One-Carbon metabolism gene polymorphisms are associated with cognitive trajectory among African-American adults

Abstract

The sex-specific link between longitudinal annual rate of cognitive change (LARCC) and polymorphisms in One-carbon metabolism enzymatic genes remains unclear, particularly among African-American adults. We tested associations of fourteen single nucleotide polymorphisms (SNPs) from MTHFR, MTRR, MTR and SHMT genes and select MTHFR haplotypes and latent classes (SNPHAP/SNPLC) with LARCC. Up to 797 African-American participants in the Healthy Aging in Neighborhoods of Diversity across the Lifespan (HANDLS) study (Age:30–64y, 52% women) had 1.6–1.7 (i.e. 1 or 2) repeated measures (Follow-up time, mean=4.69y) on 9 cognitive test scores, reflecting verbal and visual memory, verbal fluency, psychomotor speed, attention and executive function: California Verbal Learning test [CVLT-List A (immediate recall), CVLT-DFR (delayed free recall)], Benton Visual Retention Test (BVRT), Animal Fluency (AF), Digits Span Forward and Backwards tests, and Trailmaking tests (Trails A and B). Multiple linear mixed-effects and multiple linear regression models were conducted. Overall, MTHFR SNPs rs4846051(A1317G, G>A) and rs1801131(A1298C, G>T) were associated with slower and faster declines in AF, respectively, while rs2066462(C1056T, A>G) was related to slower decline on Trails B (executive function). Among men, rs4846051(A1317G, G>A) was linked to faster decline on BVRT (visual memory), while rs2066462(C1056T, A>G) and rs9651118(C>T) were associated with slower decline on CVLT-List A and rs9651118(C>T) with faster decline on CVLT-DFR. Among women, a slower decline on the domain “verbal memory/fluency” was observed with rs1801133(C677T, A>G). MTHFR₂ SNPHAP (rs1801133(C677T, A>G)/rs1801131(A1298C,G>T): GG) was associated with slower decline on AF among women, while MTHFR₃ SNPHAP(AT) was linked with slower decline on CVLT-List A among men but faster decline on “verbal memory/fluency” among women. Similar patterns were observed for MTHFR SNPLCs. In sum, MTHFR gene variations can differentially impact longitudinal changes in multiple cognitive domains among African-American adults.

1. Introduction

As the world’s populations ages, dementia from all causes is estimated to have a prevalence of 4.7% among older adults ≥60y(Sosa-Ortiz, et al., 2012) with 4.6–7.7 million cases added annually worldwide and region-specific incidence rates ranging between 3.5 and 10.5 per 1,000. (Ferri, et al., 2005,Prince, et al., 2013,Sosa-Ortiz, et al., 2012) Alzheimer’s Disease (AD) accounts for 60–80% of all dementias (Sosa-Ortiz, et al., 2012). Known to be a neurodegenerative disorder with multi-factorial etiology, AD manifests itself with progressive episodic memory deterioration followed by impairment in several domains of cognition. (Lindeboom and Weinstein, 2004) Two key hallmarks of AD are progressive Aβ-amyloid brain deposition – “the amyloid cascade hypothesis”.(Hardy and Selkoe, 2002) and neurofibrillary tangles (NFT) arising from hyper-phosphorylated tau.(Turner, 2003) AD is also the leading cause of disability in old age (Helmer, et al., 2006) and the sixth leading cause of death in the US.(Alzheimer’s, 2016) Around 5.4 million Americans currently live with AD, a continuously rising number expected to reach 13.8 million by 2050. (Alzheimer’s, 2016) Though an effective treatment is yet to be discovered, important genetic risk factors were identified for late-onset AD, such as the APOE4 genotype (Bertram, et al., 2007) as well as mid-life modifiable risk factors including education, smoking, physical inactivity, depression, mid-life obesity, hypertension, type 2 diabetes, antioxidants (e.g. vitamin E), n-3 fatty acids, and homocysteine. (Barnes and Yaffe, 2011,Beydoun, et al., 2014)

In fact, elevated total homocysteine (tHcy) plasma concentration, a sulfur amino acid, is an established risk factor for cardiovascular and cerebrovascular disease.(Refsum, et al., 1998) The latter could mediate tHcy-dementia relationship in old age.(Ford, et al., 2012a,Haan, et al., 2007,Kim, et al., 2008,Ravaglia, et al., 2005,Seshadri, et al., 2002,Zylberstein, et al., 2011) Specifically, tHcy was directly linked to greater white matter hyperintensities and faster brain atrophy.(Bleich and Kornhuber, 2003,den Heijer, et al., 2003,Dufouil, et al., 2003,Sachdev, et al., 2002,Scott, et al., 2004)

Although the risk of homocysteinemia increases with age and diminished renal function, B-vitamin dietary intakes can influence that risk by converting tHcy it into methionine and cysteine through one-carbon metabolism (OCM) cycles.(Bottiglieri, 2005,Troesch, et al., 2016) Thus, B-vitamin interventions can potentially reduce plasma tHcy, with evidence of an inverse relationship in plasma between tHcy and folate/B-12. (Selhub, et al., 1993) Importantly, folate, vitamins B-6 and B-12 may impede cognitive decline and delay dementia onset, (Duthie, et al., 2002,Feng, et al., 2006,Kado, et al., 2005,Mooijaart, et al., 2005,Ramos, et al., 2005,Ravaglia, et al., 2005,Tucker, et al., 2005) with studies suggesting an antagonistic interaction between vitamin B-12 and tHcy in plasma with respect to age-related cognitive decline.(Haan, et al., 2007,Li, et al., 2008,Vidal, et al., 2008) Furthermore, in vitro studies show that tHcy has neurotoxic and excitotoxic properties, (Kruman, et al., 2000,Parsons, et al., 1998) suggesting direct and adverse influences on brain function.

OCM enzymatic genetic polymorphisms play an equally important role as B-vitamins in determining tHcy plasma concentrations. Mutations to methylenetetrahydrofolate reductase (MTHFR), the most widely studied OCM enzyme, (e.g. MTHFR C677T, A1793G and A1298C) are risk factors for hyperhomocysteinemia, due to encoding of an enzyme with reduced activity.(Rozen, 1997) Thus, MTHFR and its pathological sequelae, are also risks for AD, with some studies indicating synergistic interaction between hyperhomocysteinemia or hypertension and “T” allele of C677T mutation in adverse cognitive outcomes. (Bottiglieri, et al., 2001,Cai, et al., 2016,Deshmukh, et al., 2009,Elkins, et al., 2007,Ford, et al., 2012b,Guenther, et al., 1999,Kageyama, et al., 2008,Polito, et al., 2016,Rajagopalan, et al., 2012,Religa, et al., 2003,Roussotte, et al., 2017,Schiepers, et al., 2011,Schwahn and Rozen, 2001,Troesch, et al., 2016,Tsai, et al., 2011,Wakutani, et al., 2004,Weisberg, et al., 1998,Yamada, et al., 2001). Other relevant OCM mutations include Methionine Synthase, MTR (e.g. A2756G) (Chen, et al., 1997,Leclerc, et al., 1996,Troesch, et al., 2016), Methionine Synthase Reductase, MTRR (e.g. A66G, C574T) (Olteanu, et al., 2002,Troesch, et al., 2016), Cystathionine β-Synthase, CBS (e.g. 68 bp insertion at exon 8, G9276A and 31 bp variable number of tandem repeats) (Barbaux, et al., 2000,Nienaber-Rousseau, et al., 2013,Olteanu, et al., 2002,Sebastio, et al., 1995,Troesch, et al., 2016), and Serine Hydroxymethyltransferase (SHMT) (e.g. C1420T). (Lievers, et al., 2001,Troesch, et al., 2016) OCM-related genes are under-studied in relation to cognitive outcomes, particularly among middle-aged African-American adults. To date, only one cross-sectional study tested these associations among African-American adults, advocating the need to have longitudinal studies that target this socio-demographic group.(Moorthy, et al., 2012) In addition, gender differences in One-carbon metabolism resulting in higher levels of tHcy among men suggests that genetic factors linked to this metabolic pathway can alter cognitive trajectories differently between men and women.(Sadre-Marandi, et al., 2018)

This study tested associations of selected OCM enzymatic gene SNPs, SNP latent classes (SNPLC) and SNP haplotypes (SNPHAP) with longitudinal changes in cognitive performance in a large sample of African-American urban adults. Individual SNPs such as MTHFR (C677T, rs1801133), MTR(A2756G, rs1805087) and MTRR(A66G, rs1801394; C574T, rs1532268), known to be associated with increased serum tHcy (Chen, et al., 1997,Leclerc, et al., 1996,Olteanu, et al., 2002,Rozen, 1997,Troesch, et al., 2016) are expected to be directly linked with age-related cognitive decline. In an attempt to replicate findings from a previous study, (Wakutani, et al., 2004) we specifically hypothesize that the “GG” MTHFR SNPHAP (rs1801133(C677T, A>G)/rs1801131(A1298C, G>T) is associated with slower age-related cognitive decline.

2. Methods

2.1. Database

Initiated in 2004, Healthy Aging in Neighborhoods of Diversity across the Life Span (HANDLS) is an on-going population-based cohort study. Household screenings from an area probability sample of 13 neighborhoods in Baltimore City, Maryland were conducted. Neighborhoods were contiguous groups of census tracts with high participant yield filling a 4-way factorial design: race by sex by age by socioeconomic status [SES: below or above 125% of the federal poverty level]. After residential dwellings were identified within selected neighborhoods, doorstep interviews were conducted to select one or two eligible persons in each household to become HANDLS participants. Eligible persons were 30–64y old, self-identified as White or African-American, capable of informed consent, performed ≥5 tests, and present valid photo identification. Pregnancy, being within 6 months of active cancer treatment, or self-identifying as multi-ethnic were key reasons for exclusion. (Evans, et al., 2010) Written informed consent was obtained after reviewing a laymen term protocol booklet and a video detailing procedures and future re-contacts. HANDLS was approved by the National Institute on Environmental Health Sciences, National Institutes of Health Institutional Review Board.

Our present study analyzed longitudinal data from two HANDLS visits, selecting the sample of African-American participants with repeated cognitive data, and complete baseline covariate and genetic data. Time between examination visits 1 (Wave 1:2004–2009) and 2 (also known as Wave 3:2009–2013,(National Institute on Aging, 2004)) ranged from <1y to ~8y, with a mean of 4.64±0.93y.

2.2. Study subjects

HANDLS recruited 3,720 [mean±SD age(y) of 48.3±9.4, 45.3% men, and 59.1% African-Americans and 40.9% White]. Of 2,198 African-Americans baseline HANDLS participants, 1,024 had complete genetic data. Nine cognitive test scores were used in this study as was done previously (Beydoun, et al., 2012). Cognitive data completeness at either visit among African-Americans ranged between n=1,392 for CVLT-DFR and n=1,472 for BVRT, with an average of 1.7 visits/participant, excluding participants with missing baseline smoking status. Mixed-effects regression models for predicting the annualized rates of cognitive change assumed missingness at random (Ibrahim and Molenberghs, 2009) and used full information on repeated cognitive tests. The final analytic sample size ranged between n=660 participants for CVLT-DFR and n=797 participants for the verbal fluency test-categorical (VFT-C), (Figure 1).

Figure 1. Participant flowchart.

Abbreviations: AF=Animal Fluency; BVRT=Benton Visual Retention Test; CVLT-List A=California Verbal Learning Test, List A; CVLT-DFR=California Verbal Learning Test, Delayed Free Recall; DS-B=Digits Span Backwards; DS-F=Digits Span Forward; MMSE=Mini-Mental State Examination; Trails A and B= Trailmaking Test, parts A and B.

2.3. Cognitive assessment

Cognitive assessment included 6 tests with 9 test scores covering 7 domains (mental status, attention, learning/memory, executive function, visuo-spatial/visuo-construction ability, psychomotor speed, language/verbal): Mini-Mental State Examination (MMSE) [mental status], California Verbal Learning Test (CVLT) immediate (List A) and Delayed Free Recall (DFR) [learning/memory, language/verbal], Digit Span Forward and Backwards tests (DS-F and DS-B) [attention and working memory], Benton Visual Retention Test (BVRT) [figural memory and visuo-constructional abilities], Animal Fluency(AF) [semantic verbal fluency], Trailmaking Test, Parts A and B (Trails A and B) [attention and executive functioning]. BVRT, Trails A and B were coded as higher scores reflecting poorer performance (Supplemental method 1). Multiple linear mixed-effects regression models with quadratic age added among fixed effects, were conducted for cognitive score estimation at specific ages (Beydoun, et al., 2012) and for prediction of annualized rate of cognitive change at mean follow-up age, which was termed longitudinal annual rate of cognitive change (LARCC). LARCCs of 8 out of 9 cognitive test scores (excluding MMSE) were entered into a factor analytic model (Sharma, 1996):

$${\text{LARCC}}_{\text{i}}=\sum _{j=1}^k{\lambda }_{ij}\times Domai{n}_j+{\phi }_i$$

Where $\text{LARCCi}$ is standardized z-score for each cognitive test LARCC, ${\text{λ}}_{\text{ij}}$ is factor loading for each LARCC and each factor, ${\text{Domain}}_{\text{j}}$ is standardized z-score for each factor $\text{j}$, and ${\text{φ}}_{\text{i}}$ is residual error. Two factors were extracted based on eigenvalue>1 rule, and their factor loadings were rotated using varimax orthogonal rotation, creating domain-specific LARCCs that were uncorrelated. Those factors were interpreted and labelled based on loadings $\left({\text{λ}}_{\text{ij}}\ge 0.40\right)$. Domains were labeled as: “Domain 1(“Verbal memory (+) and fluency(+)”), Domain 2 (“Visual(−)/working memory(+) and executive function(−)”), whereby “+”indicates slower decline with higher score and “-“ faster decline with higher score. Most ${\mathrm{LARCC}}_{\text{i}}$ factor loadings were elevated for a single domain of two, creating an easily interpretable simple structure. Pearson’s correlation between extracted domains 1 and 2 was weak (r=0.12). In addition, the MMSE LARCCs, which was not included in the factor analysis, was inversely correlated with Domain 2 (r=−0.66), while being weakly correlated with Domain 1 (r=−0.03) (See Supplemental method 2).

All HANDLS participants completed informed consent after probing for protocol understanding. Participants were administered mental status tests, which they completed successfully, with a low scores resulting from poor literacy skills rather than signs of dementia.

2.4. OCM enzyme SNP, SNPLC and SNPHAP

HANDLS participants were genotyped using Illumina 1M genotyping arrays. A total of 1,024 individuals were successfully genotyped and passed genotype quality control criteria. Details are provided in supplemental method 3.

SNP selection was based on previously published genome-wide association studies relating cognitive function, decline or dementia to gene polymorphisms involved in the one-carbon metabolism(Coppede, 2010,Porter, et al., 2016,Rai, 2016,Rai, 2017,Sun, et al., 2015,Troesch, et al., 2016) and as an attempt to replicate a previous case-control study that examined associations between MTHFR haplotypes and AD.(Wakutani, et al., 2004) All available MTHFR SNPs were screened for imputation quality and minor allele frequency (MAF). SNPs with MAF<0.05 were excluded. Our database included most selected SNPs from each target genes. For MTHFR, 10 SNPs were selected with MAF>0.05: rs4846049 (T>G, MAF=0.472), rs1476413 (T>C, MAF=0.170), rs4846051 (A1317G, G>A, MAF=0.323), rs1801131 (A1298C, G>T, MAF=0.175), rs2066462 (C1056T, A>G, MAF=0.095), rs1801133 (C677T, A>G, MAF=0.088), rs1703796 (T>C, MAF=0.092), rs9651118 (C>T, MAF=0.058), rs17367504 (G>A, MAF=0.108), rs2066470 (A>G, MAF=0.071). Two MTRR SNPs were selected: rs1801394 (G>A, MAF=0.284) and rs1532268 (T>C, MAF=0.259). Finally, MTR and SHMT included only 1 SNP each, rs1805087 (G>A, MAF=0.261) and rs1979277 (A>G, MAF=0.362), respectively. Excluded SNPs were included in MTHFR rs2274976 and rs17375901 (MAF<0.05) and in SHMT rs5742905 (MAF<0.05 and R-square=0.007).

MTHFR SNPLCs were extracted using latent class analysis (PROC LCA in SAS version 9.3)(Iivonen, et al., 2004,Lanza, et al., 2007), using an additive mode of inheritance (i.e. SNPs: 0/1/2). Model fit was determined based on reduced BIC, allowing for 10% difference between more vs. less parsimonious model (range of latent class numbers: 4–8). (Beydoun, et al., 2012,Beydoun, et al., 2017,Beydoun, et al., 2013)

Similarly, MTHFR SNP haplotypes (SNPHAP) were considered among key predictors. Using Haploview version 4.2 (Barrett, et al., 2005,Wigginton, et al., 2005), we extracted three distinctive and common SNPHAP with frequency>5% based on two commonly studied MTHFR SNPs namely rs1801133(C677T, A>G)/ rs1801131(A1298C,G>T): MTHFR₁: GT, MTHFR₂:GG, MTHFR₃:AT, coded as: 0=having no MTHFR_x haplotype; 1=having one allele carrying the MTHFR_x haplotype; 2=having two alleles carrying the MTHFR_x haplotype.

2.5. Covariates

Three sets of covariates were assessed as potential confounders: (1) socio-demographic factors – baseline age, sex, educational attainment (years of schooling), and smoking status as a lifestyle-related factor (never, former or current smoker); (2) self-reported history of type 2 diabetes, hypertension, cardiovascular disease (stroke, congestive heart failure, non-fatal myocardial infarction or atrial fibrillation) and dyslipidemia at first-visit; and (3) Measured first-visit body mass index (BMI in kg/m²). Right and left sitting systolic and diastolic blood pressure levels (SBP and DBP) were averaged. Blood pressure was measured non-invasively using brachial artery auscultation with an aneroid manometer, a stethoscope, and an inflatable cuff. Following an overnight fast (8–12 hours) and consenting, blood was drawn and collected from an antecubital vein. Serum total cholesterol, high density lipoprotein-cholesterol (HDL-C), and glucose were assessed using a spectrophotometer (Olympus 5400). First-visit blood pressure (systolic and diastolic in mm Hg), plasma total and HDL-cholesterol, and fasting blood glucose (in mg/dL) were only analyzed for descriptive purposes.(Beydoun, et al., 2012,Beydoun, et al., 2017) Serum folate and B-12 were measured at baseline using immunoassays. [https://www.questdiagnostics.com/testcenter/BUOrderInfo.action?tc=7065&labCode=AMD&searchString=Folate,%20Serum’]. While those measures were not included in our key models given their potential mediating role in the association between One-Carbon genetic risk markers and cognitive decline, they are presented for descriptive purposes by sex and MTHFR SNP (C667T).

2.6. Statistical analysis

For each selected SNP, Hardy-Weinberg equilibrium was assessed with an exact test, while pair-wise linkage disequilibrium (LD) was computed and visualized with Haploview version 4.2. (Barrett, et al., 2005,Wigginton, et al., 2005) Weighted participant study characteristics were estimated and compared by sex and MTHFR C667T (0=Low risk, 1=Medium risk, 2=High risk) SNP, using design-based F-test. Moreover, unweighted analyses were also conducted for LARCCs to obtain their respective standard deviations and aid in interpreting effect sizes of genetic markers on cognitive change. Study characteristics were also compared according to completeness in genetic data, though not fully presented. The association between MTHFR C667T genotype and LARCCs was additionally age and sex-adjusted.

Associations of MTHFR, MTR, MTRR and SHMT SNPs and of MTHFR SNPLC and SNPHAP with LARCCs were examined by multivariable-adjusted ordinary least squares (OLS) that did not take into account sampling design complexity. SNPs (wild type w/ variant v) were operationalized as genotypes, comparing two variant genotypes (wv, vv) with wild type genotype (ww) and as dosage of variant allele (v), using an additive mode of inheritance model. P-trend was estimated when testing the linear relationship between haplotype dosage (0, 1, 2 copies) and cognitive outcomes. Effect sizes were interpreted as the effect of one additional allele copy on the LARCC and was compared to 1 SD of LARCC in the descriptive part of the analysis. An effect >10% of 1 SD was considered appreciable.

Non-random selection of participants with genetic data was accounted for using a 2-stage Heckman selection model. (Heckman, 1979) At stage 1, probit model estimated using inverse mills ratio was derived from predicted probability of being selected, conditional on covariates in that model. (Beydoun, et al., 2010) At stage 2, inverse mills ratio was included in OLS regression models, thus adjusting for related selection bias. Sex-stratified analyses were conducted, with effect modification by sex formally tested by including interaction terms.

Type I error of 0.05 was considered for all analyses, and p-values between 0.05 and 0.10 were considered as borderline significant for main effects, whereas a p-value below 0.10 was considered significant for interaction terms (Selvin, 2004), prior to multiple testing adjustments. Correction for multiple testing was done using familywise Bonferroni procedure, taking only the cognitive outcome into account. Specifically, a family was defined as a cognitive test, assuming their content-wise independence though not necessarily in their degree of correlation (Hochberg, 1987). Within each cognitive test, there were generally two test scores for which correction for multiple testing was required. This was the case for CVLT-DFR and CVLT-List A; Trails A and Trails B; DS-F and DS-B. For these cognitive tests, statistical significance criterion for p-values and p-values for trend was reduced to p=0.05/2=0.025 (marginal significance: p=0.10/2=0.05). For MMSE (a measure of global mental status), BVRT and AF, no correction was needed, an approach taken in our previous study. (Beydoun, et al., 2012) Selected findings were illustrated using plotted predictive margins with 95% CI from the associated OLS models. All analyses [except for LCA (SAS ver. 9.3) and haplotype extraction (Haploview version 4.2)] were performed using Stata version 15.0.(STATA, 2015)

3. Results

3.1. Study sample characteristics

Baseline study sample characteristics are presented in Table 1, comparing the sample by sex and MTHFR C667T genotype (0=Low risk to 2=Highest risk). Sample characteristics by completeness in genetic data are also analyzed but not fully presented. Except for the LARCCs for which samples were specific to the cognitive tests, comparisons were made for the sample that was complete on MMSE LARCC as the main criterion for selection [N=648–797 for both sexes]. More than half of the sample consisted of women. Comparable socio-demographic, lifestyle and health-related characteristic distributions were observed by genetic data availability, except for SBP being higher and cognitive decline on BVRT, CVLT-List A and DS-B being faster among those with “complete genetic data” (P<0.05). Within the latter group, while smoking was more prevalent among men, women had higher cardiovascular disease prevalence, higher mean BMI and HDL-C level. Nevertheless, cognitive decline was consistently faster among men for most cognitive tests. Most notably, the “Verbal memory and fluency” domain 1, indicated an overall faster decline among men. When comparing low risk to highest risk group for the MTHFR C667T genotype, the highest risk group had a lower mean BMI, total cholesterol, HDL-C, fasting glucose and most importantly a lower level of serum folate (P<0.001). In addition, cognitive decline was found to be faster in this group for both domains of cognition. However, after adjustment for age and sex, only domain2 remained significantly associated with the high risk group (vs. lowest risk), mainly driven by a faster decline on Trailmaking test, parts A and B and BVRT (data not shown).

Table 1.

Baseline study sample characteristics, genetic SNPs and predicted LARCC by availability by sex and MTHFR C667T SNP among eligible African-American participants, n=788; HANDLS study⁵

	By sex				By MTHFR: rs1801133(C677T,G>A)						Total with genetic data
	Men		Women				0: Low risk	1: Medium risk	2: High Risk	P	P
							%, Mean±SE			(1vs. 0) 2	(2 vs. 0) 2
	n	%, Mean±SE	n	%, Mean±SE	P 1	83.3	15.3	1.3			n	%, Mean±SE
Baseline study sample characteristics
Female, %		__		__		52.2	55.0	10.7		0.15	788	52.1
Baseline age (years)	349	48.6±0.8	439	47.0±0.8	0.18	47.8±0.6	47.7±1.5	43.3±2.9	0.94	0.091	788	47.8±0.57
Education, years	349	12.6±0.2	439	12.6±0.2	0.85	12.6±0.2	12.5±0.3	12.4±0.4	0.79	0.56	788	12.60±0.16
Smoking status, %	349		439		0.025					0.039	788
Never/former		42.2		56.4		51.5	43.4	3.9				49.6
Current		57.8		43.5		48.5	56.6	96.1				50.4
Type 2 diabetes, %	349	12.8	439	15.8	0.43	12.5	25.4	0.0		0.17	788	14.3
Hypertension, %	349	36.3	439	47.9	0.06	43.2	40.4	5.8		0.16	788	42.3
Cardiovascular disease, %3	349	10.3	439	18.1	0.022	14.6	17.1	3.9		0.60	788	14.8
Dyslipidemia, %	349	23.2	439	24.5	0.80	22.3	33.5	8.4		0.17	788	23.9
Body mass index, kg.m−2	349	27.4±0.5	439	31.6±0.8	<0.001	29.8±0.6	29.3±1.0	22.3±0.8	0.68	<0.001	788	29.6±0.5
Systolic blood pressure (mm Hg)	341	121.4±1.4	427	122.7±1.6	0.55	121.7±1.2	124.5±2.9	120.6±2.6	0.37	0.68	768	122.1±1.1
Diastolic blood pressure (mm Hg)	337	79.1±0.8	418	76.4±1.3	0.07	77.7±0.9	77.7±1.4	77.2±1.6	0.97	0.75	755	77.7±0.8
Serum total cholesterol level (mg/dL)	336	185.8±5.0	424	187.2±3.5	0.83	185.4±2.9	198.7±9.5	116.9±7.0	0.18	<0.001	760	186.5±3.0
Serum HDL-C (mg/dL)	336	49.9±1.9	424	57.2±1.7	0.004	54.1±1.5	52.7±1.9	40.9±2.0	0.55	<0.001	760	53.7±1.3
Fasting plasma glucose (mg/dL)	336	104.5±3.4	424	103.9±3.1	0.88	103.0±2.3	112.0±7.7	88.7±1.9	0.26	<0.001	760	104.2±2.3
Serum folate (ng/mL)	338	15.9±1.9	424	13.5±0.5	0.23	14.8±1.1	14.2±0.9	9.5±0.5	0.66	<0.001	762	14.7±1.0
Serum vitamin B-12 (pg/mL)	338	527±19	424	594±25	0.032	570±19	531±30	442±107	0.27	0.19	762	562±16
Genetic SNPs
MTHFR: rs4846049(T>G)	349	0.89±0.07	439	0.83±0.06	0.48	0.73±0.05	1.48±0.07	2.00±0.00	<0.001	<0.001	788	0.86±0.05
MTHFR: rs1476413(T>C)	349	1.56±0.06	439	1.59±0.05	0.77	1.52±0.04	1.85±0.04	2.00±0.00	<0.001	<0.001	788	1.58±0.04
MTHFR: rs4846051(A1317G, G>A)	349	1.36±0.05	439	1.38±0.05	0.78	1.31±0.04	1.65±0.06	2.00±0.00	<0.001	<0.001	788	1.37±0.04
MTHFR: rs1801131(A1298C,G>T)	349	1.56±0.05	439	1.57±0.05	0.87	1.51±0.04	1.86±0.04	2.00±0.00	<0.001	<0.001	788	1.57±0.04
MTHFR: rs2066462(C1056T,A>G)	349	1.79±0.04	439	1.74±0.05	0.44	1.73±0.04	1.92±0.03	2.00±0.00	<0.001	<0.001	788	1.76±0.03
MTHFR: rs1801133(C677T,A>G)	349	1.81±0.05	439	1.83±0.03	0.67	2.00±0.00	1.00±0.00	0.00±0.00	<0.001	<0.001	788	1.82±0.03
MTHFR: rs1703796(T>C)	349	1.83±0.03	439	1.76±0.04	0.22	1.77±0.03	1.93±0.03	2.00±0.00	<0.001	<0.001	788	1.80±0.03
MTHFR: rs9651118(C>T)	349	1.90±0.02	439	1.88±0.02	0.46	1.89±0.02	1.88±0.04	2.00±0.00	0.77	<0.001	788	1.89±0.01
MTHFR: rs17367504 (G>A)	349	1.74±0.04	439	1.72±0.05	0.81	1.70±0.04	1.92±0.03	2.00±0.00	<0.001	<0.001	788	1.73±0.03
MTHFR: rs2066470(A>G)	349	1.87±0.03	439	1.81±0.04	0.20	1.82±0.03	1.93±0.03	2.00±0.00	0.005	<0.001	788	1.84±0.03
MTRR: rs1801394 (G>A)	349	1.32±0.06	439	1.37±0.05	0.47	1.35±0.04	1.39±0.08	0.96±0.05	0.63	<0.001	788	1.35±0.04
MTRR: rs1532268 (T>C)	349	1.39±0.06	439	1.57±0.05	0.041	1.47±0.05	1.61±0.07	1.10±0.13	0.097	0.003	788	1.48±0.04
MTR:rs1805087(G>A)	349	1.46±0.05	439	1.47±0.06	0.86	1.48±0.04	1.35±0.10	1.90±0.13	0.22	<0.001	788	1.47±0.04
SHMT:rs1979277(A>G)	349	1.18±0.07	439	1.27±0.06	0.35	1.24±0.05	1.11±0.12	1.92±0.09	0.34	<0.001	788	1.23±0.05
Predicted LARCC 4
MMSE	349	−0.046±0.002	439	−0.034±0.002	<0.001	−0.040±0.001	−0.040±0.003	−0.034±0.001	0.87	0.005	788	−0.040±0.001
		(SD: 0.022)		(SD: 0.018)		(SD:0.020)	(SD:0.024)	(SD:0.010)				(SD:0.021)
BVRT	350	+0.218±0.003	432	+0.173±0.004	<0.001	+0.195±0.003	+0.193±0.007	+0.222±0.005	0.22	0.77	782	+0.195±0.003
		(SD:0.045)		(SD: 0.046)		(SD:0.050)	(SD:0.053)	(SD:0.037)				(SD:0.050)
CVLT-List A	295	−0.292±0.002	385	−0.270±0.002	<0.001	−0.281±0.002	−0.278±0.003	−0.283±0.003	0.27	0.57	680	−0.281±0.001
		(SD:0.018)		(SD:0.019)		(SD:0.021)	(SD:0.019)	(SD:0.014)				(SD:0.021)
CVLT-DFR	284	−0.138±0.001	376	−0.119±0.001	<0.001	−0.128±0.001	−0.127±0.001	−0.134±0.002	0.60	0.002	660	−0.128±0.001
		(SD:0.008)		(SD:0.009)		(SD:0.012)	(SD:0.012)	(SD:0.009)				(SD:0.012)
AF	356	−0.067±0.002	441	−0.045±0.002	<0.001	−0.056±0.002	−0.057±0.003	−0.049±0.002	0.67	0.016	797	−0.056±0.002
		(SD:0.022)		(SD:0.020)		(SD:0.023)	(SD:0.020)	(SD:0.010)				(SD:0.023)
Trails A	326	+0.745±0.072	419	+0.855±0.117	0.43	+0.774±0.078	+0.945±0.182	+0.993±0.192	0.39	0.24	745	+0.803±0.071
		(SD:1.023)		(SD: 1.224)		(SD:1.080)	(SD:1.230)	(SD:3.720)				(SD:1.140)
Trails B	326	+4.865±0.261	419	+4.137±0.194	0.026	+4.309±0.157	+4.856±0.310	+10.515±1.587	0.115	<0.001	745	+4.480±0.163
		(SD:2.607)		(SD:2.600)		(SD:2.549)	(SD:2.884)	(SD:4.037)				(SD:2.625)
DS-F	351	−0.030±0.001	431	−0.015±0.001	<0.001	−0.023±0.001	−0.020±0.001	−0.032±0.003	0.10	0.002	782	−0.022±0.001
		(SD:0.011)		(SD:0.010)		(SD:0.013)	(SD:0.0124)	(SD:0.010)				(SD:0.013)
DS-B	351	−0.027±0.002	424	−0.017±0.001	<0.001	−0.023±0.001	−0.018±0.002	−0.030±0.002	0.048	0.001	775	−0.022±0.001
		(SD:0.016)		(SD: 0.017)		(SD:0.018)	(SD:0.018)	(SD:0.013)				(SD:0.018)
Cognitive domain 1	277	−0.848±0.062	371	+0.674±0.048	<0.001	−0.034±0.08	+0.021±0.108	−0.514±0.046	0.69	<0.001	648	−0.033±0.070
		(SD:0.55)		(SD:0.55)		(SD: 0.91)	(SD:0.79)	(SD:0.31)				(SD:0.89)
Cognitive domain 2	277	−0.002±0.081	371	−0.128±0.087	0.29	−0.088±0.07	−0.037±0.130	+0.462±0.112	0.73	<0.001	648	−0.069±0.061
		(SD:0.86)		(SD:0.87)		(SD: 0.86)	(SD: 0.93)	(SD: 0.64)				(SD:0.87)

Abbreviations: AF=Animal Fluency; BMI=body mass index (calculated as weight in kg/square of height in meters); BVRT=Benton Visual Retention Test; CVLT-List A=California Verbal Learning Test, List A; CVLT-DFR=California Verbal Learning Test, Delayed Free Recall; DS-B=Digits Span Backwards; DS-F=Digits Span Forward; HDL-C= high density lipoprotein cholesterol; MMSE=Mini-Mental State Examination; Trails A and B= Trailmaking Test, parts A and B.

¹P-value for null hypothesis of no difference between sexes among those with complete genetic data. Note that this analysis was done on African-American participants with complete baseline covariates, including baseline MMSE scores.

²P-value for null hypothesis of no difference between MTHFR C667T SNP (0: Low risk, 1: Medium risk, 2: High risk) using design-based F-test.

³Reported any of the following conditions at first visit: stroke, congestive heart failure, nonfatal myocardial infarction, or atrial fibrillation.

⁴Cognitive scores were predicted at mean age at follow-up prior to onset of dementia or for all time points using a linear mixed model controlling for sex, race/ethnicity, education (years), and smoking status, with age (centered at 50y) added among the fixed effect variables to allow for quadratic non-linear change, while age (centered at 50) was added to the random effects to allow for individual-level variation in slopes. The slope or annual rate of change was predicted from these models at the mean age at follow-up (i.e. between age 50 and individual mean age of follow-up for each cognitive test). Using factor analysis, two factor scores were estimated and were labeled as LARCC in the following domains: Domain 1: “Verbal memory and fluency”, Domain 2: “Visual/working memory and executive function” (See Supplemental method 1).

⁵Socio-demographic, lifestyle, health-related factors are presented among participants with complete data on those variables, as well as complete data on MMSE LARCC. LARCC measures are presented for eligible subjects with complete data on covariates entered into subsequent models as well as complete data on each of the cognitive test scores at either baseline or follow-up wave. Unreliable data from each cognitive test score was excluded. Serum folate and vitamin B-12 were not included into main models as they could be mediators.

All examined SNPs were in Hardy-Weinberg equilibrium (P > 0.002). Variants within each MTHFR and MTRR gene were deemed in low linkage equilibrium (r²<0.30). Genotypic frequencies suggested that one genotype in each SNP had prevalence of >45% and thus was dominant compared to other genotypes (Figures 2, S1–S3). Table 2 presents MTHFR SNPLC (determined by LCA) and SNPHAP [overall frequency and for 0, 1 or 2 copies] distributions. While SNPLCs yield mutually exclusive categories, SNHAPs are independent and non-mutually exclusive, reflecting allelic combinations for the selected individuals.

Figure 2.

(A) Schematic representation of the MTHFR gene. The SNP and gene coordinates are based on NCBI build 37 (hg19, May 2013, Phase 3 of the 1000 Genomes Project). The MTHFR gene on chromsome 1 composed of 12 exons and 20.4 kilobase pairs in size; (B) Genotype frequencies (%) of selected MTHFR SNPs of original sample with complete genetic data (n=1,024)

Abbreviations: hg=human genome; MTHFR=Methylenetetrahydrofolate reductase; RefSeq=Reference Sequence; SNP=Single Nucleotide Polymorphism; vv=variant-variant; wv=wild type-variant; ww=wild type-wild type.

Note: MTHFR SNPs are in the direction v → w: 0=vv, 1=wv, 2=ww.

Table 2.

Findings from latent class analysis and haplotype analysis: definitions and distributions of SNPLC and SNPHAP for the selected MTHFR SNPs, n=1,024¹

	SNP Haplotypes (SNPHAP)		SNP Latent Classes (SNPLC)
	Definitions	%	Definitions	%
MTHFR	rs1801133(C677T, A>G)/ rs1801131(A1298C,G>T)		rs4846049(T>G)/rs1476413(T>C)/rs4846051(A1317G, G>A)/rs1801131(A1298C,G>T)/rs2066462(C1056T,A>G) /rs1801133(C677T,A>G)/rs1703796(T>C)/rs9651118(C>T)/rs17367504 (A>G)/rs2066470(A>G)]
Overall	MTHFR1: GT	73.7	MTHFR1: TT/--/-A/--/--/GG/-C/TT/--/GG	10.7
	MTHFR2: GG	17.3	MTHFR2: T-/CC/--/TT/GG/GG/CC/TT/-A/GG	11.4
	MTHFR3: AT	8.8	MTHFR3: -G/-C/AA/-T/GG/--/CC/-−/−A/GG	19.8
			MTHFR4: TG/-C/-A/TT/GG/-G/CC/-T/-A/GG	34.4
			MTHFR5: TG/-C/-A/GT/-G/-G/-C/-T/-A/GG	9.6
			MTHFR6: --/--/-A/--/--/-G/T-/-T/--/--	14.1
Allele copies
MTHFR1
0		4.2
1		41.6
2		54.2
MTHFR2
0		71.2
1		25.3
2		3.5
MTHFR2
0		85.6
1		13.7
2		0.7

Abbreviations: MTHFR=Methylenetetrahydrofolate Reductase; SNP=Single Nucleotide Polymorphism; SNPLC= Single Nucleotide Polymorphism Latent Class;

SNPHAP= Single Nucleotide Polymorphism Haplotype.

3.2. MTHFR SNPs and their associations with LARCC

Supplemental Table 1 presents findings from multiple OLS models examining associations between MTHFR SNPs (entered alternatively, models A-J) and LARCC, overall and stratifying by sex. After adjustment for multiple testing, overall, rs4846051(A1317G, G>A) and rs1801131(A1298C,G>T) were associated with slower (β=+0.0023±0.0014, p=0.014) and faster(β=−0.0027±0.0011, p=0.036) decline on AF (SD=0.023 for AF, Table 1), respectively. The latter association was also found among women (β=−0.0042±0.0018, p=0.017). Among men, rs4846051(A1317G, G>A) was associated with a faster decline on the BVRT(β=+0.0049±0.0024, p=0.042), while rs2066462(C1056T,A>G) and rs9651118(C>T) were associated with a slower decline on CVLT-List A (β=+0.0013±0.0005, p=0.005; β=+0.0012±0.0005, p=0.020, respectively). However, those latter SNPs, particularly rs9651118(C>T), were linked to a faster decline on CVLT-DFR, among men (β=−0.0021±0.0009, p=0.023). When examining further those associations among women, rs1801133(C677T,A>G) was linked to a slower decline on cognitive domain 1, (β=+0.0840±0.0364, p=0.022) combining mainly longitudinal changes in tests of verbal memory and verbal fluency. This net effect is the equivalent of an increase by 15% of 1 SD (SD=0.55 among women) in domain 1 and thus is considered to be appreciable.

3.3. MTHFR and MTRR SNPLC and their associations with LARCC: sex-stratified findings

In Table 3, we present findings from OLS regression models whereby SNPLC predicted LARCC among men and women, separately. Following multiple-testing adjustments, among men, MTRR:rs1532268(T>C) was associated with faster BVRT decline (β=+0.0066±0.0027, p=0.013), while MTRR: rs1801394(G>A) was associated with slower DS-B decline (β=+0.0031±0.0012, p=0.014). Compared with MTHFR4 SNPLC, MTHFR5 was associated with faster CVLT-List A decline in men (β=−0.0015±0.0006, p=0.018). Among women, MTRR: rs1532268(T>C) was associated with slower Trails B decline (β=−0.3948±0.1689, p=0.020), as reflected by an inverse relationship with Domain 2 (β=−0.1136±0.00574, p=0.049). Furthermore, among women, compared with MTHFR₄, MTHFR₁ was associated with slower AF decline (β=+0.0075±0.0033, p=0.023). No significant associations were detected between MTR/SHMT SNPs and LARCCs.

Table 3.

MTHFR, MTRR SNP latent classes (SNPLC), MTR and SHMT SNPs’ associations with predicted longitudinal annual rate of cognitive change (LARCC) by sex: Multiple OLS regression analysis (n=648–788); HANDLS study

	Predicted LARCC1
	Men			Women
	N	β±SE2	P	n	β±SE2	P
MMSE
MTHFR1 vs. MTHFR4	349	+0.0022±0.0031	0.46	439	−0.0010±0.0022	0.66
MTHFR2 vs. MTHFR4	349	−0.0003±0.0033	0.92	439	−0.0012±0.0021	0.56
MTHFR3 vs. MTHFR4	349	+0.0014±0.0026	0.61	439	−0.0002±0.002	0.92
MTHFR5 vs. MTHFR4	349	−0.0005±0.0031	0.88	439	+0.0030±0.0025	0.23
MTHFR6 vs. MTHFR4	349	+0.0015±0.0031	0.62	439	−0.0001±0.0020	0.95
MTRR: rs1801394 (G>A)	349	+0.0002±0.0015	0.91	439	+0.0008±0.0010	0.43
MTRR: rs1532268 (T>C)	349	−0.0012±0.0016	0.47	439	−0.0001±0.0010	0.89
MTR:rs1805087(G>A)	349	−0.0006±0.0015	0.67	439	+0.0008±0.0010	0.46
SHMT:rs1979277(A>G)	349	−0.0012±0.0014	0.36	439	−0.0000±0.0009	0.99
BVRT
MTHFR1 vs. MTHFR4	350	+0.0066±0.0050	0.19	432	+0.0017±0.0051	0.74
MTHFR2 vs. MTHFR4	350	+0.0027±0.0056	0.63	432	+0.0054±0.0048	0.26
MTHFR3 vs. MTHFR4	350	+0.0043±0.0045	0.34	432	+0.0065±0.0042	0.12
MTHFR5 vs. MTHFR4	350	+0.0064±0.0052	0.23	432	+0.0088±0.0056	0.12
MTHFR6 vs. MTHFR4	350	+0.0072±0.0053	0.17	432	+0.0027±0.0044	0.55
MTRR: rs1801394 (G>A)	350	−0.0014±0.0026	0.59	432	+0.0025±0.0023	0.29
MTRR: rs1532268 (T>C)	350	+0.0066±0.0027	0.013 3,4	432	−0.0031±0.0024	0.20
MTR:rs1805087(G>A)	350	+0.0000±0.0026	0.97	432	−0.0009±0.0024	0.71
SHMT:rs1979277(A>G)	350	+0.0001±0.0023	0.96	432	+0.0011±0.0021	0.59
CVLT-List A
MTHFR1 vs. MTHFR4	295	−0.0008±0.0006	0.21	385	+0.0003±0.0006	0.59
MTHFR2 vs. MTHFR4	295	−0.0009±0.0007	0.16	385	−0.0010±0.0006	0.080
MTHFR3 vs. MTHFR4	295	−0.0004±0.0005	0.39	385	−0.0002±0.0005	0.66
MTHFR5 vs. MTHFR4	295	−0.0015±0.0006	0.018 3,4	385	+0.0006±0.0007	0.39
MTHFR6 vs. MTHFR4	295	−0.0014±0.0007	0.047	385	−0.0002±0.0005	0.70
MTRR: rs1801394 (G>A)	295	−0.0002±0.0003	0.45	385	+0.0004±0.0003	0.20
MTRR: rs1532268 (T>C)	295	−0.0006±0.0003	0.073 4	385	+0.0002±0.0003	0.51
MTR:rs1805087(G>A)	295	−0.0002±0.0003	0.54	385	−0.0001±0.0003	0.66
SHMT:rs1979277(A>G)	295	−0.0006±0.0003	0.054 4	385	+0.0005±0.0002	0.074
CVLT-DFR
MTHFR1 vs. MTHFR4	284	+0.0011±0.0011	0.30	376	+0.0001±0.0012	0.93
MTHFR2 vs. MTHFR4	284	+0.0019±0.0011	0.10	376	+0.0006±0.0011	0.55
MTHFR3 vs. MTHFR4	284	+0.0008±0.0009	0.41	376	+0.0001±0.0009	0.96
MTHFR5 vs. MTHFR4	284	+0.0013±0.0011	0.24	376	−0.0011±0.0012	0.38
MTHFR6 vs. MTHFR4	284	+0.0023±0.0012	0.057	376	−0.0002±0.0010	0.85
MTRR: rs1801394 (G>A)	284	+0.0001±0.0006	0.88	376	−0.0008±0.0005	0.10
MTRR: rs1532268 (T>C)	284	+0.0005±0.0006	0.38	376	+0.0000±0.0006	0.92
MTR:rs1805087(G>A)	284	+0.0004±0.0006	0.43	376	−0.0004±0.0005	0.49
SHMT:rs1979277(A>G)	284	−0.0003±0.0005	0.57	376	−0.0008±0.0004	0.11
AF
MTHFR1 vs. MTHFR4	356	+0.0012±0.0037	0.74	441	+0.0075±0.0033	0.023 3
MTHFR2 vs. MTHFR4	356	−0.0031±0.0040	0.44	441	−0.0028±0.0030	0.36
MTHFR3 vs. MTHFR4	356	+0.0020±0.003	0.54	441	+0.0005±0.0027	0.85
MTHFR5 vs. MTHFR4	356	+0.0046±0.0039	0.24	441	+0.0042±0.0036	0.24
MTHFR6 vs. MTHFR4	356	+0.0051±0.0038	0.18	441	+0.0022±0.0028	0.43
MTRR: rs1801394 (G>A)	356	−0.0005±0.0018	0.77	441	−0.0016±0.0015	0.28
MTRR: rs1532268 (T>C)	356	+0.0016±0.0019	0.40	441	−0.0002±0.0015	0.90
MTR:rs1805087(G>A)	356	−0.0019±0.0018	0.31	441	−0.0019±0.0015	0.23
SHMT:rs1979277(A>G)	356	−0.0015±0.0017	0.37	441	+0.0006±0.0014	0.65
Trails A
MTHFR1 vs. MTHFR4	326	−0.0453±0.1931	0.82	419	+0.3476±0.2127	0.10
MTHFR2 vs. MTHFR4	326	−0.1714±0.2107	0.42	419	+0.2525±0.1969	0.20
MTHFR3 vs. MTHFR4	326	−0.0440±0.1665	0.79	419	−0.0839±0.1738	0.63
MTHFR5 vs. MTHFR4	326	−0.1429±0.1952	0.46	419	−0.1884±0.2331	0.42
MTHFR6 vs. MTHFR4	326	+0.0323±0.1919	0.87	419	−0.1076±0.1868	0.57
MTRR: rs1801394 (G>A)	326	+0.0724±0.0951	0.45	419	−0.0786±0.0960	0.41
MTRR: rs1532268 (T>C)	326	+0.0387±0.0991	0.70	419	−0.1347±0.1009	0.18
MTR:rs1805087(G>A)	326	+0.0975±0.0950	0.31	419	−0.1186±0.0994	0.23
SHMT:rs1979277(A>G)	326	+0.0308±0.0863	0.72	419	−0.1161±0.0887	0.19
Trails B
MTHFR1 vs. MTHFR4	326	+0.4533±0.3958	0.25	419	+0.5724±0.3586	0.11
MTHFR2 vs. MTHFR4	326	+0.1398±0.4350	0.75	419	−0.0050±0.3291	0.99
MTHFR3 vs. MTHFR4	326	+0.3257±0.3434	0.34	419	−0.0488±0.2937	0.87
MTHFR5 vs. MTHFR4	326	+0.1402±0.4006	0.73	419	−0.1132±0.3931	0.77
MTHFR6 vs. MTHFR4	326	+0.2980±0.3939	0.45	419	+0.5033±0.3145	0.11
MTRR: rs1801394 (G>A)	326	+0.0254±0.1958	0.90	419	+0.1122±0.1633	0.49
MTRR: rs1532268 (T>C)	326	+0.2369±0.2046	0.25	419	−0.3948±0.1689	0.020 3,4
MTR:rs1805087(G>A)	326	+0.0409±0.1951	0.83	419	−0.2059±0.1673	0.22
SHMT:rs1979277(A>G)	326	+0.0967±0.1772	0.58	419	−0.2441±0.1502	0.11
DS-F
MTHFR1 vs. MTHFR4	351	+0.0006±0.0016	0.72	431	+0.0020±0.0015	0.19
MTHFR2 vs. MTHFR4	351	−0.0005±0.0018	0.79	431	−0.0010±0.0014	0.49
MTHFR3 vs. MTHFR4	351	+0.0003±0.0014	0.83	431	−0.0016±0.0012	0.18
MTHFR5 vs. MTHFR4	351	+0.0010±0.0017	0.53	431	−0.0003±0.0016	0.84
MTHFR6 vs. MTHFR4	351	−0.0011±0.0017	0.52	431	+0.0015±0.0013	0.28
MTRR: rs1801394 (G>A)	351	+0.0002±0.0008	0.78	431	+0.0005±0.0007	0.50
MTRR: rs1532268 (T>C)	351	+0.0005±0.0008	0.54	431	−0.0008±0.0007	0.28
MTR:rs1805087(G>A)	351	−0.0016±0.0008	0.048	431	−0.0004±0.0007	0.61
SHMT:rs1979277(A>G)	351	+0.0002±0.0007	0.75	431	−0.0003±0.0006	0.67
DS-B
MTHFR1 vs. MTHFR4	351	+0.0045±0.0024	0.066	424	+0.0007±0.0027	0.80
MTHFR2 vs. MTHFR4	351	+0.0022±0.0027	0.42	424	−0.0006±0.0024	0.80
MTHFR3 vs. MTHFR4	351	+0.0003±0.0021	0.89	424	−0.0020±0.0021	0.35
MTHFR5 vs. MTHFR4	351	+0.0028±0.0025	0.27	424	+0.0015±0.0028	0.60
MTHFR6 vs. MTHFR4	351	+0.0008±0.0025	0.76	424	+0.0013±0.0023	0.57
MTRR: rs1801394 (G>A)	351	+0.0031±0.0012	0.014 3,4	424	−0.0004±0.0012	0.77
MTRR: rs1532268 (T>C)	351	+0.0013±0.0013	0.314	424	−0.0025±0.0012	0.045
MTR:rs1805087(G>A)	351	−0.0017±0.0012	0.16	424	+0.0001±0.0012	0.96
SHMT:rs1979277(A>G)	351	−0.0020±0.0011	0.068	424	−0.0013±0.0011	0.23
Cognitive domain 1
MTHFR1 vs. MTHFR4	277	+0.0462±0.0578	0.43	371	+0.0566±0.0516	0.27
MTHFR2 vs. MTHFR4	277	+0.0249±0.0590	0.67	371	−0.0543±0.0478	0.26
MTHFR3 vs. MTHFR4	277	+0.0343±0.0483	0.48	371	−0.0674±0.0420	0.11
MTHFR5 vs. MTHFR4	277	+0.0530±0.0588	0.37	371	−0.0579±0.0579	0.32
MTHFR6 vs. MTHFR4	277	+0.0813±0.0615	0.19	371	+0.0137±0.0450	0.76
MTRR: rs1801394 (G>A)	277	+0.0018±0.0293	0.95	371	−0.0090±0.0233	0.70
MTRR: rs1532268 (T>C)	277	+0.0058±0.0300	0.85	371	−0.0081±0.0251	0.75
MTR:rs1805087(G>A)	277	−0.0338±0.0284	0.24	371	−0.0212±0.0245	0.39
SHMT:rs1979277(A>G)	277	−0.0230±0.0257	0.37	371	−0.0261±0.0211	0.22
Cognitive domain 2
MTHFR1 vs. MTHFR4	277	+0.1849±0.1203	0.13	371	+0.0765±0.1179	0.52
MTHFR2 vs. MTHFR4	277	+0.0194±0.1226	0.87	371	−0.0285±0.1092	0.79
MTHFR3 vs. MTHFR4	277	+0.1168±0.1005	0.25	371	−0.0385±0.0959	0.69
MTHFR5 vs. MTHFR4	277	+0.0333±0.1224	0.79	371	+0.0138±0.1322	0.92
MTHFR6 vs. MTHFR4	277	+0.1740±0.1278	0.18	371	+0.0912±0.1028	0.38
MTRR: rs1801394 (G>A)	277	+0.0072±0.0609	0.91	371	+0.0253±0.0531	0.64
MTRR: rs1532268 (T>C)	277	+0.0523±0.0623	0.40	371	−0.1136±0.0574	0.049 3
MTR:rs1805087(G>A)	277	−0.0757±0.0591	0.20	371	−0.0742±0.0560	0.19
SHMT:rs1979277(A>G)	277	+0.0128±0.0535	0.81	371	−0.0464±0.0483	0.34

Abbreviations: AF=Animal Fluency; BMI=body mass index (calculated as weight in kg/square of height in meters); BVRT=Benton Visual Retention test; CVLT-List A=California Verbal Learning Test, List A; CVLT-DFR=California Verbal Learning Test, Delayed Free Recall; DS-B=Digits Span Backwards; DS-F=Digits Span Forward; MMSE=Mini-Mental State Examination; MTHFR=Methylenetetrahydrofolate reductase; MTR=Methionine synthase; MTRR=methionine synthase reductase; OLS=Ordinary Least Square; SHMT= Serine Hydroxymethyltransferase; SNP=Single Nucleotide polymorphism; SNPLC=SNP latent class; Trails A and B=Trailmaking test, parts A and B.

¹Cognitive scores were predicted at mean age of follow-up using a linear mixed model controlling for sex, race/ethnicity, education (years), and smoking status, with age (centered at 50y) added among the fixed effect variables to allow for quadratic non-linear change, while age (centered at 50) was added to the random effects to allow for individual-level variation in slopes. The slope or annual rate of change was predicted from these models at the mean age at follow-up. Using factor analysis, two factor scores were estimated and were labeled as LARCC in the following domains: Domain 1: “Verbal memory and fluency”, Domain 2: “Visual/working memory and executive function” (Supplemental method 1). See Table 2 for more details on definition the SNP latent classes.

²Based on multiple OLS regression models with outcome being cognitive annual rate of change. The model controlled for first-visit age, mean age at follow-up, education, first-visit smoking status, first-visit self-reported type 2 diabetes, hypertension, cardiovascular disease, BMI. The 10 principal components obtained with multidimensional scaling (See Supplemental method 3) were also added in a separate sensitivity analysis.

³Significant main effects after familywise Bonferroni correction: p<0.05 for MMSE, BVRT AF and cognitive domains and p<0.025 for other cognitive tests.

⁴P<0.05 for null hypothesis that sex×SNPLC or sex×SNP interaction term=0 in a model where main effect of sex was added.

3.5. MTHFR SNPHAP and their associations with LARCC: sex-stratified findings

Table 4 displays findings when testing net associations between each of the two MTHFR SNPHAP on the LARCC, adjusting for other SNPs and potentially confounding covariates. After multiple-testing adjustments, MTHFR₂ SNPHAP (rs1801133(C677T, A>G)/rs1801131(A1298C,G>T): GG) was associated with slower decline on AF among women, (β=+0.0040±0.0018, p=0.024) while MTHFR₃ SNPHAP(AT) was linked with slower decline on CVLT-List A among men (β=+0.0012±0.0005, p=0.019) but faster decline on “verbal memory/fluency” among women (β=−0.092±0.038, p=0.017). Figure 3 shows the predictive margins from OLS regression model with Animal Fluency LARCC as the outcome and MTHFR₂ SNPHAP (GG). It is clear from this Figure that the potential protective effect of the GG haplotype was only found among women. A sensitivity analysis which included 10 population stratification principal components into the main models did not change the main associations of interest observed in the reduced models.

Table 4.

MTHFR SNP haplotypes (SNPHAP: [rs1801133(C677T, A>G)/ rs1801131(A1298C,G>T)]) and their associations with predicted longitudinal annual rate of cognitive change (LARCC) by sex: Multiple OLS regression analysis (n=648–788); HANDLS study

	Predicted LARCC1
	Men			Women
	N	β±SE2	P	N	β±SE2	P
MMSE: Models A-C
MTHFR1 : GT	349	+0.0008±0.0016	0.61	439	+0.0002±0.0011	0.89
MTHFR2 : GG	349	−0.0014±0.0017	0.41	439	−0.0000±0.0012	0.99
MTHFR3: AT	349	−0.0024±0.0025	0.33	439	−0.0009±0.0016	0.59
BVRT: Models A-C
MTHFR1 : GT	350	−0.0032±0.0027	0.24	432	−0.0008±0.0026	0.75
MTHFR2 : GG	350	+0.0039±0.0028	0.16	432	−0.0005±0.0028	0.86
MTHFR3: AT	350	−0.0040±0.0042	0.35	432	+0.0015±0.0038	0.69
CVLT-List A: Models A-C
MTHFR1 : GT	295	+0.0001±0.0003	0.67	385	−0.0005±0.0003	0.15
MTHFR2 : GG	295	−0.0006±0.0003	0.11	385	+0.0001±0.0003	0.79
MTHFR3: AT	295	+0.0012±0.0005	0.019 3	385	+0.0006±0.0005	0.23
CVLT-DFR: Models A-C
MTHFR1 : GT	284	−0.0001±0.0006	0.88	376	+0.0005±0.0006	0.42
MTHFR2 : GG	284	+0.0006±0.0008	0.36	376	−0.0001±0.0006	0.53
MTHFR3: AT	284	−0.0012±0.0009	0.18	376	−0.0008±0.0009	0.37
AF: Models A-C
MTHFR1 : GT	356	−0.0010±0.0019	0.59	441	−0.0023±0.0016	0.17
MTHFR2 : GG	356	−0.0002±0.0020	0.91	441	+0.0040±0.0018	0.024 3,4
MTHFR3: AT	356	−0.0006±0.0031	0.85	441	−0.0030±0.0024	0.23
Trails A: Models A-C
MTHFR1 : GT	326	−0.178±0.100	0.071	419	−0.0225±0.108	0.84
MTHFR2 : GG	326	+0.0594±0.1034	0.57	419	+0.0429±0.1187	0.72
MTHFR3: AT	326	+0.2762±0.151	0.068	419	+0.0272±0.0157	0.86
Trails B: Models A-C
MTHFR1 : GT	326	−0.3788±0.2025	0.062	419	−0.1663±0.1813	0.36
MTHFR2 : GG	326	+0.1050±0.2135	0.62	419	+0.2170±0.1992	0.28
MTHFR3: AT	326	+0.5935±0.3102	0.057	419	−0.1958±0.2675	0.46
DS-F: Models A-C
MTHFR1 : GT	351	+0.0001±0.0009	0.91	431	−0.0003±0.0008	0.74
MTHFR2 : GG	351	−0.0007±0.0009	0.42	431	+0.0013±0.0008	0.12
MTHFR3: AT	351	+0.0006±0.0013	0.65	431	−0.0022±0.0011	0.052
DS-B: Models A-C
MTHFR1 : GT	351	−0.0012±0.0013	0.36	424	+0.0004±0.0013	0.78
MTHFR2 : GG	351	+0.0012±0.0013	0.35	424	+0.0017±0.0014	0.24
MTHFR3: AT	351	−0.0007±0.0021	0.72	424	−0.0039±0.0019	0.045
Cognitive domain 1: Models A-C
MTHFR1 : GT	277	−0.0032±0.031	0.92	371	−0.0364±0.0263	0.88
MTHFR2 : GG	277	−0.0007±0.0327	0.98	371	+0.0467±0.0284	0.10
MTHFR3: AT	277	−0.0123±0.0461	0.79	371	−0.092±0.0383	0.017 3,4
Cognitive domain 2: Models A-C
MTHFR1 : GT	277	−0.0651±0.0636	0.31	371	−0.0161±0.0596	0.79
MTHFR2 : GG	277	+0.0539±0.0682	0.43	371	+0.0566±0.0647	0.38
MTHFR3: AT	277	+0.0124±0.0962	0.90	371	−0.0928±0.0872	0.29

Abbreviations: AF=Animal Fluency; BMI=body mass index (calculated as weight in kg/square of height in meters); BVRT=Benton Visual Retention test; CVLT-List A=California Verbal Learning Test, List A; CVLT-DFR=California Verbal Learning Test, Delayed Free Recall; DS-B=Digits Span Backwards; DS-F=Digits Span Forward; MMSE=Mini-Mental State Examination; MTHFR=Methylenetetrahydrofolate reductase; MTR=Methionine synthase; MTRR=methionine synthase reductase; OLS=Ordinary Least Square; SHMT= Serine Hydroxymethyltransferase; SNP=Single Nucleotide polymorphism; SNPHAP=SNP haplotypes; Trails A and B=Trailmaking test, parts A and B.

¹Cognitive scores were predicted at mean age of follow-up using a linear mixed model controlling for sex, race/ethnicity, education (years), and smoking status, with age (centered at 50y) added among the fixed effect variables to allow for quadratic non-linear change, while age (centered at 50) was added to the random effects to allow for individual-level variation in slopes. The slope or annual rate of change was predicted from these models at the mean age at follow-up. Using factor analysis, two factor scores were estimated and were labeled as LARCC in the following domains: Domain 1: “Verbal memory and fluency”, Domain 2: “Visual/working memory and executive function” (Supplemental method 1). See Table 2 for more details on definition the SNP haploytpes.

²Based on multiple OLS regression models with outcome being cognitive annual rate of change. Each model included one SNPHAP (Model A: SNPHAP₁, Model B: SNPHAP₂ and Model C: SNHAP₃) and controlled for first-visit age, mean age at follow-up, education, first-visit smoking status, first-visit self-reported type 2 diabetes, hypertension, cardiovascular disease, BMI. The 10 principal components obtained with multidimensional scaling (See Supplemental method 3) were also added in a separate sensitivity analysis.

³Significant main effects after familywise Bonferroni correction: p<0.05 for MMSE, BVRT, AF and cognitive domains and p<0.025 for other cognitive tests.

⁴P<0.05 for null hypothesis that sex×SNPHAP interaction term=0 in a model where main effect of sex was added.

Figure 3.

Predictive margins with 95% CI for Animal Fluency LARCC

by MTHFR₂ SNPHAP (rs1801133(C677T, A>G)/rs1801131(A1298C,G>T): GG), for men and women: Multiple OLS regression models

Abbreviations: LARCC=Longitudinal Annual Rate of Cognitive Change; MTHFR = Methylenetetrahydrofolate reductase; SNPHAP=Single Nucleotide Polymorphism Haplotype.

Note: Predictive margins estimated from multiple linear regression model with Animal Fluency LARCC as the main outcome and MTHFR2 SNPHAP interacted with sex. For list of covariates adjusted for see Table 4.

4. Discussion

4.1. Key Findings

This study tested associations between OCM enzymatic gene variations and cognitive performance change over ~5y mean follow-up among 660–797 African-American HANDLS participants. Overall, MTHFR SNPs rs4846051(A1317G, G>A) and rs1801131(A1298C, G>T) were associated with slower and faster declines in AF, respectively, while rs2066462(C1056T, A>G) was related to slower decline on Trails B (executive function). Among men, rs4846051(A1317G, G>A) was linked to faster decline on BVRT (visual memory), while rs2066462(C1056T, A>G) and rs9651118(C>T) were associated with slower decline on CVLT-List A and rs9651118(C>T) with faster decline on CVLT-DFR. Among women, a slower decline on the domain “verbal memory/fluency” was observed with rs1801133(C677T, A>G). MTHFR₂ SNPHAP (rs1801133(C677T, A>G)/rs1801131(A1298C,G>T): GG) was associated with slower decline on AF among women, while MTHFR₃ SNPHAP(AT) was linked with slower decline on CVLT-List A among men but faster decline on Domain 1 among women, mainly driven by a decline on Digits Span test scores. Similar patterns were observed for MTHFR SNPLCs.

4.2. Previous studies

The positive association between tHcy and adverse cognitive outcomes, including dementia, mild cognitive impairment and AD, were previously shown in several studies (Annerbo, et al., 2005,Blasko, et al., 2008,Gabryelewicz, et al., 2007,Haan, et al., 2007,Kim, et al., 2007,Quadri, et al., 2004,Quadri, et al., 2005). Other studies testing associations between tHcy and cognitive domains of verbal memory (Mooijaart, et al., 2005,Schafer, et al., 2005) and executive function (Elias, et al., 2006,Eussen, et al., 2007,Garcia, et al., 2004,Mooijaart, et al., 2005,Schafer, et al., 2005) among others have reported positive findings as well. (Elias, et al., 2006,Eussen, et al., 2007,Feng, et al., 2006,Schafer, et al., 2005) A recent review and meta-analysis of modifiable risk factors for cognitive decline and impairment concluded that elevated tHcy was associated with an average 93% increase in the risk of incident AD (HR=1.93, 95% CI: 1.50–2.49).(Beydoun, et al., 2014)

For most selected gene polymorphisms, evidence on an effect on tHcy serum concentration is inconsistent (Chango, et al., 2000,Chen, et al., 2001,de Lau, et al., 2010,Fredriksen, et al., 2007,Gaughan, et al., 2001,Hanson, et al., 2001,Jacques, et al., 2003,Ravaglia, et al., 2004,von Castel-Dunwoody, et al., 2005,Yates and Lucock, 2003) except for the MTHFR 677TT variant, which was linked with an increased tHcy level in multiple recent studies. (de Lau, et al., 2010,Elkins, et al., 2007,Ford, et al., 2012b,Fredriksen, et al., 2007,Polito, et al., 2016,Tsai, et al., 2011,Ueland, et al., 2000) Results were inconclusive when examining MTHFR 1298A>C, MTR 2756A>G and CBS 844ins68 in relation to cognition.(Barbaux, et al., 2000,Ravaglia, et al., 2004,Schiepers, et al., 2011) Nevertheless, MTHFR 677TT dosage (C>T) was associated with AD risk in a recent meta-analysis, with a pooled OR of 1.29 (95% CI: 1.07, 1.56) (Rai, 2017) and several cross-sectional and longitudinal studies have found that “TT” was linked to faster decline or poorer performance on global mental status,(Elkins, et al., 2007,Ford, et al., 2012b,Tsai, et al., 2011) and on domains of psychomotor speed,(Elkins, et al., 2007) executive function,(Elkins, et al., 2007,Polito, et al., 2016) short-term memory(Tsai, et al., 2011) and concentration/mental manipulation.(Tsai, et al., 2011) The “C” allele of MTHFR 1298(A>C) was also recently linked with lower abstraction ability.(Cai, et al., 2016) Our haplotype analysis which included only 677A/G-1298G/T gave similar results as a previous study which indicated that Haplotype C(677G-1298G-1793C), of the MTHFR gene is protective against the development of AD.(Wakutani, et al., 2004) In fact, our study concluded that the supposedly “protective” haplotype “GG” was associated with slower decline on a test of verbal fluency among women. Nevertheless, further large studies are needed to replicate our findings and that of others. (Wakutani, et al., 2004)

4.3. Biological pathways

The OCM is a complex metabolic pathway in which folate’s active form (tetrahydrofolate (THF)) transfers methyl groups by acting as a co-factor to specific enzymes.(Troesch, et al., 2016) The OCM consists of a series of interrelated cycles known as methionine, thymidylate and purine cycles. (Troesch, et al., 2016) The neuro-toxic substance tHcy is metabolized by either entering the methionine or the thymidylate cycle. A negative feedback loop exists, whereby under low levels of methionine, tHcy is remethylated into methionine, through a methyl group’s transfer from methylenetetrahydrofolate (MTHF) to tHcy by methionine synthase (MTR), producing THF and methionine. (Shane, 2008,Troesch, et al., 2016) Methionine is further metabolized into S-adenosylmethionine (SAM), the principal methyl-donor in DNA methylation, and the synthesis of phospholipids, myelin and neurotransmitters.(Shane, 2008,Troesch, et al., 2016)

Following transfer of a methyl group from SAM, S-adenosylhomocysteine (SAH) is hydrolyzed back into tHcy.(Selhub, 1999,Troesch, et al., 2016) Serine and THF react together to form glycine and 5,10-methylene-THF with the help of serine hydroxymethyltransferase (SHMT).(Troesch, et al., 2016) Closing the cycle, is the reduction of 5,10-methylene-THF to 5-methylenetetrahydrofolate (MTHF), a reaction catalyzed by the enzyme methylenetetrahydrofolate reductase (MTHFR).(Troesch, et al., 2016) Under conditions where methionine is abundant and tHcy is accumulating, the latter condenses with non-essential amino acid serine to form cystathionine, and then cysteine, (Shane, 2008,Troesch, et al., 2016) a reaction termed “transsulfuration pathway”, which requires two vitamin B6-dependent enzymes (cystathionine β-synthase (CBS) and cystathionase). (Shane, 2008,Troesch, et al., 2016)

In fact, many of the OCM enzymes depend on vitamins B-2, B-6, folate (B-9) and B-12. For instance, MTR depends on the active form of vitamin B12, or methyl-cobalamin,(Shane, 2008,Troesch, et al., 2016) acting as a methyl carrier, and making it essential to the OCM cycle. (Troesch, et al., 2016) Vitamin B12 is activated back into methyl-cobalamin through methionine synthase reductase (MTRR) through re-methylation with one-carbon units from SAM. (Ludwig and Matthews, 1997,Troesch, et al., 2016) Furthermore, MTRR is a flavoprotein, (Leclerc, et al., 1998,Troesch, et al., 2016) thus depending on vitamin B-2 for its activity. Each of SHMT’s 4 sub-units uses vitamin B-6’s active form as a co-factor. (Renwick, et al., 1998,Troesch, et al., 2016) Finally, MTHFR uses flavin adenine dinucleotide (FAD), a co-factor derived from vitamin B-2.. (Leclerc, et al., 1998,Troesch, et al., 2016) An imbalance in any one B-vitamin may alter OCM cycles and tHcy homeostasis. (Troesch, et al., 2016)

4.4. Strengths and limitations

Our study has several strengths, including a relatively large sample, a longitudinal study design and the use of advanced statistical techniques by conducting both multiple linear mixed-effects regression models and OLS multiple linear regression analyses to test associations of OCM SNPs, SNPHAPs and SNPLCs with the annual rates of change in cognitive performance. Although used less frequently than haplotype analysis, LCA was conducted to examine clustering of genotypes within the key enzymes involved in the OCM cycles and the effect of that clustering on cognitive change over time.

Nevertheless, our study has notable limitations. First, the final analytic sample may have been selected in a non-random manner by oversampling certain groups from the African-American participants. A 2-stage Heckman selection model accounted for these biases (Heckman, 1979). Second, baseline age and follow-up durations varied among participants rendering data structure unbalanced. Mixed-effects regression models were therefore used to predict cognitive test scores and annual rates of change at specific ages where data was most dense to estimate the LARCC. Those models assume missingness at random, even though informative censoring may occur. However, due to a relatively younger age distribution in our study, dropout due to cognitive impairment is less likely than for older study populations. (Ibrahim and Molenberghs, 2009) The younger baseline age group, with only 1.7 repeats on average and 5y follow-up may also limit the ability to find clinically significant cognitive change. Our main OLS regression models were additionally controlled for both first-visit and mean age at follow-up. Third, serum tHcy was not available at the time of the analysis to examine tHcy-gene interaction and its potential role in mediating the association between MTHFR gene polymorphisms and age-related cognitive decline. Moreover, such interaction (as well as potential mediating role) can only be tested in larger samples with higher statistical power. Finally, positive findings may have been due to chance, residual confounding by key unmeasured factors or selection bias due to unequal probability of selection from the initial study sample of African-Americans, while negative findings may be due to low statistical power. Some of the positive findings were statistically but not necessarily clinically significant. For instance, as CVLT-List A declines on average by 0.280 units per year with a SD of 0.018, the difference between two SNPLCs (MTHFR5 vs. MTHFR4) was −0.0015, indicating ~8.3% SD in change, a weak net effect. Thus, those findings should be interpreted with caution until they are replicated elsewhere on comparable adult populations.

4.5. Conclusions

In summary, OCM enzymatic gene variations can alter age-related cognitive trajectories among African-American urban adults, specifically in domains of visual and verbal memory and in verbal fluency. Future studies should examine associations of those SNP, SNPLC and SNPHAP with incident dementia, AD and mild cognitive impairment in comparable populations, using a longitudinal study with an extended follow-up period.

Supplementary Material

Supplementary Figures 1-3

Figure S1. (A) Schematic representation of the MTR gene. The SNP and gene coordinates are based on NCBI build 37 (hg19, May 2013, Phase 3 of the 1000 Genomes Project). The MTR gene on chromosome 1 has 32 exons and 108.7 kilobase pairs in size; (B) Genotype frequencies (%) of selected MTR SNPs of original sample with complete genetic data (n=1,024)

Abbreviations: hg=human genome; MTR=Methyl synthase; SNP=Single Nucleotide Polymorphism; vv=variant-variant; wv=wild type-variant; ww=wild type-wild type.

Figure S2. (A) Schematic representation of the MTRR gene. The SNP and gene coordinates are based on NCBI build 37 (hg19, May 2013, Phase 3 of the 1000 Genomes Project). The MTR gene on chromosome 5 has 15 exons and 49.9 kilobase pairs in size; (B) Genotype frequencies (%) of selected MTRR SNPs of original sample with complete genetic data (n=1,024)

Abbreviations: hg=human genome; MTRR=Methyl synthase reductase; SNP=Single Nucleotide Polymorphism; vv=variant-variant; wv=wild type-variant; ww=wild type-wild type.

Figure S3. (A) Schematic representation of the SHMT gene. The SNP and gene coordinates are based on NCBI build 37 (hg19, May 2013, Phase 3 of the 1000 Genomes Project). The SHMT gene on chromosome 17 has 17 exons and ~23.2 kilobase pairs in size; (B) Genotype frequencies (%) of selected SHMT SNPs of original sample with complete genetic data (n=1,024)

Abbreviations: hg=human genome; SHMT=Serine hydroxymethyltransferase; SNP=Single Nucleotide Polymorphism; vv=variant-variant; wv=wild type-variant; ww=wild type-wild type.

Highlights.

• Verbal memory/fluency domain 1 declined slower in women with MTHFR rs1801133(A>G).

• Verbal fluency declined slower in women with “GG” MTHFR SNPHAP (rs1801133/rs1801131).

• The “AT” MTHFR SNPHAP’s association with domain 1 rate of change was sex-specific.

• Similar findings were observed for MTHFR SNPLCs.

LIST OF ABBREVIATIONS

AD

Alzheimer’s Disease

AF

Animal Fluency

ANOVA

Analysis of Variance

BMI

Body Mass Index

BVRT

Benton Visual Retention Test

CVLT

California Verbal Learning Test

DNA

Deoxyribonucleic acid

DS-B

Digits Span-Backwards

DS-F

Digits Span-Forward

HANDLS

Healthy Aging in Neighborhoods of Diversity Across the LifeSpan

HAP

Haplotype

tHcy

Total Homocysteine

HDL-C

High Density Lipoprotein-Cholesterol

LARCC

Longitudinal Annual Rate of Cognitive Change

LCA

Latent Class Analysis

LD

Linkage Disequilibrium

LOD

logarithm of the odds

MMSE

Mini-Mental State Examination

MTHFR

Methylenetetrahydrofolate Reductase

MTR

Methionine synthase

MTRR

methionine synthase reductase

OLS

Ordinary Least Square

RefSeq

Reference Sequence

SE

Standard Error

SHMT

Serine Hydroxymethyltransferase

SNP

Single Nucleotide Polymorphism

SNPHAP

Single Nucleotide Polymorphism Haplotype

SNPLC

Single Nucleotide Polymorphism Latent Class

Trails A

Trailmaking test, Part A

Trails B

Trailmaking test, Part B