Plasma choline metabolites and colorectal cancer risk in the Women’s Health Initiative Observational Study

Study population

The WHI-OS is a prospective cohort study to investigate the predictors and causes of morbidity and mortality in postmenopausal women (20, 21). The study enrolled 93,676 postmenopausal women, aged 50–79 y, at 40 centers throughout the US between 1993 and 1998. Women were excluded if they had medical conditions with a predicted survival of <3 y; if they had adherence/retention issues; or if they were participating in another clinical trial.

For the present study, incident CRC cases were selected as of April 24 2008, and the average time from baseline to CRC diagnosis was 5.2 ± 3.1 y (mean ± SD) (11, 22). Women were excluded if they had a history of CRC or in situ CRC; if they had no available biospecimens; or if a death certificate provided the only report of CRC. Controls who were free of cancer at the time of case diagnosis were selected from the WHI-OS by using risk-set sampling. Cases and controls were matched on age (±3 y), race/ethnicity, timing of baseline blood draw (±6 months), enrollment date (±1 y) and baseline hysterectomy status (11, 22). Thus, the present study included 835 incident CRC cases and 835 matched controls. Approval for conducting the study was obtained from human subject review committees at the Fred Hutchinson Cancer Research Center (WHI Clinical Coordinating Center), as well as at all 40 clinical centers. Written informed consent was obtained from all participants.

Data collection

Demographic and health-related characteristics were collected at baseline using standardized questionnaires (20). Height and weight were measured using a standardized protocol, and body mass index (BMI) was calculated as weight (kg)/height (m²). CRC was annually assessed using self-administered questionnaires collected from each participant by mail and during an in-person clinical follow-up visit at year 3 (23). All CRC cases were confirmed by physician adjudicators. The International Classification of Diseases for Oncology, second edition codes were used to identify CRC cases based on tumor site as previously described (11). The Surveillance Epidemiology and End Results (SEER) program guidelines of the National Cancer Institute were used for classifications of cancer cases (23).

Analytic measurements

Blood samples were drawn at study baseline after at least 12 hours of fasting. Samples were kept at 4°C for up to 1 hour prior to centrifugation. Plasma and serum were collected and stored at −70°C until analysis (22). Plasma concentrations of choline and its metabolites (betaine, DMG, and TMAO) were measured in de-identified samples using liquid chromatography-tandem mass spectrometry methodology with modifications based on our instrumentation (24). Plasma and red blood cell (RBC) folate as well as plasma vitamin B12 were measured by radioassays (SimulTRAC; MP Biomedicals); plasma pyridoxal-5′-phosphate (PLP) was analyzed by high-pressure liquid chromatography (HPLC) with fluorescence detection (25); and total plasma homocysteine was determined by HPLC with post-column fluorescence detection (26). Inter-assay coefficients of variance of the blind duplicate control samples for each of the assays were as follows: choline, 7%; betaine, 5%; DMG, 9%; TMAO, 6%; plasma folate, 5%; RBC folate, 10%; vitamin B12, 6%; PLP, 6%; and homocysteine, 7%.

Statistical analysis

Baseline characteristics of CRC cases and controls were compared using (i) t tests for normally distributed continuous variables; (ii) Wilcoxon tests for non-normally distributed continuous variables; and (iii) chi-square tests for categorical variables. Associations among plasma concentrations of choline metabolites were assessed using Spearman correlation analysis. Plasma choline metabolites were divided into quartiles based on the distribution of the controls. Conditional logistic regression models were used to estimate odds ratios (ORs) and 95% confidence intervals (CIs) of CRC risk among quartiles of choline metabolites, using the lowest quartiles as reference groups. Because risk-set sampling was used for selecting matched controls, the conditional ORs yielded estimates of the incidence rate ratio in a full cohort study. We further explored the associations between the ratios of choline metabolites (i.e., betaine:choline, DMG:choline and DMG:betaine) and CRC risk, because the ratios of these metabolites (vs. individual metabolite alone) are suggested to be stronger predictors of metabolic disturbances (27). The models were first adjusted only for age (continuous) and then further adjusted for baseline confounding factors selected a priori: BMI, pack-years of smoking, physical activity, use of postmenopausal hormone therapy, history of colonoscopy, RBC folate, plasma vitamin B12, PLP and homocysteine. All of these factors were added in the model as continuous variables except for postmenopausal hormone therapy use (categorical: never, past or current). Tests of linear trend across increasing quartiles of choline metabolites were conducted by the Wald test, using the median value for each quartile as a single continuous variable.

To explore whether the associations between choline metabolites and CRC risk were modified by B-vitamins involved in one-carbon metabolism, we conducted analyses stratified into high/low plasma concentrations of folate, PLP and vitamin B12 based on median values among controls. We also examined the influence of FA fortification by stratifying into the following FA fortification periods based on the timing of baseline blood draw: pre-fortification (1994–1995), peri-fortification (1996–1997; when initial fortification began, but was not yet mandated), and post-fortification (1998) (28). The Wald test was used to evaluate the effect modification including a 2-way interaction term between the ordinal trend variables (choline metabolites) and effect modifiers (B-vitamins or FA fortification period). Because the matching was broken, unconditional multiple logistic regression models were used in these stratified analyses, further adjusting for days to CRC diagnosis and ethnicity. Significance was defined as P<0.05, and all statistical tests were 2-sided. Analyses were conducted by SAS version 9.3 (SAS Institute Inc).

Characteristics of the study population

Baseline characteristics of the CRC cases and controls are shown in Table 1. Compared to the controls, the cases had a higher BMI, a greater number of cigarettes smoked among current smokers, fewer weekly minutes of moderate or strenuous physical activity, and had a different distribution pattern of postmenopausal hormone therapy use. The CRC group also had a lower percentage of previous colonoscopy, but a higher percentage of having history of a colon polyp removed.

Plasma choline, betaine and DMG concentrations did not differ between cases and controls (Table 1). However, the cases (vs. controls) had higher (P=0.005) median plasma concentrations of TMAO (4.0 vs. 3.8 μmol/L) and tended to have a lower (P=0.07) mean plasma betaine:choline ratio (2.9 vs. 3.0). In addition, the cases had lower median plasma folate, PLP and vitamin B12 as well as higher median plasma homocysteine.

Among the cases, tumors were classified by tumor site (proximal, distal, or rectal) and stage (local/regional or metastatic). More than half (59%; n=489) of the tumors were proximal followed by distal (21%; n=177) and rectal (19%; n=155). Two percent (n=14) of the tumors were not classified by tumor site because they were unknown or had overlapping lesions. In addition, when stratified by tumor stage, the majority of the cases (85%; n=712) had localized or regional tumors, whereas 12% of the cases (n=104) had distant metastases. Two percent (n=18) of the tumors were not stratified by tumor stage because their stages were unknown or not determined.

Correlations among plasma concentrations of choline metabolites

Spearman correlation coefficients (r) were computed to examine associations among plasma choline metabolites. There were statistically significant, but modest, positive associations of plasma choline with plasma betaine (r=0.22; P<0.001), DMG (r=0.21; P<0.001) and TMAO (r=0.18; P<0.001). Plasma betaine was also positively correlated with plasma DMG (r=0.39; P<0.001).

Associations between plasma choline metabolites and CRC risk

In multivariable-adjusted analyses, women in the highest (vs. lowest) choline quartile were at an estimated 2.4 times greater risk of rectal cancer [P-trend=0.08] (Table 2). Conversely, women in the highest (vs. lowest) betaine quartile were at 32% lower CRC risk overall [OR (95% CI)_{highest vs. lowest quartile}=0.68 (0.47–0.99); P-trend=0.01], 36% lower risk of local/regional tumors [0.64 (0.42–0.99); P-trend=0.009] and 31% lower risk of proximal tumors [0.69 (0.43–1.10); P-trend=0.05] (Table 3). No association between DMG quartiles and CRC risk was observed (Supplementary Table S1).

Notably, after controlling for covariates, women in the highest (vs. lowest) quartile of the plasma betaine:choline ratio were at an estimated 44% lower CRC risk overall [0.56 (0.39–0.82); P-trend=0.004] as well as 34% lower risk of proximal tumors [0.66 (0.41–1.06); P-trend=0.049], 73% lower risk of rectal tumors [0.27 (0.10–0.78); P-trend=0.02] and 50% lower risk of local/regional tumors [0.50 (0.33–0.76); P-trend=0.001] (Table 4). The plasma DMG:choline ratio tended to be inversely associated with CRC risk overall [0.69 (0.48–0.98); P-trend=0.06] (Supplementary Table S2). The inverse association was statistically significant for local/regional tumors [0.62 (0.42–0.91); P-trend=0.04] and borderline significant for proximal tumors [0.57 (0.36–0.93); P-trend=0.07]. Last, the DMG:betaine ratio tended to be positively associated with rectal cancer risk [2.56 (0.98–6.64); P-trend=0.09] (Supplementary Table S3).

Plasma TMAO, an oxidative derivative of choline produced by intestinal bacteria, was positively associated with CRC risk in age-adjusted analyses [1.78 (1.32–2.40); P-trend=0.005] (Table 5). Women in the highest (vs. lowest) TMAO quartile were at approximately 1.9 times greater risk of proximal tumors [P-trend=0.04], 2.3 times greater risk of rectal tumors [P-trend=0.02] and 1.8 times greater risk of local/regional tumors [P-trend=0.008]. After controlling for covariates, the positive association remained strong and statistically significant for rectal cancer with approximately 3.4 times greater risk among women in the highest (vs. lowest) TMAO quartile [P-trend=0.02]. The borderline significant positive association was also observed for local/regional tumors with approximately 1.8 times greater risk in the highest (vs. lowest) TMAO quartile [P-trend=0.08]. Notably, although the linear trend across TMAO quartiles was not statistically significant, higher risk was observed from the second (vs. lowest) quartile of TMAO for CRC overall [1.90 (1.36–2.64)] and for proximal tumors [2.37 (1.52–3.70)]. Similarly, women in the second (vs. lowest) quartile of TMAO were at an estimated 1.9 times higher risk for local/regional tumors and 3.6 times higher risk for metastatic tumors, but this was not consistently observed in the other quartiles.

Associations of choline metabolites with CRC risk according to plasma B-vitamin concentrations

To further explore whether B-vitamins (folate, PLP and vitamin B12) modified the associations between choline metabolites and CRC risk, we stratified into high/low plasma concentrations of B-vitamins and assessed the interaction. After controlling for covariates, vitamin B12 status modified the association between plasma TMAO and CRC risk (P-interaction=0.003) (Table 6). Specifically, higher CRC risk was observed with higher TMAO quartiles among women with low plasma vitamin B12 (i.e., ≤505 pg/mL) [P-trend=0.001], but not among those with high B12 levels. Other than this finding, no effect modifications by B-vitamins were observed on the associations of plasma choline metabolites and their ratios with CRC risk (data not shown).

Associations of choline metabolites with CRC risk according to FA fortification period

We next explored the possible effect modification by FA fortification. The association of plasma choline, DMG, TMAO and the ratios of choline metabolites with CRC risk did not differ by fortification periods (data not shown). However, after controlling for covariates, plasma betaine tended to interact with FA fortification period in association with CRC risk (P-interaction=0.08) (Table 7). Specifically, lower CRC risk was observed with higher plasma betaine during the pre-[P-trend=0.02] and peri-[P-trend=0.02] fortification periods, but not during the post-fortification period.