A Population-Based Study of Dietary Acrylamide and Prostate Cancer Risk

IntroductionIn 2002 researchers from the Swedish National Food Administration reported the detection ofhigh levels of acrylamide in commonly consumed baked and fried foods. (SNFA,2002) Acrylamide isclassified as a probable human carcinogen (IARC, 1994), so the discovery that the compound isformed during the preparation of many foods caused alarm. Prior to 2002 sources of non-occupational acrylamide exposure were thought to be limited to tobacco products and drinking water.The data establishing acrylamide as a carcinogen comes from animal studies, which showincreased cancer rates in rats given acrylamide in water. (Johnson, 1986; Friedman, 1995) It isunclear whether human cancer risk is affected by chronic, low-level exposure to acrylamide throughfoods. Several epidemiological studies have examined the association between dietary acrylamideintake, as measured by a food-frequency questionnaire (FFQ), and cancer risk at various sites.(Mucci, 2003a and b, 2004, 2005, 2006; Pelucchi 2003) To date, no published report has found anysignificant increase in risk associated with higher acrylamide intake. However, the ability of FFQs tomeasure dietary acrylamide intake has not been established. One study reported significantcorrelations between acrylamide adducts to hemoglobin and questionnaire-calculated acrylamideintakes in men and in smoking women, but not in non-smoking women. (Wirfalt, 2007) In addition,no study has examined blood levels of acrylamide, a well-established biomarker for exposure inoccupational settings, and cancer risk in humans.In this study, we address these limitations using data from a population-based case-controlstudy of prostate cancer in Sweden. First, we measured blood acrylamide in a subset of men toexamine the correlation between this biomarker of exposure and intake measured by a food-frequency questionnaire. Second, we used both the blood data and the FFQ data on acrylamide toconduct case-control analyses of acrylamide exposure and prostate cancer risk. MethodsStudy population. The Cancer of the Prostate in Sweden (CAPS) study is a population-based case-control study of prostate cancer. Cases were drawn from four of the six national cancer registries inSweden between 2001 and 2002. Participants from the northern and central regions were between35 and 79 years old, and those from the southern regions were between 35 and 65 years. Caseswere incident cases of pathologically or cytologically verified prostate cancer. Cases were informedabout the study and asked to participate through their treatment physicians. Clinical data on TNM(tumor, nodes, and metastasis) stage, Gleason score, and serum prostate-specific antigen (PSA)level at diagnosis were obtained from linkage to the National Prostate Cancer Registry. This datawas available for 95% of cases in the study.Controls were randomly selected from the Swedish Population Registry and were frequencymatched to cases by five-year age groups and region of residence. Controls were contacted by mailand received the same information as cases.Overall, 1895 prostate cancer cases were invited to participate. Of those, 1499 (79%)completed the questionnaire and 1400 (74%) donated a blood sample. Of 1684 invited controls,1130 (67%) completed the questionnaire and 879 (52%) donated blood.All study participants gave informed consent at the time of enrollment in the study. The studywas approved by the ethics committees at Karolinska Institute and Umeå University. Dietary assessment. Participants completed a self-administered 261-item food frequencyquestionnaire that assessed usual intake of foods over the previous 12 months. Data from theSwedish National Food Administration on energy and nutrient content of foods was used to calculatetotal energy intake and intake of nutrients.Data on the acrylamide content of foods was collected from the Swedish National FoodAdministration and used to calculate usual intake of acrylamide. The acrylamide content of 18 foodswas used in the calculation of acrylamide intake. Acrylamide intake was calculated by multiplying theacrylamide content of a serving of the food by the frequency of consumption of that food and summing across all acrylamide-containing items on the FFQ. The resulting acrylamide intake is inmicrograms per day. Acrylamide and other intakes were calorie-adjusted using the residual method.(something by Willett)2617 men completed the FFQ component of the questionnaire. Sixteen were excludedbecause of unreasonably high or low energy intakes, and one was excluded because of missingacrylamide data. After exclusions, we had 1499 cases and 1118 controls with dietary acrylamideinformation. Measurement of blood acrylamide.As a biomarker of acrylamide exposure, acrylamide adducts to hemoglobin were measured inblood samples from a random sample of 377 men in the CAPS study. This biomarker has beenshown to correlate with levels of occupational exposure to acrylamide (Tornqvist, 2006). Itrepresents acrylamide exposure over the previous four months, or the half life of red blood cells.Participants received a blood sampling kit along with the questionnaire and informed consentform. They were given four tubes (heparin, plasma and EDTA-treated) and were instructed to donateblood at the nearest clinic. Unprocessed blood samples were sent by overnight mail to the UmeåBiobank, where samples were divided into plasma, serum, white and red blood cell components andstored in a freezer at -80°C until the time of analysis.Red blood cell samples were analyzed for acrylamide adducts to hemoglobin as describedelsewhere (Tornqvist, 1986; Bergmark, 1997)The samples were processed in four batches of approximately 100 samples each. Laboratorypersonnel were blinded to the case/control status of the samples. By chance, more case sampleswere included in batches one and two, and more controls were in batches three and four. Meanacrylamide adduct levels decreased over the four batches. Within each batch, mean adduct levelswere similar for cases and controls, suggesting that the difference between batches was due tolaboratory drift. Laboratory batch was adjusted for in all analyses.Eleven samples were not processed because the cells were clotted. 34 men who reportedusing tobacco products (cigarettes, pipes, or snuff) at the time of the questionnaire were excludedfrom the blood analysis, as tobacco users are exposed to much higher levels of acrylamide throughtobacco than through the diet. Mean blood acrylamide in these men was 152 pmol/g globincompared to 54 pmol/g in non-smoking men. After exclusions, 175 cases and 168 controls wereused in the analysis of blood acrylamide. Statistical analysis.For the validation of FFQ acrylamide intake, the correlation between calculated acrylamideintake and acrylamide adducts to hemoglobin was calculated in the subset of men with bloodmeasurements. The correlation was adjusted for age, region, and laboratory batch.To measure the association between blood acrylamide and risk of prostate cancer, we usedunconditional logistic regression models with indicator variables for quartile of blood acrylamide level.Quartiles were created based on the distribution among the controls. Age group and region, whichwere matching factors in this study, were included in all models, as was laboratory batch. Fullyadjusted models also include variables for BMI (continuous) and former smoking.To measure the association between dietary acrylamide intake and risk of prostate cancer,we used unconditional logistic regression models with indicator variables for quintiles of calorie-adjusted acrylamide intake. Quintiles were created based on the distribution of intake among thecontrols. Age group and region were included in all models. Fully adjusted models also includevariables for BMI (continuous), former and current smoking, education (four categories), zinc intake(ordinal quartiles), and total energy intake. Employment status and civil (marital) status were alsoconsidered as potential confounders. Several other nutrients and foods were considered as potentialconfounders, as well, including: alcohol, alpha-linolenic acid, calcium, vitamin D, folate,phytoestrogens, red meat, fish, and tomato. None of these was included in the final models, as theyhad little effect on the acrylamide effect estimates or precision. Data on these confounders werecollected in the self-administered mailed questionnaire. To test for a dose-response trend across quantiles of acrylamide, we modeled acrylamide asa continuous variable using the median intake in each quantile. The p-value of this continuousvariable was used to determine the significance of any linear trend across quantiles of intake. Allstatistical analysis was done using SAS 9.1. ResultsCharacteristics of the study population.Cases and controls were similar in acrylamide intake calculated from the FFQ and in bloodlevels of acrylamide (Table 1). Mean intake was 44.5 mcg/day among controls and 43.8 mcg/dayamong cases. In the subset of men with blood acrylamide measurements, the mean adduct levelwas 49.3 pmol/g globin among controls and 51.6 pmol/g globin among cases. Cases were morelikely to come from the Northern regions of Sweden. Cases and controls were similar in age,education, BMI, height, smoking status, and diet including daily intakes of dairy, red meat, fish, fruits,and vegetables and total energy intake.Acrylamide intake ranged from 8 to 125 mcg/day, or 0.08 to 1.59 mcg/ kilogram body weightper day. The top food contributors to acrylamide intake were crispbread, coffee, other bread, friedpotatoes, and buns and cakes (Figure 1). There was a significant (p<0.0001) correlation betweenacrylamide intake and intake of carbohydrates, fiber, and zinc (all positive) and alcohol (negative).Acrylamide intake was not correlated with age or height and was mildly correlated with BMI (r=0.05,p=0.01). Validation of FFQ acrylamide measurement.The partial correlation between dietary acrylamide intake and blood acrylamide (asacrylamide adducts to hemoglobin) was 0.25 (p<0.0001), adjusted for age, region, energy intake, andlaboratory batch (Table 2). Among controls only, the correlation was 0.35 (p<0.0001). Amongcases, it was 0.16 (p=0.05). Correlations between blood acrylamide and acrylamide intakemeasured in mcg/kg body weight per day were almost identical to the correlations with acrylamideintake in mcg/day (data not shown). Adjustment for energy intake improved the correlations byreducing within-person measurement error. Without adjusting for calories, the correlation betweenFFQ and blood acrylamide was 0.18 (p=0.0009), adjusted for age, region, and batch. Association between blood acrylamide and CaP risk.As shown in Table 3, no significant association was seen between quartile of bloodacrylamide and prostate cancer risk. Adjusting for age, region, BMI, former smoking, and laboratorybatch, the relative risk for the highest versus lowest quartile of blood acrylamide was 1.02 (95% CI:0.46-2.24). For a 10 pmol/g globin increase in blood acrylamide, the relative risk of prostate cancerwas 0.97 (CI: 0.81-1.17, p for trend=0.77).No association was found between blood acrylamide and specific prostate cancer endpointsincluding advanced disease, localized disease, highor low-grade disease, or highor low-PSAdisease (Table 4). Association between dietary acrylamide and CaP risk.For dietary acrylamide calculated from the FFQ, there was no association between higherintakes and prostate cancer risk (Table 5). The relative risk for the highest versus lowest quintile ofacrylamide intakes was 0.97 (CI: 0.75-1.27), adjusting for age, region, BMI, education, smoking, andzinc and energy intake. For a 10 mcg/day increase in acrylamide intake, the RR of prostate cancerwas 0.99 (CI: 0.92-1.06, p for trend=0.67). Similar results were seen when the analysis was limitedto non-smokers only. Among non-smokers, a 10 mcg/day increase in acrylamide intake wasassociated with a relative risk of 1.02 (CI: 0.94-1.09, p for trend=0.68).No association was found between acrylamide intake and specific prostate cancer endpointsincluding mortality, advanced disease, localized disease, highor low-grade disease, or highor low-PSA disease (Table 6). Again, these results were similar when restricted to non-smokers only. Intake of the five foods that contribute most to acrylamide intake in this study population wasexamined (Table 7). There was no association between crispbread, other bread, or coffee intakeand prostate cancer risk. There was a suggestion of increased risk for men in the highest tertile ofintake of fried potatoes and buns/cakes. However, this increased risk was unchanged when dietaryacrylamide was also included in the models, suggesting that any association between these foodsand prostate cancer risk is due to chance or components other than acrylamide. DiscussionValidation of FFQ Measurement.The correlation between blood acrylamide and FFQ acrylamide was highly significant. It wasnotably higher among controls than among cases. This difference may be due to less accuratereporting of diet among cases, or to recent changes in the diets of cases after diagnosis. Ourfindings are in line with the other published report comparing FFQ-calculated acrylamide intake andblood levels of acrylamide adducts. In the Malmo Diet and Cancer Cohort, Wirfalt, et al. (2007) founda correlation of 0.36 in smokers and 0.43 in non-smokers.Our correlations are highly statistically significant, and they are in line with the magnitude ofcorrelations seen for intake of some nutrients when compared to biomarkers of intake. In addition,the observed correlations compare favorably to the correlation of 0.47 observed between intensity oftobacco use and acrylamide adducts in the Malmo Diet and Cancer Cohort. This is reassuring, giventhat self-reported smoking habits are considered a valid assessment of exposure, and bloodacrylamide adducts have been shown to vary with smoking intensity. (Bergmark 2007; Schettgen2004)It is also worth noting that acrylamide adducts to hemoglobin have not been definitivelyestablished as a valid biomarker of dietary acrylamide exposure. Acrylamide adducts areestablished as a valid marker of occupational levels of exposure, but it is not clear how responsiveadduct levels are to the much lower levels of acrylamide found in the diet. It is known that the levelof variation in acrylamide adducts in non-smoking, non-occupationally exposed humans is lower thanthe estimated variation in dietary acrylamide content. (Hagmar, 2005) One small pilot feeding study(Vesper, 2005) showed mixed results in assessing change in adduct levels after increasing intake ofacrylamide; however, the study was too small and too short to draw conclusions. To fully evaluatethe results of FFQ validation studies such as ours, it will be necessary to see how a larger andlonger-term feeding study affects acrylamide adduct levels.Our results are promising considering that only acrylamide adducts were measured, and notadducts of glycidamide, the major metabolite of acrylamide. The degree of conversion toglycidamide varies between individuals and by acrylamide intake (Boettcher, 2005; Schettgen, 2004),so a combined measure of acrylamide and glycidamide adducts to hemoglobin is probably the bestbiomarker of acrylamide exposure. Therefore, our correlation likely underestimates the true validityof the FFQ. Blood acrylamide and prostate cancer risk.This is the first study to look at the association between blood acrylamide and cancer risk inhumans. We found no association between blood levels of acrylamide and risk of overall prostatecancer or specific prostate cancer endpoints. Our ability to study specific endpoints such asadvanced or localized disease was limited by low sample sizes in the subgroup with bloodacrylamide data.As discussed above, our ability to study this association was limited by the lack ofglycidamide adduct data. In addition, we have only one blood sample per participant, so we cannotaccurately measure levels of acrylamide exposure over time. These limitations would likely result innon-differential misclassification of blood acrylamide, reducing our ability to see an increased risk ofcancer associated with blood acrylamide levels. Finally, for cases the blood sample was collectedafter diagnosis, so it may not reflect typical levels before the cancer. It is unclear how this source oferror might affect our estimates. Dietary acrylamide and prostate cancer risk.We found no association between FFQ-measured acrylamide intake and risk of overallprostate cancer or specific prostate cancer endpoints. We also found no association between intakeof the major acrylamide-contributing foods and prostate cancer risk. This is the first study to examineprostate cancer risk and acrylamide intake. Several other cancer sites have been examined, and allreports to date have been null. (Mucci, 2003a and b, 2004, 2005, 2006)Random measurement error in acrylamide intake is likely. The FFQ was not originallydesigned to measure acrylamide intake, so food items with different cooking methods and quitedifferent acrylamide content are sometimes grouped into a single FFQ question. This makes it moredifficult to measure intake accurately and likely biases our results towards the null.Differential measurement error is also a possibility. The difference between cases andcontrols in the validation component of this study suggest that FFQ accuracy may vary by diseasestatus. If so, it is not clear how such recall bias might affect our results. Further study of acrylamideintake in prospective cohorts will be necessary to eliminate the possibility of recall bias.In conclusion, we find no association between acrylamide exposure as measured in blood orby FFQ and risk of prostate cancer. Epidemiological studies do not have the power to rule out verysmall increases in risk associated with high acrylamide intake. However, the concordance of theblood and FFQ results, along with previous epidemiological studies, suggest that the levels ofacrylamide taken in the diet are not responsible for a significant increase in cancer risk in humans.