Environmental Pollution 231 (2017) 671e680
Contents lists available at ScienceDirect
Environmental Pollution journal homepage: www.elsevier.com/locate/envpol
Reductions in fish-community contamination following lowhead dam removal linked more to shifts in food-web structure than sediment pollution* Robert P. Davis*, S. Ma zeika P. Sullivan, Kay C. Stefanik School of Environment and Natural Resources, The Ohio State University, 2021 Coffey Road, Columbus, OH 43210, United States
a r t i c l e i n f o
a b s t r a c t
Article history: Received 30 March 2017 Received in revised form 9 July 2017
Recent increases in dam removals have prompted research on ecological and geomorphic river responses, yet contaminant dynamics following dam removals are poorly understood. We investigated changes in sediment concentrations and fish-community body burdens of mercury (Hg), selenium (Se), polychlorinated biphenyls (PCB), and chlorinated pesticides before and after two lowhead dam removals in the Scioto and Olentangy Rivers (Columbus, Ohio). These changes were then related to documented shifts in fish food-web structure. Seven study reaches were surveyed from 2011 to 2015, including controls, upstream and downstream of the previous dams, and upstream restored vs. unrestored. For most contaminants, fish-community body burdens declined following dam removal and converged across study reaches by the last year of the study in both rivers. Aldrin and dieldrin body burdens in the Olentangy River declined more rapidly in the upstream-restored vs. the upstream-unrestored reach, but were indistinguishable by year three post dam removal. No upstream-downstream differences were observed in body burdens in the Olentangy River, but aldrin and dieldrin body burdens were 138 and 148% higher, respectively, in downstream reaches than in upstream reaches of the Scioto River following dam removal. The strongest relationships between trophic position and body burdens were observed with PCBs and Se in the Scioto River, and with dieldrin in the Olentangy River. Food-chain length e a key measure of trophic structure e was only weakly related to aldrin body burdens, and unrelated to other contaminants. Overall, we demonstrate that lowhead dam removal may effectively reduce ecosystem contamination, largely via shifts in fish food-web dynamics versus sediment contaminant concentrations. This study presents some of the first findings documenting ecosystem contamination following dam removal and will be useful in informing future dam removals. © 2017 Elsevier Ltd. All rights reserved.
Keywords: Ecosystem contamination Biomonitoring Persistent contaminants Restoration Lowhead dam removal
1. Introduction Dams serve many functions including provisioning hydropower, storing water, controlling floods, and providing recreational opportunities (Born et al., 1998; Graf, 1999). Lowhead, run of river dams (43,000; US Army Corp of Engineers, 2013). Removal of these dams is growing in popularity, as many have become obsolete, expensive to maintain, or dysfunctional (Bednarek, 2001; Doyle et al., 2005; Harris and Evans, 2014). To date, >1000 dams have been removed
*
This paper has been recommended for acceptance by Maria Cristina Fossi. * Corresponding author. E-mail address: [email protected] (R.P. Davis).
http://dx.doi.org/10.1016/j.envpol.2017.07.096 0269-7491/© 2017 Elsevier Ltd. All rights reserved.
in the United States alone (O'Connor et al., 2015). The ecological and geomorphic effects of large dams on rivers are well documented (Freedman et al., 2014; Poff et al., 2007; Pringle et al., 2000); increasing attention has also recently been afforded to lowhead dams (Csiki and Rhoads, 2014; Fencl et al., 2015; Mbaka and Mwaniki, 2015). Studies suggest that fish communities, in particular, experience significant shifts in diversity, community structure, and food-web architecture following lowhead-dam removal (Dorobek et al., 2015; Dorobek, 2016; Kornis et al., 2015). Prior to dam removal, reservoirs created by impoundments behind dams have been shown to alter pH, streamflow, sediment load, and temperature (Fairchild and Velinsky, 2006; Poff and Hart, 2002; Santucci et al., 2005). As streamflow velocity, discharge, and boundary shear stress decrease behind the dam, fine suspended sediments settle out of the water column, along with contaminants
672
R.P. Davis et al. / Environmental Pollution 231 (2017) 671e680
that readily bind to sediments and the organic compounds they contain (An et al., 2016; Green et al., 2016; Syvitski et al., 2005), potentially leading to elevated contaminant levels behind dams (Gao et al., 1998; Warren et al., 2003). Contaminants of particular concern include chlorinated pesticides, polychlorinated biphenyls (PCBs), mercury (Hg), and selenium (Se). Although some of these contaminants e such as chlorinated pesticides (e.g., DDT) and PCBs e have not seen widespread use since the 1970's, they can persist in aquatic sediments (Evans and Gottgens 2007; Hinck et al., 2008; Johnson et al., 2013) by binding to benthic organic material (Warren et al., 2003). The potential for dam removal to release contaminated sediments is a major ecological concern (Bednarek, 2001; Green et al., 2016). In the case of large dams, removal has resulted in up to 70% of the former reservoir sediment transported downstream in the initial years (Grant and Lewis, 2015; O'Connor et al., 2015; Sawaske and Freyberg, 2012), potentially transporting contaminants downstream as well (Shuman, 1995; Stanley and Doyle, 2003). As sediments are resuspended during downstream transport, contaminants may remobilize and become bioavailable to fish and other aquatic organisms. Generally, higher localized contaminant concentrations lead to elevated fish-community contaminant burdens (Madenjian et al., 2009; Odhiambo et al., 2013). However, few examples of changes in contaminant burdens in fish communities exist with regards to dam removal, in spite of documented or predicted changes in sediment dynamics (Hart et al., 2002; Poff and Hart, 2002; Stanley and Doyle, 2003). Legacy contaminants can biomagnify (i.e., the multiplicative increase in contaminant body burdens up the food chain (Miranda et al., 2008)); and remain stored in fatty tissues of fish (Borga et al., 2012). Thus, contaminant body burdens in fish are commonly associated with fish food-web properties (Bentzen et al., 1996; Rasmussen et al., 1990). Trophic position, and in particular, foodchain length (FCL) e the maximum number of energetic transfers in a food chain e can be strongly related to biomagnification in fish, whereby individuals feeding at higher trophic positions exhibit greater contaminant body burdens than those feeding at lower trophic positions (Houde et al., 2008; Schmitt et al., 2011; van der Velden et al., 2013). Of fundamental importance to the biomagnification process, FCL can be altered by the insertion or deletion of consumers feeding at distinct trophic levels (Post et al., 2000; Sabo et al., 2010), thereby creating or removing trophic steps. Hoeinghaus et al. (2008) found that FCL varied by habitat type, with impounded river reaches having the longest food chains, which they attributed to the insertion of trophic interactions occurring below top predators. Conversely, Kautza and Sullivan (2016) observed that FCL in impounded reaches of the Scioto River system was shorter (x ¼ 3.88) than free flowing reaches (x ¼ 4.19), owing to the exclusion of top predators at the impounded sites. Here, we measured concentrations of a suite of common environmental contaminants in sediment and fish communities before, as well as multiple years following the removal of two dams in two rivers of Columbus, Ohio. Previous research in this system has shown strong fish-community and food-web responses to the presence and subsequent removal of dams (see Dorobek, 2016; Dorobek et al., 2015 for details), along with downstream transport of fine sediment, suggesting that contaminant body burdens in fish following dam removal would reflect both changes in contaminant availability due to sediment release, as well as the shifting trophic composition of the fish community. Within this context, we hypothesized that changes in fishcommunity body burdens would be related to changes in contaminant concentrations in sediment, as well as fish trophic positions through time and space. Specifically, we predicted that:
(1) Contaminant concentrations in sediment would be a function of sediment grain size, organic carbon content, and streamflow velocity (which is related to flow dynamics such as boundary shear stress that influence sediment size) (Jones and de Voogt, 1999; Miller and Orbock Miller, 2007; Selin, 2009; Warren et al., 2003; Zhao et al., 2010); (2) Fish contaminant body burdens would be positively correlated with trophic position, but this relationship would be stronger before dam removal and at upstream, reservoir reaches, where greater sediment concentrations of sequestered contaminants and more predatory fish species were found; (3) Contaminant body burdens in fish would decrease in reaches where there was a loss of large predatory fish occupying high trophic levels (Dorobek, 2016); and (4) In general, contaminant body burdens in fish would be the most invariant in downstream reaches through time following dam removal, where anticipated shifts in sediment contaminant concentrations (Ashley et al., 2006; Cantwell et al., 2014) and changes in the trophic composition of fish would be minimal (Dorobek, 2016). We anticipate these findings will help inform future dam removal efforts relative to their potential impacts on contamination of resident fish communities and drivers of contaminant transfer. 2. Methods 2.1. Study reaches and experimental design Seven, 500-m river reaches were surveyed in the Olentangy and Scioto Rivers (Fig. 1) in Columbus, Ohio using a modified BeforeAfter-Control-Impact design (Kibler et al., 2011a; Stewart-Oaten et al., 1986). The Olentangy River is a 150-km, 5th-order tributary of the 370-km, 6th-order Scioto River, in the Ohio River basin. Study reaches were located upstream and downstream of two channelspanning dams: the 5th Avenue Dam (2.5 m high) in the Olentangy River and the Main Street Dam (4.1 m high) in the Scioto River. The 5th Avenue Dam was removed during low-flow conditions in late summer of 2012. Approximately 1600 m3 of sediment was removed from behind the dam and was used for the construction of riparian and wetland habitat surrounding the newly designed river channel (DLZ, 2015). The Main Street Dam was removed in November 2013. Prior to removal of the Main Street Dam, the Ohio EPA conducted an analysis of metals, pesticides, and PCBs within the sediment and found that only arsenic exceeded Residential Criteria values (Ohio EPA, 2011). Study reaches from both rivers were assigned to various treatments (e.g., upstream/downstream, upstream restored/unrestored, before/after) to assess the impacts of dam removal and subsequent restoration efforts on contaminant body burdens in fish communities. Control reaches were designated “upstream control” and “downstream control” and represent both impounded (OR1) and free-flowing (SR3) reaches. The upstream control reach (OR1, Fig. 1) is behind an intact dam of comparable size and age to the removed 5th Avenue dam; the downstream control reach (SR3, Fig. 1) is also separated from other upstream reaches by an intact lowhead dam. All other reaches were designated as “experimental”. In the Olentangy River, a 2.6-km section of the river immediately upstream of the previous dam was restored from August 2013 through winter 2014, and included channel engineering (using heavy machinery) to create heterogeneous in-stream habitat (e.g., pools and riffles), developing and reconnecting floodplain wetlands, and planting riparian vegetation (Ohio EPA, 2011). Further upstream was left unmodified. We surveyed one upstream reach in the restored section (OR3, upstream restored; Fig. 1) and one upstream reach in the unmodified section (OR2, upstream unrestored; Fig. 1). Study reaches in the Scioto River were located upstream and downstream of the Main St. Dam in downtown Columbus. Channel restoration
R.P. Davis et al. / Environmental Pollution 231 (2017) 671e680
673
Fig. 1. Study reaches on the Olentangy and Scioto Rivers in Columbus, Ohio (USA). Reaches in the Olentangy River (OR 1e4) were located upstream and downstream of the 5th Avenue Dam, removed in August of 2012. Reaches in the Scioto River (SR 1e3) were located upstream and downstream of the Main Street Dam, removed in November 2013.
was not undertaken after this dam removal; however riverbanks were modified and reconstructed to develop pubically accessible greenspaces. We recognize that replication of both control and experimental reaches would be ideal, however the designs of dam-removal studies are constrained by multiple factors. Our study features three experimental reaches in the Olentangy River and two in the Scioto River, control reaches in each river, and data collected before and after dam removal spanning multiple years. This design is on par or exceeds the designs of other published dam-removal studies (e.g., Claeson and Coffin, 2016; Kornis et al., 2015; Tullos et al., 2014). Although we recognize the methodological limitations, important information can be gleaned from this study, where we used a mixture of traditional statistical approaches in combination with the practical significance (sensu Kibler et al., 2011b) of our results to evaluate our hypotheses.
of the river reach being sampled. Sampling effort was based on predam removal efforts of 600 s per sampling event and adjusted to accommodate changes in habitat volume, in congruence with other dam-removal studies (Catalano et al., 2007; Maloney et al., 2008), by reducing sampling time concomitant with reductions in habitat volume. Three to five individuals of fish species common to multiple reaches and representing major trophic guilds (e.g., common carp as benthic omnivores) were dispatched and transported on ice to the laboratory where they were frozen at 30 C until processing. For contaminant and lipid analysis, approximately 2 g-plugs of dorsal muscle were removed and frozen until further analysis. Trophic position and FCL estimates, based on analysis of 15N and 13C natural-abundance isotopes, were generated from a companion study of the same study reaches, sampling locations, and time periods as reported in Dorobek (2016).
2.2. Fish sampling
2.3. Sediment analysis
Fish surveys were conducted from 2011 to 2015 during stable, base flows in late summer-early autumn. Control reaches (OR1 and SR3) as well as experimental reaches (OR3, SR1, and SR2) were sampled prior to dam removal. Sampling occurred across all study reaches two years following dam removal in the Scioto River and three years following dam removal in the Olentangy River. Each study reach was stratified into three transects (upstream, middle, downstream; running across the stream). At each transect, fish communities were surveyed along three upstream-downstream sections (right bank, middle, left bank; centered along transect), for a total of nine sampling locations per reach. Fish were collected via electrofishing using a Smith Root® LR-24 backpack electrofisher, 2.5 GPP Smith-Root® long-line electrofisher, or a 5 GPP SmithRoot® boat unit (Vancouver, Washington). Electrofishing units were selected based on physical characteristics (e.g., depth, conductivity)
Composite surface sediment samples were collected from each reach before and following dam removal annually in the Scioto River and 2e3 times per summer in the Olentangy River (Alberts et al., 2013; Walters et al., 2010) using a hand trowel. Sediment samples were refrigerated at ~5 C until processing for analysis. Prior to analysis, composite samples were homogenized. Sediment was analyzed at the Service Testing and Research Lab (STAR Lab, Wooster, Ohio) for total carbon, inorganic and organic carbon fractions, and total Hg (hereafter “THg”) content. Total carbon and organic carbon were analyzed using The International Organization for Standardization (1995) ISO 10694 method and inorganic carbon was determined using EPA Method 9060A (US EPA, 2014). Sediment samples were analyzed at the Michigan State University Diagnostic Center for Population and Animal Health for PCBs and chlorinated pesticides via gas-chromatography tandem-quadrupole mass
674
R.P. Davis et al. / Environmental Pollution 231 (2017) 671e680
spectrometry (GC-MS/MS). THg content was analyzed using cold vapor atomic fluorescence with a CETAC M8000 Mercury Analyzer (CETAC Technologies, Omaha, Nebraska). Mean sediment sizes reported are from Comes (2016), collected using a McNeil Sampler. 2.4. Streamflow velocity As a proxy for boundary shear stress, streamflow velocity measurements were calculated for each stream reach using an Acoustic Doppler Current Profiler (ADCP) (M9 RiverSurveyor; SonTek, San Diego, California). Discharge, a function of streamflow velocity and cross-sectional area of a channel, has a non-linear relationship with boundary shear stress, with boundary shear stress generally increasing with discharge (Pitlick and Wilcock, 2001). Data-collection methods are detailed in Comes (2016). Transects within the processed data were selected at what was approximately the upstream, middle, and downstream of each reach and velocity was calculated using the summary statistic tools in ArcGIS. Velocities were calculated at all collection events in Comes (2016), and averaged by transect. 2.5. Lipid analysis Lipid extractions were performed to determine total lipid content of fish muscle tissue at ALS Environmental Laboratories (Kelso, Washington) following protocols from Bligh and Dyer (1959). Lipid percentage was determined on a wet-weight basis. Subsets of muscle tissue samples from three individuals representing each species included in the present study were analyzed. In cases where there were less than three individuals of a given species, tissue samples from all individuals of that species were analyzed. 2.6. Fish contaminant analysis Contaminant analysis of fish-muscle tissue samples was performed at the Michigan State University Diagnostic Center for Population and Animal Health (East Lansing, Michigan). Fish-tissue samples were analyzed for a toxic elements panel, a suite of chlorinated pesticides, and PCBs (Table SI1). From this more comprehensive analysis, elements such as THg and Se were selected for statistical analysis based on relevancy given overall toxicity of the element and propensity of these contaminants to bioaccumulate, along with the chlorinated pesticides aldrin and dieldrin due to their frequency of detection in our samples. PCB burdens were considered cumulatively, not by representative congeners. PCB and pesticide body burdens were lipid normalized by species before any further analysis to account for differences in fat content, and thus differences in contaminant storage capabilities of each species. Fish with burdens below detection limits were assigned a value at half the detection limit. Detection limits were 1 ng g1 for aldrin and dieldrin, and 10 ng g1 for PCBs. Se and THg body burdens in fish were analyzed using Inductively Coupled Plasma Mass Spectroscopy (ICP/MS) and reported as wet-weight concentrations (mg g1) (Lehner et al., 2013; Perrault et al., 2014). A subset of fish positive for detectable levels of THg was analyzed for MeHg to investigate the contribution of MeHg to the THg body burdens in fish from this study. Samples were analyzed at ALS Environmental (Kelso, Washington) for MeHg using a procedure based on EPA method 1630 (US EPA, 1998), using a Brooks-Rand Model III gas chromatography cold-vapor atomicfluorescence spectrometer (Seattle, Washington). Analyses of chlorinated pesticides and PCBs were conducted on 1-g muscletissue samples using gas-chromatography tandem-quadrupole mass spectrometry (GC-MS/MS) as described by Bokhart et al. (2015) and reported as parts per billion (ng g1) wet weight.
2.7. Statistical analysis All statistical analyses were conducted in R version 3.3.1 (R Core Team, 2016). Prior to analysis, data sets were screened for outliers, as well as assumptions of homogeneity of variance and normality via box-and-whisker plots, q-q plots, and model residuals. PCB, aldrin, and dieldrin fish body burdens were log transformed to improve distributional properties of the data set. A value of one was added to PCB body burdens due to the presence of zeros in the data before log transformation (Quinn and Keough, 2002). THg was converted to presence-absence data because a majority of the samples were below detection limits, which can violate model assumptions (Croghan and Egeghy, 2003). All samples with measurable THg body burdens were considered positive detections, coded as a 1; all samples with amounts below detection limits were coded as a 0, resulting in a binomial distribution (hereafter, “THg Counts”). After initial processing, the data for each contaminant were partitioned by river system, given differences in the characteristics of the rivers and the resident fish communities. Owing to the potential lack of independence of study reaches, contaminant body burdens and THg sediment concentrations were assessed for spatial autocorrelation with Moran's I. Simple linear regression was used to investigate relationships between THg sediment concentrations and detection rates of THg in fish, FCL and mean contaminant body burdens in fish, trophic position and MeHg body burdens, contaminant body burdens and trophic position both for all reaches and time steps, as well as contaminant body burden and before/after, upstream/downstream and control/experimental reaches. Analysis of variance (ANOVA) paired with a Tukey's HSD post-hoc analysis was used to determine if differences existed between THg sediment concentrations by study reach, with years as internal replicates. A paired t-test was used to determine if sediment THg concentrations were different before and after dam removal, using reach means across time before and after dam removal as replicates. Generalized linear models (GZLMs) with a binomial distribution were used to assess the relationship between THg detections and trophic position. ANOVA was used to examine potential differences in slopes between treatments. Multiple regression, in a model-selection approach based on Akaike Information Criterion adjusted for small sample sizes (AICc) (Burnham and Anderson, 2002, 2004), was used to examine the influence of organic carbon content, streamflow velocity, and sediment size on sediment contaminant concentrations. Akaike weights were also calculated to assess the relative strength of the model within its set. Highly correlated variables (jrj > 0.70) were not included in the same model set (Allen and Vaughn, 2010; Burnham and Anderson, 2002). To assess the potential differences in contaminant body burdens in fish over time and across study reaches, we used linear mixed models by restricted maximum likelihood with the package “lme4” (Bates et al., 2015) in R. Time, reach, and their interaction were included as fixed effects with a categorical trophic level variable assigned as a random effect. Here, we assigned trophic position to bins at the nearest half trophic step to create a categorical trophic-level variable (e.g., trophic position of 3.2 assigned to 3.0, trophic position of 3.4 assigned to 3.5) to account for using individuals of different trophic levels from each reach. The marginal and conditional R2 (Nakagawa and Schielzeth, 2013) were calculated in the “MuMIn” package (Barton, 2016), which describe the amount of variation that can be explained by the fixed effects alone (marginal) and the fixed and random effects combined (conditional). Simple and orthogonal contrasts were then used to evaluate specific treatment comparisons (e.g., restored/unrestored, upstream/downstream) through time using “multcomp” in R (Hothorn et al., 2008). For THg, lack of fit due to
R.P. Davis et al. / Environmental Pollution 231 (2017) 671e680
overdispersion (too many 0 values) made the models untenable. Instead, we calculated detection rates of THg body burdens in fish sampled at each reach by year combination and compared pre-dam removal rates to post-dam removal rates using paired t-tests, and compared study reaches (with years as internal replicates) using analysis of variance (ANOVA). In most cases, significance was determined as p < 0.05, and evidence of a trend as p < 0.10 (e.g., Sullivan et al., 2016). However, in the cases of multiple tests (e.g., some of the simple linear regressions and ANOVAs), significance was determined using a Bonferroni correction (Bonferroni, 1936; Frane, 2015); corrected p values are listed in relevant figure captions. 3. Results 3.1. Sediment contaminants Across all study reaches and time steps, a total of 42 sediment samples were analyzed for PCBs and chlorinated pesticides; 114 sediment samples were analyzed for THg. Detectable levels of PCB congeners (limit of detection [LOD] ¼ 10 ng g1) were found in only 9.5% of sediment samples, with concentrations averaging 76.18 ng g1 (SD ¼ 48.30 ng g1). Aldrin and dieldrin were below detection in all samples (LOD ¼ 1 ng g1). DDT metabolites were found in 69% of sediment samples and chlordane metabolites in 88% of samples. Detectable levels of THg were found in all sediment samples and ranged from 16.36 to 397.02 ng g1 (x ¼ 64.62 ng g1, SD ¼ 59.096 ng g1, Table SI2). THg sediment concentrations were found to differ by reach (ANOVA: F ¼ 21.46, p ¼ 0.05). In general, contaminant concentrations decreased through time and converged in the last year of the study. Mean aldrin and dieldrin body burdens varied through time in both the Olentangy (linear mixed model: aldrin e F ¼ 134.70, p < 0.001; dieldrin e F ¼ 63.38, p < 0.001) and Scioto (aldrin e F ¼ 334.74, p < 0.001; dieldrin e F ¼ 137.19, p < 0.001) Rivers, and generally decreased across all reaches in the 2e3 years following dam removal (Table 1). Post-hoc contrasts revealed that aldrin body burdens significantly decreased in the last year across all reaches (Table SI4). In the Olentangy River, both aldrin and dieldrin body burdens were significantly lower in the upstream restored reach after vs. before dam removal (post-hoc contrasts: aldrin p < 0.001, dieldrin, p < 0.001, Fig. 2), and were significantly lower (aldrin, p ¼ 0.021, dieldrin p ¼ 0.033) in the upstream restored reach compared to the upstream unrestored reach (Fig. SI4). Mean PCB body burdens were relatively unchanged throughout the course of the study (Fig. 2). Se body burdens increased in the Scioto River (linear mixed model: F ¼ 5.06, p ¼ 0.010; Fig. 2) through time, but patterns were less evident in the Olentangy River (Fig. 2). THg body burden detection rates were not significantly different pre and post dam removal (ttest: t ¼ 0.961, p ¼ 0.359) or among reaches across study years (ANOVA: F ¼ 0.59, p ¼ 0.737, df ¼ 6,15). 3.3. Fish body burdens and trophic dynamics Fish surveyed represented five feeding guilds and spanned trophic positions 1.77 to 4.20. In the Olentangy River, dieldrin body burdens exhibited a positive relationship with trophic position (R2 ¼ 0.21, F ¼ 12.71, p < 0.001, n ¼ 50, Fig. SI5). In the Scioto River, Se body burdens were negatively related to trophic position (R2 ¼ 0.21, F ¼ 15.04, p < 0.001, n ¼ 58, Fig. SI5) and a positive trend was evident between trophic position and PCB concentrations (R2 ¼ 0.30, F ¼ 10.75, p ¼ 0.003, n ¼ 27; Fig. SI5). In the Olentangy River, relationships between contaminant body burdens and trophic position were stronger before dam removal (e.g., Dieldrin, R2 ¼ 0.76 before dam removal, R2 ¼ 0.15 after removal), however the interactions were not significant with the exception of selenium (Table SI5). Aldrin (R2 ¼ 0.84, F ¼ 31.6 p ¼ 0.001, n ¼ 8) and dieldrin (R2 ¼ 0.88, F ¼ 44.1 p < 0.001, n ¼ 8) body burdens exhibited significant relationships in the upstream reach of the Scioto River (Fig. SI6). System-wide, the proportion of MeHg also increased with trophic position (R2 ¼ 0.24, F ¼ 6.34 p ¼ 0.002, n ¼ 22) (Fig. SI7). FCL ranged from 2.5 to 4.3 across study reaches, with the longest food chain at OR3 before dam removal (4.3). FCLs were the shortest in the Olentangy River following dam removal: 2.5 (OR1, year 2), 2.6 and 2.7 (OR2, years 2 and 3, respectively), and 2.6 (OR3 year 1) (Dorobek, 2016). Although there was evidence of a trend between mean aldrin body burdens and FCL across all reaches and times (R2 ¼ 0.33, F ¼ 8.24 p ¼ 0.011, n ¼ 19; Fig. 3), FCL was not a significant predictor of PCBs, dieldrin, or Se body burdens, or THg detection rates (p > 0.02; Fig. 3). 4. Discussion The objective of this study was to determine exposure of fish communities to a suite of contaminants following lowhead dam removal, an important consideration in both dam-removal planning and management activities. Fish contaminant body burdens are often related to the level of contamination in the sediment of their environment (Madenjian et al., 2009), and were used in this study because they integrate exposure over space through their daily foraging and movement patterns (Gewurtz et al., 2011). In the
676
R.P. Davis et al. / Environmental Pollution 231 (2017) 671e680
Table 1 Results from linear mixed models of fish body burdens with Study Reach, Year, and Study ReachYear interaction as fixed effects. * indicates a significant effect (p < 0.05). PCBs ¼ polychlorinated biphenyls, Se ¼ selenium. n
R2m
R2c
Study Reach
Year
Study ReachYear
F
df
p
F
df
p
F
df
p
Olentangy River PCBs 50 Aldrin 51 Dieldrin 51 Se 82
0.40 0.94 0.90 0.23
0.40 0.95 0.90 0.23
5.52 13.07 4.96 2.40
4,37 3,37 3,37 4,69
0.003*
Contents lists available at ScienceDirect
Environmental Pollution journal homepage: www.elsevier.com/locate/envpol
Reductions in fish-community contamination following lowhead dam removal linked more to shifts in food-web structure than sediment pollution* Robert P. Davis*, S. Ma zeika P. Sullivan, Kay C. Stefanik School of Environment and Natural Resources, The Ohio State University, 2021 Coffey Road, Columbus, OH 43210, United States
a r t i c l e i n f o
a b s t r a c t
Article history: Received 30 March 2017 Received in revised form 9 July 2017
Recent increases in dam removals have prompted research on ecological and geomorphic river responses, yet contaminant dynamics following dam removals are poorly understood. We investigated changes in sediment concentrations and fish-community body burdens of mercury (Hg), selenium (Se), polychlorinated biphenyls (PCB), and chlorinated pesticides before and after two lowhead dam removals in the Scioto and Olentangy Rivers (Columbus, Ohio). These changes were then related to documented shifts in fish food-web structure. Seven study reaches were surveyed from 2011 to 2015, including controls, upstream and downstream of the previous dams, and upstream restored vs. unrestored. For most contaminants, fish-community body burdens declined following dam removal and converged across study reaches by the last year of the study in both rivers. Aldrin and dieldrin body burdens in the Olentangy River declined more rapidly in the upstream-restored vs. the upstream-unrestored reach, but were indistinguishable by year three post dam removal. No upstream-downstream differences were observed in body burdens in the Olentangy River, but aldrin and dieldrin body burdens were 138 and 148% higher, respectively, in downstream reaches than in upstream reaches of the Scioto River following dam removal. The strongest relationships between trophic position and body burdens were observed with PCBs and Se in the Scioto River, and with dieldrin in the Olentangy River. Food-chain length e a key measure of trophic structure e was only weakly related to aldrin body burdens, and unrelated to other contaminants. Overall, we demonstrate that lowhead dam removal may effectively reduce ecosystem contamination, largely via shifts in fish food-web dynamics versus sediment contaminant concentrations. This study presents some of the first findings documenting ecosystem contamination following dam removal and will be useful in informing future dam removals. © 2017 Elsevier Ltd. All rights reserved.
Keywords: Ecosystem contamination Biomonitoring Persistent contaminants Restoration Lowhead dam removal
1. Introduction Dams serve many functions including provisioning hydropower, storing water, controlling floods, and providing recreational opportunities (Born et al., 1998; Graf, 1999). Lowhead, run of river dams (43,000; US Army Corp of Engineers, 2013). Removal of these dams is growing in popularity, as many have become obsolete, expensive to maintain, or dysfunctional (Bednarek, 2001; Doyle et al., 2005; Harris and Evans, 2014). To date, >1000 dams have been removed
*
This paper has been recommended for acceptance by Maria Cristina Fossi. * Corresponding author. E-mail address: [email protected] (R.P. Davis).
http://dx.doi.org/10.1016/j.envpol.2017.07.096 0269-7491/© 2017 Elsevier Ltd. All rights reserved.
in the United States alone (O'Connor et al., 2015). The ecological and geomorphic effects of large dams on rivers are well documented (Freedman et al., 2014; Poff et al., 2007; Pringle et al., 2000); increasing attention has also recently been afforded to lowhead dams (Csiki and Rhoads, 2014; Fencl et al., 2015; Mbaka and Mwaniki, 2015). Studies suggest that fish communities, in particular, experience significant shifts in diversity, community structure, and food-web architecture following lowhead-dam removal (Dorobek et al., 2015; Dorobek, 2016; Kornis et al., 2015). Prior to dam removal, reservoirs created by impoundments behind dams have been shown to alter pH, streamflow, sediment load, and temperature (Fairchild and Velinsky, 2006; Poff and Hart, 2002; Santucci et al., 2005). As streamflow velocity, discharge, and boundary shear stress decrease behind the dam, fine suspended sediments settle out of the water column, along with contaminants
672
R.P. Davis et al. / Environmental Pollution 231 (2017) 671e680
that readily bind to sediments and the organic compounds they contain (An et al., 2016; Green et al., 2016; Syvitski et al., 2005), potentially leading to elevated contaminant levels behind dams (Gao et al., 1998; Warren et al., 2003). Contaminants of particular concern include chlorinated pesticides, polychlorinated biphenyls (PCBs), mercury (Hg), and selenium (Se). Although some of these contaminants e such as chlorinated pesticides (e.g., DDT) and PCBs e have not seen widespread use since the 1970's, they can persist in aquatic sediments (Evans and Gottgens 2007; Hinck et al., 2008; Johnson et al., 2013) by binding to benthic organic material (Warren et al., 2003). The potential for dam removal to release contaminated sediments is a major ecological concern (Bednarek, 2001; Green et al., 2016). In the case of large dams, removal has resulted in up to 70% of the former reservoir sediment transported downstream in the initial years (Grant and Lewis, 2015; O'Connor et al., 2015; Sawaske and Freyberg, 2012), potentially transporting contaminants downstream as well (Shuman, 1995; Stanley and Doyle, 2003). As sediments are resuspended during downstream transport, contaminants may remobilize and become bioavailable to fish and other aquatic organisms. Generally, higher localized contaminant concentrations lead to elevated fish-community contaminant burdens (Madenjian et al., 2009; Odhiambo et al., 2013). However, few examples of changes in contaminant burdens in fish communities exist with regards to dam removal, in spite of documented or predicted changes in sediment dynamics (Hart et al., 2002; Poff and Hart, 2002; Stanley and Doyle, 2003). Legacy contaminants can biomagnify (i.e., the multiplicative increase in contaminant body burdens up the food chain (Miranda et al., 2008)); and remain stored in fatty tissues of fish (Borga et al., 2012). Thus, contaminant body burdens in fish are commonly associated with fish food-web properties (Bentzen et al., 1996; Rasmussen et al., 1990). Trophic position, and in particular, foodchain length (FCL) e the maximum number of energetic transfers in a food chain e can be strongly related to biomagnification in fish, whereby individuals feeding at higher trophic positions exhibit greater contaminant body burdens than those feeding at lower trophic positions (Houde et al., 2008; Schmitt et al., 2011; van der Velden et al., 2013). Of fundamental importance to the biomagnification process, FCL can be altered by the insertion or deletion of consumers feeding at distinct trophic levels (Post et al., 2000; Sabo et al., 2010), thereby creating or removing trophic steps. Hoeinghaus et al. (2008) found that FCL varied by habitat type, with impounded river reaches having the longest food chains, which they attributed to the insertion of trophic interactions occurring below top predators. Conversely, Kautza and Sullivan (2016) observed that FCL in impounded reaches of the Scioto River system was shorter (x ¼ 3.88) than free flowing reaches (x ¼ 4.19), owing to the exclusion of top predators at the impounded sites. Here, we measured concentrations of a suite of common environmental contaminants in sediment and fish communities before, as well as multiple years following the removal of two dams in two rivers of Columbus, Ohio. Previous research in this system has shown strong fish-community and food-web responses to the presence and subsequent removal of dams (see Dorobek, 2016; Dorobek et al., 2015 for details), along with downstream transport of fine sediment, suggesting that contaminant body burdens in fish following dam removal would reflect both changes in contaminant availability due to sediment release, as well as the shifting trophic composition of the fish community. Within this context, we hypothesized that changes in fishcommunity body burdens would be related to changes in contaminant concentrations in sediment, as well as fish trophic positions through time and space. Specifically, we predicted that:
(1) Contaminant concentrations in sediment would be a function of sediment grain size, organic carbon content, and streamflow velocity (which is related to flow dynamics such as boundary shear stress that influence sediment size) (Jones and de Voogt, 1999; Miller and Orbock Miller, 2007; Selin, 2009; Warren et al., 2003; Zhao et al., 2010); (2) Fish contaminant body burdens would be positively correlated with trophic position, but this relationship would be stronger before dam removal and at upstream, reservoir reaches, where greater sediment concentrations of sequestered contaminants and more predatory fish species were found; (3) Contaminant body burdens in fish would decrease in reaches where there was a loss of large predatory fish occupying high trophic levels (Dorobek, 2016); and (4) In general, contaminant body burdens in fish would be the most invariant in downstream reaches through time following dam removal, where anticipated shifts in sediment contaminant concentrations (Ashley et al., 2006; Cantwell et al., 2014) and changes in the trophic composition of fish would be minimal (Dorobek, 2016). We anticipate these findings will help inform future dam removal efforts relative to their potential impacts on contamination of resident fish communities and drivers of contaminant transfer. 2. Methods 2.1. Study reaches and experimental design Seven, 500-m river reaches were surveyed in the Olentangy and Scioto Rivers (Fig. 1) in Columbus, Ohio using a modified BeforeAfter-Control-Impact design (Kibler et al., 2011a; Stewart-Oaten et al., 1986). The Olentangy River is a 150-km, 5th-order tributary of the 370-km, 6th-order Scioto River, in the Ohio River basin. Study reaches were located upstream and downstream of two channelspanning dams: the 5th Avenue Dam (2.5 m high) in the Olentangy River and the Main Street Dam (4.1 m high) in the Scioto River. The 5th Avenue Dam was removed during low-flow conditions in late summer of 2012. Approximately 1600 m3 of sediment was removed from behind the dam and was used for the construction of riparian and wetland habitat surrounding the newly designed river channel (DLZ, 2015). The Main Street Dam was removed in November 2013. Prior to removal of the Main Street Dam, the Ohio EPA conducted an analysis of metals, pesticides, and PCBs within the sediment and found that only arsenic exceeded Residential Criteria values (Ohio EPA, 2011). Study reaches from both rivers were assigned to various treatments (e.g., upstream/downstream, upstream restored/unrestored, before/after) to assess the impacts of dam removal and subsequent restoration efforts on contaminant body burdens in fish communities. Control reaches were designated “upstream control” and “downstream control” and represent both impounded (OR1) and free-flowing (SR3) reaches. The upstream control reach (OR1, Fig. 1) is behind an intact dam of comparable size and age to the removed 5th Avenue dam; the downstream control reach (SR3, Fig. 1) is also separated from other upstream reaches by an intact lowhead dam. All other reaches were designated as “experimental”. In the Olentangy River, a 2.6-km section of the river immediately upstream of the previous dam was restored from August 2013 through winter 2014, and included channel engineering (using heavy machinery) to create heterogeneous in-stream habitat (e.g., pools and riffles), developing and reconnecting floodplain wetlands, and planting riparian vegetation (Ohio EPA, 2011). Further upstream was left unmodified. We surveyed one upstream reach in the restored section (OR3, upstream restored; Fig. 1) and one upstream reach in the unmodified section (OR2, upstream unrestored; Fig. 1). Study reaches in the Scioto River were located upstream and downstream of the Main St. Dam in downtown Columbus. Channel restoration
R.P. Davis et al. / Environmental Pollution 231 (2017) 671e680
673
Fig. 1. Study reaches on the Olentangy and Scioto Rivers in Columbus, Ohio (USA). Reaches in the Olentangy River (OR 1e4) were located upstream and downstream of the 5th Avenue Dam, removed in August of 2012. Reaches in the Scioto River (SR 1e3) were located upstream and downstream of the Main Street Dam, removed in November 2013.
was not undertaken after this dam removal; however riverbanks were modified and reconstructed to develop pubically accessible greenspaces. We recognize that replication of both control and experimental reaches would be ideal, however the designs of dam-removal studies are constrained by multiple factors. Our study features three experimental reaches in the Olentangy River and two in the Scioto River, control reaches in each river, and data collected before and after dam removal spanning multiple years. This design is on par or exceeds the designs of other published dam-removal studies (e.g., Claeson and Coffin, 2016; Kornis et al., 2015; Tullos et al., 2014). Although we recognize the methodological limitations, important information can be gleaned from this study, where we used a mixture of traditional statistical approaches in combination with the practical significance (sensu Kibler et al., 2011b) of our results to evaluate our hypotheses.
of the river reach being sampled. Sampling effort was based on predam removal efforts of 600 s per sampling event and adjusted to accommodate changes in habitat volume, in congruence with other dam-removal studies (Catalano et al., 2007; Maloney et al., 2008), by reducing sampling time concomitant with reductions in habitat volume. Three to five individuals of fish species common to multiple reaches and representing major trophic guilds (e.g., common carp as benthic omnivores) were dispatched and transported on ice to the laboratory where they were frozen at 30 C until processing. For contaminant and lipid analysis, approximately 2 g-plugs of dorsal muscle were removed and frozen until further analysis. Trophic position and FCL estimates, based on analysis of 15N and 13C natural-abundance isotopes, were generated from a companion study of the same study reaches, sampling locations, and time periods as reported in Dorobek (2016).
2.2. Fish sampling
2.3. Sediment analysis
Fish surveys were conducted from 2011 to 2015 during stable, base flows in late summer-early autumn. Control reaches (OR1 and SR3) as well as experimental reaches (OR3, SR1, and SR2) were sampled prior to dam removal. Sampling occurred across all study reaches two years following dam removal in the Scioto River and three years following dam removal in the Olentangy River. Each study reach was stratified into three transects (upstream, middle, downstream; running across the stream). At each transect, fish communities were surveyed along three upstream-downstream sections (right bank, middle, left bank; centered along transect), for a total of nine sampling locations per reach. Fish were collected via electrofishing using a Smith Root® LR-24 backpack electrofisher, 2.5 GPP Smith-Root® long-line electrofisher, or a 5 GPP SmithRoot® boat unit (Vancouver, Washington). Electrofishing units were selected based on physical characteristics (e.g., depth, conductivity)
Composite surface sediment samples were collected from each reach before and following dam removal annually in the Scioto River and 2e3 times per summer in the Olentangy River (Alberts et al., 2013; Walters et al., 2010) using a hand trowel. Sediment samples were refrigerated at ~5 C until processing for analysis. Prior to analysis, composite samples were homogenized. Sediment was analyzed at the Service Testing and Research Lab (STAR Lab, Wooster, Ohio) for total carbon, inorganic and organic carbon fractions, and total Hg (hereafter “THg”) content. Total carbon and organic carbon were analyzed using The International Organization for Standardization (1995) ISO 10694 method and inorganic carbon was determined using EPA Method 9060A (US EPA, 2014). Sediment samples were analyzed at the Michigan State University Diagnostic Center for Population and Animal Health for PCBs and chlorinated pesticides via gas-chromatography tandem-quadrupole mass
674
R.P. Davis et al. / Environmental Pollution 231 (2017) 671e680
spectrometry (GC-MS/MS). THg content was analyzed using cold vapor atomic fluorescence with a CETAC M8000 Mercury Analyzer (CETAC Technologies, Omaha, Nebraska). Mean sediment sizes reported are from Comes (2016), collected using a McNeil Sampler. 2.4. Streamflow velocity As a proxy for boundary shear stress, streamflow velocity measurements were calculated for each stream reach using an Acoustic Doppler Current Profiler (ADCP) (M9 RiverSurveyor; SonTek, San Diego, California). Discharge, a function of streamflow velocity and cross-sectional area of a channel, has a non-linear relationship with boundary shear stress, with boundary shear stress generally increasing with discharge (Pitlick and Wilcock, 2001). Data-collection methods are detailed in Comes (2016). Transects within the processed data were selected at what was approximately the upstream, middle, and downstream of each reach and velocity was calculated using the summary statistic tools in ArcGIS. Velocities were calculated at all collection events in Comes (2016), and averaged by transect. 2.5. Lipid analysis Lipid extractions were performed to determine total lipid content of fish muscle tissue at ALS Environmental Laboratories (Kelso, Washington) following protocols from Bligh and Dyer (1959). Lipid percentage was determined on a wet-weight basis. Subsets of muscle tissue samples from three individuals representing each species included in the present study were analyzed. In cases where there were less than three individuals of a given species, tissue samples from all individuals of that species were analyzed. 2.6. Fish contaminant analysis Contaminant analysis of fish-muscle tissue samples was performed at the Michigan State University Diagnostic Center for Population and Animal Health (East Lansing, Michigan). Fish-tissue samples were analyzed for a toxic elements panel, a suite of chlorinated pesticides, and PCBs (Table SI1). From this more comprehensive analysis, elements such as THg and Se were selected for statistical analysis based on relevancy given overall toxicity of the element and propensity of these contaminants to bioaccumulate, along with the chlorinated pesticides aldrin and dieldrin due to their frequency of detection in our samples. PCB burdens were considered cumulatively, not by representative congeners. PCB and pesticide body burdens were lipid normalized by species before any further analysis to account for differences in fat content, and thus differences in contaminant storage capabilities of each species. Fish with burdens below detection limits were assigned a value at half the detection limit. Detection limits were 1 ng g1 for aldrin and dieldrin, and 10 ng g1 for PCBs. Se and THg body burdens in fish were analyzed using Inductively Coupled Plasma Mass Spectroscopy (ICP/MS) and reported as wet-weight concentrations (mg g1) (Lehner et al., 2013; Perrault et al., 2014). A subset of fish positive for detectable levels of THg was analyzed for MeHg to investigate the contribution of MeHg to the THg body burdens in fish from this study. Samples were analyzed at ALS Environmental (Kelso, Washington) for MeHg using a procedure based on EPA method 1630 (US EPA, 1998), using a Brooks-Rand Model III gas chromatography cold-vapor atomicfluorescence spectrometer (Seattle, Washington). Analyses of chlorinated pesticides and PCBs were conducted on 1-g muscletissue samples using gas-chromatography tandem-quadrupole mass spectrometry (GC-MS/MS) as described by Bokhart et al. (2015) and reported as parts per billion (ng g1) wet weight.
2.7. Statistical analysis All statistical analyses were conducted in R version 3.3.1 (R Core Team, 2016). Prior to analysis, data sets were screened for outliers, as well as assumptions of homogeneity of variance and normality via box-and-whisker plots, q-q plots, and model residuals. PCB, aldrin, and dieldrin fish body burdens were log transformed to improve distributional properties of the data set. A value of one was added to PCB body burdens due to the presence of zeros in the data before log transformation (Quinn and Keough, 2002). THg was converted to presence-absence data because a majority of the samples were below detection limits, which can violate model assumptions (Croghan and Egeghy, 2003). All samples with measurable THg body burdens were considered positive detections, coded as a 1; all samples with amounts below detection limits were coded as a 0, resulting in a binomial distribution (hereafter, “THg Counts”). After initial processing, the data for each contaminant were partitioned by river system, given differences in the characteristics of the rivers and the resident fish communities. Owing to the potential lack of independence of study reaches, contaminant body burdens and THg sediment concentrations were assessed for spatial autocorrelation with Moran's I. Simple linear regression was used to investigate relationships between THg sediment concentrations and detection rates of THg in fish, FCL and mean contaminant body burdens in fish, trophic position and MeHg body burdens, contaminant body burdens and trophic position both for all reaches and time steps, as well as contaminant body burden and before/after, upstream/downstream and control/experimental reaches. Analysis of variance (ANOVA) paired with a Tukey's HSD post-hoc analysis was used to determine if differences existed between THg sediment concentrations by study reach, with years as internal replicates. A paired t-test was used to determine if sediment THg concentrations were different before and after dam removal, using reach means across time before and after dam removal as replicates. Generalized linear models (GZLMs) with a binomial distribution were used to assess the relationship between THg detections and trophic position. ANOVA was used to examine potential differences in slopes between treatments. Multiple regression, in a model-selection approach based on Akaike Information Criterion adjusted for small sample sizes (AICc) (Burnham and Anderson, 2002, 2004), was used to examine the influence of organic carbon content, streamflow velocity, and sediment size on sediment contaminant concentrations. Akaike weights were also calculated to assess the relative strength of the model within its set. Highly correlated variables (jrj > 0.70) were not included in the same model set (Allen and Vaughn, 2010; Burnham and Anderson, 2002). To assess the potential differences in contaminant body burdens in fish over time and across study reaches, we used linear mixed models by restricted maximum likelihood with the package “lme4” (Bates et al., 2015) in R. Time, reach, and their interaction were included as fixed effects with a categorical trophic level variable assigned as a random effect. Here, we assigned trophic position to bins at the nearest half trophic step to create a categorical trophic-level variable (e.g., trophic position of 3.2 assigned to 3.0, trophic position of 3.4 assigned to 3.5) to account for using individuals of different trophic levels from each reach. The marginal and conditional R2 (Nakagawa and Schielzeth, 2013) were calculated in the “MuMIn” package (Barton, 2016), which describe the amount of variation that can be explained by the fixed effects alone (marginal) and the fixed and random effects combined (conditional). Simple and orthogonal contrasts were then used to evaluate specific treatment comparisons (e.g., restored/unrestored, upstream/downstream) through time using “multcomp” in R (Hothorn et al., 2008). For THg, lack of fit due to
R.P. Davis et al. / Environmental Pollution 231 (2017) 671e680
overdispersion (too many 0 values) made the models untenable. Instead, we calculated detection rates of THg body burdens in fish sampled at each reach by year combination and compared pre-dam removal rates to post-dam removal rates using paired t-tests, and compared study reaches (with years as internal replicates) using analysis of variance (ANOVA). In most cases, significance was determined as p < 0.05, and evidence of a trend as p < 0.10 (e.g., Sullivan et al., 2016). However, in the cases of multiple tests (e.g., some of the simple linear regressions and ANOVAs), significance was determined using a Bonferroni correction (Bonferroni, 1936; Frane, 2015); corrected p values are listed in relevant figure captions. 3. Results 3.1. Sediment contaminants Across all study reaches and time steps, a total of 42 sediment samples were analyzed for PCBs and chlorinated pesticides; 114 sediment samples were analyzed for THg. Detectable levels of PCB congeners (limit of detection [LOD] ¼ 10 ng g1) were found in only 9.5% of sediment samples, with concentrations averaging 76.18 ng g1 (SD ¼ 48.30 ng g1). Aldrin and dieldrin were below detection in all samples (LOD ¼ 1 ng g1). DDT metabolites were found in 69% of sediment samples and chlordane metabolites in 88% of samples. Detectable levels of THg were found in all sediment samples and ranged from 16.36 to 397.02 ng g1 (x ¼ 64.62 ng g1, SD ¼ 59.096 ng g1, Table SI2). THg sediment concentrations were found to differ by reach (ANOVA: F ¼ 21.46, p ¼ 0.05). In general, contaminant concentrations decreased through time and converged in the last year of the study. Mean aldrin and dieldrin body burdens varied through time in both the Olentangy (linear mixed model: aldrin e F ¼ 134.70, p < 0.001; dieldrin e F ¼ 63.38, p < 0.001) and Scioto (aldrin e F ¼ 334.74, p < 0.001; dieldrin e F ¼ 137.19, p < 0.001) Rivers, and generally decreased across all reaches in the 2e3 years following dam removal (Table 1). Post-hoc contrasts revealed that aldrin body burdens significantly decreased in the last year across all reaches (Table SI4). In the Olentangy River, both aldrin and dieldrin body burdens were significantly lower in the upstream restored reach after vs. before dam removal (post-hoc contrasts: aldrin p < 0.001, dieldrin, p < 0.001, Fig. 2), and were significantly lower (aldrin, p ¼ 0.021, dieldrin p ¼ 0.033) in the upstream restored reach compared to the upstream unrestored reach (Fig. SI4). Mean PCB body burdens were relatively unchanged throughout the course of the study (Fig. 2). Se body burdens increased in the Scioto River (linear mixed model: F ¼ 5.06, p ¼ 0.010; Fig. 2) through time, but patterns were less evident in the Olentangy River (Fig. 2). THg body burden detection rates were not significantly different pre and post dam removal (ttest: t ¼ 0.961, p ¼ 0.359) or among reaches across study years (ANOVA: F ¼ 0.59, p ¼ 0.737, df ¼ 6,15). 3.3. Fish body burdens and trophic dynamics Fish surveyed represented five feeding guilds and spanned trophic positions 1.77 to 4.20. In the Olentangy River, dieldrin body burdens exhibited a positive relationship with trophic position (R2 ¼ 0.21, F ¼ 12.71, p < 0.001, n ¼ 50, Fig. SI5). In the Scioto River, Se body burdens were negatively related to trophic position (R2 ¼ 0.21, F ¼ 15.04, p < 0.001, n ¼ 58, Fig. SI5) and a positive trend was evident between trophic position and PCB concentrations (R2 ¼ 0.30, F ¼ 10.75, p ¼ 0.003, n ¼ 27; Fig. SI5). In the Olentangy River, relationships between contaminant body burdens and trophic position were stronger before dam removal (e.g., Dieldrin, R2 ¼ 0.76 before dam removal, R2 ¼ 0.15 after removal), however the interactions were not significant with the exception of selenium (Table SI5). Aldrin (R2 ¼ 0.84, F ¼ 31.6 p ¼ 0.001, n ¼ 8) and dieldrin (R2 ¼ 0.88, F ¼ 44.1 p < 0.001, n ¼ 8) body burdens exhibited significant relationships in the upstream reach of the Scioto River (Fig. SI6). System-wide, the proportion of MeHg also increased with trophic position (R2 ¼ 0.24, F ¼ 6.34 p ¼ 0.002, n ¼ 22) (Fig. SI7). FCL ranged from 2.5 to 4.3 across study reaches, with the longest food chain at OR3 before dam removal (4.3). FCLs were the shortest in the Olentangy River following dam removal: 2.5 (OR1, year 2), 2.6 and 2.7 (OR2, years 2 and 3, respectively), and 2.6 (OR3 year 1) (Dorobek, 2016). Although there was evidence of a trend between mean aldrin body burdens and FCL across all reaches and times (R2 ¼ 0.33, F ¼ 8.24 p ¼ 0.011, n ¼ 19; Fig. 3), FCL was not a significant predictor of PCBs, dieldrin, or Se body burdens, or THg detection rates (p > 0.02; Fig. 3). 4. Discussion The objective of this study was to determine exposure of fish communities to a suite of contaminants following lowhead dam removal, an important consideration in both dam-removal planning and management activities. Fish contaminant body burdens are often related to the level of contamination in the sediment of their environment (Madenjian et al., 2009), and were used in this study because they integrate exposure over space through their daily foraging and movement patterns (Gewurtz et al., 2011). In the
676
R.P. Davis et al. / Environmental Pollution 231 (2017) 671e680
Table 1 Results from linear mixed models of fish body burdens with Study Reach, Year, and Study ReachYear interaction as fixed effects. * indicates a significant effect (p < 0.05). PCBs ¼ polychlorinated biphenyls, Se ¼ selenium. n
R2m
R2c
Study Reach
Year
Study ReachYear
F
df
p
F
df
p
F
df
p
Olentangy River PCBs 50 Aldrin 51 Dieldrin 51 Se 82
0.40 0.94 0.90 0.23
0.40 0.95 0.90 0.23
5.52 13.07 4.96 2.40
4,37 3,37 3,37 4,69
0.003*
