Environmental Research 126 (2013) 125–133

Contents lists available at ScienceDirect

Environmental Research

0013-93http://d

n CorrE-m

journal homepage: www.elsevier.com/locate/envres

Identification of the main pesticide residue mixtures to which theFrench population is exposed

A. Crépet a,n, J. Tressou b, V. Graillot a,c, C. Béchaux a, S. Pierlot a, F. Héraud a, J. Ch Leblanc a

a ANSES, French Agency for Food, Environmental and Occupational Health & Safety, 27-31 Avenue du général Leclerc, 94701 Maisons-Alfort, Franceb INRA-UR 1204 Met@risk, Food Risk Analysis Methodologies, National Institute for Agronomic Research, 16 rue Claude Bernard, 75231 Paris, Francec INRA, UMR1331, Toxalim, Research Centre in Food Toxicology, INP, ENVT, EIP, UPS, National Institute for Agronomic Research, F-31027 Toulouse, France

a r t i c l e i n f o

Article history:Received 30 November 2011Received in revised form25 March 2013Accepted 27 March 2013Available online 15 June 2013

Keywords:Pesticide residue mixturesCo-exposure clusteringMultiple exposureFood risk assessmentCombined effects

51/$ - see front matter & 2013 Elsevier Inc. Ax.doi.org/10.1016/j.envres.2013.03.008

esponding author. Fax: +33 1 49 773892.ail addresses: amelie.crepet@anses.fr, crepet7@

a b s t r a c t

Owing to the intensive use of pesticides and their potential persistence in the environment, variouspesticide residues can be found in the diet. Consumers are therefore exposed to complex pesticidemixtures which may have combined adverse effects on human health. By modelling food exposure tomultiple pesticides, this paper aims to determine the main mixtures to which the general population isexposed in France. Dietary exposure of 3337 individuals from the INCA2 French national consumptionsurvey was assessed for 79 pesticide residues, based on results of the 2006 French food monitoringprogrammes. Individuals were divided into groups with similar patterns of co-exposure using theclustering ability of a Bayesian nonparametric model. In the 5 groups of individuals with the highestexposure, mixtures are formed by pairs of pesticides with correlations above 0.7. Seven mixtures of 2–6pesticides each were characterised. We identified the commodities that contributed the most toexposure. Pesticide mixtures can either be components of a single plant protection product appliedtogether on the same crop or be from separate products that are consumed together during a meal. Ofthe 25 pesticides forming the mixtures, two—DDT and Dieldrin—are known persistent organic pollutants.The approach developed is generic and can be applied to all types of substances found in the diet in orderto characterise the mixtures that should be studied first because of their adverse effects on health.

& 2013 Elsevier Inc. All rights reserved.

1. Introduction

The use of pesticides, which has become essential over the lastfew decades owing to intensive farming, results in the widespreadpresence of residues in water, soil and foodstuffs. In its last report,the European Food Safety Authority acknowledged that pesticideresidues were detected in 46.7% of the 67,887 food samplesanalysed throughout the European Union in 2008. Residues of atleast two pesticides were found in 26.7% of the analysed samples,one third of which contained more than 4 pesticide residues(EFSA, 2010a). Consumers are thus exposed to complex mixturesof pesticides which are a suspected risk to human health (EFSA,2008; WHO, 2009). Currently, there is an international consensuson the need to consider chemical mixtures when characterisingthe risk of human co-exposures (ACROPOLIS, 2010–2013;European Commission, 2011; Meek et al., 2011). The risk assess-ments conducted worldwide focus on chemicals belonging to thesame chemical family (carbamates, organophosphorus pesticides,triazoles) and/or sharing the same mechanisms of action

ll rights reserved.

gmail.com (A. Crépet).

(ACROPOLIS, 2010–2013; Boobis et al., 2008; Bosgra et al., 2009;EFSA, 2009; EPA, 2002, 2007; Müller et al., 2009; Reffstrup et al.,2010). They are based on cumulative exposure and on thecumulative effects of these substances, assuming that interactionsbetween substances are unlikely to have a significant effect onhealth. However, interactions between pesticides cannot be ruledout (EFSA, 2012). Moreover such studies are of limited interest forhealth risk assessment, as they may not reflect real mixtures ofpesticides likely to occur in actual diets. Indeed, individuals areprobably exposed to pesticides from different chemical families.

The PERICLES research programme coordinated by the FrenchAgency for Food, Environmental and Occupational Health & Safety,aims firstly to identify pesticide mixtures to which the Frenchpopulation is exposed through diet and secondly to investigate thepossible combined effects of their components on human cells.Within this programme, an original Bayesian nonparametricmodel was developed to define the combination of pesticides.The main advantage of the Bayesian nonparametric approach isthat no assumption is made on the form of the distribution of co-exposure to pesticides while it is based on a random mixture ofparametric distributions. The number of components of themixture is automatically inferred from the data in contrast withthe classical approaches that require the specification of the

A. Crépet et al. / Environmental Research 126 (2013) 125–133126

number of mixture components. This eventually allows represent-ing any kind of complex distributions, possibly multimodal.Pesticides forming mixtures are selected based on their probabilityof occurring simultaneously, considering both the pesticide resi-due patterns in food and the population's dietary habits. In Crépetand Tressou (2011), the mathematical description of the modelwas specified and applied to co-exposure of the French adult andchild population to 79 pesticides. For the adult population, thisstudy resulted in the selection of 20 mixtures composed of 3–17pesticides. In this first application, the non-detected values ofpesticide residues in food were assumed to be uniformly distrib-uted between 0 and the limit of reporting, which corresponded tothe analytical limit of detection or quantification. However thismajor assumption was unrealistic when pesticides were not usedon crops. New calculations were then conducted, assuming anabsence of residues in food samples with analytical values belowthe limit of reporting. This paper describes the mixtures obtainedon the basis of this assumption, the proportions of each pesticidecomposing the mixtures and the foodstuff which contributed mostto exposure to the different mixtures.

2. Material and methods

2.1. Pesticide selection and residual concentration

The data on pesticide residues in food and drinking water are based on annualcontrol and monitoring programmes implemented in 2006 by the French minis-tries in charge of consumer affairs, agriculture and health. These programmesprovide analytical results for up to 300 different pesticide residues screened for inabout 150 types of raw agricultural commodities, water included. Sampling ofmonitoring programmes was carried out in order to collect commodities repre-senting those sold on the French market. Results from control programmes are notrepresentative but were included because for most commodities the levels do notdiffer from those of the monitoring programmes (ANSES, 2009a, 2010a). The year2006 was chosen because it corresponds to that of the food consumption survey.Residues of 79 pesticides, for which at least 10% of analytical results werequantified, were retained for the study. They were analysed in 120 raw agriculturalcommodities, representing 306,899 analytical results and 8364 food/residuecombinations. For each combination, the observed residue levels were modelledusing a histogram distribution. Each histogram interval was constructed using twoconsecutive observed residue levels, and the probability of falling within theinterval was estimated using the proportion of corresponding observed values. Anull residue level was attributed to analytical results reported as “below the limit ofreporting” as recommended by the European Food Safety Authority (EFSA, 2010b)when the proportion of censored data exceeded 80%.

2.2. Individual food consumption data

Consumption data was extracted from the second “Individual and NationalStudy on Food Consumption”, INCA2 survey, carried out by the French Food SafetyAgency (ANSES, 2009b; Dubuisson et al., 2010; Lioret et al., 2010). This survey wasconducted between 2005 and 2007 and took into account seasonal variations. Twoindependent population groups were included in the study: 2624 adults aged 18–79 and 1455 children aged 3–17. Participants were selected using a three-stagerandom probability design stratified by region of residence, size of urban area andpopulation category (adults or children). Each participant was asked to complete ananonymous seven-day food diary. Foods were subsequently categorised into 1280“as consumed” food items (INCA2 classification). To match the consumption datawith the pesticide residue levels in food, the “as consumed” foods were brokendown into 181 raw agricultural commodities. To do this, 763 standardized recipesdefined by the French Food Safety Agency that take into account industrialprocesses, home cooking habits and edible portions for the INCA2 survey, wereused (ANSES, 2010b). Based on the sampling weights provided for each individualwhich represented their frequency in the entire French population, two samples of1898 adults and 1439 children were built from the original samples, by carrying outrandom trials with replacement. Only individuals whose energy needs are coveredby the declared consumption were included. The time window for short-termexposure calculation was a 24-h day. To avoid autocorrelation problems, 1 day wasrandomly selected for each individual from the records of 7 consecutive days.

2.3. Dietary co-exposure assessment

For each individual surveyed in INCA2, the consumed quantities of commod-ities containing a given pesticide were multiplied by the associated residue levelsand cumulated across commodities to obtain the total daily exposure to thepesticide. Exposure was expressed in micrograms of the chemical per kilogram ofconsumer body weight for the selected day (mg/kg BW/d). A series of 100 possibledaily exposure values to the 79 pesticides was calculated for each individual byrandomly selecting residue levels from each commodity/residue histogram. Theabove series of exposures represented the total uncertainty of the exposure of anindividual during a given day. One hundred values are a sufficient number toaccurately estimate the exposure distribution as was tested in Crépet and Tressou(2011). To make comparisons between exposure levels possible, exposures werelog-transformed, centred around the mean and rescaled by the standard deviationfor each pesticide and across individuals. Finally, the 95th percentile of exposure toeach pesticide for each individual was retained to form the final data set.

2.4. Individual co-exposure clustering

As diet is a combination of various foods consumed by individuals, theunivariate distribution of the population exposure to a pesticide is multimodal.This implies that the population can be clustered into groups according to the levelof exposure to a pesticide. Moreover, each individual is exposed to severalpesticides each day. The clustering should be performed by considering the levelsof exposure to the 79 selected pesticides. Therefore, individuals can be clusteredinto a certain number of groups which are homogeneous in terms of exposureprofile. Using the Bayesian nonparametric approach detailed and validated inCrépet and Tressou (2011), co-exposure was modelled as a mixture of multivariatenormal distributions that made it possible to account for the correlations betweenpesticides in each group of individuals. The mixing distribution indicating thegroup to which each individual was assigned was modelled with a Dirichlet process(Ferguson, 1973; Lo, 1984). Dirichlet processes are the natural extension of Dirichletdistributions used to model vectors of proportions summing to one when thelength of the vector is unknown. In our case, this implies that each individual willbelong to a certain group of the population in terms of its co-exposure topesticides, the number of groups being unknown. The means and the covariancematrices of the multivariate normal distributions are drawn from a Wishart-Normal distribution as this is the classically chosen conjugate prior for themultivariate normal distribution. This is a technical choice simplifying computationthat does not imply any specific assumption regarding the form of the distributionof co-exposure.

2.5. Correlations between pesticide exposures

For each group of individuals, the model produced a specific vector of meanexposures to the 79 pesticides and a matrix of covariance. Groups of individualswith higher exposure, i.e. with a majority of means exceeding the population mean,were selected to define mixtures of pesticides. The covariance matrices associatedwith the selected groups were studied to determine the pesticides to include in themixtures. For each selected group of individuals, pesticides with at least onecorrelation above an arbitrary fixed value of 0.7 were selected. This choice isdiscussed in Section 4. The proportion of each residue from a mixture wascalculated using the ratio of its mean exposure level and the sum of the meanexposures of all residues composing this mixture. For mixtures shared by severalgroups of individuals, the mean of the different proportions obtained from eachgroup was used.

2.6. Main food contributors to the pesticide exposure mixtures

In order to determine the raw agricultural commodities that explain exposureto the mixtures, a principal component analysis was performed for each mixturewhile considering exposure of the general adult population. Individual exposure toa pesticide was then broken down into individual exposures by raw agriculturalcommodities used as supplementary variables in the analysis. The commoditiesmost correlated with the main axes were those that most explain the population'sexposure to the mixture (Fisher test with p-value o0.05). Therefore, there was astrong relationship between exposure to the mixture and the consumption of aparticular raw agricultural commodity. This was either due to the presence of ahigh level of a single pesticide of the mixture in this commodity, or because therewere several pesticides of the mixture in this commodity. Finally, for each selectedcommodity mean exposures to each pesticide forming a mixture were summed toidentify the commodities with the highest levels of exposure to a mixture. Pleasenote that this sum had no biological significance but was useful to identify thecontribution of a specific commodity to the total mixture exposure.

A. Crépet et al. / Environmental Research 126 (2013) 125–133 127

3. Results

3.1. Groups of individuals with similar patterns of pesticide exposure

The model clustered the co-exposure of the child populationinto 30 groups. The majority of children were placed in three maingroups composed of 699, 238 and 239 individuals. The 27remaining groups were not analysed here as they did not cover alarge enough part of the child population (fewer than 10 indivi-duals per group). The first two groups consisted of children withhigh exposure to a large number of pesticides, whereas the lastgroup was composed of children exposed to the lowest levels ofthe 3 groups. Fig. 1 represents the co-exposure to the 79 pesticidesof the children from group 2 (light blue lines) and the childrenfrom group 3 (dark blue lines). The different co-exposure patternsof both groups are clearly visible on this figure. Since the data werecentred, the mean exposure of the whole child population is equalto zero. High exposure of a population group to a pesticide wasthus identified considering that all exposure values, or at least themedian exposure of this group, were above 0. To illustrate this, thebox plots for the three main groups of children are shown. Forexample, in Fig. A.1 in the appendix, the box plot of exposure toAzoxystrobin for group 2 individuals is above zero, which indicates“high” exposure to this pesticide in this group. In contrast, the boxplots for the individuals from group 3 (Fig. A.2 in the appendix) arebelow zero, indicating “low” exposure to the pesticide. The adultpopulation is clustered into 4 main groups of 256, 655, 302 and489 individuals respectively. The remaining groups only contain

Fig. 1. Parallel coordinates of the co-exposures to the 79 pesticides for two children grou238 children of group no. 2 (highly exposed) and the dark blue lines represent the co-expreferences to color in this figure legend, the reader is referred to the web version of th

a small number of adults and, hence, are not taken into considera-tion to identify mixtures. The first three groups had the highestexposures to a large number of pesticides.

3.2. Mixtures of pesticide residues

After screening the correlations between exposures to the 79pesticides in the 2 most exposed groups of children and the 3 mostexposed groups of adults, 25 pesticides with at least one correla-tion above 0.7 were selected. Fig. 2 shows correlation levelsbetween exposures to the selected pesticides for child group 2.This figure makes it possible to determine mixtures 2, 5 and 6. Forexample, most children clustered in group 2 are simultaneouslyexposed to chlorfenvinphos, ethion and linuron which are com-bined into one mixture (mixture 2 in Fig. 2). In this group,exposure to DDT is also highly correlated to exposure to dieldrin:another mixture is therefore created (mixture 5 in Fig. 2). Mixtures1, 4, 7 and part of mixture 3 are also shown in Fig. 2 withcorrelation factors under 0.7. These mixtures are based on theexposure correlations of child group 1 and of the adult groups.Following this procedure, 7 mixtures composed of 2–6 pesticidesare defined and listed in Table 1 for children and in Table 2 foradults. The groups of individuals from which each mixture hasbeen created have a “+” sign. Except for mixture 1, which onlyconcerns the child population, all mixtures are common to bothpopulations. Mixtures 2, 5–7 made up of 2 or 3 pesticides areshared by 4 different groups, whereas mixtures 1, 3 and 4, madeup of 5 or 6 pesticides, have only been determined from the

ps obtained with the model. The light blue lines represent the co-exposures of theosures of the 239 children of group no. 3 (lowly exposed). (For interpretation of theis article.)

DDTFenitrothionTriadimenolQuinoxyfenPenconazoleChlorprophamMaleic hydrazideEthionLinuronChlorfenvinphosFenhexamidLambda-CyhalothrinFludioxonilPyrimethanilProcymidoneIprodioneCyprodinil

Cap

tan

Phos

alon

ePr

opar

gite

Toly

lfluan

idD

iphe

nyla

min

eM

ethi

dath

ion

Imaz

alil

Die

ldrin

DD

TFe

nitro

thio

nTr

iadi

men

olQ

uino

xyfe

nPe

ncon

azol

eC

hlor

prop

ham

Mal

eic

hydr

azid

eEt

hion

Linu

ron

Chl

orfe

nvin

phos

Fenh

exam

idLa

mbd

a-C

yhal

othr

inFl

udio

xoni

lPy

rimet

hani

lPr

ocym

idon

eIp

rodi

one

Cyp

rodi

nil

CaptanPhosalonePropargiteTolylf luanidDiphenylamineMethidathionImazalilDieldrinDDT

Fig. 2. Correlations between the pesticide exposures of the 238 children of the group no. 2. Underline cocktails 2, 5 and 6, result from the pesticide exposure combinations ofthis group. They are composed of pesticides with correlations upper than 0.7. The five greener colours indicate the different correlation levels broken as o0.5; 0.5–0.7; 0.7–0.8; 0.8–0.9; 0.9–1. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

A. Crépet et al. / Environmental Research 126 (2013) 125–133128

correlations of adult group 3. For each mixture, one or twopesticides are found in high proportions. For example mixture7 consists of 97% imazalil and only 3% methidathion. Exposurelevels to the pesticides forming the mixtures are higher in thegroups from which they have been created than in the generalpopulation (results not shown).

Apples and pears are food items with significant correlations tothe pesticides in mixture 1. Regarding exposure levels, apples arethe main vector of exposure to mixture 1. The raw agriculturalcommodities with significant correlations to the pesticides ofmixture 2 are carrots, parsley, turnips, celeriac and herbs. Howeveronly carrots are associated with high exposure to the pesticides ofmixture 2 and can thus be considered as a main vector of exposureto this mixture. Only one main contributor is found for mixtures3 and 5: table grapes and fish products respectively. The fooditems which contribute most to exposure to mixture 4 are shownby decreasing levels of exposure: salad greens other than lettuce,table grapes, peaches, lettuces, strawberries and tomatoes. Pota-toes and onions have significant correlations with pesticides ofmixture 6 but potatoes are vectors of higher exposure levels thanonions. The main contributors to the pesticides of mixture 7 arecitrus fruits, listed here by decreasing levels of exposure: oranges,mandarin oranges and grapefruit.

4. Discussion

This work resulted in the characterisation of 7 mixtures com-prising 2–6 pesticide residues. Several factors may explain thesemixtures. First, the active substances the residues of which form a

mixture can be components of the same plant protection product.For example, 8 out of the 19 plant protection products currentlyauthorised in France and made with cyprodynil also containfludioxonil (Ministère de l'agriculture et de la pêche, 2011). Bothare components of mixture 4. Active substances the residues ofwhich form a mixture can also be components of different plantprotection products used on the same crop. For example, the rawagricultural commodity that contributes the most to exposure tomixture 3 is table grapes. Two fungal diseases can contaminategrapevines: oidium and grey mould. Penconazole and quinoxyfenare both used to treat oidium whereas pyrimethanil and fenhex-amid are applied to eliminate grey mould. These 4 pesticidesbelong to mixture 3. Chlorpropham and maleic-hydrazide whichmake up mixture 6 can both be used to inhibit germination ofpotatoes, one of the main foods that contribute to mixture6 exposure. More generally, certain components of plant protec-tion products used on different raw agricultural commodities areconsumed together during a single day by individuals from thegroup for which the mixture has been defined. For example,linuron is a herbicide authorised on carrots, herbs and celeriacwhich are the main contributors of mixture 2. This can begeneralised to mixture 4: cyprodinil, fludioxonil, and iprodioneare authorised as fungicides and λ-cyhalothrin as an insecticide onall foods contributing to mixture 4. Mixture 5 contains formerpesticides DDT and dieldrin, known today as persistent organo-chlorine pollutants, for which foods of animal origin such as fishproducts were shown to be the main contributors to totalexposure (Van Audenhaege et al., 2009). Thus, mixtures foundwith this statistical model seem to be consistent with agriculturalpractices and/or contamination of the environment in France.

Table 1Exposure to the mixtures of pesticides expressed in mg/kg body weight/day of the two groups of children with the highest levels of co-exposure obtained with the clustering model.

Pesticide residue Chemical family Percentageper mixture

Exposure of child population (lg/kg body weight/day)Group 1: 699 children Group 2: 238 children

Mean Standarddeviation

50th 97.5th 99th Maxi-mum

Mean Standarddeviation

50th 97.5th 99th Maxi-mum

Mixture 1 + + + + + +

Captan Phtalamide 3% 0.07 0.25 1.00E�7 0.50 0.94 3.71 0.14 0.16 0.08 0.61 0.80 0.95

Diphenylamine Amine 40% 0.86 1.79 1.00E�7 6.38 8.49 17.59 1.31 1.46 0.95 4.87 6.95 10.70

Phosalone Organophosphate 13% 0.28 0.57 3.27E�5 1.90 2.41 5.86 0.48 0.52 0.33 2.11 2.57 3.07

Propargite Sulphite ester 42% 0.90 1.78 1.00E�7 5.67 7.51 14.02 1.58 1.83 1.03 7.29 8.48 10.18

Tolylfluanid Sulphamide 2% 0.05 0.11 1.00E�7 0.35 0.59 1.01 0.11 0.13 0.07 0.46 0.60 0.82

Mixture 2 + + + + + + + + + + + +

Chlorfenvinphos Organophosphate 61% 0.06 0.12 0.01 0.41 0.60 0.86 0.08 0.19 0.01 0.45 0.60 2.02

Ethion Organophosphate 25% 0.03 0.06 4.02E�04 0.20 0.28 0.50 0.04 0.08 1.00E-07 0.25 0.35 0.60

Linuron Urea 14% 0.01 0.03 1.24E�03 0.09 0.15 0.22 0.02 0.04 1.40E-03 0.12 0.21 0.28

Mixture 3

Fenhexamid Hydroxyanilide – 0.36 2.44 1.00E�07 3.54 12.73 34.72 0.98 4.71 0.06 7.08 15.00 58.61

Fenitrothion Organophosphate – 0.00 0.00 1.00E�07 4.79E-3 0.01 0.03 0.05 0.19 1.76E-03 0.54 0.86 1.99

Penconazole Triazole – 0.00 0.00 1.00E�07 1.00E-07 1.00E-07 2.96E-03 0.01 1.00E-07 0.04 0.08 0.11

Pyrimethanil Anilinopyrimidine – 0.02 0.04 4.59E�03 0.12 0.18 0.48 0.27 0.61 0.10 1.67 2.75 5.67

Quinoxyfen Quinoline – 0.00 0.00 1.00E�07 0.00 0.00 0.00 0.01 0.03 1.00E-07 0.09 0.17 0.26

Triadimenol Triazole – 0.02 0.12 1.00E�07 0.21 0.59 1.74 0.05 0.13 2.18E-04 0.53 0.69 0.98

Mixture 4

Cyprodinil Anilinopyrimidine – 0.16 0.36 1.24E�03 1.14 1.75 3.32 0.36 0.75 0.11 2.10 3.06 7.60

Fludioxonil Phenylpyrrole – 0.17 0.47 1.00E�07 1.61 2.00 4.74 0.23 0.41 0.06 1.55 1.90 2.47

Iprodione Dicarboximide – 0.73 1.32 0.20 4.73 6.48 10.95 1.06 1.60 0.45 4.71 8.57 12.74

λ-Cyhalothrin Pyrethroid – 0.02 0.05 0.01 0.17 0.23 0.35 0.04 0.06 0.02 0.19 0.27 0.51

Procymidone Dicarboximide – 0.45 0.79 0.14 2.50 3.36 8.01 1.02 1.68 0.52 6.22 8.34 14.88

Mixture 5 + + + + + +

DDT Organochlorine 95% 1.35E-03 2.98E-03 8.50E�06 0.01 0.01 0.04 2.45E-03 4.83E-03 4.63E-04 0.02 0.02 0.05

Dieldrin Chlorinatedhydrocarbon

5% 5.72E-05 1.84E-04 1.00E�07 5.23E-04 8.49E-04 2.33E-03 1.16E-04 2.71E-04 1.00E-07 9.50E-04 1.26E-03 2.27E-03

A.Crépet

etal./

Environmental

Research

126(2013)

125–133

129

Table 1 (continued )

Pesticide residue Chemical family Percentageper mixture

Exposure of child population (lg/kg body weight/day)Group 1: 699 children Group 2: 238 children

Mean Standarddeviation

50th 97.5th 99th Maxi-mum

Mean Standarddeviation

50th 97.5th 99th Maxi-mum

Mixture 6 + + + + + + + + + + + +

Chlorpropham Carbamate 26% 6.42 9.23 2.34 33.33 40.01 68.13 7.45 12.35 2.06 38.55 60.86 81.45

Maleic hydrazide Pyridazine 74% 17.86 24.07 8.24 79.95 106.17 177.42 20.56 30.23 8.91 90.13 136.01 215.28

Mixture 7 + + + + + +

Imazalil Imidazole 97% 3.30 5.69 0.85 17.29 28.08 61.40 4.58 6.44 2.23 24.38 33.50 38.63

Methidathion Organophosphate 3% 0.08 0.24 3.84E�03 0.58 1.18 2.68 0.14 0.28 0.03 0.96 1.42 1.98

The + sign indicates group(s) from which the mixture has been defined.

Table 2Exposure to the mixtures of pesticides expressed in mg/kg body weight/day of the three groups of adults with the highest levels of co-exposure obtained with the clustering model.

Pesticide residue Percentage permixture

Exposure of adult population (lg/kg body weight/day)Group 1: 256 adults Group 2: 655 adults Group 3: 302 adults

Mean Sd 50th 97.5th 99th Maximum Mean Sd 50th 97.5th 99th Maximum Mean Sd 50th 97.5th 99th Maximum

Mixture 1

Captan – 0.10 0.13 0.05 0.43 0.64 0.82 0.07 0.13 0.02 0.41 0.68 1.06 0.07 0.23 0.01 0.37 0.48 3.57

Diphenylamine – 1.05 1.37 0.59 3.68 6.21 12.56 0.90 1.47 0.21 4.91 6.43 11.73 0.51 1.00 2.37E-03 3.45 4.46 6.41

Phosalone – 0.31 0.37 0.18 1.12 1.46 3.61 0.28 0.39 0.12 1.34 1.72 2.48 0.18 0.32 0.01 1.25 1.53 1.85

Propargite – 1.05 1.32 0.57 4.90 5.93 9.13 0.92 1.35 0.22 4.60 5.46 9.96 0.58 1.03 0.05 3.45 4.94 6.13

Tolylfluanid – 0.07 0.09 0.04 0.32 0.41 0.70 0.05 0.09 0.01 0.30 0.46 0.93 0.04 0.11 3.55E-03 0.31 0.52 1.01

Mixture 2 + + + + + + + + + + + +

Chlorfenvinphos 61% 0.06 0.11 0.02 0.36 0.54 0.85 0.05 0.09 0.01 0.33 0.42 0.80 0.05 0.08 0.01 0.27 0.37 0.42

Ethion 25% 0.03 0.04 0.01 0.14 0.18 0.32 0.02 0.04 0.00 0.15 0.19 0.27 0.02 0.04 1.84E-03 0.12 0.19 0.41

Linuron 14% 0.01 0.03 4.18E-03 0.08 0.12 0.26 0.01 0.02 9.72E-04 0.08 0.09 0.23 0.01 0.02 2.74E-03 0.06 0.09 0.14

Mixture 3 + + + + + +

A.Crépet

etal./

Environmental

Research

126(2013)

125–133

130

Table 2 (continued )

Pesticide residue Percentage permixture

Exposure of adult population (lg/kg body weight/day)Group 1: 256 adults Group 2: 655 adults Group 3: 302 adults

Mean Sd 50th 97.5th 99th Maximum Mean Sd 50th 97.5th 99th Maximum Mean Sd 50th 97.5th 99th Maximum

Fenhexamid 52% 0.39 1.84 0.02 4.83 10.44 17.83 0.23 1.55 0 3.72 8.42 25.27 0.84 1.59 0.12 5.24 8.39 12.03

Fenitrothion 9% 2.00E-03 0.01 1.04E-04 0.01 0.02 0.23 1.07E-03 0.01 0 0.01 0.02 0.20 0.15 0.25 0.02 0.94 1.15 1.39

Penconazole 1% 0 0 0 0 0 0 0 0 0 0 0 0 0.01 0.02 6.66E-04 0.06 0.09 0.15

Pyrimethanil 31% 0.09 0.16 0.04 0.46 0.95 1.56 0.02 0.11 3.38E-03 0.15 0.23 2.18 0.51 0.86 0.11 3.00 4.31 6.09

Quinoxyfen 2% 0 0 0 0 0 0 0 0 0 0 0 0 0.03 0.05 4.08E-03 0.19 0.28 0.37

Triadimenol 6% 0.01 0.03 0 0.11 0.15 0.23 0.01 0.06 0 0.08 0.30 0.81 0.09 0.16 0.02 0.60 0.74 1.03

Mixture 4 + + + + + +

Cyprodinil 16% 0.32 0.46 0.08 1.72 1.99 2.81 0.19 0.35 0.01 1.11 1.37 2.58 0.59 0.76 0.37 3.03 3.90 4.83

Fludioxonil 9% 0.36 0.52 0.09 1.76 2.30 3.19 0.19 0.39 0.00 1.35 1.58 2.92 0.33 0.44 0.12 1.59 1.82 1.94

Iprodione 33% 1.16 1.49 0.61 5.43 5.75 11.58 0.74 1.10 0.23 3.64 5.25 10.14 1.24 1.26 1.00 4.48 5.03 8.10

λ-Cyhalothrin 1% 0.04 0.04 0.03 0.12 0.19 0.31 0.03 0.04 0.01 0.12 0.18 0.31 0.05 0.06 0.04 0.19 0.28 0.34

Procymidone 42% 0.83 1.21 0.37 4.45 5.80 8.11 0.43 0.70 0.14 2.70 3.41 5.39 1.59 2.23 0.81 6.75 9.82 19.88

Mixture 5 + + + + + + + + + + + + + + + + + +

DDT 95% 1.29E-03 1.78E-03 2.89E-04 5.77E-03 7.43E-03 7.97E-03 1.26E-03 2.29E-03 1.16E-04 0.01 0.01 0.03 1.80E-03 3.01E-03 7.47E-04 0.01 0.02 0.02

Dieldrin 5% 7.92E-05 1.80E-04 0 5.45E-04 8.35E-04 1.39E-03 6.05E-05 1.45E-04 0 5.09E-04 6.55E-04 1.25E-03 7.65E-05 1.64E-04 0 5.31E-04 6.43E-04 1.59E-03

Mixture 6 + + + + + + + + + + + +

Chlorpropham 26% 4.77 6.29 2.71 21.94 25.44 41.11 4.28 6.26 1.70 20.34 26.73 48.08 4.22 6.14 1.11 19.08 22.87 41.79

Maleic hydrazide 74% 13.24 15.61 8.61 57.24 63.68 87.98 12.09 15.94 6.14 55.16 65.09 98.53 12.00 16.02 5.96 49.79 61.46 111.28

Mixture 7 + + + + + + + + + + + + + + + + + +

Imazalil 97% 1.76 2.27 0.67 6.92 10.17 14.61 1.73 2.46 0.50 7.87 12.31 18.79 1.58 2.38 0.48 7.59 11.77 13.89

Methidathion 3% 0.05 0.08 0.02 0.30 0.38 0.68 0.05 0.14 0.01 0.37 0.58 2.11 0.04 0.09 0.01 0.32 0.45 0.53

The + sign indicates the sub-population(s) from which the mixture has been defined.Sd: Standard deviation.

A.Crépet

etal./

Environmental

Research

126(2013)

125–133

131

A. Crépet et al. / Environmental Research 126 (2013) 125–133132

Nevertheless, due to worldwide food trade, consumers can beexposed to pesticides which are not used in their country butwhich may be used in others. Thus pesticides such as ethion formixture 2, non-authorised in France in 2006, could also be foundin the mixtures.

These mixtures may rapidly evolve over time, from 1 year toanother in accordance with pest pressure. For example, insectpressure was very low in France in 2006. As a consequence, onlyone authorised insecticide (λ-Cyhalothrin) was found in themixtures (Agreste, 2009). Another cause of changes in mixturecomponents is the change in their authorisation of use. Somepesticides found in the mixtures are no longer authorised inFrance. This is the case for phosalone and tolylfluanid in mixture 1,chlorfenvinphos and fenitrothion in mixture 3, procymidone inmixture 4 and methidathion in mixture 7. Some substances suchas propargite and diphenylamine are currently undergoing a non-approval process. Within the framework of this study, the timewindow chosen for exposure assessment was a day and only 1 yearof the monitoring programme was taken into consideration. Thestability of mixtures for short-term exposure requires furtherinvestigation; the calculation needs to be updated with other yearsof monitoring. This also raises the question of the relevance ofdefining mixtures of pesticide residues for long-term exposure. Themethod developed was also applied to chronic exposure in Crépetand Tressou (2011). However, the algorithm failed to cluster thepopulation. In fact, in the chronic context, co-exposure profiles wereestimated using mean consumption and mean residue levels, whichyield a more homogeneous distribution but are more difficult tocluster. Nevertheless, correlations between the pesticides formingthe mixtures obtained in short-term contexts are also found to behigh in the chronic exposure context.

Some of the identified mixtures contain residues associatedwith pesticides which have similar crop-protection effects. Forexample, this is the case for cyprodinil, fludioxonyl and iprodione,or for chlorpropham and maleic hydrazide. Being simultaneouslyexposed to these compounds does not seem realistic in a short-term exposure framework. Indeed, it is unlikely that a farmerwould apply to crops different pesticides having similar proper-ties. However, because monitoring programmes are performed oncomposite samples, i.e. food samples including several units thatmay come from different farms,.residues of pesticides with similarcrop-protection effects can be found in some of these compositesamples. Moreover, the main reason is probably linked to themethod used to calculate exposure as described in Section 2.3. Infact, exposures to the 79 pesticide residues were determinedindependently one from another, i.e. without taking into accountcorrelations between residue levels within each food sample. Toincorporate these correlations, methods such as that of (Iman andConover, 1982) used to generate correlated random variables inMonte Carlo simulations are available. However, in order toimplement this, it is necessary that all the pesticide residues beanalysed in each food sample, which is not the case for mostof the monitoring data used in this work. One solution, studiedin ACROPOLIS, (2010–2013), would be to perform a pesticideusage survey in addition to monitoring data in order to providethe percentages of crops simultaneously treated with furtherpesticides.

Another consequence of the independent calculation of indivi-dual exposure to the 79 pesticides is the way the individual vectorof co-exposure is built. In this study, an arbitrary choice was madeto retain the 95th percentile of the exposure distribution for eachpesticide. In co-exposure assessment, this calculation choiceseems unrealistic, as the probability of being exposed to highvalues for all pesticides over a day is very low. To account for theuncertainty of pesticide exposure over a day, the series of 100possible daily exposures to a pesticide was used in Crépet and

Tressou (2011). However, due to the high uncertainty of individualexposure, an individual can frequently be exposed to a low level ofone pesticide and to a high level of another one. The clustering ofsuch co-exposure profiles is complex and the correlations betweenpesticide exposures are very low. The inclusion of exposureuncertainty thus makes it somewhat difficult to define mixturesof pesticides and requires further statistical developments.

Variability factors between food units and processing factors ofthe raw agricultural commodities are two parameters not includedin the present co-exposure assessment that could influenceexposure levels. Variability factors can be introduced to integratethe fact that analytical samples in monitoring programmes arecomposite samples made up of many units of a single food (EFSA,2007). Thus, the analysed (average) composite residue level maycontain lower residue levels than some of the individual units ofthe sample, and the exposure assessed without accounting forthese factors may be underestimated. On the other hand, proces-sing (washing, peeling, baking, etc.) raw agricultural commoditiesinto consumed products can reduce residue levels and conse-quently exposure (Boon et al., 2009a, 2009b). However, we do notexpect that including the variability and processing factors wouldsignificantly modify the mixtures.

This model also relies on two major choices which may have animpact on the final results. The first choice concerns theway in which non-detected residues are handled. In the presentstudy, to highlight pesticides observed most often in the diet,concentrations found below the limit of reporting were set at 0. Inprevious studies, it was considered that these samples mightcontain traces of pesticides, and residue level values were ran-domly and uniformly selected between 0 and the correspondinglimit of reporting. However, the high number of censored dataleads to a high dependency between exposure estimates andvalues of the limit of reporting. With this scenario, Crépet andTressou (2011) concluded that clustering was mostly related to theconsumption behaviour of individuals rather than contaminationlevels. With this method, the individuals from the main groupswere simulated as being either highly exposed to all pesticides orslightly exposed to all pesticides. In addition, mixtures definedfrom exposures estimated with this treatment contained a largenumber of substances. For example, 5 mixtures out of the 20defined for the adult population were composed of more than 10substances. The other choice involves the correlation levelretained to define the mixtures (0.7 in the present article). Thisvalue was chosen to balance the number of mixtures with theirnumber of components. The use of a lower or higher correlationlevel will impact the number of mixtures and the number of theircomponents: the higher the correlation level, the lower thenumber of mixtures and the number of their components.

Within the framework of the PERICLES research programme,mixtures defined by the clustering model presented in this articlehave been studied for their short-term effects on human cells.Unspecific responses including cytotoxicity (Takakura et al., inpress), apoptosis and oxidative stress, endocrine disruptor effects(analysis of PXR transactivation and target gene regulation) andgenotoxic effects (Comet assay, H2AX phosphorylation, Graillotet al. (2012)) have been tested on human intestinal, hepatic, renal,colon and nerve cells. Effects of single pesticide residues of themixtures were also studied to compare the mixture effect with theones estimated from concentration addition or independent actionmodels.

5. Conclusion

This study applies a statistical Bayesian non-parametricapproach to characterise mixtures of pesticide residues to which

A. Crépet et al. / Environmental Research 126 (2013) 125–133 133

the French population is simultaneously highly exposed. We canextend this generic approach to all types of chemicals found infood. Applied to 79 pesticide residues, it results in 7 mixturescomposed of 2–6 pesticide residues each. In order to consolidateand validate this global approach, further statistical developmentsand more effective collection of data on pesticide residue levels infood are needed. Nevertheless, these initial results show that themain pesticide residues to which the French population is exposeddo not necessarily belong to the same chemical family and do notappear to share the same modes of action.

Funding sources

This work is part of the framework of the 2009–2011 PERICLESno. 2008-CESA 01601 research programme, granted by the FrenchNational Research Agency (Agence Nationale de la Recherche) andthe former French Agency for Environmental and OccupationalHealth Safety (Agence française de sécurité sanitaire de l'environne-ment et du travail).

Acknowledgements

The residue data are provided by the French ministries incharge of agriculture, consumer affairs and health. The authorswish to thank Rémy Maximilien and all the participants in thePERICLES research programme for their valuable input.

Appendix A. Supplementary materials

Supplementary materials associated with this article can befound in the online version at http://dx.doi.org/10.1016/j.envres.2013.03.008.

References

ACROPOLIS, EU-project coordinated by the National Institute for Public Health andthe Environment (RIVM). Aggregate and Cumulative Risk of Pesticides: an on-line integrated Strategy, 2010–2013, ⟨http://www.acropolis-eu.com/⟩.

Agreste, Chiffres et données. Pratiques culturales en viticulture. Série Agricultureno. 204. 2009.

ANSES, 2009a. (Agence française de sécurité sanitaire, alimentation, environne-ment, tavail). Appui scientifique et technique de l'ANSES relatif à la stratégie desurveillance des résidus de pesticides, Saisine n 2009-SA-171.

ANSES, (Agence française de sécurité sanitaire, alimentation, environnement,tavail). INCA 2 (2006–2007). Report of the Individual and the National Studyon Food Consumption. 2009b, pp. 199.

ANSES, 2010a. (Agence française de sécurité sanitaire, alimentation, environne-ment, tavail). Avis de l'Agence nationale de sécurité sanitaire de l'alimentation,de l'environnement et du travail, relatif au programme 2011 de surveillance desrésidus de pesticides dans les aliments, Saisine n 2010-SA-0110, pp. 22.

Boobis, A.R., Ossendorp, B.C., Banasiak, U., Hamey, P.Y., Sebestyen, I., Moretto, A.,2008. Cumulative risk assessment of pesticide residues in food. Toxicol. Lett. 15,137–150.

Boon, P.E., Ruprich, J., Petersen, A., Moussavian, S., Debegnach, F., van Klaveren, J.D.,2009a. Harmonisation of food consumption data format for dietary exposureassessments of chemicals analysed in raw agricultural commodities. FoodChem. Toxicol. 47, 2883–2889.

Boon, P.E., Svensson, K., Moussavian, S., van der Voet, H., Petersen, A., uprich, J.,Debegnach, F., de Boer, W.J., van Donkersgoed, G., Brera, C., van Klaveren, J.D.,Busk, L., 2009b. Probabilistic acute dietary exposure assessments to captan andtolylfluanid using several European food consumption and pesticide concen-tration databases. Food Chem. Toxicol. 47, 2890–2898.

Bosgra, S., Voet, H.v.d., Boon, P.E., Slob, W., 2009. An integrated probabilisticframework for cumulative risk assessment of common mechanism chemicalsin food: an example with organophosphorus pesticides. Regul. Toxicol. Phar-macol. 54, 124–133.

Crépet, A., Tressou, J., 2011. Bayesian nonparametric model for clustering individualco-exposure to pesticides found in the French diet. Bayesian Anal. 6, 127–144.

Dubuisson, C., Lioret, S., Touvier, M., Dufour, A., Calamassi-Tran, G., Volatier, J.L.,Lafay, L., 2010. Trends in food and nutritional intakes of French adults from1999 to 2007: results from the INCA surveys. Br. J. Nutr. 103, 1035–1048.

EFSA, (European Food Safety Agency), 2007. Opinion of the Scientific Panel on Plantprotection products and their Residues on a request from the Commission onacute dietary intake assessment of pesticide residues in fruit and vegetables.EFSA J. 538, pp.1–88. ⟨http://www.efsa⟩.

EFSA, (European Food Safety Agency), 2008. Opinion of the Scientific Panel on Plantprotection products and their Residues to evaluate the suitability of existingmethodologies and, if appropriate, the identification of new approaches toassess cumulative and synergistic risks from pesticides to human health with aview to set MRLs for those pesticides in the frame of Regulation (EC) 396/20051. vol. 704. pp. 1–84.

EFSA, (European Food Safety Agency), 2009. Scientific Opinion on Risk Assessmentfor a Selected Group of Pesticides from the Triazole Group to Test PossibleMethodologies to Assess Cumulative Effects from Exposure through Food fromthese Pesticides on Human Health.Technical Report 7 (9). http://dx.doi.org/10.2903/j.efsa.2009.1167 (187 pp.). Available from: ⟨www.efsa.europa.eu⟩.

EFSA, (European Food Safety Agency), 2010a. Annual Report on Pesticide Residuesaccording to Article 32 of Regulation (EC) no. 396/2005. http://dx.doi.org/0.2903/j.efsa.2010.1646. EFSA Journal 8, 1646. (442 pp.). Available from: ⟨www.efsa.europa.eu⟩.

EFSA, (European Food Safety Agency), 2010b. Management of left-censored data indietary exposure assessment of chemical substances. Technical Report 8 (3).http://dx.doi.org/10.2903/j.efsa.2010.1557. (96 pp.). Available from: ⟨www.efsa.europa.eu⟩.

EFSA, (European Food Safety Agency), 2012. Scientific Report on Investigation of thestate of the science on combined actions of chemicals in food throughdissimilar modes of action and proposal for science-based approach forperforming related cumulative risk assessment. Supporting Publications2012:EN-232. Available from: ⟨www.efsa.europa.eu⟩.

EPA, (Environmental Protection Agency), 2002. Revised Organophosphorus Pesti-cide Cumulative Risk Assessment. Office of Pesticide Programs, US Environ-mental Protection Agency, Washington, DC.

EPA, (Environmental Protection Agency), 2007. Revised N-Methyl CarbamateCumulative Risk Assessment. Office of Pesticide Programs, US EnvironmentalProtection Agency, Washington, DC.

European Commission, 2011. Opinion on Toxicity and Assessment of ChemicalMixtures. Available from: ⟨http://ec.europa.eu/health/scientific_committees/environmental_risks/docs/scher_o_150.pdf⟩.

Ferguson, T.S., 1973. A Bayesian analysis of some nonparametric problems. Ann.Stat. 1, 209–230.

Graillot, V., Takakura, N., Hegarat, L.L., Fessard, V., Audebert, M., Cravedi, J.P., 2012.Genotoxicity of pesticide mixtures present in the diet of the French population.Environ. Mol. Mutagen. 53, 173–184.

Iman, R.L., Conover, W.J., 1982. A distribution-free approach to inducing rankcorrelation among input variables. Commun. Stat. 11, 311–334.

Lioret, S., Dubuisson, C., Dufour, A., Touvier, M., Calamassi-Tran, G., Maire, B.,Volatier, J.L., Lafay, L., 2010. Trends in food intake in French children from 1999to 2007: results from the INCA (étude Individuelle Nationale des Consomma-tions Alimentaires) dietary surveys. Br. J. Nutr. 103, 585–601.

Lo, A.Y., 1984. On a class of Bayesian nonparametric estimates: I. Density estimates.Ann. Stat. 12, 351–357.

Meek, M.E.B., Boobis, A.R., Crofton, K.M., Heinemeyer, G., Van Raaij, M., Vickers, C.,2011. Risk assessment of combined exposure to multiple chemicals: a WHO/IPCS framework. Regulatory Toxicology and Pharmacology 60, S1–S14.

Ministère de l'agriculture et de la pêche, (Ministry of agriculture and fisheries) E-phy. Le catalogue des produits phytopharmaceutiques et de leurs usages desmatières fertilisantes et des supports de culture homologués en France.Available from : ⟨http://e-phy.agriculture.gouv.fr/⟩. Last consultation 22 February2011.

Müller, A.K., Bosgra, S., Boon, P.E., Voet, H.v.d., Nielsen, E., Ladefoged, O., 2009.Probabilistic cumulative risk assessment of anti-androgenic pesticides in food.Food Chem. Toxicol. 47, 2951–2962.

Reffstrup, T.K., Larsen, J.C., Meyer, O., 2010. Risk assessment of mixtures ofpesticides. Current approaches and future strategies. Regul. Toxicol. Pharmacol.56, 174–192.

Takakura, N., Sanders, P., Fessard, V., Le Hegarat, L., 2013. In vitro combinedcytotoxic effects of pesticide cocktails simultaneously found in the French diet.Food and Chemical Toxicology 52, 153–162.

Van Audenhaege, M., Heraud, F., Menard, C., Bouyrie, J., Morois, S., Calamassi-Tran,G., Lesterle, S., Volatier, J.L., Leblanc, J.C., 2009. Impact of food consumptionhabits on the pesticide dietary intake: Comparison between a French vegetar-ian and the general population. Food Additives and Contaminants. Part A Chem.Anal. Control Exposure Risk Assessment 26, 1372–1388.

WHO, (World Health Organization), 2009. Assessment of combined exposures tomultiple chemicals: report of a WHO/IPCS international workshop on aggre-gate/cumulative risk assessment.