Skip to main content
NCBI home page
Alzheimer's & Dementia logoLink to Alzheimer's & Dementia
. 2025 Nov 14;21(11):e70898. doi: 10.1002/alz.70898

Genetic regulation of the metabolome differs by sex, Alzheimer's disease stage, and plasma biomarker status

Jaclyn M Eissman 1,2, Min Qiao 1,2, Vrinda Kalia 3, Marielba Zerlin‐Esteves 2, Dolly Reyes‐Dumeyer 1,2, Angel Piriz 1,2, Saurabh Dubey 3, Renu Nandakumar 3, Annie J Lee 1,2, Rafael A Lantigua 4, Martin Medrano 5, Diones Rivera Mejia 6,7, Lawrence S Honig 1,2,4, Clifton L Dalgard 8,9, Gary W Miller 3, Richard Mayeux 1,2,4, Badri N Vardarajan 1,2,4,✉
PMCID: PMC12617352  PMID: 41236724

Abstract

INTRODUCTION

We investigated genetic regulators of circulating plasma metabolites to identify pathways underlying biochemical changes in clinical and biomarker‐supported Alzheimer's disease (AD).

METHODS

We computed metabolite quantitative trait loci (QTL) with whole‐genome sequencing (WGS) and small molecule plasma metabolites from 229 older adults with clinical AD and 322 age‐matched healthy controls. Unbiased associations between 6881 metabolites and 332,772 common genetic variants were tested, adjusted for age, sex, and both metabolomic and genomic principal components.

RESULTS

We identified 72 SNP‐metabolite associations spanning 66 genes and 12 metabolite classes, including PYROXD2/N6‐methyllysine, FAAH/myristoylglycine, and FADS2/arachidonic acid. Additionally, we found differences in genetic regulation of metabolites among individuals with clinically‐defined AD compared to biomarker‐defined AD based on a published plasma P‐tau181 cutoff. We also found more SNP‐metabolite associations among males compared to females.

DISCUSSION

In summary, we identified sex‐ and disease‐specific genetic regulators of plasma metabolites, revealing unique biological mechanisms of genetic perturbations in AD.

Highlights

  • Genetic regulators of the metabolome spanned 66 genes and 12 metabolite classes.

  • Clinically versus biomarker‐defined Alzheimer's disease (AD) affects genetic regulators of the metabolome.

  • Most metabolite quantitative trait loci (QTLs) are not shared between males and females.

Keywords: Alzheimer's disease, genetic regulation, genomics, metabolites, metabolomics, quantitative trait loci

1. BACKGROUND

Metabolic changes are part of Alzheimer's disease (AD) pathogenesis, beginning very early in the disease, 1 with a decline in glucose metabolism in the brain. 2 Throughout disease, other metabolic shifts include changes in lipids, phosphatidylcholines, ceramides, lysophosphatidylcholines (lysoPC), bile acid metabolism, 3 methylhistidine metabolism, and fatty acid metabolism. 3 , 4 Prior studies have shown that metabolic signatures can accurately discriminate healthy controls fromindividuals with mild cognitive impairment (MCI) and those with AD, especially between the latter two groups. 5 , 6 , 7 , 8

Twin studies show that endogenous metabolic features may be heritable, with estimates of individual locus contributions at a median of 6.9% and a maximum of 62%, 9 and likewise, single nucleotide polymorphism (SNP) ‐based estimates find a strong median SNP‐based heritability of 19.7%. 10 Multiple quantitative trait loci (QTL) studies have found robust genetic loci associated with the metabolome. 10 , 11 , 12 , 13 , 14 , 15 However, most of these studies have not focused on aging‐ and AD‐specific genetic regulation, and have been predominantly in non‐Hispanic White individuals, limiting the generalizability of findings to more diverse populations. 15 It is important to explore genetic regulation of the metabolome in diverse groups, as AD risk differs by race and ethnicity 16 and metabolic changes in those with AD likewise differ by racial and ethnic groups. 17

In addition to racial and ethnic differences, sex differences are apparent in AD risk and pathogenesis, with robust sex differences in prevalence, 16 lifetime risk, 16 as well as neuropathologic burden and its relation to clinical AD presentation. 18 , 19 Evidence also suggests sex differences in the metabolome 20 and in metabolic dysregulation in AD. 3 , 7 , 21 Arnold et al., 3 identified sex differences in associations of metabolites and AD biomarkers, which included acylcarnitines and amino acids showing sex‐specific associations with cerebrospinal fluid (CSF) P‐tau and fluorodeoxyglucose positron emission tomography (FDG‐PET). 3 Furthermore, modules containing metabolites such as phosphatidylcholines, acylcarnitines, lipids, amino acids, and sphingomyelins showed sex‐specific associations with AD brain endophenotypes. 1 Genome‐wide association studies (GWAS) studies have also identified sex‐specific loci associated with AD endophentoypes, 22 , 23 , 24 , 25 , 26 but the intersection of genetic regulation and the metabolome in AD has not been fully elucidated in a sex‐specific manner.

The goal of this study was to expand our understanding of genetic regulation of the metabolome in AD and to elucidate the impact of sex and AD pathology on this relationship. We conducted a genome‐ and metabolome‐wide QTL analysis in a Caribbean Hispanic cohort of aging and AD. We conducted metabolome QTL analyses in the full cohort, by clinical and P‐tau181–assisted diagnosis subgroups, and by biological sex. This study adds to the current literature by clarifying the relationship of genetic regulation of the metabolome in a diverse sample, by sex, and by clinically and biomarker‐assisted AD diagnosis.

2. METHODS

2.1. Study participants

This study included participants from the Estudio Familiar de Influencia Genetica en Alzheimer study (EFIGA). 27 Recruitment for EFIGA began in 1998, following participants every 2 years, and enrolled individuals of Caribbean Hispanic ancestry from the Dominican Republic and the Washington Heights area of New York. All study participants had late‐onset AD (LOAD; ≥60 age cutoff) or a family history of AD, and were given a standardized evaluation, including a neurological test battery, structured medical and neurological exams, and a depression assessment. 28 , 29 Please note, only unrelated individuals were included in this analysis. AD diagnoses were determined from the NINCDS‐ADRDA criteria 30 , 31 for probable or possible LOAD, 32 and the Clinical Dementia Rating 33 , 34 , 35 was included to determined disease severity. Hixson and Vernier 36 modified criteria 37 and TaqMan genotyping were leveraged to determine each participant's apolipoprotein E (APOE) genotype.

2.2. Whole genome sequencing data collection

Whole‐genome sequencing (WGS) was performed at the Uniformed Services University of the Health Sciences, leveraging an Illumina polymerase chain reaction (PCR) ‐free library protocol, sequencing the data on the Illumina NovaSeq platform. WGS was generated among individuals with clinically diagnosed AD and age‐matched healthy controls using DNA extracted from PAXgene tubes at a mean depth of coverage of 30×. Analysis of WGS data were performed using an automated pipeline which is in line with recommendations from the Centers for Common Disease Genomics (CCDG) and the Trans‐Omics for Precision Medicine (TOPMed) platforms (note—although updated pipelines are available for more accurate calling of rare variants, indels and structural variants, for common variants as used in this analysis, the accuracy of the pipeline used is quite high). 38 Reads were aligned to human reference hs38DH with BWA‐MEM v0.7.15, and variant calling was conducted according to recommendations from the Genome Analysis Toolkit. 39

2.3. WGS quality control

Using BCFtools v1.19, we split multi‐allelic variants, aligned indels, retained variants that passed all quality filters, and set missing genotypes to the reference genotype. VCF files were converted to PLINK (v2.0 and v1.9) binary file sets retaining only biallelic SNPs, filtering for mean depth > 10 and genotype quality > 20. On the binary file sets, we performed variant‐level filtering, including filtering out variants with a missing genotyping rate > 5% and retaining common SNPs with a minor allele frequency (MAF) of > 5%. Then we performed sample‐level filtering, including filtering out samples with > 1% sample missingness and dropping duplicate samples. We conducted identity‐by‐descent relatedness calculations, dropping both samples in a pair if a pi‐hat estimate was > 0.9, and one sample of a pair if a pi‐hat estimate was between 0.25 and 0.9. Specific X‐chromosome processing included removing the pseudo‐autosomal region, sex check and sex imputation (for missing sex), as well as a differential missingness test between sexes (p < 10−7). A Hardy–Weinberg Equilibrium (HWE) exact test was conducted in all samples (p < 10−6) and among females for the X‐chromosome, filtering male samples accordingly. Additionally, to ensure only common variants were retained, we compared variant frequencies to gnomAD, 40 keeping variants with > 5% gnomAD frequency. Finally, we performed a principal component analysis (PCA; with PC‐AiR) to assess genetic ancestry and cryptic relatedness, removing sample outliers with an iterative outlier removal procedure. The final cleaned genetic data consisted of 619 unrelated samples (452/619 females and 167/619 males; 474/619 clinical controls, 143/619 clinical AD cases, and 2/619 with missing diagnosis) and 6,665,147 variants. For more extensive information regarding our WGS data and quality control (QC) pipeline, please see the Supplementary Methods located in the Supplementary Materials.

1. RESEARCH IN CONTEXT

  1. Systematic review: Previous studies have identified numerous robust genetic regulators—quantitative trait loci (QTLs)—of metabolite levels. However, most studies have not examined QTLs in diverse cohorts or explored sex‐ and disease‐specific QTLs, limiting generalizability of findings.

  2. Interpretation: We identified 72 QTLs of the plasma metabolome in an aging and Alzheimer's disease (AD) cohort of Hispanic individuals. Notably, we were able to both validate top QTLs in previous studies, demonstrating the applicability to Hispanic individuals, as well as identify hundreds of novel QTLs, including those specific to individuals with clinical AD, biomarker‐supported AD, and by sex.

  3. Future directions: Future work should continue to evaluate QTLs in diverse samples to better understanding molecular perturbations in groups disproportionately affected by AD. Additionally, future work should continue to identify genetic influences of metabolic changes that are specific to each disease stage and to each sex.

2.4. Plasma metabolomics data generation

The protocol for metabolite data generation was previously described. 41 Plasma was collected by venipuncture in K2EDTA tubes, and by 2h of collection was centrifugated, prepared, and store at −80°C. 41 , 42 Metabolites were extracted with acetonitrile, and injected in triplicate into two chromatographic columns: a hydrophilic interaction column under positive ionization (HILIC+) 43 and a C18 column under negative ionization (C18–), 44 which resulted in three technical replicates per sample per column. Columns were coupled to a Thermo Orbitrap HFX Q‐Exactive mass spectrometer and scanned for 85–1250 kDa molecules. To process the metabolites, feature detection and peak alignment were performed with apLCMS 45 and xMSanalyzer 46 software. Feature tables were produced that included mass‐to‐charge ratio, retention time, and median summarized abundance/intensity of each ion (i.e., metabolic feature) for each sample. Then an empirical Bayesian framework batch correction was implemented with ComBat. 47 If a feature had zero‐intensity, it was deemed to be below the detection limit. For each of these features, one‐half of the minimum intensity of the observed metabolic feature was used to impute their value. Metabolic features were retained if present in at least 70% of the samples, resulting in 3253 features and 3628 features from the HILIC+ and C18– columns, respectively. 41 All features were log‐transformed, quantile normalized, and autoscaled. 41

2.5. Metabolite annotation

Processed metabolic features were annotated leveraging the Human Metabolome Database (HMDB) and a multi‐stage clustering algorithm from the R package, xMSannotator (v.1.3.2). 41 , 46 Annotation was implemented for metabolic features in order to get pathway associations, intensity profiles, retention time, mass defect, and isotope/adduct patterns. A confidence level of 1 to 5 was assigned with each annotation, which was determined based on a previous published protocol 48 and levels 1–3 (where 1 = most confident) were used in this analysis. To assign a singular annotation when multiple annotations matched one feature, a protocol was followed for annotation that included the following rules: (1) annotations were chosen based on the highest confidence score. (2) If scores were similar, then the lowest difference between expected versus observed mass was chosen. (3) If these strategies did not resolve annotations, then a feature was labeled as having multiple matches or unknown. In total, 1340 C18‐ and 1070 HILIC+ features had an assigned confidence level of 1–3, and thus were retained for this analysis.

2.6. Statistical analyses

2.6.1. Metabolome QTL analyses

Prior to performing the metabolome‐wide QTL analyses, we retained individuals who had both WGS data and metabolomics data, which resulted in 551 individuals. To ensure robust data quality among this specific sample, we performed additional quality control steps, including a specific MAF filter (>5%) and a HWE test (p < 10−6) among the N = 551. Due to the high multiple testing burden and non‐independence between variants (linkage disequilibrium [LD]), we LD‐pruned the genetic data prior to analysis, resulting in 332,772 variants carried forward to analysis. All QTL analyses were performed with the MatrixEQTL R package (v. 2.3) 49 applying the linear association model and a genome‐wide false‐discovery rate (FDR) adjustment, with a priori significance set at FDR < 0.05. QTL models included each metabolite as the outcome and age at blood draw, sex, a binary diagnosis variable (cognitively unimpaired or AD), and APOE ε4 carrier status (binary variable) as covariates. Additional covariates included the first three genetic ancestry principal components and the first three principal components of the metabolomics data.

Met‐QTL subgroup analyses included (1) sex‐specific models, stratifying by sex (male or female), and (2) diagnosis‐specific models, stratifying by binarized clinical diagnosis (cognitively unimpaired or AD). Exploratory models stratified by P‐tau181 status, using a previously defined cutoff (please see Supplementary Methods accompanying this section, 2.6.1, located in the Supplementary Materials), 41 , 42 created two groups: biomarker‐negative control and biomarker‐supported AD (Table 1). Notably, about 56% of biomarker‐supported AD cases (N = 91/163) are also clinically‐defined AD cases, and about 40% of clinical AD cases (N = 91/229) are also biomarker‐supported AD cases. Additionally, we validated Met‐QTLs within P‐tau181‐defined strata by defining biomarker positivity using P‐tau217. P‐tau217 levels < 0.39 pg/mL were considered healthy controls and those with plasma P‐tau217 ≥ 0.39 pg/mL were considered biomarker positive for AD (Table 1). Due to sample size constraints of individuals with biomarker data and the concern of spurious associations, we did not evaluate QTLs stratified by both biomarker status and sex.

TABLE 1.

Participant characteristics.

Clinical controls

(N = 322)

Clinical cases

(N = 229)

Full sample

(N = 551)
Age at blood draw, (years)
Mean (SD) 68.71 ± 7.02 74.38 ± 7.86 71.08 ± 7.89
Sex
Male, N (%) 81 (25.16%) 72 (31.44%) 153 (27.77%)
Female, N (%) 241 (74.84%) 157 (68.56%) 398 (72.23%)
APOE ε4 carrier, (≥ 1 ε4 alleles)
Non‐carrier, N (%) 221 (68.63%) 132 (57.64%) 353 (64.07%)
Carrier, N (%) 99 (30.75%) 96 (41.92%) 195 (35.39%)
Missing, N (%) 2 (0.62%) 1 (0.44%) 3 (0.54%)
Biomarker‐supported diagnosis, P‐tau181‐cutoff: 2.63
Control (< 2.63), N (%) 247 (76.71%) 137 (59.83%) 384 (69.69%)
Male, N (%) 65 (20.19%) 40 (17.47%) 105 (19.06%)
Female, N (%) 182 (56.52%) 97 (42.36%) 279 (50.64%)
Case, (≥ 2.63), N (%) 72 (22.36%) 91 (39.74%) 163 (29.58%)
Male, N (%) 14 (4.35%) 31 (13.54%) 45 (8.17%)
Female, N (%) 58 (18.01%) 60 (26.20%) 118 (21.42%)
Biomarker‐supported diagnosis, P‐tau217‐cutoff: 0.39
Control, N (%) 220 (68.32%) 127 (55.46%) 347 (62.98%)
Male, N (%) 61 (18.94%) 38 (16.59%) 99 (17.97%)
Female, N (%) 159 (49.38%) 89 (38.86%) 248 (45.01%)
Case, N (%) 96 (29.81%) 101 (44.10%) 197 (35.75%)
Male, N (%) 16 (4.97%) 33 (14.41%) 49 (8.89%)
Female, N (%) 80 (24.84%) 68 (29.69%) 148 (26.86%)
Open in a new tab

Abbreviations: APOE, apolipoprotein E; SD, standard deviation.

2.6.2. Candidate Met‐QTL analysis of published AD GWAS loci

We tested whether previously published AD risk and protective loci were associated with metabolites, potentially elucidating biological mechanisms to these loci by understanding their influence on the metabolome. To conduct these tests, we compiled AD GWAS loci from the Bellenguez et al., 50 manuscript. In total, we tested 76 AD loci, utilizing the exact index SNPs identified by Bellenguez et al., 50 to clarify the relationships of these loci to the metabolome. We included these SNPs in a separate candidate analysis to ensure inclusions of all 76 SNPs and to reduce multiple testing burden. Due to the nature of the candidate analysis, we considered an association significant with a p < 0.05 (i.e., nominal significance). Filtering criterion (please see next section) for Met‐QTLs was applied to this candidate analysis solely at the metabolite level (not at the gene level).

2.6.3. Prioritization of Met‐QTLs

We applied a four‐step filtering procedure to prioritize Met‐QTLs. We retained QTLs with both a p‐value surviving genome‐wide FDR correction and that mapped to a metabolite with an annotation confidence level of 1–3. 48 Additionally, we mapped metabolites to previously known classes, and if an annotation was missing, we filtered out the QTL. We also retained QTLs that mapped to a known gene (ANNOVAR, 2020‐06‐07 release). 51 If variants were mapped to multiple genes, we selected the gene that was in closest proximity to the variant location, or if ambiguous, we selected a listed gene. If a variant mapped to genes > 1 Mb away, we set the gene as missing. For sex‐specific analyses, we performed one more filtering step, removing Met‐QTL pairs whereby the metabolite levels were significantly different between sexes. This retained only QTLs where the genetic regulation of metabolites differed by sex and not the levels of the metabolite itself. Group differences were determined by comparing mean levels between males and females through Welch's two‐sample t‐tests (in R), whereby a metabolite was considered as significantly different between sexes if the t‐test had a p < 0.05. In addition, we compared QTLs across subgroups, and this was based on comparing gene–metabolite pairs from QTLs surviving all post‐hoc filtering detailed above.

2.6.4. Validation of significant QTLs from Met‐QTL analyses

In an effort to validate our main effects results, we surveyed large blood, plasma, CSF, or brain metabolite QTL studies including Chen et al., 10 Yin et al., 11 Hysi et al., 12 Long et al, 13 Lotta et al., 14 and Wang et al. 15 Our study mapped QTLs to genes; thus, we compared our set of prioritized genes from the significant QTLs to that of each study above. If a gene matched and the associated metabolite was identical or of the same metabolite class as in our study, we considered it as evidence for validation.

3. RESULTS

3.1. Study participants

After both genetic and metabolomic data quality control, 551 EFIGA (Table 1) study participants were included in this analysis, with 229 clinical AD cases and 322 age‐matched healthy clinical controls. The full sample consisted of 153 men (27.77%) and 398 women (72.23%). The average age was 71.08 ± 7.89 years among the full sample. Furthermore, 353 participants (64.07%) were APOE ε4 non‐carriers and 195 (35.39%) were APOE ε4 carriers and 3 (0.54%) were missing APOE genotype information. Additionally, 384 participants (69.69%) were classified as biomarker‐negative controls (P‐tau181 < 2.63) and 163 participants (29.58%) were classified as biomarker‐supported AD (P‐tau181 ≥ 2.63; and 4 [0.73%] were missing biomarker status). 42

3.2. Genome‐wide metabolite QTL analysis

Metabolite QTL (Met‐QTL; Tables S1–S13; Supplementary Figures–located in the Supplementary Materials) analyses were performed separately for C18– and HILIC+ associated metabolites, adjusting for a genome‐wide FDR of < 0.05 among each set of tests. Overall, we identified 72 Met‐QTLs that survived both adjustment for multiple comparisons (FDR < 0.05) and post‐hoc filtering criteria (Figure 1A; Figure S1; Table S1). These 72 QTLs spanned 66 unique genes and 12 unique metabolite classes. Four significant Met‐QTL pairs were also observed in previously published QTL studies (Table S13). The strongest association, PYROXD2 and N6‐methyllysine (amino acid), was validated in published data from three QTL studies, 10 , 11 , 15 whereby each study also identified PYROXD2 and N6‐methyllysine. Similarly, one more study 12 previously identified PYROXD2 and observed an association also with a lysine derivative, N‐methylpipecolate. We additionally validated the association between FAAH and myristoylglycine (fatty amide, N‐acyl amine) in data from three previously published QTL studies 10 , 12 , 13 all of whom identified associations between FAAH and acylglycines (N‐acyl amines). We found an association with PDXDC1 and trans‐2‐dodecenoylcarnitine, an acylcarnitine, and previous work, 10 which likewise identified an association between PDXDC1 and an acylcarnitine. Interestingly, FADS2 regulates lysoPC that bind to or interact with arachidonic acid. The association between FADS2 and arachidonic acid was observed in both our study and a recent, published QTL study, 12 in another study between FADS2 and both a related molecule to arachidonic acid and multiple phospholipids, 13 between FADS2 and lysoPC in another study, 14 and finally between FADS2 and a glycerophosphocholine, 1‐palmitoyl‐2‐dihomo‐linolenoyl‐GPC (16:0/20:3n3 or 6) in a very recent QTL study. 15 Additionally, please see Table S12 for a candidate Met‐QTL analysis of known AD GWAS loci. All associations with a p < 0.05 are presented. Of the 76 loci tested with the 3628 C18‐ and 3253 HILIC+ features, respectively, 1746 associations were nominally significant (p < 0.05). Notably, ABCA7 appeared frequently in association with metabolites spanning multiple classes.

FIGURE 1.

FIGURE 1

Open in a new tab

Genome and metabolome‐wide quantitative trait loci (QTL) study in full sample and by diagnosis. Manhattan plots depicting metabolome‐wide QTL study results among the full sample (A), among individuals who are clinically healthy, (B) and those with clinical Alzheimer's disease (AD). (C) The red dashed lines depict the significance threshold, determined based on a genome‐wide false‐discovery rate threshold (FDR < 0.05).

3.3. Genome‐wide metabolite QTL subgroup analysis

We identified Met‐QTLs in participants stratified by clinical diagnosis of cognitively unimpaired or clinical AD (Figure 1B, C; Tables S2–S3). Among clinical controls, we identified 308 Met‐QTLs surviving adjustment for multiple comparisons (FDR < 0.05) and post‐hoc filtering criteria, with 274 non‐overlapping QTLs with AD QTLs. Among those with clinical AD, we identified 507 Met‐QTLs surviving adjustment for multiple comparisons (FDR < 0.05) and post‐hoc filtering criteria, of which 463 were not observed in healthy participants. Next, in exploratory analyses, we stratified the sample based on AD diagnosis defined by P‐tau181 levels (Figure S2; Tables S4–S5). Among biomarker‐negative controls, we identified 190 Met‐QTLs surviving adjustment for multiple comparisons (FDR < 0.05) and post‐hoc filtering criteria, with 168 non‐overlapping QTLs with biomarker‐supported AD. Among those with biomarker‐supported AD, we identified 866 Met‐QTLs surviving adjustment for multiple comparisons (FDR < 0.05) and post‐hoc filtering criteria, with 799 non‐overlapping QTLs with biomarker‐negative controls. Additional validation using P‐tau217‐defined criteria for AD found high concordance (Figure S3, Tables S4–S5) with top hits from the corresponding P‐tau181‐defined‐stratum.

Notably, there were significant QTLs that were present among both clinical AD cases and individuals with biomarker‐supported AD, but not among either control group. One noticeable difference between both healthy control strata as compared to clinical AD and biomarker‐supported AD strata was that we found nearly one‐third more QTLs associated with fatty acids among clinical AD compared to healthy controls and > 7‐fold more QTLs associated with fatty acids among biomarker‐supported AD cases as compared to biomarker‐negative controls. For example, we observed SNP associations with eicosapentaenoic fatty acid in clinical AD and biomarker‐supported AD groups but not in either healthy‐control strata.

3.4. Sex‐stratified genome‐wide metabolite quantitative trait loci analysis

We performed sex‐stratified Met‐QTL analyses in all males and females andclinically healthy and  clinical AD subgroups (Figure 2; Tables S6–S11). Among all males (Figure 2A; Table S6), 687 Met‐QTLs survived adjustment for multiple comparisons (FDR < 0.05) and post‐hoc filtering criteria, with 632 non‐overlapping QTLs with females. Among all females (Figure 2B; Table S9), 137 Met‐QTLs survived adjustment for multiple comparisons (FDR < 0.05) and post‐hoc filtering criteria, with 120 non‐overlapping QTLs with males. Types of QTLs that tended to differ between sexes included those associated with fatty acids and glycerophospholipids, both of which were more prevalent among males, and moreover the types of fatty acids associated also differed between sexes. Further, there were significant differences between the QTLs identified in male and female clinically healthy participants and males and females with clinically diagnosed AD (Figure 3A–D, Tables S6‐11).

FIGURE 2.

FIGURE 2

Open in a new tab

Sex‐specific genome and metabolome‐wide quantitative trait loci (QTL) study in full sample. Manhattan plots depicting metabolome‐wide QTL study among males (blue) and females (pink). The top row shows results in the full sample of males (A) and bottom row shows results in the full sample of females (B). The black dashed lines depict the significance threshold, determined based on a genome‐wide false‐discovery rate threshold (FDR < 0.05).

FIGURE 3.

FIGURE 3

Open in a new tab

Sex‐specific genome and metabolome‐wide quantitative trait loci (QTL) study by diagnosis. Manhattan plots depicting metabolome‐wide QTL study among males (blue) and females (pink). The top row shows results in males who are clinical controls (A) and males who have clinical Alzheimer's disease (AD) (B). The bottom row shows results in females who are clinical controls (C) and females with clinical AD. (D) The black dashed lines depict the significance threshold, determined based on a genome‐wide false‐discovery rate threshold (FDR < 0.05).

4. DISCUSSION

We performed a genome‐ and metabolome‐wide association study in a cohort of older adults with and without AD. We identified 72 robust QTLs across 66 genes spanning the genome, and these QTLs were associated with metabolites across 12 unique metabolic classes. Notably, we validated some of the top gene‐metabolite pairs in previous QTL studies, 10 , 11 , 12 , 13 , 14 , 15 and our study clarified that genetic regulation of these metabolites may be implicated in AD pathogenesis. Sex‐ and diagnosis‐specific analyses clarified that most QTLs in the metabolome do not appear to be shared across sexes and diagnostic subgroups (Figure 4; Figure S4), providing evidence that genetic regulation of metabolic changes with aging and in AD possibly differ by sex and disease stage.

FIGURE 4.

FIGURE 4

Open in a new tab

Unique and shared metabolite quantitative trait loci (QTLs) across sexes and diagnosis subgroups. Upset plots (A) and (C) illustrate the overlapping and non‐overlapping QTLs among diagnosis subgroups (A) and among sexes (C). Blue dots connected by lines under each set of bars show which subgroups share QTLs, whereas a blue dot with no line under a bar indicates those QTLs are not shared across any subgroups. Overlap was determined by comparing gene‐metabolite pairs from QTLs surviving all post‐hoc filtering. (B, D) The percentage of QTLs that belong to each metabolite class. AD, Alzheimer's disease; C18–, C18 column under negative ionization; CCDG, Centers for Common Disease Genomics; EFIGA, Estudio Familiar de Influencia Genetica en Alzheimer; FDR, false‐discovery rate; GWAS, genome‐wide association study; HILIC+, hydrophilic interaction under positive ionization; HMDB, human metabolome database; HWE, Hardy–Weinberg equilibrium; LD, linkage disequilibrium; LOAD, late‐onset AD; lysoPC, lysophosphatidylcholines; MAF, minor allele frequency; MCI, mild cognitive impairment; Met‐QTL, metabolome quantitative trait locus; PCA, principal component analysis; TOPMed, trans‐omics for precision medicine; WGS, whole‐genome sequencing.

4.1. Top Met‐QTLs are implicated in AD pathogenesis

Validated Met‐QTL associations (Tables S1 and S13) include, PYROXD2 and N6‐methyllysine, 10 , 11 , 12 , 15 FAAH and myristoylglycine, 10 , 12 , 13 FADS2 and arachidonic acid (LysoPC[20:4]), 12 , 13 , 14 , 15 and PDXDC1 and an acylcarnitine (trans‐2‐dodecenoylcarnitine). 10 (Additionally, please see Table S12 presents a candidate Met‐QTL analysis of known AD GWAS loci). PYROXD2 has been identified in multiple QTL studies in association with N6‐methyllysine or methyl‐L‐lysine derivatives. 52 Lysine metabolism plays a reported role in AD pathology, including that methylation of lysine residues is a post‐translational modifier of tau in neurofibrillary lesions, 53 , 54 and may harbor some protection against tau pathological aggregation. 54 One study further demonstrates that lysine metabolic changes can differentiate cognitively unimpaired individuals from those with MCI or AD. 5 Interestingly, in our study, the PYROXD2/N6‐methyllysine pair was present in both controls and AD groups, as well as among biomarker‐negative controls and biomarker‐supported AD groups, but showed more significant associations among both healthy control strata (Tables S1–S5).

The validated association between FAAH and myristoylglycine was observed among clinical and biomarker‐negative controls as compared to those with clinical AD and biomarker‐supported AD (Tables S1–S5). N‐acyl glycines, such as myristoylglycine, are upregulated in AD. 5 Furthermore, fatty acid amide hydrolase (FAAH) enzyme inhibitors play a known role in AD. 55 FAAH is found in the brain and is part of the endocannabinoid pathway, a well‐established pathway in AD etiology, whereby endocannabinoid levels correlate with AD biomarkers—especially Aβ pathology—and memory performance. 55 Interestingly, the FAAH and myristoylglycine was one of the strongest associations among females, and also is significant among females who were clinical controls or had clinical AD (Table S9–S11). This reveals that the main effects association was likely driven by females, suggestive that this metabolic change may be important in the clinical manifestation of AD among females more than males. It also reemphasizes the importance of sex‐specific analysis for not only identifying novel genetic loci but for uncovering information about associations identified in a main effects analysis.

FADS2 and arachidonic acid (LysoPC[20:4]), a top, validated association in this study, does not appear associated with clinical or biomarker‐negative controls nor clinical AD strata (Tables S2–S4). However, this association does appear in the biomarker‐supported AD strata (Table S5). The FADS gene cluster, which includes FADS2, is associated with cognition, whereby a Mendelian randomization colocalization analysis illustrates that FADS1 and FADS2 expression appear to have a causal effect on cognition. 56 Furthermore, genetic variation in the FAD gene cluster is associated with arachidonic acid levels, and arachidonic acid interacts with Aβ40 and Aβ42 pathology. 57 Since FADS2–arachidonic acid was a top association in the main analysis and in biomarker‐supported AD only, and arachidonic acid plays a role in AD pathology, this may mean that genetic regulation of the FAD gene cluster and arachidonic acid play an important role once AD pathology is present, but not before.

PDXDC1 and trans‐2‐dodecenoylcarnitine, an acylcarnitine, was a validated association in the main effects analysis and then appeared only in clinical controls (Tables S1–S5). Acylcarnitine levels in plasma are increased in individuals with AD, 4 and furthermore these metabolites show a pattern of consistent decrease from preclinical to clinical AD and may be able to identify AD converters before onset of disease. 4 , 58 One key feature of acylcarnitines is that they tend to show sex‐specific effects, 1 , 3 and notably in our sex‐stratified analyses, both among males and females, we identified QTLs associated with various acylcarnitines (Table S6–S11). Taken together, future studies should continue to investigate the sex‐and disease‐specific role of individual acylcarnitines in AD etiology.

4.2. Genetic regulation of the metabolome differs by sex

Our analysis sheds light on sex‐biased genetic regulation of the metabolome (Figures 2, 3). Sex differences in metabolic changes in AD 3 , 7 , 21 are well understood, but the question remains if shared genetic variation or if differing genetic variation mainly contributes to these observed phenotypic differences. This analysis provides evidence that most genetic variation relating to the metabolome in AD differs between sexes, as shown in our study (Figure 4; Figure S4), very few QTLs are shared between sexes. Importantly, these sex‐specific QTLs are not due to metabolite level differences by sex, as we filtered out any Met‐QTLs whereby the metabolite levels showed significant differences between sexes. In addition, the sex‐specific QTLs were spread throughout the genome, spanning beyond sex chromosome complement differences. One possible contribution to the observed sex differences is sex hormones, as a sex‐specific relationship already exists between sex hormones, aging, and AD. 59 Future studies should further investigate the sex‐specific crosstalk between the genome, the metabolome, and sex hormones in aging and AD.

4.3. Strengths and limitations

Strengths of this study included the use of an untargeted metabolomics approach, which both allows for more metabolites to be included and is a less biased approach as compared to targeted metabolomics. Additional strengths include the diversity of the sample, as most large Met‐QTL studies have only been performed in non‐Hispanic White individuals. The inclusion of sex‐specific models, including the X‐chromosome, allowed for a more complete understanding of the association of the genome and metabolome within each sex. We identified more Met‐QTLs from the biomarker‐defined analyses (top hits validated by P‐tau217–stratified analyses), potentially due to the more homogenous nature of these strata compared to the heterogenous pathologies present in clinically‐defined AD cases. Weaknesses of this study included the sample size, as we had 551 individuals in this study and recent QTL studies had sample sizes in the thousands. The sex‐stratified and diagnosis‐stratified models also had sample size imbalances; for example, we had fewer males compared to females, which could have partially influenced this analysis. Furthermore, we did not report QTL associations for metabolites that were not properly annotated and of an unknown class, but these metabolites do have meaning and should be investigated in future studies. Leveraging one biomarker (i.e., P‐tau181) to define biological AD with population‐specific cutoffs could limit generalizability to other population groups. Lastly, in this study, we covaried for APOE ε4 carrier status to investigate the genetic regulation of the metabolome above and beyond the well‐characterized APOE locus, but we do know that APOE is involved in lipid metabolism and shows sex differences, making it a key player in the sex‐specific relationship of the genome and metabolome.

5. CONCLUSIONS

Overall, this analysis identified (and validated) novel genetic regulators of the metabolome in aging and AD within a Hispanic cohort. We provided hundreds of novel sex‐specific and disease‐specific Met‐QTL findings that have not previously been the focus of most QTL studies to date, especially in the context of AD. Future studies should continue to investigate the relationship between the genome and the metabolome in AD through a precision medicine lens to continue to better understand the totality of the genetic contribution to AD at the molecular level.

CONFLICT OF INTEREST STATEMENT

The authors do not have any conflict of interest with the research presented in this investigation. Author disclosures are available in the Supporting Information.

CONSENT STATEMENT

All participants of the EFIGA study provided informed consent and all secondary analyses were approved by the Columbia University Medical Center Institutional Review Board.

Supporting information

Supporting Information

ALZ-21-e70898-s001.docx (1.1MB, docx)

Supporting Information

ALZ-21-e70898-s002.xlsx (1.2MB, xlsx)

Supporting Information

ALZ-21-e70898-s003.pdf (2.2MB, pdf)

ACKNOWLEDGMENTS

We acknowledge the services of CEDIMAT for collaborating with sample collection and processing in the EFIGA cohort. The metabolomics core that generated the metabolomics data for the project is supported by the National Center for Advancing Translational Sciences. Additionally, we thank Drs. Carlos Cruchaga and Postdocs, as well as Ph.D. students for their valuable input in validating the results from the Met‐QTL analyses in this dataset. The EFIGA study is supported by NIA grants R56AG063908, R01AG067501 and RF1AG015473. The metabolomics core is supported by grant 5UL1TR001873.

Eissman JM, Qiao M, Kalia V, et al. Genetic regulation of the metabolome differs by sex, Alzheimer's disease stage, and plasma biomarker status. Alzheimer's Dement. 2025;21:e70898. 10.1002/alz.70898

DATA AVAILABILITY STATEMENT

All results are included in the main text or in the supplementary materials of the present manuscript. The raw whole genome sequencing, metabolomics and biomarker data can be shared with qualified investigators using the request form available here: https://cumc.co1.qualtrics.com/jfe/form/SV_dmck0uV3A91pmzb. The WGS data are also available via the Alzheimer's Disease Sequencing Project: https://dss.niagads.org/datasets/ng00067/. Code from this study is made available on GitHub: https://github.com/jaclyn‐eissman/Metabolome‐Wide‐QTL‐Analysis.

REFERENCES

  • 1. González Zarzar T, Lee B, Coughlin R, Kim D, Shen L, Hall MA. Sex differences in the metabolome of Alzheimer's disease progression. Front Radiol. 2022;2:782864. doi: 10.3389/fradi.2022.782864 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2. Wilkins JM, Trushina E. Application of metabolomics in Alzheimer's disease. Front Neurol. 2018;8:719. doi: 10.3389/fneur.2017.00719 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3. Arnold M, Nho K, Kueider‐Paisley A, et al. Sex and APOE ε4 genotype modify the Alzheimer's disease serum metabolome. Nat Commun. 2020;11(1):1148. doi: 10.1038/s41467-020-14959-w [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4. Kalecký K, German DC, Montillo AA, Bottiglieri T. Targeted metabolomic analysis in Alzheimer's disease plasma and brain tissue in non‐Hispanic Whites. J Alzheimers Dis. 2022;86(4):1875–1895. doi: 10.3233/JAD-215448 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5. Trushina E, Dutta T, Persson XMT, Mielke MM, Petersen RC. Identification of altered metabolic pathways in plasma and CSF in mild cognitive impairment and Alzheimer's disease using metabolomics. PLOS ONE. 2013;8(5):e63644. doi: 10.1371/journal.pone.0063644 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6. Jääskeläinen O, Hall A, Tiainen M, et al. Metabolic profiles help discriminate mild cognitive impairment from dementia stage in Alzheimer's disease. J Alzheimers Dis. 2020;74(1):277–286. doi: 10.3233/JAD-191226 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7. Berezhnoy G, Laske C, Trautwein C. Metabolomic profiling of CSF and blood serum elucidates general and sex‐specific patterns for mild cognitive impairment and Alzheimer's disease patients. Front Aging Neurosci. 2023;15:1219718. doi: 10.3389/fnagi.2023.1219718 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8. Snowden SG, Ebshiana AA, Hye A, et al. Association between fatty acid metabolism in the brain and Alzheimer disease neuropathology and cognitive performance: a nontargeted metabolomic study. PLOS Med. 2017;14(3):e1002266. doi: 10.1371/journal.pmed.1002266 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9. Shin SY, Fauman EB, Petersen AK, et al. An atlas of genetic influences on human blood metabolites. Nat Genet. 2014;46(6):543–550. doi: 10.1038/ng.2982 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10. Chen Y, Lu T, Pettersson‐Kymmer U, et al. Genomic atlas of the plasma metabolome prioritizes metabolites implicated in human diseases. Nat Genet. 2023;55(1):44–53. doi: 10.1038/s41588-022-01270-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11. Yin X, Chan LS, Bose D, et al. Genome‐wide association studies of metabolites in Finnish men identify disease‐relevant loci. Nat Commun. 2022;13:1644. doi: 10.1038/s41467-022-29143-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12. Hysi PG, Mangino M, Christofidou P, et al. Metabolome Genome‐Wide Association Study identifies 74 novel genomic regions influencing plasma metabolites levels. Metabolites. 2022;12(1):61. doi: 10.3390/metabo12010061 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13. Long T, Hicks M, Yu HC, et al. Whole‐genome sequencing identifies common‐to‐rare variants associated with human blood metabolites. Nat Genet. 2017;49(4):568–578. doi: 10.1038/ng.3809 [DOI] [PubMed] [Google Scholar]
  • 14. Lotta LA, Pietzner M, Stewart ID, et al. A cross‐platform approach identifies genetic regulators of human metabolism and health. Nat Genet. 2021;53(1):54–64. doi: 10.1038/s41588-020-00751-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15. Wang C, Yang C, Western D, et al. Genetic architecture of cerebrospinal fluid and brain metabolite levels and the genetic colocalization of metabolites with human traits. Nat Genet. 2024;56(12):2685–2695. doi: 10.1038/s41588-024-01973-7. Published online November 11, 2024:1‐11. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16. 2024 Alzheimer's disease facts and figures. Alzheimers Dement. 2024;20(5):3708–3821. doi: 10.1002/alz.13809 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17. Vardarajan B, Kalia V, Manly J, et al. Differences in plasma metabolites related to Alzheimer's disease, APOE ε4 status, and ethnicity. Alzheimers Dement Transl Res Clin Interv. 2020;6(1):e12025. doi: 10.1002/trc2.12025 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18. Barnes LL, Wilson RS, Bienias JL, Schneider JA, Evans DA, Bennett DA. Sex differences in the clinical manifestations of Alzheimer disease pathology. Arch Gen Psychiatry. 2005;62(6):685–691. doi: 10.1001/archpsyc.62.6.685 [DOI] [PubMed] [Google Scholar]
  • 19. Oveisgharan S, Arvanitakis Z, Yu L, Farfel J, Schneider JA, Bennett DA. Sex differences in Alzheimer's disease and common neuropathologies of aging. Acta Neuropathol (Berl). 2018;136(6):887. doi: 10.1007/s00401-018-1920-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20. Wiese CB, Avetisyan R, Reue K. The impact of chromosomal sex on cardiometabolic health and disease. Trends Endocrinol Metab. 2023;34(10):652–665. doi: 10.1016/j.tem.2023.07.003 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21. Berezhnoy G, Laske C, Trautwein C. Quantitative NMR‐based lipoprotein analysis identifies elevated HDL‐4 and triglycerides in the serum of Alzheimer's disease patients. Int J Mol Sci. 2022;23(20):12472. doi: 10.3390/ijms232012472 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22. Eissman JM, Dumitrescu L, Mahoney ER, et al. Sex differences in the genetic architecture of cognitive resilience to Alzheimer's disease. Brain. 2022;145(7):2541–2554. doi: 10.1093/brain/awac177 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23. Eissman JM, Archer DB, Mukherjee S, et al. Sex‐specific genetic architecture of late‐life memory performance. Alzheimers Dement. 2023;20(2):1250–1267. doi: 10.1002/alz.13507 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24. Dumitrescu L, Mayeda ER, Sharman K, Moore AM, Hohman TJ. Sex differences in the genetic architecture of Alzheimer's disease. Curr Genet Med Rep. 2019;7(1):13–21. doi: 10.1007/s40142-019-0157-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25. Dumitrescu L, Barnes LL, Thambisetty M, et al. Sex differences in the genetic predictors of Alzheimer's pathology. Brain. 2019;142(9):2581–2589. doi: 10.1093/brain/awz206 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26. Deming Y, Dumitrescu L, Barnes LL, et al. Sex‐specific genetic predictors of Alzheimer's disease biomarkers. Acta Neuropathol (Berl). 2018;136(6):857–872. doi: 10.1007/s00401-018-1881-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27. Vardarajan BN, Faber KM, Bird TD, et al. Age‐specific incidence rates for dementia and Alzheimer disease in NIA‐LOAD/NCRAD and EFIGA families: National Institute on Aging Genetics Initiative for Late‐Onset Alzheimer Disease/National Cell Repository for Alzheimer Disease (NIA‐LOAD/NCRAD) and Estudio Familiar de Influencia Genetica en Alzheimer (EFIGA). JAMA Neurol. 2014;71(3):315–323. doi: 10.1001/jamaneurol.2013.5570 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28. Ruiz‐Grosso P, Loret de Mola C, Vega‐Dienstmaier JM, et al. Validation of the Spanish center for epidemiological studies depression and Zung self‐rating depression scales: a comparative validation study. PLoS ONE. 2012;7(10):e45413. doi: 10.1371/journal.pone.0045413 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29. Stallones L, Marx MB, Garrity TF. Prevalence and correlates of depressive symptoms among older U.S. adults. Am J Prev Med. 1990;6(5):295–303. [PubMed] [Google Scholar]
  • 30. McKhann GM, Knopman DS, Chertkow H, 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(3):263–269. doi: 10.1016/j.jalz.2011.03.005 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31. Bennett DA, Schneider JA, Aggarwal NT, et al. Decision rules guiding the clinical diagnosis of Alzheimer's disease in two community‐based cohort studies compared to standard practice in a clinic‐based cohort study. Neuroepidemiology. 2006;27(3):169–176. doi: 10.1159/000096129 [DOI] [PubMed] [Google Scholar]
  • 32. McKhann G, Drachman D, Folstein M, Katzman R, Price D, Stadlan EM. Clinical diagnosis of Alzheimer's disease. Neurology. 1984;34(7):939–939. doi: 10.1212/WNL.34.7.939 [DOI] [PubMed] [Google Scholar]
  • 33. Hughes CP, Berg L, Danziger W, Coben LA, Martin RL. A new clinical scale for the staging of dementia. Br J Psychiatry. 1982;140(6):566–572. doi: 10.1192/bjp.140.6.566 [DOI] [PubMed] [Google Scholar]
  • 34. Morris JC. The clinical dementia rating (CDR). Neurology. 1993;43(11):2412–2414. doi: 10.1212/WNL.43.11.2412-a [DOI] [PubMed] [Google Scholar]
  • 35. Morris JC, Ernesto C, Schafer K, et al. Clinical dementia rating training and reliability in multicenter studies. Neurology. 1997;48(6):1508–1510. doi: 10.1212/WNL.48.6.1508 [DOI] [PubMed] [Google Scholar]
  • 36. Hixson J, Vernier D. Restriction isotyping of human apolipoprotein E by gene amplification and cleavage with HhaI. J Lipid Res. 1990;31(3):545–548. doi: 10.1016/S0022-2275(20)43176-1 [DOI] [PubMed] [Google Scholar]
  • 37. Maestre G, Ottman R, Stern Y, et al. Apolipoprotein E and Alzheimer's disease: ethnic variation in genotypic risks. Ann Neurol. 1995;37(2):254–259. doi: 10.1002/ana.410370217 [DOI] [PubMed] [Google Scholar]
  • 38. Regier AA, Farjoun Y, Larson DE, et al. Functional equivalence of genome sequencing analysis pipelines enables harmonized variant calling across human genetics projects. Nat Commun. 2018;9(1):4038. doi: 10.1038/s41467-018-06159-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39. McKenna A, Hanna M, Banks E, et al. The Genome Analysis Toolkit: a MapReduce framework for analyzing next‐generation DNA sequencing data. Genome Res. 2010;20(9):1297–1303. doi: 10.1101/gr.107524.110 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40. Chen S, Francioli LC, Goodrich JK, et al. A genomic mutational constraint map using variation in 76,156 human genomes. Nature. 2024;625(7993):92–100. doi: 10.1038/s41586-023-06045-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41. Kalia V, Reyes‐Dumeyer D, Dubey S, et al. Lysophosphatidylcholines are associated with P‐tau181 levels in early stages of Alzheimer's disease. medRxiv. 2023. doi: 10.1101/2023.08.24.23294581. Preprint posted online August 25, 2023.08.24.23294581. [DOI] [Google Scholar]
  • 42. Honig LS, Kang MS, Lee AJ, et al. Evaluation of plasma biomarkers for A/T/N classification of Alzheimer disease among adults of Caribbean Hispanic ethnicity. JAMA Netw Open. 2023;6(4):e238214. doi: 10.1001/jamanetworkopen.2023.8214 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43. Buszewski B, Noga S. Hydrophilic interaction liquid chromatography (HILIC)—a powerful separation technique. Anal Bioanal Chem. 2012;402(1):231–247. doi: 10.1007/s00216-011-5308-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44. Yamada T, Uchikata T, Sakamoto S, et al. Supercritical fluid chromatography/orbitrap mass spectrometry based lipidomics platform coupled with automated lipid identification software for accurate lipid profiling. J Chromatogr A. 2013;1301:237–242. doi: 10.1016/j.chroma.2013.05.057 [DOI] [PubMed] [Google Scholar]
  • 45. Yu T, Park Y, Johnson JM, Jones DP. apLCMS—adaptive processing of high‐resolution LC/MS data. Bioinformatics. 2009;25(15):1930–1936. doi: 10.1093/bioinformatics/btp291 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46. Uppal K, Walker DI, Jones DP. xMSannotator: an R package for network‐based annotation of high‐resolution metabolomics data. Anal Chem. 2017;89(2):1063–1067. doi: 10.1021/acs.analchem.6b01214 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47. Leek JT, Johnson WE, Parker HS, Jaffe AE, Storey JD. The sva package for removing batch effects and other unwanted variation in high‐throughput experiments. Bioinformatics. 2012;28(6):882–883. doi: 10.1093/bioinformatics/bts034 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 48. Schymanski EL, Jeon J, Gulde R, et al. Identifying small molecules via high resolution mass spectrometry: communicating confidence. Environ Sci Technol. 2014;48(4):2097–2098. doi: 10.1021/es5002105 [DOI] [PubMed] [Google Scholar]
  • 49. Shabalin AA. Matrix eQTL: ultra‐fast eQTL analysis via large matrix operations. Bioinformatics. 2012;28(10):1353–1358. doi: 10.1093/bioinformatics/bts163 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 50. Bellenguez C, Küçükali F, Jansen IE, et al. New insights into the genetic etiology of Alzheimer's disease and related dementias. Nat Genet. 2022;54(4):412–436. doi: 10.1038/s41588-022-01024-z [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 51. Wang K, Li M, Hakonarson H. ANNOVAR: functional annotation of genetic variants from high‐throughput sequencing data. Nucleic Acids Res. 2010;38(16):e164. doi: 10.1093/nar/gkq603 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 52. Reus LM, Boltz T, Francia M, 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(8):3478–3490. doi: 10.1101/2023.09.26.559021 [DOI] [PubMed] [Google Scholar]
  • 53. Thomas SN, Funk KE, Wan Y, et al. Dual modification of Alzheimer's disease PHF‐tau protein by lysine methylation and ubiquitylation: a mass spectrometry approach. Acta Neuropathol (Berl). 2012;123(1):105–117. doi: 10.1007/s00401-011-0893-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 54. Funk KE, Thomas SN, Schafer KN, et al. Lysine methylation is an endogenous post‐translational modification of tau protein in human brain and a modulator of aggregation propensity. Biochem J. 2014;462(1):77–88. doi: 10.1042/BJ20140372 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 55. Jain S, Bisht A, Verma K, Negi S, Paliwal S, Sharma S. The role of fatty acid amide hydrolase enzyme inhibitors in Alzheimer's disease. Cell Biochem Funct. 2022;40(2):106–117. doi: 10.1002/cbf.3680 [DOI] [PubMed] [Google Scholar]
  • 56. Wu X, Jiang L, Qi H, et al. Brain tissue‐ and cell type‐specific eQTL Mendelian randomization reveals efficacy of FADS1 and FADS2 on cognitive function. Transl Psychiatry. 2024;14(1):1–9. doi: 10.1038/s41398-024-02784-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 57. El Shatshat A, Pham AT, Rao PPN. Interactions of polyunsaturated fatty acids with amyloid peptides Aβ40 and Aβ42. Arch Biochem Biophys. 2019;663:34–43. doi: 10.1016/j.abb.2018.12.027 [DOI] [PubMed] [Google Scholar]
  • 58. Cristofano A, Sapere N, Marca GL, et al. Serum levels of Acyl‐carnitines along the continuum from normal to Alzheimer's dementia. PLOS ONE. 2016;11(5):e0155694. doi: 10.1371/journal.pone.0155694 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 59. Rosario ER, Chang L, Head EH, Stanczyk FZ, Pike CJ. Brain levels of sex steroid hormones in men and women during normal aging and in Alzheimer's disease. Neurobiol Aging. 2011;32(4):604–613. doi: 10.1016/j.neurobiolaging.2009.04.008 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supporting Information

ALZ-21-e70898-s001.docx (1.1MB, docx)

Supporting Information

ALZ-21-e70898-s002.xlsx (1.2MB, xlsx)

Supporting Information

ALZ-21-e70898-s003.pdf (2.2MB, pdf)

Data Availability Statement

All results are included in the main text or in the supplementary materials of the present manuscript. The raw whole genome sequencing, metabolomics and biomarker data can be shared with qualified investigators using the request form available here: https://cumc.co1.qualtrics.com/jfe/form/SV_dmck0uV3A91pmzb. The WGS data are also available via the Alzheimer's Disease Sequencing Project: https://dss.niagads.org/datasets/ng00067/. Code from this study is made available on GitHub: https://github.com/jaclyn‐eissman/Metabolome‐Wide‐QTL‐Analysis.

Articles from Alzheimer's & Dementia are provided here courtesy of Wiley

RESOURCES