The following strains commonly present in the oral microbiome (Aas et al., 2005; Maddi and Scannapieco, 2013; Loozen et al., 2014) were obtained from the American Type Culture Collection (ATCC): A. actinomycetemcomitans (ATCC 43718), F. nucleatum (ATCC 10953), P. gingivalis (ATCC 33277), P. intermedia (ATCC 25611), S. mutans (ATCC 25175), S. sobrinus (ATCC 33478), T. forsythia (ATCC 43037), Actinomyces naeslundii (ATCC 51655), Capnocytophaga sputigena (ATCC 33612), Streptococcus gordonii (ATCC 49818), Actinomyces viscosus (ATCC 15987), and S. mitis (ATCC 49456). Veillonella parvula was obtained from the Leibniz Institute DSMZ-German Collection of Microorganisms and Cell Cultures (DSM 2007); S. sanguinis was acquired from the BCCM/LMG Bacteria Collection (LMG 14657). S. salivarius strain TOVE-R (Tanzer et al., 1985) was utilized. Bacterial strains were classified as either commensal (indigenous), potentially health-associated or disease-associated, as indicated in Table 1.

Biolog phenotype microarrays (PM, Biolog, Inc., Hayward, CA, USA) are a commercial high-throughput system for the analysis of microbial cellular phenotypes. The assay uses the reduction of a tetrazolium dye by NADH as a sensitive indicator of microbial respiration in response to the presence of individual nutrients or chemicals (Bochner et al., 2001). However, this methodology does not allow for estimating population size or bacterial abundance, because only a small portion of all compounds may be actively engaged in cell growth (Posch et al., 1997; Smith and Del Giorgio, 2003). PM plates 1–8, which contained 190 sole carbon sources, 95 sole nitrogen sources, 59 sole phosphate sources, 35 sole sulfur sources, 95 nutrient supplements, and 285 sole peptide-nitrogen sources were used in this analysis. All bacteria, excepting T. forsythia, were grown on blood agar plates (Blood Agar Base No 2, Oxoid, Hampshire, UK) supplemented with 5 μg/ml hemin (Sigma, Diegem, Belgium), 1 μg/ml menadione (Sigma, Diegem, Belgium), and 5% defibrinated horse blood (E&O Laboratories Limited, Bonnybridge, Scotland). T. forsythia was grown brain-heart infusion agar (Oxoid, Hampshire, UK) supplemented with 0.5% (w/v) of yeast extract (Oxoid, Hampshire, UK), 0.1% (w/v) of L-cysteine (Merck, Darmstadt, Germany), 5.0 mg/L of hemin (Sigma, Diegem, Belgium), 0.05 mg/L of menadione (Sigma, Diegem, Belgium), 0.001% (w/v) of N-acetylmuramic acid (Sigma, Diegem, Belgium), 5% Fetal Bovine Serum (Sigma, Diegem, Belgium), and 5% defibrinated horse blood (E&O Laboratories Limited, Bonnybridge, Scotland). A. actinomycetemcomitans (Aa), S. gordonii (Sgord), S. mitis (Smit), S. mutans (Smut), S. salivarius (Ssal), S. sanguinis (Ssang), and S. sobrinus (Ssob) were incubated at 37°C in a 5% CO₂ environment. A. naeslundii (Anaes), A. viscosus (Avisc), C. sputigena (Csput), F. nucleatum (Fn), P. gingivalis (Pg), P. intermedia (Pi), T. forsythia (Tf), and V. parvula (Vp) were incubated at 37°C under anaerobic conditions (10% CO₂, 10% H₂, 80% N₂). Anaerobic incubation was performed to recreate the conditions in the periodontal pockets (Marsh et al., 2015). PM assays were performed following the manufacturer's instructions using Biolog Redox Dye Mix D. Metabolic activity was estimated based on the optical density (OD), which was measured after 0, 24, and 48 h of incubation under anaerobic conditions at 37°C at 492 nm using a microplate reader (Multiskan Ascent, Thermo Fisher Scientific Oy, Vantaa, Finland). Incubations were performed in triplicate, under anaerobic conditions (10% CO₂, 10% H₂, 80% N₂), because all bacterial strains were facultative or anaerobic.

All the metabolites were categorized according to the respiratory activity triggered on each bacterial species, regarding the control (no nutrient added). Mean OD was calculated for each metabolite, based on the OD readings from the 15 bacterial species. If the OD reading on a species was above the 0.975 quantile, the metabolite was considered to generate high respiratory activity (H-RA) in that strain. Metabolites whose OD value was below the 0.025 quantile were considered to trigger low respiratory activity (L-RA). If the OD was between the 0.025 and 0.975 quantiles, it was considered not significantly different from the control (NS); control OD readings were included within this category (Hernandez-Sanabria et al., 2010). The frequency of the activity triggered by each metabolite within a group of bacteria (commensal, disease-associated, or health-associated) was compared using a Chi-square analysis. In this way, 3 × 2 contingency tables of cross-classification containing the respiratory activity (H, L, or NS) were produced for each of the bacterial groups. When the count of any of the cells was below 5, Fisher Exact test (R Core Team, 2000) was used to calculate the table probability (Hernandez-Sanabria et al., 2013), and activity-associated metabolites were highlighted.

Multiple Correspondence Analysis (MCA) was applied to detect similarities in the respiratory activity of the 15 species. The coordinates best representing bacterial species on two dimensions were retained to create clusters of bacteria with similarities in the use of the PM microarray metabolites. Thus, Hierarchical Clustering of Principal Components (HCPC) was performed, followed by construction of trees using the Ward's method (double factor hierarchical clustering) (Husson et al., 2010). Further, the Calinski-Harabasz (CH) index on k-means clustering was used to evaluate the cluster validity based on the average between- and within- cluster sum of squares, using the vegan package in R (Oksanen et al., 2013). The package NbClust in R was employed to calculate the CH, Silhouette, and Dunn indices (Charrad et al., 2014). Scores were visualized with the package factoextra (Kassambara and Mundt, 2016). Multiple Factor Analysis (MFA) was employed to discriminate particular metabolites triggering respiratory activity (H-RA, L-RA, NS) on each bacterial group at both time points. The function MFA from the FactoMineR package (Lê et al., 2008) was performed in R and the square cosine values (cos²) extracted. Results were plotted in a Principal Coordinate Analysis (PCoA) (Oksanen et al., 2013) and visualized (Wickham, 2016).

Further proof-of-principle validation was performed in selected species grown in complex medium, to determine whether activity can nevertheless be increased by specific compounds. Cell number and membrane integrity were used as indicators of activity (De Roy et al., 2012). Culturing of F. nucleatum, A. actinomycetemcomitans, and S. salivarius was in modified brain-heart infusion (Alvarez et al., 2013), using the Hungate tube method and maintaining the media in anaerobic conditions (10% CO₂, 10% H₂, 80% N₂). Optical density of the culture used for inoculation was measured after 18 h. Growth analyses on single nutrient sources were performed in 8.0 ml of medium, supplemented with 1% v/v of either L-Asparagine, L-Aspartic Acid, or D-Sorbitol (Sigma, Diegem, Belgium), all at 25mM. Triplicate tubes were inoculated with 100 μl of a 18-h culture diluted 10 × and further incubated anaerobically at 37°C. Flow cytometry was performed at 0, 24, and 48 h. SYBR Green I (SG, Invitrogen, Merelbeke, Belgium) was used alone or in combination with Propidium Iodide (Live/dead BacLight Bacterial Viability kit, Invitrogen, Merelbeke, Belgium), to quantify the total number of bacteria and to differentiate between cells with intact and damaged cytoplasmic membranes, respectively (Berney et al., 2007). Stock staining solutions were prepared as described elsewhere (Van Nevel et al., 2013). Under anaerobic conditions, samples were placed in 96-well plates, diluted 100 × with PBS (pH 7.0) and stained with 2 μl of staining solution and 2 μl of 500 mM Na₂EDTA (pH 8.0) for outer membrane permeabilization. Prior to flow cytometry analysis, the stained samples were incubated for 13 min in the dark at 37°C. Cell count analysis was performed using an Accuri C6 flow cytometer (BD Biosciences, Erembodegem, Belgium) equipped with an autoloader, at a flow rate of 66 μl min⁻¹ in 25 μl (Van Nevel et al., 2013). Processing of the flow cytometry results was performed with the Summit v4.3 software. Differences in total and intact number of bacterial cells supplemented with different metabolites were compared using a mixed model with time as a repeated measure, in SAS version 9.3 (Sas, 2012). Statistical significance was assumed at P < 0.05.

The results of the tetrazolium dye indicator assay were used as estimates of the respiratory activity. Changes regarding the control were classified into High respiratory activity (H-RA), Low respiratory activity (L-RA), or Not Significantly different (NS) from the control using a Chi-square procedure. In this way, 3 × 2 contingency tables of cross-classification containing the respiratory activity (H, L, or NS) were produced for each of the bacterial groups (health-associated, disease-associated, or commensals). The objective of this analysis was to detect common metabolites that promote similar respiratory activity on an entire bacterial group.

Thus, Val-Ala increased the respiratory activity in five of the disease-associated species but the effect was not significantly different from the control for F. nucleatum and in P. gingivalis. Additional 17 compounds tended to decrease the respiratory activity of disease-associated at 24 h (n = 7, P < 0.10, Figure 1A and Table 2A). From these, 11 were carbon sources, one nitrogen and one nutrient source, and four peptides. The respiratory activity of all the commensal species (n = 3), was increased with sources of nitrogen and peptides. Only carbon sources triggered high respiratory activity (P < 0.0001) in all the health-associated (n = 5). At 48 h, 2-aminoethanol increased the respiratory activity of 5 disease-associated bacteria, but not in P. gingivalis and A. actinomycetemcomitans (Figure 1B). Similarly, glucuronamide stimulated respiration in all disease-associated except for F. nucleatum and P. gingivalis. A total of 14 carbon, four nitrogen sources, and three peptides tended to decrease the respiratory activity of all disease-associated, but not significantly (Table 2B). For the commensals, 20 compounds promoted their respiratory activity, mainly peptides (P < 0.0001). Three sources of carbon increased the respiratory activity of all health-associated species (P < 0.0001). Three molecules triggered activity at both 24 and 48 h in commensals, whereas α-hydroxyglutaric acid γ-lactone had similar impact in health-associated. None of the evaluated compounds significantly increased activity in disease-associated at both time points (Figure 1B). Our results indicated that the metabolite requirements of disease-associated bacteria are dissimilar, suggesting that collective modulation of their respiratory activity may be challenging. However, one molecule may consistently support the respiratory activity of all species considered health-associated.

We first focused in detecting the metabolites that collectively increased the respiratory activity on each of the bacterial groups tested, using the frequency analysis. Then, MCA was employed to determine whether bacteria could be grouped by overall metabolic capacities and requirements, regardless of their initial classification (health-associated, disease-associated or commensal). The loadings of the first five components of the MCA were retained and clustered using the k-means hierarchical clustering (Supplementary Table 1). At 24 h, commensals and health-associated were grouped together while a second cluster included five disease-associated, one commensal, and one health-associated (Figure 2A). C. sputigena showed functional similarities with P. gingivalis at 48 h and their activity was significantly different from the bacteria grouped in the other clusters (P < 0.05, Figure 2B). S. mitis, S. sanguinis, and V. parvula were clustered with disease-associated, whereas A. naeslundii, A. viscosus, S. gordonii, and S. salivarius remained together over time. This clustering analysis suggested that there may be overall metabolic similarities among bacterial species. Following, MFA was applied to explain why bacteria considered as commensal or health-associated were grouped with disease-associated. Based on the respiratory activity triggered by all the screened metabolites, health-associated were included in dimensions 3 and 4 at 24 h (Figure 3A). Commensal bacteria were also included in Dim 3, where the highest scores from the metabolic variables corresponded to the carbon and nitrogen sources (Supplementary Table 2A). This result indicates that carbon and nitrogen sources promote respiratory activity in both health-associated and commensals. At 48 h, health-associated bacteria were spread in Dimension 4 and 5, while disease-associated were in Dimension 3 and commensals in Dim. 2. (Figure 3B). The location of health-associated bacteria in the plot, regarding disease-associated, indicated that the differences in the metabolic activity among both groups were triggered by phosphorus and sulfur (Supplementary Table 2B). These findings confirmed that metabolic requirements in health-associated species fluctuate over time but they are shared among the members of a specific bacterial community.

MFA was additionally used to discriminate metabolites that exclusively triggered respiratory activity in a single group of bacteria, and to uncover correlations among these metabolites. The square correlation ratios (cos²) measure the degree of association between variables and any given dimension of the factor analysis. Thus, the cos² between the coordinates of the bacterial species and the metabolites revealed that carbon and nitrogen sources were the main metabolic triggers for health-associated bacteria at 24 h 1 (cos² > 0.8, Supplementary Table 2A). Ala-Gly, urea, D, L-α-Amino-N-Butyric Acid, L-Phenylalanine, nitrate, L-Cysteine, α-Amino-N-Valeric Acid, L-Citrulline, L-Asparagine, D-Lysine, D-Glucosamine, L-Lysine, and ammonia had the highest contributions to dimensions 3 and 4 of the MFA, which both described health-associated bacteria. This suggested that the above were the metabolites that triggered high respiratory activity only in all the health-associated (Supplementary Table 2). After 48 h, carbon and P and S sources were the metabolites that contributed to discriminate between disease-associated and health-associated bacteria. L-Alanyl-Glycine, L-Methionine, Ala-Gly, γ-Hydroxy Butyric Acid, and D-Glucose-1-Phosphate, were described in Dimension 3 with disease-associated, while Glycyl-L-Proline, Glycyl-L-Aspartic Acid, Glycyl-L-Glutamic Acid, Gly-Asn, L-Leucine, gelatine, and L-Alanyl-Glycine, 2-Hydroxyethane Sulfonic Acid, and L-Cysteine Sulfinic Acid, were included in Dimension 4, with health-associated species (Supplementary Table 2B, P < 0.05). L-Valine, L-Isoleucine, and O-Phospho-D-Tyrosine were positively associated with both health-associated and disease-associated, indicating that both groups increase their respiratory activity with these compounds, suggesting potential competitive interactions. L-Glutamine and N-Acetyl-L-Cysteine were negatively associated with disease-associated and Adenosine-3',5'-cyclic monophosphate, and Taurine were partially associated with health-associated, because their cos² values were high on either dimension 4 or 5 (Supplementary Table 2B). These results explain why collective modulation of respiratory activity may result challenging and thus, alternative strategies may need to be developed.

Flow cytometry was used to validate whether variations in the respiratory activity were a result of increased cell number. Bacterial cells were quantified following supplementation with compounds that either promoted respiratory activity (L-Asparagine, in health-associated) or that were not found to have effect on the respiratory activity of neither group (D-Sorbitol and L-Aspartic Acid). SYBR Green I (SG) allows for quantification of the total cell number irrespective of viability or cultivability (Zipper et al., 2004), while Propidium iodide (PI) is appropriate for investigating oxidative damage to cells. We used them in combination to differentiate between intact and damaged cells. Significant differences in total cell number in comparison with the control were identified only when L-Aspartic Acid was supplemented (P < 0.05, Table 3). PI allowed for determination of the impact that diverse metabolites may have on the membrane integrity. L-Asparagine and L-Aspartic Acid decreased the number of intact cells in F. nucleatum and S. salivarius at 48 h. However, no significant difference in cell numbers of F. nucleatum was detected with either stain at 24 h. Intact cells of S. salivarius were less when sources of amino acids were added to the medium, while the opposite effect was observed in A. actinomycetemcomitans (P < 0.05). These results indicate that the metabolic activity detected in the Biolog plates may not be a direct result of the number of bacterial cells. In addition, damage to the cell membrane may not influence overall respiratory activity.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.