in the health-derived microbiomes, and 75–135 species, 32–52 genera and 6–8 phyla in the periodontitis-derived microbiomes. The relative abundances and detection frequencies of identified phyla, genera and species in each of the samples are listed in Supplementary Files 1–3, respectively, while the average taxonomic profiles in the inocula as well as in vitro microbiomes as a function of serum concentration, incubation period and shaking are presented in Supplementary Figs. 1 and 2, for the phylum- and genus-level, respectively.
Firmicutes, Fusobacterium, and Bacteroidetes, in this order of abundance, were the most dominant phyla in health in both the clinical inocula as well as in vitro microbiomes. The same phyla were also the most abundant in periodontitis but in the order of Bacteroidetes, Firmicutes and Fusobacterium by abundance. However, in both health and periodontitis, these three phyla were over-represented in the in vitro microbiomes at the expense of the Saccharibacteria, Proteobacteria and Actinobacteria (Supplementary Fig. 3). Chloroflexi was exclusively detected in the periodontitis inoculum but not in the respective in vitro microbiomes. At the genus level, Fusobacterium, Streptococcus, Porphyromonas, Prevotella, and Alloprevotella were the most abundant overall, although their relative abundances differed between health and disease and between the clinical inocula and the in vitro microbiomes (Supplementary Figs. 2). The latter showed enrichment of Fusobacterium and Prevotella in addition to Mogibacteria, Catonella and Bacteroides at the expense of Leptotrichia, Rothia, Haemophilus, Capnocytophaga and TM7 genera 1 and 5 (Supplementary Fig. 3).
At the species level, the dominant species in the health-derived microbiomes on average were Fusobacterium periodonticum, Fusobacterium nucleatum, Streptococcus dentisani, Mogibacterium diversum, Porphyromonas endodontalis, Alloprevotella tannerae, Porphyromonas oral taxon 278, Prevotella intermedia, Streptococcus oral taxon058, Catonella morbi, and Veillonella parvula_group, while periodontitis derived microbiomes were dominated by Prevotella intermedia, Porphyromonas gingivalis, Bacteroides heparinolyticus, Bacteroides zoogleoformans, Fusobacterium nucleatum, Fusobacterium periodonticum, Streptococcus tigurinus, and Peptoniphilaceae oral taxon 790.
In vitro microbiomes replicate subgingival normobiosis and dysbiosis
Regardless of growth conditions, the health- and periodontitis-derived microbiomes along with the respective clinical inocula formed two separate main clusters in beta diversity analysis, accounting for ~ 32% variation along principal component 1 (Fig. 2A) – the biological replicates formed sub-clusters within each cluster and accounted for less variation (14% along principal component 2), primarily in periodontitis. Similarly, the health- and periodontitis-derived microbiomes reflected the differences between the respective inocula in terms of biomass, species richness (Chao index), alpha diversity (Shannon index) and dysbiosis (SMDI) (Fig. 2B), with all being significantly higher (with the exception of Shannon index) in the periodontitis-derived microbiomes.
More importantly, differential abundance analysis by MaAsLin2 identified microbial differences between the health- and periodontitis-derived in vitro microbiomes that are largely consistent with known differences between periodontitis and health in vivo (Fig. 2C). For example, P. gingivalis, P. intermedia, Treponema denticola, Filifactor alocis, Fretibacterium fastidiosum, Pyramidobacter piscolens, and Mogibacterium timidum, which have been consistently implicated as pathogens in periodontitis, were all significantly enriched in the periodontitis-derived microbiomes. Likewise, species such as Porphyromonas catoniae, Streptococcus dentisani, S. sanguinis, Streptococcus oral taxon 58, Catonella morbi and Granulicatella adiacens, which are typically health-associated species, were significantly enriched in the health-derived microbiomes. Figure 3 presents the relative abundances of selected differentially abundant genera and species. The latter were chosen to represent sister species (i.e. two species within the same genus) that showed opposite enrichment in the health- and periodontitis-derived microbiomes consistent with differences demonstrated in their preponderance in vivo.
Higher serum concentration and longer incubation time promote dysbiosis
Generalized linear modelling or MaAslin2, as appropriate, were used to identify the independent effects of serum and incubation time on the different microbial parameters assessed, applying a false discovery rate (FDR) cutoff of 0.1 when applicable. Biofilm biomass significantly increased with time and with increasing serum concentrations for both the health- and periodontitis-derived microbiomes (Fig. 4A). Species richness (Chao index) did not change by time and serum concentration in the health-derived microbiomes, but it significantly dropped in 5% serum and after days 10 and 13 incubation in the periodontitis-derived microbiomes (Fig. 4B). Alpha diversity (Shannon index) significantly decreased in days 7 and 10, increased at 2% and 3.5% serum but dropped at 5% in the health-derived microbiomes; however, the magnitude of changes was small (Fig. 4C). In the periodontitis-derived microbiome, the Shannon index substantially increased with time but markedly decreased as a function of serum concentration. Dysbiosis (SMDI) increased proportionally as a function of time and serum concentration, in both the health-derived and periodontitis-derived microbiomes grown in vitro (Fig. 4D), being closest to the respective clinical inocula in the health-derived-microbiomes when grown in 1% serum for 4 days (Median SMDI of -1 in the in vitro microbiomes compared to -2.2 in the health inoculum), and in the periodontitis-derived-microbiomes after growth in 1% serum for 13 days (Median SMDI of 1.33 in the microbiomes compared to 1.38 in the periodontitis inoculum); those grown in 3.5%-5% for 7 days or 2% for 10 days came next (SMDI ~ 1.25). Beta diversity analysis for health and periodontitis separately resulted in two main clusters by biological replicate along PC1 and sub-clusters by growth time along PC2 (Fig. 4E); analysis of the distance-matrix revealed that the health-derived microbiomes grown for 4 days at 1% serum concentration and the periodontitis-derived microbiomes grown for 4 days in 5% serum (followed by those grown in 5% for 7 days or 2% for 10 days) were the closest to the respective clinical inocula.
The relative abundances of phyla and genera that significantly changed as a function of time and serum concentration are shown in Supplementary Fig. 4 and Fig. 5, respectively. At the phylum level, serum resulted in a dose-dependent increase in Bacteroidetes at the expense of Firmicutes and Fusobacteria, while a prolonged growth period was associated with an increase of Actinomyces and Spirochetes and slight decrease in Firmicutes, Fusobacteria and Bacteroidetes. At the genus-level, the major changes included substantial enrichment of Porphyromonas and Alloprevotella as a function of serum concentration at the expense of Streptococcus, Fusobacterium and Prevotella, and an increase in Porphyromonas, Bacteriodes and Mogibacterium accompanied by a decrease in Prevotella, Catonella and Gemella as a function of time. Figure 6 presents selected sister species that responded in opposite directions to increased serum concentration and incubation period.
Shaking had limited effect on composition of the in vitro microbiomes
The independent effects of shaking on the growing microbiomes is shown in Supplementary Fig. 5. Shaking increased biomass of the health-derived microbiome but not of the periodontitis-derived microbiomes. Statistically significant differences were observed for species richness and dysbiosis, but the magnitude of change was minor. Namely, shaking slightly increased in Chao index in the health-derived microbiomes and slightly decreased it in the periodontitis-derived microbiomes, while it marginally increased SMDI in both the health- and periodontitis-derived microbiomes, probably because of enrichment of genera Treponema and Pyramidobacter at the expense of Gemella and Granulicatella (Supplementary Fig. 6). Shaking did not affect alpha diversity (Shannon index) in either microbiome type.