Relationships between land use, predicted pollution loadings, and ecotoxicological assays in constructed wetlands

Environmental degradation related to uncontrolled development resulted in the passage of the United States Clean Water Act (CWA) in 1972, with the stated purpose “to restore and maintain the integrity of the nation’s waters”. Implementation of the CWA leads to increased research to develop multimetric indicators to better measure and understand the complex patterns of ecological responses to stress occurring across levels of biological, spatial, and temporal organization. One area of research is the use of integrated indices of chemical risk, ecotoxicological risk, and ecological risk to assess the impact of human activity across disturbance gradients of urbanization. Selecting relevant metrics for constructing a multimetric index requires identifying bioindicator organisms with capacities to detect signals from anthropogenic disturbances. This study explored the potential efficacy of a suite of higher plant ecotoxicological assays for use as bioindicators in ecological risk assessment along a gradient of urbanization in a wetland ecosystem. The study was conducted in the Pike River watershed (Racine, Wisconsin, USA) in six wetlands selected across a gradient of dominant land-use types (agricultural, commercial, residential, undeveloped, and industrial). MicroBioTest PhytotoxkitTM ecotoxicological assays, based on growth inhibition of three plants (Sinapis, Sorghum, and Lepidium) were used to assess sediment toxicity. The relationships between Phytotoxkit™ responses and predicted pollutant loadings calculated from surrounding land use provided clear signals of stress from watershed pollutants draining into the wetland sites. The potential for these ecotoxicological indicators to serve as biological response signatures is strong, and further research and calibration in field and microcosms studies will assist in calibrating responses for use in integrated monitoring efforts.


INTRODUCTION
Urbanization is one of the major drivers of degraded surface water quality [1], attributable to increasing impervious surface area contributing to higher stormwater runoff into local streams, rivers, lakes, and wetlands [2]. This runoff in turn carries increased concentrations and loadings of nutrient and heavy metal pollutants contributing to a deterioration in the quality of the receiving waters [3] (Fig. 1). Increased public awareness of the interconnections between changes in land cover and surface water quality contributed to the passing of the US Clean Water Act (CWA) in 1972, establishing quality standards for surface waters and setting limits for the discharge of pollutants and excess nutrients [4,5]. But CWA created a challenge for bridging the gap between the science of environmental monitoring, social-ecological domains of designated uses (e.g. fishable-swimmable), and the interconnected biogeochemical cycles affecting protective numerical criteria (e.g. phosphate standards) [6]. To this end, it became imperative to incorporate diverse disciplinary perspectives in the selection of metrics and indicators for use in monitoring programs [7]. The U.S. Environmental Protection Agency (USEPA) initially promoted the use of monitoring strategies that integrated metrics of water quality parameters, whole-effluent toxicity testing, and ambient biological assays [8]. This "3-legged stool" approach has proven to be limited in its capacity to characterize ecological integrity across diverse environmental contexts [9][10][11]. To develop more effective and robust monitoring strategies, the use of biological assays and bioindicators has increased steadily [8,9]. Concurrent with the increase in available tools and data, analytical approaches have increasingly focused on detecting "biological response signatures" [10] as a way to characterize the complex patterns of ecological responses to stress occurring across levels of biological, spatial, and temporal organization [7]. In a study, Yoder and Rankin used the term "biological response signatures" to describe the variety of ways that indicators in aquatic ecosystems may respond to different types of environmental stressors [10]. Their work reframes the question away from looking for distinct cause-effect relationships towards identifying signals of response amidst the complex noise of potential causes [12,13]. Ecological Risk Assessment approaches have been used extensively for both monitoring the effects of development (ex-post impact assessment) and predicting the likely effects of proposed projects (ex-ante impact assessment). To this end, constructed wetlands are often used to address water quantity and quality problems and mitigate the environmental impacts of historical urbanization and minimize the impact of new construction [14]. In addition to capturing sediment and pollutants that flow off surrounding landscapes [15], constructed stormwater wetlands can play a critical role in managing nutrients generated from agricultural and urban runoff [16][17][18]. Besides, constructed wetlands are effective in reducing heavy metal contamination generated from industrial sources [19][20][21].
Biomonitoring is measuring and evaluating the conditions of a living system [22] . Since the passage of CWA, biomonitoring has become an essential component for monitoring the ecological integrity and condition of watersheds [22] and bioindicators developed to serve as tools for assessing attainment of and adherence to water quality standards [10]. Bioindicators developed for wetland sediments are particularly sensitive in detecting ecological changes in watersheds [23,24] and for conducting sediment risk assessments from pollutants such as metals or nutrients [25][26][27]. routine evaluations as bioindicators [31,32]. The question addressed in this paper is, does variation in growth inhibition of PhytoTox™ ecotoxicological assays (Sorghum saccharatum, Lepidium sativum, and Sinapis alba) correlate with variation in pollution-related stressors as predicted loadings estimate that enter wetlands from their surrounding watersheds?

Study-system, land use, and site characteristics
This study was conducted in the Pike River watershed (Racine County, Wisconsin USA) utilizing a series of stormwater wetlands that were constructed between 2001 and 2008 as structural features in a flood-control plan implemented by the Village of Mount Pleasant. The plan included significant modifications in channel morphology, the creation of riparian wetland-pond systems, and the installation of fish habitat along an 8 km stretch of the river [33][34][35]. The wetlands were excavated to receive runoff from adjacent catchments that comprised of a combination of agricultural, commercial, residential, undeveloped, and industrial land uses [33][34][35]. Six individual wetlands were selected for this study to capture a gradient of dominant land cover types (Fig.  2). The catchment area and percent land use (residential, commercial, industrial, agricultural, and undeveloped) for each wetland were determined from Southeastern Wisconsin Regional Planning Commission or SEWRPC, 2010 Racine County map book [36] and are shown in Table 1.   Fig. 3 and 4). By default, agricultural lands were incorporated into the category of the undeveloped land in SLAMM, due to its use as an urban planning model [33,37]. Therefore, land classifications were manually re-coded to agricultural land uses by cross-comparison with the 2010 SEWRPC land cover data [36]. All land use measurements were converted from acres as provide by SLAMM to square meters. The percent of land uses was calculated with respect to the total area of the land cover draining into the wetlands. Values for each of the land use categories (residential, industrial, commercial, undeveloped, and agricultural) are the summation of the source subcategories (e.g. roofs, street area, parking, driveways, sidewalks and landscaped area) ( Fig. 3 and 4) [38].
Predictions for pollutant loadings (nitrate + nitrite, phosphate, Zn, Pb, Cu, and Cd) were estimated using the geometric mean of values measured from studies reported in the literature [39][40][41][42][43][44] by the source area subcategories (e.g. roofs, street area, parking, driveways, sidewalks, and landscaped area) of each land use category (residential, industrial, commercial, undeveloped and agricultural) ( Table 2). Due to the inadequacy of data the loading estimates of nitrate and metals like Ag, As, Hg, and Ni could not be calculated. Then the total pollutant loadings in Kg/year were calculated by multiplying the pollutant loading estimates from the literature by the source area (m 2 ) subcategories (e.g. roofs, street area, parking, driveways, sidewalks, and landscaped area) of each land use category (residential, industrial, commercial, undeveloped and agricultural) in a year. This produces the total pollutant loading at each wetland site by land use category (residential, industrial, commercial, undeveloped, and agricultural) in a year. These calculated loadings are shown in Figure 4 and Table 2. Each error bar is constructed using one standard error (± 1 SE) from the mean.

Data Analyses
Data distributions were examined for normality and were transformed as necessary to meet the assumptions of statistical tests. Count and length data were transformed using a log transformation (log10 (X + 1)) while proportional data were transformed using an arcsine transformation [44] before statistical analyses conducted using JMP® 14 [45].

Effect of predicted nutrients and metals on Growth Inhibition
Multifactor Analysis of variance (ANOVA) was used to examine the effects of predicted pollutants (nutrients and metals) loadings and seed species on growth inhibition of Lepidium sativum, Sinapis alba, and Sorghum saccharatum. Land use was assumed to not have changed significantly over the course of the study, and as such, the ANOVA tests for the effect of predicted loadings from the surrounding land use on the stem and root growth inhibitions (the dependent response variable) included loadings and seed species and year as independent variables. ANOVA tests for the effect of predicted loading of individual (Cd, Cu, Pb, Zn) metals from the surrounding land use on growth inhibition (the dependent response variable) included loadings and seed species and year as independent variables.

Ecotoxicological bioindicators
Proportion root and stem growth inhibition values are calculated relative to growth in control sediments (clean silica sand) so that positive values indicate inhibition (i.e. reduced growth = inhibition) whereas negative values indicate growth stimulation (i.e. increased growth = stimulation). For Lepidium sativum, root inhibition ranged from -1.5 to +1.5 and stem inhibition ranged from -0.75 to +1.25. For Sinapis alba, root inhibition varied from -1.5 to +1.25 and stem inhibition ranged from -1 to +1.25. For Sorghum saccharatum, the proportion root inhibition ranged from -1.5 to +1.25 and stem inhibition ranged from -3.5 to +1.5. Responses varied among wetland sites and between years. Sorghum exhibited consistently higher growth inhibition for both roots and stem across the study compared to the other two-bioindicator species (Fig. 5). Wetland 1 exhibited consistently the lowest inhibition (highest stimulation) values for Lepidium sativum and Sinapis alba, whereas wetlands 3 and 4 exhibited higher inhibition (Fig. 5). Ecotoxicological bioindicator, responses to pollution stress a. Nutrient Effects For predicted nutrient loading ANOVA models the dependent variables were the growth inhibitions and the independent x variables included predicted total nutrient loadings and seed species. This model initially considered the year as an independent variable but as no significant effect of this variable was observed, the year effect was not considered in the final model (Fig. 6). This final model detected no statistically significant effects of predicted nutrient loadings or the seed species on the root growth inhibition of Lepidium sativum, Sinapis alba, Sorghum saccharatum (Fig. 6, P-values: seed species = 0.5024, Nitrate and nitrite loading = 0.4916, phosphate loading = 0.8761, nutrient interaction = 0.9162). Although with an increase in the nitrate + nitrite and phosphate loading decrease in the root growth inhibition was observed. This suggests that the root inhibition was negatively affected by the predicted nutrient loadings. There were significant results for stem growth inhibition. There was a significant effect of the seed species (P<0.0001) with the highest inhibition in Sorghum saccharatum, nitrate and nitrite loading (P = 0.0041), and the nutrient interaction (P = 0.0116) on stem growth inhibition of Lepidium sativum, Sinapis alba, Sorghum saccharatum. However, the effect of phosphate loading was not significant for stem growth inhibition (P=0.0898) (Fig. 6). Stem growth inhibition was observed to be decreasing with nitrate and nitrite loading, but increasing with phosphate loading (Fig. 6). Implying the negative effect of nitrate and nitrite loading on the stem growth inhibitions.

Seed Species
Predicted total nutrient loading (Kg/year)

Nitrate+nitrite Phosphate
Nitrate+nitrite Phosphate Fig. 6. Prediction profiles from ANOVA showing the effects of seed species and the predicted total loadings of total nitrate + nitrite and phosphate (kg/year) on the growth inhibitions of stems and roots for the bioindicator species Lepidium sativum, Sinapis alba, Sorghum saccharatum. The bluelined area in each profile represents the 95% confidence prediction interval of the response variable.
The profiler is set for nitrate + nitrite at 1.96 kg/year, phosphate at 26.23 kg/year in case of root growth inhibition, and nitrate + nitrite at 10.91 kg/year, phosphate at 26.23 kg/year in case of stem growth inhibition.
b. Metal Effects For predicted metal loading ANOVA models the dependent variables were the growth inhibitions and the independent x variables included predicted total metal loadings and seed species. This model initially considered the year as an independent variable but as no significant effect of this variable was observed, the year effect was not considered in the final model (Fig. 7). The effects of heavy metal loadings predicted by the land cover on root inhibition were not statistically significant except for Pb ( Fig. 7). P-values for root inhibition: seed species = 0.4359, Cd loading = 0.3064, Cu loading = 0.9990. Pb loading = 0.0168, Zn loading = 0.6119, metal loading interactions = 0.4625). Decreased root inhibition (i.e. facilitated root growth) was associated with an increase in Pb loading (Figure 7). Likewise, the effects of heavy metal loadings predicted by land cover on stem inhibition were not significant (P-values for effect on stem inhibition: Cd loading = 0.3167, Cu loading = 0.6489, Pb loading = 0.1512, Zn loading = 0.9076, metal loading interaction = 0.4629). Suggesting that suggest that the Pb loading as predicted (in Kg/year) in these wetland sites (1-6) did not affect these three plant bioindicator species of Lepidium sativum, Sinapis alba, and Sorghum saccharatum negatively. Seed species responded differently to predicted metal loadings. For stem inhibition, the effect of seed species effect was significant (P <.0001) with highest inhibition observed in Sorghum saccharatum (Fig. 7). This study was designed to explore the potential of PhytoTox™ ecotoxicological tests to serve as possible bioindicators for predicted pollution loading from surrounding land uses for wetland ponds located in urbanizing watersheds. Agricultural and residential land uses both produce runoffs rich in nutrients such as phosphate and nitrate due to the presence of fertilizers and pesticides applied to lawns, gardens, and agricultural fields. These fertilizers and pesticides especially when rich in nutrients affect plant growth [46][47][48][49][50][51][52][53][54]. The predicted loadings were a constant measure over a while (a year). The results suggest a possibility of interactions between the loadings of nutrients (e.g. from fertilizers) and loadings of metals associated with these pesticides resulting in various levels responses from Lepidium sativum, Sinapis alba, Sorghum saccharatum such as root inhibition was negatively affected by the predicted nutrient loadings, the negative effect of nitrate + nitrite loading on the stem growth inhibitions but the Pb loading as predicted (in Kg/year) in these wetland sites (1-6) did not affect negatively. The factors contributing to the differing responses by the different ecotoxicological bioindicator plant species is grounds for further study. As in this paper, herbicides and metals are well-known to affect the growth and development of Sorghum saccharatum [49,54]. In comparison, however, Sinapis alba and Lepidium sativum frequently exhibited negative inhibition (stimulation) for root and stem growth in this paper.

CONCLUSIONS
One of the challenges for monitoring environmental impacts in terms of the Clean Water Act is to identify and develop indicators that can capture and integrate the effects of pollutants or stressors across various (sometimes mismatched) spatial and temporal scales. Chronic stressors such as baseline nutrient loading from agricultural fields provide fundamentally different signals to detect compared to acute events such as a manure spill or pesticide application whose detection by direct chemical measurement may be missed between monitoring sessions. The situation is made more complicated by the fact that interactions among different stressors in nature may result in complex response patterns that can result in the interpretation of the patterns detected being very context-dependent. The results of this paper provide signals of stress from watershed pollutants draining into the wetland sites, which should be further explored with real measurements in the wetland sites (1-6). A major character of a biological sub-metrics that it should be able to detect biological responses to human activities across different scales, these ecotoxicological bioindicators demonstrated evidence of stress across different spatial scales of six different wetlands. We estimated the chemical risk (nutrient and metals) with the use of plant ecotoxicological bioindicators. Our results detected a correlation in the ecotoxicological bioindicators with watershed pollutants that were predicted.