Proteomics of Porphyromonas gingivalis within a model oral microbial community

Background Porphyromonas gingivalis is a periodontal pathogen that resides in a complex multispecies microbial biofilm community known as dental plaque. Confocal laser scanning microscopy showed that P. gingivalis can assemble into communities in vitro with Streptococcus gordonii and Fusobacterium nucleatum, common constituents of dental plaque. Whole cell quantitative proteomics, along with mutant construction and analysis, were conducted to investigate how P. gingivalis adapts to this three species community. Results 1156 P. gingivalis proteins were detected qualitatively during comparison of the three species model community with P. gingivalis incubated alone under the same conditions. Integration of spectral counting and summed signal intensity analyses of the dataset showed that 403 proteins were down-regulated and 89 proteins up-regulated. The proteomics results were inspected manually and an ontology analysis conducted using DAVID. Significant decreases were seen in proteins involved in cell shape and the formation of the cell envelope, as well as thiamine, cobalamin, and pyrimidine synthesis and DNA repair. An overall increase was seen in proteins involved in protein synthesis. HmuR, a TonB dependent outer membrane receptor, was up-regulated in the community and an hmuR deficient mutant was deficient in three species community formation, but was unimpaired in its ability to form mono- or dual-species biofilms. Conclusion Collectively, these results indicate that P. gingivalis can assemble into a heterotypic community with F. nucleatum and S. gordonii, and that a community lifestyle provides physiologic support for P. gingivalis. Proteins such as HmuR, that are up-regulated, can be necessary for community structure.


Background
The microbial communities that exist on oral surfaces are complex and dynamic biofilms that develop through temporally distinct patterns of microbial colonization [1,2]. For example, initial colonizers of the salivary pellicle on the coronal tooth surface are principally commensal oral streptococci such as S. gordonii and related species. Establishment of these organisms facilitates the subsequent colonization of additional gram-positives along with gramnegatives such as Fusobacterium nucleatum. As the biofilm extends below the gum line and becomes subgingival plaque, further maturation is characterized by the colonization of more pathogenic gram-negative anaerobes including Porphyromonas gingivalis [2][3][4]. While organisms such as P. gingivalis are considered responsible for destruction of periodontal tissues, pathogenicity is only expressed in the context of mixed microbial communities. Periodontal diseases, therefore, are essentially microbial community diseases, and the interactions among the constituents of these communities and between the communities and host cells and tissues, are of fundamental importance for determining the health or disease status of the periodontium.
Oral biofilm developmental pathways are driven by coadhesive, signaling and metabolic interactions among the participating organisms. Pioneer bacteria provide a substratum and appropriate metabolic support for succeeding organisms. Complex consortia then accumulate through recognition and communication systems. These interbacterial signaling processes can be based on cell-cell contact, short range soluble mediators, AI-2, or nutritional stimuli [2,[5][6][7][8]. In general, bacterial adaptation to the community lifestyle is accompanied by distinct patterns of gene and protein expression [9,10]. In S. gordonii for example, arginine biosynthesis genes are regulated in communities with Actinomyces naeslundii which enables aerobic growth when exogenous arginine is limited [11]. Over 30 genes are differentially regulated in P. gingivalis following community formation with S. gordonii but not with S. mutans [12], whereas in monospecies P. gingivalis biofilm communities there are changes in abundance of over 80 envelope proteins [13].
While over 700 species or phylotypes of bacteria can be recovered from the oral cavity, in any one individual there are closer to 200 species [14] and the diversity of bacteria assembled in dense consortia will be further limited by nutritional and other compatibility constraints. P. gingivalis can accumulate into single species biofilms and mixed species consortia with S. gordonii and related oral streptococci [15][16][17]. Moreover, introduction of P. gingivalis into the mouths of human volunteers results in almost exclusive localization in areas of streptococcal-rich plaque [18]. Development of more complex multi-species communi-ties in aerated environments such as supragingival tooth surfaces may require oxygen scavenging by F. nucleatum [19]. F. nucleatum is also able to coaggregate with P. gingivalis and with oral streptococci [19][20][21]. Hence communities of S. gordonii, F. nucleatum and P. gingivalis are likely to be favored in vivo; however, community formation by these three organisms has not been investigated. The aim of this study was to examine the ability of S. gordonii, F. nucleatum and P. gingivalis to form multispecies communities in vitro, and to utilize a global proteomic approach to investigate differential protein expression in P. gingivalis in response to presence of these organisms.

Results and discussion
Assembly of P. gingivalis-F. nucleatum-S. gordonii communities in vitro Confocal laser scanning microscopy (CLSM) was used to investigate the ability of P. gingivalis to assemble into communities with S. gordonii and F. nucleatum. In order to mimic the temporal progression of events in vivo, S. gordonii cells were first cultured on a glass surface and this streptococcal substratum was then reacted in succession with F. nucleatum and P. gingivalis. The F. nucleatum and P. gingivalis cells were maintained in the absence of growth media in order to be able to detect any metabolic support being provided by the other organisms in the community. A 3D reconstruction of the heterotypic community is shown in Fig. 1. Both P. gingivalis and F. nucleatum formed discrete accumulations and could be either separate from each other or interdigitated, consistent with the concept that the later gram-negative colonizers such as P. gingivalis and F. nucleatum initially establish themselves on the streptococcal rich supragingival plaque [4,18]. The results demonstrate the mutual compatibility of these three organisms for heterotypic community development, an early step in the overall process of plaque biofilm accumulation. Participation in multispecies communities may provide a basis for synergistic interactions in virulence. For example, mixed infections of P. gingivalis and F. nucleatum are more pathogenic in animal models than either species alone [22], and F. nucleatum can enhance the ability of P. gingivalis to invade host cells [23].

Proteome of P. gingivalis in a three species community
To begin to investigate the mechanisms of adaptation of P. gingivalis to a community environment, the proteome of non-growing P. gingivalis cells incorporated into a community with F. nucleatum and S. gordonii was compared to the proteome of non-growing P. gingivalis cells alone. The expressed proteome of P. gingivalis in a community consisted of 1156 annotated gene products detected qualitatively. Based on spectral counting, 271 gene products showed evidence of relative abundance change at a qvalue of 0.01: 109 proteins at higher relative abundance and 162 at lower relative abundance, using P. gingivalis alone as a reference state. Spectral counting is a conservative measure of protein abundance change that tends to generate low FDRs [24][25][26] but that often suffers from high FNRs in studies of the kind described here [27]. Less conservative calculations based on intensity measurements [27] found 458 gene products with evidence of relative abundance change at a q-value of 0.01: 72 proteins at higher relative abundance, and 386 proteins at lower relative abundance. Spectral counting and protein intensity measurements were examined for common trends. Trends tended to be consistent across both biological replicates, but the magnitudes of the abundance ratios showed significant scatter, similar to most published expression data at either the mRNA or protein level [27]. In most cases the abundance ratio trends were the same, using both quantitation methods, although not necessarily significantly so. In only eight cases were the spectral counting trend and summed intensity trend significantly in opposite directions for the same protein (PGN 0329, 0501, 1094, 1341, 1637, 1733, 2065). The integrated relative abundance trends found 403 gene products with evidence of lower relative abundance change and 89 at higher relative abundance. For purposes of examining the totals for combined trends, if an abundance change was called as significant (red or green in Additional file 1: Table ST1) in one measurement, it was considered significant for the above combined totals only if the ratio of the other measurement showed the same direction of abundance change, with a log 2 ratio of ± 0.1 or greater regardless of the q-value in the second measurement. The experimental data for differential protein abundance are shown in Fig. 2 as a pseudo M/A plot [28,29] with a LOW-ESS curve fit [30]. The same data are plotted in Fig. 3 as open reading frames according to PGN numbers from the ATCC 33277 genome annotation [31]. A complete listing of all proteins, their abundance ratios relative to P. gingivalis controls incubated alone under the same conditions as determined by spectral counting and summed signal Confocal laser scanning microscopy of P. gingivalis-F. nuclea-tum-S. gordonii community Figure 1 Confocal laser scanning microscopy of P. gingivalis-F. nucleatum-S. gordonii community. S. gordonii cells (red, stained with hexidium iodide) were cultured on a glass plate. FITC-labeled F. nucleatum cells (green), followed by DAPI labeled P. gingivalis cells (blue), were reacted sequentially with the S. gordonii substratum. Bacterial accumulations were examined on a Bio-Rad Radiance 2100 confocal laser scanning microscope. A series of fluorescent optical x-y sections in the z-plane to the maximum vertical extent of the accumulation were collected with Laser Sharp software. Images were digitally reconstructed with Imaris software. Image is representative of three independent experiments.
Pseudo M versus A plot [28,29] of the average protein abun-dance ratios over all replicates for the P. gingivalis-F. nuclea-tum-S. gordonii/P. gingivalis comparison versus total abundance as estimated by spectral counting Figure 2 Pseudo M versus A plot [28,29] of the average protein abundance ratios over all replicates for the P. gingivalis-F. nucleatum-S. gordonii/P. gingivalis comparison versus total abundance as estimated by spectral counting. Color codes: red, P. gingivalis protein is overexpressed in the P. gingivalis-F. nucleatum-S. gordonii community relative to P. gingivalis alone; green, P. gingivalis protein is under-expressed in the community relative to P. gingivalis alone; black, no significant abundance change. Solid black lines represent a LOWESS curve fit [30] to the biological replicates of P. gingivalis alone, and represent the upper and lower boundaries of the experimentally observed error regions or null distributions associated with the relative abundance ratio calculations. Proteins coded as either red or green were determined to be significantly changed at the q-value [24] cut-off value of 0.01. Thus, the G-test predictions [56] were in good agreement with the curve fitting procedure. Details regarding hypothesis testing procedures can be found in Methods and in the explanatory notes to the data tables [see Additional File 1]. log 2 spectral counts log 2 abundance ratio Genomic representation of the P. gingivalis proteome, showing changes in relative abundance for the P. gingivalis-F. nucleatum-S. gordonii/P. gingivalis comparison by spectral counting Figure 3 Genomic representation of the P. gingivalis proteome, showing changes in relative abundance for the P. gingivalis-F. nucleatum-S. gordonii/P. gingivalis comparison by spectral counting. Each dot represents a PGN ORF number in the order followed by the ATCC 33277 strain annotation. Color codes: red, over-expression in the P. gingivalis-F. nucleatum-S. gordonii community relative to P. gingivalis alone; green, under-expression in the community relative to P. gingivalis alone; yellow, protein was detected qualitatively, but did not change in abundance; gray, proteins that were qualitative non-detects; gaps indicate ORFs that were not common to both the ATCC 33277 and W83 annotations according to a master cross-reference compiled by LANL (G. Xie, personal communication).
intensity [27,32,33], and q-values, are given in Additional file 1: Table ST1. Qualitative identifications for proteins secreted by P. gingivalis in the 3-species community but not by P. gingivalis alone are given Additional file 1: Table  ST2. Additional file 1: Figs. SF1, SF2, SF3, SF4, SF5 and SF6 and explanatory notes provide more detailed technical information regarding reproducibility of the biological replicates and the adequacy of sampling depth. To assess global sampling depth, average spectral counts were calculated by summing all spectral count numbers for all P. gingivalis proteins in the FileMaker script output described under Methods and dividing by the total number of P. gingivalis proteins in that file. The average redundant spectral count number for peptides unique to a given ORF for P. gingivalis alone was 80, for P. gingivalis in the community it was 64. The lower number of counts observed for P. gingivalis proteins in the community is consistent with the added sampling demands placed on the analytical system by sequence overlaps in the proteomes of all three microbes and thus the smaller number of unique proteolytic fragments predicted. More discussion of this topic is given in the explanatory notes [see Additional File 1].
Spectral count values for individual proteins are given in data Additional file 1: Table ST1. Details regarding access to mass spectrometry data for individual peptides and their SEQUEST database searching scores [34], p-values and q-values are given in the notes to the data tables [see Additional File 1].

Proteins and functions differentially regulated by P. gingivalis in a community
Cell envelope and cell structure In bacterial communities significant surface-surface contact occurs both within and among accumulations of the constituent species, as was also observed in the P. gingivalis-F. nucleatum-S. gordonii consortia. Regulation of outer membrane constituents of P. gingivalis would thus be predicted in the context of a community and this was borne out by the proteomic results. Overall, 84 proteins annotated as involved in the cell envelope were detected, and 40 of these showed reduced abundance in the three species community, indicating an extensive change to the cell envelope. Only four proteins showed increased abundance, two OmpH proteins (PGN0300, PGN0301) and two lipoproteins (PGN1037, PGN1998). MreB (PGN0234), a bacterial actin homologue that plays a role in determining cell shape, showed almost a 2-fold decrease in community derived P. gingivalis. Expression of MreB has been found to decrease under stress or during stationary phase in Vibrio paraheamolyticus [35]. However, stress-related proteins were generally reduced in P. gingivalis cells in the community (see below) so stress is an unlikely explanation for the change in MreB. Rather, the decrease in MreB abundance may be due to the P. gingivalis cells entering a state resembling stationary phase or responding in a previously unseen way to the formation of the three species community.

Protein synthesis
Extensive changes were observed in ribosomal proteins and in translation elongation and initiation proteins. While overall more proteins showed reduced abundance in the three species community, the changes to the translational machinery were almost exclusively increases in abundance. Of 49 ribosomal proteins detected, 27 showed increased abundance, while only one showed decreased abundance. Of nine translation elongation and initiation proteins detected, none showed significant abundance decreases but five showed increased abundance (EfG (PGN1870), putative EfG (PGN1014), EfTs (PGN1587), EfTu (PGN1578), and If2 (PGN0255)). This represents not only a substantial portion of the translational machinery but also a large portion, 36%, of the proteins showing increased abundance. It is well known that ribosomal content is generally proportional to growth rate [36]; however, given that the cells were not in culture medium during the assay, rapid growth is an unlikely explanation for these results. The increased ribosomal content presumably indicates increased translation, consistent with the community providing physiologic support to P. gingivalis and allowing higher levels of protein synthesis.

Vitamin synthesis
Pathways for synthesizing several vitamins showed reduced protein abundance in the three species community. Most of the proteins involved in thiamine diphosphate (vitamin B1) biosynthesis were downregulated (Fig.  4). Thiamine is a cofactor for the 2-oxoglutarate dehydrogenase complex that converts 2-oxoglutarate to succinyl-CoA and for the transketolase reactions of the anaerobic pentose phosphate pathway [37]. However, transketolase (PGN1689, Tkt) showed no abundance change while of the three components of the 2-oxoglutarate dehydrogenase complex (PGN1755, KorB) only the beta subunit showed an abundance increase.
Only incomplete pathways have been identified for many of the other vitamin biosynthesis activities in P. gingivalis. However, cobalamin (vitamin B12) synthesis [38] can be predicted to be decreased in the community, with five (PGN0010, CobC; PGN0316, CbiG; PGN0317, CobL; PGN0318, CobH/CbiC; PGN0735, CobU) of the seven identified proteins having statistically significant reductions. Less complete population of pathways was observed for pyridoxal phosphate (vitamin B6) and biotin synthesis. Only two of the four detected proteins for vitamin B6 synthesis showed reduced abundance (PGN1359, PdxB and PGN2055, PdxA). For biotin synthesis, three of the six detected proteins showed reduced abundance (PGN0133, BioA; PGN1721, BioF; PGN1997, BioD). None of the vitamin/cofactor synthesis pathways showed any indication of increased protein levels in the three species community.
The decrease in several vitamin/cofactor pathways could be due to a decreased utilization of those cofactors. However, in the case of thiamine, the proteins that utilize this cofactor showed no decrease, and a possible increase in abundance, implying that demand for vitamin B1 was unchanged. A more likely explanation for the reduced cofactor pathways is therefore nutrient transfer. Either one or both of the other organisms in the three species community could be providing P. gingivalis with cofactors, allowing reduced cofactor synthesis without reducing expression of the cofactor dependent pathways. Nutritional cross-feeding among members of oral biofilms is well established [5], and indeed P. gingivalis has been found to utilize succinate produced by T. denticola [39].

Nucleotide synthesis
Pyrimidine biosynthesis appeared to be reduced in the three species community (Fig. 5) as many of the proteins leading to the production of finished pyrimidine nucleotides have decreased abundance. However, the proteins responsible for incorporating finished ribonucleotides into RNA show unchanged or increased abundance. As with vitamin biosynthesis this may be the result of nutrient transfer from the other organisms in the community. P. gingivalis can acquire nucleosides and nucleobases and it has even been suggested that they may represent an important nutrient source for P. gingivalis [40]. Consistent with uptake of nucleosides and their precursors, uracil permease (PGN1223) shows increased expression in the three species community.
Purine biosynthesis does not appear to be significantly effected in the three species community (Fig. 6). A few proteins showed reduced abundance, but the central biosynthesis pathway was primarily unchanged. Only two stress proteins showed increased abundance, and then only 30% increases, the molecular chaperone DnaK (PGN1208) and a PhoH family protein possibly involved in oxidation protection (PGN0090).

Role of the differentially regulated P. gingivalis protein HmuR
To begin to test the functional relevance of proteins identified as differentially regulated in the three species community, we undertook a mutational analysis. For this Pyrimidine biosynthetic pathway, showing protein abundance changes for the P. gingivalis-F. nucleatum-S. gordonii/P. gingivalis comparison Figure 5 Pyrimidine biosynthetic pathway, showing protein abundance changes for the P. gingivalis-F. nucleatum-S. gordonii/P. gingivalis comparison. The protein names follow the same conventions as in Fig. 4 purpose it was important to target a protein that directly effectuates a biological function and lacks homologs in the genome. HmuR, a major hemin uptake protein, and potential adhesin [42], was selected. As shown in Fig. 7A, while wild type P. gingivalis cells are abundant within a S. gordonii-F. nucleatum-P. gingivalis community, P. gingivalis cells lacking HmuR are deficient in community formation. Biovolume analysis showed a 70% reduction in community formation by the HmuR mutant (Fig. 7C). Furthermore, this effect was specific for the three species community as a decrease in accumulation by the HmuR deficient mutant was not observed in monospecies biofilms, or in two species communities of P. gingivalis with either S. gordonii or F. nucleatum (Fig. 7B, D-G). Hence loss of HmuR, that is up-regulated by P. gingivalis when the organism is associated with S. gordonii and F. nucleatum, results in a phenotype that is restricted to three species community formation. P. gingivalis cells were first cultured in hemin excess, under which conditions the hmu operon is expressed at a basal level [42]. As the three species model system involves metabolically quiescent P. gingivalis cells in buffer, it is unlikely that the role of HmuR is related to its hemin uptake capacity. However, TonB dependent receptors can exhibit functions distinct from transport across the outer membrane. For example, in E. coli the TonB dependent catecholate siderophore receptor  HmuR mutant of P. gingivalis is deficient in community accumulation Iha confers an adhesin function and contributes to colonization and virulence in the mouse urinary tract [43]. Hence, HmuR may have a cohesive function in community formation by P. gingivalis although further studies are necessary to resolve this issue.

Conclusion
Complex multi-species biofilms such as pathogenic dental plaque accumulate through a series of developmental steps involving attachment, recruitment, maturation and detachment. Choreographed patterns of gene and protein expression characterize each of these steps. In this study we developed a model of the early stages of plaque development whereby three compatible species accreted into simple communities. P. gingivalis increased in biomass due to attachment and recruitment, and this allowed us to catalog differential protein expression in P. gingivalis consequent to contact dependent interbacterial signaling and communication through short range soluble mediators. The proteomic analysis indicated that around 40% of P. gingivalis proteins exhibit changes in abundance in a community with F. nucleatum and S. gordonii, implying extensive interactions among the organisms. The proteomic results were consistent with the formation of a favorable environment in a P. gingivalis-F. nucleatum-S. gordonii community, wherein P. gingivalis showed evidence of increased protein synthesis and decreased stress. Moreover, nutrient transfer may occur among the constituents of the community. As evidenced by HmuR, these proteins may have a functional role in the development of multispecies communities and ultimately shape the pathogenic potential of plaque.

Proteomics of model bacterial communities
High density bacterial communities were generated by the method of Merritt et al. [44]. Bacteria were cultured to mid-log phase, harvested by centrifugation and resuspended in pre-reduced PBS (rPBS). 1 × 10 9 cells of P. gin-givalis were mixed with an equal number of S. gordonii and F. nucleatum as a combination of the three species. P. gingivalis cells alone were also used as a control. Two independent biological replicates from separate experiments comprised of at least two technical replicates were analyzed. Bacteria were centrifuged at 3000 g for 5 min, and pellets were held in 1 ml pre-reduced PBS in an anaerobic chamber at 37°C for 18 h. The bacterial cells remain viable under these conditions, as determined by both colony counts and live/dead fluorescent staining. Supernatant and bacterial cells were separated and processed separately. Bacterial cells were lysed with ice cold sterile distilled water and proteins were digested with trypsin as previously described for P. gingivalis [33], then fractionated on a 2.0 mm × 150 mm YMC polymer C18 column. There were five pre-fractions collected for each cellular sample, with a final volume of 50 μl for each fraction. The 2D capillary HPLC/MS/MS analyses [32,45,46] were conducted using an in-house fabricated semi-automated system, consisting of a Thermo LTQ mass spectrometer   [50]) concatenated with the reversed sequences. After data processing, the genome sequence for strain 33277 became available [31] and the data were subsequently cross-referenced to PGN numbers from the 33277 specific FASTA database provided by LANL (personal communication with G. Xie). Although Naito et al. [31] reported extensive genome re-arrangements between W83 and ATCC 33277, the actual protein amino acid sequences are sufficiently similar across the proteome that the use of a database based on W83 was not expected to greatly impact the analysis. Our proteomic methods are not sensitive to genome re-arrangements, only to changes in amino acid sequence for a given protein. The reversed sequences were used for purposes of calculating a peptide level qualitative FDR using the published method [51,52]. The SEQUEST peptide level search results were filtered and grouped by protein using DTASelect [53], then input into a FileMaker script developed in-house [32,33] for further processing. The DTASelect Ver. 1.9 filter parameters were: peptides were fully tryptic; ΔCn/Xcorr values for different peptide charge states were 0.08/1.9 for +1, 0.08/2.0 for + 2, and 0.08/3.3 for +3; all spectra detected for each sequence were retained (t = 0). Only peptides that were unique to a given ORF were used in the calculations, ignoring tryptic fragments that were common to more than one ORF or more than one organism, or both. In practice this had the consequence of reducing our sampling depth from what we have achieved with single organism studies [27,32,33], because the gene sequence overlap among the three organisms is significant. A bioinformatic analysis (data not shown) of inferred protein sequence overlaps between P. gingivalis and S. gordonii or F. nucleatum suggested the reduction in the number of predicted tryptic fragments unique to P. gingivalis would not be sufficient to impact the analysis of more than a small number of proteins. The qualitative peptide level FDR was controlled to approximately 5% for all conditions by selecting a minimum non-redundant spectral count cut-off number appropriate to the complexity of each condition, P. gingivalis alone or the P. gingivalis-F. nucleatum-S. gordonii community.

Protein abundance ratio calculations
Protein relative abundances were estimated on the basis of summed intensity or spectral count values [27,32,33] for proteins meeting the requirements for qualitative identification described above. Summed intensity refers to the summation of all processed parent ion (peptide) intensity measurements (MS 1 ) for which a confirming CID spectrum (MS 2 ) was acquired according to the DTASelect filter files. For spectral counts, the redundant numbers of peptides uniquely associated with each ORF were taken from the DTAselect filter table (t = 0). Spectral counting is a frequency measurement that has been demonstrated in the literature to correlate with protein abundance [54]. These two ways of estimating protein relative abundance, that avoid the need for stable isotope labeling, have been discussed in a recent review [27] with specific reference to microbial systems. To calculate protein abundance ratios, a normalization scheme was applied such that the total spectral counts or total intensities for all P. gingivalis proteins in each condition were set equal for each comparison. This normalization also had the effect of zero centering the log 2 transformed relative abundance ratios, see Fig. 2 (and also the frequency histograms in Additional file 1: Figs. SF5 and SF6). The normalized data for each abundance ratio comparison was tested for significance using either a global G-test or a global paired t-test for each condition, the details of which have been published for this type of proteomics data in which all biological replicates are compared against each other [55,56], and are also described in the explanatory notes [see Additional File 1]. Both of these testing procedures weigh deviation from the null hypothesis of zero abundance change and random scatter in the data to derive a probability or p-value that the observed change is a random event, i.e. that the null hypothesis of no abundance change is true. Each hypothesis test generated a p-value that in turn was used to generate a q-value as described [24,32], using the R package QVALUE [26]. The q-value in this context is a measure of quantitative FDR [25] that contains a correction for multiple hypothesis testing. A q cut-off value of 0.01 was used for all ratios reported in Additional file 1: Table ST1. All statistical calculations were done in R (Ver. 2.5.0), using source code that has been published [32,33,55]. Only proteins with data consisting of confirmed high scoring MS 2 mass spectra (high scoring qualitative database matches as described above) present in both the numerator and denominator of the abundance ratio comparison were listed as significantly changed in Additional file 1. Certain proteins listed in the tables with q-values = 0.01 are still coded yellow for no significant abundance change due to missing data in either the numerator or the denominator.

Ontology analysis
An overall list of detected proteins as well as lists of proteins that showed increased or decreased levels in the three species community were prepared using Entrez gene identifiers. Ontology analyses were then conducted using the DAVID [57] functional annotation clustering feature with the default databases. Both increased and decreased protein level lists were analyzed using the overall list of detected proteins as the background. Potentially interesting clusters identified by DAVID were then examined manually.

Construction of P. gingivalis HmuR mutant
A mutation in the hmuR gene was generated using ligation-independent cloning of PCR mediated mutagenesis (LIC-PCR) [58]. A 2.1-kb ermF-ermAM cassette was introduced into the hmuR gene by three steps of PCR to yield a hmuR-erm-hmuR DNA fragment as described previously [59]. The fragment was then introduced into P. gingivalis 33277 by electroporation. The hmuR deficient mutant (ΔhmuR) was generated via a double crossover event that replaces hmuR with the hmuR-erm-hmuR DNA fragment in the 33277 chromosome. The mutants were selected on TSB plates containing erythromycin (5 μg/ml), and the mutation was confirmed by PCR analysis. Growth rates of mutant and parent strains were equivalent.
Quantitative community development assays i) Crystal violet assay. Homotypic community formation by P. gingivalis was quantified by a microtiter plate assay [60], as adapted for P. gingivalis [61]. Parental and mutant strains in early log phase (2 × 10 8 cells) were incubated at 37°C anaerobically for 24 h. Wells were washed, stained with 1% crystal violet and destained with 95% ethanol. Absorbance at 595 nm was determined in a Benchmark microplate reader. ii) ELISA. F. nucleatum was incubated at 37°C anaerobically for 36 h in microtiter plate wells. After washing, parental and mutant P. gingivalis strains (2 × 10 6 cells) were incubated with the fusobacterial biofilm at 37°C anaerobically for 24 h. P. gingivalis accumulation was detected with antibodies to whole cells (1:10,000) followed by peroxidase-conjugated secondary antibody (1:3,000), each for 1 h at 37°C. Antigen-antibody binding was determined by a colorimetric reaction using the 3,3',5,5'-tetramethylbenzidine (TMB) liquid substrate, and absorbance at 655 nm. P. gingivalis antibody binding to the fusobacterial biofilm alone was subtracted as background. iii) Confocal microscopy assay. A. Single species. P. gingivalis was stained with 4',6-diamidino-2-phenylindole (50 μg ml -1 ) and 2 × 10 6 cells in rPBS incubated anaerobically at 37°C for 16 h with rocking in individual chambers of the CultureWell coverglass system (Grace Bio Labs). Chambers were washed three times in rPBS. B. Dual species. Heterotypic P. gingivalis-S. gordonii communities were generated as described previously [15]. S. gordonii cells were labeled with hexidium iodide (15 μg ml -1 ), then cultured anaerobically at 37°C for 16 h with rocking in CultureWell chambers. P. gingivalis was stained with 5-(and-6)-carboxyfluorescein, succinimidyl ester (10 μg ml -1 ), and 2 × 10 6 cells in rPBS were reacted with the surface attached S. gordonii for 24 h anaerobically at 37°C with rocking. C) Three species. Surface attached hexidium iodide-stained S. gordonii were generated as above. Fluorescein stained F. nucleatum (2 × 10 6 cells in rPBS) reacted with S. gordonii for 24 h anaerobically at 37°C with rocking. The coverglass was then washed with rPBS to remove non-attached bacteria. P. gingivalis was stained with 4',6diamidino-2-phenylindole (50 μg ml -1 ) and 2 × 10 6 cells in rPBS were added and further incubated for 24 h anaerobically at 37°C with rocking. Communities were observed on a Bio-Rad Radiance 2100 confocal laser scanning microscope (Blue Diode/Ar/HeNe) system with an Nicon ECLIPSE TE300 inverted light microscope and 40 × objective using reflected laser light of combined 405, 488 and 543 nm wavelengths where appropriate. A series of fluorescent optical x-y sections were collected to create digitally reconstructed images (z-projection of x-y sections) of the communities with Image J V1.34s (National Institutes of Health) or Laser Sharp software (Bio-Rad). Z stacks of the x-y sections of CLSM were converted to composite images with "Iso Surface" functions of the "Surpass" option on Imaris 5.0.1 (Bitplane AG; Zurich, Switzerland) software. Iso Surface images of P. gingivalis were created at threshold of 20 and smoothed with Gaussian Filter function at 0.5 width, and P. gingivalis biovolume was calculated.
Biofilm assays were repeated independently three times with each strain in triplicate. Crystal violet results were compared by t-tests. Biovolume calculations were compared with a t-test using the SPSS statistics software.

Authors' contributions
MK carried out the community construction and analysis by confocal microscopy; ELH did the pathway analysis; QX and TW performed the protein biochemistry, separations and mass spectrometry; HX constructed the hmuR mutant; MH and RJL conceived the experiments. MH, ELH, MK and RJL wrote the manuscript. MK and ELH contributed equally.