Dietary Cryptoxanthin and Reduced Risk of Lung Cancer: The Singapore Chinese Health Study¹

Vegetables and fruit are associated with a reduced risk of lung cancer (1). The possible protective compounds in vegetables and fruit may include carotenoids and other antioxidant vitamins. The main carotenoids are α-carotene, β-carotene, β-cryptoxanthin, lycopene, lutein, and zeaxanthin (2). They are potent quenchers of free radicals originating from cigarette smoke, by far the single most important etiological factor for lung cancer (3). Historically, β-carotene had drawn a great deal of attention within the cancer research community, especially after the publication of the classical review article by Peto et al. in 1981 (4), who hypothesized that the association between vitamin A and reduced cancer risk is because of the portion of the vitamin A derived from β-carotene. Although prospective observational studies have consistently shown that people with high baseline serum β-carotene concentrations exhibit reduced risk of lung cancer relative to those with low β-carotene levels (5, 6, 7, 8, 9), three large, double-blind, placebo-controlled intervention trials have failed to demonstrate any observable reduction in lung cancer risk among subjects after prolonged (4–12 years) administration of high-dose β-carotene supplementation (10, 11, 12).

A possible explanation for the discrepancy in the β-carotene-lung cancer association between observational studies and intervention trials is that the inverse association in the former is an indirect one, attributable merely to association of β-carotene ingestion with some truly protective dietary habit(s) and/or other nondietary component(s). β-Carotene is highly correlated with other main carotenoids because they share the same food sources (vegetables and fruit). Data on specific carotenoids other than β-carotene and lung cancer risk are scarce, because comprehensive information on these noncarotene carotenoids have been available relatively recently. We reported recently statistically significant, inverse relationships of prediagnostic serum levels of specific carotenoids (α-carotene, β-carotene, β-cryptoxanthin, lycopene, and lutein/zeaxanthin) with lung cancer incidence in a prospective cohort of middle-aged and older Chinese men in Shanghai, China (13). Whereas the effects of most carotenoids (α-carotene, β-carotenes, and lycopene) were explained by cigarette smoking, the inverse association between β-cryptoxanthin and lung cancer risk remained statistically significant after adjustment for cigarette smoking and other potential confounders (13). In the present study, we assessed the effects of specific dietary carotenoids, vitamins A, C, and E, and folate on lung cancer risk within the Singapore Chinese Health Study, a residential cohort of >60,000 middle-aged and older Chinese men and women in Singapore.

The design of the Singapore Chinese Health Study has been described (14). Briefly, the cohort was drawn from permanent residents or citizens of Singapore who resided in government-built housing estates (86% of the Singapore population reside in such facilities). The eligible age range for cohort enrollment was 45–74 years. We restricted study subjects belonging to the two major dialect groups of Chinese in Singapore: the Hokkiens and the Cantonese. Between April 1993 and December 1998, 63,257 subjects (∼85% of eligible subjects) were recruited. The gender-dialect breakdown is as follows: 15,617 (24.7%) Hokkien men, 18,356 (29.0%) Hokkien women, 12,342 (19.5%) Cantonese men, and 16,942 (26.8%) Cantonese women. Among cohort participants, 865 reported a history of cancer at baseline and, thus, were excluded from the present data analysis. The present study included 62,392 subjects. The study was approved by the Institutional Review Boards of the University of Southern California and the National University of Singapore.

At recruitment, an in-person interview was conducted in the home of the subject by a trained interviewer using a structured questionnaire. Details of the validated, 165-item food frequency questionnaire assessing usual intake pattern during the past 12 months have been described (14). Main contributors to dietary carotenoids and vitamins in our study population include green- and orange-colored vegetables, yellow- and orange-colored fruit, and their juices. Specifically, the main contributors to β-cryptoxanthin are papaya, tangerine, orange, mango, and their juices. In addition to these food items, green vegetables are main contributors to vitamin C. The questionnaire also requested information on demographics, lifetime use of tobacco (cigarettes and water-pipe), current physical activity, reproductive history (women only), occupational exposure, medical history, and family history of cancer.

Identification of incident cancer cases and deaths among cohort members was accomplished through record linkage of cohort files with databases of the nationwide Singapore Cancer Registry (15) and the Singapore Registry of Births and Deaths. Migration out of Singapore, especially among housing estate residents, is negligible (Department of Statistics, Singapore Ministry of Trade and Industry, 1997). As of December 31, 2000 (an average of 5.3 years of follow-up), 2815 cohort participants who were free of cancer at baseline had developed cancer, including 482 lung cancers.

For each subject, person-years of follow-up were counted from the date of recruitment to the date of diagnosis of lung cancer, death, or December 31, 2000, whichever occurred first. The person-year distribution of the entire cohort was used as an internal standard in the computation of age-adjusted incidence rates of lung cancer in both sexes.

Nutrient levels for individual subjects were computed from nutrient contents of raw and cooked food items in the Singapore Food Composition Table (14). To adjust for energy intake, all of the food groups and nutrients were expressed in weight unit/1000 kcal. Only 3.2% (n = 1974) of cohort participants used vitamin supplements on a regular basis. Study results were similar with or without the inclusion of supplemental nutrients. Therefore, all of the results presented in this report were derived from food-based nutrients only.

We used analysis of covariance methods (16) to calculate the means and SEs³ of dietary variables within specific stratum of cigarette smoking status, with adjustment for age and sex. The proportional hazards regression methods (17) were used to examine the associations between dietary exposure levels and lung cancer risk. The associations were measured by RRs, and their corresponding 95% CIs and Ps. Study subjects were grouped into tertiles, quintiles, or dectiles based on the distribution of the entire cohort. The linear trend tests for exposure-disease associations were based on median values within each quintile. RRs also were computed for individual antioxidants that were associated with changes between the 10th and 90th percentiles of intakes. Analyses were performed for men and women separately and for both sexes combined. Because all of the studied exposure-lung cancer risk associations were comparable between men and women, all of the presented results were for both sexes combined with adjustment for gender.

In all of the analyses, we adjusted for smoking by including covariate terms for average number of cigarettes smoked per day, number of years of smoking, and number of years since quitting smoking for former smokers at baseline. Other potential confounders included in the multivariate Cox proportional hazards models were age at baseline, sex, dialect group (Hokkiens or Cantonese), year of interview (1993–1995 or 1996–1998), level of education (primary school or less, middle school, or A-level school or university), and BMI (<20, 20–<24, 24–<28, or ≥28 kg/m²).

We used the regression calibration method (18) to adjust for measurement errors in dietary assessment, and computed calibration-adjusted risk estimates for lung cancer in relation to individual dietary factors. To reduce the right-skewed distribution of nutrient values, original dietary intake values for each individual were replaced by the median values within quintile levels. The regression coefficients derived from all of the calibration study subjects were used in computing the calibration-adjusted risk estimates for lung cancer (14).

Consumption of fruit, vegetables, and antioxidants was inversely associated with self-reports of cigarette smoking levels in the present study (see “Results” section below). Therefore, residual confounding of smoking remains a possible explanation for our observed β-cryptoxanthin-lung cancer association. We used various measurement error models described in Stram et al. (19) to adjust for residual confounding of cigarette smoking on the dietary β-cryptoxanthin-lung cancer association. The models of Stram et al. require specification of 3 statistical parameters: (a) the correlation (r_a) between self-reported and true (biologically relevant) smoking exposure in smokers; (b) the (negative) correlation (r_b) assumed between true smoking exposure and the nutrient (e.g. β-cryptoxanthin) of interest in smokers; and (c) the shape of the observed dose-response model relating self-reports of smoking exposure to lung cancer risk in current smokers. A conservative model that assumes a correlation coefficient (r_a) of 0.55 between self-reported and biologically relevant doses of cigarette smoking to the lung was used in the present analysis. The value 0.55 is approximately the mid-range of published correlations between serum cotinine and self-reports of smoking level by smokers, and Stram et al. (19) treat this as a reasonable lower bound of r_a. We derived r_b by dividing the observed correlation between self-reports of smoking and β-cryptoxanthin by 0.55 (i.e., r_a), which corrects for attenuation because of measurement errors in reported smoking (the value −0.12 was obtained in this fashion for β-cryptoxanthin). Finally (point “c” above), we considered linear and linear-quadratic dose-response relations between number of cigarettes smoked per day and lung cancer risk as described by Doll and Peto (20). We also considered a set of modified linear and linear-quadratic risk models that assumed a RR of 10 for lung cancer at a dose level of 20 cigarettes per day, which more closely resembled the observed smoking-risk association in the present study. The risk models of Doll and Peto (20) assumed a RR of 60 for lung cancer at a dose level of 40 cigarettes per day.

The Stram et al. (19) method shows that when self-reported rather than the true smoking exposure is adjusted for in the statistical analysis, residual confounding would lead to a non-null RR between smokers reporting “high” versus “low” levels of the dietary nutrient of interest when the diet-risk association is actually null. This TRR was used to correct for residual confounding in our analyses by dividing the observed RR by TRR to form a corrected RR estimate. The upper and lower boundaries of the CIs for the observed RR are corrected similarly. Here, high and low intakes of dietary β-cryptoxanthin were defined as the 90th and 10th percentiles of dietary β-cryptoxanthin, respectively.

Statistical computing was conducted using SAS version 8.2 (SAS Institute Inc., Cary, NC) and Epilog windows version 1.0 (Epicenter Software, Pasadena, CA) statistical software packages. All of the Ps quoted are two-sided.

As of December 31, 2000, 482 cohort subjects (330 males and 152 females) developed incident lung cancer. Among them, 115 (24%) were squamous-cell cancers, 179 (37%) were adenocarcinomas, 48 (10%) were small-cell cancers, 31 (6%) were large-cell cancers, and 12 (2%) were other cell-type cancers. Fifty-two (11%) patients had lung cancer of unknown cell type. We were unable to retrieve pathological reports to confirm the histology on 45 (9%) cases. The mean age at cancer diagnosis was 63 years for both sexes. The mean time interval between cohort enrollment and cancer diagnosis was 3.3 years (range, 1 month to 7.7 years).

The cohort accrued 329,560 person-years by the end of 2000. Incidence rate of lung cancer in males was ∼3-fold higher than that in females after adjustment for age (230 versus 82 per 100,000 person-years). At baseline, 35% and 23% of the men were current and former cigarette smokers, respectively. The corresponding figures in women were 6% and 3%, respectively.

Compared with never-smokers, current smokers experienced an ∼5-fold increased incidence rate of lung cancer (315 versus 69 per 100,000 person-years). Risk of lung cancer increased monotonically with increasing number of cigarettes smoked per day and number of years of smoking. Former smokers had an incidence rate of lung cancer intermediate between never- and current smokers. Number of years since quitting smoking habits was associated with monotonically reduced lung cancer risk (Table 1).

Compared with never-smokers, current smokers consumed lowest levels of fruit and vegetables, and carotenoids and antioxidant vitamins, followed by former smokers, after adjustment for age, sex, and total energy. Among current smokers, intake levels of fruit, vegetables, and antioxidants decreased with increasing number of cigarettes smoked per day (Table 2).

Level of education was inversely related to lung cancer risk in men, but not in women. After adjustment for age and smoking, men with a high school or college education had a RR of 0.59 (95% CI, 0.41–0.85) compared with those with no formal schooling. Low BMI (weight in kg divided by height in m²) was associated with elevated risk of lung cancer, particularly in men (P for trend = 0.01) after adjustment for age and smoking.

The relationships of dietary antioxidants and folate with risk of lung cancer are presented in Table 3. Before adjustment for cigarette smoking, lung cancer incidence was significantly lower for individuals who consumed higher levels of β-cryptoxanthin, vitamin C, and folate than those with lower intake levels of these nutrients, respectively (all of the Ps for trend <0.01). Adjustment for smoking weakened these inverse associations. The smoking-adjusted β-cryptoxanthin-lung cancer association remained statistically significant (P for trend = 0.02); a 30% reduction in risk of lung cancer was associated with a 366-μg/1000 kcal increase in consumption of β-cryptoxanthin (the range from the 10th to the 90th percentile). The inverse associations between intake of vitamin C and folate, and lung cancer risk were of borderline statistical significance after adjustment for smoking (Table 3). Dietary α-carotene, β-carotene, lycopene, lutein/zeaxanthin, and vitamins A, C, and E were not significantly associated with risk of lung cancer with or without adjustment for cigarette smoking (Table 3).

Table 3 also shows the calibration-adjusted RR of lung cancer in relation to dietary antioxidants and folate. Adjustment for measurement errors in dietary assessment strengthened the relationships between nutrients and lung cancer risk (Table 3).

The associations between dietary antioxidants and lung cancer risk within smoking strata are presented in Table 4. High levels of dietary β-cryptoxanthin and vitamin C were associated with statistically significantly reduced risk of lung cancer in current smokers, but not in never- or former smokers. Among current smokers, a statistically significant, 35–40% reduction in risk of lung cancer was associated with increasing consumption of β-cryptoxanthin or vitamin C from the 10th to the 90th percentile level. After adjustment for dietary measurement errors, 10th to 90th percentile levels of β-cryptoxanthin and vitamin C were associated with roughly 50–60% reduction in risk of lung cancer (both Ps for trend <0.05). All of the other dietary antioxidants, including α-carotene, β-carotene, lycopene, lutein/zeaxanthin, vitamins A and E, and folate, were not associated with risk of lung cancer in never-, former, or current smokers (Table 4).

β-Cryptoxanthin was highly correlated with vitamin C (Pearson correlation coefficient = 0.68). The smoking-adjusted inverse association between dietary vitamin C and lung cancer risk was largely explained by β-cryptoxanthin; the β-cryptoxanthin-adjusted RR for lung cancer associated with an 83-mg/1000 kcal increase in intake of vitamin C was 0.98 (95% CI, 0.65–1.48). Within each tertile level of β-cryptoxanthin, increasing intake level of vitamin C was not associated with a reduced risk of lung cancer (Table 5). On the other hand, a 27% reduction in risk of lung cancer was associated with a 366-μg/1000 kcal increase in consumption of β-cryptoxanthin after adjustment for vitamin C (RR, 0.73, 95% CI, 0.48–1.09). Within each tertile level of vitamin C, there was a monotonic reduction in lung cancer risk with increasing β-cryptoxanthin intake (Table 5).

Table 6 presents the TRRs for lung cancer associated with dietary β-cryptoxanthin when adjustments were made to account for the residual confounding effect of smoking because of inherent discrepancy between self-reported and biologically relevant doses to the lung. A conservative model assuming a high degree of such discrepancy was used (19). Assuming a linear dose-response relation between number of cigarettes per day and lung cancer risk, high dietary β-cryptoxanthin had statistically borderline significant 30–40% reduction in lung cancer risk (Table 6). When we assumed a linear-quadratic dose-response relation, the protective effect of dietary β-cryptoxanthin on lung cancer diminished; about 15–25% reduction in risk of lung cancer was seen for the highest versus lowest 10th percentile of β-cryptoxanthin (Table 6).

Risk estimates by histological classifications were relatively unstable because of small sample sizes upon stratification. Nonetheless, we examined the associations between dietary nutrients and risk of lung cancer according to histological categories and did not detect any meaningful variations in disease risk by histology (data not shown).

Of the 482 lung cancers, 50 (10%) were diagnosed based on radiography and clinical symptoms. We repeated all of the analyses on histologically confirmed cases only. Results were similar to those based on all of the lung cancer cases (data not shown).

Incident lung cancer cases diagnosed within a short time period after entry into the study might have changed their dietary habits already (because of disease symptoms) at the time of baseline interview. Therefore, we repeated all of the analyses after excluding cases diagnosed during the first year of follow-up and the corresponding person-year contributions from all of the study subjects. Results were similar to those base on the entire data set (data not shown).

The present study demonstrates a statistically significant, inverse association between dietary β-cryptoxanthin and lung cancer risk that is independent of cigarette smoking at the level of detail measured in this study. Limited data from nonhuman in vivo experimental studies showed that β-cryptoxanthin has anticarcinogenic effect on the lung. After initial treatment with a tobacco carcinogen, 4-(methylnitrosamino)-1-(3-pyridyl)-1-butanone, mice given a concentrated Mandarin (Citrus unshiu Marc.) juice (3.9 mg of β-cryptoxanthin per 100 g of juice) as drinking water had a statistically significant 26% reduction in incidence of lung tumor compared with 4-(methylnitrosamino)-1-(3-pyridyl)-1-butanone-treated animals given water only (65% versus 91%; Ref. 21). A number of epidemiological studies have examined the association between serum or dietary β-cryptoxanthin levels and lung cancer risk (9, 13, 22). In a cohort of middle-aged and older men in Shanghai, China, we found a statistically significant, inverse association between prediagnostic serum level of β-cryptoxanthin and lung cancer risk; men in the highest quartile of serum β-cryptoxanthin at baseline experienced a statistically significant 50% reduction in risk of lung cancer compared with those in the lowest quartile after adjustment for cigarette smoking (13). Similar results were noted in a prospective study in the United States (9). In a cohort of 58,279 middle-aged and older men in the Netherlands, a statistically significant, inverse association between dietary β-cryptoxanthin and risk of lung cancer was observed, particularly among current smokers (22). An experimental study also showed an inverse association between plasma β-cryptoxanthin and a marker of oxidative DNA damage (8-OHdG). Among 47 subjects fed controlled diets for 14 days, highly significant correlation was noted between changes in β-cryptoxanthin and 8-OHdG levels during the course of the intervention. On the other hand, no association was seen between changes in carotenes and 8-OHdG levels in those study subjects (23).

Random measurement errors in dietary assessment bias any estimates of diet-disease risk associations toward the null. In the present study, the measurement error-corrected RR associated with a 10th-90th percentile difference in β-cryptoxanthin intake was 0.50. Among current smokers, this figure was 0.36. Both RRs were statistically significant.

To our knowledge, this study was the first to attempt correction for residual confounding by self-reported smoking in diet-cancer associations using statistical modeling techniques. Our results suggest that β-cryptoxanthin is an independent protective factor for smoking-related cancer. The modeling results are corroborated by the findings in lifelong nonsmokers whose reduction in risk was comparable in magnitude as the current smokers (Table 4). However, one should be aware of the inherent limitations in all of the observational studies including ours. β-Cryptoxanthin could be a mere marker for some as-yet unidentified agent(s) or healthy lifestyle factors that exert protective effects on lung carcinogenesis.

A stronger association for lung cancer with β-cryptoxanthin than with β-carotene was noted in the present study. These results are consistent with our previous findings of a null association between serum β-carotene and lung cancer risk after adjustment for cigarette smoking in Chinese men in Shanghai (13). The present study data also are in line with those from three large intervention trials that failed to show any protection against lung cancer development after β-carotene supplementation (10, 11, 12), suggesting that β-carotene may not be a chemopreventive agent for lung cancer prevention.

A number of epidemiological studies have examined the relationships for lung cancer with dietary α-carotene, lycopene, and lutein/zeaxanthin. Some studies noted an inverse association for lung cancer with α-carotene (1, 24, 25), whereas others did not (22, 26, 27). Among six studies that examined the association between dietary lutein/zeaxanthin intake and lung cancer risk (1, 22, 24, 25, 26, 27), only two found a statistically significant inverse association (24, 25). None of those previous studies reported a significant association between dietary lycopene and lung cancer risk (1, 22, 24, 25, 26, 27). Our results are generally consistent with those reported previously. After adjustment for cigarette smoking and other confounding factors, a null association for lung cancer with dietary α-carotene, lycopene, or lutein/zeaxanthin was seen in the present study.

In a previous study, we observed a threshold effect of serum retinol level on lung cancer risk in middle-aged and older men in Shanghai who possessed a serum retinol level only 50–75% that of comparable western populations (13). In that study, men in the 2nd-4th quartiles of serum retinol (i.e., ≥40 μg/dl) had a 40% reduction in risk of lung cancer compared with those in the lowest quartile (13). It is not surprising to us that the present study failed to detect a protective effect of vitamin A on lung cancer. Our study population shows comparable vitamin A intake level as people in the United States (5014 versus 4960 IU/day; Ref. 28). This null association is consistent with results of previous studies in well-nourished western populations (5, 28, 29, 30, 31).

A number of epidemiological studies have found an inverse association for lung cancer with dietary (22, 28, 32) but not serum vitamin C levels (8, 9, 31). In a cohort study in the Netherlands, male current smokers in the high vitamin C category showed a statistically significant 45% reduction in risk of lung cancer compared with their counterparts in the low intake category (22). In the present study, we observed a similar magnitude in risk reduction for lung cancer for high versus low vitamin C intake in current smokers. However, this inverse vitamin C-lung cancer association could be completely explained by dietary consumption of β-cryptoxanthin. However, we recognize that it is inherently difficult in epidemiological analyses to determine which of two highly correlated variables (vitamin C or β-cryptoxanthin) is directly (as opposed to indirectly) associated with a risk reduction.

Numerous epidemiological studies have examined the association between serum or dietary vitamin E and lung cancer risk (5, 8, 9, 13, 22, 33, 34). In general, null or weak inverse associations were observed. One intervention trial also failed to show any protective effect of vitamin E supplementation on lung cancer development (10). Consistent with most of the previous studies, the present study did not observe any effect of dietary vitamin E on lung cancer risk.

Folate may have a role in carcinogenesis, given its involvement in DNA synthesis, methylation, and repair (35). A number of epidemiological studies have examined the association between serum or dietary folate and lung cancer risk (22, 27, 32, 36). An inverse folate-lung cancer association was found in some studies (22, 32), but not in others (27, 36). In a cohort of middle-aged and older men in the Netherlands, high dietary folate intake was associated with a statistically significant 40% reduction in risk of lung cancer. This association was independent of cigarette smoking and other dietary antioxidants including vitamin C and β-cryptoxanthin (22). In contrast, another prospective study failed to detect an inverse association between baseline serum folate levels and lung cancer risk in a cohort of men in Finland after adjustment for cigarette smoking (36). Consistent with results of the latter, the present study found that the inverse association between dietary folate and lung cancer risk could be explained by cigarette smoking.

In summary, our study demonstrates a statistically significant inverse association between dietary β-cryptoxanthin and lung cancer development, independent of cigarette smoking, among middle-aged and older Chinese men and women in Singapore. The protective effect of β-cryptoxanthin against lung cancer was mainly confined to current smokers. The present study gives additional credence to prior experimental and epidemiological data in support of the hypothesis that dietary β-cryptoxanthin is a chemopreventive agent for lung cancer in humans.

We thanks Siew-Hong Low of the National University of Singapore for supervising the field work of the Singapore Chinese Health Study.