Gingimaps: Protein Localization in the Oral Pathogen Porphyromonas gingivalis

Porphyromonas gingivalis is an oral pathogen involved in the widespread disease periodontitis. In recent years, however, this bacterium has been implicated in the etiology of another common disorder, the autoimmune disease rheumatoid arthritis. Periodontitis and rheumatoid arthritis were known to correlate for decades, but only recently a possible molecular connection underlying this association has been unveiled.

SUMMARY

Porphyromonas gingivalis is an oral pathogen involved in the widespread disease periodontitis. In recent years, however, this bacterium has been implicated in the etiology of another common disorder, the autoimmune disease rheumatoid arthritis. Periodontitis and rheumatoid arthritis were known to correlate for decades, but only recently a possible molecular connection underlying this association has been unveiled. P. gingivalis possesses an enzyme that citrullinates certain host proteins and, potentially, elicits autoimmune antibodies against such citrullinated proteins. These autoantibodies are highly specific for rheumatoid arthritis and have been purported both as a symptom and a potential cause of the disease. The citrullinating enzyme and other major virulence factors of P. gingivalis, including some that were implicated in the etiology of rheumatoid arthritis, are targeted to the host tissue as secreted or outer-membrane-bound proteins. These targeting events play pivotal roles in the interactions between the pathogen and its human host. Accordingly, the overall protein sorting and secretion events in P. gingivalis are of prime relevance for understanding its full disease-causing potential and for developing preventive and therapeutic approaches. The aim of this review is therefore to offer a comprehensive overview of the subcellular and extracellular localization of all proteins in three reference strains and four clinical isolates of P. gingivalis, as well as the mechanisms employed to reach these destinations.

INTRODUCTION

Porphyromonas gingivalis is a Gram-negative, black-pigmented anaerobic bacterium belonging to the Bacteroidetes phylum. Although this bacterium is often described as rod-shaped, its appearance is more reminiscent of a little sausage (Fig. 1). P. gingivalis initially garnered interest as a model organism for bacteria of the Cytophaga-Flavobacterium-Bacteroides (CFB) group and, later on, as an oral pathogen. The focus on P. gingivalis has recently peaked with the discovery of a new protein secretion system (1) and with the evidence of its involvement in Alzheimer’s disease (2). However, this bacterium is best known as a major etiological agent of the oral disease periodontitis (3, 4), being present in almost 85% of severe cases (5–7). Periodontitis is an inflammatory disorder affecting the tissue surrounding the teeth, the periodontium, potentially leading to tooth loss. Severe forms of periodontitis have a global prevalence of ∼11%. However, depending on the degree of severity, socioeconomic status, and oral hygiene, this disease can affect up to 57% of particular populations (8, 9). In the United States, for example, 46% of adults are affected by this disorder, with 8.9% presenting severe forms (9). This extremely high incidence establishes periodontitis as one of the most common diseases and as the main cause of tooth loss worldwide (9, 10).

FIG 1.

P. gingivalis: electron micrographs of P. gingivalis type strain W83 (A) and the clinical strains 505700 (B), 512915 (C), 505759 (D), and MDS33 (E and F). Note the capture of OMV formation in panels A, B, C, and E (marked by white arrows).

Interestingly, periodontitis has been associated with several health conditions, such as diabetes, heart diseases, Alzheimer’s disease, and rheumatoid arthritis (RA). In the case of diabetes, a two-way relationship was proposed, where the inflammatory mediators released in response to a periodontal infection would have an adverse effect on glycemic control, while diabetes-driven factors such as impaired chemotaxis, reduced collagen synthesis, and increased collagenase production would, in turn, enhance the severity of periodontitis (11–16). The association between periodontitis and heart diseases, on the other hand, is more tenuous than the one with diabetes, and no potential mechanistic links are currently known (17–19). Investigations on the association of periodontitis with dementia support the potential involvement of periodontitis in this cognitive disorder both at the immunomodulatory level, which would relate to the systemic inflammatory responses caused by this oral disease, and at the physiological level, which could relate to possible micronutrient deficiencies (e.g., for thiamine and vitamin B₁₂) that may arise from dietary changes as a consequence of tooth loss and that potentially lead to cognitive impairment (20, 21). A special case has been made for the most common type of dementia, Alzheimer’s disease, where P. gingivalis has been proposed to play a significant role (2, 20–23). In particular, it was suggested that the secretion of particular cysteine proteases called gingipains may cause neuronal damage, which would be supported by the fact that these proteases, along with bacterial DNA, were detected in the brains of Alzheimer’s disease patients (2). Lastly, the association of periodontitis with RA has been studied most intensively (24–42). RA is an inflammatory autoimmune disorder for which the etiology is still not fully understood and that is clinically associated with periodontitis. In several countries, the prevalence of periodontitis was reported to be increased among RA patients in comparison with the general population (24, 29, 36, 37, 40, 43, 44). Correspondingly, RA was found to be more prevalent among patients with periodontitis (35–37, 40, 44), which supports the hypothesis that an intimate connection exists between the two disorders.

The suspected role of P. gingivalis in the interplay between periodontitis and RA has drawn attention to the bacterium’s citrullinating enzyme (25, 27, 28, 32, 34, 41). This enzyme, a peptidylarginine deiminase (PAD), catalyzes the conversion of arginine into citrulline residues in a posttranslational protein modification called citrullination. Citrullination can alter the net charge of a substrate protein, possibly leading to severe changes in its structure and function (27). Although citrullination is a physiological process that takes place in a wide variety of healthy tissues as a general regulatory mechanism, especially during apoptosis, it is also associated with inflammatory processes.

Although peptidylarginine deiminases are highly conserved in mammals, only three bacteria of the genus Porphyromonas are known to produce such enzymes (27, 38, 45–47). The PAD of P. gingivalis (PPAD) and the homologous enzymes from Porphyromonas loveana and Porphyromonas gulae share no evolutionary relationship with the mammalian PADs (47, 48). Remarkably, PPAD is believed to citrullinate certain human host proteins that, especially in genetically predisposed subjects (27, 34), can stimulate the production of anti-citrullinated protein antibodies (ACPAs) (27, 31, 38, 39, 45). These ACPAs have 95% specificity and 68% sensitivity for RA (49, 50). Interestingly, like many other bacterial virulence factors, the citrullinating enzyme of P. gingivalis is targeted to the host milieu. In particular, PPAD was detected in gingival tissue of patients with severe periodontitis (51). In vitro studies have shown that P. gingivalis secretes PPAD both in a secreted soluble state and in an outer membrane vesicle (OMV)-bound state (45, 52, 53). In addition, a substantial portion of PPAD remains associated with the bacterial cell in an outer membrane (OM)-bound state. Other proteins that play a role in P. gingivalis colonization of the periodontal pockets, such as hemagglutinins and fimbrial components, or that have been implicated in RA etiology, such as several cysteine proteases, are exposed on the bacterial cell surface (54). These findings and the possible roles of P. gingivalis in RA and other diseases focus interest on the mechanisms and pathways responsible for protein sorting and export in this bacterium. In this context, it is noteworthy that P. gingivalis is an extremely successful oral pathogen that does not only take advantage of proteinaceous virulence factors, but also of nonproteinaceous virulence factors, such as capsule (55, 56) and lipopolysaccharides (57), to manifest itself as a “keystone” species within subgingival biofilms (58, 59).

All factors that contribute to the success of P. gingivalis in the periodontal pockets, where it mainly resides, are contained in the bacterium’s proteome. Importantly, the presence of particular bacterial proteins in a specific subcellular compartment is related to their biological function. Secreted proteins, for example, are involved in processes that take place at the cell surface or beyond, such as nutrient acquisition, cell motility, cell-cell communication, or host colonization and invasion. Accordingly, bacterial cell surface proteins represent excellent targets for drugs or vaccines (60). Moreover, knowledge of the subcellular localization of proteins is an invaluable tool for genome annotation and the interpretation of proteomics data. This review is therefore aimed at providing a comprehensive overview on protein localization in P. gingivalis making use of an in-depth bioinformatic reappraisal of previously published biochemical, genomic, and proteomic studies (52, 61–87).

GENERAL ARCHITECTURE AND SUBCELLULAR COMPARTMENTS

To predict the subcellular localization of proteins in a bacterium, it is necessary to first gather information on this bacterium’s cellular architecture. The knowledge of subcellular compartments is in fact required to develop a species-specific prediction strategy. In this review, protein localization in a total of seven P. gingivalis strains was evaluated, including three reference strains and four clinical isolates. The three investigated reference strains—W83, TDC60, and ATCC 33277—are the main and best-studied P. gingivalis strains with publicly available genome sequences that were manually curated. Their proteomes were accessed and downloaded from UniprotKB (88) on 11 March 2018: W83 (UP000000588), ATCC 33277 (UP000008842), and TDC60 (UP000009221). The included clinical isolates (20655, MDS140, MDS33, and 512915) (46, 53) can be divided in PPAD sorting types I and type II (53). This classification concerns the differential sorting of PPAD as recently detected in one of our studies on clinical isolates (53). Compared to sorting type I isolates, sorting type II isolates display an extremely hampered production of OM- and OMV-bound PPAD, which appears to be due to a Gln to Lys amino acid substitution at position 373 of this protein (53). The two sorting type I isolates, 20655 and MDS140, were obtained from a patient with severe periodontitis but no RA, and a healthy carrier, respectively (46). The sorting type II isolates (512915 and MDS33), on the other hand, were isolated from a periodontitis patient without RA and a patient with severe periodontitis and RA, respectively (53). Of note, the previous study during which the sorting type I and II isolates were identified showed that neither the association of PPAD with vesicles nor the vesiculation of P. gingivalis by itself is a critical determinant for interactions of P. gingivalis with its human host (53) because the sorting type I or II distinction could not be reconciled with the severity of periodontitis according to the Dutch periodontal screening index (25).

Consistent with its status as a Gram-negative bacterium, the protein-containing subcellular compartments of a P. gingivalis cell can be divided into cytoplasm, inner membrane (IM), periplasm, and OM (Fig. 2). Nascent bacteriophages could, in principle, represent a separate intracellular compartment, but to date no bacteriophages have been described for P. gingivalis (89, 90). In addition, some of the proteins like PPAD are targeted to the extracellular milieu, in particular the periodontium of the human host (51). As mentioned above, proteins can be secreted either in a soluble state or bound to OMVs (Fig. 2) (47, 53, 91, 92). Gram-negative bacteria produce OMVs by natural “blebbings” of their outer membrane. Accordingly, OMVs consist of a single membrane originating from the OM, contain OM proteins, lipopolysaccharide (LPS), and other lipids. The cargo of OMVs also includes cytoplasmic and periplasmic proteins (93), but they appear enriched in virulence factors (91). The OMVs of P. gingivalis were shown to be involved in biofilm formation and to have invasive capabilities (92, 94, 95). In particular, the P. gingivalis OMVs were shown to enter host epithelial cells and degrade key receptor proteins using the aforementioned gingipains (96–98). These cysteine proteases are essential for virulence in animal models where they were shown to degrade many host proteins, thereby impairing cellular functions and the host immune response (99, 100). Other studies have implicated OMVs of P. gingivalis in selective tumor necrosis factor tolerance (101), inflammasome activation, and pyroptosis in macrophages (102). Therefore, OMVs represent nonviable satellite compartments of the P. gingivalis cell (Fig. 2 and 3). Of note, the precise mechanisms underlying the formation of OMVs in P. gingivalis are still unknown, and therefore no bioinformatic tool exists or can be created yet to predict which proteins are localized in this peculiar extracellular compartment. Although biochemical studies have investigated the OMV cargo proteins (91), the results are presently still limited to the one strain analyzed. For this reason, the OMV compartment has not been taken into account in our bioinformatic reappraisal of the available data.

FIG 2.

Protein sorting mechanisms in P. gingivalis. An overview diagram shows the cellular architecture and protein transport systems that occur in P. gingivalis. The indicated protein transport systems were identified by domain searches for major components of known transport systems previously observed in Gram-negative bacteria. The Lol, Bam, T4SS, and T5SS systems for which only a limited number of known potential components were identified in P. gingivalis are indicated in parentheses. SP, signal peptide.

FIG 3.

OMVs are the “satellite compartments” of P. gingivalis. A transmission electron micrograph shows the purified outer membrane vesicles of P. gingivalis type strain W83.

SYSTEMS FOR PROTEIN EXPORT FROM THE CYTOPLASM, MEMBRANE INSERTION, AND SECRETION IN P. GINGIVALIS

Knowledge of the subcellular compartments present in P. gingivalis is required for the identification of protein transport, secretion, and membrane insertion systems. Uncovering the suite of such systems used by different P. gingivalis strains will grant a deeper understanding of the ways in which general virulence factors and particular toxins are exported from the cytoplasm and delivered to cells and tissues of the host. This will also highlight possible strain-specific differences. Importantly, certain transport systems are dedicated to the export of proteins with specific functions in virulence, but such systems can also serve other functions. The latter is showcased by the type IV protein secretion system that can also facilitate conjugation (103), or the type IX secretion system that is also used in gliding motility (104). Therefore, a careful dissection of the occurrence of different types of protein sorting and secretion systems may lead to a detailed understanding of a strain’s array of biological capabilities and virulence potential.

In general, Gram-negative bacteria possess an IM and an OM and, for this reason, several export and membrane insertion systems are utilized to translocate proteins across these two membranes and to sort them to their rightful destinations. The vast majority of extracytoplasmic proteins in Gram-negative bacteria is translocated across the cytoplasmic membrane in an unfolded state by the Sec translocase. This includes integral IM proteins and lipoproteins that remain associated with the IM (105) (Fig. 2). A relatively small number of proteins traverses the cytoplasmic membrane by the Tat system, which is specific for cargo proteins in a prefolded state that usually contain cofactors (106–109). Among the proteins that reach the periplasm, the β-barrel proteins can be inserted into the OM by the “β-barrel assembly machinery” (BAM) complex (110, 111), while lipoproteins are inserted into the OM by the “localization of lipoprotein” (Lol) system (112, 113) (Fig. 2). Due to their major roles in protein sorting, these systems are broadly conserved among Gram-negative bacteria and their genes are thus easily recognizable by automated pipelines. Key components of the Sec, Tat, BAM, and Lol systems can be promptly identified by looking for the homologues of known members of these systems in other species. In addition, specific domain searches can be utilized (Table 1).

TABLE 1.

Presence or absence of known protein transport and membrane insertion systems in P. gingivalis^a

Secretion system	Domain	Protein	ATCC 33277	W83	TDC60
Sec	COG0653	SecA	PGN_1458	PG0514	PGTDC60_1633
	Tigr00963	SecA	PGN_1458	PG0514	PGTDC60_1633
	IPR003708	SecB			PGTDC60_1688
	IPR035958	SecB
	IPR027398	SecD first TM region
	IPR005791	SecD	PGN_1702	PG1762	PGTDC60_1374
	IPR005665	SecF	PGN_1702	PG1762	PGTDC60_1374
	IPR022645	SecD/F
	COG0690	SecE	PGN_1577	Present*	PGTDC60_1503
	Tigr00964	SecE	PGN_1577	Present*	PGTDC60_1503
	COG1314	SecG	PGN_0258	PG0144	PGTDC60_0422
	Tigr00810	SecG	PGN_0258	PG0144	PGTDC60_0422
	COG0201	SecY	PGN_1848	PG1918	PGTDC60_0188
	Tigr00967	SecY	PGN_1848	PG1918	PGTDC60_0188
	COG0706	YidC	PGN_1446	PG0526	PGTDC60_1645
	Tigr03592	YidC	PGN_1446	PG0526	PGTDC60_1645
	Tigr03593	YidC	PGN_1446	PG0526	PGTDC60_1645
	COG1862	YajC	PGN_1485	PG0485	PGTDC60_1601
	Tigr00739	YajC	PGN_1485	PG0485	PGTDC60_1601
Sec			+	+	+
TAT	COG0805	TatC
	Tigr00945	TatC
	pfam00902	TatC
	COG1826	TatA/E
	Tigr01411	TatA/E
	Tigr01410	TatB
	Tigr01409	Tat signal
	Tigr01412	Tat signal
	pfam10518	Tat signal
TAT			–	–	–
SRP	IPR004780	Ffh	PGN_1205	PG1115	PGTDC60_1100
	IPR004390	FtsY	PGN_0264	PG0151	PGTDC60_0428
SRP			+	+	+
BAM	Tigr03303	BamA	PGN_0299	PG0191	PGTDC60_0462
	IPR023707	BamA
	Tigr03300	BamB
	IPR017687	BamB
	IPR014524	BamC
	Tigr03302	BamD	PGN_1354	PG1215	PGTDC60_1188
	IPR017689	BamD	PGN_1354	PG1215	PGTDC60_1188
	pfam06804	BamD
	pfam04355	BamE
	IPR026592	BamE
BAM			±	±	±
LOL	Tigr00547	LolA
	pfam03548	LolA
	COG2834	LolA
	Tigr00548	LolB
	COG3017	LolB
	pfam03550	LolB
	Tigr02212	LolC
	Tigr02211	LolD
	Tigr02213	LolE
	COG4591	LolE	PGN_0718	PG0682	PGTDC60_0845
			PGN_0719	PG0683	PGTDC60_1224
			PGN_1025	PG0922	PGTDC60_1807
			PGN_1387	PG1252	PGTDC60_1808
LOL			±	±	±
T1SS	Tigr01842	PrtD
	IPR010128
	Tigr01843	HlyD
	IPR010129
	Tigr01844	TolC
	IPR010130
	Tigr01846	HlyB
	IPR010132
	Tigr03375	LssB
	IPR017750
	pfam02321	Outer membrane efflux protein	PGN_0444	PG0063	PGTDC60_0345
			PGN_0715	PG0094	PGTDC60_0374
			PGN_1432	PG0285	PGTDC60_0631
			PGN_1539	PG0538	PGTDC60_1397
	IPR003423		PGN_1679	PG0679	PGTDC60_1540
			PGN_2012	PG1667	PGTDC60_1656
			PGN_2041		PGTDC60_1804
	pfam03412	Bacteriocin exporter family (peptidase C39 family)			PGTDC60_1000
	IPR005074				PGTDC60_1973
T1SS			–	–	–
T2SS	COG1450	PulD
	COG2804	PulE
	COG1459	PulF
	COG2165	PulG
	IPR013545	PulG
	Tigr02517	Type II secretion system protein D (GspD)
	IPR013356
T2bSS	Tigr02519	Pilus (MSHA type) biogenesis protein MshL
	IPR013358
	Tigr02515	Type IV pilus secretin (or competence protein) PilQ
	IPR013355
	pfam07655	Secretin N-terminal domain
	IPR011514
	pfam07660	Secretin and TonB N terminus short domain
	IPR011662
T2a-cSS, T3aSS	pfam00263	Bacterial type II and III secretion system protein (secretin)
	IPR004846
	pfam03958	Bacterial type II/III secretion system short domain
	IPR005644
T2SS			-	-	-
T3SS	COG1157	FliI
	IPR032463	FliI
	COG1766	FliF
	IPR000067	FliF
	COG1886	FliN
	IPR012826	FliN
T2a-cSS, T3aSS	pfam00263	Bacterial type II and III secretion system protein (secretin)
T2a-bSS, T3aSS	pfam03958	Bacterial type II/III secretion system short domain
T3aSS	Tigr02516	Type III secretion outer membrane pore, YscC/HrcC family
	IPR003522
T3bSS	pfam02107	Flagellar L-ring protein (FlgH)
	IPR000527
T3SS			–	–	–
T4SS	COG3838	VirB2
	IPR007039	VirB2
	COG3702	VirB3
	IPR007792	VirB3
	COG3451	VirB4	PGN_0065	PG1481	PGTDC60_1018
					PGTDC60_1993
	COG3704	VirB6
	IPR007688	VirB6
	COG3736	VirB8
	IPR007430	VirB8	PGN_0062	Present*	PGTDC60_1021
	IPR026264	Type IV secretion system protein VirB8/PtlE
	COG3504	VirB9
	IPR014148	VirB9
	COG2948	VirB10
	IPR005498	VirB10
	COG0630	VirB11
	IPR014155	VirB11
	COG3505	VirD4	PGN_0076	PG1490	PGTDC60_1006
			PGN_0579		PGTDC60_1984
	IPR003688	Type IV secretion system protein TraG/VirD4		PG1490	PGTDC60_1984
T4bSS	pfam03524	Conjugal transfer protein
	IPR010258
	Tigr02756	Type F conjugative transfer system secretin TraK
	IPR014126
	pfam06586	TraK protein
	IPR010563
T4SS			±	±	±
T5SS	COG3468	Adhesin AidA
	COG5295	Autotransporter adhesin
	COG5571	Autotransporter β-barrel domain
T5cSS	pfam03895	YadA-like C-terminal region
	IPR005594
T5aSS	pfam03797	Autotransporter β domain
	IPR005546	Autotransporter β domain	PGN_0129	PG1823	PGTDC60_0070
			PGN_0178	PG2130	PGTDC60_1255
			PGN_1744	PG2168	PGTDC60_1292
T5dSS	pfam07244	Surface Ag VNR domain (PlpD POTRA motif)	PGN_0299	PG0191	PGTDC60_0462
	IPR010827
	pfam01103	Bacterial surface Ag domain (PlpD β-barrel domain)	PGN_0147	PG0980	PGTDC60_0900
	IPR000184		PGN_0973	PG2095	PGTDC60_1324
T5SS			–	–	–
T5dSS			±	±	±
T6SS	Tigr03345	Type VI secretion ATPase, ClpV1 family
	IPR017729
	Tigr03347	Type VI secretion protein, VC_A0111 family
	IPR010732
	Tigr03350	Type VI secretion system OmpA/MotB family protein
	IPR017733
	Tigr03352	Type VI secretion lipoprotein, VC_A0113 family
	IPR017734
	Tigr03353	Type VI secretion protein, VC_A0114 family
	IPR010263
	Tigr03354	Type VI secretion system FHA domain protein
	IPR017735
	Tigr03355	Type VI secretion protein, EvpB/VC_A0108 family
	IPR010269
	Tigr03358	Type VI secretion protein, VC_A0107 family
	IPR008312
	Tigr03362	Type VI secretion-associated protein, VC_A0119 family
	IPR017739
	Tigr03373	Type VI secretion-associated protein, BMA_A0400 family
	IPR017748
T6SS			–	–	–
T7SS	Tigr03919	Type VII secretion protein EccB
	IPR007795
	Tigr03920	Type VII secretion integral membrane protein EccD
	IPR006707
	Tigr03921	Type VII secretion-associated serine protease mycosin
	IPR023834
	Tigr03922	Type VII secretion AAA-ATPase EccA
	IPR023835
	Tigr03923	Type VII secretion protein EccE
	IPR021368
	Tigr03924	Type VII secretion protein EccCa
	IPR023836
	Tigr03925	Type VII secretion protein EccCb
	IPR023837
	Tigr03926	Type VII secretion protein EssB
	IPR018778
	Tigr03927	Type VII secretion protein EssA/YueC
	IPR018920
	pfam10661	WXG100 protein secretion system (Wss), EssA
	IPR034026
	Tigr03928	Type VII secretion protein EssC
	IPR023839
	IPR022206
	Tigr03931	Type VII secretion-associated protein, Rv3446c family
	IPR023840
	pfam00577	Fimbrial usher protein
	IPR000015
	pfam06013	Proteins of 100 residues with WXG
	IPR010310
T7SS			–	–	–
T8SS (ENP)	pfam03783	Curli production assembly/transport component CsgG
	IPR005534
	pfam07012	Curlin-associated repeat
	IPR009742
	pfam10614	Tafi-CsgF
	IPR018893
	pfam10627	CsgE
	IPR018900
T8SS (ENP)			–	–	–
PorSS		PorK	PGN_1676	PG0288	PGTDC60_1400
		PorL	PGN_1675	PG0289	PGTDC60_1401
		PorM	PGN_1674	PG0290	PGTDC60_1402
		PorN	PGN_1673	PG0291	PGTDC60_1403
		PorP	PGN_1677	PG0287	PGTDC60_1399
		PorQ	PGN_0645	PG0602	PGTDC60_1728
	pfam13568	PorT	PGN_0778	PG0751	PGTDC60_1868
		PorU	PGN_0022	PG0026	PGTDC60_0023
		PorV (PG27, LptO)	PGN_0023	PG0027	PGTDC60_0024
		PorW	PGN_1877	PG1947	PGTDC60_0218
	pfam14349	Sov	PGN_0832	PG0809	PGTDC60_1927
		PorX	PGN_1019	PG0928	PGTDC60_0851
		PorY	PGN_2001	PG0052	PGTDC60_0334
		Lipoprotein; TPRd, WD40d, CRDd, OmpA family domain	PGN_1296	PG1058	PGTDC60_0980
		PorZ	PGN_0906	PG1064	PGTDC60_1144
	Orthology	PorZ	PGN_0509	PG1604	PGTDC60_0697
		β-Barrel protein	PGN_0297	PG0189	PGTDC60_0460
		TonB-dependent receptor; β-barrel protein	PGN_1437	PG0534	PGTDC60_1652
		Omp17; OmpH-like	PGN_0300	PG0192	PGTDC60_0463
		SigP	PGN_0274	PG0162	PGTDC60_0438
PorSS			+	+	+

^aKey members of protein transport systems and membrane protein insertion systems in the three P. gingivalis reference strains were identified by domain searches and secondary verification of the presence of particular orthologs. The presence or absence of the main components of each protein transport system is indicated in the shaded rows by + (presence), − (absence), or ± (potential/partial presence). *, not annotated as protein.

With the exceptions of SecE and SecG, either missing among some clinical strains or poorly annotated, all the analyzed strains of P. gingivalis possess the components of the SecYEG-DFyajC system. Intriguingly, however, they lack the known Tat translocase components, showing that the Tat system is not conserved in this bacterium. This is consistent with the outcome of domain searches using motifs identifying Tat signals (Tigr01409, Tigr01412, and pfam10518), which yield no matches (Table 1), and with previous analyses reported in the literature (63). Moreover, only two members of the BAM system (BamA and BamC) appear to be present in the strains studied, as shown by domain searches and similarity analyses (Table 1). No homologues of BamB, BamD, and BamE are present in P. gingivalis. This is noteworthy, because only BamA and BamD are regarded universally essential for functionality of the Bam system (114). Similarly, the Lol system is only partially represented in P. gingivalis as merely four proteins with a LolE motif (COG4591) are detectable in the P. gingivalis reference strains. Some of these proteins display moderate levels of similarity to LolE proteins from other Gram-negative species, as judged by the presence of potential LolCE motifs (tigr002212 and tigr002213). These proteins are predicted to reside in the IM, which would be consistent with the localization of the LolCE proteins of Escherichia coli, and half of them belong to the core proteome of P. gingivalis. However, canonical members of the Lol system, especially LolA, LolB, LolC, and LolD, are absent from P. gingivalis. These observations suggest that analogous Lol and BAM systems may, respectively, be operational in the IM and OM of this bacterium (Fig. 2), while the prototype Lol and BAM systems are lacking.

Gram-negative bacteria can also possess other common systems enabling the translocation of proteins across the OM (115). These secretion systems vary from type I to type VIII (T1SS to T8SS), with the recent addition of a type IX secretion system specific to certain members of the Bacteroidetes phylum (63). The type IX secretion system is also referred to as T9SS, or porin secretion system (PorSS), and the latter designation is most frequently used in the context of protein export in P. gingivalis. Unfortunately, secretion systems are usually not well annotated by automated pipelines, mainly because certain members of different secretion systems (e.g., T2SS and T4SS) share higher sequence similarity with one another than with functionally equivalent members of the same secretion system (e.g., pilin proteins). Moreover, many secretion systems are still poorly characterized, leading to difficulties in finding the most suited domains for a domain search. Fortunately, the genes encoding members of these systems usually colocalize on the genome, thus facilitating the identification of system components.

The potential presence of known secretion systems in P. gingivalis was evaluated by domain searches, literature and genome context analyses, and similarity searches across the P. gingivalis reference strains. All three analyzed reference strains lack the vast majority of secretion systems commonly encountered in Gram-negative bacteria (Table 1). Nevertheless, proteins containing two motifs belonging to members of the type I secretion system, pfam02321 and pfam03412, were found. The pfam02321 motif was detected in multiple proteins across all strains while pfam03412 was present in only two proteins for the TDC60 strain. Interestingly, these two proteins display no significant similarity to proteins in the other P. gingivalis reference strains, as opposed to a significant similarity shared with proteins belonging to other species in the same phylum. Of note, the pfam02321 motif can also detect OM components of drug and metal efflux pumps, suggesting that the identified proteins do not necessarily belong to a functional type I secretion system. Conversely, the pfam03412 motif was used in combination with tigr01193 to identify bacteriocin exporters. The two proteins identified in strain TDC60 appear to possess both motifs, suggesting a possible role in bacteriocin secretion. However, the similarity scores for the tigr01193 motifs are significantly lower than those for the pfam03412 motifs. In conclusion, it appears that a canonical type I secretion system is absent from P. gingivalis (63).

None of the known protein components of type II and III secretion systems was found in P. gingivalis, including members of subclasses a, b, and c of the type II secretion systems and subclasses a and b of the type III secretion systems. Conversely, domain searches for three major components of the type IV secretion system, named VirB4, VirB8, and VirD4, showed multiple matches across the three reference strains. The VirB4 domain is present in two TDC60 proteins that share a relatively high level of similarity, while the VirD4 domain is present in two W83 and two TDC60 proteins. At least one gene per strain encoding these proteins colocalizes with a VirB4 motif gene on the P. gingivalis chromosome, with a distance between the respective genes of about 5 kb in the W83 strain, about 7 kb in the ATCC 33277 strain, and 10 to 11 kb for the TDC60 strain. The VirB8 domain is present in one gene per reference strain, but it should be noted that the respective gene has not been annotated for the W83 strain and that the presence of the VirB8 domain was only discovered upon closer inspection of the genome sequence.

Although no matches were identified for signature domains of key components of the type V secretion system, one protein in every reference strain was found to display a PlpD motif (pfam07244). This motif identifies components of subclass d of the type V secretion system. Moreover, these proteins appear to possess also a second PlpD motif used in T5dSS searches, pfam01103, although this second motif was identified with a sensibly lower score. On this basis, it is difficult to predict the activity of T4SS and T5dSS in the analyzed P. gingivalis strains. In the canonical T4SS, VirB4 and VirD4 are two of the three ATPases that energize the secretion machinery (116). This could imply that the VirB4- and VirD4-like proteins of P. gingivalis may be involved in another secretion system or that they serve a different function. In contrast, the possibility that a T4SS could function in P. gingivalis in the absence of other key members of this type of secretion system appears less likely. Another piece of evidence in line with the latter view relates to the fact that the VirB8 domain, used for the present similarity searches, also recognizes conjugal transfer proteins, such as TrbF and TraK. Nonetheless, VirB8 is generally responsible for forming the channel through which the T4SS cargo proteins are translocated across the IM. Hence, the detected P. gingivalis proteins containing a VirB8 domain could potentially offer an alternative pathway to the Sec system for protein passage across the IM of this bacterium (Fig. 2).

No proteins belonging to the type VI, VII, or VIII secretion systems were detectable in the analyzed P. gingivalis strains, which is in agreement with previous literature (63). On the other hand, P. gingivalis strongly relies on a novel secretion system shared by members of the Bacteroidetes phylum, the aforementioned PorSS (63), whose prominence in Bacteroidetes has recently been highlighted (1) (Fig. 2). The PorSS comprises several proteins broadly conserved throughout the Bacteroidetes group and is also involved in gliding motility in many species of this phylum, albeit not in P. gingivalis. As of now, there is general consensus that 17 proteins are essential for the PorSS function, although two additional proteins are probably required as well (86, 87). Four of these proteins, PorK to PorN, form the PorSS core membrane complex. PorM, the main component of this complex, appears to localize in the IM together with PorL. However, thanks to its long periplasmic domain, PorM is capable of interacting with the rest of the complex comprising the OM-bound lipoprotein PorK and the periplasmic OM-bound protein PorN (63, 117). Cargo proteins of the PorSS are targeted for secretion by conserved C-terminal domains (CTDs) (63), which can be identified by the TIGR04131 (IPR026341) and TIGR04183 (IPR026444) motifs. The presence of a Sec-type N-terminal signal peptide in proteins exported via the PorSS suggests that these proteins are translocated across the IM by the Sec machinery. Importantly, all known members of the PorSS are present in all of the strains evaluated here, highlighting the major role that this export system plays in P. gingivalis (Fig. 2).

It was demonstrated by different localization studies that certain CTD-containing proteins cross the OM with the help of the PorSS, subsequently appearing both in the OM and in the extracellular milieu (52, 118, 119). According to the current models, the CTD is cleaved during export of the “CTD proteins” by a sortase-like mechanism, and the resulting mature proteins are secreted or reattached to the OM via A-LPS modification, with the A-LPS acting as an anchor to the bacterial surface (86, 87). Consequently, the CTD is lacking from the mature soluble forms of these proteins (120, 121), and it has not been detected in the OM-associated mature forms, which are extensively A-LPS modified (122–124). Clearly, especially due to the two possible destinations of CTD proteins (i.e., OM insertion or secretion), it is challenging to predict their precise localization by bioinformatic approaches.

In addition to the classical secretion systems, the aforementioned release of OMVs with cargo proteins (Fig. 3) should be regarded as a specialized protein secretion pathway dedicated to virulence and the capture of nutrients (91). Indeed, the mechanism triggering the blebbing process that leads to these nanostructures, albeit poorly understood, is not random (95). In addition, the proteins secreted via this pathway seem to empanel mostly periplasmic and OM proteins, which serve as virulence factors (51, 91). Among the latter, proteases, especially the gingipains, appear to be most abundant (91). The prevalence of proteases in the OMVs might serve several purposes. First, it could be a way to deliver them to their foreign targets, especially proteins of phagocytic cells (51). Second, encasing the proteases within the membrane of the OMVs can protect them and/or the rest of the OMV cargo from the outside environment, either physically or by rendering proteolytic sites on these proteins inaccessible. Third, this feature might have evolved to protect P. gingivalis proteins, e.g., bound to the OM, from the bacterium’s own highly proteolytic potential. Lastly, OMVs and the OMV-associated PPAD could serve a decoy function in immune evasion by P. gingivalis (51). In light of these different scenarios, the observed phenomenon of extracellular compartmentalization through vesiculation might be categorized as a “protective secretion” behavior. In fact, the attachment of CTD proteins, like PPAD, to the OM and to OMVs via A-LPS modification seems to protect them from proteolysis by P. gingivalis’ own proteases, as evidenced by the recent observation that OMV disruption by ultrasound results in PPAD degradation (125). In addition, the finding that the PPAD proteins of sorting type II isolates, which are ineffectively attached to the OM and OMVs, are processed to a 37-kDa form is consistent with the idea that the OMVs serve to protect cargo against proteolysis (53).

SIGNAL PEPTIDASES

The identification of the suite of secretion systems in P. gingivalis warranted a further investigation on the signal peptidases involved. First, the Sec system utilizes two different types of signal peptidases. In general, cargo proteins of the Sec pathway are processed by signal peptidase I, which belongs to the S26 Merops family (126–128). As is the case for all living cells, the P. gingivalis genome encodes signal peptidase I, as identified through COG0681 and pfam00717 searches. Of note, a recent study from Bochtler et al. showed that over 60% of signal peptidase I substrates in P. gingivalis display a glutamine residue immediately downstream of the signal peptidase I cleavage site (in position +1), irrespective of their subcellular localizations (129). These glutamine residues are cyclized to pyroglutamate residues by the glutaminyl cyclase PG2157 (alternatively called PG_RS09565), a lipoprotein most likely located in the IM (129). This high frequency of signal peptidase I substrates with a glutamine residue in position +1 is a common feature of most Bacteroidetes species (129).

Lipoproteins have N-terminal signal peptides recognized and removed by the signal peptidase II, which takes place after the invariant cysteine residue at position +1 relative to the cleavage site has been diacyl-glyceryl modified by the diacyl-glyceryl transferase Lgt (127). Signal peptidase II belongs to the A08 Merops family and is detectable in P. gingivalis through domain searches for the pfam01252 and COG0597 motifs. Likewise, Lgt is conserved in all investigated P. gingivalis strains, as confirmed by BLAST searches. In E. coli, the N-terminal amino group of the diacyl-glyceryl-modified cysteine of the mature lipoprotein is acylated by the N-acyl transferase Lnt (127). This may not be the case in P. gingivalis, since no homologues of the E. coli Lnt were detected in the investigated strains. However, the possibility of N-acylation of the mature lipoprotein upon cleavage by Lsp cannot be fully excluded, since it was shown that N-acylation by an as-yet-unidentified enzyme takes place in Staphylococcus aureus (130). Interestingly, the N-acylation of staphylococcal lipoproteins has been invoked in the silencing of innate and adaptive immune responses (130), which is a trait that could enhance the fitness and pathogenicity of P. gingivalis as well.

AVAILABLE ALGORITHMS FOR GENOME-WIDE IDENTIFICATION OF EXPORTED BACTERIAL PROTEINS

Genome-wide prediction of the subcellular localization of proteins is a relatively recent endeavor in proteomics that has garnered increasing attention because it provides valuable insights into the biological functions of the sorted proteins, even if their precise function is still unknown (131–134). Various bioinformatic tools have been designed to identify signal peptides, such as SignalP (135), Predisi (136), and Phobius (137). These algorithms are generally used to predict signal peptides cleaved by signal peptidase I, but they do not readily recognize the lipoprotein signal peptides that are cleaved by signal peptidase II. To address this issue, lipoproteins have to be identified first by predictors capable of recognizing lipoprotein signals, such as LipoP (138) and Lipo (139). The subsequently developed PSORT I represented the first comprehensive bacterial protein localization predictor. Since then, several prediction tools for protein localization have been developed and implemented, rendering bioinformatic approaches a viable alternative to biochemical localization studies (134, 140, 141). All of these studies involved the development of a complex network of subcellular localization predictors that were tailored to a specific bacterium in order to predict, as accurately as possible, the position of each protein in the proteome. One of these studies (141) has been taken into particular consideration for this review, and its workflow was adapted to review the overall protein localization in P. gingivalis.

It should be noted that all publicly available prediction tools for subcellular protein localization have particular pros and cons. One of the difficulties in selecting the most suited programs for a bacterium of interest lies in the fact that publicly available predictors may quickly cease to be maintained, are subject to major modifications, or even become obsolete. This, coupled with the fact that certain programs may be more suited to bacteria of a certain group, makes it difficult to implement strategies previously developed for major model organisms, such as Escherichia coli or Bacillus subtilis (127, 131, 133, 142). Aside from public access, another important parameter determining our choice of programs was availability of a batch submission option, which grants fast genome-wide analyses. Moreover, to further refine the selection of prediction programs for a comprehensive overview of subcellular protein localization in P. gingivalis, tools with a high level of specialization were used as listed in Table 2. In most cases, such tools were single function predictors with few limitations, especially limitations that could have been offset by the application of other programs.

TABLE 2.

List of localization predictors^a

Name	Use	Limitation(s)
LipoP	Primarily prediction of Sec signal peptides that are cleaved by SpII but also provides prediction of inner membrane or cytoplasmic localization as well as SpI cleavage	Does not detect Tat substrates
Lipo	Prediction of Sec signal peptides cleaved by SpII	Does not detect Tat substrates
SignalP	Prediction of Sec signal peptides cleaved by SpI	Does not detect Tat substrates
Predisi	Prediction of Sec signal peptides cleaved by SpI	Does not detect Tat substrates
Phobius	Prediction of alpha helices in inner membrane proteins, distinguishing N-terminal TM from signal peptides
TmHmm	Prediction of alpha helices in inner membrane proteins	Signal peptides often considered TM spans
Bomp	Prediction of beta-barrel spans in outer membrane proteins
SecretomeP	Prediction of ECP	Limited number of sequences per batch
Interpro	Functional analysis of proteins by classification into families, domain and site prediction by combination of protein signatures from a number of member databases

^aAn overview of localization predictors, membrane insertion detectors, and other programs used in this study and their relative strengths and weaknesses is presented.

Interestingly, different predictors occasionally assigned the same proteins to different subcellular compartments, even in case of programs with the same specific functions. Disagreements in localization between different programs underscore the notion that some predictors may be more accurate or, at the very least, better suited than others to chart the proteins of a specific bacterium. Moreover, these discrepancies reveal the levels of uncertainty of bioinformatics predictions and the need for an organized method encompassing all the chosen tools that can exploit all the strengths and balance the limitations of each program. On the other hand, it must be acknowledged that protein sorting mechanisms in a living bacterium do not usually operate with a fidelity of 100%, which means that proteins that are generally secreted are detectable within different cellular compartments, while proteins that are meant to be retained in the cell (e.g., cytoplasmic proteins, lipoproteins, or cell wall-bound proteins) can be encountered in the extracellular environment. The protein sorting ambiguities encountered in silico are thus perhaps an unintended reflection of the imperfections of sorting systems employed by a bacterial cell in vivo. Clearly, as long as these imperfections have no bearing on the competitive success of a bacterium, they do not matter.

To meet the need for biologically relevant predictions of protein sorting, a decision tree (Fig. 4) was devised which organizes the predictors and sorts proteins through them with the purpose of assigning them to their rightful subcellular compartment. The first challenge in a prediction analysis is to localize the components of the export, secretion, and membrane insertion systems themselves, which relates to the difficulty in recognizing their signal peptides by predictors. The level of difficulty depends on the system examined, with more common and conserved systems being more easily localized. For example, some components of the recently discovered Por secretion system have an uncertain localization. Second, the identification of lipoproteins has priority, especially in view of the inability of different predictors to distinguish Sec signal peptides cleaved by the lipoprotein-specific signal peptidase II from Sec signal peptides cleaved by signal peptidase I. Notably, localization tools generally distinguish between IM and OM lipoproteins, utilizing data from extensive research on the widely favored model Gram-negative bacterium E. coli. These studies have shown that lipoproteins possessing an aspartic acid in the +2 position of the mature protein become IM lipoproteins (i.e., the “D⁺² rule”), while all other lipoproteins are presented to the OM by the Lol system (143). Intriguingly, several exceptions to this rule have been observed in other species (129, 142, 144–147), presenting the possibility that it is only obeyed in Enterobacteriaceae. Analyzing known OM lipoproteins of P. gingivalis by applying the D⁺² rule, in fact, resulted in a faulty prediction for the subcellular location of the vast majority of lipoproteins. Conversely, inspection of lipoproteins of known subcellular localization showed a preferential glycine residue at the +2 or +3 positions of the mature form for IM lipoproteins. The present evaluation of lipoprotein localization in P. gingivalis therefore relied on inspection of the +2 and +3 residues combined with specific domain searches. Following the designation of the “lipoproteome,” investigation of proteins with transmembrane helices and Sec signal peptides was performed, in this order (Fig. 4). This relates to the fact that predictors of membrane spanning regions occasionally mistake relatively longer signal peptides for transmembrane spans (Table 2).

FIG 4.

Bioinformatics pipeline to unravel protein sorting events in P. gingivalis. The flowchart depicts the different steps used to assess the subcellular localization of proteins in the analyzed P. gingivalis strains.

Excretion of cytoplasmic proteins (ECP), also termed nonclassical or leaderless secretion, is a highly discussed topic and a way to explain the presence in the extracellular milieu of proteins that lack a known signal peptide and a dedicated transport system of the categories described above (148, 149). These features apply to the bulk of cytoplasmic proteins. Accordingly, cell lysis was for a long time the most accredited hypothesis to explain the presence of cytoplasmic proteins in the extracellular milieu (150). This view is supported by the observation that ECP can be associated with autolysin and phage activity, or the production of cytotoxic peptides (149, 151). Nonetheless, the existence of dedicated “nonclassical” secretion systems for proteins deprived of known signal peptides cannot be excluded, as underpinned by the relatively recent discovery of the Tat and type VII secretion systems (152). Such hidden treasures are likely to be buried in the exoproteome haystack until uncovered by the application of molecular biological or mass spectrometric approaches to assess bacterial protein secretion. In fact, with increasing sensitivity of mass spectrometric measurements, more and more signal peptide-less proteins have been identified in bacterial exoproteomes. This is exemplified by a recent investigation on the exoproteome of P. gingivalis, where many signal peptide-less proteins were identified in the growth medium fraction (54). In fact, the latter analysis highlights two remarkable features. First, signal peptide-less extracellular proteins were overrepresented among the low-abundance extracellular proteins and, second, the detection of these proteins was most variable between the investigated strains. This is suggestive of an unspecific export mechanism, such as cell lysis. However, also among the most abundantly detectable and invariant exoproteins of P. gingivalis there are proteins lacking signal peptides, which is suggestive of specific export, stable extracellular maintenance in the presence of gingipains, and a possible function in the bacterial life cycle. As to possible functions, it has been shown that proteins with important roles in the cytoplasm, like elongation factors and proteins involved in central carbon metabolism, can serve important extracytoplasmic “moonlighting” functions in bacterial adhesion to mammalian cells and tissues (149, 153, 154). Altogether, it seems that ECP in Gram-negative bacteria may be more complex than initially thought, with several distinct pathways present (148). Importantly, proteins subject to ECP can be predicted by homology using SecretomeP 2.0 (155).

DEDICATED PIPELINE TO APPROXIMATE SUBCELLULAR PROTEIN LOCALIZATION IN P. GINGIVALIS

To approximate subcellular protein localization in P. gingivalis with the ultimate objective of better understanding which proteins are targeted to the bacterial cell envelope or the host milieu, an in-house script implementing the decision tree presented in Fig. 4 was developed. In addition to the aforementioned algorithms, TMHMM (156) was included to predict transmembrane helices, BOMP (157) was included to predict β-barrel OM proteins, and InterPro Scan (158) version 5.27 was included to detect particular domains in the InterPro consortium database (159) version 66. Based on P. gingivalis proteins of known location (see Table S1 in the supplemental material [https://bitbucket.org/teto1991/gingimaps/src/master/]), three lists of domains specific for PorSS cargo, IM proteins, and OM proteins were established (Table S2). Such mainly structural domains were chosen to be as specific as possible, in order to avoid biases. Using the software listed in Table 2, localization data were generated for all the seven revisited P. gingivalis strains. Further, following the flow scheme presented in Fig. 4, a knowledge-based approach was implemented that is grounded on the currently available understanding of protein sorting systems active in P. gingivalis, as detailed in the aforementioned sections. Importantly, the hierarchy of decisions in this pipeline was tailored to minimize mistakes and biases, and to maximize compensation for possible software weaknesses.

Proteins displaying at least one of the selected PorSS cargo-specific domains (Table S2) were immediately designated as secreted via the PorSS (Fig. 4), since these signatures are highly reliable in predicting secretion via this pathway. On this basis, the inspected P. gingivalis strains potentially secrete between 19 and 24 proteins specifically via the PorSS (Table 3). Despite its high specificity, this approach does not guarantee the identification of all PorSS cargo proteins, because some proteins exported via the PorSS may lack the selected PorSS cargo domains. Further, the present listing of predicted PorSS cargo proteins (Table S3) may represent an underestimation since potential misidentifications were not manually curated in order to avoid bias. This explains why the present number of potential PorSS cargo proteins is lower than the previously proposed 30 to 35 cargo proteins (87, 160), which may include some proteins whose secretion is indirectly related to the PorSS.

TABLE 3.

Summary of predicted protein localizations^a

Protein	No. of protein localizations for various strains
	ATCC 33277	W83	TDC60	MDS33	MDS140	512915	20655
CYT	1,286	1,193	1,451	1,375	1,362	1,426	1,366
ECP	95	80	109	94	97	103	115
IM	3	3	3	3	3	3	3
IM LP	18	19	18	20	22	20	19
IM TM	333	316	348	332	318	342	363
OM	57	58	60	61	56	62	64
OM LP	67	46	54	54	55	61	56
PERI	135	118	118	138	125	131	136
PorSS	22	22	24	20	21	19	20
UNK	6	8	9	9	6	5	5
Total	2,022	1,863	2,194	2,106	2,065	2,172	2,147
TOT EXTRA	117	102	133	114	118	122	135
TOT IM	354	338	369	355	343	365	385
TOT OM	124	104	114	115	111	123	120

^aAn overview of the protein localization predictions for each strain enumerating all proteins present in the different subcellular compartments is presented. CYT, cytosolic protein; ECP, ECP protein; IM, inner membrane protein; IM LP, inner membrane lipoprotein; OM, outer membrane protein; OM LP, outer membrane lipoprotein; PERI, periplasmic protein; PorSS, PorSS secreted protein; UNK, protein of unknown localization; TOT EXTRA, total of extracellular proteins (ECP + PorSS); TOT IM, total of inner membrane proteins (IM + IM LP + IM TM); TOT OM, total of outer membrane proteins (OM + OM LP).

In the second step of the prediction pipeline, both the LipoP and Lipo algorithms were used to identify lipoproteins among those proteins that were not assigned as PorSS cargo (Fig. 4). Since there was no possibility for a majority vote, the LipoP predictions were given priority in case of disagreement. The same approach was used to assess the signal peptidase II cleavage sites, which was necessary to pinpoint amino acid residues at positions +2 and +3 of the mature lipoproteins. If glycine residues were absent from these positions, the respective protein was predicted to be an OM lipoprotein (OM LP). If a glycine residue was present at the +2 or +3 position, an additional control was performed by assessing the presence of a known OM domain. If an OM domain was detected (Table S2), the “G^+2/+3 rule” was ignored and the protein was still predicted as an OM lipoprotein (OM LP). Conversely, the apparent lack of OM domains resulted in a protein’s designation as an IM lipoprotein (IM LP). Per investigated strain, the numbers of predicted IM lipoproteins ranged from 18 to 22, and the numbers of OM lipoproteins from 46 to 67 (Table 3). The relatively large variation in the numbers of OM lipoproteins predicted for different strains may relate to previously observed genomic rearrangements in P. gingivalis (161).

For nonlipoproteins, an agreed number of two or more transmembrane helices as identified by TMHMM and Phobius was used to predict IM transmembrane proteins (Fig. 4). When instead both programs agreed on the presence of at least one transmembrane helix, a signal peptide check was performed in order to reduce the number of false positively predicted helices caused by the presence of a signal peptide. When the signal peptide prediction consensus was one or lower (i.e., one or none of the three applied programs predicted a signal peptide) the predicted helix was considered a “true positive,” and the respective protein was therefore predicted to have a transmembrane localization in the IM. In case of a signal peptide consensus equal to or higher than two (i.e., at least two of the three applied programs predicted a signal peptide) and an agreed number of transmembrane helices as predicted by TMHMM and Phobius was equal to or lower than one, the protein was considered to have a signal peptide. In this case, it was further analyzed to determine a possible IM, OM, or periplasmic localization. Conversely, when the signal peptide consensus was one or zero, and the agreed number of transmembrane helices as predicted by TMHMM and Phobius was one or zero, the protein was considered to lack a signal peptide. Thus, despite being allegedly unable to cross the IM via Sec, such a protein was further analyzed for a possible cytoplasmic localization, or a potential IM, OM, or extracellular localization via ECP. In all other cases, i.e., when the outcome of the predictions of transmembrane helices and the presence of a signal peptide were in conflict, the predicted localization of the respective protein was designated as unknown (Table 3). Merely five to nine proteins with unknown localization were encountered for the presently evaluated strains, suggesting that the adopted approach was extremely discriminative and robust.

Proteins with a signal peptide are able to cross the IM, ending up in the periplasm. The additional presence of either a β-barrel, indicated by a BOMP score higher than three, or of at least one OM domain (Table S2) can be considered an indicator for subsequent association with or insertion into the OM. The latter proteins were thus predicted to localize in the OM compartment. In case a transmembrane domain, β-barrel, or OM domain were absent, while a signal peptide was present, the respective protein was designated as having a periplasmic localization. Since the presence of a Phobius-predicted transmembrane domain is indicative of protein retention in the IM, such proteins were designated as IM resident-proteins.

In the absence of a canonical signal peptide, a protein can be retained in the cytosol, be secreted through noncanonical or unknown pathways thereby ending up in the OM or extracellular milieu, or be inserted into or associated with the IM. For such reasons, all proteins lacking a predicted signal peptide were checked for the presence of OM domains (Table S2). If one or more of such OM domains were present, the respective protein was designated to have an OM localization. Analogously, the presence of an IM-related domain (Table S2) was used as an indicator for IM localization.

A SecretomeP analysis was performed as the last verification step in the prediction pipeline (Fig. 4) because of its knowledge-based nature. At this juncture, the remaining proteins exhibit no relevant feature as discussed above and, accordingly, their predicted sorting destination could only be the cytoplasm or the extracellular milieu due to noncanonical or unknown ECP pathways. Therefore, in the presence of a SecretomeP score equal or higher than 0.75, proteins were predicted to undergo ECP. Instead, a lower score pointed at a cytosolic localization, since none of the applied predictors suggested the possibility of the respective protein leaving the cytosol. The overall outcome of the predicted protein localization in P. gingivalis is listed in Table S3 in the supplemental material (https://bitbucket.org/teto1991/gingimaps/src/master/), while Table 3 presents an overview of these predictions.

CORE AND VARIANT EXOPROTEOME ANALYSES

Interestingly, analysis of the P. gingivalis exoproteome highlighted strain-specific variations (54), which were also encountered in the present inspection of subcellular protein localization. This was the incentive for a bioinformatics-based appraisal of the core and variant (exo)proteome of P. gingivalis. Thus, to identify orthologues in the proteome of different strains, reciprocal best hits (RBHs) were calculated. In brief, Galaxy (162) was used to perform reciprocal protein BLAST searches (NCBI BLAST+, v2.3.0 [163]). Default parameters (minimum percentage identity, 70%; minimum high scoring pair coverage, 50%) were used, and all redundancies were removed prior to the BLAST search. RBHs were then calculated by blasting the deduced amino acid sequences of all investigated strains against those of P. gingivalis ATCC 33277. Despite P. gingivalis W83 being the most used reference strain in the field, the ATCC 33277 strain was adopted as a reference for the present analyses after the realization that many proteins were actually encoded by the W83 genome sequence while the respective genes were never annotated (data not shown). Tblastn was used to identify some of these proteins, being part of the main secretion complexes, and they are reported in Table 1. The core proteome was thus defined by the set of proteins having an ortholog in all six strains analyzed against the ATCC 33277 reference strain (Table S4 [https://bitbucket.org/teto1991/gingimaps/src/master/]). The remaining protein complement identified for each strain is regarded as the respective variable proteome (Table S5). Of note, considering possible misannotations of the used genome sequences, the presently proposed distinction between the P. gingivalis core and variable proteomes should be regarded as an approximation rather than an absolute distinction.

To predict the core exoproteome, the proteins in the core proteome were divided according to their possible subcellular localizations, as per our prediction pipeline, and two categories were pulled together: (i) proteins of the OM compartment (OM_LP and OM proteins) and (ii) PorSS cargo proteins (Table S6 [https://bitbucket.org/teto1991/gingimaps/src/master/]). The GO terms associated with the domains detected by InterPro for these exoproteins were taken into account for each strain. The obtained GO terms were then used in a REVIGO (164) analysis, to unravel the network of biological pathways created by the core exoproteome. It should be noted that the potential ECP complement as designated by our pipeline was excluded from the exoproteome classification due to its high variability between strains (Table S3). The remaining predicted exoproteins were, instead, almost entirely predicted to make up the core exoproteome. The limited suitability of SecretomeP for our P. gingivalis data set is probably due to the fact that ECP predictors cannot be tailored on specific transport systems or bacterial species due to their intrinsic nature. GO term analysis of the core exoproteome predicted for each inspected strain yielded close to identical results (Fig. 5) with some marginal differences observed for strain MDS33 (data not shown). The latter may relate to minor discrepancies in the genome annotation of strain MDS33 or to some small potential differences in the core orthologs of this strain. The identified core exoproteins operated in eight different major biological pathways, namely, putrescine biosynthesis, intracellular protein transport, membrane assembly, protein folding, metabolism, carbohydrate metabolism, proteolysis, and oxidation-reduction processes (Fig. 5).

FIG 5.

Biological pathways represented in the P. gingivalis core exoproteome. A REVIGO tree map depicts the outcome of a GO term analysis of cellular pathways involving the proteins predicted to define the core exoproteome of the P. gingivalis strains under examination.

Since these results slightly differed from previous observations on the core exoproteome of a different and smaller set of samples (54), mainly for the lack of a pathogenesis GO term cluster, we also analyzed the variable exoproteome (Table S7 [https://bitbucket.org/teto1991/gingimaps/src/master/]). The simple absence of one virulence factor from one strain, in fact, would eliminate the protein from the core exoproteome and relegate it to the variable exoproteome. As expected, the GO term analyses of the variable exoproteomes revealed a sizable amount of extracellular proteins involved in pathogenesis in all P. gingivalis strains, except MDS140. The latter strain happens to be isolated from a healthy carrier. It is therefore tempting to speculate that the MDS140 strain could lack a number of virulence factors. In addition, only the “transport” and “proteolysis” labels were assigned to predicted exoproteins of the MDS140 isolate (Fig. 6A), in contrast to the various other functional labels assigned to exoproteins from the other investigated strains (Fig. 6B to D).

FIG 6.

Biological pathways represented by the P. gingivalis variable exoproteome. REVIGO tree maps represent the outcomes of GO term analyses of the cellular pathways in which variable exoproteomes of different P. gingivalis strains are involved. (A) MDS140; (B) W83; (C) TDC60, MDS33, and 512915; (D) ATCC 33277 and 20655.

Lastly, all the protein sorting information gathered by reviewing the available literature and predicting subcellular protein localization in P. gingivalis has been combined in Fig. 7, which presents the total numbers of proteins predicted per subcellular compartment. Of note, this overview image distinguishes the core and variant proteomes of each compartment.

FIG 7.

Overview of subcellular localization of core and variant P. gingivalis proteins. The numbers of proteins residing at a particular subcellular location, or extracellularly, are indicated for the core proteome and the variable proteome of each examined P. gingivalis strain. Specifically, these include strains ATCC 33277, W83, TDC60, MDS33, MDS140, 512915, and 20655. The numbers of cytoplasmic proteins are indicated in blue, IM lipoproteins in red, IM proteins in black, periplasmic proteins in green, OM lipoproteins in orange, OM proteins in yellow, PorSS secreted proteins in gray, and ECP-secreted proteins in cyan.

CONCLUSION

This review is focused on protein localization in the oral pathogen P. gingivalis. It integrates the results of published biochemical studies (52, 61–63) and a tailored in silico evaluation of published genome sequences that is grounded on established bioinformatic approaches (133, 134, 141). Considering the broad spectrum of interests that P. gingivalis elicits, especially in the fields of periodontology, rheumatology, and microbiology, this review will serve as an important lead for many upcoming studies concerning this bacterium. In fact, a compendium of the different subcellular and extracellular destination(s) that each individual protein may reach constitutes a treasure trove of invaluable information for any kind of research involving the biology and virulence of this bacterium. This view is underscored by the importance of the exoproteome in bacterial virulence, adhesion, and biofilm development, as well as diagnostic and therapeutic applications. The presently highlighted pathways for subcellular protein localization and secretion combined with the predicted protein addresses in P. gingivalis—in short, the “Gingimaps”—could therefore be used to devise diagnostic or therapeutic antibodies targeting specific surface proteins, to create vaccines, and to discover druggable targets. This view is supported by the finding that, in a mouse model, oral infection by P. gingivalis and bacterial dissemination to arthritic joints can be inhibited with an anti-FimA antibody (165). In addition, since several proteins of P. gingivalis are the subject of ongoing studies, the availability of data regarding the proteins belonging to the same subcellular compartments is a significant advantage when looking for targets, inhibitors, or possible cofactors. A simple example of this is utilizing exoproteomic data to narrow down the list of possible targets of the citrullinating enzyme PPAD. The same can be applied to gingipains, whose high proteolytic potential is under investigation in multiple fields, both clinical and biochemical.

Lastly, all the known and yet unknown mechanisms responsible for the status of P. gingivalis as a successful oral pathogen implicated in a variety of diseases rely directly or indirectly on proteins. A direct impact of P. gingivalis proteins in disease is highlighted by the biological functions of PPAD, gingipains, hemagglutinins, and fimbriae, but there are likely to be many more. The indirect relationships of P. gingivalis proteins with human diseases are underpinned by the machinery needed to synthesize lipopolysaccharides and capsular components. Consequently, the present “Gingimaps” may hold the key to a better understanding of causal or indirect relationships between this bacterium and the disorders to which it is linked.

Biographies

Giorgio Gabarrini obtained his BSc degree in Biology at the University of Milano-Bicocca, Italy, in 2010. He subsequently studied Molecular Biology and Genetics at the University of Pavia, Italy, where he obtained his MSc degree in 2012. Afterward, he moved to Groningen, the Netherlands, to pursue the interest in oral microbiology he developed during his studies. At the University of Groningen, he earned his PhD in Medical Sciences with a dissertation on the role of Porphyromonas gingivalis and its peptidylarginine deiminase in rheumatoid arthritis. He continued to work in the same university for his first postdoctoral research until 2019, when he moved to the Karolinska Institute, Sweden, where he is currently a postdoctoral researcher investigating the role of the oral microbiome in pancreatic cancer.

Stefano Grasso is a PhD student and Marie Sklodowska Curie fellow at the University Medical Center Groningen and DSM Food Specialties B.V. since 2015. He obtained his BSc degree in Biotechnology and his MSc degree in Plant and Animal Biotechnology at the University of Udine, in 2013 and 2015, respectively. There, he also attended the Scuola Superiore, a parallel institution to the traditional University, from which he graduated in 2016. Prior to his PhD studies, he has been working on Plant Molecular Genetics. Since then, he has been interested in the secretion mechanisms of bacteria, with a special focus on Bacillus subtilis, and their possible exploitation for industrial purposes. Believing in engineering-like approaches to biology, he has also been working on the functional annotation and prediction of wild-type and designed proteins.

Arie Jan van Winkelhoff has been a Professor of Oral and Medical Microbiology at the University of Groningen and the University Medical Center Groningen, the Netherlands, since 2008. He earned his PhD in Medical Sciences in 1986 at the Vrije Universiteit of Amsterdam on the role of black-pigmented Bacteroides in human oral infections. In 1987, he became Assistant Professor in the Department of Oral Microbiology of the Academic Centre for Dentistry Amsterdam (ACTA). From 1986 to 1989 he was appointed Visiting Professor at the Department of Oral Biology, Dental Faculty, University of New York at Buffalo, New York. In 1989, he became Associate Professor at the Department of Oral Microbiology (ACTA), and in 2003 he was appointed as Full Professor in Oral Microbiology (ACTA). His research has been focused on virulence factors of oral bacterial pathogens, the application of antibiotics in the treatment of periodontitis, and the link between oral diseases and systemic disorders, such rheumatoid arthritis.

Jan Maarten van Dijl is Professor of Molecular Bacteriology at the University of Groningen and the University Medical Center Groningen, the Netherlands. He earned his PhD degree in Mathematics and Natural Sciences in 1990 under supervision of Gerard Venema at the University of Groningen. As a postdoctoral researcher, he joined the group of Gottfried Schatz at the Biozentrum of the University of Basel, Switzerland, in the period between 1994 and 1996. He became Assistant Professor in Pharmaceutical Biology in 1998 and has been a Full Professor of Molecular Bacteriology since 2004. His research addresses fundamental and applied aspects of bacterial gene expression, protein secretion, virulence, and antimicrobial drug resistance. He has been a coordinator of various national and European research programs on protein production systems and the development of novel antimicrobial therapies.