CSF and plasma lipid profiling predicts risk of cognitive impairment in cognitively unimpaired older adults

Takahisa Kanekiyo; Launia J White; Hongping Ye; Yingxue Ren; Jason D Ulrich; Ana‐Caroline Raulin; Alla Alnobani; Chengjie Xiong; Yan W Asmann; Maria Vassilaki; Alicia Algeciras‐Schimnich; Junmin Peng; Fatima S Amtashar; Suzanne E Schindler; David M Holtzman; Guojun Bu; Ronald C Petersen; Michael G Heckman; Xianlin Han

doi:10.1002/alz.71924

. 2026 Oct 27;22(10):e71924. doi: 10.1002/alz.71924

CSF and plasma lipid profiling predicts risk of cognitive impairment in cognitively unimpaired older adults

Takahisa Kanekiyo ^1,^✉, Launia J White ², Hongping Ye ³, Yingxue Ren ², Jason D Ulrich ^4,⁵, Ana‐Caroline Raulin ¹, Alla Alnobani ¹, Chengjie Xiong ⁵, Yan W Asmann ², Maria Vassilaki ⁶, Alicia Algeciras‐Schimnich ⁷, Junmin Peng ⁸, Fatima S Amtashar ⁴, Suzanne E Schindler ⁴, David M Holtzman ⁴, Guojun Bu ^1,⁹, Ronald C Petersen ¹⁰, Michael G Heckman ², Xianlin Han ^3,^✉

PMCID: PMC13714092 PMID: 42947286

Abstract

INTRODUCTION

Age‐related cognitive decline and Alzheimer's disease (AD) are strongly influenced by lipid metabolism, yet how central and peripheral lipid profiles relate to disease risk before symptom onset remains unclear.

METHODS

We performed comprehensive shotgun lipidomics in cerebrospinal fluid (CSF) and plasma from cognitively unimpaired older adults, with longitudinal follow‐up for Clinical Dementia Rating (CDR) conversion.

RESULTS

CSF lipidomes were primarily associated with age and established CSF AD biomarkers, whereas plasma lipid profiles were strongly influenced by sex. Specific CSF phosphatidylethanolamine subspecies showed robust positive associations with CSF total tau, phosphorylated tau 181, neurogranin, and chitinase‐3 like‐1. CSF triacylglycerol TAG(C49:6), phosphatidylglycerol PG(18:0‐18:1), cholesteryl ester CE(20:3), and phosphatidylethanolamine PE(D16:0‐18:1) independently predicted CDR conversion, improving risk prediction. Plasma sphingomyelin subspecies were modestly associated with cognitive decline risk.

DISCUSSION

These findings identify distinct CSF and plasma lipid signatures linked to preclinical cognitive decline and highlight brain lipid metabolism as a promising source of predictive biomarkers and therapeutic targets.

Keywords: Alzheimer's disease, apolipoprotein E (APOE), biomarker, cerebrospinal fluid, cognitive decline, lipidomics, plasma

Highlights

Comprehensive lipidomic profiling in CSF and plasma reveals distinct central and peripheral lipid signatures in cognitively unimpaired older adults.
Specific CSF lipid species are strongly associated with CSF AD biomarkers and predict risk of cognitive decline.
Findings support lipid metabolism as a key contributor to early disease processes and highlight its potential as a diagnostic and therapeutic target for age‐related cognitive decline and AD.

1. BACKGROUND

Lipids are vital components of the brain, accounting for approximately 50% of its dry weight, ¹ and play essential roles in maintaining brain homeostasis by modulating cellular membrane structure, synaptic plasticity, myelination, energy regulation, immune responses, and diverse signaling pathways. ² , ³ Given their central role in brain function, disturbances in lipid metabolism have been increasingly implicated in the pathogenesis of neurodegenerative diseases, including Alzheimer's disease (AD). ² , ³ , ⁴ Notably, several well‐established AD genetic risk genes such as apolipoprotein E (APOE), CLU, and ABCA7 are related to lipid metabolism. ³ A genetic meta‐analysis also revealed that AD risk loci were enriched in pathways related to lipid processing and immunity as well as amyloid beta (Aβ) and tau pathology. ⁵ Furthermore, a recent study estimated that up to 40% of dementia cases may be attributed to 12 modifiable risk factors across the life course, including early life factors (<45 years; less education), midlife factors (45 to 65 years; hypertension, obesity, hearing loss, traumatic brain injury, and alcohol misuse), and late‐life factors (>65 years; smoking, depression, physical inactivity, social isolation, diabetes, and air pollution). ⁶ Importantly, many of these risk factors are closely linked to metabolic syndrome, in which lipid metabolism is directly or indirectly involved in disease pathophysiology. Therefore, a deeper understanding of brain lipid dyshomeostasis in humans is essential for the development of effective diagnostic and therapeutic strategies for age‐related cognitive decline and dementia.

In this study, we aimed to define lipid profiles in the cerebrospinal fluid (CSF) and plasma samples from cognitively unimpaired participants aged over 65 years in the Mayo Clinic Study of Aging (MCSA) ⁷ using shotgun lipidomics. We investigated the influence of APOE genotype, age, sex, and modifiable dementia risk factors (education, obesity, smoking, hypertension, diabetes, and dyslipidemia) on the CSF and plasma lipidomes. Additionally, we examined the relationship between these lipid profiles and established AD fluid biomarkers, such as CSF Aβ42, total tau, phosphorylated tau 181 (p‐tau181), ⁸ neurofilament light chain (NfL), ⁹ , ¹⁰ synaptosomal‐associated protein 25 (SNAP25), ¹¹ neurogranin (Ng), ¹² , ¹³ and chitinase‐3 like‐1, cartilage glycoprotein‐39 (YKL40), ¹⁴ , ¹⁵ as well as plasma p‐tau181, Aβ40, Aβ42, NfL, and glial fibrillary acidic protein (GFAP). ¹⁶ To explore the potential contributions of brain lipids to cognitive decline, we evaluated the associations of baseline lipid markers and subsequent progression to cognitive impairment, defined by changes in Clinical Dementia Rating (CDR) scale. ¹⁷ Our results demonstrated that CSF lipid species are primarily associated with age and CSF AD biomarkers, whereas plasma lipid profiles are predominantly influenced by sex. Finally, our findings suggest that CSF and plasma lipid markers may serve as potential biomarkers to improve the prediction of cognitive decline in older adults.

2. METHODS

2.1. Study design

CSF or plasma samples from cognitively unimpaired participants (CDR global score = 0, >65 years old) with available APOE genotype were obtained from the MCSA ⁷ or the Knight Alzheimer's Disease Research Center (ADRC) at Washington University (WashU) School of Medicine in St. Louis. Among the cohort, longitudinal CDR conversion analysis included cases with CDR Sum of Boxes (CDR‐SB) = 0 at the baseline visit and at least one follow‐up visit for cognitive testing. CDR conversion was defined as the occurrence of the first CDR‐SB greater than 0. The MCSA includes participants from Olmsted County, Minnesota, USA. Research participants enrolled in studies of memory and aging at the Knight ADRC consisted of community‐dwelling older adults including participants with and without cognitive impairment, who were recruited primarily from the greater St. Louis, Missouri metropolitan region. All participants were unrelated, non‐Hispanic/Latino, and White. The participants underwent a battery of examinations: a neurological evaluation, risk factor ascertainment, neuropsychological testing, and self‐ and informant‐reported neuropsychiatric symptoms. The MCSA has been approved by the Mayo Clinic Institutional Review Boards and Olmsted Medical Center. All procedures at the Knight ADRC were approved by the WashU Human Research Protection Office. These studies follow Health Insurance Portability and Accountability Act guidelines and obtained written informed consent from all participants.

RESEARCH IN CONTEXT

Systematic review: The authors reviewed the literature using traditional sources (e.g., PubMed). While the importance of lipid metabolism in AD pathogenesis has been established, associations of CSF and plasma lipid profiling with AD fluid markers and cognitive impairment risk in older adults have not been comprehensively characterized.
Interpretation: Our findings support an integrated hypothesis that alterations in lipid metabolism reflect early pathogenic processes underlying age‐related cognitive decline and AD. This hypothesis is consistent with existing experimental and clinical evidence implicating contributions of lipid pathways to the pathogenesis of neurodegenerative diseases.
Future directions: The manuscript proposes a framework to generate new hypotheses and guide future studies. These include further investigation of (a) the mechanistic roles of specific lipid species in amyloid and tau pathology, neurodegeneration, and neuroinflammation; (b) the interplay between central and peripheral lipid metabolism and their contributions to disease progression; (c) longitudinal changes in lipid profiles and their potential utility as fluid biomarkers; and (d) the therapeutic potential of targeting lipid metabolism to prevent or treat age‐related cognitive decline and AD.

2.2. AD biomarker measurements and lipidomics in CSF and plasma samples

CSF samples were collected through lumbar puncture and centrifuged at 2000 × g for 10 min to remove leukocytes. To avoid potential blood contamination, the first 1 to 2 mL of CSF was not used for the biomarker measurements. No further quality control testing was conducted to evaluate potential blood contamination. Datasets for CSF Aβ42, total tau, and p‐tau181 levels measured by automated electrochemiluminescence Elecsys immunoassays (Roche Diagnostics) and for CSF NfL levels measured by ELISA were obtained from MCSA. ¹⁸ Datasets for CSF Ng and SNAP25 levels measured by microparticle‐based immunoassays using Single Molecule Counting technology (Singulex® and now part of EMD Millipore, Burlington, MA, USA) ¹² , ¹³ and CSF YKL40 level measured by ELISA (Quidel MicroVue Bone, San Diego, CA) ¹¹ were obtained from our previous publication. ¹⁹ Datasets for plasma Aβ40, Aβ42, NfL, GFAP, and p‐tau181 levels measured by Quanterix Simoa analyzer were obtained from MCSA. Participants were encouraged to fast prior to sample collection. No standardized protocol was implemented regarding the timing of sample collection.

Shotgun lipidomic analysis was conducted through the multidimensional mass spectrometry system as described previously. ²⁰ , ²¹ The median sample storage duration was 12 years, and samples underwent one freeze‐thaw cycle before lipidomic analysis. CSF and plasma samples (100 µL) with the addition of appropriate amounts of lipid internal standards (one per each lipid class) as a premixture were subjected to lipid extraction through modified Bligh and Dyer procedure. ²² Each lipid extract was reconstituted in organic solvent composed of chloroform and methanol (1:1 volume ratio, 100 µL/100 µL samples) and further diluted to a final total lipid concentration of ∼500 fmol/µL prior to lipidomics analysis. Mass spectrometric analysis was performed on a triple quadrupole mass spectrometer (TSQ Altis, Thermo Fisher Scientific) and a Q Exactive mass spectrometer (Thermo Fisher Scientific), which were connected with an automated nanospray device (TriVersa NanoMate, Advion Inc). ²³ Data processing based on the principles of shotgun lipidomics such as selective ionization, low concentration of the lipid solution to avoid lipid aggregation, and correction for differential isotopologue patterns and kinetics of fragmentation was conducted as previously described. ²⁴

2.3. Statistical analysis

Continuous variables were summarized with the sample median and range. Categorical variables were summarized with the number and percentage of subjects. Comparisons of subject characteristics according to APOE genotype group were made using a Kruskal–Wallis rank‐sum test (continuous variables) or Fisher exact test (categorical variables). For CSF biomarkers, NfL, Ng, and SNAP25 were examined on a logarithmic scale in all regression analyses due to their skewed distributions, while no transformations were utilized for other CSF biomarkers. For plasma biomarkers, a natural logarithm transformation was utilized for p‐tau181, GFAP, and NfL in all regression analyses to address skewed distributions; no transformations were employed for other plasma biomarkers. To address distributional skewness and outliers, as well as to examine each analyte on the same scale in order to enhance the interpretability of linear regression results, a rank transformation ²⁵ was applied following quantile normalization for all analytes and lipid class sums for use in regression analysis; a higher rank corresponds to a higher value. Following the aforementioned transformations, CSF biomarkers, plasma biomarkers, CSF analytes and lipid class sums, and plasma analytes and lipid class sums were all scaled to have mean = 0 and standard deviation = 1 in order to enhance the interpretability of results of regression analyses.

Associations of CSF and plasma lipid subspecies, CSF and plasma lipid class sums, and CSF/plasma lipid ratios with APOE genotype group (APOE2 vs APOE3, APOE4 vs APOE3, and ordinal APOE group: APOE2 = 1, APOE3 = 2, APOE4 = 3, where higher values represent a more “severe” APOE genotype), age, sex, years of education, obesity, smoking, hypertension, diabetes, dyslipidemia, and CSF or plasma AD biomarkers were evaluated using unadjusted and multivariable linear regression models. Multivariable models were adjusted for APOE genotype group, age, sex, years of education, obesity, smoking, hypertension, diabetes, and dyslipidemia. Regression coefficients (denoted by β) and 95% confidence intervals (CIs) were estimated and are interpreted as the change in the mean analyte (following rank transformation and subsequent scaling) in comparison to a reference group (categorical variables) or corresponding to a specified increase (continuous variables). Associations between CSF and plasma lipids were also evaluated using the aforementioned unadjusted and multivariable regression models.

The Kaplan–Meier method was used to estimate the cumulative incidence of CDR conversion after the baseline visit, where censoring occurred at the date of the final visit. Associations of CSF and plasma lipid subspecies, CSF and plasma lipid class sums, CSF lipid ratios, plasma lipid ratios, AD biomarkers, age, sex, APOE, and modifiable dementia risk factors with CDR conversion were evaluated using unadjusted and multivariable Cox proportional hazards regression models. Multivariable models were adjusted for APOE genotype group, age, sex, years of education, obesity, smoking, hypertension, diabetes, and dyslipidemia in analyses of CSF data and for APOE genotype group, age, and sex in analyses of plasma data and the WashU series where the number of CDR conversions was lower. Hazard ratios (HRs) and 95% CIs were estimated and corresponded to each 1‐standard deviation increase in the given analyte or lipid class sum (following rank transformation). To assess the ability of CSF and plasma lipid subspecies, CSF and plasma lipid class sums, CSF lipid ratios, plasma lipid ratios, AD biomarkers, demographic variables, and dementia risk factors to accurately predict CDR conversion, we first used unadjusted Cox proportional hazards regression models in an exploratory analysis, where of primary interest was the estimation of the model c‐index; a c‐index of 1.0 indicates perfect predictive ability, whereas a c‐index of 0.5 indicates predictive ability equal to chance. Subsequently, we utilized multivariable Cox proportional hazards regression models applying a forward selection approach for variable selection with a p < 0.01 criterion for model entry (except when analyzing CSF lipid ratios and plasma lipid ratios, where a p < 0.0001 criterion for model entry was utilized). Variables that were considered in forward selection included all clinical characteristics, as well as analytes and lipid class sums that were associated with CDR conversion with a p value < 0.01 (or < 0.0001 for CSF and plasma lipid ratios) when adjusting for all clinical characteristics; c‐index was again estimated. In further examination, we also included the AD biomarkers in forward selection analysis, though such an analysis was considered secondary due to the reduced sample size caused by missing data for some of the biomarkers. To avoid providing an overly optimistic estimate of the ability of a given forward selection model to predict CDR conversion, we calculated internally validated c‐indexes using five‐fold cross validation. ²⁶ Comparisons of observed c‐index (i.e., the c‐index observed in our data, without internal validation) between different Cox regression models were made using the compareC function in R (version 4.4.1; R Foundation for Statistical Computing, Vienna, Austria). ²⁷

A Bonferroni correction for multiple testing was utilized to account for 156 CSF analyte outcomes (145 analytes and 11 lipid class sums), 166 plasma analyte outcomes (155 analytes and 11 lipid class sums), 113 CSF/plasma analyte outcomes (104 analytes and nine lipid class sums), 10,440 pair‐wise CSF lipid analyte ratio outcomes (CSF lipid class sums were not assessed), and 11,935 pair‐wise plasma analyte lipid ratio outcomes (plasma lipid class sums were not assessed). Correspondingly, p values < 0.00032 for CSF, < 0.00030 for plasma, < 0.00044 for CSF/plasma, < 4.79E‐06 for CSF lipid ratios, and < 4.19E‐06 for plasma lipid ratios were considered statistically significant. However, since applying a correction for multiple testing limits the probability of a Type I error (i.e., false‐positive finding), it correspondingly increases the likelihood of a Type II error (i.e., a false‐negative finding), and therefore, we considered p values < 0.01 (or < 0.0001 for CSF lipid ratios and plasma lipid ratios) as displaying suggestive evidence of an association. All statistical tests were two‐sided. Statistical analyses were performed using SAS (version 9.4; SAS Institute, Inc., Cary, NC, USA).

2.4. Weighted correlation network analysis

A weighted correlation network analysis was performed using the WGCNA R package. ²⁸ The raw lipid abundances were quantile normalized and log2 transformed. The power of 6 was chosen to build scale‐free topology using a signed hybrid network. ²⁹ Hybrid dynamic tree cutting was utilized based on a minimum module size of five lipids and a minimum height for merging modules of 0.15. Each module was summarized by module eigengene and assigned a unique color identifier. Lipids with high connectivity in the respective modules were considered hub lipids. Intra‐modular lipid‐lipid connection was visualized using VisANT version 5.53. ³⁰ Given the exploratory nature of the WGCNA module‐trait analyses, nominal p values from Pearson correlations are reported without additional adjustment for multiple comparisons.

3. RESULTS

3.1. Associations between CSF and plasma lipids in cognitively unimpaired participants

A total of 282 study subjects of cognitively unimpaired individuals from the MCSA were included in this study. Participants were selected based on the following baseline criteria: CDR = 0, age > 65 years, and the availability of CSF samples and APOE genotype status. The median age was 74 years (range: 65 to 93 years), and 168 subjects (59.7%) were male. APOE genotype distribution was 39 subjects with ε2/ε3, 171 subjects with ε3/ε3, 68 subjects with ε3/ε4, and four subjects with ε4/ε4. For all analyses, APOE genotype was categorized into three groups: APOE2 (ε2/ε3, N = 39), APOE3 (ε3/ε3, N = 171), or APOE4 (ε3/ε4 or ε4/ε4, N = 72) (Table S1). Within this cohort, 216 paired plasma samples were also analyzed (Table S2). There were no evident differences in modifiable dementia risk factors including years of education, body mass index (BMI), smoking, hypertension, diabetes, and dyslipidemia between APOE groups (Tables S1 and S2). While CSF levels of Aβ42 and SNAP25 differed significantly between the three APOE genotype groups (p < 0.001), ¹⁹ the effects on plasma biomarkers were modest (Tables S1 and S2). Using shotgun lipidomics, we identified and quantified 145 lipid species across 11 different lipid classes including cholesterol esters (CEs), non‐esterified fatty acids (FAs), total cholesterol (TC), lysophosphatidylcholine (LPC), phosphatidic acid (PA), phosphatidylcholine (PC), phosphatidylethanolamine (PE), phosphatidylglycerol (PG), phosphatidylserine (PS), sphingomyelin (SM), and triacylglycerol (TAG) in CSF and plasma samples. In CSF, some TAG subspecies were positively correlated with FA subspecies, while there were negative associations of TAG or FA subspecies with major lipid classes, particularly SM (Figure 1A). On the other hand, plasma lipids showed positive associations among major lipid classes with some exceptions such as those of TAG or FA with SM subspecies (Figure 1B). The associations between CSF and plasma samples were modest (Figure 1C), although significant positive associations in CE(20:5), CE(22:6), and PC subspecies were detected (Table S3).

Correlative heatmaps between CSF and plasma lipids. Heatmaps for pairwise Pearson correlation coefficients (r) between lipid species measured in CSF (A) and plasma (B) and between CSF and plasma lipid species (C) across subjects are shown. Lipids are grouped and ordered by class. Statistical significance was assessed using Pearson correlation tests followed by Benjamini–Hochberg false discovery rate (FDR) correction. *FDR‐adjusted q < 0.05.

3.2. Impact of APOE on CSF and plasma lipids in cognitively unimpaired participants

In this and all subsequently described analyses, p < 0.00032 (CSF analysis), p < 0.00030 (plasma analysis), and p < 0.00044 (CSF/plasma ratio analysis) were considered statistically significant after correcting for multiple testing, while p < 0.01 was considered suggestive. When making comparisons with APOE3 separately for APOE2 and APOE4 subjects, suggestive differences in multivariable analysis adjusting for potential confounding variables were observed between APOE2 and APOE3 subjects for PC(D16:1‐16:1/D14:1‐18:1) (β: 0.48, p = 0.008) and CE(14:0) (β: 0.64, p = 0.0005) in CSF. Lower CE(14:0) and SM(N14:0) and higher PE(D16:0‐18:2/D16:1‐18:1) and PE(A18:0‐18:1) CSF levels were detected in APOE4 subjects compared to APOE2 subjects. Additionally, while plasma FA(C20:4) levels were higher in APOE2 subjects than APOE3 subjects (β: 0.56, P = 0.0088), plasma PC(P18:2‐20:1/P20:1‐18:2/A18:0‐20:4) was increased in APOE4 subjects (β: 0.43, P = 0.0068). We also observed a suggestive decrease in CSF/plasma ratio of LPC(18:1) in APOE2 subjects and suggestive increases in CSF/plasma lipid ratios of TAG subspecies and FA(C18:2) in APOE4 subjects compared to APOE3 subjects. When APOE4 subjects were compared with APOE2 subjects, suggestive increases in CSF/plasma ratio of LPC(18:1) and PC(D18:2‐18:2/D16:0‐20:4) were observed (Table 1).

TABLE 1.

Associations of CSF and plasma lipids with APOE group.

	APOE2 versus APOE3 (reference group)				APOE4 versus APOE3 (reference group)				APOE4 versus APOE2 (reference group)
	Unadjusted analysis		Adjusting for age, sex, years of education, BMI, smoking, hypertension, diabetes, and dyslipidemia		Unadjusted analysis		Adjusting for age, sex, years of education, BMI, smoking, hypertension, diabetes, and dyslipidemia		Unadjusted analysis		Adjusting for age, sex, years of education, BMI, smoking, hypertension, diabetes, and dyslipidemia
	β (95% CI)	p value	β (95% CI)	p value	β (95% CI)	p value	β (95% CI)	p value	β (95% CI)	p value	β (95% CI)	p value
CSF lipids
PC(D16:1‐16:1/D14:1‐18:1)	0.41 (0.07, 0.76)	0.018	0.48 (0.13, 0.83)	0.0080	0.21 (−0.07, 0.49)	0.15	0.20 (−0.08, 0.48)	0.16	−0.21 (−0.59, 0.17)	0.28	−0.34 (−0.75, 0.07)	0.11
CE(14:0)	0.64 (0.30, 0.99)	0.0003	0.64 (0.28, 1.00)	0.0005	−0.04 (−0.32, 0.24)	0.77	−0.06 (−0.34, 0.22)	0.69	−0.68 (−1.04, −0.33)	0.0002	−0.66 (−1.05, −0.27)	0.0012
PE(D16:0‐18:2/D16:1‐18:1)	−0.40 (−0.74, −0.05)	0.024	−0.43 (−0.79, −0.08)	0.017	0.27 (−0.00, 0.54)	0.050	0.25 (−0.03, 0.52)	0.077	0.67 (0.28, 1.06)	0.0009	0.57 (0.15, 0.99)	0.0079
PE(A18:0‐18:1)	−0.14 (−0.49, 0.21)	0.45	−0.15 (−0.52, 0.22)	0.43	0.29 (0.01, 0.56)	0.039	0.30 (0.02, 0.58)	0.035	0.42 (0.03, 0.82)	0.036	0.59 (0.17, 1.01)	0.0063
SM(N14:0)	0.39 (0.04, 0.74)	0.030	0.47 (0.11, 0.84)	0.011	−0.03 (−0.31, 0.25)	0.84	−0.03 (−0.31, 0.25)	0.83	−0.42 (−0.79, −0.05)	0.028	−0.62 (−1.02, −0.22)	0.0027
Plasma lipids
FA(C20:4)	0.52 (0.10, 0.93)	0.016	0.56 (0.14, 0.98)	0.0088	0.17 (−0.14, 0.48)	0.27	0.15 (−0.15, 0.45)	0.34	−0.34 (−0.81, 0.12)	0.14	−0.43 (−0.89, 0.03)	0.068
PC(P18:2‐20:1/P20:1‐18:2/A18:0‐20:4)	−0.10 (−0.50, 0.31)	0.64	0.06 (−0.37, 0.48)	0.79	0.44 (0.13, 0.74)	0.0050	0.43 (0.12, 0.74)	0.0068	0.53 (0.05, 1.02)	0.031	0.34 (−0.20, 0.87)	0.22
CSF/plasma ratio
LPC(18:1)	−0.66 (−1.10, −0.23)	0.0032	−0.75 (−1.21, −0.29)	0.0014	−0.00 (−0.31, 0.31)	1.00	0.01 (−0.29, 0.32)	0.92	0.66 (0.21, 1.11)	0.0045	0.70 (0.21, 1.20)	0.0061
TAG(C50:2/C51:9)	0.24 (−0.19, 0.67)	0.27	0.34 (−0.11, 0.80)	0.14	0.53 (0.22, 0.84)	0.0010	0.53 (0.22, 0.85)	0.0010	0.29 (−0.17, 0.75)	0.22	0.39 (−0.13, 0.90)	0.14
TAG(C52:3/C53:10)	0.31 (−0.13, 0.75)	0.16	0.32 (−0.15, 0.79)	0.18	0.44 (0.12, 0.75)	0.0072	0.45 (0.12, 0.77)	0.0072	0.13 (−0.32, 0.57)	0.57	0.20 (−0.28, 0.68)	0.41
TAG(C54:4/C55:11)	0.25 (−0.18, 0.68)	0.26	0.22 (−0.22, 0.67)	0.32	0.42 (0.10, 0.73)	0.010	0.44 (0.12, 0.76)	0.0072	0.17 (−0.30, 0.64)	0.48	0.27 (−0.21, 0.76)	0.27
FA(C18:2)	0.26 (−0.18, 0.71)	0.24	0.22 (−0.23, 0.67)	0.34	0.37 (0.06, 0.69)	0.021	0.45 (0.14, 0.76)	0.0046	0.11 (−0.33, 0.55)	0.61	0.24 (−0.22, 0.70)	0.30
TAG(C52:5)	0.24 (−0.19, 0.67)	0.27	0.21 (−0.25, 0.67)	0.37	0.50 (0.18, 0.81)	0.0022	0.52 (0.20, 0.84)	0.0016	0.25 (−0.19, 0.70)	0.26	0.34 (−0.15, 0.84)	0.17
TAG(C54:5/C55:12)	0.20 (−0.23, 0.64)	0.36	0.14 (−0.30, 0.59)	0.53	0.40 (0.08, 0.73)	0.014	0.44 (0.12, 0.77)	0.0076	0.20 (−0.24, 0.65)	0.37	0.30 (−0.18, 0.79)	0.22
TAG(C48:3)	−0.03 (−0.46, 0.41)	0.90	−0.10 (−0.56, 0.36)	0.68	0.52 (0.20, 0.83)	0.0015	0.52 (0.20, 0.84)	0.0015	0.54 (0.10, 0.98)	0.017	0.64 (0.16, 1.12)	0.010
PC(D18:2‐18:2/D16:0‐20:4)	−0.50 (−0.94, −0.06)	0.026	−0.35 (−0.81, 0.12)	0.14	0.24 (−0.08, 0.56)	0.14	0.24 (−0.08, 0.56)	0.14	0.74 (0.32, 1.16)	0.0007	0.88 (0.42, 1.35)	0.0003

Open in a new tab

Note: β values, 95% CIs, and p values result from linear regression models. β values are interpreted as the difference in the mean analyte/lipid class sum (following rank transformation and subsequent scaling to mean = 0 and standard deviation = 1) in comparison to the reference group. Only analytes and lipid class sums with p < 0.01 in adjusted analysis are displayed. P < 0.00032 (CSF lipids), p < 0.00030 (plasma lipids), and p < 0.00044 (CSF/plasma ratio) were considered statistically significant after applying a Bonferroni correction for multiple testing.

Abbreviations: APOE, apolipoprotein E; BMI, body mass index; CI, confidence interval.

3.3. Impact of age, sex, and modifiable dementia risk factors on CSF and plasma lipids in cognitively unimpaired participants

Regarding demographic and modifiable risk factors, we found that age predominantly influenced CSF lipids, with subspecies of TAG and LPC in the top‐ranked lipids in multivariable analysis. Whereas older age was negatively associated with TAG(C49:1/C50:8) (β: −0.51, p = 1.32E‐07), TAG(C50:0/C51:7) (β: −0.40, p = 0.0001), TAG(C46:1/C47:8) (β: −0.39, p = 0.0001), and TAG(C49:2/C50:9) (β: −0.36, p = 0.0003), there were positive associations with LPC(18:1) (β: 0.42, p = 1.81E‐05), LPC(16:0) (β: 0.41, p = 3.69E‐05), and total LPC (β: 0.39, p = 0.0001) (Table 2). In addition, a significantly lower PC(D16:1‐16:0/D14:1‐18:0) (β: −0.52, p = 2.63E‐05) was observed in males. There were also several suggestive associations between sex and PC, with males having lower levels of total PC and PC subspecies, including PC(D16:0‐18:1), PC(D18:0‐18:2/D18:1‐18:1), and PC(P16:0‐18:1/P18:1‐16:0) (Table 2). On the other hand, plasma lipids were strongly affected by sex, where males had significantly lower plasma levels of SM lipid class sum (β: −1.07, p = 8.78E‐17) and other SM species as well as TC (β: −0.94, p = 1.22E‐13) than females. Significant negative associations of age with plasma PE(P16:0‐20:4) (β: −0.52, p = 1.47E‐05) and suggestive negative associations with plasma PE species were also detected (Table 2). CSF/plasma lipid ratios of total FA and TAG(C55:0/C56:7) were decreased and increased with age, respectively. CSF/plasma lipid ratios of CE(20:5) and CE(18:0) were increased in males (Table 2). While no evident correlations of modifiable dementia risk factors with CSF lipids were observed in the cohort (Table S4), there were significant negative associations of diabetes with plasma levels of CE(18:2) (β: −0.67, p = 0.0001), TC (β: −0.64, p = 0.0001), and SM(N24:2) (β: −0.58, p = 0.0003) and positive associations of dyslipidemia with plasma levels of PE(D18:0‐22:6/D18:1‐22:5) (β: 0.80, p = 1.47E‐05) and PE(D16:0‐22:6) (β: 0.71, p = 0.0001) (Table S5). There were no notable associations between modifiable risk factors and CSF/plasma lipid ratios (Table S6).

TABLE 2.

Associations of CSF and plasma lipids with age and sex.

	Unadjusted analysis		Adjusting for age, sex, years of education, BMI, smoking, hypertension, diabetes, dyslipidemia, and APOE group
	β (95% CI)	p value	β (95% CI)	p value
CSF lipids
Associations with age
TAG(C49:1/C50:8)	−0.55 (−0.72, −0.38)	1.19E‐09	−0.51 (−0.69, −0.32)	1.32E‐07
LPC(18:1)	0.46 (0.28, 0.63)	6.77E‐07	0.42 (0.23, 0.62)	1.81E‐05
LPC(16:0)	0.44 (0.26, 0.61)	2.02E‐06	0.41 (0.22, 0.60)	3.69E‐05
TAG(C50:0/C51:7)	−0.35 (−0.53, −0.17)	0.0002	−0.40 (−0.60, −0.21)	0.0001
LPC lipid class sum	0.43 (0.25, 0.60)	3.44E‐06	0.39 (0.20, 0.58)	0.0001
TAG(C46:1/C47:8)	−0.41 (−0.58, −0.23)	1.05E‐05	−0.39 (−0.58, −0.19)	0.0001
CE(22:5)	0.43 (0.25, 0.61)	2.95E‐06	0.37 (0.18, 0.56)	0.0002
PE(D18:0‐20:4/D16:0‐22:4)	−0.39 (−0.57, −0.22)	1.98E‐05	−0.36 (−0.55, −0.17)	0.0002
PE(P16:0‐20:4)	−0.34 (−0.52, −0.16)	0.0002	−0.36 (−0.55, −0.17)	0.0002
TAG(C49:2/C50:9)	−0.37 (−0.55, −0.19)	0.0001	−0.36 (−0.55, −0.17)	0.0003
Associations with sex
PC(D16:1‐16:0/D14:1‐18:0)	−0.49 (−0.72, −0.25)	4.88E‐05	−0.52 (−0.76, −0.28)	2.63E‐05
PC lipid class sum	−0.43 (−0.66, −0.20)	0.0004	−0.43 (−0.67, −0.20)	0.0004
PC(D16:0‐18:1)	−0.40 (−0.63, −0.16)	0.0010	−0.41 (−0.65, −0.17)	0.0009
PC(D18:0‐18:2/D18:1‐18:1)	−0.37 (−0.61, −0.14)	0.0019	−0.37 (−0.61, −0.13)	0.0028
PC(P16:0‐18:1/P18:1‐16:0)	0.32 (0.08, 0.55)	0.0090	0.36 (0.12, 0.60)	0.0030
LPC(18:0)	0.39 (0.15, 0.62)	0.0014	0.33 (0.09, 0.57)	0.0064
PC(D18:0‐18:1)	−0.32 (−0.56, −0.08)	0.0081	−0.33 (−0.57, −0.09)	0.0070
PC(D18:2‐18:2/D16:0‐20:4)	0.27 (0.03, 0.51)	0.027	0.33 (0.08, 0.57)	0.0089
PE(D18:0‐20:4/D16:0‐22:4)	−0.33 (−0.56, −0.09)	0.0068	−0.32 (−0.55, −0.08)	0.0093
Plasma lipids
Associations with age
PE(P16:0‐20:4)	−0.52 (−0.74, −0.29)	1.12E‐05	−0.52 (−0.75, −0.29)	1.47E‐05
SM(N24:0)	−0.42 (−0.65, −0.19)	0.0004	−0.41 (−0.64, −0.18)	0.0006
SM(N22:0)	−0.44 (−0.67, −0.21)	0.0002	−0.40 (−0.63, −0.17)	0.0006
SM(N22:2)	0.25 (0.02, 0.48)	0.037	0.33 (0.12, 0.53)	0.0018
SM(N21:1)	0.19 (−0.04, 0.43)	0.11	0.30 (0.10, 0.50)	0.0042
PC(D18:1‐22:6/D18:2‐22:5)	0.36 (0.13, 0.59)	0.0023	0.33 (0.10, 0.56)	0.0051
PE(P18:0‐20:4/P16:0‐22:4/P18:1‐20:3)	−0.34 (−0.57, −0.11)	0.0041	−0.34 (−0.58, −0.10)	0.0052
PE(P18:1‐20:4/P16:0‐22:5)	−0.32 (−0.56, −0.09)	0.0066	−0.34 (−0.57, −0.10)	0.0056
PE(P16:0‐20:3/P18:1‐18:2)	−0.31 (−0.54, −0.08)	0.0088	−0.33 (−0.56, −0.10)	0.0058
Associations with sex
SM(N16:1)	−1.35 (−1.55, −1.15)	2.03E‐29	−1.36 (−1.57, −1.15)	6.78E‐28
SM(N22:1)	−1.22 (−1.44, −1.00)	8.29E‐23	−1.17 (−1.39, −0.95)	6.03E‐21
SM(N23:1)	−1.20 (−1.42, −0.98)	8.21E‐22	−1.16 (−1.38, −0.93)	7.81E‐20
SM(N20:1)	−1.16 (−1.39, −0.94)	1.78E‐20	−1.13 (−1.36, −0.90)	3.85E‐18
SM(N24:2)	−1.14 (−1.36, −0.91)	2.20E‐19	−1.06 (−1.28, −0.83)	2.71E‐17
SM(N18:1)	−1.11 (−1.33, −0.88)	2.97E‐18	−1.11 (−1.35, −0.88)	2.86E‐17
SM lipid class sum	−1.11 (−1.34, −0.88)	1.93E‐18	−1.07 (−1.30, −0.84)	8.78E‐17
SM(N21:1)	−1.04 (−1.27, −0.80)	6.45E‐16	−1.00 (−1.24, −0.76)	1.23E‐14
SM(N22:2)	−1.02 (−1.26, −0.79)	2.12E‐15	−0.96 (−1.20, −0.72)	1.12E‐13
TC	−1.03 (−1.26, −0.80)	1.11E‐15	−0.94 (−1.17, −0.70)	1.22E‐13
CSF/plasma ratio
Associations with age
FA lipid class sum	−0.39 (−0.63, −0.15)	0.0018	−0.36 (−0.61, −0.11)	0.0046
TAG(C55:0/C56:7)	0.37 (0.13, 0.61)	0.0029	0.37 (0.11, 0.62)	0.0046
Associations with sex
CE(20:5)	0.36 (0.08, 0.64)	0.012	0.41 (0.12, 0.70)	0.0059
CE(18:0)	0.34 (0.07, 0.62)	0.016	0.41 (0.11, 0.70)	0.0066

Open in a new tab

Note: β values, 95% CIs, and p values result from linear regression models. β values are interpreted as the change in the mean analyte/lipid class sum (following rank transformation and subsequent scaling to mean = 0 and standard deviation = 1) corresponding to each 10‐year increase in age or as the difference in the mean analyte/lipid class sum rank for males in comparison to females. Only analytes and lipid class sums with p < 0.01 in adjusted analysis are displayed. P < 0.00032 (CSF lipids), p < 0.00030 (plasma lipids), and p < 0.00044 (CSF/plasma ratio) were considered statistically significant after applying a Bonferroni correction for multiple testing; statistically significant findings are shown in bold. The 10 analytes/lipid class sums that had the lowest p values are displayed for the association with age (CSF lipids) or sex (plasma lipids).

Abbreviations: Aβ, amyloid beta; APOE, apolipoprotein E; BMI, body mass index; CSF, cerebrospinal fluid; CI, confidence interval; FA, fatty acid.

3.4. Associations of CSF and plasma lipids with CSF AD biomarkers in cognitively unimpaired participants

Significant associations between CSF lipids and each AD biomarker were identified except for NfL. Specifically, in multivariable analysis adjusting for demographic variables and modifiable dementia risk factors, higher CSF Aβ42 was associated with increased PS(20:0‐22:5) but decreased TAG(C48:0/C49:7) and TAG(C49:1/C50:8). In addition, PE(D18:0‐22:6/D18:1‐22:5), PE(D20:0‐20:4/D18:0‐22:4), and PE(D18:0‐20:4/D16:0‐22:4) were identified as common top‐ranked lipids for total tau, p‐tau181, Ng, and YKL40, where positive correlations were noted in relation to each CSF AD marker. Increased PE(P18:0‐22:6/P18:1‐22:5/D18:0‐20:0) was associated with higher CSF levels of SNAP25 as well as total tau, p‐tau181, and Ng. SNAP25 was also positively correlated with PG(18:0‐18:1), PG(16:0‐18:1), and SM(N18:0). There were negative correlations of PC(P16:0‐18:1/P18:1‐16:0) with SNAP25 and YKL40 (Table 3). Significant associations of plasma lipids with CSF and plasma biomarkers were not detected (Table S7).

TABLE 3.

Associations of CSF lipids with CSF AD biomarkers.

	Unadjusted analysis		Adjusting for age, sex, years of education, BMI, smoking, hypertension, diabetes, dyslipidemia, and APOE group
CSF lipids	β (95% CI)	p value	β (95% CI)	p value
Associations with CSF Aβ42
PS(20:0‐22:5)	0.23 (0.11, 0.35)	0.0002	0.24 (0.12, 0.37)	0.0002
TAG(C48:0/C49:7)	−0.21 (−0.33, −0.10)	0.0005	−0.23 (−0.35, −0.11)	0.0003
TAG(C49:1/C50:8)	−0.25 (−0.37, −0.13)	4.22E‐05	−0.22 (−0.33, −0.10)	0.0003
PE(D16:0‐16:1/D14:0‐18:1)	−0.19 (−0.31, −0.07)	0.0019	−0.23 (−0.36, −0.10)	0.0004
PE(D20:0‐20:4/D18:0‐22:4)	0.20 (0.08, 0.32)	0.0009	0.23 (0.10, 0.35)	0.0004
Associations with CSF total tau
PE(D18:0‐22:6/D18:1‐22:5)	0.44 (0.33, 0.55)	2.99E‐14	0.55 (0.44, 0.67)	6.55E‐19
PE(D20:0‐20:4/D18:0‐22:4)	0.33 (0.22, 0.44)	1.75E‐08	0.46 (0.34, 0.58)	2.05E‐13
PE(D16:0‐22:6)	0.28 (0.17, 0.39)	2.18E‐06	0.42 (0.30, 0.54)	3.08E‐11
PE(P18:0‐22:6/P18:1‐22:5/D18:0‐20:0)	0.31 (0.19, 0.42)	1.68E‐07	0.42 (0.30, 0.54)	4.13E‐11
PE(D18:0‐20:4/D16:0‐22:4)	0.25 (0.14, 0.37)	2.64E‐05	0.40 (0.28, 0.52)	8.45E‐11
Associations with CSF p‐tau181
PE(D18:0‐22:6/D18:1‐22:5)	0.40 (0.29, 0.51)	5.63E‐12	0.50 (0.39, 0.62)	8.53E‐16
PE(D20:0‐20:4/D18:0‐22:4)	0.29 (0.17, 0.40)	1.08E‐06	0.40 (0.28, 0.52)	1.80E‐10
PE(P18:0‐22:6/P18:1‐22:5/D18:0‐20:0)	0.29 (0.17, 0.40)	1.08E‐06	0.39 (0.27, 0.51)	7.63E‐10
PE(D16:0‐22:6)	0.25 (0.14, 0.37)	2.04E‐05	0.38 (0.26, 0.50)	1.83E‐09
PE(D18:0‐20:4/D16:0‐22:4)	0.22 (0.10, 0.34)	0.0002	0.36 (0.24, 0.47)	8.64E‐09
Associations with CSF NfL
LPC(P16:1)	−0.25 (−0.37, −0.13)	4.58E‐05	−0.18 (−0.31, −0.05)	0.0086
Associations with CSF SNAP25
PE(P18:0‐22:6/P18:1‐22:5/D18:0‐20:0)	0.29 (0.18, 0.40)	8.90E‐07	0.38 (0.25, 0.50)	3.93E‐09
PG(18:0‐18:1)	0.33 (0.21, 0.44)	2.53E‐08	0.37 (0.25, 0.50)	6.90E‐09
SM(N18:0)	0.30 (0.19, 0.42)	2.37E‐07	0.34 (0.22, 0.47)	7.53E‐08
PG(16:0‐18:1)	0.30 (0.19, 0.41)	3.68E‐07	0.33 (0.21, 0.46)	2.90E‐07
PC(P16:0‐18:1/P18:1‐16:0)	−0.26 (−0.37, −0.14)	1.51E‐05	−0.32 (−0.44, −0.20)	4.50E‐07
Associations with CSF Ng
PE(D18:0‐22:6/D18:1‐22:5)	0.43 (0.32, 0.54)	3.68E‐14	0.47 (0.36, 0.57)	1.62E‐15
PE(D20:0‐20:4/D18:0‐22:4)	0.34 (0.23, 0.45)	5.72E‐09	0.39 (0.28, 0.50)	3.58E‐11
PE(D18:0‐20:4/D16:0‐22:4)	0.31 (0.20, 0.42)	8.93E‐08	0.37 (0.27, 0.48)	9.66E‐11
PE(D18:0‐22:5/D18:1‐22:4)	0.33 (0.22, 0.44)	1.11E‐08	0.37 (0.25, 0.48)	1.43E‐09
PE(P18:0‐22:6/P18:1‐22:5/D18:0‐20:0)	0.28 (0.16, 0.39)	2.41E‐06	0.33 (0.21, 0.44)	6.14E‐08
Associations with CSF YKL40
PE(D18:0‐22:6/D18:1‐22:5)	0.24 (0.13, 0.36)	3.34E‐05	0.37 (0.24, 0.49)	1.79E‐08
PC(P16:0‐18:1/P18:1‐16:0)	−0.25 (−0.36, −0.13)	3.06E‐05	−0.29 (−0.41, −0.16)	1.05E‐05
PE(D20:0‐20:4/D18:0‐22:4)	0.14 (0.03, 0.26)	0.016	0.29 (0.16, 0.42)	1.21E‐05
PE(D18:0‐22:5/D18:1‐22:4)	0.19 (0.08, 0.31)	0.0013	0.29 (0.16, 0.42)	2.11E‐05
PE(D18:0‐20:4/D16:0‐22:4)	0.10 (−0.02, 0.22)	0.094	0.27 (0.15, 0.40)	2.47E‐05

Open in a new tab

Note: β values, 95% CIs, and p values result from linear regression models. β values are interpreted as the change in the mean analyte/lipid class sum (following rank transformation and subsequent scaling to mean = 0 and standard deviation = 1) corresponding to each 1‐standard deviation increase in the given AD biomarker (following natural logarithm transformation for NfL, SNAP25, and Ng). P < 0.00032 were considered as statistically significant after applying a Bonferroni correction for multiple testing; statistically significant findings are shown in bold. Only analytes and lipid class sums with p < 0.01 in adjusted analysis are shown. If more than 5 analytes/lipid class sums had p < 0.01 in adjusted analysis, the five analytes/lipid class sums that had the lowest p values are displayed.

Abbreviations: Aβ, amyloid beta; APOE, apolipoprotein E; BMI, body mass index; CI, confidence interval; CSF, cerebrospinal fluid.

Next, a weighted correlation network analysis of lipidomics data was performed to investigate the correlation between CSF and plasma lipids and APOE, age, sex, modifiable dementia risk factors, and CSF AD biomarkers. We identified four modules (MEturquoise, MEblue, MEbrown, and MEyellow) in CSF lipidomes (Figure 2A). Module yellow enriched for TAG subspecies with a total of 46 to 51 carbon atoms was negatively associated with age and CSF levels of Aβ42, total tau, Ng, and YKL40. BMI showed a positive correlation with Module turquoise enriched for TAG subspecies with a total of 50 to 55 carbon atoms, but a negative correlation with module blue enriched for SM. The hub lipids in module brown included PC subclasses, which were negatively associated with the presence of dyslipidemia and increased CSF Ng levels (Figure 2B). When plasma lipidomes are similarly analyzed, three modules (MEbrown, MEblue, and MEturquoise) were identified (Figure 2C). While the three modules were positively associated with BMI, hypertension, diabetes, and/or dyslipidemia, module turquoise was mainly composed of TAG subspecies with a total of 52 to 56 carbon atoms. Module blue enriched for TAG subspecies with a total of 48 to 52 carbon atoms was negatively associated with age and CSF levels of SNAP25 and YKL40 (Figure 2D).

Weighted correlation network analysis between CSF and plasma lipids, *APOE*, age, sex, modifiable dementia risk factors, and AD biomarkers. The correlation between module eigengenes (MEs), *APOE*, age, sex, modifiable dementia risk factors, and CSF AD biomarkers through weighted correlation network analysis in the CSF (A and B) and plasma (C and D) lipidomics data is shown. The values in the heatmap are Pearson correlation coefficients (with nominal p values). Network plots of the top 10 lipids with the highest intramodular connectivity in each module in the CSF (B) and plasma (D) lipidomics are shown.

3.5. Prediction for CDR conversion based on CSF lipids

In the subset of 270 cognitively unimpaired participants with longitudinal CDR information, which overlaps with our published study, ¹⁹ CDR conversion from 0 to >0 was observed in 79 subjects (29.3%). The median number of follow‐up visits was seven (range: 2 to 13 visits), and the median follow‐up length after the baseline visit was 7.8 years (range: 1.2 to 16.5 years) (Table S8). In multivariable analysis, there was a strong association between older age and an increased risk of CDR conversion (HR = 2.90, p = 1.56E‐07), where the c‐index for age in predicting CDR conversion was 0.708. Additionally, increased CSF Aβ42 was notably associated with a lower risk of CDR conversion (HR = 0.63, p = 0.0002, c‐index = 0.608), whereas higher CSF NfL was associated with a greater risk of CDR conversion (HR = 1.41, p = 0.0026, c‐index = 0.676) consistent with our previous work. ¹⁹ No significant associations with CDR conversion were observed in multivariable analysis for other modifiable dementia risk factors or CSF AD biomarkers (Table S9). With regard to associations between CSF lipids and CDR conversion, in the multivariable analysis adjusting for demographic variables and dementia risk factors, there was a statistically significant (p < 0.00032) association between TAG(C49:6) and risk of CDR conversion (HR = 0.63, p = 0.0001), where the c‐index for this analyte was 0.624. Additionally, suggestive (p < 0.01) associations with CDR conversion were observed for SM(N16:1) (HR = 0.67, p = 0.002, c‐index = 0.536), PG(18:0‐18:1) (HR = 0.67, p = 0.002, c‐index = 0.557), CE(20:3) (HR = 1.48, p = 0.004, c‐index = 0.597), CE(20:0) (HR = 1.43, p = 0.005, c‐index = 0.602), PE(D16:0‐18:1) (HR = 1.39, p = 0.008, c‐index = 0.571), and SM(N18:1) (HR = 0.73, p = 0.009). Of note, when further adjusting multivariable models for CSF levels of Aβ42 and NfL (note that NfL was unavailable in 29 subjects, resulting in lower power in the smaller sample), the associations with CDR conversion for TAG(C49:6) (HR = 0.63, p = 0.0004) and CE(20:3) (HR = 1.44, p = 0.0063) remained relatively consistent (Table S10).

In assessments of all pair‐wise ratios of CSF lipid subspecies (p < 4.79E‐06 considered significant, p < 0.0001 considered suggestive), suggestive associations with CDR conversion were detected for SM(N16:1)/CE(20:3) (HR = 0.58, p = 1.05E‐05), CE(20:5)/TAG(C49:6) (HR = 1.71, p = 1.56E‐05), TAG(C47:1/C48:8)/TAG(C49:6) (HR = 1.65, p = 2.39E‐05), CE(18:0)/TAG(C49:6) (HR = 1.71, p = 2.92E‐05), CE(20:3)/TAG(C49:6) (HR = 1.64, p = 3.22E‐05), CE(20:3)/PG(18:0‐18:1) (HR = 1.74, p = 3.25E‐05), CE(20:0)/TAG(C49:6) (HR = 1.67, p = 3.79E‐05), TAG(C49:6)/TAG(C51:1/C52:8) (HR = 0.62, p = 4.30E‐05), CE(16:0)/TAG(C49:6) (HR = 1.62, p = 5.95E‐05), CE(18:3)/TAG(C49:6) (HR = 1.62, p = 7.65E‐05), and CE(14:0)/TAG(C49:6) (HR = 1.60, p = 9.61E‐05), where c‐indexes ranged from 0.595 to 0.641 (Table S11). The internally validated c‐index for a model including all 11 analyte ratios was 0.585. For the validation, we similarly performed lipidomics in CSF samples (N = 129) from cognitively unimpaired participants (CDR = 0, over 65 years old) in Knight ADRC at WashU School of Medicine in St. Louis (Tables S12 and S13). Among the six lipids and 11 lipid ratios, CSF level of TAG(C49:6) (HR = 0.75, p = 0.074) (Table S14), CE(20:5)/TAG(C49:6) ratio (HR = 1.36, p = 0.068), CE(18:0)/TAG(C49:6) ratio (HR = 1.40, p = 0.061), TAG(C49:6)/TAG(C51:1/C52:8) ratio (HR = 0.71, p = 0.058), and CE(14:0)/TAG(C49:6) ratio (HR = 1.43, p = 0.081) (Table S15) showed non‐significant trends in predicting CDR conversion risk in a consistent direction to the results from the MCSA cohort.

We subsequently included age, APOE, sex, modifiable dementia risk factors, and the six lipid subspecies that were suggestively associated with CDR conversion in the aforementioned multivariable analysis in a forward selection procedure in order to identify the strongest independent predictors of CDR conversion in the MCSA cohort. In that analysis, statistically significant or suggestive associations were observed for age (HR = 2.90, p = 1.79E‐08), TAG(C49:6) (HR = 0.62, p = 0.0002, Figure 3A), PG(18:0‐18:1) (HR = 0.64, p = 0.0004, Figure 3B), CE(20:3) (HR = 1.49, p = 0.001, Figure 3C), and PE(D16:0‐18:1) (HR = 1.40, p = 0.009, Figure 3D). The internally validated c‐index of that model was 0.683 (Figure 3E). When considering observed c‐index, though there was a clinically significant improvement from a model including age alone to the full model including age and the aforementioned four lipid subspecies (observed c‐index: 0.753 vs 0.708), this did not quite reach statistical significance (p = 0.074). When performing forward selection including age, APOE, sex, modifiable dementia risk factors, the six lipid subspecies, and the seven AD biomarkers in the subset of 241 individuals with all seven biomarkers available, the resulting model identified significant or suggestive associations with CDR conversion for age (HR = 2.21, p = 3.30E‐05), diabetes (HR = 2.13, p = 0.005), TAG(C49:6) (HR = 0.59, p = 9.00E‐05), PG(18:0‐18:1) (HR = 0.70, p = 0.006), Aβ42 (HR = 0.53, p = 2.13E‐06), and NfL (HR = 1.57, p = 7.43E‐05), with an internally validated c‐index of 0.730 (Figure 3F). We also included all demographic variables and dementia risk factors and the 11 CSF lipid ratios with p < 0.0001 that were associated with CDR conversion in the aforementioned multivariable analysis in a forward selection procedure. Independent associations were observed for age (HR = 2.88, p = 6.28E‐09) and SM(N16:1)/CE(20:3) ratio (HR = 0.60, p = 2.18E‐05), and the internally validated c‐index of a model that includes these two variables is 0.600 (Figure 3G,H). However, adding SM(N16:1)/CE(20:3) ratio did not improve the prediction of CDR conversion to that observed for only age and CSF levels of Aβ42 and NfL (observed c‐index: 0.772 vs 0.760, p = 0.28).

Association of CSF and plasma lipids with cumulative incidence of CDR conversion after baseline visit. Kaplan–Meier curves of conversion from CDR = 0 to CDR > 0 after baseline visit according to lipids in CSF and plasma are shown. (A–D) The cohort used for CSF lipidomics was dichotomized by the median value of TAG(C49:6) (A), PG(18:0‐18:1) (B), CE(20:3) (C), or PE(D16:0‐18:1) (D) in CSF. (E and F) Kaplan–Meier curve of CDR conversion was plotted based on the number of risk factors including age and the four CSF lipids, TAG(C49:6), PG(18:0‐18:1), CE(20:3), and PE(D16:0‐18:1) (E), or the number of selected risk factors and CSF biomarkers/lipids including age, diabetes, Aβ42, NfL, TAG(C49:6), and PG(18:0‐18:1) (F) for subjects without any missing data for any of these variables. (G) The cohort used for CSF lipidomics was dichotomized by the median value of CSF SM(N16:1)/CE(20:3) ratio. (H) Kaplan–Meier curve of the CDR conversion was plotted based on the number of risk factors including age and CSF SM(N16:1)/CE(20:3) ratio. (I–K) The cohort used for plasma lipidomics was dichotomized by the median value of SM(N16:0) (I), SM(N21:1) (J), or SM(N22:1) (K) in plasma. (L) Kaplan–Meier curve of the CDR conversion was plotted based on the number of risk factors including age and plasma SM(N16:0).

3.6. Prediction for CDR conversion based on plasma lipids

We also investigated whether plasma lipid levels are associated with the CDR conversion risk in the subset of 209 cognitively unimpaired participants with longitudinal CDR information from MCSA (Table S16). With a median length of follow‐up after the baseline visit of 7.7 years (range: 1.0 to 14.7 years), CDR conversion was observed in 40 subjects (19.1%). In multivariable analysis, there was a significant association between older age and an increased risk of CDR conversion (HR = 2.88, p = 3.33E‐05, c‐index = 0.691), while higher plasma NfL was associated with a greater risk of CDR conversion (HR = 1.58, p = 0.0080, c‐index = 0.641) (Table S17). In multivariable analysis adjusting for age, sex, and APOE, suggestive associations with CDR conversion were observed for plasma levels of SM(N16:0) (HR = 0.61, p = 0.0057; Figure 3G), SM(N21:1) (HR = 0.60, p = 0.0093; Figure 3H), and SM(N22:1) (HR = 0.58, p = 0.0095; Figure 3I), with c‐indexes of these three analytes ranging from 0.614 to 0.637. Associations of analytes with CDR conversion were similar after adjusting models for plasma NfL (Table S18). When performing forward selection including age, APOE, sex, modifiable dementia risk factors, and these three lipid subspecies to identify the strongest independent associations with CDR conversion, statistically significant or suggestive associations were noted for age (HR = 2.96, p = 3.24E‐05) and plasma SM(N16:0) (HR = 0.59, p = 0.0013) with an internally validated c‐index of the model including both of these measures of 0.608 (Figure 3J). There was not a significant improvement in observed c‐index for the full model including age and SM(N16:0) compared to a reduced model including age alone (observed c‐index: 0.723 vs 0.691, p = 0.35). Additional inclusion of plasma NfL in this full model did not noticeably improve ability to predict CDR conversion (observed c‐index: 0.724 vs 0.723, p = 0.99). There were no significant (p < 4.19E‐06) or suggestive (p < 0.0001) associations between any plasma lipid ratios and CDR conversion.

4. DISCUSSION

Brain lipid metabolism represents central pathways affected in various neurodegenerative diseases. Although the amounts of total lipids in CSF are much lower than those in serum despite the abundance of lipid components in the brain, ³¹ , ³² altered CSF lipid profiles have been identified in multiple sclerosis, ³³ amyotrophic lateral sclerosis, ³² Parkinson's disease, ³⁴ and AD. ³⁵ , ³⁶ Our findings demonstrate that CSF levels of TAG(C49:6), PG(18:0‐18:1), CE(20:3), and PE(D16:0‐18:1) are associated with the risk of CDR conversion independently of age, APOE genotype, sex, and modifiable dementia risk factors in cognitively unimpaired individuals. While previous studies have established age, CSF Aβ42, and NfL as strong predictors of cognitive decline, ¹⁸ , ¹⁹ , ³⁷ it is noteworthy that the inclusion of TAG(C49:6) and PG(18:0‐18:1) improves the predictive power of the models already accounting for age, diabetes, CSF Aβ42, and NfL. Removing the two lipid factors from the model reduced the c‐index to 0.770, indicating that their inclusion provides an improvement of 0.033 in predictive accuracy. We found that lower CSF levels of TAG(C49:6) were associated with higher risk of CDR conversion. Several TAG subspecies are also negatively associated with age. A longitudinal study showed that higher plasma/serum TAG levels within the normal to high‐normal range were associated with lower risk of dementia and slower progression of cognitive decline in older participants without dementia. ³⁸ However, the relationships between TAG and dementia are controversial. ³⁹ Notably, plasma TAG subspecies were not associated with CDR conversion risk in our study. Instead, the weighted correlation network analysis in plasma lipidomics found that modules enriched for TAG subspecies positively correlated with modifiable dementia risk factors such as high BMI, hypertension, diabetes, and dyslipidemia. While prolonged high TAG levels in blood might correlate with metabolic syndrome and vascular damage contributing to dementia risk, a modest increase of TAG in the brain may induce neurotrophic effects by improving energy metabolism in older adults. PG(18:0‐18:1) also has a negative association with CDR conversion risk. PG(18:0‐18:1) is a key subspecies for de novo cardiolipin synthesis, whereas cardiolipin is crucial for mitochondria function and homeostasis. ⁴⁰ , ⁴¹ Thus, higher CSF levels of PG(18:0‐18:1) may indicate better mitochondria functions. Although TAG in CSF is relatively less explored, one study reported that CSF levels of ether‐linked triglyceride TAG(O52:2) were positively associated with the progression from mild cognitive impairment (MCI) to AD. ³⁶ Ether‐linked lipids and CE are major components of lipid droplets, the accumulation of which is often detected in the brain of neurodegenerative diseases. ⁴² Indeed, we also found a positive association between CE(20:3) and CDR conversion risk.

A set of plasma phospholipids composed of PC, LPC, and acylcarnitine subspecies have been shown to predict the conversion risk to MCI or AD in the elderly. ⁴³ Our study found that lower plasma levels of SM(N16:0), SM(N21:1), and SM(N22:1), but not PC, LPC, and acylcarnitine subspecies, were associated with the increased risk of cognitive decline in the old cognitively unimpaired participants, supported by the decreased plasma SM subspecies in AD cases compared to control. ⁴⁴ Higher plasma levels of SM and SM/ceramide ratio have also been shown to associate with slower progression of cognitive decline in AD. ⁴⁵ However, there is a conflicting report showing the positive association between serum levels of total SM and ceramides and the incident of cognitive impairment in old females. ⁴⁶ Higher plasma SM(d18:1/16:0) levels were related to a faster cognitive decline in community‐dwelling old adults, although higher plasma SM(d18:1/22:0) levels correlated with slower decline. ⁴⁷ While our lipidomics do not distinguish SM(d16:1) and SM(d18:1), plasma d18:1 species are likely increased in AD despite reduced plasma d16:1 species in vascular dementia. ⁴⁸ These results suggest that altered sphingolipid metabolism is involved in the pathogenesis of age‐related cognitive decline and AD. Sphingomyelinase hydrolyzes SM into phosphorylcholine and ceramide, which can be similarly reversed converted to SM with substrates PC and ceramide by SM synthase. ⁴⁹ As increased ceramides induce cellular stress, ⁵⁰ higher plasma ceramide C22:0 and C24:0 levels likely associate with exacerbated memory loss and right hippocampal volume loss in MCI subjects. ⁵¹ Developing more sensitive assays to monitor sphingolipid metabolism and/or the enzymatic activities in plasma may be necessary to establish effective sphingolipid‐targeted biomarkers for the pathogenic conditions. In addition, we found that plasma levels of total SM and subspecies are significantly lower in males than females, indicating the importance in considering sex‐dependent effects on the sphingolipid‐mediated pathways in the pathogenesis.

AD is characterized by brain deposition of Aβ and tau, accompanied by synaptic damage, neurodegeneration, and neuroinflammation, ⁵² , ⁵³ supported by significant changes in fluid biomarkers related to those neuropathological features. Interestingly, we found PE subspecies, including PE(D18:0‐22:6/D18:1‐22:5), PE(D20:0‐20:4/D18:0‐22:4), and PE(D18:0‐20:4/D16:0‐22:4), in CSF had strong positive associations with CSF p‐tau181, Ng, and YKL40. Supporting our findings, a meta‐analysis revealed that higher plasma/serum levels of PE including PE(18:0‐22:5) and PE(16:1‐22:5) were associated with an increased risk of AD incidence. ⁵⁴ PE is the second most abundant lipid accounting for 15% to 25% of total lipids in mammalian cells next to PC. While PE is ubiquitously detected in membrane lipid bilayers, particularly in the inner layer of the plasma and mitochondrial membranes, ⁵⁵ , ⁵⁶ a multi‐omics study showed the abundant presence of PE in inhibitory and excitatory neurons in adult human brains. ⁵⁷ As PE is enriched in synaptic vesicles, ⁵⁸ the increased CSF levels of PE and the AD biomarkers may represent synaptic damage or excess release of extracellular vesicles from synapses. PE is synthesized from ethanolamine through the CDP‐ethanolamine pathway in the endoplasmic reticulum (ER) or from PS through the decarboxylase pathway in mitochondria. ⁵⁶ PE subspecies with polyunsaturated fatty acids at position sn‐2 are likely more predominant in mitochondria than in the ER. ⁵⁹ Thus, our results suggest that CSF AD biomarkers sense synaptic mitochondrial injuries indicated by the increase of PE subspecies. Furthermore, the weighted correlation network analysis found a negative association of TAG subspecies with 46 to 51 carbon atoms with age and some of the CSF AD biomarkers but a positive association between BMI and TAG subspecies with 50 to 55 carbon atoms. Future studies should also dissect potentially different roles of TAG depending on their acyl chain compositions in the development of metabolic syndrome and age‐related cognitive decline. In addition, high TAG levels may be beneficial for brain activity, particularly when cells can effectively utilize those lipids. Excess TAG accumulates in intracellular spaces beyond their metabolic capacity may instead cause lipotoxicity, although further studies are needed. In wild‐type mice, there is a trend of age‐dependent decreases of TAG(C49:1), TAG(C50:0), TAG(C46:1), and TAG(C49:2) in the brains after 16 weeks of age ⁶⁰ consistent with our finding from human CSF samples.

The APOE genotype substantially impacts the risk of age‐related cognitive decline and AD. ⁶¹ , ⁶² As apoE isoforms differently regulate brain lipid metabolism, human induced pluripotent stem cell models have demonstrated that APOE4 exacerbates intracellular lipid accumulation and lipid droplet formation compared with APOE3 in astrocytes, microglia, and oligodendrocytes. ⁶³ , ⁶⁴ , ⁶⁵ Mouse models also showed aggravated glial lipid accumulation in MAPT P301S transgenic mice with APOE4, and administration of an LXR agonist increased ABCA1 expression, enhanced lipid efflux, and ameliorated tauopathy and neurodegeneration. ⁶⁶ Although APOE effects on CSF lipids were modest in our cohort, CE(14:0) was higher in the APOE2 group but lower in the APOE4 group than the APOE3 group. The observation may suggest the less effective ability of apoE4 to eliminate CE with shorter acyl chains from lipid droplets. Interestingly, we also found the higher CSF/plasma ratios of various TAG subspecies in APOE4 group than APOE3 group. Further studies to determine how APOE4 alters TAG balance between brain and blood may provide a clue for better understanding of the pathogenic mechanisms by which APOE4 increases the risk of cognitive impairment and AD.

Our study is not without its limitations. Specifically, although the sample sizes utilized were not overly small, the possibility of a Type II error (i.e., a false‐negative finding) is still important to consider for certain analyses, specifically analysis of CDR conversion in the plasma lipids and CSF replication datasets, where power to detect associations is lower due to the smaller number of subjects who experienced CDR conversion; we cannot conclude that a true association does not exist simply due to the occurrence of a non‐significant p value in our study, and emphasis is best placed on 95% confidence limits when interpreting results. Nonetheless, we observed similar trends in predicting CDR conversion risk for CSF TAG(C49:6) levels and the ratios of CE(20:5)/TAG(C49:6), CE(18:0)/TAG(C49:6), TAG(C49:6)/TAG(C51:1/C52:8), and CE(14:0)/TAG(C49:6) in the Knight ADRC cohort at WashU School of Medicine in St. Louis, supporting the findings from the MCSA cohort. In addition, as this study was conducted in geographically defined non‐Hispanic/Latino and White cohorts, our findings may not accurately capture lipid metabolism across broader populations. Furthermore, the lack of standardization regarding sample collection timing may have influenced lipid profiles and should be considered when interpreting the findings.

In summary, our study provides novel evidence that lipid metabolism is involved in the pathogenic cascade of cognitive decline during aging. CSF and plasma lipids can be potential biomarkers to improve the prediction of age‐related cognitive decline. While CSF and plasma levels of some lipids such as CE and PC subclasses correlated well, CSF and plasma lipids were differently associated with age, sex, APOE, modifiable dementia risk factors, and fluid AD biomarkers. Thus, brain lipid metabolism may be regulated independently of the peripheral system. Defining the contributions of specific lipid species to the pathogenic pathways will enable the development of new therapeutic interventions. Since our findings are based on cohorts with relatively less diversity, further studies are required to confirm our findings in more diverse populations.

AUTHOR CONTRIBUTIONS

TK, JDU, JP, DMH, GB, and XH designed the research studies. TK, LJW, YR, JDU, CX, YWA, and MGH analyzed all data. HY and XH performed lipidomics. ACR, AA, and FSA acquired data. TK, YR, and MGH visualized results. MV, AAS, SES, DMH, and RCP administered the project. TK, MGH, and XH wrote the first draft. All authors contributed to writing the final manuscript.

CONSENT STATEMENT

We used deidentified datasets and CSF and plasma samples from MCSA and Knight ADRC. All participants signed informed and repository consent.

CONFLICT OF INTEREST STATEMENT

M.V. served as a consultant for F. Hoffmann‐La Roche Ltd., unrelated to this manuscript and has equity ownership in Amgen, Johnson & Johnson, and Medtronic. A.A.S. served on scientific advisory boards for Roche Diagnostics, Fujirebio Diagnostics, Siemens Healthineers, and Beckman Coulter. S.E.S. has recently received honoraria for educational presentations and/or advising from Medscape, PeerView, the Academy for Continued Healthcare Learning, Med Learning Group, the Global Learning Collaborative, and multiple academic institutions. S.E.S. has provided unpaid advising to Acumen, Biogen, Cognito Therapeutics, Danaher, Eisai, Eli Lilly, Johnson & Johnson Innovative Medicine, Roche, Sanofi, and Siemens. D.M.H. co‐founded and is on the scientific advisory board of C2N Diagnostics. D.M.H. is on the scientific advisory board of Denali, Cajal Neuroscience, and Genentech and consults for Asteroid Therapeutics. R.C.P. is a consultant for Roche, Inc., Genentech, Inc., Eli Lilly and Co., Eisai, Inc., Nestle, Inc., and Biogen. R.C.P. receives royalties from Oxford University Press and UpToDate and educational materials for Medscape. T.K. has served on scientific advisory boards for Path Biotech LLC. Other authors declare no competing interests. Author disclosures are available in the Supporting Information.

Supporting information

Supporting Information: alz71924‐sup‐0001‐SuppMat.pdf

ALZ-22-e71924-s003.pdf^{(383.6KB, pdf)}

Supporting Information: alz71924‐sup‐0002‐IcmjeDisclosureForm.pdf

ALZ-22-e71924-s002.pdf^{(775.3KB, pdf)}

Supporting Information

ALZ-22-e71924-s001.zip^{(1.4MB, zip)}

ACKNOWLEDGMENTS

We gratefully thank all the participants and their families. We acknowledge the contributions of the staff in Administration, Clinical, Biomarker, and Biostatistics Cores at MCSA and Mayo Clinic Alzheimer's Disease Research Center (U01 AG006786, P30 AG062677) and the contributions of the staff at the Clinical and Fluid Biomarker Cores at the Knight Alzheimer Disease Research Center (P01 AG03991, P30 AG066444, P01 AG026276). The Functional Lipidomics Core at the University of Texas Health San Antonio was partially supported by National Institute on Aging grants (P30AG013319 and P30AG044271). National Institutes of Health grant U19 AG069701 (TK, DMH) National Institutes of Health grant R01 AG061729 (XH) National Institutes of Health grant R01 AG071226 (TK) National Institutes of Health grant R01 AG057181 (TK) National Institutes of Health grant R01 AG068034 (TK) National Institutes of Health grant R01 AG081203 (TK) National Institutes of Health grant U01 AG006786 (RCP) National Institutes of Health grant P30 AG062677 (RCP) National Institutes of Health grant U19 AG102336 (RCP)

Contributor Information

Takahisa Kanekiyo, Email: kanekiyo.takahisa@mayo.edu.

Xianlin Han, Email: HanX@uthscsa.edu.

DATA AVAILABILITY STATEMENT

All data are available in the main text or the supplementary materials. Visualizations of associations between APOE and CSF/plasma lipids are available at the EPAAD website (https://epaad.org/lipid‐abundance‐database). The raw data are available upon reasonable request.

REFERENCES

1. Sastry PS, Lipids of nervous tissue: composition and metabolism. Prog Lipid Res. 1985;24:69‐176. [DOI] [PubMed] [Google Scholar]
2. Tracey TJ, Steyn FJ, Wolvetang EJ, Ngo ST, Neuronal lipid metabolism: multiple pathways driving functional outcomes in health and disease. Front Mol Neurosci. 2018;11:10. [DOI] [PMC free article] [PubMed] [Google Scholar]
3. Chew H, Solomon VA, Fonteh AN, Involvement of lipids in Alzheimer's disease pathology and potential therapies. Front Physiol. 2020;11:598. [DOI] [PMC free article] [PubMed] [Google Scholar]
4. Yoon JH, Seo Y, Jo YS, et al. Brain lipidomics: from functional landscape to clinical significance. Sci Adv. 2022;8:eadc9317. [DOI] [PMC free article] [PubMed] [Google Scholar]
5. Kunkle BW, Grenier‐Boley B, Sims R, et al. Genetic meta‐analysis of diagnosed Alzheimer's disease identifies new risk loci and implicates Abeta, tau, immunity and lipid processing. Nat Genet. 2019;51:414‐430. [DOI] [PMC free article] [PubMed] [Google Scholar]
6. Livingston G, Huntley J, Sommerlad A, et al. Dementia prevention, intervention, and care: 2020 report of the Lancet commission. Lancet. 2020;396:413‐446. [DOI] [PMC free article] [PubMed] [Google Scholar]
7. Roberts RO, Geda YE, Knopman DS, et al. The Mayo clinic study of aging: design and sampling, participation, baseline measures and sample characteristics. Neuroepidemiology. 2008;30:58‐69. [DOI] [PMC free article] [PubMed] [Google Scholar]
8. Olsson B, Lautner R, Andreasson U, et al. CSF and blood biomarkers for the diagnosis of Alzheimer's disease: a systematic review and meta‐analysis. Lancet Neurol. 2016;15:673‐684. [DOI] [PubMed] [Google Scholar]
9. Zetterberg H, Bendlin BB, Biomarkers for Alzheimer's disease‐preparing for a new era of disease‐modifying therapies. Mol Psychiatry. 2021;26:296‐308. [DOI] [PMC free article] [PubMed] [Google Scholar]
10. Park SA, Han SM, Kim CE, New fluid biomarkers tracking non‐amyloid‐beta and non‐tau pathology in Alzheimer's disease. Exp Mol Med. 2020;52:556‐568. [DOI] [PMC free article] [PubMed] [Google Scholar]
11. Sutphen CL, Jasielec MS, Shah AR, et al. Longitudinal cerebrospinal fluid biomarker changes in preclinical Alzheimer disease during middle age. JAMA Neurol. 2015;72:1029‐1042. [DOI] [PMC free article] [PubMed] [Google Scholar]
12. Schindler SE, Li Y, Todd KW, et al. Emerging cerebrospinal fluid biomarkers in autosomal dominant Alzheimer's disease. Alzheimers Dement. 2019;15:655‐665. [DOI] [PMC free article] [PubMed] [Google Scholar]
13. Sutphen CL, McCue L, Herries EM, et al. Longitudinal decreases in multiple cerebrospinal fluid biomarkers of neuronal injury in symptomatic late onset Alzheimer's disease. Alzheimers Dement. 2018;14:869‐879. [DOI] [PMC free article] [PubMed] [Google Scholar]
14. Hellwig K, Kvartsberg H, Portelius E, et al. Neurogranin and YKL‐40: independent markers of synaptic degeneration and neuroinflammation in Alzheimer's disease. Alzheimers Res Ther. 2015;7:74. [DOI] [PMC free article] [PubMed] [Google Scholar]
15. Janelidze S, Mattsson N, Stomrud E, et al. CSF biomarkers of neuroinflammation and cerebrovascular dysfunction in early Alzheimer disease. Neurology. 2018;91:e867‐e877. [DOI] [PMC free article] [PubMed] [Google Scholar]
16. Teunissen CE, Verberk IMW, Thijssen EH, et al. Blood‐based biomarkers for Alzheimer's disease: towards clinical implementation. Lancet Neurol. 2022;21:66‐77. [DOI] [PubMed] [Google Scholar]
17. Morris JC, The clinical dementia rating (CDR): current version and scoring rules. Neurology. 1993;43:2412‐2414. [DOI] [PubMed] [Google Scholar]
18. Kern S, Syrjanen JA, Blennow K, et al. Association of cerebrospinal fluid neurofilament light protein with risk of mild cognitive impairment among individuals without cognitive impairment. JAMA Neurol. 2019;76:187‐193. [DOI] [PMC free article] [PubMed] [Google Scholar]
19. Shue F, White LJ, Hendrix R, et al. CSF biomarkers of immune activation and Alzheimer's disease for predicting cognitive impairment risk in the elderly. Sci Adv. 2024;10:eadk3674. [DOI] [PMC free article] [PubMed] [Google Scholar]
20. Yang K, Cheng H, Gross RW, Han X, Automated lipid identification and quantification by multidimensional mass spectrometry‐based shotgun lipidomics. Anal Chem. 2009;81:4356‐4368. [DOI] [PMC free article] [PubMed] [Google Scholar]
21. Kawatani K, Holm ML, Starling SC, et al. ABCA7 deficiency causes neuronal dysregulation by altering mitochondrial lipid metabolism. Mol Psychiatry. 2023;3(29):809‐819. [DOI] [PMC free article] [PubMed] [Google Scholar]
22. Wang M, Han X, Multidimensional mass spectrometry‐based shotgun lipidomics. Methods Mol Biol. 2014;1198:203‐220. [DOI] [PMC free article] [PubMed] [Google Scholar]
23. Han X, Yang K, Gross RW, Microfluidics‐based electrospray ionization enhances the intrasource separation of lipid classes and extends identification of individual molecular species through multi‐dimensional mass spectrometry: development of an automated high‐throughput platform for shotgun lipidomics. Rapid Commun Mass Spectrom. 2008;22:2115‐2124. [DOI] [PMC free article] [PubMed] [Google Scholar]
24. Wang M, Wang C, Han X, Selection of internal standards for accurate quantification of complex lipid species in biological extracts by electrospray ionization mass spectrometry‐What, how and why? Mass Spectrom Rev. 2017;36:693‐714. [DOI] [PMC free article] [PubMed] [Google Scholar]
25. Iman RL, Conover WJ, Use of the Rank Transform in Regression. Technometrics. 1979;21:499‐509. [Google Scholar]
26. Moons KG, Altman DG, Reitsma JB, et al. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): explanation and elaboration. Ann Intern Med. 2015;162:W1‐73. [DOI] [PubMed] [Google Scholar]
27. Kang L, Chen W, Petrick NA, Gallas BD, Comparing two correlated C indices with right‐censored survival outcome: a one‐shot nonparametric approach. Stat Med. 2015;34:685‐703. [DOI] [PMC free article] [PubMed] [Google Scholar]
28. Langfelder P, Horvath S, WGCNA: an R package for weighted correlation network analysis. BMC Bioinformatics. 2008;9:559. [DOI] [PMC free article] [PubMed] [Google Scholar]
29. Matarin M, Salih DA, Yasvoina M, et al. A genome‐wide gene‐expression analysis and database in transgenic mice during development of amyloid or tau pathology. Cell Rep. 2015;10:633‐644. [DOI] [PubMed] [Google Scholar]
30. Hu Z, Mellor J, Wu J, DeLisi C, VisANT: an online visualization and analysis tool for biological interaction data. BMC Bioinformatics. 2004;5:17. [DOI] [PMC free article] [PubMed] [Google Scholar]
31. Seyer A, Boudah S, Broudin S, Junot C, Colsch B, Annotation of the human cerebrospinal fluid lipidome using high resolution mass spectrometry and a dedicated data processing workflow. Metabolomics. 2016;12:91. [DOI] [PMC free article] [PubMed] [Google Scholar]
32. Sol J, Jove M, Povedano M, et al. Lipidomic traits of plasma and cerebrospinal fluid in amyotrophic lateral sclerosis correlate with disease progression. Brain Commun. 2021;3:fcab143. [DOI] [PMC free article] [PubMed] [Google Scholar]
33. Pieragostino D, Cicalini I, Lanuti P, et al. Enhanced release of acid sphingomyelinase‐enriched exosomes generates a lipidomics signature in CSF of Multiple Sclerosis patients. Sci Rep. 2018;8:3071. [DOI] [PMC free article] [PubMed] [Google Scholar]
34. Fernandez‐Irigoyen J, Cartas‐Cejudo P, Iruarrizaga‐Lejarreta M, Santamaria E, Alteration in the cerebrospinal fluid lipidome in Parkinson's disease: a post‐mortem pilot study. Biomedicines. 2021;9(5):491. [DOI] [PMC free article] [PubMed] [Google Scholar]
35. Byeon SK, Madugundu AK, Jain AP, et al. Cerebrospinal fluid lipidomics for biomarkers of Alzheimer's disease. Mol Omics. 2021;17:454‐463. [DOI] [PMC free article] [PubMed] [Google Scholar]
36. Dakterzada F, Jove M, Huerto R, et al. Cerebrospinal fluid neutral lipids predict progression from mild cognitive impairment to Alzheimer's disease. Geroscience. 2024;46:683‐696. [DOI] [PMC free article] [PubMed] [Google Scholar]
37. Roe CM, Ances BM, Head D, et al. Incident cognitive impairment: longitudinal changes in molecular, structural and cognitive biomarkers. Brain. 2018;141:3233‐3248. [DOI] [PMC free article] [PubMed] [Google Scholar]
38. Zhou Z, Ryan J, Tonkin AM, et al. Association between triglycerides and risk of dementia in community‐dwelling older adults: a prospective cohort study. Neurology. 2023;101:e2288‐e2299. [DOI] [PMC free article] [PubMed] [Google Scholar]
39. Dimache AM, Salaru DL, Sascau R, Statescu C, The role of high triglycerides level in predicting cognitive impairment: a review of current evidence. Nutrients. 2021;13(8):2118. [DOI] [PMC free article] [PubMed] [Google Scholar]
40. He Q, Han X, Cardiolipin remodeling in diabetic heart. Chem Phys Lipids. 2014;179:75‐81. [DOI] [PubMed] [Google Scholar]
41. Falabella M, Vernon HJ, Hanna MG, Claypool SM, Pitceathly RDS. Cardiolipin, mitochondria, and neurological disease. Trends Endocrinol Metab. 2021;32:224‐237. [DOI] [PMC free article] [PubMed] [Google Scholar]
42. Farmer BC, Walsh AE, Kluemper JC, Johnson LA, Lipid droplets in neurodegenerative disorders. Front Neurosci. 2020;14:742. [DOI] [PMC free article] [PubMed] [Google Scholar]
43. Mapstone M, Cheema AK, Fiandaca MS, et al. Plasma phospholipids identify antecedent memory impairment in older adults. Nat Med. 2014;20:415‐418. [DOI] [PMC free article] [PubMed] [Google Scholar]
44. Liu Y, Thalamuthu A, Mather KA, et al. Plasma lipidome is dysregulated in Alzheimer's disease and is associated with disease risk genes. Transl Psychiatry. 2021;11:344. [DOI] [PMC free article] [PubMed] [Google Scholar]
45. Mielke MM, Haughey NJ, Bandaru VV, et al. Plasma sphingomyelins are associated with cognitive progression in Alzheimer's disease. J Alzheimers Dis. 2011;27:259‐269. [DOI] [PMC free article] [PubMed] [Google Scholar]
46. Mielke MM, Bandaru VV, Haughey NJ, Rabins PV, Lyketsos CG, Carlson MC, Serum sphingomyelins and ceramides are early predictors of memory impairment. Neurobiol Aging. 2010;31:17‐24. [DOI] [PMC free article] [PubMed] [Google Scholar]
47. Moseholm KF, Cronje HT, Koch M, et al. Circulating sphingolipids in relation to cognitive decline and incident dementia: the cardiovascular health study. Alzheimers Dement (Amst). 2024;16:e12623. [DOI] [PMC free article] [PubMed] [Google Scholar]
48. Chua XY, Torta F, Chong JR, et al. Lipidomics profiling reveals distinct patterns of plasma sphingolipid alterations in Alzheimer's disease and vascular dementia. Alzheimers Res Ther. 2023;15:214. [DOI] [PMC free article] [PubMed] [Google Scholar]
49. Kuo A, Hla T, Regulation of cellular and systemic sphingolipid homeostasis. Nat Rev Mol Cell Biol. 2024;25:802‐821. [DOI] [PMC free article] [PubMed] [Google Scholar]
50. Turpin‐Nolan SM, Bruning JC, The role of ceramides in metabolic disorders: when size and localization matters. Nat Rev Endocrinol. 2020;16:224‐233. [DOI] [PubMed] [Google Scholar]
51. Mielke MM, Haughey NJ, Bandaru VV, et al. Plasma ceramides are altered in mild cognitive impairment and predict cognitive decline and hippocampal volume loss. Alzheimers Dement. 2010;6:378‐385. [DOI] [PMC free article] [PubMed] [Google Scholar]
52. Alzheimer'sAssociation . 2023 Alzheimer's disease facts and figures. Alzheimers Dement. 2023;19:1598‐1695. [DOI] [PubMed] [Google Scholar]
53. Jack CR Jr, Bennett DA, Blennow K, et al. NIA‐AA research framework: toward a biological definition of Alzheimer's disease. Alzheimers Dement. 2018;14:535‐562. [DOI] [PMC free article] [PubMed] [Google Scholar]
54. Huynh K, Lim WLF, Giles C, et al. Concordant peripheral lipidome signatures in two large clinical studies of Alzheimer's disease. Nat Commun. 2020;11:5698. [DOI] [PMC free article] [PubMed] [Google Scholar]
55. Vance JE, Phospholipid synthesis and transport in mammalian cells. Traffic. 2015;16:1‐18. [DOI] [PubMed] [Google Scholar]
56. St Germain M, Iraji R, Bakovic M, Phosphatidylethanolamine homeostasis under conditions of impaired CDP‐ethanolamine pathway or phosphatidylserine decarboxylation. Front Nutr. 2022;9:1094273. [DOI] [PMC free article] [PubMed] [Google Scholar]
57. Osetrova M, Tkachev A, Mair W, et al. Lipidome atlas of the adult human brain. Nat Commun. 2024;15:4455. [DOI] [PMC free article] [PubMed] [Google Scholar]
58. Binotti B, Jahn R, Perez‐Lara A, An overview of the synaptic vesicle lipid composition. Arch Biochem Biophys. 2021;709:108966. [DOI] [PubMed] [Google Scholar]
59. Bleijerveld OB, Brouwers J, Vaandrager AB, Helms JB, Houweling M, The CDP‐ethanolamine pathway and phosphatidylserine decarboxylation generate different phosphatidylethanolamine molecular species. J Biol Chem. 2007;282:28362‐28372. [DOI] [PubMed] [Google Scholar]
60. Ding J, Ji J, Rabow Z, et al. A metabolome atlas of the aging mouse brain. Nat Commun. 2021;12:6021. [DOI] [PMC free article] [PubMed] [Google Scholar]
61. Martens YA, Zhao N, Liu CC, et al. ApoE Cascade Hypothesis in the pathogenesis of Alzheimer's disease and related dementias. Neuron. 2022;110:1304‐1317. [DOI] [PMC free article] [PubMed] [Google Scholar]
62. 2024 Alzheimer's disease facts and figures. Alzheimers Dement. 2024;20:3708‐3821. [DOI] [PMC free article] [PubMed] [Google Scholar]
63. Sienski G, Narayan P, Bonner JM, et al. APOE4 disrupts intracellular lipid homeostasis in human iPSC‐derived glia. Sci Transl Med. 2021;13(583):eaaz4564. [DOI] [PMC free article] [PubMed] [Google Scholar]
64. Tcw J, Qian L, Pipalia NH, et al. Cholesterol and matrisome pathways dysregulated in astrocytes and microglia. Cell. 2022;185:2213‐2233.e25. [DOI] [PMC free article] [PubMed] [Google Scholar]
65. Blanchard JW, Akay LA, Davila‐Velderrain J, et al. APOE4 impairs myelination via cholesterol dysregulation in oligodendrocytes. Nature. 2022;611:769‐779. [DOI] [PMC free article] [PubMed] [Google Scholar]
66. Litvinchuk A, Suh JH, Guo JL, et al. Amelioration of Tau and ApoE4‐linked glial lipid accumulation and neurodegeneration with an LXR agonist. Neuron. 2024;112:384‐403 e8. [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: alz71924‐sup‐0001‐SuppMat.pdf

ALZ-22-e71924-s003.pdf^{(383.6KB, pdf)}

Supporting Information: alz71924‐sup‐0002‐IcmjeDisclosureForm.pdf

ALZ-22-e71924-s002.pdf^{(775.3KB, pdf)}

Supporting Information

ALZ-22-e71924-s001.zip^{(1.4MB, zip)}

Data Availability Statement

[alz71924-bib-0001] 1. Sastry PS, Lipids of nervous tissue: composition and metabolism. Prog Lipid Res. 1985;24:69‐176. [DOI] [PubMed] [Google Scholar]

[alz71924-bib-0002] 2. Tracey TJ, Steyn FJ, Wolvetang EJ, Ngo ST, Neuronal lipid metabolism: multiple pathways driving functional outcomes in health and disease. Front Mol Neurosci. 2018;11:10. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0003] 3. Chew H, Solomon VA, Fonteh AN, Involvement of lipids in Alzheimer's disease pathology and potential therapies. Front Physiol. 2020;11:598. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0004] 4. Yoon JH, Seo Y, Jo YS, et al. Brain lipidomics: from functional landscape to clinical significance. Sci Adv. 2022;8:eadc9317. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0005] 5. Kunkle BW, Grenier‐Boley B, Sims R, et al. Genetic meta‐analysis of diagnosed Alzheimer's disease identifies new risk loci and implicates Abeta, tau, immunity and lipid processing. Nat Genet. 2019;51:414‐430. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0006] 6. Livingston G, Huntley J, Sommerlad A, et al. Dementia prevention, intervention, and care: 2020 report of the Lancet commission. Lancet. 2020;396:413‐446. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0007] 7. Roberts RO, Geda YE, Knopman DS, et al. The Mayo clinic study of aging: design and sampling, participation, baseline measures and sample characteristics. Neuroepidemiology. 2008;30:58‐69. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0008] 8. Olsson B, Lautner R, Andreasson U, et al. CSF and blood biomarkers for the diagnosis of Alzheimer's disease: a systematic review and meta‐analysis. Lancet Neurol. 2016;15:673‐684. [DOI] [PubMed] [Google Scholar]

[alz71924-bib-0009] 9. Zetterberg H, Bendlin BB, Biomarkers for Alzheimer's disease‐preparing for a new era of disease‐modifying therapies. Mol Psychiatry. 2021;26:296‐308. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0010] 10. Park SA, Han SM, Kim CE, New fluid biomarkers tracking non‐amyloid‐beta and non‐tau pathology in Alzheimer's disease. Exp Mol Med. 2020;52:556‐568. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0011] 11. Sutphen CL, Jasielec MS, Shah AR, et al. Longitudinal cerebrospinal fluid biomarker changes in preclinical Alzheimer disease during middle age. JAMA Neurol. 2015;72:1029‐1042. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0012] 12. Schindler SE, Li Y, Todd KW, et al. Emerging cerebrospinal fluid biomarkers in autosomal dominant Alzheimer's disease. Alzheimers Dement. 2019;15:655‐665. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0013] 13. Sutphen CL, McCue L, Herries EM, et al. Longitudinal decreases in multiple cerebrospinal fluid biomarkers of neuronal injury in symptomatic late onset Alzheimer's disease. Alzheimers Dement. 2018;14:869‐879. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0014] 14. Hellwig K, Kvartsberg H, Portelius E, et al. Neurogranin and YKL‐40: independent markers of synaptic degeneration and neuroinflammation in Alzheimer's disease. Alzheimers Res Ther. 2015;7:74. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0015] 15. Janelidze S, Mattsson N, Stomrud E, et al. CSF biomarkers of neuroinflammation and cerebrovascular dysfunction in early Alzheimer disease. Neurology. 2018;91:e867‐e877. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0016] 16. Teunissen CE, Verberk IMW, Thijssen EH, et al. Blood‐based biomarkers for Alzheimer's disease: towards clinical implementation. Lancet Neurol. 2022;21:66‐77. [DOI] [PubMed] [Google Scholar]

[alz71924-bib-0017] 17. Morris JC, The clinical dementia rating (CDR): current version and scoring rules. Neurology. 1993;43:2412‐2414. [DOI] [PubMed] [Google Scholar]

[alz71924-bib-0018] 18. Kern S, Syrjanen JA, Blennow K, et al. Association of cerebrospinal fluid neurofilament light protein with risk of mild cognitive impairment among individuals without cognitive impairment. JAMA Neurol. 2019;76:187‐193. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0019] 19. Shue F, White LJ, Hendrix R, et al. CSF biomarkers of immune activation and Alzheimer's disease for predicting cognitive impairment risk in the elderly. Sci Adv. 2024;10:eadk3674. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0020] 20. Yang K, Cheng H, Gross RW, Han X, Automated lipid identification and quantification by multidimensional mass spectrometry‐based shotgun lipidomics. Anal Chem. 2009;81:4356‐4368. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0021] 21. Kawatani K, Holm ML, Starling SC, et al. ABCA7 deficiency causes neuronal dysregulation by altering mitochondrial lipid metabolism. Mol Psychiatry. 2023;3(29):809‐819. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0022] 22. Wang M, Han X, Multidimensional mass spectrometry‐based shotgun lipidomics. Methods Mol Biol. 2014;1198:203‐220. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0023] 23. Han X, Yang K, Gross RW, Microfluidics‐based electrospray ionization enhances the intrasource separation of lipid classes and extends identification of individual molecular species through multi‐dimensional mass spectrometry: development of an automated high‐throughput platform for shotgun lipidomics. Rapid Commun Mass Spectrom. 2008;22:2115‐2124. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0024] 24. Wang M, Wang C, Han X, Selection of internal standards for accurate quantification of complex lipid species in biological extracts by electrospray ionization mass spectrometry‐What, how and why? Mass Spectrom Rev. 2017;36:693‐714. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0025] 25. Iman RL, Conover WJ, Use of the Rank Transform in Regression. Technometrics. 1979;21:499‐509. [Google Scholar]

[alz71924-bib-0026] 26. Moons KG, Altman DG, Reitsma JB, et al. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): explanation and elaboration. Ann Intern Med. 2015;162:W1‐73. [DOI] [PubMed] [Google Scholar]

[alz71924-bib-0027] 27. Kang L, Chen W, Petrick NA, Gallas BD, Comparing two correlated C indices with right‐censored survival outcome: a one‐shot nonparametric approach. Stat Med. 2015;34:685‐703. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0028] 28. Langfelder P, Horvath S, WGCNA: an R package for weighted correlation network analysis. BMC Bioinformatics. 2008;9:559. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0029] 29. Matarin M, Salih DA, Yasvoina M, et al. A genome‐wide gene‐expression analysis and database in transgenic mice during development of amyloid or tau pathology. Cell Rep. 2015;10:633‐644. [DOI] [PubMed] [Google Scholar]

[alz71924-bib-0030] 30. Hu Z, Mellor J, Wu J, DeLisi C, VisANT: an online visualization and analysis tool for biological interaction data. BMC Bioinformatics. 2004;5:17. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0031] 31. Seyer A, Boudah S, Broudin S, Junot C, Colsch B, Annotation of the human cerebrospinal fluid lipidome using high resolution mass spectrometry and a dedicated data processing workflow. Metabolomics. 2016;12:91. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0032] 32. Sol J, Jove M, Povedano M, et al. Lipidomic traits of plasma and cerebrospinal fluid in amyotrophic lateral sclerosis correlate with disease progression. Brain Commun. 2021;3:fcab143. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0033] 33. Pieragostino D, Cicalini I, Lanuti P, et al. Enhanced release of acid sphingomyelinase‐enriched exosomes generates a lipidomics signature in CSF of Multiple Sclerosis patients. Sci Rep. 2018;8:3071. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0034] 34. Fernandez‐Irigoyen J, Cartas‐Cejudo P, Iruarrizaga‐Lejarreta M, Santamaria E, Alteration in the cerebrospinal fluid lipidome in Parkinson's disease: a post‐mortem pilot study. Biomedicines. 2021;9(5):491. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0035] 35. Byeon SK, Madugundu AK, Jain AP, et al. Cerebrospinal fluid lipidomics for biomarkers of Alzheimer's disease. Mol Omics. 2021;17:454‐463. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0036] 36. Dakterzada F, Jove M, Huerto R, et al. Cerebrospinal fluid neutral lipids predict progression from mild cognitive impairment to Alzheimer's disease. Geroscience. 2024;46:683‐696. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0037] 37. Roe CM, Ances BM, Head D, et al. Incident cognitive impairment: longitudinal changes in molecular, structural and cognitive biomarkers. Brain. 2018;141:3233‐3248. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0038] 38. Zhou Z, Ryan J, Tonkin AM, et al. Association between triglycerides and risk of dementia in community‐dwelling older adults: a prospective cohort study. Neurology. 2023;101:e2288‐e2299. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0039] 39. Dimache AM, Salaru DL, Sascau R, Statescu C, The role of high triglycerides level in predicting cognitive impairment: a review of current evidence. Nutrients. 2021;13(8):2118. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0040] 40. He Q, Han X, Cardiolipin remodeling in diabetic heart. Chem Phys Lipids. 2014;179:75‐81. [DOI] [PubMed] [Google Scholar]

[alz71924-bib-0041] 41. Falabella M, Vernon HJ, Hanna MG, Claypool SM, Pitceathly RDS. Cardiolipin, mitochondria, and neurological disease. Trends Endocrinol Metab. 2021;32:224‐237. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0042] 42. Farmer BC, Walsh AE, Kluemper JC, Johnson LA, Lipid droplets in neurodegenerative disorders. Front Neurosci. 2020;14:742. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0043] 43. Mapstone M, Cheema AK, Fiandaca MS, et al. Plasma phospholipids identify antecedent memory impairment in older adults. Nat Med. 2014;20:415‐418. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0044] 44. Liu Y, Thalamuthu A, Mather KA, et al. Plasma lipidome is dysregulated in Alzheimer's disease and is associated with disease risk genes. Transl Psychiatry. 2021;11:344. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0045] 45. Mielke MM, Haughey NJ, Bandaru VV, et al. Plasma sphingomyelins are associated with cognitive progression in Alzheimer's disease. J Alzheimers Dis. 2011;27:259‐269. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0046] 46. Mielke MM, Bandaru VV, Haughey NJ, Rabins PV, Lyketsos CG, Carlson MC, Serum sphingomyelins and ceramides are early predictors of memory impairment. Neurobiol Aging. 2010;31:17‐24. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0047] 47. Moseholm KF, Cronje HT, Koch M, et al. Circulating sphingolipids in relation to cognitive decline and incident dementia: the cardiovascular health study. Alzheimers Dement (Amst). 2024;16:e12623. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0048] 48. Chua XY, Torta F, Chong JR, et al. Lipidomics profiling reveals distinct patterns of plasma sphingolipid alterations in Alzheimer's disease and vascular dementia. Alzheimers Res Ther. 2023;15:214. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0049] 49. Kuo A, Hla T, Regulation of cellular and systemic sphingolipid homeostasis. Nat Rev Mol Cell Biol. 2024;25:802‐821. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0050] 50. Turpin‐Nolan SM, Bruning JC, The role of ceramides in metabolic disorders: when size and localization matters. Nat Rev Endocrinol. 2020;16:224‐233. [DOI] [PubMed] [Google Scholar]

[alz71924-bib-0051] 51. Mielke MM, Haughey NJ, Bandaru VV, et al. Plasma ceramides are altered in mild cognitive impairment and predict cognitive decline and hippocampal volume loss. Alzheimers Dement. 2010;6:378‐385. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0052] 52. Alzheimer'sAssociation . 2023 Alzheimer's disease facts and figures. Alzheimers Dement. 2023;19:1598‐1695. [DOI] [PubMed] [Google Scholar]

[alz71924-bib-0053] 53. Jack CR Jr, Bennett DA, Blennow K, et al. NIA‐AA research framework: toward a biological definition of Alzheimer's disease. Alzheimers Dement. 2018;14:535‐562. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0054] 54. Huynh K, Lim WLF, Giles C, et al. Concordant peripheral lipidome signatures in two large clinical studies of Alzheimer's disease. Nat Commun. 2020;11:5698. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0055] 55. Vance JE, Phospholipid synthesis and transport in mammalian cells. Traffic. 2015;16:1‐18. [DOI] [PubMed] [Google Scholar]

[alz71924-bib-0056] 56. St Germain M, Iraji R, Bakovic M, Phosphatidylethanolamine homeostasis under conditions of impaired CDP‐ethanolamine pathway or phosphatidylserine decarboxylation. Front Nutr. 2022;9:1094273. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0057] 57. Osetrova M, Tkachev A, Mair W, et al. Lipidome atlas of the adult human brain. Nat Commun. 2024;15:4455. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0058] 58. Binotti B, Jahn R, Perez‐Lara A, An overview of the synaptic vesicle lipid composition. Arch Biochem Biophys. 2021;709:108966. [DOI] [PubMed] [Google Scholar]

[alz71924-bib-0059] 59. Bleijerveld OB, Brouwers J, Vaandrager AB, Helms JB, Houweling M, The CDP‐ethanolamine pathway and phosphatidylserine decarboxylation generate different phosphatidylethanolamine molecular species. J Biol Chem. 2007;282:28362‐28372. [DOI] [PubMed] [Google Scholar]

[alz71924-bib-0060] 60. Ding J, Ji J, Rabow Z, et al. A metabolome atlas of the aging mouse brain. Nat Commun. 2021;12:6021. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0061] 61. Martens YA, Zhao N, Liu CC, et al. ApoE Cascade Hypothesis in the pathogenesis of Alzheimer's disease and related dementias. Neuron. 2022;110:1304‐1317. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0062] 62. 2024 Alzheimer's disease facts and figures. Alzheimers Dement. 2024;20:3708‐3821. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0063] 63. Sienski G, Narayan P, Bonner JM, et al. APOE4 disrupts intracellular lipid homeostasis in human iPSC‐derived glia. Sci Transl Med. 2021;13(583):eaaz4564. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0064] 64. Tcw J, Qian L, Pipalia NH, et al. Cholesterol and matrisome pathways dysregulated in astrocytes and microglia. Cell. 2022;185:2213‐2233.e25. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0065] 65. Blanchard JW, Akay LA, Davila‐Velderrain J, et al. APOE4 impairs myelination via cholesterol dysregulation in oligodendrocytes. Nature. 2022;611:769‐779. [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71924-bib-0066] 66. Litvinchuk A, Suh JH, Guo JL, et al. Amelioration of Tau and ApoE4‐linked glial lipid accumulation and neurodegeneration with an LXR agonist. Neuron. 2024;112:384‐403 e8. [DOI] [PMC free article] [PubMed] [Google Scholar]

PERMALINK

CSF and plasma lipid profiling predicts risk of cognitive impairment in cognitively unimpaired older adults

Takahisa Kanekiyo

Launia J White

Hongping Ye

Yingxue Ren

Jason D Ulrich

Ana‐Caroline Raulin

Alla Alnobani

Chengjie Xiong

Yan W Asmann

Maria Vassilaki

Alicia Algeciras‐Schimnich

Junmin Peng

Fatima S Amtashar

Suzanne E Schindler

David M Holtzman

Guojun Bu

Ronald C Petersen

Michael G Heckman

Xianlin Han

Abstract

INTRODUCTION

METHODS

RESULTS

DISCUSSION

Highlights

1. BACKGROUND

2. METHODS

2.1. Study design

RESEARCH IN CONTEXT

2.2. AD biomarker measurements and lipidomics in CSF and plasma samples

2.3. Statistical analysis

2.4. Weighted correlation network analysis

3. RESULTS

3.1. Associations between CSF and plasma lipids in cognitively unimpaired participants

FIGURE 1.

3.2. Impact of APOE on CSF and plasma lipids in cognitively unimpaired participants

TABLE 1.

3.3. Impact of age, sex, and modifiable dementia risk factors on CSF and plasma lipids in cognitively unimpaired participants

TABLE 2.

3.4. Associations of CSF and plasma lipids with CSF AD biomarkers in cognitively unimpaired participants

TABLE 3.

FIGURE 2.

3.5. Prediction for CDR conversion based on CSF lipids

FIGURE 3.

3.6. Prediction for CDR conversion based on plasma lipids

4. DISCUSSION

AUTHOR CONTRIBUTIONS

CONSENT STATEMENT

CONFLICT OF INTEREST STATEMENT

Supporting information

ACKNOWLEDGMENTS

Contributor Information

DATA AVAILABILITY STATEMENT

REFERENCES

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases