{"paper_id":"56b8f6e0-98fd-4b40-9747-910067caad41","body_text":"The concept of the\nexposome was introduced in 2005 to encourage\nmore research on the environmental determinants of disease. 1 , 2  Environmental chemicals have long been known to be human disease\nrisk factors, 3  and relevant exposures today\ninclude ambient air pollution and mixtures of contaminants ingested\nfrom food, water, and dust. 4 − 7  Analytical methods are now evolving to measure the\n“chemical exposome”, broadly defined as all environmental\nchemical exposures throughout the life course. 8 − 10  With this ambitious\nscope of chemical exposomics come several practical methodological\nchallenges, including to comprehensively quantify a broad range of\npriority target analytes, to discover and identify novel exposures\nby nontarget workflows, to scale up sample throughput, and to achieve\nhigh sensitivity in the small volumes of biofluids available in typical\ncohort studies. 11\nTraditional analytical\nmethods for human biomonitoring are mostly\ntargeted and mass spectrometry-based, employing either liquid chromatography\n(LC) or gas chromatography (GC). 12 − 19  Such methods are very sensitive and quantitative but can require\nlarge sample volumes and laborious sample processing steps and ultimately\nreveal only a limited fraction of the chemical exposome, defined  a priori  by the investigators. Today, there are over 350,000\nchemicals in global commerce, 20  but only\n5% of these have ever been analyzed in environmental media, 21  and likely fewer have been biomonitored in humans.\nIn this context, modern instrumental advances in LC- and GC-high-resolution\nmass spectrometry (HRMS) provide great potential for chemical exposomics,\nas several commercial instruments now combine high mass spectral resolving\npower, high mass accuracy, high scanning frequency, good sensitivity,\nand a wide dynamic range in full scan mode. 10 , 22 , 23  These instruments are therefore well suited\nto perform parallel target and nontarget data acquisition, but methodological\nchallenges remain for sample preparation. One specific challenge is\nto quantitatively extract diverse analytes from complex biological\nsamples while minimizing interferences. In human blood, known organic\nenvironmental contaminants vary greatly in their hydrophobicity, spanning\n17 orders of magnitude in octanol–water partition coefficient\n( K ow ), and their concentrations range\nover 11 orders of magnitude (i.e., 160 fM–140 mM) 24 , 25  in a matrix dominated by major lipid classes and complex mixtures\nof endogenous metabolites 26  that may overshadow\nsmall signals from environmental chemicals. Multiclass target exposomes\nare now being reported, as well as methods for combined target and\nnontarget analysis, but the sample preparation methods are often adapted\ndirectly from metabolomics, 27 − 29  and not specifically designed\nor optimized for chemical exposomics, which strives to profile small\nmolecules present at 1000× lower concentrations than endogenous\nsubstances. 25\nMost reported chemical\nexposomics methods for blood have so far\nbeen LC-based and thus focused on the polar environmental chemical\nfraction. 22 , 30 , 31  However, the\ncommercial availability of GC-Orbitrap HRMS instruments 32  has enabled recent development of methods for\nlow-polarity and semivolatile analytes. 29 , 33 , 34  The associated sample preparation methods for blood\nserum and plasma are simple, rapid, and potentially scalable, including\nliquid–liquid extraction 29  and liquid–liquid\nextraction with dispersive powders 34  (i.e.,\nQuEChERS). Nevertheless, the performance of existing methods to minimize\nmajor blood lipid coextractives has not been directly evaluated, and\nthere is concern that abundant lipids may negatively influence method\nrobustness, lower method sensitivity, and interfere with molecular\ndiscovery. Of particular relevance is that the resolving power of\nOrbitrap HRMS analyzers is adversely impacted by the abundant matrix\nsignal, such that the auto gain control function dynamically lowers\nthe ion injection time to minimize space-charging when an abundant\nsignal is detected. 35 , 36  Therefore, we hypothesized that\nby minimizing lipid coextractives, trace analytes would be more easily\ndetected by GC-Orbitrap HRMS; moreover, larger volumes could be injected\nto improve method sensitivity.\nBuilding on existing GC-based\nchemical exposomics methods for human\nblood, here, we explore methods to achieve improved sensitivity for\nplasma chemical exposomics by a GC-Orbitrap HRMS workflow. Specifically,\nwe aimed to minimize coextracted plasma lipids at the extraction step\nto achieve dual benefits to method performance from (i) minimizing\nmatrix interference and (ii) enabling large-volume injections (LVIs)\nof plasma extracts. A simple and potentially scalable protocol, which\nuses isohexane (H) to liquid–liquid extract acetonitrile-plasma\n(A-P), was developed and validated for small samples of human plasma\n(100–200 μL) while optimizing for lipid removal and method\nsensitivity for 103 priority target analytes. The validated method,\ntermed the HA-P method, was applied to a subset of adult plasma samples\n( n  = 32) in a combined target and nontarget analysis,\nand method performance was examined with respect to target analytes\nand molecular discoveries.\n\nA representative\nlist of 103 target analytes from 6 chemical classes was selected based\non higher detection frequencies in major historical biomonitoring\ninitiatives (e.g., HBM4 EU, US NHANES), and also considering a broad\nrange of physiochemical properties for use in method development,\nand their native and isotopic labeled standards were acquired from\ncommercial suppliers ( Table S1 ). These\nincluded 33 polychlorinated biphenyls (PCBs), 8 polybrominated diphenyl\nethers (BDEs), 10 polychlorinated dibenzodioxins/dibenzofurans (PCDD/Fs),\n16 polycyclic aromatic hydrocarbons (PAHs), 24 organochlorine pesticides\n(OCPs), and 12 phthalates. Toluene, formic acid,  n -hexane, and silica gel (high-purity grade, pore size 60 Å,\n230–400 mesh particle size) were from Merck (Germany). Isohexane,\nethyl acetate, acetonitrile, dichloromethane, chloroform, and water\nwere from Fisher Scientific (USA); isohexane was used instead of  n- hexane because it has similar physiochemical properties\nbut lower toxicity. 37 , 38  MgSO 4  and a dispersive\nsolid-phase extraction powder, Bond Elut EMR-Lipid, were purchased\nfrom Agilent (USA).\nFor method\ndevelopment and validation,\nsterile-filtered human serum was obtained commercially (Merck, Germany,\nhuman male AB plasma-derived, origin USA). Standard reference material\n(SRM 1958, fortified and lyophilized human serum) was purchased from\nNIST (USA). The optimized exposomics workflow was applied to 32 individual\nhuman plasma samples (100 μL) from the Västerbotten Intervention\nProgramme (VIP) cohort, 39  a cardiovascular\nstudy cohort launched in 1985 in north Sweden. The samples (16 men\nand 16 women), which were collected between 1991 and 2013 ( Table S2 ) and stored at −80 °C, were\nrandomly selected among participants whose samples (separate aliquots)\nhad previously been analyzed for persistent organic pollutants. 6  Participant smoking and moisture snuff consumption\nas well as dietary intake of meat and fish were self-reported. As\ndescribed previously, 31  a Swedish pooled\nplasma reference sample was also prepared in-house from residual heparinized\nplasma of 953 adults (male and female) from the VIP cohort. Approval\nfrom the Swedish Ethical Review Authority [Dnr 2020-03301] was granted\nfor work with these plasma samples.\nAll glassware\nwas newly furnaced to minimize background contamination, and sample\npreparation was in a positive-pressure clean laboratory with high-efficiency\nparticulate air filtration. Plastic containers and materials (e.g.,\ndisposable pipet tips) were avoided during sample preparation and\nanalysis to minimize contamination of the samples and associated extracts.\nMethod development included various tests of lipid removal techniques,\nincluding liquid–liquid and solid-phase extraction (see the  Supporting Information , SI). The optimized method\nfor human plasma and serum can be used for a range of small volumes\ntypically available in cohort studies (100–200 μL). The\nplasma/serum was added to a conical borosilicate-glass centrifuge\ntube (10 mL, Pyrex, Corning, USA), and each sample was then spiked\nwith 8 μL of a 100 ng/mL surrogate internal standard mixture\ncontaining 26 isotopically labeled chemicals ( Table S1 ). For method development, small volumes of the internal\nstandard solution were spiked onto each sample using precise Hamilton\nsyringes (10 μL). For application to cohort plasma samples,\nthe internal standard solution was indirectly added to larger volumes\nof acetonitrile and later added for protein precipitation using glass\ntip pipettes (200–2000 μL, Socorex). Procedural blanks\ncomposed of LC water, substituted for plasma, were prepared with all\nexperiments and batches. For protein precipitation, a volume of acetonitrile\nwas added that was 4 × sample volume (e.g., 400 μL of acetonitrile\nadded to a 100 μL sample or 800 μL of acetonitrile added\nto a 200 μL sample), vortexed at 1400 rpm for 1 min (multitube\nvortex mixer, Ohaus, USA) and then centrifuged (5804R, Eppendorf)\nat 4400 g  for 5 min at room temperature. The supernatant\nwas transferred to a new borosilicate-glass tube, and 1.2 mL of isohexane\nwas added, followed by vortexing at 1400 rpm (1 min) and centrifugation\nat 3000 g  (1 min). The upper isohexane layer was transferred\nto a new borosilicate-glass tube; another 400 μL of isohexane\nwas added to the extraction test tube, and the extraction step was\nrepeated. The two isohexane layers were combined and evaporated to\n100 μL by gentle nitrogen flow at room temperature (TurboVap\nLV, Biotage, Sweden), followed by adding 10 μL of 20 ng/mL volumetric\ninternal standard (methoxychlor-D14) by a Hamilton syringe (700 series,\n100 μL), vortexing, and transferring to an amber glass\nvial with an insert (0.3 mL, Thermo Scientific) for GC-HRMS analysis.\nAll 32 individual samples (VIP cohort) were extracted in one daily\nbatch along with 3 procedural blanks and 3 Swedish pooled plasma reference\nsamples.\nFor comparison, plasma samples were also prepared by\na leading literature chemical exposomics method designed for GC-Orbitrap\nHRMS. 29  Briefly, 8 μL of 100 ng/mL\nsurrogate internal standard was added to 200 μL of Swedish pooled\nplasma ( n  = 3) in conical glass tubes, followed by\nadding 50 μL of formic acid and 200 μL of  n -hexane: ethyl acetate (v:v = 2:1). This was vortexed for 1 h on\nice and then centrifuged at 4400 g  at 4 °C (10\nmin). The organic supernatant was transferred to a new tube with 25\nmg of MgSO 4 , vortexed, and centrifuged at 4400 g  (10 min). The final supernatant was transferred to an injection\nvial with 6 μL of 100 ng/mL volumetric internal standard (methoxychlor-D14)\nfor instrumental analysis.\nPlasma extracts\nwere analyzed by GC\n(TRACE 1300 Series, Thermo Fisher Scientific, US) interfaced to an\nOrbitrap HRMS (Q-Exactive, Thermo Fisher Scientific, US) operating\nin full scan (34–750  m / z )\nin electron ionization mode. Nominal resolution was set to 60,000,\nwith AGC Target at 1 × 10 6 , and Maximum IT set to\nauto. The Orbitrap was calibrated with the internal calibration gas\nand evaluated before every sequence, typically every 1–3 days.\nThe ion source and transfer line temperature were 300 °C. The\noptimized method utilized a DB-5MS column (30 m × 0.25 mm ×\n0.25 μm, Agilent) and temperature gradient program starting\nat 30 °C for 1 min, increasing to 50 °C at 20 °C/min,\nthen ramping to 170 °C at 25 °C/min, to 250 °C at 6\n°C/min, and then increasing to 315 °C/min at 25 °C/min\nwith a 12 min hold. The carrier gas was helium at a constant flow\nof 1.3 mL/min. Injections of 25 μL of extract were made to a\nprogrammable temperature vaporizer (PTV) injector with a baffle liner\nin large-volume mode ( Table S3a ). The temperature\nprograms on the injection port and GC oven were coordinated. The starting\ntemperature was held at 30 °C to evaporate the injection solvent\n(isohexane) while retaining the semivolatile analytes. During method\ndevelopment, a shorter DB-5MS column (15 m) was compared, based on\ninstrumental detection limits (IDLs) of the target analytes, and 1\nμL of each solution was injected into each system in PTV splitless\nmode, with injection temperature increasing from 30 to 315 °C\nat 7.3 °C/s, and oven and MS conditions were the same as in the\noptimized method. The IDL was defined as the lowest concentration\nin an 11-point calibration curve (range 0.0025–50 ng/mL, triplicate)\nwith the signal-to-noise ratio >3, and at least 3 data points across\nthe peak. For LVI, an optimal injection volume was decided by comparing\nresponses for various injection conditions: 1 and 5 μL injections\nwith a small syringe (10 μL) and 5, 10, 25, 30, and 40 μL\ninjections with a larger syringe (100 μL); parameters for autosampler\nare shown in  Table S3b . For investigation\nof lipid removal effects, the extracts prepared by the literature\nmethod were analyzed by the same instrumental method as above, but\nonly with 2 and 5 μL injection volumes.\nFor application\nto cohort samples, calibration curves of target analytes were run\n3 times during the injection sequence (beginning, middle, and end).\nKovats retention index (RI) was applied, consisting of  n -alkane standards (C7–40 mixture, Merck, 25 ng/mL in isohexane)\ninjected separately at the beginning and end of the sequence. The\nentire sequence, including samples, calibration solutions, procedural\nand instrumental blanks, and Swedish pooled reference samples, lasted\napproximately 71 h (35 min gradient plus 15 min re-equilibration per\ninjection), while all samples were maintained at 10 °C in the\nautosampler. The instrumental blanks consisted of a clean solvent\nand were run multiple times in the sequence to monitor for carryover.\nMethod robustness was examined by\n60 continuous injections of the same spiked (2 ng/mL) serum extract\n(split into 12 vials), with an additional 12 injections of isohexane\nspread throughout the sequence, with one solvent injection after every\n5 sample injections. The whole sequence lasted 60 h, with no interruptions\nto the sequence. Target analyte recoveries, precision, and matrix\neffects were evaluated in triplicate spiked samples at 0.1, 1, and\n10 ng/mL for all analytes in 200 μL of the commercial serum.\nAn additional spike recovery experiment was also conducted at 75 ng/mL\n( n  = 3) for those analytes with higher background\nconcentrations. During method development, absolute recoveries (without\ninternal standard correction) were calculated from the peak areas\nof native analytes spiked to serum before sample preparation compared\nto the same spike added to processed serum extracts immediately before\nGC-HRMS analysis. For method validation, final recoveries of spiked\nnative analytes were calculated by the relative response to internal\nstandards spiked before extraction. Method precision was evaluated\nby relative standard deviations of the triplicate recovery experiments\nat 1 ng/mL (if not available because of the native serum background,\na value at a higher concentration was taken). For evaluating matrix\neffects, peak areas of spiked native analytes were compared between\n(i) extracts of serum and (ii) extracts of blank water. For analytes\nnot present in method blanks, the method limit of quantification (MLOQ)\nwas defined as the lowest concentration spiked to commercial serum\n(8 points between 0.005 and 10 ng/mL,  n  = 4), in\nwhich the peak area of the analytes had relative standard deviations\nbelow 20% with no internal standard correction but corrected for the\nvolumetric internal standard. In the cases where background levels\nof the analytes in commercial serum interfered, the response of the\ncorresponding isotopic labeled standard (spiked at 4 points between\n0.005 and 5 ng/mL,  n  = 4) was used to calculate the\nMLOQ when possible; otherwise, solvent-based calibration curves were\nused (for 3 analytes). In the cases where the analyte was present\nin method blanks, the MLOQ was alternatively calculated from the average\nblank response plus 10 standard deviations. The linearity was evaluated\nby the  R 2  of standard solvent calibration\ncurves (MLOQ–5 ng/mL,  n  = 3). Method accuracy\nwas examined in two ways, including by comparison of calculated target\nanalyte concentrations to a previous target analysis of the same VIP\nplasma samples (different aliquots) 6  and\nby comparison of target analyte concentrations in NIST SRM 1958 with\ncertified reference values.\nExternal\ncalibration curves\n(7 points, 0.005–5 ng/mL,  n  = 3) were used\nto quantify the 103 target analytes. TraceFinder (v.5.0, Thermo Scientific)\nwas used for peak detection, identification, and quantification of\nthe target analytes. Isotopically labeled surrogate internal standards\n( n  = 26) were used to correct the recoveries and\nvariations during sample preparation ( Table S4 ). Target analytes were generally absent in procedural blanks; however,\npeaks were detectable for certain phthalates and PAHs, and in this\ncase, analytes were only considered detected if sample peak areas\nwere 3 times higher than the corresponding procedural blank; concentrations\nwere also blank-subtracted in these cases.\nFor nontarget analysis,\nMS-DIAL 40  (v.4.9.221218, parameters in  Table S5 ) was used to align features and deconvolute\ncorresponding electron ionization spectra. Each nontarget feature\nhad a corresponding retention time (RT), RI, deconvoluted spectrum,\nand peak area. Mass spectral library matching was performed from the\ncombination of an in-house GC-HRMS Orbitrap library (244 chemicals),\nMassBankEU, 41  and NIST20 (version 2.4,\npurchased from Thermo Fisher Scientific, USA). Annotation confidence\nlevels were applied and defined according to the annotation scoring\nframework for GC-HRMS, 42  with RI matching\nscores considered.\nAnalyzed data were further processed and\nvisualized in Excel (Microsoft\nOffice 2019), Python (v.3.7.3), 43  Jupyter\nNotebook (v.5.7.8), 44  and R (v.4.3.2) 45  and RStudio (v.2023.12.1 + 402) 46  with ggplot2 (v.3.4.4).\nStatistical tests were performed\nin R (v.4.4.1) using the rstatix\npackage (v.0.7.2). Normality was evaluated by the Shapiro–Wilk\ntest, equal variance was evaluated by the F-test (Data Analysis Tool,\nExcel), and group differences were evaluated by two-tailed  t  tests, with  p -values adjusted for multiple\ntesting by the Bonferroni method, if the data were normally distributed\nand had equal variance. Otherwise, statistical differences were evaluated\nby the two-tailed Wilcoxon–Mann–Whitney test, with the  p -value adjusted by the Bonferroni method. For tests of\ncorrelation, the normality of the data was first tested by the Shapiro–Wilk\nmethod. Spearman's correlations were tested in R (stats package\nv.4.4.1).\n\nIn initial instrumental optimization,\nthe IDLs of the 69 halogenated analytes (PCBs, BDEs, OCPs, and PCDD/Fs)\nwere compared on two different GC column lengths (15 and 30 m; 0.25\nmm × 0.25 μm DB-5MS, Agilent) using 1 μL injections\nof standard mixtures between 0.0025 and 50 ng/mL ( Figure S1 , original data in  Table S6 ); phthalates and PAHs were not considered for this aim because of\nthe low-level background contamination for certain analytes in these\nclasses, making IDLs more difficult to determine on either column.\nFor approximately half of the analytes ( n  = 35),\nthe 30 m column resulted in better sensitivity ( Figure S1 , green), while for approximately one-third of the\nanalytes ( n  = 22), the column length had no significant\neffect on detection limits ( Figure S1 ,\ngray), and only 10 target analytes had better sensitivity on the shorter\n15 m column ( Figure S1 , red); two analytes\n(pentachlorobenzene and PCB-3) had very low IDLs (i.e., ≪2.5\nfg) that could not be adequately compared in the concentration ranges\ntested. Those analytes with lower detection limits on the 15 m column\nwere mostly late-eluting (RI > 2570) substances with higher boiling\npoints, including one OCP, five BDEs, and four PCDD/Fs. The highly\nbrominated BDEs are known to degrade on longer or thicker GC columns\nand thus perform better on shorter columns. 47  Nevertheless, for this multiclass chemical exposomics method, we\ndecided to use the 30 m column because of the 2–5 fold increased\nsensitivity for a majority (51.5%) of the analytes examined and considering\nthat this may lead to greater sensitivity for nontarget molecular\ndiscovery over most of the retention range.\nFor further\nmethod development, commercial human serum was spiked with 103 priority\ntarget analytes ( Table S4 ) representing\nthe wide chemical space of hydrophobic environmental contaminants\ntargeted by GC-based methodologies in national 48  or international biomonitoring programs. 49  These analytes belonged to 6 chemical classes (PCBs, BDEs,\nOCPs, PCDD/Fs, PAHs, and phthalates), ranged in molecular weight from\n128 (i.e., naphthalene) to 637 (i.e., BDE-154) Da, and had log  K ow  ranging from 1.7 (i.e., dimethyl phthalate)\nto 11.2 (i.e., dechlorane 603). 50\nFor the traditional GC-MS sample preparation of blood plasma extracts,\nit is a common strategy for major lipid interferences to be removed\nby destructive techniques under acidic conditions, such as by addition\nof concentrated sulfuric acid 51 , 52  or using chromatographic\ncleanup on acidified silica gel. 53  For\nthe suite of multiclass target analytes here, we briefly tested their\nrecoveries in acidified silica gel columns by loading standards in\nisohexane (10–13.5 ng/mL). However, this step was relatively\nlaborious and resulted in very low recoveries for a majority of PAHs\nand phthalates (<5%,  Figure S2 ), likely\ndue to degradation or hydrolysis under acidic conditions. 54 , 55  To avoid these issues, this step was abandoned, and we focused on\nsolvent extraction conditions that would minimize lipid coextractives.\nAcidic conditions were also avoided in further development; thus,\nprior to solvent extraction, we denatured and precipitated plasma\nproteins by addition of acetonitrile, rather than by formic acid. 29  Plasma protein precipitation by acetonitrile\nis a common technique in metabolomics protocols and is also compatible\nwith some LC-HRMS-based chemical exposomics methods, 31  thereby opening the future possibility to split deproteinized\nsupernatants for dual analysis by LC- and GC-based exposomics.\nFollowing the protein precipitation and centrifugation step, further\nmethod development focused on minimizing lipid coextractives from\nthe A-P supernatant. We first tested the effect of adding 40 mg of\ndispersive solid-phase extraction material (Bond Elut EMR-Lipid, preconditioned\nwith 250 μL of water) on relative recoveries into isohexane\nextracts (detailed in the SI). Although EMR-Lipid reduced major sterol\nlipids in the chromatograms ( Figure S3a , RT = 24.5 min), 25 late-eluting analytes (i.e., RT > 20 min)\nwere\nalso lost (average 20% absolute loss, range 10–28%). Moreover,\nthe precision was much lower using EMR, and few improvements to the\nrelative recoveries were evident ( Figure S3b ). Subsequent tests focused solely on solvent types and their ratios\nunder liquid–liquid extraction conditions, including pure isohexane\nand mixtures of isohexane with more polar solvents (isohexane: toluene\n(9:1); isohexane: ethyl acetate (2:1); isohexane: chloroform (9:1))\n( Figure S4 ). However, the addition of polar\nsolvents always resulted in decreased recoveries, in particular with\nethyl acetate, and especially for the more polar early-eluting analytes,\nlikely because the tested polar solvents partition significantly into\nthe acetonitrile phase. Moreover, the addition of toluene resulted\nin three layers and was abandoned. In all cases, the isohexane-containing\nextract layer (on the top) also contained more interference peaks\nof lipids or fatty acids when polar solvents were included.\nTherefore, it was concluded that 100% isohexane (H) was the optimum\nextraction solvent for the A-P supernatant; in subsequent sections,\nwe refer to this as the HA-P method. It is germane for us to note\nthat a recent study investigating various extraction conditions for\nplasma exposomics reported the best quantitative performance for a\nhexane-acetonitrile-plasma extraction condition, but the authors chose\nalternate methods due to the assumption that the hexane layer would\ncontain lipid interferences; 56  as discussed\nlater, the hexane layer in the HA-P method was in fact free from most\nlipid interferences.\nIn final HA-P method optimization, we compared\ntarget analyte recoveries\nwith different volumes of isohexane for the primary extraction (600\nμL, 1.2, and 2 mL). Although the larger volume (2 mL) slightly\nimproved recovery for some analytes (average 3%), precision was lower\n(i.e., see higher standard deviations in  Figure S5a ). Therefore, 1.2 mL was determined to be the optimum volume\nof isohexane. Moreover, we tested the effect of performing a second\nfollow-up extraction with the addition of isohexane. The addition\nof this secondary isohexane extraction (400 μL) after the primary\nextraction (0.6 mL) increased the absolute recoveries on average by\n8.5% (range 1.1–14.4%,  Figure S5b ), particularly for later-eluting analytes. Thus, this method was\nincluded in the optimized HA-P method for validation tests.\nIn order to increase the sensitivity\nfor trace levels of contaminants, an LVI method was applied and optimized.\nThe initial parameters (temperature gradient ramp, split flow rates,\netc.) were first optimized with pure standards ( Table S4 ), and the optimal injection volume was selected based\non performance with a commercial serum extract, prepared by the optimized\nHA-P method and spiked at 10 ng/mL. Peak areas increased with larger\ninjection volumes between 1 and 40 μL ( Figure S6 ), but at 30 and 40 μL, the peak areas did not further\nincrease linearly, and some analytes had greater standard deviations\nand split peaks, suggesting overloading at these injection volumes.\nThus, 25 μL injections were chosen as optimal for method validation\ntests. Under these conditions, method precision and robustness were\ntested over 4 days by 60 continuous injections of the same spiked\nsample extract (2 ng/mL). Although peak areas for many target analytes\ndecreased slowly over the entire sequence (mean of 28%), this minor\neffect was adequately controlled by internal standard correction.\nThe median RSD was 5.1% for target analytes over 4 days (range 1.6–24.5%,  Figure S7 ).\nUsing the optimized HA-P method with\nLVI (25 μL), we performed method validation for all target analytes,\nincluding determination of matrix effects, internal standard corrected\nrecoveries, calibration linearity, and MLOQs ( Table S7 ). The internal standard for correcting each native\nanalyte was selected based on similar retention, absolute recovery,\nand matrix effect. The median MLOQ was 0.088 ng/mL (range 0.005–4.83\nng/mL;  Figure  1 a),\nand for most analytes (99 out of 103), calibration curve linearity\nwas good ( R 2  > 0.99) between the MLOQ\nand 5 ng/mL. Two phthalates (diethyl phthalate and dibutyl phthalate)\nhad relatively lower  R 2  values (0.98 each),\nand diisobutyl phthalate and di(2-ethylhexyl) phthalate were semiquantified\nby single-point calibration because of background interference at\nlower contamination (detailed in  Table S7 ). When all analytes were spiked to native serum at 1 ng/mL, among\nthe 91 detectable analytes (out of 103), the mean matrix effect was\nnull (i.e., mean 100% response; range 76–140%) and the median\ninternal standard recovery was 104% (range 22–132%;  Figure  1 b). For 22 analytes\nwith interfering background levels in method blanks or commercial\nserum, spiking experiments at 10 or 75 ng/mL showed a mean recovery\nof 96% and a median matrix effect of 110% ( Table S7 ). The lowest recoveries (22–41%) and worst matrix\neffect (513%) were found for three phthalates, specifically dimethyl\nphthalate, diethyl phthalate, and diethoxyethyl phthalate, likely\nowing to their lower hydrophobicity (log  K ow  = 1.6, 2.5, and 2.1) 57  and their presumed\npreferential partitioning to the plasma-acetonitrile layer during\nextraction. All other spiked target analytes have log  K ow  > 3; thus, we suggest that the HA-P method is only\nappropriate for analytes in this more hydrophobic range. This limitation\nof the HA-P method is not a limitation for chemical exposomics in\ngeneral, for example, dimethyl phthalate and diethyl phthalate are\nmore sensitive and quantitative by LC-HRMS-based chemical exposomics, 58  and it is understood that GC-HRMS will need\nto be applied together with LC-HRMS for comprehensive analysis of\nthe exposome. 9  Finally, it is noteworthy\nthat the relatively low recovery of β-HCH and relatively high\nrecovery of α-HCH ( Figure  1 b) are likely due to interconversion of these isomers\nin the presence of water (i.e., plasma) 59  and therefore not necessarily a limitation of the analytical method.\nMethod\nvalidation results for multiclass target analytes, arranged\nby classes, spiked to 200 μL of human serum and analyzed by\nthe HA-P method with LVI, showing (a) MLOQ ( n  = 4)\nand (b) internal standard corrected analyte recoveries ( n  = 3, 1 ng/mL spiking levels). Panel (a): MLOQ of BDE-154 was >10\nng/mL and is not plotted. Panel (b): 13 analyte recoveries were calculated\nat 10 or 75 ng/mL (detailed in  Table S7 ).\nThe performance of the HA-P method\nwith LVI was also examined by\nanalysis of NIST SRM 1958 (i.e., a freeze-dried fortified human serum),\nwhich has certified or noncertified reference values, for 33 of the\nanalytes targeted here. For most of these ( n  = 23),\nquantified concentrations by the HA-P method were within 75–125%\nof certified/noncertified values (i.e., ratio 0.75–1.25,  Figure S8 ), and the mean ratio for all analytes\nwas 0.98 (range 0.32–1.43). Notable outliers were again HCH\nisomers (i.e., low concentrations of β-HCH and high concentrations\nof γ-HCH), which have noncertified values in the SRM and are\nknown to interconvert as described above. The authors of the literature\nmethod reported that the ratios of 20 target analytes were within\n75–125% of certified/noncertified values and that the mean\nratio for all analytes was 0.9 (range 0.3–1.79). 29  Overall, the current chemical exposomics method\nperforms similarly as well as the literature chemical exposomics method\nfor target analysis based on analysis of the same SRM (see  Figure S8  for comparison). 29\nDuring\nmethod development, it was noted that the total ion chromatograms\nof HA-P method extracts were relatively clean (visually) and free\nfrom large interfering peaks in GC-HRMS analysis ( Figure S3a ). This suggested that few lipid species had been\ncoextracted and may explain why dispersive solid-phase extraction\nby EMR-Lipid had no beneficial effect in our method development tests.\nThe major lipids in human plasma include glycerolipids, glycerophospholipids,\nand sterol lipids such as cholesterol esters. 26  Nonpolar bulk lipids (i.e., glycerolipids and some sterol esters)\ngenerally have very low solubility in polar solvents such as acetonitrile\nand are known to be removed with proteins during protein precipitation. 60  Most phospholipids remain in the plasma-acetonitrile\nphase during extraction, and here, we found no traces of phospholipids\nin the isohexane extract ( Figure S9 , cyan).\nTo fully understand the relative extent of lipid interferences,\nand the impact of these on chemical exposomics, we compared plasma\nextracts from the HA-P method (in 100 μL of isohexane) to extracts\nof the same plasma by a literature method (in 200 μL of ethyl\nacetate). 29  The relatively clear appearance\nof extracts from the HA-P method was evident, relative to yellow-colored\nextracts by the literature method ( Figure  2 ). Consistent with visual appearances, after\ninjecting 2 μL of each extract ( Figure  2 a,b), the corresponding total ion chromatograms\nrevealed a comparably complex matrix for the literature method, with\nmany abundant coextracted substances. The relatively clean total ion\nchromatogram of the HA-P extract is noteworthy considering that twice\nas much plasma equivalents were injected on-column. The major chromatographic\npeaks for the literature method extract included long-chain fatty\nacids (RT = 11.19, 13.77, 16.46 min) and sterol lipids (RT = 16.06,\n20.96–24.61 min,  Figure S10b ), which\nwere either absent or much lower in the HA-P method, even with 10-fold\nmore plasma equivalents injected on-column ( Figure S10a ).\nTotal ion chromatograms and photos of extracts from pooled\nSwedish\nplasma (200 μL,  n  = 3 each) prepared by the\nHA-P method (cyan, left panels) and a literature method (brown, right\npanels) injected with various volumes to GC-HRMS. Chromatograms for\nthe HA-P method extract are shown for a 2 μL injection (a) and\na 25 μL injection (c), corresponding to 4 and 50 μL of\nplasma equivalents on-column, respectively, and these can be contrasted\nwith chromatograms for the literature method extract for a 2 μL\ninjection (b) and a 5 μL injection (d), corresponding to 2 and\n5 μL of plasma equivalents on-column. Photos of the associated\nsolvent extracts are also shown for (a) HA-P method in 100 μL\nof isohexane and (b) from the literature method in 200 μL of\nethyl acetate, with 100 μL taken for photography.\nFor the extracts from the literature method, the MS response\nwas\nsaturated for many of the largest peaks, as maximum peak heights were\nin the range of 2 × 10 10  and did not increase with\nincreasing injection volumes from 2 to 5 μL. The largest peaks\nin TICs of both extracts were for sterol lipids (RT = 24.5 min); thus,\nboth methods are still similarly prone to interference from these\nmajor plasma metabolites. When the injection volume of the plasma\nextract was increased from the literature method ( Figure  2 d), the major chromatographic\npeaks became visibly broader, suggesting that the GC column was overloaded,\nand larger injection volumes were not attempted. In comparison, a\nlarger injection volume of the HA-P extract (25 μL, corresponding\nto 50 μL plasma equivalents on-column,  Figure  2 c) was nevertheless applied, and the total\nion chromatogram was still relatively clean but did show elevated\nbaseline at later retention times (>25 min). This increased background\noverlaps with the retention range of 18 (17.5%) of the target analytes;\nthus, its potential as a minor interference cannot be discounted.\nAs shown in extracted ion chromatograms of example target analytes\nin the HA-P method ( Figure  3 a,b), the increased injection volume resulted in larger target\nanalyte peaks and also revealed new detectable analyte peaks that\nwere not evident at lower injection volumes (e.g., trans-nonachlor,  m / z  408.7840 at 16.04 min, and PCB-105,  m / z  325.8810 at 18.25 min), demonstrating\nenhanced method sensitivity by LVI for low-abundance analytes in plasma.\nUsing 200 μL plasma aliquots of pooled Swedish plasma, 16 target\nPCBs/OCPs (0.02–3.3 ng/mL) were consistently detected in triplicate\n( n  = 3/3) samples by the HA-P method (25 μL\ninjection), compared to only 5 target analytes (0.17–2.8 ng/mL)\nby the literature sample preparation method (2 μL injection)\nand the same instrumental settings ( Figure  3 e,f). Among the 11 PCBs/OCPs not consistently\ndetected by the literature method, most had plasma concentrations\nbelow 0.2 ng/mL, according to the HA-P method, except for PCB-180\n(0.45 ng/mL), which was inconsistently detected (2 of 3 replicates)\nby the literature method, despite relatively higher concentrations.\nExample\nextracted ion chromatograms (EICs, a–d), as well\nas concentrations and detection frequencies (e, f) of analytes in\ntriplicate extracts of pooled Swedish plasma (200 μL aliquots)\nprepared by the HA-P method and literature method and injected to\nGC-HRMS with various injection volumes. Panels (a) and (b) present\nthe HA-P method and injection volumes of 2 and 25 μL. Panels\n(c) and (d) present the literature method and injection volumes of\n2 and 5 μL, respectively. [Top row (panels a–d) is the\nEIC of ion  m / z  221.9998 for PCB\n13 (Level 4 annotation, RT = 11.39 min), PCB-15 (RT = 11.89 min),\nand other dechlorinated PCBs. Middle row (panels a–d) is the\nEIC of  m / z  408.7840, corresponding\nto trans-nonachlor (RT = 16.03 min). Bottom row (panels a–d)\nis the EIC of ion  m / z  325.8810 for\nPCB-118 (RT = 17.54 min) and PCB-105 (RT = 18.25 min).] Panel (e)\nshows general agreement of quantified targeted analyte concentrations\nby the two methods above 0.2 ng/mL, but below this approximate threshold,\nthe literature method resulted in nondetection for most analytes;\nmean concentrations and detection frequencies are listed in panel\n(f).\nAll target analytes detected and\nquantified in SRM 1958 ( n  = 33 analytes) had similar\naccuracies by the HA-P method\nand literature method, including those analytes present at moderate\nto high concentrations (i.e., 293–1250 ng/kg, certified values)\nand at lower concentrations, such as for PCB-123 and PCB-114 (0.0525,\n0.0466 ng/kg, noncertified values) ( Figure S8 ). When analyzing the pooled Swedish plasma in triplicate by the\nHA-P and literature methods, the detection and quantification were\nsimilar by both methods when concentrations were above 0.2 ng/mL ( Figure  3 e,f). The major discrepancies\nwere at lower concentrations, where most analytes were nondetectable\nby the literature method, thereby demonstrating enhanced sensitivity\nby the HA-P extraction method and LVI.\nAn additional observation\nin the analysis of the extracts produced\nby the literature method was that, when increasing injection volumes\nfrom 2 to 5 μL, some analyte peaks disappeared (e.g., PCB-13\nat 11.39 min,  m / z  221.9998) ( Figure  3 c,d). This may be\ndue to abundant interferences ( Figure  4 a), and the auto gain control function of the Orbitrap\nmass spectrometer, which applies a dynamic ion injection time (i.e.,\nto the C-trap and subsequently to the Orbitrap analyzer) throughout\nthe analytical run to balance sensitivity and mass spectral resolving\npower. 35 , 36  For a very clean injection, as shown for\nanalysis of instrumental blanks composed only of isohexane ( Figure  4 b, black), the ion\ninjection time was initially maximal (i.e., 112 ms), thereby allowing\nmaximum signal to the Orbitrap analyzer, but declined to approximately\n20 ms after 22 min due to increasing background signal from column-bleed\nat higher temperatures (300 °C at 22 min); overall, the mean\nion injection time throughout the analysis of isohexane instrumental\nblanks was high (mean = 73 ms; 97 ms before 25 min). In comparison,\nlower ion injection times were observed with LVIs of the HA-P extract\n(mean: 38.6 ms, 56 ms before 25 min,  Figure  4 b, cyan). Nevertheless, these ion injection\ntimes were still substantially higher than that for the extract produced\nby the literature method (mean: 3.22 ms, 3.85 before 25 min,  Figure  4 b, brown), even considering\na 10× more plasma-equivalent volume injected on-column from the\nHA-P extract ( Figure  4 b; 50 μL plasma equivalents by HA-P (25 μL injection)\nand 5 μL plasma equivalents by the literature method (5 μL\ninjection)). Lower ion injection times predictably corresponded to\nregions of the chromatograms with a higher total ion signal ( Figure  4 a). For analytes\nthat are still detectable, a correction factor is applied by the software\nto maintain quantitative analysis, 61  but\nfor trace analytes near detection limits, the signal can become nondetectable\ndue to the lower ion injection time, as shown for PCB-13 in the extract\nfrom the literature method ( Figure  3 d).\nComparison of (a) total ion chromatograms and (b) ion\ninjection\ntimes for extracts of pooled Swedish plasma prepared by the HA-P method\n(cyan, 50 μL plasma equivalents on-column) and by a literature\nmethod (brown, 5 μL plasma equivalents on-column), both injected\nto the same GC-HRMS. In both plots, the instrumental blank is also\nshown (black, 25 μL isohexane) for comparison.\nThe HA-P method with LVI was applied to 32 individual plasma samples\nof Swedish adults as well as to pooled Swedish plasma for reference\nand quality assurance. Among all samples, 51 (out of 103) target analytes\nwere detected in at least one individual ( Table S8 ). The detected analytes included 7 dioxin-like PCBs (#105,\n#114, #118, #123, #156, #157, and #167), 14 nondioxin-like PCBs (#1,\n#3, #4, #19, #15, #28, #52, #37, #101, #138, #153, #202, #180, and\n#205), 9 PAHs, 12 OCPs, 1 BDE, and 8 phthalates. Among these, 28 analytes\nhad detection frequencies above 20%, the distributions of which are\nshown by sex ( Figure  5 ), and 9 analytes (β-HCH, DEHP, PCB-156, pyrene, p,p’-DDT,\nDiBP, PCB-167, HCB, and acenaphthene) showed gender differences ( p  < 0.05), but they were not significant anymore after\nthe Bonferroni correction. Contrasting the current results to previous\ntarget analyses of the same plasma samples (separate aliquots) 6  showed linear associations between the two methods\n( Figure S11 ).\nViolin plots showing\nconcentrations and distributions for detected\ntarget analytes in 32 individual Swedish adult plasma samples by sex\n(female in purple, male in green). Panel (a) shows 28 analytes with\ndetection frequencies >20%, while panel (b) shows 23 analytes detected\nat lower frequencies (i.e., in 1–6 samples). Black “+”\nsymbols indicate the target analyte MLOQ. For data visualization,\ndetectable signals below the MLOQ are plotted as MLOQ/2, and nondetectable\nsignals are plotted as MLOQ/4. For analytes marked with black “*”\nsymbols, the MLOQs were calculated by methods other than the native\nstandard in the matrix-matched calibration curve, detailed in  Table S7 . Values of DBP were extrapolated from\nthe calibration curves, and DiBP and DEHP were semiquantified using\none point.\nCorrelations between concentration\nand sampling year were tested\nby the Spearman method as most data were not normally distributed.\nTwelve target analytes showed statistically significant temporal associations\nbetween 1990 and 2013 (test method and statistics in  Table S8 ), six of which are shown in  Figure  6 . Particularly, PCBs and OCPs were negatively\nassociated with sampling year, consistent with bans and restrictions\nthat started in the 1970s. 62 , 63  To the contrary, DEHP,\na commonly used phthalate plasticizer, showed a positive association\nwith sampling year, from below 10 ng/L in the early 1990s to >40\nng/L\nby 2005, and possibly lower concentrations thereafter. We acknowledge\nthat no field blanks were available in this cohort to rule out phthalate\ncontamination from medical sampling equipment, but similar or higher\nlevels of DEHP have been reported in other studies. For example, mean\nDEHP was 180 ng/mL (maximum 1030 ng/mL) in healthy women in the 2000s 64  and 65 ng/mL in the serum of women diagnosed\nwith endometriosis in the 2010s. 65\nConcentrations\nof example target analytes, with statistically significant\nnegative associations with sampling year; the Spearman correlation\ncoefficient (ρ) and statistical significance ( p ) are shown in each plot.\nWe found no significant associations between the concentrations\nof these target analytes and individual metadata such as birth year,\nsampling age, meat consumption (Spearman correlation test), or smoking\nstatus (two-tailed Wilcoxon mean value test). Nevertheless, concentrations\nof several PCBs (#15, #156, and #180), as well as β-HCH and\nHCB, were significantly higher in snuff users than nonsnuff users\n(means: 0.008 vs 0.002 ( p  < 0.01), 0.07 vs 0.05\n( p  = 0.02), 0.37 vs 0.05 ( p  = 0.02),\n0.06 vs 0.05 ( p  = 0.03), and 0.15 vs 0.11 ng/mL ( p  = 0.05), respectively, for PCB-15, 156, 180, β-HCH,\nand HCB), whereas DEHP was lower in snuff users (mean: 8.5 vs 21.4\nng/mL;  p  < 0.01) by the two-tailed Wilcoxon mean\nvalue test. However, snuff users were not equally distributed over\nsampling years (with fewer snuff users in later years); thus, these\nresults are likely confounded by the associated temporal trends.\nWe\nadditionally evaluated the suitability of the HA-P method with LVI\nfor the discovery of unexpected substances in a nontarget exposomics\nworkflow. After data processing by MS-DIAL and blank filtration, a\ntotal of 875 molecular features were detectable among all individual\nplasma samples (see  Table S9 , including\nrelative responses). Among these features, 112 were matched to reference\nlibrary spectra, corresponding to a relatively high annotation rate\nof 12.8%. The annotations included 30 of the target analytes (confirmed\nLevel 1, 42  ΔRT < 1%), as well\nas 82 new annotations (Level 2 confidence, 42  ΔRI < 50) that included 28 prospective environmental substances.\nAuthentic standards for 8 of these environmental chemicals were purchased,\nresulting in 7 confirmed identifications (Level 1, ΔRT <\n1%, chromatograms and spectra in  Figures S12–18 ). These included the related analytes 2,4-di- tert -butylphenol ( Figure S12 ) and tris(2,4-di- tert -butylphenyl) phosphite ( Figure S18 ), which have been used as antioxidants and UV stabilizers\nin rubber and plastics; the co-occurrence of the two was reported\nin indoor dust in 2018. 66  Although 2,4-di- tert -butylphenol, which is a degradation product of the\nlatter 67  and other substances, has been\ndetected in human blood and urine previously, 68 , 69  we are not aware that tris(2,4-di- tert -butylphenyl)\nphosphite has been reported previously in human biomonitoring. This\nlatter substance was previously reported in chemical migration tests\nfrom polymeric materials to water and simulated foods. 66 , 67 , 70 , 71  This analyte has a high boiling point and very late elution time\nin our method (26.14 min) and in some methods may suffer from high\nbackground interferences. The discovery of this substance and, in\ngeneral, the high nontarget annotation and confirmation rates are\nlikely due to a combination of compounding factors, including higher\nsensitivity from LVI and lower matrix interference in the HA-P extracts.\nThe low matrix interference by the HA-P method may not only result\nin higher sensitivity (i.e., due to higher ion injection times) but\ncould also improve the performance of the  in silico  spectral deconvolution (i.e., in MS-DIAL), resulting in higher quality\nspectra that will have better matches to spectral libraries.\nIn this study, a comprehensive multiclass target and nontarget chemical\nexposomics method was developed and validated for human plasma. With\nthe combination of lipid removal at the extraction stage and LVI,\nthe method sensitivity was significantly improved in comparison to\na literature GC-HRMS method without these steps. These developments\nresulted in more target analytes being consistently detected and a\nhigher rate of molecular annotation and confirmation in nontarget\nanalysis. An important reason for the enhanced method performance\nwas the longer ion injection times achieved in the Orbitrap instrument\nthroughout the GC run time owing to a much lower signal from coextracted\nlipids. The method was rugged and can be applied to larger studies\nof the chemical exposome in combination with LC-HRMS-based analysis.","source_license":"CC-BY-4.0","license_restricted":false}