For this study, surplus samples from a previous research project were employed. Blood samples were collected for reasons unrelated to this study, due to the decision of the veterinary specialist in charge of the animal. In this study, only the serum remaining after biochemical analysis was used.

Paired blood serum and saliva samples of a total of 30 (9 male and 21 female) dogs with a mean age of 9.2 years (SD 3.7 years) were included in the present study (Supplementary Table S1). All animals were privately owned dogs of various breeds presented to private veterinary clinics of the Murcia Region, Spain. Based on clinical evaluation and body condition score, animals were divided into three groups (n = 10 per group): healthy normal-weight (HN), healthy obese (HO), and dogs with mammary tumours (MT).

In all the cases, samples were collected after at least 8 h of fasting. Saliva was collected using a previously reported non-invasive method (24). For saliva sampling, a piece of sponge was placed in the dog’s mouth for 1–2 min and then passed into a Salivette device (Salivette®, Sarstedt AG & Co., Nümbrecht, Germany) for centrifugation (3,000 × g, 10 min, 4 °C), and saliva supernatants were transferred to 1.5 mL polypropylene tubes. Saliva sampling blanks were prepared by sampling a single anonymous individual using a sponge and by collecting saliva via direct drooling into the polypropylene tubes; both procedures were performed in triplicate. Blood samples were collected by the venipuncture of the jugular or cephalic vein in tubes containing a coagulation activator and a gel separator (Tapval, Aquisel, Spain) and maintained at room temperature (25 °C) until visible clot reaction. Samples were centrifuged (3,500 × g, 5 min, 4 °C), and serum was transferred to 1.5 mL polypropylene tubes. Samples were stored at −80 °C and shipped on dry ice until analysis at the Department of Environmental Science, Stockholm University, Sweden.

Sample collection and usage were reviewed and approved by the Animal Experimentation Ethics Committee (CEEA) of the University of Murcia and the General Directorate of Livestock, Fisheries and Aquaculture of the Region of Murcia, Spain, under approval number A13170503, in accordance with the European Council Directives on the protection of animals used for scientific purposes. In addition, the study was conducted in compliance with the ARRIVE (Animal Research: Reporting of In Vivo Experiments) guidelines for animal care and use.

The analytical method was based on a previously published method for the determination of phenolic compounds in human blood (7), which describes the analysis of both the conjugated and the free fraction of phenolic analytes in 1 mL of serum. As the volume of serum and saliva samples from the dogs was limited, the method needed some adaptations. First, only the free fraction of SPAs and PFAS was determined in this study. Second, ca 200 μL of serum or saliva were analysed, and the volume of solvents and chemicals was reduced (no re-extractions) to minimise background contamination. Traditionally, PFAS analysis is carried out using polypropylene tubes to avoid glass absorption. In this study, we used glass tubes to prevent SPA contamination from plastic material. Much effort was spent on minimising background contamination due to the widespread use of these chemicals. The glassware was burned at 300 °C for 24 h, the solvent was distilled, and the clean-up matrix was washed.

Details regarding chemical information, analytical standards, instrumental settings can be found in Engelhardt et al. (7) for SPA analysis, and in Engelhardt et al. (25) for PFAS analysis.

Sample preparation: In detail, to ca 200 μL sample (serum or saliva), 1 ng of the internal standards were added (Supplementary Table S2) and left overnight to equilibrate with the sample. The saliva volume available was, in some cases low, ranging from 64 μL to 210 μL. Two mL of distilled acetonitrile (ACN, HiPerSolv Chromanorm for HPLC, VWR International bv, Leuven, Belgium) was added to the sample for denaturation and extraction. The sample was vortexed, cradled for 5 min, and centrifuged at 3,500 relative centrifugal force (rcf) for 5 min at 16 °C. The supernatant was transferred to a new glass tube containing 50 mg of prewashed (2 times with 1 mL distilled ACN) Enhanced Matrix Removal-Lipid powder (Bond elut EMR-Lipid, Agilent Technologies, Santa Clara, United States). The tube was then vortexed for 10 s and centrifuged at 3500 rcf for 5 min at 16 °C. The supernatant was transferred to a new glass tube, and 200 mg anhydrous magnesium sulphate (MgSO₄) was added to the extract to remove any remaining water, vortexed, and centrifuged at 3500 rcf for 5 min at 16 °C. The supernatant was transferred to a new glass tube and evaporated with a gentle stream of nitrogen gas to ca 100 μL. The volumetric standards M8PFOS (1 ng) and CB207 (0.5 ng) were added for the quantification of the recovery of the internal standards using LC–MS and GC–MS analysis, respectively. The sample extract was stored in the freezer until analysis.

As quality control (QC) samples, method blanks (n = 7), saliva sampling blanks (n = 3 + 3), and certified reference material (CRM, human serum NIST 1957, n = 3) were analysed. The method blank sample matrix was prepared by adding a drop of previously analysed olive oil to 10 mL distilled ACN. Background contamination from saliva sampling was assessed by subtracting the analyte concentrations found in the drool samples from those found in the sponge-collected samples.

Instrumental analysis: Short, the sample extract was first injected (1 μL) on a Trace 1,300 gas chromatograph coupled to a triple quadrupole mass spectrometer (GC-TSQ 9000, Thermo Fisher Scientific Inc.). The injection temperature was 280 °C, and extracts were injected in a splitless mode. The chromatographic separation was achieved using a DB-35MS UI column (30 m, diameter 0.25 mm with a 0.25 μm film, Agilent J&W GC columns Agilent Technologies) with helium as carrier gas. The temperature programme was 90 °C (1.5 min hold), 10 °C/min up to 310 °C (5.5 min hold). The ionization mode was set to positive electron ionization in multiple reaction monitoring mode using timed events [see details in (7)].

The matrix-matched calibration curves were used to quantify SPA analytes on the GC-TSQ, utilising commercially obtained charcoal-stripped fetal bovine serum (Merck), as matrix effect could be observed for the analytes (Supplementary Table S3). The SPA concentration of the calibration curve ranged from 0.03 to 300 ng/g using 10 points, with eight points distributed in the lower concentrations. The two high calibration points were added to cover the concentration of some analytes with high concentrations. The calibration curves were accepted if the linearity criteria r² > 0.975 were met.

The same extract as analysed on the GC-TSQ was further analysed using a Dionex UltiMate™ 3,000 ultra-high performance liquid chromatography combined with a Q Exactive™ HF hybrid Quadrupole-Orbitrap™ mass spectrometer (LC-Orbitrap, Thermo Fisher Scientific Inc.). Background contamination from the mobile phases was delayed using an isolator column XBridge™ C18 (2.1 × 50 mm, 3.5 μm particle size, Waters), mounted before the injector. The injection volume was 5 μL, and the flow rate of the mobile phases was set to 0.4 mL/min. A BEH C18 column (2.1 × 50 mm, 1.7 μm particle size, Waters) and a guard column BEH C18 (2.1 × 5 mm, 1.7 μm particle size, Waters) were used for chromatographic separation. For PFAS analysis, an 8-point calibration curve was used with a concentration span of 0.02–15 ng/g.

The mobile gradient started with 90% A (2 mmol/L ammonium acetate in 95% water and 5% ACN) and 10% B (2 mmol/L ammonium acetate in 99% ACN and 1% water), 0.5 min of hold time, mobile phase B increased to 100% during 7.5 min, 3 min of hold time and then switched back to 90% A to equilibrate the column. The analysis was run in negative ionization mode using full scan and data-dependent acquisition with an inclusion list of more than 30 target analytes and 40 suspect features [see details in (7)]. The peak integration was performed using Tracefinder (version 4.1, Thermo Fisher Scientific Inc).

Only PFAS and SPAs detected in dog saliva or serum are reported here. The instrumental limit of quantification (LOQ) was determined as the lowest calibration point with a bell-shaped peak (Supplementary Tables S3, S4). The method LOQ (MLOQ) and the sampling LOQ (SLOQ) were calculated as the average blank level plus one standard deviation (Supplementary Tables S3, S4). The saliva sampling was tested for background contamination, but not the serum sampling. If the analyte was present and stable in the blank (RSD < 30%), the average blank was subtracted from the reported levels.

The levels are reported based on weight to be able to compare with previously reported levels. Statistical evaluation was conducted on analytes with a quantification frequency (QF) of at least 30% of the samples in the two matrices. For statistical evaluation and creation of the figures, results below LOQ were set to LOQ/2, and non-detects (n.d.) to LOQ/10.

All statistical analyses were performed using R v3.2.2 (R Core Team, 2013). Data normality was assessed using both the Shapiro–Wilk test and visual inspection of Q–Q plots. Given the sample size and the Shapiro–Wilk results (p < 0.05), all variables were considered non-normally distributed. Therefore, non-parametric tests were applied.

Correlations between paired serum and saliva were evaluated using Spearman’s rank correlation test. Comparisons among analytes and the three health groups were conducted using the Kruskal–Wallis test, followed by Dunn’s post hoc test. To reduce the risk of type II error in the context of limited sample size, post hoc testing was also performed for variables with Kruskal–Wallis p ≤ 0.3. Categorical variables (i.e., sex) were analysed in relation to serum and saliva analytes using the Mann–Whitney U test. In all analyses, p < 0.05 were considered statistically significant.

In this pilot study, a previously established method developed for the analysis of a wide range of synthetic phenolic compounds in blood serum was used, with minor adaptations. No recovery study on neither saliva nor blood serum was performed here, but previously established recoveries of SPAs in blood serum are reported in Supplementary Table S3. The recoveries of the internal standards DBP-d21 and the labelled PFAS were similar in blood serum and saliva (Supplementary Table S2). The CRM sample (NIST 1957) was analysed in triplicates and compared to previously reported SPA levels reported in the same material (Supplementary Table S3) (7). The agreement was found between 84 and 220%, except for 2,6-di-tert-butyl-4-(hydroxymethyl)phenol (BHT-OH). BHT-OH in this study was 10 times higher than that reported earlier. Although, in the previous study, 2.7 ng/g serum of BHT-OH was found in the conjugated fraction but not analysed here. The repeatability was measured by the relative standard deviation (RSD) of the CRM sample analysed in triplicates, which was good (below 12%).

The recoveries of the PFAS internal standards were satisfactory (Supplementary Table S2). Compared to the levels reported by NIST (26), PFAS in the CRM (NIST 1957) were within 30% deviation (82–127%) on an average (Supplementary Table S3). Based on this finding, it was judged that the applied sample preparation method developed for phenol analyses was also suitable for PFAS analysis.

In Table 1 and Supplementary Tables S3, S4 analyte concentrations are reported above the limit of quantification, and levels below are reported with their corresponding LOQ, or as non-detects (n.d.) when no peak was indicated. Fourteen SPAs (including five BHT metabolites) and eight PFAS were determined in dog’s saliva and/or serum. In the dog saliva, 2,4-di-tert-butylphenol (2,4-DBP. <LOQ-860 ng/g saliva), 3,5-di-tert-butyl-4-hydroxybenzaldehyde (BHT-CHO. <LOQ-11 ng/g saliva), and perfluorononanoic acid (PFNA. n.d.-0.82 ng/g saliva) were quantified in >90% of the samples, followed by BHA (n.d.-11 ng/g saliva) and perfluorooctanesulfonic acid (PFOS, linear, and branched; <LOQ-6.2 ng/g) quantified in >70% of the samples (Table 1). 2,4-DBP (130–2,100 ng/g serum), BHA (0.35–12 ng/g serum), 2,6-Bis-(1,1-dimethylethyl)-2,5-cyclohexadiene-1,4-dione (BHT-Q. 19–710 ng/g serum) and PFOS (linear and branched, 1.0–19 ng/g) were quantified in all dog blood serum samples (QF 100%).

2,4-DBP was detected at the highest level, with a median concentration of 200 ng/g saliva and 340 ng/g serum. A background contamination (53 ng/g sample, RSD 27%) was identified, likely originating from the sample pretreatment (method blank, Supplementary Table S3). Levels found in the saliva sampling QC sample were linked to the sample pretreatment and not to the sampling equipment. Unfortunately, no dog serum sampling blank was available in this study. In the previously published method (7), the sampling blank for 2,4-DBP analyzing human serum was below the method blank, which supports the notion that the contamination originated from the chemicals used in the pre-treatment method and not during sampling. Instead, the EMR powder was previously reported to be the primary source of 2,4-DBP contamination in the validated method, despite washing the EMR powder twice with distilled ACN (7).

The two approved feed additives, BHA and BHT, could be found in all dogs. BHA levels were similar in saliva and serum samples (median 1.8–1.9 ng/g and range n.d.-12 ng/g). BHT were primarily found as its metabolites, BHT-CHO, 3,5-di-tert-butyl-4-hydroxybenzoic acid (BHT-COOH), BHT-OH, BHT-Q, and 2,6-di-tert-butyl-4-hydroxy-4-methyl-2,5-cyclohexadienone (BHT-quinol). The median levels were below LOQ, but average levels of octyl 3-(3,5-ditert-butyl-4-hydroxyphenyl) propanoate (AO1135) were similar in saliva and serum (9–11 ng/g) and could be quantified in 31% of the saliva samples (range n.d.-140 ng/g saliva) and 48% of the serum samples (n.d.-87 ng/g serum).

Eight PFAS congeners were quantified in dog’s saliva and/or serum (Table 1). Primarily, PFNA (n.d.-0.82 ng/g saliva) and PFOS (linear and branched, <LOQ − 3.3 and n.d.-2.9, respectively) could be quantified in dog’s saliva, in 93 and 73% of the saliva samples, respectively. PFBS was quantified in only one saliva sample (0.14 ng/g saliva). In dog blood serum, all eight PFAS could be quantified in >20% of the samples, at levels of up to 19 ng of PFOS (linear and branched)/g of serum.

In Figure 1, all SPAs (n = 14, Figure 1A) and all PFAS (n = 8, Figure 1B) are summed to compare the total exposure in dog saliva and blood serum of each individual dog. The major SPA contributor to sum SPA was 2,4-DBP, constituting, on average, 83% in saliva and 72% in blood serum. The major PFAS contributor to sum PFAS was PFOS (linear and branched), constituting on an average 62% in saliva and 57% in blood serum.

Only analytes quantified in more than 30% of the samples were further evaluated statistically, resulting in 7 SPAs and 3 PFAS in saliva and 10 SPAs and 7 PFAS in blood serum (Supplementary Table S5).

There were no correlations between any of the analytes in the paired saliva and blood samples. There were no statistically (Kruskal–Wallis) significant differences between the dogs with the three different health status; healthy normal-weight (HN), healthy obese (HO), and dogs with mammary tumours (MT) (Figures 2, 3 showing the analytes quantified in >70% in both matrices). In addition, sex, age, weight, and body condition score did not correlate to the target analytes.

The analytes were tested for correlation within the two matrices (Supplementary Table S5). In saliva, 2,4-DBP correlates positively with BHT-CHO (r = 0.448, p = 0.015) and PFNA (r = 0.613, p = 0.000) but negatively to BHT-Q (r = −0.391, p = 0.04). AO1135 correlates negatively to PFOS (r = −0.441, p = 0.017). In serum, BHT-COOH correlates positively to AO1135 (r = 0.561, p = 0.008). BHT-Q correlates positively to 2,4-DBP (r = 0.452, p = 0.041), and BHA (r = 0.518, p = 0.017) but negatively to 2,4,6-tris-tert-butylphenol (AO246. r = −0.595, p = 0.004). BHT-quinol correlates negatively to 4,4′-methylenebis(2,6-di-tert-butylphenol) (AO4426. r = −0.489, p = 0.024), BHT-OH (r = −0.449, p = 0.041), but positively to BHT-Q (r = 0.532, p = 0.013). PFNA correlates positively to all PFAS (r = 0.442–0.69, p < 0.015).

This is the first time SPAs and PFAS levels have been reported in dog’s saliva. Despite limitations regarding sample size and method validation, this study confirms saliva to be a potential alternative non-invasive matrix to blood serum that motivates further investigation. It is recommended to verify these results in larger cohorts, as well as further studies on kinetics, using controlled exposure experiments.

Considering the lack of knowledge regarding the toxicity of the SPAS, the accumulated concentration in dog’s blood serum is concerningly high, reaching levels above 4 μg/g serum. As AO4426 and 2,4-DBP are currently under assessment as endocrine disruptors further motivates investigations of these high-production volume chemicals.

No correlations between analyte levels and health status could be statistically determined, which was expected due to the limited samples size. The comparative exposure assessment showed that humans and dogs are exposed to similar levels and that using dogs as sentinels for human exposure is a promising strategy. Comparative and translational studies on dogs as model organisms for endocrine disease, is a non-experimental animal study design within One Health and 3R.