ARTICLE pubs.acs.org/est
Assessing Exposure to Transformation Products of Soil-Applied Organic Contaminants in Surface Water: Comparison of Model Predictions and Field Data Susanne Kern,†,‡ Heinz Singer,† Juliane Hollender,† Rene P. Schwarzenbach,‡ and Kathrin Fenner*,†,‡ † ‡
Eawag, Swiss Federal Institute of Aquatic Science and Technology, 8600 D€ubendorf, Switzerland Institute of Biogeochemistry and Pollutant Dynamics (IBP), ETH Zurich, 8092 Z€urich, Switzerland
bS Supporting Information ABSTRACT: Transformation products (TPs) of chemicals released to soil, for example, pesticides, are regularly detected in surface and groundwater with some TPs even dominating observed pesticide levels. Given the large number of TPs potentially formed in the environment, straightforward prioritization methods based on available data and simple, evaluative models are required to identify TPs with a high aquatic exposure potential. While different such methods exist, none of them has so far been systematically evaluated against field data. Using a dynamic multimedia, multispecies model for TP prioritization, we compared the predicted relative surface water exposure potential of pesticides and their TPs with experimental data for 16 pesticides and 46 TPs measured in a small river draining a Swiss agricultural catchment. Twenty TPs were determined quantitatively using solid-phase extraction liquid chromatography mass spectrometry (SPE-LC-MS/MS), whereas the remaining 26 TPs could only be detected qualitatively because of the lack of analytical reference standards. Accordingly, the two sets of TPs were used for quantitative and qualitative model evaluation, respectively. Quantitative comparison of predicted with measured surface water exposure ratios for 20 pairs of TPs and parent pesticides indicated agreement within a factor of 10, except for chloridazon-desphenyl and chloridazon-methyl-desphenyl. The latter two TPs were found to be present in elevated concentrations during baseflow conditions and in groundwater samples across Switzerland, pointing toward high concentrations in exfiltrating groundwater. A simple leaching relationship was shown to qualitatively agree with the observed baseflow concentrations and to thus be useful in identifying TPs for which the simple prioritization model might underestimate actual surface water concentrations. Application of the model to the 26 qualitatively analyzed TPs showed that most of those TPs categorized as exhibiting a high aquatic exposure potential could be confirmed to be present in the majority of water samples investigated. On the basis of these results, we propose a generally applicable, model-based approach to identify those TPs of soil-applied organic contaminants that exhibit a high aquatic exposure potential to prioritize them for higher-tier, experimental investigations.
’ INTRODUCTION Many biologically active compounds, such as pharmaceuticals and biocides contained in biosolids or pesticides used for plant protection, enter the environment through their application to agricultural soils. From soil, these chemicals are transported to varying degrees into surface waters where they can negatively impact aquatic organisms. Transport pathways into surface waters include fast processes, such as surface runoff or macropore flow into drainage systems, following intense rain events,1-3 but also slow, more steady inputs from contaminants that have leached into groundwater,4,5 which, during baseflow conditions, exfiltrates to make up the major part of surface water flow. r 2011 American Chemical Society
While degradation through biotic and abiotic transformation processes attenuates soil concentrations of these compounds and reduces their availability for transport to surface water, transformation products (TPs) of comparable persistence, and oftentimes higher mobility as the parent compound (PC) might be formed. Such TPs are also transported to surface water and, depending on their ecotoxicological potential, add to the risk posed by the release of the PC.6,7 For certain well-studied classes Received: July 26, 2010 Accepted: January 20, 2011 Revised: January 17, 2011 Published: March 03, 2011 2833
dx.doi.org/10.1021/es102537b | Environ. Sci. Technol. 2011, 45, 2833–2841
Environmental Science & Technology of pesticides, such as the triazine and chloroacetanilide herbicides, a large body of monitoring data demonstrates that their TPs are indeed present in surface water and groundwater more frequently and often in similar or even higher concentrations than their PCs.8-10 TPs of other widely used pesticides have also been shown to be of concern because of their high surface water or groundwater concentrations or because of their high toxicity toward aquatic organisms.5,11,12 These findings demonstrate that TPs of soil-applied, biologically active chemicals cannot be disregarded a priori in water quality and prospective chemical risk assessment. However, except for plant protection products,13 little to no guidance is given on how to include TPs into chemical risk assessment. A tiered strategy that at a first tier uses models to prioritize TPs for more detailed, experimental investigations has been advocated for to deal with the large number of potential TPs in regulatory risk assessment.14,15 Several methods to prioritize TPs with respect to their aquatic exposure potential have been suggested, ranging from scoring methods16,17 to generic multimedia, multispecies models (ref 18 and FOCUS Steps 1-219), to site-specific soil transport models (FOCUS Step 319). Here, we argue that because of the lack of sufficient experimental fate data for TPs to suitably parametrize those latter models and the ensuing prediction uncertainties,20 their use at the initial stages of prospective chemical risk assessment is inefficient. In agreement with most of the existing prioritization methods, we further argue that, at the prioritization stage, it is most appropriate to assess the exposure potential of TPs relative to their PC: (i) From a risk assessment perspective, it seems appropriate to base the decision to consider a TP for more detailed assessment on its risk relative to the PC, which is defined as the product of relative exposure and relative ecotoxicological potential;21 (ii) relative exposure assessment cancels out the impact of uncertain parameters that affect PC and TP fate in a similar manner such as application rates of the PC or environmental conditions, thus reducing prediction uncertainty; and (iii) using the usually well-determined predicted or experimental exposure concentrations for the PC and the predicted relative exposure potential for the TPs, absolute exposure concentrations of the TPs can easily be estimated, for example, for comparison to water quality criteria. The objective of this study, therefore, was to evaluate the suitability of an earlier developed, generic multimedia, multispecies model as a first tier approach to identify TPs that have a comparable or higher potential to be present in surface waters than their PC. For this purpose, we compared model predictions of relative surface water exposure potential for altogether 46 TPs of 16 pesticides against field data measured in a small river draining an agricultural catchment in Switzerland. We further explored the relative importance of groundwater exfiltration, which so far has not been considered in any of the prioritization methods, by comparing predictions from a simple leaching relationship with baseflow concentrations and mean groundwater concentrations of the investigated TPs.
’ METHODS Model Description. A dynamic (level IV), two-box multispecies model with the compartments soil and air was used to calculate the relative surface water exposure potential of TPs and their PC. A schematic of the model, model equations, and a description and quantification of all processes included in the model are given in the Supporting Information (SI), pp S2-S4.
ARTICLE
The model output considered for comparison to measured concentration ratios was the ratio of mass fluxes of TPs and PCs from soil into surface water (fsw) (for details see section Comparison of Measured and Predicted Aquatic Exposure Potential). This simplification of the model compared to the model described in Gasser et al.,18 that is, a three-box model coupled to a river model, was possible because fate in the river considered was negligible because of small travel times (50 kg/year in the catchment investigated (see below) based on data from an earlier survey.25 For 20 of the 46 TPs, analytical reference standards were available that allowed quantification of their concentrations in surface water (marked in Table S4 in the SI). These 20 TPs and their 10 corresponding PCs were therefore used for a quantitative evaluation of model predictions. The remaining 26 TPs had been reported to be formed in soil degradation studies, but analytical reference standards for those TPs are typically not available. Here, their absence or presence in 2834
dx.doi.org/10.1021/es102537b |Environ. Sci. Technol. 2011, 45, 2833–2841
Environmental Science & Technology the field study samples determined by high-resolution mass spectrometry (HR-MS)26 was used for a qualitative evaluation of model predictions. Environmental fate data needed as compound-specific model input parameters for pesticides and their TPs were compiled predominately from the EFSA DARs (European Food Safety Authority Draft Assessment Reports27). When available, the data from the EFSA DARs were used exclusively and only in cases of missing data were they complemented with data from other reregistration dossiers,28-31 the Footprint database,32 and the scientific literature. FFs values were estimated from soil degradation studies assuming first-order degradation of PC and TP. Typically, data from one to two studies for each type of environmental fate endpoint, with the exception of atmospheric half-lives, were available for most TPs, whereas considerably larger numbers of studies for each fate endpoint could be found for PCs. In cases of multiple data for the same fate endpoints, geometric means were used as model inputs. Where no experimental data were available at all, half-lives and partition coefficients were predicted using EPI Suite33 and generic FFs values were estimated based on qualitative information given in the dossiers. All fate data used as model input parameters are listed in the SI, Table S4, except for atmospheric half-lives, which were always estimated using AOPWIN from EPI Suite33 and assuming OH radical concentrations of 5 105 molecules/cm3. Field Study. Surface water samples were taken from La Petite Gl^ane, a small river draining an agriculture-dominated catchment in the Western part of Switzerland. The catchment has an area of 90 km2 (catchment description in SI, Figure S2) and the main crops are cereals (wheat, barley, maize), sugar beets, potatoes, and rapeseed. Samples were taken with a flow-proportional autosampler (ISCO 6700, Teledyne Inc., USA) between May and October 2008. Daily discharge at the sampling point was calculated from measured pressure and velocity using a pressure transducer and a Doppler sensor (ISCO, Teledyne Inc., USA) and the exact cross section of the river. Daily rain and discharge data during the sampling period are shown in Figure 1a. The catchment exhibited a rather high baseflow index of 0.78 during the sampling period (see SI, p. S13). Since major chemical inputs of pesticides and TPs into surface water could be expected to occur upon intense rainfall during or following the application period,1-3 20 samples were taken during high discharge events. To establish background concentrations, three additional samples were taken during baseflow conditions (for timing of all samples see Figure 1a). The water samples were transferred from plastic bottles of the autosampler to 1 L glass bottles and were stored at 4 °C for maximally three months before workup and analysis. Separate storage experiments and duplicate measurements of the same sample stored up to three months indicated no significant loss of the analytes under these conditions. Quantitative chemical analysis was performed by solid phase extraction (SPE) and subsequent liquid chromatography tandem mass spectrometry (LC-MS/MS) on an LTQ-Orbitrap as described in previous studies.26,34 For quantification, 100 ng of isotope-labeled internal standards were added to 500 mL surface water sample after pH adjustment. For recovery calculations, samples were spiked with external standard mix (1 mg/L) to yield concentrations of 200 and 600 ng/L, respectively. Concentrations of substances without identical internal standards were corrected for recoveries where deviations were more than
ARTICLE
30%. The blank sample contained nanopure water with internal standard. For comparison to model predictions, measured values between the limit of quantification (LOQ) and limit of detection (LOD) were calculated as LOQ/2, values below LOD as LOD/2 (data on LOQ and LOD are given in the SI, Table S5). Further details on materials and the analytical methods can be found in the SI, pp. S14-15. For detection of TPs for which no analytical reference standards were available, a procedure based on the high-resolution (HR) capabilities of the LTQ-Orbitrap, involving extraction at the exact mass, evaluation of retention time, and analysis of LC-HR-MS/MS spectra, was used as described in detail in ref 26. Comparison of Measured and Predicted Aquatic Exposure Potential. Whereas the field study yielded concentrations of PCs and TPs in individual samples taken during discharge events of various durations, the model provided continuously changing mass fluxes of the PC and its TPs from soil to surface water. For quantitative model evaluation, average mass ratios of TP/PC during discharge events were selected as a basis for comparing predicted with experimental data. Since in some cases several samples had been taken over the duration of one discharge event, measured mass loadings of individual compounds over the duration of the entire event (mevent) [ng] were calculated using eq 4. event
mevent ¼
∑ ðcsample 3 Qsample 3 Δtsample Þ
i¼1
i
i
i
ð4Þ
where csamplei is the measured concentration [ng/L] in sample i, and Qsamplei [L/s] and Δtsamplei [s] are the average discharge and duration, respectively, of the time period during which sample i was taken. Measured ratios of TP/PC during each event (revent.meas) were then calculated based on these mass loadings according to eq 5. revent:meas ¼
mTP event mPC event
ð5Þ
Predicted ratios for each event (revent.mod) were calculated as the ratio of mass fluxes from soil to surface water averaged over the duration (Δtevent) of the actual discharge events ( fsw.event): R event fsw ðtÞdt fsw:event ¼ t ð6Þ Δtevent revent:mod ¼
f TP sw:event PC fsw:event
ð7Þ
The comparison of model predictions with measurements was carried out for three time periods; the application period in May/ June, a post-application period in July, and a harvesting period from August to October. For these periods, TP/PC ratios over all events in that period were averaged (i.e., five events during application, four events during post-application period, and three events during harvesting period). For TPs without analytical reference standard, no measured mass ratios, revent.meas, could be determined, but rather how frequently they were detected in those 20 samples taken during discharge events. At the same time, based on an extensive evaluation of the analytical procedure used (unpublished data), we could assume that a positive detection of a TP meant that it was present in the sample in concentrations above 1-10 ng/L. Therefore, to compare model predictions with measurements, predicted TP/PC mass ratios for a given period were multiplied 2835
dx.doi.org/10.1021/es102537b |Environ. Sci. Technol. 2011, 45, 2833–2841
Environmental Science & Technology
ARTICLE
Figure 1. (a) Discharge of the river La Petite Gl^ane and precipitation during the sampling campaign (May-October 2008). Time periods over which samples were taken are indicated and assigned to the three periods for model evaluation: application period (red), postapplication period (green), and harvesting period (blue). Baseflow samples are indicated in gray. (b-e) Time courses of measured concentrations of the herbicide metamitron and its TP metamitron-desamino, and the herbicide chloridazon and its TP chloridazon-methyl-desphenyl.
with the average measured concentrations of the PCs in those periods to yield predicted TP concentrations. These average predicted TP concentrations were compared to the number of samples a given TP was actually detected in. If average predicted concentrations were below 1 ng/L, detection in hardly any sample was expected, and, conversely, if they were above 10 ng/L, detection in most samples was expected. Finally, concentrations of chlorothalonil TPs were predicted relative to the major TP chlorothalonil-4-hydroxy because the parent pesticide chlorothalonil itself could not be measured by LC-MS due to a low ionization efficiency with electrospray ionization (ESI).35
’ RESULTS AND DISCUSSION Measured Concentration Patterns. As an illustration of measured concentrations in the river La Petite Gl^ane from May to October 2008, field data for two herbicides are shown in Figure 1, that is, metamitron and its major TP metamitrondesamino (Figures 1b and 1c), and chloridazon and one of its
major TPs, chloridazon-methyl-desphenyl (Figures 1d and 1e). Median, minimal and maximal concentrations during application, post-application and harvesting period for all 16 parent pesticides and the 20 TPs that were analyzed quantitatively are summarized in the SI, Table S6. The two PC/TP pairs shown in Figure 1 are representative of two types of relative behavior between PC and TP. The two parent pesticides followed a very similar concentration pattern with peak concentrations of 2.7 and 1.2 μg/L for metamitron and chloridazon, respectively, at the beginning of June. With the exception of diuron, which was applied later in the season, the same pattern of highest concentrations in samples from discharge events during and shortly after application in May and June, followed by at least 10-fold drops in concentration levels to well below 100 ng/L during harvesting period were found for all PCs (SI, Table S6). In contrast, concentration patterns between TPs varied much more. Some TPs, such as metamitron-desamino (Figure 1c), CMBA or terbuthylazine-desethyl, exhibited a clear pattern of 2836
dx.doi.org/10.1021/es102537b |Environ. Sci. Technol. 2011, 45, 2833–2841
Environmental Science & Technology
ARTICLE
Figure 2. Comparison of predicted with measured ratios of surface water exposure potential for 20 TP/PC pairs during application, post-application, and harvesting period.
increasing and decreasing concentrations over time. Peak concentrations were usually slightly shifted to later times relative to those of the PC, depending on the difference in absolute and relative soil half-lives between the PC and the TP. Other TPs, including chloridazon-methyl-desphenyl (Figure 1e), but also chloridazon-desphenyl and metolachlor ethanesulfonic acid (metolachlor-ESA), were found to exhibit much smaller seasonal fluctuations than their PCs. These more even levels of TP concentrations are due to high, constant background concentrations of these TPs as demonstrated by their concentrations in the baseflow samples (Figure 1e and SI, Table S7). Also, persistent TPs can be remobilized from soil during strong precipitation events even late in the season as has, for instance, been demonstrated in detail for metolachlor-ESA,9 and can also be observed for ESA and oxanilic acid (OXA) TPs of metolachlor and flufenacet in this study. In all cases, combinations of these different factors resulted in a steady increase of the TP/PC ratios from the application period through to the harvesting period as also shown by others.8,36 While for some TPs, that is, chloridazon-desphenyl, chloridazonmethyl-desphenyl, metamitron-desamino, flufenacet-ESA, and metolachlor-ESA, this led to TP/PC ratios >1 during the postapplication and/or harvesting period in this study, there were also several groups of TPs which never exceeded PC concentrations. For example, the TPs of the triazine herbicides terbuthylazine and atrazine always exhibited lower concentrations compared to their PCs, as described previously for atrazine and its TPs in Swiss lakes 37 and rivers.38 We further found consistently lower levels of TPs in comparison to their PCs for azoxystrobin acid, DCPMU, DCPU, flufenacet-OXA, metolachlor-morpholinone, metribuzin-desamino, and CMBA. Comparison of Measured TP/PC Ratios with Model Predictions. The comparison of measured and predicted mass ratios of TP and PC during application, post-application, and harvesting period is given in Figure 2 for all 20 pairs of TPs and PCs for which a quantitative comparison was possible. The
results in Figure 2 indicate that, while measured mass ratios across all compounds and periods span four orders of magnitude, the model used in this study in combination with data that can be extracted from registration information predicts the relative levels of TPs and their parent pesticides correctly within one order of magnitude, with only two exceptions, chloridazondesphenyl/chloridazon (pair 6) and chloridazon-methyl-desphenyl/chloridazon (pair 7). For these pairs, the predicted ratios underestimate the measured ratios by more than a factor of 10. This general agreement between model predictions and measurements is considered satisfactory in light of reported uncertainty ranges of at least one order of magnitude for estimating exposure to TPs using multimedia models because of uncertainty and variability in chemical fate parameters.39 To evaluate the influence of groundwater exfiltration on the results presented in Figure 2, Table 1 summarizes the baseflow concentrations of TPs during the harvesting period and compares them to (i) mean concentrations from a 2007/2008 survey of 22 diverse Swiss groundwater sites 40 and (ii) the leaching indicators, cni and rgw (eqs 1-3), calculated in this study. The good rank correlation between the calculated leaching potentials, the observed mean groundwater concentrations, and the baseflow concentrations in the La Petite Gl^ane river confirms that for highly persistent and mobile TPs that leach into groundwater in high amounts, the groundwater contributes a major part to their overall load in surface water, as has been demonstrated before.8,9,36,41 Since the evaluative model used in this study does not account for leaching of compounds into groundwater and exfiltration of groundwater into surface water, it underestimated the presence of the two chloridazon TPs that have a high absolute leaching potential, cn > 1, and are also stronger leachers than their PCs, rgw > 1. Two more TPs, namely metolachlor-ESA (Table 1) and flufenacet-ESA (SI, Table S7), are predicted to show both high individual leaching potential (cn > 1) and a high groundwater ratio (rgw > 1). While for metolachlor-ESA TP/PC ratios in surface water were also consistently underestimated with our 2837
dx.doi.org/10.1021/es102537b |Environ. Sci. Technol. 2011, 45, 2833–2841
Environmental Science & Technology
ARTICLE
Table 1. Comparison of Baseflow Concentrations with Groundwater Concentrations from a Swiss Groundwater Survey40a pesticides and TPs
concentration baseflow
mean concentration
harvesting period (ng/L)
groundwater (max. conc.) (ng/L)
cn (-)
rgw (-)
chloridazon-desphenyl
721
126 (>600)
1-10
>10
chloridazon-methyl-desphenyl
282
42 (>600)
1-10
1-10
metolachlor-ESA
149
29 (270)
>10
1-10
atrazine-desethyl
39
27 (150)
0.1-1
metamitron-desamino
27
0.2 (5)
0.1-1
metolachlor-OXA
16
10 (180)
atrazine-2-hydroxy DCPU
6 5
7 (37) NA b
0.1-1
