An integrated approach to evaluating water contaminants and evaporation in agricultural water distribution systems

Abstract

This study presents an innovative approach for assessing water quality in agricultural irrigation networks, integrating stable isotope analysis, in vivo zebrafish screening, and comprehensive chemical profiling to investigate the occurrence, transformation, and potential toxicity of organic contaminants. Stable isotope analysis was used to measure evaporation as a proxy for water residence time in the canal, while liquid chromatography-high resolution mass spectrometry (LC-HRMS) identified a range of organic compounds in water samples collected from both the irrigation canal and its source river. Results indicated a reduction in contaminant levels in the canal compared to the river, with the most significant evaporation and concentration changes occurring at a holding reservoir, suggesting that managing residence time could help reduce water loss in arid irrigation networks. The data also highlighted how evaporation, particularly during the dry, hot season, influences contaminant dynamics. Hierarchical clustering of LC-HRMS results showed notable differences between the chemical profiles of canal and river samples, indicating that irrigation systems may contribute to the degradation or removal of certain compounds. Over 60 % of detected compounds were naturally derived, with anthropogenic contaminants like pesticides and personal care products further highlighting human impacts. Priority contaminants, including DEET and 2-naphthalene sulfonic acid, likely originated from urban activities upstream. Initial screening using zebrafish embryos showed bioactivity across sites, confirming the presence of contaminants needing further examination. Correlation analysis linked natural compounds to evaporation rates, suggesting that flora and fauna play significant roles in the chemical makeup of canal water. Overall, this approach provides a comprehensive framework for monitoring irrigation water, offering insights into contaminant behavior and supporting the development of standardized methods for assessing chemical fate and ecological risks in agricultural irrigation systems.

1. Introduction

Irrigation networks are a critical infrastructure for water distribution and food production in agricultural systems (Hrozencik, 2022). In Oregon, over 15 million acres are devoted to agriculture and 80 % of the available water resources in the state are used on agricultural lands (Brewer et al., 2008; Schumacher et al., 2022). Farmers rely on irrigation networks to maintain this vital part of the economy, especially in the arid eastern part of the state (Hrozencik, 2022). Many of the irrigation networks were constructed over 70 years ago and have not been updated since (Evett et al., 2020; Sojka et al., 2002). Assessing and increasing the efficiency of aging irrigation systems is a regional priority (Farmers Conservation Alliance, 2020a, 2020b). In addition to infrastructure challenges, drought, and increased evaporation due to warmer temperatures associated with climate change have led to water scarcity and reliance on marginal water supplies (Dehghani et al., 2022; Iqbal et al., 2002; Sivapragasam et al., 2009). Water scarcity and the increasing demand on limited water resources are forcing the agricultural community to adopt new water resource management tools for water distribution and preservation. Consequently, there is a need to advance water quantity and quality monitoring in irrigation networks, as well as improve efficiency of these complex hydrological systems.

Anthropogenic chemicals are a threat to freshwater resources, agricultural sustainability and healthy ecosystems and communities. Currently, there are no standard assessment methods for organic chemicals in irrigation systems. The routine monitoring efforts that have been implemented for organic chemical pollutants in agricultural systems are often limited to pesticides runoff (Lothrop et al., 2018; Tian et al., 2021). In recent years, as treated wastewater has been explored as a source for irrigation water, more attention has been given to water quality, effects and fate of organic chemicals on crop and soil health (Alygizakis et al., 2023; Brock et al., 2003; Deviller et al., 2020; Margenat et al., 2017). These studies have utilized mass spectrometry methodologies in the measurement of pesticides and other organic contaminants from wastewater (Beretsou et al., 2022; Calderón-Preciado et al., 2011; Gravert et al., 2021). To our knowledge, no studies have utilized suspect and nontarget screening approaches to assess the overall composition of organic chemical contaminants in irrigation water distributions systems in the United States.

Water distribution systems fed by urban or semi-urban watersheds or near industrial areas, such as the river in this study, may contain contaminants leading to crop and environmental health degradation (Margenat et al., 2017; Peng et al., 2017). Chemical sources of pollutants in waterways include industrial emissions, agricultural runoff, and chemicals from consumer products (Schwarzenbach, 2006). Some efforts aimed to understand the potential negative effects of these contaminants on human health during food consumption or recreational exposure have been published (Brock et al., 2003; Margenat et al., 2017). However, there are currently no standardized methods to monitor and assess organic contaminants and their potential toxicity in irrigation water and water distribution systems. Our study aims to fill this gap by applying state-of-the-art organic chemical analysis with high throughput biological assays to assess water quality and toxicity.

Ultra-high performance liquid chromatography (UPLC) in tandem with high-resolution mass spectrometry (HRMS) is a well-established technique used for detection and quantification of organic chemicals in environmental and biological samples (Bletsou et al., 2015; Hernández et al., 2019). The technological advancement of mass analyzers has led to increased mass accuracy and resolving power allowing for the reliable and reproducible measurement of low concentration compounds (Aceña et al., 2015; Pereira et al., 2021; Pérez-Fernández et al., 2017). This has resulted in an expansion of qualitative suspect and nontarget screening approaches. Suspect screening relies on exact m/z (mass-to-charge) values and spectral library matching to identify compounds in a sample (González-Gaya et al., 2021). Nontarget screening (NTS) is an unbiased approach that uses unassigned chemical features to describe trends in the system and simultaneously works to identify the compounds that might be driving those trends using computational approaches (González-Gaya et al., 2021; Hohrenk et al., 2020; Schymanski et al., 2015).

In this study, we aimed to assess the chemical composition and transport in a regional irrigation canal and its source river. To address these questions, we utilized UPLC-HRMS to assess the organic chemical composition of water samples collected from an irrigation canal and its source river. In addition, we performed cavity ring down mass spectrometry to measure hydrogen and oxygen isotopes in the canal and calculate evaporation rates and determine how water residence time may impact the chemical composition. Finally, we tested the bioactivity of collected water using a zebrafish embryonic platform and assessed the hazard of annotated features using the EPA Hazard Cheminformatics Module which compiles experimental and in silico response data. Isotopic ratios and flow rates were used to explore the trends associated with evaporation in the canal and assess the effect of evaporation on the chemical profiles. We expected the chemical profiles to differ between the irrigation canal and river. We hypothesized that evaporation may be transforming and concentrating organic compounds in irrigation canals resulting in changes to the downstream canal chemical profile. These changes in the chemical profile may alter the toxicity of water contaminants and subsequent ecosystems health effects. The ultimate goal, shown in the conceptual diagram (Fig. 1), is to use these data streams to inform water management and risk assessment.

Fig. 1.

Conceptual model describing data generated and potential uses of these data streams for assessment of irrigation water quality monitoring and improvement.

2. Materials and methods

2.1. Sites and water collection

Surface water samples were collected from the Deschutes River and the Northern Unit Irrigation District (NUID) canal between Bend and Warm Springs, OR. The North Unit Irrigation District (NUID) serves approximately 60,000 productive agricultural acres and 970 users in Jefferson County OR (North Unit Irrigation District, 2023). Sampling was completed August 14–15, 2021, within a 24-hour period. Our study area was focused on the NUID canal, which is diverted from the Deschutes River in Bend, OR and flows north 105 km through the desert. Left over water not utilized for irrigation is returned to the Deschutes River north of Warm Springs, OR. There is a small holding reservoir in the canal 73 kilometer from the diversion. Sites were spaced equally in each system and designated based on distance and ease of access (Fig. 2). Public bridges were used as primary access points to enable collection from the middle of the water way, with water collected using a 5-gallon bucket. Water was divided for LC-HRMS organic chemical analysis, toxicity assessments, and stable isotope analysis. A summary of the collected samples and the analysis performed is represented in Table S1. Water for chemical analysis was collected at 11 sites, 5 in the canal and 6 in the river including one before the canal diversion in the river (represented as Start site). Six replicates were collected at each site for chemical analysis. Water was collected for the zebrafish analysis at 3 of the chemical collection sites. The one before the diversion and the last site in each system. For zebrafish analysis, we collected 4 replicates and pooled them into 2 samples. For stable isotope analysis, we collected 5 mL of water across 20 sites, which included those for chemical analysis and some additional sites for broader coverage. Twelve of the collected samples were used to estimate evaporation in the canal. Samples were stored on ice after collection and water for chemical analysis and zebrafish analysis was filtered through a 0.45 μm nylon filter and stored at 4°C until analysis.

Fig. 2.

Map of study region, sampled systems and sample types.

2.2. Standards and reagents information

Reagents and standards include water, methanol and formic acid were acquired from ThermoFisher Scientific (Waltham, MA, USA). Salicylic acid-D4 and Fluoxetine-D6 were acquired from Millipore Sigma (St. Louis, MO, USA). Verapamil-D3 was obtained from Cayman chemical (Ann Arbor, MI). Diclofenac-D4, DEET-D6, Imdiacloprid-D4 and 2,4,6-trichlorophenol-D2 were obtained from Cambridge isotopes (Andover, MA). DEET and 2-naphthalene sulfonic acid were acquired from HPC standards (Atlanta, GA, USA) and ThermoFisher Scientific (Waltham, MA, USA). respectively.

2.3. Stable isotope analysis

Samples for stable isotope analysis were collected in study sites along the river and canal in glass vials, sealed with Parafilm, and refrigerated until analysis as previously described in Jameel et al., (2016). Analysis of the ²H/¹H and ¹⁸O/¹⁶O isotope ratios of water were determined at the Stable Isotope Ratios for Environmental Research (SIRFER) facility at the University of Utah. Cavity Ring-Down Spectroscopy (CRDS; Picarro L2130-I, Santa Clara, CA) was used for sample measurements including corrections for memory and drift effects as outlined in Good et al., (2014). Sample values were reported in δ notation, where δ=R_sample/R_standard - 1, where R=²H/¹H or ¹⁸O/¹⁶O with V-SMOW as the water reference standard. Four injections of each sample were measured and averaged along with three laboratory reference waters for quality assurance and quality control. Analytical precision, calculated as the standard deviation of the mean calibrated EV values over a batch of samples is expected to be about 0.2 ‰ and 0.04 ‰; for δ²H and δ¹⁸O, respectively.

These permil values (‰) are the stable isotopic composition reported as a change in the enrichment or depletions relative to a standard composition in part per thousands (Jameel et al., 2016). Evaporation over inflow (E/I) ratios were estimated independently for each sample following previously published methods (Skrzypek et al., 2015):

$$\frac{E}{I}=\left[\frac{{\delta }_L-{\delta }_I}{\left({\delta }^{*}-{\delta }_L\right)\times m}\right]$$ (1)

where ${\delta }_L$ is isotopic composition of the water in the canal at a given sample point, ${\delta }_I$ is the is input water isotopic composition, ${\delta }^{*}$ is the limiting isotopic composition and m is the enrichment slope. The isotopic ratio of atmospheric moisture, needed for calculation of fractionation within the Craig-Gordon equation (Craig and Gordon, 1965) is considered to be in equilibrium with local precipitation isotope ratios according to the Online Isotopes in Precipitation Calculator (OPIC) following Brooks et al., (2014). The approach above (Eq. 1) is a steady-state approach that assumes inflows and water levels are relatively constant, with inputs balancing outputs (evaporation, outflows/tailwater, and seepage losses (Skrzypek et al., 2015). Evaporation effects on the isotope ratio are cumulative in nature, and thus estimated E/I values will be reflective of the total amount fraction of evaporation that water translating through the network has been subjected to relative to that of the source water (primarily the Deschutes River for the NUID). If a unit volume (V) of water within the canal is considered as plug-flow system, and all samples are assumed to be subjected to similar evaporation rates throughout the system, then the residence time of water within the network, τ = V/I = (E/I)*(V/E), will be proportional to the E/I ratio since V and E are assumed similar for all samples (Brooks et al., 2014). Note, because of other water inputs from tributaries during the course of the river, the calculation of E/I (and its proportionality to τ) violates the assumptions inherent in Eq. (1) and are not estimated.

2.4. Analysis of organic compounds

2.4.1. Water sample preparation

Chemical compounds in filtered water samples were preconcentrated using solid phase extraction (SPE). SPE was performed using an Oasis HLB LP 96 well plate (60 mg) from Waters (Milford, MA). The plate was placed over a collection plate on a manifold, conditioned with 1.5 mL methanol, and equilibrated with 1.5 mL LCMS grade water. A total of 4.5 mL of each water sample was loaded into the extraction plate in 3 consecutives 1.5 mL aliquots. Last aliquot was spiked pre-extraction with 10 μL of the internal standard mix 1 (Imidacloprid-D4, Diclofenac-D4, DEET-D6, and Verapimil-D3 each at 1 μg/mL). Vacuum was applied between each water aliquot. The cartridge was washed 4 times with 250 μL of water and allowed to dry for 5 minutes. Samples were eluted four times with 250 μL MeOH into a 1 mL glass collection vial (Waters, Milford, MA) nested in a 96 well collection plate. Samples were then transferred to 5 mL glass centrifuge tubes and evaporated to 300 μL using a Turbovap LV (Biotage, Uppsala, Sweden) automated blowdown apparatus with a gentle nitrogen stream, the water bath was kept at 30°C. Samples were filtered again with a 0.22 μm filter to limit potential matrix interferences. Samples were transferred to autosampler vials and spiked with 10 μL of the internal standard mix 2 (2,4,6-trichlorophenol-D2, Salicylic acid-D4 and Fluoxetine-D6, each at 1 μg/mL in mix) before LC-HRMS analysis. Internal standard mixtures were used for monitoring extraction recoveries (mix 1), and injection volume and instrument performance (mix 2). The spiked compounds chosen represented commonly reported chemicals in our environmental matrices across a few classes including as pesticides, pharmaceuticals, and personal care products.

2.4.2. UPLC-HRMS

Chromatographic separation or organic compounds was performed using a XBridge C18 column (2.1 × 50 mm, 3.5 μm, Waters, Milford, MA) with a Waters Cartridge Guard Column (XBridge BEH, C18, 3.5 μm, Waters, Milford, MA) in a Shimadzu Nexera ultra-high performance liquid chromatography (UHPLC) system. Samples were maintained at 10°C and the column oven was set to 25°C. Mobile phase A was 100% ultrapure water and mobile phase B was 100% methanol, both acidified with 0.1% formic acid. The linear gradient in percent B was 10% at 0 minutes, 50% at 4 minutes, 95% at 17 minutes and held at 95% until 25 minutes and then returned to 10% until 30 minutes. The flow rate was 0.3 mL/min, and the injection volume was 15 μL. Separation is followed by ionization and mass analysis using a high-resolution AB Sciex 5600 Triple Time of Flight mass spectrometer (HR-TOF-MS). Samples were run in data dependent acquisition (DDA) in both positive and negative electrospray ionization modes (ESI). Both MS1 and MS2 spectra were acquired for masses ranging from m/z 70–1000 Da for all samples, including extraction blanks, instrument blanks and pooled quality control samples. We ran standards using the same parameters for external validation.

2.4.3. Data processing and statistical analysis

A list of suspect compounds was generated from lists available in the NORMAN Suspect List Exchange and additional compounds of interest from the NIST mass spectral libraries (Mohammed Taha et al., 2022; NIST, 2023). A suspect list was created in SciexOS and added to the peak alignment and data curation processing in MS-DIAL. Raw data files were converted using Reifycs ABF converter (v 111, Tokyo, Japan). Files were then imported into MS-DIAL, version 4.0.7 (Tsugawa et al., 2015). Parameters for the MS-DIAL alignment are included in the Supplementary Information (excel file: Supplementary_Information.xlsx). Positive and negative ESI data was processed separately. Library matches were performed by comparing MSMS spectra from the MS-DIAL metabolomics MSP spectral kit version 17. Mass differences greater than 10 ppm between experimental and library mass were filtered out and not considered. Features were only considered if the relative abundance of the peak was above 1000. Blank subtraction was performed by filtering out features that were less than three times higher than the intensity of the blanks. Tentative matches were identified, and duplicates were removed based on those with less complete spectral information and with lower relative abundances. Positive and negative ESI modes were combined, and duplicate features were removed based on the quality of the peak and feature intensity. Features with peak intensity 3 or more times higher in the samples than the instrument blanks were kept. MetaboAnalyst version 5.0 was used to perform statistical analysis (Pang et al., 2022). Data was log transformed and pareto scaled (mean centered and divided by the square root of the standard deviation of each feature). Kruskal-Wallis One-Way Analysis of Variance was performed as a nonparametric test to determine which features were statistically different amongst sites. Hierarchical clustering analysis was utilized to explore relationships between sites and a heatmap of these chemical features was generated to describe the trends and patterns of chemical features amongst sites. Principal component analysis (PCA) was performed to describe the variance between sites. A covariate linear model and pearson’s correlation were performed to determine the relationship between E/I and the chemical features in the canal. Further data analysis was performed using R studio (v. 2022.02.1+461) (R Core Team, 2023). After compound annotation was completed, we utilized the EPA Cheminformatics Hazard Module to identify hazards of annotated xenobiotic compounds (US EPA, 2023). Very high, high and medium outcomes were scored, and hazard categories were summed to generate a hazard score. We used the compounds with high hazard scores to prioritize compounds. Categorization of compound use class was determined using PubChem.

2.5. In vivo zebrafish bioassays

In vivo toxicity assessments were performed with tropical 5D line zebrafish (Danio rerio) at the Oregon State University Sinnhuber Aquatic Research Laboratory (SARL, Corvallis OR) and husbandry protocols followed those outlined in Wilson et al., (2022). Assays were performed with raw water from three of the sampling locations without undergoing any concentration. 5 % embryo media (pH 7.3, μS 2500) was added to the water diluting the sample to 0.95 times concentration which adjusted the pH to 7.3 and conductivity to 2500 μS. The sites include the start location in Bend, OR and canal site 5 and river site 5. The chorion on the embryo was enzymatically removed following established procedures described by Mandrell et al., (2012). Embryos were then transferred into a 96-well plate containing 100 μL of 0.95x sample with embryo media with 32 embryos per treatment. Exposures began at approximately 4 hours post fertilization (hpf) (Truong et al., 2014).

Embryo photomotor response behavior (EPR) was assessed at 24 hpf using an automated photomotor response tool which records frames of each well as the embryos are exposed to two pulses of light with periods of darkness before, between and after (Reif et al., 2016). The larval photomotor response (LPR) behavior is assessed at 120 hpf with free swimming larvae in a 96 well-plate. The swim distance is assessed during a series of light dark cycles and total swim distance between each cycle is compared. The response to light stimulus at both periods indicates the perturbations to neuronal development from an exposure (Knecht et al., 2017). In addition to the behavioral assays, mortality and morphological endpoints were assessed. Mortality counts were performed at 24 hpf and 120 hpf. A total of 13 morphological endpoints were assessed.

3. Results

3.1. Stable isotope analysis

The ²H and ¹⁸O stable isotope ratios and evaporation over inflow (E/I) ratios increased from the beginning to the end of the canal, which could be an indication of hydrogeochemical processes leading to a concentration of heavier water isotopes. While the permil values (‰) of isotopes were linear, they were not unidirectional, with some upstream sites more enriched in heavy isotopes (Fig. 3a) than downstream sites. The correlation between the distance along the canal and E/I is 0.76. This positive correlation indicates an accumulation of heavy isotopes implying few inputs in the canal from other systems or precipitation. When comparing the evaporation over inflow (E/I) to the distance along the canal (Fig. 3b), the first six water samples had an E/I between 0.5 and 1. After that point, the values and the slope increased. This change in slope occurred after a small holding reservoir that irrigation water is temporally stored in. That change in E/I associated with the reservoir location, may indicate evapo-concentration of the stable isotopes at this location.

Fig. 3.

Stable isotope analysis values and evaporation percentage along the length of the canal. A. Change in stable O isotopes by change in stable H isotopes with global meteoric water line (GMWL) for reference. Average evaporation percentage calculated for canal sites. River E/I is not calculated due to variability caused by inputs from run off and other rivers. B. Distance from start of canal by average evaporation percentage (E/I) highlighting how E/I can be used as a proxy for residence time.

3.2. Mass spectrometry data analysis and statistics

Across all samples we extracted 55,286 aligned features in the positive ESI mode, and 13,316 aligned featured in the negative ESI mode. A principal component analysis shows that QC samples such as instrument blanks and internal standards cluster separately from pooled quality control samples and experimental sample data (Figure S1). From those, there were 303 and 140 reference matched features, respectively, with levels of confidence ranging from 2b to 5 (Schymanski et al., 2014). Final data curation of the reference matched features using retention time, accurate mass and MS2 spectra, resulted in a total of 398 unique features, after combining the positive and negative ESI modes. Details for these features including annotation confidence levels are reported in the document Supplementary_Information.xlsx

Overall, 60 % of the annotated compounds in the river and 78 % of the annotated compounds in the canal decreased in abundance from the beginning to the end of the river and canal. 29 % of the annotated features had the opposite direction of change in the river compared to the canal. The chemical profile of the river and canal was similar in upstream locations, where they are closely connected. The start site, upstream of where the canal water is removed from the river, clustered between the upstream river and canal groups in the PCA (Fig. 4a). In the downstream locations of the river and canal, the groupings of the chemical profiles separated from each other, and the start group was closer to the canal group (Fig. 4b). These differences are even more clearly shown in the heatmap (Fig. 5). Kruskal-Wallis one-way ANOVA indicated there were 106 significantly different compounds (26 %) between at least two groups. The heatmap (Fig. 5) shows a hierarchical clustering analysis of all the sites for the 106 significant features in the Kruskal Wallis Analysis. Geographically close sites were clustered together. The heat map shows that the samples from Start, River 1 and River 2 clustered together and have similar enrichment of compounds (Fig. 5). In addition, the samples from Canal sites 1, 2 and 3 clustered together (Fig. 5). We compared the log-2 fold change of features between the river and canal (Fig. 6) samples. The annotated compounds are represented in orange and all the chemical features in green. The annotated compounds showed a similar trend to the unannotated features with the latter having larger fold changes and more variability.

Fig. 4.

Principal component analysis for sites at the beginning and end of systems. a. Comparison of Canal 1 and River 1 and the Start site by PCA; b. Comparison of Canal 5 and River 5 and the Start site by PCA.

Fig. 5.

Heatmap showing the relative intensities of 106 significant features between sites. Hierarchical clustering performed with Euclidean distances and Ward’s linkage to show relationship between groups.

Fig. 6.

Comparison of annotated features and all chemical features detected in the river and canal. Log-2 fold change of annotated compounds in orange and all chemical features detected in green.

Identified compound use classes included natural products and dietary supplements, food additives, personal care products, industrial products, pharmaceuticals, illicit drugs, pesticides, and transformation products from those listed above. Natural products made up 63 % of the annotated compounds. A covariate linear model was performed to explore the influence of the evaporation in the canal and the chemical profile. There are 106 significant features between sites as observed in the Kruskal Wallis analysis. When adjusting for evaporation, there were no additional significant features. Pearson correlation identified features that are most correlated to E/I (Fig. 7). Of the top 25 correlated features, 19 are natural products.

Fig. 7.

Top 25 chemical features correlated with E/I. Features with green stars are natural products.

3.3. In vivo toxicity assessments

To investigate potential toxic effects in vivo, we screened collected water in the high-throughput embryonic zebrafish platform. There were no significant morphological changes observed at either the 24 or 120 hpf time points (Figure S3). Mortality was observed at both time points, though results were not statistically significant when comparing treated animals with non-treated controls (Figure S3). None of the water samples showed significantly different behavior in the embryo photomotor response compared to the negative control animals. In the larval photomotor response assay, larvae exposed to water from all three test sites showed significant (p < 0.001) hyperactivate swimming behavior during the dark phase of the cycle (Figure S4).

3.4. Hazards assessment

The EPA Hazard Cheminformatics Module provides a summary of data to assess the hazard of large data sets. We performed the hazard assessment with only the xenobiotic compounds annotated in the data. Of the 152 xenobiotics, the module had records for 116 compounds (Figure S2). Fourteen compounds that had high hazard scores were prioritized for future monitoring (Table 1). These included many pesticides and industrial products. Aquatic toxicity was the most commonly observed effect and was reported in 58% of the compounds. Of the 13 prioritized compounds 5 were most abundant at site 2 in the river. The rest of the prioritized compounds were distributed amongst other sites in the river and canal. The presence of DEET and 2-naphthalene sulfonic acid were confirmed by analytical reference standards (Figures S5 and S6).

Table 1.

Prioritized compounds based on abundance and hazard endpoints from the EPA Hazard Cheminformatics Module.

CAS number	Compound Name	Use Classes	Annotation Confidence Level (Schymanski et al., 2014)	Most Enriched Site	High Hazard Categories
834–12–8	Ametryn	Pesticide	Level 3	Start	Acute Aquatic Toxicity, Chronic Aquatic Toxicity
1593–77–7	Dodemorph	Pesticide	Level 3	Start	Skin Irritation, Acute Aquatic Toxicity, Chronic Aquatic Toxicity
69–72–7	Salicylic acid	Personal care product	Level 4	River 5	Genotoxicity Mutagenicity, Eye Irritation
120–18–3	2-naphthalene sulfonic acid	Industrial Product	Level 2b	River 2	Bioaccumulation, Acute Aquatic Toxicity, Chronic Aquatic Toxicity
732–26–3	2,4,6-Tris(tert-butyl) phenol	Industrial Product	Level 3	River 2	Acute Aquatic Toxicity, Chronic Aquatic Toxicity
116–06–3	Aldicarb	Pesticide	Level 5	River 2	Oral Toxicity, Inhalation Toxicity, Acute Aquatic Toxicity, Chronic Aquatic Toxicity
84–65–1	Anthraquinone	Industrial Product	Level 3	River 2	Carcinogenicity, Genotoxicity Mutagenicity, Acute Aquatic Toxicity, Chronic Aquatic Toxicity
126–73–8	Tributyl phosphate	Industrial Product	Level 4	River 2	Inhalation Toxicity, Carcinogenicity, Eye Irritation
134–62–3	Diethyltoluamide (DEET)	Personal care product, Pesticide	Level 2b	River 1	Oral Toxicity, Skin Irritation, Carcinogenicity, Acute Aquatic Toxicity, Chronic Aquatic Toxicity
112–18–5	N,N-Dimethyldodecan–1-amine	Personal care product	Level 3	River 1	Skin Irritation, Eye Irritation, Chronic Aquatic Toxicity
65–85–0	Benzoic acid	Food Additive	Level 2b	Canal 3	Genotoxicity Mutagenicity, Eye Irritation, Exposure Risk
10453–86–8	Resmethrin	Pesticide	Level 3	Canal 3	Acute Aquatic Toxicity, Chronic Aquatic Toxicity
55406–53–6	Iodocarb	Pesticide, Preservative	Level 2b	Canal 2	Genotoxicity Mutagenicity, Eye Irritation, Acute Aquatic Toxicity, Chronic Aquatic Toxicity

4. Discussion

In this study, we tested a novel water quality assessment paradigm which included stable isotope analysis to measure evaporation as a proxy for in canal residence time, an initial in vivo screening of collected water, a comprehensive UPLC-HRMS organic chemical analysis, and in silico toxicity predictions of annotated features using the EPA Hazard Cheminformatics Module. Subsequently, compounds were prioritized based on occurrence, relative abundance, and predicted toxicity.

4.1. Assessing evaporation with stable isotope analysis

The use of stable isotopes to assess evaporative loss in surface and ground water is widely described (Al-Oqaili et al., 2020; Chen and Tian, 2021; Koeniger et al., 2021; Simpson and Herczeg, 1991). Quantifying evaporation in irrigation networks allows for identification of inefficiencies and assists in the establishment of management strategies to minimize losses (Kulkarni and Nagarajan, 2019; Steinfeld et al., 2020). Novel approaches for quantify evaporation in canals such as software packages to better assess isotope data, imagery-based models, and in situ automatic sampling systems have been proposed (Heinz et al., 2014; Kulkarni and Nagarajan, 2019; Skrzypek et al., 2015). Measuring evaporation rates in irrigation networks allows for correlations between hydrologic conditions of the canal and other measured factors such as the organic chemical composition of water samples (Koeniger et al., 2021). E/I values can also be a simple proxy for water residence time within the canal. Water residence time can depend on the season and catchment characteristics (Zhou et al., 2021). Water residence time can provide additional insights into hydrological processes and chemical movement in the system.

In this study we collected samples from more locations for stable isotope analysis than for chemical and bioactivity analysis to better capture the evaporation trends occurring in the system. The correlation between the isotopic trends and the evaporation rate in the river are difficult to determine because of the many water inputs in and out of the system throughout the watershed (Skrzypek et al., 2015). Rivers tend to have many inputs limiting the interpretation of E/I values in those systems. Alternatively, irrigation canals in this area tend to have minimal inputs after the initial diversion, which favors our ability to accurately determine the correlation between the stable isotope ratio and the evaporation trends. The largest changes in the E/I values occurred at a holding reservoir at the south end of the district. Evaporative loss from reservoirs is a known issue for water managers in arid regions and it is estimated to increase along with increased solar radiation (Friedrich et al., 2018; G. Zhao and Gao, 2019). Samples were collected in August 2021 during a particularly dry and hot year (White et al., 2023). This led to the reduction and cessation of allocations to water users several months before the standard water shut off date. Examples such as this highlight the need for new technologies to reduce evaporative loss such as solar panels near or over irrigation systems and changes to management strategies have been proposed to limit water loss especially in warming climates (El-Nashar and Elyamany, 2023; McKuin et al., 2021; Muñoz-Cerón et al., 2023). Describing evaporation and water residence time with stable isotopes can support our understanding of hydrologic processes which may impact contaminant dynamics and water quality in irrigation canals.

4.2. UPLC-HRMS analysis to identify organic compounds

High resolution mass spectrometry can be used to assess contaminants in urban and semi-urban surface water (Albergamo et al., 2019; Picó et al., 2020; Zhao et al., 2022). These methods can help to identify sources of pollutants and assess whether a water source will be safe for consumers (Du et al., 2020; Margenat et al., 2017). However, data analysis is limited to available spectral libraries which often do not contain transformation products and emerging contaminants (Schulze et al., 2020). Despite these limitations, we can use our data to explore the variability in the chemical fingerprint (features and abundance) at each site and prioritize identified compounds for further testing and confirmation. More compounds decreased in abundance in the canal than in the river which may indicate that components of the canal aids in degradation of chemical compounds. We used principal component analysis to explore how the chemical fingerprint varied between the two systems. We observed that the start group was between the clusters for site 1 in the canal and river which reflects the systems being recently connected at the start (Fig. 4a). At the end of the canal and river the start site group is closer to the canal (Fig. 4b) which may be due to the canal not having any inputs through the system after it diverges near the start site and the river having many inputs over the stretch we measured. We observed that most of the annotated compounds decreased in abundance from the beginning to the end of both systems. There were 106 compounds statistically significant between at least two sites as described by the Kruskal-Wallis test. Using hierarchical clustering (Fig. 5), we explored the relationship between sites and the relative intensity of features between the sites for the significantly different compounds observed in the Kruskal-Wallis test. The first two river sites and the upstream start site, which was in the river as well, clustered together. The first three canal sites formed a cluster, and different chemical features were enriched in the canal sites compared to the upstream river sites. Other studies have shown that hierarchical clustering analysis can be used to identify chemical features which may be predictive of water source and region (Du et al., 2020). We hypothesized that canal sites might be more concentrated in organic chemical compounds due to evaporation particularly given that sampling occurred in the driest and warmest part of the growing season. These results indicate that did not occur, in fact downstream canal sites were less enriched in compounds overall which may be explained by degradation or movement out of the system during irrigation processes or into the groundwater. The river was more enriched in annotated compounds overall compared to the canal. This could be driven by higher abundance of natural products from a more diverse flora and fauna in the river system (Pietra, 2002; Shi et al., 2023). Alternatively, point source introductions of compounds could be contributing to differences within and between the systems (Ruff et al., 2015).

To further understand the potential source of the annotated compounds, we categorized the compounds by use class. Over 60% of the annotated compounds were natural products. Many of these compounds are derived from plants and microbial metabolism in the environment but some may be anthropogenically introduced. For example, carnosol is found in plants including sage of which there is an abundance in the region but kahweol is a coffee specific diterpenoid which is likely anthropogenically introduced (Cavin et al., 2002; Jassbi et al., 2016). Other annotated compound classes include food additives like stevia and personal care products like salicylic acid. The annotated compounds make up a small proportion of the total detected features. In order to compare the two data sets, we plotted the log-2 fold change of the river and canal for the annotated and unannotated data in the same plot (Fig. 6). The trends are similar for both the annotated and unannotated data sets but there are greater overall changes in the unannotated data. The annotated compounds may adequately describe the global trends between the systems. However, we cannot determine if the annotated features accurately reflect the distribution of compound classes present in the unannotated data. Nontarget LC-HRMS approaches produce large multivariate datasets and require prioritization and reduction methods must be used to aid in interpretation of the results (Hollender et al., 2017). The cheminformatic module from the EPA is a helpful tool to prioritize features for testing and monitoring. Several chemicals associated with industrial processes, agricultural practices and personal care products were shown to be most hazardous by having a higher toxicity risk reported across several endpoints (Table 1). Many of the prioritized features were most abundant at site 2 in the river. This site is at popular state park and is downstream from the town of Bend Oregon, the only major population area the river ran through in this study. This increase in priority chemical features is likely due to the urban influence of Bend (Liu et al., 2023; Riva et al., 2019). Many of the prioritized pesticides are not currently registered for sale in the United States. However, the use of unregistered pesticides does occur with previously purchased chemicals. We confirmed DEET and 2-naphthalene sulfonic acid by reference standard. Spectra are available in the supplementary figures, S5 and S6. These compounds are well described, persistent, environmental contaminants (Pan et al., 2008; Weeks et al., 2012). Compounds shown to elicit carcinogenicity and genotoxicity should be assessed at environmental concentrations and in mixtures to determine the risk of their combined effects in the environment.

4.3. In vivo screening of collected water using the zebrafish model

The zebrafish model is a robust tool for assessing bioactivity (Rericha et al., 2021; Wilken et al., 2020). Investigators typically model complex health disorders in zebrafish (Axton et al., 2019; García-Jaramillo et al., 2020; Shams et al., 2018). Scientists increasingly take advantage of standardized high-throughput behavioral performance assays to conduct behavioral neuroscience research with zebrafish (Bailey et al., 2013; Bugel et al., 2014; Shams et al., 2018; Stewart et al., 2014). Zebrafish studies provide important insights into human biology. If a chemical is bioactive in zebrafish, the probability is ~70% that it will also be bioactive in rodents and mammals (Wiley et al., 2017). The initial screening showed bioactivity in the zebrafish model across the three sites tested. All sites showed significant hyperactivity in the larval photomotor response (LPR) according to methods outlined in Knecht et al., (2017). Despite the limited results, observed bioactivity of non-concentrated water is a meaningful corroboration of the presence of contaminants or bioactive natural products that should be prioritized for further interrogation (Shao et al., 2019). Observed effects may be an indicator of the presence of bioactive chemicals and can help with the prioritization of chemical compounds and sampling sites for further investigation based on previously reported effects (Allan et al., 2012). From the prioritized compound, only ametryn and resmithrin were found to be more abundant at the start of the river and canal than at any other studied location. While prioritization of suspect compounds is important for hypothesis generation it is likely that the bioactivity can be attributed to a mixture of chemicals rather than one compound (Altenburger et al., 2019; Bradley et al., 2021).

4.4. Linking evaporation and organic chemical fingerprints

After assessing separately, the isotope data and the presence of organic chemical features in the canal, we performed a linear model to explore the relationship between the two data sets. There were no additional significant features between groups when accounting for E/I. The Pearson’s correlation indicated that most features correlated to E/I were natural products (Fig. 7). This suggests that compounds from flora and fauna within the system might be the primary features driving trends in the canal. The relatively low volume of water in the canal could also contribute to the greater concentration of natural products in the system. Resmethrin, a prioritized pesticide from the hazard module was positively correlated with E/I and should be further monitored in the irrigation network. Often, target and nontarget water screening studies focus on anthropogenically derived compounds or specific compound classes such as organophosphates or pesticides (Gong et al., 2022; Huang et al., 2021). However, to fully understand the fate of chemicals in the system and use this data to forecast the behavior of relevant contaminants for management decisions, we benefit from measuring a broader chemical space. Natural substances make up a large portion of compounds in environmental samples and their movement can be predictive of the behavior of xenobiotic compounds (Gros et al., 2021; Qian et al., 2021). The bioactivity of natural substances is also important to consider as they contribute to the complexity of chemical mixtures present in the environment. Additionally, the bioactivity of contaminants may be synergized or antagonized by natural compounds present in the water (Rosenmai et al., 2018). Variable trends of the observed chemical features can uncover potential physical or hydrologic parameters that may influence the chemical profile and inform the potential bioactivity throughout the canal.

5. Conclusions

Monitoring organic contaminants in water distribution systems is of critical importance for the sustainability of agricultural water resources, ensuring crop yield, and environmental and human health. Broad monitoring efforts can also help to identify and mitigate sources of pollution. As novel strategies for maintaining water quality are developed, we propose the implementation of recurrent, reliable, and robust water monitoring programs using UPLC-HRMS and suspect and non-target screening approaches. Mass spectrometry analysis are still relatively expensive to operate and requires trained personnel which adds to the challenges associated with implementation in monitoring programs. Complementary to the identification of contaminants, in vivo toxicity data generated for collected water samples is shown to be a fast and robust platform for initial testing biological effects. Our long-term goal is to develop a predictive framework able to forecast the presence and degradation of relevant chemical contaminants in irrigation networks with different physical and morphological attributes.

Data Availability

Data will be made available on request.