Genome-wide association study of circulating folate one-carbon metabolites

Abstract

Experimental, observational and clinical trials support a critical role of folate one-carbon metabolism (FOCM) in colorectal cancer (CRC) development. In this report, we focus on understanding the relationship between common genetic variants and metabolites of FOCM. We conducted a genome-wide association study of FOCM biomarkers among 1788 unaffected (without CRC) individuals of European ancestry from the Colon Cancer Family Registry. Twelve metabolites, including 5-methyltetrahydrofolate, vitamin B2 (flavin mononucleotide and riboflavin), vitamin B6 (4-Pyridoxic acid, pyridoxal and pyridoxamine), total homocysteine, methionine, S-adenosylmethionine, S-adenosylhomocysteine, cystathionine, and creatinine were measured from plasma using liquid chromatography-mass spectrometry (LC-MS) or LC-MS/MS. For each individual biomarker, we estimated genotype array-specific associations followed by a fixed-effect meta-analysis. We identified the variant rs35976024 (at 2p11.2 and intronic of ATOH8) associated with total homocysteine (P=4.9×10⁻⁰⁸). We found a group of six highly correlated variants on chromosome 15q14 associated with cystathionine (all P<5×10⁻⁰⁸), with the most significant variant rs28391580 (P=2.8×10⁻⁰⁸). Two variants (rs139435405 and rs149119426) on chromosome 14q13 showed significant (P<5×10⁻⁸) associations with S-adenosylhomocysteine. These three biomarkers with significant associations are closely involved in homocysteine metabolism. Furthermore, when assessing the principal components (PCs) derived from seven individual biomarkers, we identified the variant rs12665366 (at 6p25.3 and intronic of EXOC2) associated with the first PC (P=2.3×10⁻⁰⁸). Our data suggest that common genetic variants may play an important role in FOCM, particularly in homocysteine metabolism.

Introduction

Folate one-carbon metabolism (FOCM) is a metabolic process in which folate activates and transfers one carbon units to support a wide range of biological processes including nucleotide synthesis and methylation reactions (Ducker & Rabinowitz, 2017). Due to the essential role of FOCM in maintaining genome function and integrity, it is biologically plausible that folate and associated one-carbon metabolites may play an important role in cancer development, in particular for colorectal cancer (CRC). Indeed, the role between these biological reactions and carcinogenesis has been defined in vitro, in vivo (Giovannucci, 2002; Kim, 2003; Mason & Choi, 2000), and even in silico (Nijhout, Reed, Budu, & Ulrich, 2004; Reed, Nijhout, Sparks, & Ulrich, 2004; Ulrich et al., 2008). Epidemiological studies have also shown dietary folates are associated with decreased colorectal neoplasia risk (Giovannucci, 2002; Kennedy et al., 2011). Although supporting evidence from both experimental and observational studies makes folate and one-carbon metabolic cycle a good target to probe as a biomarker in CRC development, the exact role that folate plays is complicated with evidence of a dual role in colorectal carcinogenesis-protection prior to development of neoplastic lesions but promotion of growth after tumor development) (Kim, 2003). Furthermore, Mason et al. documented a significant trend toward increasing CRC incidence in the US and Canada coinciding with the fortification of grain products with folic acid (Mason et al., 2007). Moreover, some recent human studies, including randomized clinical trials, suggest risks for colon, breast or prostate cancers may be increased in those taking high doses of folic acid from supplements (Charles, Ness, Campbell, Davey Smith, & Hall, 2004; Cole et al., 2007; Figueiredo et al., 2009; Hirsch et al., 2009; Stolzenberg-Solomon et al., 2006) in the context of a folic acid fortified diet. Notably, a randomized clinical trial of colorectal adenoma suggested the risk of recurrent advanced colorectal adenoma or multiple adenomas increased with excessive levels of folate (Rees et al., 2017). Several other trials reported no such effect (Logan, Grainge, Shepherd, Armitage, & Muir, 2008; Song et al., 2012; Wu et al., 2009), although they had shorter follow-up periods.

The ability to combine genomics and metabolism data is a powerful tool to quantify and identify potential mechanisms within the biological system in question (Fiehn, 2002; Nicholson, Lindon, & Holmes, 1999). It has been extensively used in biomarker discovery to facilitate disease diagnosis (Madsen, Lundstedt, & Trygg, 2010) and mechanistic dissection of disease pathophysiology (Li et al., 2008). Targeted metabolomics is commonly considered to facilitate the accurate measure of selected endogenous metabolites in the biological samples. With the emergence of liquid chromatography mass spectrometry (LCMS)-based metabolomics, it is possible to profile and even quantify the analytes found in a particular pathway. In this study, we combined genome-wide genotype data with targeted metabolomics to profile the baseline relationships in the FOCM pathway among individuals disease-free of CRC. The goal of this study is to improve our understanding of the genetic contribution to the FOCM pathway in order to better elucidate its role in colorectal carcinogenesis.

Materials and Methods

Study participants.

The study population consisted of disease-free controls (i.e. individuals without CRC) enrolled in the Colon Cancer Family Registry (CCFR) (Newcomb et al., 2007). The CCFR is an international consortium of six study centers, including Sinai Health System (Ontario), Fred Hutchinson Cancer Research Center (FHCRC), Mayo Clinic, University of Hawaii (UHI, not included in this study), University of Southern California/Cedars-Sinai Medical Center (USC/CSMC) consortium, and the University of Melbourne (Australia). Details of the study design have been reported previously (Newcomb et al., 2007). Briefly, CCFR Phase I (1998–2002) focused on recruitment of CRC cases identified from population-based cancer registries and/or clinical centers. Disease-free controls were either age-and sex-matched population-based (Australia, FHCRC, Ontario), CRC case spouses (Mayo) or same-generation (sibling or cousin) family-based (USC/CSMC) consortium participants. CCFR Phase II (2002–2007) focused on recruitment of either CRC cases diagnosed under 50 years of age or clinically identified multi-case families. CCFR Phase II controls were age-and sex-matched from the general population (FHCRC) or case spouses (Mayo) and had no personal history of CRC. The study population had previously participated in a previous genome-wide association study (GWAS) of CRC (Schmit et al., 2018; Schumacher et al., 2015). All controls self-reported as non-Hispanic White. From the available controls with genotyping data, a total of 1,788 disease-free controls were included in the measurement of plasma FOCM and contributed to the analysis. Participants provided written informed consent, and the Institutional Review Boards at each center approved the study.

Genotyping and imputation.

Details of sample collection, genotyping, quality control (QC) and imputation have been reported elsewhere (Schmit et al., 2018). In brief, genotype data was generated from germline DNA on the Affymetrix Axiom, Illumina 1M/1M-Duo, Omni1 and OncoArray. Standard QC filters were applied to the high-density genotype array data at both the individual participant and SNP levels. Quality-controlled genotype data was imputed to the 1,000 Genomes Project (1KGP) Phase 1 multiethnic reference panel (March 2012 release, N=1,092) using SHAPE-IT/IMPUTE2 (Affymetrix Axiom; Illumina 1M/1M-Duo, and Omni1) (Eyre et al., 2012; Howie, Donnelly, & Marchini, 2009) or the 1KGP Phase 3 reference panel (Illumina OncoArray) (Amos et al., 2017). Imputation quality (info score> 0.3) and minor allele frequency filters (MAF ≥ 1%) were imposed on variants prior to the analysis phase. Approximately 44.4% (N=794) of the participants were run on Illumina 1M/1M-Duo while 12.1% (N=217) were on Illumina OncoArray; Affymetrix Axiom and Illumina Omni1 contributed similar proportions (20.5% (N=366) and 23.0% (N=411), respectively).

Biomarker measurement.

Blood samples were collected from participants at study entry to quantify circulating FOCM biomarkers (N=12), including folate (5-methyltetrahydrofolate), vitamin B2 (flavin mononucleotide and riboflavin), vitamin B6 (4-Pyridoxic acid, pyridoxal and pyridoxamine), total homocysteine (tHCY), methionine (MET), S-adenosylmethionine (SAM), S-adenosylhomocysteine (SAH), cystathionine (CYSTA), and creatinine. Detailed information regarding biomarker measurement, including QC is included in Supplementary Materials. We used internal standardization with controls to remove any batch effect.

Statistical analysis.

SNPs with allele frequency < 5% were excluded from the analysis. We only considered a SNP if it was identified in at least three arrays. Principal components (PCs) were calculated with EIGENSTRAT 6.1.4 (version) (Price et al., 2006) and used for ancestral adjustment in analyses. The percentage of samples returning below detection limit (BDL) measurements varied across the biomarkers (0.1%-47%). For biomarkers with a BDL <10%, including creatinine, MET, tHCY, SAH, SAM and CYSTA, we excluded samples with levels falling below the detection limit. We performed a sensitivity analysis replacing all the values falling below the detection limit with the minimal detectable values and results remained similar for biomarkers with a BDL <10% (data not shown). Biomarkers with BDL <10% were analyzed as continuous to achieve optimal power. Biomarkers with a BDL >10%, including FMN, 5-MTHF, pyridoxamine, pyridoxal, 4-pyridoxic acid and riboflavin, were dichotomized as high vs. low using median values as cutoffs. Due to skewed distributions, all the biomarkers were log-transformed except creatinine. Because the ratio of SAM to SAH is considered to reflect methylation potential, including histone methylation (Mentch et al., 2015), we also evaluated the ratio of SAM to SAH in this study. Furthermore, three additional ratios that are involved in FOCM, including the ratios of SAH to tHCY, tHCY to MET, and tHCY to CYSTA (Pacana et al., 2015; Stabler et al., 2013; Yi et al., 2000) were also assessed. The four ratios were dichotomized by median values.

As there is the potential for differential coverage and imputation quality by array, we first estimated genotype array-specific associations followed by a fixed-effect meta-analysis of the four arrays. Associations between imputed genetic dosage and plasma biomarkers were performed for each individual biomarker. Linear (for biomarkers treated as continuous) or logistic (for biomarkers treated as binary) regression models, adjusting for age, sex, centers and PCs of ancestry were used to obtain array-specific association estimates (regression coefficients and standard errors). All the biomarkers except creatinine were further adjusted for estimated Glomerular filtration rate (GFR) which was inferred from creatinine levels, age, sex and race using the Chronic Kidney Disease Epidemiology Collaboration equation (Levey et al., 2009) because GFR is associated with circulating one-carbon metabolite levels, particularly for tHCY (Francis, Eggers, Hostetter, & Briggs, 2004). An array-wide meta-analysis using fixed-effect models with inverse variance weighting was implemented in METAL (Willer, Li, & Abecasis, 2010).

Because the biomarkers were moderately correlated in this study (Supplementary Figure 1), we further evaluated biomarkers (those with BDL <15%, including creatinine, tHCY, MET, SAH, SAM, FMN and CYSTA) using principal component analysis (PCA) to derive patterns that best explained the maximal variation in these one-carbon metabolites. Since biomarker levels varied by study centers (Table 1 and Supplementary Table 2), for the PCA we first regressed each individual biomarker on variables known to be correlated with the biomarker levels (i.e. age, sex, current alcohol consumption status, current cigarette smoking status and body mass index [BMI]) in each study center and extracted the residuals from the multiple linear regression models. PCA was then performed on the residuals pooled from all study centers and principal components of the biomarkers were derived. We then evaluated the association between imputed genetic dosage and the biomarker principal components using similar analysis as we did for individual biomarker. We conducted analysis for the top four biomarker components accounting for approximately 70% of the variance in biomarker levels (20.3%, 18.5%, 15.6% and 14.7% respectively), respectively.

Table 1.

Study population characteristics and circulating biomarker levels according to arrays used for genotyping.

	GWAS Array
	Affymetrix Axiom (N=366)		Illumina 1M/1M Duo (N=794)		Illumina Omni1 (N=411)		Illumina OncoArray (N=217)
	Mean	SD	Mean	SD	Mean	SD	Mean	SD	P for differences
Age, yrs	52.3	11.5	59.1	11.0	55.1	10.9	50.0	10.6	<0.001
BMI, kg/m2	26.8	5.3	26.6	5.0	26.8	4.9	27.2	5.4	0.50
	N	%	N	%	N	%	N	%
Female	200	54.6	410	51.6	228	55.5	101	46.5	0.14
Current alcohol intake status, yes	223	60.9	517	65.1	207	50.4	104	47.9	<0.001
Current smoking status, yes	49	13.4	84	10.6	58	14.1	20	9.2	0.13
A history of adenomas	83	22.7	91	11.5	115	28.0	15	6.9	<0.001
Center									<0.001
Ontario	90	24.6	384	48.4	104	25.3	0
USC/CSMC consortium	79	21.6	0	0	114	27.7	0
Australia	93	25.4	173	21.8	10	2.4	0
Mayo	37	10.1	0	0	98	23.8	0
Seattle	67	18.3	237	29.8	85	20.7	217	100
	Median	IQR	Median	IQR	Median	IQR	Median	IQR
Creatinine, mg/dl	2.4	1.3	2.3	1.0	2.3	1.3	2.0	0.8	<0.001
SAH, ng/ml	9.4	12.7	11.2	15.9	11.4	12.9	5.1	7.3	<0.001
tHCY, umol/l	13.2	11.2	11.5	8.2	11.6	8.8	8.7	8.5	<0.001
METH, umol/l	15.4	16.5	31.4	24.7	14.9	16.9	22.8	20.6	<0.001
CYSTA, ng/ml	35.8	60.4	47.0	66.3	51.2	65.8	86.0	96.9	<0.001
SAM, ng/ml	90.5	219.4	142.5	208.3	56.7	170.8	98.8	195.5	<0.001
Riboflavin, nmol/l	1.1	11.7	0.6	18.0	1.9	16.2	15.3	43.3	<0.001
FMN, ng/ml	2.7	12.0	11.4	30.0	3.0	15.7	6.2	11.5	<0.001
Pyridoxic acid, ng/ml	2.5	13.9	2.1	24.4	7.2	23.1	7.3	23.2	0.0003
Pyridoxal, ng/ml	2.5	38.0	4.1	50.7	0.6	5.5	2.9	10.3	<0.001
Pyridoxamine, ng/ml	1.9	2.1	2.1	1.9	2.1	2.5	1.5	2.0	0.001
5-MTHF, ng/ml	6.3	16.2	9.0	19.0	10.0	17.1	21.3	24.4	<0.001
Ratio of SAM to SAH	5.2	20.4	12.7	33.8	2.5	12.4	13.9	30.8	<0.001
Ratio of SAH to HCY	0.005	0.01	0.01	0.01	0.01	0.01	0.005	0.01	<0.001
Ratio of tHCY to METH	0.7	1.2	0.3	0.5	0.6	0.8	0.3	0.3	<0.001
Ratio of tHCY to CYSTA	54.8	139.9	30.8	68.3	35.9	90.5	15.0	28.9	<0.001

Abbreviation: SD, standard deviation. IQR, interquartile range. SAH, S-adenosylhomocysteine. SAM, S-adenosylmethionine. tHCY, total Homocysteine.

CYSTA, Cystathionine. FMN, Flavin-mononucleotide. 5-MTHF, 5-methyltetrahydrofolate.

Finally, as there are candidate genes known to play important roles in FOCM, we selected 24 genes (Supplementary Table 1) involved in FOCM according to prior studies (Cheng et al., 2015; Figueiredo, Levine, Crott, Baurley, & Haile, 2013; Levine et al., 2010; Ose et al., 2018). We used the ANNOVAR software to characterize the SNPs with gene-based annotation (K. Wang, Li, & Hakonarson, 2010) and included coding exonic, intronic and non-coding variants (Supplementary Table 1). A total of 1,316 SNPs associated with the 24 FOCM-related genes were identified and included in the analysis. For all the SNPs within a single gene, P values for SNPs were adjusted for multiple testing using Pact which takes into account the correlation among SNPs within a gene (Conneely & Boehnke, 2007). For a test within a single gene, an α-level of 0.05 after implementing Pact was used to determine statistical significance; across all 33 genes tested, a Bonferroni corrected α-level of 0.002 (0.05/33) was considered as the threshold.

For the identified genome-wide significant variants, we used ANNOVAR (K. Wang et al., 2010), HaploReg v4.1 (Ward & Kellis, 2012) and RegulomeDB (Boyle et al., 2012) to annotate the functional aspects of the variants. All statistical analysis was performed using R software (version 3.3.3).

Results

A total of 1,788 participants who had both imputed genetic data and biomarker measurements contributed to the analysis. We compared participants’ characteristics and plasma biomarker levels across the four arrays (Table 1 and Supplementary Table 2). The distributions of sex, BMI and current smoking status were similar across arrays while age, current alcohol consumption status, a history of adenomas and study center were significantly different by array. For instance, participants genotyped using Illumina 1M/1M-Duo array were older than those genotyped on the other three arrays and more likely to be current alcohol drinkers than those genotyped on Omni1 or OncoArray (Supplementary Table 2). This difference reflects study center differences – for example the Affymetrix Axiom included participants from all 5 participating study centers in this study while the OncoArray data for this investigation had participants from the Seattle center only. Plasma levels of all the biomarkers as well as the ratios of individual metabolites were significantly different across arrays, although the levels were similar between Illumina 1M/1M-Duo and Omni1. Small to moderate correlations were observed among seven one-carbon metabolites (Supplementary Figure 1). The strongest correlation was observed between SAH and MET (Spearman r = 0.31; P < 0.05). Plasma tHCY was significantly inversely correlated with MET (Spearman r = −0.25) and one of the vitamin B2 co-factors (i.e. FMN, Spearman r = −0.2), and positively correlated with SAM (Spearman r = 0.19).

Several genome-wide significant associations were identified with plasma tHCY, CYSTA and SAH in the meta-analysis of the four arrays. Variant rs35976024, located at 2p11.2 and intronic of ATOH8, was associated with tHCY levels: the A allele was linked to an approximately 10% decrease in tHCY levels (P = 4.89E-08; Table 2, Supplementary Table 3 and Figure 1). The inverse association was found in participants tested in all four arrays with similar effect estimates (P_{heterogeneity} = 0.91). Six variants, located at Chr15, were found significantly associated with CYSTA levels. Variant rs28391580 was the most significant (P = 2.82E-08; Table 2 and Figure 1) and was highly correlated with the other 5 SNPs (r²: 0.86 – 1.0; the highest correlation with rs28416399: r² = 1.0). rs28391580 maps to 15q14 with the nearest gene TMCO5A (approximately 200kb away; Supplementary Table 3 and Supplementary Figure 2). In the meta-analysis, the effect estimates of all the 6 SNPs for plasma CYSTA were very similar (the effect allele was associated with approximately 20% decreased CYSTA levels). Two variants (moderately correlated in this study, r² = 0.57), located at chr14, were identified as significantly associated with SAH levels in the meta-analysis of three of the arrays except the OncoArray (SNPs excluded due to poor quality). The stronger association was seen with variant rs139435405, which maps to 14q13 with two nearest genes being PTCSC3 and MBIP (approximately 5kb and 112kb away, respectively; Supplementary Table 3 and Supplementary Figure 2). The effect alleles of both SNPs (i.e. T and C for variants rs139435405 and rs149119426, respectively) was associated with about 26% increased SAH levels. Regional association plots for the genome-wide significant variants in Supplementary Figure 2 depict the meta-analysis of GWAS results in the context of their surrounding linkage disequilibrium (LD) structures and nearby genes. When assessing the four ratios (i.e. SAM to SAH, SAH to tHCY, tHCY to MET and tHCY to CYSTA) which were treated as high vs. low, we did not find genome-wide significant variants that were associated with any of the four ratios (Supplementary Figure 3).

Table 2.

Genome-wide significant variants^† for circulating total HCY, CYSTA and SAH levels.

						Meta-analysis			Affymetrix Axiom			Illumina 1M/1M Duo			Illumina Omni1			Illumina OncoArray
rsID	Chr	Position (BP)	Gene	Allele Eff/Ref	Eff allele freq	Beta	S.E.	P	Beta	S.E.	P	Beta	S.E.	P	Beta	S.E.	P	Beta	S.E.	P	Phet
tHCY
rs35976024	2	86003971	ATOH8	A/G	0.32	−0.10	0.02	4.89E-08	−0.10	0.05	0.06	−0.11	0.02	2.11E-06	−0.11	0.05	0.04	−0.04	0.09	0.62	0.91
CYSTA
rs28391580	15	38033956		T/C	0.27	−0.24	0.04	2.82E-08	−0.24	0.10	0.02	−0.22	0.06	1.62E-04	−0.34	0.11	0.001	−0.18	0.11	0.11	0.75
rs28416399	15	38028526		G/A	0.27	−0.24	0.04	2.97E-08	−0.24	0.10	0.02	−0.22	0.06	1.75E-04	−0.34	0.11	0.001	−0.18	0.11	0.11	0.74
rs765683	15	38018791		T/C	0.30	−0.23	0.04	3.20E-08	−0.27	0.10	0.01	−0.18	0.06	1.21E-03	−0.35	0.10	0.001	−0.21	0.11	0.05	0.53
rs4924190	15	38017922		T/C	0.25	−0.24	0.04	3.30E-08	−0.23	0.10	0.03	−0.22	0.06	2.69E-04	−0.39	0.11	3.26E-04	−0.17	0.11	0.12	0.52
rs11856650	15	38025137		A/G	0.27	−0.23	0.04	3.75E-08	−0.24	0.10	0.02	−0.22	0.06	2.24E-04	−0.34	0.11	0.001	−0.18	0.11	0.11	0.74
rs8023982	15	38021714		A/G	0.27	−0.23	0.04	4.08E-08	−0.24	0.10	0.02	−0.22	0.06	2.23E-04	−0.33	0.11	0.001	−0.18	0.11	0.11	0.76
SAH
rs139435405	14	36655451		C/T	0.75	0.23	0.04	1.95E-08	0.27	0.09	0.002	0.19	0.05	5.43E-04	0.27	0.08	0.001	N/A	N/A	N/A	0.62
rs149119426	14	36656744		T/C	0.63	0.23	0.04	4.26E-08	0.20	0.09	0.02	0.20	0.06	3.73E-04	0.30	0.08	2.02E-04	N/A	N/A	N/A	0.61
Biomarker PC1
rs12665366	6	541907	EXOC2	G/A	0.12	−0.35	0.06	2.33E-08	−0.44	0.15	0.003	−0.26	0.08	0.001	−0.43	0.16	0.005	−0.80	0.26	0.002	0.18

Abbreviation: Chr, chromosome. S.E., standard error. PC, principal component. tHCY, total Homocysteine. CYSTA, Cystathionine. SAH, S-adenosylhomocysteine. P_het. P_{heterogeneity.}

^†Variants identified in at least three arrays.

Figure 1.

Meta-analysis of three or more arrays for circulating folate one-carbon metabolites. Manhattan plots for a) total Homocysteine, b) Cystathionine and c) S-adenosylhomocysteine and d) biomarker PC1.

When the seven metabolites (i.e. creatinine, MET, SAH, SAM, tHCY, CYSTA and FMN) were analyzed as principal components, the first PC correlated positively with all seven biomarkers. The strongest correlation was with tHCY (eigenvector = 0.63), followed by creatinine, SAM and SAH (eigenvector = 0.49, 0.38 and 0.32, respectively) (Supplementary Table 4). The G allele of variant rs12665366, located at 6p25.3 and intronic of EXOC2, was inversely associated with PC1 (P = 2.33E-08; Table 2 and Figure 1). No significant heterogeneity in array-specific associations was observed except the effect estimate from the OncoArray was stronger (P_{heterogeneity} = 0.18).

We further evaluated the 1,316 SNPs in 24 genes that are known to play important roles in FOCM according to prior studies (Cheng et al., 2015; Levine et al., 2010; Ose et al., 2018). Out of the 1,316 SNPs, 39 were mapped to coding exonic variants and the rest (1,277) were intronic or non-coding variants (Supplementary Table 1). After taking into account multiple comparisons, we observed several SNPs within a single gene that were significantly (Pact < 0.05) associated with a specific one-carbon metabolite (Table 3a and 3b). For instance, six SNPs of MTHFR, including one of the well-studied SNP (rs1801131, located at 1p36.3 and exonic of MTHFR), were associated with plasma tHCY levels. The A allele of variant rs1801131 was associated with approximately 7% increased tHCY levels in the meta-analysis and similar positive associations were observed across the four arrays.

Table 3.

Candidate gene SNPs involved in folate one-carbon metabolism and circulating folate one-carbon metabolites^† in meta analysis (Table 3a) and by each array (Table 3b). Table 3a.

Table 3a.
							Meta-analysis
rsID	Chr	Position (BP)	Gene	Location	Allele Eff/Ref	Eff allele freq	Beta	S.E.	P	Pact
Creatinine
rs12485165	22	31020341	TCN2	Intronic	T/G	0.13	0.16	0.04	3.61E-04	0.01
rs150167372	22	31020114	TCN2	Intronic	C/T	0.88	0.16	0.05	4.92E-04	0.01
rs9621049	22	31013419	TCN2	Exonic	T/C	0.12	0.16	0.05	5.01E-04	0.01
rs12168392	22	31022301	TCN2	Intronic	A/G	0.12	0.16	0.05	5.72E-04	0.01
rs4820886	22	31016539	TCN2	Intronic	G/T	0.88	0.16	0.05	6.11E-04	0.01
rs1004474	18	660383	TYMS	Intronic	G/A	0.54	−0.08	0.03	4.84E-03	0.04
HCY
rs1801131	1	11854476	MTHFR	Exonic	G/T	0.69	0.07	0.02	7.89E-04	0.01
rs7526128	1	11857240	MTHFR	Intronic	T/C	0.61	−0.06	0.02	2.36E-03	0.03
rs1476413	1	11852300	MTHFR	Intronic	T/C	0.26	0.06	0.02	2.54E-03	0.03
rs6541005	1	11857525	MTHFR	Intronic	A/T	0.43	−0.06	0.02	2.68E-03	0.03
rs3818762	1	11851003	MTHFR	Intronic	C/G	0.27	0.06	0.02	4.08E-03	0.04
rs17367504	1	11862778	MTHFR	Intronic	A/G	0.85	0.07	0.03	5.14E-03	0.04
SAM
rs4911253	20	31352585	DNMT3B	Intronic	A/G	0.56	−0.17	0.06	3.00E-03	0.02
rs2424905	20	31352927	DNMT3B	Intronic	T/C	0.57	−0.16	0.06	4.00E-03	0.03
rs6087995	20	31358253	DNMT3B	Intronic	A/C	0.57	−0.16	0.06	5.00E-03	0.04
rs72765189	9	140525434	EHMT1	Intronic	A/G	0.11	0.28	0.09	2.19E-03	0.02
rs190756635	9	140530183	EHMT1	Intronic	G/A	0.89	0.28	0.09	2.66E-03	0.03
rs111275198	9	140530274	EHMT1	Intronic	A/G	0.89	0.27	0.09	2.80E-03	0.03
rs57705093	5	162944172	MAT2B	Intronic	T/A	0.73	−0.23	0.06	3.25E-04	0.005
5-MethyTHF
rs2235523	1	14096457	PRDM2	Intronic	A/T	0.37	0.29	0.09	1.22E-03	0.04
rs3891167	18	658423	TYMS	Intronic	G/A	0.75	−0.33	0.11	2.09E-03	0.02
METH
rs34973186	11	59598262	GIF	Intronic	T/C	0.07	0.17	0.05	1.92E-03	0.01
Pyridoxal acid
rs3788202	21	46957218	SLC19A1	Intronic	G/A	0.26	0.28	0.09	2.65E-03	0.03
rs12801088	11	4116335	RRM1	Intronic	A/C	0.08	−0.42	0.16	7.07E-03	0.04
Flavin mononucleotide
rs34973186	11	59598262	GIF	Intronic	T/C	0.07	−0.29	0.11	1.10E-02	0.03
Pyridoxamine
rs2244171‡	21	44496594	CBS	Intronic	T/G	0.08	0.58	0.22	8.08E-03	N/A
rs3784621	15	48633092	DUT	Intronic	C/T	0.82	−0.42	0.15	4.11E-03	0.01
rs10851465	15	48629884	DUT	Intronic	T/C	0.18	−0.41	0.15	4.96E-03	0.01
rs11637235	15	48633153	DUT	Intronic	T/C	0.78	−0.36	0.14	8.48E-03	0.02
rs3784619	15	48625938	DUT	Intronic	A/G	0.85	−0.33	0.15	2.80E-02	0.04
rs12592155	15	48623500	DUT	Intronic	A/C	0.15	0.33	0.15	2.85E-02	0.03
rs12592157	15	48623524	DUT	Intronic	C/G	0.85	0.33	0.15	2.92E-02	0.03

Table 3b.
	Affymetrix Axiom			Illumina 1M/1M Duo			Illumina Omni1			Illumina OncoArray
rsID	Beta	S.E.	P	Beta	S.E.	P	Beta	S.E.	P	Beta	S.E.	P
Creatinine
rs12485165	0.18	0.13	0.18	0.10	0.06	0.10	0.22	0.10	0.03	0.22	0.10	0.02
rs150167372	0.17	0.14	0.23	0.13	0.07	0.05	0.20	0.10	0.05	0.20	0.10	0.05
rs9621049	0.17	0.14	0.21	0.12	0.06	0.05	0.20	0.10	0.05	0.20	0.10	0.05
rs12168392	0.17	0.14	0.20	0.12	0.06	0.06	0.19	0.10	0.05	0.20	0.10	0.05
rs4820886	0.17	0.14	0.21	0.12	0.06	0.05	0.19	0.10	0.06	0.20	0.10	0.05
rs1004474	−0.12	0.08	0.13	−0.09	0.04	0.03	−0.005	0.07	0.94	−0.09	0.06	0.13
HCY
rs1801131	0.14	0.05	0.01	0.06	0.02	0.02	0.05	0.06	0.38	0.07	0.09	0.41
rs7526128	−0.12	0.05	0.01	−0.04	0.02	0.06	−0.05	0.05	0.28	−0.07	0.08	0.41
rs1476413	0.14	0.06	0.01	0.05	0.03	0.04	0.03	0.06	0.65	0.08	0.09	0.37
rs6541005	−0.14	0.05	0.005	−0.04	0.02	0.09	−0.05	0.05	0.35	−0.09	0.09	0.30
rs3818762	0.14	0.05	0.01	0.05	0.03	0.05	0.04	0.06	0.51	0.04	0.09	0.62
rs17367504	0.06	0.07	0.44	0.08	0.03	0.01	0.09	0.07	0.19	−0.01	0.11	0.95
SAM
rs4911253	−0.15	0.13	0.26	−0.23	0.08	0.002	−0.16	0.14	0.24	0.12	0.18	0.49
rs2424905	−0.14	0.13	0.27	−0.23	0.08	0.003	−0.15	0.14	0.27	0.15	0.18	0.39
rs6087995	−0.16	0.13	0.22	−0.22	0.08	0.004	−0.13	0.14	0.33	0.11	0.18	0.55
rs72765189	0.32	0.21	0.11	0.34	0.13	0.01	0.29	0.24	0.20	−0.01	0.27	0.97
rs190756635	0.33	0.21	0.10	0.35	0.13	0.01	0.18	0.22	0.40	0.02	0.27	0.94
rs111275198	0.32	0.21	0.12	0.33	0.13	0.01	0.23	0.22	0.28	0.01	0.27	0.96
rs57705093	0.03	0.15	0.82	−0.38	0.09	8.3E-06	−0.14	0.15	0.34	−0.02	0.19	0.92
5-MethyTHF
rs2235523	0.31	0.21	0.13	0.20	0.13	0.13	0.33	0.20	0.09	0.59	0.27	0.02
rs3891167	−0.65	0.25	0.01	−0.15	0.16	0.33	−0.49	0.25	0.05	−0.32	0.28	0.26
METH
rs34973186	0.07	0.13	0.57	0.20	0.07	0.01	0.07	0.14	0.62	0.23	0.13	0.06
Pyridoxal acid
rs3788202	0.63	0.21	0.003	0.21	0.14	0.13	0.30	0.20	0.13	0.01	0.25	0.98
rs12801088	−0.29	0.35	0.42	−0.27	0.22	0.22	−0.84	0.36	0.02	−0.60	0.45	0.17
Flavin mononucleotide
rs34973186	−0.11	0.20	0.57	−0.36	0.19	0.06	−0.34	0.27	0.20	−0.42	0.28	0.13
Pyridoxamine
rs2244171‡	−0.01	0.42	0.97	0.99	0.36	0.004	0.75	0.61	0.21	0.54	0.47	0.26
rs3784621	−0.21	0.37	0.58	−0.51	0.22	0.02	−0.08	0.37	0.83	−0.58	0.29	0.05
rs10851465	−0.15	0.37	0.69	−0.52	0.22	0.02	−0.06	0.38	0.88	−0.59	0.29	0.05
rs11637235	−0.14	0.35	0.68	−0.35	0.20	0.08	−0.33	0.35	0.34	−0.54	0.28	0.06
rs3784619	−0.12	0.38	0.75	−0.43	0.23	0.05	−0.06	0.40	0.88	−0.45	0.30	0.14
rs12592155	0.13	0.38	0.73	0.43	0.23	0.05	0.06	0.40	0.89	0.43	0.30	0.15
rs12592157	0.13	0.38	0.73	0.43	0.23	0.05	0.05	0.40	0.90	0.43	0.30	0.15

Abbreviation: SNP, Single nucleotide polymorphism. tHCY, total Homocysteine. SAM, S-adenosylmethionine. FMN, Flavin-mononucleotide. 5-MTHF, 5-methyltetrahydrofolate. MET: Methionine.

^†Only SNPs with Pact < 0.05 presented.

^‡1 single SNP was included for CBS.

Finally, since several SNPs have been found significantly associated with circulating HCY, vitamin B12 and vitamin B6 levels in prior GWA studies (Hazra et al., 2009; Hazra et al., 2008; Tanaka et al., 2009), we assessed those variants in this study. We were only able to assess prior identified SNPs for plasma tHCY because vitamin B12 was not evaluated in our study and different metabolites of vitamin B6 were assessed in our study compared to previous ones (Hazra et al., 2009; Tanaka et al., 2009). However, we did not observe significant associations between previously identified SNPs and plasma tHCY levels (Supplementary Table 5).

Discussion

Genetic factors have long been hypothesized to influence circulating levels of folate and associated metabolites (Hustad et al., 2007; Nilsson, Read, Berg, & Johansson, 2009; Siva et al., 2007; Thuesen et al., 2010), however, only a few studies conducted genome-wide assessment of the genetic determinants of biomarkers involved in FOCM (Hazra et al., 2009; Hazra et al., 2008; Tanaka et al., 2009). Furthermore, none of the prior studies evaluated genetic determinants for metabolites other than B vitamins involved in FOCM, including MET, CYSTA and SAM. Circulating concentrations of several those metabolites, such as MET, have been found to be associated with CRC risk, individually (Myte et al., 2016) or together with folate levels (Nitter et al., 2014). Thus, we performed a GWAS to identify common genetic variants that influence plasma levels of 12 FOCM biomarkers among 1,788 unaffected (free of cancer) participants. We identified variant rs35976024 (located at 2p11.2 and intronic of ATOH8), 6 variants on chromosome 15q14 and 2 variants on chromosome 14q13 which demonstrated significant (P < 5 × 10⁻⁸) associations with tHCY, CYSTA and SAH, respectively. Additionally, when assessing principal components derived from 7 individual metabolites, we found that variant rs12665366 (located at 6p25.3 and intronic of EXOC2) was significantly associated with the first principal component, a component characterized with a strong positive correlation with plasma tHCY and moderate correlation with SAM, SAH, CYSTA and creatinine.

The locus (rs35976024) which was found to be significantly associated with plasma tHCY is located at chromosome 2 (2p11.2) and intronic of ATOH8. The variant rs35976024 overlaps with an enhancer (i.e. H3K4me1) detected in colon tissue and is likely to affect transcription factor binding site and result in the binding motif changes (Supplementary Table 3). ATOH8 basic-helix-loop-helix transcription factor involved in embryogenesis (B. Wang, Balakrishnan-Renuka, Napirei, Theiss, & Brand-Saberi, 2015) and the development of multiple tissues (Fang et al., 2014; Guttsches et al., 2015; Inoue et al., 2001; Lynn, Sanchez, Gomis, German, & Gasa, 2008). It has been reported to be associated with tumor progression in several types of cancer, including CRC (Ye et al., 2017). ATOH8 expression measured by immunohistochemistry in CRC tumor tissue was significantly higher than that in tumor-adjacent normal tissue and was associated with a worse overall survival (Ye et al., 2017). Copy number amplification of ATOH8 was found in glioblastoma tissue (Freire et al., 2008). On the other hand, ATOH8 mRNA expression was found substantially lower in tumor tissue than in tumor-adjacent normal tissue in other types of cancer, including hepatocellular carcinoma (Song et al., 2015) and Nasopharyngeal carcinoma (Z. Wang et al., 2016). In addition, the decreased expression of ATOH8 in hepatocellular carcinoma was associated with a significant reduction in disease-free survival (Song et al., 2015). However, despite the apparent role of the ATOH8 gene in tumor progression, whether it is involved in FOCM remains largely unknown. Variants identified for plasma CYSTA or SAH are intergenic. A group (N=6) of highly correlated SNPs at chromosome 15 (15q14) were found to be associated with circulating CYSTA levels. Although it is likely that this specific region may influence CYSTA levels, the functional analysis of this region is largely unknown. These 6 variants appeared not to influence regulatory elements, such as enhancer or promoter elements, although some of them may lead to binding motif changes (Supplementary Table 3). The nearest gene to the 6 variants is TMCO5A which is approximately 200kb upstream of them. For plasma SAH, although the strongest SNP (rs139435405) is also intergenic, it is located approximately 5kb downstream of PTCSC3 which is a thyroid-specific long non-coding RNA. PTCSC3 is substantially down-regulated in papillary thyroid carcinoma (PTC) and appears to act as a tumor suppressor gene (Fan et al., 2013; Jendrzejewski et al., 2012). Despite no apparent association between folate status and thyroid cancer, the C677T polymorphism in MTHFR gene (rs1801133), a known SNP that affects folate metabolism, has been reported to be associated with thyroid cancer risk (Vu-Phan & Koenig, 2014). However, this variant rs139435405 does not appear to alter regulatory elements (Supplementary Table 3).

Given the correlation between metabolites involved in FOCM, we assessed 7 markers (i.e. those with BDL <15%, including creatinine, tHCY, MET, SAH, SAM, FMN and CYSTA) using PCA to derive patterns that best capture the maximal variation in these metabolites. We identified a SNP rs12665366, located at chromosome 6 (6p25.3) and intronic of EXOC2, that was significantly associated with the first PC. However, no evidence has shown that the variant rs12665366 may substantially affect regulatory elements (e.g. enhancer or promoter; Supplementary Table 3). The first PC accounts approximately 20.3% variation of the 7 metabolites and was positively correlated most strongly with tHCY. A SNP rs1540771, located at the same 6p25.3 region and lying between IRF4 and EXOC2, was identified in prior GWAS to be significantly associated with the presence of freckles in Europeans (Sulem et al., 2007), but it is not correlated with rs12665366 (r² < 0.01; ~76kb away from it). Furthermore, another intronic variant in EXOC2 (rs9328342) which was approximately 192kb away from, but again not correlated with rs12665366 (r² < 0.01), has been reported to be associated with serum 25-hydroxyvitamin D levels (Saternus et al., 2015). Therefore, if the association of rs12665366 with biomarkers of FOCM, particularly tHCY, is replicated in future studies, it suggests that 6p25.3 region may be biologically-relevant to metabolism of multiple nutrients.

Of the 12 one-carbon metabolites assessed individually or using principal component analysis, we identified genome-wide significant variants that are associated with plasma tHCY, CYSTA or SAH. Results from our study suggest that genetic factors may play an important role in HCY metabolism. HCY is metabolized through two major pathways: re-methylation and transsulfuration (Selhub, 1999). In re-methylation, HCY acquires a methyl group from 5-MTHF and vitamin B12 folate to form MET. Alternatively, it acquires the methyl group from betaine to form MET, independent of vitamin B12. In transsulfuration, HCY condenses with serine to irreversibly form CYSTA, which requires pyridoxal-5’-phosphate (the active form of vitamin B6). A substantial proportion of MET is activated to form SAM, the universal methyl group carrier. SAH which is generated by demethylation of SAM is then hydrolyzed to form HCY, which becomes available for a new cycle of methyl group transfer. In addition, changes in MET levels can lead to changes in SAM/SAH ratio (Mentch et al., 2015) that has a profound impact on methylation reactions. HCY which is not re-methylated to MET or transsulfurated to CYSTA is quickly exported to circulation. Thus, our findings generally support that genetic variants may influence pathways related to HCY metabolism. Furthermore, our results are consistent with findings of relatively high heritability of circulating levels of tHCY in previous studies. Heritability calculated from utilizing monozygotic (MZ) and dizygotic (DZ) twins is estimated to be more than 50% for circulating folate (Nilsson et al., 2009) and approximately 60% for circulating tHCY (Nilsson et al., 2009; Siva et al., 2007). Previous studies focusing on key genes involved in FOCM found a variant in MTHFR (i.e. C677T, rs1801133) to be associated with circulating folate and tHCY levels (Hustad et al., 2007; Thuesen et al., 2010). Thus, if replicated in future studies, our results provide new insights in the genetic influence on folate-mediated metabolism, particularly HCY metabolism. This is interesting given that high tHCY levels have been reported to be associated with CRC risk (Miller et al., 2013).

A major strength of this study is the genome-wide assessment of genetic determinants not only for B vitamins (i.e. folate, vitamin B2, vitamin B6) but also other components involved in FOCM, while previous studies focused mostly on B vitamins. Another major strength is the use of a single laboratory for the measurement of all the biomarkers despite participants in this study being recruited from multiple sites. Furthermore, we (Louie and Asante) have developed a validated and more sensitive and precise assay to quantify blood B vitamins. This validated multi-analyte LC-MS method can simultaneously measure the endogenous plasma levels of metabolites involved in FOCM (Asante et al., 2018). However, our study also has several limitations. A major limitation is the lack of validation of the identified GWAS SNPs in our studies. However, to our best knowledge, there seem no existing studies which conducted GWA analysis on biomarkers involved in one-carbon metabolism in addition to B vitamins. Another limitation was that participants were genotyped on 4 different arrays with a very small number of overlapping SNPs (N=26,797). However, we conducted a rigorous QC on genotyped SNPs and thus were able to generate accurately imputed SNP data. Furthermore, although this is a large study for the measurement of biomarkers, the sample size is small for a GWAS, possibly contributing to the lack of variants identified to have associations with plasma B vitamins. Other limitations include lack of racial/ethnic diversity (i.e. European ancestry only) in the study population and lack of information on dietary folic and methionine intake. FOCM-related dietary intake may modify the association of FOCM-related genetic variants with colon cancer risk (Liu et al., 2013); however, we were not able to perform the G-E interaction tests in this study (e.g. assess whether dietary folic and/or methionine intake may modify the SNP -FOCM biomarker association). This may be warranted further investigation in future larger-scale studies with information on FOCM-related dietary intake.

In conclusion, in this genome-wide association analysis of biomarkers involved in FOCM, some of which were assessed for the first time to our knowledge, we identified several genetic variants or regions which are associated with circulating tHCY, SAH and CYSTA. As these biomarkers are specifically involved in the pathway of HCY metabolism, if replicated in future studies, results from our study provides support that common gene variants may play an important role in FOCM, particularly HCY.

Abbreviations:

GWAS

Genome-wide Association Study

FOCM

Folate One-carbon Metabolism

MET

Methionine

tHCY

total Homocysteine

SAH

S-adenosylhomocysteine

SAM

S-adenosylmethionine

CYSTA

Cystathionine

FMN

Flavin-mononucleotide

5-MTHF

5-methyltetrahydrofolate

CRC

colorectal cancer

CCFR

Colon Cancer Family Registry

BDL

Below Detection Limit

LD

Linkage Disequilibrium

PTC

Papillary Thyroid Carcinoma

PCA

Principal Component Analysis

LC-MS

Liquid chromatography–mass spectrometry