Network Protein Interaction in the Link between Stroke and Periodontitis Interplay: A Pilot Bioinformatic Analysis

The clinical interaction between stroke and periodontitis has been consistently studied and confirmed. Hence, exploring potentially new protein interactions in this association using bioinformatic strategies presents potential interest. In this exploratory study, we conducted a protein–protein network interaction (PPI) search with documented encoded proteins for both stroke and periodontitis. Genes of interest were collected via GWAS database. The STRING database was used to predict the PPI networks, first in a sensitivity purpose (confidence cut-off of 0.7), and then with a highest confidence cut-off (0.9). Genes over-representation was inspected in the final network. As a result, we foresee a prospective protein network of interaction between stroke and periodontitis. Inflammation, pro-coagulant/pro-thrombotic state and, ultimately, atheroma plaque rupture is the main biological mechanism derived from the network. These pilot results may pave the way to future molecular and therapeutic studies to further comprehend the mechanisms between these two conditions.


Introduction
Stroke is a highly prevalent neurological disease affecting one in four adults worldwide [1,2]. It is the second leading cause of death (accounting for almost 6 million deaths/year) and the third leading cause of disability in the world [1,2]. Most stroke cases are ischemic due to reduced blood flow caused by arterial occlusion. The remaining stroke presentations are haemorrhagic, resulting from rupture of cerebral arteries [3]. A recent economic analysis in Europe reported that stroke cost EUR 60 billion, with health care accounting for EUR 27 billion (45%), representing 1.7% of health expenditure [4].
Periodontitis, a chronic oral inflammatory and infectious disease that is characterized by destruction of soft and hard gingival tissues, affects more than half of adults [5]. It has been estimated that the most advanced forms of this disease affect around 11% of the adult population worldwide [6]. More importantly, periodontitis not only affects the gingival tissues but also contributes to the body's overall inflammatory burden, which makes this disease a potential risk factor for atherosclerotic inflammatory vascular diseases such as stroke [7].
A link between periodontitis and stroke, mainly ischemic stroke, has been established over the last few decades [8]. Moreover, stroke survivors with periodontitis are more prone to have worse prognosis and more likely to suffer recurrent vascular events than those without periodontitis [9,10]. Potential mechanisms proposed behind the association between periodontitis and ischemic stroke might include activation of innate immune system, systemic inflammation leading to vascular endothelial dysfunction, increased cholesterol biosynthesis, and prothrombotic state. All of them may exacerbate atherosclerotic lesions and increase the risk of atherosclerosis, thus promoting cerebrovascular events including large and small vessel cerebral infarcts [11]. However, a recent Mendelian randomization study failed to demonstrate a causal genetic association between both diseases where only five single nucleotide polymorphisms were used from a number of genome-wide association studies [12].
To better understand the contribution of genetics to this relationship, genetic risk factors should be analysed and integrated in terms of biological pathways and functions [13]. For this purpose, protein interaction data, derived from a wide range of cellular and biochemical model systems, can be used for building protein-protein interaction (PPI) networks from genes associated with both periodontitis and stroke. PPI network analysis could be, therefore, a powerful time-and cost-effective approach to identify potential biological pathways, key players, or candidate genes involved in the periodontitis-stroke link.
Hence, the aim of this study was to identify the key proteins and the biological regulatory pathways involved in both periodontitis and stroke physiopathology.
For stroke, we used summary GWAS statistics performed in over 2,000,000 individuals of European, Asian, American, Sub-Saharan African, Caribbean and other ancestries  (Supplementary S2). In this, we included the link where genes were reported in the GWAS search [63]. Both GWAS data for stroke and periodontitis were resulting from different populations as none of the included studies had data combining both conditions.

Protein-Protein Interaction Networks Functional Enrichment Analysis
To forecast potential PPI networks, we used the STRING (Search Tool for the Retrieval of Interacting Genes/Proteins) database, through heuristic methods of association and analysis. This platform renders possible protein networks of interaction via high-throughput experimental data, literature, and predictions grounded on genomic context analysis [64,65]. Five main sources contribute to this bioinformatic output: Genomic Context Predictions, High-throughput Lab Experiments, (Conserved) Co-Expression, Automated Textmining and Previous Knowledge in Databases. The Universal Protein Resource provided the characteristics of all proteins [66]. Then, the STRING tool provides a node-node interaction score calculated as detailed in [67].

Data Management, Test Methods and Analysis
After download the GWAS datasets of periodontitis and stroke, data was handled through Microsoft Office Excel. Then, we used the 'Multiple protein' to render PPI networks at STRING version 10.5. We carried a first analysis with a confidence cut-off of 0.7 to serve as sensitivity analysis aiming to infer the dependency on the choice of the confidence cut-off. Then, we defined a cut-off of 0.9, the highest confidence in the final interaction examination. In the resultant PPI network, proteins are presented as nodes related through lines, and nodes thickness increase reveals higher confidence level. Furthermore, 'ggplot' for R was used to assemble a gene interaction heatmap.

Gene Enrichment Analysis
Then, to investigate whether there were genes over-represented in the final network and its possible effect, we run the list of proteins in The Geneontology Resource (http: //geneontology.org/ accessed on February 2021). A 'Reactome pathways' was executed, and false discovery rates were computed using Fisher's Exact. Functional enrichment analysis was further confirmed using the WebGestalt tool (http://www.webgestalt.org/ accessed on April 2021).

Protein-Protein Interaction Analysis
According to the STRING results, a sensitivity analysis with a confidence cut-off of 0.7 (Supplementary S3) exposed dependency on the confidence cut-off, as a more concise network emerged with a confidence cut-off of 0.9 ( Figure 1). Overall, 148 nodes with 41 PPI relationships (from 32 expected edges) were found ( Figure 1). The PPI enrichment significance was 0.0795, indicating the current set of proteins is small and the result of these proteins have not been studied very much. The average node degree was 0.554, and the average local average local clustering coefficient was of 0.194.
The casted network display possible PPIs between known associated genes stroke and periodontitis ( Figure 1, Table 1). Furthermore, the physiological characteristics and localization of each encoded protein is presented in Table 2. The gene interaction was further confirmed via heatmap plot ( Figure 2) and Genemania network corroborated a PPI similar network obtained in STRING database (Supplementary S4). Table 1. Interaction weight between stroke and periodontitis genes identified in the network.

Genes for Stroke (Regulation)
Genes for Periodontitis (Regulation) Score  Metalloprotease that may play a role in the degradation of COMP. Cleaves also α-2 macroglobulin and aggregan. Has anti-tumorigenic properties Nucleoli and mitochondria

Gene Enrichment Assessment and Gene Ontology
The over-representation analysis with "Reactome pathways" revealed some overrepresentation in terms of: fibrin clot formation; CREB phosphorylation through the activation of CaMKII; Ras activation upon Ca 2+ influx through NMDA receptor; unblocking of NMDA receptors, glutamate binding and activation; defective B3GALTL causes Peters-plus syndrome (

Gene Enrichment Assessment and Gene Ontology
The over-representation analysis with "Reactome pathways" revealed some overrepresentation in terms of: fibrin clot formation; CREB phosphorylation through the activation of CaMKII; Ras activation upon Ca2+ influx through NMDA receptor; unblocking of NMDA receptors, glutamate binding and activation; defective B3GALTL causes Petersplus syndrome (PpS); O-glycosylation of TSR domain-containing proteins; Platelet degranulation; Platelet activation, signalling and aggregation; haemostasis (Supplementary S5). The WebGestalt tool confirmed the gene enrichment assessment performed.
Furthermore, upon the genet ontology of these significant genes, the most significant molecular functions were: titin binding (strength = 2.

Discussion
In this exploratory bioinformatic analysis, a likely PPI network between stroke and periodontitis was foreseen using open-access catalogues of GWAS studies in humans. This forecasted PPI network needs careful interpretation and warrants preclinical and clinical validations, nevertheless shows new evidence that will empower new studies on the association between stroke and periodontitis. Below we present different hypothetical biological scenarios based on our PPI network through which periodontitis can increased the risk of stroke.
The largest node found with the PPI network is related to the atherosclerotic process. In the early stages of atherogenesis, two α actinins associated with periodontitis, namely ACTN-1 and -2, can bind to intracellular adhesion molecule 1 (ICAM-1) providing a "firm foothold" for leukocytes undergoing migration between endothelial cells into the intima layer of the artery and the underlying inflammatory lesion [65]. In this phase, endothelial dysfunction can be seen as the endothelial lining partially has lost its integrity. In this sense, ACTN-1 and -2 interact with SERPINA1, which encodes α 1-antitrypsin (A1AT), a protease inhibitor that protects surrounding tissues at sites with inflammation and has been found in human atherosclerotic lesions [66]. On the other hand, after pro-inflammatory cytokines and matrix metalloproteinases (MMPs) and its tissue inhibitor (e.g., TIMP-1) upregulation takes place due to lipid streaks and calcification, the atheroma plaque starts the maturation process. This process together with development of a compensatory blood supply within the atherosclerotic lesion, will lead to additional activation and proliferation of pro-inflammatory cells and production of thrombin (interaction TIMP-1 with F5). In this stage, the endothelial lining is all lost and fibrinogen is enzymatically generated from fibrin. This clotting cascade (interaction F2 and F5 with FGG) will result in formation of thrombus (i.e., thrombosis) and atherosclerotic cerebrovascular disease will develop latter on. MMPs overexpression leads to degradation of collagen, weakening of the vessel's strength and fissures in the atheroma (interaction between TIMP-1 with SERPINA1). In some cases, where atheroma keeps progressing, a large necrotic core is exposed to the vasculature with the lesion which results in contact with platelets, coagulation is initiated and ultimately plaque rupture (i.e., vulnerable lesions) (interaction between FGG and SERPINA1).
In the interrelation of pro-coagulant state and formation of thrombus, another interesting node was observed in the PPI network mainly consisted of metallopeptidase with thrombospondin type 1 motif 2 and 12 (ADAMTS2 and 12) and Thrombospondin type 1 domain containing 4 (TSH4). In addition, they are produced by vascular smooth muscle cells and platelets, ADAMTS have also a key role in maintaining cardiovascular haemostasis and down-regulating coagulation by inhibition of thrombin [67]. For example, an increase in ADAMTS12 activity would hypothetically lead to less inhibition of thrombin and an elevated risk for ischemic stroke. On the other hand, genetic variations of the ADAMTS family such as ADAMTS2 and 12 have shown to reduce the integrity of the endothelial lining, which together with inflammatory processes and defective vascular remodelling might play a key role in cerebral aneurysms pathogenesis [68]. Thrombospondin 1 is elevated in gingival tissues with periodontitis and this overexpression is induced by lipopolysaccharide from Porphyromonas gingivalis (a keystone pathogen) via innate immunity system activation and inflammatory responses [69]. Therefore, biologically, it would be plausible that periodontitis through the interaction of TSH4 and ADAMTS2 and 12 is associated with stroke and cerebral aneurysms.
Calcium/calmodulin dependent protein kinase II (CAMK2) has been related to ischemic neuronal death due to glutamate-mediated excitotoxicity [70]. In the present PPI network ACTN2 (a protein which main role is to anchor actin with several intracellular molecules) was associated with CAMK2D, which is involved in the regulation of Ca 2+ homeostasis and excitation-contraction. Concentrations of some amino acids such as arginine and glutamate were altered in gingival crevicular fluid of periodontal patients [71]. Therefore, the pathway ACTN2-CAMK2 deserved further investigation, as it could be an interesting and relevant biological mechanism behind the relationship between periodontal destruction and cerebral ischemia.
Other associations derived from the present PPI network includes complement system activation (C5AR1) and increased pro-coagulant/thrombotic (F2) state via melanin. Increased local activation of complement products such as C5 in the periodontal tissues increases the intensity of the local inflammatory response, resulting in enhanced vascular permeability and vasodilatation and recruitment of inflammatory cells, which, in turn, will lead to excessive release of reactive oxygen species and interleukins which promotes accumulation of immune cells [72]. Hence, a potential link between activation of immune systemic via C5 activation and overproduction of for instance fibrinogen, a molecule related to systemic inflammation and intimately involved in the development of atherosclerosis could be feasible. Research needs to be done investigating the mechanism underlying this finding as well as to study the exact role of melanin in this link.
The last node to be highlighted is the one which relates ubiquitination, autophagy and coagulation. For instance, in our PPI network we found that Smad ubiquitination regulatory factor-2 (SMURF2), an ubiquitin E3 ligase responsible for proteasome-mediated degradation of enhancer of zeste homolog 2 EZH2 which is a process required for neuron differentiation [73] is present in periodontitis. This ubiquitin was associated with the process of autophagy in stroke and ultimately with promotion of coagulation. Again, a potential biological pathway has been described for the perio-stroke relationship.
Importantly, we highlight that all these proposed interactions are hypothetical and demand further validation. Furthermore, the inflammatory machinery is inherent in the vast majority of diseases and is still far from being fully understood and, therefore, is very difficult to be established as a definitive criterion of interaction between these two conditions. Thus, the reader must bear in mind that the aforementioned lines of interaction are, at this stage, contingent on further evidence.
Overall, this exploratory report foresees a comprehensive analysis using free and large outputs. However, a number of shortcomings are relevant to discuss. First, this sort of analysis is dependent on the number of genes included in GWAS, and future studies with a higher number of genes disclosed will result in new paths of interaction. Second, network protein interaction approaches have limited clinical predictive value [68], but unveil hypothetical new paths of interaction. On the other hand, the quantity of SNPs of interest in these datasets has combined a diversity of ethnical background allowing the generalization of our results. Beyond these limitations, our sample size (over 2 million people) makes the results persuasive. Furthermore, we have also provided a proteinenrichment analysis, to overlook the over-representation level e in the observed PPI.

Conclusions
This exploratory bioinformatic analysis shows a potential network of interaction between proteins related to stroke and periodontitis, respectively. The main biological pathway identified relies on inflammatory response, pro-coagulant/pro-thrombotic state and subsequent mechanisms of atheroma plaque rupture. These results may shed light on future molecular and therapeutic studies to further grasp the association between these two conditions.

Conflicts of Interest:
The authors declare no conflict of interest.