Table 2.
Results of analyses associating metabolites with dementia risk.
| References | Analytical platform, metabolite targets | Statistical analysis | Covariates in fully adjusted model | Statistically significant and/or selected metabolites, adjusted HR (95% CI) for cognitive impairment per SD* | Key findings |
|---|---|---|---|---|---|
| Mielke et al., 2010 | ESI/MS/MS, targeted (SM and ceramides) | Cox proportional hazards regression model | Age, glucose and BMI |
HVLT-delayed recall (per tertile): Total SM: 1.88 (1.05–3.38) Ceramides: C16:0: 2.17 (1.17–3.84) C22:0: 2.12 (1.10–4.08) C24:0: 2.23 (1.23–4.05) Lactosyl C12:0: 2.26 (1.15–4.44) Stearoyl: 1.86 (1.05–3.27) Sulfatide: 2.06 (1.08–3.96) |
High levels of serum SM could predict incident impairment in asymptomatic individuals, and be biomarkers of AD progression. |
| Oresic et al., 2011 | UPLC-MS, untargeted (139 lipids: phospholipids, sphingolipids, and neutral lipids); GC × GC-TOFMS, untargeted (544 small polar metabolites: amino acids, free fatty acids, ketoacids, organic acids, sterols, and sugars) |
Logistic regression model | Age, APOE ε4 | Combination of three metabolits (PC (16:0/16:0), an unidentified carboxylic acid and 2,4-dihydroxybutanoic acid): OR, 8.0 (90% CI: 2.7–27.6) per unit increase | (1) Concentrations of ribose-5-phosphate was decreased, whereas 2,4-dihydroxybutanoic acid and lactic acid were upregulated in converters; (2) Combination of PC (16:0/16:0), an unidentified carboxylic acid and 2,4-dihydroxybutanoic acid predicted AD reasonably well, with AUC = 0.77 (90% CI: 0.65–0.87). |
| Mapstone et al., 2014 | UPLC-ESI-QTOF-MS, untargeted (lipidomic profiling, 2,700 positive-mode features and 1,900 negative-mode features); SID-MRM-MS, targeted (184 small molecules and lipids) | LASSO penalty for putative metabolites selection, and logistic regression model for prediction analysis | Age, gender, education, and visit matched, and additional adjusted for APOE in prediction model | PC diacyl (aa) C36:6, PC aa C38:0, PC aa C38:6, PC aa C40:1, PC aa C40:2, PC aa C40:6, PC acyl-alkyl (ae) C40:6, lysoPC a C18:2, and AC (Propionyl AC (C3) and C16:1-OH); NR | (1) Baseline plasma levels of phosphatidylinositol, serotonin, phenylalanine, proline, lysine, PC, taurine and AC in converters were significant low; (2) A panel of lipids, comprising the 10 putative metabolites, could predict converters well, with AUC = 0.92 (95% CI: 0.87–0.98). |
| Mousavi et al., 2014 | GC-TOF-MS, targeted, 208 metabolites | OPLS-DA | Age-, sex-,and education- matched | 3,4-dihydroxybutanoic acid, docosapentaenoic acid, and uric acid; NR | Metabolites were different in serum in participants at the preclinical stage up to 5 years preceding dementia, despite that the cognitive performance were comparable with healthy controls. |
| Graham et al., 2015 | UPLC-Q-TOF-MS, untargeted (6751 spectral features) | OPLS-DA | Age-matched | 4-aminobutanal, GABA, L-ornithine, N1,N12-diacetlyspermine, N-acetylputrescine, spermine, creatine; NR | (1) Concentrations of 4-aminobutanal, GABA, L-ornithine were low, whereas N1,N12-diacetlyspermine, N-acetylputrescine, spermine, creatine were upgraded in converters relative to matched healthy controls; (2) Polyamine metabolism and L-arginine metabolism were disturbed in converters. |
| Casanova et al., 2016 | FIA-MS/MS, targeted (AC, lipids, and hexoses); HPLC-MS/MS, targeted (amino acids and biogenic amines), 187 metabolites in total. | Logistic regression model, 4 machine learning methods (EN-RLR, RF, SVM, L1-RLR) | Age and sex matched | Propionylcarnitine, glutarylcarnitine, creatinine, methionine, ornithine, serine, taurine, threonine, glucose, PC aa C36:4, PC aa C38:4, PC ae C30:2, PC ae C42:5, and PC ae C44:4; NR | (1) A panel of 10 serum metabolites found by Mapstone et al. which could detect preclinical AD within 3 years, could not be replicated in the two cohorts; (2) A modest signal was found in one cohort (BLSA) with distinct metabolites associated with preclinical AD; however, the classification accuracies were not good, with AUC = 0.64 (95% CI: NR). |
| Simpson et al., 2016 | UPLC-Q-TOF-MS, targeted (PC16:0/20:5, PC16:0/22:6, and PC18:0/22:6) | Generalized linear mixed model | Age, sex, education year, APOE ε4 | None | (1) Baseline and changes in plasma PC concentrations were not associated with longitudinal changes in cognitive performance; (2) Dysregulation of peripheral PC metabolism may be a common feature of both AD and age-associated differences in cognition. |
| Abdullah et al., 2017 | HPLC-MS, untargeted (lipidomics, including PC, PE, PI, lysoPC, and so on) | Cox proportional hazards regression model | Age, education, gender, creatinine, and treatment with statins or anti-hypertensive medications | Ratio of AA to DHA; NR | (1) High AA to DHA ratios were associated with the risk of developing MCI/AD within 3 years; (2) Combining the APOE genotypes, blood AA and DHA species and the Aβ42/Aβ40 ratio improves the accuracy for detecting preclinical MCI/AD; (3) An interaction between the ε4 status and high AA to DHA ratios was found with the risk of developing MCI/AD. |
| Bressler et al., 2017 | GC-MS and LC-MS, untargeted (118 named and 86 unnamed metabolites) | Linear regression models were used for 6-year cognitive change analyses; Cox proportional hazards models were used for incident hospitalized dementia analyses |
Age, gender, education, eGFR, DM, hypertension, BMI, LDL-C, current smoking, alcohol intake and APOE ε4 |
6-year cognitive chang: beta (SE) (DWRT) N-acetyl-1-methylhistidine: −0.656 (0.183), (DSST) Docosapentaenoate (n-6 DPA): 1.254 (0.320), (DSST) X-12844: 1.404 (0.391); Incident dementia: 4-androsten-3 beta, 17 beta-diol disulfate 1: 1.25 (1.11–1.40) pregnen-diol disulfate: 1.35 (1.17–1.56) 5 alpha-androstan-3 beta,17 beta-diol disulfate: 1.26 (1.12–1.42) X-11440: 1.37 (1.18–1.60) X-12851: 1.26 (1.12–1.43) |
(1) Basline high levels of N-acetyl-1-methylhistidine and low levels of docosapentaenoate were significantly associated with greater 6-year change in DWRT and DSST scores; (2) Three sex steroid hormones (4- androsten-3 beta, 17 beta-diol disulfate 1, 5 alpha-androstan-3 beta, 17 beta-diol disulfate and pregnen-diol disulfate) were associated with an increased risk of dementia. |
| Chouraki et al., 2017 | LC-MS, untargeted (54 amines and related metabolites, 59 organic acids and related metabolites, and 104 lipids) | Cox proportional hazards model | Age, sex, education, APOE ε4, total homocysteine, SBP, antihypertensive medication, DM, smoking, CVD, AF and left ventricular hypertrophy. | Dementia: Anthranilic acid: 1.38 (1.12–1.69) Glutamic acid: 1.33 (1.06–1.66) Taurine: 0.74 (0.59–0.92) Hypoxanthine: 0.74 (0.59–0.91); AD: Glutamic acid: 1.37 (1.04-1.79) |
(1) Higher plasma anthranilic acid levels were associated with greater risk of dementia; (2) Higher plasma glutamic acid, lower taurine and lower hypoxanthine showed possible associations with greater dementia risk. |
| Li et al., 2017 | HPLC-MS/MS and FIA-MS/MS, targeted (188 metabolites, including 40 AC, 21 amino acids, 21 biogenic amines, 15 sphingolipids, 90 glycerophospholipids, and 1 hexose) | Logistic regression models were used in prediction analysis for baseline and changes in 9 targeted metabolites and incident MCI and dementia; linear regression models were used for cognitive change analyses | Age, race, sex, APOE, education, DM, BMI, drinking, smoking, sports index, SBP, use of antihypertensive medications, CVD, HF, stroke, TC, HDL-C, and TG. |
MCI: LysoPC a C18:2: 1.66 (1.04–2.64) MCI and dementia: LysoPC a C18:2: 1.52 (1.03–2.37) Global cogintive function score change: Propionyl-L-carnitine (C3): 0.11 (0.01-0.22) |
(1) A panel of 10 serum metabolites found by Mapstone et al. which could detect preclinical AD within 3 years, was not predictive of MCI or dementia in ARIC-NS; (2) Higher concentrations of lysoPC a C18:2 was significantly associated with high risk of MCI or MCI and dementia; (3) Higher levels of propionyl-L-carnitine (C3) were significantly associated with slower decline in the global cognitive score; (4) Higher concentrations of 28 plasma amino acids, carnitines, phospholipids, and sphingomyelins were prospectively associated with MCI or dementia in African Americans. |
| Toledo et al., 2017 | ADNI: UPLC-/MS/MS and FIA-MS/MS, targeted (186 metabolites, including amino acids, biogenic amines, cylcarnitines, SM, PCs, and lysoPC) ERF: ESI/MS/MS, targeted RS: NMR, untargeted |
Cox proportional hazards model was used to evaluate the association of metabolite levels with progression from MCI to AD; Mixed effects model was used to evaluate the association of metabolites with change in ADAS-Cog13; Linear regression model was used to evaluate the association of metabolites and g-factor. |
Discovery: Age, gender, APOE ε4, and education. Validation: Age, gender, lipid-lowering medication, and education. PC ae C44:4: 0.49 SM (OH) C14:1: 0.015 SM C16:0: 0.0009 SM C20:2: 0.11 α-AAA:−0.093 Valine:−0.0006 Validation: g-factor: beta (SE, NR) (ERF) PC ae C40:3:−0.231 (ERF) SM C20:2:−0.239 (RS) Valine: positive correlation (NR) Incident AD: (RS) Valine: negative correlation (NR) |
Discovery: Progression MCI to AD: (95% CI, NR) PC ae C36:2: 1.056 PC ae C40:3: 5.98 PC ae C42:4: 1.96 PC ae C44:4: 5.89 SM (OH) C14:1: 1.08 SM C16:0: 1.004 SM C20:2: 1.9 ADAS-Cog13 Change: beta (SE, NR) C14:1: 1.39 PC ae C40:3: 0.38 PC ae C42:4: 0.15 |
(1) Six metabolites (PC ae C40:3, PC ae C42:4, PC ae C44:4, SM (OH) C14:1, SM C16:0, and SM C20:2) showed a positive association with risk of conversion from MCI to AD and cognitive score change; (2) Lower valine and α-AAA values were associated with faster cognitive decline; (3) Valine was strongly associated with a higher general cognitive ability, decrease in valine concentration was associated with risk of AD in the validation cohorts (RS). |
| Dorninger et al., 2018 | UFLC-MS/MS, targeted (lysoPC, PlsCho, and lyso-PAF) | Baseline phospholipid difference in groups: t-test; Phospholipid changes: ANCOVA; Group difference in phospholipid changes: ANCOVA |
Gender, group indicator, APOE, and intake of lipid-lowering drugs | total lysoPC, lysoPC 18:2, total PlsCho, and total lyso-PAF; NR | (1) Total lysoPC and total PlsCho levels were lower, whereas total lyso-PAF was higher at baseline in converters than those in healthy controls; (2) The levels of lysoPC, PlsCho, and lyso-PAF increase significantly during normal aging as well as in developed probable AD patients. |
| Tynkkynen et al., 2018 | NMR analysis was used in all cohorts except for FHS (LC-MS), untargeted (228 metabolites, including lipids, fatty acids, amino acids, ketone bodies, and gluconeogenesis-relatedmetabolites) | Cox proportional hazards models | Age, sex, education grade, APOE ε4, SBP, hypertension treatment, DM, smoking, and any CVD. |
Demnetia: Creatinine: 0.90 (0.83–0.97) Isoleucine: 0.87 (0.80–0.94) Leucine: 0.83 (0.76–0.90) Valine: 0.84 (0.78–0.91) L-HDL-CE-%: 1.11 (1.03–1.21) S-VLDL-C: 0.87 (0.81–0.94) XL-VLDL-C-%: 1.11 (1.02–1.20): 0.91 (0.84–0.99) AD: Isoleucine: 0.89 (0.81–0.98) Leucine: 0.88 (0.79–0.97) Valine: 0.87 (0.79–0.96) L-HDL-CE-%: 1.12 (1.01–1.23) |
(1) Lower levels of the BCAA such as valine were associated with an increased risk of both all dementia and of AD; (2) Inverse associations of creatinine, total cholesterol in S-VLDL-C, and triglycerides to total lipids ratio in very large VLDL were found associated with incident dementia, but not with AD; (3) The concentration of L-HDL-CE-% was associated with an increased risk of AD. |
| van der Lee et al., 2018 | ERF, RS, NTR, VUMC ADC, EGCUT, WHII, Finrisk 97, and DILGOM were used NMR platform, SHIP was used LC-MS/MS, FHS was used LC-MS, AgeCoDe was used GC-FID; untargeted (299 metabolites in discovery analysis, including lipids, fatty acids, amino acids, ketone bodies, and gluconeogenesis-relatedmetabolites) | Cox proportional hazards models (logistic regression was used in VUMC ADC) | Age, sex, BMI, lipid-lowering medication, and APOE ε4. |
Dementia: Small particles -HDL-free cholesterol: 0.85 Medium particles -HDL-phospholipids: 0.90 DHA: 0.91 Glutamine: 1.08 Medium particles -HDL-cholesterol esters: 0.92 Medium particles-HDL-total cholesterol: 0.92 AD: Small particles -HDL-free cholesterol: 0.87 DHA: 0.89 Glutamine: 1.11 |
|
| Varma et al., 2018 | FIA-MS/MS and HPLC-MS/MS, targeted (187 metabolites, including amino acids, biogenic amines, AC, lipids, and hexoses) | Machine-learning method (SVM and RF) was used to select potential brain metabolite signature of AD; Cox regression models were used to test blood metabolite associations with risk of conversion from normal cognition to incident AD in BLSA and risk of conversion from MCI to incident AD in ADNI. |
Age and sex |
Incident AD in BLSA: per log unit SM C16:0: 4.43 (1.70–11.52) SM C16:1: 3.46 (1.52–7.87) SM (OH) C14:1: 3.54 (1.37–9.12) SM C18:1: 2.26 (1.05–4.85) PC aa 38:4: 0.25 (0.10–0.63) PC ae C34:2: 3.06 (1.21–7.70) Cognitive performence change in BLSA: beta (SE) Attention: SM C18:1: −0.17 (0.07) PC aa C40:6: −0.12 (0.05) Language: arginine: −0.14 (0.07) lysoPC a C18:0: −0.15 (0.06) PC ae C40:1: −0.25 (0.12) SM C26:1: −0.53 (0.27) Visuospatial ability: arginine: 0.20 (0.10) spermidine: 1.22 (0.57) Converters from MCI to AD in ADNI: SM C18:1: 2.35 (1.27–4.36) PC aa 38:4: 2.38 (1.19–4.74) |
Perturbations in sphingolipid metabolism may be integral to the evolution of AD neuropathology as well as to the eventual expression of AD symptoms in cognitively normal older individuals. |
AA, arachidonic acid; AAA, aminoadipic acid; Aβ, β-amyloid; AC, acylcarnitines; ADAS-Cog13, Alzheimer's Disease Assessment Scale–Cognition, lower levels indicate better cognition; ADNI, Alzheimer's Disease Neuroimaging Initiative; AF, atrial fibrillation; AgeCoDe, German Study on Aging, Cognition, and Dementia; ANCOVA, analysis of covariance; APOE, apolipoprotein E; ARIC-NS, Atherosclerosis Risk in Communities-Neurocognitive Study; AUC, area under the curve; BCAA, branched-chain amino acids; BLSA-(NI), Baltimore Longitudinal Study of Aging-(neuroimaging substudy); BMI, body mass index; CI, confidence interval; CVD, cardiovascular disease; DHA, docosahexaenoic acid; DILGOM, Dietary, Lifestyle and Genetic determinants of Obesity and Metabolic Syndrome Study; DM, diabetes mellitus; DSST, Digit Symbol Substitution Test; DWRT, Delayed Word Recall Test; EGCUT, Estonian Biobank (Estonian Genome Center, University of Tartu); eGFR, estimated glomerular filtration rate; EN-RLR, elastic net regularized logistic regression; ERF, Erasmus Rucphen Family study; ESI/MS/MS, electrospray ionization triple stage quadruple tandem mass spectrometer; FHS, Framingham Heart Study; FIA-, flow injection analysis-; GC, gas chromatography; HDL-C, high density lipoprotein cholesterol; HF, heart failure; HPLC-, high-pressure liquid chromatography; HVLT, Hopkins Verbal Learning Test; L1-RLR, L1 regularized logistic regression; LASSO, least absolute shrinkage and selection operator; LC, liquid chromatography; LDL-C, low density lipoprotein cholesterol; L-HDL-CE-%, Cholesterol esters to total lipids ratio in large HDL; lyso-PAF, lyso-platelet activating factor; lysoPC, lysophophatidylcholine; MCI, mild cognitive impairment; MMSE, Mini-Mental State Examination; MS, mass spectrometry; NMR, nuclear magnetic resonance; NTR, Netherlands Twin Registry; NR, not reported; OPLS-DA, orthogonal projection to latent structures-discriminant analysis; PC, phosphatidylcholine; PCA, principal component analysis; PE, phosphatidylethanolamine; PI, phosphatidyl-inositol; PlsCho, choline plasmalogen; Q-TOF, quadrupole time-of-flight; RF, random forest; RS, Rotterdam Study; SBP, systolic blood pressure; SD, standard deviation; SHIP-Trend, Study of Health in Pomerania–Trend; SID-MRM-, stable isotope dilution–multiple reaction monitoring-; SM, sphingomyelins; S-VLDL-C, total cholesterol in small VLDL; SVM, support vector machines; TC, total cholesterol; TG, triglycerides; TMT, Trail Making Test; UFLC-, ultra-fast liquid chromatography; UPLC-, ultra-performance liquid chromatography coupled to-; VUMC ADC, VUMC Amsterdam Dementia Cohort; WFT, Word Fluency Test; WH II, Whitehall II study; WHAS II, Women's Health and Aging Study (WHAS) II; X-*, unnamed metabolites; XL-VLDL-C-%, total cholesterol to total lipids ratio in small VLDL; XL-VLDL-TG-%, triglycerides to total lipids ratio in very large VLDL; yrs, years; *Except where stated otherwise.