Analytical strategies for chemical exposomics: exploring limits and feasibility

Tackling the challenges of chemical exposomics will require the implementation of diverse analytical strategies and technological advancements. Herein, high-resolution mass spectrometry-based methods applied in current chemical exposome studies have been surveyed and are shown to be limited. Notably, liquid chromatography separations almost exclusively employ reversed-phase C18 columns using water-methanol gradients with formic acid additive, whilst gas chromatography is underexploited in the field at this stage. A systematic evaluation of strategies applied in related disciplines (i.e. metabolomics, proteomics, multi-residue trace analysis) was undertaken to provide practical guidance for the development of chemical exposomics. The approaches were assessed on the basis of their costs (i.e. capital expenditure, overhead and maintenance fees, expertise required, consumables) and potential benefits (i.e. improvements to sensitivity, coverage, reproducibility, throughput, ease of use) to prioritize those with promise for chemical exposomics application. Alongside a need for technological investments (e.g. advanced hardware updates), numerous low cost strategies showed high potential benefits (e.g. different column phases, enhanced sample fractionation) and are feasible for rapid adoption.


Introduction
Coupling the comprehensive analysis of chemical exposures (i.e. chemical exposomics) with other omics (i.e. metabolomics, proteomics) is a promising strategy for linking chemical exposures to specific biological responses 1-3 , chemical prioritization 2,4 and to lead exposure-based risk assessment 3,[5][6][7] . The profiling of small endogenous compounds and exogenous chemicals converge in terms of platforms used (e.g. high-resolution mass spectrometry (HRMS) [8][9][10][11][12] ), m/z range covered 3,13 and the analytical challenges faced, such as the need to increase coverage (i.e. physicochemical property range) and sensitivity. These challenges are amplified when profiling trace analytes and if working with low sample volumes/amount of complex matrix [14][15][16] .
Despite approaches to enhance the analytical sensitivity and coverage of endogenous compounds also having the potential to improve the detection of exogenous compounds 17 , many remain underexploited for the profiling of chemical exposure agents and further exchange between the fields of chemical exposomics and metabolomics should be encouraged 11 . Additionally, multi-residue methods to assay hundreds to thousands of target contaminants in complex matrices (e.g. food/plant) have recently been developed [18][19][20][21] . Although these approaches are selective for known chemicals, the breadth of chemicals assayed means sample preparation and chromatography is akin to profiling, and evidence that large-scale quantification of low abundant chemicals is feasible.
We have reviewed the gas chromatography (GC)-HRMS and liquid chromatography (LC)-HRMS approaches applied in chemical exposome studies and evaluated strategies used in related disciplines which may enhance analyte detection, coverage, sample throughput and reproducibility. The strategies that warrant further investigation have been prioritized through an assessment of their relative costs and potential benefits of implementation in chemical exposomics. The review and assessment were restricted to GC-and LC-HRMS workflows, because their use is most widespread, but we guide readers to reviews of alternative platforms that provide appealing features for chemical exposomics e.g. ion mobility-mass spectrometry 22 and supercritical fluid chromatography 23 . In this paper we focus on the laboratory-based components and do not venture into the range of bioinformatic and computational approaches necessary for data analysis.
2) Since the approaches applied in chemical exposomics are limited, we reviewed strategies used in related disciplines to identify those with potential for advancing chemical exposome research.
Particular attention was paid to those applied in metabolomics because of the analytical and conceptual similarities with chemical exposomics 24,25 , and applications from proteomics were considered.
Targeted analysis of environmental contaminants was also considered if a broad range of compounds in complex matrix were assayed.
3) The costs and benefits of the reviewed strategies were assessed and evaluated with the aim to prioritize approaches with potential for application in chemical exposomics studies. Focus was upon strategies that could be adopted in the near future. The evaluation was based on literature findings and author's experiences and a summarized perspective is presented.

Analytical approaches currently used for chemical exposomics are rather uniform
Literature was surveyed during December 2020 to identify analytical approaches applied in HRMS-based chemical exposomics. The keywords/phrases 'non target analysis' (title), 'non target analysis environmental chemicals' (topic), 'suspect screening' (title), and 'exposome' (topic) were used to search Web of Science.
Whereas targeted analysis methods selectively measure chemicals of already confirmed identity, non-target analysis and suspect screening is typically performed with the MS scanning a wide m/z range to enable broadscope chemical profiling 8,26,27 . Non-target analysis aims to detect features without any a priori information whilst suspect screening relies on some a priori considerations, seeking to detect chemicals expected to be present in a sample 8,26,27 . Unlike target analysis, non-target analysis and suspect screening provide a putative identification that requires subsequent efforts for confirmation.
The survey was a formal search of studies presented as chemical exposomics and suspect screening/non-target analyses of environmental chemicals in biological and/or environmental matrices.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60 It is acknowledged that the selected terms are not comprehensive but the survey is intended to provide an overview of the trends of recent studies in the field. The retrieved manuscripts were manually inspected by scientists with the appropriate expertise to identify studies which employed suspect screening/non-target workflows. In total, 144 publications reporting full-scan HRMS-based analysis were selected and compiled (see Supplementary Data).
From this survey, it emerged that experimental conditions commonly used in chemical exposomics are rather uniform: Solvent extraction and solid phase extraction (SPE) are the most frequently used techniques for sample extraction of human matrices 28 and the same trend is evident when extending to other (non-human) biological specimens (Table S1.1, S2.1). This was to be expected because minimally selective sample preparation/extraction methods are often advocated for chemical exposomics [27][28][29] . Notably, no study applied pre-normalization procedures to correct for matrix effects or to reduce the dynamic range of concentration (e.g. specific gravity, creatinine normalization).
Across all publications, LC is the most commonly used separation platform compared to GC: 172 LC-HRMS methods were reported for LC-HRMS (28 applied to biospecimens, 126 to other matrices, 18 to both) (Table   S3.1) while 29 methods were reported for GC-HRMS (9 applied to biospecimens, 18 to other matrices, 1 applied to both) (Table S4.1).
The majority of LC studies used ESI (electrospray ionization), with acquisitions performed in both positive and negative polarity while atmospheric pressure chemical ionization (APCI) had limited application (Table   S3.1). Reversed-phase chromatography was commonly performed by using C18-based columns and water/methanol gradients, with formic acid, at temperatures below 50 °C (Table 1, S3.1). Column lengths and particle diameters varied from 50 to 150 mm and 1.7-3.5 µm respectively, whilst an internal diameter of 2.1 mm was common (Table 1, S3.1).
Among the limited applications of GC separation, 5-type columns were generally used (Table S4.1) coupled with EI (electron impact ionization), operated in positive acquisition mode at 70 eV (Table S4.1).  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60 Downloaded from https://academic.oup.com/exposome/advance-article/doi/10.1093/exposome/osab003/6372691 by guest on 10 October 2021 The lack of versatility/diversity of current HRMS analytical approaches was recently highlighted as an obstacle limiting the comprehensive characterization of the internal chemical exposome 24 . It was evidenced that HRMS-based methodologies have limited capability to detect the wide range of environmental pollutants and consumer chemicals in human blood 18 ; due to their abundance typically being orders of magnitude lower than many endogenous compounds and drugs/pharmaceuticals 24,30-32 . Furthermore, the dynamic range of chemical concentrations in samples far exceeds that of MS detectors, meaning that multiple different extractions and enrichments will be essential for comprehensive detection. The diversification of methodologies must also be balanced with achieving highly reproducible workflows that provide a solid basis for harmonizing procedures across multiple laboratories to generate comparable data at large scale.
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. 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 glucuronidated 68 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 conditions 70 have been successfully applied for enrichment in metabolite profiling studies 70,71 , e.g. targeting amines 72 , aldehydes and ketones 73 .

Sample extraction and purification
Protein receptors can be used as probes for selection of biologically-active chemicals 74 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 and enzymatic treatments
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 compounds 75 and recommended for suspect screening to confirm the identify 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 glucuronide 77 and sulfate metabolites 78 respectively, greatly increasing analyte detection. Adoption in chemical exposomics studies would advance towards 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 substances 79 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 occur 80 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 to 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 analytes 87 and improve confidence for annotation (e.g. comparing labelled and non labelled sample fractions/features), although libraries will need to be extended to include more derivatized compounds (e.g. 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 agents 86 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 towards 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 up to 1184-fold 92 .

Microextraction techniques & passive sampling
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  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60 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 has 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 (e.g. 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 to matrix components but usually with stronger affinity for polymeric phases) 40 , and is a promising technique for in situ 104 and in vivo 112 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  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60 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 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 towards 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. Compared to LC-HRMS, GC-HRMS remains an underexploited tool in metabolomics and chemical exposomics, despite providing complementary compound coverage. GC separations can provide more reproducible retention than LC, with use of retention indices commonplace. Plus, the predominant mode of EI is largely standardized which enables highly reproducible fragmentation. The drawbacks are the limited range of compounds amenable to GC alongside the fact that EI is a "hard" (i.e. highly energetic) ionization mode and hinders annotation due to extensive in-source fragmentation. Furthermore, there is a substantial lack of open-source GC-HRMS spectral libraries 123 and spectra collected in low resolution may be not fully comparable to data acquired in HR in the abundances of certain ions 124 .

Optimizing GC
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 GC run 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 x 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 theory 128 and advances of this approach have been discussed for applications in e.g. metabolomic 129 , biomedical 130 and environmental 131 research. Briefly, two GC column are connected in series and multiple fractions/whole sample are sent from the first to the second column via a modulator. Increased commercial availability of GC x GC modulators looks set to push GC x GC towards routine use in the coming years 132 .  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60 However, the reproducibility of GC x GC remains to be demonstrated for large-scale profiling applications, particularly studies which would necessitate intervening maintenance (e.g. column replacements).

GC-MS ionization sources
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  were demonstrated 134 , whilst 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 is 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 chemical ionization are widely applied for highly sensitive targeted analysis of chemicals containing groups with high electronegativity, particularly halogens.
Although overlooked in chemical exposomics (S4.1), recent applications have shown promise for non-target analysis 140,141,142 . Liquid chromatography coupled to HRMS enables separation and detection of a wide array of compounds in a diverse spectrum of physicochemical properties, providing maybe the best compromise between broad analytical coverage and simple sample treatment 124,143 . However, results obtained in terms of analytical detection, sensitivity, and separation, largely depend on specific conditions used since different columns phases, modifiers and chromatographic run length and temperatures can remarkably affect analytical outcomes. Advances in LC have been fundamental for the development of many broad range suspect/nontarget methods. Notably, the introduction of ultra-high pressure liquid chromatography (UHPLC) with sub 2 micron columns enabled higher separation efficiencies 144 , whilst core-shell/fused-core particles extended similar advantages to conventional high pressure liquid chromatography (HPLC) 145 . To capture the large diversity of chemicals, novel developments should be addressed towards the combination/testing of different stationary and mobiles phase chemistry, chromatographic conditions and LC-MS interfaces.

Stationary Phase
Reversed-phase (RP) LC using C18 columns is typically favored because of the reliability and robustness, providing broad coverage, highly reproducibility retention times and pH stability. Periat 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60 Further diversification of reversed-phase column choice also offers a rapid way to enhance coverage. For example, a combination of C8 and T3 (C18 modified for improved retention of polar groups) columns quadrupled the number of non-polar to semi-polar standards detectable compared to use of a single conventional C18 column 149 , whilst the application of C30 column showed better performance for separation for 49 lipid standards 150 and alternate selectivity for polar and non-polar analytes 151 , yet are rarely used in comparison to traditional C18 columns. In addition, the use of porous graphitized carbon has been suggested as a reversed-phase alternative to HILIC separation, though considerations needed to maintain retention stability 152 and optimal choice between each dependent on analytes investigated and sample matrix 153 .
The benefits and drawbacks of coupling of column with different phase chemistry and separation mechanism in series, in parallel or offline have recently been discussed 154 . Notably, many metabolomic studies have applied complementary RPLC and HILIC to simultaneously screen hydrophilic and hydrophobic chemicals 147,[155][156][157] , and the approach could be similarly applied in chemical exposomics.
However, compared to microscale and capillary separations major drawbacks of nanoscale LC are lowered robustness (e.g. column connections, tubing) increasing maintenance, slow conditioning and equilibration times 171 , and the need for more careful sample preparation and purification required to avoid the clogging of the system 54166 . Each of these drawbacks reduces throughput and will need to be overcome for application to population-based chemical exposomics. To this end, applications of nano-UHPLC separations (i.e. using small diameter, small particle size columns) is increasing 171 . Recently commercialized micro pillar array columns (µPAC) with enhanced robustness and reproducibility are showing promise in proteomics 172 and the coupling of SPME with nano-LC represents a promising approach to introduce a highly purified sample 115,173 without need of extensive clean-up steps.

LC temperatures
The use of high-temperature (HT) LC to increase separation efficiency and enable faster separations is a potential approach to increase throughput 180 188 . These studies showed that reproducible HT-UPLC-MS is achievable for complex matrices (e.g. blood plasma), though the authors note HT LC still requires further development.

LC-MS ionization sources
Novel interfaces could extend LC-MS coverage and sensitivity. The use of multiple polarity and ionization modes has been evidenced to enhance coverage for metabolomics 189 and rapid polarity switching and dual/multi-mode 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 use of a single mobile phase can compromise coverage 192 .
Multi-mode 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  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 towards commercial viability [197][198][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 whilst technological advances have seen the emergence of information-dependent acquisition modes to enable automated real-time parameter optimizations during analysis (e.g. [204][205][206] ).

An assessment of costs, benefits & potential for chemical exposomics
Strategies reviewed in the previous section were assessed on the basis of their benefits and costs and a division proposed sorting out strategies with: i) low costs & high benefits; ii) low costs & low benefits; iii) high costs & high benefits; iv) high costs & low benefits (Table 3). Benefits considered were in terms of: sensitivity (enhanced recovery/sample enrichment or instrumental response); physicochemical range coverage; reproducibility; sample throughput; practicality (strategies easy to apply, potential for automation/ reducing manual operations, etc.). Costs accounted for were: capital expenditure (dividing minor and major investment for capital equipment); installation (ready to use items vs. need for installations/specialist support); operational costs associated to maintenance (long-term financial costs to keep functionality); consumables (financial cost plus environmental burden) and commercial availability (niche/in-house applications were considered demanding high efforts). Access to conventional LC and GC-HRMS instrumentation was considered a prerequisite.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60 Ideally, strategies with high benefit and limited costs/efforts should be readily tested and include alternative sample enrichment methods, more refined purification methods, chromatographic optimizations (Table 3).
Promising results in enhancing coverage and detection of low abundant compounds have been obtained with sequential extractions and fractionations, microextraction techniques and chemoselective probes which encourages their application for non-target analysis of environmental contaminants.
Some of these approaches have been widely applied in target studies while still underexploited for nontargeted analysis (e.g. SPME). However, commercially available materials (e.g. HLB SPME, novel imprinted polymers/probes) should be developed to simplify analysis and reduce manual operations. Fractionation methods, with different solvents or sorbent extraction, are included in recently developed workflows, e.g. 59,67 and offer the benefit of differential enrichment of fractions and/or analysis via multiple platforms. Whenever possible, preparation methods should merge the steps needed for sample purification. For example, interferences that ionize in multiple modes and/or have detrimental effects on columns and signal suppression (e.g. lipids) should be removed prior to fractionation. Similarly, enzymatic/chemical treatments (e.g. protease/lipase treatment/derivatization) could be implemented prior to extraction, and may enable more selective clean-up.
Optimized chromatographic settings from other LC-HRMS-based omics (e.g. ammonium fluoride vs. formic acid; PBrBz vs. C18-based separation) should be more widely tested as they show potential to be largely beneficial with virtually no costs. Target studies support the potential of these strategies (e.g. utilization of ammonium fluoride as modifier) 174,175 .
Low pressure GC has shown improved sensitivity and potential for routine implementation with minor hardware updates. This is a largely overlooked approach in chemical exposomics that should be prioritized for implementation, especially considering the possibility of loading large sample volumes that could be a valuable asset for low abundant chemicals. Remarkable improvements in instrumental sensitivity can be obtained just optimizing MS acquisition parameters with proper experimental design. Although this is commonly performed in house, harmonized guidelines on how to do this when a broad range of chemicals is investigated are lacking. The combination of GC and LC separation via workflows that separate polar and non-polar compounds prior to analysis would capitalize on the advantages of each platform. Compared to minimal sample preparation, this would allow to introduce cleaner extracts in the system, since potential matrix interferences would partition as well, and thus both coverage and sensitivity can be enhanced.
If limited resources are available, strategies with low costs and some benefits could be implemented (Table   3). These include mainly minor laboratory tips for sample handling/storage conditions and strategies to improve recovery (e.g. hot solvent & ultrasound extractions) and/or to obtain larger coverage (e.g. derivatization). Limited studies for example are available on chemical stability over storage and potential of deconjugation 76 and derivatization 92 for suspect and untargeted workflows.
Reducing pitfalls of very promising strategies currently with elevated costs (Table 3) should be the goal of next research developments. These include advanced hardware developments with innovative interfaces that allow for better ionization and instrumental sensitivity, though requiring development to become commercially available. For example, high temperature liquid chromatography appears a promising strategy to enhance mass transfer and allow for high throughput analysis, yet is currently is limited by high costs and maintenance. LC miniaturization needs to be coupled with advanced sample cleaning that can limit throughput. These strategies are largely unexplored in chemical exposomics.

Outlook and conclusions
A survey of the analytical condition used in chemical exposomics underlined the need to diversify current methods and approaches to advance towards more comprehensive measures of chemical exposures. To enhance coverage and detection of low abundant environmental chemicals, a proficient use of knowledge and experience inherited from related fields was previously advocated. Therefore, the analytical approaches used in HRMS-based metabolomics and proteomics were reviewed, along with newly emerging wide-scope multiresidue target analysis methods, to identify potential strategies suitable for chemical exposomics. A pragmatic assessment of the costs and potential benefits of implementing the strategies in chemical exposomics was  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60 performed to prioritize those with most promise (Table 3). We hope this provides a practical aid for research groups developing chemicals exposomics and guides investments to address next research.
Beside analytical challenges, data processing constrains hinder the development of a comprehensive characterization of the chemical exposome and represent a bottleneck for untargeted workflows 24,27 .
Computational tools must cope with low intensities features at noise level, scarcity of the signals across To conclude, we wish that this manuscript encourages further exchanges and discussions across experts from different fields. In our vision, sharing knowledge, tools, and analytical hints across fields is fundamental to push forward the limits of the current approaches and integrate findings. Integrating experiences from related fields is a critical step to avoid reinventing the wheel in the emerging field of chemical exposomics. A plethora of approaches and novel strategies must be developed, tested and integrated to push towards more comprehensive measures of chemical exposures and there is no time to waste.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60 34.