A Dietary Pattern Associated with LINE-1 Methylation Alters the Risk of Developing Cervical Intraepithelial Neoplasia

The importance of diet for health was emphasized more than a quarter century ago when Doll and Peto reported that approximately 35% (10%–70%) of all cancers might be attributable to dietary factors and approximately 90% of colorectal cancer may be preventable through dietary modifications (1). During the early 1990s, we had high expectations that a higher consumption of fruits and vegetables would reduce the risk of many cancers (2). However, this evidence was based primarily on results generated by case–control studies. The results generated by prospective cohort studies completed in more recent years did not confirm these findings (3). Several researchers responded to these inconsistent results between case–control and cohort studies by pointing out that those case–control studies were biased by differences in recall of fruit and vegetable intake by individuals diagnosed with cancer and healthier controls. Furthermore, even if both cases and non-cases reported their intakes similarly, non-cases who participated in these studies could have been more health conscious than non-cases who did not participate, leading to an exaggerated benefit of fruits and vegetables in cancer prevention. However, we still do not exclude the possibility that specific groups of fruits and vegetables, specific substances in some fruits and vegetables, or overall dietary patterns have important cancer protective effects. Recent studies have shown that a higher consumption of dark green and deep yellow vegetables and fruits was associated with lower risk of having cervical intraepithelial neoplasia (CIN), precursor lesions for developing cervical cancer, especially among smokers (4). Frequent consumption of fruits high in antioxidant nutrients was also shown to be associated with lower risk of CIN (5). Studies also support a role for fruit and vegetable consumption in reducing the risk of CIN, especially in women infected with a higher load of human papilloma viruses (HPVs), the main causative factor for CIN and cervical cancer (6). A study also suggested that diets rich in plant-based nutrients may lower the risk of cervical cancer (7).

Because of the obesity epidemic and its associated chronic disease risk, assessment of overall dietary patterns and their link to chronic disease risk (8) and diet-related alterations in the epigenome are becoming increasingly recognized as important. To our knowledge, only a few studies have focused on precancerous stages in relation to dietary patterns, a point where the development of cancers could be prevented by dietary modifications. Even though it is logical to assume that dietary recommendations focused on promoting a healthier overall dietary pattern rather than encouraging consumption of certain foods or food categories should be the first line of intervention for prevention of many different types of cancers, use of biomarkers to monitor the effectiveness of these interventions should be an integral part of such efforts. To our knowledge, there have been no systematic studies conducted to derive biomarkers of dietary patterns which are also associated with higher risk of developing precancerous lesions. We have recently documented that a higher degree of DNA methylation in the long interspersed nucleotide elements (L1s) of peripheral blood mononuclear cells (PBMC) was associated with 56% lower risk of being diagnosed with higher grades of CIN (CIN 2 +), a common precancerous lesion found among sexually active women exposed to carcinogenic or high-risk (HR) types of HPVs (9). The main purpose of the current study was to identify overall dietary patterns that are associated with CIN 2+ and to determine whether these dietary patterns are associated with the degree of L1 methylation in PBMCs.

The present analysis is based on 319 women enrolled in an ongoing prospective follow-up study funded by the National Cancer Institute, Bethesda, MD (R01 CA105448, Prognostic Significance of DNA & Histone Methylation). The study has been described in a previous publication (10). Briefly, all women were diagnosed with abnormal cervical cells in clinics of the Health Departments in Alabama and were referred to the University of Alabama at Birmingham (UAB), Birmingham, AL, for further examination by colposcopy and biopsy. Women were 19 to 50 years old; had no history of cervical cancer or other cancers of the lower genital tract; no history of hysterectomy or destructive therapy of the cervix; were not pregnant; were not using antifolate medications such as methotrexate, sulfasalazine, or phenytoin; and were nonvitamin supplement users. Of the 319 women, 93 women were diagnosed with CIN 2+ [cases: including CIN 2 (n = 57), CIN 3 (n = 33), or carcinoma in situ (CIS, n = 3)] and 226 women were diagnosed with ≤CIN 1 [non-cases: including normal cervical epithelium (n = 12), HPV cytopathic effect (HCE, n = 26), reactive nuclear enlargement (RNE, n = 39), or CIN 1 (n = 149)]. Both cases and controls tested positive for HR-HPV (any one of 13 types of HR-HPV, HPV 16, 18, 31, 33, 35, 39, 45, 51, 52, 56, 58, 59, and 68 based on Roche Diagnostics Linear Array results). All women included in this analysis participated in an interview that assessed sociodemographic variables and lifestyle risk factors (age, race, level of education, smoking status, use of vitamin supplements, and oral/hormonal contraceptives), physical activity (CDC questionnaire), and dietary intake (Block's food frequency questionnaire, version 98.2). The healthy eating index (HEI; Block scale of 0–100) was obtained from Block questionnaire data. Height, weight, and waist circumference (WC) measurements were obtained using standard protocols. The body mass index (BMI) was calculated as weight (kilograms) divided by height (meters squared). The study protocol and procedures were approved by the UAB Institutional Review Board.

DNA was extracted from buffy coat samples using a standard phenol–chloroform extraction method. As described below, methylation of the L1 promoter (GenBank accession no. x58075) in PBMCs was investigated by a pyrosequencing-based methylation analysis.

Bisulfite treatment of 1 μg of DNA extracted from buffy coat was completed by the EZ DNA methylation kit (Zymo Research) and the converted DNA was eluted with 30 μL TE buffer. PCR reactions were carried out using forward (5′-TTTTTTGAGTTAGGTGTGGG-3′) and reverse-biotinylated (5′-biotin-TCTCACTAAAAAATACCAAACAA-3′) primers, as described (11). The biotinylated PCR product, purified and made single stranded to act as a template, was annealed to the pyrosequencing primer (5′-GGGTGGGAGTGAT-3′; 0.4 μmol/L final concentration), and then was subjected to sequencing using an automatically generated nucleotide dispensation order for sequences to be analyzed corresponding to each reaction. The pyrograms were analyzed using allele quantification mode to determine the proportion of C/T, and hence, methylated and unmethylated cytosines at the targeted position(s). The degree of methylation was evaluated at 3 CpG methylation sites (11). The reproducibility of the assay was satisfactory with a coefficient of variance of 2.0% to 2.2%.

Usual dietary intake assessed with the Block food frequency questionnaire, version 98.2, which includes intake of phytochemicals, was used to derive dietary patterns in this population. Women with daily calorie intakes of <1,000 kcal and >5,000 kcal were excluded before deriving the dietary patterns. Food consumption frequencies were standardized into frequencies of intake per week. To reduce the number of patterns generated and to increase interpretability, we assigned each food item into a defined food group based on similarity of nutrients in a given food item, source (plant vs. animal) and how they are commonly consumed. This resulted in 35 food groups. The frequencies of intake of these foods in a given group were summed up to give the total intake per week for the food group. Some food groups contained only one food item and were entered in the model for deriving patterns as individual foods because of their unique nutrient profiles (e.g., water) or because their consumption reflects a distinct dietary pattern. We used PROC FACTOR in SAS v.9.2 (SAS Institute; 2008) to derive dietary patterns and corresponding factor loadings for each food group. We generated the SCREE plots and examined eigenvalues for each food or food group and determined the number of factors to keep. After determining the number of factors to keep, we refitted the model with NFACTOR option to limit the number of factors generated to 2 factors. An orthogonal transformation was done using the VARIMAX rotation option to produce uncorrelated dietary patterns. Factor loadings for each food group were generated to reflect the contribution of each food or food group to the dietary pattern. Food items with a factor loading of 0.30 or more were considered important components of each pattern and were used to identify and name the dietary patterns.

We observed 2 main distinct dietary patterns in our study population. As shown in Table 1, the unhealthiest dietary pattern or dietary pattern 1 mainly consisted of food items considered to be unhealthy (high sugar beverages, pasta and starchy foods, margarine, butter, refined grains, desserts and sweets, snacks, high fat dairy, fatty meat, sausages and bacon, condiments, pizza, macaroni, and cheese). Each of these food items had a factor loading 0.30 or more for the first factor. Dietary pattern 2 mainly consisted of healthier food items (seafood, beans and lentils, tofu and meat substitutes, whole grains, fresh fruits, canned fruits, vegetables, peanut butter, low fat dairy, chicken and turkey, cereals, water, yogurt, dressings and gravy, and phytochemical rich foods) each with a factor loading 0.30 or more for the second factor. This pattern was named as the healthy dietary pattern. The study participants were then ranked in ascending order according to pattern 1 or pattern 2 scores. Using PROC RANK in SAS, 2 groups per dietary pattern were derived (i.e., above or below median for a given dietary pattern). Those with a factor score rank above median for the unhealthy pattern (factor 1) and a factor score rank below median for the healthy pattern (factor 2) were classified as unhealthiest (n = 76). Those with a factor score rank above median for the healthy pattern and a factor score rank below median for the unhealthy pattern were classified as the healthiest (n = 63). These 2 extremes of the patterns (healthiest and unhealthiest) comprised 44% of the women in our population. Fifty-six percent of women had intermediary dietary patterns [i.e., higher factor score ranks for both pattern 1 and pattern 2 (intermediary dietary pattern 1, n = 96) or lower factor score ranks for both pattern 1 and pattern 2 (intermediary dietary pattern 2, n = 84)].

We tested whether participant characteristics such as age, race, level of education, BMI, smoking status, physical activity, use of oral/hormonal contraceptives, total dietary calorie intake, PBMC L1 methylation, and case status varied by the 4 dietary pattern groups (healthiest, intermediary dietary pattern 1, intermediary dietary pattern 2, and the unhealthiest). We also tested whether energy adjusted intakes of “cancer protective micronutrients” [total dietary folate equivalents (DFE), vitamins B12, B6, B2, α-carotene, β-carotene, vitamins A, C, and E] and the HEI differ by the 4 dietary patterns in the entire population and among non-cases only. We used ANOVA and the 2-sided χ² to test the statistical significance for continuous variables and categorical variables, respectively. Using the “healthiest” dietary pattern as the referent in unconditional logistic regression models, we estimated the ORs [95% confidence interval (CI)] for having CIN 2+ (yes or no) for each of the dietary pattern groups. These analyses were adjusted for age (≥median vs. <median), race (Caucasian-American vs. African-American)/level of education (less than high school education vs. high school education or greater), BMI (>25 vs. ≤25 kg/m²), smoking status (ever vs. never), physical activity (>150 vs. ≤150 min/wk), use of oral/hormonal contraceptives (ever vs. never), and total dietary calorie intake (≥median vs. <median).

We then examined the association between dietary patterns and PBMC L1 methylation among non-cases using unconditional logistic regression models. Exclusion of the cases was necessary to avoid the possibility of reverse causation (i.e., CIN 2+ status influencing dietary habits). If women were above the 50th percentile of the percentage of PBMC L1 methylation distribution, they were classified as having higher methylation; otherwise they were classified as having lower methylation. In this analysis, 3 unconditional logistic regression models were run to test the association between PBMC LI methylation and dietary pattern keeping the unhealthiest, intermediary dietary pattern 1, or intermediary dietary pattern 2 as the referent groups. All models were adjusted for age, race/level of education, BMI, smoking status, physical activity, use of oral/hormonal contraceptives, and total dietary calorie intake. Because race and level of education were highly correlated, we did not include both variables as covariates in the same model. We tested all models separately with each of these variables.

The characteristics of the study population based on the 4 dietary patterns are reported in Table 2. Race, use of oral/hormonal contraceptives, and median total dietary calorie intake per day were significantly different among the 4 dietary pattern groups (P = 0.01, P < 0.01, and P < 0.001, respectively). None of the other variables were statistically different among the 4 dietary patterns. Results from a similar univariate analysis between the healthiest dietary pattern and the unhealthiest dietary pattern showed that the unhealthiest dietary pattern was more common among African-American women (78%) than Caucasian-American women (22%; P = 0.0009). Women with the unhealthiest dietary pattern were significantly less likely to be engaged in more than 150 min/wk physical activity and more likely to use oral/hormonal contraceptives than women with the healthiest dietary pattern (P = 0.043 and 0.001, respectively). We also observed that women with the unhealthiest dietary pattern had significantly higher median total dietary calorie intake than women with healthiest dietary pattern (P < 0.001). The median PBMC L1 methylation was significantly lower in women with the unhealthiest dietary pattern than women with the healthiest dietary pattern (P = 0.021). Furthermore, 34% of women with the unhealthiest dietary pattern were cases whereas only 19% of women with the healthiest dietary pattern were cases (P = 0.046). None of the other variables were statistically different between women with the healthiest dietary pattern and women with the unhealthiest dietary pattern.

Table 3 shows the median energy adjusted intakes of cancer protective micronutrients and HEI among the 4 dietary patterns. The intake of all micronutrients (except for vitamin C) and HEI were significantly different among the 4 dietary patterns. We observed that the unhealthiest dietary pattern had significantly lower HEI and intakes of all micronutrients (except for vitamin C) than the healthiest dietary pattern (P < 0.05 for HEI and all micronutrients except for vitamin C). Among the non-cases, all micronutrients (including vitamin C) and HEI were significantly higher in women with the healthiest dietary pattern than unhealthiest dietary pattern (P < 0.05 for all comparisons, data not shown).

As shown in Fig. 1, women with the unhealthiest dietary pattern were 3.5 times more likely to be diagnosed with CIN 2+ than women with the healthiest dietary pattern (OR = 3.5; 95% CI, 1.2–10.1; P = 0.02). Compared with the healthiest dietary pattern, women with the intermediary dietary pattern 1 (OR = 1.7; 95% CI, 0.6–5.4; P = 0.34) and intermediary dietary pattern 2 (OR = 2.2; 95% CI, 0.9–5.6; P = 0.10) showed a nonsignificant positive association with CIN 2+ status.

As shown in Table 4, pre-cancer–free women with the healthiest dietary pattern were 3.3 times more likely to have higher PBMC L1 methylation than women with the unhealthiest dietary pattern (OR = 3.3; 95% CI, 1.0–10.6; P = 0.04) in a model that adjusted for age, race, BMI, smoking status, physical activity, use of oral/hormonal contraceptives, and total dietary calorie intake. Women with the healthiest dietary pattern than the intermediary dietary pattern 1 and dietary pattern 2 were 1.5 and 1.7 times more likely to have higher PBMC L1 methylation, respectively, but these associations were statistically nonsignificant (OR = 1.5; 95% CI, 0.5–4.1; P = 0.45 and OR = 1.7; 95% CI, 0.8–3.8; P = 0.20, respectively). Models yielded similar results when race was replaced with the level of education.

Even though it is logical to assume that dietary recommendations focused on promoting a healthier overall diet rather than encouraging consumption of certain foods or food categories should be the first line of intervention for prevention and control of cancer, we are currently in a weak position to do so because of lack of data on biologically meaningful overall dietary patterns that are specific for cancer prevention and control.

The most common health outcomes examined in relation to dietary patterns for some time have been all-cause mortality and cardiovascular disease risk (12). The use of a dietary pattern approach in the setting of cancer research has become more common but largely limited to cross-sectional studies of cancer risk (13–23) or survival from cancer (24, 25). A recent follow-up study showed that a dietary pattern rich in fruit and salad might protect against invasive breast cancer (26). A prospective follow-up study on prostate cancer, however, failed to identify any dietary pattern associated with risk of prostate cancer (27).

The limited number of studies which evaluated the effects of dietary patterns on pre-cancer risk has focused largely on colorectal adenomas. A dietary pattern consisting of a higher consumption of dairy products and fruits and vegetables with low alcohol consumption was associated with lower risk of colorectal adenomas in Japanese men (28). A high-fruit, low-meat diet was shown to be protective against colorectal adenomas compared with a dietary pattern of higher vegetable and meat consumption (29). A recent study showed that African-American women may be able to reduce their risk of developing colorectal adenomas by following a prudent dietary pattern (30). Our study was able to identify an unhealthy dietary pattern which might put women at higher risk for developing HR-HPV–related CIN 2+, precursor lesions for developing cervical cancer. The unhealthiest dietary pattern identified in our study is similar to a Western dietary pattern. Further, women with the unhealthiest dietary pattern were significantly more likely to use oral/hormonal contraceptives and had lower HEI, factors that may be associated with lower cancer protective micronutrient status. In fact, intakes of several micronutrients with cancer protective effects were significantly lower in women with the unhealthiest dietary pattern, showing the biologic plausibility of the observed association between unhealthiest dietary pattern and higher risk of CIN 2+. African-Americans were found to consume lower amounts of micronutrients and were reported to be at higher risk for some cancers than Caucasian-Americans (31). Interestingly, we observed that the lower micronutrient containing unhealthiest dietary pattern identified in our study was significantly more common among African-Americans than Caucasian-Americans. We also observed that women with the unhealthiest dietary pattern consumed significantly higher amount of calories and were physically less active than women with the healthiest dietary pattern. Even though the association between physical activity and cancer risk is inconsistent, higher calorie consumption is associated with higher risk of some cancers (32).

Even though only the unhealthiest dietary pattern was associated with statistically significant higher risk of CIN 2+, both intermediary dietary patterns were associated with approximately 2-fold higher risk of CIN 2+, indicating that 80% of this population do not have dietary patterns which may exert cervical cancer protective effects. Therefore, these observations suggest that the consumption of higher amounts of healthier food items along with higher amounts of unhealthier food items or the consumption of lower amounts of both healthier and unhealthier food items are unlikely to be beneficial for reducing cancer risk. Therefore, modifications toward the healthiest dietary pattern based on food categories identified by our study may exert the most beneficial effects on the prevention of cervical cancer in this population.

Our study has limitations inherent to factor analysis used to derive dietary patterns, that is, subjective judgment in deriving 35 food groups, in determining the number of patterns and possibly the interpretation of these patterns. However, evaluation of a pre-cancer–related biomarker in relation to dietary patterns is a unique aspect of this study. The biomarker we have chosen to associate with dietary patterns (L1 methylation) is a validated surrogate biomarker of genome-wide methylation changes (33). Studies have shown that methylation levels measured in L1 regions, which are easy to characterize by pyrosequencing technology, do not vary significantly with time within an individual, and therefore, changes in their methylation levels could potentially be attributed to dietary or lifestyle factors or interventions with such factors (9). A recent study has shown that a prudent dietary pattern was associated with a lower prevalence of PBMC L1 hypomethylation in a dose-dependent manner suggesting the beneficial effects of a healthier dietary pattern on L1 methylation in a cancer-free population (34). We evaluated the association between dietary patterns and PBMC L1 methylation in women free of cervical pre-cancer, but they are at higher risk for developing cervical pre-cancer or cancer because they are diagnosed with abnormal pap and tested positive for HR-HPVs. In these women, we observed that those with the healthiest dietary pattern were significantly more likely to have higher PBMC L1 methylation than those with the unhealthiest dietary pattern. We observed that the intakes of several methyl donor micronutrients (folate, vitamins B12, B2, and B6) were significantly higher in the healthiest dietary pattern than the unhealthiest dietary pattern identified in our study, indicating the biologic plausibility for higher PBMC L1 methylation observed in women with the healthiest dietary pattern identified in our study population. Our results also showed that the PBMC L1 methylation was 1.5- to 1.7-fold higher in the healthiest dietary pattern than the 2 intermediary dietary patterns. Even though these differences were statistically nonsignificant, the observed results suggest that 80% of the population may not have dietary patterns which provide adequate L1 methylation. We have previously shown that higher PBMC L1 methylation is associated with lower risk of CIN 2+ (5). Therefore, intervening to change the unhealthiest and the 2 intermediary dietary patterns toward the healthiest dietary pattern may result in lower risk of developing cervical pre-cancer in this population.

PBMC L1 methylation may serve as a unique epigenetic marker for monitoring the effectiveness of such dietary interventions. A higher degree of PBMC L1 methylation was also shown to be associated with lower risk of other cancers such as head and neck (35) and bladder (36) and also pre-cancerous conditions such as colorectal adenomas (37). Therefore, L1 methylation may serve as a biomarker for dietary pattern interventions that are targeted to reduce the risk of these cancerous and pre-cancerous conditions. To our knowledge, this is the first study to show that a pre-cancer–related biomarker is associated with a dietary pattern. Future studies are needed to confirm whether the association between dietary pattern and risk of developing CIN 2+ holds in longitudinal studies and whether L1 methylation serves as a biomarker for monitoring the effectiveness of dietary pattern–based interventions for cancer prevention.

No potential conflicts of interest were disclosed.

The authors thank the staff of the Molecular Epidemiology Laboratory of C.J. Piyathilake who assisted with the dietary data collection and Ilene Brill and David Helms who assisted with the statistical analysis.

This study is supported by R01 CA105448 and funded by the National Cancer Institute.

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.