Use of advanced cluster analysis to characterize fish consumption patterns and methylmercury dietary exposures from fish and other sea foods among pregnant women


On account of the interspecies variability in contamination and nutrient contents, consumers must balance the risks and benefits of fish consumption through their choice of species, meal size and frequency. The objectives of this study were to better characterize the risk of methylmercury (MeHg) exposure in a sample of 161 French pregnant women consuming sea food, including fish, molluscs and crustaceans, and to explore the use of unsupervised statistical learning as an advanced type of cluster analysis to identify patterns of fish consumption that could predict exposure to MeHg and the coverage of the Recommended Daily Allowance for n-3 polyunsaturated fatty acid (PUFA). The proportion of about 5% of pregnant women exposed at levels higher than the tolerable weekly intake for MeHg is similar to that observed among women of childbearing age in earlier French studies. At the same time, only about 50% of the women reached the recommended intake of 500 mg/day n-3 PUFA. Cluster analysis of the fish consumption showed that they could be grouped in five major clusters that are largely predictable of the intake of both MeHg and n-3 PUFA. This study shows that a global increase in seafood consumption could lead to MeHg exposure above the toxicological limits for pregnant women, thereby questioning the overall balance between this potential risk and potential beneficial effects of n-3 PUFA intakes. Only pregnant women consuming a high proportion of fatty fish meet the n-3 PUFA intake requirements without exceeding the toxicological limit for MeHg. The clusters identified suggest that different intervention strategies may be needed to address the dual purpose of ensuring high PUFA intakes at acceptable MeHg exposures.


Fish, molluscs and crustaceans provide considerable amounts of easily digestible protein of high biological value and, especially in the case of marine species, can constitute a good source of iodine, selenium and vitamins A and D (EFSA, 2005). They are beneficial to the development of cognition during infancy and the maintenance of cardiovascular health in the whole population because they contain long-chain polyunsaturated fatty acids of the n-3 variety (n-3 PUFA). These nutrients are crucial to fetal development; in particular, eicosapentaenoic acid (EPA), docosapentaenoic acid (DPA) and docosahexaenoic acid (DHA) appear important to ensuring optimum development of the central nervous system (Alessandri et al., 2004). The impact of fish consumption (or LC n-3 PUFA intake) on the length of gestation and birth weight was the purpose of several studies (Olsen et al., 1991, 1995; Helland et al., 2001; Smuts et al., 2003), providing different results as a function of the population studied. However, the occurrence of preterm delivery differed significantly across four groups of seafood intake, decreasing significantly when fish consumption is increasing up to at least once a week (Olsen and Secher, 2002). Oily fish consumption by the mother during pregnancy was also associated with an increased likelihood of the child having foveal stereoacuity. This may indicate that the availability of DHA to the fetus has consequences on the development of visual function (Williams et al., 2001). Finally, supplementation with fish oil during pregnancy in atopic women might be of influence on the infant's immune system and the development of subsequent allergic disease (Dunstan et al., 2004; Denburg et al., 2005). There is no unanimously Recommended Daily Allowance (RDA) for n-3 PUFAs, which are generally related to the total caloric intake: several countries have established various recommendations for the general population. In France, the recommendation for LC n-3 PUFA for the general population is 0.2% of the energy intake or 0.5 g/day (Martin, 2001). The UK Scientific Advisory Committee on Nutrition (FSA, 2004) recommends that the intake of LC n-3 PUFA should be 0.45 g/day. In the United States, a working group at the US National Institutes of Health has recommended that with an energy intake of 2000 kcal/day, 0.65 g/day of DHA plus EPA should be consumed or 0.3% of the energy (Simopoulos et al., 1999).

At the international level, the International Society for the Study of Fatty Acids and Lipids (ISSFAL, 2004) has recommended a dietary intake for EPA and DHA of 0.5 g/day for cardiovascular health. Population nutritional goals were set at an intake of 1–2% of energy as n-3 PUFA (WHO, 2003). For the purpose of this paper, an LC n-3 PUFA intake of 0.5 g/day was chosen as a limit value.

At the same time, seafood consumption is also the primary dietary source of human exposure to methylmercury (MeHg), a well-known human neurotoxicant absorbed almost exclusively from eating fish and other seafood products. Of the organic mercury compounds, MeHg is the most toxic form and is highly absorbed in humans (>95% of ingested dose) (Aberg et al., 1969). It passes easily across the placenta to the fetus and is retained with a long half-life of about 6 weeks in the nervous system (NRC, 2000; UNEP, 2002).

Epidemiological studies of populations with high fish and other seafood consumption have reported that maternal MeHg exposure during pregnancy adversely affects the developing nervous system of the fetus (McKeown-Eyssen et al., 1983; Kjellström, 1991; Lebel et al., 1996a, 1998b; Grandjean et al., 1999; Murata et al., 1999a, 1999b; Steuerwald et al., 2000; Cordier et al., 2002; Stewert et al., 2003). Neuropsychological tests have shown that children exposed to MeHg during prenatal development perform less well on several neuropsychological tests, including those representing concentration, fine motor speed and verbal memory, than non-exposed children (Grandjean et al., 1997; NRC, 2000; UNEP, 2002). Another epidemiological study performed in the Seychelles has found no consistent neurodevelopmental or neuropathological impairments (Myers et al., 2003), and enhanced child development was correlated with increasing maternal hair MeHg levels for some end points. From the same cohort, the authors recently examined the possible associations of maternal PUFAs, MeHg and infant development (Strain et al., 2008) and found a positive association between the psychomotor developmental index and the total n-3 PUFAs in children at 9 months . This association was stronger when adjusted for prenatal MeHg exposure. Even if all the results from this study are not fully consistent with this hypothesis, the various studies altogether indicate that confounding of adverse effects by nutrients or, conversely, confounding of beneficial effects of PUFAs by MeHg should be considered when evaluating data from observational studies (Oken et al., 2005; Budtz-Jørgensen et al., 2007; Strain et al., 2008).

Limit values for intake of MeHg have recently been reviewed (WHO, 1990; NRC, 2000; US EPA, 2001; AFSSA, 2004; WHO, 2004), and a new provisional tolerable weekly intake (PTWI) of 1.6 μg/kg body weight (bw)/week was proposed by the Joint Food and Agriculture Organisation/World Health Organisation Expert Committee on Food Additives and Contaminants (JECFA) (WHO, 2004, 2007). The United States Environmental Protection Agency (US EPA), using slightly different calculations from National Research Council, established an intake limit of 0.7 μg/kg per week (NRC, 2000; US EPA, 2001).

The concentrations of LC n-3 PUFA and MeHg present in different sea food species vary considerably (Mahaffey, 2004). For lean fish containing 1–4% (1–4 g for 100 g of fish) of fat, the concentration of LC n-3 PUFA is about 0.25–1 g per 100 g of fish. Fatty fish may contain more than 20% (20 g for 100 g of fish) of fat and the concentration of LC n-3 PUFA is up to around 5 g LC n-3 PUFA per 100 g of fish (FAO, 2003). As for MeHg, major predatory species, such as shark, swordfish or bluefin tuna, can be contaminated to a degree higher than the 1 mg/kg limit proposed by the Codex Alimentarius, whereas small non-predatory fish, such as herring or sardine, contains MeHg at concentrations that are one-tenth of that level or lower.

In earlier publications (Crépet et al., 2005; Verger et al., 2007), and by combining data on fish consumption and contamination, it was shown that about 3–5% of French women of childbearing age consuming fish were likely to exceed the PTWI for MeHg. In view of the variability described above regarding both food contamination and nutrient content, the choice of fish species by consumers is crucial to the balance between the risks and benefits of consuming fish (Mahaffey, 2004). The objectives of this paper were thus, first, to determine whether the probability of exceeding the PTWI for MeHg was similar in women during pregnancy as in women of childbearing age. Second, we wanted to compare modeled dietary exposure with both measurements of the hair mercury concentration of subjects and with the Benchmark Dose Level obtained from major epidemiological studies on MeHg. Finally, to contribute to the risk/benefit analysis of eating fish, we applied unsupervised statistical learning to identify different clusters of fish consumption behaviors that may be predictable of exposure to MeHg and/or coverage of the RDA for LC n-3 PUFA. The unsupervised learning task is based on a cluster analysis aimed at dividing a data set into subgroups such that those in each particular subgroup are more “similar” than those found in other subgroups. In our case, this approach results in clusters for which the consumption frequencies of individuals are simultaneously close regarding all types of fish. In this study, the consumption of fish, molluscs and crustaceans was considered but preliminary tests show that the contribution of molluscs and crustaceans to MeHg exposure and PUFA intake in the sample of French pregnant women was not significant enough to justify a separated analysis. Molluscs and crustaceans are included under the broad category “other fish species”.

Materials and methods


Between December 2005 and August 2006, the consumption of fish and other seafoods was assessed in pregnant French women attending the Nantes University Hospital, Saint-Nazaire Hospital and the Saint-Herblain Polyclinic. Nantes, Saint-Nazaire and Saint-Herblain are towns in the coastal region of Loire Atlantique where the frequency of fish consumption is higher than in other French regions distant from the sea (Credoc, 1996; Ofimer, 2005).

Participants were recruited at 12 weeks of pregnancy (gestational age was determined from the first day of the last menstrual period) at the time of their initial visit for an ultrasound examination. Those eligible needed to be capable of completing study questionnaires in French, were not planning to move out of the study area before the end of their pregnancy, were not younger than 18 years of age and had hair that was 3 cm long or longer. Criteria for exclusion from the study were African hairFootnote 1, permed or dyed hair, no fish or seafood consumption, multiple pregnancies, pathological conditions (human immunodeficiency virus, hepatitis or metabolic diseases), hormone therapy and the consumption of fish oil preparations. Of the 355 women approached, 326 agreed to take part in the study (92%) and 29 women refused to participate (mainly because of a lack of interest). Among those who agreed, 165 were ineligible (51% of those who agreed), because of dyed or permed hair (12%), pathological conditions (8%), African hair (3%), no fish and seafood consumption (3%), multiple pregnancies (2%) and other reasons (23%). Of the 161 eligible participants, 137 women (85%) agreed to return for a second visit (at 32 weeks of pregnancy), satisfactorily completed food frequency questionnaires (FFQs) and provided hair samples. At the time of enrollment, 86% of women were between 25 and 30 years of age (mean=30±4, range: 19–43). The mean body weights of women at 12 and 32 weeks of pregnancy were 59.7±11.2 and 72.6±12.1 kg, respectively.

Experimental Protocol

A meeting was held at each recruitment center with the head of the Maternity Unit and medical staff members (Figure 1). Practical aspects of the study were explained by one member of the research team. At enrollment, each participant was invited to read an information letter explaining the study, and then those who were eligible were asked to sign the informed consent form. Each participant completed the initial FFQsFootnote 2. At that time (T0), a hair sample was collected by gathering strands of hair and cutting them from the occipital area of the scalp (approximately the diameter of a matchstick) (Grandjean et al., 2002; Weihe et al., 2005). Hair samples were stored in a labeled paper envelope at room temperature. Before the end of the interview, the women received a stamped envelope containing a questionnaire for the collection of sociodemographic data, and information on their awareness to advisories, risk perceptions concerning food consumption and dietary habits.

Figure 1

Description of the experimental protocol.

During the second visit, performed at 32 weeks of gestation, participants completed a second FFQ. A hair sample was collected using the same methods as those described above.

During the study, participants were interviewed face to face by the same trained researcher. Questionnaires and hair samples were collected, coded and packaged by the same researcher. All procedures were carried out in compliance with the ethical standards for human experimentation established by the Declaration of HelsinkiFootnote 3.

Fish Consumption

Fish consumption was assessed using a detailed FFQ that had been used earlier and calibrated (CALIPSO, 2006). The questionnaire contained 48 food items included in five sections and related to the intake of molluscs and crustaceans: 34 species of fish fresh or frozen, 4 canned fish (anchovies, mackerel, sardines or pilchards and tuna), 4 smoked fish (haddock, herring, mackerel and salmon) and 5 seafood products (fish cakes, surimi, fish soup, paella and fish eggs).

Response options on the questionnaire ranged from never to once a day (never, less than once a month, once a month, two to three times a month, once a week, two to three times a week, four to six times a week and daily). Under these conditions, this type of questionnaire is deemed to be reproducible (Willett et al., 1985; Feunekes et al., 1995). Portion sizes of fresh fish and seafood products were identified using a catalog of photos (SUVIMAX, 1994). Fish consumption was normalized to a weekly consumption for each of the items; for example, we coded “2–3 times a month” as 0.625 times a week.

Estimated Intake of n-3 PUFA

The amount of each fish species eaten by each subject was also combined with the corresponding average level of n-3 PUFA (Table 1). The concentration of n-3 PUFA in fish species on the French market was estimated from the results of analyses of the total fat content in fish performed since 2004 by the French Ministry of Agriculture and Fisheries (MAAPAR, 2004). An average amount for n-3 PUFA corresponding to 25% of total fats was chosen from the available data (FAO, 2006). When the information was not available in the French database for a particular fish species, a default value was taken from the literature (Mahaffey, 2004; EFSA, 2005; Dewailly et al., 2007).

Table 1 Average concentrations in MeHg and PUFA n-3 for various fish species considered in our study.

Assessment of MeHg Exposure

Dietary Exposure Modeling

The amount of each fish species eaten by each subject was combined with the corresponding mean MeHg content (Table 1). Figures on MeHg concentrations originated from the analyses conducted by the Ministry of Agriculture, Food, Fisheries and Rural Affairs in France between 1998 and 2003 (Verger et al., 2007). To ensure comparability with earlier publications (Crépet et al., 2005; Verger et al., 2007), analytical results expressed as total mercury were converted into MeHg using published conversion factors (Cossa et al., 1989; Thibaud and Noel, 1989; Claisse et al., 2001).

Mercury in Hair

Mercury analyses were performed using atomic-absorption spectrometry (Pineau et al., 1990; Grandjean et al., 2002). After microwave digestion of an accurately weighed hair sample representing a 2-cm segment close to the scalp, the digested sample was further prepared and analyzed in duplicate. Mercury analyses were performed by flow-injection cold-vapor atomic absorption spectrometry using Perkin-Elmer apparatus (FIMS-400; Perkin-Elmer, Norwalk, CT, USA). The mercury results were read against a standard curve prepared from a mercury stock treated in the same way as digested samples. The limit of detection for the dissolved sample was estimated to be 0.42 μg/l, that is three times the standard deviation of the blank. Total analytical imprecision was estimated to be 2.0% and 3.8% at mercury concentrations of 4.7 and 11.8 μg/g, respectively. The accuracy of mercury determinations in human hair was ensured using the certified reference material CRM 397 (BCR, Brussels, Belgium) as a quality control; mercury concentrations averaged 11.80 μg/g compared with the assigned value of 11.93±0.77 μg/g.

Statistical Analysis

The first analysis consisted of comparing the results concerning individual variables (fish consumption, PUFA intake and MeHg dietary exposure) between 12 and 32 weeks of pregnancy. The results are presented as means, median and 95th percentile. A paired t-test was applied to means and a non-parametric test (Wilcoxon test) to enable comparisons between groups. The level of statistical significance was set at 5%.

Unsupervised Statistical Learning

An important goal of this paper was to better characterize the various behaviors regarding fish consumption, to identify homogeneous subgroups likely to be similarly exposed to both beneficial and hazardous chemicals from fish. This so-called unsupervised learning task was based on a cluster analysis aimed at partitioning a data set into subgroups such that those in each particular subgroup are more “similar” than those found in other subgroups. Numerous methods have been proposed and studied in the statistical literature to identify such clusters (Gordon, 1999, for a review), most of them relying on a specific distance to measure dissimilarity between pairs of objects. In the present case, the objects examined were the vectors indicating fish consumption frequencies. Two fish consumers were assumed to be “similar” if the corresponding frequency vectors were close according to Euclidian distance. This approach resulted in clusters of consumers with a simultaneously small dispersion regarding all types of fish (i.e., the consumption frequencies of individuals within each subgroup were simultaneously close regarding all types of fish).

From several candidate clustering algorithms that could minimize distance-based dispersion measurements, we have compared two classical methods (mobile centroids and ascending hierarchical classification (AHC)) with an artificial neural networks approach (Kohonen's self-organizing maps). This last one appeared to provide the strongest results. We calculated the statistics, “Wilks’ Lambda” and “Hotelling–Lawley Trace” (multidimensional test), to compare the performance of these three procedures of classification (see Table 6). We now briefly describe Kohonen's neural algorithm approach (Kohonen, 2001).

Kohonen's Self-Organizing Maps

Self-organizing maps (SOMs) belong to the family of artificial neural networks and produce low-dimensional representations (two-dimensional in our case) of (possibly very high dimensional) data sets. They are very convenient from the interpretation perspective in the sense that they preserve the topological properties of the input space (here, the space of all possible values of the vector of fish consumption frequencies): “adjacent classes” of SOMs contain objects that are close in the input space. It is then easy to see which subgroups should be eventually merged so as to obtain a parsimonious clustering of objects. An SOM is obtained by training a standard neural network algorithm on the data set. In the present case, computations were performed by implementing the SAS routines developed by Patrick Letrémy (see As a result of the Kohonen classification, macroclasses are statistically different using a multidimensional test (Wilks’ Lambda and Hotelling–Lawley Trace). The best classification corresponds to the weakest value of the Wilks statistic and the strongest value of Hotelling statistic. On the data of fish consumption at 12 weeks of pregnancy, we have chosen the classification obtained by the SOM algorithm because the AHC contains a singleton (see Table 7).

After clustering for fish consumption, the objective was to classify subjects according to three parameters: their modeled dietary exposure to MeHg, their concentration of MeHg in hair and their intake of PUFAs. For dietary exposure to MeHg, subjects were classified as a function of their exposure, below the US EPA RfD (0.7 μg/kg bw/week), above the PTWI values established by JECFA (1.6 μg/kg bw/week) or between these two values. For mercury in hair, subjects were classified arbitrarily as being below or above the value corresponding to the 75th percentile of the distribution of this concentration (0.93 μg/g hair). Similarly, for n-3 PUFAs, the subjects were classified as being below or above the value corresponding to the median of the distribution (0.43 g/day — close to the RDA of 0.5 g/day).

Finally, we compared the macroclasses resulting from clustering at 12 and 32 weeks. The differences between macroclasses at 12 and 32 weeks of pregnancy were estimated using a multidimensional test (Hotelling test). The level of statistical significant was set at 5%.


Fish and Seafood Consumption

The most frequently consumed fish species were cod, Alaska hake, tuna (canned) and salmon, and the distribution of their consumption is described in Table 2. The results were similar at 12 and 32 weeks of pregnancy. The median consumption of each of the other 32 fish species was equal to zero and the mean consumption for each of them was less than 20 g/week. The consumption of fish was similar between 12 and 32 weeks of pregnancy. For “fish only,” the t-test and the Wilcoxon tests failed to exhibit a statistically significant difference (P=0.56 and P=0.87, respectively). Similar results were observed for “fish and other sea food” (P=0.57 and P=0.99, respectively).

Table 2 Distribution of consumption regarding the four most widely consumed fish species.

Dietary Exposure to MeHg, Mercury in Hair and n-3 PUFA Intake

The mean dietary exposure to MeHg, estimated from fish consumption and MeHg concentration, was not significantly different between 12 and 32 weeks of pregnancy with 0.56 μg/kg bw/week and 0.67 μg/kg bw/week (P=0.22), respectively. The comparison of medians for the same parameter exhibits a statistically significant difference with 0.35 and 0.45 μg/kg bw/week at 12 and 32 weeks of pregnancy (P=0.01), respectively.

At the 95th percentile of the distribution curve, dietary exposure was estimated to be 1.79 and 1.66 μg/kg bw/week at 12 and 32 weeks of pregnancy, respectively (Table 3). The dietary exposure to MeHg from seafood other than fish represents 1.6% of the total.

Table 3 Comparison between modeled dietary exposure (based on fish consumption and mean mercury content in fish) and mercury concentrations in hair.

At 12 weeks of pregnancy, hair mercury concentrations ranged from 0.17 to 3.66 μg/g, with a mean of 0.82 μg/g (n=161). At 32 weeks of pregnancy, hair mercury levels ranged from 0.13 to 2.88 μg/g, with a mean of 0.81 μg/g (n=137). Both the mean and median are similar (P=0.85 and P=0.93, respectively, for t-test and Wilcoxon test).

The distribution of n-3 PUFA ingested by our group of subjects is described in Table 4. In terms of the commonly recommended intake of 0.5 g, about half of the women reached this level and about 20% of them were ingesting more than 1 g n-3 PUFA per day. The comparison of PUFA intakes at 12 and 32 weeks of pregnancy shows that both the mean and the median are similar (P=0.80 and P=0.24, respectively, for t-test and Wilcoxon test).

Table 4 Distribution of dietary intake of n-3 PUFA (based on fish consumption and mean PUFA content in fish) in women at 12 and 32 weeks of pregnancy.

Clustering of Fish Consumer Behavior

As shown in particular by the principal component analysis performed initially on consumption frequencies data, our training data sets exhibited highly non-linear features (see Figure 2) and we need to extract 16 factors to be able to explain approximately 70% of the total variance (see Table 5). It was, therefore, unlikely that these data would be grouped in a few homogeneous clusters without any preliminary transformation of the observed variables: all partitioning methods used on (non-reprocessed) data yielded unreliable numerical results. To remedy this, we chose to reduce the dimension of the data by grouping together in a single variable the consumption of both predatory fish (tuna, dogfish, shark, swordfish, grenadier, ling, marlin and grouper) and fatty fish (herring, mackerel, sardine, salmon and trout). We then added the consumption variables sequentially and one-by-one, in the decreasing order of the average consumption level (i.e., starting with the most widely consumed fish species), stopping when intraclass variance was seen to increase without any additional, significant decrease in global dispersion. In this way, we obtained a clustering based on four variables corresponding to the consumption of high-fat fish, cod, large predatory fish and Alaska hake (sorted by order of importance with respect to discriminatory power, measured in terms of Fisher P-values in the ANOVA). Other fish species are not useful variables to reduce the global dispersion of individual behaviors and to classify the subjects into clusters. However, the consumption of all fish and other seafoods is taken into consideration for all the exposure calculations.

Figure 2

Projection of individuals on the first factorial plane (Fl and F2 being identified by ‘axis 1’ and ‘axis 2’, respectively) of the principal component analysis performed on consumption frequencies at 12 weeks of pregnancy for all of the 48 sea food categories. Graphical display of individuals, no clear clustering appears.

Table 5 Results of principal component analysis on fish consumption.
Table 6 Results of multidimensional test comparing mobile centroids, ascending hierarchical classification (AHC) and Kohonen's self-organizing maps.

SOM Macroclasses at 12 Weeks

When applying the SOM method to two data sets (at 12 and 32 weeks of pregnancy), we used a grid containing 64 cells/units. The choice of the number of P units was arbitrary (commonly P<100 units). The units of the Kohonen map obtained were merged into five macroclasses (represented by colored group of boxes) on the grounds of parsimony and in accordance with the method suggested by Cottrell et al. (1999) using a hierarchical classification with Ward distance. Each of the five macroclasses corresponded to a specific fish consumption behavior.

Table 7 Macroclasses (Ci, i=1…5) obtained by each of three procedures of classification (AHC, centroids motives and SOM).

Segmentation into five classes of the population observed according to the consumption data set at 12 weeks of pregnancy explained more than 75% of total inertia.

Table 8 Consumption of different fish species by macroclass.

Figure 3 represents a grouping of consumers according to similarities of fish consumption patterns at 12 weeks of pregnancy. Colors indicate macroclasses, obtained applying a hierarchical classification to the centroids of classes. Figure 4 is a three-dimensional visualization of the consumption of different fish categories in macroclasses. These figures represent the consumption for each type of fish considered in the clustering in Figure 3 and with a similar color code. It also allows to see the transitions between macroclasses by the visualization of average consumption for each of the units of the Kohonen map.

Figure 3

SOM macroclasses at 12 weeks of pregnancy: grouping of consumers according to similarities of fish consumption patterns at 12 weeks of pregnancy. Colors and the numbers within the figure indicate macroclasses, obtained applying a hierarchical classification to the centroids of classes. Codes inside the units/cells refer to individuals belonging to each class.

Figure 4

SOM macroclasses at 12 weeks of pregnancy: three-dimensional visualization of the mean consumption of different fish categories in each of the 64 cells/units. For each cell/unit presented in Figure 3, the consumption of cod (graph 1), hake Alaska (graph 2), predatory fish (graph 3) and fatty fish (graph 4) is provided. Colors refer to the macroclasses similarly to Figure 3. Example of reading in graph 3: the consumption of predatory fish is predominant in macroclasses 5 and 2 (orange and mauve). These graphs allow prolonging descriptions of the classes and the super classes. a: Predatory fish including tuna, dogfish, shark, swordfish, grenadier, ling, marlin and grouper. b: Fatty fish including herring, mackerel, sardine, salmon and trout.

Results are summarized in Table 9. The classes could be described as follows:

Table 9 Summary table of the repartition of subjects in the five clusters at 12 weeks of pregnancy regarding MeHg dietary exposure, n-3 PUFA intake and mercury concentration in hair.

Macroclass 1 (colored pink in Figures 3 and 4): This group represented 81% of the population observed. This group was characterized by individuals regularly consuming small quantities of the four fish species used in the clustering procedure: within this subgroup, 81% of individuals ate less than the average weekly consumption (AWC)=222 g of fish. The dietary exposure to MeHg was below 0.7 μg/kg bw and below 1.6 μg/kg bw for 92% and 99% of individuals, respectively, in this macroclass based on modeled exposure. As for hair concentrations of MeHg, 77% of the individuals of this class exhibited levels below 0.93 μg/g hair (third quartile). n-3 PUFA intake in this class was below 0.43 g/day (median value) in 46% of subjects.

Macroclass 2 (colored mauve in Figures 3 and 4): This group contained 5% of the population observed and was characterized by a high predatory fish consumption (AWC=217 g; AWC of the population=51 g). Individuals in this group also consumed fatty fish, cod and hake, but in moderate quantities (see Table 8). The dietary exposure to MeHg for this group was above 0.7 μg/kg bw/week but below 1.6 μg/kg bw/week for 75% of the individuals. The dietary exposure to MeHg for other subjects in this class (25%) was above 1.6 μg/kg bw/week. One in two individuals in this class had mercury concentrations above 0.93 μg/g in their hair. A large proportion (87%) of estimated intakes of n-3 PUFA in this class was above the median value of 0.43 g/day.

Macroclass 3 (colored green in Figures 3 and 4): This class represented 6% of the population observed and was characterized by a high cod consumption (AWC of 187 g, compared with the general AWC of 51 g). The consumption of fatty fish and predatory fish was below the general mean (see Table 8). The dietary exposure to MeHg for the majority of individuals in this group (70%) was below 0.7 μg/kg bw/week. Eighty percentage of consumers in this macroclass had methylmercury concentrations in their hair lower than 0.93 μg/g. The estimated n-3 PUFA intake was higher than 0.43 g/day in 60% of individuals.

Macroclass 4 (colored blue in Figures 3 and 4): This group included 4% of the population observed, and fatty fish consumption was dominant (AWC=344 g; the mean consumption in the population=78 g). The average consumption of predatory fish was close to the average in the general population. Cod and hake consumption was higher than the general AWC. The dietary exposure to MeHg for all consumers in this macroclass was below 1.6 for MeHg, and 66% of them had methylmercury concentrations in their hair lower than 0.93 μg/g. Concentrations of omega-3 were above 0.43 g/day for all individuals in this class.

Macroclass 5 (colored orange in Figures 3 and 4): This group included about 4% of the population observed and corresponded to the consumption of large quantities of fatty fish, predatory fish, cod and hake, the AWCs of which were 380.21, 300.83, 246 and 238 g, respectively, when compared with the general AWC (see Table 8). The dietary exposure to MeHg for 83% of the consumers within this macroclass was above the PTWI of 1.6 μg/kg bw/week, and 66% of them had methylmercury concentrations higher than 0.93 μg/g in their hair. The estimated intake of omega-3 by all individuals in this class was more than 1.5 g/day, corresponding to three times the RDA.

SOM Macroclasses at 32 Weeks

After 32 weeks of pregnancy, we observed a different segmentation of the population. This segmentation explained 69% of total inertia. Only two out of five classes were similar at 12 and 32 weeks of pregnancy based on a multidimensional Hotelling test, that is macroclass 2 (P=0.79), characterized by a high consumption of predatory fish (13% of subjects at 32 weeks), and macroclass 3 (P=0.13), characterized by a high consumption of cod (8% of subjects at 32 weeks of pregnancy). Macroclass 4 was also characterized by a high consumption of fatty fish at both 12 and 32 weeks, but due to the limited number of subjects in these classes (six and eight subjects, respectively), the statistical test failed to show any similarity (P=0.018). Otherwise, we observed that the largest class (macroclass 1) included individuals consuming small quantities of the four fish species used for the clustering procedure. At 32 weeks, this class was smaller than that at 12 weeks, accounting for 53% of subjects (vs 81% at 12 weeks). This was because first, the number of observations decreased, and second, the number of individuals consuming less than the AWC of the four fish species and globally included in this class also decreased (from 68% at 12 weeks of pregnancy to 56% at 32 weeks). Finally, macroclass 5, characterized at 12 weeks by a high consumption of the four fish categories, had disappeared at 32 weeks. A “new” macroclass 5 appeared, in which women were consuming all fish species in larger quantities than subjects in macroclass 1, but less than the earlier macroclass 5 (Table 8). It could be noted that a similar analysis based only on women having participated both at 12 and 32 weeks provides similar differences between fish consumption patterns.


Organic mercury is considered to be more toxic than other forms of mercury following ingestion, and fish and other seafoods are the main sources of exposure to this element. Similarly, fish and other seafoods are the predominant sources for n-3 PUFA intake far above the other dietary sources (with the exception of supplementation with fish oil). It is, therefore, relevant to assume that the consumption of fish will be a good indicator of the uptake of both MeHg and n-3 PUFAs.

This study enabled a better characterization of the risk of MeHg exposure in pregnant French women consuming fish. On the basis of a population of women consuming fish, recruited in three maternity units, the results show that the proportion of pregnant women exposed above the PTWI for MeHg is similar (5%) to that observed among women of childbearing age belonging to a representative sample of the French population (Crépet et al., 2005; Verger et al., 2007). This similarity may indicate that about 5% of French women could exceed the PTWI for MeHg during pregnancy even if only a study fully representative of the national population of pregnant women would capture eventual regional variations. Indeed, French obstetricians frequently advise women to consume more fish during pregnancy because of the potential benefits of n-3 PUFA and are rarely aware of MeHg content of certain fish species. In view of the fact that 75% of consumers are eating fish at least once a week (CREDOC, 1998), and 800,000 babies are born in France every year (INSEE, 2007), 5% of pregnant women consuming fish exposed over the PTWI would correspond to about 30,000 newborns per year in whom exposure exceeds the limit value for MeHg. Although exceeding the PTWI or the lower limit value recommended by the US EPA may not result in any detectable adverse effect, emphasis on protecting optimum brain development would suggest that exposure to neurotoxicants should be minimized. In terms of the mercury levels found in the hair of pregnant women, results were consistent with those obtained by modeling. Thus, the percentage of women exceeding a limit value of mercury in their hair was similar to the one based on the FFQ data (4% of subjects). The uncertainty in blood-to-hair ratio has been well discussed by Budtz-Jørgensen et al. (2002) and the one in dietary exposure modeling was lengthily described by EFSA (2006). Nevertheless, the consistency between these two indirect indicators of exposure to MeHg is likely to reflect the actual situation.

The cluster analysis provides a key to understanding the consumer behavior regarding fish consumption and its relationship with MeHg exposure and n-3 PUFA intake. First of all, we found that although about 5% of subjects exceeded the PTWI for MeHg, some 50% of the pregnant women were consuming less than the recommended amount of 500 mg/day n-3 PUFA. The non-supervised clustering identified five different behavior patterns that were largely predictable of the dietary exposure to both MeHg and n-3 PUFA. The macroclass containing the majority of the subjects (class 1 in Figure 3) was characterized by a relatively low total consumption of different fish species (mean: 440 g/week). This consumption corresponded to the most widespread recommendations (eating two portions of fish per week). Within this class, 99% of individuals were below the PTWI established for MeHg and 77% were below 0.93 μg/g for mercury in hair. The distribution of n-3 PUFA intake in this class was close to that found in all the populations studied, that is one in two individuals was consuming less than nutritional guideline levels of n-3 PUFA. Subjects in class 3 could be similarly described (Figure 3): they were consuming slightly more fish than those in class 1 (mean: 474 vs 440 g/week), but a large proportion of the fish consumed was cod (mean: 187 vs 30 g/week), which is a white/lean fish. Both classes may be defined as moderate fish eaters. Subjects in class 2 were characterized by a high consumption of fish (mean: 680 g/week), including a high consumption of predatory fish (mean: 217 vs 51 g/week in class 1). These subjects were eating about three portions of fish per week — including one portion of predatory fish — and could therefore be defined as high consumers of predatory fishes. In contrast, subjects in class 4 had an even higher overall consumption of fish (mean: 890 g/week), and a high consumption of fatty fish (mean: 344 vs 78 g/week in class 1) could be defined as high consumers of fatty fishes. Finally, subjects in class 5 were the highest consumers of all fish species (mean: 1540 g/week), eating it almost every day (or more than seven times a week), and could be defined as high consumers of fish. In this group, where all subjects were exposed above the PTWI for MeHg, the additional nutritional interest of their n-3 PUFA intake, which corresponded to three times the nutritional guideline level, may be questionable.

The comparison between 12 and 32 weeks of pregnancy made it possible to document a decline in the numbers of both high (macroclass 5) and low consumers (macroclass 1) in regard to particular fish species. Interestingly, we observed an increase in the number of high consumers of predatory fish (13% vs 5%). Some individuals (9% of those observed over the two periods) were grouped in class 1 at 12 weeks, but later belonged to the class characterized by a high consumption of predatory fish. This explained why the overall results were similar in terms of fish consumption and MeHg exposure between the beginning and the end of pregnancy. The differences between 12 and 32 weeks do not correspond to specific recommendations for pregnant women or to observable seasonal variations in fish availability. These preliminary observations would possibly request further investigations.

The clusters identified suggest that different groups of women may not require the same information in regard to recommendable seafood consumption. Even in this coastal population with easy access to seafood, most women consumed less fish than needed to reach PUFA intake levels considered optimal. However, the MeHg exposure levels show that supplementary fish intake among these women should emphasize fatty fish that are low in MeHg. Among women who were eating much seafood, recommendations to abstain from eating predatory species would be appropriate.


The ultimate goal of risk analysis in food science is to guide decision making to reduce the exposure to undesirable substances and optimize the coverage of nutrient intakes according to nutritional recommendations. This study proposes a tool to relate more accurately in existing and future cohort the observed effects with the actual (dietary) exposure. On the basis of a sample of French pregnant women, it also provides evidence on risk/benefit characterization regarding fish consumption. It shows, in particular, that a global increase in this consumption during pregnancy above the generally recommended amounts leads to MeHg exposure exceeding the toxicological limits, thereby questioning the overall effects that can be expected. Although the knowledge of consumer behavior is incomplete, the dietary advisories should take into account that different clusters exist and that different messages may be needed. This underlines the complexity of communicated health benefits and also points to the difficulty that consumers face when a detailed risk/benefit information is provided (for discussion, see Blanchemanche et al., 2006; Roosen et al., 2009).


  1. 1.

    None of the previous hair–blood comparison studies have included hair of African type, and comparison with mercury in Caucasian hair and with existing dose–effect relationships would therefore involve additional uncertainty. Further, it is more difficult to obtain an accurate 2-cm segment from curly hair. Although some Africans may have straightened their hair by permanent treatment, this type of hair treatment is known to leach mercury out of the hair, thereby causing a bias.

  2. 2.

    The FFQ in French can be uploaded on our web site:

  3. 3.

    World Medical Association 1997 (



Agence Française de Sécurité Sanitaire des Aliments


average weekly intake


docosahexaenoic acid


docosapentaenoic acid


eicosapentaenoic acid


food frequency questionnaire


Joint FAO/WHO Expert Committee on Food Additives and Contaminants




provisional tolerable weekly intake


polyunsaturated fatty acid


Recommended Daily Allowance


self-organizing map


United States Environmental Protection Agency


  1. Aberg B., Ekman L., Falk R., Greitz U., Persson G., and Snihs J.O. Metabolism of methyl mercury (203Hg) compounds in man. Arch Environ Health 1969: 19 (4): 478–484.

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  2. AFSSA. Agence Française de Sécurité Sanitaire des Aliments. Avis relatif à la réévaluation des risques sanitaires du méthylmercure lies à la consommation des produits de la pêche au regard de la nouvelle dose hebdomadaire tolerable provisoire (DHTP). Saisine no2003-SA-0380, 16 March 2004, Maison-Alfort, France, 2004.

  3. Alessandri J.M., Guesnet P., Vancassel S., Astorg P., Denis I., Langelier B., Aid S., Poumes-Ballihaut C., Champeil-Potokar G., and Lavialle M. Polyunsaturated fatty acids in the central nervous system: evolution of concepts and nutritional implications throughout life. Reprod Nutr Dev 2004: 44 (6): 509–538.

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  4. Blanchemanche S., Marette S., Roosen J., and Verger P. Food risk management and regulatory experiments. The case of methylmercury in fish. In: Andersson K. (Ed.). VALDOR 2006 —VALues in Decisions On Risk – Proceedings. Stockholm, Congrex, 2006, pp. 441–448.

    Google Scholar 

  5. Budtz-Jørgensen E., Grandjean P., and Weihe P. Separation of risks and benefits of seafood intake. Environ Health Perspect 2007: 115: 323–327.

    Article  PubMed  PubMed Central  Google Scholar 

  6. Budtz-Jørgensen E., Keiding N., Grandjean P., and Weihe P. “Estimation of health effects of prenatal methylmercury exposure using structural equations model”. Environmen Health 2002: 1 (2): 145–168.

    Google Scholar 

  7. CALIPSO. Fish and seafood consumption study and biomarker of exposure to trace elements, pollutants and omega 3, 2006. Leblanc, J.Ch. coordinator.

  8. Cordier S., Garel M., Mandereau L., Morcel H., Doineau P., Gosme-Seguret S., et al. Neurodevelopmental investigation among methylmercury-exposed children in French Guiana. Environ Res 2002: 89: 1–11.

    CAS  Article  Google Scholar 

  9. Cottrell M., Gaubert P., Letremy P., and Rousset P. Analyzing and representing multidimensional quantitative and qualitative data: demographic study of the Rhone valley. The domestic consumption of the Canadian families. In: Oja E., and Kaski S. (Eds.), Finland, Kohonen Maps. ISBN-13: 978-0-444-50270-4, Elsevier, 1999, Chapter 1, pp. 1–14.

    Google Scholar 

  10. Claisse D., Cossa D., Bretaudeau-Sanjuan G., Touchard G., and Bombled B. Methylmercury in molluscs along the French coast. Mar Pollut Bull 2001: 42: 329–332.

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  11. Cossa D., Auger D., Averty B., Lucon M., Masselin P., Noel J., and San-Juan J. Atlas Des Niveaux de Concentration en Métaux Métalloïdes et Composés Organochlorés dans les Produits de la Pêche Côtière Française. Technical Report, IFREMER, Nantes, 1989.

    Google Scholar 

  12. Credoc. OCA-CREDOC. “Les disparités régionales de la consommation alimentaire des ménages français.” Rapport no. CP004. Février 1996, Paris, 1996.

  13. Credoc. OCA-CREDOC. “Vérification des régimes alimentaires proposés par un organisme non gouvernemental pour estimer des expositions aux dioxines.” Note technique no. 98.33. Septembre 1998, Paris, 1998.

  14. Crépet A., Tressou J., Verger P., and Leblanc J.C. “Management options to reduce exposure to methylmercury through the consumption of fish and fishery products by the French population”. Regul Toxicol Pharmacol 2005: 42: 179–189.

    Article  PubMed  PubMed Central  Google Scholar 

  15. Denburg J.A., Hatfield H.M., Cyr M.M., Hayes L., Holt P.G., Sehmi R., Dunstan J.A., and Prescott S.L. Fish oil supplementation in pregnancy modifies neonatal progenitors at birth in infants at risk of atopy. Pediatr Res 2005: 57: 276–281.

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  16. Dewailly E., Ayotte P., Lucas M., and Blanchet C. Risk and benefits from consuming salmon and trout: a Canadian perspective. Food Chem Toxicol 2007: 45 (8): 1343–1348.

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  17. Dunstan J.A., Roper J., Mitoulas L., Hartmann P.E., Simmer K., and Prescott S.L. The effect of supplementation with fish oil during pregnancy on breast milk immunoglobulin A, soluble CD 14, cytokine levels and fatty acid composition. Clin Exp Allergy 2004: 34: 1237–1242.

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  18. EFSA. “Opinion of the Scientific Panel on Contaminants in the Food Chain on a Request from the European Parliament Related to the Safety Assessment of Wild and Farmed Fish”. EFSA J 2005: 236: 1–118. February 2006).

    Google Scholar 

  19. EFSA (European Food Safety Authority). Guidance of the Scientific Committee on a request from EFSA related to Uncertainties in Dietary Exposure Assessment. EFSA J 2006: 438: 1–54.

    Google Scholar 

  20. FAO. Committee on World Food Security. The Role of Aquaculture in Improving Food Security and Nutrition. 29th session, Rome 2003:

  21. FAO. 2003–2008. topics fact sheets — fish composition. Text by Peter manning. In FAO Fisheries and Aquaculture Department [online]. Rome. Updated 15 09 2006. 2006 〈〉.

  22. Feunekes I.J., Van Staveren W.A., Graveland F., De Vos J., and Burema J. Reproducibility of a semiquantitative food frequency questionnaire to assess the intake of fats and cholesterol in The Netherlands. Int J Food Sci Nutr 1995: 46: 117–123.

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  23. FSA (UK Food Standards Agency). Advice on fish consumption: benefits & risks. Scientific Advisory Committee on Nutrition and Committee on toxicity, 2004 Published at:

  24. Gordon A.D. Classification, 2nd edn. Chapman & Hall/CRC, Monographs on Statistics and Applied Probability 82, 1999 ISBN 1-58488-013-9.

  25. Grandjean P., Budtz-Jørgensen E., White R.F., Jørgensen P.J., Weihe P., Debes F., and Keiding N. Methylmercury exposure biomarkers as indicators of neurotoxicity in children aged 7 years. Am J Epidemiol 1999: 150: 301–305.

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  26. Grandjean P., Jørgensen P.J., and Weihe P. Validity of mercury exposure biomarkers. In: Wilson S.H., and Suk W.A. (Eds.). Biomarkers of Environmentally Associated Disease. CRC Press/Lewis Publishers, Boca Raton, FL, 2002, pp. 235–247.

    Google Scholar 

  27. Grandjean P., Weihe P., White R.F., Debes F., Araki S., Murata K., Sørensen N., Dahl D., Yokoyama K., and Jørgensen P.J. Cognitive deficit in 7-year-old children with prenatal exposure to methylmercury. Neurotoxicol Teratol 1997: 19: 417–428.

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  28. Helland I.B., Saugstad O.D., Smith L., Saarem K., Solvoll K., Ganes T., and Drevon C.A. Similar effects on infants of n-3 and n-6 fatty acids supplementation to pregnant and lactating women. Pediatrics 2001: 108: 82–91.

    Article  Google Scholar 

  29. INSEE. La France en faits et chiffre (mise à jour 01/07), 2007:

  30. International Society for the Study of Fatty Acids and Lipids (ISSFAL). Recommendations for intake of polyunsaturated fatty acids in adults, 2004:

  31. Kjellström T. Effects on early childhood development of prenatal exposure to methylmercury. Arch Environ Health 1991: 46 (2): 118.

    Google Scholar 

  32. Kohonen T. Self-Organizing Maps. Springer Series in Information Sciences, vol. 30, Springer, Berlin, Heidelberg, New York, 2001.

    Google Scholar 

  33. Lebel J., Mergler D., Lucotte M., Amorim M., Dolbec J., Miranda D., Arantes G., Rheault I., and Pichet P. Evidence of early nervous system dysfunction in Amazonian populations exposed to low levels of methylmercury. Neurotoxicology 1996a: 17: 157–168.

    CAS  PubMed  PubMed Central  Google Scholar 

  34. Lebel J., Mergler D., Lucotte M., Amorim M., and Dolbec J. Neurotoxic effects of low-level methylmercury contamination in the Amazonian Basin. Environ Res 1998b: 79: 20–31.

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  35. MAAPAR (Ministère de l’Agriculture, de l’Alimentation, de la Pêche et des Affaires Rurales). Résultats Des Plans de Surveillance Pour Les Produits de la Mer de 2002 à 2004. MAAPAR, Paris, 2004.

  36. McKeown-Eyssen G., Ruedy J., and Neims A. Methylmercury exposure in northern Quebec. II: neurological findings in children. Am J Epidemiol 1983: 118 (4): 470–479.

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  37. Mahaffey K.R. Fish and shellfish as dietary sources of methylmercury and the ω-3 fatty acids, eicosahexaenoic acid and doxosahexaenoic acid: risks and benefits. Environ Res 2004: 95: 414–428.

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  38. Martin A. Apports Nutritionnnels Conseillés Pour La Population Française. Tec & Doc, 3e édition, Paris, 2001.

    Google Scholar 

  39. Murata K., Weihe P., Araki S., Budtz-Jorgensen E., and Grandjean P. Evoked potentials in Faroese children prenatally exposed to methylmercury. Neurotoxicol Teratol 1999a: 21 (4): 471–472.

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  40. Murata K., Weihe P., Renzoni A., Debes F., Vasconcelos R., and Zino F., et al. Delayed evoked potentials in children exposed to methylmercury from seafood. Neurotoxicol Teratol 1999b: 12 (4): 343–348.

    Article  Google Scholar 

  41. Myers G.J., Davidson P.W., Cox C., Shamlaye C.F., Palumbo D., Cernichiari E., Sloane-Reeves J., Wilding G.E., Kost J., Huang L.S., and Clarkson T.W. Prenatal methylmercury exposure from ocean fish consumption in the Seychelles child development study. Lancet 2003: 361 (9370): 1686–1692.

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  42. N.R.C. (National Research, Council). Toxicological Effects of Methylmercury. National Academy Press, Washington, DC, 2000.

  43. OFIMER. Office National Interprofessionnel des Produits de la Mer et de L’Aquaculture. Bilan Annuel 2004. Consommation des produits de la pêche et de l’aquaculture, Paris, 2005: June 2006).

  44. Oken E., Wright R.O., Kleinman K.P., Bellinger D., Amarasiriwardena C.J., Hu H., Rich-Edwards J.W., and Gillman M.W. Maternal fish consumption, hair mercury, and infant cognition in a U.S. cohort. Environ Health Perspect 2005: 113: 1376–1380.

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  45. Olsen S.F., Hansen H., Secher N.J., and Sandstroem B. Gestation length and birth weight in relation to intake of marine n-3 fatty acids. Br J Nutr 1995: 73: 397–404.

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  46. Olsen S.F., Hansen H.S., Sommer S., Jensen B., Sorensen T.I., Secher N.J., and Zachariassen P. Gestational age in relation to marine n-3 fatty acids in maternal erythrocytes: a study of women in the Faroe Islands and Denmark. Am J Obstet Gynecol 1991: 164: 1203–1209.

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  47. Olsen S.F., and Secher H.J. Low consumption of seafood in early pregnancy as a risk factor for preterm delivery: prospective cohort study. Br Med J 2002: 324: 447–450.

    Article  Google Scholar 

  48. Pineau A., Piron M., and Boiteau H-L. Determination of total mercury in human hair samples by cold vapor atomic absorption spectrometry. J Anal Toxicol 1990: 14: 235–238.

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  49. Roosen J., Marette S., Blanchemanche S., and Verger Ph. Does Health Information Matter for Modifying Consumption? A Field Experiment Measuring the Impact of Risk Information on Fish Consumption. Review of Agricultural Economics 2009: 31: 2–20.

    Article  Google Scholar 

  50. Simopoulos A.P., Leaf A., and Salem Jr N. Essentiality of and recommended dietary intakes for omega-6 and omega-3 fatty acids. Ann Nutr Metab 1999: 43: 127–130.

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  51. Smuts C.M., Huang M., Mundy D., Plasse T., Major S., and Carlson S.E. A randomized trial of docosahexaenoic acid supplementation during the third trimester of pregnancy. Obstet Gynecol 2003: 101: 469–479.

    CAS  PubMed  PubMed Central  Google Scholar 

  52. Steuerwald U., Weihe P., Jorgensen P.J., Bjerve K., Brock J., and Heinzow B., et al. Maternal seafood diet, methylmercury exposure and neonatal neurologic function. J Pediatr 2000: 136 (5): 599–605.

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  53. Stewert P.W., Reihman J., Lonky E.I., Darvill T.J., and Pagano J. Cognitive development in preschool children prenatally exposed to PCBs and MeHg. Neurotoxicol Teratol 2003: 25 (1): 11–22.

    Article  Google Scholar 

  54. Strain J.J., Davidson P.W., Bonham M.P., Duffy E.M., Stokes-Riner A., Thurston S.W., Wallace J.M., Robson P.J., Shamlaye C.F., Georger L.A., Sloane-Reeves J., Cernichiari E., Canfield R.L., Cox C., Huang L.S., Janciuras J., Myers G.J., and Clarkson T.W. Associations of maternal long-chain polyunsaturated fatty acids, methyl mercury, and infant development in the Seychelles Child Development Nutrition Study. Neurotoxicology 2008: 29 (5): 776–782.

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  55. SU.VI.MAX. (1994). Portions Alimentaires Manuel Photos Pour L’estimation Des Quantités. Polytechnica éditions, Paris..

  56. Thibaud Y., and Noel J. Evaluation Des Teneurs en Mercure, Methylmercure et Sélénium Dans Les Poissons et Coquillages Des Côtes Françaises de la Méditerranée. Rapp. DERO 89-09, IREMER, Nantes, 1989.

    Google Scholar 

  57. UNEP Global Mercury Assessment Report. New York: United Nations Environmental Programme. 2002 Available: 7 may 2004).

  58. U.S. EPA. Water Quality Criterion for the Protection of Human Health: Methyl Mercury. EPA 0823-R-01-001. U.S. Environmental Protection Agency, Washington, DC, 2001.

  59. Verger P., Houdart S., Marette S., Roosen J., and Blanchemanche S. Impact of a risk-benefit advisory on fish consumption and dietary exposure to methylmercury in France. Regul Toxicol Pharmacol 2007: 48 (3): 259–269.

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  60. Weihe P., Grandjean P., and Jørgensen P.J. Application of hair-mercury analysis to determine the impact of seafood advisory. Environ Res 2005: 97 (2): 200–207.

    Article  PubMed  PubMed Central  Google Scholar 

  61. WHO (World Health Organization). Methylmercury. World Health Organization, Environmental Health Criteria 101, Geneva, 1990.

  62. WHO (World Health Organization). Diet, Nutrition and the Prevention of Chronic Diseases. WHO Technical Report Series 916, World Health Organization, Geneva, 2003.

  63. WHO. Evaluation of Certain Food Additives and Contaminants. Prepared by the sixty-first meeting of the Joint FAO/WHO Expert Committee on Food Additives, World Health Organization, Geneva, June 2003. 2004.

  64. WHO. Evaluation of Certain Food Additives and Contaminants. Prepared by the sixty-seventh meeting of the Joint FAO/WHO Expert Committee on Food Additives, World Health Organization, Geneva, June 2006. 2007.

  65. Willett W.C., Sampson L.S., and Stampfer M.J., et al. Reproducibility and validity of a semiquantitative food frequency questionnaire. Am J Epidemiol 1985: 42: b317–b322.

    Google Scholar 

  66. Williams C., Birch E.E., Emmett P.M., Northstone K., and the ALSPAC Study team Stereoacuity at age 3.5 y in children born full-term is associated with prenatal and postnatal dietary factors: a report from a population-based cohort study. Am J Clin Nutr 2001: 73: 316–322.

    CAS  Article  PubMed  PubMed Central  Google Scholar 

Download references


We are very grateful to Dr. Stéphan Clémençon for his guidance regarding statistical analyses and to Dr. Jessica Tressou for her help in reviewing the paper. We also thank Ms. Brita Andersen and Ranja Bjerring for carrying out mercury analyses, and the women who participated in this prospective study. We are indebted to the members of Saint-Nazaire Hospital, Nantes University Hospital and the Polyclinique de l’Atlantique in Saint-Herblain. We also thank Sylvie and Céline for their assistance with generating this cohort. This research was jointly funded by grants from the Research Program on Human Nutrition and the National Institute of Agricultural Research (INRA).

Author information



Corresponding author

Correspondence to Philippe Verger.

Additional information

Human subjects review

All research methods and materials were approved by the local Ethics Committee for Pays de Loire (Nantes) regarding the protection of individuals participating in biomedical research programs.

Rights and permissions

Reprints and Permissions

About this article

Cite this article

Pouzaud, F., Ibbou, A., Blanchemanche, S. et al. Use of advanced cluster analysis to characterize fish consumption patterns and methylmercury dietary exposures from fish and other sea foods among pregnant women. J Expo Sci Environ Epidemiol 20, 54–68 (2010).

Download citation


  • methylmercury
  • pregnancy
  • fish
  • dietary exposure
  • biomarker
  • non-supervised clustering

Further reading