Ingestion has been a primary route of PCB exposure for people, especially those not working directly in industrial settings. During 2002–2004, women were recruited at delivery from two districts in eastern Slovakia: Michalovce with high PCB contamination from a chemical manufacturing plant, and Svidnik located 70 km to the northwest, having lower environmental levels of PCBs. Concentrations of 15 PCB congeners were measured in maternal serum using high-resolution gas chromatography with electron capture detection. A food frequency questionnaire was developed and validated at the Research Institute of Nutrition and School of Medicine in Bratislava, Slovakia. The questionnaire was comprised of 88 food items representative of the national dietary habits and designed to ascertain (1) the source of food items as reported by the women (locally produced or purchased from a retail outlet) and (2) quantities consumed of high-fat food categories representative of the national dietary habits. Our primary goal was to identify specific food sources, either locally produced or purchased from retail stores, that might predict serum concentrations of PCBs. We used multiple linear regression to examine the relationship of dietary fats to lipid-adjusted serum PCB levels in 948 adult women (Michalovce N=662, Svidnik N=284) who had recently given birth. We adjusted for residential district, age, body mass index, education and duration of previous lactation. Consumption of fat from locally produced foods was significantly associated with higher levels of lipid-adjusted serum PCB (β=0.06, P=0.007). Fat from foods purchased in retail outlets showed no significant association (β=−0.02, P=0.36). There was no interaction between district and diet in predicting serum PCB levels. Comparing women in Michalovce consuming 20 g of fat per day from local sources with those consuming 1 g of fat per day from local sources, lipid-adjusted serum concentrations were predicted to be higher by 81 ng/g lipid or 14.8% 630 ng/g lipid versus 549 ng/g lipid.
A major chemical plant that produced PCBs until 1984 is situated in eastern Slovakia. Improper production and disposal practices have led to environmental contamination of the region, which has been documented in wildlife, lake sediment, local food sources and human tissues (Kocan et al., 1994; Petrik et al., 2006). As PCBs have been associated with a variety of adverse health and developmental outcomes in adults and children, determination of the primary routes of exposure is of direct practical importance.
Previous research evaluating food consumption in relation to body burdens of PCBs has largely focused on populations with diets high in fish or sea mammals. Johansen et al. (2004) observed that the traditional diet of blubber from seals and whales in Greenland was the dominant source of PCB intake. Another study observed monthly intakes of fish, seal and polar bear were positively correlated with PCB concentrations in plasma from 192 residents in East Greenland (Deutch et al., 2004). A Canadian study found that fish intake was the strongest dietary predictor of PCB concentration in breast adipose tissue of women, after controlling for other possible confounders, such as age, body mass index (BMI), an index of adult weight gain, gravidity, duration of lactation, occupational exposure, place of birth, place of hospital care and energy intake (Paris-Pombo et al., 2003). Few investigators have related food other than fish and sea mammals to PCBs. Kim et al. (2004) reported concentrations of PCBs in pork, beef and chicken sampled from domestic South Korean and imported products, and found that beef had higher PCB levels compared with the other products. Very high levels of PCBs in poultry and eggs were reported in a contaminated area from Yugoslavia (Jan and Adamic, 1991). After the 1999 Belgium contamination incident, where PCB contaminated oil was inadvertently added to animal feed increased PCB levels were found in chickens, eggs and pigs. Calculations given a worst case scenario suggested that a person eating this contaminated food could double their body burden after 10–20 meals (Bernard et al., 2002).
As part of a larger study of the health effects from early life exposures to PCBs, we undertook an examination of food consumption patterns and sources of food items in a cohort of women who gave birth in the local hospitals of two Slovak districts. Our main goal was to identify which food items and sources predict serum concentrations of PCBs after adjusting for other relevant factors. This investigation is the first crucial step in the development of strategies for reducing body burdens in the residents of the contaminated area. To determine the role of individual food items and sources of food as a potentially continuing source of exposure to PCBs in eastern Slovakia, we collected complete food frequency questionnaires on 948 women who recently delivered babies in two districts and measured PCB concentrations in serum specimens drawn at the time of delivery.
Materials and methods
This report is based on an on-going cohort study being conducted by a bi-national team of investigators in Slovakia and California, USA. The participants are women recruited from two districts in eastern Slovakia at the time they came to the hospital to deliver. Both districts consist of a central town, Michalovce and Svidnik, and the surrounding villages. Michalovce has high PCB contamination in the environment from a chemical manufacturing plant, and Svidnik is located 70 km to the northeast, with somewhat lower environmental levels of PCBs. The study protocols were approved by the Institutional Review Boards (IRBs) of the University of California, Davis and the Slovak Medical University (SMU) prior to initiating the data collection.
Maternal blood specimens (approximately 20 ml) were collected at delivery by a nurse using venipuncture into the standard vacutainers (S-Monovette Serum SARSTEDT) and labeled by date, time and study identification number. Immediately after the collection of blood, tubes were stored in the refrigerator (5–10°C). Within the next 2 h, samples were transported to the Department of Biochemistry in the local hospital, where they were centrifuged (15 min, 3000 r.p.m.) and serum aliquots were divided into vials. Approximately 3 ml of blood serum were placed into an 8-ml clear screw cap glass vial (CHROMSERVIS) with solid screw cap and teflon liner for analysis of PCBs; and 0.2 ml of blood serum was placed into an Eppendorf microtube (SARSTEDT) for further lipid analysis. Samples were stored frozen (−20°C) and were regularly transported frozen in thermo-boxes to Bratislava, in the west of the country, where they were analyzed at the Research Base of the SMU.
Total serum lipids were determined utilizing the enzymatic ‘summation’ method as described previously (Akins et al., 1989). Total serum lipids (TL) were calculated from serum total cholesterol (TC), nonesterified cholesterol (FC), triglycerides (TG) and phospholipids (PL) using the expression: TL =1.677 *(TC–FC) + FC + TG + PL.
Specimens were stored at −18°C. The concentrations of 15 PCB congeners (28, 52, 101, 105, 114, 118, 123, 138, 153, 156, 157, 167, 170, 180 and 189) were determined in the maternal serum samples by high-resolution gas chromatography with electron capture detection at the Department of Toxic Organic Pollutants at the SMU. The selectivity and sensitivity of the electron capture detection method is not able to quantify other potentially toxic PCB congeners that can be measured by other methods such as high-resolution mass spectrometry. In addition, some of the 15 congeners were not always quantified because of the low levels of lipids in the blood and the limited serum volumes available. These 15 congeners have been measured in other study populations since 1992 in an ongoing program to monitor PCB trends in this region. The laboratory was accredited by the Slovak National Accreditation Service for PCB analysis in biological materials and has regularly participated since 1997 in round robin tests organized by the German Agency for Occupational and Environmental Medicine.
In summary, the procedure included standardized extraction, clean-up and quantification accompanied by a quality control program (Conka et al., 2005). Serum spiked with a known amount of PCB-174 congener, which is relatively low in the Slovak population and hence served as a recovery standard, was treated with a mixture of water and 1-propanol to denaturate the proteins. The PCBs were extracted using solid-phase extraction method with n-hexane and dichloromethane mixture. The extract was concentrated and then cleaned by passing through a Florisil-silica gel/H2SO4 column. The n-hexane and dichloromethane eluate was then evaporated to a small volume and a known amount of PCB-103 congener was added as a syringe standard. An aliquot of the mixture was injected and analyzed on a chromatography system (Agilent Technologies 6890N, Network GC system) equipped with a Ni-63 micro-electron capture detector using a 60-m DB-5 capillary column (J&W Scientific, Folsom, Massachusetts, USA). Quantification was based on the calibration curve generated by the GC peak areas of authentic PCB reference standards versus their concentration. Quality control activities consisted of analyses of samples in batches of 10 simultaneously with a blank and a quality control sample (in-house reference material). To be included in the analysis, a response for a particular congener had to be in the range of 90% to 110% using the concentration of the middle point of the calibration curves for that congener. The limit of detection (LOD) for each analyte was determined from the ratio of noise/peak height (peak of analyte should be at least three times higher than the noise).
The selection of the six PCB congeners (PCB IUPAC 118, 138, 153, 156, 170 and 180) used for our statistical analysis was based on the proportion of individual samples that were below the LOD. For these six congeners, less than 20% of the population had values below the LODs. If an individual's PCB value was below the LOD, imputation was carried out by dividing the LOD value by the square root of 2 (Persky et al., 2001; CDC, 2005; Weisskopf et al., 2005). The PCB congener values were then summed to calculate that subject's summed serum PCB.
Data collection and covariate definitions
Data on health and lifestyle were collected by in-person interviews conducted by trained study staff during the hospital stay. Specimens and data were collected after administration of informed consent. A general questionnaire focused on demographic and socio-economic characteristics, health status, lifestyle factors such as smoking, reproductive histories covering past pregnancies and previous breast-feeding, and living environment.
Each woman's age was coded in years and district of residence was coded as a binary variable. Education was coded as the total years of schooling completed. BMI was defined as weight (kg) divided by height (cm) squared. Breast-feeding was defined as the total number of months of breast-feeding across all previous children.
The food frequency questionnaire was used to obtain detailed data on dietary habits and on sources of foods: locally produced versus purchased from a retail outlet. The questionnaire comprised 88 food items, reflecting the national dietary habits. Previous work in eastern Slovakia suggested that PCBs have accumulated in locally produced foods (Chovancova et al., 2005). Questionnaire items were designed to ascertain information about the dietary habits of the women over the past 12 months, including the grams consumed per day, per week or per month, whichever was appropriate. With respect to PCB exposure, eight specific food types (milk, beef, pork, chicken, fish, lard, butter and eggs) were selected for our analysis because PCBs are fat soluble compounds and so are present in greater abundance in these foods as compared with fruits and vegetables, which have a lower fat content. The selected foods also represent common items in the typical diet of this area. Data processing included conversion into individual daily consumption units by an experienced dietitian. Subsequently, the daily intake of individual foods (g/day) was calculated using software specifically designed for the study. The average grams of each food type usually consumed per day was converted into the grams of fat based on the percentage of fat typically found in each food. The percentage of fat content for each food was obtained from the electronic version of the Slovak food composition database compiled by the Food Research Institute in Bratislava, Slovak Republic (Food Research Institute, 2005). Grams of fat from the individual food items were then aggregated based on the source (local or retail or both). If the study participant indicated that a particular food was obtained from “both” sources, the amount of fat was divided between the local fat and retail fat. The proportions attributed were based on the proportion of the remaining population that obtained each food type from each source. For example, if in the total sample of women, 40% obtained eggs from local sources, then 40% of the fat grams for that subject who had selected the “both” response category for eggs were attributed to the local source and the remaining 60% were attributed to the retail source. Then, we summed the grams of fat across the respective sources, either local or retail, to create a summed local fat source and a summed retail fat source. Since we limited the source of dietary fat to the above-mentioned food types, the aggregate dietary fat from locally produced and retail sources combined do not represent the total fat in the diet but rather the amount consumed from high-fat foods.
We also performed a sensitivity analysis by testing three different methods of apportioning the dietary fat for subjects from the “both” source category. (1) We used proportions that were found in the study population (as described above). (2) We divided the grams of fat from the “both” response category 50:50 into the local or retail categories. (3) We reversed the first method and using the same example attributed the 40% to retail when in the first method the 40% had been attributed to local.
Initial bivariate regression models related the overall sum of the six PCB congeners in serum to fat intake from individual food items, grouped food items and food sources. These initial regression analyses guided the selection of independent variables to include in the final multiple regression model. Maternal age, BMI, sum of previous periods of lactation (in months) and years of education were centered on their respective means. The summed fat values were transformed using the natural logarithm. The outcome of the sum of serum PCBs was lipid-adjusted (ng/g lipids) and then transformed using the natural logarithm. To examine the effects of potential outliers on the fitted model, we compared results from models fitted with and without the outliers. The models provided essentially the same results. Thus, the final model presented below is based on all the data. Additionally, diagnostic checks were performed and results did not indicate violation of model assumptions. Because of substantial overlap in the serum PCB distributions and the similarity in food consumption patterns, we analyzed all subjects together and controlled for district as a covariate.
The linear regression model predicted lipid-adjusted serum concentration of the sum of six PCBs as a function of the following: total grams of fat from locally produced sources, total grams of fat from retail sources, district of her home residence, and the woman's prepregnancy BMI, years of education and previous breast-feeding. The analyses were implemented in SAS/STAT® software, version 9.1 of the SAS system for windows (SAS, 2004).
During the period from October 2002 to December 2004, a total of 2653 participants were invited to participate in this study. Of these, 1134 (42.7%) were eligible, 494 (18.6%) were not eligible and 1026 (38.7%) with unknown eligibility declined to participate. Of the total 1134 participants, 811 (71.5%) were from Michalovce and 323 (28.5%) from Svidnik. From this total, 948 had measurements of maternal serum PCB concentrations, diet information and relevant covariates data to be included in the analysis.
The participant characteristics from this eastern Slovak cohort are described in Table 1. The women ranged in age from 18 to 44 years in Michalovce and 18 to 43 years in Svidnik. The median age was 25 years in both districts. The median BMI in both districts was 21 with a range of 15–41 in Michalovce and 15–38 in Svidnik. The median years of education in both districts was 12, with a minimum of 2 years of education in Michalovce and a minimum of 8 in Svidnik. The maximum education in both districts was 18 years. Forty-nine percent of the women in Michalovce and 51% of the women in Svidnik had previously breast-fed one or more of their children. The median total previous nursing time was 6 months.
The mean grams of the eight food types consumed per day by district and the corresponding grams of fat typically found in each food are described in Table 2. The patterns of food consumption for the two districts were similar. Butter was the largest contributor of fat in the diet, followed by milk, pork and eggs. The grams of fat for local and retail sources by the district and food type are also indicated in Table 2. On average, more than 1/3 of the fat consumed by these women from pork, lard and eggs came from locally produced foods. The minimum, 25th percentile, median, mean, 75th percentile, 95th percentile and maximum values for PCB wet weight, lipid-adjusted PCB, retail and local fat are shown in Table 3. PCBs were higher in Michalovce, although distributions showed considerable overlap.
The primary relationship between serum PCB levels and source-specific dietary fat is summarized by a regression model with PCB concentrations as the outcome, and adjustment for district, age, duration of breast-feeding of previous children, BMI and education level (Table 4). The analysis results indicate that the (total) fat consumed from locally produced sources was a significant predictor of lipid-adjusted serum PCB whereas total fat consumption from retail sources was not. It is informative to compare the predicted lipid-adjusted serum PCB levels for an individual who does not consume local fat or consumes a very small quantity of local fat (e.g. 1 g/day) relative to an individual who consumes a high quantity of local fat (e.g. 20 g/day). The model predictions of lipid-adjusted serum PCB levels for Michalovce residents consuming 0, 1 and 20 g of total local fat per day, in addition to a diet that already consists of a median amount of retail fat (of 17 g/day), are 527, 549 and 630 ng PCB per gram of lipid, respectively. In Svidnik these values are 267, 278 and 319 ng PCB per gram of lipid, respectively. For individuals who consume a high amount of local fat of 20 g, relative to those consuming very low to no local fats (≤1 g/day) the predicted lipid-adjusted serum PCB increased by a range of 14.8% to 19.5%. As expected, the model demonstrates significantly lower levels of lipid-adjusted serum PCBs in residents of Svidnik. Women with higher levels of education or a longer history of breast-feeding also had significantly lower serum lipid-adjusted serum PCB concentrations. In this study, BMI was not a significant predictor of lipid-adjusted serum PCB levels.
Our sensitivity analysis in which we reassigned the proportions of locally produced fat versus fat purchased in a retail outlet in women, who reported consuming both (for a given food item), yielded very similar results.
This study focused on the contribution of locally produced high-fat foods to serum PCB levels in two districts within Slovakia, Svidnik and Michalovce. Both districts consist of a central town and surrounding villages. What distinguishes them is the major chemical plant, located in Michalovce, which in the past manufactured PCBs for 25 years. Both districts have similar diets, which include some locally produced food items and some food items purchased from retail outlets. In this study population, the consumption of locally produced high-fat food items led to increased serum PCB levels. Thus, we found that locally produced food products high in fats represent an ongoing PCB exposure hazard that continues to have an impact on the residents of these districts.
Our utilization of the fat content in a range of high-fat foods is somewhat unique among studies that assessed the contribution of diet to serum PCB levels. Most previous similar investigations focused on populations with high consumption of contaminated fish or sea mammals, and measures used were number of meals from those products (Rylander et al., 1995, 1996; Bjerregaard et al., 2001). Our population was different, in that fish and seafood were not a major component of the diet. Moreover, we approached the question of estimating dietary contribution to serum PCB levels in greater detail. Because PCBs are lipophilic and would likely be distributed in the fat of multiple food sources, we reasoned that grams of fat would be a more sensitive indicator of possible PCB exposure than the total grams of the whole food or a simple frequency measure of consumption. We obtained a complete food frequency questionnaire, which made possible this detailed analysis across a range of food products. Our approach is, in principle, generalizable to populations where a wider diversity or different composition of food items might contribute to PCB exposures.
In this study, district, age, education and lifetime lactation (months) were associated with serum PCB levels. The Michalovce district has higher PCB contamination in the environment as the result of a chemical manufacturing plant being located in this area as compared to Svidnik, which is located 70 km to the northeast and having lower environmental levels of PCBs (Kocan et al., 2001). Older age predicted higher serum PCB levels. This is consistent with what other researchers have noted as indicative of the accumulation of exposure (Furberg et al., 2002) and/or a cohort effect whereby younger women have less exposure to PCB because of decreasing PCB use and levels in the environment (Glynn et al., 2007). Lopez-Carrillo et al. (2001) found that breast-feeding is a route for the excretion of metabolites and that 1–2 years of lactation could reduce DDE serum levels in mothers by 50%. In our study, the number of total months that a woman had breast-fed her children was also inversely associated with serum PCB levels, consistent with this route as an efficient means of excretion from the mother but a substantial source of exposure for the infant. Some studies have found an inverse association between pre-pregnancy BMI and serum PCB levels in pregnant women (James et al., 2002; Glynn et al., 2007). Glynn et al. (2007) attribute this to a dilution effect associated with the rapid weight gain during pregnancy, which might be greater in those with high BMI to begin with. Furberg et al. (2002) found a positive association between serum concentration of PCB and BMI in middle-aged women. In our study, BMI was not a significant predictor of serum PCB levels. We did find women with higher education to have lower serum PCB levels.
The refusal rate at the recruitment stage was 51% in Michalovce and 27% in Svidnik, although many of these were likely to have been ineligible. Whether those who participated differed systematically from the (eligible) refusals is unclear, as is the extent to which those differences might affect relationships between dietary habits and serum PCB levels. Nevertheless, the food consumption habits reported by participants are consistent with the traditions of the region. When women were asked about the source of their food purchases they were provided three possible responses: bought that food item from local sources only, from retail outlets only, or from both the sources. For food items obtained from both the sources, the lack of information on how much they bought from each of the two sources (local or retail) is a potential limitation, although the sensitivity analysis suggests that the results here are robust to the allocation proportions used in these cases.
The current analysis is one of the largest investigations yet published of individual-level dietary habits in relation to serum PCB concentrations. Whereas similar studies focused on populations with a high consumption of fish or sea mammals, our study region consisted of the towns Michalovce and Svidnik and the surrounding villages covering areas of 590 and 840 km2, respectively, which represent a population whose food consumption includes a greater range of dietary sources for these persistent pollutants. For comparison of PCB levels in this study with other published results, we calculated the median for the sum of congeners nos. 118,138,153 and 180. (This value is different from the six congeners selected for the regression model and reported in Table 3.) In our study, the median for these four congeners was 367 ng/g lipid and comparable to those of Faroe Islanders in 1994–1995 (median for the sum of 118, 138, 153 and 180 congeners =640 ng/g lipid) (Steuerwald et al., 2000). Other researchers have reported median PCB-153 values of: 30 (ng/g lipid) (Korrick et al., 2000) in US/Massachusetts (1993–1998); 140 (ng/g lipid) (Walkowiak et al., 2001) in Germany/Dusseldorf (1993–1995); 100 (ng/g lipid) (Muckle et al., 2001) in Canada/Northern Quebec (1995–1998); and 133 (ng/g lipid) (James et al., 2002) in US California (1964–1967). In our study the median serum PCB-153 was 141 (ng/g lipid). Caution must be used when comparing PCB levels across studies due to differences in the methods used to determine the PCB values. Longnecker et al. (2003) focused on comparing PCB-153 across several studies highlighting the complications of such comparisons.
The fat intake in the Slovak Republic has decreased in recent years, but it still exceeds the recommended dietary allowances (Babinska and Bederova, 2002). In our study, locally produced fat was directly associated with higher serum PCB levels and it appears likely that a decrease in intake of locally produced fat may help to not only improve health generally, but also to reduce the PCB body burdens.
This project has been funded by the following sources: NIH no. R01-CA96525, no. P01-ES11269, no. R01-ES015359 and EPA STAR Grant no. R829388, and no. RD-83154001, and no. UL1 RR024146 from the National Center for Research Resources. We gratefully acknowledge the ongoing assistance from our Scientific Advisory Board, including Dr. Michael Alavanja, Aake Bergman, Allen Silverstone, John Vena, Don Patterson and Gerhard Winneke.
About this article
Environmental Science and Pollution Research (2015)