Diagnostic Models for Screening of Periodontitis with Inflammatory Mediators and Microbial Profiles in Saliva

This study aims to investigate and assess salivary biomarkers and microbial profiles as a means of diagnosing periodontitis. A total of 121 subjects were included: 28 periodontally healthy subjects, 24 with Stage I periodontitis, 24 with Stage II, 23 with Stage III, and 22 with Stage IV. Salivary proteins (including active matrix metalloproteinase-8 (MMP-8), pro-MMP-8, total MMP-8, C-reactive protein, secretory immunoglobulin A) and planktonic bacteria (including Aggregatibacter actinomycetemcomitans, Porphyromonas gingivalis, Tannerella forsythia, Treponema denticola, Fusobacterium nucleatum, Prevotella intermedia, Porphyromonas nigrescens, Parvimonas micra, Campylobacter rectus, Eubacterium nodatum, Eikenella corrodens, Streptococcus mutans, Staphylococcus aureus, Enterococcus faecalis, and Actinomyces viscosus) were measured from salivary samples. The performance of the diagnostic models was assessed by receiver operating characteristics (ROCs) and area under the ROC curve (AUC) analysis. The diagnostic models were constructed based on the subjects’ proteins and/or microbial profiles, resulting in two potential diagnosis models that achieved better diagnostic powers, with an AUC value > 0.750 for the diagnosis of Stages II, III, and IV periodontitis (Model PA-I; AUC: 0.796, sensitivity: 0.754, specificity: 0.712) and for the diagnosis of Stages III and IV periodontitis (Model PA-II; AUC: 0.796, sensitivity: 0.756, specificity: 0.868). This study can contribute to screening for periodontitis based on salivary biomarkers.


Introduction
The diagnosis of periodontitis is conventionally based on clinical evaluation, including probing pocket depth, bleeding on probing, clinical attachment level, periodontal index and gingival index, and radiographic examinations. However, the conventional method has a limitation in that some

Subjects
This clinical study was approved by the Institutional Review Board of Seoul National University Dental Hospital (CRI19004), approved on 4 April 2019. The subjects in this study were recruited from July 2019 to December 2019. Upon receiving written consent, 121 human subjects, aged 18 years or older, were evaluated at the Department of Periodontology in the Seoul National University Dental Hospital. All subjects involved in this study were required to have 20 or more teeth. In addition, the exclusion criteria applied was as follows: (a) use of any antibiotics or anti-inflammatory drugs within 3 months of registration; (b) use of an immunosuppressant (methotrexate, leflunomide, tacrolimus, cyclosporin, azathioprine) or adrenal cortical hormone (oral or injection) within 3 months of registration; (c) having less than 20 teeth; (d) having uncontrolled hypertension or diabetes; (e) subjects who have serious cardiovascular disease, respiratory system disease, kidney disease, liver disease, digestive system disease, blood system disease, or neuropsychiatric disease; (f) subjects with hyperthyroidism or hypothyroidism; (g) women who were pregnant or planning to become pregnant; (h) subjects with autoimmune diseases; (i) subjects with a history or presence of malignant tumors in the jawbone; (j) subjects who have had a history of or are currently using drugs or alcohol abuse within one year; (k) subjects with other inflammatory diseases in the oral cavity besides periodontitis, such as stomatitis (including ulcerative, blistering, erosive) or oral cancer; (l) subjects with other inflammatory diseases in the oral cavity besides periodontitis, such as ulcers, simple herpes and shingles, or fungal or bacterial infections; (m) subjects whose participation was judged by the researcher to be inappropriate because their involvement may cause ethical problems or seriously affect the research results.

Clinical and Radiographic Examination
A consent form was signed by and obtained from each subject following a sufficient explanation of the study. To identify if the subject was suitable for this study, demographic information such as gender and date of birth was obtained, and the systemic conditions of the participants were also examined.
On the first visit, clinical examinations were performed, and the following parameters were recorded. The gingival index (GI) [24] and plaque index (PI) [25] were examined on the buccal and lingual surfaces of the teeth. Additionally, the probing pocket depth (PPD), gingival recession (GR), and clinical attachment level (CAL) were measured at the six sites around the tooth. The amount of tooth loss (TL), sites of bleeding on probing (BOP), tooth mobility (TM), and furcation involvement (FI) were recorded per tooth. Alveolar bone loss at the mesial and distal site of the tooth was measured with periapical radiographs. The subjects were classified as healthy or Stages I, II, III, or IV depending on the severity, extent, and complexity of their periodontitis [23].

Preparation of Solution for Saliva Storage
Approximately 0.1 M phenylmethylsulphonyl fluoride (PMSF) stock solution, dissolved in isopropanol, was stored at room temperature and 0.5 M ethylenediaminetetraacetic acid (EDTA) stock solution, dissolved in distilled water (DW), was refrigerated. The two stock solutions were stored independently due to the instability of PMSF when mixed with 1X phosphate-buffered saline (PBS).

Whole Saliva with Draining Method
Unstimulated saliva was collected with the participant's head tilted slightly forward in a sitting position by drooling into a funnel-shaped test tube. The sampling was performed for 15 min and was stopped when the amount collected reached 5 mL. Subsequently, the saliva sample was placed on ice and supplemented with 1X PBS 4930 µL, 20 µL EDTA solution, and 50 µL PMSF stock solution; then, vortexing was performed. The samples were stored in a deep freezer at a temperature of −80 • C immediately after collection for the preservation of biomarkers.

Saliva Collection for Oral Microbial Identification
After collecting GCF, subjects took another break for 5 min. The subjects gargled and rinsed with gargle solution (EasyperiO kit, YD Life Science Company, Gyeonggi-do, Korea) for 30 s and then spit it into the sample container. The sample container cap was closed tightly. Samples were stored in a refrigerator (4 • C) and transported to the analytical company (YD Life Science company, Gyeonggi-do, Korea) on dry ice.

Sample Size Calculation
The sample size was established according to the previous study by Mauramo M et al. [26], reporting the diagnostic performance of MMP-8 for periodontitis with an AUC value of 0.67. With a 95% confidence interval and 80% power, the sample size was calculated as a total of 108 subjects [27]. Give a 10% drop-out rate, 30 subjects were to be collected for each group of periodontitis.

Statistical Analysis
Continuous data were represented with means and standard deviation for each subject group. Group comparisons were made using one-way ANOVA in SPSS version 17 (IBM Software, Armonk, NY, USA). Dichotomized data were represented with a number and percentage for each subject group. Group comparisons were made with Fisher's exact tests. Differences were considered statistically significant when p-values < 0.05. Seven diagnostic models of periodontitis (Model PD), namely, Stage I periodontitis (Model PD-I), Stage II periodontitis (Model PD-II), Stage III periodontitis (Model PD-III), Stage IV periodontitis (Model PD-IV), periodontitis above Stage I (Stages II, III, and IV; Model PA-I), and periodontitis above Stage II (Stages III and IV; Model PA-II), were constructed according to the concentration of proteins and microbial profiles based on a forward stepwise logistic regression analysis using SPSS statistics software (version 21.0, IBM Software, Armonk, NY, USA). The diagnostic models were evaluated by sensitivity, specificity, and ROC curve analysis using Excel (Microsoft 365, Redmond, WA, USA).

Demographic and Clinical Characteristics of Subjects
Thirty-eight male (30.4%) and eighty-seven female (69.6%) subjects, ranging in age from 20 to 79 years, were enrolled in the study. Following the recording of clinical and radiographic parameters, the subjects were allocated into the five groups of periodontal status. One subject was excluded due to the lack of teeth. The data of three subjects were excluded because the proteins in their whole-saliva samples, including active MMP-8, pro-MMP-8, and total MMP-8, did not show detectable levels. A total of 121 subjects were included in the final analysis ( Figure 1). Analysis of the data obtained from the healthy (n = 28) and periodontitis populations (n = 93; Stage I: 24, Stage II: 24, Stage III: 23, and Stage IV: 22) was performed in this study.
The demographic data ( Table 2) for systemic disease, including osteoporosis and hepatitis, were balanced among the five groups. However, age, sex, hypertension, diabetes mellitus, and smoking were statistically and significantly different among the groups. The subjects with hypertension, diabetes, and smoking were not found in the periodontally healthy group; however, they were significantly associated with Stage III and Stage IV periodontitis. In addition, the older participants were significantly found to have higher stages of periodontitis. Figure 1. Consolidated standards of reporting trials (CONSORT) flow diagram of the study. 125 subjects were screened and one subject was excluded due to lack of teeth. Finally 124 subjects were included in this study and allocated into the five groups of periodontal status including healthy, periodontitis stage I, II, III, and IV. Three subjects were excluded in the analysis due to failure in enzyme-linked immunosorbent assay (ELISA) dectection. A total 121 subjects were included in the final analysis.  Dental and periodontal data ( Table 2)

Construction of Diagnostic Models Based on Inflammatory Mediators in Saliva
Diagnostic models of periodontitis (Model PD), Stage I periodontitis (Model PD-I), Stage II periodontitis (Model PD-II), Stage III periodontitis (Model PD-III), Stage IV periodontitis (Model PD-IV), periodontitis above Stage I (Stages II, III, and IV; Model PA-I), and periodontitis above Stage II (Stages III, and IV; Model PA-II) were constructed using the proteins and microbial profiles that showed significant differences.
The Model PD mathematical formula, with an accuracy of 0.802 (Table 3), was constructed to discriminate the periodontitis groups from the healthy group using logistic regression analysis (Equation (1)). This model showed high sensitivity (1.000), but low specificity (0.143) ( Table 3). The Model PD-I mathematical formula, with an accuracy of 0.785 (Table 3), was constructed to diagnose the Stage I periodontitis group from the healthy group and Stage II, III, and IV periodontitis patients using logistic regression analysis (Equation (2)). This model showed high specificity (0.969) but low sensitivity (0.042) ( Table 3).
The Model PD-II mathematical formula, with an accuracy of 0.760 (Table 3), was constructed to diagnose the Stage II periodontitis group from the healthy group and Stage I, III, and IV periodontitis patients using logistic regression analysis (Equation (3)). This model showed high specificity (0.948) but low sensitivity (0.000) ( Table 3).
The Model PD-III mathematical formula, with an accuracy of 0.835 (Table 3), was constructed to diagnose the Stage III periodontitis group from the healthy group and Stage I, II, and IV periodontitis patients using logistic regression analysis (Equation (4)). This model showed high specificity (0.959) but low sensitivity (0.304) ( Table 3).
The Model PD-IV mathematical formula, with an accuracy of 0.860 (Table 3), was constructed to diagnose the Stage IV periodontitis group from the healthy group and Stage I, II, and III periodontitis patients using logistic regression analysis (Equation (5)). This model showed high specificity (0.980) but low sensitivity (0.318) ( Table 3).
The Model PA-I mathematical formula, with an accuracy of 0.736 (Table 3), was constructed to discriminate the Stage II, III, and IV periodontitis groups from the healthy group and Stage I periodontitis patients using logistic regression analysis (Equation (6)). This model showed improved sensitivity (0.754) and specificity (0.712) ( Table 3).
The Model PA-II mathematical formula, with an accuracy of 0.826 (Table 3), was constructed to discriminate the Stage III and IV periodontitis groups from the healthy group and Stage I and II periodontitis patients using logistic regression analysis (Equation (7)). This model showed good specificity (0.756) and sensitivity (0.868) (

Validation of Multianalyte Models
Among the seven models, Model PA-I and Model PA-II, which showed good sensitivity and specificity, were further investigated with the ROC curves ( Figure 3). It was found that the models combining salivary biomarkers and microbial profiles were useful for discriminating periodontal status. The AUC values of pro-MMP-8, C. rectus, and E. nodatum were 0.720, 0.685, and 0.733, respectively. In contrast, the AUC value of Model PA-I, which combined the markers, was 0.796. The AUC values of sIgA, P. gingivalis, T. forsythia, P. micra, E. nodatum, and A. viscosus were 0.695, 0.653, 0.694, 0.736, 0.831, and 0.673, respectively, while Model PA-II, which combined them, showed an AUC value of 0.894.

Discussion
In this study, we built the seven diagnostic models for periodontitis by combining meaningful salivary biomarkers (including inflammatory mediators and microbial profiles) that significantly changed with the stages of periodontitis. Among the seven models, PA-I and PA-II were both highly sensitive and specific compared with PD-I, -II, -III, and -IV models. In Model PA-I, E. nodatum and C. rectus were included; these bacteria belong to the orange complex, which has a role in linking early colonizing and pathogenic bacteria of periodontitis. Therefore, it is thought that these bridging species are very important in the diagnosis of periodontitis with Stages II, III, and IV of the disease. Meanwhile, in Model PA-II, the purple complex (such as A. viscosus), orange complex (such as P. micra and E. nodatum), and red complex (such as P. gingivalis and T. forsythia) were included. The high accuracy of Model PA-II means that overall planktonic bacterial species can increase when the severity of periodontitis increases because various species consisting of early colonizer, bridging species, and pathogenic bacteria are included. Simple counting of bacteria also showed that there were more groups of streptococcus and A. viscosus in Stage IV, meaning the more severe the stage of periodontitis is, the more overall bacteria there are.
To our knowledge, there is little literature to validate a new classification system of periodontitis in the development of a diagnostic method, including salivary biomarkers. In 2017, the new classification system for case definitions of periodontitis was developed and suggested based on cumulative studies for almost 20 years [28]. In the workshop, three obviously different forms of periodontitis based on pathophysiology were categorized: necrotizing periodontitis, periodontitis as a direct manifestation of systemic disease, and periodontitis. Herein, we included the latter form. The periodontitis stage increased according to the severity, complexity, extent, and distribution of the disease. Our prediction models for periodontitis showed high accuracy based on this new classification system by using limited amounts of significant biomarkers.
In this study, we included both pro-and active forms of MMP-8. Several studies found that the active forms of MMP could distinguish periodontitis [19,[29][30][31][32][33][34]; however, in this study, two forms were significantly different among all stages of periodontitis. The increase was considered to reflect the increased leakage of all types of MMP-8 according to the severity of periodontitis. Interestingly, CRP did not show a significant difference, which was not consistent with other studies [35,36]. This might be because the detective capacity of ELISA is not sensitive enough to detect the changes in the level of CRP [37,38]. Therefore, to monitor the level of CRP, more sensitive techniques should be used.
Contrary to our expectations, all periodontopathogenic bacteria were not detected in the saliva in periodontitis patients. This could be a result of the fact that almost all the patients showed periodontitis in a "localized" form, and the amount of planktonic bacteria was too low to be detected. Further studies should be performed in patients with "generalized" periodontitis.
SIgA indicates adaptive immunity and is widely used for a diagnosis of periodontitis. Mesa et al. reported similar results showing sIgA was not significantly different between healthy patients and periodontitis groups [39]. Chronic inflammation increased cortisol levels and, inversely, decreased sIgA [40]. In the present study, sIgA seemed to show higher levels in Stages III and IV of periodontitis compared with other groups, but there were no statistical differences.
In spite of the limitations of the cross-sectional study, our findings suggest that whole saliva might be used as a diagnostic tool for periodontitis. Additionally, the proper selection of biomarkers in whole saliva is important in order to increase the sensitivity and specificity of the diagnosis of periodontitis [41,42]. On the other hand, this method had limitations in the early diagnosis of Stage I. Other studies also showed similar limitations, where the screening test could not distinguish early-stage periodontitis [43]. By definition, Stage I periodontitis is a bridge between gingivitis and periodontitis; thus, it can show lower levels of biomarkers than the severe forms. Other researchers have tried to discriminate gingivitis and periodontitis through macrophage inflammatory protein-1α [44,45]. To overcome the problem, it is necessary to find additional biomarkers or develop more sensitive techniques for early detection.
The diagnosis of periodontitis with clinical parameters is a very effective tool, but it is time-consuming and labor-intensive. The evaluation of clinical parameters is somewhat difficult to standardize and cannot monitor the real-time changes in periodontal disease progression [41,42]. Therefore, the method of whole-saliva analysis could be an easier and simpler diagnostic tool for the detection of periodontitis. Additionally, salivary biomarkers can be a very prospective screening tool when one considers the high correlations between periodontitis and systemic disease [29,46,47].

Conclusions
This study can contribute to screening for periodontitis based on salivary biomarkers. Two potential diagnosis models, PA-I for the diagnosis of Stage II, III, and IV periodontitis and PA-II for the diagnosis of Stage III and IV periodontitis, showed the highest performance with biomarkers in whole saliva.