The salivary metabolome of children and parental caregivers in a large-scale family environment study

Metabolomics enables broad assay of thousands of small molecules across human and microbial pathways using minimally invasive biofluids such as saliva. Saliva contains host- and microbially-derived compounds and can reveal biomarkers of disease and environmental stress. This study focuses on intra- and inter-family relationships in the salivary metabolome and its associations with measures of environmental tobacco smoke exposure (cotinine), antioxidant potential (uric acid), inflammation (C-reactive protein, CRP), metabolic regulation (adiponectin), and salivary metals (chromium, copper, lithium, manganese, zinc). Prior work shows cohabiting family members share oral health attributes and similar oral microbiomes, and targeted studies report familial similarities in select metabolic biomarkers. However, targeted approaches may miss broader metabolic relationships and unknown biomarkers. Here, untargeted LC-MS metabolomics in a large cohort of children and caregivers from the Family Life Project is used to: (1) quantify within- and between-family metabolomic variation, (2) identify population-level metabotypes, and (3) test associations between salivary metabolites and biomeasures/metals, to understand how the family environment and exposures shape salivary metabolism.

Previous studies demonstrate saliva’s utility for metabolomics and biomarker discovery and highlight the role of the oral microbiome in shaping salivary chemistry. Cohabiting family members tend to have more similar oral microbiomes and targeted metabolic profiles than unrelated individuals. Yet, many prior studies emphasized conventional biomarkers (e.g., cholesterol, amino acids, hormones) and may not capture the full spectrum of familial metabolic relationships or unknown oral health biomarkers. The concept of oral metabotypes, analogous to microbiome enterotypes/stomatotypes, has been applied in other contexts, revealing clusters associated with physiological or lifestyle factors. Environmental tobacco smoke and metals exposures are known to perturb metabolism, but their relationships with the salivary metabolome in large, family-based cohorts remain underexplored.

Study population and samples: Saliva from elementary school-aged children and their caregivers participating in the Family Life Project was analyzed. In total, 1425 saliva samples were processed (children N=719; caregivers N=706), with 1344 samples paired into 672 caregiver–child dyads. Demographic and smoking status data were collected. Ethical approvals were obtained, and all data were deidentified. Biomeasure assays (children unless noted): Samples were stored at −80 °C. After thawing, vortexing, and centrifugation (15 min, 3500 RPM), supernatants were used for assays. Adiponectin (N=701) was measured via MSD Human Adiponectin kit with 5× dilution and four-log standard curve; sensitivity 0.06–1000 ng/mL. CRP (N=616) was assayed in duplicate using MSD V-Plex Human CRP with 5× or 10× dilution; sensitivity 1.33–46,600 pg/mL. Cotinine was measured in both children (N=714) and caregivers (N=672) using Salimetrics ELISA; 20 µL analyzed in duplicate; some samples diluted 10× for reported nicotine use; sensitivity 0.15–200 ng/mL (neat) or 1.5–2000 ng/mL (10× dilution). Values below detection were imputed as half the lower limit for 316 samples (138 caregivers, 178 children). Uric acid (N=630) was measured with Salimetrics kit; sensitivity 0.07–20 mg/dL. Metals data: Salivary metals (children) were obtained from Gatzke-Kopp et al. (2023), using ICP-OES on matched aliquots: chromium (N=237), copper (N=237), manganese (N=237), zinc (N=237), lithium (N=204). Metabolomics sample prep and data acquisition: Frozen saliva was extracted (100 µL) with 3:3:2 acetonitrile:isopropanol:water at the West Coast Metabolomics Center (UC Davis). Samples were evaporated and analyzed by HILIC LC coupled to an Agilent 6530 Q-TOF MS/MS in positive and negative ion modes. Quality control pool samples and method blanks were run identically. Identification used MassBank of North America. Peak areas were normalized with SERRF. Stable isotope internal standards (iSTDs) were added for retention time calibration. Data processing and statistics: iSTD ions were removed; ions with average relative peak height less than twofold over blanks were filtered. Data were normalized to within-sample relative peak heights. Community ecology-inspired “metabology” methods were applied. Diversity metrics (Shannon, Bray–Curtis) and NMDS ordinations were computed (ggplot2/patchwork). PERMANOVA (Adonis, 999 permutations) tested categorical variables (participant type, child sex, caregiver smoking status, child ETS category by cotinine <1 vs >1 ng/mL). Linear mixed-effects models (lmerTest) with dyad as random effect tested differential abundance for ions above thresholds (≥0.01% or ≥0.1% relative peak height as specified). For metals, concentrations were split into tertiles for PERMANOVA. Spearman correlations (Hmisc) assessed associations between ion intensities and biomeasures/metals; P-values were adjusted by Benjamini–Hochberg. Distance-based redundancy analysis (db-RDA, vegan) quantified continuous variable contributions. Metabotype analysis used Jensen–Shannon distances and PAM clustering (ade4, cluster); optimal cluster number was estimated by silhouette values; univariate comparisons used mixed models with multiple-testing correction. Generalized linear models tested caregiver smoking group differences and children’s ETS categories on individual metabolites.

- Data yield: 2881 unique ions detected (281 identified by name). After filtering, 2046 ions remained (210 named). The top 104 ions accounted for 50.02% of total relative peak height; identified metabolites comprised 15.56% of total intensity vs 84.45% unknown. - Pairing: 1426 samples paired into 672 caregiver–child dyads. - Children vs caregivers: PERMANOVA showed small but significant differences by participant category (F=37.2, R²=0.02, P<0.001), while dyad explained most variation (F=1.8, R²=0.62, P<0.001). Alpha diversity did not differ (H(1)=1.6, P=0.20). Ninety-five identified metabolites differed between caregivers and children (Padj<0.05); 20 were highly significant (Padj<2×10^−15), including theophylline−, caffeine+, 4-imidazoleacrylic acid±, ribose-5-phosphate−, inosine, agmatine+, 2-amino-1-phenylethanol+, and multiple amino acids/derivatives. - Metabotypes: PAM clustering of 1344 metabolomes (dyads) indicated two overlapping clusters (silhouette=0.12). Beta diversity differed between clusters (F=189.6, R²=0.12, P<0.001). Twenty-five identified metabolites significantly differed (Padj<0.05), with many showing very small adjusted P-values. Notable differences included hypoxanthine+, oxypurinol−, 2-amino-1-phenylethanol+, 4-imidazoleacrylic acid+, histamine+, nudifloramide+, phenylacetaldehyde B+, and amino acid-related features. - Child sex: No differences in alpha diversity (H(1)=0.44, P=0.51) or beta diversity (F=1.14, R²=0.002, P=0.25); no significant univariate ion associations with sex. - Biomeasures (children): db-RDA with adiponectin, CRP, cotinine, and uric acid (N=538 with all) explained 5.3% of variation (Overall F=7.5, P<0.001); each measure significantly associated (P<0.026). Spearman correlations: adiponectin with 41 metabolites (N=701; −0.16<ρ<0.35; Padj<0.05); CRP with 22 (N=616; −0.13<ρ<0.15; Padj<0.05); uric acid with 94 (N=630; −0.49<ρ<0.57; Padj<0.05). - ETS exposure: Children categorized by cotinine (<1 vs >1 ng/µL) showed no significant differences in alpha (H(1)=0.25, P=0.61) or beta diversity (F=1.1, R²=0.002, P=0.29). Only nicotine+ differed between ETS groups (Padj<0.001). Continuous cotinine (children) correlated with 14 metabolites (N=714; −0.11<ρ<0.26; Padj<0.05). - Caregivers’ smoking: Beta diversity differed slightly by self-reported smoking (F=8.6, R²=0.01, P<0.001); alpha diversity not different (H(1)=0.44, P=0.51). Twenty-two identified metabolites differed between smoking vs nonsmoking caregivers (Padj<0.05), including nicotine+, 4-imidazoleacrylic acid+, phenylethanolamine B+, 5-aminosalicylic acid+, N-omega-acetylhistamine+, histamine+, 2′-deoxyguanosine+, theophylline, phenylacetaldehyde B+, 2′-deoxyadenosine+, inosine, hypoxanthine+, choline+, and nudifloramide+. Caregiver cotinine correlated with 39 identified metabolites (−0.32<ρ<0.67; Padj<0.05). - Metals (children): Mean (range, µg/L): Li 13.9 (0.11–845.5), Cr 7.3 (0.07–22.5), Cu 29.0 (0.23–844.2), Mn 8.6 (0.04–122.5), Zn 60.9 (0.09–566.0). Metals tertiles associated with beta diversity: Li R²=0.01 F=1.5 P=0.044; Cr R²=0.01 F=1.5 P=0.039; Cu R²=0.02 F=2.1 P=0.004; Mn R²=0.02 F=2.7 P<0.001; Zn R²=0.01 F=1.6 P=0.038. db-RDA (N=204 with all metals) explained 4.6% of variation (Overall F=1.9, P<0.001); Cr, Mn, Cu significant contributors (P<0.05), Li and Zn not (P>0.05). Spearman correlations (Padj<0.05): Li with 6 metabolites (|ρ| up to ~0.21), Cr with 9 (|ρ| up to ~0.18), Cu with 42 (−0.46<ρ<0.44), Mn with 52 (−0.47<ρ<0.38), Zn with 29 (−0.43<ρ<0.31).

The study demonstrates that cohabitation/family dyad is the dominant determinant of salivary metabolome variation, underscoring the impact of shared environment, behaviors, and possibly shared microbiota on metabolism. While children and caregivers exhibit broadly similar metabolomes, specific lifestyle- and exposure-related metabolites (e.g., nicotine, caffeine, medications) and dipeptides differ, suggesting age- and behavior-related biochemical contrasts. Population-wide clustering revealed two overlapping metabotypes not explained by recorded metadata, with differences involving dipeptides, urea cycle intermediates, and unidentified ions, implying underlying physiological or microbial drivers that warrant multi-omic follow-up. Kit-based biomeasures tracked expected biochemical pathways: uric acid inversely with purine intermediates and positively with acyl-carnitines/creatine; adiponectin with free dipeptides potentially reflecting immune activity and oxidative stress; CRP showed weak salivary metabolome associations, consistent with its nonspecificity. ETS exposure modestly altered caregivers’ salivary metabolomes and correlated with metabolites linked to cholinergic signaling, catecholamine metabolism, creatine kinase activity, polyamine turnover, and histamine-mediated inflammation; passive ETS in children showed minimal global effects but some specific correlations. Salivary metals associated with subtle but significant shifts in children’s metabolomes, with patterns suggesting interactions with proteolysis and antioxidant processes, and synergistic effects across co-occurring metals. Collectively, these findings support saliva as an informative matrix for population, dyadic, and exposure-related metabolic phenotyping.

Untargeted salivary metabolomics in a large family-based cohort reveals both population-level patterns and subtle within-family differences, with caregiver–child dyad explaining the majority of metabolomic variation. Metabotype clustering suggests shared metabolic features across individuals that are not captured by basic metadata. Kit-based biomeasures (uric acid, adiponectin, CRP, cotinine) correspond to expected metabolic pathways and exposures, and metals are detectably associated with children’s salivary metabolomes. Tobacco smoke influences primary users’ salivary metabolism more than passively exposed children. These discovery-based results highlight saliva’s value for assessing environmental exposures, oxidative stress, inflammation, and metabolic regulation. Future work should incorporate targeted and multi-omic approaches, including integration with the oral microbiome, to identify causal drivers, validate biomarkers, and improve generalizability across diverse populations.

- Large proportion of unknown (unidentified) ions limits biological interpretation and may include indistinct spectral features. - Cohort targeted to specific locations/demographics, reducing generalizability. - Single chromatographic mode (HILIC) likely overrepresents biogenic amines and underrepresents other classes; matrix effects and ionization efficiencies may bias signals. - Magnitude of group differences is small, despite statistical significance. - PAM clustering drivers could not be pinpointed; clusters did not align with available metadata; unmeasured factors (diet, personal habits) may confound. - Cross-sectional, correlative analyses cannot infer causality; metals and ETS associations are associations only. - Limited smoking exposure quantification (e.g., no detailed consumption data) constrains interpretation of dose–response in caregivers/children. - Variability and noise evident in overlapping clusters may affect interpretations.