{"paper_id":"5baaefad-b3c2-4ce4-81ff-b78a25842d64","body_text":"Identified\nas emerging contaminants, endocrine-disrupting chemicals\n(EDCs) are exogenous substances that interfere with hormone actions\nin the endocrine system. \n , \n  Such chemicals have\nbeen widely found in consumer products, including food, beverages,\npersonal care products, detergents, and plasticware, entering human\nbodies through inhalation, ingestion, and dermal absorption. \n − \n \n  Human exposure to EDCs leads to increasing risks of health concerns,\nincluding neurological disorders, cardiovascular diseases, metabolic\nissues, obesity, and diabetes. \n , , \n  Despite global regulatory efforts, the mass production of the synthetic\nEDCs results in their ubiquitous and persistent presence in the environment\nthrough airborne emissions, landfills, and wastewater release. \n − \n \n \n  The complex sources of human EDC contamination have led to significant\nresearch attention focused on analytical method development and biomonitoring\nin human fluid samples such as urine, blood, and saliva. \n −\nOvarian follicular fluid contains hormones that provide the\nmicroenvironment\nfor the developing oocyte, whose quality is imperative for embryo\ndevelopment and subsequent fertility.  Chemicals absorbed into bodies enter follicular fluid by passing\nthrough the blood-follicle barrier (BFB).  Growing evidence has shown a direct correlation between high EDC\nconcentrations in follicular fluid and reduced women’s fecundity. \n , \n  A case-control study by Tian et al. suggested a potentially higher\nrisk of diminished ovarian reserve related to increased levels of\n21 EDCs detected in follicular fluid.  Two more recent works by Hofmann-Dishon et al.  and Li et al.  investigated\nparticipants’ reproductive outcomes, based on measurements\nof 24 and 64 EDCs in follicular fluid samples, respectively. While\nall these studies listed phthalates and phenolic compounds as EDC\nspecies of concern, \n , , \n  such targeted analyses could only envelop a limited number of known\nchemical contaminants. Thus, a more comprehensive screening method\nwill improve the understanding of EDC distribution in follicular fluid,\nespecially for compounds without prior knowledge or available pure\nreference materials.\nNon-targeted analysis (NTA) of human biological\nsamples for exposure\nassessments using high-resolution mass spectrometry (HRMS) has been\ndeveloping rapidly in the past decade.  Typically, NTA workflows begin with considering the range of chemicals\nto be detected and identified (i.e., a “chemical space”)  and developing a sample preparation strategy\nfor optimal extraction. Due to rapid biotransformation processes in\nthe human body, circulating xenobiotics such as phthalates and phenols\nfound in urine and blood are likely metabolized. \n − \n \n \n \n \n \n  Specifically, using parabens as an example ( Figure  \n ), typical metabolic processes include reduction,\noxidation, or hydrolysis (phase I metabolism), as well as conjugation\n(phase II metabolism). \n − \n \n \n  To facilitate chemical analysis, glucuronides and sulfates are often\ndeconjugated to their original form via sample hydrolysis, due to\nthe unavailability of conjugate reference standards. \n , \n  After that, sample cleanup procedures such as solid-phase extraction\n(SPE) are often performed to eliminate interfering components in the\nsample matrix, notably large proteins and lipids in biological fluids. \n ,\nExamples\nof biochemical reactions occurring in the proposed metabolism\nof parabens. Species shown in the upper half are phase I metabolites,\nand the structures in the lower half are phase II metabolites.\nThe next stage in the NTA workflow involves liquid\nchromatography\n(LC)-HRMS data acquisition, which consists of two modes: data-dependent\nacquisition (DDA) and data-independent acquisition (DIA). The DDA\nis a product ion scan method that only fragments isolated precursor\nions that meet preset requirements (typically a minimum abundance\nthreshold) in a higher-energy collisional dissociation (HCD) cell. \n , , \n  While this approach significantly\nreduces the spectral background, its constraints of a limited selection\nof precursor ions could overlook analytes with low concentrations,\nespecially when multiple ions coelute. On the other hand, DIA fragments\nall precursor ions in the selected mass range, regardless of signal\nintensity. \n , − \n \n  While DIA methods\ncollect a higher volume of fragmentation data, the absence of ion\nselectivity can lead to high noise and lower sensitivity, along with\nchallenges in pairing product and precursor ions in complex samples.\nAs a result, DDA results are often supported by well-developed software,\nwhereas unbiased DIA requires complex deconvolution methods.  Nevertheless, the impacts of acquisition methods\nvary significantly among studies and should be examined in individual\nNTA method development. \n −\nData analysis is the final\ncomponent in the NTA workflow. Here,\nmass spectra containing thousands of molecular features are first\nfiltered through multiple criteria (e.g., peak intensity, mass tolerance,\nelemental composition, etc.) to prioritize candidate precursor ions. \n , \n  For further molecular formula assignment and structural elucidation,\nMS2 library searches and/or spectral matching with  in silico  fragmentation have been widely used in NTA. \n , \n  However, recent critics have highlighted the frequency of mismatches,\nand the reliability of identified molecular structures remains uncertain. \n − \n \n  Accordingly, additional confirmation tools have been established\nto avoid false positives, thereby enhancing the identification confidence.\nFor example, our previous studies deployed (i) retention time ( RT ) matching via a prediction model,  and (ii) manual selection of diagnostic fragment ions as\nconstraints for monophthalate identification. \n , \n  In particular, the latter allows rapid precursor ion prioritization\nas all such phthalate metabolites follow similar fragmentation mechanisms. \n , \n  To expand this NTA approach of extracting chemical “fingerprints”\nto other types of EDCs, we hypothesize that parabens (also known as\n4-hydroxybenzoates) and their metabolites can likewise be identified\nusing diagnostic fragment ions. This is due to the fact that parabens\nare similar to phthalates in terms of their homologous chemical structure\nand phase I metabolism (see  Figure  \n ).  Therefore, we have\nfocused on stable phenols and metabolized organic acids as our chemical\nspace in this study, and our main objectives are to (1) characterize\nthe fragmentation patterns for parabens and metabolites; (2) apply\ndiagnostic ions to identify parabens and their metabolic pathways\nin follicular fluid; (3) compare DDA and DIA workflows for this application;\nand (4) evaluate the performance of NTA through computational tools\nfor identifying a broader range of EDCs in follicular fluid.\n\nThe detailed list of spiking\nchemicals, including 64 EDC reference standards and 11 isotopically\nlabeled standards, is summarized in  Tables S1 and S2 , respectively. Sixteen additional standards used for\nretention time modeling, in-house spectral database, and/or NTA structural\nconfirmation are listed in  Table S3 . Both\nβ-glucuronidase (≥300,000 units/g) and sulfatase (≥10,000\nunits/g solid) enzymes were type H-1 from  Helix pomatia  (Sigma-Aldrich). Used as received, the following chemicals were\nHPLC or LC-MS grade: glacial acetic acid (Fisher Chemical), ammonium\nacetate (Sigma-Aldrich), acetonitrile (Fisher Chemical), methanol\n(Fisher Chemical), and water (Fisher Chemical). 800 μg/mL solutions\nof each standard were prepared in methanol, prior to their combination\nand further dilution with methanol for working solutions of native\nstandards (20, 100, and 500 ng/mL) and a working solution of isotopically\nlabeled standards (200 ng/mL).\nThis study was approved by the Research\nEthics Board of Health Canada (REB 2017–023H). The follicular\nfluid samples were collected from 211 female patients receiving reproductive\ntreatment between 2016 and 2017 by ONE Fertility and McMaster University\nin Burlington, ON, Canada, for the study of the molecular physiology\nof human granulosa cells (HiREB 11–252-T). In this study group,\nthe average patient age (years) was 35.7 (SD = 4.3, range = 23–45).\nThe proportions of patients diagnosed with at least one of the following\ninfertility conditions were: polycystic ovary syndrome (PCOS) –\n6%; endometriosis – 9%; tubal factor – 11%; diminished\novarian reserve (DOR) – 29%; other female factor – 27%;\nmale factor – 38%; unexplained – 10%. A pooled sample\nwas prepared by combining an equal volume of follicular fluid from\nall of the individual patients. All samples were stored at −80\n°C prior to treatment.\nThe sample treatment\nprocedure was\nmodified based on our previous work on urine \n , , , \n  to accommodate\nthe removal of phospholipids and triglycerides in follicular fluid. \n , \n  A schematic diagram of the entire workflow designed for this study\nis shown in  Figure  \n . Briefly, thawed at room temperature, each 100 μL pooled follicular\nfluid aliquot was added with 5 μL of 200 ng/mL isotopically\nlabeled EDC standard solution. To hydrolyze the conjugated species\nin follicular fluid, 100 μL of 1 M ammonium acetate solution\n(pH = 5.0) containing 2800 U/mL β-glucuronidase and 280 U/mL\nsulfatase was added. The hydrolysis was done by incubating the mixture\nat 37 °C for 120 min in a dry block incubator (Eppendorf Thermomixer\nR) at a mixing speed of 550 rpm, followed by incubation at 60 °C\nfor 15 min to deactivate the enzymes. After cooling to room temperature,\nthe hydrolyzed samples were acidified by an addition of 20 μL\nof acetic acid and 880 μL of cold acetonitrile (−20 °C)\nto facilitate protein precipitation. These steps were also applied\nto three levels of spiked samples (1, 5, and 25 ng/mL follicular fluid),\nwhich were prepared by spiking 5 μL of 20, 100, and 500 ng/mL\nmixed EDC standard solution respectively, to the 100 μL follicular\nfluid aliquots. 100 μL LC-MS grade water was used as procedural\nblanks. Three replicates were prepared for blanks, nonspiked samples\nand samples at each spiking level.\nSchematic diagram of the designed workflow\nhighlighting the key\nsteps in this study.\nAfter the addition of\nacetonitrile, each sample was further incubated\nin the −20 °C freezer for at least 60 min and was centrifuged\nat 5000 rpm for 15 min (Eppendorf centrifuge 5424). The supernatant\nwas further filtered through a Captiva EMR-Lipid cartridge (Agilent,\n40 mg/1 mL) that was preconditioned with 1 mL of 80% acetonitrile\nin water (0.1% acetic acid) for passthrough removal of lipids. The\nprecipitated protein pellet was further washed with 200 μL of\n80% acetonitrile in water (0.1% acetic acid), and the wash solution\nwas also passed through the same cartridge for collection. After this\ncleanup process, the extracts were evaporated to dryness with a gentle\nflow of nitrogen gas at room temperature (Thermo Scientific Reacti-Vap\nevaporator). Lastly, the dried samples were reconstituted with 200\nμL of 50% methanol in water (0.1% acetic acid), followed by\nfiltration by 0.22 μM centrifugal filter (PTFE membrane) and\ncentrifuging at 10000 rpm for 2 min (Eppendorf centrifuge 5424). To\nimprove the spectral quality for unknown product identification, additional\n500 μL nonspiked pooled follicular fluid samples were treated\nwith and without enzymatic hydrolysis. Detailed procedures are described\nin Section 2 of the  Supporting Information (SI) .\nAll data acquisition was completed\non a Q-Exactive Orbitrap mass spectrometer (Thermo Fisher Scientific)\ncoupled to a Vanquish Horizon UHPLC system (Thermo Fisher Scientific).\nChromatographic separation was done on an Accucore reversed phase\nbiphenyl column (2.1 × 100 mm, 2.6 μm, Thermo Fisher Scientific)\ninstalled after an Accucore biphenyl guard cartridge (2.1 × 10\nmm, 2.6 μm, Thermo Fisher Scientific). The associated mobile\nphases consisted of A (0.1% acetic acid in water) and B (0.1% acetic\nacid in methanol). With a flow rate of 0.25 mL/min, the linear solvent\ngradient (curve setting 5) was: 90% A and 10% B for the first minute,\n40% A and 60% B at 10th minute and hold for 2 min, 15% A and 85% B\nat 15th minute and hold for 1 min, 10% A and 90% B at 20th minute\nand hold for 5 min, 90% A and 10% B at 30th minute with an additional\n10 min of equilibrium time. As for the HESI ion source, monophthalates\nand phenolic compounds favor negative ionization, producing deprotonated\nmolecular ions represented by [M-H] − .  The major settings were: −2.70 kV spray\nvoltage, 300 °C capillary temperature, 375 °C auxiliary\ngas heater temperature, 50.0 S-lens RF level, 40, 10, and 2 arbitrary\nunits of sheath gas, auxiliary gas, and sweep gas flow, respectively.\nMS data acquisition was done through three injections with the\nfollowing scan modes: 1) full scan only, 2) DDA with full scan/ddMS2\nmode, and 3) DIA with full scan/all ion fragmentation (AIF) mode.\nThe full scan settings in all three modes were: 80 – 1000  m / z  scan range, 70,000 resolution, 1e6\nautomatic gain control (AGC) target, and 100 ms maximum injection\ntime. In DDA, ddMS2 settings were: 17,500 resolution, 1e4 minimum\nAGC target, 1e5 AGC target, 55 ms maximum injection time, 1.5  m / z  isolation window, 5 loop count (TopN),\nand 7.0 s dynamic exclusion. In DIA, AIF settings were: 75 –\n1000  m / z  scan range, 35,000 resolution,\n2e5 AGC target, 55 ms maximum injection time. Three scan events with\n10, 20, and 40 eV collision energies (CEs) were applied in both DDA\nand DIA modes to maximize fragmentation ion coverage. Note that a\nstepped normalized CE (NCE) was not used to allow a clearer observation\nof fragmentation patterns at incremental CEs.\nTo identify\nthe fragmentation behavior of parabens and other EDCs, a standard\nsolution containing 50 ng/mL of 64 EDC native standards ( Table S1 ) and 11 isotopically labeled standards\n( Table S2 ) was prepared in solvent. MS/MS\nspectra were acquired through the DDA mode described above with a\npredefined inclusion list covering precursor ion  m / z  values and an  RT  window of 0.4\nmin. For each analyte, three MS/MS spectra scanned at 10, 20, and\n40 eV collision energies were saved into the mzVault software (Thermo\nFisher Scientific) as in-house MS2 database for NTA.\nStructures\nof parabens and alkyl protocatechuates can be retrieved\nfrom DDA and DIA data by matching their diagnostic fragment ions in\nthe FreeStyle Software (Thermo Fisher Scientific). Based on the investigation\nof their fragmentation patterns to be discussed in this study, product\nions at  m / z  91.0184, 95.0133, and\n108.0211 were selected as the common diagnostic ions for both classes\nof chemicals. Additional diagnostic ions were  m / z  92.0262, 93.0340, 136.0160, and 137.0239 for parabens,\nand  m / z  109.0290, 111.0082, 124.0160,\n152.0110, and 153.0188 for alkyl protocatechuates. In follicular fluid\nsamples, prioritized precursor ions were selected when 1) peak intensity\nwas at least 3 times higher than that of the process blanks, and 2)\nat least 4 diagnostic fragment ions (mass error <10 ppm) were found\nin the extracted ion chromatograms (EICs). Next, on Freestyle, the\npreselected precursor ions were further constrained through the elemental\ncomposition restrictions: 7–30 carbon, 6–60 hydrogen,\n3–10 oxygen, 0–1 sulfur, 5–10 ring double bond\nequivalents (RDBE), and 5 ppm mass tolerance in monoisotopic mass.\nFor DIA data, candidate precursor ions (≥10 5  intensity\nat CE = 0 eV) were selected at the retention times of observed diagnostic\nfragment ions if their intensity consistently decreased with increasing\nCEs.  These precursor ions were then added\nto an inclusion list for a separate full scan-ddMS2 acquisition to\nreaffirm their fragmentation for structure proposals. Lastly, if a\nnew compound without a reference standard was proposed, retention\ntime prediction was applied. Briefly described in Section 2 of the  SI , an in-house retention time prediction model  was built by converting the structures of EDC\nstandards into molecular descriptors (inputs). A linear relationship\nbetween measured  RT s (outputs) and a selected number\nof molecular descriptors was established for  RT  prediction\nof unknown compounds. This process is better known as a quantitative\nstructure-retention relationship (QSRR) model.\nSince not all groups of endocrine-disrupting chemicals\nyield characteristic fragments in MS/MS, more molecular features were\nextracted via Compound Discoverer software 3.3.3.200 (Thermo Fisher\nScientific) for a wider range of EDC identification in follicular\nfluid. The detailed software settings are summarized in  Figure S2  and Section 2 of the  SI.  Here, DDA files of blanks and samples were input into\nthe workflow. The spectral databases included the online mzCloud MS/MS\nlibrary (Thermo Fisher Scientific) and the in-house mzVault library\ncompiled in this study. For compounds without sufficient fragmentation\ndata in spectral libraries, structural prioritization was performed\nthrough matching Human Metabolome Database (HMDB 5.0)  mass lists with  in silico  fragmentation\n(Mass Frontier). For molecular structure annotation, a 70 best match\n(mzCloud or mzVault) score or 70 Fragment Ion Search (FISh) scoring\nwas selected as a minimum threshold in selecting the candidate precursor\nions in general. \n , , \n  Finally, retention time model predictions were applied to the suspect\ncompounds without available standards. Identification confidence was\nranked through the Schymanski scale.  In\nthis study, level 1 refers to structural confirmation through reference\nstandards. A level 2 confidence was assigned when the gap between\nmodeled and experimental  RT s (Δ RT ) was ≤ 1.0 min. A Δ RT  value above\nthis threshold was denoted as level 3 confidence.\nThe Orbitrap mass spectrometer was calibrated\nand evaluated weekly with the Pierce ESI negative ion calibration\nsolution (Thermo Fisher Scientific, product no. 88324) through direct\ninfusion. An eight-point external standard calibration consisting\nof 0.2, 0.5, 1, 2, 5, 10, 25, and 50 ng/mL mixed native standards\n( Table S1 ) was prepared in 50% methanol\nin water (0.1% acetic acid) with 5 ng/mL internal standards ( Table S2 ). The linearity ( R \n 2 ) of calibration curves was >0.99 for all standards within\ntheir calibration ranges. For every 6 randomly injected samples, at\nleast one solvent blank was placed in the sequence to monitor carryover\neffects, along with one 10 ng/mL standard for instrument stability\nchecks (i.e., peak intensity variation <15%;  RT  shift <0.1 min and mass error <5 ppm). The instrument detection\nlimit (IDL, ng/mL in solvent) was estimated as the minimal analyte\nconcentration at which the precursor ion signal-to-noise ratio (S/N)\nexceeded 3.  Given the lack of standardized\nquantitative methods in NTA, \n , \n  calculating the method\ndetection limit (MDL) remains challenging. For the purpose of assessing\ncompound identification capability, we estimated the MDL (upper limit)\nas the lowest spiking level (1, 5, or 25 ng/mL in follicular fluid)\nwhen (1) the intensity of molecular ion signal was at least 3 times\nhigher than that in the process blanks, and (2) a clear MS2 spectrum\nwas generated in DDA. Method reproducibility was assessed through\nthe relative standard deviation (RSD) of internal standard signal\nresponse in calibration standards ( n  = 8) and follicular\nfluid samples ( n  = 12). Processed in TraceFinder\nsoftware (Thermo Fisher Scientific), detailed calculations on recoveries \n , \n  and matrix effects \n , \n  in spiked follicular fluid are\nincluded in Section 2 of the  SI.\n\nBroadly found in human biological samples,\nparabens are bioactive\nphenols with estrogenic impacts. \n , \n  However, most\nof the existing studies are heavily focused on the distribution of\na few common parent species such as MeP, EtP, PrP, and BuP. \n , \n  To expand our knowledge of paraben identification in NTA, we first\ninvestigated the fragmentation pathways. For parent parabens,  Figure  \n a–e presents\nthe MS/MS spectra of 5 representative standards (linear, branched,\nand benzyl parabens) at 20 eV CE, and  Table S6  lists the distribution of major fragment ions at incremental CEs\n(10, 20, and 40 eV) for all species. Based on our experimental evidence,\nthe proposed mechanisms are illustrated in  Figure  \n . Specifically, at low CEs (10 and 20 eV),\nfragmentation of paraben precursor ions is initiated by the heterolytic\nand homolytic cleavage of the R (alkyl or benzyl) group,  producing 4-hydroxybenzoate ions at  m / z  137.0239 ( A1 ) and distonic\nions at  m / z  136.0160 ( A2 ), respectively. Note that [RH] loss is not structurally\npossible for MeP and BzP, and therefore, only  A2  is observed\nin  Figure  \n a,e. As\nCE increases,  A1  and  A2  further breakdown\nvia α-elimination of CO 2 , \n , \n  forming phenolate ions at 93.0340 ( B1 ) and  m / z  92.0262 ( B2 ). However,\nthe intensity of  B2  tends to be significantly lower than\nthat of  B1 , presumably due to its second radical (•H)\nloss as an intermediate species. The resulting fragment at  m / z  91.0184 ( B3 ) is known\nas a didehydrophenoxide biradical anion, as proposed by Schmidt et\nal.  However, the continuous CO loss of  B3  indicated by Schmidt et al.  was not observed in this study. To support the proposed structures\nof the product ions in  Figure  \n , we performed mass spectral fragmentation of an isotopically\nlabeled standard (PrP-d4). The MS2 spectrum in  Figure S3  shows identical deuterated 4-hydroxybenzoate ions\nand phenolate ions, the location of which may differ.\nMS2 spectra of (a–e)\nselected parabens at 20 eV CE, (f,\ng) alkyl protocatechuates at 20 eV CE and (h, i) phenolic acids at\n10 eV CE. The analyte concentration was 50 ng/mL, and the collision\nenergies selected for the spectra shown were best conditions showing\nboth precursor and fragment ions. See  Table S6  for a complete list of fragment ions for all 14 standards at 10,\n20, and 40 eV CEs. Abbreviations used in this work:  MeP  – methyl paraben;  EtP  – ethyl paraben;  PrP  – propyl paraben;  iPrP  – isopropyl\nparaben;  BzP  – benzyl paraben;  BuP  – butyl paraben;  iBuP  –  i sobutyl paraben;  iPeP  – isopentyl paraben;  2-EtHeP  – 2-ethylhexyl paraben;  OcP  –\noctyl paraben;  4-HB  – 4-hydroxybenzoic acid;  3,4-DHB  – 3,4-dihydroxybenzoic acid;  OH–MeP  – methyl 3,4-dihydroxybenzoate;  OH-EtP  –\nethyl 3,4-dihydroxybenzoate.\nProposed\nfragmentation pathways of parabens and alkyl protocatechuates.\nThe listed mass-to-charge ratios correspond to the accurate monoisotopic\nmasses (calculations via ChemCalc ).\nAdditional common fragmentation pathways for parabens\nare also\nobserved in  Figure  \n a–e. For example,  A2  could transform into a resonance-stabilized\ncyclic carbonyl structure  A3 , which favors CO elimination\nthat yields an ion at  m / z  108.0211\n( C ). \n , \n  In addition, the product ion\nat  m / z  123.0082 ( D2 ) corresponds to  A2  with an additional CH loss. Its\nproposed structure is supported by the observation of a deuterium\nloss from the aromatic ring in PrP-d4 ( m / z  = 126.0276 in  Figure S3 ). A\nsubsequent CO elimination produces an ion at  m / z  95.0133 ( E ),  which was detected in the spectra of all parabens in this work.\nOverall, we identified 8 characteristic product ions that can be used\nas diagnostic ions for screening parabens. This observed fragmentation\nmechanism complements the previous paraben measurements involving\ntriple-quadrupole (TQ) and quadrupole time-of-flight (qTOF) mass spectrometers.\nSpecifically, only  m / z  92, 93, 136,\nand 137 have been reported as fragments in multiple reaction monitoring\n(MRM) transitions for targeted analysis, \n − \n \n \n  and a recent non-targeted workflow\ndeveloped by Chi et al. was limited to two diagnostic ions ( m / z  92 and 93).  In our present work, while we also observed these fragments starting\nat 10 eV CE, a higher CE (20 or 40 eV) facilitated further dissociation,\nyielding product ions at  m / z  91,\n95, 108, and 123. This highlights the differences between traditional\nlow-energy collision-induced dissociation (CID) and higher-energy\ncollisional dissociation (HCD) pathways in Orbitrap mass spectrometers.\nSimilar distinctions between CID and HCD spectra have been recorded\nin proteomic identifications. \n , \n  Moreover, we also matched\nthe experimental MS2 spectra in  Figure  \n a–e with those published in the mzCloud library\nand the simulated spectra generated by the  in silico  algorithm (Mass Frontier). At the time of writing, the online database\nhad limited coverage of paraben species acquired in the negative ESI\nmode, while only two common ions at  m / z  92 and 136 were matched by the simulator (an example of MeP shown\nin  Figure S4 ). Such discrepancies have\nbeen noted in other recent evaluations. \n , \n  Thus, we expect\nthat the inclusion of experimentally confirmed diagnostic ions would\nsignificantly enhance the annotation confidence of parabens.\nAs a part of phase I metabolism (see  Figure  \n , upper section), hydroxylation is an important\nreactive pathway for parabens absorbed by the human body. \n , \n  The resulting products, alkyl protocatechuates (also known as 3,4-dihydroxybenzoates),\nhave been identified as urinary biomarkers for paraben exposure analyses. \n , , \n  In fact, higher concentrations\nof alkyl protocatechuates have been reported in urine than their respective\nparent parabens.  However, biomonitoring\nstudies on hydroxylated metabolites are limited to OH–MeP and\nOH-EtP due to the availability of reference materials, potentially\nunderestimating the total cumulative paraben exposure. Inspired by\nthe identification of diagnostic fragmentation pathways of parabens\nin this work, we also hypothesized that alkyl protocatechuates would\nfollow similar patterns, with a unique set of product ions. To test\nthis, tandem MS was performed on OH–MeP and OH-EtP (i.e., the\nonly available standards). As expected ( Figure  \n f,g), the fragmentation mechanisms follow\nthe patterns of parabens as many observed ions contained an additional\noxygen (Δ m / z  = 15.9949). Notably,\nas proposed in  Figure  \n , the product ions at  m / z  153.0188\n( F1 ) and/or 152.0110 ( F2 ) undergo a decarboxylation\nprocess to form 2-hydroxyphenolate ions at  m / z  109.0290 ( G1 ), 108.0211 ( G2 ),\nand 107.0133 ( G3 ). Furthermore, similar to the observed\nion  D2  and  E  in parabens, a single and double\nneutral loss of CO from  m / z  139.0031\n( H ) yield ions at  m / z  111.0082 ( J ) and 83.0133 ( K ), respectively.\nIt is noteworthy that some identical characteristic ions were identified\nin both paraben parents and alkyl protocatechuate metabolites. Specifically,\nthe ion at  m / z  124.0160 ( D1 ), formed via a neutral CO elimination from  F3 , is proposed\nto undergo successive losses of a hydrogen radical and a CO, \n , \n  yielding ions at  m / z  123.0082\n( D2 ) and  m / z  95.0133\n( E ). Additionally, the characteristic ion at  m / z  91.0184 ( B3 ) is attributable\nto a H 2 O loss from  G1 . In fact, from a metabolomics\nperspective, this fragmentation pattern provides important structural\nand metabolism site  information as it\nconfirms hydroxylation occurring on the aromatic ring. Lastly, to\ninclude all phase I paraben metabolites, we also investigated the\nfragmentation pathways of phenolic acids. For 4-HB (precursor ion\nidentical to  A1 ), only  m / z  93.0340 ( B1 ) was observed at 10, 20, and 40 eV CEs\n( Figure  \n h and  Table S6 ), and the absence of  B2  and  B3  in the observed spectra indicates that •H\nelimination is not feasible for  B1 . Conversely, a loss\nof CO 2  from the deprotonated 3,4-DHB ( F1 )\nproduces  m / z  109.0290 ( G1 ) at 10 eV CE ( Figure  \n i). However, at 20 and 40 eV CEs ( Table S6 ), it is subject to further •H or H 2 O elimination\nthat yields  G2  and  B3 . In summary, we have\nidentified a total of 17 common fragment ions covering paraben parents\nand their phase I metabolites through tandem MS of 14 standards. It\nshould be noted that phase II metabolites ( Figure  \n , lower section) were not investigated in\nthis manner due to the lack of reference standards. However, the diagnostic\nfragment ions described in  Figure  \n  are still useful for identifying paraben conjugates\n(vide infra).\nTo evaluate the workflow performance,\nwe first assessed the analytical sensitivity via detection limit estimations\nof compounds within the chemical space. Specifically, as listed in  Table S7 , 60% of the 64 EDC standards had IDLs\nbelow the lowest calibrant concentration (0.2 ng/mL), while the detection\nof some bisphenols tended to be much less sensitive. This was not\nsurprising as bisphenols are well-known for their poor ionization\nefficiency in negative ESI, which requires mobile phase additives\nwith high gas-phase proton affinity.  In\nthe follicular fluid sample matrix, approximately 40% and 75% of the\nanalytes were reliably detected with abundant precursor ion intensities\nand clear MS2 spectra in ddMS2 (as a measure of MDL estimation), at\n1 and 5 ng/mL spiking levels, respectively. A few compounds, such\nas 2,4-dichlorophenol and 2,4,6-trichlorophenol, showed low IDLs but\nhigh MDLs, likely due to possible procedural loss during extraction.\nVarying from 12 to 20%, method reproducibility was calculated through\nthe RSD of precursor peak areas of the internal standards added in\nfollicular fluid samples ( Table S8 ). In\nterms of accuracy,  the recovery range\nwas between 75 and 125% for approximately 30, 60, and 70% of the analytes\nat the 1, 5, and 25 ng/mL spiking levels, respectively ( Table S9 ). Lastly, while thorough cleanup steps\nwere performed during sample treatment, the sample matrix still caused\nslight to moderate signal suppression (10–30%) for most analytes\nbased on matrix effect (ME) calculations ( Table S9 ). This could be due to coeluting ions (including residual\nmatrix ions) competing for the limited capacity of the C-trap.\nAfter identifying the recovered EDCs in spiked follicular fluid by\ntheir precursor ions, we then assessed the feasibility of compound\nidentification based on fragmentation patterns. First, with the use\nof diagnostic paraben fragment ions characterized in the previous\nsection,  Figure  \n  shows\nthe extracted ion chromatograms (EICs) of these ions in the MS2 spectra\nof the spiked sample (25 ng/mL). Overall, both DDA ( Figure  \n a–g) and DIA ( Figure  \n h–n) acquisition\nmodes were able to demonstrate the presence of spiked parabens and\nmetabolites, but some variations were observed. Specifically, as seen\nin the EICs of diagnostic ions for both parabens and alkyl protocatechuate\n( Figure  \n a–c,h–j),\nwhile the DIA spectra illustrate smoother peaks with higher peak intensity,\nDDA results are more informative. For example, coeluting compounds\n(MeP and OH-EtP; BzP and iPeP) are more easily identified in DDA due\nto its direct selection of precursor ions. Similar differences between\nthe two modes have been previously reported in NTA of monophthalates.\nExtracted ion chromatograms (EICs) of paraben diagnostic\nfragment\nions in (a–g) DDA and (h–n) DIA data of a pooled follicular\nfluid sample with a spiking concentration of 25 ng/mL. The diagnostic\nion at  m / z  123.0082 is not displayed\ndue to its relatively weak peak intensity. The chromatograms shown\nin DDA correspond to a combination of spectra collected at 10, 20,\nand 40 eV CEs, while for DIA, a preferred CE (20 or 40 eV) was chosen\nto demonstrate the best data quality. B1–B6 correspond to the\nspiked benzophenone standards: B1 – benzophenone-2; B2 –\n2,4,4 trihydroxybenzophenone; B3 – 4-hydroxybenzophenone; B4\n– benzophenone-1; B5 – dioxybenzone; and B6 –\nbenzophenone-6.\nIt is noteworthy that strong interfering\npeaks (denoted as B1–B6)\nwere found in the EICs of the diagnostic ions discussed above, especially\nfor  m / z  91. Confirmed with the fragmentation\nprofiles of pure standards, these peaks resulted from benzophenones\nthat were among the 64 EDCs in the spiked samples. Thus, to minimize\nthe effects of other interfering phenolic compounds, the inclusion\nof a fourth (or more) diagnostic ion could reduce the chance of false\npositives. Accordingly,  Figure  \n d–g,k–n shows EICs of additional diagnostic\nions for parabens, and similarly, DDA spectra were more informative,\nespecially for illustrating weak signals such as  m / z  92 ( Figure  \n d,k). Moreover, EICs of  m / z  136 and 137 had the cleanest background for this follicular\nfluid matrix ( Figure  \n f,g,m,n).  Figure S5  compares DDA and DIA\nin spiked samples at 1 ng/mL, and it is not surprising that fewer\npeaks were observed in both modes. This poses a challenge for precursor\nion isolation in DIA data, since the spectral background significantly\ninterferes with the diagnostic ions in EICs. Comparatively, although\nDIA can be effective for identifying a higher number of analytes, \n , , \n  our DDA data exhibited sufficient\nsensitivity for paraben identification when four diagnostic fragment\nions were set as a threshold for prioritizing precursors.\nGiven\nthat not all EDC chemicals can be traced to a known diagnostic\nfragmentation pattern, we then evaluated compound annotation for\na wider range of EDCs through mass spectral library matching in Compound\nDiscoverer, including the use of an in-house database built in this\nwork and  in silico  fragmentation. With 70 as the\nthreshold matching score, the true positive identification rates,\ndefined as the proportion of matched compounds out of 64 spiked EDCs,  were 41, 70, and 83% for samples with 1, 5,\nand 25 ng/mL spiking levels, respectively ( Table S10 ). This result is similar to our previous work on chemical\nidentification in spiked urine (∼50% rate at 1 ng/mL spiking\nlevel).  A common issue for extracted\nprecursor ions not being identified through library matching was the\nfailure of triggering fragmentation in DDA scans. \n , \n  This was due to their signal intensity near or below the preset\nthreshold value. To improve the chances of analyte selection and fragmentation,\nwe attempted to adjust scan parameters such as lowering the signal\nintensity threshold to 1.5E5 (via reducing the minimum AGC target\nto 8.0E3) and increasing the loop count (or topN) to 6, 8, and 10.\nHowever, from our observations, more coeluting matrix ions entered\nthe C-trap simultaneously for scanning in the Orbitrap, thereby increasing\nthe spectral noise and lowering the overall compound matching scores.\nThe newly\ndeveloped NTA workflow was applied to nonspiked follicular fluid samples,\nwith and without enzymatic treatment, to assess its capability for\nidentifying unknown compounds. Here, 500 μL pooled samples were\nprocessed (see Section 2 of  SI ) to facilitate\nthe detection of ultralow-concentration species. In hydrolyzed samples,\nbased on the selection criterion of at least four diagnostic fragment\nions in DDA data, 5 candidate precursor ions were identified as parabens\n(MeP, EtP, PrP) and protocatechuates (OH–MeP and OH-EtP). Together\nwith their metabolized acids (4-HB and 3,4-DHB), all compounds were\nannotated with level 1 confidence (Schymanski scale ) after retention time confirmation with standards ( Table S11 ). Tentatively assigned as level 3,\nseveral positional isomers of these species, which share identical\nMS and MS2 profiles but differ in retention times, were also proposed.\nThe use of the Compound Discoverer further enabled structural deconvolution\nof some suspect compounds, as discussed later in this section.\nThe extraction of paraben diagnostic ions also allowed us to identify\nthe speciation of phase II metabolites present in unhydrolyzed follicular\nfluid. As demonstrated in  Figure  \n a–f, six precursor ions were identified as sulfate\nconjugates of parabens and their phase I metabolites, with detailed\nfragmentation profiles provided in  Table S12 . This identification was supported by the formation of the SO 3 \n   ion at  m / z  80,  yielding deprotonated ions of free\nparabens (or their phase I metabolites) that continued to fragment\nvia their diagnostic pathways. In some cases, the HSO 4 \n –  ion at  m / z  97 was\nalso observed.  The measured  RT s of the proposed conjugates were compared against the model prediction\nor standards ( Table S12 ). Notably, five\nof six identified metabolites were assigned level 2 confidence (Δ RT  ≤ 0.5 min), while PrP sulfate was further confirmed\nusing its standard, reaching level 1 confidence. Conversely, only\n3,4-DHB sulfate was assigned level 3 confidence, owing to a larger\ndiscrepancy between the measured and predicted  RT s.\n(Left side) MS2 profiles of proposed phase II paraben metabolites\n(sulfates) identified in unhydrolyzed pooled follicular fluid samples:\n(a) methyl paraben sulfate, (b) ethyl paraben sulfate, (c) propyl\nparaben sulfate, (d) 4-hydroxybenzoic acid sulfate, (e) 3,4-dihydroxybenzoic\nacid sulfate, and (f) methyl 3,4-dihydroxybenzoate sulfate. (Right\nside) Extracted ion chromatograms (EICs) of corresponding free-from\nchemicals (precursor ions with mass error ± 5 ppm) in hydrolyzed\nand unhydrolyzed follicular fluid samples: (g) methyl paraben, (h)\nethyl paraben, (i) propyl paraben, (j) 4-hydroxybenzoic acid, (k)\n3,4-dihydroxybenzoic acid, and (l) methyl 3,4-dihydroxybenzoate.\nMore importantly, given that none of the proposed\nsulfate conjugates\nwas detected in hydrolyzed samples, a comparison of free-from chemicals\nbetween hydrolyzed and unhydrolyzed samples can reveal the distribution\nof paraben speciation in follicular fluid. Accordingly, from the EIC\nchromatograms in  Figure  \n g–l, while free parabens such as MeP and EtP were detected\nin unhydrolyzed samples, enzymatic hydrolysis increased their intensity.\nIn particular, 3,4-DHB and OH–MeP ( Figure  \n k,l) appeared to exist predominantly in their\nconjugate forms in the original fluid. Without any evidence of detecting\nparaben-based glucuronides in this study, we thus infer that sulfation\nis likely a dominant phase II metabolism for EDCs of this type. Although\nconjugated parabens have long been reported in biological matrices,  detailed structural evidence of individual species\nremains rare. Our results complement the recent NTA works that also\nsuggest the significance of sulfation metabolism, as shown through\nurine  and human milk analyses.\nFinally, more compounds were identified\nthrough mass spectral library\nmatching in Compound Discoverer. In the analysis of hydrolyzed samples,\nthe software detected more than 6000 molecular features, the vast\nmajority of which had only exact mass information without MS2 spectral\nmatches, corresponding to level 4–5 confidence.  Within our chemical space, we selected 37 features\nbased on a minimum match score of 70 (mzCloud, mzVault, or FISh),\nallowing structure annotations at level 3 confidence or higher. A\ndetailed summary is listed in  Table S13 . Briefly, a total of 14 features were identified as level 1 confidence.\nApart from parabens and metabolites mentioned above in  Table S11 , monophthalates (monomethyl-, monoethyl-,\nand monoisobutyl phthalate), UV filters (benzophenone-1) and metabolized\norganic acids were among the other potential EDCs found in this pooled\nfollicular fluid sample. One compound, initially proposed as a 4-HB\nisomer ( RT  = 8.1 min) through diagnostic fragmentation\nions ( Table S11 ), was later confirmed to\nbe salicylic acid (2-HB) through database search and  RT  match with a standard. Notably, Chen et al. detected elevated levels\nof salicylic acid in follicular fluid that were four times higher\nin patients with polycystic ovary syndrome (PCOS) than in the healthy\ncontrol group,  due to its association\nwith phenylalanine metabolism that affects oocyte quality. \n , \n  As well, hippuric acid, a glycine conjugate of benzoic acid used\nas a biomarker for metabolic health  and\ntoluene,  was also annotated as level\n1. Lastly, as explained in  Figure S6 , three\nmolecular features matched the MS2 fragmentation patterns of 4-hydroxyhippuric\nacid (4-HHA, structure shown in  Figure  \n ), an amino acid derivative of 4-HB. Further retention\ntime validation using a pure standard confirmed level 1 identification\nfor the precursor ion eluting at 3.1 min. Thus, the use of library\nsearch facilitated the identification of another potential phase II\nparaben metabolite, \n , \n  whose MS2 fragmentation pattern\ndiffers from that of the parent parabens. Although Compound Discoverer\nsuggested a match to 2-hydroxyhippuric acid (2-HHA) for the other\ntwo features shown in  Figure S6 , this identification\nwas not supported by the  RT  model prediction ( Table S13 ) for assignments at level 2 confidence.\n\nEDCs in follicular fluid are important indicators\nin evaluating\noocyte quality and reproductive health, \n , \n  and thus,\na robust analytical method detecting a wide range of species is valuable\nfor biomonitoring studies. In the present work, we developed a new\nnon-targeted workflow consisting of sample treatment, data acquisition,\nand data analysis to meet this need. Our experimental results demonstrated\nthat parabens (and protocatechuates) followed a characteristic fragmentation\npattern, yielding unique diagnostic ions in HCD cells. With high sensitivity\nand precision, the extraction of such mass spectral fingerprints enables\nrapid prioritization of paraben precursor ions in DDA. Conversely,\nDIA requires complex deconvolution for interpretation at this point,\nbut ongoing development of computational tools could enhance its capabilities\nin the future.  More importantly, our\nparaben annotation approach is also much more reliable than database\nmatching through online library and  in silico  fragmentation.\nIn particular, we examined that the latter algorithm led to high errors\nin predicting the MS2 spectra, an observation consistent with the\nrecent criticisms calling for cautious use of simulator tools. \n , \n  Nevertheless, the application of diagnostic evidence is still limited\nto specific groups of chemicals, and broader compound identifications\nthrough spectral database matching remain an important aspect of NTA.\nIn this work, species of potential concern, such as monophthalates,\nUV filters, and metabolized organic acids, were confirmed in follicular\nfluid, highlighting the utility of this method for identifying EDC\nbioaccumulation linked to human exposure.\nThis NTA work not\nonly revealed the distribution of exogenous substances\nin human biological samples but also implied the possible metabolic\nprocesses occurring in the human body. To the best of our knowledge,\nthis is the first study that identifies parabens, as well as their\nphase I and phase II metabolites, simultaneously in follicular fluid\nsamples. Specifically, hydrolysis and hydroxylation products were\nidentified and confirmed using standards. Annotated at level 1 and\n2 confidence, sulfate and glycine conjugates were the detected forms\nof phase II metabolic biomarkers for parabens. These conjugates have\nrarely been characterized in biological samples due to the limited\navailability of their reference standards. In this work, although\nno firm evidence was found for paraben glucuronidation, the ability\nto screen such large molecules through diagnostic fragment ions improves\nour understanding of metabolite speciation for more accurate exposure\nassessments.\nLastly, some limitations remain in the workflow\nand are worth further\nresearch attention. First of all, the present method was designed\nfor extracting and detecting phenolic compounds and metabolized organic\nacids; its analysis capacity over other major groups of EDCs (e.g.,\npolychlorinated biphenyls)  remains to\nbe validated. Within the chemical space, while high identification\nrates were attained for compounds such as parabens and monophthalates,\nalong with satisfactory recoveries and reproducibility, the current\nmethod underperformed in extracting and identifying (1) bisphenols\nand halogenated phenols and (2) some EDCs with ultralow concentrations\n(i.e., ≤1 ng/mL). Thus, inspired by recent comprehensive optimization\nworks on Orbitrap instruments, \n , \n  our future work will\nfocus on sensitivity improvements in NTA data acquisition. Furthermore,\nthe present NTA workflow is limited to qualitative identification\nin follicular fluid. Similar to our in-house retention time model,\nfuture improvements on quantitative non-targeted analysis (qNTA) can\nbe developed through modeling with the molecular descriptors. \n , \n  Such quantitative data from individual patient analyses, obtained\nthrough both targeted analysis and qNTA, will help us determine any\npotential correlations between chemical exposure and infertility conditions.","source_license":"CC-BY-4.0","license_restricted":false}