Salivary Proteins and Metabolites as Caries Biomarkers in Adolescents

Dental caries is a multifactorial disease involving both intrinsic and extrinsic factors. Host factors such as the individual’s immune competence, tooth morphology, saliva flow rate and saliva buffering capacity, as well as bacterial composition have a direct impact on tooth decay progression [1, 2]. The development of dental caries is not solely dependent on internal factors but also to a large extent influenced by environmental factors, for example, diet, oral hygiene, fluoride exposure, socioeconomic, and medical parameters [3, 4].

Despite a well-documented decline of the disease over the past decades, dental caries still affects a significant proportion of children and adolescents. Recent research shows a skew distribution, and few individuals carry the greatest burden of the disease [5, 6]. In data from 2022, retrieved from the national SKaPa registry (Swedish Quality Registry for Caries and Periodontal Disease), approximately 40% of the Swedish 19 years olds were caries-free [7]. Overall, dental caries causes suffering and pain in individuals’ lives, rendering societal costs worldwide [8] as well as in Sweden [6]. This suggests a need of early and precise caries risk assessment [5, 6].

The pathogenesis of dental caries is a result of a dysbiotic, cariogenic microbial community that interacts with the intrinsic and extrinsic factors of the disease [9]. The caries process could be described as a constant interaction between the metabolic processes in the biofilm at the tooth surface and the saliva surrounding it; therefore, an understanding of salivary functions and composition is important when investigating potential factors involved in caries development [3, 10]. In addition, saliva is an easy-to-collect, reproducible sample, making it a convenient fluid to develop health monitoring diagnostics [11], identifying compounds associated with dental caries that could be used as biomarkers of the disease [12].

Methods for measurements of salivary compounds have improved radically over the last decades, for example, by the development of high-sensitivity and multiplex protein quantification methodologies that require smaller amounts of saliva. Multiplex immunoassays, for instance, allow the detection of multiple targets in a single reaction well that can be distinguished from each other and quantified simultaneously [13].

Another set of powerful techniques include those aiming at describing the metabolomic composition of oral samples, such as mass spectrometry or nuclear magnetic resonance (NMR) spectroscopy. NMR is a robust and quantitative method for determining molecular concentrations in small samples with minimal sample preparation [14], and previous studies on NMR have yielded promising results regarding metabolite profiling in caries from different sources, such as dental plaque [15] and saliva [16]. Several metaproteomic or metabolomic studies of dental plaque based on mass spectrometry identified several human or bacterial compounds that appeared to differentiate caries-free from caries-experienced individuals [17, 18].

In addition to these open-ended approaches, targeted determination of specific proteins has commonly been used to detect or validate potential caries biomarkers, mainly in the adult population or in children with early childhood caries [12, 19]. In some cases, this failed to identify proteins with discriminatory power, and in others, several potential biomarkers were found which maximized the diagnostic distinction between conditions of health and disease. Typically, the selected compounds are related to immune competence, adhesion capacity of microorganisms, and acid production or pH buffering. Although some of these tests have shown positive results in the adult population [20], the same biomarkers have failed to show significant differences in children [21], which is where caries risk assessment could prove most useful for prevention programs. This supports the use of age-specific oral biomarkers. In addition, there are few studies aiming to evaluate caries-associated salivary compounds in adolescents with a young permanent dentition, and no reliable biomarkers to distinguish between individuals with and without caries have been found [22].

The aim of the present study was to identify potential caries biomarkers related to host factors and to the metabolic output of bacterial communities, in saliva collected from adolescents with and without caries experience. In addition, we have aimed to evaluate different factors that could influence biomarkers’ concentration in saliva, such as time of sampling or gender, as well as information about diet and oral hygiene habits, which altogether could hinder the identification of appropriate predictive markers.

The study population comprised 40 adolescents with permanent dentition, between 14 and 18 years of age receiving treatment at two Public Dental Clinics in the county of Jönköping, Sweden. A strategic sample of study subjects, an equal number of boys and girls in both groups were recruited. The participants were divided into two groups; 20 adolescents were caries-free and 20 had caries.

The inclusion criteria in caries group were ≥3 surfaces (DiMFS) with initial or manifest proximal and/or buccal caries lesions. The exclusion criteria for participating in the study were (1) using antibiotics or other systemic medications within the last 3 months; (2) reporting daily use of tobacco or alcohol; (3) routine use of antiseptics. None of the participants had periodontal disease, and presence of gingivitis was registered and did not vary between the groups. All recruitment, examinations and data collection were performed by one specialist in pediatric dentistry (K.H.).

Ethical approval was obtained from the Regional Ethics Committee for Human Research at Linköping University, Sweden, 2017/599-31. All participants and their parents (if participants were younger than 15 years of age) signed up an informed consent form prior to the initiation of the study. The study was conducted in line with the Helsinki Declaration [23].

An intraoral examination, including four posterior bite-wing radiographic examination, registration of caries lesions through gentle probing, dental plaque [24], and gingivitis [25], were performed by the same dental health professional (K.H.). Initial and manifest caries were registered (clinical and radiographic) according to the criteria by Koch [26] and Alm [27]. For intra-individual calibration, all radiographics were reviewed after an interval of 3 months, made by the first author (K.H.). The weighted kappa value was 0.61 (CI: 0.54–0.68). The percentage agreement was 89.7%. In connection with the clinical examination, all participants answered five questions regarding oral hygiene, use of fluoride supplements, diet intake frequency, and food content. These questions were included in the ordinary anamnesis and were based on earlier questionnaire [28]. The questionnaire was not included in the caries risk assessment modeling, where only continuous variables (i.e., metabolite concentration values) were used.

Saliva sampling was performed over a 5-month period between October 2018 and February 2019. All participants were instructed to avoid tooth brushing 24 h before sampling. Sampling was performed twice a day, between 07:30 and 09:00 h, and between 12:30 and 15:30 h, and at least 1 h after food intake.

Unstimulated whole saliva (2.0 mL) from each participant was collected in test tubes. Saliva was transferred to microcentrifuge tubes in 500 μL aliquots. The tubes were sealed and transported on ice to the laboratory, then stored at −80°C until analysis of salivary compounds, which included colorimetric analysis of ions, ELISA tests, multiplex immunoassays, NMR, and enzymatic tests. A flowchart of the study design is shown in Figure 1. On the day of the analyses, the corresponding saliva aliquots were thawed at room temperature, vortexed, and centrifuged at 1,500 g for 15 min, following recommendation by Salimetrics on sample handling and preparation (Salimetrics, Carlsbad, CA, USA). Supragingival plaque samples from the same individuals were also used for comparison (see Metabolomics section below).

Saliva samples were thawed at RT for 1 h, then centrifuged at 10,000 g for 15 min at 4°C. 180 µL of each saliva supernatant was transferred to a deep well plate containing 20 μL buffer (400 mm potassium phosphate, pD 6.95, 0.1% 3-(trimethylsilyl)propionic-2,2,3,3-d₄ acid (TSPd₄), 0.13% sodium azide, all in 99.8% deuterium oxide) per well, using a SamplePro Tube L (Bruker BioSpin, Ettlingen, Germany) liquid handler. The DWP was shaken at 800 rpm for 1 min at 9°C before being spun down at 2,250 g, 1 min at 4°C. The Sample Pro robot was used to transfer the mixed NMR samples to 3 mm SampleJet tubes. The temperatures of the cooled DWP and SampleJet rack positions in the robot were set to 2°C during the whole preparation procedure. An 800 MHz Oxford magnet equipped with a Bruker Avance III HD console, a 3 mm TCI cryoprobe and a cooled SampleJet sample changer was used to acquire 1D 1H data with the standard pulse sequence “zgespe,” incorporating water suppression through excitation sculpting and a perfect echo sequence. 64 scans were acquired into 64k data points at a receiver gain setting of 181, with an acquisition time of 2.04 s, a relaxation delay of 2 s and a sweep width of 20 ppm. Data acquisition and processing was done in TopSpin3.5pL7 (Bruker BioSpin, Ettlingen, Germany). Zero filling to 128k points and an exponential line broadening of 0.3 Hz was applied before Fourier transformation. Automatic phasing, baseline correction, and referencing to the TSPd₄ peak were applied before importing data into MATLAB (MathWorks Inc., Natick, USA) using in-house developed scripts. Spectral features were aligned with icoshift 1.2 [29] and integrated to a linear baseline, peak by peak. Peaks were annotated using Chenomx 8.4 (Chenomx Inc., Edmonton, Canada) and the public spectral database HMDB [30]. The comparison of morning and afternoon data was done on PQ-normalized [31] data.

NMR-derived metabolomic data from plaque samples collected on the same day in the same individuals were taken from [15]. In short, interproximal plaque was collected using a sterile QuickStick (Dab Dental AB), with three sites per quadrant and a total of twelve sites per person. Sampling was not conducted if neighboring tooth or contact point were missing. Plaque samples were then collected in a separate sterile microfuge tube, sealed, and transported under cold conditions to the laboratory within 24 h, where the metabolome profile after sucrose exposure was obtained by NMR [15].

Cortisol levels (μg/dL) in saliva were determined in duplicates using the Salivary Cortisol EIA Kit (Catalog No. 1-3002; Salimetrics, USA). The assay was performed based on the manufacturer’s protocol. The test volume was 25 μL, lower limit of sensitivity 0.007 μg/dL, range of calibrators 0.007–3.0 μg/dL. Cortisol values used in the statistical analyses were averaged across duplicates in line with standard practice. Alpha-amylase was determined in duplicates using alpha-amylase kinetic reaction assay kit (Catalog No. 1-1902; Salimetrics USA). The assay was performed based on the manufacturer’s protocol. Saliva samples were diluted to 1:200 with the provided diluent before analyzed. The lower limit of sensitivity was 2.0 U/mL, range of calibrators 2.0–360.0 U/mL. Alpha-amylase values used in the statistical analyses were averaged across duplicates.

Human immunoglobulins IgG₁, IgG₂, IgG₃, IgG₄, lgA, lgM, and IgE in saliva were analyzed with multiplex fluorochrome technique (Luminex; Bio-Rad Laboratories, Hercules, CA, USA). Immunoglobulins were analyzed in morning as well as afternoon samples using a Human Isotyping Panel (Bio-Rad Laboratories). A Bio-Plex 200 TM system (Luminex xMAPTM Technology, Austin, TX, USA) was used for identification and quantification of each immunoglobulin, and the threshold was set to a minimum of 50 individual microspheres per region. Raw data (median fluorescence intensity) for each reaction were analyzed using Bio-Plex Manager^TM Software 6.0. To obtain sample concentration values, a five-parameter logistic equation was used to calculate each standard curve. The cutoff value enforced for minimum detectable concentrations for each immunological marker was as follows: IgG₁ (0.37 ng/mL), IgG₂ (0.26 ng/mL), IgG₃ (0.03 ng/mL), IgG₄ (0.04 ng/mL), lgA (0.34 ng/mL), lgM (0.33 ng/mL), and IgE (0.04 ng/mL).

Saliva was coated on 96-well plates (3912, Falcon, Franklin Lakes, NJ, USA) and placed in a humid chamber over night at room temperature. The plates were then rinsed three times with PBS containing 0.05% (v/v) Tween-20 (Bio-Rad, Hercules, CA, USA) to remove non-adherent molecules. The plates were incubated for 1 h in a blocking solution of PBS containing 0.05% (v/v) Tween-20 with 1% BSA (Sigma Chemical Co., St Louis, MO, USA). The plates where then incubated for 1 h with horseradish peroxidase (HRP)-conjugated Sambucus nigra lectin (recognizing sialic acid α [2–6]) diluted in blocking solution with 0.1 mm CaCl₂. The plates were then rinsed again with washing solution to remove unbound lectins. SigmaFast O-phenylenediamine-dihydrocloride (Sigma-Aldrich, St Louis, MO, USA) was used as substrate to visualize the lectins. The reactions were recorded in a plate reader, CLARIOstar (BMG, Cary, NC, USA). Carbohydrate reactivity was expressed as absorbance at 450 nm. Subsequent readings were performed at 30 min.

The salivary and immune metabolic analysis (SIMMA) comprises a set of biomarkers for morning and afternoon samples that were found to have discriminatory power between caries-free and caries-active adults [20]. From the SIMMA adult biomarkers, those with the highest statistical significance were selected for the current study, namely, statherin and fibronectin in the morning, and LL37 and beta-defensin 2 in the afternoon. These proteins were measured in the saliva supernatants with ELISA commercial kits using dilutions between 1:2 and 1:50 using the protocol recommended by the manufacturer adapted to the levels of these compounds in children, as previously described [21]. Formate and lactate, also part of the SIMMA test, were determined by NMR together with other organic acids, as described above.

Nitrate, nitrite, and pH measurements were performed with the RQflex^® 10 Reflectoquant^® (Merck Millipore) reflectometer, following [32]. Accuracy of all reflectometer strips was confirmed with standard solutions (Merck Millipore) with known concentrations. Saliva samples were diluted when the concentrations were above the working range of the strips. Salivary pH was measured in the lab after one defrost of the samples. Salivary flow rate was not measured.

For ammonium measurements, 20 μL aliquots were used, and the level of NH₄ estimated by the Nessler colorimetric method for each individual sample. Specifically, 100 μL of ten-fold diluted saliva were mixed with 20 μL of Nessler reagent and 80 μL of water, incubated during 5 min in dark, and measured in an optical density lector at 405 nm by duplicate. The averaged values obtained were substituted in the equation of the standard curve, obtained with standard solutions of known concentrations (Merck Millipore).

R programming language was used for statistical comparisons (R Development Core Team, 2016). Statistical differences in metabolites concentration were calculated using non-parametric Wilcoxon paired tests with multiple test correction (false discovery rate). Association analyses between metabolite concentration in saliva and plaque samples were performed using the sPLS method implemented in mixOmics R library. In short, metabolites are represented as vectors in a principal components map provided by sPLS and associations between two features are calculated as the inner product of their vectors on this map.

In order to find the best group of metabolites to classify samples as CA (caries) or CF (caries-free), we used the feature selection algorithm Boruta combined with cross-validation [33]. Boruta works as a wrapper around the learning method for classification of random forest. Briefly, it adds randomness to the given data set by creating shuffled copies of all features and rejects those features that exhibit less importance given by random forest than their corresponding shuffled copies. Then, the potential biomarkers confirmed by Boruta were used in a cross-validation procedure to select the best model based on the median area under the curve (AUC). In addition, we tested the distribution of data in each dataset using three indices: skewness, kurtosis, and the p value for Kolmogorov-Smirnov (K-S) test. According to these, an ideal normal distribution would have skewness = 0, kurtosis = 0, and K-S p = 1. Data on oral hygiene and dietary factors were analyzed by χ²-test by use of the statistical package SPSS for Windows, version 10 (SPSS Inc., Chicago, IL, USA). The PCA and OPLS-EP comparisons of morning and afternoon samples were calculated in SIMCA version 17.0 (Sartorius Data Analytics AB) using standard cross-validation settings leaving every seventh sample out in each round.

The mean age of the participants was 16.3 ± 1.49 years in the caries group and 15.6 ± 1.23 years in the caries-free group. In the caries group, all participants displayed proximal initial caries lesions, and ten participants had manifest proximal lesions, with a distribution of 1–20 affected tooth surfaces. The mean and median caries data, which included both initial and manifest caries lesions as well as dental restorations (D_i+mFS), were 17.25 and 15.00, respectively. Data regarding detailed caries status in these individuals have earlier been published [15].

There were statistical differences between the two groups in oral hygiene and diet habits. Most participants in the caries-free group brushed their teeth more often (two or three times per day) and had a significantly lower intake frequency of sweetened beverages, soda, energy drinks, and candies compared to individuals in the caries group (Table 1).

The levels of salivary statherin and fibronectin, two proteins influencing adhesion of oral bacteria, did not vary between caries-free and caries individuals. Sialic acid, which is also a target for the attachment of several oral bacteria, had statistically higher concentrations in the caries group (Fig. 2).

In relation to immune-related proteins, LL37 and beta-defensin 2 did not vary between caries and caries-free groups (online suppl. Fig. 1; for all online suppl. material, see https://doi.org/10.1159/000540090). Salivary antibodies did not vary significantly between groups either, but there was a consistent trend for different IgG types to be at higher concentrations in the caries group in the afternoon samples (p values were 0.052, 0.071, and 0.052 for IgG2, IgG3, and IgG4, respectively). However, these values were mainly driven by girls, as the tendencies were absent when boys were analyzed separately, whereas in girls, IgG4, IgG3, and IgG1 showed a significantly higher concentration in the caries group (Fig. 3). A similar sex-associated pattern was found for the levels of alpha-amylase, which were significantly higher in the caries group, but in this case only for boys (Fig. 3). Similarly, morning IgE levels had a trend for higher values in the caries group, mainly driven by boys (p = 0.065), whereas no trend was detected for girls (p = 0.573).

In both morning and afternoon samples, average pH levels were lower in caries-experienced than in caries-free individuals, but the difference was not statistically significant (Fig. 4). The concentrations of nitrate, nitrite, or the nitrite:nitrate ratios were not statistically different between caries and caries-free groups. Similarly, no significant differences were found in the levels of ammonia for both morning and afternoon samples (Fig. 4). The levels of arginine could not be determined due to severe signal overlap and generally low intensity of the arginine signals in the NMR spectra.

A total of 244 peaks could be identified by NMR analysis, from which 36 metabolites could be assigned (online suppl. Table 1). These included organic acids like lactate, acetate, formate, butyrate, or succinate, as well as amino acids, or several secondary metabolites. The metabolomic profile assessed by NMR was different between the morning and afternoon samples (Fig. 5), showing a variation in the presence and concentration of multiple compounds. This was especially apparent for individuals R3, R8, R9, and R10 due to a high increase in proline, maltose, lactate, succinate, acetoin, citrate, cadaverine, taurine, ethanolamine, and creatinine (Fig. 5a). Also, a number of metabolites appeared to increase or decrease consistently during the day (Fig. 5c). Among others, histidine, proline, tyramine, and cadaverine typically increased, whereas methylamine, trimethylamine, choline and acetate decreased (Fig. 5d).

A total of 4 proteins or NMR peaks of metabolites were significantly different between caries-experienced and caries-free individuals in the morning (Fig. 6a), and 6 in the afternoon samples (online suppl. Fig. 1). These included IgG2, IgG4, as well as 8 metabolites that could not be unequivocally assigned to known compounds. Ratios between several organic acids and ethanol, which have been shown to correlate with caries experience in other studies, were calculated, but only the lactate:succinate ratio from the morning samples was significantly different, with higher values in the caries group (Fig. 6b).

The comparison of metabolome data in the saliva samples and their corresponding plaque samples in the same individuals showed a poor correlation in the concentration of the same compounds between both niches (Fig. 7). In addition, a total of 45 salivary compounds showed significant positive or negative correlations with other metabolites in dental plaque (online suppl. Fig. 2).

Given that no individual metabolite or protein was effective in discriminating samples from caries-free and caries-active individuals, a Boruta algorithm with cross-validation was used to score the different metabolites according to their multivariate predictive capacity. The accuracy of the model was then assessed by the area under curve (AUC) values (Fig. 8). This random forest modeling shows that the best model accuracy in the morning samples is obtained by the combination of 4 metabolites (AUC median: 0.88), whereas in the afternoon samples, the best model is obtained by a combination of 3 different metabolites (AUC median: 0.93). The suggestion that the afternoon samples have a better discrimination capacity to classify caries-active and caries-free individuals is reinforced by the results of a PERMANOVA test applied to the metabolite selection datasets (p = 0.068 for morning samples and 0.005 for the afternoon samples).

Identifying individuals at high risk of developing dental caries would provide a powerful tool to implement preventive and treatment strategies in those patients with higher risk. However, the multifactorial nature of dental caries could make this task difficult [34]. Different teams have focused on bacterial composition or activity, proteins levels, buffering capacity, or the concentration of different ions, salivary proteins, organic acids, or other metabolites as potential biomarkers of dental caries [10, 17, 18, 35]. Moreover, the results are variable and not always consistent between studies, probably as a consequence of the multiple factors influencing the disease (e.g., host, microbial, and environmental features) and due to the lack of standardization of sampling procedure and analyses [34, 36, 37]. In the current manuscript, we focused on the adolescent population and have made a large effort to analyze the largest number of salivary components possible. For this, we used different methodologies that ranged from NMR to ELISA tests, multiplex capillary immunoassays, colorimetric tests, or reflectometry, giving rise to over 100 compounds that included salivary proteins, enzymes, hormones, ions, organic acids, amino acids, and other metabolites. In addition, basic data on oral hygiene and diet were also collected, with the purpose of providing a greater understanding in the differences between caries-free and caries-experienced individuals and to assess the value of anamnestic questions used in day-to-day clinical practice. However, integrating data from entirely different nature, such as metabolite concentrations, bacterial composition, or dietary habits, can be statistically challenging.

Although the collected data on diet and hygiene habits were limited, there were clear associations of sugar intake and frequency of tooth brushing with caries incidence (Table 1), suggesting that a simple questionnaire can provide valuable information about the environmental components of caries risk [38]. Caries development is considered to increase under the confluence of host, microbial, and environmental factors [34]. This means that even if we quantify a large set of salivary components that could influence caries risk, these could be masked or over-shadowed by external factors like an excellent/deficient oral hygiene or a high/low sugar intake. Thus, an effort should be made to include environmental factors into risk models, and statistical procedures should be developed to integrate information from different data types in the prediction.

Our study identified several factors that impeded the search for efficient biomarkers. One of them was gender, as several compounds appeared to efficiently discriminate between caries-experienced and caries-free individuals, but only in girls (cortisol or IgGs) or boys (alpha-amylase, IgE). A sex-related effect for caries incidence has been detected in the past in some studies. In some cases, differences in diet intake [39, 40] or hygiene habits [41] between adult males and females have been reported as influencing oral health. This sex-associated effect could be even more relevant for the adolescent population, for example, as a consequence of different rates of immune and hormonal maturation in boys and girls [42], and an effort should be made to identify biomarkers that could be unaffected by this.

Our data also underline the importance of standardized sampling time, with important differences between morning and afternoon samples in the concentration of salivary metabolites. In our dataset, afternoon samples provided better predictive capability. These differences for samples collected on the same day could be due to natural fluctuation of molecules with respect to circadian rhythms, as well as to the effect of daily activity such as eating or drinking [20, 43]. For practical reasons, we chose to have two saliva sampling times per day, but other studies have shown that more sampling occasions may reveal additional variation [20]. It is well established that a reference condition, commonly fasting conditions, is crucial in blood and urine tests to establish standardized threshold concentrations for health and disease [44, 45]. Thus, the potential need for fasting saliva sampling should be established, or the selection of a time of the day and circumstances that could minimize variability and enhance the discriminatory power of the selected biomarkers [20].

While the use of salivary proteins and metabolites as caries biomarkers is still in the early stages of research, it holds promise as a non-invasive, cost-effective way to detect and monitor caries [20, 46]. However, it has to be kept in mind that dental caries takes place in the teeth, and presumably the protein and metabolite composition at the dental biofilm itself could be more informative to identify caries-associated biomarkers [15, 18]. Our data indeed show that the concentration of NMR-identified metabolites in plaque and saliva from the same individuals do not correlate. This is likely to be the result of salivary components being provided by all tissues in the oral cavity, and therefore only those compounds exclusively associated to the teeth would correlate between plaque and saliva. This could represent an important drawback for using saliva in caries prediction, as it has already been reported for microbiome studies [47].

Given the importance of biofilm-mediated pH-buffering capacity to prevent caries, it was unexpected that the levels of ammonia (a pH-buffering base produced from urea, arginine or nitrate metabolism) did not vary between caries and caries-free individuals. In other studies, the arginolytic or urease activity in both adults and children has been found to correlate well with caries experience [34, 48]. These studies are however enzymatic, where the biofilm activity in the presence of the substrate is measured after incubation in vitro. Thus, although the metabolic potential of the microbial community to produce ammonia appears to be related to caries risk [34], the salivary levels of its final product (ammonia) at a given time point may not correlate with caries, and the same could apply for nitrite or other metabolites that could play a role in caries prevention only when the substrate is available or when environmental conditions are appropriate, such as under an acidic pH. For example, the discriminatory power of organic acids as caries markers improved after a sugar rinse [20]. Thus, the possibility of using enzymatic assays or to enrich saliva in a given substrate before biomarker quantification to stimulate a given enzymatic reaction associated with the caries process or its prevention should be evaluated in the future. Our data also suggest the previously reported caries biomarkers in adults may not be valid for other age groups. For instance, proteins involved in bacterial adhesion, like statherin or fibronectin, were not found to differ between the two adolescent groups, whereas in other studies they were selected as informative caries biomarkers in the adult population [37, 49]. A similar result was obtained when applying the Salivary Immune and Metabolic Marker (SIMMA) test, whose biomarkers were selected in adults, to children [29].

Despite all the above confounding factors, our data identified several components of saliva with significant differences between caries-free and caries-experienced individuals. These compounds included citrate, ethanolamine, or 2-hydroxyvalerate, as well as several metabolites that could not be unequivocally assigned to known compounds and that should be investigated in the future. The lactate:succinate ratio from the morning samples was also significantly different, with higher values in the caries group. However, all these parameters partially overlapped and therefore not a single compound showed statistical power on its own to discriminate between the two groups. Therefore, we used random forest modeling and biomarker discovery algorithms to identify combinations of compounds that together could improve the ability to differentiate between caries and caries-free adolescents. Our data suggested that a combination of several compounds could significantly improve the prediction, reaching AUC values over 0.9 for a combination of three to five compounds. Thus, despite the difficulties to obtain robust biomarkers due to factors described above (i.e., sex-effects, lack of correlation between saliva and plaque concentrations, variability during the day), our data suggest that a combination of salivary components could still provide a promising strategy for discriminating between caries-active and caries-free individuals. Unfortunately, the NMR profile peaks corresponding to some of these salivary biomarkers could not be unequivocally assigned to known metabolites, and future efforts should be directed towards identifying those compounds.

There are other limitations to this study. First, we are aware that the sample size was small and may not be representative of the population variability. The general rationale of this work was to test a large number of compounds with potential diagnostic value in a limited number of samples and validate the selection of potential biomarkers in a larger population size, ideally from different geographic locations. To predict an appropriate sample size, some power estimates were done before the study based on existing data on salivary pH values and salivary concentrations of IgA in previous studies. Considering an alpha value of 0.05 and 80% power, a sample size of n = 16 per group was calculated to detect a potential pH difference of 0.25 between the two groups (e.g., from pH = 7 to pH = 6.75), and a size of n = 23 to detect a smaller difference of 0.2 points. For IgA, the required sample size for detecting a 10% difference in mean values between caries-experienced and caries-free individuals would be n = 20 with the same power. We therefore used this sample size as a starting point. However, some of the measured metabolites showed larger variability, which considerably increased the desirable sample size. We hope that the current work stimulates further research in the field and the validation of the compounds with more discriminatory signal in larger cohorts, taking into account the sampling conditions (e.g., time of the day). Finally, we have to keep in mind that the association of several salivary components with caries does not mean that those compounds will be able to predict future appearance of the disease. To address these limitations, future studies should incorporate longitudinal analysis and consider a combination of host factors (including diet and oral hygiene), microbial, behavioral and environmental factors for screening precise biomarkers related to caries progression.

In the current work, even after performing an extensive effort to identify potential salivary biomarkers for dental caries, a single compound that discriminates between caries-free and caries-experienced individuals could not be found. However, a combination of several markers did show discriminatory power, especially when samples were collected in the afternoon. Our work also identifies the underlying reasons for this difficulty in finding robust disease markers in adolescents, such as gender differences, lack of saliva-plaque correlations, or differences in oral hygiene and diet, which should be included in caries prediction models in the future.

The authors are grateful to all participants and dental professionals at the Public Dental Service Clinics in Råslätt and Vetlanda, county of Jönköping, Sweden. We also thank Sandra Garcia (FISABIO Foundation) for technical assistance during ELISA essays and quantification of ions.

Ethical approval was obtained from the Regional Ethics Committee for Human Research at Linköping University, Sweden, 2017/599-31. All participants provided written informed consent, and in cases when study participants were under 15 years of age, it was obtained from the parent/legal guardian of participants prior to the study.

The authors have no conflicts of interest to declare.

This project was funded by grants 807631 and 931593 funded by the Medical Research Council of Southeast Sweden; grant 931659 funded by Futurum Academy for Health and Care, Jönköping County Council, and Foresight multidisciplinary program at Malmö University.

K.H. contributed to conception and design, collected samples, performed dental examinations, wrote manuscript writing, and drafted and critically revised the manuscript. M.C.-D. performed data analysis, elaborated figures and performed statistical analysis, and critically revised the manuscript. M.S., H.J., H.I., and C.W. contributed to conception, design, data acquisition and interpretation, and drafted and critically revised the manuscript. A.B.N. and D.M. performed N.M.R. experimental work, contributed to data analysis, statistical analysis, and critically revised the manuscript. E.C. performed multiplex fluorochrome technique, contributed to data analysis, and critically revised the manuscript. A.M. performed experimental work, contributed to conception, design, data acquisition, and interpretation, and drafted the manuscript.

All data generated or analyzed during this study are included in this article and its supplementary material files. Further inquiries regarding raw data can be directed to the corresponding author.