Plasma Alkylresorcinol Metabolites as Biomarkers for Whole-Grain Intake and Their Association with Prostate Cancer: A Swedish Nested Case–Control Study Free

This article is featured in Highlights of This Issue, p. 1

Whole grains have been linked to decreased risk of many diet-related diseases, including cardiovascular disease (1), type II diabetes (2), and certain cancers (3). Possible mechanisms for the protective effect are that whole grains slow the digestion, lower insulin secretion, and could potentially improve insulin sensitivity (4). Whole grains are also a major source of fiber, vitamins, minerals, lignans, and other phenolic compounds, all of which may have beneficial health effects (5). Whole-grain products are relatively high in antioxidant activity, which may protect DNA from oxidative damage and mutation that can lead to cancer (5). Lignans are hormonally active compounds that may protect against certain hormonally mediated diseases, including prostate cancer (6). Diet has been suggested as one of many environmental factors affecting prostate cancer development and progression (7, 8), but the only established risk factors to date are age, ethnicity, and family history of prostate cancer (9). Small clinical trials and animal studies suggest that whole grains might have a beneficial effect on prostate cancer progression (10–12), mediated by increased apoptosis and lower insulin secretion. Landberg and colleagues also showed that whole grains from rye compared with cellulose-supplemented refined wheat lowered plasma prostate-specific antigen (PSA) among patients with prostate cancer (11). The hypothesis that whole grains may lower the risk of prostate cancer has been examined in several epidemiologic studies. Results from cohort studies have mostly found null associations (13–16). Dietary assessment methods used in large-scale observational studies are associated with a varying degree of random and systematic measurement errors that may attenuate (or bias) diet–disease associations. For this reason, dietary biomarkers can be used as additional estimates of dietary intake. Alkylresorcinols are phenolic lipids present in the bran fraction of wheat, rye, and barley kernels. About half (45%–71%) of alkylresorcinols are absorbed, and intact alkylresorcinols and its metabolites can be measured in plasma (17). The two main metabolites of alkylresorcinol, 3,5-dihydroxybenzoic acid (DHBA) and 3,5-dihydroxyphenylpropanoic acid (DHPPA), have been suggested as biomarkers of whole-grain intake in Nordic populations (18, 19). There is, however, limited data on their usefulness in large observational studies. The aim of this study was therefore to identify major dietary and lifestyle determinants of plasma alkylresorcinol metabolites in a nested case–control study of men in the Malmö Diet and Cancer (MDC) study and to examine whether alkylresorcinol metabolite concentration is associated with incident prostate cancer.

The MDC study is a population-based prospective cohort study set in Sweden. The cohort has been described in detail elsewhere (20–22). In short, baseline examinations were carried out from March 1991 to October 1996. All men born between 1923 and 1945 and all women born between 1923 and 1950 living in Malmö were invited to participate in the study. Baseline examinations included an extensive lifestyle and socioeconomic questionnaire, dietary assessment, direct anthropometric measurements, and collection of blood samples. With a participation rate of approximately 40%, 28,098 individuals (11,063 men) with complete data represent the cohort. The MDC study was approved by the Ethics Committee at Lund University (Lund, Sweden) and informed written consent was obtained from all participants.

Incident prostate cancer cases were identified by linkage of personal identification numbers with the Swedish Cancer Register (SCR). Information on vital status of the study population was obtained from the Swedish Cause of Death Registry and the Swedish Tax Agency. All men with prevalent cancers at baseline were excluded, except nonmelanoma skin cancer. A total of 1,016 incident prostate cancer cases (as first cancer diagnosis) occurred until December 31, 2009. Data on the clinical and histologic characteristics of the tumors were collected from the National Prostate Cancer Register (NPCR). Data on cases (n = 54) that occurred between 1991 and 1995 were manually extracted from medical records by using standard routines. The SCR is known to be 98% complete. A validation of NPCR data from another region showed high validity for all variables, including the variables used in the classification of tumors by case severity in this study (23). High-risk prostate cancer cases were classified as local clinical tumor stage T3 or T4, the presence of lymph node metastasis (N1) or bone metastasis (M1), a Gleason score 8, or serum PSA concentrations 50 ng/mL (24). Tumors were also classified as high-risk cases if the World Health Organization grade was 3 and the Gleason score was unavailable (n = 6). Cases diagnosed after subjects presented with malignancy-related symptoms or lower urinary tract symptoms were classified as symptomatic cases. The cases were diagnosed 0 to 18 years (mean, 9 years) after baseline examinations and the mean age at diagnosis was 70 years (range 50–85 years). For each case, on average two controls were selected, matched by age and date of study entry (± 90 days) from the cohort members at risk at the time of diagnosis of the case (n = 1,828). Successful blood analysis of alkylresorcinol metabolites (see below) was available for 2,827 subjects, leaving 1,010 cases and 1,817 controls for the analyses.

The sum of DHBA and DHPPA was used to approximate total plasma alkylresorcinols. The nonfasting blood samples collected at baseline were processed and separated for plasma within 1 hour. The samples have been stored at −80°C until analysis 2012. The quality control program of the biobank in the MDC study has been described previously (25). Alkylresorcinol metabolites were analyzed at Folkhälsan Research Center in Helsinki by high-performance liquid chromatography (HPLC) with coulometric electrode array detection (ESA Biosciences, Inc.) as described by Koskela and colleagues (26). In brief, plasma samples (100 μL) were hydrolyzed overnight at 37°C with β-glucuronidase and sulfatase, and the sample was extracted with diethyl ether. Separation of diethyl ether and water phase was done by freezing. The combined organic phases were evaporated to dryness. The sample was reconstituted in 50 μL methanol, and 100 μL HPLC mobile phase was added. The sample was filtered through Gelman GHP 0.2 μm filter and analyzed with HPLC-coulometric electrode array detection. The method is considered accurate, specific, and reproducible (26).

The dietary assessment methodology has been described in detail elsewhere (27–30). It combined (i) a 7-day menu book that collected detailed description of lunches and dinners, cold beverages, medications, and dietary supplements, (ii) a 168-item dietary questionnaire, including frequencies and portion sizes, and (iii) a 1-hour interview. Food intakes were converted to nutrient intake data using the MDC Food and Nutrient Database, originating from PC KOST2-93 of the Swedish National Food Administration. The nutrient variables examined in this study were energy (kcal), fiber (grams per day), and percentage of nonalcohol energy contributed by protein, fat, and carbohydrates. The food groups (grams per day) examined were: cereals (grains, cereals, and flours), low-fiber bread, high-fiber bread, and rice and pasta. The selection of these food variables was based on their potential contribution of whole grains. Whole-grain intake (grams per day) was estimated using the detailed information about consumed foods/brands in the abovementioned food groups and using the Food and Nutrient Database of the Swedish National Food Administration (available from: www.slv.se).

Trained project staff measured participants' weight (kg), height (cm), and waist (cm). Body mass index (BMI) was defined as kg/m². Lifestyle and socioeconomic variables were obtained through the standardized questionnaire. Educational status was categorized into elementary, primary and secondary, upper secondary, further education without a degree, and university degree. Smoking habits were categorized into current smokers (including irregular), former smokers, and nonsmokers. A total physical activity score was obtained by combining information on leisure-time physical activity, domestic activity, and work activity (13). Individuals with no consumption of alcohol in the menu book and those who reported no consumption of alcohol during the previous year in the questionnaire were categorized as zero consumers. The other subjects were categorized into three groups according to their alcohol consumption: <20 g alcohol per day (low), 20–40 g (medium), and >40 g (high). Prevalent diabetes at baseline was confirmed through national and regional registries. Participants were classified as potential underreporters, adequate reporters, or overreporters of energy intake, as described in detail previously (31). Dietary change in the past (yes or no) was derived from the questionnaire item “have you substantially changed your eating habits in the past due to illness or other reasons?” (32).

Alkylresorcinol metabolite concentrations were highly skewed (Fig. 1). The differences in total alkylresorcinol metabolite concentration (DHBA+DHPPA) and participant characteristics between cases and controls were tested using Wilcoxon rank-sum test (alkylresorcinol metabolite concentrations and whole-grain intake), Student t test (other continuous variables), and χ² test (categorical variables). Participant characteristics across quintiles of plasma alkylresorcinol metabolites were also examined. Median alkylresorcinol metabolite concentration by age, lifestyle factors (smoking status, educational level, alcohol consumption, BMI categories, and physical activity), diabetes status, past food habit change, and energy misreporting was calculated among the controls (n = 1,817). Because of skewed data and the large number of zero values (n = 408) for plasma alkylresorcinol metabolites, normal distribution could not be approximated by natural log transformation, and decile ranking was used instead. Differences in plasma alkylresorcinol metabolite concentration (assigning the median value for controls in each decile) by these variables were tested, using a general linear model controlling for age and date of study entry.

Food and nutrient variables were energy adjusted by regressing the intake (log_e-transformed, dependent variable) on the total energy intake (log_e-transformed, independent variable). All food variables (i.e., whole grains, cereals, low-fiber bread, high-fiber bread, and rice and pasta) were divided into approximate deciles based on their residual ranking. Partial correlations between alkylresorcinol metabolite concentration and energy-adjusted dietary variables and body composition were computed among the controls (n = 1,817), controlling for age, date of study entry, and total energy. We also formulated a minimally adjusted model with all food variables included simultaneously and adjusting for age, date of study entry, and total energy. Finally, in an exploratory analysis, all food variables, whole grains, body composition variables, smoking status, educational level, alcohol consumption, and physical activity were included in a linear regression model followed by stepwise backward elimination of variables with P > 0.10. The model was adjusted for age, date of study entry, and total energy.

We examined a potential nonlinear relation between plasma alkylresorcinol metabolites and prostate cancer risk by fitting a restricted cubic spline with five knots (placed at the 10th, 25th, 50th, 75th, and 90th percentiles of total plasma alkylresorcinol metabolite concentrations) to a logistic regression model using a continuous variable for alkylresorcinol metabolites and adjusting for all other covariates. Given the long natural history of prostate cancer and the strong age dependence, we assessed potential competing risks from nonprostate cancer causes of death by fitting a restricted cubic spline to a logistic regression model for nonprostate cancer mortality. The Wald χ² test was used to test the overall significance (4 df). Linearity was tested by using the Wald χ² test (2 df), for which P < 0.05 is consistent with nonlinearity.

For examination of the association between alkylresorcinol metabolites and prostate cancer, participants were ranked into quintiles (Q) of their plasma concentration. To make full use of the data without restriction to the matched cases and controls, we used unconditional logistic regression to estimate ORs and 95% confidence intervals (CI) for prostate cancer incidence adjusting for matching variables (age and date of study entry). However, estimated ORs were virtually the same when using conditional logistic regression (data not shown). Additional analyses included adjustment for total energy, height, waist, smoking status, and educational level. These covariates were identified through the literature and were graphically evaluated for being potential confounders. Because few individuals (n = 12) had missing values on any of the covariates, these were excluded from analysis. Inclusion of potential dietary confounders (including calcium, vitamin D, selenium, monosaccharides, sucrose, saturated fatty acids, polyunsaturated fatty acids, total carbohydrates, dairy products, soft drinks, vegetables and processed meats) did not affect risk estimates (data not shown). In addition to a main effect analysis, the models were stratified by low-risk and high-risk cases to examine whether associations between alkylresorcinol metabolites and prostate cancer incidence differed according to clinical stage. To assess the impact of opportunistic PSA testing, we also investigated symptomatic cases separately.

We conducted a series of sensitivity analyses. The analyses were repeated with exclusion of individuals who reported dietary change in the past (to include only individuals more likely to have stable food habits and consequently more stable long-term whole-grain intake and alkylresorcinol metabolite concentrations). Further exclusions included men with prevalent diabetes at baseline and potential energy misreporters. We also investigated the association between plasma DHBA and DHPPA and risk of prostate cancer separately.

Stata/SE 12.0 (StataCorp LP) was used for all statistical analyses. All tests were two sided and P < 0.05 was considered statistically significant.

Median concentration of alkylresorcinol metabolites (DHBA+DHPPA) in plasma was 45.8 nmol/L among cases and 43.1 nmol/L among controls (Table 1). A lower frequency of current smokers was observed among the cases compared with controls.

Alkylresorcinol metabolite concentration was positively associated with age (P = 0.001) and educational level (P < 0.0001). Nonsmokers had 30% higher median alkylresorcinol metabolite concentration than smokers (P < 0.001), and men with normal weight had 16% higher concentration than obese men (P = 0.001). There was no difference in metabolite concentration by alcohol consumption, physical activity, or diabetes status at baseline. Past food habit changers had 21% higher median metabolite concentration (P = 0.023; Table 2). The alkylresorcinol metabolite concentration was positively correlated with fiber intake (r = 0.27; P < 0.001) and whole-grain intake (r = 0.31; P < 0.001). Potential food sources of alkylresorcinols were also positively correlated with metabolite concentration, including cereals (r = 0.15; P < 0.001) and high-fiber bread (r = 0.34; P < 0.001), whereas low-fiber bread was negatively correlated with alkylresorcinol metabolite concentration (r = −0.17; P < 0.001) when examined in separate models. BMI and waist circumference were negatively correlated with metabolite concentration (Table 3). When all food variables were included simultaneously in the model, high-fiber bread (r = 0.30) and cereals (r = 0.12) were positively correlated with alkylresorcinol metabolite concentrations. In the model, including all food variables (model 1) and various potential confounders, cereals and high-fiber bread as well as height were significantly positively associated with alkylresorcinol metabolite concentration, and waist circumference was negatively associated with metabolite concentration (Table 3). Including only whole grain as a food source (model 2) of alkylresorcinol in the multivariate model showed that whole grain was significantly positively associated with plasma alkylresorcinol metabolites (Table 3). Educational level and nonsmoking were also positively associated with metabolite concentrations (data not shown). When adjusting for age and date of study entry, whole-grain intake, waist circumference, smoking status, and educational level explained 13.1% of the variation in alkylresorcinol metabolite concentrations. Whole grains alone explained 10.9% of the variation in AR metabolite concentration. Correlation coefficients for food variables and plasma DHBA and DHPPA, when examined separately, showed similar associations as that for total plasma alkylresorcinol metabolites (Supplementary Table S1). The correlation between DHBA and DHPPA was moderate (r = 0.68) when adjusting for age and date of study entry. Baseline characteristics of the study population by quintiles of plasma alkylresorcinol metabolites are presented in Supplementary Table S2.

The restricted cubic spline regression model (Fig. 2) showed a nonlinear association between alkylresorcinol metabolites and prostate cancer incidence (P_linearity = 0.04). Higher risks were seen for low and high concentrations compared with moderate concentration. On the basis of this observation, the third quintile was used as the reference category in subsequent analyses. In the multivariate model, an increased prostate cancer risk was observed for the highest (Q5) compared with the median quintile (Q3) of plasma alkylresorcinol metabolites (OR = 1.41; 95% CI, 1.10–1.80). Similar associations were seen when investigating the association by case severity (low- and high-risk cases) as well as symptomatic cases. However, the increased risk seen in Q5 compared with Q3 was only significant for, and seemed to be driven by, low-risk prostate cancer (OR = 1.39; 95% CI, 1.05–1.85); Table 4). Among the potential confounding factors, smoking status and educational level had an attenuating influence on the risk estimates.

We investigated several potential dietary confounders, but their effect on the observed risk estimates was marginal and the associations remained significant (data not shown). Among plausible dietary confounders, we found that plasma alkylresorcinol metabolites were positively correlated with primarily calcium, monosaccharide, and total carbohydrate intake (Supplementary Table S3). When we excluded men with prevalent diabetes at baseline, past food habit changers and energy misreporters, the associations between alkylresorcinol metabolite concentration and prostate cancer were strengthened and the OR in Q5 compared with Q3 was 1.50 (95% CI, 1.10–2.04) for total prostate cancer incidence (data not tabulated). We investigated potential effect modification by smoking and waist circumference on the association between plasma alkylresorcinol metabolites and prostate cancer, but we observed no significant interaction (all P > 0.05; data not shown). The association between the two metabolites (i.e., DHBA and DHPPA) and prostate cancer did not differ significantly, but the associations seen with prostate cancer seemed to be driven by DHBA concentration rather that DHPPA concentration (data not shown).

Plasma alkylresorcinol metabolite concentration was linearly associated with decreased risk of nonprostate cancer mortality (P_{overall effect} = 0.03; P_linearity = 0.58). The OR for nonprostate cancer death in the highest quintile of alkylresorcinol metabolite concentration was 0.66 (95% CI, 0.46–0.95) compared with the lowest quintile. The reduced risk for nonprostate cancer death remained significant in strata-specific analysis by case–control status.

In this large observational study, we investigated the usefulness of plasma alkylresorcinol metabolites as a biomarker for whole-grain intake in relation to disease incidence. We observed that alkylresorcinol metabolites positively correlated with whole-grain intake. We found no evidence for a protective association between whole grains and prostate cancer incidence, but high alkylresorcinol metabolite concentration was associated with lower risk of nonprostate cancer mortality. Unexpectedly, we found that very high levels of plasma alkylresorcinol metabolites were associated with an increased risk of prostate cancer (primarily low-risk prostate cancer) compared with moderate concentrations.

The underlying concept of using a biomarker for dietary intake is that it may improve ranking of subjects for exposure compared with dietary assessment methods relying on self-report. The coefficients of correlation between plasma alkylresorcinol metabolites, whole grains, and dietary fiber were significant, but moderate. In a free-living population with self-selected diets, the correlation between plasma alkylresorcinol metabolites and self-reported intake of whole grains is however, within the expected range. In an intervention study where subjects are assigned a high whole-grain versus no whole-grain diet, correlation coefficients are likely to be much higher, as previously reported (11). In addition, alkylresorcinols are only present in appreciable amounts in whole grains from rye and wheat, and therefore intake of other cereals (e.g., oats) may deflate the correlation. Absorbed alkylresorcinols are thought to be metabolized similarly to tocopherols (17). The phase I metabolism of alkylresorcinols leads to the formation of DHBA and DHPPA. We measured the plasma concentration of DHBA and DHPPA due to their longer estimated half-lives (≈11–16 hours; ref. 33) compared with intact alkylresorcinol (≈5 hours; ref. 34). The reliability of plasma alkylresorcinol (35) and its metabolites (36) over 4-month time has been investigated recently. The intraclass correlation coefficients for DHBA and DHPPA were estimated to 0.23 and 0.33, respectively, among men. Although the mean concentrations of the metabolites did not differ significantly over the 4-month period, these results indicate a fairly poor reliability. The results from the study by Montonen and colleagues may however, not be transferable to the current study with a different population and a different methodology for the analyses of metabolite concentrations (36). However, additional studies are needed on the validity of alkylresorcinol metabolites as a biomarker for whole grains. Because the estimated half-life seems longer for DHPPA (33) and its reliability is potentially higher compared with DHBA, it is of concern that the association between the alkylresorcinol metabolites and prostate cancer seems to be driven by DHBA. Among the investigated determinants of alkylresorcinol metabolites, the concentrations were mainly affected by whole grains and fiber intake. Additional studies are needed to assess the impact of various factors that could potentially affect the plasma concentrations, including dietary intakes, anthropometric factors, and drugs. In this study, we found that whole-grain intake alone explained approximately 11% of the variation in total plasma alkylresorcinol metabolite variation. The large proportion of unexplained variation may be due to the use of nonfasting blood samples, and to the measurement error of dietary intake that combined with a single blood sample is likely to attenuate the true correlation between whole grains and alkylresorcinol metabolites.

The use of a quantifiable biomarker confirms and strengthens the previous conclusion that whole grain is not associated with decreased risk of prostate cancer incidence in this population (13). Given the high colinearity between dietary factors, we cannot exclude the possibility that the observed increased risk of prostate cancer seen with high plasma alkylresorcinol metabolites is an effect of some other dietary factor or environmental factor that is correlated with whole-grain intake. Although adjustment for various dietary confounders did not affect risk estimates, it is virtually impossible to isolate the effect of a single dietary factor. Furthermore, whole-grain intake may also reflect exposure to dietary cadmium, which has been suggested to increase the risk of primarily localized prostate cancer (37). Although we had no information on dietary cadmium exposure, the major food sources for dietary cadmium and whole grains are the same, primarily bread and cereals. It should also be noted that plasma alkylresorcinol metabolites are subject to confounding by the same risk factors that are associated with self-reported whole-grain intakes. Higher plasma alkylresorcinol metabolite concentrations were observed, as expected, among older and lean men, nonsmokers, and men with higher educational level. This finding strengthens the conclusion that alkylresorcinol metabolites are useful biomarkers for whole-grain intake in this population. The lower than expected median intake among men in the highest age group (ages 65–73) when not adjusting for date of study entry is likely explained by a lower total energy intake among older men as well as a change in the coding routines for the dietary assessment method that occurred in September 1994 (38). Furthermore, because the MDC study recruited slightly older men during the last years of baseline examination, this may also be a cohort effect. When adjusting for date of study entry (which takes into account method and cohort effects), age was positively associated with metabolite concentration.

A previous study found that high whole-grain intake was associated with increased risk of prostate cancer, however, the association was attenuated and not significant when limiting analysis to cases occurring in the pre-PSA screening era (14). The association seen between alkylresorcinol metabolites and prostate cancer in this study seems to be stronger among low-risk cases. There is currently no formal screening program in place in Sweden, and the effect of opportunistic PSA testing is difficult to estimate. However, similar associations (but nonsignificant) were seen also for high-risk and symptomatic prostate cancer risk. Although it is highly plausible that diagnostic bias to some extent affected the associations seen, it is not likely to fully explain the association seen between plasma alkylresorcinol metabolites and prostate cancer. Interestingly, a high alkylresorcinol metabolite concentration was associated with a decreased risk of nonprostate cancer mortality. Although this finding must be interpreted with caution, because the current study population is a nested case–control study aimed at investigating prostate cancer, the association indicates a health-positive effect of whole grains, which requires further confirmation in a different study setting. Men with high intake of whole grains are potentially more likely to live to an age when prostate cancer is most commonly diagnosed. Thus it cannot be ruled out that the increased risk of prostate cancer seen with high alkylresorcinol metabolite concentration is an effect of competing risks. Residual confounding might explain the nonlinear association seen between plasma alkylresorcinol metabolites and prostate cancer because men with very low as well as very high whole-grain intake are likely to differ in regards to various lifestyle, environmental, and socioeconomic factors.

In this nested case–control study of Swedish men, we found that plasma alkylresorcinol metabolites were primarily determined by reported whole-grain intake. The main conclusion of this study is that high whole-grain intake is unlikely to lower risk of incident prostate cancer but may be associated with lower mortality from nonprostate cancer causes. The results from this study also highlight the problem of excluding alternative explanations to findings between diet, lifestyle, and environmental factors in relation to prostate cancer risk. Residual confounding, detection bias, and undue influences of competing risks may cloud the true associations between lifestyle-related risk factors and this disease.

M.J. Tikkanen has a commercial research grant from Fazer Inc. E. Wirfält is a consultant/advisory board member of the working group coordinating the revision of the Nordic Nutrition Recommendations (5th edition) under the Nordic Council of Ministers. No potential conflicts of interest were disclosed by the other authors.

Conception and design: I. Drake, E. Wirfält, P. Wallström

Development of methodology: B. Gullberg, H. Adlercreutz, M.J. Tikkanen

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Bjartell, P. Wallström

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): I. Drake, B. Gullberg, P. Wallström

Writing, review, and/or revision of the manuscript: I. Drake, E. Sonestedt, B. Gullberg, A. Bjartell, H. Olsson, H. Adlercreutz, M.J. Tikkanen, E. Wirfält, P. Wallström

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M.J. Tikkanen, P. Wallström

Study supervision: E. Wirfält, P. Wallström

The authors thank Adile Samaletdin for technical assistance and the participants of the MDC cohort.

This study was supported by the Swedish Council for Working Life and Social Research, the Swedish Cancer Society, the Albert Påhlsson Foundation for Scientific Research, the Gunnar Nilsson Cancer Foundation, Skåne University Hospital – Foundations and Donations, the Malmö General Hospital Foundation for the Combating of Cancer, and the Ernhold Lundström Foundation for Scientific Research.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.