Due to the large number of chemicals found in the environment, individuals are daily exposed to complex mixtures of chemicals which can interact and cause health problems. One of the main/major challenges facing risk assessment is the risk related to mixtures, a particularly difficult task due to the multitude of possible combinations of chemicals for which it is unrealistic to test combined toxicological effects. For this reason, risk is usually assessed for chemicals belonging to the same chemical family (such as dioxins and furans (Van den Berg et al., 2006) or triazoles (EFSA, 2009b)), that share a similar mode of action (Bosgra et al., 2009, Kortenkamp and Faust, 2010). It may also be assessed by grouping substances by their organ toxicity or specific effects (EFSA, 2013; RIVM et al., 2016). Recently, Orton et al. (2014) proposed an approach based on a combination of 30 androgen receptor antagonists composed of 13 pesticides and 17 non pesticides. The mixture analysed in Orton’s study contained substances with similar mechanisms of action and covered a wide range of sources and exposure routes, but the co-occurrence of their exposure was not considered. Thus, one important question remains: do the mixtures defined on the basis of toxicological properties reflect the reality of exposure? The present work proposes to identify chemical mixtures from individuals’ exposure to different foods.
Dietary exposure is commonly assessed by combining data on the quantity of food consumed with chemical concentrations in the food. Total diet studies (TDS) aim to provide data on concentrations in food consumed by the general population for a wide range of substances as well as the corresponding exposure. The second French TDS (hereafter referred to simply as TDS 2) (Sirot et al., 2009) investigated around 440 chemicals and used consumption data provided by INCA 2, the second French national food consumption survey (Dubuisson et al., 2010, Lioret et al., 2010). One hundred and fifty-three substances were detected in at least one food, showing the multitude and diversity of substances to which the French population is exposed through its diet. As it is not realistic to test all the possible combinations of these substances for their combined effects, it is important to extract the main mixtures to be evaluated as a priority by toxicological and epidemiological studies.
Crépet and co-authors (Crépet and Tressou, 2011, Crépet et al., 2013a, Crépet et al., 2013b) proposed a statistical method based on a Bayesian non-parametric model to determine major mixtures from dietary exposures. This method is used to classify the population regarding their exposure profiles and then study correlations between pesticide exposures to define mixtures. A second approach presented in Béchaux et al. (2013) and based on non-negative matrix factorisation (NMF) (Lee and Seung, 2001), consists in reducing the size of the dataset before classification. This work was conducted on 26 priority pesticides from TDS 2. The peculiarity of this approach was to characterise not only the mixtures but also the main foods contributing to exposure to these mixtures. Indeed, the NMF method has previously been used to define dietary patterns and clusters of individual diets by Zetlaoui et al., 2011, Sy et al., 2013 and, more recently, Gazan et al. (2016).
The objective of this work was to apply the approach based on NMF (Béchaux et al., 2013) to the 153 substances detected in TDS 2 (some of which contained pesticides and others not) in order to identify chemical mixtures. The NMF analysis was completed by hierarchical clustering to classify groups with similar dietary behaviour and a similar co-exposure profile. The defined groups were characterised by socio-demographic data (age, body mass index and the monthly household income) and levels of exposure to the substances in the mixtures. Thus, this analysis of the 153 substances led to the definition of the major chemical mixtures to which the French population is exposed and their main food vectors.
Dining and Cooking