Method performance evaluation

For the instrument detection limits (IDLs), 3 replicates of spiked solutions in LC-MS grade water or 5 replicates in methanol were analyzed and IDLs were estimated as the lowest concentration value with S/N > 3, in targeted ion extraction mode using Xcalibur software, thus providing instrumental sensitivity when using either the online SPE or direct LC-HRMS method. The IDL results for each compound in the QC mix are shown in Supplementary Table S5. Overall, for 13 of the 17 QC compounds, the IDL for the direct injection method was 0.1 µg/L, except for sucralose and gemfibrozil which were 1 µg/L and hydrochlorothiazide at 4 µg/L. The IDL for the online-SPE ranged from 1.9 to 38 ng/L with 14 of the 17 analytes having an IDL of 1.9 ng/L, except for caffeine (9.5 ng/L), hydrochlorothiazide (19 ng/L), and sucralose (38 ng/L).

The internal standard mixture (IS) used in this study contained a total of 22 isotopically labeled chemicals amenable for positive and/or negative detection modes. To verify which labeled standards would be adequate and should be monitored for all types of samples (ones that were constantly detected by the different procedures), it was assessed their presence in the analyzed samples at the concentration of 500 ng/L for online SPE (10.5 µL of 0.5 µg/mL IS mix) and 500 µg/L for direct LC-HRMS (100 µL of 5 µg/mL IS mix). The internal standard mixture was added to a total of 22 samples, including laboratory blanks, quality control solutions, indoor dust, food, and urine samples. The IS detection frequency ranged from 23% (for paroxetine-d4) to 100%, in which trimethoprim-d9, albuterol-d9, atenolol-d7, and valsartan-d3 were detected in positive mode in all 22 analyzed samples with an intensity between 106 and 109 (as seen in Supplementary Table S6). Due to the limited commercial availability of valsartan-d3 and albuterol-d9, trimethoprim-d9 and atenolol-d7 were selected as internal labeled standards for the positive mode. Among the 5 internal standards (IS) that were amenable in negative mode, glipizide-d11 and warfarin-d5 were selected since they showed the highest detection frequency of 86.4% and 95.5% (Supplementary Table S6), respectively, with an intensity of 106.

For the evaluation of the NTA method for sensitivity, selectivity, accuracy, and precision, native chemicals in the QC samples (at a concentration of 380 ng/L for the SPE method and 200 µg/L for the direct LC-HRMS method) were evaluated over 11 different days. The average TPR (sensitivity) for the developed NTA method was 0.711, the average TNR (selectivity) was 0.984, the average precision was 0.203, and the average accuracy was 0.982. The selectivity and accuracy showed the best performance with values greater than 0.98, indicating that the method was accurate and specific, at the tested concentration. The sensitivity was above 0.7, which is deemed acceptable and above the 70% threshold established for the analysis. The precision was the lowest metric observed at 0.203. Overall, the developed and optimized NTA method has shown adequate performance.

Method optimization for urine

The purpose of determining the urine dilution factor was to find an optimal condition where matrix effects are not too pronounced to interfere with compound identification and that the dilution is not too much to significantly impair compound detection. Dilution factors of 2, 5, 10, 20, and 50 were tested for the optimization. The diluted urine samples were spiked with 20 µL of the 0.2 µg/mL QC working solutions and analyzed by online SPE-LC-HRMS. The averages of retention times and peak areas were calculated for QC samples prepared in LC-MS grade water. For each urine sample, a comparison of the individual retention time and peak area with the average was conducted. If the retention time shift was more than 0.5 min or the peak area varied more than 50% of the average [8], it would be considered as retention time fail or peak area fail, as seen in Supplementary Table S7. Among the 17 compounds present in the QC mixture, a maximum of 15 QC compounds were observed in the urine samples when the dilution factor was 20, whereas 86.7% of the QC compounds passed the retention time check and 80% passed the peak area check. Compared to others, the 20- and 50- times dilution factors showed the best results with less pronounced matrix effects. Considering that the 50 times dilution also led to a reduced number of compounds detected, possibly due to sensitivity issues, a dilution factor of 20 was selected for the NTA of urine samples. Initially, very few features were detected in non-hydrolyzed urine (321) and the KMD plot showed that features were mostly in the lower mass ranges (100–400). When comparing the effect of adding the enzymatic hydrolysis step to the urine NTA method, it was clear that the use of the β-glucuronidase/arylsulfatase enzyme enables the identification of more annotated features (hydrolyzed urine had 823 features, Fig. 1), which were before (non-hydrolyzed urine sample) likely either glucuronidated or sulfonated and therefore not identified in the databases used.

Fig. 1: KMD plot and Venn diagram comparing unhydrolyzed and hydrolyzed urine samples.

Blue dots represent features detected in the hydrolyzed urine and Orange circles represent features detected in the non-hydrolyzed urine. The Venn Diagram shows the number of features detected in each urine treatment, highlighting the number of features found in common (98 features).

Comparison of ASE and USE for soil extractions

To evaluate the efficiency and performance of ASE and USE for the comprehensive extraction of chemicals of interest from soil samples in the NTA context, we have tested spiked and unspiked soil samples. For unspiked soil samples, a total of 251 features (155 in positive and 96 in negative mode) were identified only by USE and 91 (56 positive and 35 negative mode) were identified only by ASE, with some overlap; a total of 40 tentatively identified features (24 positive and 16 negative) were found by both methods as shown in Supplementary Fig. S3, suggesting a potential higher extraction efficiency for USE. In the soil samples spiked with the QC analytes, we observed that both extractions were able to successfully recover all 17 QC analytes with half of the analytes having similar responses (Table 1), except for sucralose, hydrochlorothiazide, caffeine, norcocaine, diltiazem, diclofenac, and mefenamic acid that showed higher response (represented as peak area) by ASE, and gemfibrozil that was improved by USE. Overall, the methods were deemed comparable in terms of QC performance, nevertheless, taking into consideration the practicality in terms of the reduced time of analysis and semi-automated sample preparation steps (extraction and cleanup are performed simultaneously inside the cell) associated with the high temperature and pressure that enable a faster diffusion rate of compounds into solvent solution, ASE was selected as the method of choice for the soil and dust samples.

Table 1 ASE versus USE comparisons for the soil extractions.Method optimization for food samples

To optimize food extraction performance, ACN and MEOH were tested in spiked (with QC mixture) and unspiked food (consisting of a homogenized mixture of lettuce, rice, milk, and bread). For the unspiked samples, a total of 334 (202 in positive and 132 in negative mode) features were identified only in the ACN extraction compared to 147 (94 positive and 53 negative) features in MEOH, while having 87 (68 positive and 19 negative) features identified in both extraction solvents (Venn diagram in Supplementary Fig. S4). Overall, extraction performance was comparable, but acetonitrile as extraction solvent showed not only higher number of compounds detected (level 2a) as well as a higher response for 16 out of the 17 QC analytes spiked in the samples; only lincomycin had a higher response in methanol (Table 2). In addition, the use of acetonitrile led to clearer final extracts (methanol extracts were cloudy even after cleanup step), therefore ACN was selected as the extraction solvent for further QuEChERS food assessments.

Table 2 Comparison of acetonitrile (ACN) and methanol (MEOH) solvents in the extraction of QC analytes by a QuEChERS in food sample matrices.Prioritization and identification of chemicals in soil samples

To visualize the tentative identified chemicals obtained in this study, which encompassed a total of 10 soil samples to which small children have access and contact to, the data was plotted in a Kendrick mass defect (KMD) plot [29, 30]. A KMD plot is a visualization tool in mass spectrometry used to compare molecular weight distribution in complex mixtures offering a simplified way to visualize data and identify difference between samples and is graphically represented by the difference between the nominal mass and exact Kendrick mass against the Kendrick nominal mass (KNM). This difference reduces the massive spectral data obtained by restricting compounds within the same homologous series to a fixed mass unit intervals (the most used is 14 for CH2), allowing in some cases the observation of distinct patterns [29,30,31]. As seen in Fig. 2, features are distributed in the KMD plot between the masses of 100 and 800 [32], but most features showed higher overlap among different samples at lower mass region (KNM 100–400), forasmuch as unique features are more frequently observed at higher mass ranges (400–800). Features with negative KMD (−0.6 to −0.1) were observed in soil samples, indicative of polyhalogenated compounds which tend to exhibit a negative mass defect [18]. There is no specific pattern identified in the KMD plot displayed in Fig. 2, in fact it can be difficult to identify homologous series in complex mixtures and samples with many detected features [31, 33]. Therefore, often employed alongside KMD plots is the Van Krevelen diagram (VKD), in which the atomic ratio of hydrogen to carbon (H/C) is plotted in the x-axis against the atomic ratio of oxygen to carbon (O/C) in the y-axis of a specific compound [31]. VKD is a valuable tool in understanding the chemical composition of organic compounds, separating them based on their degree of saturation (aromaticity) and by oxygen‐containing classes. Using VKD, for example, aromatic compounds will be distinctively found along the y‐axis of H:C, whereas per- or polyfluorinated compounds (PFAS), in which most H atoms are replaced with fluorine, will shift to the lower region of the VKD [19]. We have previously identified regions in the VKD associated with anthropogenic chemicals such as legacy and emerging organic contaminants of concern using the EPA DSSTox library [33] and applied the concept to our samples, as seen in Fig. 3. According to the VKD, the regions/boxes heavily populated are of aromatic hydrocarbons (region 1), polyethylene glycol/polypropylene glycol (PEG/PPG) (region 3), surfactants (region 4), and pesticides, bisphenol, and phthalates (region 5), however considering that aromatic hydrocarbons are not amenable to ESI and that this tool is for broader application, including GC-HRMS, it’s not expected that this class of compounds will be detected in the samples by the methodology used.

Fig. 2: Kendrick mass defect plot of soil samples from different participants (N = 10).

The dots of different colors represents the features detected in the soil samples analyzed; gray dot = S001, red dot = S002, green dot = S003, yellow dot = S004 and blue dot = S005. KMD: Kendrick mass defect, KNM: Kendrick Nominal Mass.

Fig. 3: Van Krevelen plot of soil samples from the different participants.

Numbered boxes comprise (1) aromatic hydrocarbons; (2) polychlorinated biphenyls; (3) polyethylene glycol/polypropylene glycol; (4) surfactants; (5) pesticides, bisphenols, and phthalates; (6) polybrominated diphenyl ethers; and (7) per-and polyfluoroalkyl substances. The dots of different colors represents the features detected in the soil samples analyzed; gray dot = S001, red dot = S002, green dot = S003, yellow dot = S004 and blue dot = S005.

A total of 2239 features were detected in soil samples, in which 107 annotated features were commonly detected in more than 50 % of the samples (Supplementary Table S8). Information on feature classifications was further searched at PubChem, ChemSpider, EPA ECOTOX database, and literature references. Among the 107 tentatively identified chemicals, 35% were classified as natural product, followed by 16% of pharmaceuticals, 14% of industrial products and less than 5% each of pesticides and personal care products (Supplementary Fig. S5). Interestingly, 2 % of the features were identified as per- or polyfluorinated compounds (PFAS), corroborating with the few detections in the VKD PFAS region. The top 10 most abundant features detected in the soil are included in Table 3.

Table 3 Most abundant features in each type of samples.Prioritization and identification of chemicals in indoor dust samples

The KMD and VKD plot of the features of dust samples are plotted in Supplementary Figs. S6 and S7. A total of 3218 features were detected in dust samples, having 85 commonly detected features in more than 50% samples (Supplementary Table S8). Distribution of the features in the KMD plot ranged from masses of 150 and 800, with the majority overlapping in the 150–500 mass range, similarly to the pattern observed in the soil samples. Only two features had negative KMD, suggesting the presence of few polyhalogenated compounds identified in the samples. In the VKD, features of dust samples were mostly aggregated in region 4 and 5, indicating a high proportion of surfactants, and pesticides, bisphenols, and phthalates within the analyzed samples. Tentatively identified features were composed predominantly by 29 % of natural products and 25% of surfactants, which confirms the high number of features in this region of the VKD (Supplementary Fig. S8). Also, chemicals used in industrial products (13%), phthalates (8%), multiple use chemicals (8%), and personal care products (8%) were detected in the dust samples. The top 10 most abundant features detected in indoor dust are shown in Table 3.

Prioritization and identification of chemicals in urine samples

It was observed the largest number of features in the urine, which lead to a total of 5121 features, in which 265 were commonly detected in more than 60% of the urine samples (seen in Supplementary Table S8). The KMD and VKD plot of the features observed in the urine samples are shown in Supplementary Figs. S9 and S10, respectively. The urine samples showed detected features distributed between the masses of 100–800, with high overlapping in a wider range of KNM (100–600) than previously seen in other matrices, suggesting that compounds with higher molecular weight have been commonly found in urine. Similar to what was previously observed, the majority of the features are highly populated in regions 4 and 5, representing the presence of surfactants, pesticides and products containing plastic (bisphenol) and plasticizers (phthalates), and with very few features in the PFAS region (region 7). The presence of some features in the region 6 corresponding to polybrominated diphenyl ethers (PBDE) it’s unexpected as this class of compounds is not amenable to LC-ESI-HRMS, therefore these features likely correspond to another class of anthropogenic organic contaminants not included in this VKD (for example, brominated flame retardants or hydroxylated derivatives of polybrominated diphenyl ethers). Tentatively identified features in urine were composed predominantly by natural products (35%) and 17% of pharmaceuticals/drugs (Supplementary Fig. S11). Also, chemicals observed in the urine samples in minor proportion were pesticides (7%), personal care products (7%), multiple use chemicals (4%), and industrial products (3%). The most abundant and frequently detected features in urine (Top 10) can be found in Table 3.

Prioritization and identification of chemicals in food samples

A total of 2552 features were detected in the food samples, in which 39 annotated features were frequently observed in the samples (50–90%) and listed in Supplementary Table S8. The KMD was plotted in Supplementary Fig. S12, showing features commonly detected in the mass range of 100–500, with sample S002 having more features in the high KNM region (500–800), and S004 and S005 having few features in negative KMD. As observed in the VKD displayed in Supplementary Fig. S13, regions 4 (surfactants) and 5 (pesticides, bisphenol and phthalates) were heavily overlapped among all samples, with fewer features detected in the PFAS and PEG/PPG boxes. The majority of the features identified in the food were natural products (52%), followed by 19% of food additives, 13% of industrial products, 7% of personal care products and a small proportion of chemicals with multiple uses (3%) (Supplementary Fig. S14). The list of the top 10 detected features in food samples are presented in Table 3.

Prioritization and identification of chemicals in water samples

A total of 788 features were detected in the water samples provided by the participants, in which 20 annotated features were commonly detected in at least 50% of the samples (Supplementary Table S8). The KMD plot illustrated in Supplementary Fig. S15 shows that features detected in sample S005 comprised mostly KNM between 150 and 250, while others were spread out between the masses of 150 and 600, showing the detection of compounds with a wider range of molecular weight, as also observed for the urine samples. However, few features were identified at masses higher than 500. A few features from samples S002, S003 and S005 showed potential halogenated compounds with negative mass defects (between −0.4 and −0.6). The VKD displayed in Supplementary Fig. S16 showed the majority of the features overlapping in regions 3 (PEG/PPG), 4 (surfactants), and 5 (pesticides and plasticizers), and fewer and more spread detected features in region 7 (PFAS). The predominant composition of the tentatively identified chemicals was natural product (28%), followed by pharmaceuticals/drugs (22%), food additives (17%), pesticides (11%), and industrial products (11%) (Supplementary Fig. S17). A list of the top 9 compounds detected in the drinking water samples are listed in Table 3.

Identification of common features with children’s urine

To better understand children exposure to organic contaminants and potential associated toxicological concerns, features frequently identified in all different ingestion sources were combined and illustrated in a Venn diagram shown in Fig. 4, to identify correlations between the chemicals found in the possible ingestion sources and in children’s urine. The data was found to be not normally distributed when applying Shapiro Wilk’s test, and therefore, Spearman correlations were performed as shown in Fig. 5. It was observed a strong positive correlation between the compounds found in common in food and water with urine, whereas a very weak correlation was found for dust and soil, which reinforces that diet, including water consumption, is the major exposure pathway of organic chemicals in children. The tentative identity of the common features in each sample are shown in Supplementary Table S9. Compounds identified in food and urine samples were mostly natural products (Abscisic acid, 3-hydroxy-N-(1-hydroxy-3-methylpentan-2-yl)-5-oxohexanamide, F-36316 C, Hexanoylcarnitine, Naringenin, Piperanine, Streptazone F, 4-Indolecarbaldehyde), but also included pharmaceuticals (Dobutamine, Pactamycin, Phenacetin). Common features in water and urine samples were the natural product cuminaldehyde, the pesticide naphthaleneacetamide, and the industrial product isophorone. Compounds identified in soil and urine samples contained industrial product (Caprolactam), natural product (Dibutyl ethylmalonate), and pharmaceutical (Oseltamivir). Common features in dust and urine samples were natural products (3-[(3-Hydroxydecanoyl)oxy]decanoic acid, Piperanine, and Uric acid), the personal care product Tetraacetylethylenediamine and the industrial product 3,6,9,12,15,18-Hexaoxaicosane-1,20-diol. Further confirmation of the identified chemicals by acquisition of authentic standards and quantification are still necessary and would bring a better understanding for environmental and human risk assessments, including estimation of children’s health risks.

Fig. 4: Venn diagram of combined features and intersections found between the different matrices and urine.

The Venn diagram shows the number of features detected in each matrix and the intersections between the circles shows the number of features detected in common with urine. Circles size are proportional to the number of features detected.

Fig. 5: Spearman correlations between chemicals found in possible ingestion sources and in children’s urine.

NA means not enough data was found in common to perform correlations.