Multi-omic responses of fish exposed to complex chemical mixtures in the Shenandoah River watershed

Abstract

To evaluate relationships between different anthropogenic impacts, contaminant occurrence, and fish health, we conducted in situ fish exposures across the Shenandoah River watershed at five sites with different land use. Exposure water was analyzed for over 500 chemical constituents, and organismal, metabolomic, and transcriptomic endpoints were measured in fathead minnows. Adverse reproductive outcomes were observed in fish exposed in the upper watershed at both wastewater treatment plant (WWTP) effluent- and agriculture-impacted sites, including decreased gonadosomatic index and altered secondary sex characteristics. This was accompanied with increased mortality at the site most impacted by agricultural activities. Molecular biomarkers of estrogen exposure were unchanged and consistent with low or non-detectable concentrations of common estrogens, indicating that alternative mechanisms were involved in organismal adverse outcomes. Hepatic metabolomic and transcriptomic profiles were altered in a site-specific manner, consistent with variation in land use and contaminant profiles. Integrated biomarker response data were useful for evaluating mechanistic linkages between contaminants and adverse outcomes, suggesting that reproductive endocrine disruption, altered lipid processes, and immunosuppression may have been involved in these organismal impacts. This study demonstrated linkages between human-impact, contaminant occurrence, and exposure effects in the Shenandoah River watershed and showed increased risk of adverse outcomes in fathead minnows exposed to complex mixtures at sites impacted by municipal wastewater discharges and agricultural practices.

Graphical Abstract

1. Introduction

Surface waters globally are impacted by anthropogenic contaminants (Focazio et al., 2008; Kolpin et al., 2002; Bradley et al., 2017) including pharmaceuticals (Daughton, 2001), pesticides (Kolpin et al., 2013), hormones (Ternes et al., 1999; Lange et al., 2002), and industrial chemicals (Fromme et al., 2002; Blackburn and Waldock, 1995; Suja et al., 2009). Many individual contaminants are known to cause adverse outcomes in aquatic organisms, including abnormal growth and development (Hayes et al., 2010; Vajda and Norris, 2011), altered behavior (Painter et al., 2009), reproductive disruption (Schwindt et al., 2014; McGree et al., 2010), and population collapse (Kidd et al., 2007). However, humans and wildlife encounter contaminants as components of complex mixtures. Effects from mixture exposure can deviate from predicted responses to individual constituents due to additive or antagonistic interactions (Kortenkamp, 2007; Altenburger et al., 2012). A better understanding of the occurrence and biological consequences of environmental mixtures and the pathways through which they cause adverse effects is needed to protect human and ecosystem health.

Assessing the risks associated with environmental mixture exposures presents several challenges. There are thousands of potential chemical contaminants, many with limited existing toxicity data. This makes chemical monitoring alone insufficient for mixture risk assessment, despite significant improvements in target and non-target analytical chemistry. In addition, the last several decades of ecotoxicity research have revealed a variety of sublethal exposure effects that still represent risks to individual health and population sustainability (Ankley and Edwards, 2018). Given these challenges, predicting effects a priori and choosing relevant endpoints to measure is difficult, as important biological perturbations may be missed by monitoring only endpoints of overt toxicity. Furthermore, the composition and concentration of chemical mixtures is subject to continuous variability, both temporally, as human activity and hydrology change, and spatially, as organisms move through habitats.

To address these challenges, we conducted integrated site assessments, employing broad chemical monitoring combined with multi-omic, cellular, and organismal endpoints to assess the effects of exposure to environmental contaminant mixtures. This approach provides comprehensive data on the effects of realistic environmental contaminant mixtures using completely in vivo endpoints. Omics data can be especially valuable in complex mixture assessment because they provide holistic information for a broad range of molecular biomarkers and pathways that may be mechanistically linked to chemical exposure. This information is useful for identifying non-lethal effects of exposure and for interpreting exposure effects using known or hypothesized adverse outcome pathways (AOPs) (Ankley and Edwards, 2018; Ankley et al., 2010). Combining omics with broad chemical monitoring and measurement of cellular and organismal endpoints provides data at multiple points along an AOP, from a molecular initiating event (MIE) to the adverse outcome itself. Through this approach, intermediate key events, such as changes in transcript and metabolite abundance, may serve as early responding biomarkers for both “upstream” molecular initiating events and “downstream” adverse outcomes. Thus, omics data may be useful for identifying specific chemicals or sub-mixtures with related mechanisms of action that contribute to adverse organismal or population level outcomes.

Transcriptomic and metabolomic profiling have been used to assess exposure risks from wastewater treatment plant (WWTP) effluents (Berninger et al., 2014; Garcia-Reyero et al., 2008; Martinović-Weigelt et al., 2014; Bahamonde et al., 2015; Garcia-Reyero et al., 2011; Ekman et al., 2018; Davis et al., 2016), pulp mill effluents (Costigan et al., 2012; Davis et al., 2013), urban streams (Ekman et al., 2018; Rodriguez-Jorquera et al., 2015), and agriculturally impacted waters (Sellin Jeffries et al., 2012; Hook et al., 2017). However, these studies typically focused on the effects of a single source of environmental contaminant mixtures (i.e., agricultural runoff, WWTP effluent, etc.), often at highly impacted sites, whereas the majority of riverine habitats lie along a gradient between highly-impacted and pristine and are contaminated with diverse classes of chemicals originating from multiple point and non-point sources (Daughton, 2001; Neumann et al., 2002). To fully address the risk posed by complex environmental contaminant mixtures, effects assessments are needed in habitats that are moderately impacted by multiple human activities.

To help fill in these gaps surrounding complex mixture exposures, we deployed in situ fish exposure laboratories to sites throughout the Shenandoah River watershed (SRW) in Virginia, USA in 2014 and 2015 (Fig. 1). The SRW contains a diversity of land uses, including mixed urban and built-up land, agricultural, and forested areas. The region has a history of fish health issues that have been, in part, associated with pollution and endocrine disruption (Blazer et al., 2010; Blazer et al., 2012). Fish exposure experiments were conducted using fathead minnows (FHM; Pimephales promelas) at sites with a gradient of chemical impacts originating from wastewater effluents, agricultural land use, and/or urban and built-up land.

Fig. 1.

Map of the Shenandoah River watershed showing locations of mobile laboratory exposure units and surrounding land use characteristics. The vast majority of low herbaceous vegetation is agricultural, either row-crop or pasture. [PC = Passage Creek (reference), NRA = North River above wastewater treatment plant (agricultural), NRW = North River wastewater (wastewater treatment plant), SFSR = South Fork Shenandoah River (mixed-use), NFSRB2 = North Fork Shenandoah River (mixed-use). NRA and NRW are separated by a few hundred meters and thus share a yellow square on this map.]

We hypothesized that contaminant profiles would vary between locations with different land use, and that exposure to these mixtures would result in site-specific biological responses. These multi-site assessments provide a broad characterization of exposure effects across a variety of land uses, a necessary step in expanding our understanding of risks from complex mixtures to broader spatial scales. Additionally, this integrated approach offers the opportunity to identify potential contaminants and/or MIE responsible for adverse outcomes. A companion paper presents the hydrological and chemical contaminant results and watershed-scale risk assessment (Barber et al., 2022); here we focus on the biological effects of exposure.

2. Methods

Five sites across the SRW with varying land use were selected for integrated landscape, contaminant, and biological characterization, including the on-site deployment of mobile fish exposure laboratories (Fig. 1). Exposures were conducted as two separate experiments in August 2014 and May–June 2015, with some sites being sampled in both years and other sites only being sampled in one year (Fig. S1). Flow conditions were lower during the August experiments than during May–June. Mobile laboratories were deployed at a forested reference site at Passage Creek (PC; 2014 and 2015), mixed-use sites (agriculture, developed land, and forests) in large watersheds including the South Fork Shenandoah River (SFSR; 2014 and 2015) and the North Fork Shenandoah River (NFSRB2; 2015), and two sites in the upper watershed including a WWTP-dominated site, North River wastewater (NRW; 2014), and an agriculture-impacted site (poultry houses and row-crops) on the North River above the WWTP (NRA; 2014). A limited number of exposure units and access to sites prevented sampling of all sites in both years. Detailed land use metrics for these sites are presented in Table S1.

2.1. Chemical monitoring

Grab samples were collected weekly at each site from within the mobile laboratories and analyzed individually for 534 chemical constituents, including trace elements, pharmaceuticals, hormones, phytoestrogens, and pesticides. In addition to grab samples, basic water quality was monitored daily in exposure aquaria using a YSI Professional Plus multiparameter meter (YSI, Yellow Springs, OH, USA), including measurements of temperature, pH, oxygen saturation, ammonium, nitrate, and specific conductance. Interpretation of the chemical results and the complete analytical methodologies and dataset are presented elsewhere (Barber et al., 2022; Barber et al., 2019).

2.2. Mobile laboratory exposure and sampling

Fish exposure experiments were conducted with reproductively mature male and female FHM using flow-through mobile exposure laboratories in August–September 2014 (PC, SFSR, NRW, NRA) and May–June 2015 (PC, SFSR, NFSRB2). Water was continuously pumped from the stream to stainless steel storage tanks at on-site mobile laboratories. The source water then flowed by gravity to stainless steel splitter tanks, where it was heated to 25 ± 1 °C, and then distributed to 10 L glass aquaria housing the fish. All surfaces in contact with exposure water were glass, stainless steel, or Teflon^®. Adult male and female fathead minnows (FHM) were provided by the U.S. Geological Survey Columbia Environmental Research Center (CERC). Animal care and handling was in accordance with the Institutional Animal Care and Use Committee of the University of Colorado Denver (Protocol # 00698). Fish were randomly divided into exposure aquaria with 5 males or females per aquarium. Water was aerated in exposure aquaria (∼100 bubbles/min, >90 % saturation). Continuous water flow to each aquarium was maintained at ∼200 mL min⁻¹. Fish were fed laboratory fish chow provided by CERC. The photoperiod was maintained at 14-h light:10-h dark.

To assess the starting condition of organisms, an initial control cohort (IC; n = 10/sex) was sampled on day-0 of each experiment (2014 and 2015), before exposure to stream water began. The IC fish were only exposed to CERC water. Stream-water exposed FHM were necropsied on day-7 and day-21 of exposure (n = 10 fish/site/sex from 2 or 3 separate aquaria). Tissues from IC fish and stream-water exposed fish were analyzed for the same endpoints. Fish were anesthetized with 100 mg L⁻¹ tricaine methanesulfonate. Fish weight (0.1 g) and total length (mm) were recorded. Nuptial tubercles were counted for male fish (Smith, 1974; Vajda et al., 2011). For both male and female fish, blood was collected from the caudal vein into heparinized tubes and stored on ice for <4 h until it was centrifuged. The plasma fraction was collected, frozen on dry ice and stored at −80 °C. Livers were dissected, weighed (0.001 g), and divided into three equal fragments for different preservation methods. For this study, two liver fragments were separately snap frozen on dry ice, and then stored at −80 °C for microarray and metabolomic analysis. Gonads were dissected, weighed and preserved in 10 % neutral-buffered formalin until processed for paraffin histology using standard methods (Presnell, 1997).

2.3. Organismal, cellular, and individual molecular endpoints

Condition factor, secondary sex characteristics, gonadosomatic index (GSI), and gonadal histology were evaluated in day-0 and day-21 fish. Plasma vitellogenin protein (Vtg) and hepatic vitellogenin mRNA (vtg) were measured in all sampled fish (day-0, day-7, day-21). Liver metabolomic profiles were evaluated in 2014 for day-7 and day-21 fish (n = 9–10/site/exposure duration/sex). Liver transcriptomic responses were evaluated in a subset of day-7 male fish from 2014 and 2015 (n = 7–8/site). For 2014, transcriptomic responses were measured from a subset of the same individuals used for metabolomics, using different liver fragments.

Condition factor was calculated based on fish wet weight and total length [(weight (g)/length³ (mm)) × 100]. Plasma Vtg concentration was measured by enzyme-linked immunosorbent assay (ELISA) using FHM-specific kits according to the manufacturer’s instructions (Biosense, Bergen, Norway). GSI was calculated [(gonad weight (g)/body weight (g)) × 100]. For gonadal histology, ten cross-sections from each gonad were evaluated for histopathological abnormalities (Blazer, 2002) and indexed on the amount of mature sperm present or predominant ovarian stages as described previously (Pawlowski et al., 2004).

2.4. Liver metabolomics and transcriptomics

Endogenous hepatic metabolite profiles were measured by liquid chromatography/tandem mass spectrometry (LC-MS/MS) for male and female fish exposed in 2014 for both 7- and 21-day cohorts. Metabolites were extracted using a dual phase method as reported previously (Davis et al., 2013). Relative intensities of endogenous metabolite features were determined using an Accela 1250 UHPLC system coupled to a Q-Exactive LC-MS/MS (Thermo Fisher Scientific Inc., Waltham, MA, USA) via a heated electrospray ionization source (HESI-II) operating in positive and negative modes.

RNA was extracted from livers using Qiagen RNeasy plus mini kits according to the manufacturer’s instructions (Qiagen LLC, Germantown, MD, USA). RNA quality was measured with an Agilent 2100 Bioanalyzer (Agilent Technologies, Inc., Santa Clara, CA, USA). RNA integrity numbers were above 8 for all samples (mean = 9.48). Transcriptome profiles were measured from a subset (n = 7 or 8 per site) of the 7-day male cohort from 2014 and 2015 using custom FHM Agilent 60K one-color microarrays (https://www.ncbi.nlm.nih.gov/geo; platform accession: GPL15775; Agilent Technologies, Inc., Santa Clara, CA, USA). The number of samples analyzed by microarray was limited by cost. To further investigate estrogenic mechanisms of action implicated in health of fish in the Shenandoah River (Blazer et al., 2012; Blazer et al., 2007), we focused our transcriptomic analysis on male fish. We chose to measure transcriptomic responses from the day-7 exposure to measure the initial response of exposure to complex environmental mixtures instead of the delayed or compensatory responses from the day-21 exposure (Ankley et al., 2012). Technical details for metabolomic and transcriptomic analysis are provided in the SI.

2.5. Data analysis and bioinformatics

Chemical occurrence data were used to characterize contaminant impacts at different sites and to evaluate biological impacts using the AOP framework. Hierarchical clustering analysis was used to compare patterns of chemical occurrence. Average-linkage hierarchical clustering was performed separately on organic and inorganic data. Due to low-detection frequencies, organic constituents were analyzed separately from inorganic and clustered using Jaccard distance based on detect/non-detect data. Z-score standardized concentrations for a subset of inorganic trace elements with high detection frequencies (>90 %) were clustered using Euclidean distance. Uncertainty in the clustering was evaluated using approximately unbiased bootstrapped probability values (AU p), calculated in the pvclust package (Suzuki and Shimodaira, 2006) in R (R Core Team, 2017). Clusters were considered moderately supported with AU p-values between 80 and 95, and strongly supported with AU p-values > 95.

For biological endpoints, continuous variables, including condition factor, gonadosomatic index, plasma Vtg, vtg mRNA (see below), and GSI were analyzed by one-way ANOVA, followed by Tukey’s HSD for pairwise comparisons. Plasma Vtg and hepatic vtg mRNA were log transformed prior to ANOVA. One-way ANOVA was selected for plasma Vtg and hepatic vtg analysis so that IC cohort could be included in the comparisons. Count and ordinal data for nuptial tubercle number, sperm abundance index, and ovarian stage index were analyzed by Kruskal-Wallis tests, followed by pairwise Dunn test with Bonferroni correction. For condition factor, GSI, sperm abundance, and ovarian stage, comparisons were made between initial controls and the 21-day exposure cohort, as shorter duration exposures were not expected to affect these slower changing endpoints. For plasma Vtg and vtg mRNA, both the 7- and 21-day cohorts were analyzed separately.

For organismal, cellular, and vitellogenin endpoints (plasma Vtg and vtg mRNA), full pairwise comparisons between all exposure sites and IC fish were performed and are reported in the figures. In the interest of clarity and brevity, we focus our discussion on differences between impacted sites and both the IC cohort and the PC reference site. Although subject to different treatment and handling than mobile laboratory exposed fish, the IC cohort provides important information on the initial condition of fish, especially for organismal endpoints that are unlikely to change over short time periods. For omics endpoints, the PC reference site was used as the control comparison site to avoid reducing statistical power from multiple testing corrections and because it represents the most relevant site for comparison when evaluating effects from mixture exposures.

For metabolomic data, differential expression of individual metabolites was determined by two-tailed t-test, comparing NRA, NRW, and SFSR to the PC reference site. The resulting p-values were corrected for multiple hypothesis testing using the false discovery rate (FDR) approach. Further analysis was limited to metabolites that exhibited an FDR-corrected p-value < 0.05 for at least one comparison for a given sex and exposure duration. Pathway analyses were performed using the open source Python program mummichog (version 1.0.5) (Li et al., 2013). We used partial least squares (PLS) regression (SIMCA-13.0, Sartorius AG, Göttingen, Germany) to compare endogenous metabolite profiles for the 21-day exposures with organic contaminant concentrations. This modeling approach, as described in the SI, has previously been used to identify environmental stressors that have a significant association with metabolomic responses (Davis et al., 2016; Collette et al., 2019). Metabolomics data are available in the U.S. Environmental Protection Agency Science Hub database (https://catalog.data.gov/organization/epagov; doi:10.23719/1529242).

For transcriptomic data, one-way ANOVA was used to identify differentially expressed genes (DEGs) by comparing transcript abundance levels at NRA, NRW, SFSR, and NFSRB2 to the PC reference site within each year. Probes with an FDR-corrected p-value < 0.05 were considered differentially expressed. ANOVA was performed in JMP Genomics (SAS Institute, V6.0). A subset of contaminant-responsive genes was selected a priori for focused analysis (Table S2). Differential abundance was determined by ANOVA, followed by Tukey’s HSD for pairwise comparisons. Gene set enrichment analysis (Subramanian et al., 2005) (GSEA) and sub-network enrichment analysis (SNEA) were performed in Pathway Studio 9.0 (Elsevier) using the ResNet 10.0 database (Nikitin et al., 2003). GSEA was performed on gene sets in the categories: cell signaling, receptor signaling, cell process pathways, and metabolic pathways. SNEA was conducted using cellular processes as the seeds for sub-network construction. Average linkage hierarchical clustering analysis was performed to identify patterns in transcript abundance. AU p-values were calculated in the same manner as chemical clustering. For microarray verification, vtg and a subset of genes showing differential transcript abundance were analyzed by qPCR. Primers were designed using Primer-BLAST (Table S3, www.ncbi.nlm.nih.gov) and abundance was measured using a SYBRgreen qPCR protocol (SI, Extended Methods). Microarray data were uploaded to the Gene Expression Omnibus (https://www.ncbi.nlm.nih.gov/geo, Series GSE111967).

3. Results and discussion

3.1. Contaminant overview

Of the 534 measured chemical constituents, over 200 were detected in exposure waters in at least one grab sample at one or more sites (Barber et al., 2022; Barber et al., 2019). Many of these detections were of nutrients, inorganic trace elements, or natural organic compounds. To help gauge the level of potential human impact at each site, the number of unambiguously anthropogenic organic contaminants detected at each site was calculated (Bradley et al., 2017). This includes synthetic chemicals that are almost certain to have anthropogenic origin such as pharmaceuticals, herbicides, insecticides, and fungicides, but excludes steroid hormones, phytoestrogens, nutrients, and inorganic compounds. There were 53 unambiguously anthropogenic organic contaminants detected at one or more sites in 2014 and 48 detected in 2015. Concentrations were generally in the ng L⁻¹ range, but occasionally reached the low μg L⁻¹ range.

In 2014, the average number of unambiguously anthropogenic organic contaminants detected (across repeat samplings, n = 4) was highest at NRW (wastewater) at 33, followed by SFSR (mixed) at 17, NRA (agriculture) at 9, and PC (forested) at 4. NRW averaged 7 pesticides or degradants per sampling, 17 pharmaceuticals, 4 other consumer or industrial chemicals, and 9 halogenated disinfection byproducts. NRA averaged 4 pesticides, 3 pharmaceuticals, 1 other consumer or industrial chemicals, and 0 halogenated disinfection by products. SFSR averaged 4 pesticides, 11 pharmaceuticals, 2 other consumer or industrial chemicals, and 0 halogenated disinfection by products. PC averaged 1 pesticide, 2 pharmaceuticals, 0 other consumer or industrial chemicals, and 0 halogenated disinfection by products.

In 2015, NFSRB2 (mixed) averaged 25 detections of unambiguously anthropogenic organic contaminants followed by SFSR at 21 and PC at 9. NFSRB2 averaged 6 pesticides, 6 pharmaceuticals, 4 other consumer or industrial chemicals, and 0 halogenated disinfection by products. SFSR averaged 6 pesticides, 8 pharmaceuticals, 6 other consumer or industrial chemicals, and 0 halogenated disinfection by products. Finally, PC averaged 4 pesticides, 2 pharmaceuticals, 3 other consumer or industrial chemicals, and 0 halogenated disinfection by products.

The occurrence of different classes of chemicals is generally consistent with the level and type of anthropogenic impact at these sites, with the wastewater impacted NRW having the highest detections of pesticides, pharmaceuticals and disinfection byproducts. It was somewhat surprising that the agriculturally impacted NRA did not have higher detections of pesticides, although concentrations at the site were often higher than other sites (Barber et al., 2022; Barber et al., 2019). The mixed-use SFSR and NFSRB2 both had a moderate number of detections of chemicals from the different classes, consistent with mixed impacts. The PC reference site had the lowest number of chemical detections and often had the lowest concentrations of individual chemicals. PC also had the lowest concentrations of nutrients in both years, consistent with relatively low anthropogenic impact. Detections of unambiguously anthropogenic organic contaminants help to establish the level of human impact at each site but may underestimate risk to fish health which can be impacted by chemicals with both natural and anthropogenic sources (e.g.,phytoestrogens and steroid hormones).

3.2. Organismal, cellular, and targeted molecular endpoints

Fish exposure experiments in the SRW resulted in sex- and site-specific adverse organismal outcomes, including elevated mortality and reproductive disruption. To streamline interpretation of the large number of measured biological endpoints and accompanying statistical comparisons, discussion is focused on sites that were significantly different from the IC and/or PC. Both the PC and IC cohorts have advantages and limitations for use as the statistical “baseline” when evaluating exposure effects. The IC fish were sampled on day-0 of the experiment and although they were not subjected to the mobile laboratory exposure conditions, they are valuable for evaluating the starting condition of fish and for quantifying the changes in slowly responding organismal and organ level biomarkers. The PC fish were subject to the same handling and mobile laboratory conditions as all other sites. However, the water at PC was not 100 % pristine, as is likely the case with most surface waters in the U.S. (Kolpin et al., 2002; Bradley et al., 2017), and perhaps globally. Nevertheless, our chemistry data indicate that PC had a low level of anthropogenic contamination compared to other sites and its use as a reference allows for comparisons of exposure effects to realistic surface waters across a gradient of contaminant impacts. Pair-wise comparison of organismal and molecular biomarkers for all sites and IC fish are shown in Fig. 2 and SI Figs. S2–S4. Readers are encouraged to evaluate biomarker data holistically, without relying too heavily on arbitrary statistical cutoffs.

Fig. 2.

Overview of organismal and molecular biomarkers from initial controls and fish exposed for 21 days at different sites in the Shenandoah River watershed for A) males in 2014, B) males in 2015, C) females in 2014, and D) females in 2015. [Colors are based on Z-scores for individual biomarkers, blue indicates an increase in a given biomarker (row) and red indicates a decrease, relative to the grand mean. Z-scores were calculated independently for all biomarkers, years, and sexes. Different letters indicate significant pairwise differences between sites for an individual biomarker. IC = day-0 initial controls, PC = Passage Creek (reference), NRA = North River above (agricultural), NRW = North River wastewater (wastewater treatment plant), SFSR = South Fork Shenandoah River (mixed-use), NFSRB2 = North Fork Shenandoah River (mixed-use).]

At the agriculture-impacted NRA (2014 only), about one-half (52 %) of the male fish died during the 21-day exposure (Fig. 3). Survivorship was >90 % at all other sites, and in females exposed contemporaneously at NRA. The 21-day cohort of surviving males at NRA had reduced condition factor (Tukey’s HSD, p = 0.049, Figs. 2A & S2A) and GSI compared to IC fish (Tukey’s HSD, p < 0.001; Figs. 2A & S3A) along with decreased nuptial tubercle number relative to PC fish (Dunn test, p < 0.001, Figs. 2A & S4A). Females exposed at NRA had reduced GSI after 21 days compared to both IC and PC fish (Tukey’s HSD, IC p < 0.0001, PC p = 0.015; Figs. 2C & S3B) and reduced plasma Vtg at 7 and 21 days compared to PC (Tukey’s HSD, 7-day p = 0.039, 21-day p < 0.001; Figs. 2C & S5B).

Fig. 3.

Mortality from male fish exposed at sites in the Shenandoah River watershed in 2014. [PC = Passage Creek (reference), NRA = North River above (agricultural), NRW = North River wastewater (wastewater treatment plant), SFSR = South Fork Shenandoah River (mixed-use).]

Exposure at the WWTP-impacted NRW (2014 only) site resulted in impacts on reproductive biomarkers. GSI was decreased in males (gonad mass decreased) compared to IC fish (Tukey’s HSD, p < 0.001; Figs. 2A & S3A), as was sperm abundance (Dunn test, p = 0.018; Figs. 2A & S4C). Nuptial tubercle number was decreased relative to both IC and PC fish (Dunn test, IC p = 0.014, PC p = 0.001; Figs. 2A & S4A). Male fish exposed at this site showed decreased abundance of plasma Vtg after 21 days compared to PC fish (Tukey’s HSD, p = 0.003; Figs. 2A & S5A) and decreased vtg mRNA compared to IC fish (Tukey’s HSD, p = 0.027, Figs. 2A & S5E). Compared to IC and PC fish, females exposed at NRW also had decreased GSI (Tukey’s HSD, IC p < 0.001, PC p = 0.006; Figs. 2C & S2B) and vtg mRNA at day 21 (Tukey’s HSD, IC p < 0.001, PC p < 0.001; Figs. 2C & S5F) but not significantly decreased plasma Vtg (Figs. 2C & S5B).

Organismal effects were less severe at the SFSR and NFSRB2 mixed-use sites. In 2014, male fish at SFSR had increased condition factor (Tukey’s HSD, p = 0.015, Figs. 2A & S2A) and decreased GSI (Tukey’s HSD, p = 0.005, Figs. 2A & S3A) compared to IC fish. Females exposed at SFSR in 2014 had decreased ovarian stage index relative to PC fish (Dunn test, p = 0.010, Figs. 2C & S4E), and increased plasma Vtg compared to IC fish after 21-days (Tukey’s HSD, p = 0.041, Figs. 2C & S5B). In 2015, male fish at SFSR showed decreased plasma Vtg concentrations at day-7 compared to IC fish (Tukey’s HSD, p = 0.003; Figs. 2B & S5C), and vtg mRNA levels were decreased at day-21 compared to IC and PC fish (Tukey’s HDS, IC p = 0.035, PC p = 0.003; Figs. 2B & S5G). At NFSRB2 plasma Vtg was decreased in males compared to IC fish after 21 days (Tukey’s HSD, p = 0.009; Figs. 2B & S5C). At the PC reference site, the only significant differences relative to IC fish were an increased number of nuptial tubercles for males in 2014 (Dunn test, p = 0.019, Figs. 2A & S4A) and increased plasma Vtg in females at day-21 in 2014 (Tukey’s HSD, p = 0.043, Figs. 2C & S5B).

3.3. Liver metabolomics

Hepatic metabolites were analyzed only for 2014 FHM. Approximately 1200 polar hepatic metabolites were measured; the only significant change detected at 7-day was in one unidentified metabolite among NRA males (p < 0.05); no other 7-day exposure resulted in any impacted metabolites. Thus, we restricted further reporting of the metabolomic data to the 21-day exposures, where a considerable number of metabolites were found to have been impacted (Fig. S6). There were marked differences between male and female metabolomic responses for the 21-day exposures (Fig. S7). For example, the largest response for males was observed at NRW (384 significantly changed metabolites) whereas with females, the largest response (161 significantly changed metabolites) was observed at SFSR. There were no significantly impacted metabolites observed for males at SFSR. Pathway analysis of differentially expressed metabolites identified 47 differentially expressed pathways at NRW, 7 at NRA, and 0 at SFSR for males. For females, there was 1 differentially expressed pathway at NRW, 9 at NRA, and 35 at SFSR. Differentially expressed pathways were generally related to carbohydrate and amino acid metabolism (Table S4).

For a systematic analysis of the relationship between contaminants and exposure effects, we used PLS regression to estimate covariances between contaminant concentrations and metabolomic responses to exposure (Davis et al., 2016; Collette et al., 2019). The final PLS regression model for males contained 58 of the 77 detected organic chemicals that significantly covaried with endogenous metabolite profiles; the final model for females contained 53 (Tables S5 and S6). The contaminants in the final models were categorized into three groups based on CV-ANOVA values, referred to as exhibiting either relatively strong, moderate, or weak covariance with endogenous metabolite changes (Tables S5 and S6). The contaminants identified by PLS regression differ markedly between males and females. Of the 15 chemicals that strongly covaried with female endogenous metabolites, 11 were either omitted from the male model or were found to covary only weakly with male metabolites. There were also some similarities among the male and female final PLS models. For example, the herbicide metolachlor strongly covaried with metabolite profiles in both sexes. Thiabendazole, simazine, and atrazine also were found to either strongly or moderately covary in both models. Nine contaminants were omitted from both models due to little or no covariance with metabolite profiles: 4-tert-octylphenoltriethoxylate, biochanin A, caffeine, daidzein, equilenin, fexofenadine, formononetin, medroxyprogesterone, and metalaxyl.

3.4. Differentially expressed genes (DEGs)

Individual genes were identified as up- or downregulated, using PC as the reference, from the 7-day male cohorts from the 2014 and 2015 experiments. The qPCR analysis produced similar results (directional and fold change) to the microarray data which are consistent with previously published microarray papers (Fig. S8) (Bahamonde et al., 2015; Martyniuk et al., 2016; Villeneuve et al., 2017). In 2014, there were 202 DEGs in fish at NRA, 297 at NRW and 219 at SFSR. In 2015, there were 530 DEGs at SFSR and 1224 at NFSRB2. The higher number of DEGs in 2015 is surprising, given that organismal impacts were more severe in 2014. Sites in 2015 had higher overall contaminant loads than 2014 (Barber et al., 2019), in terms of number of detections and overall concentration. Thus, it is possible that fish in 2015 expressed broader transcriptional responses that did not translate to higher-level impacts following exposure to compounds with limited toxicity, whereas fish in 2014 were exposed to specific contaminants that initiated directed AOPs leading to adverse organismal outcomes. In the field of genomic toxicology, it remains unclear how the total number of DEGs is related to the physiological status of an organism. The relationship of the transcriptome to higher level endpoints is expected to depend more upon the specific pathways that are altered in an organism rather than based upon absolute numbers of DEGs (Bahamonde et al., 2016). Toward this end, discussion below is focused on genes from known AOPs and pathway level analysis (GSEA and SNEA).

A subset of transcripts was selected for focused analysis based on their relevance to environmental contaminant exposures (Table S2). Consistent with the Vtg data, there was no differential abundance of reproduction related transcripts (e.g., estrogen and androgen receptors) in 2014 (Barber et al., 2019). Among selected transcripts in the aryl-hydrocarbon receptor (AhR) pathway, cytochrome p450 1a1 (cyp1a1) was upregulated 2.8-fold at NRW (Tukey’s HSD, p < 0.001, Fig. S9E). Cyp1a1 encodes an enzyme involved in the transformation of certain organic contaminants. Of the chemicals detected only at NRW, the pesticide diuron is known to increase the expression of cyp1a1 in mammals (Ihlaseh et al., 2011). Diuron was also found to strongly covary with metabolite profile changes for the 21-day exposure of male fish (Table S5). Additional chemicals detected at NRW and other sites have also been shown to increase the expression of cyp1a1 including 4-nonylphenol (Lee et al., 2005), and carbamazepine (Oscarson et al., 2006). Of these, carbamazepine strongly covaried with metabolomic changes in the male fish, but 4-nonylphenol was omitted from the final PLS model for metabolomics due to low covariance.

3.5. Differentially expressed pathways

GSEA and SNEA were used to identify differentially expressed molecular pathways and cellular processes, providing a functional biological context for transcriptomics results. Pathways involved in lipid transport and metabolism, inflammation, and immune function were differentially regulated in a site-specific manner (Tables S7–S11). Differentially expressed pathways can be responsive to different chemical exposures as well as other differences in the water between sites.

All sites in both years showed differential expression of lipid related pathways, including lipid transport, storage, metabolism, and cholesterol synthesis (Tables S7–S11). In 2014, NRA and NRW had the highest number of differentially regulated lipid pathways. Transcriptomic changes to lipid related processes are frequently observed in environmental mixture studies, including in FHM exposed at agriculture-impacted sites (Sellin Jeffries et al., 2012), WWTP-impacted waters (Martinović-Weigelt et al., 2014; Garcia-Reyero et al., 2009) and in wild barramundi (Lates calcarifer) from an agricultural drainage (Hook et al., 2017). Although separated temporally from the transcript abundance data (i.e., 7-day vs. 21-day), metabolite changes observed at NRA and NRW in 2014 further suggested impacts on lipid pathways. For example, l-carnitine, an amino acid associated with mitochondrial transport of fatty acids (Perna et al., 2007), was affected by exposure at NRA. l-carnitine was not affected by exposure at NRW, but the concentration of its primary precursor, trimethyllysine, decreased as did S-adenosylhomocysteine, another component in the l-carnitine biosynthesis pathway (Perna et al., 2007). Other lipid related metabolites were also affected by exposure at NRW in 2014, including phosphoglyceric acid, glycerol-3-phosphate, nonanedioic acid, and phosphocholine.

Pathways involved in inflammation and immune function were largely downregulated at NRA in 2014 (Table S7) and upregulated at SFSR in both years (Tables S9 and S11). The most enriched immune-related pathways at NRA were in the IL-6/STAT5B signaling family as well as classical, alternative, and lectin-induced complement pathways. At SFSR, several interleukin/STAT pathways in the common γ-chain family were up-regulated in 2014. The lectin-induced and classical complement pathways, T-cell activation, and several NF-κβ signaling pathways were also differentially regulated, as were sub-networks for immune cell differentiation, humoral immune response, and inflammatory response.

3.6. Hierarchical clustering of chemical and transcriptome profiles

Hierarchical clustering of organic constituents, inorganic constituents, and transcriptome profiles was used to explore patterns in these data (Fig. 4, Fig. S10). In general, individual sites formed unique clusters for both chemical occurrence and transcriptomic response, indicating that sites had overall unique contaminant profiles and exposure to these mixtures produced unique transcriptomic responses, consistent with our hypothesized linkages between land use, contaminant occurrence, and exposure effects. Sites with unique land uses generally form distinct clusters, whereas sites with similar land use characteristics, such as SFSR and NFSRB2, did not form unique contaminant or transcriptional clusters in 2015.

Fig. 4.

Average-linkage hierarchical clustering of differentially expressed genes from male fish exposed at different sites in the Shenandoah River watershed in A) 2014 and B) 2015. [PC = Passage Creek (reference), NRA = North River above (agricultural), NRW = North River wastewater (wastewater treatment plant), SFSR = South Fork Shenandoah River (mixed-use), NFSRB2 = North Fork Shenandoah River (mixed-use). Numbers at nodes are the approximately unbiased bootstrap probability values. Numbers following site names represent an arbitrary code for individual fish.]

In 2014, hierarchical clustering of organic and inorganic contaminant profiles showed a pattern of clustering by site. For organic contaminants, all site-level clusters received strong support (AU p > 95; Fig. S10A). Trace elements in 2014 also showed site-specific clusters with strong or moderate support (AU p > 95 or 80, respectively), except for one sampling event from NRA (Fig. S10C). In 2015, the organic profiles from PC formed a unique cluster with moderate support (AU p = 88) while SFSR and NFSRB2 did not form unique clusters (Fig. S10B). The inorganic constituents from 2015 showed a similar pattern with 3 of 4 PC samples clustering together (AU p = 100), but NFSRB2 and SFSR samples did not form unique groupings (Fig. S10D). The chemical data from both years indicate variation in contaminant profiles at sites with varying land use and wastewater impacts.

In 2014, DEG profiles showed a pattern of clustering by site, with PC- and NRW-exposed fish forming strongly- and moderately-supported clusters, respectively, with the exception of one fish from NRW (Fig. 4A). The transcriptome profiles from PC fish were the most distinct from other profiles. In 2015, transcript data again showed site-specific patterns, with PC-exposed fish forming a moderately supported cluster (Fig. 4B). The two mixed-use sites together formed a moderately supported cluster, but they did not cluster together as individual sites. This cluster is consistent with variation in human impact and land use between sites.

3.7. Comparison of transcriptomics and metabolomics results

Metabolomic data were available for both males and females exposed for 7 and 21 days in 2014, while transcriptomic data were available only for males exposed for 7 days in 2014 and 2015. These two techniques have large differences in number of individual endpoints that are measured and in the types of biological information gathered. Thus, pathway analysis of metabolomic and transcriptomics datasets are not expected to produce identical results. However, if these techniques are indeed useful for elucidating effects of mixture exposure, they should reveal similar insights about a given exposure.

At day-7, transcriptomic data showed moderate disruption (200–300 DEGs) at each site, whereas metabolomic indicated minimal impact with only one differentially expressed metabolite at SFSR. However, after 21-days of exposure there were hundreds of differentially expressed metabolites in males between the various exposure sites. The difference in response between the two techniques is consistent with the slower response of higher order metabolic endpoints compared to transcripts due to variation in their underlying mechanisms. Additionally, the larger metabolomic response on day-21 vs. day-7 could be influenced by bioconcentration leading to higher in vivo concentrations of contaminants. The early response of transcriptomics may be useful for early detection and screening of potential adverse outcomes. For both techniques, the highest number of differentially expressed entities was observed in males at the WWTP-impacted NRW site. At the pathway level, both metabolomic and transcriptomic data show differential expression of metabolism-related processes. Of the 47 metabolomic pathways differentially expressed at NRW, 14 had analogous pathways that were differentially expressed in transcriptomic data. Of the 7 differentially expressed metabolomic pathways at NRA, 4 had analogous pathways differentially expressed in transcriptomics data. The pathways in common between metabolomics and transcriptomics were related to carbohydrate, amino acid, and lipid metabolism.

3.8. Integrated effects assessment of relevant AOPs

Exposure to complex mixtures produced site- and sex-specific effects at multiple levels of organization. Adverse outcomes were most pronounced at the agriculturally-impacted NRA and the WWTP-impacted NRW sites. The most striking organismal result was the increased mortality of male fish at NRA in 2014. There was no obvious single variable that explained this mortality within the chemical, omic, or organismal datasets. The NRA site did not appear exceptional in the occurrence and concentration of measured contaminants. The numbers of differentially expressed genes and genetic pathways prior (day-7) to the mortality event (day-16 and day-21) were not higher at NRA compared to other impacted sites. Metabolite data from day-21 male fish show less of an impact at NRA compared to NRW. At the organismal level, day-21 male NRA fish showed evidence of reproductive disruption in the form of reduced GSI and decreased nuptial tubercles, but many of these effects were also observed at NRW. Given that day-21 data are only available for surviving animals, it is possible that genetic or other inter-individual variation in susceptibility contributed to mortality and therefore relevant physiological data are censored from our observations. Alternatively, stream gages near the NRA site indicate a rain event impacting streamflow occurred on day-9 of exposure in 2014, between the day-7 and day-21 water and fish samplings. This event may have mobilized a pulse of contaminants that was undetected during weekly water sampling, but that may have induced latent effects on survival. It is also possible that factors other than contaminant exposure were involved; however, the mortality occurred independently across six aquaria and males at NRA were housed in the same mobile exposure unit as females and NRW fish, suggesting that experimental conditions were unlikely to explain this outcome.

Changes in reproductive endpoints were observed at several sites, consistent with potential endocrine disruption. At NRW and NRA, male fish displayed decreased GSI, sperm abundance, and nuptial tubercles, consistent with an estrogenic MIE, which has been reported in studies of WWTP-impacted sites (Ekman et al., 2018; Vajda et al., 2011; Jobling et al., 1998; Hicks et al., 2017) and has been suspected to occur throughout the SRW. Risk quotients, where measured contaminant concentrations were compared to benchmark toxicity values, indicated potential risk from environmental estrogens, including bisphenol A (BPA) and 4-nonylphenoldiethoxylate (Barber et al., 2022). BPA was found to covary moderately with metabolite profiles in males while 4-nonylphenoldiethoxylate showed weak covariance. However, molecular changes at the gene, protein, and pathway level were not consistent with estrogen receptor agonism, especially in the case of plasma Vtg, which was reduced in both males and females.

The finding of reduced Vtg in males at NRW is surprising given that wastewater effluents are frequently estrogenic and results in increases in Vtg. The magnitude of variation in Vtg expression between all sites was small compared with exposures to known estrogens and estrogenic effluents, which often involve inductions of several orders of magnitude (Vajda and Norris, 2011; Jobling et al., 1998). It is also possible that Vtg was slightly elevated in initial control males, and subsided toward baseline levels at different rates at the different exposure sites due to the unique chemical mixtures present. However, this would be unlikely to explain the large decreases in either plasma Vtg or hepatic vtg mRNA observed in female fish at NRW and NRA. Disruption of androgen signaling pathways could explain the reduced number of nuptial tubercles, sperm abundance, and GSI in males (Ankley et al., 2020; Ankley et al., 2004) and is also consistent with the large metabolomic impact to male fish compared to females at NRW. Fluconazole, a common antiandrogenic fungicide (Draskau and Svingen, 2022) detected at NRW and SFSR, was found by PLS regression to covary strongly with metabolomic profiles in males. BPA, identified recently as an androgen receptor antagonist (Ekman et al., 2012), was detected at all sites in 2014 and moderately covaried with male metabolomic profiles.

Female fish exposed at NRA and NRW showed decreased GSI, hepatic vtg transcript, and NRA female fish also showed decreased plasma Vtg. These effects are consistent with several endocrine disruption AOPs including aromatase inhibition, estrogen receptor antagonism, and androgen receptor agonism all leading to reproductive disruption (Vajda and Norris, 2011). Additionally, non-endocrine mechanisms, such as metabolic or immune disruption, could have impacted reproductive endpoints. Taken together, the suite of biological effects from multiple levels of organization indicate that fish experienced multiple MIEs with concurrent activation of different pathways leading to a network of mixture effects.

Although omics measurements have been employed in previous studies of environmental chemical mixtures, the current study is, to our knowledge, the first to employ extended 21-day exposures in controlled conditions with multiple sampling points and multiple omics technologies. This design revealed several important features of mixture exposure. At the organismal level, impacts at NRA and NRW appear similar, with both sites showing decreased GSI in males and females, lower numbers of nuptial tubercles in males, and a general trend of decreased vtg in females. However, transcriptomic and metabolomic data show these sites produced very different responses at the molecular level. For example, the number and identities of DEG varied between the sites, the NRA site had many immune related pathways differentially regulated while NRW did not, and the number of differentially expressed metabolites varied dramatically between these two sites. This variation in molecular response is consistent with variation in contaminant profiles between these sites. Hence, omics endpoints were valuable in differentiating underlying molecular mechanisms in cases where mixture exposure produced somewhat convergent organismal outcomes through different AOPs.

3.9. Risk assessment and contaminant prioritization

As expected, exposure effects were most pronounced at sites with the highest level of human impact, including the agricultural-impacted NRA and the WWTP-impacted NRW. This is consistent with previous studies in the SRW and greater Chesapeake Bay watershed that have associated endocrine disruption in fish with both agricultural land use and wastewater effluents (Blazer et al., 2012; Blazer et al., 2007; Blazer et al., 2021). We also observed organismal and molecular impacts at the mixed-use sites. For example, male fish exposed at the mixed-use SFSR site in 2014 had decreased condition factor and GSI, and females had decreased plasma Vtg and ovarian stage index. These findings suggest that there is some increased risk from complex mixture exposures at moderately impacted sites in the SRW.

The ubiquitous occurrence of chemical mixtures across the landscape represents an inherently complex and challenging scenario for risk assessment, where multiple stressors impact diverse species through multiple mechanisms. In this study and the companion chemical mixture study (Barber et al., 2022) we employed different approaches to parse this complexity. Fish exposures provided detailed physiological response data for one species. This approach captures real-world exposure scenarios and integrates mixture effects through in vivo toxicokinetic and toxicodynamic processes. However, the insights produced by this approach apply to a narrow taxonomic range and the results of aggregate mixture exposure are difficult to apply to existing regulatory frameworks. The risk ranking approach used in the companion paper compared measured concentrations of contaminants to established or predicted risk benchmarks, which provides greater taxonomic coverage and implicates individual contaminants through established risk assessment methods, but at the cost of decreased environmental and biological realism.

The risk ranking approach and the PLS regression approach both seek to identify individual contaminants that contribute the greatest risk to exposed organisms, therefore comparison of the results of these approaches may be valuable for guiding future complex chemical mixture studies. Under the risk ranking approach, 6 contaminants had risk quotients (RQ) >0.1 when literature and database sources were used to establish fish-specific predicted no effects concentration equivalents (PNEC_equivalent). Of those 6, only estrone was detected at any sites in 2014, the year for which PLS regression was available. Estrone had an RQ of 0.14, indicating the potential for increased risk in exposed fish, and PLS regression revealed strong covariance between estrone concentrations and metabolite response in females but weak covariance in males. However, the molecular and organismal responses observed in males and females in 2014 are not concordant with the estrogenic AOP expected for estrone.

When benchmark toxicity concentrations were estimated from quantitative structure activity relationship models, none of the contaminants had RQ values >0.01. There was some overlap between the 25 contaminants with the highest modeled RQ values and contaminants with strong or moderate covariance with metabolite profiles as estimated by PLS regression. For the male PLS model those contaminants were galaxolide, triclosan, bupropion, 1,4-dichlorobenzene, and bisphenol A. For the female PLS model, the overlapping contaminants were 4-nonylphenol and fipronil. Previous complex mixture studies have employed similar conceptual approaches to identify associations between contaminant occurrence, benchmark toxicity concentrations, and/or observed exposure effects. These prioritization studies also identified estrone (Perkins et al., 2017; Schroeder et al., 2017; Blackwell et al., 2017), triclosan (Perkins et al., 2017; Blackwell et al., 2017), bisphenol A (Perkins et al., 2017; Blackwell et al., 2017), 4-nonylphenol (Perkins et al., 2017; Blackwell et al., 2017), and 1,4-dichlorobenzene (Perkins et al., 2017) as potentially contributing to increased risk for exposed fish. These results demonstrate how various prioritization-based approaches are valuable for identifying mixture constituents that may be responsible for increased risk in exposed organisms. Continued application of these approaches in complex chemical mixture studies is expected to yield comprehensive data for informing regulatory decisions and management of aquatic habitats.

3.10. Implications for fish health in the SRW

There are interesting similarities and differences between the results seen here and observations from wild fish in the SRW. Mass mortality events and reproductive disruption have been reported in wild centrarchid species throughout the SRW beginning in 2004 (Blazer et al., 2010; Blazer et al., 2007; Fink, 2014). The current mobile-laboratory exposures also demonstrated reproductive disruption and elevated mortality, but largely at sites in tributaries in the upper watershed (NRW and NRA) characterized by distinct land use and contaminant profiles. Reproductive disruption in wild bass (Micropterus spp.) has occurred in the form of elevated Vtg concentrations and a high prevalence of testicular oocytes in male fish, leading to a focus on estrogenic endocrine disruption (Blazer et al., 2012; Iwanowicz et al., 2009). Much of the focus on estrogenic contaminants in the SRW stemmed from observations of testicular oocytes in male fish, which were assumed to be the result of endocrine disruption. However, recent studies of museum specimens have documented testicular oocytes in male smallmouth bass dating as far back as the 1880s (Christiana Grim et al., 2020), before the widespread discharge of synthetic estrogenic chemicals in wastewater effluents. It is likely that some frequency of testicular oocytes in smallmouth bass populations is a natural phenomenon, and more research is necessary to distinguish between naturally occurring testicular oocytes and those induced by exposure to anthropogenic contaminants.

Here, reproductive disruption was observed at NRA and NRW, but implicated a more complex biological response than exposure to estrogens alone. These findings demonstrate the value of our integrated biomarker approach, that was able to capture these complex biological responses along multiple levels of organization and thus enable evaluation of multiple potential AOPs. These complex responses may have been missed by more traditional hypothesis-based approaches.

The wild fish mortality in the Shenandoah and Potomac Rivers was accompanied by external lesions, high abundances of opportunistic pathogens, and site-specific immunomodulation, leading to the hypothesis that immunosuppression was contributing to fish kills (Blazer et al., 2010; Ripley et al., 2008). The elevated mortality observed in males at NRA during this study was accompanied by down-regulation of immune pathways in surviving males, suggesting that immunosuppression may have been involved, although the mortality reported here only occurred in males and there is no evidence of sex-biased mortality in wild fish.

There are multiple potential explanations for differences between FHM mobile laboratory experiments and observations in wild bass. For example, variation in responses to contaminant exposure between species is often attributed to differential sensitivity between taxa: however, the specific etiology of this variation remains poorly understood. Potential factors influencing species sensitivity include physiological factors such as receptor binding affinity, gene regulatory architecture, or toxicokinetic variation. However, ecological factors should also be considered, including dietary exposures at different trophic levels, movement behavior, and spawning habitats, all of which could affect exposure dynamics. For example, far-ranging fish such as bass (Todd and Rabeni, 1989) are exposed to different chemical mixtures as they move through different habitats. This movement may obscure the relationship between the spatial patterning of contaminants and biological phenomena in a watershed, creating a potential scenario where localized contaminant exposure leads to widespread adverse outcomes in far-ranging fish species. This is consistent with the current study where adverse outcomes were most prevalent at tributaries in the upper watershed (NRW and NRA) characterized by distinct local land use and point-source impacts, suggesting a disproportionate contribution of local sources to widespread adverse effects in wild fish. Future studies are needed to explicitly evaluate how spatial scaling impacts observed landscape/contaminant/biology relationships. Mobile laboratory exposures serve as a complement to wild fish surveys by reducing confounding variables and by placing mixture exposures in a spatially explicit context for exploration of landscape-scale processes.

The increased mortality at NRA and the reproductive disruption at NRA and NRW are a cause for concern given the potential for population-level consequences in wild fish. These findings implicate complex mixtures in the adverse fish health outcomes at agricultural- and wastewater- impacted sites in the Shenandoah River watershed. The candidate list of potential causative agents identified here, based on their covariance with metabolomic responses and their concordance with known AOPs, can inform watershed management efforts. Further investigation is needed to better understand how concurrent activation of multiple AOPs contribute to complex responses across levels of biological organization.

Supplementary Material

Supplement1

Table S1. Geographic metadata for mobile laboratory exposure sites in 2014 and 2015.

Table S2. Genes selected a priori for analysis based on known responses to environmental contaminants.

Table S3. Primer sequences used for qPCR assays.

Table S4. Differentially expressed metabolomic pathways.

Table S5. Chemicals included in the final, optimized PLS regression model for relating chemical occurrence levels to changes in polar hepatic metabolite profiles in male fathead minnows exposed in 2014.

Table S6. Chemicals included in the final, optimized PLS regression model for relating chemical occurrence levels to changes in polar hepatic metabolite profiles in female fathead minnows exposed in 2014.

Table S7. Selected differentially expressed pathways in livers of fathead minnows exposed at NRA in 2014.

Table S8. Selected differentially expressed pathways in livers of fathead minnows exposed at NRW in 2014.

Table S9. Selected differentially expressed pathways in livers of fathead minnows exposed at SFSR in 2014.

Table S10. Selected differentially expressed pathways in livers of fathead minnows exposed at NFSRB2 in 2015.

Table S11. Selected differentially expressed pathways in livers of fathead minnows exposed at SFSR in 2015.

Table S12. All differentially expressed pathways identified by gene set enrichment analysis (GSEA).

Table S13. All differentially expressed sub-networks identified by sub-network enrichment analysis (SNEA).

Data availability

Data will be made available on request.