Background
The methionine, purine, and thymidylate cycles together encompass one-carbon metabolism in the cytosol [
1]. Both genetic (i.e., polymorphisms in
MTHFR) and environmental factors influence one-carbon metabolism; however, genetic factors are thought to play a minor role, whereas dietary intake of folate, vitamin B
6, and vitamin B
12 explains 35% of the variation [
2].
The most-studied biomarker of one-carbon metabolism is homocysteine (Hcy), a sulfur-containing amino acid generated by the metabolism of methionine [
3]. Inherited deficiencies in cystathionine synthase, methionine synthase, or methylene-tetrahydrofolate reductase enzymes causing homocystinuria and severe atherosclerotic plaques were discovered in children in the mid-20th century [
4]. More recently, hyperhomocysteinemia has been identified as a risk factor for a multitude of conditions, including cognitive decline and Alzheimer dementia [
5‐
10]. Elevated plasma Hcy is linked with deficiency in vitamins B
6, B
12, and folate, as well as oxidative damage [
11,
12]. However, Hcy-lowering trials with B vitamins have met with mixed results that have dampened enthusiasm for targeting Hcy in the prevention of cognitive decline or Alzheimer disease (AD) progression [
13]. There may be several explanations for this inconsistency, but one often overlooked is that emphasis on Hcy specifically may be insufficient in magnitude of effect under the constraints of current clinical trial designs. A role for other participants in the one-carbon cycle that are highly interactive with Hcy may have been underappreciated. For example,
S-adenosylmethionine (SAM),
S-adenosyl-L-homocysteine (SAH), and 5-methyltetrahydrofolate (5-MTHF) balance may operate on several mechanisms that support cognitive function, ranging from epigenetic modulation of synaptic function in the hippocampus to methylation reactions in the liver that convert phosphatidylethanolamine to phosphatidylcholine and facilitate the delivery of essential fatty acids (i.e., docosahexaenoic acid) to the plasma and the brain [
14‐
18].
The ε4 allele of apolipoprotein E (
APOE) gene is a well-known and major genetic risk factor for AD [
19]. Several recent studies have shown the interaction of
APOE genotype and environmental factors, such as diet, on the risk of dementia and AD [
20]. Nutrient intake effect on cognitive function might be influenced by
APOE status, and the association between B
12 and cognitive function was indeed shown to be stronger in
APOE ε4 carriers [
21]. The effects of vitamin B
12 and Hcy on gray matter volume may also be influenced by
APOE genotype in AD [
22]. The clear interaction between Hcy, its cofactors, and
APOE remains to be more broadly investigated.
Metabolomics is an approach with the potential to target and identify perturbations in specific pathways of interest by quantifying a wide range of biochemical compounds in tissue [
23]. We leveraged this approach to test the hypothesis that comprehensive assessment of one-carbon metabolism in cerebrospinal fluid (CSF) and plasma would better explain cognitive impairment and CSF measures of AD pathology in older adults.
Methods
Study population
The participants with cognitive impairment were recruited among outpatients who were referred to the memory clinics, departments of psychiatry, and the Leenaards Memory Center, Department of Clinical Neurosciences, University Hospitals of Lausanne (Switzerland). Cognitively intact participants were recruited from the community through advertisement or among the spouses of memory clinic patients. In total, 120 subjects were enrolled in the biomarker study.
CSF and plasma sample collection
Venous and lumbar punctures were performed between 8:30 and 9:30 a.m. after overnight fasting. For lumbar puncture, a standardized technique with a 22-gauge “atraumatic” spinal needle and the subject in a sitting or lying position was applied [
24]. A volume of 10–12 ml of CSF was collected in polypropylene tubes. CSF samples were centrifuged, frozen in aliquots, and stored at −80 °C before further use. Blood was drawn into ethylenediaminetetraacetic acid-containing vacutainers. After 20–30 minutes on ice, the tubes were centrifuged. Plasma samples were aliquoted in polypropylene tubes and stored at −80 °C.
CSF β-amyloid 1–42 peptide chain, tau, tau phosphorylated at threonine 181, and APOE ε4 genotyping
The measurements were performed using commercially available enzyme-linked immunosorbent assay kits and TaqMan assays (Applied Biosystems, Foster City, CA, USA) as described in Additional file
1: Supplementary Methods.
Liquid chromatography-tandem mass spectrometry (LC-MS/MS) was performed to measure absolute concentrations of metabolites and cofactors of one-carbon metabolism in CSF and blood plasma as described previously [
23] (
see Additional file
1: Supplementary Methods).
Primary outcome measures
Cognitive impairment
Diagnosis of mild cognitive impairment (MCI) or dementia was based on neuropsychological and clinical evaluation and made by a consensus conference of psychiatrists and/or neurologists and neuropsychologists prior to subject inclusion. The participants had no major psychiatric or neurological disorders, nor did they have substance abuse or severe or unstable physical illness that might contribute to cognitive impairment. Magnetic resonance imaging and computed tomographic scans were used to exclude cerebral pathologies possibly interfering with the cognitive performance. Subjects known to take folate and B-vitamin supplementation were excluded from the study. MCI was diagnosed according to widely used consensus recommendations [
25]. Subjects with MCI had memory impairment [
26] and/or impairment in another cognitive domain such as executive tasks, as well as a Clinical Dementia Rating (CDR) [
27] score of 0.5. The diagnosis of probable AD dementia was defined according to the clinical diagnostic criteria for probable dementia due to AD according to the recommendations of the National Institute on Aging and the Alzheimer’s Association [
28] as well as the criteria of the
Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, for dementia of the Alzheimer type [
29] with a CDR of 1. In the present study, we had only nine subjects with AD. Because there is a clinical continuum between MCI and mild dementia, and because the participants with cognitive impairment were patients from memory clinics recruited in the same way regardless of MCI or mild dementia classification, these subjects were collapsed with the MCI group and labeled as cognitively impaired. The participants without cognitive impairment had no history or evidence of cognitive deficits and required a CDR of 0. We therefore defined two categories of subjects on the basis of their CDR (i.e., CDR 0 or >0 [i.e., 0.5 or 1]) (Table
1).
Table 1
Demographics and clinical characteristics
Age, years | 70.4 (7.9) | 66.0 (7.4) | 73.3 (6.9)a
|
Male sex, n (%) | 43 (35.83%) | 17 (35.42%) | 26 (36.11%) |
Education, years | 12.4 (2.6) | 13.2 (2.3) | 11.8 (2.7)a
|
MMSE score, mean | 26.9 (3.1) | 28.5 (1.4) | 25.9 (3.5)a
|
APOE ε4 carrier, n (%) | 37 (30.83%) | 11 (22.92%) | 26 (36.11%)a
|
CSF parameters |
Aβ1–42, pg/ml | 847.4 (265.1) | 957.4 (194.0) | 774.0 (281.5)a
|
tau, pg/ml | 371.3 (278.6) | 221.5 (82.9) | 471.1 (316.6)a
|
p-tau181, pg/ml | 62.0 (35.2) | 45.9 (13.3) | 72.7 (40.9)a
|
p-tau181/Aβ1–42
| 0.088 (0.082) | 0.049 (0.015) | 0.114 (0.097)a
|
Albumin indexb
| 6.1 (2.4) | 5.3 (1.9) | 6.6 (2.5)a
|
Choline, μM | 2.7 (0.5) | 2.6 (0.4) | 2.8 (0.6)a
|
Cystathionine, nM | 35.8 (17.0) | 34.4 (13.8) | 36.7 (18.9) |
Methionine, μM | 3.6 (0.8) | 3.3 (0.5) | 3.9 (0.9)a
|
S-adenosylhomocysteine, nM | 15.1 (5.8) | 13.5 (4.3) | 16.1 (6.3)a
|
S-adenosylmethionine, nM | 183.0 (42.0) | 188.2 (41.1) | 179.4 (42.5) |
Serine, μM | 27.1 (4.7) | 25.7 (4.1) | 28.0 (4.8)a
|
Cysteine, μM | 1.2 (0.4) | 1.2 (0.4) | 1.3 (0.4) |
5-Methyltetrahydrofolate, nM | 45.1 (12.5) | 47.2 (10.9) | 43.8 (13.4) |
Plasma parameters |
Cystathionine, nM | 252.3 (232.4) | 267.5 (290.0) | 242.1 (185.9) |
Glycine, μM | 234.5 (60.6) | 235.9 (59.7) | 233.6 (61.5) |
Methionine, μM | 19.7 (3.6) | 19.4 (2.9) | 19.9 (4.1) |
S-adenosylhomocysteine, nM | 20.9 (9.1) | 18.3 (6.3) | 22.7 (10.2)a
|
S-adenosylmethionine, nM | 71.9 (20.7) | 67.6 (17.7) | 74.7 (22.2) |
Serine, μM | 115.3 (19.9) | 111.5 (16.9) | 117.9 (21.3) |
Cysteine, μM | 158.3 (20.3) | 150.7 (18.4) | 163.3 (20.1)a
|
Homocysteine, μM | 5.6 (1.6) | 5.3 (1.5) | 5.8 (1.7) |
CSF profiles of AD pathology
Subjects were also classified into two groups on the basis of their CSF tau phosphorylated at threonine 181 (p-tau181)/β-amyloid 1–42 peptide chain (Aβ
1–42) ratio: “low” when p-tau181/Aβ
1–42 was ≤0.0779 or “high” when p-tau181/Aβ
1–42 was >0.0779, considered as negative and positive CSF profiles of AD pathology, respectively. This threshold optimized the Youden index [
30] of the receiver operating characteristic (ROC) curve for the prediction of CDR categories (CDR 0 versus CDR >0) [
31]. It was similar to previously reported findings [
32]. Forty-two and 78 patients had positive and negative CSF profiles of AD pathology, respectively.
Biomarker quality control
Metabolites with >5% missingness or below limits of quantification were excluded, which left eight CSF and plasma metabolites (Table
1). Data were log
10-transformed to tend toward a Gaussian distribution and standardized to null average and SD of 1 prior to statistical analyses. One low-quality CSF sample was not analyzed. CSF and plasma metabolite data were available for 119 and 120 subjects, respectively.
Statistical analyses
Using least absolute shrinkage and selection operator (LASSO) logistic regression [
33], we selected biomarkers predict both cognitive impairment and CSF profiles of AD pathology. A reference model was initially generated, testing variables that are likely to be available to the clinician and known risk factors for AD to provide a benchmark for comparison with the models that included one-carbon metabolites. These inputs included age, sex, years of education, presence of
APOE ε4 allele, CSF Aβ
1–42, CSF tau, and CSF p-tau181 concentrations for the prediction of cognitive impairment; and age, sex, years of education, and presence of
APOE ε4 allele for the prediction of AD CSF profiles. In addition to all variables used to make the reference models, all metabolite measurements and CSF albumin index were then included in building so-called best models. The best models were built using LASSO regression by entering the CSF and plasma metabolites separately, thereby producing the two best models. A tenfold cross-validation process was performed for each LASSO analysis using the glmnet package in R [
34], which allows estimating the confidence interval (CI) of the misclassification error for each value of the regularization parameter λ. The LASSO analyses were repeated 100 times (1000 times for the reference model). The models that minimized the upper limit of the cross-validated misclassification error CI across the 100 runs were selected. Their performance was assessed by ROC area under the curve (AUC) estimation using a bootstrap approach with 1000 iterations [
35]. Results were compared visually and formally tested for significance against the reference model using ROC AUC [
36] and accuracy using the McNemar test.
Two-sided correlation analyses between metabolites and CSF Aβ1–42, tau, and p-tau181 were performed with Pearson’s statistics and Bonferroni-corrected for multiple comparisons. Significant correlations between the metabolites and cofactors (p < 0.05) with CSF tau or p-tau181 were studied further in a linear regression model with tau or p-tau181 as the dependent variable and the following explanatory covariates: each metabolite, age, sex, APOE genotype (presence versus absence of an APOE ε4 allele), CDR (0 versus >0), and CSF albumin index (only for plasma metabolites). Values of the regression terms were reported, and their differences from 0 were assessed with t tests. Interaction terms between the metabolites and age, sex, CDR, or APOE genotype were identified in a type II analysis of variance (ANOVA) with an F test.
Discussion
In the present study, we deployed a LC-MS/MS-based method to expand previous work [
38], targeting one-carbon metabolism in both CSF and plasma, and to better understand its role in cognitive impairment and AD pathology. We identified CSF and plasma profiles including several one-carbon metabolites beyond Hcy associated with cognitive impairment in older adults. The association between several CSF and plasma one-carbon metabolites with CSF tau and p-tau181 warrants further research. Although we were unable to predict an a priori-stated CSF profile of AD pathology in the total cohort, the plasma one-carbon metabolites markedly improved the classification of CSF profiles of AD pathology in APOE ε4 carriers.
Our results highlight the contribution of methionine, serine, choline, and cysteine in addition to the more commonly assessed and reported one-carbon metabolites (i.e., SAM, SAH, and Hcy). Methionine supplementation in wild-type mice can induce neurotoxicity, higher levels of tau phosphorylation and Aβ peptides in the brain, and memory loss [
39]. These mechanisms may explain our observation of higher CSF methionine in cognitive impairment. Protective effects of choline against age-related cognitive deficits were previously evidenced in animal models [
40], and better memory performance was related to a higher concurrent choline intake in a large, nondemented, community-based human population [
41]. However, we observed higher CSF choline in cognitive impairment, an inconsistency that may reflect neurodegeneration and the breakdown of synaptic membranes enriched with choline [
42]. We showed significantly increased CSF and plasma SAH in cognitive impairment, which is consistent with previous findings [
14,
43].
Several metabolites of the one-carbon cycle measured in CSF and plasma were associated with tau metabolism, reflected by total tau and p-tau181 in the CSF, after controlling for potential confounders. Higher CSF 5-MTHF appeared protective, whereas higher CSF SAH appeared to promote tau aggregation and neuronal injury. These results are consistent with reports demonstrating higher CSF SAH and lower CSF 5-MTHF associated with higher CSF p-tau181 [
14]. Accordingly, impairment of the one-carbon metabolism, induced by feeding with a high-methionine/low-folate diet, was shown to increase SAH levels, downregulate methyltransferase, reduce protein phosphatase 2A methylation, and induce accumulation of neurofibrillary tangle pathology in mice [
44]. Higher levels of tau phosphorylation were also observed in the brains of wild-type mice fed a methionine-enriched diet, putatively inducing the fragmentation of tau from microtubules and leading to fibrillization [
39]. Together, these results suggested that altered one-carbon metabolism may contribute to microtubule-associated tau protein hyperphosphorylation and neurodegeneration in general. Our inability to observe a relationship between any one-carbon metabolites with CSF Aβ
1–42 was remarkable and consistent with some studies [
14,
45], but in contradiction with other in vitro and in vivo preclinical investigations [
46,
47]. A recent study of human adults with normal cognition found that disturbed one-carbon metabolism may be related to increased CSF levels of Aβ
1–42 and soluble amyloid precursor protein forms, suggesting a contribution to the accumulation of cerebral amyloid pathology [
48], making it difficult to draw conclusions at this time.
We were unable to predict CSF profiles of AD pathology until an interaction with
APOE genotype was identified. For example, whereas plasma Hcy appeared insubstantial to CSF tau in the total sample, hyperhomocysteinemia was relevant to CSF tau in the
APOE ε4 carriers. This was further corroborated by our prediction of positive AD pathology using the p-tau181/Aβ
1–42 ratio after restricting the analysis to
APOE ε4 carriers only. Although this interaction between Hcy,
APOE, and tau pathology was clear in our analysis, the biological rationale behind this interaction is less clear. A recent report demonstrated that plasma Hcy predicts the conversion from MCI to AD only in
APOE ε4 carriers [
49], adding a further level of consistency to our findings. Other metabolites of the one-carbon cycle, such as CSF SAM [
50], may also be more relevant in
APOE ε4 carriers and should be further explored.
Taken together, our results show that one-carbon metabolism is relevant to cognition independent of and beyond AD core pathology. One-carbon metabolism disturbances may cripple DNA repair, methylation, and/or synthesis, as well as reduce the availability of neurotransmitters, phospholipids, and myelin [
51]. Several of these mechanisms may be involved in cognitive dysfunction and underlying neurobiology.
The present study has limitations. We used measures of cognition and CSF measures of AD in a cross-sectional analysis, and the temporality of the associations cannot be extrapolated. Nevertheless, we present the most comprehensive mapping of the one-carbon pathway in both CSF and plasma in older adults. The results are encouraging and in the future should be replicated in an independent cohort as well as confirmed in relation to longitudinal change in cognition and the incidence of AD.