Sample storage, stability, and handling are crucial variables affecting analytical outcomes. The impact of pre-analytical procedures on coverage of endogenous metabolites has been comprehensively reviewed^33–35 and guidelines for biological specimens elaborated.^36,37 Chemical lability has been of minor concern for the measurement of exogenous compounds because the traditional chemicals of focus are persistent (e.g. persistent organic chemicals) or relatively stable (e.g. pesticides). These chemicals could potentially be measured many (tens of) years after sample collection with minimal provisions undertaken for storage. Yet, more comprehensive characterization of the chemical exposome extends the focus toward more polar and labile contaminants.³⁸ Cohort studies often involve long storage time (tens of years) and implementing greater protective measures for storage would enable the prospective integration with assays of less stable molecules (e.g. proteins). Hebels et al. reported that so long as consistent procedures have been implemented, biobank samples (stored at –80°C) would be suitable for meaningful multiomics analysis even after long storage periods (∼15 years).³⁹

Studies comparing the long-term impact of different storage conditions are limited but some generic measures beneficial to preserve sample integrity include (1) quenching (e.g. dry ice, liquid nitrogen) immediately after sample collection and appropriate sample sealing to prevent losses of volatile compounds⁴⁰; (2) storage at −80°C or −196°C using liquid nitrogen, to prevent biotransformation reactions and prevent potential losses^41,42; (3) limiting storage time and frosting/defrosting cycles (e.g. by aliquoting subsamples in advance) and minimizing sample handling.^40,42

Further investigations to assess the long-term impacts of storage conditions on chemical components of biospecimen and environmental matrices are encouraged, and are especially critical for large-scale studies, where the analyzed samples have often been stored for different periods of time. The diversity of intracellular biochemical reactions that exogenously-derived chemicals are involved is typically less than for endogenous compounds⁴³ which facilitates comparatively more accurate prediction of biotransformation and modeling of chemical exposure than for metabolic networks. Approaches similar to the ones used to estimate chemical biodegradation could be attempted to model biotransformation reactions in biospecimens over storage. For example, Quantitative Structure-Biodegradability Relationships (QSBRs) are widely used to predict chemical biodegradability by using molecular descriptors.⁴⁴ Time and cost-efficient approaches to extrapolate half-lives reading across different environmental media have been recently proposed for small molecules, where activated sludges were used as accelerated reactors to calculate half-lives of various agrochemicals and regression models developed to extrapolate to soil half-lives.⁴⁵ Similarly, in vitro “reactors” with biological extracts enriched with degradative molecules and enzymes or genetically modified microorganisms (e.g. to over-express cytochrome P450 enzymes) could be used to provide biotransformation rates of chemicals in biological specimens to eventually extrapolate with temperature corrections (e.g. by Arrhenius equation)⁴⁶ or correlations with measurements at lower temperatures.

To reduce variability across biological samples and obtain comparable data, sample normalization is fundamental, especially when approaching quantification.^47,48 Otherwise, the correct interpretation of the results can be confounded by large inter-individual and intra-individual variation. For example, dilution factors for urine may vary up to a factor of 15,⁴⁹ leading to different detection limits and potential matrix interferences. For more details on normalization strategies for biological specimens (e.g. protein/lipid normalization, specific gravity/osmolarity, weight/volume base methods), numerous reviews are available.^34,47–49

Pre-analytical normalization is more relevant for exocrine-related glands/tissues/fluids (e.g. urine, kidney, gastrointestinal tract) compared to endocrine (e.g. blood, thyroid, ovaries, etc.); a consequence of the role in maintaining homeodynamic regulation. High variability limits comparative analysis of relative abundances due to varying limits of detection, and is further complicated, e.g. by signal ratio bias from non-linear ESI responses. This is particularly problematic for low abundant features because it reduces detection frequencies and sparsity cannot easily be corrected post-acquisition. On the other side, pre-analytical procedures to minimize signal variability typically involve dilution of samples, reducing the overall number of compounds detectable. Analyzing serial dilutions of representative samples to assess potential matrix effects by evaluating response linearity was recently recommended over procedures with extensive sample manipulation, with direct injection demonstrated to provide more reliable semi-quantification in many cases.^50,51

Post-analytical normalization strategies by using dedicated software and statistical approaches can be used to correct for sensitivity drifts and reduce variability across samples.⁴⁸ However, exogenously-derived compounds are present at lower abundance and with reduced frequency compared to constitutive metabolites and proteins, limiting applicability and validity of typical corrective procedures (e.g. noise thresholds and detection frequency filters, missing value imputation).⁵²

Longitudinal studies become even more complex because analytical batch variation is often larger than temporal changes, so randomizing measurements may not be enough to ensure data comparability. In addition, at the point of analysis, it is common that samples have been stored for varying lengths of time, adding further complexity. Consequently, the implementation of normalization procedures requires careful consideration for each chemical exposomics study and more long-term storage studies would also help to advance identification of suitable normalization parameters.

Sample fractionation has been highlighted as a potential approach for the in-depth characterization of biospecimen because greater enrichment factors can be achieved through concentration of fractions and a combination of platforms can be used for more focused analysis of different fractions.²⁴ However, each stage of purification entails losses and the initial sample amount available is the main limiting factor.

Solvent partitions allow for the separation of analytes in different fractions based on their hydrophobicity that can be processed separately or purified and then pooled together. For example, Pourchet et al. proposed a non-targeted workflow for investigating halogenated compounds in human milk after protein and lipid removal by using solvent partition to separate analytes in two different fractions for GC-HRMS and LC-HRMS analysis.⁵⁵ Furthermore, hot solvent mixtures,¹⁴ microwave,⁵⁶ and ultrasound⁵⁷ can be used to enhance solvent extraction efficiency. SPE-based fractionations, via sequential elution^58,59 or the use of sequential phases,⁶⁰ have been evidenced to increase metabolome coverage⁶¹ and can equally be applied for enhancing coverage of the chemical exposome, capitalizing on the wide range of SPE sorbents are commercially available. For example, step-wise elution from mixed-mode SPE was used to separate neutral and acidic polycyclic aromatic compounds in runoff water samples and fractions screened via GC- and LC-MS, uncovering numerous emerging biotransformation products,⁶² while a two-stage hydrophilic–lipophilic balance (HLB)-activated carbon SPE purification was applied to screen hundreds of chemicals in waters.⁶³ Plates and cartridges provide an alternative to the traditional precipitation techniques for removing proteins and lipids.^64,65 For example, a phospholipid depletion followed by cation exchange SPE enabled greater concentration of plasma sample extracts prior to instrumental analysis compared to classical protein precipitation, showing increased sensitivity.⁵⁴

The use of molecularly imprinted polymers (MIPs) represents a promising approach to selectively purify specific structural classes of compounds, e.g. halogenated,⁶⁶ organophosphate,⁶⁷ and glucuronidated⁶⁸ chemicals, and shows greater cleanup efficiency to enable enhanced concentration of low abundant analytes.⁶⁹ Similarly, chemoselective probes that consist of a solid support containing a probe with specific sites to bind target groups and cleavage linkers to release the compound in certain conditions⁷⁰ have been successfully applied for enrichment in metabolite profiling studies,^70,71 e.g. targeting amines,⁷² aldehydes, and ketones.⁷³ Protein receptors can be used as probes for selection of biologically-active chemicals⁷⁴ and multiple MIPs and/or probes can be used in tandem to enable highly resolved fractionation. However, applications are limited by the lack of commercially available chemistries.

Chemical reactions (e.g. deconjugation, derivatization) can promote analyte release from matrix components and/or enhance ionization efficiency. Deconjugation has been commonly used in target assays (e.g. for bisphenols) to measure the total concentration of the parent compounds⁷⁵ and recommended for suspect screening to confirm the identity of parent substances by comparing non-deconjugated and deconjugated samples.⁷⁶ Similar comparative profiling has also been conducted using glucuronidase and sulfatase with high activity and promiscuity for non-target analysis of glucuronide⁷⁷ and sulfate metabolites,⁷⁸ respectively, greatly increasing analyte detection. Adoption in chemical exposomics studies would advance toward more comprehensive screening of Phase 2 detoxification metabolism. However, obtaining such high activity requires multistep purification from commercial enzyme extracts, thus requiring a great level of expertise. Alternatively, typically more affordable chemical deconjugations could be used but their decreased specificity complicates annotation/identification.

Different reaction rates mean it is difficult to obtain complete deconjugation for a wide range of substances⁷⁹ and challenges to standardize procedures and reaction conditions often leads to large intra- and inter-lab variability.²⁷ Furthermore, unexpected transformation products and degradation of parent compounds can occur⁸⁰ and information of biotransformation patterns can be lost, which is detrimental for fate and toxicity assessment across the chemical life cycle.

Besides deconjugation, enzymes can be used to assist extraction by degrading cell/membranes and matrix components.⁸¹ For example, Wawrzyniak et al. applied a mild proteinase K digestion to human plasma to misfold proteins and induce the release of associated metabolites.⁸² The method increased metabolome coverage compared to direct precipitation, more than doubling the number of annotated, reproducible features, though the formation of non-specific protein fragments was also noted. Application of a protease treatment step could also be beneficial for the enhanced detection of environmental chemicals since many are known to associate with proteins, such as per- and poly-fluorinated substances.⁸³ Similarly, the potential of other biological enzymes (e.g. lipases) could be explored as milder digestion alternatives to traditional strong acid/alkali treatments often employed.⁸⁴

Chemical derivatization can be used to make compounds more volatile for GC,⁸⁵ extend the hydrophobicity range covered by LC (e.g. for better separation of very polar compounds)⁸⁶ and increase detection sensitivity through the addition of moieties that improve ionization, e.g. halogens or alkyl groups. Various derivatization agents can be combined to target diverse groups of analytes⁸⁷ and improve confidence for annotation (e.g. comparing labeled and non labeled sample fractions/features), although libraries will need to be extended to include more derivatized compounds.⁸⁸ For example, Zhao et al.⁸⁷ proposed a 4-channel chemical isotope labeling method to quantify hydroxyls, amines/phenols, carboxylic acids, and ketones/aldehydes. However, optimization of reaction for each sub-class is complex. The slow reaction speeds can limit throughput for use of some derivatization agents⁸⁶ and the potential instability of derivates makes it challenging to ensure reproducibility without automatization.^89,90 Although usually performed in the latest stage of sample preparation,⁸⁶ derivatization before sample extraction can be used to increase the affinity of polar compounds toward polymeric phases (e.g. SPE cartridges)⁹¹ or solvents. The potential of this approach has been demonstrated for assessment of the urinary chemical exposome, with enhanced detection sensitivities from 2- to 1184-fold.⁹²

The main microextraction techniques can be classified as solved-based i.e. liquid-phase microextraction (LPME), or sorbent-based, i.e. solid-phase microextraction (SPME).⁹³ LPME consists liquid–liquid extractions using small solvent volumes, with the most common approaches being (1) dispersive liquid–liquid microextraction (DLLME), where droplets of a non-polar solvent are finely dispersed in an aqueous sample allowing for a rapid extraction; (2) hollow fiber-protected liquid-phase microextraction (HF-LPME), where analytes are extracted by a solvent placed in the pores of a hollow fiber and then transferred in an acceptor phase and (3) electromembrane microextraction (EME), which is similar to HF-LPME but supported by the application an electrical field. LPME present numerous advantages, such as reduced solvent consumption, low cost, broad selectivity, and rapidity. However, working with such small volume extracts can pose challenges for automation. More details on these techniques and applications can be found in recent reviews^94,95 and their potential to improve selectivity and efficiency in sample clean-up should be further investigated for chemical exposomics. The advantages of DLLME for broad extraction for multiple novel psychoactive compounds have been shown.⁹⁶

SPME has been more widely applied in metabolomics research,^40,97 often showing performance comparable or even better than traditional approaches.^98–101 Non-exhaustive extraction methods, such as SPME, can provide better sensitivity for low abundant species since suppression from matrix components is reduced.^98,99 Comparatively, a limited number of studies have been published on non-target analysis of environmental chemicals using passive sampling approaches,^102–107 but applications in targeted studies show their potential for concentrating low abundant chemicals from complex matrixes (e.g. hydrophobic organic chemicals in waters or soil/sediment porewaters).^108,109 Passive sampling (e.g. by SPME, silicone rods, etc.) provides many appealing features. First, sample extraction, clean-up, and concentration are conjugated in one step, enhancing reproducibility for sample preparation and throughput. Second, if operated in non-depletive mode, passive sampling allows insights of chemical activity because spontaneous processes such as partitioning will be driven by activity gradients until equilibrium is reached.^110,111 Furthermore, the use of passive sampling devices provides a good balance between hydrophilic (mainly in freely dissolved form) and hydrophobic species (mainly associated with matrix components but usually with stronger affinity for polymeric phases),⁴⁰ and is a promising technique for in situ¹⁰⁴ and in vivo¹¹² applications. The potential of wearable passive samplers to measure a large array of chemicals across individuals was recently reviewed.¹¹³ Their low-cost, noninvasive nature make them practical for large-scale deployment and integration with ecological momentary assessments enables the spatiotemporal variability of an individual’s chemical exposure to be captured. However, attention must be paid to potential artifacts with careful consideration to placement and sampling rate alteration, e.g. through obstruction such as coverage by clothing, personal care product use if in contact with skin, air velocity alterations for individual behaviors, etc.¹¹³

Sorbent choice is critical and most metabolomics applications have favored commercially available polydimethylsiloxane/carboxen/divinylbenzene (PDMS/Car/DVB) to provide broad coverage. Although currently limited to coatings made in lab, (HLB) particles have been shown to provide good performances in terms of sensitivity and extend coverage toward more polar analytes.^100,114,115 For example, Vasiljevic et al. developed biocompatible SPME minitips coated with HLB particles and demonstrated their application for metabolomics and the targeted analysis of drugs of abuse,¹¹⁵ a combination polystyrene-divinylbenzene-weak anion exchange (PS-DVB-WAX)/HLB (50/50, w/w) fiber was demonstrated to cover a wide range of hydrophobicity (−7 < log P < 17) and molecular weight (100-1000 Da)¹¹⁶ and Gionfriddo et al. proposed a polytetrafluoroethylene amorphous fluoroplastics (PTFE AF)/HLB combination suitable for both GC and LC applications.¹¹⁴ Xu et al. developed a biocompatible and stable polystyrene–polydopamine–glutaraldehyde SPME coating for in vivo analysis on pharmaceuticals in fish muscle and demonstrated sensitivities from 26 up to 42 times larger than traditional PDMS fibers,¹¹⁷ evidencing the potential for application of novel sorbents for the in vivo sampling of environmental chemicals.¹¹⁸

The main pitfalls of passive sampling for non-target applications are selectivity (similarly to SPE or extraction by solvents), extraction of unbound (freely dissolved) molecules, and lengthy sampling times; though these can be tackled via addition of matrix modifiers (e.g. salts, surfactants) and high temperatures can speed up sampling rate,^119–122 albeit with risk of increased degradation. When used for qualitative analysis or measurements of total concentrations, pushing passive sampling boundaries toward depletive mode by increasing the volume of the sampler (e.g. thickness of the SPME fiber) and or enhancing desorption from matrix (e.g. by using surfactants) to increase the amount of chemical extracted, can improve sensitivity.

Optimizing gas (and liquid) chromatography typically requires finding the best compromise between peak capacity/separation and sample throughput, selecting the best performing stationary phase and column characteristics (e.g. length, diameter, particle size), fitting the purposes of the work. An often overlooked approach to speedup the GC run and providing large sensitivity is the low pressure (so-called “vacuum-outlet”) GC. This consists in connecting a restrictor in the inlet to a short (10-15 m) wide (typically i.d. ≥ 0.5 mm) column as vacuum outlet for the MS.^18,125 High vacuum reduces the viscosity of the gas mobile phase allowing for a higher flow rate and mass transfer. Compared to other fast-GC approaches like using short capillary columns, fast thermal gradients or high flow rates, this method is much more robust because of the use of large columns. The numerous benefits of low-pressure GC led some authors to recommend it as the standard approach for GC analysis,¹⁸ with promising applications to HRMS suspect/non-target screening.¹²⁶ Possible drawbacks of this method regards increase of the column bleed, and potential narrowing of the peaks to durations incompatible with MS detector scan speed.¹²⁶

Comprehensive two-dimensional GC (GC × GC) has been identified as a promising strategy for non-target analysis of environmental chemicals, providing greater peak capacity separations and often greater sensitivity.¹²⁷ The theory¹²⁸ and advances of this approach have been discussed for applications in, e.g. metabolomic,¹²⁹ biomedical,¹³⁰ and environmental¹³¹ research. Briefly, two GC columns are connected in series and multiple fractions/whole sample is sent from the first to the second column via a modulator. The increased commercial availability of GC × GC modulators looks set to push GC × GC towards routine use in the coming years.¹³² However, the reproducibility of GC × GC remains to be demonstrated for large-scale profiling applications, particularly studies that would necessitate intervening maintenance (e.g. column replacements).

The high fragmentation and absence/low intensity of a molecular ion when using hard ionization (e.g. EI) poses challenges for processing and annotation workflows. Using lower ionization energies in EI has been proposed to reduce fragmentation.¹²⁷ However, conflicting results have been reported in literature with this approach shown to be highly dependent upon source design.¹³³ For example, significant improvements of molecular ion peak detection at 30 vs. 70 eV with a quadrupole-time of flight-mass spectrometer (Q-ToF-MS) were demonstrated,¹³⁴ while just slight differences between 12 and 70 eV were observed via GC-Orbitrap.¹³⁵ Another approach to enhance molecular ions is the use of Cold EI whereby a supersonic molecular beam interfaces the GC and MS and EI+ conducted on cold molecules.¹³⁶ This was later pushed further toward soft ionization by reducing electron energy (from 70 eV to 18 eV) and lowering helium pressure to limit collisions via increased nozzle-skimmer distance, which effectively limited in source fragmentation and was successful to obtain mainly molecular ions.¹³⁷ Cold EI has subsequently been linked to both LC and GC and shown favorable to ESI and conventional EI for identification,¹³⁸ yet only tested at low resolution.

The use of “soft” (i.e. low energy) ionization such as chemical ionization (CI), field ionization (FI), photoionization (PI), or APCI is relatively widespread. Recently the coverage domain of EI and CI (positive and negative mode) was investigated on a commercial set of metabolites and shown to be complementary.¹³⁹ Notably, of the 330 total compounds detected 81, 39, and 28 were unique for EI+, positive CI, and negative CI, respectively. However, sensitivity and coverage of soft ionization are on average reduced compared to EI+, and application usually additional rather than replacement for comprehensive profiling, decreasing throughput.¹²⁷ Notably, electron capture negative ionization/negative CI is widely applied for highly sensitive targeted analysis of chemicals containing groups with high electronegativity, particularly halogens. Although overlooked in chemical exposomics (Table S4.1), recent applications have shown promise for non-target analysis.^140–142

Novel interfaces could extend LC-MS coverage and sensitivity. The use of multiple polarity and ionization modes has been evidenced to enhance coverage for metabolomics¹⁸⁹ and rapid polarity switching and dual/multimode sources common.¹⁹⁰ Rapid polarity switching can be less robust than separate acquisition (e.g. through disturbed spray) and increases cycle times limiting scans per peak; though dual emitters help tackle each problem.¹⁹¹ The main limitation is that the use of a single mobile phase can compromise coverage.¹⁹² Multimode ionization tends to show a reduction in sensitivity.¹⁹³ Alternatively, rapid switching between ESI and APCI sources was recently demonstrated without ion abundance loss and is a promising option for further development.¹⁹⁴ Novel ionization processes have been recently reviewed.¹⁹⁵ In particular, the electrospray ionization inlet (ESII) shows promise, offering increased ion abundance and lower background signal when compared to ESI, able to provide similar advantages as nano-ESI at microscale flowrates.¹⁹⁶

Finally, research advances coupling LC with EI ionization are rapidly progressing toward commercial viability.^197–199 Combining the advantages of LC separations with EI ionization could significantly enhance coverage and provides more reproducible fragmentation though, at present, sensitivity requires improvement to be competitive with ESI. Optimizing source parameters (e.g. by using design of experiment, DoE) can improve analyte detection.^15,200 For example, high in-source fragmentation was recently demonstrated to increase the proportion of annotated coverage.²⁰¹ Notably, in silico method development enables rapid testing and optimization of acquisition parameters^202,203 while technological advances have seen the emergence of information-dependent acquisition modes to enable automated real-time parameter optimizations during analysis.^204–206