LC-MS grade solvents were employed throughout all experimental steps. Serial dilution of a multicomponent stock solution of analytical standards containing 216 analytes was conducted for building a calibration curve covering seven concentration levels. Concentration ranges were compound-dependent and were previously optimized by Jamnik et al.¹⁵ and Gu et al.,³⁷ hence, we refer to concentration levels rather than to actual concentration values in this work. An overview on the analytes included, alongside their working concentrations are reported in Table S1. Labeled bisphenol A (bisphenol-A-diphenyl-¹³C₁₂), was purchased from Sigma-Aldrich, labeled zearalenone (U-[¹³C₁₈]-zearalenone), from Romer Labs. All chemicals and solutions were stored at −20°C until use. Detailed information on the chemical suppliers is provided by Gu et al.³⁷

Breast milk samples from Austria were provided by the Semmelweis Women’s Clinic milk bank in Vienna. Samples of more than 150 women were collected in 2015, pooled and stored at −20°C.⁹ For preparing the DMS, 40 µL of pooled breast milk sample were pipetted onto 903 Whatman protein saver cards and allowed to dry for 3 h at room temperature. The dried spots were then prepared using a 1.2 cm diameter punch, ensuring the entire spot was utilized. The spots were put into 1.5 mL tubes and 1 mL of two tested extraction solutions was added (ACN/MeOH/H₂O 40:40:20 v/v versus ACN/MeOH/MTBE/H₂O 30:30:20:20 v/v). The samples were extracted at 1200 rpm for 5 min at room temperature on a thermo mixer, followed by sonication for 10 min at room temperature and additional shaking for 5 min under the same conditions. After centrifugation at 9,500g and 4°C for 5 min, 800 µL of the supernatant were transferred to another tube. The samples were placed into a vacuum concentrator set at 20°C until complete dryness. The remaining residue was reconstituted in 80 µL of H₂O/ACN (90:10 v/v). After shaking for 10 min, the samples were centrifuged for 5 min, the supernatant transferred to glass vials and measured afterwards. For evaluating extraction recoveries and matrix effects, spiked samples were utilized. Samples spiked before extraction (“pre-spiked”) were prepared by spiking an aliquot of the pooled breast milk with the working solution of standards prior to spotting in the cards. A medium concentration spiking level was chosen, aiming at obtaining a final extract at theoretical concentration Level 23" (see Table S1), assuming 100% recovery. Post-spiked samples were prepared by reconstituting the residue of non-spiked DMS samples in a spiked reconstitution solution prepared with the same mix of compounds to achieve the same nominal concentration level. For each extraction solution tested, five replicates of both pre- and post-spiked samples were used. In addition to the pooled breast milk samples, standard reference material (SRM 1954) from the National Institute of Standards and Technology (NIST, USA) were spotted and processed in parallel following the same protocol. Five replicates were extracted for each sample type.

An Agilent 1290 Infinity II LC connected to a SCIEX QTrap 7500 system with a heated electrospray ionization source (ESI, OptiFlow Pro) was used. The LC-MS method was recently developed and validated in-house and detailed information is available at Gu et al.³⁷ In short, data was obtained in scheduled multiple reaction monitoring mode (sMRM) using a previously validated method. MRM transitions and RT are described in Table S2. Chromatographic separation was achieved using a Waters Acquity HSST3 (100 mm × 2.1 mm, 1.8 μm) paired with a Waters VanGuard Acquity HSST3 precolumn (50 × 2.1 mm, 1.8 μm). The column compartment was kept at a temperature of 40°C, and the autosampler was maintained at 7°C. The injection volume was 5 μL and the flow rate was set to 0.4 mL/min. The mobile phases used were water + 0.3 mmol/L NH₄F (eluent A) and acetonitrile (eluent B). Details of the settings, including chromatographic gradient and ion source settings can be found in Table S3.

Several quality control (QC) measures were applied: isotopically labeled internal standards were added to all extraction solutions (¹³C₁₈-zearalenone, 0.3 ng/mL) and to the reconstitution solution (¹³C₁₂-bisphenol A, 2 ng/mL) for controlling procedural/instrumental variability (Table S4). QC samples were prepared by extracting additional non-spiked samples, pooled after reconstitution. For the method optimization, five samples of each extraction solutions tested were merged. For both the stability study and the comparison between DMS and Mitra samples, a new set of five freshly prepared DMS samples was employed. Background contaminant levels were evaluated for working solutions (extraction and reconstitution), process blanks (extraction protocol without any matrix) and paper blanks (filter papers without breast milk undergone extraction protocol).

Pre-spiked DMS were prepared and extracted in the same manner as described above. In summary, 2450 µL of milk were spiked with 50 µL of the stock solution mix to produce DMS spiked at concentration “Level 20”. Twelve independent sample sets (paper cards), each containing three samples and one blank spot were prepared on the same day. Four of these sets were allocated to each tested temperature condition for evaluation at the 2-month time point. The remaining sets were stored at -80°C until needed. To ensure all samples were extracted and analyzed simultaneously, additional sets were moved from -80°C to the respective test temperatures either 2 weeks or 2 days before the end of the 2-month period. This allowed all samples to complete the intended duration by the final time point. In addition, one reference sample set with a larger number of replicates (2 paper cards containing 8 samples and 2 blanks) was stored continuously at −80°C for 2 months. In total, 14 paper cards were prepared, containing 34 samples and 14 blank spots, and used for the stability assessment. Extraction of the DMS was carried out using the final extraction solution ACN/MeOH/H₂O (40:40:20 v/v).

The performance of DMS and Mitra tips (Trajan Scientific Americas Inc.) was compared in terms of extraction recovery, matrix effects (SSE, signal suppression/enhancement), and blank contamination. A total of five pre-spiked 10 µL-Mitra sticks and 10 µL-DMS were prepared using pre-spiked milk. The same number of replicates were prepared for post-spiked samples by reconstituting the residue of non-spiked pooled samples in a spiked reconstitution solution at the same nominal concentration (Level 8, see Table S1). The Mitra sticks were prepared by briefly contacting the surface of the milk with the sampler tip for 8 s without complete submersion, as recommended by the vendor. The DMS were prepared by pipetting 10 µL of pooled human milk onto the Whatman cards. Both sample types were allowed to dry for 3 h at room temperature and protected from direct light. The final extraction solution (ACN/MeOH/H₂O 40:40:20 v/v) was used for sample preparation. For Mitra sticks, the tip of the sampler was inserted into tubes and treated as previously described.

Data analysis was performed using SCIEX OS (v3.0). First, automatic peak integration was performed using AutoPeak algorithm, followed by visual inspection of possible missing peaks, wrong peak assignment or peak integration of false positives. Next, calibration curves were built based on linear regression with 1/x weighting. Average Peak areas of samples spiked before and after extraction were used to calculate analyte recovery and matrix effect.³⁸ Compound recoveries were assessed based on the ratio of average peak area of pre- and post-spiked samples (n = 5 each). SSE was calculated as the ratio of average areas between post-spiked average and solvent standard at same spiking level. Average peak areas of non-spiked samples were subtracted from both pre- and post-spiked samples to minimize the effects of previously present xenobiotics. All figures of merit were calculated in Excel 16.0.

Concentrations of detected xenobiotics in both pooled Austrian breast milk and SRM 1954 were calculated by external calibration in solvent. For compounds detected in the reconstitution solution (used for sample reconstitution and serial dilution of the calibration curve) standard addition approach was used to calculate concentration. Next, the theoretical concentrations for the calibration curve were obtained by summing the concentration of the reconstitution solution to the original values. Compound concentrations were then calculated based on the corrected calibration curve. Finally, compound concentrations were corrected for previously estimated recoveries and matrix effects.

Samples spiked before extraction were used for the estimation of the methods limit of detection (LOD). The choice of pre-spiked samples was intended to accurately represent the method performance, i.e., considering recoveries and matrix effects. Signal-to-noise ratios (S/N) were obtained using SCIEX OS software with MQ4 algorithm for peak integration and Peak to Peak algorithm for computing S/N. For this purpose, a noise region was defined for every analyte as the remaining region within the MRM window that did not contain the peak of interest. Mean S/N were determined to calculate the ratio corresponding to a S/N of 3 to estimate the LOD. The respective concentrations of the individual compounds were then used to determine the actual concentrations at which an S/N of 3 was reached.

For a preliminary stability assessment, compounds with RSD > 30% in the −80°C reference samples (measured throughout data acquisition) were excluded in order to ensure data reliability, retaining a total of 137 out of 216 compounds for subsequent statistical analysis. This approach was favored over employing RSD calculations on QC samples since only a few compounds were detected in the non-spiked QC samples used in this study, limiting RSD calculations for all compounds. Initially, principal component analysis (PCA) was performed for exploratory data visualization using autoscaled data. For assessing compound stability, across all storage temperatures (−20°C, 4°C, 18°C and 37°C) and time points (2 days, 2 weeks, 2 months), independent temperature-based one-way ANOVA tests were performed, which included the reference group (−80°C) against all time points, for each temperature. The data was log10-transformed (no scaling) and both parametric and non-parametric one-way ANOVA (Kruskal–Wallis test) were performed. Compounds presenting a P-value < 0.05 after false discovery rate (FDR) correction in both tests were considered to be significant (ie, not fully stable). This approach was taken in order to ensure robustness of results in a low-sample scenario. Finally, a spearman rank correlation between peak areas and time profile within each tested condition was performed in order to assess the most prominent time-dependent patterns (correlation coefficient > 0.9, with either positive or negative correlation). All analyses were performed in MetaboAnalyst 6.0 using raw peak areas as input.

A proof-of-principle analysis was performed to demonstrate the suitability of the method for the detection of a variety of xenobiotics in DMS. In total, 30 compounds were detected in all replicates of the SRM 1954 pooled samples, while 22 of those were present in the Austrian pooled sample (Table 1), including air pollutants (cotinine), plastic-related chemicals (phthalates and bisphenols), flame retardants (TBBPA and TCBPA), perfluoroalkyl substances (PFOA and PFOS), personal care products ingredients (parabens) and drug-related compounds (acetaminophen and fluconazole). Reported concentrations were corrected for recovery and matrix effects and blank subtracted if necessary. Both sample types showed similar detection profiles for plastic components, though concentrations varied. The majority of these compounds were phthalates with diethyl phthalate (DEP) being the most abundant in both samples. In addition to the phthalates, N-butylbenzenesulfonamide was also consistently detected, while BPS was only present in the Austrian pooled milk samples. When compared to the ranges reported by Jamnik et al.,¹⁴ BPS concentrations in this study are approximately tenfold higher, whereas the N-butylbenzenesulfonamide levels fall within the reported ranges.

Personal care product related compounds were the second most abundant group in both milk types. Among these, five parabens were quantified with methylparaben showing the highest concentrations: 11 ng/mL in SRM 1954 and 9.7 ng/mL in the Austrian pooled milk. Overall, paraben levels were found to be higher in the SRM 1954 samples. This observation is supported by the work of Iribarne-Durán et al.,⁴⁴ who reported that parabens concentrations tend to be higher in human milk samples from the United States compared to those from Europe. In addition to the parabens, 4-methylbenzophenone was detected in both sample types and triclosan was exclusively detected in SRM 1954.

For phytoestrogens, genistein was present in both samples, while daidzein and glycitein were only detected in SRM 1954. Since phytoestrogens are secondary plant metabolites, their levels in the human body are strongly diet-related, explaining the varying ranges detected.⁴⁵

Two pharmaceutical drugs were identified, namely acetaminophen and fluconazole. Acetaminophen was detected in the SRM 1954 at a concentration approximately 1000 times higher than in the Austrian pooled milk. For the Austrian samples, the detected concentration was below the lowest standard, so the value needs to be interpreted with caution. Acetaminophen glucuronide, a major metabolite, was only detected in the SRM 1954, which is consistent with the substantially higher concentration of the parent compound.

In SRM 1954 samples, fluconazole levels exceeded the upper limit of the calibration curve (6 ng/mL), preventing accurate quantification, whereas the concentration of the Austrian pooled milk sample was calculated with 0.075 ng/mL. Fluconazole is a widely used antifungal, with reported breast milk concentration up to 2.9 ng/mL in breast milk 2h after a single 150 mg oral dose.⁴⁶ It remains the most frequently prescribed outpatient antifungal in the USA (over 17 million prescriptions annually)⁴⁷ and is similarly prevalent in Europe.⁴⁸ While usage data for lactating women are limited, there is documented evidence of fluconazole being used in this population to treat thrush infections.^49,50

Two out of seven analyzed PFAS could be detected in both sample types, namely PFOS and PFOA. For both substances, concentrations were higher in the SRM 1954 samples, with concentrations at 0.14 ± 0.013 ng/mL and 0.26 ± 0.10 ng/mL for PFOA and PFOS, respectively. The PFOA concentration aligns closely with the consensus values reported by Keller et al. (0.13 ± 0.04 ng/mg), whereas the PFOS concentration is somewhat elevated compared to the reported consensus (0.16 ± 0.03 ng/mg), however, the values overlap within the respective standard deviations.⁵¹ In the Austrian milk pool, concentrations of 0.080 ± 0.010 ng/mL (PFOA) and 0.17 ± 0.064 ng/mL (PFOS) were determined. These levels fall within the ranges reported by Hartmann et al.⁵² who analyzed PFAS in 40 Austrian breast milk samples collected between 2013 and 2016, with concentrations up to 0.08 ng/mL for PFOA and 0.31 ng/mL for PFOS.

Among the analyzed flame retardants, tetrabromobisphenol A (TBBPA, 6.5 ± 3.8 ng/mL) and tetrachlorobisphenol A (TCBPA, 3.7 ± 1.6 ng/mL) were detected exclusively in the SRM 1954 samples. Notably, 12 halogenated phenolic compounds were spiked into SRM 1954 at a concentration of 0.5 ng/mL,⁵³ which may have included TBBPA and TCBPA. However, the measured concentrations significantly exceeded this spiking level, suggesting additional exposure sources. These elevated concentrations may reflect regional differences in environmental exposure, industrial application or regulatory policies concerning flame retardants. In 2024, the European Food Safety Authority (EFSA) published an updated scientific opinion on TBBPA and its derivates in food, establishing a tolerable daily intake (TDI) of 0.7 µg/mL body weight per day.⁵⁴ Although this TDI is not legally applicable in the United States, it provides a health-based benchmark for assessing potential risks. Based on this value, the TBBPA concentration in SRM 1954 would not be expected to result in infant exposure exceeding the TDI.

Cotinine, a biomarker of nicotine exposure, was detected at 0.58 ng/mL in SRM 1954 and 0.13 ng/mL in the Austrian pooled milk, while its metabolite trans-3-hydroxycotinine was only found in SRM 1954. Cotinine levels for passive smokers were previously reported as around 17 ng/mL and 54 ng/mL.^55,56 The cotinine levels observed in our study are approximately two orders of magnitude lower than those reported for passive smokers, suggesting minimal environmental tobacco exposure among the milk donors. Considering that the SRM 1954 was obtained from six milk banks across the USA,⁵³ our finding is consistent with the donor screening policies of the Human Milk Banking Association North America, which excludes individuals who smoke or use tobacco products.⁵⁷ Concentrations for all detected compounds are displayed in Table 1, with illustrative MRM chromatograms depicted in Figure 2 Detailed results on the calculation of concentrations for each compound is depicted in Table S7.

To assess analyte stability under different storage conditions, samples stored at –80°C for 2 months were used as reference, since all breast milk samples were initially kept at this temperature before transfer to the respective test conditions. In addition to –80°C, storage at –20°C, 4°C, and 18°C was evaluated, as these conditions are more practical for routine use. Demonstrating sufficient stability under such conditions would reduce reliance on ultra-low freezers and increase the feasibility of large-scale studies. To mimic further extreme conditions for in-field sample collection and storage, an additional set of samples was stored at 37°C.

Compounds were considered to be fully stable if FDR-corrected P-value > 0.05 for both parametric and non-parametric one-way ANOVA, aiming to ensure the robustness of results. In total, 137 (100%), 96 (70%), 101 (74%) and 86 (63%) compounds fulfilled these criteria for –20°C, 4°C, 18°C and 37°C, respectively, first demonstrating no significant degradation at –20°C for the entire compound panel at all storage times tested. PCA analysis revealed a trend towards more pronounced separation from –80°C samples (ie, larger differences in either PC1 or PC2) with increasing temperature, especially for 2 months at non-refrigerated conditions (18°C and 37°C), as observed in Figure S8.1. To determine which groups in specific presented a significant difference against −80°C-samples, a Tukey’s HSD post-hoc test was conducted for the parametric ANOVA. Around 80% of all compounds stored at 4°C, 18°C and 37°C remained stable when stored for up to 2 days, while up to 70% of the compounds remained stable after 2 months at 37°C.

In order to highlight the more robust findings (considering the relatively small sample size used in the statistical analysis), we next focused on key compounds/classes based on the overlap between the 10 lowest raw P-values and spearman rank correlation >0.9 against the time profile for 18°C (with either positive or negative correlation). This temperature was chosen as the most typical storage/transport condition for dried matrix spots. Parabens were the most prevalent class considering both criteria, with isobutyl-, benzyl- and heptylparaben, alongside with mono-2-ethylhexyl phthalate. Butylparaben also presented a high correlation (and significance as 12^th lowest P-value), while short-chain parabens showed no significant changes over time or across temperature. Figure 3 highlights the consistent decreasing trend for storage at 4°C, 18°C and 37°C for long-chain parabens.

Mono-2-ethylhexyl phthalate, on the other hand, demonstrated an increasing trend, possibly due to migration from the packaging material or due to degradation of its parent compound di-2-ethylhexyl phthalate (DEHP).⁵⁹ Apart from the parabens, no structural patterns could be observed with regard to chemical class or functional groups. To further investigate potential predictors of instability, we evaluated the available information on biological half-life for several analytes. Urinary half-lives have been reported for methyl-, ethyl-, and propylparaben as well as triclosan,⁶⁰ among these, triclosan has the longest half-life but was found to be unstable. Similarly, TBBPA and avobenzone were classified as unstable in DMS, despite having longer biological half-lives than the previously mentioned analytes.^61,62 Together, these findings indicate that in-vivo half-life is not a suitable predictor of stability in DMS. A complete overview of the results obtained in the performed statistical analysis is provided in Table S8.

To the best of our knowledge, no previous work has evaluated the feasibility of Mitra devices for the analysis of human breast milk. Their analytical performance was compared to DMS by assessing recovery and matrix effects at a lower spiking level (nominal 8 versus 23 in the previous experiment). For consistency, 10 µL of milk was applied to both Mitra tips and DMS.

Analyte recovery could be assessed for 196 (DMS) and 195 (Mitra) compounds. For both sampling methods, nearly 50% of the reported compounds were found to be within an acceptable recovery range of 42%–134%, compared to 80% in the previous experiment, likely due to the reduced volumes and concentrations employed (see Table S9 for detailed results). As shown in Figure 4, recoveries were similar across the two sampling methods and varied primarily in an analyte-dependent manner. For aflatoxins, differences in recovery between filter paper and Mitra tips, were generally within 20%, with the exception of AFP₁ and AFQ₁, which showed greater variability. A similar trend was observed for bisphenols, with recovery differences not exceeding 20% for the majority of compounds (BPB, BPE and BPF). In contrast, BPFL, BPM and BPS exhibited poor recovery and greater variability. PFAS recoveries were likewise consistent across both sampling types (DMS x Mitra): PFOS (103 ± 18% × 97 ± 18%), PFOA (118 ± 10% × 107 ± 19%), PFHxA (94 ± 7% × 95 ± 16%) and PFNA (109 ± 16% × 108 ± 17%). Phthalates were largely comparable and showed satisfactory recovery results across both sampling types. For diethyl phthalate and monobenzyl phthalate, DMS showed better performance with recoveries of 106 ± 53% and 114 ± 20%, respectively, compared to 200 ± 57% and 181 ± 27% obtained with Mitra tips. Matrix effects were evaluated for 192 (DMS) and 194 (Mitra) compounds. About 30% of the compounds were within the acceptance range of 60%–140%, compared to 50% in the previous experiment. Aflatoxins consistently exceeded the acceptance range for both DMS and Mitra tips, with partly high variance, particularly for AFG₂, AFM₁ and AFQ₁. Bisphenols showed a heterogeneous behaviour with BPA, BPAF, BPB and BPF being within the acceptance range for both DMS and Mitra tips. In contrast, BPFL and BPM showed notable signal suppression, while BPS was more affected by signal enhancement. All PFAS, except for PFBS, fell within the acceptance range, including PFOS (107 ± 15% x 106 ± 14%), PFOA (135 ± 15% x 141 ± 20%) and PFDA (103 ± 12% x 109 ± 13%), for DMS and Mitra tips, respectively. Background levels of contaminants were assessed as in the first experiment (see section Background contamination in paper substrate) for blank paper substrate, working solutions (extraction- and reconstitution solution), as well as for blank Mitra tips. In general, there was a good agreement in calculated concentrations between the first and second experiment for the paper substrate background. The majority of compounds that presented substantial concentration difference in the paper blank between both experiments (mostly lower in second experiment) were mainly originating from the extraction process and/or analytical system rather than the paper substrate itself, evidenced by the fact that a similar trend was observed in the process blanks levels. A total of 19 compounds were detected in the Mitra tips, which largely demonstrated lower concentrations compared to filter paper. Compounds such as methylparaben, and triphenyl phosphate were present at significantly higher levels in the filter paper, with concentrations up to 10x greater than in the Mitra tips. Results are summarized in detail in Table S10 along with previously reported values by Hernandes et al.²⁸