Epidemiologic evidence linking dietary intake of heterocyclic amines (HCAs) with cancers of the colon (Augustsson et al., 1999; Butler et al., 2003; Gunter et al., 2005; Nowell et al., 2002; Shin et al., 2007; Sinha et al., 2001; Wu et al., 2010), breast (Delfino et al., 2000; Sinha et al., 2000), pancreas (Anderson et al., 2005; Li et al., 2007), and prostate (Cross et al., 2005) is inconsistent, possibly due to differences in exposure assessment methods and the associated measurement error (Alaejos et al., 2008; Butler et al., 2003; Cantwell et al., 2004; Cross et al., 2011; Norat and Riboli, 2001; Turesky and Le, 2011). The standard method for assessing dietary HCA intake in epidemiologic studies of cancer combines meat intake information from food frequency questionnaires (FFQs) with a database of mean concentrations of HCAs measured in cooked meats (Kazerouni et al., 2001; Keating et al., 2007; Sinha et al., 2005). Strengths of this method include low participant burden and the ability to provide an estimate of usual intake of HCAs, typically over the previous year. Possible limitations include imperfect recall of usual intake, recall bias (in retrospective studies), and differing accuracy of reporting of meat intake information by gender, social desirability (the tendency to respond to questions in a manner to avoid embarrassment or present oneself favorably), or body mass index (Scagliusi et al., 2009). In addition, the FFQ may poorly represent individual-level exposure because HCA concentrations in meats vary with many aspects of cooking that may not be included in the FFQ or HCA database, such as pre-cooking in the microwave, frequency of flipping meat, concentrations of precursors, or use of marinades (Sinha and Rothman, 1997; Skog, 2002; Salmon, 2000). Despite the importance of exposure classification in epidemiologic evaluation of PhIP and reliance of the FFQ for this purpose, little is known of its reliability and comparability with other methods.

Diet diaries offer another strategy for classifying exposure. An advantage of diet diaries is that they don’t rely on recall, but accuracy of recording may differ by the same individual characteristics as FFQs (Scagliusi et al., 2009). Furthermore, the number of collection days required to obtain accurate estimates of usual intake increases with increasing intra-individual variability in intake of the target food items. Increasing the number of days of diary collection decreases compliance (Willett, 1998). Like FFQs, diaries must be paired with measured values of mean concentrations of HCAs to obtain individual-level estimates of intake. As with FFQs, there has been little previous work examining the comparability of diary-based estimated HCA intake with other methods.

Biomarkers reflect the body burden from exposure without the measurement error introduced from self-reported dietary intake (Cantwell et al., 2004). HCA biomarkers likely have high specificity for the dietary route, because diet is the major source of HCA exposure (Sugimura et al., 2004). Urinary concentration of one of the most mass-abundant HCAs, 2-amino-1-methyl-6-phenylimidazo[4,5b]pyridine (PhIP) and its metabolites, is sensitive to increases in dietary PhIP as demonstrated in controlled feeding studies (Friesen et al., 2001; Kulp et al., 2000; Reistad et al., 1997; Stillwell et al., 1997; Strickland et al., 2002). Urinary PhIP and its metabolites have been detected at background levels in populations on unrestricted diets (Kidd et al., 1999; Ushiyama et al., 1991). One of the limitations of urinary PhIP for estimating long-term exposure indicative of cancer risk is its short half-life (less than 12 hr) (Boobis et al., 1994; Turteltaub et al., 1999). Although multiple days of urine samples can be used to estimate usual exposure, like diaries, this can present issues of participant burden and compliance. While longer-term HCA biomarkers (e.g. DNA, hemoglobin, albumin adducts, hair levels) are more likely to capture the more relevant exposure period in epidemiologic studies of cancer, these methods are not yet adequately sensitive, reliable, or feasible for such studies (Magagnotti et al., 2000; Turesky and Le, 2011).

Recognizing the importance of HCA exposure classification for evaluating linkages with cancer and the limited availability of evaluative data, we conducted a multi-day, multi-method study comparing repeated measures from an FFQ, diary, and urinary biomarker, using PhIP as a representative compound.

PhIP intakes estimated from each of the three methods are summarized in Table 1. The median (of the individual mean values) PhIP intake for the FFQ and diary methods were 32 (range: 0 to 368 ng/day) and 0 ng/day (range: 0 to 630 ng/day), respectively. The diary was relatively insensitive to detecting PhIP intake (20%) relative to the FFQ (96%). The median creatinine unadjusted and adjusted urinary PhIP concentration (of individual mean values) from the three collection periods were 3.42 pg/mL (range: <LOD to 35.6 pg/mL) and 7.23 ng/g (range: <LOD to 37.4 ng/g), respectively. PhIP was detectable in 56% of the urine samples.

The within-subject variability in intake estimates for all three methods far exceeded the between-subject variability, yielding ICCs less than 0.5 (Table 2). The FFQ was administered twice 9 months apart and, of the three methods, was the most reliable for estimating PhIP intake, though modest (ICC=0.30, 95% confidence interval [−0.02, 0.57]); Spearman correlation between two FFQs=0.50, P=0.02). The urinary biomarker, which was measured three times over 9 months, was the least reliable method of PhIP intake (ICC=0.01, 95% confidence interval [−0.16, 0.23]). However, the reliability improved substantially with creatinine-correction (ICC=0.15, 95% confidence interval [-0.04, 0.38]). These low ICCs indicate that use of a single collection of any of these metrics in a case-control study would substantially attenuate the observed OR (Table 2). For example, using a single FFQ to estimate PhIP intake in a study population in which the true OR is 2.0, the observed OR would be 1.2.

Examining the associations across all three methods, the mean diary PhIP intake and the mean urinary PhIP concentration achieved significance only when restricting the analysis to those with at least one non-zero intake estimate from the diary (n=15, r=0.75, P=0.001; Figure 1B). No association was observed when all participants were included (n=36, r=0.18, P=0.28; Figure 1Ab). Fifteen of the 21 participants (71%) with diary-based mean PhIP intake of zero had measurable urinary PhIP at one or more time points (Figure 1a). Mean PhIP intake estimated from the FFQ was not correlated with mean urinary PhIP levels (r=0.20, P=0.23) nor or mean PhIP intake estimated using the diary (r=0.17, P=0.32), but was modestly correlated with a metric that combined urinary and diary-based PhIP (r=0.30, P=0.08) (Table 3). Meat consumption (diary and FFQ reported) was not associated with mean urinary PhIP concentrations. We also looked for associations across methods by aggregating the individual-level data into tertiles with no positive trend data evident (Table 4). Analysis with and without participants who consumed fish that was broiled, grilled, or fried did not yield appreciable differences in results, so all fish eaters were retained in the analyses. Correlations were of similar magnitude and statistical significance with creatinine-adjusted PhIP, and therefore only correlations with unadjusted PhIP are presented.

We used a repeated measures design to evaluate the repeatability and comparability of three different methods for assessing dietary exposure to HCA: FFQ, diary, and the urinary biomarker PhIP. We observed relatively high within-subject variability in the repeated collections for all methods and relatively weak agreement among the different methods. The current study raises important methodological considerations for future studies.

We observed PhIP intake estimates that were lower compared to those of controls in previous case-control studies that used the same methods. For example, estimated PhIP intake from the baseline (median=26 ng/day; mean=60 ng/day) and follow-up (median=33 ng/day; mean=75 ng/day) FFQs were lower than intake reported in controls in North Carolina (mean=86.7 ng/day, n=611) (Butler et al., 2003); Maryland (mean=78.1 ng/day, n=165) (Cantwell et al., 2004); Los Angeles, Seattle, Detroit, and Iowa (median=73.8 ng/day) (Cross et al., 2006); and California (median=92 ng/day) (Delfino et al., 2000). The lower intakes observed in the current study may be due to differences in season of ascertainment (Subar et al., 1994) or the exclusion of Fridays and Saturdays from our collection days, since meat intake tends to be higher on the weekends (Haines et al., 2003).

Similar to the FFQ-based estimates, the mean estimated PhIP intake from the diary in our study (85 ng/day) was lower than the previously reported mean intake based on the diary (264 ng/day) (Cantwell et al., 2004). In addition to our exclusion of Fridays and Saturdays, this difference may also be due in part to the larger number of diaries collected by Cantwell et al. (12 days) relative to the current study (3 days). Because of the large within-subject variability, the true long-term PhIP intake is likely not well-represented by 3 days of diary collections.

The concentrations of urinary PhIP in our study population were within the range of previously reported results. The mean urinary PhIP concentration in our study (5.5 pg/mL) was 80% lower than the mean observed in pre-feed levels in a controlled feeding study (30.4 pg/mL) in which participants were asked to refrain from eating charbroiled, smoked, or fried meat, suggesting our study group had low well-done meat consumption (Strickland et al., 2001; Strickland et al., 2002). However, the geometric mean (4.1 ng/g), range (<LOD to 10.4 ng/g), and percent of samples above the LOD (56%) of creatinine-adjusted urinary PhIP concentrations in our study were somewhat higher than those reported for Caucasian participants on unrestricted diets in a prior study in Los Angeles (geometric mean=1.18 ng/g, range=<LOD to 14.7 ng/g, and %>LOD=21%) (Kidd et al., 1999). The median (0.43 ng) and range (0.080-23 ng) of the mass of PhIP excreted in our approximately 8 to 12-hr samples were similar to the median (0.31 ng) and range (0.12–2.0 ng) reported in 24-hr urine samples in ten healthy volunteers in Japan who were on unrestricted diets (Ushiyama et al., 1991).

We observed high within-subject variability in the PhIP intake estimates across all methods (ICC of 0.20, 0.30, and 0.15 for FFQ, diary, and creatinine-adjusted biomarker, respectively), which will substantially attenuate observed ORs if a single measurement is used in an epidemiologic study. The ICC of 0.30 for PhIP intake based on the FFQ was similar to the ICC of 0.36 reported for two dietary interviews asking about fried meat intake over the prior year, administered 4 to 7 years apart (Knekt et al., 1994). We previously assessed the repeatability of the urinary biomarker 1-hydroxypyrene glucuronide (1-OHPG), an indicator of exposure to the dietary carcinogens polycyclic aromatic hydrocarbons (PAHs), in the same study population (Deziel et al., 2011). The repeatability was higher for 1-OHPG concentration (ICC=0.37) than that of PhIP (ICC=0.01), possibly because PAHs are found in a wider variety of foods. In contrast to the current study, we previously observed that creatinine-adjustment decreased the repeatability of the 1-OHPG concentration (ICC=0.15 for creatinine-adjusted 1-OHPG). These contradictory findings add to the discussion about the utility of and approach for creatinine adjustment (Barr et al., 2005). The large within-subject variability in consumption of PhIP-containing foods highlights the challenges in dietary exposure assessments and provides justification for the use of a repeated measure approach or development and use of longer-term biomarkers, such as protein adducts or hair PhIP (Bessette et al., 2009; Kobayashi et al., 2007).

We did not observe an association between any of the three exposure methods examined separately when the full study population was considered. A previous study observed a modest correlation between FFQ and diary-based estimates of PhIP intake (Pearson r=0.22, 95% confidence interval: 0.07–0.36) (Cantwell et al., 2004). Two studies were identified that compared an FFQ against biomarker measurements. In a study of 129 men, Kidd et al. observed that the odds of having detectable urinary PhIP was not different in individuals with higher frequencies of intake of various meats compared to those with lower frequency of intake (Kidd et al., 1999). This study only used a single urine sample to compare with the biomarker, and the FFQ did not ask participants to report on meat cooking method or degree of doneness. Kobayashi et al. observed a Spearman correlation coefficient of 0.47 (P=0.036) between PhIP intake based on an FFQ and PhIP measured in hair among 20 volunteers (Kobayashi et al., 2007).

We did observe a modest association between mean PhIP intake based on the FFQ and a metric that combined PhIP intake based on the diary and urinary PhIP measurements. If the biomarker-based and survey-based methods to assessing PhIP intake are indeed capturing different time periods or sources of exposure, using HCA biomarkers in combination with survey-based methods could potentially reduce some of the uncertainty in HCA intake estimates in epidemiologic studies (Freedman et al., 2010). For example, Peters et al. used urine mutagenicity in combination with an FFQ and observed a stronger relationship with cancer risk than when using either method alone (Peters et al., 2003).

We did not observe correlations between meat intake and urinary PhIP, though we had observed an association between meat intake and urinary 1-OHPG concentration in our previous analysis of PAHs (Deziel et al., 2011). This may be because PhIP formation is more specific to meat preparation methods that involve direct and/or efficient transfer of heat, such as broiling, grilling, or frying at high temperatures and/or long cooking times (Skog 2002). We expected a positive correlation between diary-based PhIP and urinary PhIP measurements in the full study population because these two methods reflected a similar time period. Furthermore, we observed such a correlation for the PAH benzo[a]pyrene in our previous analysis in this population (Deziel et al., 2011). We observed several participants with diary-based PhIP intake of 0 ng/g, but detectable urinary PhIP concentrations at concurrent collections. There are many plausible explanations for the lack of correlation. Firstly, the diary and database may not capture all foods contributing to exposure. While we focused only on meats, there have been some reports in the literature that certain non-meat foods, such as smoked cheeses (Naccari et al., 2009), beer (Manabe et al., 1993b), and wine (Manabe et al., 1993b), contain HCA, though their contribution to total exposure is likely minimal.

Secondly, in addition to only including meat-based foods, the FFQ, diary, and database did not include information on factors related to cooking preparation that can influence the concentrations of PhIP in food. For example, increasing the frequency of flipping meat during cooking can reduce the formation of PhIP (Salmon et al., 2000). Use of marinades may increase or decrease the formation of PhIP, depending on the components of the marinade, contributing to increased variability in HCA concentrations in foods (Busquets et al., 2006).

Thirdly, it is possible that the biomarker is capturing non-dietary exposures. PhIP has been isolated in non-dietary environmental samples, such as tobacco smoke (Manabe et al., 1991), cooking smoke (Thiebaud et al., 1995), and diesel exhaust (Manabe et al., 1993a). The relative contribution of these other sources to total exposure is likely minimal. Furthermore, our selection criteria aimed to exclude people with potentially high exposure to these other sources.

Finally, inter-individual differences in toxicokinetics may account for some of the variability in the relationship between PhIP intake estimated from the diaries relative to PhIP urine levels. For example, in two controlled feeding studies, participants consuming the same preparation of cooked meat had urinary PhIP concentrations that varied 2.7-fold (Strickland et al., 2002) and 4.2-fold (Reistad et al., 1997). Additionally, since urinary PhIP is actually reflecting the balance of activation and detoxification pathways, the levels of excretion of PhIP conjugates may be affected by other dietary constituents, such as the amount of fruits and vegetables consumed (Murray et al., 2001; Walters et al., 2004).

Strengths of this study included the multiple methods included and the repeated measures design. However, we were limited by a relatively small sample size, the infrequent consumption of PhIP containing foods, and the low detection rates of urinary PhIP. Our study could have been strengthened by incorporating a more robust analytical technique, such as liquid chromatography tandem mass spectrometry, potentially yielding increased detection rates in our urine samples and an improved ability to compare methods. Also, additional diary and urine collections may be needed to make a fair comparison, given the episodic nature of PhIP consumption. We recognize the collection and analysis of duplicate diet samples as another strategy for directly measuring PhIP dietary exposure, however, this approach was not feasible in the context of the current investigation due to cost and participant burden. Although duplicate diet has some of its own limitations (in addition to cost and burden, there is potential for bias in foods provided for the duplicate diet), it is considered the “gold standard” for dietary exposure and therefore would be a useful method to include in future studies.

Our objective was to compare commonly used approaches for estimating PhIP intake in epidemiologic studies. Among the methods tested, there is no clear gold standard, making the poor associations observed across the three methods difficult to interpret. In general, this lack of association may be an indication of the uncertainty with which these measures are made and the need for additional research to improve assessment methods. Alternatively, the lack of association may be a reflection of the different time windows of exposure being captured by each method, rather than a flaw in any single method. Even the timing of the concurrently collected urinary PhIP concentrations and the diary-based PhIP intake may not be optimal. The very short half-life of PhIP (less than 12 hr) suggests that foods consumed early in the day of collection may not be captured by the PhIP measured in the last evening void.

This study provides insight into the interpretation of previous epidemiologic findings, suggesting that reported risk estimates, using a single FFQ to estimate exposure, could be substantially attenuated due to relatively large errors in exposure classification. The current study highlights concerns about the repeatability of current methods for assessing the episodically consumed HCA PhIP. Incorporation of repeated measures or a combination of methods in future studies could address this issue. In addition, development of a well-validated, widely applicable, longer-term biomarker for assessing dietary exposure would be useful.