Identification of a pesticide exposure based market basket suitable for cumulative dietary risk assessments and food monitoring programmes

ABSTRACT

Consideration of cumulative dietary risks is a requirement in the European legislation for setting maximum residue levels for pesticides. Current cumulative exposure assessment methodologies strongly rely on representative occurrence data from food monitoring programmes. This study provides a sensitivity analysis, utilising (i) European consumption data expressed as raw agricultural commodity (RAC) equivalents from 23 different countries as published by the European Food Safety Authority and (ii) all maximum residue levels established for pesticides under European Regulation (EC) No 396/2005. Based on two different degrees of conservatism, market baskets consisting of 16 or 41 RACs, respectively, were identified, covering the majority of the total chronic and acute daily exposure. The coverage of the exposure by these market baskets was tested by comparison of cumulative probabilistic exposure assessments for the German population using all food commodities and those using the reduced sets. It was demonstrated that ≥85% of the total chronic exposure is already covered by 16 RACs, while 41 RACs are required to reach a similarly satisfying coverage of the total acute exposure. Results from this study support resource efficient modelling of complex cumulative assessment scenarios and may help to improve the design of food monitoring programmes with respect to a more efficient assessment of potential consumer risks.

KEYWORDS:

• Pesticide
• exposure
• monitoring
• cumulative
• risk

Introduction

In regulatory frameworks for pesticides a shift from single substance evaluations towards the assessment of complex mixtures and multiple residues in foods takes place. In the European Union (EU), consideration of “cumulative and synergistic effects, when the methods to assess such effects are available” is a mandatory request when setting maximum residue levels (MRLs) for pesticide residues in foods and feeds under Regulation (EC) 396/2005 (EU 2005). However; a harmonised methodology to address the cumulative risk for consumers has not yet been implemented and is still under scientific development. To accelerate this process, large research projects like EuroMix (EuroMix 2018) were funded by the European Commission (EC) to improve the knowledge both on mixture toxicity and on methodologies to estimate the cumulative exposure to multiple components.

Mixture toxicity is often related to specific adverse outcome pathways (AOPs) (OECD 2017) and therefore strongly linked to the toxicological properties of each component in the mixture. A common approach is the grouping of potentially relevant components into cumulative assessment groups (CAGs), for example based on their capability to induce similar toxicological effects e.g. to target organs (PPR 2013) or by their structural relationship like sharing the organophosphorus moiety (EPA 2006).

From an exposure assessment point of view, the individual modes of toxicity and the grouping of components are of low relevance for the dietary risk assessment methodologies. In all cases, occurrence data for each component of interest is correlated with information on the amount of food consumed in a specific timeframe. The underlying data and the principles in the exposure models used are mostly unaffected by the size of CAGs or by the AOP they are based on. However; with increasing complexity of models, computational capacity becomes the most limiting factor, both in time and in cost. Currently, most dietary exposure models addressing multiple residues like MCRA (van der Voet et al. 2015) or DEEM (EPA 2014) are based on probabilistic approaches. The general aim is the generation of an exposure distribution for the population group of interest, from which specific percentiles are selected for decision making (e.g. percentile 99.9) (EPA 2000). The data sources used in such models are typically food consumption surveys and results from representative food monitoring programmes. Especially the latter provide a large potential for optimisation of the sampling design to enable a focused and efficient modelling of cumulative dietary risks. Since these programmes have not been optimised in regard to the sensitivity of analysed matrices for the consumer exposure, sampling resources and computation capacities could be invested with maximum benefit.

In Europe, the food monitoring for pesticide residues is organised both on national level (e.g. Germany (Bundesministerium für Ernährung und Landwirtschaft 2015)) and in an EU-wide coordinated control programme (European Parliament and the Council of the European Union 2018). The underlying scientific concepts for the Germany Food Monitoring (Sieke et al. 2008a, 2008b) and for the EU Annual Coordinated Control Programme (EFSA 2015b) focus on a food basket covering a high percentage of the average daily consumption. Although this concept ensures coverage of important foods eaten frequently, it contains no substance related information on its sensitivity for the dietary risk assessment. The contribution to the total daily dietary exposure from highly consumed food commodities like cereal grains or meat could be much lower than for less consumed foods due to different residues concentrations. Especially for cumulative risk assessments based on probabilistic modelling, the upper percentiles of exposure distributions are driven by a very limited range of food commodities. For example, the cumulative dietary risk to organophosphorus and carbamate insecticides was mainly driven by 6 different raw food commodities each, representing 73% to 89% of the total exposure (Boon et al. 2008). In view of the plethora of possible combinations between foods and residue components, consideration of all possible cases would require high efforts and very broad, detailed data but often with only very limited impact on the final result.

In the current work, the sensitivity of raw food commodities for cumulative dietary risk assessments was determined considering the pesticide MRLs established in the European Union. Based on the results, a prioritisation of foods in monitoring programmes is suggested, which allows to focus on key commodities and to identify less relevant commodities which could be omitted from the programmes. Also, in view of future approaches to establish MRLs on the basis of cumulative effects, knowledge of key commodities may help to increase the practicability.

Material and methods

In the first step, a sensitivity analysis was conducted for the contribution of all raw food commodities to the long- (chronic) and short-term (acute) daily exposure. The results were expressed as percentages of the total exposure via all food items and as percentages of the compounds’ health based guidance values (HBGVs), namely the acceptable daily intake (ADI) or the acute reference dose (ARfD), as derived by EFSA in the assessment for the approval of active substance into Annexe I of the European Regulation (EC) 1109/2007 (The European Parliament and the Council of the European Union). The information was obtained from the Pesticide Database of the European Commission (European Commission (COM)). Data management was performed with Microsoft EXCEL while all model calculations were performed using R Software (R Core Team 2019).

Consumption data

Consumption data were obtained from the EFSA raw primary commodity (RPC) model (Dujardin and Kirwan 2019). The EFSA RPC model is based on 51 dietary surveys from 23 different European countries, representing a total of over 26 million consumption records covering 111 sub-populations. Within the EFSA RPC model, consumption data were transformed into their respective raw primary commodity equivalents according to Annexe I of Regulation (EC) 396/2005, based on recipe data and under consideration of processing effects by using yield factors. The complete daily consumption of a given commodity from all sources was obtained by aggregating all entries to one single raw commodity equivalent per commodity and survey. The RPC model provides statistical parameters (e.g. mean, median and selected percentiles) for the consumption of each raw food item per survey, based either on the total population (including non-consumers) or on consumers only (individuals with a daily portion size >0). In both datasets, consumption is expressed either on a “g per day” or “g per kg bodyweight and day” basis, from which the latter was used for direct comparison with HBGVs.

From the RPC model, the “total population” was considered to calculate the relative contribution of each commodity to the chronic exposure as suggested for predicting the dietary exposure of pesticide residues (Global Environment Monitoring System – Food Contamination Monitoring and Assessment Programme (GEMS/Food) 1997). For the acute exposure, consumption data for “consumers only” were considered. This approach reflects the procedure to derive large portions in the “International Estimate of Short-Term Intake” (IESTI) approach (FAO 1999; EFSA 2015a) which is broadly used in pesticide regulatory frameworks.

Occurrence data and toxicological information

The maximum concentration of pesticides allowed in food commodities according to Regulation (EC) 396/2005 as well as corresponding HBGVs were obtained from the European Commission’s Pesticide Database (date: 31.07.2019) (European Commission (COM) 2019). MRLs set at the limit of quantification (LOQ) were considered at the numeric value of the LOQ. Compounds without established MRLs have not been considered in the calculation. Also, MRLs normally address residue components according to the residue definition for enforcement purposes, primarily selected to ensure easy and cost-efficient analysis. For risk assessment purposes, it may be necessary to considered the presence of additional residue components with toxicological relevance. The ratio between these two approach is commonly described as a conversion factor (CF) (EFSA Panel on Plant Protection Products and their Residues 2016), which may be used to extrapolate from analytical enforcement results to the residue relevant for the exposure assessment. However, since CF are very limited in their availability, differences in the residue definitions for enforcement purposes and for risk assessment were not taken into account.

For considering the sensitivity of individual foods to the cumulative exposure, ADI and ARfD values are expected to provide a conservative estimate of the risk involved. Both HBGVs are established at the lowest “no observed adverse effect level” (NOAEL) multiplied by a safety factor (typically 100). When this lowest NOAEL is related to the toxicological effect of the CAG, the individual toxicity is directly addressed. When another toxicological effect provides the lowest NOAEL, the sensitivity is even overestimated.

Sensitivity analysis model

In the current work, a model was used to address the sensitivity of each food commodity to the total long- or short-term exposure. The sensitivity was expressed either by the relative contribution to the total daily exposure or by the utilisation of health based guidance values for each commodity.

The relative contribution to the total daily exposure was expressed in percent and calculated as the ratio between the exposure arising from each individual raw food commodity and the total mean exposure from all raw food commodities (Equation 1(1) $\begin{array}{rl}& Rel.contributionfoodX\left(\mathrm{\%}\right)\\ & =100\times \frac{\left(Meanorhigh\right)Consumptio{n}_{Foodx}\times Occurrenc{e}_{Foodx}}{{\sum }_{Food1}^{n}MeanConsumptio{n}_{Foodn}\times Occurrenc{e}_{Foodn}}\end{array}$(1) ). For each individual commodity, the mean consumption (“mean approach”) and the percentile P97.5 (or the next lower available percentile P90, P75 or mean, if the P97.5 was not available in the RPC model), were used (“high approach”). This calculation was performed for each consumption survey and each pesticidal active substance, providing a large overall distribution of relative contributions from all active substances per food commodity.

(1) $\begin{array}{rl}& Rel.contributionfoodX\left(\mathrm{\%}\right)\\ & =100\times \frac{\left(Meanorhigh\right)Consumptio{n}_{Foodx}\times Occurrenc{e}_{Foodx}}{{\sum }_{Food1}^{n}MeanConsumptio{n}_{Foodn}\times Occurrenc{e}_{Foodn}}\end{array}$(1)

In addition, the utilisation of HBGVs by the exposure for individual food commodities (Equation 2(2) $\begin{array}{rl}& ContributionHBGVfoodX\left(\mathrm{\%}\right)\\ & =100\times \frac{\left(Meanorhigh\right)Consumptio{n}_{Foodx}\times Occurrenc{e}_{Foodx}}{HBGV}\end{array}$(2) ) was calculated. The mean and high approach for the “total population” was assessed against the ADI, while only the high approach based on “consumers only” was assessed against the ARfD. Compounds with a very limited range of uses could result in high relative contributions of single food commodities to the total daily exposure. However, their relevance becomes low when only a very minor percentage of the ADI value is utilised. On the other hand, single combinations of commodity and MRL may already result in acute exposures close to the ARfD although the contribution to the long-term total exposure is minor. The calculations were performed for each survey and for each pesticidal active substance.

(2) $\begin{array}{rl}& ContributionHBGVfoodX\left(\mathrm{\%}\right)\\ & =100\times \frac{\left(Meanorhigh\right)Consumptio{n}_{Foodx}\times Occurrenc{e}_{Foodx}}{HBGV}\end{array}$(2)

For both approaches, summary statistics were applied to the total distributions of the relative contributions to the exposure and of the utilisations of HBGVs. Based on the reporting frequency in the RPC model, total distributions consisted of 455 up to 56000 individual results. When HBGVs have not been established or were not considered necessary, only the relative contribution to the exposure was calculated. All food commodities either posing a significant contribution to the total daily exposure or utilising a significant percentage of the HBGVs were taken into account to derive an overall market basket.

Reassessment of cumulative risks from pesticide residues in food

Based on the overall market baskets derived by the model, the impact of the reduced number of food commodities on probabilistic cumulative dietary exposure assessments was investigated. In 2018, the cumulative exposure against pesticide residues was estimated for three German sub-populations based on food monitoring data in 115 commodities (Sieke 2018). The cumulative assessment groups were obtained from EFSA and referred to chronic and acute effects to the nervous and thyroid system (PPR 2013). It should be noted that recently the CAGs for the nervous and thyroid system (Crivellente et al. 2019a, 2019b) have been updated. However; to estimate the relative coverage of the total diet under realistic conditions, the exact composition of the CAG is not expected to make much difference and therefore adaption to the new CAGs was not performed.

The datasets from the 2018 calculation were directly used for recalculations with the full range of commodities and with the reduced market basket. All scenarios were newly calculated with the MCRA 8.2 software version released in the meantime to identify differences solely based on the food selection, not on changes of the algorithm. Results for the reduced market baskets were expressed in terms of percentage of the exposure of the whole market basket for selected percentiles (percentile 50, 90, 95, 99, 99.9 and 99.99).

Due to the high percentage of non-detects in food monitoring, two different approaches were selected. For each CAG the limit of quantification (LOQ) was either used as such or substituted by zero. The calculations were based on German consumption data for children (0.5–4 years) (Heseker et al. 2003) and the general adult population (14–80 years) (Brombach et al. 2006; Krems et al. 2006; Max Rubner-Institut (MRI) 2016), however, for the latter the size of the database did not allow computation of the acute CAG for neurotoxic effects on the motor system (“motor division”).

Results

The results of the model were evaluated by calculating the mean, median, 90^th, 95^th and 97.5^th percentile and the maximum for the contribution of each food commodity and scenario (see supplemental information). Generally, a contribution of ≥10% was considered sufficiently significant to affect the outcomes of total exposure estimations. From the calculated statistical figures, the 95^th percentile (P95) was selected to provide a robust upper-bound estimate of the relevance of each food commodity. In addition, the 97.5^th (P97.5) was selected to correlate to the degree of conservatism involved in the IESTI concept to address potential cases relevant for acute exposure scenarios.

Total population

For the total population, contributions from individual food commodities to the overall exposure for each compound ranged from close zero up to 60% for the P95 and P97.5 (Tables 1 and 2). In total, 12 food commodities were identified as major contributors to the exposure in the mean approach (descending order: cow milk ≤61.4%, oranges <60.1%, sugar cane ≤32.4%, apples ≤31.9%, sugar beet roots ≤30.0%, wine grapes ≤28.9%, potatoes <23.7%, wheat grain ≤23.0% head lettuce ≤15.3%, tomatoes ≤15.1% and olives for oil production ≤12.3%). Taking also into account the individual chronic HBGVs for each pesticide, only nine food commodities remained at the P95 and P97.5 which were major contributors to the exposure and utilised the ADI by 10% or more (descending order: cow milk ≤77.6%, apples ≤35.9%, sugar beet roots ≤35.6%, oranges ≤31.9%, wheat grain ≤27.2%, sugar cane ≤24.3%, potatoes ≤15.3%, wine grapes ≤14.5% and tomatoes ≤13.3%).

Table 1. Total population chronic contribution (95^th percentile of the results, descending).

Commodity	Mean approach – % Contribution	Mean approach – % ADI	High approach – % Contribution	High approach – % ADI	Number of combinations
Cow milk	50.9	45.1	102.5	88.4	53244
Oranges	19.2	18.2	70.2	74.9	56376
Apples	19.2	19.7	57.5	67.2	56376
Sugar beet roots	16.8	16.8	42.0	44.6	56376
Wine grapes and similar	18.0	<10	67.6	26.4	56120
Wheat grain	21.8	15.2	40.0	31.1	56840
Sugar canes	12.7	12.5	29.9	33.1	56376
Tomatoes	15.1	<10	41.2	24.9	56376
Potatoes	10.0	<10	22.3	19.3	56492
Table grapes and similar	<10	<10	27.4	17.2	55144
Head lettuces	<10	<10	21.3	10.7	55290
Mandarins and similar	<10	<10	14.4	15.7	54432
Banana	<10	<10	12.2	14.0	56260
Olives for oil production	<10	<10	13.4	12.3	56260
Pears	<10	<10	12.8	12.5	56376
Strawberries	<10	<10	11.6	10.5	56376
Barley grains	<10	<10	11.9	<10	56235
Pig fresh meat	<10	<10	10.6	<10	53012
All other commodities	<10	<10	<10	<10	variable

Table 2. Total population chronic contribution (97.5^th percentile of the results, descending).

Commodity	Mean approach – % Contribution	Mean approach – % ADI	High approach – % Contribution	High approach – % ADI	Number of combinations
Cow milk	60.1	77.6	125.6	154.5	7752
Oranges	31.9	31.9	115.7	127.8	53244
Apples	28.9	35.2	86.1	119.2	56376
Wine grapes and similar	30.0	14.5	119.5	60.1	56376
Sugar beet roots	23.0	35.6	65.5	94.5	56120
Wheat grain	32.4	27.2	58.6	54.8	56376
Sugar canes	15.1	24.3	36.3	63.0	56840
Tomatoes	23.7	13.3	63.4	37.9	56376
Potatoes	15.3	15.3	33.4	36.9	56376
Head lettuces	11.0	<10	47.2	22.2	56492
Table grapes and similar	<10	<10	44.2	30.7	55290
Olives for oil production	12.3	<10	23.6	29.2	55144
Mandarins and similar	<10	<10	23.7	30.7	56260
Banana	<10	<10	22.1	26.8	54432
Pig fresh meat	<10	<10	19.0	27.3	56260
Pears	<10	<10	21.7	20.7	53012
Strawberries	<10	<10	20.6	19.5	56376
Bovine fresh meat	<10	<10	16.2	22.6	56376
Barley grains	<10	<10	22.4	10.6	53128
Peaches and similar	<10	<10	16.6	16.3	56235
Oat grain	<10	<10	16.7	14.4	55890
Durum wheat grain	<10	<10	16.0	15.0	55257
Sunflower seeds	<10	<10	15.0	14.8	56840
Spinaches	<10	<10	17.5	10.7	56376
Rice grain	<10	<10	12.1	12.3	55775
Maize grain	<10	<10	12.6	11.8	56840
Carrots	<10	<10	12.2	11.4	56724
Rye grain	<10	<10	13.1	10.0	56260
Chicken fresh meat	<10	<10	10.0	12.0	55257
Crisp lettuces	<10	<10	21.2	<10	53012
Teas leaves, dry and/or fermented, and similar	<10	<10	11.5	<10	44620
Rapeseeds	<10	<10	11.5	11.2	55031
Pineapples	<10	<10	10.5	<10	56376
Coffee beans, green	<10	<10	10.1	<10	54805
All other commodities	<10	<10	<10	<10	Variable

In the high approach, which calculates the percentage of a large daily portion size versus the mean daily consumption, 18 food commodities were identified with more than 10% contribution (Table 1) for the P95 and 34 food commodities for the P97.5 (Table 2). The RPC model expresses all food consumptions in terms of their raw agricultural commodity equivalents. Highly processed foods or ingredients like white sugar, hard cheese or olive oil show exceptionally high sensitivity in the calculation, since yield rates are low and therefore high amounts of raw commodity are used to prepare the processed food. Of the major contributors to the exposure, 16 food commodities at the P95 and 28 food commodities at the P97.5 also resulted in utilisations of the ADI of 10% or more.

Consumers only

When only consumers of a food are considered, the contribution of portion sizes to the total daily consumption would give unrealistic results, since the frequency of consumption is not considered anymore. The sensitivity of rarely consumed food commodities would be overestimated in comparison to frequently eaten food commodities. Thus, only the sensitivity against HBGVs is calculated to identify food commodities utilising a large percentage of the compound specific ADI or ARfD value. The aim is to address consumers with high consumption frequencies and consumers of single large portions.

For the mean approach, only 9 food commodities were identified (see Table 3) utilising the ADIs by 10% or more (in descending order: nettle ≤37.7%, oranges ≤37.4%, cow milk ≤21.7%, mandarins ≤15.9%, apples ≤14.3%, wheat grain ≤12.4%, quinces ≤11.9%, sugar beet roots ≤10.7% and lamb´s lettuce ≤10.1%). The high approach resulted in a total of 14 food commodities for the P95 and 38 food commodities for the P97.5 with at least 10% ARfD utilisation.

Table 3. Consumer only HGBV contribution (95^th and 97.5^th percentile of the results, descending).

Commodity	Mean approach – % ADI		High approach – % ARfD		Number
	P95	P97.5	P95	P97.5
Oranges	18.0	37.4	54.2	125.2	53460
Cow milk	13.1	21.7	33.8	53.4	50949
Mandarins and similar	<10	15.9	21.9	50.3	46656
Apples	<10	14.3	29.6	47.9	53946
Sugar beet roots	<10	10.3	21.0	42.2	53946
Nettle	17.3	37.7	20.3	40.6	2430
Wheat grain	<10	12.4	14.9	31.2	54390
Wine grapes and similar	<10	<10	13.4	22.8	52704
Table grapes and similar	<10	<10	13.6	20.5	51240
Grapefruits	<10	<10	<10	20.0	43254
Oat grain	<10	<10	<10	19.7	47922
Strawberries	<10	<10	10.2	19.6	53460
Bilberries	<10	<10	<10	19.6	11155
Banana	<10	<10	<10	19.2	53835
Tomatoes	<10	<10	12.0	18.9	53460
Sugar canes	<10	<10	12.0	18.4	53946
Sour cherries	<10	<10	<10	17.6	26675
Pineapples	<10	<10	<10	17.1	47045
Pears	<10	<10	10.2	16.5	51516
Barley grains	<10	<10	<10	16.3	49389
Peaches and similar	<10	<10	10.0	15.8	51030
Olives for oil production	<10	<10	<10	15.4	53835
Durum wheat grain	<10	<10	<10	15.1	54390
Lamb’s lettuces	<10	10.1	<10	14.1	13095
Head lettuces	<10	<10	<10	14.0	47045
Potatoes	<10	<10	<10	12.8	54057
Cherries (sweet)	<10	<10	<10	12.3	46075
Quinces	<10	11.9	<10	12.1	4860
Cranberries	<10	<10	<10	11.9	28674
Spinaches	<10	<10	<10	11.7	47045
Lemons	<10	<10	<10	11.6	52974
Plums	<10	<10	<10	11.6	46075
Sorrel	<10	<10	<10	11.2	3888
Goat fresh meat	<10	<10	<10	10.6	458
Melons and similar	<10	<10	<10	10.4	39285
Crisp lettuces	<10	<10	<10	10.2	32980
Beans (with pods) and similar	<10	<10	<10	10.2	50440
Watermelons and similar	<10	<10	<10	10.0	32010
All other commodities	<10	<10	<10	<10	variable

Consolidated market basket

Based on the analyses for the total population and for consumers only, all food commodities with high sensitivity (≥10%) towards the total daily consumption or to the exposure were identified (Table 4). For the P95, 16 food commodities with high sensitivity were identified whereas for the more conservative P97.5 this was a total of 41 food commodities. Normally, all commodities with significant contributions to the total daily diet are also significant commodities in terms of exposure. Exceptions are tea leaves, rape seeds and coffee beans. Although these foods may add more than 10% to the total exposure, related residues are too low to represent a significant utilisation of HBGVs. Also, sugar cane and sugar beet are consumed primarily as highly processed white sugar, for which residue concentrations are normally negligible (Scholz et al. 2018). In view of estimating the dietary exposure, these raw commodities can be considered as irrelevant.

Table 4. Consolidated market basket of commodities contributing significantly to the total daily diet and dietary exposure (95^th and 97.5^th percentile of the results, alphabetic order).

	P95							P97.5
Commodity	TP Mean ≥10% cont.	TP Mean ≥10% ADI	TP High ≥10% cont.	TP High ≥10% ADI	CO Mean ≥10% ADI	CO High ≥10% ARfD	Market basket inclusion	TP Mean ≥10% cont.	TP Mean ≥10% ADI	TP High ≥10% cont.	TP High ≥10% ADI	CO Mean ≥10% ADI	CO High ≥10% ARfD	Market basket inclusion	Remarks
Apples	✓	✓	✓	✓	×	✓	✓	✓	✓	✓	✓	✓	✓	✓
Banana	×	×	✓	✓	×	×	✓	×	×	✓	✓	×	✓	✓
Barley grains	×	×	✓	×	×	×	×	×	×	✓	✓	×	✓	✓
Beans (with pods) and similar	×	×	×	×	×	×	×	×	×	×	×	×	✓	✓	2
Bilberries	×	×	×	×	×	×	×	×	×	×	×	×	✓	✓	2, 5
Bovine fresh meat	×	×	×	×	×	×	×	×	×	✓	✓	×	×	✓
Carrots	×	×	×	×	×	×	×	×	×	✓	✓	×	×	✓
Cherries (sweet, sour)	×	×	×	×	×	×	×	×	×	×	×	×	✓	✓	2
Chicken fresh meat	×	×	×	×	×	×	×	×	×	✓	✓	×	×	✓
Coffee beans, green	×	×	×	×	×	×	×	×	×	✓	×	×	×	×	1
Cow milk	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓
Cranberries	×	×	×	×	×	×	×	×	×	×	×	×	✓	✓	2, 5
Goat fresh meat	×	×	×	×	×	×	×	×	×	×	×	×	✓	✓	2
Grapefruits	×	×	×	×	×	×	×	×	×	×	×	×	✓	✓	2
Lamb’s lettuces	×	×	×	×	×	×	×	×	×	×	×	✓	✓	✓	4
Lemons	×	×	×	×	×	×	×	×	×	×	×	×	✓	✓	2
Lettuces (head, crisp)	×	×	✓	✓	×	×	✓	✓	✓	✓	✓	×	✓	✓
Maize grain	×	×	×	×	×	×	×	×	×	✓	✓	×	×	✓
Mandarins and similar	✓	✓	✓	✓	×	✓	✓	✓	✓	✓	✓	✓	✓	✓
Melons and similar	×	×	×	×	×	×	×	×	×	×	×	×	✓	✓	2
Nettle	×	×	×	×	✓	✓	✓	×	×	×	×	✓	✓	✓	5
Oat grain	×	×	×	×	×	×	×	×	×	✓	✓	×	✓	✓	4
Olives for oil production	×	×	✓	✓	×	×	✓	×	×	✓	✓	×	✓	✓
Oranges	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓
Peaches and similar	×	×	×	×	×	✓	✓	×	×	✓	✓	×	✓	✓
Pears	×	×	✓	✓	×	✓	✓	×	×	✓	✓	×	✓	✓
Pig fresh meat	×	×	✓	×	×	×	×	×	×	✓	✓	×	×	✓
Pineapples	×	×	×	×	×	×	×	×	×	×	✓	×	✓	✓
Plums	×	×	×	×	×	×	×	×	×	×	×	×	✓	✓	2
Potatoes	✓	×	✓	✓	×	×	✓	✓	✓	✓	✓	×	✓	✓
Quinces	×	×	×	×	×	×	×	×	×	×	×	✓	✓	✓	4
Rapeseeds	×	×	×	×	×	×	×	×	×	✓	×	×	×	×	1
Rice grain	×	×	×	×	×	×	×	×	×	✓	✓	×	×	✓
Rye grain	×	×	×	×	×	×	×	×	×	✓	✓	×	×	✓
Sorrel	×	×	×	×	×	×	×	×	×	×	×	×	✓	✓	2, 5
Spinaches	×	×	×	×	×	×	×	×	×	✓	✓	×	✓	✓
Strawberries	×	×	✓	✓	×	✓	✓	×	×	✓	✓	×	✓	✓
Sugar beet roots	✓	✓	✓	✓	×	✓	✓	✓	✓	✓	✓	✓	✓	✓	3
Sugar canes	✓	✓	✓	✓	×	✓	✓	✓	✓	✓	✓	×	✓	✓	3
Sunflower seeds	×	×	×	×	×	×	×	×	×	✓	✓	×	×	✓
Table grapes and similar	×	×	✓	✓	×	×	✓	✓	×	✓	✓	✓	✓	✓
Teas leaves, dry and/or fermented	×	×	×	×	×	×	×	×	×	✓	×	×	×	×	1
Tomatoes	✓	×	✓	✓	×	✓	✓	✓	✓	✓	✓	×	✓	✓
Watermelons and similar	×	×	×	×	×	×	×	×	×	×	×	×	✓	✓	2, 4
Wheat grain (common, durum)	✓	✓	✓	✓	×	✓	✓	✓	✓	✓	✓	✓	✓	✓
Wine grapes and similar	✓	×	✓	✓	×	✓	✓	✓	✓	✓	✓	×	✓	✓
All other commodities	×	×	×	×	×	×	×	×	×	×	×	×	×	×	2

Remark 1: no inclusion, only relevant in terms of contribution to the total daily consumption but not in terms of exposure.

Remark 2: inclusion, potentially relevant for short-term exposure.

Remark 3: primarily consumed as white sugar, which does not contain quantifiable residues. No consideration necessary.

Remark 4: extrapolation from other commodities according to EU Guidance for Extrapolation possible.

Remark 5: only relevant based on single RPC diets like infants or specific European regions.

✓: relevant.

×: not relevant.

TP: Total population.

CO: Consumers only.

For some commodities separately listed in the EFSA RPC model (e.g. bilberries and strawberries) only one common code exists in the Regulation (EC) No 396/2005 and the respective MRL applies to all sub-commodities. Aggregation to one representative commodity is therefore possible.

Recalculation of previous cumulative exposure assessments

Chronic CAGs

The coverage of the total exposure by the reduced market baskets was generally higher for the upper percentiles of the exposure distribution relevant for decision making like the P99.9. Since lower percentiles represent the background exposure consisting of a plethora of foods with multiple active substances, the outcome is strongly affected by any reduction of the number of commodities. In contrast, the upper percentiles are mainly driven by single food commodity and compound combinations, which remain unchanged as long as sensitive pairings are still considered.

Based on the strongly reduced P95 market basket, the median coverage of the exposure calculated for the total market basket was less than 70% for the 50^th percentile, but increased to 85–89% for the 99.9^th percentile (Figure 1). The results were in the same order of magnitude for all 11 CAGs considered for the chronic cumulative exposure. The impact of the LOQ scenario was low. Although absolute exposures differed strongly due to the high amount of “<LOQ” values substituted by zero in the “LOQx0” scenario, the relative change by the reduction of food items was identical for both approaches.

Figure 1. Percent chronic exposure based on P95 market basket relative to the total market basket.

As expected, the percent coverage for the larger P97.5 market basket was higher, ranging from 82–89% for the 50^th percentile up to 89–94% for the 99.9^th percentile (Figure 2). The LOQx1 scenario generally resulted in a higher coverage of the total exposure than the LOQx0 scenario. However, the differences were still small with less than 10% deviation between both scenarios.

Figure 2. Percent chronic exposure based on P97.5 market basket relative to the total market basket.

Acute CAGs

The recalculation of acute scenarios for the CAGs showed a much lower coverage of the total exposure for the reduced P95 market basket (Figure 3). For the LOQ×0 scenario, the 50^th and 90^th percentile of the exposure distribution represented less than 60% of the total exposure. Median levels for higher percentiles like the 99.9^th percentile were below 70%, suggesting a significant underestimation of the total acute exposure based on the 16 food commodities selected. The LOQ×1 scenario gave more robust results, showing coverages of the total exposure above 80% for all scenarios.

Figure 3. Percent acute exposure based on P95 market basket relative to the total market basket.

The larger P97.5 market basket also takes into account food commodities, which potentially utilise a significant part of the ARfD (Figure 4). Consequently, the coverage of the total acute exposure was much higher with >90% for the 90^th to 99.9^th percentiles. The 50^th percentile again resulted in a low coverage of <70% of the total acute exposure.

Figure 4. Percent acute exposure based on P97.5 market basket relative to the total market basket.

One deviation was observed for the CAG on acute effects on the autonomic system, which gave less than 60% coverage of the total exposure at higher percentiles like the 99.9th. Detailed investigation of the risk drivers in this group revealed, that the combination of oxamyl in peppers represents a main contributor to the exposure, which is not covered by the P97.5 market basket. In the underlying German monitoring data used for the probabilistic modelling, three samples with oxamyl residues above the LOQ were reported, two of them exceeding the established MRLs by a factor of up to 25. The sample with the highest residue already resulted in an exposure above the ARfD for oxamyl per se (BVL 2011). An additional calculation of the P97.5 market basket plus peppers resulted in an increased coverage of the total exposure of 73.7% at the 99.9^th percentile (Figure 5), demonstrating the high sensitivity of the probabilistic cumulative exposure calculation for these two non-MRL compliant samples.

Figure 5. Impact of peppers to the acute cumulative exposure for the CAG on motor division effects to the nervous system (LOQx0 scenario).

Discussion

The current assessment has demonstrated that for probabilistic cumulative exposure calculations a strongly reduced number of raw food commodities still covers the majority of the total exposure. For chronic scenarios, where single high findings level out for the long-term exposure, already the small P95 market basket consisting of 16 raw food commodities gave a coverage of ≥85%. However, also short-term cumulative effects are relevant and may already be induced by single compound-commodity combinations. Only the larger P97.5 market basket provides a sufficiently high coverage of the total short-term exposure (≥90%) to address this scenario. Under practical conditions, the coverage is likely to be even higher than calculated, since possible extrapolations between closely related commodities, as they are currently accepted in the EU (e.g. wheat to rye grain or tomatoes to aubergines) (EC 2017), have not been considered yet.

Consequently, when using exposure based market baskets, resources can be focused to improve the robustness and the sample size of residue data for the key commodities identified. Additional information like processing factors or information on the agricultural use can be implemented in the modelling to improve the outcome. Finally, technical resources for the modelling itself in terms of computing power and time could be redirected to put more emphasis on a broader range of sub-populations or add factors like non-dietary exposure pathways or benchmark-dose modelling of HBGVs.

In the European multi-annual control programme for pesticides (EC 2019), currently 30 plant and 6 animal food commodities are considered for the years 2020 to 2022. From the P95 market basket, all commodities except mandarins are already scheduled for analysis in the EU. In comparison to the P97.5 market basket, significant differences exist. Beans with pods, cherries, goat meat, lemons, maize grain, mandarins, pineapples, plums and sunflower seeds gave a significant contribution to the total exposure, but are currently not part of the monitoring programme. Also lamb´s lettuce, quinces and watermelons were identified as relevant, but residues in these commodities can be extrapolated from lettuce, apples/pears or melons, respectively, and do not necessarily require specific sampling. Additionally, bilberries, cranberries, nettle and sorrel, which were identified as relevant contributors to the exposure for single populations reported in the EFSA RPC model, are not covered by the EU coordinated control programme. In contrast, kiwi fruits, dry beans, aubergines, broccoli, sweet peppers, cultivated fungi, head cabbage, bovine liver and chicken eggs are included in the current monitoring programme, although the contribution of these commodities is too low to influence the total long- or short-term cumulative exposure significantly at the upper percentiles.

The European multi-annual control programme for pesticides primarily relies on the Pesticide Monitoring Programme designed by EFSA (EFSA 2015b), which ranked food commodities from multiple European Member States according to their proportions in terms of the total daily consumption. The 30 food items identified by this approach represented up to 74% of the daily consumption. However; the contribution of food items to the daily consumption is not necessarily related to their total exposure. Food commodities with large portion sizes but diminutive residues might falsely be identified as relevant while food commodities with small portion sizes but high residues might be omitted. In terms of exposure, the P95 market basket based on 16 raw food commodities already covers a higher percentage of the total exposure (≥85%) than the 30 food commodities suggested by EFSA.

Although the proposed market baskets already represent a refinement in view of consumer exposure assessment, it became obvious that the rate of non-MRL compliant samples may have a high sensitivity to the upper percentiles of the exposure distribution. Frequent findings of residues resulting in potential exceedances of the ARfD per se, like oxamyl in peppers, require consideration both in cumulative exposure modelling and in establishing monitoring programmes.

It needs to be mentioned that the current sampling approaches for pesticide residues in monitoring programmes are a hybrid between the purposes of estimating the consumer exposure and identifying potential MRL violations. For example, the Pesticide Monitoring Programme designed by EFSA or the Pesticide data programme in the USA (USDA 2019) both rely on foods which are highly consumed, especially by vulnerable sub-populations like infants and children. The focus is normally based on the raw/fresh commodity, while processed foods are only considered following simple transformations like cooking, freezing or juicing. The selection criteria are solely the contribution to the diet, not the sensitivity towards the exposure of pesticide residues. Another approach is followed for market baskets used in total diet studies (TDS). These kind of studies allow measurement of the average exposure by analysis of representative samples for the majority of foods consumed. Again, the selection of food items follows the principle to cover a high percentage of the daily consumption, but in contrast to food monitoring programmes the food list is amended by specific food items with high contribution to the exposure. For example, in the second Fresh TDS (Sirot et al. 2009) 51 additional foods have been identified adding significantly to the exposure. The main advantage of a TDS is the good approximation to the true average dietary exposure, since effects like industrial processing, blending or the composition of complex foods is already taken into account. However, total diet studies also have the drawback that their design is limited to measure mean residues and to estimate the chronic dietary exposure. It does not allow assessments on the acute dietary exposure or on the MRL compliance of individual samples (European Food Safety Authority/FAO/WHO 2011).

The newly proposed exposure based market basket keeps the principles of the established monitoring programmes to assess both the dietary exposure and the MRL compliance, but was amended with the relevance of each food towards the exposure. Exposure based market baskets might be helpful when in future cumulative and synergistic effects have to be taken into account for establishing legal limits for pesticide residues as demanded by EU legislation (EU 2005). Since consumer protection is the key intention for such a requirement, the range of commodities for which interaction between active substances needs to be regulated could be drastically reduced.

In summary, the exposure based market baskets identified in the current work allow better priorisation of raw food commodities for food monitoring programmes. While the current design of the European food monitoring relies on the contribution of each food to the daily consumption only, a more focused layout becomes possible when potential residue concentrations and HBGVs are taken into account additionally. In parallel, the proposals allow more resource efficient approaches both for cumulative and single compound based exposure assessments and provide a quantitative estimate of the uncertainties arising from a limited set of foods.

Disclaimer

This Paper presents the opinion of the author and not necessarily the regulatory views of the BfR.

Acknowledgments

The author would like to thank the MCRA team at the RIVM, especially Dr. J. van Klaveren and G. van Donkersgoed for their support.

Supplementary material

Supplemental data for this article can be on the publisher’s website.