Decline of pesticide residue on treated crop seeds: An analysis of comprehensive industry data and implications of the current risk assessment scheme for plant protection products

For plant protection products applied as seed treatments, the risk to birds and mammals possibly feeding on treated seeds must be addressed in the EU to register products for commercial use. One assumption of the Tier 1 long‐term risk assessment of the European Food Safety Authority (EFSA) is that residues of pesticides on treated seeds do not decline over time after seeding. Consequently, a time‐weighted average factor (fTWA) of 1 (i.e., no dissipation) is used to calculate residue concentrations on seeds. In contrast, for spray applications, a default dissipation half‐life (DT50) of 10 days is considered corresponding to an fTWA of 0.53. The aim of this study was to establish a default fTWA for treated seeds based on 29 industry‐conducted seed dissipation studies, providing 240 datasets covering different active substances, crops, and regions. For fTWA calculation, two approaches were used: (i) kinetic fitting and (ii) using measured data without kinetic fitting. From kinetic fitting, 145 reliable DT50 values were obtained. Because there were no significant differences in DT50 values between crops and between the central and southern EU, the DT50 data from all studies were pooled. The geometric mean DT50 was 3.8 days and the 90th percentile was 13.0 days, corresponding to 21‐day fTWA values of 0.27 and 0.59, respectively. Twenty‐one‐day fTWA values could be calculated directly from measured residues for 204 datasets. The resulting 21‐day fTWA values were comparable with those from kinetic fitting (geometric mean: 0.29, 90th percentile: 0.59). The results demonstrate that residue decline on seeds is comparable with foliar dissipation after spray applications. Therefore, the risk assessment scheme by EFSA should implement a default fTWA < 1.0 in the Tier 1 risk assessment for treated seeds (e.g., either 0.53 as for foliage or 0.59, the 90th percentile fTWA in seeds reported in this study). Integr Environ Assess Manag 2024;20:239–247. © 2023 The Authors. Integrated Environmental Assessment and Management published by Wiley Periodicals LLC on behalf of Society of Environmental Toxicology & Chemistry (SETAC).


INTRODUCTION
In modern agriculture, seed treatments are frequently used before sowing to control pathogens, insects, and other pests that can attack seeds or seedlings (Sharma et al., 2015).In the European registration process for plant protection products, a risk assessment must be conducted for birds and mammals (European Food Safety Authority [EFSA], 2023), assuming treated seeds may be ingested as part of the animal's diet.For the long-term risk assessment, exposure is calculated based on food intake rate and seed loading.Although dissipation of residues is considered by EFSA (2023) for spray applications as default in the Tier 1 risk assessment, common practice by Member States and EFSA is to assume no dissipation for treated seeds (i.e., f TWA = 1).Also, in the recently published updated EFSA guidance document for birds and mammals (EFSA, 2023), no default residue decline is proposed for treated seeds for the Tier 1 long-term risk assessment.A refinement of the risk assessment is, however, possible when substance-specific dissipation half-life (DT 50 ) data based on seed residue dissipation trials are available (with which a refined f TWA can be calculated).
For spray applications, the default foliar residue decline at Tier 1 for the long-term risk assessment is based on published data collected by Willis and McDowell (1987).Based on this dataset, a default foliar DT 50 of 10 days is currently used for the Tier 1 assessment, corresponding to a 21-day f TWA of 0.53 (EFSA, 2009).In the recently published birds and mammals guidance (EFSA, 2023), additional data from public literature were considered to further substantiate the default foliar DT 50 of 10 days (Ebeling & Wang, 2018;Fantke & Juraske, 2013;Lahr et al., 2018;Lewis & Tzilivakis, 2017).
In contrast, for treated seeds, no such dataset has been compiled to date for risk assessment purposes.The aim of this article is to derive a default residue decline factor (f TWA factor) for treated seeds for use in the Tier 1 risk assessment based on industry-conducted residue decline studies of treated seeds.These studies were conducted by scattering treated seeds on study plots in the field according to a target sowing rate.Samples of preferably ungerminated seeds are taken at defined intervals (e.g., directly after scattering, 1, 3, 7, 14, 21, 28 days after scattering) and the amount of pesticide residues on the seeds is determined.For the present evaluation, these residue data were used to calculate f TWA values by both direct calculation of the timeweighted average exposure and by fitting kinetic dissipation models (FOCUS, 2014) in order to propose a default 21-day f TWA value for the Tier 1 seed treatment risk assessment.

Database
In all, 29 industry field dissipation studies were collected in which residue decline of 19 active substances (Table 1) in seeds was measured.All studies were conducted in Europe during the typical drilling seasons of the considered crops according to good laboratory practice (GLP).Twenty-six studies were conducted outdoors, and three studies were conducted under protected conditions with small amounts of irrigation to simulate morning dew.Only in two outdoor trials (both from the same study) irrigation was considered necessary by the farmer.Seeds had either been drilled according to the proposed label or scattered manually on the soil.Heavy rainfall after sowing or scattering was avoided.Hence, the data were generated to reflect realistic conditions and according to typical agronomic practice and environmental conditions prevailing at the time of sowing.
For seed collection, fields (1-10 fields per study) were either subdivided into sampling locations (range: 12-75) or seeds were taken randomly across the study field for residue analysis.Seeds were generally collected from the surface (but in some studies, it is not explicitly mentioned from where seeds were taken).Collection of seeds from the soil surface reflects a realistic exposure situation for birds and mammals foraging on leftover seeds that remained on the soil surface after drilling.Also, residue analysis was conducted according to GLP, applying current guidelines at the time of conduct of the studies (including OECD ENV/JM/ MON0(2007)17, OECD ENV/JM/MONO(2009)31, SANCO/ 825/00/rev.8.1, SANCO/3029/99 rev.4, USEPA OCSPP Guideline No. 860.1500).Residue measurements in the studies were conducted from scattering until Days 3-28 after scattering.
Measured residues from all sampling days were extracted from the study reports along with information about the seed type (crop), the active substance, and the trial sites (i.e., country).Overall, 304 residue decline datasets (i.e., decline data of one compound from one trial) from 203 independent trials (in some trials more than one active substance was measured) were extracted.When residue concentrations for enantiomers were reported, these were summed up to obtain the overall concentration of the active substance.Only the overall concentration of an active substance was used for the evaluation.After summing residue data from single enantiomers to give the total active substance residue, 240 datasets from 19 active substances remained.From these, 187 datasets were obtained from the northern European residue zone (Germany, Poland, N-France, Austria, the Netherlands, Hungary, the UK) and 53 datasets were from the southern European residue zone (Spain, Italy, S-France, Greece, Portugal, Bulgaria).Residue zones are defined by the European Commission (2019).The UK is not included in this document but, due to similar climatic conditions, the UK is assigned to the northern zone in this evaluation.In all, 211 datasets were obtained for fungicides and 29 for insecticides.Of the 240 datasets, 204 contained seed residue measurements up to a minimum of 20 days (the remaining datasets contained fewer than 20 days).Each seed residue dataset included samples from 2 to 19 time points (but kinetic evaluations were possible only when data from at least three time points were available).An overview of crops investigated in these studies is provided in Table 2.

Calculation of 21-day f TWA
For f TWA calculation, two approaches were used: (i) f TWA calculation based on kinetic models according to FOCUS (2014), using "trigger endpoints" and (ii) f TWA calculation based on measured data (without using kinetic models).
For the first approach, kinetic models were fitted to all datasets.Kinetic modeling was conducted according to FOCUS (2014), which proposed approaches to calculating degradation kinetics for laboratory soil and water studies, field studies, and water-sediment studies.According to FOCUS (2014), kinetic modeling can be conducted to obtain "modeling endpoints" (i.e., DT 50 values that can be used in FOCUS models to calculate predicted environmental concentrations [PEC] in soil, ground water, and surface water) or "trigger endpoints" (i.e., DT 50 /DT 90 values that are used to decide whether additional studies are required or not).Because in some environmental fate models used to calculate PEC, only single first-order (SFO) kinetics are implemented (i.e., the models cannot be used with biphasic kinetics), the FOCUS flowchart for modeling endpoints favors the selection of SFO kinetics as being an acceptable description of observed residue decline.This can result in SFO kinetics being selected even when a biphasic model provides a better fit to the data.Furthermore, when a biphasic model is recommended by the flowchart for modeling endpoints, the user is directed to derive a conservative pseudo SFO DT 50 from this fitting.This pseudo SFO DT 50 value can be considerably larger than the DT 50 at which 50% of residues have dissipated from treated seeds.Again, this is because the environmental fate models for which the flowchart was created are generally compatible only with SFO dissipation.
In considering the risk assessment of birds and mammals, there is no valid scientific reason to focus only on SFO endpoints, and therefore the assumptions and biases of the modeling endpoint flowchart in FOCUS (2014) are inappropriate.Instead, it is mathematically possible to calculate a realistic exposure estimate using either SFO or biphasic kinetics, whichever one provides the best fit to observed residue decline.Such a "best fit" approach is the objective of the "trigger endpoint" flowchart in FOCUS (2014), where DT 50 values are derived to describe substance decline as accurately as possible.This is typically assessed as being the fit that provides the lowest χ 2 -error and best visual fit without systematic deviation between predicted and observed residues.Hence, in the work presented here, the methodology for trigger endpoints was followed for the calculation of DT 50 for residues on seeds.
According to the trigger endpoint flowchart: first, the fit based on an SFO model is compared with the fit of a firstorder multicompartment (FOMC) model.If the SFO model fit is considered to be better (criteria are described below) than the fit of the FOMC model, the SFO model is chosen and the DT 50 is calculated.However, if the FOMC model fit is considered better than the SFO model fit, then a double firstorder in parallel (DFOP) model is additionally fitted to the data, and its fit is compared with the FOMC model fit.Finally, the best fit (FOMC or DFOP) is chosen.The procedure is illustrated in Supporting Information: Figure S1.The quality of the fits was evaluated based on the goodness-of-fit and the reliability of parameter estimates (FOCUS, 2014).It is recommended that an χ 2 error of 15% or less indicates an acceptable fit, although, for intrinsically more variable data (such as the field studies presented here), higher values can be tolerated if the visual fit is acceptable.In the present work, a threshold of 20% was selected, because this threshold is typically used by regulators when evaluating field residue data.The visual assessment was based on the model fit to the measured concentrations and the corresponding residuals plot, where negative and positive residuals should be randomly scattered around the horizontal zero line (FOCUS, 2014).The reliability of the model parameters was assessed based on p-values from a one-sided t-test with α = 0.05.Only model fits that were visually acceptable and resulted in reliable estimation of model parameters were considered.Default dissipation half-life values were calculated numerically for all models, that is, the time points on the curve were calculated, when 50% of the substance had disappeared.From the model fits that fulfilled the conditions mentioned above, the one with the lowest χ 2 error was selected.
As recommended by FOCUS (2014), residue concentrations on seeds between the limit of detection (LOD) and the limit of quantification (LOQ) were set at the actual measured value or, if the actual measured value has not been reported, to 0.5 × (LOQ + LOD) before fitting.All samples below the LOD were set at 0.5 × LOD.Samples after the first value <LOD were omitted unless it was followed by later samples above LOQ.All parameters were allowed to freely optimize with the average residue on Day 0 being used as an initial value.The best model fit was then used to calculate the overall DT 50 value by a numerical approach.The 21-day f TWA value was calculated by determining the area under the curve (AUC) within the 21-day time interval divided by the period (21 days) and the predicted concentration at Day 0. The AUC was determined by using the trapezoidal rule where a = x 0 = Day 0, b = x N = Day 21, x k is a partition of the interval [a, b] with regular spacing, and N = 10 000 (i.e., the AUC was approximated by the area of 10 000 trapezia within the 21-day period).Kinetic fitting was conducted with KinModeller v. 1.0 (WSC Scientific GmbH, 2022).For the second approach, direct calculation of f TWA was done without kinetic modeling but by weighting residue concentrations for the sampling intervals.For this method, a step function was first created from the measured data, with the steps being located in the middle of two time points.When replicates were available, average values were used.For the first and last value of a dataset, timeweights were derived by calculating only the distance between the current and next or previous and current sampling event, respectively.A schematic presentation of the approach is presented in Figure 1.The area under this step function was calculated by multiplying function values and the associated time intervals.Subsequently, the total area was divided by the total period (21 days) and the initial concentration on Day 0 yields the f TWA .
Only datasets with residue data until at least 20 days after seed scattering (204 datasets) were considered because shorter periods would result in unreliable f TWA values.In such cases, the measured values from Day 20 were used until Day 21.

Statistics
Because the residue measurements were conducted for different substances in different crops and different residue zones, it was tested if these groups differ significantly or if they can be combined to give an overall DT 50 , and therefore an f TWA value, for the risk assessment.Default dissipation half-life values vary between substances; hence, differences between zones or seeds of different crops can only be evaluated for each substance separately.As the DT 50 values were not normally distributed (Kolmogorov-Smirnoff test), a nonparametric test was used to test if average DT 50 values differ (Mann-Whitney U-test, two-sided).

RESULTS
Regarding kinetic fitting, 145 datasets fulfilled the criteria for goodness-of-fit and reliability of parameter estimates.For most datasets, SFO resulted in the best fit.Biphasic models (FOMC, DFOP) were selected as the best fit in 36% of cases.Selected best fit kinetic models were then used to predict a continuous time series of residue values from Day 0 to Day 21 for each individual trial.From these 145 continuous time series of predicted residue values, the 10th, 50th, and 90th percentile residue values at each time point were calculated for illustration and are presented in Figure 2. Furthermore, DT 50 values were calculated from kinetic fits, and they ranged between 0.1 and 38.1 days.The geometric mean DT 50 was 3.8 days and the 90th percentile was 13.0 days.The current default foliar DT 50 of 10 days (EFSA, 2009) corresponds to the 83th percentile DT 50 in this evaluation.Default dissipation at the 90th percentile (DT 90 ) values ranged between 1.8 and 13 064 days.The geometric mean DT 90 was 17.8 days, and the 90th percentile was 53.2 days.The distribution of DT 50 and DT 90 values is presented in Figure 3.A comparison of DT 50 values between zones could only be made for spring wheat and maize (Figure 4A); as for all other crop-substance combinations, data were not available for both zones.A statistical comparison of DT 50 between northern and southern zone was conducted only for maize because of the low sample size (n < 3 for substance S27 and S26) for spring wheat in the southern zone.Statistical testing indicated that DT 50 values for maize were not significantly different between the northern and southern residue zone (Table 3).For seven active substances, DT 50 for more than one crop were available (Figure 4B).For five of the seven substances sample sizes were sufficient (n ≥ 3) to conduct statistical comparison of DT 50 between different crops per active substance.There were no statistically significant differences in DT 50 values between crop seed types (Table 4).Therefore, it was considered appropriate to pool data from crop seed types and residue zones.
As a result, the overall median and geometric mean 21-day f TWA from kinetic fitting was 0.30 and 0.27, respectively.The 90th percentile was 0.59.The second approach, calculating f TWA directly from measured residues without kinetic fitting, resulted in f TWA values that were comparable with the ones from kinetic fitting.Of the 240 datasets for active substances, residue measurements until at least 20 days after seed scattering were available for 204 datasets and hence a 21-day f TWA could be calculated.The distribution of the pooled 21-day f TWA values calculated from kinetic fits and from measured data was very similar (Figure 5).Additionally, correlating 21-day f TWA values from kinetic fitting and the corresponding values calculated from measured data indicated that both methods resulted in comparable f TWA values (Pearson: R 2 = 0.971, p < 0.001; Figure 6).Hence, the descriptive statistics of 21-day f TWA from both methods was also very similar: the median, geometric mean, and 90th percentile 21-day f TWA values are presented in Table 5.The 21-day f TWA values from kinetic fits ranged from 0.04 to 0.83, and the 21-day f TWA values from measured data ranged from 0.05 to 1.00.

DISCUSSION
This study presents the first analysis of a large dataset from field studies (29 studies with 240 datasets for 19 active substances) measuring residues on treated seeds of various crops and residue zones.A kinetic evaluation of residue dissipation trials on treated seeds demonstrated that for maize residue concentrations on seeds decline over time without significant differences between crops or zones.Hence, the calculated f TWA s seem to be equally relevant for all crop seed types across the two residue zones.Overall, the decline was rather fast with a geometric mean DT 50 of 3.8 days, corresponding to a 21-day f TWA value of 0.27.The results from kinetic fitting were based on reliable fits for 145 out of 240 datasets, where the reliability of a fit was determined by the χ 2 error, p-values from a t-test and visual inspection of the fit as recommended by FOCUS (2014).The second approach to calculation of 21-day f TWA values was based on measured data and could be conducted for many datasets (n = 204 for which data for at least 20 days were available).These 21-day f TWA values were very similar to the ones derived from kinetic fitting regarding both distribution and descriptive statistics.Additionally, values from both methods were highly correlated.The high correspondence of results from both methods indicate that the data can be considered robust.The 90th percentile f TWA value of 0.59 for seed treatments is comparable with the default f TWA = 0.53 for foliar residue decline in EFSA (2009).The corresponding default foliar DT 50 is based on analysis of approximately 450 mean foliar DT 50 values for vegetative plant materials by Willis and McDowell (1987), who stated that "most pesticides have DT 50 s below 10 days."The EFSA (2023) provided additional public literature on plant residue decline data to further substantiate the relevance of the default foliar DT 50 value for pesticides.Although the EFSA (2023) discussed findings from publications on foliar residue decline, data on seeds provided in some of these publications were not evaluated.In the following, we discuss the findings on residue decline on seeds and also other plant matrixes from literature cited in EFSA (2023) with our results on treated crop seeds.Lewis and Tzilivakis (2017) presented data on pesticide dissipation rates in various plant matrixes, including seeds (e.g., leaves, stems, seeds, fruits).For seeds, only four DT 50 values are listed, ranging from 2.65 to 10.2 days (median: 3.7 days).The median DT 50 value over all plant matrixes (i.e., leaves, stems, seeds, etc.) was 4.0 days, and the 90th percentile was 13.8 days (n = 2713).The values by Lewis and Tzilivakis (2017) both for seeds and for all plant matrixes are in line with the seed-specific DT 50 values presented here.Fantke and Juraske (2013) collected data on dissipation of pesticides in plants and on plant surfaces (including straw, fruit pulp, tree bark, but without seeds) from trials that were conducted worldwide.The EFSA (2023) calculated a median DT 50 from all plant matrixes.This median DT 50 (3.89days) was very similar to the median value of 4.2 days derived for treated seeds in the work presented here.Also the median foliar DT 50 (2.7 days) calculated by the EFSA (2023), based on the data published by Lahr et al. (2018), indicates a very fast dissipation of pesticides from plant foliage.A fast foliar residue dissipation under field conditions was also estimated by Ebeling and Wang (2018) with geometric mean and 90th percentile DT 50 values of 3.2 and 7.9 days, respectively.Overall, the pesticide residue decline in seeds presented in this study was comparable with the decline of residues on foliage.
In conclusion, this study demonstrates a residue decline of plant protection products on treated seeds under field conditions based on a large dataset of 29 studies.The derived f TWA value is similar to the EFSA default f TWA applied for spray applications at the Tier 1 risk assessment.The current assumption by EFSA (2023) of no decline at Tier 1 for the long-term risk assessment for seed treatment uses is thus highly conservative and not realistic.It is therefore proposed that the recently published EFSA guidance (2023) for birds and mammals should be updated by including a default f TWA (e.g., f TWA of 0.53 as for foliage, which corresponds to the 83th percentile f TWA in this evaluation, or f TWA of 0.59, the 90th percentile f TWA in seeds) for the Tier 1 risk assessment for treated seeds.

FIGURE 1
FIGURE 1 Example of a kinetic fit (first-order multicompartment [FOMC]; dotted line) compared with the respective step function (solid gray line) used to calculate time-weighted average factor (f TWA ) values.Dots indicate measured residues

FIGURE 4
FIGURE 4 Arithmetic mean dissipation half-life (DT 50 ) values in the northern (NZ) and southern residue zone (SZ) for individual active substances identified by substance codes (e.g., S27) (A) and in seeds of different crops for separate substances (B).The numbers next to the bars indicate the sample size

FIGURE 6
FIGURE 6 Plot of 21-day time-weighted average factor (f TWA ) values calculated from kinetic fitting against those calculated from measured data.The straight line presents the 1:1 line

TABLE 2
Number of datasets in different seed types and in the northern and southern residue zone

TABLE 3
Statistical results regarding differences in DT 50 in maize seeds between zones within individual active substances Note: Not calculated (n.c.) due to low sample size (n < 3) in at least one zone.Abbreviations: DT 50 , dissipation half-life; MWU, Mann-Whitney U-test.

TABLE 5
Summary of 21-day f TWA values from kinetic fitting and from measured data Statistic 21-day f TWA from kinetic fitting 21-day f TWA from measured data