Polycyclic aromatic hydrocarbons: determinants of urinary 1-hydroxypyrene glucuronide concentration and risk of colorectal cancer in the Shanghai Women’s Health Study

Colorectal cancer cases and matched controls were selected from among participants in the Shanghai Women’s Health Study (SWHS), a prospective cohort of approximately 75,000 women enrolled between 1997 and 2000 [14]. Of the 65,574 women in the SWHS cohort who provided spot urine samples at enrollment, 347 were diagnosed with incident colorectal cancer through December 2005. One control was individually matched to each case based on age (±2 years), date of sample collection (within 1 month), menopausal status at sample collection, time of sample collection (morning or afternoon), and time interval since last meal (within two hours). Four of the colorectal cancer cases were subsequently determined to have been misdiagnosed; these cases and the corresponding matched controls were excluded from the analyses of colorectal cancer risk, leaving 343 cases (colon cancers, N=205; rectal cancers, N=138) and matched controls.

A total of 652 controls from the SWHS cohort were included in the analyses of determinants of urinary 1-OHPG concentration (347 controls from the colorectal cancer study, and 305 controls from a nested case–control study of gastric cancer for which samples were analyzed contemporaneously at the same laboratory).

The design and collection of data and specimens in the SWHS has been described [14]. Briefly, women between 40–70 years of age were recruited from selected urban communities in Shanghai, China from 1997 to 2000. All subjects provided written informed consent, and the study protocols were approved by the Institutional Review Boards of the National Cancer Institute, Vanderbilt University, and the Shanghai Cancer Institute. After obtaining informed consent, subjects completed a self-administered questionnaire and an in-person interview. Information was collected on demographic characteristics, active and passive cigarette smoking, diet and cooking practices, occupational history, and various other factors. Height and weight were measured at the time of the in-person interview.

Subjects were also asked to provide a spot urine sample, which was kept cold and processed within 6 hours for long-term storage. Another questionnaire was completed at the time of sample collection to obtain additional information about diet, smoking, and medication use within the previous week and the 24-hour period prior to urine collection.

Colorectal cancer cases were identified through linkage with the Shanghai Cancer Registry and the Shanghai Vital Statistics Unit records, and in biennial follow-up visits with participants. Information on the date of cancer diagnosis was collected, and diagnoses were verified by reviewing medical charts and diagnostic slides. Incident cases of colon and rectal cancers diagnosed through December 2005 were included in this analysis.

Measurements of urinary concentration of 1-OHPG were performed using immunoaffinity chromatography and synchronous fluorescence spectroscopy [13]. Assays were performed in batches of 20 including laboratory QC samples and blinded duplicate samples in each batch. The coefficient of variation (CV) for replicate measurements of urinary 1-OHPG concentration across batches using aliquots from a single quality control pool was 10.6%. In an analysis of blinded duplicate samples from 30 subjects, the Spearman rank correlation coefficient for paired 1-OHPG measurements was 0.82.

The assay limit of detection was 0.1 pmol/mL. For measurements below the limit of detection (7.6% and 7.3% of cases and controls, respectively), a value of 0.05 pmol/mL was assigned.

Conditional logistic regression analyses were performed to estimate odds ratios (ORs) and 95% confidence intervals (CIs) for risk of colorectal cancer in relation to creatinine-adjusted urinary 1-OHPG concentration, which was categorized as quartiles based on the distribution in controls. Statistical tests for trend were performed by modeling the within-category medians as a continuous parameter. We also evaluated colorectal cancer risk in relation to creatinine-adjusted urinary 1-OHPG concentration as a continuous variable; as in previous studies [15], a one unit change was defined as half the difference between the 25^th and 75^th percentile measurements in controls. Analyses were performed with and without adjustment for the following covariates: age (years); education level; smoking status; aspirin use; estimated fruit, vegetable, and folate intake; measured body mass index (BMI); and leisure time and occupational physical activity [in metabolic equivalent hours per week (MET-h/wk), as estimated in ref. [16]. Because intra-individual variability in 1-hydroxypyrene levels was modestly reduced in first morning void urine samples compared to 24-hour urine samples after adjustment for creatinine [17], we used creatinine-adjusted 1-OHPG concentrations in our main analyses. However, to evaluate whether correcting 1-OHPG measurements for creatinine concentration had any effect on the resulting risk estimates, we performed sensitivity analyses using 1-OHPG concentrations without correction for creatinine, and with creatinine concentration included as an independent variable in the statistical model, as recommended by Barr et al. [18]. Several analyses were performed to assess whether the association differed by time from sample collection to case diagnosis, including an analysis restricted to cases (and corresponding matched controls) diagnosed two or more years after sample collection, and analyses stratified by duration of follow-up (below/above the median time of 3.8 years from sample collection to diagnosis, and above the 75^th percentile of follow-up time to diagnosis of 5.6 years). We also conducted analyses stratified by tumor location (colon vs. rectal) and menopausal status.

As secondary analyses, we evaluated urinary 1-OHPG concentration as a dependent variable in relation to selected exposures that were suspected a priori to contribute to PAH exposure. These analyses were conducted among a combined set of control subjects (N=652). Measurements of 1-OHPG concentration were corrected for creatinine concentration, and data were natural log-transformed to achieve a normal distribution. Results are reported as the geometric mean (GM) and 95% CI within each exposure category. Self-reported exposures from the baseline questionnaire that were evaluated included active and passive smoking status, cooking fuel type, cooking ventilation conditions, and food preparation methods. We also evaluated smoking and food preparation methods in the last week and last 24 hours prior to sample collection (as reported on the sample collection questionnaire). Independent t-tests were used to evaluate each variable separately in relation to 1-OHPG concentration. We also performed multivariate analyses for exposures reported on the baseline questionnaire (Model 1) and exposures reported for the last 24 hours on the sample collection questionnaire (Model 2); both models were adjusted for age in years, level of education, measured BMI, and study sample (i.e., controls from the colorectal cancer study or gastric cancer study).

Colorectal cancer cases and matched controls were similar in level of education, marital status, fruit and vegetable consumption, BMI, folate intake, and physical activity (Table 1). Smoking was uncommon in general among study participants, but was slightly less common among cases than among controls (2% and 5%, respectively). Urinary levels of 1-OHPG, with or without adjustment for creatinine, were not significantly different between cases and controls. The geometric means (95% CIs) of creatinine-adjusted 1-OHPG concentration in urine were 0.15 (95% CI 0.13-0.17) and 0.16 (95% CI 0.14-0.18) μmol/mol creatinine for cases and matched controls, respectively (P = 0.4, paired t-test).

No statistically significant differences in risk of colorectal cancer by creatinine-adjusted 1-OHPG concentration were observed; adjustment for selected covariates did not have any appreciable effect on risk estimates (Table 2). Results were similar when husband’s smoking status was included in the multivariate analysis, and when smoking variables were not included in the statistical model (data not shown). Risk of colon cancer, but not rectal cancer, appeared to decrease slightly with increasing levels of creatinine-adjusted 1-OHPG; however, this trend was not statistically significant when 1-OHPG was analyzed as a continuous variable. Risk estimates did not change appreciably after exclusion of cases (and corresponding matched controls) diagnosed within two years of sample collection, and we did not observe differences in risk by menopausal status or duration of follow-up.

The geometric mean (95% CI) of creatinine-adjusted 1-OHPG concentrations by categories of selected exposures from the baseline questionnaire are reported in Table 3. Among 652 controls, the median (25^th-75^th percentile) urinary 1-OHPG levels with and without adjustment for creatinine were 0.21 (0.11-0.38) μmol/mol creatinine and 2.01 (0.95-4.09) pmol/mL, respectively. Subjects who reported ever smoking had significantly higher 1-OHPG levels than non-smokers (P = 0.03). We observed a borderline significant association between husband’s smoking status and 1-OHPG concentration; women whose husbands were current smokers had higher levels of 1-OHPG than women whose husbands never smoked (P = 0.05). Results for passive smoking were similar when we restricted to women who never smoked. Mean 1-OHPG levels were approximately twice as high among subjects who reported using coal for cooking at their current residence compared to subjects who used gas or other types of fuel, though this difference did not achieve statistical significance (P = 0.08). Relative to subjects who ever ate stir-fried meats, those who did not had higher 1-OHPG levels; this difference was borderline significant (P = 0.05).

We also evaluated creatinine-adjusted 1-OHPG concentration in relation to reported recent exposures from the sample collection questionnaire (Table 4). Similar to the results of the baseline questionnaire analysis, participant smoking was associated with elevated 1-OHPG levels. We observed statistically significant trends of increasing 1-OHPG concentration by number of cigarettes smoked in the last week (P = 0.03) and the last 24 hours (P = 0.04) prior to sample collection. Subjects who reported eating fried dough (yóutiáo) in the last 24 hours also had significantly higher 1-OHPG levels than subjects who did not eat these foods (P = 0.01). No other notable differences in 1-OHPG concentration in relation to recent exposures were observed.

Results of multivariate analyses of creatinine-adjusted 1-OHPG concentration in relation to selected characteristics from the baseline questionnaire (Model 1) and reported exposures within the last 24 hours from the sample collection questionnaire (Model 2) are shown in Table 5. As in the bivariate analyses, participant smoking status was a statistically significant determinant of 1-OHPG concentration in both multivariate models. In Model 1, husband’s smoking status was associated with a slight, non-significant elevation in 1-OHPG concentration. Other factors in Model 1 that were associated with elevated levels of 1-OHPG included eating foods that were cooked well done (43% higher geometric mean relative to subjects who did not eat “browned” foods) and use of coal as a cooking fuel (63% higher geometric mean relative to subjects who used gas or other types of cooking fuel). In Model 2, recent consumption of fried dough products continued to be a significant determinant of 1-OHPG concentration after covariate adjustment (P = 0.01).

Results of all analyses of colorectal cancer risk and determinants of 1-OHPG levels were similar when 1-OHPG concentration was not corrected for creatinine and when creatinine was included as a covariate in the statistical model (data not shown). We also performed analyses of colorectal cancer risk in relation to those exposures that were associated with elevated urinary 1-OHPG levels (e.g., personal smoking status, husband smoking status, consumption of fried dough products, and cooking with a coal stove), and no statistically significant associations were observed (not shown).

The large sample size, prospective collection of urine samples, and availability of data on potential confounding factors were strengths of this study. Because samples were collected prospectively and results were similar when cases diagnosed within two years of sample collection were excluded, it is unlikely that early disease processes would have affected 1-OHPG measurements. The availability of prospectively collected data on covariates minimized potential recall bias and allowed us to control for potential confounding factors such as age, BMI, physical activity, and fruit and vegetable intake in our analyses.

There were also several limitations in this study. Urine was collected as a spot sample for practical reasons, though the interpretation of urinary creatinine levels used for the correction to ≥24 hour urine output may be problematic. To address this issue, we performed sensitivity analyses without correction for creatinine and with adjustment for creatinine concentration as a covariate in the statistical model; results were similar to what was observed in the main analyses.

As with most prospective investigations involving a single pre-diagnostic specimen, it is possible that urinary levels of 1-OHPG may not have reflected long-term PAH exposure status or exposure levels during the etiologically relevant period. In our analyses among controls, we identified associations between several reported sources of exposure from the baseline and sample collection questionnaires and urinary concentration of 1-OHPG, suggesting that 1-OHPG levels may be a reasonable reflection of both usual and recent exposure to PAHs. It should be noted that when we evaluated colorectal cancer risk in relation to these sources of PAH exposure, our findings were similarly null. The median time from urine sample collection until diagnosis of colorectal cancer in this study was 3.8 years; continued follow-up of the cohort may be useful to assess relations between long-term PAH exposure and colorectal cancer risk.

Most previous studies of determinants of 1-OHPG concentration have been based on relatively small sample sizes, and the inclusion of a large number of healthy subjects (N = 652) was a strength of the present study. However, despite the large sample size, some of the reported exposures were relatively uncommon in this population. Also, there was somewhat less heterogeneity of 1-OHPG levels relative to other highly exposed populations [25, 26], which may have limited our ability to detect associations with colorectal cancer and to evaluate determinants of PAH exposure. We were unable to assess ambient exposure to PAHs in this study, an important determinant of exposure in previous studies [30] that could possibly explain the high and relatively homogeneous levels of 1-OHPG in our study population.