P. gingivalis is a gram-negative bacterium that is widely implicated in the oral disease periodontitis [1]; however, the influence of P. gingivalis and its virulence factors extends well beyond damage to oral tissue [2]. P. gingivalis has been designated a keystone pathogen due to its ability to impair innate host immune defenses and reshape normal symbiotic microbiota into a dysbiotic state that contributes to a sustained inflammatory response in the absence of bacterial clearance [3, 4]. Various translocation mechanisms including degradation of the epithelial barrier, invasion of epithelial and endothelial cells, and dissemination via the blood permit systemic circulation of P. gingivalis periodontal pathogens and virulence factors with corresponding systemic implications [5–7]. To that end, mounting evidence suggests that P. gingivalis is a strong risk factor for Alzheimer’s dementia and other conditions, including cardiovascular disease, rheumatoid arthritis, and various neurodegenerative disorders [6–14].

P. gingivalis produces several virulence factors, including gingipains and fimbriae, proteins that are involved in adhesion, tissue degradation, and disruption of immune system responses [15, 16]. Gingipains, the primary P. gingivalis virulence factors, are cysteine proteases that are essential for P. gingivalis survival and pathogenicity, playing critical roles in host colonization, inactivation of host defenses, iron and nutrient acquisition, and tissue destruction [17]. They comprise arginine gingipain-A (RgpA) and -B (RgpB) and lysine-specific gingipain (Kgp). Fimbriae are hair-like appendages that promote bacterial motility, adhesion, and invasion [16, 18]. They comprise several types of major (or long) fimbriae (FimA) and minor (or short) fimbriae (Mfa). These and other virulence factors (e.g., glycolipid lipopolysaccharide) subvert innate immune system responses aimed at elimination of P. gingivalis [19, 20], rendering P. gingivalis mostly resistant to the innate immune system responses targeting them and prompting activation of adaptive immune system responses as a secondary line of defense [17].

The adaptive immune system utilizes lymphocytes (T and B cells) to attack specific pathogens and establish immunological memory to hasten immune response in the event of future exposure to that pathogen. Human leukocyte antigens (HLAs) are key to adaptive immunity. HLAs are cell-surface glycoproteins that play a critical role in presentation of peptides from bacterial proteins and other foreign antigens to T cells to promote their elimination. HLA-I (encoded by HLA-A, -B, -C genes) are expressed in all nucleated cells and bind short peptides (mostly 9-mer) of endogenous proteins for presentation to cytotoxic CD8+ T cells to signal destruction of infected cells. HLA-II (encoded by the HLA-DPB1, DQB1 and DRB1 genes) is expressed in professional antigen-presenting cells (APCs; e.g., macrophages, dendritic cells) and binds longer peptides (mostly 15-mer) of exogenous proteins for presentation to CD4+ helper T cells to stimulate B cell-mediated antibody production and enhance the activation of CD8+ T cells. Unlike the non-specific innate immune system responses, HLA-mediated adaptive immunity is highly specific to each foreign antigen encountered and rests on high-affinity binding of HLA molecules with foreign antigen peptides. HLA is extremely polymorphic [21] and most of the variability is localized in the binding groove [22–24], the region that binds peptides for transport to the cell surface for T cell presentation. Consequently, HLA molecules differ with regard to the repertoire of antigens that are suitable for high-affinity binding; that is, an interaction between a given HLA molecule and a given antigen peptide determines the adaptive immune system response such that strong, high-affinity HLA-peptide binding enhances immune responses targeting their elimination relative to the weak, low-affinity HLA-peptide binding, which is ineffective. The mechanisms of HLA-mediated antigen presentation to T lymphocytes are discussed in detail elsewhere [25, 26].

The strongly bound peptide-HLA (pHLA) complex moves to the surface and engages CD8+ (pHLA-I complex) and CD4+ (pHLA-II complex) T cells by attaching to their αβ T-cell receptor (TCR) [27–30]. Remarkably, a single pHLA-I complex is sufficient to activate a single CD8+ T-cell [31] with a bond lasting ~1 min [31, 32], leading to serial engagement of ~200 TCRs [33]. The pHLA-I/TCR occupancy correlates well with the release of interferon (IFN)-γ by the activated T cell [33], documenting the dependence of the latter on the former. The number of pHLA-I complexes on the cell surface can vary from a few to a few thousand, depending on the cell and the peptides involved [34]. Given that (a) a single pHLA-I complex can successfully engage ~200 TCRs, and that (b) these engagements are functionally active, it follows that the overall T-cell activation initiated by a single APC can be very high and successful in eliminating infected cells by various killing mechanisms [35].

With respect to CD4+ T cell engagement, high-affinity-bound pHLA-II complexes traffic to the APC plasma membrane where they remain very stable. Such pHLA-II complexes can last very long (half-life of t1/2 > 150 h) [36]. Importantly, this pHLA-II half-life determines the in vivo immunogenicity of the bound epitope [36]. In summary, the degree of binding affinity of a peptide to an HLA-II molecule ultimately determines the magnitude of the biological effect of this pHLA-II complex, most probably due to the long time that an antigen is presented for CD4+ T cell engagement. pHLA-II complexes activate CD4+ T cells by binding to their TCR. This effect is extremely sensitive, since a single pHLA-II complex is sufficient to activate a CD4+ T cell [37]. This high sensitivity of CD4+ engagement and the associated long life of pHLA-II presentation [36] result in very strong and long-acting CD4+ T-cell activation. Although a unique role of the HLA-II system is to initiate, by the activation of CD4+ T cells, the production of antibodies by B cells against foreign antigens [38–40], an additional important role of CD4+ T cells is to augment the function of CD8+ T cells, with respect to both their cytotoxic action and the establishment of memory T-cells [41]. Finally, in addition to the binding of pHLA complex, successful activation and clonal expansion of CD8+ and CD4+ T cells in the presence of a pathogen requires the presence on APC surface of co-stimulatory signals B7.1 and B7.2, interfacing the combined action of innate and adaptive immune systems in pathogen elimination [42–47].

In summary, high-affinity pHLA binding is a necessary but not sufficient condition for a successful adaptive immunity, for which immunocompetency and well-functioning native immunity mechanisms are assumed. Conditions of immunosuppression (e.g., by the administration of immunosuppressant drugs, HIV infection, etc.) would compromise the mounting of an effective adaptive immunity. This is also true for P. gingivalis, which has been shown to suppress adaptive immunity for immunoevasion [48].

Both HLA Class I and Class II have been implicated in periodontitis [49–54]. Although several bacteria are associated with periodontitis, P. gingivalis is well-established as a major, primary contributor [15, 55, 56]. Compared to the wealth of literature evaluating HLA variation with regard to periodontal disease, relatively few studies have evaluated HLA with respect to P. gingivalis specifically. One recent study identified higher titres of anti-P. gingivalis antibodies related to specific HLA-DRB1 alleles in patients with rheumatoid arthritis [57]. P. gingivalis has adopted several immune evasion strategies [58, 59], including suppression of adaptive immunity [48]. Notwithstanding potential adaptive immune evasion, strong HLA-peptide binding of P. gingivalis proteins would be expected to induce an immune response aimed at pathogen elimination and reduce the development of periodontitis and systemic disease sequelae. Here we evaluated in silico the predicted binding affinity of 334 HLA alleles with proteins involved in P. gingivalis and identified specific epitopes characterized by high-affinity binding.

Thirteen proteins of P. gingivalis involved in its pathogenicity and structure were tested and are shown in Table 1. Their amino acid (AA) sequences were retrieved from the UniProt database (https://www.uniprot.org/) on December 25, 2025, and are given in Table S1.

We investigated a total of 334 HLA common alleles [60], including 142 HLA-I alleles (Table S2; N = 41, 72, 29 of genes A, B, C, respectively) and 192 HLA-II alleles (Table S3; N = 41, 35, 116 of genes DPB1, DQB1, DRB1, respectively). Allele population frequencies are from [60].

Predicted binding affinities were obtained for antigen peptides using the Immune Epitope Database (IEDB) NetMHCpan (ver. 4.1 BA) tool [61, 62]; accessed on January 7, 2026. More specifically, we used the sliding window approach [63–65] to test all possible linear 9-mer and 15-mer peptides for HLA-I and HLA-II predictions, respectively; the method is illustrated in Figure 1 for 9-mer and 15-mer peptides of the Gingipain R1 protein. Peptides of 9- and 15-mer length were tested because these are the dominant peptide lengths for binding to HLA-I [66, 67] and HLA-II molecules [67–69], respectively. DPB1 and DQB1 alleles were paired with highly frequent alpha chains, namely DPA*01:03 and DQA*01:01, respectively. For each pair of pHLA molecules tested, this tool gave, as an output, the IC₅₀ of the predicted binding affinity; the smaller the IC₅₀, the stronger the binding affinity. An IC₅₀ value of < 50 nM was regarded as strong [70–73] and was called a “hit”. Given a protein of N AA length and a peptide length of k AA, there are N-k + 1 binding affinity predictions returned by the prediction tool. The numbers of 9-mer (HLA-I) and 15-mer (HLA-II) peptides tested for each protein are given in Table 1.

Standard statistical methods (e.g., Pearson and Spearman correlations) were used to analyze the data using the IBM-SPSS statistical package (version 31). All P-values reported are 2-sided, and statistical significance was set at P < 0.05.

The frequency distribution of hits per protein is shown in Figure 3. It can be seen that number of hits per protein varied substantially, reflecting the higher chance for hits in proteins with longer AA sequences (Table 1). Indeed, there was a high correlation between the number of hits and the number of AAs (Figure 4 Pearson r = 0.931, P < 0.001, N = 13 proteins).

The 13 columns of hits per allele across the 249 alleles in Table 2 constitute the “immunogenetic profiles” of the 13 P. gingivalis proteins. We computed and evaluated all possible pairwise Pearson correlations (N = 13 × 122 = 78) between the immunogenetic profiles of the 13 proteins. Remarkably, all profiles were positively and highly significantly intercorrelated, as shown in Table 4 and illustrated in Figure 5. Very similar results were obtained for nonparametric Spearman correlations.

The number of hits per HLA Class and gene is given in Table 5. It can be seen that alleles of all HLA-I genes had high percentages of hits (A > C > B), for an overall hit percentage of 81.7%. By contrast, HLA-II alleles had an overall hit percentage of 69.3%, with DRB1 gene leading (92.2%), DPB1 gene following (63.4%), and DQB1 gene totally absent. The frequency distribution of hits per HLA allele is shown in Figure 6. It can be seen that the distribution tends to exponential, with a few alleles having a lot of hits; for example, DRB1*01:18, DRB1*01:01, DRB1*01:20 and DRB1*10:01 have more than 600 hits. Alleles with zero hits were also observed in other genes, with lowest occurrence in gene A of HLA-I (4.9%) and gene DRB1 of HLA-II (7.8%). The promiscuity of HLA-II alleles (candidates for vaccine development) are given in Table S4.

The 249 rows of hits per protein across the 13 proteins in Table 2 constitute the “protein profiles” of a given allele. We computed and evaluated all possible pairwise Pearson correlations (N = 249 × 2482 = 30,876) among these protein profiles. We found the following: of the total 30,876 correlations, (a) 69 (0.223%) were practically zero; (b) 9,554 (30.943%) were negative (range: –0.0001 to –0.776) comprising 121 statistically significant correlations at the nominal (uncorrected) level of P < 0.05 (P-value range: 0.0018 to 0.0495) but nonsignificant when Bonferroni-corrected for 30,876 multiple comparisons (P < 0.05); and (c) 21,253 (68.833%) correlations were positive (range: 0.999 to 0.008), comprising 9,024 statistically significant at the nominal (uncorrected) P < 0.05 level (P-value range: P < 0.00001 to P = 0.0499), including 244 correlations with significant Bonferroni-corrected P-values (P < 0.05).

We further investigated the association between 15-mer peptides and HLA-II alleles to identify peptides suitable for development of vaccines against P. gingivalis. Indeed, we identified 798 unique 15-mer peptides (Table S5) that bound with predicted high-affinity binding to 133 HLA-II molecules in 6,343 pHLA-II combinations (Table S6) where it can be seen that there were frequently multiple hits for the same protein-allele combinations. It should be noted that 770/798 (96.5%) of these peptides were associated with alleles of the DRB1 gene (Table S5) which, unlike the DPB1 and DQB1 genes, possesses the same alpha chain. The frequency distribution of unique 15-mer, strongly binding peptides is illustrated in Figure 7A, and their association with specific HLA-II alleles is illustrated in Figure 7B.

We tested for possible overlap of the 798 unique 15-mer peptides with other human proteins by comparing them against the human proteome dataset version 24.1, provided by The Human Protein Atlas (The Human Protein Atlas. Accessed on October 22, 2025. https://www.proteinatlas.org/about/download#protein_atlas_data), comprising a total of 83,607 human proteins. We did not find any matches. This suggests that potential vaccines based on those peptides are not likely to cause autoimmune adverse effects.

The hits for DPB1 and DQB1 genes are possibly underestimated, given that only one alpha chain was used for pairing, namely DPA*01:03 and DQA*01:01, respectively. Although both of these specific alpha chains are very frequent, DP and DQ alpha chains can vary substantially, especially among different ethnicities and among different alpha-beta chain combinations, making practically infeasible the exhaustive testing of all possible DPA/DPB and DQA/DQB combinations.