Specimens were obtained from participants of the 2003–2004 cycle of the National Health and Nutrition Examination Survey (NHANES), a nationally representative survey of the noninstitutionalized civilian population of the United States. NHANES obtained a stratified, multistage probability cluster sample designed to represent the U.S. population on the basis of age, sex, and race/ethnicity. Certain subpopulations were oversampled to allow for more precise estimates. All participants in the survey gave written informed consent. Information on NHANES procedures may be found at the Centers for Disease Control and Prevention Web site (CDC 2009). The NHANES protocol for each survey period was reviewed and approved by the National Center for Health Statistics Institutional Review Board.

A total of 7,166 whole-blood samples from individuals ≥ 3 years of age were used in this study. Collection was performed by antecubital phlebotomy into EDTA-containing vacutainers and stored at −70°C before analysis. HbAA and HbGA were measured in 350 μL whole blood and analyzed by high-performance liquid chromatography/tandem mass spectrometry (HPLC/MS/MS), as described previously (Vesper et al. 2007, 2008). In brief, the adducts of acrylamide and glycidamide at the N-terminal valines of hemoglobin were cleaved from the protein chain by use of modified Edman reaction with pentafluorophenyl isothiocyanate as the Edman reagent (Mowrer et al. 1986). The resulting pentafluorophenylthiohydantoin reaction products were extracted by use of liquid–liquid extraction and analyzed by HPLC/MS/MS. Calibrators, reagent blanks, and quality-control materials were processed in the same manner as the samples. Hemoglobin adduct concentrations are reported relative to the amount of hemoglobin used in the analysis. The total hemoglobin content was determined as cyanomethhemoglobin by use of a commercial assay (Stanbio Laboratory, Boerne TX). The detection limits for HbAA and HbGA adducts were 3 and 4 pmol/g hemoglobin, respectively. The interday imprecision (n = 20 days) of this method, expressed as percentage coefficient of variation, were 13% for HbAA and 19% for HbGA, on average, determined with three blood pools [HbAA and HbGA concentrations in picomoles per gram hemoglobin: pool 1 (135 and 93); pool 2 (103 and 700); pool 3 (62 and 50), respectively. Calibrators were synthesized by Bachem (King of Prussia, PA) from octapeptides with the same amino acid sequence as the N-terminus of the β-chain of hemoglobin reacted with acrylamide and glycidamide at the valine (AA-octapeptide: 96% purity by HPLC/UV and MALDI (matrix-assisted laser desorption/ionization)/MS; and GA-octapeptide: 90% purity by HPLC/UV and MALDI/MS). Calibrators were verified for accuracy in-house by HPLC/MS and by the U.S. Food and Drug Administration (FDA)/National Center for Toxicology Research by use of HPLC/MS and that organization’s own standards. The corresponding stable isotope-labeled compounds [AA-octapeptide and GA-octapeptide with stable-isotope labeled valine (¹³C5¹⁵N)] were synthesized by the same company and used as internal standards. The Edman reagent was obtained from Fluka (St. Louis, MO), and formamide was obtained from USB Corporation (Cleveland, OH). All other reagents were purchased from Fisher Scientific (Fair Lawn, NJ).

HbAA, HbGA, and cotinine levels were log10-transformed for statistical analysis, because the data distribution is skewed to the higher levels. For selected age groups, sexes, and three race/ethnicity groups, we calculated the geometric means and distribution percentiles for HbAA and HbGA, the HbGA:HbAA-ratio, and the sum of HbAA and HbGA. Race/ethnicities were self-reported as non-Hispanic black, non-Hispanic white, and Mexican American. Individuals not included in one of these three race/ethnicity groups were included in the total population estimates. Geometric means and percentiles were calculated with SUDAAN version 9.0 (Research Triangle Institute, Research Triangle Park, NC). SUDAAN uses sample weights and calculates variance estimates that account for the complex survey design. We estimated 95% confidence intervals (CIs) for geometric means were estimated on the basis of the Taylor series linearization method (Shah et al. 1996), and adapted CIs for percentiles from the methods of Korn and Graubard (1998) and Woodruff (1952). For concentrations below the limit of detection (LOD), a level equal to the LOD divided by the square root of 2 was used (Hornung and Reed 1990). Because of the known effect of smoking on acrylamide levels, we divided the data into two groups according to cotinine levels (serum cotinine ≤ 10 ng/mL: nonsmoker group; serum cotinine >10 ng/mL: smoker group) (Pirkle et al. 2006).

We calculated covariate-adjusted geometric means for selected demographic groups separately for the nonsmoking and smoking populations, using least-squares multiple regression adjusted for the covariates of age group (3–11, 12–19, 20–39, 40–59, ≥ 60 years), sex (male and female), and self-reported race/ethnicity (non-Hispanic white, non-Hispanic black, and Mexican American), and the continuous covariates of body surface area (BSA) and the base 10 logarithm of serum cotinine. We used BSA to adjust for blood volume and thus to account for other possible variability affecting delivery of acrylamide to the blood related to body size measures and was calculated according to Mosteller (1987). To arrive at the final model, we used backward elimination with SUDAAN to eliminate the nonsignificant interactions (p > 0.05) one at a time and then the nonsignificant main effects. Nonsignificant main effects were kept in the model as confounders when they changed the beta coefficients for significant main effects or interactions by > 10%. Once the backward procedure was completed, main effects and interactions were added back into the model one at a time and kept in the final model when they were significant (p < 0.05). Hypothesis tests to compare covariate-adjusted geometric means were performed using the Contrast option in the Regress procedure of SUDAAN. For further details about the final model and observed interactions between the investigated variables, see Supplemental Material, Model Information, available online (doi:10.1289/ehp.0901021.S1 via http://dx.doi.org).

HbAA and HbGA levels were detectable in 99.9% and 97.5% of all samples, respectively, with individual levels ranging from < 3 to 910 pmol/g hemoglobin for HbAA and < 4 to 756 pmol/g hemoglobin for HbGA. Distributions were different in nonsmokers and smokers (Figure 1). Geometric mean HbAA levels were 49.1–72.5 pmol/g hemoglobin higher in smokers than in nonsmokers when compared across subgroups of sex, age, or race/ethnicity (Table 1); HbGA levels were 31.6–52.2 pmol/g hemoglobin higher. Ratios of HbGA:HbAA ranged from 0.08 to 0.25 units lower in smokers than in nonsmokers when compared across subgroups of sex, age, and race/ethnicity. The intersubject variability (difference between 90th and 10th percentile) for HbAA and HbGA in nonsmokers was 45 pmol/g hemoglobin and 66 pmol/g hemoglobin, respectively [Supplemental Material, Table 1 (doi:10.1289/ehp.0901021.S1)]. In smokers, it was much larger, with 179 pmol/g hemoglobin and 191 pmol/g Hb, for HbAA and HbGA, respectively (Supplemental Material, Table 2). The subgroups of sex, age, and race/ethnicity showed similar intersubject variability as the entire group of either smokers or nonsmokers (Supplemental Material, Figures 1 and 2).

Exposure to acrylamide, as indicated by the presence of HbAA or HbGA, was detectable in > 99% of all NHANES participants, indicating that most of the U.S. population is exposed to this chemical. However, the levels of HbAA or HbGA were not profoundly different among several population subgroups but differed highly among individuals, suggesting that individual exposures were affected mainly by factors other than age, sex, or race/ethnicity. In addition, the relationship of levels of HbAA to HbGA could be useful in studying factors related to the metabolic conversion of acrylamide to the more toxic glycidamide.

The HbAA levels in children were modestly higher than those in two older age groups, and the HbGA levels in children were higher than those in all four older age groups even after adjustment for body size, cotinine, and sex. The higher levels in children may be due to the known larger intake of food per body mass in children, particularly acrylamide-rich foods such as french fries and potato chips, as suggested elsewhere (Dybing et al. 2005; Mucci and Wilson 2008). In addition, children have the highest HbGA:HbAA ratios, whereas adults ≥ 60 years of age have the lowest ratios. Smokers show similar patterns, with biomarker values and ratios being higher at young age. This suggests differences in the acrylamide metabolism or metabolic rate in these age groups. The reasons for these different ratios are not fully understood. Metabolism and clearance of drugs are higher in children in part because of the larger liver-to-body weight ratio and the higher blood flow through the liver compared with older adults, in whom liver volume and liver blood flow are decreased (Greim and Snyder 2008). Further investigation is needed to assess possible health effects associated with acrylamide exposure in children.

The observed differences in acrylamide exposure between race/ethnic groups, with Mexican Americans having the highest HbAA and HbGA levels and non-Hispanic blacks having the lowest HbGA levels and HbGA:HbAA ratios in nonsmokers and smokers, have not been described before. The differences observed in nonsmokers could be explained by different food consumption patterns. The 2003–2004 U.S. Department of Agriculture (USDA) food consumption survey reported differences in food consumption patterns for total sugars and carbohydrate-rich food among non-Hispanic whites, non-Hispanic blacks, and Mexican Americans (USDA-ARS 2007), and this pattern appears similar to the pattern observed for HbAA and HbGA. Carbohydrates are precursors for acrylamide formation during food processing (Mottram et al. 2002; Stadler et al. 2002). Thus, elevated consumption of carbohydrate-rich food in Mexican Americans may explain in part the elevated HbAA and HbGA levels in this group. The significantly lower HbGA:HbAA ratios and HbGA formation rate (per unit increase of HbAA) in non-Hispanic blacks may indicate differences in polymorphisms of the genes for CYP2E1 or of the genes for glutathione S-transferase (GST), which are involved in phase II detoxification of acrylamide and glycidamide. Different frequencies in polymorphisms in the genes for CYP2E1 in different race/ethnicity groups have been mentioned in literature (Bartsch et al. 2000; Bolt et al. 2003), and associations between polymorphisms in the genes for GST and the HbGA:HbAA ratio have been reported (Duale et al. 2009). Medications and alcohol are also metabolized by CYP2E1 and thus may affect CYP2E1 activity and metabolism of acrylamide to glycidamide. These factors could not be assessed in this study. Therefore, further studies are needed to better explain these differences in the observed HbGA:HbAA ratio.

The ranges of HbAA and HbGA levels measured in this study are similar to those reported in other study populations (Bjellaas et al. 2007; Chevolleau et al. 2007; Hartmann et al. 2008; Heudorf et al. 2008; Kütting et al. 2008; Olesen et al. 2008; Vesper et al. 2007, 2008; Wilson et al. 2008; Wirfält et al. 2008). The mean and median levels reported in these study populations for nonsmokers differ, ranging between 19 pmol/g and 47 pmol/g for HbAA and between 17 pmol/g and 49 pmol/g for HbGA. These differences in mean and median levels across different studies may be explained by differences in the demographics of the populations, differences in analytical methods used in these studies, and different acrylamide exposures in the different study populations. One recent study, investigating acrylamide exposure in groups of adults 41–60 years of age from nine European countries and using the same analytical method, found median levels among nonsmoking groups ranging between 35 pmol/g and 65 pmol/g for HbAA and 27 pmol/g and 58 pmol/g for HbGA, with the Dutch and British groups having the highest median levels (Vesper et al. 2008). The median HbAA and HbGA levels of the comparable age group in this study are similar to those found in the Dutch group (HbAA: 48.6 pmol/g hemoglobin; HbGA: 49.6 pmol/g hemoglobin).

Several investigations have attempted to compute a daily intake of acrylamide from measured levels of hemoglobin adducts (Fennell et al. 2005). In general, these are roughly comparable with intakes based on dietary estimates (Dybing et al. 2005; FDA 2006; Mucci and Wilson 2008; Svensson et al. 2003; World Health Organization 2002). Using the computation method of Fennell et al. (2005), our adduct data predict a nonsmoking overall population mean intake of 0.8 μg/kg body weight/day, which is broadly similar to some of these other estimates. Our findings indicate that secondhand smoke can increase HbAA and HbGA levels. Therefore, exposure from food may not be the only relevant exposure source in nonsmokers. Some limitations are inherent in these types of calculations, including using high-dose acute kinetic studies to compute low-dose chronic exposure estimates and difficulty in accurately accounting for additional (nonadduct) and variable pathways of acrylamide metabolism and elimination. Although such estimates provide an approximate magnitude of acrylamide exposure, further studies are needed to assess the applicability of this model for chronic exposures and for estimating subtle differences between subpopulations or exposure sources.

Because acrylamide mutagenicity depends on its conversion to glycidamide, it is of interest to determine HbGA levels and to assess the relationship between HbAA and HbGA. Therefore, we applied a multiple regression analysis to predict HbGA levels based on HbAA levels, cotinine, sex, and BSA. A 10% increase in HbAA levels was associated with an HbGA increase ranging between 7.55% and 9.65% for different population subgroups. This increase was similar for smokers and nonsmokers. Because of the high intersubject variability in acrylamide metabolism, this relationship is applicable only for population groups and cannot be applied to individuals.

The reasons for negative associations between BSA and HbAA levels in nonsmokers are not fully understood. Reported associations between body mass index and HbAA have been inconsistent (Vesper et al. 2008; Wilson et al. 2008). On the basis of observations in animal studies, Osterman-Golkar et al. (2003) suggested that increased body mass leads to an increase in blood volume and the amount of hemoglobin available for adduct formation, and thus to a dilution of the adducts at similar intake doses. This dilution effect also affects serum cotinine levels, complicating further assessments of the impact of smoking on HbAA levels, as shown in our model for smokers, where increased BSA levels tend to attenuate the effect of serum cotinine on HbAA.

The study population is representative of U.S. noninstitutionalized individuals, and the study results may not be applicable to institutionalized individuals or to other countries. A large number of statistical comparisons have been made in the analysis of the data, and some findings may be the result of random chance. The point estimates of biomarker exposure are estimates of the various subpopulation exposures to acrylamide and not estimates of individualized exposures or specific exposure sources.

We have characterized U.S. population blood acrylamide and glycidamide adduct levels. Sex, age, and race/ethnicity do not strongly predict adduct levels, suggesting that the large variability in adduct levels among nonsmoking individuals is likely a result of individual intake of acrylamide from food and probably, to a lesser extent, of exposure to secondhand smoke. Dietary, metabolic, or other factors such as alcohol intake were not examined in this data set. Knowledge of all factors that affect the formation of glycidamide, as described in this study, will be important with respect to estimating risk from this known carcinogen.