Feasibility of salivary lipid screening for caries biomarkers in young adults Authors: Zhenzhen Li, Kai Zhang, Dong Wang, Jie Hu, Shanshan Liu, Li Xu Published: 2025 ABSTRACT BackgroundTo identify distinct lipid metabolic profiles that differentiate young adults with dental caries (DC) from their caries-free (CF) counterparts and to investigate salivary lipid biomarkers indicative of DC status.MethodsWe analyzed salivary lipid profiles from 100 caries-free (CF) and 100 dental caries (DC) college students aged 17-20 years old. Lipidomic analysis was performed using liquid chromatography-electrospray ionization tandem mass spectrometry (LC-ESI-MS/MS). Data were analyzed using orthogonal partial least squares discriminant analysis (OPLS-DA), variable importance in projection (VIP), and fold change (FC) analysis to identify lipid metabolites that distinguished caries status. Receiver operating characteristic (ROC) curve analysis was performed to compare lipid profiles between groups, assess potential associations with caries, and calculate the area under the curve (AUC), positive predictive value (PPV), and negative predictive value (NPV) for each lipid species.ResultsAlterations in glycerophospholipid and sphingolipid metabolism were evident in participants with caries. Five salivary lipid metabolites, including ceramide (Cer) (d24:1/18:0(2OH)), phosphatidic acid (PA)(21:0_24:1), PA (25:0_20:1), phosphatidylcholine (PC)(16:0_24:0), and sphingosine (SPH)(d18:0), showed significant differences in both DC1 (decayed, missing, and filled teeth (DMFT) = 4-6) and DC2 (DMFT > 6) groups compared with CF controls (P < 0.001). Cer (d24:1/18:0(2OH)), PA (25:0_20:1), PA (21:0_24:1), and PC (16:0_24:0) were upregulated, whereas SPH (d18:0) was downregulated. Although lipid levels tended to be higher in DC2 than in DC1, the differences between these subgroups were not statistically significant. In receiver operating characteristic analysis, PA (25:0_20:1) demonstrated perfect discrimination between CF and DC2 (AUC = 1.000, PPV = 100%, NPV = 100%). For CF versus DC1, PA (25:0_20:1) achieved an AUC of 0.976, a PPV of 100%, and an NPV of 97.37%, while PC (16:0_24:0) yielded an AUC of 0.952, a PPV of 100%, and an NPV of 94.87%.ConclusionSalivary lipidomic profiling enables noninvasive differentiation between DC and CF individuals, highlighting specific lipid biomarkers with potential utility for diagnosis and disease monitoring. These findings could inform personalized prevention and management strategies.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12903-025-07319-x. Introduction Dental caries is a chronic, bacteria-driven infection that damages the hard tissues of the tooth surface and remains one of the most widespread diseases affecting global oral health. Left untreated, it can progress to severe complications such as pulpitis, periapical inflammation, tooth loss, and, in advanced cases, systemic infections including sepsis [1, 2]. Globally, caries constitutes a major public health concern. Data reported in The Lancet show that caries in permanent teeth ranks first in prevalence and third in incidence among 354 diseases assessed worldwide [2]. Despite preventive measures such as fluoride applications and sealant programs [3], the 2015 Fourth National Oral Health Epidemiological Survey reported a significant rise in the dental caries rate for permanent teeth in 12-year-olds, which reached 38.5%. This figure is still far from the target of reducing the caries rate to below 25% by 2030 [4]. These realities highlight the pressing need for reliable, easily obtainable biomarkers to enable caries detection. Such markers could not only support timely treatment but also shed light on the biological pathways underlying disease onset and progression. Streptococci and Lactobacilli have long been identified as primary microbial agents in the development of early childhood caries, dominating the dental plaque biofilm. Through the metabolism of dietary carbohydrates, these bacteria generate organic acids that demineralize tooth enamel and dentin, ultimately leading to carious lesions [5, 6]. Conventional methods of assessing caries risk have largely relied on visible clinical indicators and poor oral hygiene behaviors, such as infrequent tooth brushing or frequent consumption of sugary snacks and beverages [7–10]. Evidence suggests that healthier dietary habits, such as maintaining consistent meal schedules and reducing intake of sugar-sweetened drinks, are associated with a lower incidence of cavities [11]. In recent years, caries risk assessment has evolved with the emergence of advanced analytical approaches, including biomarker identification, oral microbiome profiling, and metabolomic analysis [12–14]. Metabolites such as histamine, L-histidine, and succinate are significantly elevated in the saliva of individuals with caries, implicating them in the metabolic processes that may influence disease progression [12]. Yet, the exact biological roles these metabolites play remain incompletely defined. Importantly, children with active carious lesions show salivary metabolomic patterns distinct from those of healthy peers, marked by pronounced shifts in carbohydrate and amino acid concentrations. Elevated taurine and mannose levels in particular have emerged as promising biomarkers for assessing caries risk, reinforcing the potential of salivary metabolomics as a foundation for personalized preventive strategies [13]. Taken together, these findings expand the understanding of caries pathogenesis and provide a basis for developing precision-based prevention and management approaches tailored to individual risk profiles. Dietary patterns, such as maintaining regular meal schedules and limiting the intake of sugary drinks, are associated with a reduced risk of cavities [11]. IFurthermore, several investigations have reported a significant negative correlation between salivary lipid levels (particularly triglycerides and phospholipids) and the development of caries in children [15]. These findings suggest that salivary lipids play a key role in the pathological processes underlying caries and serve as valuable indicators for exploring their biological mechanisms. Saliva, a biofluid in direct contact with the oral cavity, has recently gained increasing attention for its potential in disease prediction and early diagnosis. Its ease of noninvasive collection, affordability, and prolonged biochemical stability make saliva a compelling alternative to blood or tissue sampling. Importantly, it has been effectively utilized to track systemic metabolic disorders such as diabetes and obesity, as well as autoimmune conditions, including rheumatoid arthritis and systemic lupus erythematosus [16–19]. Lipidomics, an essential branch of metabolomics, covers a remarkably broad spectrum of structurally diverse biomolecules and allows for the quantitative measurement of individual lipids, their subclasses, and specific molecular species. This analytical strategy supports the detection of metabolic variations and the reconstruction of biological pathways [20]. To the best of our knowledge, no previous research has employed lipidomics to investigate salivary metabolic differences between healthy young adults and those with caries. In the present study, we examined a homogeneous cohort of university students that included both healthy participants and individuals with caries. Saliva samples were analyzed using ultra-high-performance liquid chromatography coupled with liquid chromatography–electrospray ionization tandem mass spectrometry (LC-ESI-MS/MS) to perform comprehensive targeted lipidomic profiling. The primary objectives were to identify distinct lipid metabolic patterns that distinguish caries-affected individuals from their healthy peers and to investigate salivary lipid biomarkers indicative of caries status. The outcomes of this work aim to establish a robust scientific basis for predicting caries risk, supporting early intervention, and guiding the development of targeted prevention strategies. Methods Study population This study was conducted in strict accordance with the Declaration of Helsinki and approved by the Ethics Committee of Bengbu Medical University (Approval No. [2021] KY172). All participants were aged 17–20 years, and they were enrolled in 2021 at Bengbu Medical University. All participants provided their written informed consent before their participation. The subject exclusion criteria included dental fluorosis, recurrent oral ulcers, periodontitis, other oral diseases, systemic or genetic disorders, long-term use of prescription antibiotics, smoking or alcohol consumption within the past 6 months, and an inability to cooperate with oral examinations or saliva collection. To ensure methodological rigor, the minimum sample size was calculated by using a prevalence-based formula, with reference to data from the Fourth National Oral Health Survey in China on the prevalence of permanent teeth caries in a 15-year-old [4]. The formula used was as follows:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N\;=\;\left(Z_{\alpha/2}/\delta\right)\ast2\ast\;P\ast\left(1-P\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha\;=\;0.05,\;Z_{\alpha/2}\;=\;1.96,\;\delta;=\;5\%,\;P\;=\;44.42\%$$\end{document} The minimum sample size was calculated to be 385. Clinical examination and questionnaire Oral examinations were conducted at the clinical training center of the School of Stomatology, Bengbu Medical University, under the supervision of experienced clinicians from the Department of Stomatology at the First Affiliated Hospital. Each examination was conducted by a calibrated team comprising one examiner and one recorder. A total of 10 examiners participated in the study. Before the survey, the examiners underwent standardized consistency training, achieving a Kappa > 0.8, which ensured high intra- and inter-examiner reliability. Standardized equipment, including dental chairs, disposable mouth mirrors, and Community Periodontal Index (CPI) probes (WHO-probe, LDM, Ltd., Tokyo, Japan), was used in accordance with the World Health Organization (WHO) protocols. Dental caries were diagnosed based on the WHO criteria for epidemiological studies, with reference to the International Caries Detection and Assessment System II (ICDAS-II) to enhance diagnostic reliability and international comparability. Diagnostic criteria included visible cavitated lesions on the pits, fissures, or smooth surfaces; softened or demineralized tissues at the lesion base or walls; and translucent black shadows beneath the enamel [21]. Caries severity was assessed by using the decayed, missing, and filled teeth (DMFT) index. A total of 971 patients aged 17–20 years participated in the clinical survey, surpassing the minimum sample size required for cross-sectional, population-based studies. A questionnaire was then administered to the participants to gather information on their parental education level, use of fluoride toothpaste, toothbrushing and flossing frequency, and sugar intake frequency, with the aim of controlling the potential confounders affecting caries prevalence [Appendix Table1] [22, 23]. Saliva collection and sample preparation All participants were instructed to refrain from eating or drinking for at least 2 h before their saliva collection. Unstimulated whole saliva samples were then collected in the morning under fasting conditions using the passive drool method. Immediately after collection, the samples were placed on an ice bath and transported to the laboratory within 30 min for processing. All specimens were stored at −80 °C until further analyses. For lipidomic extraction, samples were thawed on ice until they had fully liquefied. A 200 µL aliquot of the saliva was then mixed with 1 mL of lipid-extraction solvent (methyl tert-butyl ether: methanol = 3:1, v/v) and vortexed thoroughly. After adding 100 µL of purified water, the mixture was centrifuged at 12,000 rpm for 10 min at 4 °C. The upper organic phase (500 µL) was carefully collected and vacuum-dried. For LC-ESI-MS/MS analysis, the dried extract was reconstituted in 200 µL of the mobile phase B, vortexed, and re-centrifuged at 12,000 rpm for 3 min at 4 °C. The resulting supernatant was then transferred into vials for further analysis. Lipid extraction and LC-MS/MS analysis The chromatographic analysis was performed on an Ultra performance liquid chromatography (UPLC) column Thermo AccucoreTM C30 at a flow rate of 0.35 ml/min and an injection volume of 2 ul. The mobile phase system consisted of Solvent A (60% acetonitrile, 40% water, 10 mM ammonium formate) and Solvent B (90% isopropanol, 10% acetonitrile, 10 mM ammonium formate). The gradient elution program was as follows: 0.0 min (80:20, A: B); 2.0 min (70:30); 4.0 min (40:60); 9.0 min (15:85); 14.0 min (10:90); 15.0–17.3 min (5:95); and 17.5–20.0 min (80:20) for re-equilibration. Mass spectrometry was performed in both the positive and negative electrospray ionization (ESI) modes by using the multiple reaction monitoring (MRM) acquisition method under the following ion source parameters: source temperature at 500 °C, IonSpray Voltage at + 5500 V/−4500 V, curtain gas (CUR) at 35 psi, Gas 1 (GS1) at 45 psi, and Gas 2 (GS2) at 55 psi. Instrument tuning and mass calibration were performed using 10 µmol/L and 100 µmol/L polypropylene glycol solutions in QQQ and LIT modes, respectively. Quality control (QC) samples, prepared by pooling equal aliquots from all individual samples, were inserted at regular intervals (every 10 samples) throughout the analytical sequence so as to assess the stability of the instrument and its analytical reproducibility. Quality control and data processing In this study, qualitative lipid identification was performed using an in-house MetWare Database (MWDB), which integrates retention time (RT) and precursor–product ion pairs specific to individual lipid species. Quantitative analysis was performed on a QTRAP triple quadrupole mass spectrometer operating in Multiple Reaction Monitoring (MRM) mode. Raw data were processed using Analyst 1.6.3 software, and peak area integration and correction were performed with MultiQuant to ensure data accuracy and consistency. To evaluate technical reproducibility, pooled QC samples were prepared by mixing equal volumes (10 µL) from all experimental samples. QC samples were then injected at regular intervals (every 10 samples) throughout the LC-ESI-MS/MS run. Total ion chromatograms (TICs) of the QC samples showed high overlap, indicating a strong signal stability and low instrumental variability. Blank samples were interspersed throughout the run, with only minimal signals from internal standards observed, implying negligible carryover or cross-contamination. Known concentrations of internal standards were spiked into QC samples, and their response coefficients were monitored. A coefficient of variation (CV) ≤ 15% among internal standards was considered indicative of high detection stability. Lipid species with a CV of > 50% across QC replicates in either positive or negative ionization mode were excluded from subsequent analyses so as to improve the robustness. Before differential analysis, principal component analysis (PCA) was performed to assess the within-group variance and between-group separation. Orthogonal Partial Least Squares Discriminant Analysis (OPLS-DA), combined with Orthogonal Signal Correction (OSC) and Partial Least Squares Discriminant Analysis (PLS-DA), was used for supervised classification and biomarker selection. OPLS-DA decomposes the X matrix (quantitative metabolite data) into Y-predictive and Y-orthogonal components, thereby maximizing group separation and facilitating the identification of discriminative lipids. Before modeling, data were log2-transformed and mean-centered. The OPLS-DA model was implemented using the OPLSR. Anal function from the MetaboAnalyst R package in R. Model evaluation parameters included R²X (variance explained in X), R²Y (variance explained in Y), and Q² (predictive ability). A Q² value > 0.5 indicates a valid model, whereas a Q² >0.9 reflects excellent predictive performance. In this study, model robustness was further validated through 1000 permutation tests. Statistical and bioinformatics analysis SPSS version 20.0 was used to assess associations between dental caries and potential influencing factors, including gender, parental education level, fluoride use, tooth brushing frequency, flossing frequency, and sugar intake frequency. Bivariate logistic regression models were applied to evaluate these relationships. For lipidomics data, variable importance in projection (VIP) scores derived from the OPLS-DA model were initially used to identify candidate differential lipids between the groups. Subsequent univariate analyses, including fold change (FC), P-values, and false discovery rate (FDR) correction, were performed to assess any statistical significance. Lipids were considered significantly altered if they met all of the following criteria: VIP > 1.0, FC > 1.8 or < 0.56, and FDR < 0.05. To visualize differences in the lipid abundance and significance among groups, volcano plots and box plots were generated. Significantly altered metabolites were annotated using the KEGG Compound Database (https://www.kegg.jp/kegg/compound/) and mapped to relevant biological pathways via the KEGG Pathway Database (https://www.kegg.jp/kegg/pathway.html). Pathways demonstrating significant regulatory impact (P < 0.05) were further analyzed using metabolite set enrichment analysis (MSEA), with significance determined by hypergeometric testing. Finally, GraphPad Prism (version 8.0.2) was employed to construct the trend plots and receiver operating characteristic (ROC) curves. The area under the curve (AUC), positive predictive value (PPV), and negative predictive value (NPV) were calculated for each candidate biomarker for the evaluation of its diagnostic performance. Result In total, 971 college students were initially screened for oral examination, of whom 957 completed the full questionnaire and were included in the final analysis. Among these participants, 637 (66.6%) were diagnosed with caries. Logistic regression analysis indicated that female students had significantly greater odds of developing caries than males (odds ratio = 2.556; P < 0.001). No statistically significant associations were observed between DC status and other demographic or behavioral variables. For lipidomic profiling, 100 individuals with a DMFT index greater than 3 were assigned to the DC group. Of these, 62 participants were classified as DC1 (DMFT = 4–6) and 38 as DC2 (DMFT > 6). An additional 100 caries-free (CF) participants were enrolled as controls. A schematic representation of the study design and analytical workflow is shown in Fig. 1. Fig. 1A workflow depiction of the current study protocol A workflow depiction of the current study protocol Quality assessment and classification of salivary lipid metabolites Broad-targeted salivary lipidomic analysis was conducted on the DC and CF groups using LC-ESI-MS/MS. Total ion chromatograms from pooled QC samples showed highly consistent retention times and peak intensities across runs, confirming excellent instrument stability throughout data acquisition. Across the 200 samples analyzed, 906 lipid metabolites were identified. The CV for all metabolites in QC samples was below 0.5, and 888 lipids with CV values under 0.3 were retained for subsequent statistical and biological evaluation. Based on chemical structure and biosynthetic pathways defined by the MAPS lipid database, the detected metabolites were categorized into six major classes: glycerophospholipids, glycerolipids, sphingolipids (SP), prenol lipids, sterol lipids, and fatty acyls. Among these, glycerophospholipids comprised 47.07%, glycerolipids 22.97%, and SP 20.80%, reflecting the predominance of membrane-associated lipid species in human saliva (Table 1).Table 1The saliva targeted lipidomics, 888 lipid metabolitesCLASS ⅠN%CLASS ⅡN%FA(Fatty acid)566.31CAR(Acylcarnitine)2442.86Eicosanoid47.14FFA(Free fatty acid)2850.00GL(Glycerol)20422.97DG(Diacylglycerol)2612.75DG-O(Ether-linked diacylglycerol)31.47DGDG(Digalactosyldiacylglycerol)73.43MG(Monoacylglycerol)31.47MGDG(Monogalactosyldiacylglycerol)52.45TG(Triacylglycerol)16078.43GP(Glycerolipids)41847.07LNAPE(Diacylglycerophosphoethanolamines)61.44LPA(Lysophosphatidic acid)10.24LPC(Lysophophatidylcholine)174.07LPC-O(alkyl-lysophophatidylcholine)102.39LPE(Lysophosphatidylethanolamine)133.11LPE-P(alkenyl-lysophosphatidylethanolamine)61.44LPG(Lysophosphatidylglycerol)20.48LPI(Lysophosphatidylinositol)10.24LPS(Lysophosphatidylserine)51.20PA(Phosphatidic acid)276.46PC(Phosphatidylcholine)5813.88PC-O(alkyl-glycerophosphocholines)245.74PE(Phosphatidylethanolamine)409.57PE-O(alkyl-glycerophosphoethanolamine)5212.44PE-P(alkenyl-glycerophosphoethanolamine)5112.20PG(Phosphatidylglycerol)368.61PI(Phosphatidylinositol)255.98PS(Phosphatidylserine)4510.77PR(Prenol lipids)20.23CoQ(Coenzyme Q)2100.00SP(Sphingolipid)18220.5Cer-ADS(Ceramide)10.55Cer-AP(Ceramide)84.40Cer-AS(Ceramide)189.89Cer-NDS(Ceramide)126.59Cer-NP(Ceramide)158.24Cer-NS(Ceramide)4725.82CerP(Ceramide 1-phosphates)42.20Hex2Cer(Hexosylceramide)42.20HexCer-AP(Hexosylceramide)2010.99HexCer-NS(Hexosylceramide)116.04SHeCer21.10SM(Sphingomyelin)3720.33SPH(sphingosine)31.65ST(Sterol lipid)111.24BA(Bile Acid)763.64CE(Cholesteryl ester)327.27Cho(Cholesterol)19.09TG(Triacylglycerol)151.69TG(Triacylglycerol)15100.00 Table 1The saliva targeted lipidomics, 888 lipid metabolitesCLASS ⅠN%CLASS ⅡN%FA(Fatty acid)566.31CAR(Acylcarnitine)2442.86Eicosanoid47.14FFA(Free fatty acid)2850.00GL(Glycerol)20422.97DG(Diacylglycerol)2612.75DG-O(Ether-linked diacylglycerol)31.47DGDG(Digalactosyldiacylglycerol)73.43MG(Monoacylglycerol)31.47MGDG(Monogalactosyldiacylglycerol)52.45TG(Triacylglycerol)16078.43GP(Glycerolipids)41847.07LNAPE(Diacylglycerophosphoethanolamines)61.44LPA(Lysophosphatidic acid)10.24LPC(Lysophophatidylcholine)174.07LPC-O(alkyl-lysophophatidylcholine)102.39LPE(Lysophosphatidylethanolamine)133.11LPE-P(alkenyl-lysophosphatidylethanolamine)61.44LPG(Lysophosphatidylglycerol)20.48LPI(Lysophosphatidylinositol)10.24LPS(Lysophosphatidylserine)51.20PA(Phosphatidic acid)276.46PC(Phosphatidylcholine)5813.88PC-O(alkyl-glycerophosphocholines)245.74PE(Phosphatidylethanolamine)409.57PE-O(alkyl-glycerophosphoethanolamine)5212.44PE-P(alkenyl-glycerophosphoethanolamine)5112.20PG(Phosphatidylglycerol)368.61PI(Phosphatidylinositol)255.98PS(Phosphatidylserine)4510.77PR(Prenol lipids)20.23CoQ(Coenzyme Q)2100.00SP(Sphingolipid)18220.5Cer-ADS(Ceramide)10.55Cer-AP(Ceramide)84.40Cer-AS(Ceramide)189.89Cer-NDS(Ceramide)126.59Cer-NP(Ceramide)158.24Cer-NS(Ceramide)4725.82CerP(Ceramide 1-phosphates)42.20Hex2Cer(Hexosylceramide)42.20HexCer-AP(Hexosylceramide)2010.99HexCer-NS(Hexosylceramide)116.04SHeCer21.10SM(Sphingomyelin)3720.33SPH(sphingosine)31.65ST(Sterol lipid)111.24BA(Bile Acid)763.64CE(Cholesteryl ester)327.27Cho(Cholesterol)19.09TG(Triacylglycerol)151.69TG(Triacylglycerol)15100.00 The saliva targeted lipidomics, 888 lipid metabolites Multivariate analysis and identification of differential lipid metabolites OPLS-DA was applied to evaluate global lipidomic differences between the DC and CF groups. This supervised modeling technique efficiently filters out noninformative variation, improving group separation and facilitating biomarker identification. The OPLS-DA score plot demonstrated distinct separation between DC and CF samples, indicating pronounced alterations in salivary lipid composition (Fig. 2a). Model performance was evaluated using standard parameters: R²X and R²Y indicated explanatory capacity for predictor and response variables, respectively, while Q² reflected predictive accuracy. A 1000-permutation test for the DC versus CF model confirmed statistical robustness (P < 0.001), supporting the model’s reliability and validity (Fig. 2b).Fig. 2Lipid Profiling and Screening Differences between Dental Caries (DC) and Caries-Free (CF) Groups. a OPLS-DA Model: OPLS-DA analysis differentiates DC and CF groups based on their lipidomic profiles, showing clear separation between the groups. b OPLS-DA Model Validation: Cross-validation of the OPLS-DA model confirms its reliability, with R² and Q² values indicating a good fit and predictive power. c Differential Lipid Volcano Plot: Volcano plot displaying significant lipid differences between DC and CF groups. The X-axis represents log₂ fold change, and the Y-axis represents -log₁₀ P-values. d Comparison of the relative abundance of 13 differential metabolites: Bar graph comparing the relative abundance of 13 significantly altered lipid metabolites between the two groups Lipid Profiling and Screening Differences between Dental Caries (DC) and Caries-Free (CF) Groups. a OPLS-DA Model: OPLS-DA analysis differentiates DC and CF groups based on their lipidomic profiles, showing clear separation between the groups. b OPLS-DA Model Validation: Cross-validation of the OPLS-DA model confirms its reliability, with R² and Q² values indicating a good fit and predictive power. c Differential Lipid Volcano Plot: Volcano plot displaying significant lipid differences between DC and CF groups. The X-axis represents log₂ fold change, and the Y-axis represents -log₁₀ P-values. d Comparison of the relative abundance of 13 differential metabolites: Bar graph comparing the relative abundance of 13 significantly altered lipid metabolites between the two groups Differential lipid metabolites were identified using combined selection thresholds: VIP > 1.0 (from OPLS-DA), FC > 1.8 or < 0.56, and FDR < 0.05. Student’s t-tests subsequently revealed 13 significantly altered lipid species, of which 12 were upregulated and 1 was downregulated (Fig. 2c, d). Diagnostic evaluation of differential lipids and group discrimination by OPLS-DA To evaluate the diagnostic potential of individual lipid metabolites, Student’s t-tests were combined with ROC curve analysis. ROC curves were generated for each of the 13 significantly altered lipids to assess their ability to distinguish between the DC and CF groups. Several metabolites exhibited excellent diagnostic performance, with AUC values above 0.97 (Table 2).Table 2Statistics of 13 lipid metabolites distinguish DC from CFCompoundsVIPFDRFCLog2FCAUC95%CITypeCarnitine C18-OH5.488< 0.0011756.80210.7790.9800.961–0.999upCer(d24:1/18:0(2OH))2.926< 0.0012.2561.1740.8060.745–0.868upLPS(24:1/0:0)5.407< 0.0011284.52510.3270.9700.947–0.993upPA(18:0_23:0)5.573< 0.0011102.15310.1060.9850.968–1.000upPA(21:0_24:1)5.200< 0.001980.5779.9370.9350.902–0.968upPA(25:0_20:1)5.591< 0.001590.9609.2070.9850.968–1.000upPC(16:0_24:0)5.345< 0.0011211.68610.2430.9550.927–0.983upPC(24:0_16:1)4.985< 0.001924.3519.8520.9050.866–0.944upPE(22:0_18:1)5.186< 0.001989.8939.9510.9300.896–0.964upPE(O-18:2_24:2)5.272< 0.0011140.44210.1550.9450.914–0.976upPE(O-20:1_20:1)5.435< 0.001992.2559.9550.9700.947–0.993upPS(21:0_24:1)5.209< 0.001777.6699.6030.9350.902–0.968upSPH(d18:0)4.159< 0.0010.554−0.8510.8790.830–0.928downAUC Area Under the Curve, represents the classification performance of a lipid in distinguishing between the experimental and control groups, VIP Variable Importance in Projection, reflects the contribution of a variable (lipid) to the predictive ability of the model, FDR False Discovery Rate, the proportion of false positives among the results identified as significant, CI Confidence Interval, FC Fold Change, refers to the ratio of lipid abundance between two groups Table 2Statistics of 13 lipid metabolites distinguish DC from CFCompoundsVIPFDRFCLog2FCAUC95%CITypeCarnitine C18-OH5.488< 0.0011756.80210.7790.9800.961–0.999upCer(d24:1/18:0(2OH))2.926< 0.0012.2561.1740.8060.745–0.868upLPS(24:1/0:0)5.407< 0.0011284.52510.3270.9700.947–0.993upPA(18:0_23:0)5.573< 0.0011102.15310.1060.9850.968–1.000upPA(21:0_24:1)5.200< 0.001980.5779.9370.9350.902–0.968upPA(25:0_20:1)5.591< 0.001590.9609.2070.9850.968–1.000upPC(16:0_24:0)5.345< 0.0011211.68610.2430.9550.927–0.983upPC(24:0_16:1)4.985< 0.001924.3519.8520.9050.866–0.944upPE(22:0_18:1)5.186< 0.001989.8939.9510.9300.896–0.964upPE(O-18:2_24:2)5.272< 0.0011140.44210.1550.9450.914–0.976upPE(O-20:1_20:1)5.435< 0.001992.2559.9550.9700.947–0.993upPS(21:0_24:1)5.209< 0.001777.6699.6030.9350.902–0.968upSPH(d18:0)4.159< 0.0010.554−0.8510.8790.830–0.928downAUC Area Under the Curve, represents the classification performance of a lipid in distinguishing between the experimental and control groups, VIP Variable Importance in Projection, reflects the contribution of a variable (lipid) to the predictive ability of the model, FDR False Discovery Rate, the proportion of false positives among the results identified as significant, CI Confidence Interval, FC Fold Change, refers to the ratio of lipid abundance between two groups Statistics of 13 lipid metabolites distinguish DC from CF AUC Area Under the Curve, represents the classification performance of a lipid in distinguishing between the experimental and control groups, VIP Variable Importance in Projection, reflects the contribution of a variable (lipid) to the predictive ability of the model, FDR False Discovery Rate, the proportion of false positives among the results identified as significant, CI Confidence Interval, FC Fold Change, refers to the ratio of lipid abundance between two groups To investigate subgroup-specific lipidomic variations, additional OPLS-DA models were constructed for the comparisons DC1 versus CF, DC2 versus CF, and DC1 versus DC2. The DC1 versus CF and DC2 versus CF models demonstrated clear group separation and strong explanatory and predictive capacity (P < 0.005). In contrast, the DC1 versus DC2 model showed limited discrimination (P > 0.005), indicating relatively minor metabolic differences between moderate and severe DC subgroups. Further univariate analyses comparing CF with both DC1 and DC2 identified five lipid metabolites with statistically significant differences, which were classified as robust markers potentially implicated in DC progression (Fig. 3).Fig. 3OPLS-DA Model and Validation for Lipid Screening Across Groups. a–f The OPLS-DA model and OPLS-DA model validation were used to identify differentially changing lipids between the groups. g Venn Diagram: The Venn diagram shows the overlap of differentially expressed lipid metabolites among the DC1, DC2, and CF groups, highlighting the unique and shared lipid features across the groups OPLS-DA Model and Validation for Lipid Screening Across Groups. a–f The OPLS-DA model and OPLS-DA model validation were used to identify differentially changing lipids between the groups. g Venn Diagram: The Venn diagram shows the overlap of differentially expressed lipid metabolites among the DC1, DC2, and CF groups, highlighting the unique and shared lipid features across the groups Comparative abundance of key lipid metabolites across groups The relative concentrations of five lipid metabolites were compared among the CF, DC1, and DC2 groups to assess their relationship with DC severity (Table 3). The five representative lipids, Cer (d24:1/18:0(2OH)), PA (21:0_24:1), PA (25:0_20:1), PC (16:0_24:0), and SPH (d18:0), were all significantly altered in individuals with DC. Compared with the CF group, all five lipids showed highly significant differences in both DC1 and DC2 participants (P < 0.001), underscoring their strong association with DC onset and progression.Table 35 differential lipid metabolites among DC1, DC2, and CFCompoundsRatiosDC/CFDC1/CFDC2/CFDC2/DC1Cer(d24:1/18:0(2OH))2.2562.3372.1190.907PA(21:0_24:1)980.577932.6571062.1721.139PA(25:0_20:1)590.960574.932618.2501.075PC(16:0_24:0)1211.6861079.9461436.0001.330SPH(d18:0)0.5540.5570.5490.985 Table 35 differential lipid metabolites among DC1, DC2, and CFCompoundsRatiosDC/CFDC1/CFDC2/CFDC2/DC1Cer(d24:1/18:0(2OH))2.2562.3372.1190.907PA(21:0_24:1)980.577932.6571062.1721.139PA(25:0_20:1)590.960574.932618.2501.075PC(16:0_24:0)1211.6861079.9461436.0001.330SPH(d18:0)0.5540.5570.5490.985 5 differential lipid metabolites among DC1, DC2, and CF However, no statistically significant differences were observed between the DC1 and DC2 subgroups, suggesting that while these metabolites are closely linked to the presence of DC, they may not serve as reliable indicators of disease severity. Notably, SPH (d18:0) was the only metabolite found to be significantly elevated in the CF group compared with both DC groups, indicating a possible protective or inhibitory role in the development of caries (Fig. 4).Fig. 4Differential Lipid Abundance between CF, DC1, and DC2 Groups. The figure demonstrates the relative abundance of five differential lipids across caries-free (CF), DC1(4 < decayed, missing, and filled teeth (DMFT) index ≤ 6), and DC2 (decayed, missing, and filled teeth (DMFT) index > 6) groups. ***P < 0.001, indicates a highly significant difference, while ns indicates no significant difference between the groups. (a) Cer (d24:1/18:0(2OH)), (b) PA (25:0_20:1), (c) PA (21:0_24:1), (d) PC (16:0_24:0), (e) SPH (d18:0). Differential Lipid Abundance between CF, DC1, and DC2 Groups. The figure demonstrates the relative abundance of five differential lipids across caries-free (CF), DC1(4 < decayed, missing, and filled teeth (DMFT) index ≤ 6), and DC2 (decayed, missing, and filled teeth (DMFT) index > 6) groups. ***P < 0.001, indicates a highly significant difference, while ns indicates no significant difference between the groups. (a) Cer (d24:1/18:0(2OH)), (b) PA (25:0_20:1), (c) PA (21:0_24:1), (d) PC (16:0_24:0), (e) SPH (d18:0). KEGG pathway enrichment analysis KEGG pathway enrichment analysis was conducted to identify whether the metabolic and signaling pathways are significantly affected by caries. Comparative assessments were performed for DC versus CF, DC1 versus CF, and DC2 versus CF using a combination of pathway enrichment and topological analysis. Significantly enriched pathways were selected according to P-values, and the number of mapped differential metabolites, with thresholds of P < 0.05, adjusted P < 0.10. Differential lipid metabolites were mapped to several KEGG pathways, revealing notable enrichment in pancreatic cancer, Fc gamma R-mediated phagocytosis, gonadotropin-releasing hormone signaling, cyclic adenosine monophosphate (cAMP) signaling, phospholipase D signaling, pathways in cancer, the phosphatidylinositol signaling system, and choline metabolism in cancer (Fig. 5). Among these pathways, the phosphatidylinositol signaling system, a central regulator of lipid metabolism and membrane restructuring, drew particular attention owing to its direct role in controlling phosphatidic acid (PA) and phosphatidylcholine (PC) levels. Fig. 5Differential Lipid KEGG Pathway Enrichment Analysis (P < 0.05) An illustration of the KEGG pathway enrichment analysis for differentially expressed lipids in the (a) DC vs. CF, (b) DC1 vs. CF, and (c) DC2 vs. CF comparisons. Bar plots exhibiting the top enriched KEGG pathways, with the percentage of differential lipids in each pathway indicated along the X-axis. The bubble plots to the right show the Rich Factor (X-axis) and P-value (color scale) for each enriched pathway. The size of each bubble represents the count of differential lipids involved in that pathway. P < 0.05 indicates statistically significant enrichment of the pathway in the respective group comparison Differential Lipid KEGG Pathway Enrichment Analysis (P < 0.05) An illustration of the KEGG pathway enrichment analysis for differentially expressed lipids in the (a) DC vs. CF, (b) DC1 vs. CF, and (c) DC2 vs. CF comparisons. Bar plots exhibiting the top enriched KEGG pathways, with the percentage of differential lipids in each pathway indicated along the X-axis. The bubble plots to the right show the Rich Factor (X-axis) and P-value (color scale) for each enriched pathway. The size of each bubble represents the count of differential lipids involved in that pathway. P < 0.05 indicates statistically significant enrichment of the pathway in the respective group comparison Diagnostic performance of lipid biomarkers in differentiating CF and DC individuals The findings provide a detailed assessment of differential lipid biomarkers distinguishing the CF and DC groups, with particular emphasis on the DC1 and DC2 subgroups. The data highlight promising lipidomic profiles capable of reliably differentiating individuals according to their caries status. Figures 6a and b present the ROC curves for the CF versus DC1 and CF versus DC2 comparisons, respectively. The analyses reveal that several lipid species, notably PA (25:0_20:1), PA (21:0_24:1), and PC (16:0_24:0), possess high sensitivity and specificity, underscoring their potential as robust biomarkers for DC detection. In the CF versus DC1 analysis (Fig. 6a), PA (25:0_20:1) and PC (16:0_24:0) demonstrate excellent diagnostic accuracy, with AUC values of 0.976 and 0.952, respectively, confirming their ability to distinguish between CF individuals and those with moderate disease. Similarly, in the CF versus DC2 comparison (Fig. 6b), PA (25:0_20:1) achieves perfect discrimination (AUC = 1), while PA (21:0_24:1) and PC (16:0_24:0) each register AUCs of 0.959, indicating strong predictive capacity for severe DC.Fig. 6Differential Lipid Metabolites and Receiver Operating Characteristic (ROC) Curve Analysis. The figure presents the relative expressions and diagnostic performance of five differential metabolites among the CF, DC1, and DC2 groups. (a) Receiver operating characteristic (ROC) curves showing the sensitivity and 1 – specificity of five lipid species for discriminating between CF and DC1. (b) Receiver operating characteristic (ROC) curves showing the sensitivity and 1 – specificity of five lipid species for discriminating between CF and DC2. (c) Performance metrics of the selected lipids. The corresponding AUC values, sensitivity, specificity, and predictive values for each lipid are detailed in the table. Differential Lipid Metabolites and Receiver Operating Characteristic (ROC) Curve Analysis. The figure presents the relative expressions and diagnostic performance of five differential metabolites among the CF, DC1, and DC2 groups. (a) Receiver operating characteristic (ROC) curves showing the sensitivity and 1 – specificity of five lipid species for discriminating between CF and DC1. (b) Receiver operating characteristic (ROC) curves showing the sensitivity and 1 – specificity of five lipid species for discriminating between CF and DC2. (c) Performance metrics of the selected lipids. The corresponding AUC values, sensitivity, specificity, and predictive values for each lipid are detailed in the table. Figure 6 also summarizes the diagnostic performance of these lipid biomarkers, including sensitivity, specificity, NPV, and PPV. PA (25:0_20:1) demonstrates exceptional sensitivity (100%) and specificity (100%) in both DC1 and DC2 comparisons, with flawless classification in the CF versus DC2 analysis. Likewise, PA (21:0_24:1) and PC (16:0_24:0) consistently yield high specificity and predictive strength, validating their diagnostic reliability for both moderate and severe cases. Although SPH (d18:0) is less discriminatory in the CF versus DC1 comparison, it still exhibits notable diagnostic utility, achieving 98.41% specificity and an NPV of 85.68% in this subgroup. Discussion Advances such as whole metagenome shotgun sequencing, high-throughput sequencing, and other molecular technologies have enabled more detailed insights into the pathogenesis and progression of caries [24]. Our previous studies have revealed differences in the salivary metabolites between individuals with and without caries [25]. However, to date, no research has employed high-throughput sequencing methods to analyze the lipidomic characteristics of saliva in caries-affected populations. The specific role of oral lipid metabolism in individuals with caries remains unclear. In the present study, LC-ESI-MS/MS combined with a high-throughput, broad-targeted lipidomics approach was applied to systematically analyze unstimulated saliva samples from 200 young adults with and without caries. Comparative analysis identified five lipid metabolites, PA (21:0_24:1), PA (25:0_20:1), PC (16:0_24:0), Cer (d24:1/18:0(2OH)), and SPH (d18:0), that showed strong associations with caries-related biomarkers. These findings suggest that lipid metabolism in DC-affected young individuals is dysregulated, implicating specific metabolic pathways in the initiation and progression of caries. To our knowledge, this is the first systematic characterization of the salivary lipidome in young individuals with caries using a lipidomics-based approach. The results provide novel insights into the metabolic underpinnings of caries and identify lipid metabolites with potential as biomarkers for early detection, offering valuable implications for clinical management and public health. PA and PC are not only essential components of bacterial cell membranes but also intermediates in phospholipid metabolism, typically formed from membrane phospholipids through the catalytic activities of phospholipase and choline kinase [26]. Bacterial metabolism produces organic acids that demineralize tooth hard tissues, leading to the formation of carious lesions. As early as 1982, Slomiany et al. compared the salivary lipid composition of the parotid and submandibular glands in adults with high caries susceptibility (DMFS >15) and those resistant to DC (DMFS = 0) [27]. They found significantly higher levels of total lipids, triglycerides, neutral lipids, and phospholipids in the saliva of caries-susceptible individuals compared with caries-resistant ones (P < 0.001), suggesting a strong association between altered salivary lipid metabolism and caries susceptibility. More recently, Sharma et al. (2019) reported that Indian children with early childhood caries had significantly higher salivary concentrations of triglycerides and total lipids than caries-free children, with triglyceride levels positively correlated with DMFT indices (P < 0.001) [28]. Despite the small sample sizes of those studies, these findings are consistent with the overall trend of our results and suggest that alterations in salivary lipid composition may emerge in the early stages of caries and can already be observed during early childhood, offering preliminary support for incorporating lipidomics into caries-associated prediction. Changes in the lipid content of saliva also affect the role of cariogenic bacteria in the process of caries formation. Abnormally elevated salivary lipid levels are often accompanied by a significant decrease in saliva flow, reducing oral self-cleaning ability and promoting a cariogenic microenvironment, while the increased lipid content may accumulate in dental plaque, enhancing its hydrophobicity and thereby strengthening and prolonging its adhesion to the tooth surface [29, 30]. Moreover, metabolic byproducts such as lactic acid can be trapped within the plaque matrix, further enhancing bacterial cariogenic potential [31]. The lipid composition of bacterial cell membranes can also influence the synthesis and activity of glucosyltransferase, a key enzyme in biofilm formation. Alterations in membrane phospholipids and fatty acids may regulate glucosyltransferase expression and function, thereby modulating the cariogenicity of Streptococcus mutans [32]. Bacteria produce organic acids that demineralize tooth hard tissues, ultimately leading to the formation of carious lesions [33]. The development of caries involves microbial dysbiosis, chronic local inflammation, and heightened bacterial metabolism [34, 35], all of which may significantly influence the host’s lipid metabolic pathways. In this study, the elevated levels of PA and PC observed in caries-affected individuals may reflect local inflammatory responses and could also indicate systemic disturbances in lipid metabolism associated with inflammation and metabolic imbalance. Previous research has shown that serum PC concentrations are generally increased in conditions such as nephrotic syndrome [36]and nonalcoholic fatty liver disease [37], where they are linked to insulin resistance, lipid accumulation, and systemic inflammation. In contrast, decreased PC levels have been reported in ulcerative colitis [38], Alzheimer’s disease [39], Parkinson’s disease [40], and coronary atherosclerotic heart disease [41, 42]. These divergent patterns may reflect differences in disease stage, tissue type, and variability in inflammatory–immune responses. It is therefore plausible that the salivary lipid profile is closely connected to systemic metabolic and inflammatory status, and that marked increases in PA and PC could serve as early, noninvasive indicators of metabolic-inflammatory dysregulation. Ceramide (Cer), a central mediator in sphingolipid metabolism, plays an essential role in regulating the physical properties of biological membranes, including the formation of membrane microdomains [43]. Various pathogens, including bacteria, viruses, and parasites, can exploit the acidic or neutral sphingomyelinase–Cer systems to facilitate infection of mammalian cells. For instance, infection with West Nile virus, a mosquito-borne pathogen, leads to significant increases in blood Cer levels in infected mice [44]. Similarly, lipidomic profiling in rotavirus-infected animals has revealed substantial elevations in Cer, and mechanistic studies suggest that Cer promotes apoptosis and is essential for the viral replication cycle [45]. Bacteria can also activate acidic sphingomyelinase to form lipid rafts enriched in SP and cholesterol, which mediate apoptosis and modulate cytokine signaling [46]. Elevated Cer levels have been reported in infections caused by Neisseria gonorrhoeae and Neisseria meningitidis, and inhibition of the acidic sphingomyelinase–cholesterol system has been shown to mitigate disease symptoms [47, 48]. Additional studies suggest that targeting related lipid-modifying enzymes can also reduce pathogenic effects [49]. Cer-enriched membrane platforms are recognized as critical mediators in bacterial infections caused by pathogens such as Pseudomonas aeruginosa, Staphylococcus aureus, and Escherichia coli [50]. Although these mechanisms are well documented in systemic infections, the dynamics of bacterial adhesion within dental biofilms may differ considerably, and further research is warranted to clarify the role of Cer in infections involving S. mutans, particularly in the context of biofilm-enamel interactions. A notable finding from this study is the first-time detection of a significant reduction in salivary sphingosine (SPH) levels in individuals with DC. While no prior research has examined SPH in DC populations, a 2015 study by Valentijn-Benz et al. demonstrated that both SPH and plant sphingosine (PHS) markedly inhibited citric acid–induced demineralization of hydroxyapatite, a key mineral component of enamel, with inhibition rates of 80 ± 17% and 78 ± 17%, respectively [51]. Atomic force microscopy revealed that SPH forms a protective barrier that limits H⁺ion diffusion, effectively preventing acid penetration and bacterial invasion. Building on this, Sauro et al. (2016) investigated the effects of PHS, di-calcium phosphate (DCP), and their combination. They found that DCP occluded dentinal tubules and may convert to a more stable calcium phosphate phase, whereas PHS alone had no significant effect [52]. This may be explained by the strong electrostatic interaction between the positively charged amino group of SPH and the negatively charged hydroxyapatite surfaces. Consistent with our findings, reduced SPH levels have also been observed in patients with atopic dermatitis infected with Staphylococcus aureus and in individuals with COVID-19 [53, 54]. Previous studies have further shown that SPH possesses broad antimicrobial properties, directly inhibiting pathogens such as N. gonorrhoeae, P. aeruginosa, Candida spp., Aspergillus spp., and the influenza virus [55–58]. Although our study observed a decrease in SPH levels in caries patients, direct evidence linking SPH to caries inhibition or antimicrobial activity is still lacking. In general, the effects and mechanisms of changes in salivary lipid content on the colonization, metabolism, and cariogenic potential of cariogenic bacteria still need to be validated through in vitro experiments. KEGG pathway analysis indicated that the lipid differences between caries-affected and caries-free groups are primarily involved in the cAMP signaling pathway, phospholipase D signaling, and the phosphoinositide signaling system. Notably, the phospholipase D and phosphoinositide pathways are likely related to the observed increases in PA and PC, suggesting that altered lipid metabolism in either bacteria or the host may contribute to these changes. The cAMP pathway, in particular, plays a pivotal role in oral bacterial metabolism and dental plaque formation through its regulation of glucosyltransferase expression and activity. The present study found that the prevalence of caries was significantly lower in males than in females. However, no sex-related differences were observed in salivary lipid profiles, suggesting that sex does not have a meaningful effect on salivary lipid composition in caries-prone populations. Previous research has explored the relationship between caries severity and salivary lipids, reporting that the degree of inflammation in caries is positively correlated with triglyceride levels [28]. In contrast, the present study did not identify any differential lipids capable of distinguishing between different stages of caries severity. Similarly, in our earlier investigation of 100 patients with dental fluorosis (selected from a cohort of 957 participants), salivary microbiome analysis revealed no significant association between dental fluorosis severity and salivary microbial composition [59]. This may be due to the limitations of the DMFT index in accurately reflecting the clinical severity of caries. For example, it would be inappropriate to assume that a patient with five shallow lesions (DMFT = 5) necessarily has more severe disease than another patient with a single deep lesion (DMFT = 1). Such nuances underscore the need for further research into more precise measures of disease severity. In this study, a panel of five salivary lipids achieved robust discrimination between caries and control groups as a predictive signature. Consistent with the screening use case, we therefore emphasize model-level metrics (AUC, sensitivity/specificity, PPV/NPV). Together, these results support the feasibility of a salivary lipid-based screening signature for caries in young adults, while cautioning against causal interpretation of individual lipid effects. While our findings point to a potential link between lipid metabolic alterations and caries, several limitations must be considered. First, given the cross-sectional design and modest sample size, the results should be interpreted with caution, and the observed associations should not be taken as evidence of causality, particularly with respect to metabolic pathways. Large-scale, longitudinal cohort studies with clinical validation are essential to determine the reliability and predictive value of lipid metabolites as biomarkers for caries. Second, body mass index (BMI) was not assessed before lipid testing, and an elevated body mass index could influence salivary lipid composition. This factor will be evaluated in future follow-up studies. Third, although all participants were healthy university students with broadly similar lifestyles, the absence of formal plaque index measurements represents another limitation, as periodontal and gingival health could influence salivary composition. Including standardized oral health assessments in future research would strengthen the validity of such studies. Conclusion This study shown lipid metabolic imbalances between the caries-affected group and healthy individuals, and identified potential specific lipid biomarkers in caries patients. These findings provide a theoretical basis for further etiological research and may inform the development of preventive strategies. Supplementary Information Supplementary Material 1. Supplementary Material 1.