Thyroid Function and Perfluoroalkyl Acids in Children Living Near a Chemical Plant

Background: Animal studies suggest that some perfluoroalkyl acids (PFAAs), including perfluorooctanoate (PFOA), perfluorooctane sulfonate (PFOS), and perfluorononanoic acid (PFNA) may impair thyroid function. Epidemiological findings, mostly related to adults, are inconsistent. Objectives: We investigated whether concentrations of PFAAs were associated with thyroid function among 10,725 children (1–17 years of age) living near a Teflon manufacturing facility in the Mid-Ohio Valley (USA). Methods: Serum levels of thyroid-stimulating hormone (TSH), total thyroxine (TT4), and PFAAs were measured during 2005–2006, and information on diagnosed thyroid disease was collected by questionnaire. Modeled in utero PFOA concentrations were based on historical information on PFOA releases, environmental distribution, pharmacokinetic modeling, and residential histories. We performed multivariate regression analyses. Results: Median concentrations of modeled in utero PFOA and measured serum PFOA, PFOS, and PFNA were 12, 29, 20, and 1.5 ng/mL, respectively. The odds ratio for hypothyroidism (n = 39) was 1.54 [95% confidence interval (CI): 1.00, 2.37] for an interquartile range (IQR) contrast of 13 to 68 ng/mL in serum PFOA measured in 2005–2006. However, an IQR shift in serum PFOA was not associated with TSH or TT4 levels in all children combined. IQR shifts in serum PFOS (15 to 28 ng/mL) and serum PFNA (1.2 to 2.0 ng/mL) were both associated with a 1.1% increase in TT4 in children 1–17 years old (95% CIs: 0.6, 1.5 and 0.7, 1.5 respectively). Conclusions: This is the first large-scale report in children suggesting associations of serum PFOS and PFNA with thyroid hormone levels and of serum PFOA and hypothyroidism.

volume 120 | number 7 | July 2012 • Environmental Health Perspectives Research | Children's Health Thyroid hormones play important roles in regulating metabolism, growth, and develop ment, especially in normal brain maturation and development (Porterfield and Hendrich 1993). They regulate the processes of neuro genesis, denditric and axonal growth, synap togenesis, and myelination (Bernal 2007). It is now recognized that even slight differences in the concentration of thyroid hormones dur ing pregnancy or after delivery may be associ ated with neurological impairment (Freire et al. 2010;Pop et al. 1999). Thyroid hor mones are also essential for children because some neurodevelopmental processes, such as myelination, are not completed until adoles cence (Rice and Barone 2000), and they are also important for the behavior and cognitive function of the young and adolescent brain (Anderson 2001). In addition, thyroid hor mone deficiency causes growth delay, preco cious puberty in both sexes, and hirsutism in females (Papi et al. 2007).
There is a growing concern that environ mental toxicants may be related to thyroid impairment (Boas et al. 2009). Animal stud ies have suggested that exposure to some perfluoroalkyl acids (PFAAs), including perfluorooctanoate (PFOA, also called C8), perfluorooctane sulfonate (PFOS), and per fluorononanoic acid (PFNA), may interfere with thyroid status (Lau et al. 2007;Liu et al. 2011). There had been some concern that these apparent associations may derive from the analysis of analog free thyroxine (FT 4 ) being affected by the presence of PFOS, as observed in rats highly exposed to PFOS (Chang et al. 2007). However, such bias has not been observed in a human population with typical U.S. serum PFOS concentrations but higher PFOA concentrations (Lopez Espinosa et al. 2011a).
Published epidemiological findings are not consistent (Steenland et al. 2010), have generally focused on adults, and have been crosssectional in nature-leaving a gap in our understanding of possible PFAA effects in children. Whereas associations were reported for some PFAAs and thyroid hormones in some studies of nonoccupational exposure (Dallaire et al. 2009;Kim et al. 2011;Knox et al. 2011), others found little or no evidence of associations (Bloom et al. 2010;Chan et al. 2011;Emmett et al. 2006). Studies in occupational settings also reported some sta tistically significant associations with thyroid hormones, although at high PFAA concen trations (Olsen and Zobel 2007;Olsen et al. 2003). A higher odds ratio (OR) of reported thyroid disease was recently associated with PFOA or PFOS exposure in the National Health and Nutrition Examination Survey (NHANES) population (Melzer et al. 2010).
PFOA has been used in the manufac ture of fluoropolymers at a chemical plant in Parkersburg, West Virginia, since 1951. In 2001, a group of residents from the Ohio and West Virginia communities in the vicin ity of the Washington Works plant filed a classaction lawsuit, alleging health damage due to contamination of human drinking water supplies with PFOA. The settlement of this lawsuit led to a baseline survey, called the C8 Health Project, conducted during 2005 and 2006 on residents who lived in six contaminated water districts surrounding the chemical plant, as described by Frisbee et al. (2009) (Kato et al. 2011).
Assessment of possible PFAA effects on thyroid hormone function in children is of special interest for the reasons outlined above.
Accordingly, we designed the present study to estimate associations of thyroid function with a) modeled in utero PFOA concentrations and b) measured serum PFOA, PFOS, andPFNA concentrations collected during 2005-2006 in children 1-17 years of age from these Ohio and West Virginia communities.

Methods
Study population. The C8 Health Project enrolled participants between August 2005 and July 2006. All participants gave written informed consent before inclusion: Parents or guardians provided consent on behalf of children. The London School of Hygiene & Tropical Medicine Ethics Committee approved this study. The purpose of the Project was to collect health data from members of the class action lawsuit through questionnaires and blood tests, including measurements of PFAAs. Individuals were eligible to partici pate in the C8 Health Project if they had con sumed water for at least 1 year between 1950 and 2004 from six contaminated water dis tricts or private wells in proximity to a Teflon manufacturing facility. The C8 Health Project collected data on 69,030 people, of whom 12,476 were 1-17 years of age at enroll ment. Participation rates for age groups 5-10, 11-14, and 15-19 years, and residing in the area at the time of the survey, were 77%, 87%, and 95%, respectively (Frisbee et al. 2009). Of the 12,476 children, 10,725 (86%) had serum PFAA and thyroid hormone measure ments or information on reported thyroid dis eases (from questionnaire responses), and were included in the present analyses. Within this population, 4,713 children were successfully matched to their mothers (also participating in the C8 Health Project) (Mondal et al. 2012); effects on child thyroid function in relation to modeled in utero PFOA exposure was also estimated for this subsample.
PFAA determinations. Laboratory analy ses of PFAA were conducted by a commer cial laboratory (Exygen, State College, PA, USA). Samples collected at survey were analyzed for 10 PFAAs including PFOA, PFOS, and PFNA. The laboratory analyti cal methods and quality control procedures have been described elsewhere (Frisbee et al. 2009). Briefly, serum concentrations of PFAA were determined using liquid chromatog raphy separation with detection by tandem mass spectrometry. Estimates of precision for PFOA were within ± 10% for multiple replicates over the range of 0.5-40 ng/mL, with a more precise relative precision meas ure of approximately 1% for highly fortified (10,000 ng/mL) samples. Relative precision estimates for PFOS and PFNA were similar to those for PFOA. The detection limit (LOD) was 0.5 ng/mL, and observations below the LOD were assigned a value of 0.25 ng/mL (n = 0, n = 16, and n = 107 in the case of PFOA, PFOS, and PFNA, respectively, for this study population).
Historical PFOA exposures for all par ticipants in the C8 Health Project were esti mated through environmental, exposure, and pharmaco kinetic modeling in conjunction with selfreported residential histories. Information on plant operations and chemical releases was combined with environmental characteristics of the region through a series of linked models to estimate air and water concentrations of PFOA from 1951 to 2008 (Shin et al. 2011a). Based on estimates of individual air and water intake rates and linkage of residential geocodes for participant address histories to public water distribution systems and private wells, yearly PFOA serum concentrations were estimated for each participant in the C8 Health Project (Shin et al. 2011b). Historical individual mod eled serum PFOA was calibrated by factors derived from comparisons of observed with predicted serum concentrations in [2005][2006]. The ratio of observed to predicted (before cali bration) estimates for these mothers showed a geometric mean (GM) of 1.36, and the inter quartile range (IQR) of these ratios was 0.7 to 2.2. In utero, modeled exposures for each child were estimated as the modeled serum concen trations in the mother who had been success fully matched to the child, at the time of the first trimester of pregnancy (n = 4,713); these pregnancies occurred from 1987 to 2005.
We generated categories of subclinical hypothyroidism and hyperthyroidism based on the measured thyroid hormone levels, after excluding participants who reported any thy roid disease and/or medication. Subclinical hypo/hyperthyroidism cutoffs are preferably based on normal laboratory reference ranges of TSH and FT 4 , but in the absence of FT 4 we used TSH and TT 4 , as reported previ ously (Wu et al. 2006). Based on hormone levels, we classified children in the category of subclinical hypothyroidism if the TSH value was above the upper bound of the normal reference range given for the labora tory (which varies according to age groups, i.e., TSH > 5.97, > 4.84, > 4.5 μIU/mL in children < 6, 6-10, > 10 years of age, respec tively), whereas TT 4 was within its reference range (4.5-12 μg/dL). The number of children with high TSH (> 10 μIU/mL) and low TT 4 (< 4.5 μg/dL) was too small (n = 8) for sepa rate analyses, and these children were included in the category of subclinical hypo thyroidism. We classified children in the subclinical hyper thyroidism category if the TSH value was below the lower bound of the normal refer ence range given by the laboratory (which var ies according to age groups, i.e., TSH < 0.7, < 0.6, < 0.45 μIU/mL in children < 6, 6-10, > 10 years of age, respectively) and TT 4 within the normal reference range (4.5-12 μg/dL). Children with low TSH (< 0.1 μIU/mL) and high TT 4 (> 12 μg/dL) (n = 4) were included in the subclinical hyperthyroidism category.
Self-reported thyroid disease and medication. Parents or legal guardians completed a questionnaire, including information on diag noses for thyroid disease. Respondents were asked whether they had ever been told by a healthcare provider that the child had thyroid disease. If the answer was yes, they were asked to select between one of these types of thy roid diseases: goiter, Hashimoto's thyroiditis, Graves disease, or others. In the last category, they were asked to provide the type of thyroid disease (most of whom noted hypothyroid ism). We considered three classifications of thyroid diseases: a) reported diagnosis with any thyroid disease; b) reported Hashimoto's thyroiditis or hypothyroidism; and c) a nar rower selfreported thyroid disease definition formed by combining report of any type of thyroid disease diagnosis with reported current use of one of the following medications com monly used to treat thyroid disease: Armour thyroid, Levothroid, Levothyroxine, Levoxil, Methimazole, or Synthroid.
In models of in utero exposure, for a sub sample, we also had information on newborn's birth weight (grams) and gestational age (weeks) and maternal weight gain (pounds), volume 120 | number 7 | July 2012 • Environmental Health Perspectives smoking habit (yes or no), and alcohol con sumption (yes or no) during pregnancy.
Statistical analyses. We conducted a regression analysis among participants 1-17 years of age at survey to assess the relation ship between thyroid function and modeled in utero PFOA concentrations or measured serum PFOA/PFOS/PFNA concentrations in samples collected during [2005][2006]. Levels of TSH and PFAAs showed a nonnormal distribution and were natural log-transformed before inclusion in the models. We used sim ple Pearson correlations to describe pairwise relationships between thyroid hormones and also between PFAAs.
We ran linear regression analyses after exclusion of individuals with reported thyroid disease and/or thyroid medication, to calcu late the regression coefficient (beta) and 95% confidence intervals (CIs) for thyroid hor mone levels and PFAA quartiles or ln(PFAA) concentrations (the later stratified by sex and age groups). Adjusted differences in thyroid hormone levels between quartile groups of PFAA exposure were expressed as percentages relative to the lowest exposure quartile, calcu lated as the complement of the exponentiated regression coefficient {100 × [exp(beta)-1]} for TSH and the ratio of beta to the mean [(beta/mean) × 100], for TT 4 . After fitting models of ln(PFAA) on ln(TSH) or TT 4 , regression coefficients were transformed to represent percent change in TSH or TT 4 associated with the IQR-the 75th compared with the 25th percentile for each PFAA expo sure estimate. IQRs were calculated for each sex/age group. For TSH, this is the exponen tiated value of the product of the coefficient for the interquartile difference in ln(PFAA). For nonlogtransformed TT 4 , we estimated the absolute change in TT 4 associated with one IQR of ln(PFAA) as the coefficient times the IQR and expressed this as a percent of the mean TT 4 . We also fit linear regression models including other PFAAs. In addition, we performed a sensitivity analysis including children with untreated thyroid diseases (i.e., without reported thyroid medication use), but because no major differences were found, we did not present these results.
We ran logistic regression models to cal culate ORs and 95% CIs for three categories of reported disease (any thyroid disease, hypo thyroidism, and thyroid disease plus medica tion) and for subclinical hypothyroidism or hyperthyroidism (based on measured thyroid hormone levels) in association with IQR shifts in PFAA concentrations. Finally, we assessed modeled in utero and measured serum PFOA in the same models.
We adjusted final models by child age and sex (when not stratified by this variable) and month of sampling [because there was a trend in measured PFAA during the collection year as well as seasonal variations in thyroid hormone levels (Maes et al. 1997)]. No other variables considered met our operational defi nition of confounder because there was < 10% change in the PFAA coefficients when includ ing or excluding them from the final regression models. These included maternal (age, weight gain, smoking habit, and alcohol consumption during pregnancy) and child (birth weight, gestational age, BMI, average household family income, race/ethnicity, and smoking habit and alcohol intake) variables. For variables with missing values (Table 1), the above criterion was applied for the subsample of participants without missing values. We used the statisti cal software package STATA for all statistical analyses (STATA Statistical Software, release 12; StataCorp, College Station, TX, USA). Where associations are referred to as statistically significant, this implies a pvalue of < 0.05. Table 1 shows the characteristics of the study population. A slight majority of participants were boys (52%), and the mean age in the population was 11.4 years. Most of the popu lation (97.4%) was white, whereas other reported race/ethnicity groups were black (1.2%), Hispanic (0.2%), Asian (0.1%), American Indian (0.2%), and other (0.9%). Of the 10,725 children 1-17 years of age in this study, 61 individuals (0.6%) reported a diagnosis of thyroid disease and 39 of the 61 reported a diagnosis of hypothyroidism (including Hashimoto's thyroiditis). A total  of 53 children who reported thyroid disease were > 10 years old, and 46 out of 61 were girls. Using the stricter definition of reported thyroid disease diagnosis plus use of thyroid medication, 0.4% of children (n = 40) were counted as cases. Table 2 shows the levels of thyroid hor mones and PFAAs in children. The distribu tion of TT 4 levels was close to normality with mean and median values of 7.5 and 7.4 μg/dL, respectively. TSH mean and median values were 2.1 and 1.8 μIU/mL. TSH was negatively correlated with TT 4 (r = -0.07, p < 0.001).
Associations between quartiles of PFAAs and ln(TSH) and TT 4 are shown in Table 3, where the lowest quartile is the reference exposure category. Associations between IQR contrasts in ln(PFAAs) are shown in Table 4. There was little evidence for an association of PFOA with either ln(TSH) or TT 4 in children 1-17 years of age (Tables 3 and 4). However, an IQR contrast of 10 to 57 ng/mL for modeled in utero PFOA was associated with a 2% increase in TT 4 in children up to 5 years of age (95% CI: 0.1, 3.9), with similar but less precise estimates for boys and girls separately in this age group. A change in mea sured serum PFOA from 16 to 83 ng/mL was associated with a 4% drop in TSH in all chil dren ≤ 5 years, but the association appeared to be limited to girls (Table 4). In children ≤ 5 years, associations between measured serum PFOA and TSH remained significant after adjusting for PFOS (-5.7% change; 95% CI: -9.8, -1.4) and PFNA (-4.7% change; 95% CI: -8.9, -0.4).
Associations between PFOS or PFNA and TT 4 were found in children 1-17 years of age (Tables 3 and 4). Interquartile contrasts of 15 to 28 ng/mL in PFOS and 1.2 to 2.0 ng/mL in PFNA were both associated with a 1.1% increase in TT 4 (95% CIs: 0.6, 1.5 and 0.7, 1.5, respectively). The association was evident in both girls and boys 10-17 years of age in the case of PFOS and for PFNA associations overall were significant for both boys and girls (Table 4). In addition, associations between PFOS or PFNA and TT 4 were similar after adjustment by other PFAAs percent change for PFOS adjusted by PFNA: 0.7, 95% CI: 0.2, 1.2; or PFOA: 1.1, 95% CI: 0.6, 1.6; percent change for PFNA adjusted by PFOS: TSH was natural log-transformed. Adjusted differences in thyroid hormone levels between quartile (Q) groups of PFAA exposure were expressed as percentages relative to the lowest exposure quartile, calculated from the exponentiated regression coefficient for TSH and from the ratio of beta to the mean for TT 4 .  PFAAs and TSH were natural log-transformed. Change in TSH calculated from the exponentiated regression coefficient and change in TT 4 as a percent of the mean. a Number of children with PFAA measurements after excluding those who reported thyroid disease and/or thyroid medication (also excluded from models): boys, n = 5,526; girls, n = 5,199. Models adjusted by sex b , age b,c , and month of sampling b,c .
volume 120 | number 7 | July 2012 • Environmental Health Perspectives 0.8, 95% CI: 0.4, 1.3; or PFOA: 1.1, 95% CI: 0.6, 1.5). We did not find evidence of associations between serum PFOS or PFNA and TSH. The OR for thyroid disease (n = 61) was 1.44 (95% CI: 1.02, 2.03) ( Table 5). Most of the children with thyroid disease were reported to have hypothyroidism (n = 39; OR = 1.54; 95% CI: 1.00, 2.37). The asso ciation was similar for the strictest definition of reported thyroid disease plus use of thy roid medication (OR = 1.61; 95% CI: 1.07, 2.51). Associations were similar for modeled estimates of in utero PFOA concentrations, although CIs were a little wider (Table 5). We also performed analyses including both modeled in utero and measured PFOA in the same model, and in both cases ORs were attenuated and CIs were wider (OR = 1.29; 95% CI: 0.87, 1.92; and OR = 1.27; 95% CI: 0.74, 2.19 for thyroid disease vs. PFOA modeled in utero or at survey, respectively). We did not find associations between concen trations of any of the PFAAs and subclinical hypo/hyperthyroidism (Table 5).

Discussion
To the best of our knowledge, this is the first largescale report of an association between PFAAs in serum and thyroid function impair ment in children 1-17 years of age. In a group of 10,725 children from the MidOhio Valley, measured serum PFOA concentra tions in 2005-2006 were positively associated with thyroid disease (mostly hypothyroidism). In addition, serum concentrations of PFOS and PFNA, but not PFOA, were positively associated with TT 4 levels in children 1-17 years of age.
Serum PFOS and PFNA concentrations were associated with slightly higher levels of TT 4 in children 1-17 years of age, but were not positively associated with subclinical hyperthyroidism. PFOA was not associated with TSH or TT 4 in all children combined, though subgroup analyses suggested possible associations with PFOA measured at survey and modeled in utero, in children ≤ 5 years. However, given the lack of effect in the other age groups, this may be a chance finding. Previous literature on child thyroid func tion and PFAA exposure is limited. A recent study in Korea (n = 29) reported a positive association between maternal serum PFOA (median = 1.46 ng/mL) concentrations and cord serum TSH levels. No association was found between PFOA, PFOS, or PFNA and T 4 (Kim et al. 2011). Another small study of 15 Japanese mother-child pairs reported no apparent correlation between maternal blood PFOS (median = 8.1 ng/mL) concen trations and neonatal blood TSH or FT 4 (Inoue et al. 2004). In agreement with our results, a recently published study of 52,296 adults from the C8 Health Project found a positive association between serum PFOS and TT 4 . They also found a positive association between serum PFOA and TT 4 in women of all ages, and men > 50 years of age (Knox et al. 2011). In addition, a positive associ ation was reported between plasma PFOS (GM = 18.3 ng/mL) and FT 4 levels and inverse with TSH in environmentally exposed adults (n = 621) from Nunavik, Quebec, Canada (Dallaire et al. 2009). On the con trary, a modest inverse association between serum PFOA (median = 1.1 μg/mL) and FT 4 but not TT 4 or TSH levels was reported in an occupational study among 506 employees in one Belgian and two American factories (Olsen and Zobel 2007). Some adult studies have not found associations between PFAA exposures and thyroid level alteration, includ ing a study of Canadian pregnant women with hypothyroxinemia (n = 96; PFOA and PFOS medians = 3.9 and 15.5 nmol/L) whose serum PFAA concentrations were com parable to matched controls (n = 175; PFOA and PFOS medians = 3.6 and 16.4 nmol/L) (Chan et al. 2011). There was no evidence of an association between elevated serum PFOA (median = 354 ng/mL) and TSH lev els in residents (n = 371) from the same area as the present study (Emmett et al. 2006). In a small study (n = 31) of anglers from New York State, no association was reported between PFOA, PFOS, or PFNA concentra tions (GM = 1.3, 19.6, and 0.79 ng/mL) and serum TSH or FT 4 levels (Bloom et al. 2010).
Altered thyroid hormone levels follow ing exposure to PFAAs have also been found in experimental animal studies where pre and postnatal longterm exposure to PFOS decreased serum levels of TT 4 and FT 4 in pregnant dams and pups, without a concomi tant rise in TSH (Lau et al. 2007). PFNA exposure did not alter TT 4 levels in zebrafish juveniles treated with PFNA until maturity (Liu et al. 2011). In adult monkeys, PFOA exposure during 6 months, which led to serum levels up to 158 μg/mL, did not alter TT 4 or TSH (Butenhoff et al. 2002).
We found higher odds of reporting thy roid disease (mostly hypothyroidism) with increased measured PFOA concentrations, even when we restricted cases to children with reported disease plus use of thyroid medica tion. However, PFOA concentrations were not associated with subclinical hypo/hyper thyroidism based on individual hormone levels or thyroid hormone levels (as continu ous variables) in children 1-17 years of age. Therefore, the association between reported thyroid disease and PFOA exposure should be considered with caution. Nevertheless, these results are comparable to a cross sectional analysis of PFOA/PFOS concentra tions and reported thyroid disease in adults in NHANES for 1999NHANES for -2000NHANES for , 2003NHANES for -2004NHANES for , and 2005NHANES for -2006). An OR of 2.2 (95% CI: 1.4, 3.7) was estimated for thyroid disease in association with the highest versus first and second quartiles of serum PFOA in females (mean = 3.77 ng/mL), and a similar associa tion with serum PFOS (mean = 25.1 ng/mL) was reported for males (OR = 2.7; 95% CI: 1.0, 7.0) (Melzer et al. 2010).
The main strengths of the present study are the large sample size and the data for both measured and modeled in utero PFOA con centrations. A further strength is the high rate of participation, diminishing concern about potential selection biases. It is believed that this population is representative, given the high participation rates in the C8 Health Project, of all those children who drank con taminated water in the MidOhio Valley. Moreover, we were able to adjust for a num ber of potential confounders.
The mostly crosssectional design of the present study is a major limitation because the single measurements preclude determination of the time sequence between PFAA exposure and outcome, and some associations may have been attributable to chance or uncontrolled sources of bias. One further limitation is the absence of measurements of child triiodothyronine and FT 4 levels, which would have yielded more comprehensive information concerning the PFAAs were natural log-transformed. Models were adjusted by age and sex. a Based on hormonal levels at survey after excluding people who self-reported thyroid disease and/or thyroid medication.
child's thyroid regulatory system. Another limitation is the reliance on recall for thyroid diagnosis, although we also investigated a more stringent case definition of report plus medica tion use and obtained similar results. However, the prevalence of hypothyroidism reported in our study population (0.4%) was higher than the range (0.04-0.14%) of the estimated preva lence in several children and adolescent popula tions (Hunter et al. 2000). Based on TSH and TT 4 levels, the prevalence of subclinical hypo thyroidism in the 1988-1994 NHANES pop ulation (Wu et al. 2006) was lower than that of the present study (1.7% vs. 2.9%, respectively, in children > 12 years of age), whereas subclini cal hyperthyroidism was more prevalent (2.3% vs. 1.2% in children > 12 years).

Conclusions
In summary, this is the first largescale report on thyroid hormone function and PFAA concentrations among children 1-17 years of age. Results suggest that serum PFOS and PFNA concentrations are associated with thyroid hormone levels, and serum PFOA concentrations are associated with reported hypothyroidism. However, further studies to understand the effect of pre or postnatal exposure to PFAAs on thyroid hormones in children are warranted.