One-carbon metabolite ratios as functional B-vitamin markers and in relation to colorectal cancer risk

Study participants

This was a case–control study nested within the Northern Sweden Health and Disease Study (NSHDS). Participants from two population-based subcohorts were included, the Västerbotten Intervention Programme (VIP, 78% of the study participants) and the Mammography Screening Project in Västerbotten (MSP, 22% of the study participants).30 In the VIP, established in 1985, residents of Västerbotten County are invited to participate in a health survey upon turning 30 (years 1990–1996), 40, 50 and 60 years of age. The health survey consists of laboratory tests and a medical examination, donation of a fasting blood sample and completion of an extensive lifestyle questionnaire. The VIP included 115,147 blood samples from 85,877 individuals as of March 31, 2009, the final date for case identification for our study (Supporting Information Fig. S1). Selection bias has been found to be low,31 and comparisons of cancer incidence in the cohort with expected incidence (based on data from the Swedish Cancer Registry) support the population-based structure of the VIP cohort.32 The MSP was established in 1995 and concluded in 2006, and invited women residing in Västerbotten County and about 50–70 years of age, to complete a lifestyle questionnaire and donate a blood sample while attending mammography screening. The MSP includes a total of 54,401 blood samples from 28,802 women.

CRC case participants diagnosed between October 17, 1986, and March 31, 2009, were identified by linkage with the essentially complete Cancer Registry of Northern Sweden (ICD-10 C18.0 and C18.2–C18.9 for colon, C19.9 and C20.9 for rectum). A single pathologist specialized in gastrointestinal pathology verified all cases and tumor data. Patient records were used to verify tumor site. Exclusion criteria were previous cancer diagnosis other than non-melanoma skin cancer (68 cases), insufficient volume of stored plasma samples (46 cases, nine controls), location of the primary tumor outside of the colorectum (18 cases), samples prioritized to other studies (eight cases, 18 controls), serious infectious disease (one case, excluded for the safety of the laboratory staff) or no matching control available (three cases). Two control participants were randomly selected for each case, matched by age at and year of blood sampling and data collection, cohort, sex and fasting status. The same exclusion criteria were applied to the control participants as the case participants. Furthermore, controls were all alive and had no diagnosed cancer other than non-melanoma skin cancer at the time of diagnosis of their index cases. After exclusions (81 cases and 41 controls), 613 cases and 1,190 controls were in included for data analyses (Supporting Information Fig. S1).

The data handling procedures and study protocol were approved by the Research Ethics Committee of Umeå University, Umeå, Sweden (Dnr 03-186). All participants gave a written informed consent at the time of recruitment to the cohorts. All data and samples were de-identified.

Blood sampling and analysis

The blood samples for our study were collected in EDTA sample tubes, separated into plasma, erythrocyte fractions and buffy coat and cryopreserved at −80°C at a central location. Samples were frozen within 1 hr of collection, either at −80°C or at −20°C for up to 1 week before transfer to a −80°C freezer. In the VIP samples are collected in the morning, and only 2% had fasted <4 hr, and 21% <8 hr. In the MSP, blood samples were collected throughout the day, and 96% had fasted for <4 hr.

All biochemical analyses were performed at Bevital A/S (http://www.bevital.no), Bergen, Norway, in 2011. Plasma concentrations of betaine, riboflavin, creatinine, neopterin, symmetric dimethylarginine (SDMA) and PLP were measured with liquid chromatography–mass spectrometry methods (between-day coefficient of variation (CV): 3–13%).33, 34 Plasma concentrations of tHcy, tCys, kynurenine and tryptophan were measured using an isotope dilution gas chromatography–mass spectrometry method (between-day CV: 2–9%).35 Folate and cobalamin concentrations were determined by a microbiological method using Lactobacillus casei and Lactobacillus leichmannii, respectively, adapted to a microtiter plate format and carried out by a robotic workstation (between-day CV: 5%),36, 37 MTHFR rs1801133 genotype was determined using MALDI-TOF mass spectrometry (estimated average error rate of ≤0.1% in duplicated samples).38 The investigators and laboratory staff were blinded to case and control status.

Statistical analyses

All computations were conducted in R v.3.3.2 (R Foundation for Statistical Computing, Vienna, Austria). All tests were two-sided, and p-values < 0.05 were considered statistically significant. Mann–Whitney U-test or χ² tests were used to test for univariate differences in variable distributions between groups.

We calculated total B-vitamin status score by summing log-transformed and standardized (z = x-mean/standard deviation (SD)) plasma concentrations of folate, cobalamin, betaine, PLP and riboflavin. The influence of the B-vitamin score and other variables on tHcy and B-vitamin markers was then evaluated in multiple linear regression models. The relative contribution to total explained variance (R²) for tHcy and B-vitamin markers were assessed using the lmg algorithm in the relaimbo R package. The lmg algorithm decomposes total R² into a contribution for each predictor in a regression model by taking the average R² contribution of each predictor over all possible model sizes and predictor combinations.39 Four linear regression models were fitted, with log-transformed standardized tHcy, Hcy:Cys, Hcy:Cre and Hcy:Cys:Cre as response variables, and sex, age, cohort (VIP or MSP), fasting status (<4 hr, 4–8 hr, ≥8 hr), sample year, smoking status (non-, current-, former-smoker), BMI, kynurenine-to-tryptophan-ratio (KTR), neopterin, SDMA and B-vitamin score as predictor variables. We also evaluated the contribution of individual B-vitamins by estimating the same models but with log-transformed standardized B-vitamins included as separate variables. To compare associations between B-vitamin score and ratio-based B-vitamin markers, we evaluated regression-based sensitivity (defined as B-vitamin marker variance explained by B-vitamin score), regression-based specificity (defined as B-vitamin marker variance explained by B-vitamin score divided by total variance explained) and performance (defined as sensitivity multiplied by specificity). In addition, we also calculated more clinically applied metrics, i.e., diagnostic sensitivity, specificity and area under the receiver operating characteristic (ROC) curve (AUC) for B-vitamin deficiency defined as a B-vitamin score below the lowest 5th percentile in control participants.

To further evaluate the specificity of the ratio-based B-vitamin markers in relation to total B-vitamin status and not as markers of kidney dysfunction and inflammation, we calculated a combined measure of kidney dysfunction and inflammation by summing log-transformed standardized plasma concentrations of the sensitive kidney function marker SDMA,40 and the inflammatory markers KTR and neopterin.41 The score was then used in ROC analysis by defining a dichotomous kidney dysfunction and inflammation variable using the highest 95 percentile of the score in controls as the cut-off.

Odds ratios (ORs) for CRC risk by biomarker levels were calculated using conditional logistic regression, taking the matched control design into account. The markers were stratified by tertiles, with cut-off values based on the distribution of the controls, or as standardized log-transformed continuous variables. ORs were obtained from crude models adjusted only for matching (Model 1), as well as from models adjusted for variables with a plausible link to both exposure and CRC risk. Model 2 was adjusted for smoking status (current-, former-, non-smoker), body mass index (BMI) (<25, 25–30, ≥30 kg/m2), alcohol intake (zero intake, above/below the sex-specific median of self-reported grams of alcohol/day), dietary fiber intake (above/below sex and age-specific (10-year groups)) median of self-reported intake in grams/2,000 kcal), recreational- and occupational physical activity (self-reported on a scale from 1 to 5, ranging between 1 for sedentary and 5 for highly active),18 and kidney dysfunction/inflammation score. Model 3 was additionally adjusted for MTHFR rs1801133 genotype.

For plasma biomarkers, participants with missing data for a variable were excluded from analysis (betaine: n = 8; cobalamin: n = 27; creatinine: n = 8; PLP: n = 9; riboflavin: n = 9; SDMA: n = 8; B-vitamin score: n = 38 and kidney dysfunction/inflammation score: n = 13). Missing values for potential confounders were assigned to a separate category in the multivariable analyses.

The interaction between MTHFR rs1801133 genotype and biomarkers were tested using a likelihood ratio test comparing the interaction model to a model with a main effects model without interaction term. Heterogeneity of the risk associations by tumor subgroups was tested with likelihood ratio tests, comparing a model in which the risk association could vary across subgroups to a model in which all associations were held constant.42

Baseline characteristics

The study cohort consisted of 41% males with a median age of approximately 60 years, most of whom were from the VIP cohort (78%) (Table 1). About 20% of both cases and controls were current smokers, whereas the frequency of former smokers was slightly higher in cases (p = 0.01). Baseline data for BMI, physical activity (occupational and recreational), alcohol intake and dietary fibre intake were comparable for cases and controls. Of the B-vitamins, plasma concentrations of cobalamin, betaine and riboflavin were significantly lower in cases than in controls, and, consequently, the total B-vitamin score was also significantly lower in cases than in controls. Plasma concentrations of the inflammatory marker neopterin were significantly higher in CRC cases than in controls. Of the ratio-based B-vitamin markers, Hcy:Cys was significantly higher in CRC cases than in controls. The median age at diagnosis for CRC cases was 65.2 years (5th–95th percentile: 50.3–76.2 years), and the median follow-up time was 8.2 years from sampling to diagnosis. The CRC cases were approximately equally distributed with respect to tumor location in the right colon (30%), left colon (35%) and rectum (35%), as well as tumor stage (53% stages I–II, 47% stages III–IV).

Associations between B-vitamin markers and B-vitamin score

The predictors (sex, age, cohort, fasting status, smoking, BMI, KTR, neopterin, SDMA and total B-vitamin score) explained about one-third of the variance of tHcy and the ratio-based B-vitamin markers in both cases and controls (Fig. 1). Although contributions to explained variance were slightly higher in cases for some variables (total B-vitamin score and neopterin), there were overall no large differences in between cases and controls. As expected, total B-vitamin score had the strongest association with tHcy and the ratio-based B-vitamin markers (inverse association). Regression-based sensitivity (defined as variance explained by total B-vitamin score) and specificity (defined as variance explained by B-vitamin score divided by total variance explained by model) were 16.6 and 55.9% for tHcy, 23.5 and 72.7% for Hcy:Cys, 21.5 and 73.7% for Hcy:Cre and 27.0 and 83.3% for Hcy:Cys:Cre, respectively, in controls and similar in cases (Table 2). Thus, overall performance (sensitivity multiplied by specificity) was better for the ratio-based B-vitamin markers compared to tHcy (performances: 9.3, 17.1, 15.8 and 22.5 for tHcy, Hcy:Cys, Hcy:Cre and Hcy:Cys:Cre, respectively, in controls). In models with separate B-vitamins as predictors, folate was the strongest contributor for all ratio-based B-vitamin markers (explaining 10–15% of the variance for tHcy and the ratio-based B-vitamin markers), followed by cobalamin, betaine, PLP and riboflavin (each explaining 1–10% of variance) in controls. No other predictors explained >5% of the total variation for tHcy or the ratio-based B-vitamin markers, including markers of kidney function or inflammation.

Regression-based sensitivity, specificity and performance were higher for both tHcy and the ratio-based B-vitamin markers in participants with the homozygous variant MTHFR rs1801133 (TT) genotype (Supporting Information Fig. S2, Table S1). However, differences by genotype were smaller for the ratios, in particular, ratios with Hcy:Cys and Hcy:Cys:Cre, for which performance was above 15 for all genotypes.

In the ROC analysis, results consistent with the regression-based analyses were observed (Fig. 2). The ratio-based B-vitamin markers, of which Hcy:Cys:Cre performed best, were better than tHcy at predicting low total B-vitamin score (AUC 0.75–0.80 for B-vitamin markers, 0.70 for tHcy in controls). AUCs were slightly higher in cases, as in the linear regression analyses (data not shown). The ratio-based B-vitamin markers were weaker predictors of kidney dysfunction/inflammation (AUCs 0.50–0.62) than tHcy (AUC 0.68), again confirming the specificity of the ratios as markers of total B-vitamin status observed in the regression-based analyses.

Associations between B-vitamin markers and CRC risk

ORs for CRC risk by tertiles (categorical) and per standard deviation (SD, continuous) increase in B-vitamin score, and decrease in Hcy:Cys, Hcy:Cre and Hcy:Cys:Cre, are shown in Figure 3. Higher B-vitamin status was associated with a reduced risk of CRC, independently of the observed potential confounders. The negative association for total B-vitamin score was not weakened by the removal of folate from the B-vitamin score calculation (Supporting Information Fig. S3). Associations for ratio-based B-vitamin markers were similar to those for the B-vitamin score. None of the ratio-based marker ORs differed significantly from each other in z-tests based on the estimated standardized regression coefficients and covariance matrices (all p > 0.25). tHcy, cysteine or creatinine included as separate variables were not associated with CRC risk, regardless of whether the variables were mutually adjusted for each other or not (ORs: 0.85–0.90, all p > 0.30, data not shown). We observed no large differences with respect to follow-up time between sampling and diagnosis (p_{heterogeneity} = 0.15–0.76, data not shown).

Next, we evaluated interactions between MTHFR rs1801133 genotype and the total B-vitamin score and the ratio-based B-vitamin markers. Homozygous or heterozygous variant genotypes are weakly associated with a reduced CRC risk in this cohort, as previously reported.19 The group with common homozygous rs1801133 genotype in combination with low B-vitamin status as estimated by total B-vitamin score or the ratio-based markers, had the highest CRC risk in all interaction models. However, there was no multiplicative increase in CRC risk in participants with both homozygous or heterozygous variant rs1801133 genotypes and high B-vitamin status (p_interaction = 0.16 to 0.76, Supporting Information Fig. S4).