Serum Perfluorooctanoate (PFOA) and Perfluorooctane Sulfonate (PFOS) Concentrations and Liver Function Biomarkers in a Population with Elevated PFOA Exposure

Background: Perfluorooctanoate (PFOA) and perfluorooctane sulfonate (PFOS) persist in the environment and are found in relatively high concentrations in animal livers. Studies in humans have reported inconsistent associations between PFOA and liver enzymes. Objectives: We examined the cross-sectional association between serum PFOA and PFOS concentrations with markers of liver function in adults. Methods: The C8 Health Project collected data on 69,030 persons; of these, a total of 47,092 adults were included in the present analysis. Linear regression models were fitted for natural log (ln)-transformed values of alanine transaminase (ALT), γ-glutamyltransferase (GGT), and direct bilirubin on PFOA, PFOS, and potential confounders. Logistic regression models were fitted comparing deciles of PFOA or PFOS in relation to high biomarker levels. A multilevel analysis comparing the evidence for association of PFOA with liver function at the individual level within water districts to that at the population level between water districts was also performed. Results: ln-PFOA and ln-PFOS were associated with ln-ALT in linear regression models [PFOA: coefficient, 0.022; 95% confidence interval (CI): 0.018, 0.025; PFOS: coefficient, 0.020; 95% CI: 0.014, 0.026] and with raised ALT in logistic regression models [with a steady increase in the odds ratio (OR) estimates across deciles of PFOA and PFOS; PFOA: OR = 1.10; 95% CI: 1.07, 1.13; PFOS: OR = 1.13; 95% CI: 1.07, 1.18]. There was less consistent evidence of an association of PFOA and GGT or bilirubin. The relationship with bilirubin appears to rise at low levels of PFOA and to fall again at higher levels. Conclusions: These results show a positive association between PFOA and PFOS concentrations and serum ALT level, a marker of hepatocellular damage.

Perfluorooctanoate (PFOA) and perfluoro octane sulfonate (PFOS) are two members of the perfluoroalkyl acid (PFAA) class of chemicals, manmade compounds used in the manufacture of fluoro polymers, including those used for non stick cookware and breath able, waterproof fabrics. PFOA and PFOS can also result from the metabolism of fluori nated telomers, compounds used for food package coatings, carpet treatments, and stainresistant fabric treatment. PFOA and PFOS persist in the environment (Kato et al. 2011;Tao et al. 2006); potential sources of exposure to PFOA and PFOS in humans include drinking water, dust, breast milk, food packaging, ambient air, and occupation (D'Hollander et al. 2010;Goosey and Harrad 2011;Kim et al. 2011;Lau et al. 2007).
In rodents and non human primates, both PFOA and PFOS have been found in relatively high concentrations in the liver and have been associated with liver enlargement. In rats, these compounds have been also associated with hepatocellular adenomas (Lau et al. 2007). In mice, one of the biological effects of PFAAs is the activation of the peroxisome proliferator activated receptorα (PPARα), a ligand activated transcription factor that regulates gene expression, lipid modulation, glucose homeo stasis, cell proliferation, and inflammation (Pyper et al. 2010). Although some effects in experimental studies are mediated by PPARα binding, some other effects occur indepen dently of this receptor (Bjork et al. 2011;Ren et al. 2009). PPARα is also induced by PFOA and PFOS in transiently transfected human fibroblastlike cell line COS1, in a concentra tiondependent and in a roughly chainlengthdependent fashion (Wolf et al. 2008).
Studies in humans have reported incon sistent associations between PFOA or PFOS and liver enzymes. Transaminase levels have been positively associated with PFOA con centrations in some occupational studies (Costa et al. 2009;Olsen and Zobel 2007;Sakr et al. 2007b) but not in others (Costa et al. 2009;Sakr et al. 2007b). Similarly, γglutamyltransferase (GGT) levels have been inconsistently associated with PFOA concen trations in occupational studies (Costa et al. 2009;Olsen and Zobel 2007;Sakr et al. 2007aSakr et al. , 2007b. In a large populationbased survey, PFOA but not PFOS was associated with both transaminase and GGT levels (Lin et al. 2010), although those findings differed from those of an earlier, much smaller popula tionbased study of subjects heavily exposed to PFOA (Emmett et al. 2006). Direct bilirubin has been found to be negatively associated with PFOA concentrations in a few occupational studies (Costa et al. 2009;Olsen and Zobel 2007;Sakr et al. 2007b) but not in others (Costa et al. 2009;Sakr et al. 2007a). No asso ciation between direct bilirubin and PFOA or PFOS was observed in populationbased stud ies (Emmett et al. 2006;Lin et al. 2010).
From 1950 through 2005, a chemical plant in the MidOhio Valley, West Virginia (USA), was responsible for emitting PFOA into the surrounding environment. In 2001, a group of residents filed a class action lawsuit alleging Background: Perfluorooctanoate (PFOA) and perfluorooctane sulfonate (PFOS) persist in the environment and are found in relatively high concentrations in animal livers. Studies in humans have reported inconsistent associations between PFOA and liver enzymes. oBjectives: We examined the cross-sectional association between serum PFOA and PFOS concentrations with markers of liver function in adults. Methods: The C8 Health Project collected data on 69,030 persons; of these, a total of 47,092 adults were included in the present analysis. Linear regression models were fitted for natural log (ln)-transformed values of alanine transaminase (ALT), γ-glutamyltransferase (GGT), and direct bilirubin on PFOA, PFOS, and potential confounders. Logistic regression models were fitted comparing deciles of PFOA or PFOS in relation to high biomarker levels. A multilevel analysis comparing the evidence for association of PFOA with liver function at the individual level within water districts to that at the population level between water districts was also performed. results: ln-PFOA and ln-PFOS were associated with ln-ALT in linear regression models [PFOA: coefficient, 0.022; 95% confidence interval (CI): 0.018, 0.025; PFOS: coefficient, 0.020; 95% CI: 0.014, 0.026] and with raised ALT in logistic regression models [with a steady increase in the odds ratio (OR) estimates across deciles of PFOA and PFOS; PFOA: OR = 1.10; 95% CI: 1.07, 1.13; PFOS: OR = 1.13; 95% CI: 1.07, 1.18]. There was less consistent evidence of an association of PFOA and GGT or bilirubin. The relationship with bilirubin appears to rise at low levels of PFOA and to fall again at higher levels. conclusions: These results show a positive association between PFOA and PFOS concentrations and serum ALT level, a marker of hepatocellular damage. key words: C8, cross-sectional study, liver function biomarkers, PFOA, PFOS, population-based survey.  (Frisbee et al. 2009). Part of the pre trial settlement of the class action lawsuit included a baseline survey, the C8 Health Project, con ducted in [2005][2006], that gathered data from > 69,000 persons from six contaminated water districts surrounding the plant (Frisbee et al. 2009). In this population, overall PFOA lev els were much higher [mean, 83.0 ng/mL; interquartile range (IQR), 13.4-70.6 ng/mL] than in corresponding U.S. population surveys (National Health and Nutrition Examination Survey in the same year: mean, 3.9 ng/mL; IQR, 2.7-5.8 ng/mL) (Frisbee et al. 2009;Kato et al. 2011). However, the mean PFOS (23.3 ng/mL; IQR, 13.8-29.0 ng/mL) closely resembled the U.S. population mean (20.7 ng/mL; IQR, 14.6-29.9 ng/mL) (Kato et al. 2011). The present study used these data to examine the crosssectional association between serum PFOA or PFOS concentrations and markers of liver function in adults.

Methods
The study population. This study was approved by the London School of Hygiene and Tropical Medicine Ethics Committee and is one of the C8 Science Panel studies and used information from questionnaires and blood tests collected in the C8 Health Project, supplemented by further information on classification by water district developed in a companion C8 Science Panel study (Shin et al. 2011).
The C8 Health Project enrolled eligible subjects between August 2005 and August 2006. Individuals were eligible to partici pate if they had consumed water for at least 1 year between 1950 and 2004 while living, working, or going to school in one of the six water districts, or private water sources, or areas of documented PFOA contamina tion. The between and withingroup regression analysis was restricted to subjects living in one of the six contaminated water districts at the time of survey [for additional details on water districts, see Supplemental Material (http:// dx.doi.org/10.1289/ehp.1104436)]. Details of the study enrollment process, including consenting procedures, have been described elsewhere (Frisbee et al. 2009).
The C8 Health Project collected data on 69,030 persons. Its participation rate, based on U.S. census numbers, has been estimated at around 80% (Frisbee et al. 2009). In this population, the strongest predictor of PFOA serum concentration was residence in one of the contaminated water districts (Steenland et al. 2009), whereas serum levels of other PFAAs did not show such geographic variation. Of the population, 56,554 adults (≥ 18 years of age) were considered for this analysis, and a total of 46,452 of those adults (82.1%) were included in the final analysis after exclusion of subjects with missing data on socioeconomic status, alcohol consumption, or cigarette smoking or other potential confounding variables or without PFAAs or liver enzymes measurements.
Choice of parameters and laboratory analyses. Blood samples were obtained and processed at individual data collection sites. Samples were drawn into four tubes per participant, with a maximum of 35 mL blood collected. Samples were centrifuged, aliquoted, and refrigerated until shipping. Processed samples were shipped on dry ice daily from each data collection site to the laboratory (Frisbee et al. 2009). Participants were not asked to fast before blood sample withdrawal, but fasting status was recorded.
Laboratory analyses of PFAAs were con ducted by the Exygen Research Inc. (State College, PA, USA). using an automated solid phase extraction combined with reversephase highperformance liquid chromatography/ mass spectrometry (Kuklenyik et al. 2004). An intralaboratory quality assurance program was carried out by analysis of duplicate samples at AXYS Analytical Service Ltd. (Sidney, BC, Canada) (Frisbee et al. 2009). The intralabora tory coefficient of variation for both PFOA and PFOS measurements was 0.1; the inter laboratory comparison coefficient of varia tion was 0.2 for PFOA and 0.1 for PFOS (Frisbee et al. 2009). The detection limit was 0.5 ng/mL for both PFOA and PFOS, and observations below this limit were assigned a value of 0.25 ng/mL (for this study popula tion, n = 32 for PFOA, n = 230 for PFOS). Both PFOA and PFOS concentration distri butions were skewed to the right. The liver parameters we measured were alanine aminostransferase (ALT) and aspartate aminostransferase (AST), GGT, alkaline phos phatase (ALP), and direct bilirubin (also known as "conjugated bilirubin"). Both transaminases (AST and ALT) are enzymes released after liver parenchymal cell injury and are elevated in serum during acute liver damage; the correla tion between ALT and AST in the present population is r = 0.79. To limit multiple com parisons and to be consistent with the most recent published literature on the same topic (Lin et al. 2010), we restricted our analysis to ALT, GGT, and direct bilirubin as markers of liver function. Elevated ALT has been used as a proxy for hepatocellular injury in previous studies (Clark et al. 2003;Ioannou et al. 2005;Lin et al. 2010;Ruhl and Everhart 2003) because it is more specific for hepatic damage than is AST. Elevation of GGT occurs at an early stage and is more persistent than that of ALP in cholestatic disorders (Whitfield 2001); the correlation between GGT and ALP in the present population was r = 0.29. Bilirubin is mostly derived from the metabolism of hemo globin; the increase in serum of the direct bili rubin component is highly specific for liver or bile duct disease (Sedlak and Snyder 2004).
ALT, GGT, and direct bilirubin were measured using a Roche/Hitachi MODULAR automated analyzer (Roche Diagnostics, Indianapolis, IN, USA), and the analyses were performed at a large, independent, accredited clinical diagnostic laboratory (LabCorp, Inc., Burlington, NC, USA) (Frisbee et al. 2009). The homeostasis model assessment of insulin resistance (HOMAIR) index was calculated as the product of basal glucose and insulin levels divided by 2.25; it is used as a surrogate measure for insulin resistance (Matthews et al. 1985).
Methods and results are described accord ing to the STrengthening the Reporting of OBservational studies in Epidemiology-Molecular Epidemiology (STROBEME) guidelines (Gallo et al. 2011).
The association between ALT, GGT, or direct bilirubin and PFOA or PFOS was assessed using linear regression models. First, we fitted a simple model including only age and sex (model 1), followed by a model additionally including alcohol consumption, socioeconomic status, fasting status, race, and month of blood sample collection (model 2), and then a model additionally including smoking status, BMI, physical activity, and insulin resistance (i.e., HOMAIR) in deciles of distribution (model 3).
Also, specific cutoff values were used to fit logistic regression models to estimate the impact of PFOA and PFOS on being above these values. Cutoff values used were 45 IU/L in men and 34 IU/L in women for ALT (Schumann et al. 2002a), giving a total of 5,194 persons (11.2% of the 47,092 eligible participants) with abovenormal values; 55 IU/L in men and 38 IU/L in women for GGT (Schumann et al. 2002b), giving 5,990 persons (12.9%) with abovenormal values; and 0.3 mg/dL in men and women for direct bilirubin (Kratz et al. 2004), giving 506 persons (1.1%) with abovenormal values.
Prior research showed that individual serum levels of PFOA were strongly associated with the water district of residence (Steenland et al. 2009). This geographic clustering implies a potential for an ecological confounding by other uncontrolled and/or unobserved water district-specific factors. On the other hand, there is also potential for confounding at the individual level. We expect a priori some heterogeneity in between and withingroup relationships; therefore, we applied an ana lytic approach in the multilevel framework [between and withingroup regression (Davis et al. 1961)] to disentangle between and withindistrict effects of PFOA. In statisti cal modeling, this is realized by simultane ously incorporating both individual PFOA serum concentrations and the means of PFOA within water districts. Details of this analy sis are reported in the Supplemental Material (http://dx.doi.org/10.1289/ehp.1104436).

Results
General characteristics of the population are summarized in Table 1. Women had sig nificantly lower values of liver function bio markers and of PFOA and PFOS. PFOA and PFOS concentrations were associated with all potential confounders considered.
lnTransformed values of ALT were signifi cantly associated with lnPFOA and lnPFOS in linear regression models [fully adjusted (model 3) coefficient: PFOA, 0.022; 95% confidence interval (CI): 0.018, 0.025; PFOS, 0.020; 95% CI: 0.014, 0.026], with a partial R 2 greater for the association with PFOA (partial R 2 = 0.002) than for that with PFOS (partial R 2 < 0.001; Table 2). In Figure 1, mean ALT levels are plot ted against deciles of PFOA and PFOS concen trations, adjusting for covariates by setting these to their means. A steady increase in fitted levels of ALT per decile of PFOA is shown, with a possible leveling off effect after approximately  Figure 1A,B). This positive asso ciation was also observed in logistic regression models with a steady increase in the odds ratio (OR) estimates across deciles of both PFOA and PFOS (pvalue for trends across deciles in both models < 0.001), and a significant OR for both lnunit of PFOA (OR = 1.10; 95% CI: 1.07, 1.13) and lnunit of PFOS (OR = 1.13; 95% CI: 1.07, 1.18; The association between lnGGT and lnPFOA reached a significant level in the fully adjusted model (model 3; coefficient, 0.015; 95% CI: 0.010, 0.019) mainly due to the con tribution of insulin resistance and BMI, which appeared to be highly associated with lnGGT ( Table 2); fitted values of GGT by deciles of PFOA showed an apparent positive associa tion ( Figure 1C), although it was less clear than that shown for ALT. The suggested association with PFOA, however, was not confirmed in the logistic regression model, in which no trend across deciles was observed (p = 0.213), or for the linear lnunits of PFOA values (OR = 1.01; 95% CI: 0.99, 1.04; Table 3). For PFOS, there was some evidence of a slight inverse associa tion with GGT in the minimally adjusted lin ear regression model (model 1), which was lost after adjusting for additional confound ers (model 3; Table 2). The logistic regression by deciles of PFOS suggested a weak negative trend across deciles, although all ORs were close to 1, and no overall association with lnPFOS was observed (Table 3). Overall, fitted values of GGT were unchanged across deciles of PFOS ( Figure 1D). Subjects with abnormally high GGT values were more frequent in one district (chisquare test, p = 0.062) and had significantly higher PFOA serum concentrations (93.7 vs. 86.1 IU/L, p = 0.004) but not PFOS serum concentrations (23.5 vs. 23.4 IU/L, p = 0.499).
For direct bilirubin, there was a suggestion of an inverse Ushaped relationship with PFOA, with increasing levels of bilirubin per increas ing levels of PFOA at low PFOA levels, and decreasing bilirubin levels for concentrations of PFOA above about 40 ng/mL ( Figure 1E). Overall, the linear regression relationship failed  to show any association in the adjusted model (Table 2), and the likelihood ratio test after introducing the quadratic term was statistically significant (p < 0.001). In accordance, no pat tern was evident in the logistic regression mod els of high bilirubin (Table 3). By contrast, for PFOS, a clear positive association was shown in linear regression models for direct biliru bin (coefficient, 0.029; 95% CI: 0.024, 0.034; Table 2, Figure 1F), although this was not evi dent in logistic regression models (Table 3). However, logistic regression results for direct bilirubin should be interpreted cautiously because only 1.1% of the whole sample had high levels and CIs were wide. Multilevel analysis was restricted to subjects living in water districts supplied by contami nated water (n = 26,777) and excluding those with private wells. The fitted values for each of the outcomes of ALT, GGT, and direct bili rubin versus mean PFOA serum level for the six water districts are graphed in Figure 2. In Table 4, the regression coefficients per lnunit of PFOA for a model between water districts comparing district averages of lnPFOA (B) and for a model within water districts com paring differences from the average (W) are presented. There was a significant difference between the between and withindistrict com ponents (pvalue B/WΔ) for ALT and direct bilirubin; however, each outcome showed dif ferent patterns. The between waterdistrict regression coefficient from linear regression of lnPFOA and ALT (0.010; 95% CI: -0.001, 0.020) was lower than the withinwaterdistrict coefficient (0.027; 95% CI: 0.022, 0.031). However, both coefficients were significant or borderline significant, in the same direc tion, and consistent with a positive associa tion between ALT and PFOA levels. Although the regression coefficient for the association between GGT and lnPFOA within water dis trict was positive and significant (0.016; 95% CI: 0.010, 0.023), this association was not sig nificant in the betweenwaterdistrict analysis (0.005; 95% CI: -0.009, 0.018); the pvalue for interaction between/within water district was not significant (p = 0.108). Conversely, there was a significant inverse relationship between lnPFOA and direct bilirubin between water districts (-0.013; 95% CI: -0.022, -0.005) but not within water districts (0.0001; 95% CI: -0.004, 0.004).

Discussion
These results show a positive association between PFOA and PFOS concentrations and ALT serum levels, a marker of hepatocellular damage. The linear association, consistently replicated in all analyses, showed a monotonic increase in logistic regression. Furthermore, the presence of a consistent relationship in regres sion analysis both between water districts and among individuals within districts increases strength of evidence for causal association. The proportion of laboratory abnormal values rises in relation to PFOA, but the small amount of variation in outcomes explained by PFOA (partial R 2 ≤ 0.1%) suggests caution.
The observed associations between PFOA or PFOS and ALT are generally consistent in terms of direction and magnitude with previ ous findings of occupational studies. A cross sectional study of 1,025 workers at the same plant leading to the C8 Health Project reported an association between PFOA and ALT levels with a similar coefficient (mean ± SE coeffi cient of logtransformed values, 0.023 ± 0.015; p = 0.124). A significant linear regression coef ficient between logtransformed-PFOA and logALT (coefficient, 0.025 ± 0.013; p = 0.006) was described in a crosssectional occupational surveillance conducted in three 3M plants in the United States and Belgium (Olsen and Zobel 2007). A small occupational study con ducted in Italy showed a significant positive association between PFOA and ALT levels among workers exposed to PFOA (coefficient, 0.116; 95% CI: 0.054, 0.177; p < 0.01) (Costa et al. 2009). Notably, these results are also consistent with data coming from a general population survey (mean ± SE linear regression coefficients with ALT: PFOA, 1.86 ± 0.62, p = 0.005; PFOS, 1.01 ± 0.53, p = 0.066) where background concentrations of PFOA are much lower than those reported here (Lin et al. 2010). Only one previous study of this same population reported no association with ALT (regression coefficient = -0.00416, p = 0.65); however, the observation was based on a much smaller sample (most of whom were likely to be included in the present analysis) and based on figures coming from models including non logtransformed exposure or outcome variables (Emmett et al. 2006).
Evidence for an association between PFOA and PFOS and GGT is not so clear. Although there was some suggestion of an association in the linear regression models with PFOA, it was not replicated in logistic regression models. The instability of linear regression coefficients (which are very sensitive to the inclusion of additional covariates) might be due to a con founding effect of diet, or residual confound ing of alcohol consumption, which causes a direct increase of GGT. Finally, the absence of any trend across districts-even though there are large contrasts in betweendistrict mean exposures-might be indicative of some con founding factor acting at the individual level. As shown recently, GGT levels are positively associated with alcohol and meat intake after adjusting for all potential non dietary confound ers (Lee et al. 2004). To support this, there is indirect evidence from an occupational study that reported a significant association between lnGGT and lnPFOA in models adjusted by lnage, lnBMI, and lnalcohol (mean ± SE coef ficient, 0.033 ± 0.017, p = 0.05), which drops if