Pesticide contamination and associated risk factors at public playgrounds near intensively managed apple and wine orchards

Site description

We identified 125 potential playgrounds in the valleys Venosta (Vinschgau), Isarco (Eisack), Adige (Etsch), and Low Adige (Südtiroler Unterland) using official web pages listing public playgrounds [60, 78] and 71 of these playgrounds were randomly selected for sampling (Fig. 1). Randomisation and sampling site selection were done separately for each geographical area, considering an equal distribution of the mean distance to agricultural fields in the four sampling regions (Additional file 2: Fig S2).

Generally, the four investigated valleys differ in land use, mean air temperature, mean precipitation and altitude (Additional file 1: Fig S1, Additional file 2: Fig S2, Additional file 3: Fig S3). The Adige valley and the Low Adige have mean annual temperatures of 12 °C and mean annual precipitation of 720 mm and 856 mm, respectively. Sampled playgrounds in the Adige valley were located at a mean altitude of 287 m above sea level (a.s.l.; min 250, max 350) and in the Low Adige at 293 m a.s.l. (min 200, max 610, Additional file 1: Fig.S1). The Low Adige is dominated by vine yards (20%), while the agriculture of the Adige valley is dominated by apple orchards (32%, Additional file 3: Fig. S3). In comparison, the Venosta and the Isarco valley exhibit lower mean annual temperatures (8.8 °C and 10.1 °C, respectively) and annual precipitation (566 and 670 mm, respectively), while the sampled playgrounds in these valleys were located at higher altitudes, with 688 m a.s.l. (min 555, max 890) and 737 m a.s.l. (min 515, max 1130, Additional file 1: Fig S1), respectively. Apple cultivation is the dominant land use type in the Venosta and the Isarco valley (20–30% of the total land cover), while cultivation of wine has a less important role (Additional file 3: Fig. S3).

Geographical data including elevation, distance to nearest orchards and direction of closest or chards were obtained from the regional geographic information system Geobrowser [6]. Meteorological data including temperature, global irradiance, wind speed, wind direction and precipitation were obtained from the nearest meteorological stations calculated for 14 and 5 days prior to sampling provided by the regional hydrographic service [36].

Pesticide spraying is recommended during wind speeds < 3 m s⁻¹ [1, 21]. Therefore, data on wind was additionally separated in hours while spraying was recommended (i.e. < 3 m s⁻¹) and hours while spraying was not recommended (i.e. ≥ 3 m s⁻¹).

Land use data within a 500 m radius around sampled sites was provided by the Federal Agricultural Office of the Autonomous Province of Bolzano-South Tyrol. This data included detailed information on the area of apple and wine or chards other fruit orchards, other agricultural areas (e.g., arable crops, grassland), forests, buildings and infrastructure, fallow land and waterbodies (m²).

According to the distance to nearest agricultural fields, sampled playgrounds were classified as NEAR (≤ 50 m distance) and FAR (> 50 m distance). The distance was measured from the centre of the playground to the first apple tree or grape vine of the adjacent agricultural area. The agricultural areas considered in this study included both, conventionally and organically farmed apple orchards and vineyards. The final randomized selection included 21 sampling sites in the Venosta (10 NEAR, 11 FAR), 20 in the Adige valley (10 NEAR, 10 FAR), 20 in Low Adige (10 NEAR, 10 FAR) and 10 in the Isarco valley (4 NEAR, 6 FAR), resulting in a total of 34 NEAR and 37 FAR sites.

Sampling and analysis

Plant sampling was conducted between May 16-23, 2017, on dry days after a period of at least five rainless days, by the accredited office BioProgramm from Padova [11]. For this purpose, 3 to 4 randomly selected grass samples per playground were taken by hand-wearing laboratory gloves and grouped to one 300–400 g (fresh mass) composite sample. Plant material was vacuum-packed in freezer bags and stored at room temperature until analysis by the laboratory for food analysis of the Autonomous Province of Bolzano. After acetonitrile extraction (701881-Acetonitrile for LC–MS, ITW Reagents PanReac AppliChem) according to the dispersive SPE–QuEChERS-method [3], pesticide concentrations of grass samples were analysed using Agilent, 7000C Triple Quadrupole GC/MS series system, (Agilent, Waldbronn, Germany) and API 4000™LC-MC/MS series system (AB Sciex, Concord ON, Canada), following the European standard UNI EN 15662:2009 [64] to identify up to 315 substances. Measured pesticide concentrations are based on wet weight of fresh grass samples.

Statistical analyses

We conducted the statistical analysis according to a pre-defined statistical analysis plan to identify environmental mediators of pesticide concentrations and to predict the environmental conditions associated with the lowest levels of pesticide pollution. Continuous variables were presented as medians with interquartile range (IQR) and categorical variables as percentages. Descriptive statistics, analyses of variance (ANOVA) and regression analysis were performed using R v. 3.5.1 [65], considering normal distribution of data or homogeneity of variances. Non-normal distributed factors, irradiance and wind, were log-transformed before analysis.

Chi squared test and Fisher’s exact test for count data was applied for the categorical data number of detected pesticides and independent 2-group Mann–Whitney U Test for zero-inflated continuous data on pesticide concentrations between NEAR and FAR sites. Kruskal–Wallis Test, one way ANOVA by ranks and, if main effects were found, a post hoc test (Tukey and Dunn’s test) was used to compare pesticide contamination among the four different regions. An initial univariate hurdle model, with distance as predictor, was used to (i) analyse the association of spatial proximity and pesticide concentration levels (linear model using non-zero values) and (ii) to assess how distance is associated with the probability of playgrounds to be contaminated (binomial model).

Based on this, a multivariate weighted linear regression analysis was performed, to estimate the effect of environmental factors on the measured pesticide concentrations (mg kg⁻¹) weighted for number of detected pesticides. The model containing first-order interactions evaluated the impact of the following environmental factors: precipitation (mm), air temperature (°C), global irradiance (W m⁻²), altitude (m a.s.l.), wind direction, wind speed (m s⁻¹), orientation of the playground to the surrounding agricultural sites (16 cardinal points), hours of possible drift (h) and the relative area covered by forest, apple and wine orchards within a radius of 500 m (%).

Additionally, the sum of windy hours and averaged maximum wind speed per wind direction (from or to the agricultural field) were calculated for the 5- and 14-day period before sampling. Further, hours with wind speeds higher and lower than 3 m s⁻¹ were summed separately as spraying is not allowed with wind speeds > 3 m s⁻¹.

The weighting of the dependent variable pesticide concentration was based on the number of detected pesticide species per playground. Thus, the dependent variable—the total measured pesticide concentration—was weighted for the detected number of different pesticides per playground. Stepwise variable selection was conducted using the Akaike information criterion (AIC) and the Bayesian information criterion (BIC) for model building, considering multi-collinearity and variance inflation factors (VIFs; only variables with a VIF < 3 were included in the final model; [88]). For the analysis, all descriptors were centred and scaled for model analytics, but were retransformed for result presentation via effect- and interaction plots.

Model evaluation and control for over-dispersion was done by analysis of deviance with the anova() function [88]. Homoscedasticity of the models was checked by Breusch–Pagan test. Model evaluation and final selection was based on comparisons of R², adjusted R², AIC and BIC. Correction method for multiple testing (post hoc tests and Welch correction) was applied where necessary. Two-sided p values < 0.05 were considered statistically significant.

Sites NEAR vs. FAR from agricultural sites

Median and inter-quartile ranges of pesticide concentrations in NEAR sites (0.04 mg kg⁻¹, 0.03–0.08) were significantly higher than in FAR sites (0.01 mg kg⁻¹, 0.01–0.04, Mann–Whitney U = 58, n₁= 21, n₂= 11, p  = 0.024, r  = 0.4, Table 2). Similarly, the number of sites containing pesticides was significantly higher in NEAR (61.8%) compared to FAR sites (29.7%, p _χ ² =0.013) (Table 1).

More specifically, NEAR sites showed significantly more pesticides and higher concentrations of fluazinam (p _χ ²  = 0.008 and P_MannWhitneyU = 0.003, resp.) and penthiopyrad (P_FishersExact = 0.009, P_MannWhitneyU = 0.008), than FAR sites (Tables 1 and 2).

Endocrine-active pesticides were found more often and in higher concentrations in NEAR (50% and median 0.03 mg kg⁻¹) as compared to FAR sites (24%, and median 0.01 mg kg⁻¹, P_FishersExact = 0.029, Table 1 and P_MannWhitneyU = 0.006, Table 2).

Pesticide records in the four study regions

Sampled playgrounds are stratified according to the four investigated valleys. Fisher’s-exact test and post hoc pairwise test for nominal data (adj. Fisher’s Exact for multiple comparisons).

The insecticide phosmet was detected on 13 playgrounds in Venosta, being significantly more frequent (P_FishersExact< 0.001, Table 3) and higher concentrated (P_{KruskalWallis}< 0.001, Table 4) as compared to other regions.

The fungicide fluazinam was significantly more frequent (10 detections) in the Venosta valley, but with 0.013 mg kg⁻¹ not significantly higher concentrated than in the other regions (P_FishersExact = 0.048, Table 3 and P_{KruskalWallis} = 0.903, Table 4, respectively).

Penthiopyrad, however, was detected more often in the Isarco valley (4 detections, P_FishersExact = 0.004, Table 3) with significantly higher concentrations (0.036 mg kg⁻¹, P_{Kruskalwallis} = 0.001, Post Hoc test: p < 0.05) than in other regions. (Table 3).

Difenoconazol was only detected in the Isarco valley at two playgrounds and benzalkoniumchloride was only detected in the Adige valley.

Pesticide contamination and local conditions

Pesticide concentration, weighted for the number of pesticides detected per playground, was strongly dependent on the following local environmental parameters (R² = 0.8, Table 5): global irradiance (W m⁻²), amount of apple orchards (%), mean wind blowing to the agricultural field and away from the playground (h), rainfall (mm), distance from agricultural field (m), first grade interaction of distance from agricultural field (m) with maximum wind speed (m s⁻¹), hours of wind when spraying is not recommended (speed < 3 m s⁻¹), and the first grade interaction between the hours when spraying is allowed and wind speed (m s⁻¹) (Table 5 and Fig. 2).

OLS linear regression model	Estimates (β-weight)	95% CI [LL, UL]	p-value
Intercept	1.0	0.35 to 1.65	0.004
Distance from agricultural field (m)	− 1.07	− 1.99 to − 0.16	0.023
Apple orchard (%)	0.45	0.07 to 0.83	0.022
Global irradiance (W m−2)	− 1.22	− 1.82 to − 0.63	< 0.001
Time when spraying not allowed (h)a	− 1.03	− 1.45 to − 0.61	< 0.001
Wind blowing away from playground (h)	− 1.23	− 1.92 to − 0.55	0.001
Rainfall (mm)	0.62	0.20 to 1.04	0.006
Distance: high wind speed (m s−1)	1.25	0.48 to 2.02	0.003
Time when spraying allowed (h): Wind speed (m s−1)a	0.81	0.31 to 1.31	0.003

Basic model information and model fit statistics
N	F (8,21)	AIC	BIC	R2 and CI	adj. R2	Residual standard error	Homoscedasticity (Breusch-Pagan)
69	14.75	84.53	98.54	0.85 0.59–0.87	0.79	1.12	Assumption not violated (P = 0.066)

The model contains first-order interactions showing estimates (β weights) and 95% confidence intervals (95% CI) for all predictors, as well as indicators for model quality (R², AIC, BIC). All included predictors are significantly associated with pesticide concentration on playgrounds (mg kg⁻¹ = ppm). The dependent outcome variable ‘pesticide concentration (mg kg⁻¹)’ on playgrounds was weighted by the number of detected pesticides per playground. LL, lower level; UL, upper level

A significant b-weight indicates that semi-partial correlation is also significant. LL and UL indicate the lower and upper limits of a confidence interval, respectively

^aPesticide spraying is allowed during wind speeds < 3 m s⁻¹

Due to the higher degree of explained variance in pesticide concentrations, environmental data of the 5-day period before sampling were taken into consideration rather than those of the longer period. Air temperature showed a strong bivariate correlation with the area of apple and wine orchards within a radius of 500 m, respectively, representing the different weather and microclimate structures of the four valleys, and was thus excluded from the model. Playgrounds’ altitude (m a.s.l.) showed high VIFs, correlated with meteorological data (wind, temperature, irradiance) and was excluded as well. According to the final model, increasing pesticide contamination of playgrounds was best explained by increasing proportions of apple orchards around the playgrounds and increasing rainfall.

Pesticide contamination was also positively associated with interactions between distance and maximum wind speed, as well as interactions between hours allowed for spraying and prevailing wind speed. In contrast, significant negative associations for pesticide contamination was found for irradiance, wind blowing away from the playgrounds, hours when spraying is not recommended, and the distance to agricultural fields (Table 5 and Fig. 3).

Endocrine-active pesticides on playgrounds—possible impact on human health

This study is among the first demonstrating pesticide contamination in public playgrounds near intensively managed agricultural areas. Pesticide residues were found in 45% of the sampled playgrounds, while 24% of the playgrounds were contaminated with more than one compound; in total 12 different pesticides were detected. Overall, 11 of the 12 identified pesticides are endocrine-active substances, which were present in 24% of the investigated playgrounds [26]. Three detected pesticides, cypermethrin, oxadiazon and tetraconazole, are endocrine disruptors (Category 1, ED confirmed in animal studies, WHO/IPCS definition); chlorpyrifos-methyl, fluazinam, penthiopyrad and methoxyfenozide are suspected endocrine disruptors (Category 2, ED in–vitro confirmed) and difenoconazole, imidacloprid, penconazole and phosmet are endocrine-active substances or potential ED belonging to Category 3 [27, 54, 62].

The insecticide phosmet and the broad-scale fungicide fluazinam were the most abundant pesticides showing higher concentrations, particularly in the Venosta valley. Both pesticides are used to treat apple trees and grapevines. This unevenly distributed occurrence suggests an application of higher concentrations due to either different climatic conditions in that particular valley (Additional file 1: Fig. S1), and/or higher pest infestations during or prior to the sampling period.

Phosmet and fluazinam are endocrine-active substances and are reported on the endocrine disruptive compounds (EDC) criteria with level 3 and 2, respectively [4, 26]. According to the EDC screening program, fluazinam is an oestrogen receptor (ER)-antagonist and shows ER bioactivity [71].

Difenoconazole and penthiopyrad, both fungicides, were only found in the Isarco valley, a colder and more rain-laden region, suggesting that the orchards were affected by fruit-specific fungi before and during sampling period.

Dodine was the only detected pesticide without a link to possible ED activity. However, dodine was related to toxicity to humans, including carcinogenicity, reproductive and developmental toxicity, neurotoxicity, and acute toxicity [8, 12, 62].

Not much is known about effects by exposure to multiple substances. It is known, however, that the exposure to an increasing number of EDs in the environment is associated with an increasing incidence of hormone-dependent cancers like breast, prostate and thyroid and a decreasing sperm quality in the European and US population [24, 68, 69, 86]. It is important to emphasize that this study only analysed active components and that no consideration was given to metabolites and adjuvants, although these have also been classified as highly toxic [33, 50]. Children may get in contact through grass and soil at the playground and might additionally be exposed to pesticides through inhalation. Furthermore, places like playgrounds are visited especially by vulnerable groups (i.e., children and pregnant women) and the exposure time may be longer than in other areas [25, 81]. So far, however, no pesticide threshold levels for air, non-target vegetation or soil are defined. Although detected concentrations are relatively low, measured concentrations were close to permitted residual levels of fruits and vegetables: the maximum residual levels of fluazinam range from 0.01 ppm for citrus fruit to 0.3 ppm for apples (European [29]); median levels of fluazinam on playgrounds were 0.01 ppm and maximum levels were 0.26 ppm. One playground in the Venosta and one playground in the Isarco valley had higher concentrations than the residual levels for citrus fruits and similar to the residual levels of apples. Both playgrounds were near to agricultural sites. Also, the measured levels of tetraconazole and oxadiazon were very close to maximum residual levels (European [29]). Although it is unlikely that substantial amounts of contaminated grass are eaten by children, they may nevertheless get in contact with these substances. The concentrations reported here only allow for an overall statement that this concentrations might lead to relevant exposure scenarios for children. An important recommendation for future assessments would be to identify possible uptake pathways by direct ingestion and by uptake via the skin to find meaningful transfer factors, which allow for comparisons with relevant endocrine disruptor concentrations.

The obtained residual levels of these playgrounds are also likely for private gardens or grazing land next to agricultural sites. Therefore, also herbivore productive livestock feeding up to 100 kg and more grass per day are likely exposed to pesticide residues [32]. Concerning pesticide types and concentrations, our findings are comparable with those published in a recent Austrian study based on field margins [49], showing that insects like butterflies are endangered through pesticide exposure. Furthermore, earlier studies already showed that pesticide application decreased plant biodiversity [67]. Such pollutants disturb the natural processes and can even disrupt ecosystem functioning [15, 23]. However, future studies are needed to shed light on the effects of these substances in the environment.

More important than comparing the findings with permitted residue levels in food is the fact that the majority of pesticides on playgrounds are considered endocrine active because these substances can act like natural hormones in concentrations several orders of magnitude below common residue thresholds [76]. Additional mixtures of EDs at low dose produce significant combination effects also at levels well below the so-called no observed adverse effect level (NOAEL) [47] and it is still discussed whether the long-term exposure of low-dose ED mixtures is associated with chronic diseases [9, 34, 38]. Due to the low-dose effect of EDs, the non-monotonic dose–response relationship, and the interaction of EDCs with endogenous hormones, it is questionable whether even low doses of EDC can ever be considered as safe [57, 76]. Thus, a zero tolerance of pesticide residues with ED activity on non-targeted-areas would be comprehensible. So far, the current testing of pesticides with ED activity did not include low-dose and mixture effects, although new testing approaches are described [19]. With this study, we provide additional data on environmental contamination levels pointing to the necessity of future tests.

Pesticide contamination at playgrounds was only once investigated in Europe [66] and at a school yard in the USA [51] so far. Sapcanin et al. [66] found 42 different pesticides with concentrations up to 0.1 mg kg⁻¹ in soil samples, and Miersma et al. [51] reported up to 0.06 mg kg⁻¹ of DDE in soil samples of school yards [51]. However, both studies did not assess pesticide drift from nearby agricultural sites. In addition, the divergence in the detected pesticide types can be attributed to the diverse geographical locations including different pesticide regulations and probably to the type of analysed sample (grass), exhibiting differing degradation processes as compared to soil samples.

Besides pesticides, we also found one preservative agent (2-phenylphenol) and one germicide (benzalkonium chloride). The preservative agent 2-phenylphenol is used for post-harvesting treatment of citrus fruits to protect against microbial damage during storage and distribution [46]. The preservative agent was only detected once in Low Adige near Bolzano at a playground located 100 metres away from an agricultural site. Although this substance is no longer permitted as food additive in Italy [26], it is widely used from households to hospitals, farms and food processing plants and is used for disinfection of seed boxes and food packages and in the production of other fungicides. Furthermore, 2-phenylphenol is found in low concentrations in some household products, such as spray disinfectants and aerosols or deodorants [16, 59]. Similarly, benzalkonium chloride is a general disinfectant also present in sanitizing agents of all kinds.

Effects of non-occupational, long-term and low-dose pesticide exposure, however, are difficult to establish, as first many toxicological models do not comprehensively consider the numerous variables that can lead to diseases [34] and second it is difficult to account for the total pesticide exposure through all possible exposure routes (e.g., wind, rain, food, work). Most studies on non-occupational pesticide exposure have focussed on pesticide residues in food products and exposure via nutrition [37, 40, 72], whereas the pesticide exposure via drift is often neglected [77, 87].

Drift of pesticides—an underestimated fact

This study demonstrated that topographical and meteorological conditions like land use, rainfall, wind speed and its direction influence pesticide drift. Further, precipitation was shown to be positively associated with pesticide concentration. Besides the known direct deposition of pesticides via rainfall [63, 79], local weather conditions (i.e., higher precipitation and wet situations) might in addition motivate farmers to apply higher amounts of pesticides to counteract pest infestations, especially of fungi. As this study found a negative relationship between pesticide concentration and the distance to agricultural areas and a positive association between distance and the interaction with wind, we assume that direct drift via wind might be the dominant vector for pesticides ending up on the playgrounds. Hence, our results strongly suggest that other areas in similar distance to agricultural sites, such as organic fields, private gardens or other public spaces, are likely to be contaminated with similar residues and concentration ranges [81].

Our results show that the risk of pesticide contamination on playgrounds is lower at times with elevated wind speeds, when spraying is not allowed (Fig. 3g). In contrast, the combination of long periods of lower wind speeds (0–3 m s⁻¹), together with rising wind speed up to 3 m s⁻¹, significantly increased the contamination risk, suggesting that the farmers comply with the spray regulations regarding wind. The finding that global irradiance is negatively associated with pesticide contamination refers to pesticide degradation. The fact that the 5-day period prior to sampling had a stronger impact on pesticide concentration than data from the longer period lets suggest that the pesticides detected in this survey might degrade in a sufficient way. However, additional spraying events during the growing season may lead to accumulation of certain pesticides over the vegetation period.

The possibility that the pesticide residues on the playgrounds are partly due to a direct application can be excluded [39, 52], as the quantified pesticide concentrations are strongly associated with location (i.e., proportion of apple orchards in the vicinity) and local environmental conditions (i.e., hours of wind conditions when spraying is allowed) and are thus pointing towards a contamination by the surrounding pesticide-treated agricultural areas. Furthermore, most of the detected substances are only allowed for use in agriculture [26, 53, 62] that theoretically precludes the possibility of contamination by non-commercial use as the application on nearby private land or by community workers.

Besides drift, the aerial transport of volatile compounds is another pesticide pathway to non-target areas. Due to missing information about spraying events and frequencies, however, we could not include volatilisation data of the monitored compounds. In general, existing regulations demand the minimization of volatilization during pesticide spraying and application [28]. There are databases on compound degradation modelling metabolism in soils [35], but data on real compound amounts, as the European Commission regulation 283/2013 claims for, are still lacking. Our study reduces this knowledge gap. To comply this regulation, however, it is still essential that European member states initiate and support more comprehensive field studies.

The predicted impact of each environmental factor shows how pesticide contamination can be reduced with a careful planning, which considers local weather conditions. Our results show that a distance of at least 100 meters is necessary to obtain zero pesticide contamination (Fig. 3e). However, upper confidence intervals indicate that pesticide contamination is also likely within a distance of more than 300 m and that the additional impact of strong winds can lead to a maximum pesticide concentration up to 5 or more mg kg⁻¹ (Fig. 3e, f). Therefore, we highly recommend to take into account at least a distance of minimum 100 m to agricultural sites (Fig. 3e) and also to consider typical and local wind directions when planning and building public places like playgrounds, schools, pre-school and market areas. In addition, higher irradiance and dry weather conditions both reduced pesticide contamination (Fig. 3a, d). That is why we highly recommend to plan and conduct spraying events only during a longer period of sunshine without rainfall and low wind.