Application of Non-Targeted Analysis for the Identification of Novel Environmental Contaminants in Amniotic Fluid

Abstract

Background

Amniotic fluid (AF) provides a direct window into the prenatal exposome, yet it remains understudied despite the heightened vulnerability of the developing fetus to chemical and biological perturbations. Characterizing both endogenous metabolites and exogenous contaminants in AF is essential for understanding fetal chemical exposures.

Objective

To profile endogenous and exogenous chemical features in second-trimester AF and assess potential interactions between environmental contaminants and fetal metabolic pathways.

Methods

Non-targeted analysis (NTA) using liquid chromatography–high-resolution mass spectrometry (LC-HRMS) was applied to AF samples (n = 40) collected from participants in the San Francisco Bay Area. Compound annotation involved library matching and evaluation of diagnostic fragments. Correlation network analysis was performed to identify associations between exogenous chemicals and endogenous metabolites. Kendrick mass defect (KMD) analysis was used to distinguish chemical classes and prioritize unknown features for structural interpretation.

Results

A total of 290 compounds were annotated, including ten confirmed environmental contaminants. Nine of these ten contaminants, i.e., 1,3-diphenylguanidine, N,N-dimethyltetradecylamine-N-oxide, quinoline, phthalic anhydride, acetyl tributyl citrate, 4-hydroxybenzenesulfonic acid, lauryl sulfate, 4-dodecylbenzenesulfonic acid, and dibutyl adipate, are reported in AF for the first time. These chemicals spanned surfactants, plasticizers, and industrial intermediates, indicating diverse anthropogenic sources. Correlation analysis revealed links between several contaminants (e.g., tetradecyl sulfate, phthalic anhydride, quinoline) and endogenous lipid metabolites, suggesting possible metabolic disruption. KMD analysis further differentiated chemical classes and showed that many unknown features exhibited KMD signatures consistent with exogenous esters and amine/amide-containing compounds.

Significance

This study demonstrates the power of NTA to reveal previously undocumented prenatal exposures and uncover potential interactions between environmental chemicals and fetal metabolic processes. These findings expand our understanding of the prenatal exposome and highlight the need for continued investigation into environmental influences on fetal health.

Impact

This study presents the first application of LC-HRMS non-targeted analysis to human amniotic fluid, uncovering nine environmental contaminants that have never previously been reported in this matrix. These newly detected surfactants, plasticizers, and industrial chemicals reveal a wider range of fetal exposures than currently recognized. By expanding the chemical landscape of the prenatal environment, this work highlights the importance of NTA for discovering emerging contaminants and underscores the need to evaluate their potential impacts on fetal development.

Introduction

Pregnancy is marked by extensive metabolic and physiological remodeling to support fetal growth and maintain maternal homeostasis ¹. Exposure to environmental chemicals during pregnancy can disrupt this natural metabolic trajectory, interfering with processes such as lipid oxidation, hormone synthesis, and nutrient transfer ². Current biomonitoring efforts capture only a small fraction of the total exposome. For example, the U.S. National Health and Nutrition Examination Survey (NHANES) routinely measures fewer than 400 chemicals, whereas the U.S. Environmental Protection Agency’s CompTox Dashboard lists over 1.2 million chemicals of environmental relevance. Notably, less than 2% of these chemicals are available as analytical standards, substantially limiting the ability to identify and quantify unknown exposures ³.

Mid-pregnancy, a period of rapid fetal growth and placental maturation, may be particularly sensitive to such perturbations ⁴. During this stage, amniotic fluid (AF) reflects both maternal and fetal metabolic activity and can serve as a window into early biochemical disruptions caused by chemical exposures ^{5, 6}. AF is a complex and dynamic biological matrix that surrounds the fetus throughout gestation, serving as both a protective cushion and a medium for biochemical exchange between the mother, placenta, and fetus ^{7, 8}. Compared with more commonly studied matrices such as maternal blood and urine, AF provides a more direct reflection of the fetal metabolic state and intrauterine environment ^{6, 9, 10}.

Non-targeted analysis (NTA) using liquid chromatography HRMS (LC-HRMS) has been widely applied to maternal blood and urine, cord blood, and breast milk, which has enabled the characterization of the metabolome ^11-14 and exposome ^15-17 and facilitated the identification of novel endogenous metabolites and environmental chemicals. AF has received little attention in exposomic research. To date, only one comprehensive gas chromatography HRMS (GC-HRMS) effect-directed study has examined the nonpolar fraction of full-term human AF (pooled samples collected at amniotomy from four healthy pregnancies) ¹⁸. Using combined targeted and non-targeted workflows, this study profiled nonpolar endocrine-disrupting chemicals (EDCs), detecting 42 known nonpolar EDCs and 69 prioritized suspected EDCs, of which 12 were biologically active and 9 were chemically confirmed in AF ¹⁸. Therefore, a comprehensive LC-HRMS–based exposomics approach has the potential to identify xenobiotics that cross the placenta, and to elucidate how these compounds accumulate in the fetal compartment and potentially interfere with endogenous metabolic pathways.

In this study, we applied LC-HRMS–based NTA of AF collected during mid-pregnancy to pursue two main objectives: (1) uncover previously unknown or lesser-known exogenous environmental chemicals; (2) explore the potential relationships between exogenous chemicals and endogenous metabolites.

Materials and Methods

Collection of Amniotic Fluid

AF samples (n=40) were collected during the second trimester at UCSF Hospital, which serves women living in the general San Francisco Bay Area, CA during amniocentesis procedures performed for clinical indications such as the assessment of trisomy or aneuploidy associated with advanced maternal age. The samples were transported on dry ice to the Environmental Chemistry Laboratory at the Department of Toxic Substances Control, California Environmental Protection Agency, in Berkeley, California. Upon arrival, they were stored at −20 °C until sample preparation and chemical analysis.

Sample Preparation and Instrumental Analysis

Frozen amniotic fluid samples were fully thawed at room temperature (~20 °C). The samples were randomly divided into four batches, each containing 10 individual samples. From each sample, 250 μL was aliquoted for individual analysis, and 100 μL from each was combined to create a pooled quality control (QC) sample. Each sample was spiked with 25 μL of a 500 ng/mL solution containing five isotope-labeled internal standards (ISTDs, details provided in Supplementary Spreadsheet S1). Individual and pooled samples were extracted with 2 mL and 4 mL of methanol containing 0.1% formic acid (precooled at −18°C), respectively. The mixtures were shaken at 1200 rpm for 15 minutes using an Analog Multi-Tube Vortexer (Fisher Scientific, NH), followed by centrifugation at 3000 rpm for 20 minutes using a Sorvall X4 Pro-MD Centrifuge (Thermo, MA). The supernatants were collected and concentrated under a gentle stream of nitrogen (N₂) to a volume of less than 0.5 mL. The concentrates were then transferred to 1.5 mL microcentrifuge tubes, dried under N₂ to near dryness, and reconstituted in 150 μL (individual samples) or 250 μL (pooled samples) of methanol: water (1:1, v:v). Finally, the reconstituted extracts were centrifuged again for 20 minutes, and the supernatants were transferred to autosampler vials for instrumental analysis.

Sample analysis was performed using an Agilent 1290 Infinity UHPLC system coupled to an Agilent 6550 quadrupole time-of-flight mass spectrometer (Q-TOF-MS, Agilent Technologies, CA). Each sample was injected twice and analyzed both in ESI+ and ESI− modes. Chromatographic separation was achieved on a reversed-phase EclipsePlus C18 column (2.1 × 100 mm, 1.8 μm particle size, Agilent Technologies) maintained at 40°C. A binary solvent system was used, consisting of water (mobile phase A) and methanol (mobile phase B), each containing 5 mM ammonium acetate. The gradient program initiated at 10% B, linearly increased to 100% B over 15 minutes, and was followed by a re-equilibration period from 15 to 26 minutes. The flow rate was maintained at 0.3 mL/min throughout the run. Detailed LC gradient parameters and Q-TOF tuning settings are provided in Supplementary Spreadsheet S2.

For MS¹ profiling, all individual and QC pooled samples were analyzed in MS-only mode across an m/z range of 50–1200. QC pooled samples were injected periodically within the analytical batch: immediately after the instrument blank, and at the midpoint and end of the sequence for batch effect correction. MS² spectra were acquired exclusively from QC pooled samples using Auto MS/MS mode. To enhance spectral coverage, precursor ion m/z windows were segmented into six overlapping ranges: 60–180, 170–300, 290–450, 440–600, 590–900, and 890–1200 Da. For Auto MS/MS acquisition, the full scan ranges were set to m/z 50–1200 for MS1 and m/z 25–1200 for MS/MS. Features not captured by the Auto MS/MS acquisition were subsequently analyzed using targeted MS/MS with an inclusion list.

For quality assurance (QA) and quality control (QC), all glassware and pipettes were baked at 450°C prior to use. After extraction, samples were immediately analyzed using the instrument. Each analytical sequence began with pooled QC samples, with additional pooled QC samples placed in the middle and at the end of the sequence to support retention time (RT) correction and batch correction of abundance data. For matrix-spiked QC samples, AF QC materials were prepared by pooling excess samples from UCSF, as no certified reference material is currently available. We spiked the pooled AF with 17 analytical standards (7 in positive mode and 10 in negative mode) at two concentrations (100 ng/mL and 400 ng/mL). These matrix-spiked QC samples were analyzed in the middle and at the end of each sequence to monitor instrumental sensitivity and stability. The AF QC material was also used as a laboratory blank and run immediately after the instrument blank at the beginning of each batch. For each individual sample, although ISTDs cannot represent all ions detected in NTA, they were used to monitor the consistency of the extraction procedure within each batch. It should be noted that the matrix-spiked QC samples were analyzed only in full-scan (MS¹) mode. The measured abundances of the spiked analytes in the QC matrix samples and the ISTDs in individual samples were checked using Agilent MassHunter Qualitative Analysis 10.0 with a mass accuracy tolerance of 10 ppm and are provided in Spreadsheets S3 and S4.

Non-Targeted Analysis (NTA)

The complete workflow, organized by subchapter, is illustrated in Figure 1. Data analysis was conducted using R (version 4.4.2) and Python (version 3.12.3) on the Windows platform, within the RStudio (version 2024.12.0) and Spyder (version 5.0.6) environments. All R/Python packages used are summarized in Spreadsheet S5.

Figure 1.

Overview of the NTA workflow for amniotic fluid sample processing and data analysis.

Feature Alignments, Data Imputation and Batch Effect Correction

The raw data files were imported using the MSnbase R package (version 2.23.0) and processed for alignment with the XCMS R package (version 4.4.0). Detailed alignment parameters are provided in Spreadsheet S5. Briefly, the centWave algorithm was applied with a mass accuracy of 20 ppm, a peak width of 5–30 seconds, a signal-to-noise threshold of 6, a minimum noise level of 500, and a prefilter requiring at least 3 peaks with a minimum intensity of 3000.

Within each batch, peaks were initially grouped, and RT correction was performed using the Obiwarp algorithm. Cross-batch alignment was carried out using pooled QC samples and the PeakGroupsParam function, which applies a robust fit based on a symmetric family and a smoothing span of 0.5. Features were retained if they were either (a) absent from all blank samples, or (b) present with a sample-to-blank intensity ratio exceeding 2. Adduct annotation was performed by the CAMERA R package (version 3.21).

The aligned dataset across samples was first filtered using a detection frequency threshold of 70%. Missing values (i.e., those below instrumental background noise) were imputed using the k-nearest neighbors (k = 5) algorithm with Euclidean distance as the metric. After imputation, batch effects were corrected in two steps: (1) intra-batch correction using QC-pooled samples with LOESS normalization, where within each batch a LOESS curve was fitted to the QC samples as a function of injection order and used to remove systematic drift, centered on the median QC intensity, followed by z-score normalization of each feature within batch; and (2) inter-batch correction across four batches using the limma::removeBatchEffect function in R ¹⁹, which adjusts feature intensities by fitting a linear model with batch as a covariate and removing the estimated batch-associated component. The corrected data were then used for downstream analyses.

Features Annotation and Identifying Underreported Chemicals

For endogenous metabolites, the MS/MS spectra were first annotated using the online version of the metabolic reaction network (MRN)-based recursive algorithm MetDNA (http://metdna.zhulab.cn/). This tool identifies initial seed metabolites using a small library of MS² spectra and then utilizes their experimental MS² spectra as surrogate spectra to annotate their reaction-paired neighbor metabolites ²⁰. Additional annotations were performed by comparing results with the Human Metabolome Database (HMDB, https://www.hmdb.ca/downloads).

For exogenous chemicals, we compared the spectra against the MassBank of North America and the MassBank of Europe. Spectral matching against reference databases was carried out using the Python package matchms (v0.28.1). Experimental MS/MS spectra were preprocessed through intensity normalization and m/z filtering (50–1000 Da). Spectral matching was then performed using cosine similarity (threshold > 0.7) with a 5-ppm precursor mass tolerance. Annotations were assigned based on the highest-scoring reference spectrum.

Unknown features were annotated using in silico fragmentation with the Java command-line tool MetFrag (v2.6.5), which was run via Python. Precursor m/z values from experimental MS/MS spectra were searched against the PubChem, HMDB, and KEGG databases using a 5-ppm mass tolerance. Fragmentation patterns were scored with the FragmenterScore algorithm, applying a 5.0 ppm fragment mass deviation. Candidate structures for each feature were manually evaluated, and the structure with the highest score was selected. When no plausible structure was identified using MetFrag, the prediction tool CFM-ID (https://cfmid.wishartlab.com) was manually consulted for MS/MS spectrum interpretation.

Chemical identifications and annotations were ranked based on the system proposed by Schymanski, Jeon et al ²¹.

The various confidence levels are as follows:

Level 1, structure confirmed by a chemical standard with MS/MS and retention time (RT) matching;

Level 2, probable structure assigned by spectral library matching with at least three informative fragment ions;

Level 3.1, tentative structure supported by limited MS/MS evidence (fewer than three informative fragment ions) or partial spectral matching without in silico ranking;

Level 3.2, tentative structures supported by in silico candidate ranking with MS/MS similarity scores ≥ 90%, in the absence of authentic standards or high-confidence library matches;

Level 4, an unequivocal molecular formula can be assigned based on the spectral information but there is not enough information to propose possible structures;

Level 5, only exact mass (m/z) with insufficient information to assign a formula.

Level 4 and 5 features were not considered in the current study.

We queried the confirmed exogenous contaminants in the Blood Exposome Database (BEDB) to identify chemicals that are not widely reported (i.e., reported in only a few publications) or have not been previously documented. To further investigate the sources, uses, toxicological information, and reported presence of these compounds in serum, urine, or AF, we conducted systematic searches across the CompTox Chemicals Dashboard, PubChem, Wikipedia, Google Scholar, ScienceDirect, and ACS Publications. Each compound name and its synonyms (including CAS number and InChIKey) were used as search keywords, combined with terms such as “human exposure,” “toxicity,” “biomonitoring,” and specific matrices (e.g., “serum,” “urine,” “amniotic fluid”). Information from multiple sources was cross-referenced and compiled to identify consistent evidence for each compound. When conflicting or multiple sources were found, priority was given to peer-reviewed studies and authoritative databases.

Association Analysis Between Exogenous and Endogenous Compounds

A correlation network was constructed to identify associations between confirmed exogenous chemicals and endogenous metabolites present in >70% of samples. Spearman rank correlation coefficients were calculated between the abundances of exogenous and metabolomic features using the Hmisc R package. All features with a Benjamini–Hochberg (BH)-adjusted p-value < 0.05 were considered statistically significant. Among these, we focused on correlations with an absolute coefficient (∣r∣) ⩾ 0.7 for further analysis. The resulting correlation matrix was filtered to retain only exogenous–metabolite interactions. Significant correlations were visualized as an undirected network using the igraph R package. In the network, edge weights represented the absolute strength of the correlation, edge colors indicated statistical significance, and node sizes were scaled by degree of connectivity.

Kendrick Mass Defect (KMD) Analysis

KMD analysis was performed using a CH₂-based Kendrick mass transformation to evaluate homologous relationships and characterize chemical class patterns among confirmed contaminants, annotated exogenous chemicals, annotated endogenous metabolites, and unknown ions. Exact masses for all detected features were converted to Kendrick masses and KMD values, which were visualized as KMD versus m/z plots for each compound category. Assigned adducts and isotope patterns from preprocessing were retained to ensure accurate mass interpretation. KMD distributions were compared across chemical classes to identify clustering behavior and overlap between known and unknown features, enabling rapid assessment of potential structural analogs and prioritization of unknowns for downstream identification.

Semi-quantitative Analysis of Confirmed Chemicals

After confirmation of chemicals using authentic standards, an external standard calibration approach was applied. Six concentrations of standards (20, 50, 100, 150, 250, and 500 ng/mL) were analyzed together with the individual samples across two batches. Batch-specific calibration curves were constructed and used to estimate the semi-quantitative concentrations of each confirmed chemical in each AF sample.

Results

Alignment, annotation, and confirmation

In total, 983 features were aligned in the ESI+ mode and 1,299 features in the ESI⁻ mode after applying a 70% detection frequency cutoff. Following annotation, a total of 132 (ESI⁺) and 159 (ESI⁻) features were assigned to confidence levels 1–3 (Spreadsheets S6 and S7). Among these, 31 features in ESI⁺ and 72 features in ESI⁻ were annotated as endogenous metabolites. In total, ten compounds were confirmed at Level 1 in the ESI⁺ mode and six compounds at Level 1 in the ESI− mode (examples of chromatogram are shown in Figures S1-S10, and MS² data for confirmation is shown in Spreadsheet S8). The compounds identified in ESI+ mode were creatinine, caffeine, L-tryptophan, 1,3-diphenylguanidine (DPG), N,N-dimethyltetradecylamine N-oxide, mono(2-ethylhexyl) phthalate (MEHP), triphenyl phosphate (TPP), quinoline, phthalic anhydride, and acetyl tributyl citrate (ATBC). In ESI– mode, the confirmed compounds included phenylalanine, L-tryptophan, 4-hydroxybenzenesulfonic acid, lauryl sulfate, 4-dodecylbenzenesulfonic acid, and dibutyl adipate. Notably, L-tryptophan was confirmed in both ionization modes.

Among the confirmed compounds, ten were identified as environmental contaminants, including DPG, N,N-dimethyltetradecylamine N-oxide, MEHP, quinoline, phthalic anhydride, ATBC, 4-hydroxybenzenesulfonic acid, lauryl sulfate, 4-dodecylbenzenesulfonic acid, and dibutyl adipate (detailed in Table 1). Although TPP can be considered an environmental contaminant, this feature was also detected in the laboratory control blanks. Because its levels could not be distinguished from those observed in the laboratory controls and AF samples, this compound was not included in our findings. To date, MEHP remains the only compound previously reported in AF, while the other compounds have not been previously reported, to the best of our knowledge.

Table 1.

Confirmed environmental contaminants detected in amniotic fluid, including chemical identity, CAS number, common uses or sources, key toxicological information, and evidence of prior detection in human blood or urine or amniotic fluid.

Chemical name	CAS No.	Use / Source	Key toxicological info	Previously detected in blood/urine	Previously detected in amniotic fluid
4-Hydroxybenzenesulfonic acid	98-67-9	A redox mediator in laccase-catalyzed indigo dye degradation Oilfield chemicals Metal working Polymerization catalyst	In vitro indicate estrogen and androgen receptor pathway activity (EPA ToxCast data).	Yes/No	No
4-Dodecylbenzenesulfonic acid	121-65-3	Surfactant Polymer/stabilizer	Irritant; no known developmental toxicity 23	Yes/No	No
Dibutyl adipate	105-99-7	Plasticizer	Mild irritant; low toxicity 24	Yes/Yes	No
1,3-Diphenylguanidine (DPG)	102-06-7	Tire/tread materials	Oral LD50 in rats ~1,020 mg/kg (EPA ToxCast data)	Yes/Yes	No
Phthalic anhydride	85-44-9	Make phthalic acid derivatives, plastics, resins, coatings	Organ toxicant 25	Yes/Yes	No
Acetyl tributyl citrate (ATBC)	77-90-7	Plasticizer	not genotoxic in bacterial/mammalian, not developmental/reproductive toxicant in mice/rats 26, 27	Yes/Yes	No
Quinoline	91-22-5	Intermediate in the manufacture of dyes, pharmaceuticals, and pesticides Synthesis of chelating agents, corrosion inhibitors, and flavoring substances	Carcinogen 28, 29	Yes/Yes	No
Lauryl sulfate	151-41-7	Surfactant in consumer cleaning/personal care products.	Generally low systemic toxicity but skin and eye irritation depending on concentration 26 Long-chain alkyl sulfates seem less irritating 30	Yes/No	No
N,N-Dimethyltetradecylamine-N-oxide	3332-27-2	Surfactant/foam stabilizer	Corrosive to skin/eyes (Chemicalbook data).	No/No	No
Mono(2-ethylhexyl) phthalate (MEHP)	4376-20-9	Metabolism of di(2-ethylhexyl) phthalate (DEHP, a plasticizer) used in PVC.	Known developmental/reproductive toxicant 31, 32 Endocrine disruptor 33	Yes/Yes	Yes

After searching the BEDB ²², we found all confirmed contaminants were included in the database (Spreadsheet S9). The information in BEDB also indicated only a small number of studies for 4-dodecylbenzenesulfonic acid (n = 1), 4-hydroxybenzenesulfonic acid (n = 1), N,N-dimethyltetradecylamine-N-oxide (n = 1), and dibutyl adipate (n = 2). All confirmed contaminants have been previously identified in serum, and only 4-dodecylbenzenesulfonic acid and lauryl sulfate have been detected in urine. This pattern is consistent with the biological origin of amniotic fluid, which more closely reflects circulating maternal and fetal serum than urinary excretion profiles.

Batch correction

For the batch effect, batch-wise alignment showed that in the PC1 vs. PC2 plots, both pooled QC and spiked AF samples formed distinct clusters, separated from individual samples in both ESI+ and ESI− modes (Figure 2). This indicates that pooled QC samples represent a different intensity distribution, suitable for RT and intensity correction across batches. After intra-batch correction of intensities using pooled QC samples, strong correlations (R² values) confirmed that QC-based correction effectively reduced systematic drift while preserving biological variability (Figure S11). Substantial batch effects (p < 0.01) were observed among the four batches in the raw aligned features for both positive (Figure 3A, B) and negative modes (Figure S12A, B), as well as after QC-based intra-batch correction (Figure 3C, D for ESI+ mode, and Figure S12C, D for ESI− mode). After inter-batch correction using the linear batch correction function, batch effects in PC1 were removed (p > 0.05) across all four batches for both ESI+ (Figure 3E, F) and ESI− modes (Figure S12E, F).

Figure 2.

The plot of PC1 vs PC2 from principal component analysis (PCA) for the alignment data for amniotic fluid (AF) samples, AF matrix spiked samples, and pooled QC samples for each batch where alignment was separately processed with the detection frequency cut of 70%.

Figure 3.

Principal component analysis (PCA) plots (PC1 vs. PC2) of aligned features in positive ionization mode from the raw dataset (A), after intra-batch correction using pooled QC samples (C), and after inter-batch correction using the limma removeBatchEffect function in R (E). A boxplot of PC1 values by batch is shown (B, D and F), with statistical significance assessed by the Mann–Whitney U test. Significance levels are indicated as follows: *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001, ****p ≤ 0.0001, and ns = not significant.

Network between environmental contaminants and endogenous metabolites

An association network, also known as a molecular interaction network, was constructed to explore potential interactions (Spearman correlation results, including p-values and R values, are shown in Spreadsheets S10 and S11) between endogenous metabolites and environmental contaminants (Level 1, Figure 4). Among the metabolites, 8-hydroxyoctanoic acid, FA 18:2+2O, 2-oxophytanate, and octanoyl-carnitine were shown to correlate with certain exogenous chemicals, which included tetradecylsulfate, quinoline, and phthalic anhydride. Notably, FA 18:2+2O and 8-hydroxyoctanoic acid showed to positively correlate with tetradecylsulfate (Spearman R = 0.731 and 0.709, respectively). Similarly, 2-oxophytanate positively correlated with phthalic anhydride (R = 0.734). Quinoline was negatively correlated (R = −0.722) to octanoyl-carnitine.

Figure 4.

Association network between environmental contaminants and endogenous metabolites (BH–adjusted p < 0.01; correlation R > 0.7; details in Spreadsheet S10, S11), highlighting endogenous metabolites and their potential metabolic pathways as mapped in the KEGG database.

KMD Pattern

We applied KMD analysis to explore potential similarities in patterns among confirmed, annotated, and unknown features across chemical classes. Confirmed compounds clustered within characteristic KMD ranges according to chemical class (Figure 5A). Sulfonates (4-hydroxybenzenesulfonic acid, lauryl sulfate, 4-dodecylbenzenesulfonic acid) were confined to −0.20 to −0.14. Esters (dibutyl adipate, MEHP, phthalic anhydride) grouped between −0.11 and 0.00. Other structures, including DPG and N,N-dimethyltetradecylamine-N-oxide, showed moderately negative KMD values (−0.137 to −0.125). Unknown features displayed a broader distribution (−0.45 to +0.49), with most between −0.25 and −0.05 and subsets extending to highly negative (≤ −0.30) or positive (> 0) values.

Figure 5.

Kendrick mass defect (KMD) plot of detected chemical features. Panels show confirmed contaminants (A), annotated contaminants (levels 2/3, B), endogenous metabolites (levels 1-3, C), and unknown chemical features (level 4, with assigned adducts and isotope patterns). Compounds are colored by chemical class: sulfonates (yellow), esters (blue), amides (pink), N-heterocycles (purple), and other groups (green). Unknown features are shown in gray.

Annotated exogenous chemicals (Level 2/3) clustered narrowly between −0.25 and −0.05, encompassing esters, N-heterocycles, amides, and sulfonates (Figure 5B). Unknown features were more widely dispersed (−0.42 to +0.49), with a major subset co-localizing with annotated compounds and additional high- and low-mass-defect outliers. Endogenous metabolites (Levels 1-3), including creatinine, L-tryptophan, L-phenylalanine, and lysophosphatidylcholines, clustered tightly within −0.25 to +0.05 (Figure 5C). Unknown features spanned the full KMD range (−0.49 to +0.49), with a core subset overlapping the metabolite cluster and extreme outliers at both high (> +0.3) and low (< −0.35) KMD values.

Overall, KMD plots show that confirmed and annotated compounds occupy defined mass defect regions, whereas unknown features exhibit broader distributions, with subsets either co-clustering with known chemical classes or forming distinct high- or low-KMD outliers.

Semi-quantification of confirmed chemicals

The semi-quantitative analysis (Spreadsheet S12) revealed the following concentration ranges (ng/mL) for each compound: phthalic anhydride: 0.13–11.98 (median = 2.09); 1,3-diphenylguanidine: 6.00–45.07 (median = 17.10); quinoline: 0.16–54.20 (median = 14.87); N,N-dimethyltetradecylamine-N-oxide: 14.4–98.24 (median = 40.65); mono(2-ethylhexyl) phthalate: 14.71–83.85 (median = 41.43); acetyl tributyl citrate: 0.87–94.54 (median = 42.12); lauryl sulfate: 33.06–86.05 (median = 50.05); 4-hydroxybenzenesulfonic acid: 0.47–86.82 (median = 41.44); 4-dodecylbenzenesulfonic acid: 6.03–113.51 (median = 7.97); and dibutyl adipate: 17.20–40.96 (median = 37.11).

Discussion

The confirmation of various environmental contaminants (Level 1) in AF highlights maternal exposure to a diverse range of industrial chemicals, including plasticizers, preservatives, chemical intermediates, and surfactants/detergents. These findings indicate that chemicals commonly used in plastics, pharmaceuticals, and consumer products can reach the intrauterine environment, raising concerns about potential fetal exposure. To our knowledge, apart from MEHP, previously detected in AF ³⁴, the other nine confirmed exogenous contaminants have not been reported in AF before.

Among surfactants and detergents, 4-hydroxybenzenesulfonic acid and 4-dodecylbenzenesulfonic acid are aryl sulfonates widely employed in detergents, surfactants, and industrial cleaners ^{35, 36}. Human toxicity data for these compounds are limited; however, The EPA CompTox Chemicals Dashboard ToxCast in vitro data indicate that 4-hydroxybenzenesulfonic acid shows activity in both estrogen receptor (ER) and androgen receptor (AR) receptor binding assays, including ER agonist and antagonist modes as well as AR antagonist activity, and vendor information indicates both may cause skin and eye irritation but show no acute oral or dermal toxicity in animal studies ²³. N,N-dimethyltetradecylamine-N-oxide is a C14 zwitterionic amine oxide surfactant, which is widely used in cosmetics and personal care products for mild cleansing, foam-boosting, conditioning, and antimicrobial properties, particularly in shampoos and conditioners ³⁷. However, there is very limited information on its human reproductive toxicity and carcinogenic potential. Lauryl sulfate, widely used in cleaning and personal care products, can cause skin and eye irritation and dermatitis with repeated contact; oral ingestion at high doses may lead to gastrointestinal irritation ³⁸.

Among plasticizers, dibutyl adipate functions as a plasticizer and emollient in flexible PVC, adhesives, and cosmetics ²⁴. A previous study reported migration of plasticizers from fast-food packaging into food and demonstrated endocrine-disrupting effects in vitro, including increased estradiol production, reduced testosterone levels, and alterations in key genes and proteins involved in steroidogenesis ³⁹. Phthalic anhydride, a precursor for plasticizers, resins, coatings, dyes, and some pharmaceuticals, can cause respiratory irritation, nasal ulceration, mucosal atrophy, loss of smell, hoarseness, asthma, bronchitis, and allergic reactions upon repeated human exposure, with the lungs and kidneys being the most sensitive organs ⁴⁰. MEHP, the primary metabolite of DEHP, has endocrine-disrupting activity with anti-androgenic effects in vitro and in vivo, and can impair reproductive function ⁴¹.

Among rubber-related compounds, DPG is primarily used as a vulcanization accelerator in rubber products such as tires, footwear, hoses, and molded goods ⁴². US production in 2019 ranged from 400 to 4,600 tons per year ⁴³, and it has been detected in indoor dust in multiple countries as well as in human urine and maternal and cord serum ^{44, 45}. DPG exhibits moderate acute toxicity (oral LD₅₀ ≈1,000 mg/kg in rats), with dermal and ocular irritation; subchronic exposure can lead to hepatic and renal alterations (EPA CompTox DTXSID5020378). Overall, DPG likely poses moderate risk primarily through repeated dermal contact and environmental exposure. ATBC is widely used in medical devices, food packaging, children’s toys, cosmetics, adhesives, and personal care products, creating potential exposure through multiple pathways ⁴⁶. Due to restrictions on phthalates, non-phthalate alternatives, including adipates, sebacates, trimellitates, phosphate esters, cyclohexane dicarboxylic acids, and citrates, are increasingly employed.⁴⁷ ATBC is one such alternative. In animal studies, female mice exposed to ATBC exhibited reduced ovarian follicle counts ^{26, 48} and anti-estrogenic and anti-androgenic effects ⁴⁹, with additional reports of obesity and hepatic steatosis ⁵⁰.

Quinoline, is a chemical intermediate derived from petroleum, coal tar, and shale oil processing, and it is used in solvents, dyes, and pharmaceuticals ⁵¹. Clothing textiles have been suggested as a possible route of human exposure ⁵². Quinoline can cause skin irritation and is classified by the U.S. EPA as a Group C possible human carcinogen.

In addition to the nine Level 1 confirmed compounds, eight environmental contaminants were tentatively annotated (Spreadsheet S6 and S7) and merit further confirmation in larger cohorts. These included organophosphate esters such as diphenyl phosphate, commonly used as flame retardants and plasticizers; benzyl alcohol, which is used a preservative and an antimicrobial agent in cosmetics, pharmaceuticals, and personal care products; and tetradecylsulfate, which is a surfactant widely applied in detergents and cleaning products. Several oxidized or substituted fatty acid derivatives, 3-hydroxyundecanoic acid, omega-cyclohexylundecanoic acid, oleic acid methyl ester, and cibaric acid, were also detected, which may represent transformation products or co-occurring lipid-like contaminants of environmental origin. Together, these Level 2–3 compounds reflect a mixture of industrial and consumer-use chemicals that may contribute to the complex exposomic landscape of the intrauterine environment, emphasizing the importance of confirming their presence in future large-scale studies. Although the toxicological profiles of individual chemicals detected in amniotic fluid, such as quinoline (carcinogenic) and MEHP (endocrine disruptor), are known (Table 1), the potential combined or cumulative effects of these co-occurring contaminants on fetal development remain unclear. Future studies should explore mixture effects and their relevance to developmental outcomes.

Compared to the only NTA exposome study for nonpolar compounds in ¹⁸, although using a very different ionization approach with GC–Orbitrap HRMS, we identified two overlapping chemicals, namely TPP and methyl oleate (Spreadsheet S13). Both compounds were confirmed in the referenced study; in our study, TPP was confirmed, whereas methyl oleate was tentatively annotated as level 3.1. TPP is produced in high volume and is one of the most widely used organophosphorus flame retardants and plasticizers, and is also present in many personal care products ^{53, 54}. Dusza, Manz et al ¹⁸ reported that TPP concentrations in extraction blanks (80 ng/L) were above the limit of detection (LOD, 62 ng/L), while the average concentration detected in AF was 36.8 ng/L. This observation is consistent with previous findings that organophosphate esters such as TPP are widely detected in experimental materials, resulting in blank interference and influencing the accuracy of analytical results ⁵⁵. Although we confirmed TPP using an authentic standard, we detected this feature in the laboratory control blanks as well. Due to the limited sensitivity of the Q-TOF instrument, we were unable to reliably resolve subtle abundance differences at the low ng/mL level, which will require further targeted analyses to accurately assess concentrations. Another overlapping compound, methyl oleate, is the methyl ester of oleic acid and is classified as a fatty acid methyl ester (FAME). It is commonly formed during derivatization procedures or introduced as contamination from solvents, biodiesel, or laboratory materials ⁵⁶. Although methyl oleate is frequently generated during sample preparation via esterification, it can also be detected in plasma, with levels varying among different populations based on dietary intake ^{57, 58}. In our 40 samples, the abundance of methyl oleate was generally higher than that observed in the laboratory control blanks. Unlike TPP, which is an endocrine disruptor, methyl oleate is not widely reported for adverse effects; however, Hong, Kim et al ⁵⁹ reported that gamma irradiation of methyl oleate induced the formation of trans-octadecenoic acid and stearic acid.

The observed correlations among oxidized fatty acids and environmental contaminants suggest possible interactions or shared pathways between endogenous lipid metabolism and exogenous chemical exposures. Specifically, FA 18:2 + 2O and 8-hydroxyoctanoic acid showed strong associations with tetradecylsulfate, while 2-oxophytanate correlated with phthalic anhydride. Structurally, these environmental contaminants possess hydrophobic backbones with polar or electrophilic groups resembling those of oxidized fatty acid derivatives, which may facilitate molecular interactions within lipid compartments or metabolic processes.

These metabolites participate in key lipid oxidation pathways, including β-oxidation (octanoyl-carnitine) ⁶⁰, ω-oxidation of medium-chain fatty acids (8-hydroxyoctanoic acid) ⁶¹, polyunsaturated fatty acid (PUFA) oxidation/oxylipin metabolism (FA 18:2 + 2O) ⁶², and peroxisomal α-oxidation of branched-chain fatty acids (2-oxophytanate) ⁶³. These metabolic routes are essential for lipid metabolism and energy balance, and exposure to amphiphilic contaminants may perturb these processes.

KMD analysis revealed distinct distributions among confirmed, annotated, and unknown chemical features, providing insights into both chemical class and potential structural relationships. The observed KMD clustering patterns provide insight into the underlying structural commonalities among detected features and support the utility of KMD analysis as a chemical class-level screening tool in NTA. The confinement of sulfonates to a narrow, negative KMD space reflects their shared sulfonic acid functionality and high oxygen content, highlighting KMD's sensitivity to conserved elemental motifs across structurally related compounds. Similarly, the aggregation of ester-containing compounds near zero KMD values is consistent with repetitive aliphatic subunits and underscores how KMD can differentiate lipophilic, CH₂-rich structures from more highly functionalized chemistries. Importantly, the tight clustering of annotated exogenous compounds within a limited KMD window suggests that many environmentally relevant contaminants share common structural backbones, despite differences in specific functional groups. This pattern provides a rationale for using KMD proximity as an indicator of structural relatedness, particularly for prioritizing unknown features that co-localize with known xenobiotics. Unknown features falling within this same KMD space are therefore more likely to represent chemically related environmental contaminants or their metabolites, such as surfactants, plasticizers, or amide-containing compounds, rather than endogenous metabolites. In contrast, unknown features located outside the dominant KMD range may reflect greater structural diversity, including atypical elemental compositions, uncommon functional group combinations, or transformation products not well represented in current reference databases. These features may warrant separate prioritization strategies, as they could indicate novel or less characterized chemical classes.

Endogenous metabolites (Levels 1–3), including creatinine, L-tryptophan, L-phenylalanine, and lysophosphatidylcholines, clustered tightly between −0.25 and +0.05 (Figure 5C), reflecting the limited diversity of elemental compositions within core human metabolites. Unknown features overlapping this metabolite cluster may represent either unannotated endogenous compounds or conjugated forms of known metabolites. Conversely, extreme KMD outliers at both high (> +0.3) and low (< −0.35) values suggest features with atypical elemental compositions, such as highly halogenated, polyether, or long-chain alkyl compounds, which may be environmentally derived or result from biotransformation reactions.

Analysis of chemical features using both association networks and KMD plots provided complementary insights into the chemical space of confirmed contaminants and endogenous metabolites. Association networks revealed clusters of co-varying features, reflecting potential biochemical or exposure-related relationships. Notably, ester-containing compounds formed the most prominent clusters, consistent with their prevalence among environmental contaminants such as phthalates and adipates, as well as endogenous lipid conjugates. This pattern aligns with previous exposome studies, where homologous series of esters and other oxygenated compounds were readily observed in human serum and environmental samples ^{64, 65}. Beyond esters, KMD analysis highlighted the presence of amine- and amide-containing exogenous compounds. These features frequently overlapped with unknown chemical signals, suggesting that many unannotated features may represent structurally related xenobiotics. The distinct KMD space occupied by these compounds, largely separate from endogenous metabolites (Figure 5B and C), underscores the utility of KMD plots for differentiating chemical classes based on their repeating units, which may not be apparent from m/z values alone. Moreover, KMD patterns suggested partial overlap between surfactant-like exogenous compounds and endogenous metabolites, indicative of either co-elution or shared elemental composition. These observations underscore both the strength of KMD analysis in structural grouping and its limitations, highlighting the need for orthogonal annotation approaches, such as MS/MS fragmentation and retention time prediction, to confidently assign chemical identities in future studies.

Conclusion and Limitations

Our NTA study identified nine environmental chemicals not previously detected in AF, expanding current knowledge of prenatal chemical exposure. These findings highlight the complexity of the human exposome and demonstrate how diverse environmental contaminants can potentially interfere with endogenous metabolic processes. Additionally, unknown features overlap with exogenous esters and amine/amide-based compounds, providing a targeted chemical space for subsequent structural elucidation.

This work represents an exposome-focused investigation rather than an epidemiological study, emphasizing chemical characterization and biological context rather than population-level associations. Future investigations with larger cohorts and concurrent blood and urine sampling are warranted to evaluate matrix-specific differences and further elucidate the utility of AF in disease-related exposomic studies, where additional candidate features can be validated and explored for biological relevance. Our semi-quantification of confirmed contaminants provides an estimated concentration range. However, further validation with isotope-labeled internal standards and a low-resolution mass spectrometer (e.g., Triple Quadrupole) is needed to achieve better sensitivity and accurate absolute concentrations. Therefore, we refrain from extensively discussing the concentrations in AF samples here. In addition, the current chemical coverage is limited to compounds amenable to soft ionization using LC–ESI in positive and negative modes; therefore, non-polar and volatile chemicals that are better detected by GC-HRMS with hard ionization may not be captured in this study and warrant investigation in future work.

Data Availability

All underlying data are presented in the Supplementary Materials spreadsheets and all code has been made available on: https://github.com/jixiaowen4321/AF-project