Sage Journals: Discover world-class research

Abstract

Background

Cerebrospinal fluid (CSF) metabolomics offers an opportunity to investigate in vivo biological pathways impacted in the human brain by Alzheimer's disease (AD). While impairments in brain glucose metabolism and lipid homeostasis are implicated in AD, the underlying metabolic pathways remain unclear. Genotype information can also be leveraged to study associations between CSF metabolites and AD genetic risk.

Objective

To evaluate how CSF metabolomic profiles and genetic risk are associated with AD pathology as reflected by established CSF biomarkers (Aβ, P-Tau, and T-Tau).

Methods

We collected CSF mass spectrometry measurements of 678 metabolites and 4865 unnamed compounds, as well as genome-wide genotype data from 487 individuals in the Amsterdam Dementia cohort. Polygenic risk scores (PRS) for AD were calculated. Elastic net regression models were used to predict AD biomarker levels with CSF metabolites, and pathway enrichment analysis was performed to assess the metabolic pathways involved.

Results

98 CSF metabolites were found to be significantly correlated with P-Tau or T-Tau, but none with Aβ CSF levels. Elastic net regression models identified 42 and 34 metabolites predicting P-Tau and T-Tau, respectively, including novel associations with Anserine and Fucose. Pathway enrichment analysis implicated Pentose and Glucuronate Interconversions, Glycerophospholipid Metabolism, and ABC Transporters in AD pathology. PRS analysis highlighted four CSF phosphatidylcholines significantly associated with AD genetic risk.

Conclusions

CSF metabolites demonstrate a lack of Aβ levels associations, contrasting with multiple significant findings for P-Tau and T-Tau. Novel associations with Anserine and Fucose may provide new insights into metabolic pathways impacted by AD pathology.

Keywords

Alzheimer's disease amyloid-beta biomarkers cerebrospinal fluid genetics metabolomics tau

Introduction

Cerebrospinal fluid (CSF) has emerged as a valuable source for studying ongoing neurobiological processes due to its proximity to the brain. Biomarkers for Alzheimer's disease (AD), such as decreased amyloid-β (Aβ) 1-42 levels and increased levels of phosphorylated tau (P-Tau) and total tau (T-Tau),¹ can be measured from CSF in vivo. These biomarkers are associated with disease status and are used for diagnostic purposes,² reflecting the presence of extracellular Aβ plaques, intracellular tau tangles, and neurodegeneration in AD patients.^3,4 Metabolic dysfunction has been increasingly implicated in AD pathogenesis, with prior studies linking impairments in glucose metabolism⁵ and lipid homeostasis⁶ to AD. However, the underlying metabolic pathways remain largely unknown. Metabolomics, the comprehensive analysis of thousands of metabolites, offers a powerful tool to study these metabolic pathways and their components simultaneously.⁷ Plasma metabolomics, in particular, has been widely used to examine metabolic alterations associated with AD, revealing disruptions in various biochemical processes. For instance, studies have identified links between AD and alterations in amines and oxidative stress markers,⁸ disruptions in cholesterol and sphingolipid transport,⁹ and abnormalities in ammonia and glutamate metabolism.¹⁰ Additionally, plasma lipid profiles have been used to predict AD progression,^11,12 highlighting the potential of metabolomic approaches in understanding disease mechanisms.

Despite these insights, plasma and CSF metabolomics provide distinct perspectives on AD-related metabolic changes. Previous studies have shown that only a subset of metabolites exhibit significant correlations between these biofluids,^13–15 likely due to the restrictive nature of the blood-brain barrier, which limits molecular exchange between the bloodstream and CSF.¹⁶ As a result, CSF metabolomics offers a more direct window into metabolic processes occurring within the central nervous system, potentially capturing AD-related changes that may not be reflected in plasma. Previous studies have reported associations between CSF metabolites and P-Tau and T-Tau CSF levels, identifying metabolic pathways involved in tau pathology and neurodegeneration.^15,17,18 However, these studies did not find a similar pattern for CSF Aβ levels in AD patients. Additionally, these prior studies were limited by small sample sizes (288 individuals) and the number of metabolites tested (326), lacking the ability to provide a comprehensive picture of brain metabolic pathways. A more comprehensive approach, with a larger sample size and a broader range of metabolites, is needed to fully assess the connection between AD CSF biomarkers and in vivo metabolic pathways in the human central nervous system.

Metabolites reflect intricate biological processes and hold the potential to gauge the collective impact of a disease, pinpoint biomarkers, and chart disease progression.^17,19 While metabolite levels are influenced by an individual's physiological state and environmental exposures, it has been shown that genetic factors also play a role. Genome-wide association studies (GWAS) of human blood metabolites have mapped multiple loci influencing the metabolome,^20,21 with the largest study identifying associations with 690 metabolites at 248 loci.²² In the context of CSF, GWAS have shown that CSF metabolite levels are also associated with genetic variants, including those affecting monoamine levels,²³ amino acids, lipids, and nucleotides.^24–26 These SNP associations have been used to identify CSF metabolites linked to brain-related phenotypes.²⁴ AD is a highly heritable disorder, with the APOE ε4 allele being the strongest genetic risk factor.^27,28 In addition to the APOE locus, GWAS of AD have identified more than 80 genetic risk loci, each having a small association with the disease.²⁹ The SNP-based heritability of late-onset AD is estimated to be between 36% and 59% in pathologically or CSF-confirmed cases,³⁰ with the APOE ε4 and APOE ε2 alleles explaining up to 25% of the genetic variance.³¹ CSF Aβ, P-Tau, and T-Tau levels are also heritable traits, as underscored by previous GWAS.^4,32 While the APOE locus predominantly predicts Aβ CSF levels, it is also associated with P-Tau and T-Tau levels, even though other genetic loci are more strongly associated with the Tau biomarkers.⁴ Genetic loci associated with AD can be harnessed to construct polygenic risk scores (PRS), which aggregate the genetic effects of these regions to evaluate an individual's susceptibility to AD.³³ Integrating these PRS measurements with metabolites can help identify metabolic profiles associated with a disease's genetic risk, highlighting potential metabolism regulators intertwined with disease mechanisms. For instance, an atlas of polygenic scores from the UK Biobank identified associations between the PRS for body mass index and coronary heart disease with metabolite levels of lipoproteins.³⁴

Our study leveraged AD CSF biomarkers as indicators of AD neuropathologic changes, integrating CSF metabolomics and genetic risk factors (i.e., polygenic risk scores, PRS) to uncover metabolic pathways involved in AD etiology and pathophysiology. For this purpose, we collected 5543 CSF mass spectrometry measurements from AD dementia patients, mild cognitive impairment (MCI) patients, and cognitively unimpaired subjects recruited from Alzheimer Amsterdam-based cohorts,^35–37 along with genome-wide genotype data. Through correlation and elastic net regression analyses, we examined which CSF metabolites are associated with P-Tau, T-Tau, and Aβ CSF levels. Additionally, following an assessment of the contribution of APOE to AD PRS, we employed linear models to gauge the predictive capabilities of AD PRS for P-Tau, T-Tau, and Aβ CSF levels, and CSF metabolites. This study expands on past CSF metabolomics studies through increases in both sample size and panel of CSF metabolites measurements, identifying novel metabolites associations with P-Tau and T-Tau CSF levels that could point to metabolic pathways impacted by AD pathology.

Methods

Sample information and processing

Study participants

A total of 977 study samples (with an average age of 52.7 ± 16.6 years and 35.9% female) were used in this analysis, drawn from both a memory clinic cohort and a group of cognitively healthy subjects in the Netherlands. The memory clinic cohort was used throughout all the analysis presented in the paper, whereas the cognitively healthy cohort was only used for the polygenic risk score analysis.

The memory clinic cohort comprised samples from three cohorts affiliated with the Alzheimer Center Amsterdam. These cohorts include the Amsterdam Dementia Cohort (ADC),³⁵ the 90+ Study,³⁶ and the Twin Study.³⁷ The ADC, initiated in the year 2000, is an ongoing observational follow-up study of patients who visited the memory clinic at Amsterdam UMC, location VU University Medical Center (VUmc), with dementia diagnoses made according to established guidelines for neurodegenerative diseases.^38–41 The 90+ Study, part of the Innovative Medicine Initiative European Information Framework for AD (EMIF-AD), focuses on cognitively healthy individuals aged 90 and above, aiming to identify factors associated with resilience to cognitive impairment in the old.³⁶ For the Twin Study, monozygotic twins (one subject per twin pair) were recruited from the Netherlands Twin Register⁴² to participate in the PreclinAD study, a component of the EMIF-AD project (http://www.emif.eu/).37 Exclusion criteria for these cohorts included the use of high-dose antidepressants, epilepsy requiring anti-epileptic drugs, and a diagnosis of schizophrenia or bipolar disorder. Cognitive status for all participants was assessed using the Mini-Mental State Examination (MMSE), an 11-item questionnaire that evaluates various cognitive domains, with scores ranging from 0 (severe impairment) to 30 (no impairment).⁴³

Cohort of cognitively healthy subjects

The recruitment of these neurotypical subjects is described before.²³ In short, inclusion involved patients undergoing spinal anesthesia for minor elective surgical procedures, ages between 18 and 60, and with all four grandparents born in The Netherlands or other Northwestern European countries (Belgium, Germany, UK, France, and Denmark). Each potential participant underwent a telephone interview to screen for self-reported psychotic or major neurological disorders (such as stroke, brain tumors, or neurodegenerative diseases) and to document any use of psychotropic medication.

An overview of the characteristics for the memory clinic cohort (N = 487) and the cognitively healthy subjects cohort (N = 449, all cognitive healthy controls) can be found in Table 1 and Supplemental Table 1. The ADC samples comprises n = 220 cognitively unimpaired subjects, n = 87 subjects with MCI, and n = 180 patients with AD-type dementia. As expected, patient groups in the ADC differed from each other, with the AD-type dementia group having more APOE ε4 carriers, fewer APOE ε2 carriers, and exhibiting more abnormal AD CSF biomarkers compared to the MCI and cognitively unimpaired subjects (Supplemental Figure 1). Cognitively healthy subjects, in comparison to memory clinic subjects, were less frequently female, younger, had a lower APOE ε4 frequency, and a higher APOE ε2 frequency.

Table 1.

Cohort characteristics.

Characteristic	Clinical Cohort (n = 610)	Samples with CSF and Genotype (n = 487)
Age, mean (SD)	65.5 (9.4)	65.5 (9.7)
Female, n (%)	272 (44.6)	207 (42.5)
Diagnosis
Controls, n (%)	297 (48.7)	220 (45.2)
Age, mean (SD)	-	65 (12.0)
MMSE, mean (SD)	-	28 (1.4)
MCI, n (%)	103 (16.9)	87 (17.9)
Age, mean (SD)	-	67 (7.7)
MMSE, mean (SD)	-	26 (2.0)
Dementia, n (%)	210 (34.4)	180 (36.9)
Age, mean (SD)	-	65 (7.2)
MMSE, mean (SD)	-	21 (4.2)
APOE genotype
ɛ2/ɛ2, n (%)	-	-
ɛ2/ɛ3, n (%)	-	29 (6.0)
ɛ2/ɛ4, n (%)	-	8 (1.6)
ɛ3/ɛ3, n (%)	-	206 (42.3)
ɛ3/ɛ4, n (%)	-	170 (34.9)
ɛ4/ɛ4, n (%)	-	74 (15.2)
CSF Aβ42 (pg/ml), n (%)	-	399 (81.9)
CSF P-Tau (pg/ml), n (%)	-	407 (83.6)
CSF T-Tau (pg/ml), n (%)	-	407 (83.6)

Controls include both individuals with normal cognition and subjective cognitive decline. MCI: mild cognitive impairment; P-Tau: phosphorylated tau; T-Tau: total tau; MMSE: Mini-Mental State Examination

All participating studies were approved by their respective Medical Ethics Committees, including approval from the METC Amsterdam UMC and UCLA Institutional Review Board (IRB#20-002064). Informed consent, either from the patient or from the legal representative, was obtained from all participants. All methods were performed in accordance with the relevant guidelines and regulations.

CSF data collection

Memory clinic cohort: CSF samples were acquired through lumbar puncture using a 25-gauge needle and syringe, followed by storage at −80°C. Aβ, T-Tau, and P-Tau levels were assessed as part of the diagnostic work-up, employing enzyme-linked immunosorbent assays (ELISA) (Innotest: Fujirebio, Ghent, Belgium).³⁵ In the ADC cohort, CSF Aβ values were adjusted for drift over time, as detailed previously.⁴⁴ In this cohort, the biomarker abnormality cut-offs used were CSF Aβ₄₂ < 813 pg/ml, CSF t-tau > 375 pg/ml, and CSF p-tau > 52 pg/ml. Cognitively healthy subject cohort. Each subject provided a 6 ml CSF sample through lumbar puncture, immediately stored in fractions of 0.5 and 1 ml at −80°C, as previously described.²³

CSF metabolite processing

As previously described,²⁵ a total of 5543 CSF metabolites were assessed for this study. CSF extraction followed a previously published protocol,⁴⁵ using a mixture of acetonitrile/water/isopropanol (2:2:3) to precipitate proteins and extract metabolites. CSF metabolites were measured using gas chromatography (GC) mass spectrometry (MS) and two methods of liquid chromatography (LC) tandem mass spectrometry (MS/MS) at the West Coast Metabolomics Center, UC Davis. These methodologies captured different aspects of the metabolome, including GC-time-of-flight (GC-TOF) MS for primary metabolism (393 metabolites), LC charged surface hybrid quadrupole time-of-flight (CSH-QTOF) MS/MS for complex lipids (3532 metabolites), and hydrophilic interaction liquid chromatography quadrupole time-of-flight (HILIC-QTOF) MS/MS for biogenic amines (1618 metabolites). For GC-TOF MS, the 7890A Agilent™ instrument (column: Restek Rtx-5Sil MS) was used, with fragments analyzed by the Leco Pegasus IV (TOF) to obtain accurate mass information at high resolution.⁴⁶ For MS/MS (i.e., CSH-QTOF and HILIC-QTOF), the Agilent 6530a™ was used for positive electrospray ionization (ESI) and the Agilent 6550™ for negative ESI.

Raw data processing was performed using Mass Spectrometry-Data Independent Analysis (MS-DIAL) version 4.6 and mass spectral feature list optimizer (MS-FLO).^47,48 The data were provided as peak heights for the quantification ion (m/z value) at the specific retention time (rt value). Peak heights were used instead of peak areas because they offer greater precision for low-abundant metabolites, as baseline determinations have a larger influence on peak areas. To ensure reliable signal detection, measurements with intensities below a 10:1 ratio compared to blank controls were excluded. All metabolite intensities were normalized using Systematic Error Removal using Random Forest (SERRF), which employs pooled aliquots to normalize metabolite intensity.⁴⁹ Each compound was identified by a unique identifier consisting of retention time (rt) and mass-to-charge ratio (m/z), annotated using the West Coast Metabolomics Center's in-house m/z-rt databases (LipidBlast spectral library for CSF-QTOF) and MS/MS spectral matching to libraries from the National Institute of Standards and Technology (NIST)⁵⁰ and the MassBank of North America (MoNA). In total, we measured 5543 CSF compounds, including 814 annotated metabolites and 4729 unannotated compounds (Supplemental Excel file Metabolomics_meta_data includes complete mass spectrum characteristics for each metabolite, retention time, mass-to-charge ratio [m/z] and Metabolomics Standards Initiative (MSI) levels for the annotations).

To address missingness, metabolite levels were initially examined across each cohort, and metabolites with missing data for over 20% of individuals were excluded, removing 282 compounds. For the remaining 5261 CSF compounds, missing values were imputed to half the minimum value for the corresponding metabolite across the cohorts. This decision was based on the assumption that these metabolites likely exist in quantities too low to be detected in these individuals, a method consistent with previous approaches.¹⁷ Subsequently, inverse-rank normalization was applied to all metabolites to ensure normality for downstream analysis.

Genotyping and imputation

Memory clinic samples underwent genotyping using the Illumina Global Screening Array (GSA), while genotype data for cognitively healthy controls were obtained through the OmniExpress Exome array. Initial filtering of autosomal genotypes involved excluding SNPs with <2% SNP-missingness and >5% minor allele frequency (MAF) using plink,⁵¹ performed separately per cohort. Individuals with a call rate below 98% were excluded. Following this, genotype VCF files were uploaded to the TopMed server for imputation and liftover to hg38. Post-imputation quality was assessed by filtering variants with imputation R2 > 0.3, resulting in approximately 8 million SNPs per cohort for subsequent analyses. For the polygenic risk score analysis, imputed genotypes were subsequently merged between the two cohorts, and once again filtered for variants with <2% SNP-missingness and >5% MAF.

After correcting for ancestry PCs and age outliers (individuals that were over 3 standard deviations away from the average of the first ancestry PC, and over 100 years of age), 26 samples were removed from the memory clinic cohort, with 461 individuals remaining for the statistical analysis.

Statistical analyses

Correlations between AD biomarkers and CSF metabolites

Using the residuals of linear models that included diagnosis, age and sex, individual metabolites were tested for Spearman correlation with CSF phosphorylated tau, total tau, and Aβ₁₋₄₂ using the cor.test function in R. Any correlations that passed Bonferroni multiple testing correction were considered significant correlations.

Elastic net regression

In order to restrict the number of metabolites to those most relevant to predicting AD biomarkers, elastic net regression was utilized using the train function from the caret package in R.⁵² Due to the lack of AD biomarker data in the cognitively healthy cohort, this analysis was limited to the Amsterdam based cohort. This method was implemented to select the most important metabolites out of the 5261 metabolites for predicting CSF phosphorylated tau, total tau, and Aβ₁₋₄₂. In the elastic net model, lambda and alpha were selected via 10-fold cross-validation. To account for variability in predictions due to differences in data partitioning, the dataset was randomly split into training (80%) and testing (20%) sets across 1000 iterations. In each iteration, a model was trained using 80% of the data to predict AD biomarkers, while the remaining 20% was used to assess prediction accuracy. To determine the added predictive value of CSF metabolites, baseline models were tested using only age and sex. Metabolites that were included in over 80% of the 1000 models were considered highly predictive. Since this approach aggregates results from multiple iterations, individual best-performing lambda and alpha values for each model are not reported. Additionally, to verify the importance of these metabolites, the rest of the metabolites that were not considered highly predictive were corrected for the effects of the highly predictive metabolites.

Pathway enrichment analysis

Metabolite Set Enrichment Analysis (MSEA) and PaIRKAT⁵³ were employed to elucidate the biological pathways represented in metabolites of interest. MSEA through the “Pathway Enrichment” tool on MetaboAnalyst⁵⁴ was performed on metabolites correlated with either total tau or phosphorylated tau. MSEA identified pathways whose metabolites appeared more often than chance among metabolites that correlated with tau. PaIRKAT integrates known biological connections between metabolites in pathways into a semi-parametric kernel machine regression model to associate entire pathways with phenotypes. All named metabolites that were found in the KEGG database from the three platforms were included to create networks of metabolites representing pathways. These pathways were tested for association with both phosphorylated tau and total tau while corrected for sex and age on https://csevern.shinyapps.io/pairkat/.

Polygenic risk scores calculation

Polygenic risk scores for AD were calculated using PRScs software⁵⁵ and the latest AD GWAS that did not contain the cohort used in this study.⁵⁶ We used the European reference panel from 1000 Genomes Phase 3 to model LD in the polygenic risk score computation. To improve prediction accuracy of the AD PRS, we followed the approach suggested by Leonenko et al.,³³ in which the APOE locus (Chr19 43.4M-47.5 M) is removed prior to the calculation of the AD PRS, and then the APOE effect is added as a weighted sum to the AD PRS, with effect sizes of −0.47 for the ε2 allele and 1.12 for the ε4 allele. The region for the APOE locus was chosen after comparing multiple genetic windows and the resulting AD PRS across individuals with different APOE ε4 allele counts.

Polygenic scores for other brain disorders and traits such as schizophrenia,⁵⁷ attention-deficit/hyperactivity disorder,⁵⁸ bipolar disorder,⁵⁹ insomnia,⁶⁰ migraine,⁶¹ and alcohol use disorder⁶² were also calculated in both cohorts using PRScs and the respective GWAS summary statistics.

Association of AD PRS with CSF metabolites and AD biomarkers

Linear regression was used to identify CSF metabolites that could be predicted with the AD PRS. Using the lm function from R we generated linear models predicting CSF metabolite levels using the weighted AD PRS and as cofactors age and sex. We also compared these models to using the AD PRS calculated without the APOE regions, as well as including the APOE region. This approach was extended to associate Aβ₁₋₄₂, phosphorylated tau and total tau levels with the AD PRS when available for the sample. AD biomarkers levels were log normalized for linear regression analyses.

Results

Metabolome-wide correlation of AD biomarkers

Abnormal levels of AD CSF biomarkers are indicative of broad AD neuropathologic changes.⁶³ To assess how AD pathology influences physiological processes in vivo, we examined the relationships between CSF metabolites and CSF levels of P-Tau, T-Tau, and Aβ. The study design is outlined in Figure 1. CSF samples were collected and processed from 601 individuals recruited from three previously established Alzheimer Amsterdam-based cohorts (see Methods section for descriptions of these cohorts).^35–37 Participants included individuals with normal cognition (NC), subjective cognitive decline (SCD), MCI, and AD. Table 1 summarizes the characteristics of the cohort. Untargeted metabolomic data were collected from three distinct platforms capturing metabolites across primary metabolism,⁶⁴ complex lipids,⁶⁵ and biogenic amines⁶⁶ (see Methods section for full details). After removing 282 compounds due to missingness, we analyzed the remaining 5261 CSF compounds (678 named metabolites, 4583 unnamed compounds). Of these, 271 (120 named metabolites) demonstrated significant correlations with T-Tau CSF levels after correcting for diagnosis, age, and sex, with Bonferroni correction for multiple testing (Figure 2(a)). Among them, 258 exhibited positive correlations, while 13 showed negative correlations. Similarly, 255 compounds (108 named metabolites) were correlated with P-Tau, with 241 showing positive correlations and 14 showing negative correlations (Figure 2(b)). Table 2 summarizes the strongest positive and negative correlations of named CSF metabolites for both T-Tau and P-Tau.

Figure 1.

Study design outline. Single nucleotide polymorphisms (SNPs), Alzheimer's disease (AD), cerebrospinal Fluid (CSF), amyloid-β₁₋₄₂ (Aβ), phosphorylated tau (P-Tau).

Figure 2.

CSF metabolites correlations across CSF AD biomarkers. Correlation results between 5261 CSF compounds and CSF AD biomarkers after correcting for age and sex using the residuals from linear models. (a) shows results for CSF Total Tau, (b) for CSF Phosphorylated Tau, and c for CSF Aβ. X-axis represents correlation values as an adjusted R. Y-axis represents log transformed p-values. Dots indicate positive (red) and negative (blue) significantly correlated metabolites. Dotted line represents the Bonferroni adjusted significance threshold of -log10(0.05).

Table 2.

Summary of the strongest CSF metabolite correlations for T-Tau and P-Tau CSF levels.

Metabolites Correlations
	CSF Phosphorylated Tau		CSF Total Tau
CSF Metabolite	R	Adjusted p	R	Adjusted p
PC 34:0	0.539	1.28E-32	0.523	3.16E-30
PC 34:1	0.528	5.99E-31	0.538	1.80E-32
PC 32:0	0.526	1.07E-30	0.524	2.01E-30
PC 40:4	0.493	4.30E-26	0.477	3.95E-24
PI 40:6	0.486	3.43E-25	0.462	3.44E-22
PE 36:1	0.486	3.52E-25	0.510	2.11E-28
N-Acetylneuraminic acid	0.483	6.81E-25	0.476	6.40E-24
PC 36:1	0.482	1.01E-25	0.482	9.32E-25
Nicotinoylglycine	0.473	1.48E-23	0.453	3.91E-21
PI 38:5	0.471	2.2E-23	0.467	8.29E-23
Glu-Gln	-0.247	3.38E-04	-0.250	2.40E-04
Anserine	-0.390	1.39E-12	-0.366	1.78E-12

Summary of the strongest named CSF metabolite correlations with Total Tau and Phosphorylated Tau CSF levels. The first ten rows represent significant positive correlations, while the last two rows represent significant negative correlations. A black line separates the positive and negative correlations for clarity.

A total of 242 compounds (98 named metabolites) were correlated with both P-Tau and T-Tau CSF levels, all in the same direction (Supplemental Figure 2). 13 compounds were uniquely associated with P-Tau, while 29 were uniquely associated with T-Tau (full list available in the Supplemental Material). Notably, no metabolites or compounds were significantly correlated with CSF Aβ levels (Figure 2(c)). To determine whether these associations were specific to AD, we stratified the cohort by cognitive status and repeated the correlation analysis. The relationships between CSF metabolites and CSF T-Tau/P-Tau levels remained largely consistent between cases (MCI and AD) and controls (NC and SCD) (Supplemental Figure 3). No significant differences were observed in correlation coefficients across groups for any measured CSF metabolites. Furthermore, when performing the correlation analysis without accounting for diagnosis, 208 of the significant CSF compounds associated with T-Tau and 214 with P-Tau were replicated. This suggests that most metabolite associations with CSF T-Tau and P-Tau levels are independent of AD case-control status in this cohort.

CSF metabolites can predict AD biomarkers

To further refine our findings and identify combinations of CSF compounds that could jointly predict P-Tau and T-Tau CSF levels, we employed elastic net regression models. These models utilized our CSF compounds measurements as predictors, alongside age and sex, for P-Tau and T-Tau CSF levels. To identify consistently predictive CSF compounds, we constructed these elastic net regression models over a thousand iterations of training and testing sets. In each iteration, 80% of the samples from our cohort were randomly selected for the training set, while the remaining 20% were used as the test set. Metabolites present in 80% of these models were considered consistent predictors. Base models using only age and sex as predictors were also constructed for comparison.

Models using only age and sex as predictors had an average R of 0.402 for P-Tau (Figure 3(a)) and 0.410 for T-Tau (Figure 3(c)) across the 1000 iterations. Upon adjusting for age and sex, elastic net regression models incorporating CSF compounds demonstrated an average R of 0.717 for P-Tau (Figure 3(a)) and 0.677 for T-Tau (Figure 3(c)), a significant increase compared to the average from the base model (p < 2e-16 for both P-Tau and T-Tau). For models predicting P-Tau CSF levels, 41 compounds were identified as consistent predictors, with 13 being named metabolites (Figure 3(b)). Of these, 7 were identified on the biogenic amines platform (Glu-Gln, Isoxanthopterin, Anserine, Arabitol, N-Methylvaline, 3-Methylcrotonylglycine, Alpha-Galactosamine-1-phosphate), 3 from the complex lipids platform (PI 40:6, PI 38:5, PI 36:4), and 3 from the primary metabolism platform (Isothreonic acid, Gluconic acid, Fucose). For models predicting T-Tau CSF levels, 33 compounds were identified as consistent predictors, with 8 being named metabolites (Figure 3(d)) (full list provided in the Supplemental Material). Except for the SM d34:2 metabolite, a phosphosphingolipid, all consistent named metabolites predicting T-Tau CSF levels were also highly predictive of P-Tau CSF levels. We also applied this approach to Aβ CSF levels to assess if this method could identify CSF metabolites predictive of Aβ CSF levels. Although including these CSF compounds in the elastic net models provided additional predictive power for P-Tau and T-Tau CSF levels, they did not significantly increase the average prediction accuracy of Aβ CSF levels compared to the base model using only age and sex (Supplemental Figure 4). Due to the limitations of including unknown CSF compounds, we repeated this analysis using only identified CSF metabolites (n = 678). Models based solely on this subset showed lower average predictive power compared to those using all available measurements (n = 5261) (Supplemental Figure 5), which is expected given the smaller number of variables. However, this approach successfully replicated the same set of metabolites for T-Tau CSF levels and nearly all for P-Tau levels, with the exception of PI 36:4. A summary of these results are provided in the Supplemental Material.

Figure 3.

Prediction of P-Tau and T-Tau CSF levels using CSF metabolites. Elastic net prediction results of P-Tau (a) and T-Tau (c) CSF levels using only age and sex, all the CSF compounds (n = 5261), and a combined model of both. Y-axis represents R values for the models. (b) and (d) show all named CSF metabolites that consistently came up as predictors for P-Tau (n = 13) and T-Tau (n = 8) respectively. ***p-value below 0.0001.

Pathway analysis of CSF metabolites correlated with P-tau and T-tau

Having identified metabolites and compounds that are correlated with, and provided consistent predictions for P-Tau and T-Tau CSF levels, our next step was to test which metabolic pathways are associated with these metabolites. Based on the 242 unique compounds that correlated with either P-Tau or T-Tau, we conducted an overrepresentation analysis (see methods section for details). This analysis revealed three significantly enriched pathways: the Pentose and Glucuronate Interconversions, Ascorbate and Aldarate Metabolism, and Amino Sugar and Nucleotide Sugar Metabolism (Supplemental Figure 6). To corroborate these findings, we employed PaIRKAT,⁵³ a pathway-integrated regression-based kernel association test that associates entire pathways with a phenotype. This analysis reaffirmed our results, identifying the Pentose and Glucuronate Interconversions pathway, as well as additional pathways that were significantly associated with both P-Tau (Figure 4(a)) and T-Tau (Figure 4(b)) levels. Many of the highly enriched pathways overlapped between the P-Tau and T-Tau results, including Glycerophospholipid Metabolism, ABC Transporters, Linoleic Acid and Arachidonic Acid Metabolism, Retrograde Endocannabinoid Signaling, Alpha-Linoleic Acid Metabolism, and Choline Metabolism in Cancer (full list of enriched metabolites in the Supplemental Material).

Figure 4.

Pathway analysis of metabolites correlated with P-Tau and T-Tau. Results of PaIRKAT pathway enrichment analysis. All 678 named CSF metabolites were included to create networks of metabolites representing pathways that were tested for association with either P-Tau (a), or T-Tau CSF levels (b). X-axis represents the kernel association test enrichment score; Y-axis represents log transformed p-values. Color indicates the number of metabolites that were included for that particular pathway.

Prediction of AD biomarkers and CSF metabolites using polygenic risk scores

Previously, it has been shown that AD biomarker levels in CSF (i.e., P-Tau, T-Tau, and Aβ) are associated with the polygenic risk of AD.³ To investigate similar associations in our clinical cohort, we constructed a PRS for AD based on the most recent GWAS that did not include our study cohort.⁵⁶ Due to the significant impact of the APOE locus on the PRS (Supplemental Figure 7), we used an approach that initially generates the AD PRS without the APOE locus and subsequently incorporates the effects of the APOE ε4 and ε2 alleles as a weighted sum (APOE-weighted PRS).³³ After adjusting for age and sex, linear models revealed that the APOE-weighted AD PRS was significantly correlated with Aβ CSF levels (Adjusted R² = 0.220; p = 9.11E-14) but yielded only a modest association signal with P-Tau (Adjusted R² = 0.137; p = 0.014) and T-Tau (Adjusted R² = 0.144; p = 0.007) CSF levels.

We then compared various models, incorporating different approaches to account for the effects of the APOE alleles. These approaches included using (i) only the APOE ε4 allele counts, (ii) the AD PRS without the APOE region, and (iii) a weighted risk score for APOE based on autopsy-confirmed AD cases (APOE-npScore).⁶⁷ The summarized results (Table 3) revealed that across all three biomarkers, the AD PRS without the APOE region explained the least variance (Aβ R² = 0.102, P-Tau R² = 0.129, and T-Tau R² = 0.134). Additionally, the AIC (Akaike information criterion) and BIC (Bayesian information criterion) model selection criteria supported the APOE-weighted AD PRS over all other tested models (Aβ AIC = 186.94, BIC = 206.7; P-Tau AIC = 554.45, BIC = 573.4; and T-Tau AIC = 771.05, BIC = 790.9). However, its improvement for the variance explained over solely using the APOE ε4 allele counts was marginal for P-Tau (R² = 0.137 versus 0.135) and T-Tau (R² = 0.144 versus 0.142), and equal for Aβ (R² = 0.220 versus 0.220).

Table 3.

AD biomarkers prediction results.

	CSF Amyloid-Beta						CSF Phosphorylated Tau						CSF Total Tau
Model	n	R2	AIC	BIC	Beta	p	n	R2	AIC	BIC	Beta	p	n	R2	AIC	BIC	Beta	p
ε4 allele counts	394	0.220	187.04	206.8	−0.346	9.59E-14	402	0.135	555.23	575.1	0.108	0.021	402	0.142	771.94	791.8	0.119	0.011
APOE weighted AD PRS	394	0.220	186.94	206.7	-0.347	9.11E-14	402	0.137	554.45	574.3	0.116	0.014	402	0.144	771.05	790.9	0.126	0.007
APOE-npScore	394	0.217	188.42	208.1	-0.343	1.91E-13	402	0.136	555.09	574.9	0.110	0.0197	402	0.142	771.72	791.5	0.121	9.97E-03
AD PRS without APOE region	394	0.102	242.55	262.3	-0.036	0.4500	402	0.129	557.89	577.7	0.076	0.1023	402	0.134	775.62	795.46	0.078	0.0955

Summary of results from linear regression models, after adjusting for age and sex, using different approaches to account for the APOE ε4 allele and genetic architecture of AD. R2: R-squared; AIC: Akaike information criterion; BIC: Bayesian information criterion.

While metabolomics can offer dynamic insights into an individual's physiological state and environmental exposures, multiple CSF metabolites are under genetic control.^23,24 To investigate metabolic processes associated with the genetic architecture of AD, we applied linear models to predict the levels of 5261 CSF compounds using the APOE-weighted AD PRS, adjusting for age and sex. This analysis identified four significant associations (FDR < 0.05) between phosphatidylcholines (PC) and the APOE-weighted AD PRS: PC 30:0 (R² = 0.130, p = 8.20e-07), PC 32:0 (R² = 0.135, p = 3.56e-07), PC 32:0 (5.60_792.58) (R² = 0.107, p = 7.30e-07), and PC 34:0 (R² = 0.123, p = 1.30e-09). A follow-up analysis using APOE ε4 allele counts instead of the APOE-weighted AD PRS yielded similar results, suggesting that these associations were primarily driven by the APOE ε4 allele (Supplemental Figure 8 and the Supplemental Material). To validate these results and assess whether these associations between the APOE-weighted AD PRS and CSF metabolites are specific to a cohort with an AD genetic background, we replicated our analysis on an unrelated, cognitively healthy cohort (n = 449) with the same panel of CSF compounds measurements.²³ These participants underwent spinal anesthesia for minor elective surgical procedures and were screened for self-reported psychotic or major neurological disorders (Further cohort description in the Methods section and Supplemental Table 1). Linear modeling in this cohort failed to reveal any significant associations between CSF metabolites and the APOE-weighted AD PRS, both nominally and after multiple testing corrections (Supplemental Figure 8 and the Supplemental Material). To investigate if the absence of associations was exclusive to the AD PRS, we expanded our analysis to include other brain disorders and traits such as schizophrenia,⁵⁷ attention-deficit/hyperactivity disorder (ADHD),⁵⁸ bipolar disorder,⁵⁹ Insomnia,⁶⁰ migraine,⁶¹ and alcohol use disorder.⁶² A summary of these associations can be viewed in Supplemental Figure 8. None of these polygenic scores yielded significant associations with CSF metabolites in either the clinical cohort or the cognitively healthy cohort.

The absence of associations may also be influenced by the age difference between cohorts, as the cognitively healthy cohort was younger (mean age 39.7 ± 11.4 years) than the clinical cohort (mean age 65.5 ± 9.7 years). Given that aging is a key contributing factor to AD pathology,⁶⁸ we repeated the analysis after stratifying the clinical cohort by age: younger individuals (< 60 years, N = 124) and older individuals (>= 60 years, N = 337). In this stratified analysis, only the association between PC 34:0 and the AD PRS remained significant in the older subgroup, while no significant associations were observed in the younger subgroup. The lack of associations in the younger subgroup may reflect reduced statistical power due to sample size, but it may also suggest that these associations are age-dependent. To further investigate whether these findings are specific to AD rather than general age-related effects, we stratified the older subgroup by AD diagnostic status into cases (N = 206) and controls (N = 131). No significant associations were identified in either group, likely due to limited statistical power. Full results of these analyses are provided in the Supplemental Material.

Discussion

By leveraging genotype data, an extensive panel of 5261 CSF metabolites, and AD CSF biomarkers in a large cohort, in this study we aimed to elucidate biological pathways impacted by the neuropathologic changes associated with AD. Our correlation and elastic net regression analyses revealed 242 unique CSF compounds linked with P-Tau and T-Tau CSF levels, yet no metabolites demonstrated associations with Aβ CSF levels. Among the CSF metabolites predictive of both P-Tau and T-Tau levels found in this study, we identified Anserine and Fucose as novel associations. The pathway enrichment analysis of these CSF metabolites associated with P-Tau and T-Tau levels consistently highlighted pentose and glucuronate interconversions, as well as glycerophospholipid metabolism as significant pathways for both AD biomarkers. Furthermore, we have also identified 3 unique CSF metabolites associated with PRS of AD adjusted for the substantial effect size of APOE alleles, all of which were phosphatidylcholines. Linear regression models did not replicate these associations between CSF metabolites levels and the APOE-weighted AD PRS in our cognitively healthy cohort, nor further associations were identified for additional polygenic scores related to various brain traits and disorders. With this AD PRS, we also observed a significant prediction for Aβ CSF levels but with only limited and modest predictive value to either P-Tau and T-Tau CSF levels. Notably, the AD PRS without the APOE locus yielded the lowest-performing predictions across all three biomarkers, thereby reiterating the known importance of the APOE locus as an AD risk factor.

After assessing the strength and directionality of the associations between CSF compounds and AD CSF biomarkers, we identified 204 CSF compounds that significantly correlated with P-Tau CSF levels and also with T-Tau CSF levels, reflecting the high correlation between these two biomarkers.⁶⁹ This number represents a significant increase compared to previous studies, which identified 38 CSF metabolites associated with both P-Tau and T-Tau CSF levels.^15,17 While these prior studies investigated CSF metabolites in relation to AD biomarkers, differences in metabolite panels present a challenge for direct comparisons across cohorts. Nonetheless, we independently replicated 14 of these previously reported associations and confirmed that metabolites such as Cholesterol, N-acetylserine, and Arabitol are associated with these AD biomarkers (see the Spplemental Material for the full comparison). Previous metabolomic studies on plasma have also identified associations between CSF P-Tau, T-tau and metabolites, such as glutamine,⁷⁰ kynurenine and picolinic acid,⁷¹ and sphingolipids.⁷² However, the presence of the blood-brain barrier creates a challenge for direct comparisons between the biofluids,¹⁶ as other studies have shown different degrees of correlations between metabolite levels.^13–15 Interestingly, we found that most of the significant correlations between CSF metabolites and P-Tau/T-Tau CSF levels were independent of diagnosis in this cohort. This lack of a diagnosis effect may reflect how these biomarkers influence cognition. While CSF AD biomarkers capture the pathological changes in the brain associated with AD, their relationship with different cognitive phenotypes remains under investigation.^69,73 A previous study highlighted that in MCI individuals, CSF levels of Aβ and P-Tau differentially predicted cognitive functions such as learning, working memory, and processing speed.⁷⁴ As diagnosis categories primarily reflect broad cognitive impairments rather than specific cognitive dimensions, future studies assessing the relationship between CSF Tau, P-Tau, and associated metabolites with distinct cognitive functions could provide insights into biological pathways contributing to specific cognitive deficits in AD.

Among the CSF metabolite associations identified in our study using both correlation analysis and elastic net regression, we also identified two novel associations: Anserine and Fucose. Anserine is the methylated analogue of carnosine, an endogenous dipeptide present in various mammalian tissues, including the brain.⁷⁵ Due to its antioxidant and anti-inflammatory functions, carnosine could potentially modulate biological pathways affected by AD.⁷⁶ Supplementation with anserine and carnosine has demonstrated improvements in AD symptoms in both AD mouse models⁷⁷ and elderly healthy human subjects,⁷⁸ particularly in alleviating memory deficits. The observed strong negative correlation between anserine and P-Tau/T-Tau aligns with a potential protective mechanism, though further studies are essential to elucidate the interactions of anserine with these AD biomarkers. The other newly identified metabolite, Fucose, is a sugar component of glycolipids and glycoproteins.⁷⁹ Fucose's incorporation into glycolipids has been linked to effects on learning, long-term potentiation, and synapse formation in animal models.^80,81 Direct delivery of fucose into the brain improved retention of learned behavior in rats⁸² and mitigated neuroinflammation by inhibiting microglial cell activity in mice.⁸³ While these functions require validation in humans, a transcriptomic study of brain samples from an AD cohort identified the upregulation of the FUT8 gene in AD subjects.⁸⁴ The protein product of FUT8 catalyzes fucose transfer to glycolipids,⁸⁵ highlighting the potential role of fucosylation in AD.

The pathway enrichment analysis of the CSF metabolites associated with P-Tau and T-Tau levels consistently highlighted glycerophospholipid metabolism as a significant pathway for both AD biomarkers. Glycerophospholipids, major components of cellular membranes, play a diverse role in altering the functional properties of cells, including signal transduction, vesicle trafficking, and membrane fluidity.⁸⁶ In the brain, phospholipases actively catabolize glycerophospholipids through hydrolysis.⁸⁷ Abnormal lipid metabolism, particularly changes in glycerophospholipids, has been linked to AD pathogenic features such as amyloidogenesis, oxidative stress, and neuroinflammation.⁸⁸ A postmortem lipidomics study on brain tissue identified glycerophospholipids among the lipid subclasses significantly perturbed in AD cases.⁸⁹ Among the metabolites identified by elastic net regression, three glycerophosphoinositols (PI 40:6, PI 38:5, PI 36:4) exhibited the highest positive correlations with P-Tau and T-Tau. These metabolites are derived from membrane phosphoinositides via the enzymatic activity of phospholipase A2 and lysophospholipase.⁹⁰ While abnormal phospholipase A2 activity has been proposed as a predictor of cardiovascular disease, its role in dementias such as AD remains unclear.⁹¹ The enrichment of CSF metabolites in glycerophospholipid metabolism observed in our study underscores the relevance of this biological pathway in AD pathology. Another enriched pathway was pentose and glucuronate interconversions, previously associated with metabolites linked to P-Tau/T-Tau in an independent but much smaller AD cohort.¹⁷ Although dysregulation of glucose and pentose brain levels has been associated with AD pathology,^92,93 the specific interactions of these pathways and associated metabolites with Tau in AD remain to be elucidated. The enrichment of these pathways in metabolites strongly correlated with P-Tau/T-Tau suggests a potential association between elevated levels of these AD biomarkers and processes specifically impacting neuronal activity and viability. A prior study, integrating proteomic, genomic, and imaging data, demonstrated that distinct proteomic profiles and associated pathways were linked to varying levels of P-Tau/T-Tau within individuals with AD.⁹⁴ Further exploration of the data generated in this study offers an opportunity to assess whether different metabolites and pathways are associated with diverse levels of these AD biomarkers.

The absence of associations between CSF metabolites and Aβ levels may be attributed to the early accumulation of Aβ in the brain.¹ Although Aβ is an established biomarker for AD, its accumulation occurs early in the disease process and is not strongly predictive of disease progression, unlike P-Tau and T-Tau CSF levels.^95,96 These temporal characteristics suggest that the metabolic impact of Aβ may be more detectable in younger individuals, prior to the onset of AD. Future metabolomic studies in younger or preclinical cohorts could capture the early metabolic alterations related to Aβ. Consistent with previous research using smaller cohorts, our larger study also found no significant associations between CSF metabolites and Aβ levels.^15,17 Despite significantly expanding the metabolite panel by nearly tenfold compared to prior work, it remains possible that specific Aβ metabolic signals were missed. The effects of Aβ on the CSF metabolome may be subtle, requiring even larger cohorts or more sensitive detection methods to identify. Larger, and longitudinal metabolomic cohorts that can capture disease severity progression at an individual level, as well as improvements in metabolite profiling technologies, may hold promise for elucidating the intricate relationship between Aβ accumulation and CSF metabolites.

In our analysis of the genetic architecture of AD and AD CSF biomarkers, the ε4 allele counts alone significantly predicted P-Tau, T-Tau and Aβ CSF levels, corroborating findings from previous studies.^67,97 The APOE gene, encoding a glycoprotein lipid transporter, exhibits isoform-specific impacts on its function.⁹⁸ These results align with the proposed relationship between Aβ and the APOE locus, where Aβ plaque levels are influenced by different APOE alleles.⁹⁹ Regarding Aβ, it is hypothesized that the ε4 allele may influence Aβ levels by promoting plaque formation¹⁰⁰ and hindering its clearance from the brain.¹⁰¹ APOE alleles have also been related to differences in Tau neurofibrillary tangles CSF levels,¹⁰² although the precise association between APOE and Tau is still under investigation. Our findings align with these known interactions and support prior reports identifying the APOE locus as a major genetic risk factor for Aβ, T-Tau, and P-Tau CSF levels.^3,4,32 While this study directly compares the AD PRS with AD CSF biomarkers, alternative multi-omics approaches can provide valuable insights. For example, a previous study demonstrated the benefits of integrating metabolomics and genetic data, showing that a model combining an AD polygenic risk score with a metabolomic score derived from plasma metabolites yielded the highest hazard ratio for AD.¹⁰³

Metabolite levels can be heritable traits, as demonstrated by prior QTL mapping studies.^{20,22,25,26,104} For instance, polygenic scores of trait measures, such as body mass index, have been associated with metabolite levels.³⁴ Our analysis revealed four significant associations between CSF metabolite levels and the APOE-weighted AD PRS, with no additional associations observed for other polygenic scores related to brain traits or disorders. Notably, the identified metabolites were phosphatidylcholines (PC 30:0, PC 32:0, and PC 34:0), which are among the most abundant phospholipids in mammalian cell membranes and play key roles in lipoprotein transport, cell signaling, and energy metabolism.¹⁰⁵ Phosphatidylcholines have also been implicated in neuroplasticity, either through direct activation of intracellular neuronal signaling pathways or by modulating membrane function.¹⁰⁶ Altered phosphatidylcholine metabolism may represent a component of the broader dysregulation of lipid metabolism implicated in AD pathophysiology. Plasma lipidomics studies have reported decreased phosphatidylcholine levels in AD,^107,108 and CSF lipidomics studies have identified significant differences in specific phosphatidylcholine species between AD and controls,¹⁰⁷ including PC 32:0.¹⁰⁹ Further analysis revealed that APOE ε4 allele counts were the primary driver for these associations. Modifications of lipid metabolism have been linked to the different isoforms of the APOE,¹⁰⁶ as these are known to impact its function.⁹⁸ For example, the APOE ε4 allele has been linked to cholesterol and triglyceride accumulation in glial cells.¹¹⁰ Although the direct interaction between APOE isoforms and phosphatidylcholines remains unclear, a biochemical study suggested that phosphatidylcholine clearance rates differ among APOE isoforms.¹¹¹ While there is evidence linking phosphatidylcholines to AD and APOE genotype, we were unable to replicate our findings in an independent, cognitively healthy cohort. A likely explanation is the demographic differences between cohorts, as the cognitively healthy cohort was younger (mean age: 39.7 ± 11.4 vs. 65.5 ± 9.7 years), predominantly male (27.6% female vs. 42.5%), and had fewer APOE ε4 allele carriers (33.9% vs. 51.7%). To explore the impact of age, we stratified the clinical cohort into younger (< 60 years) and older (>= 60 years) subgroups. This analysis showed that only the association between PC 34:0 and the AD PRS remained significant in the older group, with no significant associations found in the younger group. Further stratification of the older individuals into AD cases and controls to disentangle general age-related effects from those specific to AD did not yield any significant associations, likely due to limited statistical power. Although the current dataset could not be applied to distinguish between age-related and AD specific effects, these results highlight the impact of age on our associations. In addition to age, both sex and APOE genotype are known to influence lipid metabolism, including phosphatidylcholine levels.^110,112,113 A study of 1517 individuals further demonstrated that phosphatidylcholine serum levels are significantly associated with both sex and APOE genotype.¹¹⁴ Although the metabolite panels were comparable between cohorts, these demographic disparities likely contributed to the differences in findings. We encourage future studies to consider these demographic factors when attempting to replicate metabolomic associations in independent cohorts.

Our correlation analyses showed both positive and negative relationships between CSF metabolites and AD biomarkers, while the elastic net regression and pathway enrichment analyses added further evidence for their biological relevance. However, determining whether these metabolic patterns reflect downstream responses to these AD biomarkers will require comprehensive investigations into the broader pathways in which these metabolites are part of. Furthermore, while our multi-level analyses provide converging evidence supporting the relevance of these metabolites in relation to both P-Tau and T-Tau, as well as to AD pathology more broadly, our study does not directly assess the mechanistic interactions between these metabolites and AD. To establish causal links and functional relevance, future research should incorporate molecular and cellular studies aimed at dissecting how these metabolic changes are triggered or influenced by neurodegenerative processes that are part of AD. Another primary limitation of this study is the current inability to validate the strength of our predictors for P-Tau and T-Tau on an independent cohort. Although we attempted to mitigate this limitation by generating 1000 predictive models and summarizing the results, our findings still represent an overfit to this particular dataset. Furthermore, due to the lack of overlap in the panels of metabolites used, direct comparisons across cohorts used in different publications are particularly challenging. Standardizing the panels used in metabolomic studies and applying a meta-analysis approach could improve the detection of biological signals associated with AD. The metabolome encompasses chemicals from core nutrient metabolism, lipids, diet-derived compounds, and pharmaceuticals,¹¹⁵ capturing multiple exposures that influence an individual's metabolic profile. In this study, we adjusted for age and sex—two major determinants of metabolic variation—but were unable to account for polypharmacy. Polypharmacy, defined as the concurrent use of five or more medications,¹¹⁶ affects up to two-thirds of older adults worldwide.¹¹⁷ In the context of AD, it often includes cognitive-enhancing drugs (e.g., acetylcholinesterase inhibitors, NMDA receptor antagonists) alongside psychotropic medications such as antidepressants, antiepileptics, antipsychotics, opioids, and benzodiazepine receptor agonists.¹¹⁸ The effects of polypharmacy on both the metabolome and AD pathology remain poorly understood, as research has primarily been limited to animal models,^119,120 and sex-specific consequences in humans are still under investigation.¹²¹ Future studies with emerging precision medicine approaches that integrate metabolomics and detailed pharmacological profiles¹²² may offer new insights into the interplay between polypharmacy and AD. Another limitation is that the cohort utilized in this study comprises exclusively individuals of European ancestry, restricting the generalizability of the results to other ancestry groups. Both the genetic architecture of AD and the performance of AD biomarkers exhibit variations across populations of different ancestries, such as in the case of African American populations. Although the number and size of GWAS studies on African American individuals with AD are limited,¹²³ evidence suggests that, besides the APOE locus, the ABCA7 locus shows strong associations with AD in this population.¹²⁴ Similarly, the performance of T-Tau and P-Tau as biomarkers for AD is inconsistent in African American populations.¹²⁵ One study demonstrated significantly lower levels of these metabolites in African Americans compared to non-Hispanic white individuals.¹²⁶ Future AD studies should aim to broaden their cohorts to include diverse populations, enhancing our understanding of the disease across different ancestry groups.

In this study, our aim was to assess how distinct biological layers, including genetic risk and metabolomic profiles, contribute to capturing various aspects of AD pathology associated with the established clinical biomarkers of disease. We successfully identified and replicated findings regarding the impact of the APOE ε4 allele on Aβ, P-Tau, and T-Tau CSF levels. Additionally, we compared the effectiveness of different approaches in modeling the genetic architecture of AD. We have also extended prior CSF metabolite investigations by substantially increasing the cohort size as well as by significantly expanding the scope of the metabolomic measurements (with more than five thousand compounds). This increased effort resulted in the identification of 242 unique CSF compounds associated with CSF levels of P-Tau and T-Tau, but also highlighted the lack of any significant findings of an association between CSF metabolites and Aβ. We identified novel associations between the CSF metabolites Anserine and Fucose, and P-Tau/T-Tau. Pathway analysis of these metabolites further supports their involvement in established biological pathways affected by AD. This makes these metabolites potential targets for studying and gaining a deeper understanding of how AD progression impacts brain physiology.

Supplemental Material

sj-docx-1-alz-10.1177_13872877251389924 - Supplemental material for Integrative analysis of cerebrospinal fluid biomarkers, metabolomics, and polygenic risk reveals novel metabolite associations with Alzheimer's disease

Supplemental material, sj-docx-1-alz-10.1177_13872877251389924 for Integrative analysis of cerebrospinal fluid biomarkers, metabolomics, and polygenic risk reveals novel metabolite associations with Alzheimer's disease by Marcelo Francia, Naren Ramesh, Toni Boltz, Merel Bot, Wiesje M van der Flier, Pieter Jelle Visser, Sven van der Lee, Charlotte E Teunissen, Yolande AL Pijnenburg, Anouk den Braber, Loes Olde Loohuis, Lianne M Reus, Betty M Tijms, and Roel A Ophoff in Journal of Alzheimer's Disease

Footnotes

Acknowledgements

We are greatly appreciative to those individuals who donated the CSF samples on which this study was based. We thank Jurjen Luykx for his help with the data recruitment of the cognitively healthy controls from Utrecht and UCLA.

Research of Alzheimer Center Amsterdam is part of the neurodegeneration research program of Amsterdam Neuroscience. Alzheimer Center Amsterdam is supported by Stichting Alzheimer Nederland and Stichting VUmc Steun Alzheimercentrum Amsterdam. The clinical database structure was developed with funding from Stichting Dioraphte.

Ethical considerations

All participating studies were approved by their respective Medical Ethics Committees, including approval from the METC Amsterdam UMC and UCLA Institutional Review Board (IRB#20-002064). All methods were performed in accordance with the relevant guidelines and regulations.

Consent to participate

Informed consent, either from the patient or from the legal representative, was obtained from all participants.

Author contribution(s)

Marcelo Francia: Conceptualization; Data curation; Formal analysis; Writing – original draft; Writing – review & editing.

Naren Ramesh: Formal analysis; Writing – original draft; Writing – review & editing.

Toni Boltz: Data curation; Formal analysis; Investigation; Methodology; Writing – original draft; Writing – review & editing.

Merel Bot: Data curation; Investigation; Resources; Writing – review & editing.

Wiesje M van der Flier: Resources; Writing – review & editing.

Pieter Jelle Visser: Resources; Writing – review & editing.

Sven van der Lee: Resources; Writing – review & editing.

Charlotte E Teunissen: Resources; Writing – review & editing.

Yolande AL Pijnenburg: Resources; Writing – review & editing.

Anouk den Braber: Resources; Writing – review & editing.

Loes Olde Loohuis: Conceptualization; Supervision; Writing – review & editing.

Lianne M Reus: Conceptualization; Data curation; Methodology; Supervision; Writing – review & editing.

Betty M Tijms: Conceptualization; Investigation; Project administration; Writing – review & editing.

Roel A Ophoff: Conceptualization; Data curation; Funding acquisition; Investigation; Project administration; Supervision; Writing – review & editing.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was funded by the National Institute of Mental Health, grant number MH115676, the National Institute of Neurological Disorders and Stroke, grant number T32NS048004, and the National Institute on Aging, grant numbers R21AG072390 & RF1AG058484.

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data availability statement

The data supporting the findings of this study are available within the article and its supplemental material.

Supplemental material

Supplemental material for this article is available online.

References

Jack Jr

Knopman

Jagust

, et al. Tracking pathophysiological processes in Alzheimer’s disease: an updated hypothetical model of dynamic biomarkers. Lancet Neurol 2013; 12: 207–216.

Bouwman

Frisoni

Johnson

, et al. Clinical application of CSF biomarkers for Alzheimer’s disease: from rationale to ratios. Alzheimers Dement 2022; 14: e12314.

Deming

Kapoor

, et al. Genome-wide association study identifies four novel loci associated with Alzheimer’s endophenotypes and disease modifiers. Acta Neuropathol 2017; 133: 839–856.

Jansen

van der Lee

Gomez-Fonseca

, et al. Genome-wide meta-analysis for Alzheimer’s disease cerebrospinal fluid biomarkers. Acta Neuropathol 2022; 144: 821–842.

Varma

, et al. Evidence for brain glucose dysregulation in Alzheimer’s disease. Alzheimers Dement 2018; 14: 318–329.

Kao

Y-C

P-C

Y-K

, et al. Lipids and Alzheimer’s disease. Int J Mol Sci 2020; 21: 1505.

Clish

. Metabolomics: an emerging but powerful tool for precision medicine. Cold Spring Harb Mol Case Stud 2015; 1: a000588.

de Leeuw

Peeters

CFW

Kester

, et al. Blood-based metabolic signatures in Alzheimer’s disease. Alzheimers Dement 2017; 8: 196–207.

Trushina

Dutta

Persson

X-MT

, et al. Identification of altered metabolic pathways in plasma and CSF in mild cognitive impairment and Alzheimer’s disease using metabolomics. PLoS One 2013; 8: e63644.

10.

Dumas

M-E

Davidovic

. Metabolic profiling and phenotyping of central nervous system diseases: metabolites bring insights into brain dysfunctions. J Neuroimmune Pharmacol 2015; 10: 402–424.

11.

Mapstone

Cheema

Fiandaca

, et al. Plasma phospholipids identify antecedent memory impairment in older adults. Nat Med 2014; 20: 415–418.

12.

Qiang

Y-X

You

X-Y

, et al. Plasma metabolic profiles predict future dementia and dementia subtypes: a prospective analysis of 274,160 participants. Alzheimers Res Ther 2024; 16: 16.

13.

Tikhonova

Zhanaeva

Shvaikovskaya

, et al. Neurospecific molecules measured in periphery in humans: how do they correlate with the brain levels? A systematic review. Int J Mol Sci 2022; 23: 9193.

14.

Schön

Naviaux

, et al. Cerebrospinal fluid and plasma metabolomics of acute endurance exercise. FASEB J 2022; 36: e22408.

15.

Darst

Johnson

, et al. Integrated analysis of genomics, longitudinal metabolomics, and Alzheimer’s risk factors among 1,111 cohort participants. Genet Epidemiol 2019; 43: 657–674.

16.

Tumani

Huss

Bachhuber

. The cerebrospinal fluid and barriers - anatomic and physiologic considerations. Handb Clin Neurol 2017; 146: 21–32.

17.

Dong

Darst

Deming

, et al. CSF Metabolites associate with CSF tau and improve prediction of Alzheimer’s disease status. Alzheimers Dement 2021; 13: e12167.

18.

Panyard

McKetney

Deming

, et al. Large-scale proteome and metabolome analysis of CSF implicates altered glucose and carbon metabolism and succinylcarnitine in Alzheimer’s disease. Alzheimers Dement 2023; 19: 5447–5470.

19.

Kremer

Taylor

Siebourg-Polster

, et al. Longitudinal analysis of multiple neurotransmitter metabolites in cerebrospinal fluid in early Parkinson’s disease. Mov Disord 2021; 36: 1972–1978.

20.

Lotta

Pietzner

Stewart

, et al. A cross-platform approach identifies genetic regulators of human metabolism and health. Nat Genet 2021; 53: 54–64.

21.

Long

Hicks

H-C

, et al. Whole-genome sequencing identifies common-to-rare variants associated with human blood metabolites. Nat Genet 2017; 49: 568–578.

22.

Chen

Pettersson-Kymmer

, et al. Genomic atlas of the plasma metabolome prioritizes metabolites implicated in human diseases. Nat Genet 2023; 55: 44–53.

23.

Luykx

Bakker

Lentjes

, et al. Genome-wide association study of monoamine metabolite levels in human cerebrospinal fluid. Mol Psychiatry 2014; 19: 228–234.

24.

Panyard

Kim

Darst

, et al. Cerebrospinal fluid metabolomics identifies 19 brain-related phenotype associations. Commun Biol 2021; 4: 63.

25.

Reus

Boltz

Francia

, et al. Quantitative trait loci mapping of circulating metabolites in cerebrospinal fluid to uncover biological mechanisms involved in brain-related phenotypes. Mol Psychiatry 2025; 30: 3478–3490.

26.

Wang

Yang

Western

, et al. Genetic architecture of cerebrospinal fluid and brain metabolite levels and the genetic colocalization of metabolites with human traits. Nat Genet 2024; 56: 2685–2695.

27.

Corder

Saunders

Strittmatter

, et al. Gene dose of apolipoprotein E type 4 allele and the risk of Alzheimer’s disease in late onset families. Science 1993; 261: 921–923.

28.

Berkowitz

Mosconi

Rahman

, et al. Clinical application of APOE in Alzheimer’s prevention: a precision medicine approach. J Prev Alzheimers Dis 2018; 5: 245–252.

29.

Bellenguez

Küçükali

Jansen

, et al. New insights into the genetic etiology of Alzheimer’s disease and related dementias. Nat Genet 2022; 54: 412–436.

30.

Baker

Leonenko

Schmidt

, et al.

What does heritability of Alzheimer’s disease represent?

PLoS One 2023; 18: e0281440.

31.

Ridge

Hoyt

Boehme

, et al. Assessment of the genetic variance of late-onset Alzheimer’s disease. Neurobiol Aging 2016; 41: 200.e13–200.e20.

32.

Cruchaga

Kauwe

JSK

Harari

, et al. GWAS Of cerebrospinal fluid tau levels identifies risk variants for Alzheimer’s disease. Neuron 2013; 78: 256–268.

33.

Leonenko

Baker

Stevenson-Hoare

, et al. Identifying individuals with high risk of Alzheimer’s disease using polygenic risk scores. Nat Commun 2021; 12: 4506.

34.

Fang

Holmes

Gaunt

, et al. Constructing an atlas of associations between polygenic scores from across the human phenome and circulating metabolic biomarkers. Elife 2022; 11: e73951.

35.

van der Flier

Scheltens

. Amsterdam Dementia Cohort: performing research to optimize care. J Alzheimers Dis 2018; 62: 1091–1111.

36.

Legdeur

Badissi

Carter

, et al. Resilience to cognitive impairment in the oldest-old: design of the EMIF-AD 90+study. BMC Geriatr 2018; 18: 289.

37.

Konijnenberg

Carter

Ten Kate

, et al. The EMIF-AD PreclinAD study: study design and baseline cohort overview. Alzheimers Res Ther 2018; 10: 75.

38.

Albert

DeKosky

Dickson

, et al. The diagnosis of mild cognitive impairment due to Alzheimer’s disease: recommendations from the National Institute on Aging-Alzheimer's Association workgroups on diagnostic guidelines for Alzheimer's disease. Alzheimers Dement 2011; 7: 270–279.

39.

McKeith

Boeve

Dickson

, et al. Diagnosis and management of dementia with Lewy bodies: fourth consensus report of the DLB Consortium. Neurology 2017; 89: 88–100.

40.

McKhann

Knopman

Chertkow

, et al. The diagnosis of dementia due to Alzheimer’s disease: recommendations from the National Institute on Aging-Alzheimer's Association workgroups on diagnostic guidelines for Alzheimer's disease. Alzheimers Dement 2011; 7: 263–269.

41.

Rascovsky

Hodges

Knopman

, et al. Sensitivity of revised diagnostic criteria for the behavioural variant of frontotemporal dementia. Brain 2011; 134: 2456–2477.

42.

Boomsma

de Geus

EJC

Vink

, et al. Netherlands Twin Register: from twins to twin families. Twin Res Hum Genet 2006; 9: 849–857.

43.

Folstein

McHugh

. Mini-mental state’. A practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res 1975; 12: 189–198.

44.

Tijms

Willemse

EAJ

Zwan

, et al. Unbiased approach to counteract upward drift in cerebrospinal fluid amyloid-β 1-42 analysis results. Clin Chem 2018; 64: 576–585.

45.

Matyash

Liebisch

Kurzchalia

, et al. Lipid extraction by methyl-tert-butyl ether for high-throughput lipidomics. J Lipid Res 2008; 49: 1137–1146.

46.

Fiehn

Wohlgemuth

Scholz

, et al. Quality control for plant metabolomics: reporting MSI-compliant studies. Plant J 2008; 53: 691–704.

47.

Tsugawa

Cajka

Kind

, et al. MS-DIAL: data-independent MS/MS deconvolution for comprehensive metabolome analysis. Nat Methods 2015; 12: 523–526.

48.

DeFelice

Mehta

Samra

, et al. Mass spectral feature list optimizer (MS-FLO): a tool to minimize false positive peak reports in untargeted liquid chromatography-mass spectroscopy (LC-MS) data processing. Anal Chem 2017; 89: 3250–3255.

49.

Fan

Kind

Cajka

, et al. Systematic error removal using random forest for normalizing large-scale untargeted lipidomics data. Anal Chem 2019; 91: 3590–3596.

50.

Yang

Stein

. Quality control for building libraries from electrospray ionization tandem mass spectra. Anal Chem 2014; 86: 6393–6400.

51.

Purcell

Neale

Todd-Brown

, et al. PLINK: a tool set for whole-genome association and population-based linkage analyses. Am J Hum Genet 2007; 81: 559–575.

52.

Kuhn

. Building predictive models in R using the caret package. J Stat Softw 2008; 28: 1–26.

53.

Carpenter

Zhang

Gillenwater

, et al. PaIRKAT: a pathway integrated regression-based kernel association test with applications to metabolomics and COPD phenotypes. PLoS Comput Biol 2021; 17: e1008986.

54.

Pang

Chong

Zhou

, et al. Metaboanalyst 5.0: narrowing the gap between raw spectra and functional insights. Nucleic Acids Res 2021; 49: W388–W396.

55.

Chen

C-Y

, et al. Polygenic prediction via Bayesian regression and continuous shrinkage priors. Nat Commun 2019; 10: 1776.

56.

Wightman

Jansen

Savage

, et al. A genome-wide association study with 1,126,563 individuals identifies new risk loci for Alzheimer’s disease. Nat Genet 2021; 53: 1276–1282.

57.

Trubetskoy

Pardiñas

, et al. Mapping genomic loci implicates genes and synaptic biology in schizophrenia. Nature 2022; 604: 502–508.

58.

Demontis

Walters

Martin

, et al. Discovery of the first genome-wide significant risk loci for attention deficit/hyperactivity disorder. Nat Genet 2019; 51: 63–75.

59.

Mullins

Forstner

O’Connell

, et al. Genome-wide association study of more than 40,000 bipolar disorder cases provides new insights into the underlying biology. Nat Genet 2021; 53: 817–829.

60.

Watanabe

Jansen

Savage

, et al. Genome-wide meta-analysis of insomnia prioritizes genes associated with metabolic and psychiatric pathways. Nat Genet 2022; 54: 1125–1132.

61.

Hautakangas

Winsvold

Ruotsalainen

, et al. Genome-wide analysis of 102,084 migraine cases identifies 123 risk loci and subtype-specific risk alleles. Nat Genet 2022; 54: 152–160.

62.

Sanchez-Roige

Palmer

Fontanillas

, et al. Genome-wide association study meta-analysis of the Alcohol Use Disorders Identification Test (AUDIT) in two population-based cohorts. Am J Psychiatry 2019; 176: 107–118.

63.

Jack Jr

Andrews

Beach

, et al. Revised criteria for diagnosis and staging of Alzheimer’s disease: Alzheimer's Association Workgroup. Alzheimers Dement 2024; 20: 5143–5169.

64.

S Haglund

Löfstrand

Siek

, et al. Powerful GC-TOF-MS techniques for screening, identification and quantification of halogenated natural products. Mass Spectrom 2013; 2: S0018.

65.

Dhariwal

Maan

Baghel

, et al. Systematic untargeted UHPLC-Q-TOF-MS based lipidomics workflow for improved detection and annotation of lipid sub-classes in serum. Metabolomics 2023; 19: 24.

66.

Paglia

Astarita

. A high-throughput HILIC-MS-based metabolomic assay for the analysis of polar metabolites. Methods Mol Biol 2022; 2396: 137–159.

67.

Deming

Vasiljevic

Morrow

, et al. Neuropathology-based APOE genetic risk score better quantifies Alzheimer’s risk. Alzheimers Dement 2023; 19: 3406–3416.

68.

DeTure

Dickson

. The neuropathological diagnosis of Alzheimer’s disease. Mol Neurodegener 2019; 14: 32.

69.

Wattmo

Blennow

Hansson

. Cerebro-spinal fluid biomarker levels: phosphorylated tau (T) and total tau (N) as markers for rate of progression in Alzheimer’s disease. BMC Neurol 2020; 20: 10.

70.

Niedzwiecki

Walker

Howell

, et al. High-resolution metabolomic profiling of Alzheimer’s disease in plasma. Ann Clin Transl Neurol 2020; 7: 36–45.

71.

Jacobs

Lim

Blennow

, et al. Correlation between plasma and CSF concentrations of kynurenine pathway metabolites in Alzheimer’s disease and relationship to amyloid-β and tau. Neurobiol Aging 2019; 80: 11–20.

72.

Chatterjee

Lim

WLF

Shui

, et al. Plasma phospholipid and sphingolipid alterations in Presenilin1 mutation carriers: a pilot study. J Alzheimers Dis 2016; 50: 887–894.

73.

Simrén

Brum

Ashton

, et al. CSF Tau368/total-tau ratio reflects cognitive performance and neocortical tau better compared to p-tau181 and p-tau217 in cognitively impaired individuals. Alzheimers Res Ther 2022; 14: 192.

74.

Malpas

Saling

Velakoulis

, et al. Tau and amyloid-β cerebrospinal fluid biomarkers have differential relationships with cognition in mild cognitive impairment. J Alzheimers Dis 2015; 47: 965–975.

75.

Hipkiss

Preston

Himsworth

, et al. Pluripotent protective effects of carnosine, a naturally occurring dipeptide. Ann N Y Acad Sci 1998; 854: 37–53.

76.

Caruso

Caraci

Jolivet

. Pivotal role of carnosine in the modulation of brain cells activity: multimodal mechanism of action and therapeutic potential in neurodegenerative disorders. Prog Neurobiol 2019; 175: 35–53.

77.

Kaneko

Enya

Enomoto

, et al. Anserine (beta-alanyl-3-methyl-L-histidine) improves neurovascular-unit dysfunction and spatial memory in aged AβPPswe/PSEN1dE9 Alzheimer’s-model mice. Sci Rep 2017; 7: 12571.

78.

Ding

Tanigawa

Kaneko

, et al. Anserine/carnosine supplementation preserves blood flow in the prefrontal brain of elderly people carrying APOE e4. Aging Dis 2018; 9: 334–345.

79.

Schneider

Al-Shareffi

Haltiwanger

. Biological functions of fucose in mammals. Glycobiology 2017; 27: 601–618.

80.

Pohle

Acosta

Rüthrich

, et al. Incorporation of [3H]fucose in rat hippocampal structures after conditioning by perforant path stimulation and after LTP-producing tetanization. Brain Res 1987; 410: 245–256.

81.

Kalovidouris

Gama

Lee

, et al. A role for fucose alpha(1-2) galactose carbohydrates in neuronal growth. J Am Chem Soc 2005; 127: 1340–1341.

82.

Wetzel

Popov

Lössner

, et al. Effect of L-fucose on brain protein metabolism and retention of a learned behavior in rats. Pharmacol Biochem Behav 1980; 13: 765–771.

83.

Fukuda

Takai

, et al. Exogenous l-fucose attenuates neuroinflammation induced by lipopolysaccharide. J Biol Chem 2024; 300: 105513.

84.

Tang

Tena

Di Lucente

, et al. Transcriptomic and glycomic analyses highlight pathway-specific glycosylation alterations unique to Alzheimer’s disease. Sci Rep 2023; 13: 7816.

85.

C-F

F-A

L-H

, et al. Quantitative glycoproteomics analysis identifies novel FUT8 targets and signaling networks critical for breast cancer cell invasiveness. Breast Cancer Res 2022; 24: 21.

86.

Hishikawa

Hashidate

Shimizu

, et al. Diversity and function of membrane glycerophospholipids generated by the remodeling pathway in mammalian cells. J Lipid Res 2014; 55: 799–807.

87.

Farooqui

Horrocks

Farooqui

. Glycerophospholipids in brain: their metabolism, incorporation into membranes, functions, and involvement in neurological disorders. Chem Phys Lipids 2000; 106: 1–29.

88.

Yin

. Lipid metabolism and Alzheimer’s disease: clinical evidence, mechanistic link and therapeutic promise. FEBS J 2023; 290: 1420–1453.

89.

Akyol

Ugur

Yilmaz

, et al. Lipid profiling of Alzheimer’s disease brain highlights enrichment in glycerol(phospho)lipid, and sphingolipid metabolism. Cells 2021; 10: 2591.

90.

Patrussi

Mariggiò

Corda

, et al. The glycerophosphoinositols: from lipid metabolites to modulators of T-cell signaling. Front Immunol 2013; 4: 213.

91.

Zuliani

Marsillach

Trentini

, et al. Lipoprotein-associated phospholipase A2 activity as potential biomarker of vascular dementia. Antioxidants (Basel) 2023; 12: 597.

92.

Begley

Church

, et al. Graded perturbations of metabolism in multiple regions of human brain in Alzheimer’s disease: snapshot of a pervasive metabolic disorder. Biochim Biophys Acta 2016; 1862: 1084–1092.

93.

Burns

Chen

Kaszniak

, et al. Higher serum glucose levels are associated with cerebral hypometabolism in Alzheimer regions. Neurology 2013; 80: 1557–1564.

94.

Visser

Reus

Gobom

, et al. Cerebrospinal fluid tau levels are associated with abnormal neuronal plasticity markers in Alzheimer’s disease. Mol Neurodegener 2022; 17: 27.

95.

van der Kant

Goldstein

LSB

Ossenkoppele

. Amyloid-β-independent regulators of tau pathology in Alzheimer disease. Nat Rev Neurosci 2020; 21: 21–35.

96.

Kent

Spires-Jones

Durrant

. The physiological roles of tau and Aβ: implications for Alzheimer’s disease pathology and therapeutics. Acta Neuropathol 2020; 140: 417–447.

97.

Darst

Koscik

Racine

, et al. Pathway-specific polygenic risk scores as predictors of amyloid-β deposition and cognitive function in a sample at increased risk for Alzheimer’s disease. J Alzheimers Dis 2017; 55: 473–484.

98.

Raulin

A-C

Doss

Trottier

, et al. Apoe in Alzheimer’s disease: pathophysiology and therapeutic strategies. Mol Neurodegener 2022; 17: 72.

99.

Lautner

Palmqvist

Mattsson

, et al. Apolipoprotein E genotype and the diagnostic accuracy of cerebrospinal fluid biomarkers for Alzheimer disease. JAMA Psychiatry 2014; 71: 1183–1191.

100.

Mahan

Wang

Bao

, et al. Selective reduction of astrocyte apoE3 and apoΕ4 strongly reduces Aβ accumulation and plaque-related pathology in a mouse model of amyloidosis. Mol Neurodegener 2022; 17: 13.

101.

Wildsmith

Holley

Savage

, et al. Evidence for impaired amyloid β clearance in Alzheimer’s disease. Alzheimers Res Ther 2013; 5: 33.

102.

Therriault

Benedet

Pascoal

, et al. Association of apolipoprotein E e4 with medial temporal tau independent of amyloid-β. JAMA Neurol 2020; 77: 470–479.

103.

Nightingale Health Biobank Collaborative Group . Metabolomic and genomic prediction of common diseases in 700,217 participants in three national biobanks. Nat Commun 2024; 15: 10092.

104.

Oka

Matsuzawa

Tsuneyoshi

, et al. Multiomics analysis to explore blood metabolite biomarkers in an Alzheimer’s Disease Neuroimaging Initiative cohort. Sci Rep 2024; 14: 6797.

105.

van der Veen

Kennelly

Wan

, et al. The critical role of phosphatidylcholine and phosphatidylethanolamine metabolism in health and disease. Biochim Biophys Acta 2017; 1859: 1558–1572.

106.

Conway

Seidler

Barrow

. Unlocking choline’s potential in Alzheimer's disease: a narrative review exploring the neuroprotective and neurotrophic role of phosphatidylcholine and assessing its impact on memory and learning. Clin Nutr ESPEN 2024; 64: 177–195.

107.

Wong

Braidy

Poljak

, et al. Dysregulation of lipids in Alzheimer’s disease and their role as potential biomarkers. Alzheimers Dement 2017; 13: 810–827.

108.

Whiley

Sen

Heaton

, et al. Evidence of altered phosphatidylcholine metabolism in Alzheimer’s disease. Neurobiol Aging 2014; 35: 271–278.

109.

Byeon

Madugundu

Jain

, et al. Cerebrospinal fluid lipidomics for biomarkers of Alzheimer’s disease. Mol Omics 2021; 17: 454–463.

110.

Yang

March

Stephenson

, et al. Apolipoprotein E in lipid metabolism and neurodegenerative disease. Trends Endocrinol Metab 2023; 34: 430–445.

111.

Segall

Dhanasekaran

Baldwin

, et al. Influence of apoE domain structure and polymorphism on the kinetics of phospholipid vesicle solubilization. J Lipid Res 2002; 43: 1688–1700.

112.

Ledesma

Martin

Dotti

. Lipid changes in the aged brain: effect on synaptic function and neuronal survival. Prog Lipid Res 2012; 51: 23–35.

113.

Tabassum

Widén

Ripatti

. Effect of biological sex on human circulating lipidome: an overview of the literature. Atherosclerosis 2023; 384: 117274.

114.

Arnold

Nho

Kueider-Paisley

, et al. Sex and APOE e4 genotype modify the Alzheimer’s disease serum metabolome. Nat Commun 2020; 11: 1148.

115.

Wishart

Guo

Oler

, et al. HMDB 5.0: the human metabolome database for 2022. Nucleic Acids Res 2021; 50: D622–D631.

116.

Gnjidic

Hilmer

Blyth

, et al. Polypharmacy cutoff and outcomes: five or more medicines were used to identify community-dwelling older men at risk of different adverse outcomes. J Clin Epidemiol 2012; 65: 989–995.

117.

Wastesson

Morin

Tan

ECK

, et al. An update on the clinical consequences of polypharmacy in older adults: a narrative review. Expert Opin Drug Saf 2018; 17: 1185–1196.

118.

Esumi

Ushio

Zamami

. Polypharmacy in older adults with Alzheimer’s disease. Medicina (Kaunas) 2022; 58: 1445.

119.

Winardi

Mach

McKay

, et al. Chronic polypharmacy, monotherapy, and deprescribing: understanding complex effects on the hepatic proteome of aging mice. Aging Cell 2025; 24: e14357.

120.

Eroli

Johnell

Latorre-Leal

, et al. Long-term exposure to polypharmacy impairs cognitive functions in young adult female mice. Aging (Albany NY) 2021; 13: 14729–14744.

121.

Rochon

Petrovic

Cherubini

, et al. Polypharmacy, inappropriate prescribing, and deprescribing in older people: through a sex and gender lens. Lancet Healthy Longev 2021; 2: e290–e300.

122.

Fujita

Masnoon

Mach

, et al. Polypharmacy and precision medicine. Camb Prism Precis Med 2023; 1: e22.

123.

Stepler

Gillyard

Reed

, et al. ABCA7, A genetic risk factor associated with Alzheimer’s disease risk in African Americans. J Alzheimers Dis 2022; 86: 5–19.

124.

Reitz

Jun

Naj

, et al. Variants in the ATP-binding cassette transporter (ABCA7), apolipoprotein E ɛ4,and the risk of late-onset Alzheimer disease in African Americans. JAMA 2013; 309: 1483–1492.

125.

Barnes

. Alzheimer disease in African American individuals: increased incidence or not enough data? Nat Rev Neurol 2022; 18: 56–62.

126.

Morris

Schindler

McCue

, et al. Assessment of racial disparities in biomarkers for Alzheimer disease. JAMA Neurol 2019; 76: 264–273.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

7.41 MB

8.01 MB

Integrative analysis of cerebrospinal fluid biomarkers,metabolomics,and polygenic risk reveals novel metabolite associations with Alzheimer's disease

Abstract

Background

Objective

Methods

Results

Conclusions

Keywords

Introduction

Methods

Sample information and processing

Study participants

Cohort of cognitively healthy subjects

CSF data collection

CSF metabolite processing

Genotyping and imputation

Statistical analyses

Correlations between AD biomarkers and CSF metabolites

Elastic net regression

Pathway enrichment analysis

Polygenic risk scores calculation

Association of AD PRS with CSF metabolites and AD biomarkers

Results

Metabolome-wide correlation of AD biomarkers

CSF metabolites can predict AD biomarkers

Pathway analysis of CSF metabolites correlated with P-tau and T-tau

Prediction of AD biomarkers and CSF metabolites using polygenic risk scores

Discussion

Supplemental Material

sj-docx-1-alz-10.1177_13872877251389924 - Supplemental material for Integrative analysis of cerebrospinal fluid biomarkers, metabolomics, and polygenic risk reveals novel metabolite associations with Alzheimer's disease

Footnotes

Acknowledgements

Ethical considerations

Consent to participate

Author contribution(s)

Funding

Declaration of conflicting interests

Data availability statement

Supplemental material

References

Supplementary Material