Dietary Advanced Glycation End-products (AGE) and Risk of Breast Cancer in the Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial (PLCO)

Breast cancer is the most commonly diagnosed cancer and second leading cause of cancer-related death among women in the United States (1). The chronic triggering of an inflammatory response is a biologic process that is recognized to promote carcinogenesis (2–4) and the presence of circulating inflammatory biomarkers promotes tumor development and progression (5). Advanced glycation end-products (AGEs) are complex compounds formed by the irreversible glycation of proteins or lipids with reducing sugars (6–8). The adverse biological effects of AGEs are propagated through the direct effects on tissues or activation of the receptor for AGE (RAGE), inducing oxidative stress and chronic inflammation (9–13). AGEs occur naturally in the body and are also contained in food products (6, 8). Food preparation and processing can influence the formation of AGEs. High AGE levels have been observed in foods prepared using high temperatures (such as frying, grilling, or roasting) or cooked for prolonged periods of time (6, 8) whereas raw fruits and vegetables generally have low AGE content (6). There is a growing body of evidence to suggest a role of AGEs in cancer development and progression (14, 15). High levels of AGEs have been detected in serum of women with breast cancer compared with healthy women (16, 17). Also, a differential effect of AGE on breast cancer subtypes by hormone receptor status has been hypothesized (16, 18).

While there is some evidence in preclinical experimental models, there is a paucity of data on dietary AGE intake in relation to breast cancer risk among human populations. We examined the association between dietary AGE intake and breast cancer risk using data from the large prospective Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial (PLCO). We also examined whether the relationship differed by race/ethnicity, fruit and vegetable intake, obesity status, in situ versus invasive breast cancer, and breast cancer hormone receptor status. We hypothesize that increasing levels of dietary AGE intake are positively associated with breast cancer risk.

Our study utilized data from the intervention arm of PLCO, a randomized controlled trial designed to assess the efficacy of cancer screening modalities in the United States. Enrollment occurred from November 1993 to September 2001 across 10 screening centers. The study enrolled over 78,000 women aged between 55 and 74 years who were randomized into an intervention or control arm (Supplementary Fig. S1). Participants in the intervention group received chest X-ray, flexible sigmoidoscopy, CA 125, and transvaginal ultrasound screenings at regular intervals, while women in the control arm received their usual medical care. Participants were excluded from the trial if they were undergoing treatment for cancers (except for basal or squamous cell melanoma), enrolled in other clinical trials, or unable to provide informed consent (19). Study participants were followed-up until December 31, 2009. The study was conducted in accordance with the U.S. Common Rule, the study protocol was approved by the Institutional Review Board of the NCI (Bethesda, MD) and screening centers, and written informed consent was obtained from all study participants.

Participants enrolled in the intervention arm were requested to complete baseline questionnaires (BQ) and dietary questionnaires (DQX) administered at baseline. The BQ obtained information on demographics, height and weight from which body mass index (BMI) was calculated, smoking, and medical history. Usual dietary intake in the previous year was captured through the DQX including information on meat preparation practices and vitamin/supplement intake. The DQX is a food frequency questionnaire (FFQ) developed for the PLCO, which solicits information on intake frequency on 137 food items and estimated portion sizes of 77 food items (20). A separate diet history questionnaire (DHQ) was introduced 5 years into the trial to both the intervention and control arms. The DHQ solicited dietary information in the previous year on intake frequency of 114 food items and estimated portion sizes of 109 food items. We utilized only data from the intervention arm to ensure consistency in dietary assessment as well as control for potential confounding effects of the intervention. Utilizing this group of women is recommended by the PLCO to promote uniformity of the dietary information obtained. Nutrient intake was calculated on the basis of DQX responses using the U.S. Department of Agriculture's pyramid food group and national dietary database (21). We excluded women with a personal history of cancer except nonmelanoma skin cancer, incomplete BQ, and invalid DQX (extreme energy intake, missing ≥ 8 FFQ responses, and incomplete DQX) at baseline (Supplementary Fig. S1). Exclusions were made for participants with missing covariates on: BMI (n = 253), age at menopause (n = 200), use of oral contraceptives (OC; n = 13), age at menarche (n = 19), family history of breast cancer (n = 171), use of postmenopausal hormones (PMH; n = 14), oophorectomy (n = 3), hysterectomy (n = 4), number of pregnancies (n = 39), number of live births (n = 2), education (n = 13), marital status (n = 7), age at first birth (n = 67), and hours spent in vigorous activity (n = 111). The final analytical sample contained 27,464 participants from which 1,592 breast cancer cases were diagnosed during the follow-up period. In analyses examining the risk of specific breast cancer subtypes, women who had missing information on breast cancer hormone receptor status were excluded (n = 378).

A published database containing N^ε-carboxymethyl-lysine (CML) content of 549 foods commonly consumed in the northeastern metropolitan region of the United States was used to assign CML-AGE values to each food item on the DQX (6). In the database, AGE content was estimated using enzyme-linked immunosorbent assay (ELISA) based on monoclonal anti-CML antibody (6). The CML-AGE content has been used in previous studies to estimate dietary intake and is the more stable and commonly measured of all the AGEs (22). CML-AGE values were appropriately assigned to match the food items in the DQX (Fig. 1) as has been done in a previous study (15). Where CML-AGE values for foods were not available in the database, values of similar foods or the average CML-AGE of foods within the food group were used. For example, a DQX response of canned tomatoes was assigned CML-AGE values of tomato sauce, while a DQX response of fresh plums was assigned the average CML-AGE value of all fruits in the database. For foods with multiple preparation methods, CML-AGE values in the database were averaged. For example, CML-AGE of “beef steak, pan fried w/olive oil;” “beef steak, strips, stir fried with 1 T canola oil, 15 minutes;” and “beef steak, strips, stir fried without oil, 7 minutes” were averaged to create a value for pan-fried steak. Where CML-AGE values of mixed foods were not available, values of individual food components were summed to create a CML-AGE value for the mixed dish. Total CML-AGE values were adjusted for energy intake using the energy density method and then categorized into quintiles for the main analyses and tertiles for the stratified analyses, with the lowest quintile/tertile representing the lowest intake level and serving as the referent group in regression analyses.

Potential confounders were identified through extensive literature review on breast cancer risk factors. Confounders identified include age (years), BMI (kg/m²; <18.5, 18.5–<25, 25–<30, or ≥30), hours spent in vigorous activity such as swimming or brisk walking (none, <1 hour/week, 1 hour/week, 2 hours/week, 3 hours/week, or 4+ hours/week), race/ethnicity [non-Hispanic white (NHW), non-Hispanic black (NHB), or other races/ethnicities), marital status (married/living as married, widowed, divorced/separated, or never married), education (less than high school, high school and some college, college graduate, or postgraduate degree), study center, smoking status (never, former, or current), first-degree family history of breast cancer (no, yes, or possibly), age at menarche (<10, 10–11, 12–13, 14–15, or 16+ years), age at menopause (<40, 40–44, 45–49, 50–54, or 55+ years), number of live births (0, 1, 2, 3, 4, or 5 or more), age at first birth (nulliparous, ≤19, 20–24, 25–29, 30–34, or >35 years), PMH use (never, former, or current), OC use (no or yes), oophorectomy (no/don't know, one ovary, or both ovaries), hysterectomy (no/don't know or yes), alcohol intake (g/day), total energy intake (kcal/day), total fat from diet (g/day), and red meat intake (g/day).

Breast cancer incidence was assessed by self-report using mailed annual study updates for in situ and invasive breast cancers confirmed by state cancer registries or reports from physicians and next of kin, and the national death index for fatal cases (23). Breast cancer supplemental forms were administered starting from 2007 to obtain detailed clinical characteristics of breast cancer tumors such as tumor grade, stage, and subtype.

Descriptive statistics estimated means and SD for continuous variables and frequencies and percentages for categorical variables. Categories were created for similar food products: fruits, vegetables, grains, dairy and eggs, red meat, white meat, processed meat, mixed foods, fish, fat and oil, nuts, or others, and the contribution of food products to total CML-AGE intake was estimated as percentages. Pearson correlation coefficient was used to estimate the correlation between CML-AGE intake and nutrients suggested to contribute to dietary AGE formation. Cox proportional hazards regression was used to estimate hazard ratios (HR) and 95% confidence interval (CI) of breast cancer by quintiles and tertiles of CML-AGE intake, and person-time was calculated from the date of completion of baseline questionnaires to the end of follow-up or breast cancer diagnosis. Simple models were adjusted for age and energy intake. A second multivariable Cox model adjusted for age, energy intake, alcohol, BMI, hours spent in vigorous activity, race/ethnicity, marital status, education, study center, smoking status, family history of breast cancer, age at menarche, age at menopause, age at first birth, number of live birth, PMH use, OC use, oophorectomy, and hysterectomy. A third multivariable model adjusted for all aforementioned covariates in addition to intakes of total fat and red meat. Nonlinear and linear relationships between CML-AGE intake and breast cancer risk were assessed (24). Linear trends were assessed for breast cancer risk across quintiles of CML-AGE intake by using the median CML-AGE intake value in each quintile. In addition, Cox proportional hazards models were used to assess the risks of breast cancer by invasiveness and hormone receptor status [estrogen receptor (ER) and progesterone receptor (PR); ref. 25]. Effect modification by race/ethnicity, obesity status, and fruit and vegetable intake was examined in stratified analyses. Sensitivity analysis was conducted by excluding 279 women with breast cancer diagnosed less than 2 years after enrollment into the PLCO. Schoenfeld residual test was used to evaluate violation of proportional hazards assumption (26) and no major violations were observed. All statistical analyses were conducted using SAS version 9.4 and statistical significance was set at α = 0.05.

During a median follow-up time of 11.5 years, there were 1,592 breast cancer cases diagnosed in 27,464 women at risk (Supplementary Fig. S1). The average daily CML-AGE consumption was 6,105 ± 2,691 KU/1,000 kcal and ranged between 867 KU/1,000 kcal and 43,387 KU/1,000 kcal (Table 1). A large proportion of women enrolled in the study was NHW and had no family history of breast cancer. Women in the highest quintile of CML-AGE intake were younger, had higher daily energy intake, higher intake of animal protein, total fat from diet, processed meat and red meat, and lower daily alcohol intake compared with the lowest quintile (Table 1). Women in the highest quintile of CML-AGE intake also had lower intakes of carbohydrates, fructose, calcium, fruits, and vegetables. A higher proportion of women in the highest quintile of CML-AGE intake were obese, divorced or separated, current cigarette smokers, had five or more live births, were OC users, had undergone a hysterectomy and oophorectomy of both ovaries, were ages 14–15 years at menarche, and were ages <44 years or 55+ years at menopause. Women in the highest quintile of CML-AGE intake were least likely to engage in vigorous activity 4+ hours a week, have a graduate education or higher, be current PMH users, and be aged 35 years or more at first birth. Similar distributions of covariates were observed across the tertiles of CML-AGE intake as seen across the quintiles (Supplementary Table S1). Fat and oil, and red meat were the food groups that contributed the most to total CML-AGE intake while nuts, grains, and fruits contributed the least (Fig. 2). As shown in Supplementary Table S2, positive correlations were observed between energy-adjusted CML-AGE intake and dietary sources of animal protein (r = 0.46), saturated fatty acids (r = 0.41), monounsaturated fatty acids (r = 0.39), and polyunsaturated fatty acid (r = 0.25), while negative correlations were observed between energy-adjusted CML-AGE and fructose (r = −0.25), carbohydrates (r = −0.19), calcium (r = −0.14), plant protein (r = −0.12), and protein from dairy (r = −0.07).

The results of the multivariable Cox proportional hazard regression are presented in Table 2. In the adjusted model 2, positive associations were observed in the upper quintiles of CML-AGE intake (HR_Q4VSQ1, 1.19; 95% CI, 1.01–1.40 and HR_Q5VSQ1, 1.23; 95% CI, 1.04–1.46; P_trend, 0.02). After further adjustment for dietary fat and red meat intake, model 3, the positive associations persisted (HR_Q4VSQ1, 1.24; 95% CI, 1.02–1.50 and HR_Q5VSQ1, 1.30; 95% CI, 1.04–1.62; P_trend, 0.04). For the fully adjusted model 3, there was no evidence of a nonlinear association between continuous CML-AGE intake and breast cancer (P = 0.29). However, there was some evidence of a linear relationship (P = 0.05) showing an increase in breast cancer risk with increasing CML-AGE intake.

Because of the small number of cases across CML-AGE quintiles and strata of effect modifiers, the tertile categorization of CML-AGE was applied in stratified analyses instead of quintile categorization (Tables 3 and 4; Supplementary Table S3). In the overall study population, a positive association was observed between tertiles of CML-AGE intake and breast cancer (HR _T2VST1, 1.14; 95% CI, 1.00–1.31 and HR _T3VST1, 1.19; 95% CI, 1.00–1.40). Positive associations in the upper tertiles were observed for all race/ethnicity groups but were only statistically significant in NHW women (HR_T3VST1, 1.21; 95% CI, 1.02–1.44). As shown in Table 4, higher intakes of CML-AGE were associated with increased risk of in situ and invasive breast cancer but were statistically significant only for in situ breast cancers (HR_T3VST1, 1.49; 95% CI, 1.11–2.01). Weak positive associations were observed for the risks of ER⁺ cancers (HR_T3VST1, 1.18; 95% CI, 0.97–1.45) and ER⁻ cancers (HR_T3VST1, 1.21; 95% CI, 0.82–1.79). Higher intake levels of CML-AGE were significantly associated with increased risk of PR⁺ breast cancer (HR_T3VST1, 1.24; 95% CI, 1.01–1.53) but not PR⁻ breast cancer. When ER and PR status were combined, there was a significant increased risk of ER⁺/PR⁺ breast cancer in the highest tertile of CML-AGE intake (HR_T3VST1, 1.24; 95% CI, 1.01–1.53). On the other hand, weak inverse associations were observed for the risk of ER⁺/PR⁻ breast cancers and the CI was wide (HR_T3VST1, 0.83; 95% CI, 0.53–1.31). There were too few cases with ER⁻/PR⁻ or ER⁻/PR⁺ tumors to examine associations separately for these groups.

Associations between CML-AGE and breast cancer were strongest among obese subjects as compared with associations among normal weight or overweight subjects but were not statistically significant (obese: HR_T3VST1, 1.32; 95% CI, 0.93–1.87; Supplementary Table S3). In the associations between CML-AGE and breast cancer by fruit and vegetable intake, the associations were strongest among high consumers (HR_T3VST1, 1.36; 95% CI, 1.00–1.85) or low consumers (HR_T3VST1, 1.27; 95% CI, 0.95–1.71) as compared with associations among moderate consumers of fruits and vegetables but also were not statistically significant (Supplementary Table S3). In sensitivity analysis excluding women who were diagnosed with breast cancer within the first 2 years of study follow-up (Supplementary Table S4), the association between CML-AGE intake and breast cancer was slightly higher compared with the association observed in the initial sample and was consistent with the overall conclusion (HR_Q5VSQ1, 1.37; 95% CI, 1.07–1.76).

In these secondary analyses of the large prospective PLCO study, increasing levels of CML-AGE intake were positively associated with breast cancer risk. The increased risk was more prominent in NHW women and women diagnosed with in situ and hormone receptor–positive (ER⁺/PR⁺) breast cancer. We utilized a published AGE database containing CML-AGE values of foods to derive estimates of total CML-AGE intake. The estimated average daily CML-AGE intake reported in a different study of healthy older women using the Uribarri and colleagues AGE database were quite similar to the estimates in our study (15, 27). Foods rich in fats and protein such as meat have been reported to contain high AGE content especially if cooked at high temperatures (6, 8, 28). In our study, fats and oil (such as margarine and cheese sauce) and red meat were the highest contributors to total daily CML-AGE intake, while fruits contributed the least amount. Animal protein and dietary fatty acids were positively correlated with CML-AGE, while fructose, plant protein, dairy protein, carbohydrates, and calcium were negatively correlated.

Few large epidemiologic studies have been published on the association between dietary AGEs and cancer risk. In analyses of the NIH-American Association of Retired Persons (NIH-AARP) Diet and Health Study, a positive relationship between CML-AGE intake and pancreatic cancer risk in men was observed (15). More recently, the NIH-AARP Diet and Health Study was used to examine associations between CML-AGE intake and breast cancer, in which an increased risk of invasive breast cancer was reported but the association was attenuated to the null after adjusting for potential confounders including intakes of dietary fat and total meat (14). In contrast, we observed an increased risk in overall breast cancer which persisted even with adjustment for multiple confounders including intake of total fat and red meat. In the joint Cox model assessing the risk by breast cancer subtypes, the positive association was stronger for in situ breast cancer and weaker for invasive breast cancer. Our study also showed stronger associations in NHW women, but null associations were found in women of other races/ethnicities (NHB and other races/ethnicities). The null associations could have been due to the smaller number of NHB subjects and women of other races/ethnicities enrolled in the PLCO.

We also observed stronger but nonsignificant positive associations among obese participants and high or low consumers of fruits and vegetables. Obesity is a risk factor for breast cancer among postmenopausal women (29) and may act synergistically with diet to increase risk. Fruits are natural sources of fructose and high consumption may contribute to serum CML-AGE levels (30, 31). Nevertheless, fruits and vegetables contain high amounts of antioxidants and fiber, and may be protective of breast cancer (32). The positive associations observed among high or low consumers of fruits and vegetables may be driven more by the overall pattern of intake of CML-AGE-rich foods and less by the intake of fruits and vegetables.

When deposited in tissue cells, AGEs induce damage to protein structure or activate RAGE to promote oxidative stress and chronic inflammation through secretion of inflammatory biomarkers which could induce DNA damage (9–12, 33). RAGE has been markedly expressed in breast tumors (34) and increased RAGE expression enhances the proliferation and invasion of breast cancer cells (35, 36). High accumulation of CML in breast cancer tissues (18) and elevated serum CML levels have been observed in women with breast cancer (16). Previous studies have reported higher circulating levels of AGEs in patients with ER⁺ breast cancer compared with ER⁻ breast cancers (16), and high CML accumulation in breast cancer tissue cell lines was related to ER expression (18). We found significant increased risks for PR⁺ breast tumors and for ER⁺/PR⁺ combinations, which support the potential role of dietary AGE in promoting hormone receptor–positive breast tumors.

The composition of the gut microbiota may affect the absorption of AGEs and consuming AGE-rich foods may potentially modify the gut microbiota. A randomized controlled trial in adolescent males found that higher CML intake was negatively correlated with Lactobacilli, a gut bacteria which may have beneficial properties (37) suggesting that the gut microbiota may play a role in AGE absorption. Because of lack of microbiota data in the PLCO dataset, we were unable to examine the role of individual variations in the gut microbiota on CML-AGE exposure and metabolism, although this represents a promising area for future research.

Our study is strengthened by the use of a large prospective dataset with an adequate sample size, long follow-up period (median 11.5 years), and information on multiple confounders. Measurement error from the FFQ is a possibility, although the error is likely to be nondifferential because dietary information was obtained prior to the occurrence of breast cancer. We utilized only the DQX completed at baseline among the intervention arm to enhance uniformity of the dietary assessment within the PLCO dataset. We used the database developed by Uribarri and colleagues which assessed AGE content of more than 500 foods and beverages using ELISA to estimate total CML-AGE intake (6). A definitive approach to measuring AGEs has not yet been established as some analytic methods produced variations in CML contents for certain foods (38). Thus, the true AGE content present in food may be under- or overestimated (22, 38). In some previous studies, estimated dietary AGE intake was correlated with serum AGE levels (39, 40), although one study which averaged two measurements taken 13 weeks apart of serum CML measured by ELISA reported no correlation between intake of foods considered high in dietary AGE and serum CML-AGE (41). The Uribarri and colleagues AGE database is the most frequently utilized method to estimate dietary AGE intake from FFQ responses in epidemiologic studies (14, 15), which will enhance comparability of our results with other studies. As in any prospective cohort study, selection bias from loss to follow-up was a potential limitation because participants were enrolled in the study for a long period of time. While we adjusted for multiple potential confounders, residual or unmeasured confounding cannot be entirely ruled out in this observational study. Finally, the race-stratified analyses were limited by small sample sizes among NHB and other racial/ethnic groups. Future studies with more diverse populations are warranted.

Our findings suggest that dietary AGEs may contribute to the risk of developing breast cancer. Future studies to confirm the effects of AGEs in breast cancer and explore the differential associations among racial/ethnic groups and breast cancer subtypes are warranted. Incorporating cooking methods involving low temperatures and moisture such as boiling, and marinating foods prior to cooking are strategies that could reduce AGE content in food (6, 8), and may be important in the prevention of breast cancer.

O.O. Omofuma reports grants from Susan G. Komen during the conduct of the study. D.P. Turner reports grants from NIH/NCI during the conduct of the study. S.E. Steck reports grants from Susan G. Komen during the conduct of the study. No potential conflicts of interest were disclosed by the other authors.

Conception and design: O.O. Omofuma, D.P. Turner, L.L. Peterson, A.T. Merchant, S.E. Steck

Development of methodology: O.O. Omofuma, L.L. Peterson, A.T. Merchant, J. Zhang, S.E. Steck

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): O.O. Omofuma, D.P. Turner, L.L. Peterson, A.T. Merchant, J. Zhang, S.E. Steck

Writing, review, and/or revision of the manuscript: O.O. Omofuma, D.P. Turner, L.L. Peterson, J. Zhang, S.E. Steck

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): O.O. Omofuma

Study supervision: S.E. Steck

This study was supported by a Graduate Training in Disparities Research grant from Susan G. Komen (GTDR17500160; principal investigator: S.E. Steck; Trainee: O.O. Omofuma); grants R21 CA194469 and U54 CA21096 from the NIH/NCI (to D.P. Turner); and a Mentored Research Scholar Grant from the American Cancer Society (MRSG-18-199-01-NEC; to L.L. Peterson). The authors would like to acknowledge Yanan Zhang for contributing to the statistical analyses.

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.