Assessment of pre- and postnatal exposure to polychlorinated biphenyls: lessons from the Inuit Cohort Study.

Polychlorinated biphenyls (PCBs) are food-chain contaminants that have been shown to induce adverse developmental effects in humans. In the course of an epidemiologic study established to investigate neurodevelopmental deficits induced by environmental PCB exposure in the Inuit population of northern Québec (Nunavik, Canada), we compared three biomarkers of prenatal exposure and models to predict PCB plasma concentration at 6 months postpartum. Concentrations of 14 PCB congeners were measured by high-resolution gas chromatography with electron capture detection in lipids extracted from maternal plasma, cord plasma, breast milk (collected at approximately 1 month postpartum), and 6-month-old infant plasma samples. Similar congener profiles were observed in all biologic samples, and PCB-153, the most abundant and persistent PCB congener, was strongly correlated with other frequently detected PCB congeners in all biologic media. When expressed on a lipid basis, maternal plasma, cord plasma, and milk concentrations of this congener were strongly intercorrelated, indicating that PCB concentration in any of these biologic media is a good indicator of prenatal exposure to PCBs. A multivariate model that included maternal PCB-153 plasma lipid concentration, breast-feeding duration, and the sum of two skin-fold thicknesses (an index of infant body fat mass) explained 72% of PCB-153 plasma concentration variance at 6 months postpartum (p < 0.001). By contrast, based on the product of breast-feeding duration times the concentration of PCBs in plasma lipids, which was used as an index of postnatal PCB exposure in several studies, only 36% of infant plasma concentration was explained.

exposure and decreased newborn behavioral function (e.g., reflexes, tonicity, and activity levels) were reported in three of the four studies (Huisman et al. 1995a;Rogan et al. 1986a;Stewart et al. 2000). Adverse neurologic effects lasting up to 18 months of age were found in the Dutch study (Huisman et al. 1995b). Both the Michigan and Dutch studies reported that prenatal PCB exposure was associated with lower birth weight and slower growth rate Jacobson et al. 1990a;Patandin et al. 1998). In the Michigan and Oswego cohorts, prenatal PCB exposure was associated with poorer visual recognition memory during infancy Jacobson et al. 1985Jacobson et al. , 1990bJacobson et al. , 1992. Deficits in psychomotor development lasting up to 24 months were noted in the most highly exposed children of the North Carolina cohort (Gladen et al. 1988;Rogan and Gladen 1991). Prenatal PCB exposure was linked to poorer intellectual function at 4 and 11 years in the children of the Michigan cohort (Jacobson et al. 1990b;Jacobson and Jacobson 1996) and at 42 months in the Dutch cohort (Patandin et al. 1999). Recently, results from a fifth cohort, from Düsseldorf, indicated that prenatal exposure to PCBs was negatively related to mental/motor development at 30 and 42 months. These effects were considered "PCB matrix" dependent because associations were noted with breast milk levels and not with cord blood concentrations. In addition, the authors reported a negative effect of postnatal PCB exposure on an intelligence test performed at 42 months (Walkowiak et al. 2001).
The latter results raise two controversial issues that researchers in this field have argued over for years. First, because associations with adverse developmental effects have been observed with PCB levels in some biologic media but not others, uncertainty has emerged regarding the most appropriate index of prenatal exposure to PCBs (Ribas-Fitó et al. 2001). Hence, there is a need to investigate the validity of various biologic measures of prenatal exposure to PCBs. Second, developmental deficits have been most consistently linked to prenatal, not postnatal, exposure (Jacobson and Jacobson 2001). However, assessing postnatal exposure to PCBs in infants is problematic because of the difficulty in obtaining a sufficient volume of plasma to perform PCB analysis. To circumvent this problem, some researchers have estimated postnatal exposure to PCBs by multiplying the concentration of PCBs in breast milk by the breast-feeding duration (Koopman-Esseboom et al. 1996;Walkowiak et al. 2001). More elaborate models have been proposed to predict PCB body burden in 42month-old children (Lanting et al. 1998;Patandin et al. 1997), but these may not be applicable to infants during the first year of life.
In 1995, we initiated The Inuit Cohort Study to investigate adverse neurodevelopmental effects induced by developmental PCB exposure in Inuit infants from Nunavik. This article is the third in a series reporting results from this epidemiologic study. The first article described determinants of PCBs and methylmercury exposure in Inuit women participating in the study (Muckle et al. 2001a). Fish and seal meat consumption was associated with increased hair Hg concentrations. Traditional food intake during pregnancy was unrelated to PCB body burden, which is more a function of lifetime consumption. The second article presented data on the magnitude of prenatal exposure of Inuit mothers and their newborns to these contaminants and to selected nutrients (Muckle et al. 2001b). To assess prenatal exposure in this study, we measured concentrations of 14 PCB congeners and 11 chlorinated pesticides in maternal plasma, umbilical cord plasma, and breast milk samples. Postnatal exposure was evaluated by quantifying PCBs in plasma samples from 6-month-old infants. Here we report on PCB profiles in the various biologic samples and the correlations between PCB levels among biologic media. Factors associated with PCB plasma lipid concentrations in 6-month-old infants are also presented. Finally, we compare the predictive value of various models to estimate PCB plasma concentrations in Inuit infants during the first year of life.

Materials and Methods
Population. Nunavik is a region of Québec located north of the 55 parallel, where 7,660 Inuit reside in 14 villages scattered along 2,000 km of the Hudson Bay and Ungava Bay shorelines ( Figure 1). Starting in November 1995, all pregnant women from three communities (Puvirnituq, Inukjuaq, Kuujjuaraapik) were invited to participate in a longitudinal study of determinants of health and development during infancy. A research assistant explained the purpose of the study, and women were enrolled after signing an informed consent form. The study protocol was reviewed and approved by the Nunavik Nutrition and Health Committee and by ethic committees of Université Laval and Wayne State University. As of November 1998, 141 mothers had completed the prenatal and postnatal interviews (84% participation rate), and PCB plasma concentrations were available from 128 of them. Data on PCB plasma concentration were available for 90 infants who had reached the age of 6 months.
Sociodemographic and physical characteristics of the mother were obtained during one prenatal and two postnatal interviews (1 month and 12 months postpartum). Infant body weight, height, and skin-fold thickness (triceps and subscapular) were measured during an additional visit at the local clinic when infants were approximately 6 months old (median = 210 days postpartum).
PCB analyses in biologic samples. Maternal blood samples (12.5 mL) were collected for the most part at delivery (median = 2 days postpartum). Cord blood samples (30 mL) were collected after the umbilical cord was severed. Infant blood samples (5 mL) were drawn when infants were approximately 6 months old (median = 206 days postpartum). Blood samples were collected in vials containing ethylenediamine tetraacetate and centrifuged (10 min, 5,000 rpm), and the plasma was transferred in glass vials prewashed with hexane. Plasma samples were stored frozen at -80°C until time of analysis. A 2-mL aliquot of plasma was first extracted with a mixture of ammonium sulfate/ethanol/hexane (1:1:3), and the lipid extract was concentrated and cleaned up on Florisil columns. Fourteen PCB congeners (International Union for Pure and Applied Chemistry no. 28,52,99,101,105,118,128,138,153,156,170,180,183,187) (Ballschmitter and Zell 1980) were quantified in the eluate using a HP-5890 series II gas chromatograph (Hewlett-Packard, Palo Alto, CA, USA) equipped with dual-capillary columns (HP Ultra I and Ultra II; Hewlett-Packard) and dual Ni-63 electron-capture detectors (Hewlett-Packard). Peaks were identified by their relative retention times obtained on the two columns, using a computer program developed in-house. Quantification was mainly performed on the Ultra-1 column. Quality control procedures for PCB analyses were described previously (Rhainds et al. 1999). The limit of detection of the method [mean of blank + (3× standard deviation of blank)] was about 0.02 µg/L for each PCB congener (0.02-0.1 µg/L in the case of infant blood samples). Percent recoveries varied from 92 to 98% for individual congeners. Coefficients of variation (n = 20, different days) ranged from 2.1 to 7.5% and biases from -10.9 to 3.8%. The laboratory of the Institut National de Santé Publique du Québec is accredited by the Canadian Association for Environmental Analytical Laboratories.
Children's Health | Ayotte et al. 1254 VOLUME 111 | NUMBER 9 | July 2003 • Environmental Health Perspectives Breast milk samples (25 mL) were collected by manual expression in polycarbonate vials approximately 1 month after delivery (median = 35 days postpartum). Milk samples were stored frozen at -20°C until time of analysis. PCB congeners were extracted from milk using a mixture of acetone/hexane, followed by hexane alone. Combined organic phases were washed with deionized water, concentrated, and purified on activated Florisil columns. A mixture of dichloromethane/hexane was used to elute the compounds, which were separated and quantified by high-resolution gas chromatography as described above. Detection limits vary from 0.08 to 0.16 µg/L for the various PCBs. Two certified milk reference materials (CRM 188 and 450; Community Bureau of Reference, Brussels, Belgium) were used to assess precision and accuracy of the results. Coefficients of variation varied from 10 to 20%, and relative biases 5 to 15%, depending on the specific congener.
Lipid analyses. Because PCBs distribute mainly in body fat, concentrations in plasma or milk samples were reported in micrograms per kilogram of lipids. Total cholesterol, free cholesterol, and triglycerides were measured in plasma samples by standard enzymatic procedures, whereas phospholipids were determined according to the enzymatic method of Takayama et al. (1977) using a commercial kit (Wako Pure Chemical Industries, Richmond, VA, USA). The concentration of total plasma lipids was estimated according to the formula developed by Phillips et al. (1989). An aliquot of the milk fat extract was weighed in order to determine the concentration of lipids in milk samples.
Statistical analysis. PCB concentrations in biologic samples followed a log-normal distribution. Hence, geometric means and confidence intervals are presented in descriptive statistics, and statistical analyses were performed using log-transformed values (log e ). Whenever chemical analysis yielded a "not detected" result, a value equal to half the limit of detection of the analytical method was entered in the database. Pearson correlation coefficients were used to test a) intercorrelations among concentrations of various PCB congeners within each biologic medium and b) intercorrelations among PCB-153 concentrations in the four biologic media. An analysis of variance followed by the Bonferroni post hoc test was used to assess the impact of exclusive breast-feeding duration (never, ≤ 3 months, > 3 months) on infant plasma lipid PCB concentrations. Maternal and infant characteristics associated with infant plasma PCB concentrations were identified using stepwise multiple regression analyses. The following variables were included in the model: maternal plasma PCB concentration (log micrograms per kilogram lipids), maternal weight before pregnancy (kilograms), maternal height (centimeters), maternal age at birth (years), parity, the average number of cigarettes smoked per day during pregnancy, gestational age (weeks), exclusive breast-feeding duration (days), mixed breast-and bottle-feeding duration (days), infant's weight (kilograms), infant's height (centimeters), infant's body mass index (kilograms per square meter), and the sum of skin-fold thicknesses (millimeters). We included in the final model only the independent variables identified as statistically significant predictors by the stepwise procedure. All statistical analyses were performed using the SPSS for Windows statistical package (version 8.0; SPSS Inc., Chicago, IL, USA).

Results
Characteristics of Inuit women enrolled in the present study and their offspring are presented in Table 1. Fourteen percent of participants were younger than 18 years, and 4.7% were older than 35 years. Twenty-two percent of women were primiparous, and 33% already had three or more children. All but nine women (5.5%) smoked during pregnancy. Few women did not breast-feed their infants (13%); 32% of infants were exclusively breast-fed for 3 months or less, whereas 55% were exclusively breast-fed for more than 3 months. Two newborns (2%) weighed less than 2,500 g at birth.
PCB congeners 99, 118, 138, 153, 180, and 187 were detected in more than 70% of maternal and infant biologic samples. Congener profiles were remarkably similar among the different biologic samples ( Figure  2). The most abundant PCB congener in all biologic samples was PCB-153, representing close to 40% of total PCB concentration defined as the sum of the 6 PCB congeners mentioned above. Congeners 138,180,99,187,and 118 followed in decreasing order,averaging,respectively,23,15,8,8, and 6% of the total PCB concentration in each of the biologic media. Furthermore, PCB-153 concentration was strongly correlated to those of other frequently detected PCB congeners whether in plasma (maternal, neonate, infant) or breast milk (all Pearson's r > 0.84; p < 0.001). Further statistical analyses were limited to PCB-153 because its concentration is a good indicator of all persistent PCBs measured in the present study. Table 2 presents concentrations of PCB-153 in all four biologic media. The highest concentrations were found in breast milk, followed by maternal plasma, cord plasma, and infant plasma. Mean cord plasma to breast milk concentration ratio was 0.63 (range, 0.22-1.71; n = 55), mean cord plasma to maternal plasma was 0.81 (range, 0.30-2.54; n = 79), and mean maternal plasma to breast milk was 0.77 (range, 0.29-1.37; n = 84). We noted very strong correlations among concentrations of PCB-153 in maternal and cord plasma and maternal milk samples, as indicated in were weakly correlated to maternal plasma, cord plasma, or breast milk concentrations (r = 0.29 -0.42; p < 0.01). We explored the relation between breastfeeding duration and plasma PCB levels in 6-month-old infants by stratifying in three categories: never breast-fed, exclusive breast-feeding for 3 months or less, and exclusive breast-feeding for more than 3 months (Table 4). Mean PCB plasma lipid concentration in infants who were breast-fed for more than 3 months was 4.3-fold greater and 6.6-fold greater, respectively (p < 0.001), than those of infants who were breast-fed for 3 months or less and never breast-fed infants.
Initially, we performed a stepwise regression analysis to identify, among several maternal and infant characteristics (see "Materials and Methods," Statistical analysis), those associated with plasma PCB concentration at 6 months of age. Maternal plasma lipid PCB-153 concentration, the duration of exclusive breast-feeding, the duration of mixed (breast + bottle) feeding, and the sum of skin-fold thicknesses were the only statistically significant predictors of infant plasma PCB-153 levels. In the final model shown in Table 5 (R 2 = 0.72; p < 0.001; n = 75), the duration of exclusive breast-feeding, the duration of mixed feeding, and maternal PCB-153 plasma lipid concentration were all positively associated with infant's PCB-153 plasma lipid concentration (p < 0.001), whereas the sum of two skin-fold thicknesses, a surrogate for infant body fat mass, was negatively associated with the dependent variable (p < 0.01). Substituting body weight for the sum of skinfold thicknesses yielded a similar but slightly weaker model (R 2 = 0.69). Maternal plasma lipid PCB-153 concentration was used in regression analyses to index maternal body burden because it was very highly correlated with milk PCB-153 concentration and available for more infants.
Various regression models to predict PCB plasma lipid concentration in Inuit infants are presented in Table 6. Almost identical models were obtained using total PCBs concentrations instead of PCB-153 concentrations (data not shown). As expected, the concentration of PCB-153 in maternal plasma lipids and the breast-feeding duration considered individually (R 2 = 0.08 and 0.53, respectively) did not predict infant PCB concentration as well as when both factors were included in the model (R 2 = 0.66). Adding the sum of skin-fold thicknesses to the later model did not improve its predictive value (data not shown). A model that incorporates the product of total breast-feeding (exclusive and mixed) duration multiplied by PCB-153 concentration in maternal plasma lipids had only a weak predictive value (36% of the variance explained). Clearly, the multivariate model composed of the independent variables presented in Table 5 provides the strongest prediction; it explained 72% of the variance in infant PCB plasma concentration.

Discussion
Our study aimed to compare different biomarkers of prenatal PCB exposure and models to predict PCB plasma lipid concentration in 6-month-old infants, using data from the Inuit Cohort Study, a large epidemiologic study taking place in Nunavik that investigates the role of pre-and postnatal exposure to PCBs and heavy metals on various developmental end points during the first year of life (Muckle et al. 2001a(Muckle et al. , 2001b. When expressed on a lipid basis, concentrations of PCB-153 in umbilical cord plasma, maternal plasma, and breast milk were all highly intercorrelated, indicating that any of these biologic media can be used to assess prenatal PCB exposure. We also found that infant PCB plasma concentration was predicted best by a model that contained measures of maternal PCB concentration in body lipids, breast-feeding duration, and infant fat mass. Previous studies have reported maternal concentrations of PCBs in milk or plasma (serum) samples and concentrations in umbilical cord plasma (serum) samples in various populations Koopman-Esseboom 1994;Rogan et al. 1986b;Skaare et al. 1988). However, failure in most of these studies to report concentrations of these lipophilic compounds on a lipid basis made it difficult to assess properly the relationships between PCB concentrations in these biologic samples. For example, Skaare et al. (1988) reported a mean PCB concentration of 10 µg/kg wet weight in 20 maternal serum samples, whereas the mean concentration in corresponding cord plasma samples was only 3 µg/kg (12 of 20 cord plasma samples had detectable levels); the authors concluded that their results were consistent with the notion that the placenta may function as a partial barrier protecting the fetus from transplacental exposure. Dekoning and Karmaus (2000) recently reviewed the literature on this subject and calculated cord-tomaternal blood ratios ranging from 0.59 to 1.1 when data from four studies were expressed on a lipid basis. In our study, cord plasma PCB concentrations expressed on a lipid basis were on average 81% of those in maternal plasma. A similar ratio (0.87) was obtained recently among Inuit women and their neonates from Greenland when total PCB levels in plasma were expressed on a lipid basis (Bjerregaard and Hansen 2000). Furthermore, in our study very high correlation coefficients were noted between maternal plasma, breast milk, and cord plasma levels (Table 3), and similar congener profiles were noted in maternal and neonates samples (Figure 2). Taken together, these results indicate that the placenta does not constitute a barrier to the transfer of major PCB congeners from the mother to the fetus and that these compounds are transported by passive diffusion to all body lipids. It follows that concentrations of PCBs in maternal plasma, milk, and umbilical cord plasma can all be used as surrogates of prenatal exposure to PCBs. Therefore, associations reported previously between neurodevelopmental outcomes and prenatal PCB exposure that appeared dependent on the "PCB matrix" (i.e., associations observed with PCB concentrations in one biologic sample but not in others) are likely due to analytical problems, as suggested by Walkowiak et al. (2001). These problems may arise from the lower concentration of lipids in Children's Health | Ayotte et al. 1256 VOLUME 111 | NUMBER 9 | July 2003 • Environmental Health Perspectives   umbilical cord plasma than in maternal plasma (2.8 g/L vs. 7.9 g/L in the present study), which makes it more difficult to measure PCBs and other lipophilic compounds reliably in the former biologic medium. To our knowledge, our study is the first to investigate factors influencing plasma PCB concentrations in infants during the first year of life. The major importance of breast-feeding as a determinant of postnatal PCB exposure in infants was demonstrated by stratifying according to the breast-feeding duration. We observed that infants who were breast-fed for more than 3 months displayed a mean plasma PCB concentration 6.6-fold higher than that of bottlefed infants (Table 4). Breast-feeding during 3 months or less increased the mean PCB concentration by only 56% compared with the group never breast-fed. Lanting et al. (1998) reported a mean PCB concentration 4.5 times higher in 42-month-old Dutch children breastfed for at least 6 weeks compared with those never breast-fed. In Dutch children of similar age, Patandin et al. (1997) reported a mean PCB plasma concentration 3-fold higher in the breast-fed group than in the bottle-fed group. In school-age children (7-10 years old) from Germany, exclusive breast-feeding during more than 3 months was associated with a doubling of PCB plasma concentrations (Karmaus et al. 2001). Hence, exposure to these lipophilic compounds through breast-feeding has a major and long-lasting influence on the offspring body burden. Using a toxicokinetic model that takes into account exposure, the diluting effect of body growth, and elimination processes, we have predicted that the influence of breast-feeding on the body burden of persistent organochlorines would last until adulthood in the Inuit population (Ayotte et al. 1996).
A multiple linear regression model that included as statistically significant predictors the exclusive breast-feeding duration, PCB concentration in maternal plasma, mixed (breast and bottle) feeding duration, and the sum of two skin-fold thicknesses explained 72% of plasma lipid PCB concentration in 6-month-old Inuit infants (Table 5). Although no other model was elaborated specifically for infants, others have looked at determinants of PCB plasma levels in children. Jacobson et al. (1989) studied determinants to PCB plasma concentration in 4-yearold Michigan children and found that maternal PCB milk level and breast-feeding duration jointly explained 60% of the variance. Lanting et al. (1998) presented a nonlinear model that explained 75% of the variance in children's levels and included as predictors breast-feeding duration (exclusive), feeding mode (breast or formula), and concentrations of PCBs in cord plasma and in breast milk. Patandin et al. (1997) presented separate models for breast-fed and bottle-fed infants. The formula-fed group model explained 36% of the variance and included as predictors maternal plasma levels, maternal age, the weight of the child, and its dietary PCB exposure. The breast-fed group model explained 63% of the variance and contained the following predictors: the duration of breast-feeding, breast milk PCB concentration, maternal plasma PCB concentration, maternal age, the weight of the child, and dietary intake of PCBs. However, only the duration of breast-feeding, breast milk PCB concentration, and the child's weight showed statistically significant associations with the child's plasma lipid concentration, similar to the model presented here for 6-month-old infants.
Throughout this study, we made the implicit assumption that the concentration of PCBs in infant body lipids is the most appropriate exposure metric to use in studying the relation of postnatal PCB exposure to developmental outcomes. Because PCBs are lipophilic and stored mainly in body fat, the higher the body fat mass, the greater the volume of distribution and the lower the plasma lipid PCB concentrations. In our model, the sum of two skin-fold thicknesses measured on the infants (triceps and subscapular), a marker of body fat mass, showed a negative association with PCB concentration in plasma lipids. Associations of both maternal plasma lipid PCB concentration and breast-feeding duration to infant plasma lipid PCB concentration become stronger when infant body fat mass is included in the model. A statistically significant negative but slightly weaker association was also noted when the infant body weight was entered in the model instead of the sum of skin-fold thicknesses. Skin-fold thicknesses are likely a better indicator of infant body fat mass than is body weight.
In view of the particular source of exposure to PCBs in the Inuit population (sea mammal fat consumption), one might question the degree to which our results are applicable to other populations environmentally exposed to these compounds. Longnecker et al. (2003) reviewed PCB exposure levels from 10 studies of PCB and neurodevelopment, including ours; the authors noted that besides PCB-153, which was the major congener present in maternal biologic samples from nearly all studies, congeners 118, 138, and 180 were also major congeners. Although similarities do exist among studies, the ratio of median PCB-118 concentration to median PCB-153 concentration was the lowest in the Inuit population (0.14), compared with values ranging from 0.18 to 0.87 in the other populations. These results suggest that some differences in PCB congener profiles exist across populations. Notwithstanding these differences, it seems reasonable to rely on a major congener such as PCB-153, which is quantified in biologic samples with relative ease, possesses a long biologic half-life, and is strongly correlated to other congeners, to assess PCB exposure within a given population.
The model developed here to predict postnatal PCB exposure (plasma lipid concentrations at 6 months of age) should be applicable to other populations environmentally exposed to PCBs. Indeed, there is no reason to believe that the factors associated with PCB plasma Children's Health | PCB exposure in Inuit infants Environmental Health Perspectives • VOLUME 111 | NUMBER 9 | July 2003 Table 5. Multiple linear regression analysis of plasma lipid PCB-153 concentration in 6-month-old Inuit infants (n = 75).