Using metabolomic profiling to inform use of surrogate species in ecological risk assessment practices

Abstract

The U.S. EPA frequently uses avian or fish toxicity data to set protective standards for amphibians in ecological risk assessments. However, this approach does not always adequately represent aquatic-dwelling and terrestrial-phase amphibian exposure data. For instance, it is accepted that early life stage tests for fish are typically sensitive enough to protect larval amphibians, however, metamorphosis from tadpole to a terrestrial-phase adult relies on endocrine cues that are less prevalent in fish but essential for amphibian life stage transitions. These differences suggest that more robust approaches are needed to adequately elucidate the impacts of pesticide exposure in amphibians across critical life stages. Therefore, in the current study, methodology is presented that can be applied to link the perturbations in the metabolomic response of larval zebrafish (Danio rerio), a surrogate species frequently used in ecotoxicological studies, to those of African clawed frog (Xenopus laevis) tadpoles following exposure to three high-use pesticides, bifenthrin, chlorothalonil, or trifluralin. Generally, D. rerio exhibited greater metabolic perturbations in both number and magnitude across the pesticide exposures as opposed to X. laevis. This suggests that screening ecological risk assessment surrogate toxicity data would sufficiently protect amphibians at the single life stage studied but care needs to be taken to understand the suite of metabolic requirements of each developing species. Ultimately, methodology presented, and data gathered herein will help inform the applicability of metabolomic profiling in establishing the risk pesticide exposure poses to amphibians and potentially other non-target species.

Graphical Abstract

Introduction

Pesticide use in the U.S. and globally has continuously increased over the past few decades and an estimated 3.5 billion kg of pesticides are used around the world annually for pest and crop management (Grube et al., 2011; Pretty and Bharucha, 2015). While the use of pesticides is essential for high crop yields and population sustainable agricultural practices, they can have a detrimental impact on non-target species and surrounding ecosystems (Brühl et al., 2013). Amphibian populations are in global decline and pesticide exposure is accepted as a major contributor to their dwindling populations (Collins and Storfer, 2003; Davidson, 2004; Brühl et al., 2013). However, there is currently no mandatory risk assessment practice for amphibians prior to or during the registration of a new pesticide product (Brühl et al., 2013).

The U.S. Environmental Protection Agency (U.S. EPA) and other regulatory organizations often utilize fish toxicity data from environmentally relevant concentrations of chemicals to assess the ecological risk posed to amphibians. The use of surrogate species in ecological risk assessments (ERAs) is standard practice to gauge the impact of chemical exposures across a range of species that cannot be tested under laboratory conditions due to either limited resources or infeasibility of testing relevant organisms (Carrascal et al., 2012; Banks et al., 2014). However, surrogate species’ physiology and life history can vary drastically from that of the other organism (Andelman and Fagan, 2000; Andelman et al., 2004; Wiens et al., 2008; Murphy et al., 2011; Romeis et al., 2013; Banks et al., 2014). Additionally, surrogate toxicity data frequently inaccurately predicts the risk towards other species or groups (Weir et al., 2010; Banks et al., 2011). While there have been previous attempts to establish effective amphibian ecotoxicology for representative species, these species frequently lack the breadth of laboratory-derived toxicity data allowing accurate estimation of risk that predictive fish models such as zebrafish (Danio rerio), fathead minnows (Pimephales promelas), or rainbow trout (Oncorhynchus mykiss) offer (Glaberman et al., 2019).

In the current study, we present methodology that can be used to assess the impact of high-use pesticides trifluralin (herbicide), chlorothalonil (fungicide), and bifenthrin (insecticide), on larval zebrafish (D. rerio) and African clawed frog tadpole’s (X. laevis) metabolomes during discrete intervals of their respective development. Efforts such as these will continue to inform the potential for ‘omics’ technologies to not only sync the important, species-specific, biochemical parameters occurring during development but also the impact of pesticide exposure on these metrics. Thus, the different biochemical perturbations from individual pesticide exposures between the two species were compared as well as their dose response in each species.

Metabolomics is an emerging analytical methodology that has become extremely useful in toxicological and environmental studies because of its unique ability to identify specific biochemical responses of xenobiotic exposure prior to the onset of overt toxicity. It has been successfully applied in eco-toxicology studies investigating organismal response to a wide range of biotic and abiotic stressors (Samuelsson and Larsson, 2008). Frequently, metabolomics is aimed at identifying novel biomarkers of exposure that are indicative of a pre-toxic response in organisms to help improve diagnostic tools and predictive models, although, metabolomics is also found useful in ecological risk assessment practices (Viant, 2009). Environmental metabolomics, an emerging field of metabolomics, focuses on elucidating the interactions between organisms and their environment and has become useful in understanding how environmental stressors impact organisms at a molecular level and further translating this knowledge from individuals to population level effects (Bundy et al., 2009; Aliferis and Chrysayi-Tokousbalides, 2011). In evaluating the use of surrogacy data in elucidating the consequences of pesticide exposure during development, metabolomics offers the ability to understand the interplay and commonality of developmentally regulated (or significant) metabolites with known physiological changes. An understanding of the biochemical perturbations across both conserved developmental mechanisms as well as those that are hormone-mediated (i.e., in amphibians) will ultimately define species-specific sensitivities in risk assessment practices (Weltje et al., 2013).

GC/MS-based metabolomics has been employed to identify the sublethal effects of acetamiprid, a neonicotinoid insecticide, and halosulfuron-methyl, an herbicide, in zebrafish (Zhang and Zhao, 2017). However, only a few studies have explored the impact of pesticides on the metabolome of pre- or post-metamorphic amphibians. In Van Meter et al. (2018), metabolomic profiling was used to demonstrate that pesticide mixtures altered the metabolome of Lithobates clamitans (green frogs) differently than single pesticide exposures, suggesting that the typical mode of action of a pesticide changes when multiple stressors are present. Melvin et al. (2018) used metabolomics to demonstrate that striped marsh frog larvae (Limnodynastes peronii) exposed to vinclozolin and propiconazole fungicides exhibited altered steroidogenesis and dysregulation in cholesterol metabolism. They also highlighted the variations in biochemical response to each pesticide suggesting alternative mechanisms of action (Melvin et al., 2018). Thus, although metabolomics is a very powerful tool for elucidating specific as well as subtle biochemical changes that occur as a result of xenobiotic exposure, limited attempts have been made to use it in cross-species comparisons to validate the use of surrogacy data in risk assessments. In addition, when metabolomics is combined with other ‘omics’ techniques and apical endpoints it can potentially provide a comprehensive snapshot of the perturbations of biological systems in response to these chemicals providing the necessary information to accurately assess risk to non-target species and surrogate similarity (Awkerman et al., 2020).

In the current study, we utilized GC/MS-based metabolomic profiling to compare the biochemical responses of larval-stage Danio rerio to those of Xenopus laevis tadpoles following exposure to three high-use pesticides frequently detected in surface waters across the U.S. We present both experimental and statistical approaches that can be applied in this and future studies to evaluate pesticide-induced biochemical perturbations more comprehensively across other life stages and in comparing the response of acute and chronic exposures in these and other species. The objectives of the study were to 1) identify the dose-responsive biochemical perturbations in each species impacted by pesticide exposure, 2) compare the metabolomic response between the two species across each individual pesticide at select, single developmental life stages, and 3) use pathway analysis to compare the predicted physiological impact of these exposures for plausible interpretation of risk.

Materials and Methods

Chemicals

The pesticides bifenthrin (CAS 82657–04-3), chlorothalonil (CAS 1897–45-6), and trifluralin (CAS 1582–09-8) were all purchased from Chem Service (West Chester, PA) at >99.4% purity and dissolved in acetone or methanol (≥99.5%; Sigma-Aldrich) at primary stock concentrations detailed in Table 1 and these were used for serial dilutions. All other extraction and analytical reagents were of highest purity (≥99%) and obtained from Fisher Scientific (Waltham, MA).

Table 1.

Summary of pesticide treatments. For D. rerio, age at exposure is given in days post fertilization (dpf). X. laevis age at exposure was at Niewkoop and Faber (NF) stage 48.

	Atrazine		Bifenthrin		Chlorothalonil		Metolachlor		Tebuconazole		Trifluralin
Parameter	D. rerio	X. laevis	D. rerio	X. laevis	D. rerio	X. laevis	D. rerio	X. laevis	D. rerio	X. laevis	D. rerio	X. laevis
Concentration Range (μg/L )	0 – 1250		0 – 0.625		0 – 12.5		0 – 100		0 – 2000	0 – 1000	0 – 240
Experimental Age	15 dpf	NF 48	15 dpf	NF 48	15 dpf	NF 48	15 dpf	NF 48	15 dpf	NF 48	15 dpf	NF 48
Technical Replicates per Concentration	5	5	5	5	5	5	5	5	5	5	5	5
Individuals per Technical Replicate	8	10	40	5	40	5	40	5	40	20	40	5
Solvent	Methanol	Acetone	Acetone	Acetone	Acetone	Acetone	Acetone	Acetone	Acetone	Acetone	Acetone	Acetone
Primary Stock Concentration (mg/L)	250	25000	12.5	12.5	250	250	2000	2000	40000	20000	4800	4800
Exposure Volume (mL)	50	1000	1000	1000	1000	1000	1000	1000	1000	1000	1000	1000
Final Solvent Amount (%) in Exposures	≤ 0.5	0.005	0.005	0.005	0.005	0.005	0.005	0.005	0.005	0.005	0.005	0.005
Exposure Chambers Pre-soaked	No	Yes	Yes	Yes	No	No	No	No	No	No	Yes	Yes
Individuals per Sample	1	1	10	1	10	1	10	1	5	1	5	1
Number of Samples per Replicate	4	2	1	2	1	2	1	2	2	2	2	2

Animal Husbandry

A detailed description of the rearing, breeding, monitoring, tank conditions, and maintenance methods for both zebrafish (Danio rerio) and the African clawed frog (Xenopus laevis) are outlined in Awkerman et al. (2020). In brief, adult zebrafish were lab-reared and 2–4 pairs of adults were placed in brood chambers to breed. Following fertilization, viable embryos were rinsed with filtered freshwater, allowed to mature to 15 dpf, and placed in replicate chambers for six separate treatment concentrations for each bifenthrin, chlorothalonil, and trifluralin exposures. X. laevis were acquired through a commercial supplier and bred at the U.S. EPA’s Gulf Ecology Division Research Laboratory (now Gulf Ecosystem Measurement and Modeling Division) in Gulf Breeze, FL. After breeding, the viable embryos were selected and allowed to mature to tadpoles (stage NF 48). Tadpoles were placed in five replicate chambers per treatment concentration for each pesticide at exposure levels within the range of concentrations listed in Table 1. Acquiring sufficient mass for metabolomic analysis precluded sampling of earlier life stages in fish in the current study. All exposure related parameters are detailed in Table 1. Exposures were static for 48 h and post-exposure, both D. rerio and X. laevis larvae or embryos were collected and stored at −80 °C until extraction for metabolomic profiling.

Exposure concentration verification

Following exposure, confirmation of exposure levels was conducted by the US EPA’s Analytical Chemistry Research Core facility in Research Triangle Park, North Carolina. Methodology and data are presented in SI “Analytical Determinations of Exposure Concentrations” and SI Table 1, respectively. Briefly, exposure water was processed via solid phase extraction and analyzed by GC/MS (bifenthrin and trifluralin) or HPLC (chlorothalonil).

Metabolomics

Whole bodies from both D. rerio and X. laevis were processed for metabolomics as outlined by (Viant, 2007). Briefly, samples were homogenized using a Qiagen Tissuelyser (Hilden, Germany) and metabolites were extracted sequentially using aqueous methanol and chloroform. After the polar (aqueous methanol) and non-polar (chloroform) phases were separated via centrifugation, each layer was transferred into pre-labeled GC vials. Samples were then placed in a Savant SpeedVac Plus SC110A evaporator (Waltham, MA) overnight to ensure that no moisture remained in the sample. Following lyophilization, samples were stored at −20 °C until being allowed to come to room temperature for derivatization.

Polar fractions were derivatized with 50 μL of methoxyamine hydrochloride (MeOX) at 20 mg/mL in pyridine and incubated at 60 °C for 2.5 h. Samples were vortexed every 30 min. After 2.5 h, the samples were allowed to cool before 50 μL of BSTFA (N,O-bistrifluorocetamide) with 10% TCMS (methyltrichlorosilane) was added to each sample. X. laevis bifenthrin exposure samples and chlorothalonil exposure samples for both X. laevis and D. rerio were derivatized with 100 μL BSTFA with 10% TCMS. Following the addition of BSTFA, samples were placed in an oven at 60 °C for 1.5 h and vortexed every 30 min. All derivatized samples were allowed to cool to room temperature before being transferred to GC vials with micro-target inserts and analyzed by GC/MS within 72 h. Approximately 10 μL from each derivatization reaction for each species and each pesticide exposure was composited into a single sample for intermittent QAQC analysis. The non-polar fractions were processed similarly and analyzed by GC/MS within 72 h of derivatization.

GC/MS Metabolomics

GC/MS-based metabolomic profiling was performed based on the methods outlined in Glinski et al. (2019). Samples were analyzed using an Agilent 6890 gas chromatograph (GC) connected to an Agilent 5973 mass selective detector (MSD) controlled by ChemStation software. All injections (2 μL) were made in splitless mode and a Rxi-5Sil MS column (30 m, 0.25 μm thickness, and 0.25 mm ID; Restek, PA, USA) was used for separation of metabolite derivatives. Injector temperature was 250 °C, transfer line temperature was maintained at 280 °C, and the source temperature was 200 °C. Helium was the carrier gas and a constant flow was maintained at 1 mL/min. The beginning oven temperature of 60 °C was held for 2 min before ramping up 8 °C/min to 300 °C with a hold time of 3 min. Blanks were run at the beginning, end, and intermittently throughout the sequence to ensure that there was no carry-over. Mass spectra were acquired over a range of 50 to 650 m/z and chromatograms were exported as netcdf files post-acquisition.

Exported spectra were then imported into MetAlign 041012 for spectral pre-processing, normalization, and alignment based on the developer’s recommended parameters for fast scan, GC/MS analysis (Lommen, 2009). Post spectral processing, data were exported as .csv files and Excel® was used to filter and truncate the aligned data based on details in Niu et al. (2014). Once truncated, each m/z:retention time pair was subjected to a Student’s t-test comparing each m/z abundance at each dose level to that of controls. If statistically different, the average response of control samples was subtracted from that of treated samples across each instrumental scan (i.e., metabolites presenting with positive values were higher in exposed samples) as demonstrated in Zhen et al. (2018). These data were used to calculate and descriptively associate the number of m/z:retention time pairs (analogous to number of metabolites perturbed) as well as the absolute sum of biochemical perturbations for each pesticide (SI Fig. 1A–L; polar and non-polar samples).

MetaboAnalyst® 3.0 was used for construction of principal components analysis (PCA) and partial least squares-discriminant analysis (PLS-DA) models, for outlier analysis and hierarchical clustering, respectively, and volcano plot analysis (used for metabolite identification in pathway analysis) as well as statistical analysis of metabolites (SAM; multigroup analysis) (Xia and Wishart, 2016). All data were filtered by mean intensity value before pareto scaling was applied when using MetaboAnalyst’s statistical packages. MetaboAnalyst does not apply scaling for univariate-based statistics. Following each analysis, a list of spectral features identified as statistically significant were exported as .csv files for peak identification. Using m/z:retention time pairs, representative spectra (n > 6 representing each exposure group) for each species and each pesticide were auto integrated and peaks searched against the NIST 2017 mass spectral database. Database hit scores >60 match similarity were compiled across samples and used for putative peak identification.

Pathway analysis

To best compare our results with pathways identified in recent literature, metabolites identified from combined PLS-DA VIP scores and SAM were uploaded into MetaboAnalyst’s Pathway Enrichment Analysis module and searched against the KEGG (Kyoto Encyclopedia of Genes and Genomes) database. Conclusions from these analyses were limited to statistically significant pathways identified (impacted p-value <0.05) as well as enrichment ratio provided by MetaboAnalyst.

Following peak annotation, metabolite names and corresponding fold-change and p-values (volcano plot analysis, MetaboAnalyst, FC > 2, p-value < 0.05) from selected doses were uploaded into the program Ingenuity Pathway Analysis (IPA; Qiagen, CA, USA) to perform functional enrichment analysis aiding in identification of the molecular and cellular functions perturbed at 1) environmentally relevant doses, 2) doses below anticipated sub-acute toxicity, as well as 3) dose that was most responsive in our test system across each chemical and species tested based on both number of metabolites changed as well as their ‘biological impact’ as described above. Analysis settings were set to consider direct and indirect relationships of high or experimentally observed confidence within the Ingenuity Knowledge Base consisting of endogenous chemicals as the reference set. IPA functional enrichment analysis results were only considered significant if the calculated p-value was less than 0.05 and contained at least three metabolite entities as conducted by Awkerman et al. (2020). IPA results were further investigated by multi-enrichment analysis using R (version 4.1.0) and the multienrichjam package (version 53.900, http://github.com/jmw86069/multienrichjam). Enriched pathways were used to generate pathway-heatmaps consisting of a subset of the top enriched pathways (n = 20 for ‘Molecular and Cellular Functions’ and n = 30 for ‘Diseases and Disorders’) for each exposure evaluated, which were clustered using Euclidian distance.

Results

The PLS-DA models were developed to aid in the visualization of the biological responses across dose of both D. rerio and X. laevis to each pesticide treatment and in each extracted biological fraction (Fig. 1A–L). Due to the number of compounds and biological replicates, models were constructed for each species:pesticide exposure to avoid potential overfitting of the data. Multivariate model parameters are summarized in Table 2 including the percent variation captured in the first two principal components (PLS-DA) as well as the number of significant metabolites identified (SAM) along with their corresponding false discovery rates (FDR). All models were cross validated with MetaboAnalyst’s Cross Validation procedure searching the first five components using leave one out cross-validation (LOOCV). Using Q2 as the metric indicating the predictive capability of the model, all presented models’ performance had to exceed 0.5 within these boundaries (see Xia and Wishart, 2016 for further interpretation). With this representation, visual inspection of the projection of class data in two-dimensional space affords the ability to summarize similarities and differences in biological response. For instance, the closer in principal component space treatment groups co-locate, afford the assumption of similar variation in metabolomic profiles and the converse is true for farthest separation.

Fig. 1.

Reconstructed PLS-DA scores plots derived from the GC/MS analysis of extracted metabolites from D. rerio (left) and X. laevis (right) exposed to pesticides in the current study (bifenthrin, A–D; chlorothalonil, E–H; trifluralin, I–L).

Table 2:

Descriptive parameters of the PLS-DA models and SAM plots constructed for each species following pesticide exposure.

		PLS-DA				SAM
		Polar models		Non-polar models		Polar models		Non-polar models
Chemical	Species	PC1a	PC2b	PC1	PC2	Significant metabolites	FDR	Significant metabolites	FDR
Bifenthrin	DR	31.7	7.7	41.6	20	54	0.001	5	0.163
	XL	16.6	24	9	10.7	7	0.242	4	1
Chlorothalonil	DR	38	10.5	30.5	11.8	3	0.48	34	0.817
	XL	45.2	26.2	12.2	9.3	7	0.176	12	1
Trifluralin	DR	24	13.9	36.5	9.1	72	0.004	40	0.007
	XL	23.2	14.8	35	12	14	0.063	1	0.05

^a.Percent variation explained by separation along principal component one.

^b.Percent variation explained by separation along principal component two.

For D. rerio exposed to bifenthrin, there is separation between treatment groups suggesting the biochemical profile was dose responsive, however there is no clear trajectory or apparent commonality between the doses (Fig. 1A). The separation along PC2 for all doses when compared to control of the polar metabolites, suggests that there is a common pattern of biological perturbation regardless of dose specifically for doses ≥0.3125 μg/L. Dose-dependent separation is more apparent in the two-dimensional scores plots in the non-polar PLS-DA models following bifenthrin exposure in D. rerio and is evident along PC1 (Fig. 1B). For X. laevis exposed to bifenthrin, less of a dose response was observed and distinctive exposure separation is not evident until doses exceed 0.3125 μg/L in hydrophilic fractions (Fig. 1C) and 0.039 μg/L (Fig. 1D) in hydrophobic fractions. Of interest, in the latter fractions, the lowest does was located farthest from control samples while the highest dose was not statistically significant from control tadpoles.

The PLS-DA plots of D. rerio and X. laevis exposed to chlorothalonil were generally similar with overlap between exposure concentrations in both models for both experimental, metabolite-containing fractions (Fig. 1E–H). The results of these models indicate that the metabolomes of both species were altered similarly as a result of chlorothalonil dose, and in general, principal component separation (via PC1) increases with increasing dose in both polar and non-polar fractions. Regardless of species or metabolite-containing fraction modeled, the highest exposure concentration was located farthest from corresponding controls in principal component space.

Based on the PLS-DA models of D. rerio and X. laevis exposed to trifluralin, the biochemical profile was indicative of dose-responsiveness in each species based on the relatively distinct separation observed along PC1 in both the polar and non-polar fractions (Fig. 1I–L). However, there is a much greater overlap in the treatments for X. laevis relative to controls which indicates that the polar metabolome was perturbed similarly regardless of dose (Fig. 1K) and only in the D. rerio non-polar PLS-DA scores plot, does the lowest dose present similar in principal component space to controls.

For putative peak identification and annotation, the variable importance in projection (VIP) was assessed for each model along principle component (PC)1. Spectral features with a VIP score > 1.0 were deemed significant and each m/z:retention time (R.T.) pair was searched against commercially available databases (described above). Annotated peaks are included in Table 3 and were singularly used for the identification of the top pathways identified as statistically impacted in each test species using MetaboAnalyst’s Pathway Enrichment Analysis module (Table 4).

Table 3.

Metabolites identified as significant from multivariate analysis of D. rerio and X. laevis exposed to three high use pesticides (from PLS-DA VIP >1.0 and SAM).

Metabolites	Bifenthrin		Chlorothalonil		Trifluralin		Total
	DR	XL	DR	XL	DR	XL	DR	XL
11-Eicosenoic acid						X	0	1
13-Octadecenoic acid		X			X		1	1
1-Monopalmitin	X	X	X	X	X	X	3	3
2-Deoxy-d-ribose		X					0	1
2-Monostearin				X			0	1
2-Oleoylglycerol		X					0	1
2-Palmitoleoylglycerol						X	0	1
2-Palmitoylglycerol				X			0	1
4-Aminobutanoic acid	X						1	0
5-Methylcytosine	X						1	0
Adenine	X						1	0
Alanine	X		X	X	X		3	1
Aminomalonic acid			X				1	0
Arachidonic acid	X	X	X	X		X	2	3
Asparagine	X	X					1	1
Aspartic acid		X			X		1	1
Boric acid		X		X	X		1	2
Butanedioic acid	X						1	0
Butanoic acid			X	X		X	1	2
Butyrylglycine			X				1	0
Cadaverine		X				X	0	2
Carbamic acid	X	X	X		X		3	1
Cholesterol		X	X	X	X		2	2
Citric acid					X		1	0
Creatinine					X		1	0
Cytosine	X	X			X		2	1
Deoxyinosine		X					0	1
Deoxyribose-1-phosphate	X						1	0
D-galactopyranoside			X				1	0
Dimethylglycine			X				1	0
Disilathiane		X					0	1
Doconexent	X		X	X	X	X	3	2
Eicosapentaenoic acid			X	X	X	X	2	2
Ethanamine		X	X	X	X		2	2
Ethanedioic acid	X						1	0
Ethanolamine		X					0	1
Ethyldiethanolamine			X				1	0
Fructose	X		X				2	0
GABA	X	X			X		2	1
Galactose	X	X		X			1	2
Gentiobiose		X	X				1	1
Glucose	X	X		X	X	X	2	3
Glutamic acid	X	X	X		X		3	1
Glyceric acid	X						1	0
Glycerol		X	X	X	X	X	2	3
glycerol monostearate		X		X			0	2
Glycerol, 2-O-galactopyranoside	X						1	0
Glycerophosphoric acid	X		X		X		3	0
Glycine	X	X	X	X	X		3	2
Guanidine	X						1	0
Guanine	X		X				2	0
Guanosine	X						1	0
Histidine	X						1	0
Hydroxybutyric acid		X					0	1
Hydroxyisobutyric acid	X	X					1	1
Hydroxypyridine		X					0	1
Hypoxanthine		X	X				1	1
Inosine	X	X	X				2	1
Inositol			X				1	0
Isoleucine	X	X	X				2	1
Itaconic acid		X					0	1
Lactic acid	X	X			X	X	2	2
Leucine		X	X		X		2	1
Linoleic acid	X		X	X		X	2	2
Lysine	X	X					1	1
Malonic acid			X		X		2	0
Maltose			X				1	0
Mandelic acid					X		1	0
Mannose	X		X	X	X		3	1
Mannose-6-phosphate	X						1	0
Mead acid		X					0	1
Methionine		X					0	1
Methylmalonic acid				X			0	1
Myo-Inositol	X	X	X	X	X	X	3	3
Myo-inositol phosphate			X				1	0
Myristic acid						X	0	1
N-acetyl lysine		X					0	1
N-Acetyl-D-glucosamine	X	X					1	1
Nonadecanoic acid	X						1	0
Nonanoic acid		X					0	1
Norleucine	X	X	X			X	2	2
Oleic acid	X		X	X	X	X	3	2
Ornithine	X						1	0
Oxalic acid		X	X	X			1	2
Palmidrol		X			X		1	1
Palmitelaidic acid		X		X			0	2
Palmitic acid	X	X	X	X	X	X	3	3
Palmitoleic acid		X					0	1
Pentadecanoic acid						X	0	1
Phenylalanine		X					0	1
Phenylethanolamine			X				1	0
Phosphoethanolamine	X						1	0
Phosphoric acid	X	X	X	X	X	X	3	3
Pipecolic acid	X						1	0
Proline	X	X	X				2	1
Propionnamide					X		1	0
Propylamine					X	X	1	1
Putrescine		X		X			0	2
Pyridine		X	X		X		2	1
Pyroglutamic acid	X	X	X		X		3	1
Pyruvic acid	X						1	0
Ribitol		X	X				1	1
Ribofuranose	X		X				2	0
Ribose	X	X	X				2	1
Ribose-5-phosphate			X				1	0
Serine	X	X	X				2	1
Stearic acid	X	X	X	X	X	X	3	3
Succinic acid	X	X	X				2	1
Taurine			X				1	0
Threonine	X	X	X				2	1
Thymine	X				X		2	0
Tryptophan	X						1	0
Tyramine			X				1	0
Tyrosine	X	X					1	1
Uracil			X				1	0
Urea	X	X	X	X	X	X	3	3
Uridine		X					0	1
Vaccenic acid				X		X	0	2
Valine	X	X	X		X	X	3	2
Xanthine	X					X	1	1
Total:	59	62	54	29	37	25	150	116

Table 4.

Top pathways identified as statistically impacted in each test species following pesticide exposure. Metabolites identified as significantly impacted from PLS-DA VIP scores and SAM analysis were used for pathway enrichment analysis.

	Metabolite set	Total	Hits	Expect	p value	FDR
Bifenthrin	D. rerio
	Urea cycle	29	6	1.56	0.003	0.13
	Glucose-alanine cycle	13	4	0.7	0.004	0.13
	Glutathione metabolism	21	5	1.13	0.004	0.13
	Ammonia recycling	32	6	1.72	0.006	0.14
	Alanine metabolism	17	4	0.91	0.011	0.19
	Glutamate metabolism	49	7	2.63	0.013	0.19
	Purine metabolism	74	9	3.97	0.014	0.19
	Arginine and proline metabolism	53	7	2.85	0.02	0.24
	Glycine and serine metabolism	59	7	3.17	0.034	0.37
	Galactose metabolism	38	5	2.04	0.048	0.47
	X. laevis
	Arginine and proline metabolism	53	7	2.95	0.024	0.93
	Glutathione metabolism	21	4	1.17	0.025	0.93
	Ammonia recycling	32	5	1.78	0.028	0.93
	Glutamate metabolism	49	6	2.73	0.049	1
Chlorothalonil	D. rerio
	Glutathione metabolism	21	5	0.96	0.002	0.19
	Alpha linolenic acid and linoleic acid metabolism	19	4	0.87	0.009	0.45
	Arginine and proline metabolism	53	6	2.43	0.03	0.76
	Alanine metabolism	17	3	0.78	0.039	0.76
	Urea cycle	29	4	1.33	0.04	0.76
	Purine metabolism	74	7	3.4	0.047	0.76
	X. laevis
	Alpha linolenic acid and linoleic acid metabolism	19	4	0.48	0.001	0.1
	Galactose metabolism	38	5	0.97	0.002	0.1
	Glutathione metabolism	21	3	0.53	0.014	0.47
	Lactose degradation	9	2	0.23	0.02	0.49
	Urea cycle	29	3	0.74	0.034	0.67
	Glucose-alanine cycle	13	2	0.33	0.041	0.67
Trifluralin	D. rerio
	Glutathione metabolism	21	5	0.7	0.0004	0.04
	Urea cycle	29	5	0.96	0.002	0.1
	Glutamate metabolism	49	6	1.63	0.004	0.14
	Glucose-alanine cycle	13	3	0.43	0.008	0.19
	Alanine metabolism	17	3	0.56	0.017	0.31
	Ammonia recycling	32	4	1.06	0.019	0.31
	Arginine and proline metabolism	53	5	1.76	0.027	0.38
	Galactose metabolism	38	4	1.26	0.034	0.4
	Warburg effect	58	5	1.93	0.038	0.4
	Malate-aspartate shuttle	10	2	0.33	0.041	0.4
	Glycerolipid metabolism	25	3	0.83	0.047	0.4
	Glycerol phosphate shuttle	11	2	0.37	0.049	0.4
	X. laevis
	Alpha linolenic acid and linoleic acid metabolism	19	4	0.45	0.0007	0.07
	Gluconeogenesis	35	3	0.82	0.045	1
	Fatty acid biosynthesis	35	3	0.82	0.045	1

Significance analysis of microarray data, or ‘metabolites’ (SAM) plots, were generated to complement the PLS-DA models (Fig. 2). Interpretation of PLS-DA models in data-rich studies such as metabolomics can potentially overpredict biological response patterns due to the algorithms being applied in these supervised classification methods. Thus, we coupled these analyses with SAM to apply a more ‘robust’ method of significant feature identification. SAM uses non-parametric statistical measures across repeated iterations of the multi-class datasets to determine statistical differences in metabolite response regardless of the assumptions of normality in the data (Efron and Tibshirani, 2007). Visual representation of statistically significant features identified with SAM analyses display these ‘metabolites’ in green (Fig. 2). For interpretation of the SAM plots, the dashed lines diagonal and parallel to the black line (expected difference) are drawn at distance delta and represent the model’s statistical inference (i.e., dots further away from the line are differentially expressed the greatest). The summary statistics for each SAM analysis are included in Table 2.

Fig. 2.

Significance Analysis of Metabolite (SAM) plots from the analysis of the metabolome of D. rerio (left) and X. laevis (right) exposed to pesticides in the current study (bifenthrin, A–D; chlorothalonil, E–H; trifluralin, I–L).

SAM analysis of both metabolite-containing fractions extracted from D. rerio exposed to bifenthrin and trifluralin (Fig. 2A (n = 54) and I (n = 72); polar and Fig. 2B (n = 5) and J (n = 40); non-polar, respectively) had a considerably higher number of metabolites significantly altered as a result of exposure compared to X. laevis and the other pesticide exposures (Fig. 2C and K, respectively). Although not fully indicative of the biological response of pesticide exposure in each species, these numerical counts can be used as descriptive indicators of potential disruptions and compared to overall magnitude of response (as described below). The number of significantly altered metabolites was similar for both D. rerio and X. laevis exposed to chlorothalonil with X. laevis displaying a slightly greater number of metabolites that were significantly changed in the polar fractions (Fig. 2E and G). Interestingly, non-polar analysis of chlorothalonil and trifluralin exposures increased the number of statistically perturbed metabolites identified for D. rerio (Fig. 2F and J). These SAM plots demonstrate that, holistically, following pesticide exposure fewer metabolites are altered in X. laevis as opposed to D. rerio and this appears to correlate with their overall biological impact.

The highest number of altered metabolites putatively identified in D. rerio and X. laevis occurred as a result of exposure to bifenthrin with 59 and 62 metabolite identifications, respectively. In addition, bifenthrin exposure resulted in the highest amount of concurrence in putatively identified, altered metabolites between the two species (n = 30; Table 3). Similarity in metabolite perturbations included changes in various amino acids (i.e., asparagine, cytosine, GABA, glutamic acid, isoleucine, lysine, and proline) as well as sugars, such as glucose and galactose. Similar perturbations were seen in the non-polar identified metabolites and changes in saturated fatty acids (i.e., stearic and palmitic acids) and a polyunsaturated fatty acid (PUFA; arachidonic acid) were statistically altered by exposure in fish and amphibians. Based on the Student’s t-test filtered spectra of the polar fractions, the number of metabolites perturbed by bifenthrin exposure in D. rerio appears indicative of overall biological impact and overall magnitude of impact positively correlates with number of metabolites (SI Fig. 1A). However, for X. laevis, at the two lowest exposure concentrations (0.039 and 0.078 μg/L), perturbations in a high number of metabolites were identified but their overall magnitude of change was significantly lower than that of the 0.3125 exposure group (SI Fig. 1C). Similar trends in metabolite number and overall impact were observed in the non-polar fractions of D. rerio exposed to bifenthrin, but the trend was reversed in the lipophilic metabolome of X. laevis (SI Fig. 1B and D) and the highest number of metabolites and corresponding maximal impact was observed at the lowest exposure concentrations.

Chlorothalonil exposure resulted in 54 and 29 perturbed metabolites in D. rerio and X. laevis, respectively. The metabolites that were altered in both species’ exposures (n = 19) included alanine, butanoic acid, glycerol, glycine, mannose, myo-inositol, oxalic acid, phosphoric acid, and urea. While fewer metabolites were identified as significant different in X. laevis, 65% are involved in the biological response of D. rerio as well. The overall biological impact and its concordance with number of perturbed metabolites in D. rerio was positive with lower doses, but alterations in greater number of metabolites appeared to cause less quantitative impact at doses exceeding 6.25 μg/L (SI Fig. 1E). However, based on interpretation of the non-polar data, number of metabolites was indicative of both increasing dose and magnitude of response in D. rerio and X. laevis (SI Fig. 1F and H).

Thirty-seven and twenty-five metabolites were perturbed in response to trifluralin exposure for D. rerio and X. laevis, respectively. The metabolites glucose, glycerol, lactic acid, myo-inositol, phosphoric acid, and valine were similarly modified in both species following trifluralin exposure (Table 3). Fourteen metabolites were identified to be in common across the two species. Trifluralin exposure in D. rerio caused the largest change in number of metabolites at 120 μg/L while the largest magnitude of change was identified in the 30 μg/L exposures (SI Fig. 1I). Interestingly, there is an overall negative correlation in the absolute sum of metabolite perturbations with the number of metabolites altered in both the polar and non-polar fractions of X. laevis (SI Fig. 1K and L).

In total, there were 150 metabolites altered in D. rerio compared to 116 metabolites that were altered in X. laevis across all pesticide treatments (Table 3). Although there appears to be limited concordance in the metabolites altered across all pesticides at the species level, when grouping the compounds by chemical class, trends in dysregulation appear. In D. rerio exposures, regardless of pesticide, monopalmitin, alanine, carbamic acid, doconexent, glyceric acid, glycerophosphoric acid, glycine, mannose, myo-inositol, oleic acid, palmitic acid, pyroglutamic acid, stearic acid, urea and valine were all altered. The same was true for monopalmitin, arachidonic acid, glutamic acid, glycerol, stearic acid, and urea in X. laevis exposures. Myo-inositol, palmitic acid, stearic acid and urea were the metabolites significantly altered in both species across all pesticide exposures. Phosphoric acid, potentially due to its role as a common co-factor, was also altered across all treatment exposures in D. rerio and X. laevis (Table 3). Overall, a number of amino acids and fatty acids were impacted by pesticide exposure, as well as metabolites involved in cellular energetics. These disruptions can facilitate the identification of biologically relevant pathways impacted and aid in understanding the similarities and differences in each species’ response. Metabolites were used to identify the top statistically responsive biological pathways using MetaboAnalyst’s Pathway Enrichment Analysis module prior to IPA’s functional enrichment analysis (Table 4). The relevance of identified pathways is included in the Discussion section below.

Functional Enrichment Analysis

Functional enrichment analysis in IPA identifies significantly altered biofunctions and can produce a z-score that indicates inhibition (z-score ≤ −1.5) or activation (z-score ≥ 1.5) given enough data on the directionality of metabolite change and sufficient understanding of the interactions of specific metabolites on pathway activity. This pathway analysis identified altered diseases/disorders, molecular/cellular functions, and physiological development/functions using fold change and p-value data for all metabolites that were significantly altered by exposure for bifenthrin, chlorothalonil, and trifluralin at three different concentrations in order to evaluate dose response and environmentally-relevant trends (Table 5).

Table 5.

Concentrations used for Ingenuity Pathway Analysis.

Chemical	Species	Environmentally relevant (μg/L)	Sub-acute toxicity (μg/L)	Maximum perturbations (μg/L)
Bifenthrin	DR	0.156a	0.078	0.3125
	XL	0.156a	0.078	0.078 (0.039)b
Chlorothalonil	DR	0.781c	6.25	0.78 (12.5)b
	XL	0.781c	6.25	6.25
Trifluralin	DR	15c	30	120
	XL	15c	240	30

^a.Stehle et al. (2019).

^b.Maximal perturbation dose already included, used 2nd maximal.

^c.Reilly et al. (2012).

In general, the response profile for bifenthrin was similar across species and the greatest of all pesticides with 555 and 530 unique biofunctions that were significantly altered across all concentrations evaluated with IPA for D. rerio and X. laevis, respectively (SI Table 4). In D. rerio, 257 and 509 unique biofunctions were significantly altered for chlorothalonil and trifluralin, respectively. X. laevis had a less robust response to chlorothalonil and trifluralin with 176 and 155 unique biofunctions altered, respectively.

Bifenthrin exposure at 0.078, 0.156, or 0.3125 μg/L resulted in significant alterations to 210, 401, and 437 biofunctions in D. rerio (SI Table 4). In X. laevis exposure to 0.039, 0.078, or 0.156 μg/L altered 393, 297, and 253 biofunctions, respectively (SI Table 4). A dose responsive increase in the number of altered biofunctions and the calculation of a directionality predictive z-score was observed in D. rerio (Fig. 3A) while a dose responsive decrease in the number of altered biofunctions was observed in X. laevis (Fig. 3B). Bifenthrin exposure elicited a similar response in D. rerio and X. laevis as the two most impacted categories of biofunctions altered by exposure were lipid metabolism and cell-to-cell signaling and interaction (Fig. 3A and B), albeit more pronounced in D. rerio. Strong evidence of activation in biofunctions related to lipid metabolism and free radical scavenging exist for D. rerio but not in X. laevis (Fig. 3A and B).

Fig. 3.

Functional enrichment analysis of molecular and cellular functional categories significantly enriched by bifenthrin exposure in Danio rerio (A) and Xenopus laevis (B). Colored bars represent categories that were significant and had an activation z score <−1.5 (blue) or >1.5 (red).

Chlorothalonil exposure at 0.781, 6.25, or 12.5 μg/L resulted in the alteration of 179, 109, and 121 biofunctions in D. rerio (SI Table 4). A z-score ≥ |1.5|, was calculated for only 11 biofunctions, 9 of which were in the lowest concentration group evaluated with IPA (Fig. 4A). In X. laevis, exposure to 6.25 μg/L chlorothalonil resulted in the significant alteration of 176 biofunctions (SI Table 4) of which only 9 had a predictive z-score. No significant biofunctions were identified in X. laevis following exposure to 0.78 and 3.125 μg/L chlorothalonil (Fig. 4B). Lipid metabolism and vitamin and mineral metabolism were the two most impacted categories for both species exposed to chlorothalonil (Fig. 4A and B).

Fig. 4.

Functional enrichment analysis of molecular and cellular functional categories significantly enriched by chlorothalonil exposure in Danio rerio and Xenopus laevis (B). Colored bars represent categories that were significant and had an activation z score <−1.5 (blue) or >1.5 (red).

Trifluralin exposure at 15, 30, or 120 μg/L resulted in significant alterations to 236, 371, and 391 functions in D. rerio (SI Table 4). In X. laevis exposure to 15, 30, or 240 μg/L trifluralin altered 74, 110, and 85 biofunctions, respectively (SI Table 4). A dose responsive increase in the number of altered biofunctions and the calculation of a directionality predictive z-score was observed in D. rerio (Fig. 5A) but not X. laevis (Fig. 5B). When investigating the response of trifluralin in both D. rerio and X. laevis, metabolomic profiling identified many of the same top molecular and cellular functions perturbed across both species. With these, vitamin and mineral metabolism, amino acid metabolism, cell-to-cell signaling, lipid metabolism, and carbohydrate metabolism were important in response to trifluralin exposure (Fig. 5A and B). Moreover, in D. rerio’s response to trifluralin exposure, dose-responsive trends are more apparent in the functions of amino acid metabolism, carbohydrate metabolism, cell cycle, and cell-to-cell signaling (Fig. 5A).

Fig. 5.

Functional enrichment analysis of molecular and cellular functional categories significantly enriched by trifluralin exposure in Danio rerio (A) and Xenopus laevis (B). Colored bars represent categories that were significant and had an activation z score <−1.5 (blue) or >1.5 (red).

While all of the IPA results are provided in the supplemental materials as outlined above, comparisons across species and pesticides were made with the results of the functional enrichment analysis using the highest dose identified as unlikely to elicit overt toxicity using data from the Ecotox database for fish and amphibians and our previous study for trifluralin (Awkerman et al., 2020) (Table 5). The biofunctions that were identified with multi-enrichment analysis as the top enriched for each species/pesticide and of specific interest to this study are highlighted for ‘Molecular and Cellular Function’ (Fig. 6A; SI Table 3) and ‘Diseases and Disorders (Fig. 6B; SI Table 3).

Fig. 6.

Multi-enrichment heatmap of selected top altered biofunctions as determined by IPA’s core metabolomics analysis in the highest concentrations identified as unlikely to elicit overt toxicity for each species and pesticide tested (Table 5). Biofunctions representing ‘Molecular and Cellular Functions are in (A) while biofunctions related to ‘Diseases and Disorders’ are in (B). Deeper red coloration denotes a smaller p-value (larger -log10(p-value) while biofunctions that were not significantly altered (p-value >0.05, -log10(p-value) >1.3) are not colored. A table of the full set of results provided in SI Table 4.

For all pesticides and in both species, evidence of altered energy production (ATP concentration, dysfunction of mitochondria, and synthesis of carbohydrate), cell death (necrosis, apoptosis, cell death of connective tissue cells, cell death of central nervous system cells, killing of cells, and cell viability), and cell cycle (cell cycle progression, interphase, s phase, arrest in G1 phase, arrest in G0 phase, and arrest in interphase) were significantly impacted (Fig. 6A). Ferroptosis, a type of cell death associated with iron accumulation and lipid peroxidation, was identified only for bifenthrin. Biofunctions related to cancer (development of malignant tumor, cancer, and malignant solid tumor) were altered for both species and across all pesticides (Fig. 6B). Hepatic steatosis was significantly altered in all exposures evaluated except for trifluralin exposed X. laevis. Other ‘Diseases and Disorders’ of note are diabetes (diabetes mellitus and insulin-dependent mellitus) in most of the exposures evaluated and neurological disorders (neuromuscular disease, movement disorders, and Parkinson’s disease) in chlorothalonil exposures (Fig. 6B).

Discussion

In the current study, larval D. rerio and X. laevis tadpoles were exposed to three high-use pesticides to determine if the metabolomic response between the two species is similar at the life stages examined. Frequently, toxicity data from D. rerio exposures are used as a surrogate to represent or estimate amphibian toxicity for ecological risk assessments. However, based on Glaberman et al. (2019), there remains concern regarding the adequacy of fish surrogacy for aquatic-phase amphibians especially in chronic exposure scenarios. To attempt to assess if D. rerio is an adequate surrogate for these higher tiered risk assessments, we utilized GC/MS-based metabolomics to ultimately determine if D. rerio elicits a similar toxic response to X. laevis at two discrete but important life stages. Data generated from this study will provide insights into future methodology and the use of ‘omics’ analysis in surrogate species data used in ecological risk assessments.

Bifenthrin

Bifenthrin, is an α-cyanogroup pyrethroid which causes irreversible blockage of sodium-gated channels leading permanent membrane depolarization. This mechanism of action is conserved across species and evidenced in this study as the predicted altered biofunctions of transmembrane potential and depolarization of cellular membranes (Fig. 6A) were found to be amongst the top altered biofunctions. Cell cycle arrest at different checkpoints were predicted to be altered in this study (Fig. 6A) suggesting that bifenthrin induces cellular stress that when not repaired can lead to programmed cell death, apoptosis, necrosis, and/or ferroptosis. In our acute toxicity studies, a large number of metabolites as well as their magnitude of change were altered in both D. rerio and X. laevis as a result of bifenthrin exposure, suggesting that multiple biochemical pathways were impacted (Table 4). These included commonality in disruptions in arginine and proline metabolism, glutathione metabolism, ammonia recycling and glutamate metabolism. In fact, based on pathway enrichment analysis, the top perturbed pathways identified in X. laevis were all disrupted in D. rerio exposures.

In addition, all four nucleotide bases, adenine, cytosine, thymine, and guanine, were altered in D. rerio as a result of bifenthrin exposure suggesting deficiencies in nucleotide metabolism which could be indicative of complications in DNA replication or synthesis. Specifically, in D. rerio and X. laevis, biofunctions related to DNA replication, recombination and repair were altered after bifenthrin exposure (Fig. 3). In Miao et al. (2017), one of the primary pathways impacted included nucleotide catabolism, and inosine, hypoxanthine, and guanosine were predictive of bifenthrin’s induction in inflammation and apoptosis in Chinook salmon brains (Magnuson et al., 2020a). In juvenile steelhead trout, similar perturbations in adenine, guanosine, and guanine were significant following toxic insult (Magnuson et al., 2020b). Trends observed in the two Oncorhynchus’ mirror the metabolomic perturbations observed in D. rerio, and changes in adenine, glutamic acid, guanosine, inosine, isoleucine, succinic acid, and uridine were all differentially regulated following bifenthrin exposure. It was shown that the increases in succinic acid (and lactic acid) accompanied by decrease of glutamine in the liver extracts of mice indicated that the energy homeostasis was perturbed (Miao et al., 2017), and this appears true for D. rerio (current study).

In the current study, thirteen amino acids were altered in X. laevis exposed to bifenthrin, and eleven of these amino acids were shared in their alterations between D. rerio except for n-acetyl-lysine and leucine. In X. laevis, cytosine was the only nucleotide that was changed, suggesting that bifenthrin’s impact in amphibians varies from those identified relevant in fish at the life-stage examined. In comparison, Southern leopard frogs (Lithobates sphenocephala) exposed to 1 ppm solution of bifenthrin exhibited defects in glutathione metabolism and ammonia recycling (Glinski et al., 2021) which was similar to observations in amphibians in this study.

Non-polar metabolites identified as significantly altered in both species as a response to bifenthrin exposure include the fatty acids: palmitic acid, stearic acid, and arachidonic acid. This was similar to results ascertained by Li et al. (2021) in X. laevis tadpoles exposed to bifenthrin. Specifically, the disruption of lipid homeostasis with decreases in pentadecanoic acid and dihomo-γ-linolenic acid with overall increases in cholesterol (and LDL) levels occurred (Li et al., 2021). Similarly, in zebrafish, significant changes in lipid biosynthesis and metabolism were identified by changes in fatty acids and their degree of saturation following maternally mediated bifenthrin exposure (Xiang et al., 2019). It has also been reported that inflammatory and immunotoxic responses in brain of juvenile trout were governed by changes in docosahexaenoic acid (termed doconexent in Table 3), frequently derived from α-linolenic acid. Interestingly, linolenic acid was significantly altered in D. rerio exposures in the current study. Ultimately, understanding both the similarities and differences in metabolomic profiles of these two species’ response to bifenthrin exposure will further develop its apparent species-specific MOA such as the alterations in calcium regulation and free radical scavenging related biofunctions predicted in D. rerio and to a lesser extent in X. laevis (Fig. 6 and SI Table 3).

Chlorothalonil

In D. rerio, chlorothalonil resulted in the perturbation of 54 metabolites and based on PLS-DA model class separation, elicited a linear response in the metabolome with increasing dose (Fig. 1E). Analysis of the identified metabolites resulted in glutathione metabolism; alpha linolenic acid and linoleic acid metabolism; and arginine and proline metabolism being identified as the most impacted pathways (based on calculated enrichment ratios). Similarly, Yang et al. (2021) concluded that disruptions in glycolysis and amino acid metabolism were indicative of chlorothalonil exposure in larval zebrafish. Gene expression analysis of adult male zebrafish exposed to environmentally relevant and sublethal concentrations of chlorothalonil exhibited decreases in hepatic genes related to DNA and cell replication, DNA damage, detoxification, and metabolic processes (Garayzar et al., 2016). However, in the current study, the only nucleotide altered in D. rerio was uracil while, in total, seven metabolites impacted were involved in purine metabolism (Table 4). Garayzar et al. (2016), observed an upregulation of phosphoglucomutase 1, a key enzyme in glycogenesis that is typically increased as a result of hypoxia, suggesting that chlorothalonil interferes with cellular respiration and elicits oxidative stress. Furthermore, da Silva Barreto et al. (2020), concluded that chlorothalonil increased lipid peroxidation and ROS by altering redox state in the gill of zebrafish. However, in the current study, in D. rerio, there was a significant, predicted inhibition of lipid peroxidation following chlorothalonil exposure which suggests that lipids could be protected from oxidative damage while in X. laevis there is a predicted activation in lipid peroxidation. Interestingly, in da Silva Barreto et al. (2020), it was concluded that potential mechanisms of metabolism of chlorothalonil (in gills) could be indicated in the protection of liver from similar oxidative stress exposure. Based on pathway analyses, fluxes in the metabolites observed and their role in glucose homeostasis and lipid metabolism are implicated in the impacts of exposure to chlorothalonil in larval zebrafish (Yang et al., 2021 and the current study) and glutathione metabolism was the top pathway impacted. Thus, it can be postulated that lipid damage (i.e., reductions in LPO levels in da Silva Barreto et al., 2020) and imbalances (current study), are likely related to antioxidant response and glutathione levels (specifically, GST activity).

X. laevis exposed to chlorothalonil exhibited fewer alterations in metabolites and identified biofunctions than D. rerio (Table 3, Table 4). Both species shared perturbations in alanine, butanoic acid, glycerol, glycine, mannose, myo-inositol, oxalic acid, phosphoric acid, and urea (Table 3) resulting in similar pathway perturbations in linolenic acid metabolism and the urea cycle. Previously, 1H NMR metabolomics has been utilized to identify biomarkers indicative of chlorothalonil exposure in earthworms, Eisenia fetida (Griffith et al., 2019) with changes associated with amino acid, energy, purine, and pyrimidine metabolism. Reduced but not statistically significant levels of glutathione were also observed, consistent with chlorothalonil’s mechanism of action since it targets glutathione and thiol-dependent enzymes (Griffith et al., 2019). In addition, there were also increased levels of glutamine, N-acetylserine, and ophthalmic acid metabolite levels which potentially indicate that chlorothalonil caused oxidative stress in the target species (Griffith et al., 2019). Similar to the mechanisms hypothesized above for ROS induction in zebrafish, changes in lipid profiles can be associated with modifications in the glutathione metabolic pathway and support this is relevant in amphibians as well. However, noted changes in galactose metabolism and glutathione as well as amino acid and nucleotide sugar metabolism were significant pathways in this study’s acute D. rerio exposures. Taken together, these studies demonstrate that chlorothalonil is acutely toxic to many fish and amphibian species and its disruptions in sugar and lipid homeostasis result in oxidative stress through mechanisms that are potentially shared amongst species, warranting further investigation.

Trifluralin

In the current study, metabolic changes were observed in citric acid, leucine, glucose, valine, lactic acid, urea, and pyroglutamic acid in D. rerio exposed to trifluralin (Table 3). Metabolomic fluxes were also noted for alanine, aspartic acid, boric acid, creatinine, cytosine, glycine, glutamic acid, malonic acid, mannose, and thymine (Table 3). In comparison albeit resulting from chronic exposures, Awkerman et al. (2020) also observed lower levels of tryptophan, leucine, isoleucine, citric acid, and oxaloacetic acid in D. rerio which suggests that the generation of acetyl-CoA could be reduced and energy production through the citric acid cycle is impaired as a result of trifluralin exposure. In the current study, X. laevis did not demonstrate the same metabolic effect in response to trifluralin exposure as observed in (Awkerman et al., 2020) most likely due to the discrete developmental stage examined herein. Both X. laevis and D. rerio exhibited increased levels of lactic acid which could potentially indicate that both species were combating the high energy demand required for detoxification (Gray et al., 2014) and this phenomenon was also observed in Awkerman et al. (2020). Furthermore, disruptions in cell cycle, DNA damage, and mitochondrial dysfunction were all implicated in the trifluralin exposure response in both species.

Cross-Species and Cross-Pesticide Pathway Analysis

Overall, six metabolites were altered across both species and all pesticide exposures: myo-inositol, a cyclohexane, which is involved in galactose and inositol metabolism; phosphoric acid (aka phosphate) which is commonly utilized in purine, glycine, and serine metabolism and is a common co-factor in numerous facets of cellular energetics; two fatty acids (stearic acid and palmitic acid) and a derivative (1-monopalmitin) which are essential for cell membranes and in the derivation of cellular energy; and urea, necessary for nitrogen metabolism and ammonia handling. For D. rerio, the amino acids alanine, glutamic acid, and pyroglutamic acid were impacted across all pesticide treatments while butanoic acid, glucose, and glycerol metabolites were altered in all pesticide exposures for X. laevis. Fatty acids, both saturated and polyunsaturated were also commonly impacted in both species across all pesticide exposures. Additional perturbed metabolites suggest further disruptions in glucogenesis/glycolysis as well as hypothesized disruptions in the glucose-alanine pathway.

Amino acid metabolism has been indicated in the response of numerous taxa to pesticide exposure including all pesticides examined in the current study. This finding, coupled with changes in TCA cycle intermediates feeding into glycolytic pathways, and disruptions in carbohydrate and lipid metabolism are common themes in our analysis corroborated by existing literature (as referenced in each section). Again, these responses in energetics across species may represent a generalized adverse response to pesticide exposure that could serve as predictive responses prior to the onset of overt toxicity.

From a pathway level analysis, common perturbed metabolic pathways include glutathione metabolism, urea cycle, alanine metabolism and arginine and proline metabolism across all pesticides in D. rerio. No pathway was similarly perturbed across all exposures in X. laevis. In response to pesticide exposure, similar pathway disruptions were identified in both species for glutathione metabolism, ammonia recycling, glutamate metabolism and arginine and proline metabolism (bifenthrin) and alpha linolenic acid and linoleic acid metabolism, urea cycle, glutathione metabolism (chlorothalonil).

Currently, there are limited studies delineating the importance of the changing amphibian metabolome during metamorphosis and the impacts of pesticide exposure on potential biochemical targets. In Ichu et al. (2014), five biological pathways were identified as relevant to the varying stages of metamorphosis including arginine and purine/pyrimidine metabolism, cysteine/methionine metabolism, sphingolipid metabolism (and hydrolysis of glycosphingolipid), eicosanoid metabolism, and the urea cycle. Ultimately, metabolites present in the urea cycle and to a lesser extent in arginine and purine/pyrimidine metabolism, were uniquely able to differentiate between pre- and post-metamorphs (Helbing, 2012; Ichu et al., 2014). In the current study, bifenthrin was the only pesticide that perturbed metabolites (i.e., uridine and proline) relevant to these pathways while bifenthrin was the singular pesticide to impact key metamorphic metabolites such as cytosine, hypoxanthine, dexoyinosine, glutamine and lysine (Ichu et al., 2014). Additional studies are needed to further investigate the biochemical consequence of pesticide exposures on these critical pathways and how the metabolite profile changes during metamorphosis (Helbing, 2012).

Conclusion

With amphibian populations declining around the world and pesticide exposure considered to be a major contributor, it is extremely important that risk assessments make use of the most robust information possible when evaluating risks to amphibians. Currently, fish toxicity data is used to represent amphibian sensitivity. In this study, we compared the biochemical responses of D. rerio and X. laevis at two-discrete life stages to three high-use pesticides by employing GC/MS-based metabolomics to evaluate if the toxic responses to pesticide exposure between the two species were similar and/or predictive. Overall, the metabolome of D. rerio exhibited similar responses to pesticide exposure as X. laevis and comparable metabolic pathways were altered in both species across pesticide treatments. The results from this study can further inform the applicability of the surrogate species approach for evaluating the aquatic toxicity of pesticides in ecological risk assessments by providing a platform suitable for further research efforts.