Intrinsic Human Elimination Half-Lives of Polychlorinated Biphenyls Derived from the Temporal Evolution of Cross-Sectional Biomonitoring Data from the United Kingdom

We use two sets of congener-specific cross-sectional biomonitoring data for PCBs. The first set consists of PCB concentration in 75 adipose tissue samples from Wales, United Kingdom, which were collected in 1990–1991 (Duarte-Davidson et al. 1994). The age of the individuals who supplied the samples ranged from 14 to 79 years. The second set consists of 154 human blood samples collected in 2003 at 13 locations in the United Kingdom (Thomas et al. 2006). The age of these individuals ranged from 22 to 80 years. To reduce the influence of outliers, we aggregated the data in 10 age groups (in both data sets). Both studies reported results in terms of lipid-normalized concentrations. Lipid-normalized concentrations derived from blood and adipose tissue samples are directly comparable because the two lipid compartments are in equilibrium (Haddad et al. 2000; Patterson et al. 1988; Sorg et al. 2009). Details about samples and analytical methods have been described elsewhere for both data sets (Duarte-Davidson et al. 1994; Thomas et al. 2006).

Congener-specific empirical daily intake data for the U.K. population were derived from total diet studies because dietary intake is the main source of PCB exposure for the general population [for detailed description and references, see Supplemental Material (doi:10.1289/ehp.1002211)].

We employed a population PK model that describes changes in body concentration of PCBs as a function of age and calendar time for multiple individuals representing different birth cohorts of the average population (Alcock et al. 2000; Pinsky and Lorber 1998). The model is a modified version of the multiindividual PK framework that was recently presented and analytically solved for a postban period (Ritter et al. 2009). The model implementation used here differs in two aspects from the earlier version: It is solved numerically and therefore is not restricted to postban conditions; and we implemented age-dependent growth of body mass and lipid mass, and age- and body-weight-dependent dietary intake, including intake by breast-feeding.

Equation 1 defines the time course of PCB concentration in one representative individual born at time t_birth [for a derivation of Equation 1 and additional information, see Supplemental Material (doi:10.1289/ehp.1002211) and Ritter et al. 2009]:

where t_age (years) is the age of the individual, C(t_age) (nanogram per gram lipid) is the lipid-normalized concentration of chemical in the body under the assumption that the chemical is present only in the lipid compartment of the body, M_lip(t_age) (kilograms lipid) is the mass of total body lipid as a function of age, k_elim (years⁻¹) is the first-order rate constant describing intrinsic elimination, and I(t_age, t_birth = constant) (ng × year⁻¹ × kg lipid × g lipid⁻¹) is the exposure trend of the representative average individual born at t_birth and is described in terms of age- and calendar-time-dependent daily intake of chemical as

where U (days × year⁻¹ × kg lipid × g lipid⁻¹) is a unit conversion factor selected to describe quantities in commonly reported units; E_a (dimensionless) describes the net absorption in the gastrointestinal tract and is set at 0.9 (Moser and McLachlan 2001); M_bw(t_age) is the body weight as a function of age in kg; I_ref (t) (ng × kg body weight⁻¹ × day⁻¹) is the reference daily intake of chemical for an adult and depends on the year of sampling, t, which can be expressed as t = t_birth + t_age (Ritter et al. 2009); and P(t_age) (dimensionless) is a proportionality factor adapting I_ref(t) to younger ages according to results from total diet studies (Alcock et al. 2000).

Equation 2 describes the daily intake for individuals with t_age > 6 months, assuming that nursing ends at that age. The daily intake of chemical during nursing is determined from the concentration in the mother [see Supplemental Material (doi:10.1289/ehp.1002211)]. The concentration of chemical in an individual at birth is set equal to the concentration in the mother.

To estimate elimination kinetics, we fit the model to measured data using a least-square optimization by adjusting three fitting parameters: adult reference intakes in the years 1970 and 2000, and k_elim. This is achieved by minimizing the sum of squared residuals weighted (SSRW). SSRW is related to the coefficient of determination, R², of a dataset with n empirical data points by

where y_i is a value of the empirical data set, f_i is the associated modeled value, and y– is the empirical sample mean. SSRW quantifies the differences between modeled and empirical values as the fraction of the sum of squares of residuals (numerator) to the total sum of squares of variability in the dataset (denominator). By minimizing SSRW, R² is maximized.

For the CSD sets from 1990 and 2003, we define SSRW_{CSD_1990} and SSRW_{CSD_2003} according to Equation 3. A third SSRW value, SSRW_Int, quantifies the difference between modeled [i.e., I_ref(t)] and empirical adult reference daily intake. I_ref(t) was defined according to the following assumptions. The intake trend between the years 1970 and 2000 is exponentially interpolated in accordance with monitoring data (Alcock et al. 2002; Bignert et al. 1998). We assume peak intake occurred in 1970. Before 1970, the shape of the modeled adult reference intake trend, I_ref(t), is defined relative to intakes in 1970 and 2000 based on information taken from historic emission inventories (Breivik et al. 2002) and exposure reconstruction studies for PCBs. In particular, the intake in the year 1950 is set to the interpolated value of 1990, which is consistent with reconstructed PCB exposures from other studies (Alcock et al. 2000; Karmaus et al. 2004). All three adjustable parameters, the adult reference intakes in 1970 and 2000 and the elimination rate constant, k_elim, are varied independently until the value of an objective function reflecting the differences between modeled and empirical data is minimized. We use two different objective functions: the function OF_{CSD_Int} reflects equally weighted contributions of each set of CSD and the empirical intake estimates,

and the function OF_{CSD_Only} fits the model by minimizing the difference between modeled and empirical age–concentration CSD only,

The two different objective functions allow us to test whether it is possible to simultaneously extract information about both exposure and elimination from multiple age–concentration CSD sets alone.

In summary, by minimizing the objective functions OF_{CSD_Only} and OF_{CSD_Int} in the course of two separated optimization runs, two different sets of results are generated. Each set consists of estimates of the intake in 1970, the intake in 2000, and the rate constant k_elim. Intrinsic elimination half-life estimates are calculated as ln(2)/k_elim.

Our results agree well with intrinsic elimination half-lives from two other studies (Table 2). Ogura (2004) used data from the general Japanese population and applied a method similar to ours. Ogura (2004) also accounted for changes in ongoing exposure derived from total diet studies and for age-dependent changes in the size of the body’s lipid compartment by using a PK model. Unfortunately, Ogura (2004) investigated different congeners than in our study, except for PCB‐105 and PCB‐118. For these two congeners, Ogura’s estimated intrinsic elimination half-lives of 5.2 years (PCB‐105) and 6.3 years (PCB‐118) are very similar to our estimates of 5.2 and 9.3 years. Grandjean et al. (2008) used a different approach and investigated intrinsic elimination using LD from a large cohort of children from 4 to 14 years old. They did not employ a PK model but used regression analysis to account for changes in ongoing exposure (i.e., the consumption of whale meat) and for changes in body weight. They achieved this by including the body mass index and the number of monthly whale dinners as covariates in the regression. Grandjean et al. (2008) found no indication that intrinsic elimination half-lives depend on age or are shorter in children than in adults after correcting for the effect of growth. A similar observation has been reported by Yakushiji et al. (1984). Table 2 therefore shows that despite the fact that the half-life estimates are based on different data types (LD and CSD), represent cohorts of different ages, and reflect background levels, consistent estimates of intrinsic elimination half-lives can be obtained if the effects of body weight changes and ongoing exposure are accounted for.

Table 3 shows apparent elimination half-lives collected from the literature. Apparent half-lives reflect the overall effect of intrinsic elimination, ongoing exposure, and body weight changes on concentrations as a function of time. Apparent half-lives are subject to a considerably larger variability (Table 3) than are estimates of intrinsic half-lives at background concentration levels (Table 2), which reflect only interindividual variability of intrinsic elimination at similar concentration levels. Apparent elimination half-lives in Table 3 differ by up to a factor of 50 for the same congener. In contrast, the estimates of intrinsic elimination half-lives at background concentration levels in Table 2 differ by less than a factor of 3 and many by less than a factor of 2, although they were derived from different data types (LD and CSD).

The large variability in apparent half-lives reflects cohort-specific differences in all three main factors that influence the observed concentration trend. First, changes in body weight influence the apparent half-life. Growth during childhood leads to growth dilution; that is, much shorter apparent half-lives are observed in infants than in adults (Milbrath et al. 2009). Further changes in body weight during adulthood may also affect the apparent elimination half-life; for example, strong weight loss may even lead to an increase in chemical concentrations with time, which correspond to a negative apparent half-life. In our estimation of intrinsic half-lives, we use a lifetime profile for changes in body weight specific to the U.K. population (Alcock et al. 2002) that takes into account growth dilution during childhood, which is a strong effect in all individuals. For adults, the body weight profile reflects the population average. Second, the rate of intrinsic elimination is faster in cohorts with initial concentrations significantly above background, for example, in incident cohorts measured soon after the exposure incident (Sorg et al. 2009). In such cases, intrinsic elimination reflects the individual status of the patient (e.g., increased elimination via skin, feces, and induced metabolism). In our estimation of intrinsic half-lives, we largely exclude this source of variability by using data from individuals exposed to background concentrations. Third, in cohorts exposed to background concentrations, ongoing exposure can lead to very long apparent elimination half-lives. In our estimation of intrinsic half-lives, we account for this effect by explicitly describing ongoing exposure in the model equation, which allows us to parameterize and quantify intrinsic elimination as a distinct process.

All three effects may contribute to the observation of increasing apparent half-lives in initially highly exposed cohorts that were observed for several decades (Masuda 2001) (Table 3). Because concentrations approach background levels, ongoing exposure becomes relevant also in these incident patients. However, it is not possible to conclude from observed concentration trends whether the increasing apparent elimination half-lives represent a slowdown of intrinsic elimination, for example, due to decreased metabolic activity at lower concentrations (Sorg et al. 2009), or whether it is due to increased confounding from ongoing exposure or body-weight loss.

The strong influence of ongoing exposure and loss of body weight on half-life estimates from incident cohorts that were measured decades after the exposure incident is demonstrated by the observation of very long and even negative apparent half-lives reported for pentachlorodibenzofuran in Yusho patients (Matsumoto et al. 2009). Increasing concentrations (i.e., negative half-lives) can be explained only by additional intake or significant weight reductions under the condition that the substance is not a metabolite synthesized within the body. Very long or infinite apparent elimination half-lives have also been reported for PCBs (Table 3).

The longest intrinsic half-lives from our study are 11.5 years for PCB‐180, 15.5 years for PCB‐170, and 14.4 years for PCB‐153 (Table 2). Other studies (Kreuzer et al. 1997; Shirai and Kissel 1996) have indicated that plausible maximum elimination half-lives of PCBs and dioxins are probably not much larger than 10 years. Our results are further evidence that a maximum intrinsic elimination half-life for persistent chemicals such as PCBs exists and is approximately 10–15 years. This half-life range likely reflects nonmetabolic elimination processes (Kreuzer et al. 1997; Rohde et al. 1999).

Most studies shown in Tables 2 and 3 report “half-lives” or “elimination half-lives” without further specification. However, to make half-life estimates usable, a conceptual and semantic distinction between apparent and intrinsic elimination half-life estimates is needed. This distinction will help to reduce the use of strongly different estimates of elimination half-lives in epidemiologic assessments, where they are needed to parameterize intrinsic elimination in PK models. An example is PCB‐153, for which half-lives selected for use differ by more than a factor of 5, including 5 years (Toft et al. 2008) and 27 years (Verner et al. 2009). Apparent half-lives reflect the overall effect of several factors, including intrinsic elimination, ongoing exposure, and changes in body weight. If ongoing exposure is small relative to concentration levels and body weight is constant, apparent half-lives may reflect intrinsic elimination, but because there is a conceptual difference between intrinsic and apparent half-lives, they will generally also have different numerical values. Importantly, both types of half-life estimates can be derived from both LD and CSD.

To evaluate the intrinsic half-lives obtained with our fitting procedure, we modified the half-lives by a factor of 1.5 and reran the PK model with these modified half-lives. Visual inspection of the results showed that with these modified intrinsic half-lives the calculated body concentrations clearly do not match the data points from the two sets of CSD. This implies that the uncertainty of the estimated intrinsic half-lives is less than a factor of 1.5. This is consistent with the interstudy variability of the intrinsic half-lives in Table 2, which is a factor of 2 or less for eight of the nine congeners and a factor of 3 for PCB‐138. Also the two estimates obtained from our two objective functions, OF_{CSD_Only} and OF_{CSD_Int}, are in good agreement (difference of less than a factor of 1.4 for seven of nine congeners and a factor of 2 for PCB‐170 and PCB‐180). We recommend the estimates based on OF_{CSD_Int}, because they integrate information from all empirical data sources.

A strength of our approach is that our estimates of intrinsic PCB elimination half-lives have a broad empirical base. In contrast to biomonitoring data types with no or only one temporal dimension (Table 1), multiple sets of empirical age–concentration CSD data can be satisfactorily fitted only if good agreement in both temporal dimensions, within each set of CSD (age) and between different sets of CSD (calender time), is achieved. This is shown in Figure 1C [and further illustrated in Supplemental Material, Figure 1 (doi:10.1289/ehp.1002211)]. The additional dimension of information provided by more than one set of CSD makes it possible to derive intrinsic elimination half-lives directly from the biomonitoring data (objective function OF_{CSD_Only}).

We did not separate empirical data according to sex because concentration differences between male and female individuals were small in our data sets [see Supplemental Material (doi:10.1289/ehp.1002211)], which is also consistent with results from other cross-sectional studies (Toms et al. 2009). In addition, a separation by sex would have reduced the size of the data set and therefore the precision of the least-square optimization. Correspondingly, we used median anthropometric data for males and females in the model (Alcock et al. 2000). We also did not separate our data according to parity and smoking status because such information is not consistently available for both data sets. Although these factors may influence apparent elimination half-lives (Milbrath et al. 2009), the influence is likely to be small for cohorts at background exposure levels, relative to the strong influence from ongoing exposure and body weight changes. For our data set from 2003 this is substantiated by the lack of significant correlations between parity and concentration in the data set (Thomas et al. 2006). Increasing efforts devoted to biomonitoring provide a promising perspective that time series of even more than two sets of age– concentration CSD that are stratified for factors such as sex, parity, or smoking status will become available for many persistent chemicals. If applied in a consistent conceptual framework that accounts for the influences of ongoing exposure and body weight changes, such stratified CSD may allow researchers to estimate statistical distributions reflecting the interindividual variability of intrinsic elimination half-lives of persistent chemicals.