Pathway analysis for intracellular Porphyromonas gingivalis using a strain ATCC 33277 specific database

Re-analysis using the P. gingivalis strain ATCC 33277 genome annotation

The proteomics data previously analyzed using the strain W83 genome annotation [GenBank: AE015924] [9] was recalculated employing the strain specific P. gingivalis strain ATCC 33277 annotation [GenBank: AP009380]. Accurately identifying a proteolytic fragment using mass spectrometry-based shotgun proteomics as coming from a particular protein requires matching the MS data to a protein sequence. Differences in amino acid sequence between the proteins expressed by strain ATCC 33277 and the protein sequences derived from the strain W83 genome annotation rendered many tryptic peptides from the whole cell digests employed unidentifiable in the original analysis [9]. Given that the quantitative power of the whole cell proteome analysis is dependent on the number of identified peptides [12, 14], the new analysis was expected to give a more complete picture of the differential proteome, an expectation that proved accurate. In addition, some proteins in the strain ATCC 33277 genome are completely absent in the strain W83 genome and were thus qualitatively undetectable in the original analysis.

Overall, 1266 proteins were detected with 396 over-expressed and 248 under-expressed proteins observed from internalized P. gingivalis cells compared to controls (Table 1). Statistics based on multiple hypothesis testing and abundance ratios for all detected proteins can be found in Additional file 1: Table S1, as well as pseudo M/A plots [15] of the entire dataset. The consensus assignment given in Additional file 1: Table S1 of increased or decreased abundance was based on two inputs, the q-values for comparisons between internalized P. gingivalis and gingival growth medium controls as determined by spectral counting and summed signal intensity from detected peptides that map to a specific ORF [9, 14, 15]. If one or the other of the spectral counting or protein intensity indicated a significant change (q ≤ 0.01) and the other measure showed at least the same direction of change with a log₂ ratio of 0.1 or better, then the consensus was considered changed in that direction, coded red for over-expression or green for under-expression. A simple "beads on a string" genomic map of the consensus calls is shown in Fig. 1.

Whole cell proteomics measurements of this type are noisy and the trade off between quantitative FDR (false discovery rate) and FNR (false negative rate) is made based on the informed judgment of the analyst, and often tends to be ad hoc and arbitrary in practice [9, 14]. The q-value cut-off of 0.01 used here for statistical significance based on formal hypothesis testing was in good agreement with experimentally derived error distributions, as illustrated by the two pseudo M/A plots given in Additional file 1. The present findings serve to show the value of examining trends in groups of proteins, both as an end in itself with respect to biological questions and as feedback in the determination of proper cut-off values for the quantitative significance testing of individual proteins. As proteomics technology improves and it becomes economically feasible to run a greater number of independent cultures (biological replicates) than what was possible here, the overall noise issue in any one set of measurements will be less of a concern, and it will be easier to distinguish biological noise from deficiencies with respect to analytical repeatability, and thus identify biological trends that are truly significant rather than stochastically driven. Nonetheless, as in our previous work [9] the trends identified here are consistent with what we know about the behavior of the organism under intracellular conditions [3, 9, 16].

Comparison between W83 and ATCC 33277 annotations for proteomics

Metabolic pathways differentially regulated in internalized P. gingivalis

The consensus assignments (see Additional file 1: Table S1) of differentially expressed proteins were used to populate metabolic pathways. The results were analyzed manually using the ATCC 33277 genome annotation [11]. In addition, an ontology analysis was done using DAVID (the Database for Annotation, Visualization and Integration Discovery) to identify over- or under-expressed ontology categories [17]. Putative changed categories were then checked manually. DAVID has proven to be useful for prokaryotes when compared with other ontology programs [18].

Energy metabolism

P. gingivalis is an asaccharolytic bacterium and cannot survive on glucose or carbohydrates alone. While some genes for carbohydrate metabolism are found in the genome, P. gingivalis derives its energy from the metabolism of amino acids [11, 13]. Takahashi and colleagues measured amino acid usage in culture and found that glutamate/glutamine and aspartate/asparagine were preferentially metabolized [13]. When grown on dipeptides of these substrates, P. gingivalis produced different amounts of metabolic byproducts. Importantly, aspartylaspartate produced significantly higher amounts of acetate, which is associated with ATP formation (Fig. 2 and Additional file 1: Table S1). Internalized P. gingivalis cells showed an increase in the energy pathway from aspartate/asparagine to acetate and energy (Fig. 2). The corollary of this trend is that the intracellular environment is energy rich for P. gingivalis. Interestingly, the protein that converts glutamate, the other favored amino acid, to 2-oxoglutarate (PGN1367, glutamate dehydrogenase) showed a decrease in abundance (Fig. 2). This may represent a preference for energy production in internalized cells or be part of a more general shift in the metabolic byproducts. We also observed a decrease in protein abundance of maltodextrin phosphorolase (PGN0733). Maltodextrin phospholase plays a role in digesting starches and, despite being an asaccharolytic organism, P. gingivalis may make some use of the starches available in the oral cavity, but restricts this activity after internalization.

Cytotoxic byproducts

P. gingivalis metabolism produces several short chain fatty acid byproducts that are cytotoxic (Fig. 2) and has been found to shift production between these compounds depending on growth conditions [13]. We have found a general increase in the pathway from 2-oxoglutarate to the cytotoxin propionate while the proteins in the pathways for production of the cytotoxin butyrate showed unchanged or reduced expression (Fig. 2). This is consistent with hints that byproduct production shifts away from butyrate and towards propionate during P. gingivalis infections [19]. The results are the opposite of what would be expected from substrate studies. As mentioned previously, the proteomics shows an increase in the aspartate/asparagine pathway and a reduction in glutamate/glutamine. Culture growth studies found that P. gingivalis grown on aspartylaspartate had significantly more butyrate production than propionate compared to cultures grown on glutamylglutamate [13]. However, a recent flux balance model of P. gingivalis metabolism predicts that there is abundant flexibility in the production of butyrate, propionate and succinate with the metabolic routes to each being equivalent with respect to redox balancing and energy production [20]. Thus a shift towards propionate could be easily explained if it presented an advantage to internalized cells. In that regard, it has been shown that butyrate is a more potent apoptosis inducing agent than propionate [21]. Hence, the diminished production of butyrate by internalized P. gingivalis may contribute to the resistance of P. gingivalis-infected GECs to apoptotic cell death [22]. There is also the question of the reduced abundance of glutamate dehydrogenase (PGN1367), the protein that converts glutamate to 2-oxoglutarate (Fig. 2). If this is the primary substrate for propionate production it could limit that production even with increased abundance in the rest of the pathway. However, 2-oxoglutarate is a common metabolic intermediate and glutamate/glutamine may not be the only source of 2-oxoglutarate for propionate production. Even if it is the primary source, given the flexibility in byproduct production, a significant shift away from butyrate production from glutamate/glutamine to propionate production could still occur in the presence of an overall reduction in glutamate/glutamine usage. Interestingly, some similar shifts are seen between planktonic cells and biofilms of P. gingivalis strain W50. A mass spectrometry analysis of planktonic cells versus biofilm cells identified 81 proteins and found several energy metabolism proteins with significant differences between planktonic and biofilm lifestyles [23]. In biofilms fumarate reductase (PGN0497, 0498) had reduced abundance while oxaloacetate decarboxylase (PGN0351) had increased abundance similar to what we see in internalized cells (Fig. 2). Obviously, biofilms and the interior of GECs are different environments, and the energy metabolism protein glyceraldehyde-3-phosphate dehydrogenase (PGN0173) was increased in biofilms [23] relative to planktonic cells, while it is decreased in internalized cells relative to external controls. A comparison between the two conditions would really require the identification of more metabolic proteins from biofilm cells, but given the relevance of biofilm formation to P. gingivalis pathogenicity in vivo [24–26], the relation between biofilm conditions and internalized cells is an interesting one that we intend to pursue further at the whole proteome level.

Translation machinery

Transcription machinery

Proteomic methods

The bacterial and gingival cell culturing, sample preparation, proteome extraction, proteolytic digestion, HPLC pre-fractionation, 2-D capillary HPLC [31, 32], LTQ linear ion trap mass spectral data acquisition parameters, Sequest database searching [33], DTASelect [34]in silico assembly of the P. gingivalis proteome, protein relative abundance calculations, statistical methods and analytical validation for FDR and FNR [14] were all as published in the previous paper [9], with the following exceptions. The processing of the raw mass spectral data differs in this report due to the genome sequence annotation specific to strain ATCC 33277 [11], [GenBank: AP009380] which served as the basis for a new ORF database prepared by LANL (Los Alamos National Laboratory, Gary Xie, private communication). The custom database prepared by LANL was combined with reversed sequences from P. gingivalis ATCC 33277, human and bovine proteins as with our W83 database [GenBank: AE015924] described previously. The total size of the combined fasta file was 116 Mbytes. The estimated random qualitative FDR for peptide identifications based on the decoy strategy [35, 36] was 3%.

Assignment of ORF numbers

Additional file 1: Table S1 is arranged in ascending order by PGN numbers assigned for the experimental strain used here by Naito et al. [11]. They have been cross referenced to the W83 PG numbers originally assigned both by TIGR-CMR and LANL, where it was possible to do so. Certain ATCC 33277 genes do not have a counterpart in the older annotations based on the W83 genome, and will thus be blank in the summary table for PG numbers.

DAVID

An overall list of detected proteins as well as lists of proteins that showed increased or decreased levels between internalized and gingival growth medium cultured cells were prepared using Entrez gene identifiers, as DAVID [17] does not recognize PGN numbers. Ontology analyses were then conducted using the DAVID 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.