Integrated Framework for Identifying Toxic Transformation Products in Complex Environmental Mixtures

Abstract

Complex environmental mixtures consist of hundreds to thousands of unknown and unregulated organic compounds that may have toxicological relevance, including transformation products (TPs) of anthropogenic organic pollutants. Non-targeted analysis and suspect screening analysis offer analytical approaches for potentially identifying these toxic transformation products. However, additional tools and strategies are needed in order to reduce the number of chemicals of interest and focus analytical efforts on chemicals that may pose risks to humans and the environment. This brief review highlights recent developments in this field and suggests an integrated framework that incorporates complementary instrumental techniques, computational chemistry, and toxicity analysis, for prioritizing and identifying toxic TPs in the environment.

Graphical Abstract

Introduction

Although analytical methods offer sensitive and selective determination of targeted pollutants, substantial data gaps exist for compounds in the environment that remain uncharacterized, but may be as toxicologically and ecologically relevant as targeted chemicals.^1–3 In addition, global chemical reform (including TSCA Reform and REACH) is resulting in a broader look at the exposure and toxicity of existing chemicals (and their transformation products) in the environment.^4–9 For these reasons, the use of non-targeted analysis (NTA) and suspect screening analysis (SSA) are rapidly expanding in the field of environmental chemistry. For example, previously unknown biogenic and anthropogenic halogenated organic compounds, including compounds related to the pesticide chlordane, were recently detected in dolphin blubber in concentrations comparable to target compounds.⁴ At least 100 of the identified compounds were not routinely monitored, suggesting that targeted analysis would only encompass a minor proportion of the detected analytes.⁴ Similarly, natural and synthetic organo-bromine compounds were reported to be increasing in concentration, over time, in the sediments of Lake Michigan.⁹ This observation was particularly concerning because some well-known natural and synthetic organo-bromine compounds, such as hydroxylated polybrominated diphenyl ethers, have been associated with adverse health effects.^10–12 These “emerging pollutants” co-exist with routinely monitored chemicals in complex environmental mixtures and are a result of decades of use in pesticides, pharmaceuticals, personal care products and as flame retardants.^{3,4,6,13–18}

Transformation products (TPs) encompass a large proportion of unregulated chemicals present in the environment and may be equally, if not more, persistent and toxic than their precursor chemicals.^3,16,19–23 While the characterization of TPs by NTA has been focused primarily on wastewater treatment plants,^24–27 prior studies have also shown the formation of TPs in chemically treated water^28–36 and in natural water.^37–42 Other toxic TPs include metabolites of pharmaceuticals and pesticides in wastewater and groundwater,^{26,27,43–51} such as persistent methylated derivatives from the antimicrobial triclosan.²³ Concerns over the formation of toxic TPs during remedial applications,^20,21,52,53 and from environmental photoxidation reactions,^22,54–56 have also been raised.

The common analytical techniques used for NTA and SSA of organic chemicals are based on gas and liquid chromatography mass spectrometry methods, for non-polar to semi-polar compounds and semi-polar to polar compounds, respectively. Because these unknown compounds exist in complex sample matrices, and may exhibit a wide range of unknown physicochemical properties, recent advancements in these analytical techniques have brought to the forefront multidimensional and hybrid applications.⁵⁷ These applications offer high throughput, high resolution and highly sensitive capabilities, including comprehensive two-dimensional gas chromatography (GC×GC) and high resolution mass spectrometry (HRMS).^43,58–64 The increased chromatographic resolution and peak capacity offered by GC×GC coupled to time-of-flight mass spectrometry (GC×GC/ToF-MS) enables high throughput screening of complex samples containing thousands of compounds.^{4,13,58,65–68} GC/Orbitrap-MS also offers high mass accuracy and high resolution^60,61 and LC-HRMS has been extensively used for NTA of polar, unknown contaminants and TPs.^{26,27,43–47} Accurate mass measurements and tandem MS (MS/MS) help narrow down potential chemical candidates by providing additional structural information.^41,43,63

Comprehensive NTA and SSA methods can be laborious, time-consuming, and computationally challenging in complex environmental matrices and mixtures, especially when there is limited prior knowledge of the chemical classes likely to be present. Moreover, because environmental samples contain hundreds to thousands of biogenic and anthropogenic organic compounds, non-targeted analytical methods that can detect a wide range of unknown compounds, and strategies for focusing the search on relevant chemicals, are needed. This brief review highlights recent developments in this field and suggests an integrated framework that incorporates complementary instrumental techniques, computational chemistry, and toxicity analysis, for prioritizing and identifying toxic TPs in the environment, as shown in Figure 1.

Figure 1.

An integrated framework for identifying toxic transformation products in complex environmental mixtures. SSA indicates use in suspect screening analysis.

Using Effects-Directed Analysis to Identify Toxicologically Relevant Transformation Products

Effects-Directed Analysis (EDA), or the integration of toxicity testing and chemical analyses for the same complex environmental mixture, is important for understanding the link between the presence of mixtures of environmental pollutants and their associated effects. Several studies suggest that most targeted priority contaminants contribute only partially to the observed toxicity.^20,21,69,70 For example, in some bioremedial applications, a decrease in routinely monitored pollutant concentrations did not coincide with decreased toxicity.^15,16,71 Realistically, both known and unknown compounds, including toxic TPs, may contribute to the overall toxicity of environmental samples.

Sample fractionation in EDA, including size exclusion, adsorption, and partition chromatography techniques, results in simplification of the complex mixture.^21,72–75 The simplification from these fractionation methods makes it easier to link compound presences with their cause and effect. An EDA approach, combining GC/MS and LC-HRMS with the AR CALUX^® assay, was successfully used to identify androgen-disrupting chemicals in soil.^76,77 A study by Marvin et al. found that high molecular weight- PAHs (MW302-PAHs) contributed significantly to the toxicity of coal tar contaminated sediment using the Salmonella typhimurium bacterial strain (YG1025) and LC/MS.⁶⁹ EDA has also been previously used to monitor the formation of toxic TPs,^21,78 such as petroleum degradation products in soil.⁷⁸ Comprehensive reviews of EDA in environmental and biological applications, including challenges and limitations have been previously reported.^79–81

Given that TPs, as well as other unknown compounds, have different modes of action and affect organisms differently, there are multiple toxicity assays that are available and have been used in EDA approaches. Nielen et al. used a yeast-based receptor gene assay to identify estrogen disruptors in urine,⁸² while Chibwe et al. used chicken DT40 lymphocyte genotoxicity and zebrafish (Danio rerio) developmental assays to assess the potential formation of toxic TPs during the bioremediation of soil.²¹ In addition, Dorn et al. used earthworm, Microtox^®, and plant germination assays to evaluate crude oil acute toxicity in soils.⁸³ Most recently, Neale et al. obtained result from, and proposed the use of, mixture-toxicity modeling to link in vitro results and organic micropollutants concentrations in surface water.⁸⁴

The information garnered from toxicity assays is invaluable, as long as a case can be made for translational relevance to human and/or ecological health. It is impossible to discern, or anticipate, the modes of action of toxins with unknown identity. This makes it important to cast a wide net in the analysis of unknown compounds. Moreover, there should be the realization that, depending on the assay(s) used, the response and performance of the bioassay will drive the identification of the unknown compounds towards a group of chemicals that exhibit particular toxic effects. In addition, even after extensive fractionation, there is still a complex chemical mixture present in the extract, which may result in mixture toxicity effects.

In our integrated framework (Figure 1), when treatment is applied in lab, pilot-plant, or field-scale, toxicity testing before and after sample treatments are necessary to differentiate measured effect between parent compounds and TPs. EDA is focused on fractions where toxicity have increased which, presumably, can be attributed to the TPs following treatment. In situation when there is no treatment applied, EDA can still play an important role in identifying the most toxic fractions of the sample. Targeted analysis has shown that parent compounds are likely to be in different fractions than the transformation products, for instance due to the introduction of more polar groups within the compounds,^21,85 indicating that the information garnered from which fractions are more toxic could still help differentiate whether the toxicity increase was due to parent compounds or TPs.

Selected Strategies for Increasing Confidence in Toxic Transformation Product Identification

The data output produced from the analysis of complex samples includes comprehensive coverage of all ions amenable to the analytical technique. This results in hundreds to thousands of peaks originating from the sample, sample matrix and background noise. Some software tools incorporate statistical modeling features, meant to unbiasedly characterize compositional differences between groups of samples. These include peak finding, peak alignment, and mass spectral deconvolution software, and these have been covered elaborately in previous reviews.^86–88 In NTA, this is beneficial in limiting the presence of false positives, reducing the time and effort spent on irrelevant peaks. Due to the scope of software tools available, the objective of this section is not to cover all available software, but to highlight certain roles in simplifying and facilitating NTA. Reviews on various NTA workflows, with different softwares and tools are available elsewhere.^38,39,74

The ChromaTOF^® Reference and Statistical Compare features (LECO Corp., St. Joseph, MI, USA) have been previously applied to facilitate the identification of compounds of interest in environmental samples in GC×GC applications.^6,89 Prebihalo et al. used the ChromaTOF^® Reference feature to investigate emerging contaminants in wastewater and soil.⁸⁹ A sample was used to create a reference method, containing a comprehensive list of peaks with accompanying retention times and mass spectra. The reference sample serves as the basis of comparison for other samples, and the software computes pairwise comparisons to determine peaks present in both the sample and reference (“match”), only in the reference sample (“found”), and only in sample (“not found/unknown”).⁸⁹ LECO’s Statistical Compare feature is useful to determine the most distinct compounds between samples, and has been applied successively to identify putative biomarkers.⁹⁰ Peaks are aligned in samples according to both retention times (1D, 2D) and mass spectral similarity, and are subsequently compared with samples in other groups (or classes). Prior to data processing, parameters to define peak tolerance (i.e. MS similarity threshold) and acceptable retention time shifts are set to account for variations and to limit false negatives. However, it should also be noted that peak alignment can be unsuccessful for poorly resolved peaks at low intensities or for highly saturated compounds.⁹⁰

MarkerView^™ (AB SCIEX Ltd., Framingham, MA, USA) similarly designed to identify structurally related and unrelated components between grouped samples for LC data. MarkerView^™ has been used to assess data in metabolomics and in biological fluids,^91–93 such as urine and plasma. Environmental applications have included the recent identification of novel fluorochemicals in firefighter foams,⁴⁷ and the discovery of quaternary triphenylphosphonium industrial contaminants in aquatic systems.⁹⁴ This software utilizes algorithms to align ions (according to retention times, mass spectral similarity, etc.) across multiple samples for comparison. Statistical tools, including principal component analysis (PCA) and principal component variable grouping (PCVG), can then be applied to the processed data to group samples and form chemical profiles. This approach is complemented by MasterView^™ (AB SCIEX Ltd., Framingham, MA, USA), to generate molecular formula and tentative identification of the candidate compounds. These tools can be effective at minimizing the presence of false peaks or at highlighting components of interest in grouped samples, while also finding candidate compounds in NTA.

Once candidate structures or molecular formula has been proposed, identity of TPs need to be identified. Mass spectral libraries and databases, such as the NIST electron ionization (EI) mass spectral library,⁹⁵ may not contain the mas spectra of the TPs of interest and soft ionization mass spectral libraries are limited. MOLGEN,^96,97 a structure generator tool, can help identify potential TPs of interest in cases where conventional mass spectral libraries do not contain the structure of TPs, regardless of whether or not there is treatment applied to the sample (Figure 1). In silico fragmentation tools, such as MetFrag and Mass Frontier^™ (Thermo Fisher Scientific Inc., Waltham, MA, USA), use mass-to-charge ratio (m/z) intensity and bond dissociation energies to predict LC-HRMS mass spectra^35,98,99 that assist in the prediction of potential structures from experimental MS and/or MS/MS fragmentation patterns, These potential structures are taken from public databases such as PubChem,¹⁰⁰ ChemSpider,¹⁰¹ MassBank,¹⁰² and METLIN.^51,103 Recently, the use of Database Assisted Identification of Organic Substances (DAIOS) and Assessment of Suspected and Unknown Anthropogenic Trace Contaminants in Aquatic Systems (STOFF-IDENT) have also been integrated in NTA.^36,49 Detailed discussion and uses of these libraries,^39,74,104 and candidate structure elucidations,^59,86,105 are available elsewhere.

Retention indices (RI) have also been used to offer an additional level of identification, and rank proposed chemical structures for the most likely candidates.^106–108 Tentative identification of the unknown compound is based on comparing how well the experimental RI matches the predicted RI from computational models. Quantitative structure retention relationship (QSRR) models, based on the number of rotatable bonds, topological polar surface, and octanol-water partition coefficient, have been used to eliminate potential structures.^109–111 Stanstrup et al. developed a retention time (RT) prediction model that predicts the RT of a compound in various LC systems, provided that the stationary phases used are based on the same separation mechanism.¹¹¹ A QSRR model for RT prediction was developed for positive and negative electrospray ionization in LC-QToF-MS, based on the selection of a set of molecular descriptors, including the presence of carbon-carbon bond.¹⁰⁹ Falchi et al. used a kernel-based and partial least squares QSRR for ultra-performance liquid chromatography (UPLC) retention time prediction for the purpose of identifying metabolites.¹¹²

RT prediction models are not restricted to LC systems alone. The Conductor-like Screening Model for Realistic Solvents (COSMO-RS) approach was used to predict the RTs of both methylated and non-methylated derivatives of monohydroxylated brominated diphenyl ethers by correlating the theoretical boiling points of the metabolites and the RTs in GC-MS.^11,113 However, because current RT prediction models tend to be chromatographic system dependent, future models that are more universal, or parameterized for a variety of different chromatographic systems, would be extremely useful.

Other methods for the identification of unknown compounds include the use of unique MS fragmentation patterns, such as the isotopic abundance of the atoms in the structures. Halogenated unknown compounds can be tentatively identified as a result of the distinctive isotopic patterns produced by Cl and/or Br atoms.^17,64,114 For example, Myer et al. used mass defect analysis to measure the biota-sediment accumulation factors in freshwater organisms exposed to unknown halogenated pollutants.¹¹⁴ Barzen-Hanson et al. used a similar approach to identify novel perfluoroalkyl sulfonates in aqueous film-forming foams and groundwater.¹⁷ Peng et al. also developed a data independent precursor isolation and characteristic fragment method and identified over 1,500 unique natural and synthetic organo-bromine compounds in sediment.^9,115 Recently, Wang et al. used NTA to investigate tap water contamination using this approach.¹¹⁶ Chemical derivatization, bromination, and subsequent dansylation of the unknown compounds in collected tap water samples were used, together with liquid chromatography coupled to quadrupole time-of-flight mass spectrometry (LC-QToF-MS), to identify hydroxylated naphthenic acids and saturated fatty acids as key pollutants. The use of mass defect screening has also been implemented for non-halogenated compounds, including a TP of venlafaxine, an antidepressant, and quarternary ammonium compounds.^40,117

Due to the general lack of commercial standards and the varying degrees of confidence in compound identification, several studies report compound identification in NTA based on hierarchal degrees of confidence (Figure 2).^4,118 Compounds confirmed with authentic standards are reported with the highest confidence (Level 1), while compounds identified only by mass spectral library or database matching are reported with lower confidence (Level 2). Compounds for which structures cannot be elucidated are not eliminated, but reported with the lowest degree of confidence (Level 5).^4,6,118,119 This ensures that the information is preserved for future studies, with the degree of confidence in the identification indicated.

Figure 2.

Illustration of communicating confidence on unknown compound identification in non-targeted analysis.¹¹⁸ Reprinted (adapted) with permission from Schymanski, E. L.; Jeon, J.; Gulde, R.; Fenner, K.; Ruff, M.; Singer, H. P.; Hollender, J. Identifying Small Molecules via High Resolution Mass Spectrometry: Communicating Confidence. Environ. Sci. Technol. 2014, 48 (4), 2097–2098. Copyright 2014 American Chemical Society

Guiding Suspect Screening of Toxic Transformation Products Using Computational Tools

Provided that there is some a priori knowledge of the classes of compounds present in a complex environmental mixture, cheminformatics can be used to predict the structure of toxic TPs.^49,120,121 A brief list of predictive degradation models is given in Table 1 and discussed here, while more comprehensive reviews are available elsewhere.^122,123

Table 1.

Examples of models that can be used to predict possible TP structures.

Type of Degradations	Tools	Developer	Description
Microbial Biodegradation	EAWAG Path Prediction System (EAWAG-PPS)	Eawag	Predicts TPs from microbial metabolism based on compound structure and transformation rules developed through data collected from known reactions, pathways, and enzymes reactions
	enviPath	University of Mainz, EAWAG	A platform created as improvement over EAWAG-PPS by addressing combinatorial explosion concerns that arise from the use of EAWAG-PPS
	Biochemical Network Integrated Computational Explorer (BNICE)	Northwestern University, EFPL	Predictive biodegradation algorithm based on generating novel xenobiotic degradation pathways
	PathPred	Kyoto University	In silico tool that determines potential reaction pathways from microbial biodegradation based on the Kyoto Encyclopedia of Genes and Genomes (KEGG) database
Human Metabolism	Meteor Nexus	Lhasa Ltd.	Commercially available microbial pathway prediction built on rules and knowledge-based reactions, which is capable to predict mammalian metabolism (CYP 450)
	Meta-PC	MultiCASE Inc.	Commercially available microbial pathway prediction built on rules and knowledge-based reactions, which is also capable to predict mammalian metabolism (CYP 450)
	Admet Predictor™	Simulation Plus Inc.	Commercially available mammalian metabolism degradation prediction (CYP 450)
	Metabolic In silico Network Expansion (MINEs)	Argonne National Laboratory	Human metabolism degradation pathway database based on the BNICE approach
Abiotic Degradation	CATALOGIC	Laboratory of Mathematical Chemistry	Commercially available program that predicts the biotic and abiotic environmental fate of chemicals based on OECD standards

The EAWAG Pathway Prediction System (EAWAG-PPS) model is used to predict products of microbial metabolism^124,125 and has been successfully used to predict TPs from pollutants in both wastewater and natural water.^37,121,126 However, the issue of combinatorial explosion, that might arise when several generations of TPs are generated within EAWAG-PPS, is a point of consideration when using this model.^127,128 A proposed solution to combinatorial explosion is potentially solved through the creation of enviPath, which limits the number of predicted TPs using machine-based learning and interfacing to enzymatic databases.¹²⁹ Other microbial biodegradation models include the Biochemical Network Integrated Computational Explorer (BNICE),^130–132 and PathPred.¹³³ Both models utilize the Kyoto Encyclopedia of Genes and Genomes (KEGG) database,¹³⁴ making them relevant in terms of enzymatic biodegradation, and can also be useful to predict novel biodegradation pathways.

Metabolite prediction models, that can predict degradation pathways due to eukaryotic metabolism, are also emerging as potentially useful models for SSA (Table 1). Commercial programs, such as Meteor Nexus (Lhasa Ltd., Leeds, UK), Meta-PC (MultiCASE Inc., Cleveland, OH, USA), and ADMET Predictor^™ software (Simulation Plus Inc., Lancaster, CA, USA), are capable of predicting mammalian metabolism based on the cytochrome P450 (CYP450) enzymes,^135–137 and are relevant to addressing the potential for mammalian toxicity.^138–141 The Metabolic In silico Network Expansion (MINEs) database is another example of a metabolism prediction model that relies on the KEGG database, as well as the yeast and Escherichia coli databases.¹⁴² In terms of abiotic degradation, CATALOGIC (Laboratory of Mathematical Chemistry, Bourgas, Bulgaria) is a model that has been used to predict the abiotic degradation of chemicals in the environment.¹⁴³

With advances in technology and computer capabilities, in silico approaches can help narrow the search for potentially toxic TPs in SSA. For example, MOLGEN-MS suggests candidate structures, based on the mass spectra fragmentation patterns.^{96,97,144,145} Schymanski et al. used this platform, and the steric energy of 1,000 randomly selected compounds, to screen candidate structures through the elimination of those that were energetically unfavorable.^74,146–148 The steric energy of the compound test set was calculated using the molecular mechanic (MM2) force field approach and suspect compounds, that fell outside the pre-determined steric energies (i.e. outside the 90% quantile), were eliminated.^146,147 When combined with other screening criteria, the steric energy algorithm serves as another step in a tiered product identification approach.

Computational chemistry software can also be used to predict the thermodynamic stability of potential TPs, and aid in compound identification. For example, Jariyasopit et al. used density functional theory (DFT), in Gaussian,¹⁴⁹ to successfully predict the formation of the most thermodynamically favorable nitro polycyclic aromatic hydrocarbon (NPAHs) formed from PAH reaction with NO₃,/N₂O₅, and NO₂ radicals.⁵⁴ A similar approach was used by Dang et al. and Borduas et al. to study the atmospheric degradation of PAHs and nitrogen containing compounds, such as nicotine, due to reaction with OH or NO₃ radicals.^150–155 Other examples of the use of Gaussian include the prediction of the UV absorption of the endocrine disrupting compound, 4-tert-butylphenol,²⁸ the prediction of the electron density of¹⁵⁶ and the chlorination and oxidation sites of PAHs.¹⁵⁷

The role and potential contribution of computational methods in SSA of TPs is currently just being realized. The thermodynamic stability consideration from computational chemistry calculations is not limited to atmospheric reactions^54,158–160 and similar approaches may also apply to the study of microbial TPs.¹⁶¹ Using the output of cheminformatics software, such as EAWAG-PPS or enviPath, the overall thermodynamic stability of potential TPs can be calculated using Gaussian or other software, such as NWChem.¹⁶² The delta Gibbs free energy of the reaction (ΔG_reaction) forming TPs can be calculated, assuming all the reactants and products are known. The resulting thermodynamic stability can then provide a priority list of the most favorable TPs likely to form, and can guide compound identification. This approach also provides an alternative in solving the issue of combinatorial explosion in that certain TPs are prioritized, or screened out, over other TPs. In addition, for biodegradation reactions, the use of molecular docking software^163–166 can improve the specificity of the TP structure predictions.

Future Directions in Identifying Toxic Transformation Products

In general, to date, most NTA and SSA studies have been focused on aqueous matrices, while studies in soil,¹⁶⁷ sediment,^{64,117,168–170} and air^171,172 are limited. This presents opportunities for future research. Furthermore, where possible, strategies that combine different analytical capabilities, such as both GC- and LC-HRMS, would be beneficial in detecting TPs with a wider range of physicochemical properties. Other researchers have suggested the incorporation of nuclear magnetic resonance (NMR) spectroscopy in the overall workflow of unknown environmental chemicals identification where TP concentrations are sufficiently high.^31,50,173 Calculation of NMR chemical shifts is also commonly predicted using computational chemistry.^174,175 Another advantage of adding NMR in the overall NTA and SSA framework is the presence of robust NMR database, such as the Spectral Database for Organic Compounds (SDBS),¹⁷⁶ and spectra prediction, such as the platform developed by scientists from Universidad del Valle and Ecole Polytechnique Fédérale de Lausanne.¹⁷⁷ For TPs that are not amenable to GC and poorly separated by LC, supercritical fluid chromatography (SFC) can be used for SSA of TPs.^12,178

Strategies to rank the most relevant compounds, either by frequency of occurrence or toxicity, are needed.^179–182 Rager et al. recently proposed the development of a chemical prioritization index, meant to select the tentatively identified compounds that warrant confirmation with reference standards, based on their project environmental and toxicological impact.¹⁸¹ Detected compounds confirmed with standards included triclocarban, which had a low detection frequency, but was prioritized based on bioactivity score, and piperine, which was prioritized based on its high detection frequency in multiple homes. Similarly, Singer et al. predicted exposure concentrations of active pharmaceutical ingredients (APIs), based on their consumption data, and APIs with high exposure potential were re-screened, resulting in the identification of new APIs that have not been previously detected.¹⁸³

An important aspect of combined toxicity and chemical methods is the verification of the toxicity of the TPs. This is certainly impeded by the general lack of commercially available authentic standards. While assumptions can be made with regard to the toxicity of the tentatively identified compounds, based on structural similarity to compounds with known toxicity, various tools are also available to predict the toxicity of suspect chemical structures. These include the predictive in silico program, VEGA-QSAR (quantitative structure-activity relationship)¹⁸⁴ and VirtualToxLab^™ (Biographics Laboratory 3R, Switzerland),^185,186 or bioactivity specific QSAR models, including the Online Chemical Database with Modeling Environment (OCHEM) and the Collaborative Estrogen Receptor Activity Prediction Project (CERAPP) platforms.¹⁸⁷ Raies and Bajic recently reviewed the workflow to generate in silico toxicology tools, which included the various methods to generate these models.¹⁸⁸ This resource may be useful for researchers who are developing in silico toxicology tools in cases where toxic TP standards are unavailable.

NTA and SSA aim to bridge the knowledge gap stemming from the presence of significant unknown toxic TPs, and other contaminants, in the environment. The magnitude of data generated from NTA and SSA for complex environmental mixtures calls for interdisciplinary, multifaceted, and collaborative efforts to prioritize the data interpretation and compound identification. The development and sharing of NTA databases for complex environmental mixtures has begun.^4,129 Databases such as this are extremely beneficial because other researchers can use them as reference and screening tools to monitor the frequency of detection of the same unknown compounds in various environmental samples.¹⁰⁴