Cerebrospinal fluid lipidomics for biomarkers of Alzheimer’s disease

Abstract

Alzheimer’s disease (AD) is the most common cause of dementia and is associated with serious neurologic sequelae resulting from neurodegenerative changes. Identification of markers of early-stage AD could be important for designing strategies to arrest the progression of disease. The brain is rich in lipids because they are crucial for signal transduction and anchoring of membrane proteins. Cerebrospinal fluid (CSF) is an excellent specimen for studying the metabolism of lipids in AD because it can reflect changes occurring in the brain. We aimed to identify CSF lipidomic alterations associated with AD, using untargeted lipidomics, carried out in positive and negative ion modes. We found CSF lipids that were significantly altered in AD cases. In addition, comparison of CSF lipid profiles between persons with mild cognitive impairment (MCI) and AD showed a strong positive correlation between the lipidomes of the MCI and AD groups. The novel lipid biomarkers identified in this study are excellent candidates for validation in a larger set of patient samples and as predictive biomarkers of AD through future longitudinal studies. Once validated, the lipid biomarkers could lead to early detection, disease monitoring and the ability to measure the efficacy of potential therapeutic interventions in AD.

Cerebrospinal fluid lipid biomarkers of Alzheimer’s disease were discovered using tandem mass spectrometry.

Graphical Abstract

Introduction

Lipids are water-insoluble molecules that are responsible for multiple cellular functions including proliferation, apoptosis, structural support, and cell signaling ^1–4. Various classes of lipids, including phospholipids, sphingolipids, sterol lipids and acylglycerols, have crucial roles in neuronal functions ⁵. These diverse and essential roles of lipids in cellular metabolism have been highlighted by the finding of abnormal lipid metabolism in various diseases including neurodegenerative disorders and cancers ^6–9. Importantly, alterations in lipids have been reported in the brain tissue of patients with Alzheimer disease (AD), a common neurodegenerative dementia associated with loss of synapses and neurons. Memory loss and progressive decline in cognitive function are commonly associated with AD, and the mortality rate related to AD has been increasing owing to the absence of disease-modifying therapy.

Pathologic hallmarks of AD include extracellular aggregation of the 42-residue form of amyloid-β peptide (Aβ) and intraneuronal fibrillary tau tangles ¹⁰. Aβ is produced by sequential cleavage of amyloid precursor protein (APP) by β- and γ-secretase, whereas intracellular neurofibrillary tangles comprise hyperphosphorylated tau protein ¹⁰. APP and secretases are integral membrane proteins and, considering their direct involvement in the development and progression of AD, membrane lipids are likely to be altered. Membrane lipids consist of phospholipids as lipid bilayers around the cells, whereas sphingolipids and cholesterols are associated with membranes of neurons and myelin ¹¹. Lipid rafts, membrane domains enriched in cholesterol and sphingolipids, promote the interaction of APP with the β-secretase responsible for generation of Aβ. Of the different classes of lipids, ceramides are related to aging and are also known to affect generation and aggregation of Aβ. Modulation of APP cleavage by lipid rafts and their components is thought to have a crucial role in amyloidogenesis ¹².

Altered metabolism of lipids in AD has been studied using different biological samples including postmortem brain tissues. Several studies have highlighted elevation of ceramide and a decrease in phospholipids in brain tissue ^13–17. In one study, alteration of several lipids including phophstidylethanolamine and diacylglycerol was observed in brain tissue with no lipid alterations in post-mortem cerebrospinal fluid (CSF) except docosahexaenoic acid ¹⁸. Another study that was restricted to studying ceramides found increased ceramide levels in CSF from living patients in agreement with observations similar published findings from brain tissue ¹⁹. This alteration in ceramide levels is also reflected in the serum of AD patients ²⁰. Overall, these findings support the idea that lipid biosynthesis is indeed altered in the context of AD.

Advanced mass spectrometric technologies can enable detection and identification of complex structures of lipids with high sensitivity and selectivity. Owing to different types of head groups and combinations of fatty acyl chains that vary in the length of hydrocarbons and degree of unsaturation, lipids are diverse in their degree of hydrophobicity. The use of chromatography can separate lipids on the basis of polarity and can distinguish isomers. Liquid chromatography coupled to tandem mass spectrometry (LC-MS/MS) is an effective analytic platform for lipidomics with high accuracy and resolution. The goal of this study was to use untargeted lipidomics to investigate alterations in the CSF lipidome in cases with AD dementia and MCI, compared to controls.

Methods

We performed global profiling of CSF to discover lipids associated with AD using the methods outlined in Fig. 1 and described below.

Fig. 1. Workflow for cerebrospinal fluid (CSF) lipidomics.

CSF samples were collected from 18 controls, 15 patients with mild cognitive impairment (MCI), and 17 patients with Alzheimer disease (AD). Lipids were extracted and analyzed by liquid chromatography coupled to tandem mass spectrometry (LC-MS/MS) in positive and negative ion modes. Identification of CSF lipids was based on precursor and product ions, and they were quantified by calculating peak areas. Statistical analysis was performed to find CSF lipids associated with AD.

CSF specimen collection/characteristics

This study was approved by the Institutional Review Board of Johns Hopkins University School of Medicine. CSF specimens were collected from cognitively normal persons and individuals with MCI enrolled in a study supported by the Johns Hopkins Alzheimer’s Disease Research Center. Cases with AD dementia were recruited from the Johns Hopkins Cerebrospinal Fluid Disorders Center. A total of 50 subjects participated in the study (AD dementia = 17, MCI = 15, controls = 18). Diagnostic criteria for the subjects with AD dementia and MCI followed the National Institute on Aging–Alzheimer’s Association criteria. For the AD group an additional criterion for inclusion was a low CSF Aβ/Aβ(1–40) ratio (<0.1) and increased phospho-tau level to ensure that patients had underlying AD pathology. The characteristics of the study groups are shown in Table 1. All AD subjects used in the study had a Global Clinical Dementia Rating Scale of 1, which is considered early or mild dementia.

Table 1.

Characteristics of Study Cohort^a

Characteristic	Controls (n=18)	MCI (n=15)	AD (n=17)
Age at CSF collection, y	70.0 ± 6.0	74.8 ± 9.0	77.1 ± 9.4
Women	11 (65%)	7 (46%)	7 (41%)
Aβ42, pg/mL	1,832 ± 519	948 ± 226	607 ± 178
Tau, pg/mL	291 ± 105	474 ± 367	667 ± 531
Phospho-tau (at Thr 181), pg/mL	40 ± 13	71 ± 64	97 ± 76

Abbreviations: Aβ, 42-residue form of amyloid-β peptide; AD, Alzheimer disease; CSF, cerebrospinal fluid; MCI, mild cognitive impairment.

^aVariables are expressed in mean with standard deviation or absolute frequency count with percentage.

All participants underwent lumbar puncture using an atraumatic needle after an overnight fast, and CSF was collected in low-bind polypropylene vials and stored at –80°C. All persons selected for inclusion in the study provided informed consent before sample collection. CSF levels of Aβ(1–42), Aβ(1–40), tau, and phospho-tau (at Thr 181) were determined using Lumipulse electrochemiluminescence assays (Fujirebio Diagnostics Ltd).

Chemicals and reagents

Chloroform used in the extraction and ammonium formate and formic acid used as modifiers in LC-MS/MS were purchased from Sigma-Aldrich. Hydrochloric acid (HCl) used in the extraction was purchased from Merck Millipore. High-performance liquid chromatography–grade methanol, acetonitrile, isopropanol, and water were from ThermoFisher Scientific and used in lipid extraction and LC-MS/MS analysis.

Lipid extraction

Lipids were extracted from CSF samples using the Bligh and Dyer method with modifications ²¹. Internal standard mixture was added to 400 μL of CSF samples before the extraction and vortexed briefly. CSF samples were concentrated using a freeze dryer; for neutral extraction, 1 mL of chloroform:methanol (1:2, v/v) was added and vortexed for 1 min. After incubation in an ice bath for 10 min, samples were centrifuged at 13,800×g for 2 min at 4°C. The lower organic layer was obtained and transferred to a new vial. To the upper aqueous layer, acidic extraction was performed by adding 750 μL of chloroform:methanol:37% 1M HCl (40:80:1, v/v/v). The sample was vortexed for 1 min and left at room temperature for 15 min. Next, 250 μL of ice-cold chloroform and 450 μL of 0.1M HCl were added, followed by vortexing for 1 min. The sample was centrifuged at 6,500×g for 2 min at 4°C. The lower organic layer was pipetted and combined with the previously collected organic layer. The combined organic layer was mixed with 300 μL of H₂O and centrifuged at 6,500×g for 2 min at 4°C for additional wash. The lower organic layer was obtained and dried under speed vacuum and stored at –80°C till further analysis.

Liquid chromatography-mass spectrometry analysis

Untargeted CSF lipidomics analysis was performed by coupling an UltiMate 3000 ultrahigh-performance liquid chromatography system to a Q Exactive Plus Hybrid Quadrupole-Orbitrap Mass Spectrometer, both from ThermoFisher Scientific. An Accucore™ C18 column (2.6 μm×2.1 mm×150 mm) from ThermoFisher Scientific was used as an analytic column under the binary gradient elution of H₂O:acetonitrile (9:1, v/v) containing 1mM of ammonium formate and 0.1% formic acid as mobile phase A and isopropanol:methanol:acetonitrile (7:2:1, v/v/v) containing 10mM of ammonium formate and 0.1% formic acid as mobile phase B. All samples were analyzed in triplicates. Each sample was loaded onto the analytical column with 100% mobile phase A at a flow rate of 250 μL/min, and mobile phase B was ramped from 0% to 65% over 1 min. Mobile phase B was linearly increased to 80% in 12 min, 85% in 12 min, 100% in 5 min, and maintained at 100% for 10 min. The column was re-equilibrated with 100% mobile phase A for 5 min. Full scan and MS/MS data acquisition were obtained in data-dependent analysis mode by alternating between positive and negative ion modes throughout the run. MS resolution of 70,000 and MS/MS resolution of 15,000 were applied. The electrospray ionization voltage was set at 3 kV. The column oven was set at 60°C and the transfer capillary at 300°C. All samples were blinded and analyzed in a randomized order.

Data analysis

Identification of CSF lipids was achieved using LipidSearch 4.1 (ThermoFisher Scientific) based on precursor and product ions. With high-resolution MS/MS, highly accurate annotation of lipids was carried out with a mass tolerance of 5 ppm for precursor ions and 8 ppm for product ions. Identified lipids were quantified using Xcalibur software (ThermoFisher Scientific), in conjunction with normalization by internal standards with identical head groups. For some classes of lipids for which such internal standards were not available, normalization was based on internal standards with the closest retention time during gradient elution. P value, principal component analysis, and volcano plot were constructed using MetaboAnalyst (http://www.metaboanalyst.ca/).

Results and Discussion

Untargeted CSF lipidomics using high-resolution LC-MS/MS

We identified and quantified a total of 245 lipids: 11 lysophosphatidylcholines (LPC), 50 phosphatidylcholines (PC), 11 lysophosphatatidylethanolamines (LPE), 44 phosphatidylethanolamines (PE), 2 cyclic phosphatidic acids (cPA), 2 phosphatidylglycerols (PG), 15 phosphatidylinositols (PI), 8 ceramides (Cer), 13 monohexosylceramide (MHC), 16 sphingomyelins (SM), 11 cholesteryl ester (ChE), 10 diacylglycerols (DG), and 52 triacylglycerols (TG) (Fig. 2A). Fig. S1 shows the base peak chromatograms of CSF lipids in positive ion mode, in which LPA, PC, Cer, MHC, and SM were analyzed in positive ion mode in protonated forms ([M+H]⁺). Neutral lipids, including ChE, DG, and TG were detected in ammonium adduct forms ([M+NH₄]⁺). One precursor ion, m/z 876.80, was detected at 38.2 min in the positive ion mode, and its MS/MS spectrum showed the typical MS/MS fragmentation pattern of TG lipids. Fragments produced by cleavage of a fatty acyl chain in the carboxylic acid form were detected (14:0, 16:0, 18:0, 18:1, and 20:0) with relatively large intensities, whereas those produced by dissociation of 2 fatty acyl chains in the forms of carboxylic acid and ketene ([M+NH₄-NH₃-RCOOH-R′CH=C=O]⁺) were seen with lower intensities (Fig. S1A). This led to a feature annotation of 52:2-TG species. The total number of carbons and double bonds in fatty acyl chains are annotated for TG lipids. In Fig. S1B, separation of CSF lipids detected in negative ion mode is shown. LPE, PE, cPA, PI, and PG were readily detected in negative ion mode by deprotonated forms ([M-H]^–). Fragment ions of fatty acyl chains in carboxylate anion forms ([RCOO]^–) were detected in negative ion mode, with deprotonated phosphoethanolamine fragment ion as a signature ion of PE (Fig. S1B).

Fig. 2. Untargeted cerebrospinal fluid (CSF) lipidomics using liquid chromatography coupled to tandem mass spectrometry (LC-MS/MS).

(A) Stacked bar graphs with the number of lipid species identified from each class written inside the parenthesis are shown. A stacked bar graph on the right is a magnified view of the upper area of the bar graph on the left. Cer indicates ceramides; ChE, cholesteryl esters; DG, diacylglycerols; MHC, monohexosylceramides; PC, phosphatidylcholines; PE, phosphatidylethanolamines; PG, phosphatidylglycerols; PI, phosphatidylinositols; SM, sphingomyelins; TG, triacylglycerols. (B) Principal component analysis on CSF lipids analyzed in the study is shown between controls (green) and patients with Alzheimer disease (AD, red). Each circle represents quantification of CSF lipidome of each individual.

We quantified CSF lipid levels by calculating the peak areas of all identified lipids, with reference to internal standards (Table S1). Principal component analysis of CSF lipids highlights the distinction between the controls and AD dementia group (Fig. 2B).

Differentially expressed CSF lipids in AD

Changes in each class of lipids were calculated by adding peak areas of individual species sharing the same head group. Total levels of several classes of phospholipids, including PC, PE, PG, PI, MHC, SM, and ChE were significantly decreased (P<.05) in patients with MCI and AD. In contrast, levels of Cer, DG, and TG were generally increased in these patients although their changes were not significant (Fig. S2). Of the 245 CSF lipids that were quantified, several species were differentially expressed (fold-change >1.5 or <0.67 and P<.05) in the AD group compared with controls (Fig. 3A). Comparison of the CSF lipidome of the AD group with that of the MCI group, by correlation analysis based on the fold-changes with reference to controls, resulted in a correlation coefficient of 0.76 (Fig. 3B; each point represents an individual peak area of CSF lipid species). This indicates that the overall lipidome of individuals with MCI and AD dementia share a similar trend of alterations. In Fig. 3B, the points in red indicate the species that were significantly increased (fold-change >1.5 and P<.05) in the AD group while those in blue indicate the species that were significantly decreased (fold-change <0.67 and P<.05) in the AD group. All of these lipids showed same trend in MCI with the coefficient of correlation for this subset of lipids being 0.91. Among these lipids, 39 species showed more significant (fold-change >1.5 or <0.67 and P<.01) changes (Table 2) and several species from phospholipid and sphingolipids were significantly altered (Fig. 4A–D). The lipids that were significantly different in AD dementia showed same trend of changes in MCI, only with lesser degree of change.

Fig. 3. Changes in cerebrospinal fluid (CSF) lipids associated with Alzheimer disease (AD).

(A) Volcano plot of CSF lipids analyzed in the study is shown with points in red indicating lipid species that were significantly increased (fold-change >1.5 and P<.05) in the AD group compared with controls. The points in blue represent species that were significantly decreased (fold-change <0.67 and P<.05). Points in gray represent lipids that were statistically unchanged (0.67<fold-change>1.5 and P>.05). DG indicates diacylglycerols; PC, phosphatidylcholines; PE, phosphatidylethanolamines; PI, phosphatidylinositols; TG, triacylglycerols. (B) Correlation analysis of AD-associated CSF lipids between mild cognitive impairment (MCI) and AD is shown with the coefficient of correlation, R². Points in red indicate lipid species that were significantly increased (fold-change >1.5 and P<.05) in the AD group compared with controls; points in blue represent species that were significantly decreased (fold-change <0.67 and P<.05). Correlation plot is based on fold-change of each species in MCI and AD, with respect to controls

Table 2.

Lipids in Cerebrospinal Fluid That Were Significantly Different^a in Patients With AD vs Controls, and Their Corresponding Changes in Patients With MCI

		AD vs Control		MCI vs Control
Molecular species	m/z	Fold-change	P value	Fold-change	P value
20:0-LysoPC	552.40	0.45	<.01	0.80	.18
16:0/14:0-PC	706.54	0.66	<.001	0.81	.03
16:0/16:0-PC	734.57	0.65	<.001	0.77	.01
16:0/18:3-PC	756.55	0.65	<.001	0.79	<.01
16:0/22:4-PC	810.60	0.65	<.01	0.76	.04
16:0e/22:4-PC	796.62	0.66	<.01	0.85	.24
18:0/18:1-PC	788.62	0.66	<.001	0.80	.02
18:0/20:2-PC	814.63	0.67	<.001	0.83	.06
18:1/18:1-PC	786.60	0.66	<.001	0.81	.04
18:1/20:3-PC	810.60	0.66	<.01	0.77	.05
18:0-LysoPE	480.31	0.63	<.01	0.80	.06
16:0/16:0-dMePE	718.54	0.63	<.001	0.83	.15
16:0e/22:4-dMePE	780.59	0.60	<.001	0.69	.02
18:0/18:2-dMePE	770.57	0.59	<.01	0.67	.01
16:0/22:6-PE	762.51	0.63	<.001	0.76	.09
18:0/20:4-PE	766.54	0.66	<.001	0.78	.07
18:0/22:4-PE	794.57	0.50	<.001	0.68	.02
18:0/22:6-PE	790.54	0.64	<.01	0.71	.05
18:0e/22:4-PE	780.59	0.60	<.001	0.69	.02
18:0e/22:6-PE	776.56	0.63	<.001	0.69	.01
18:0p/22:4-PE	778.58	0.50	<.001	0.82	.34
18:0p/22:6-PE	774.54	0.52	<.001	0.85	.35
16:0/18:1-PI	835.53	0.59	<.001	0.82	.10
18:0/18:1-PI	863.57	0.60	<.01	0.77	.07
18:0/20:4-PI	885.55	0.62	<.001	0.78	.05
18:0/22:4-PI	913.58	0.58	<.001	0.73	.02
18:0/22:5-PI	911.57	0.61	<.001	0.78	.01
18:0/22:6-PI	909.55	0.59	<.01	0.78	.12
18:1/18:2-PI	859.53	0.60	<.01	0.82	.24
18:1/20:4-PI	883.53	0.66	<.001	0.80	.08
d18:1/18:0-MHC	728.60	0.65	<.001	0.76	<.01
d18:1/20:0-MHC	756.63	0.63	<.001	0.76	.02
d18:1/24:1-MHC	810.68	0.64	<.01	0.71	<.01
d18:2/18:0-MHC	726.59	0.64	<.01	0.76	.07
d18:2/24:1-MHC	808.67	0.62	<.01	0.72	.03
d18:1/18:0-SM	731.61	0.66	<.001	0.80	.05
18:1-ChE	668.63	0.64	<.001	0.81	.02
18:2-ChE	666.62	0.65	<.001	0.85	.12
22:4-ChE	718.65	0.64	<.001	0.79	.04

Abbreviations: AD, Alzheimer disease; ChE, cholesteryl ester; dMePE, dimethsylphosphatidylethanolamine: MCI, mild cognitive impairment; MHC, monohexosylceramide; m/z, mass/charge ratio; PC, phosphatidylcholine; PE, phosphatidylethanolamine; PI, phosphatidylinositol; SM, sphingomyelin.

^aFold-change >1.5 or <0.67 and P<.01.

Fig. 4. Subset of cerebrospinal fluid (CSF) lipids with significant alterations in Alzheimer disease (AD).

Relative peak area of (A) monohexosylceramides (MHC), (B) phosphatidylcholines (PC), (C) phosphatidylethanolamines (PE), and (D) phosphatidylinositols (PI) that were altered significantly in the AD (red) group compared with controls (green) are shown, along with their corresponding changes in the mild cognitive impairment (MCI) group. Black diamond represents average peak area in each group, top and bottom whiskers indicate maximum and minimum value, respectively. The bottom of the box represents the first quartile while the top of the box represents the third quartile of relative peak area in each group.

As changes in ceramides have been observed in postmortem brain or serum of patients with AD ^{13–15, 20}, we expected to observe alterations in ceramides. While the total level of ceramide in CSF was increased in AD, that of MHC was found to be decreased (Fig. S2). We identified a subset of MHC species—d18:1/18:0-, d18:1/20:0-, d18:1/24:1-, d18:2/18:0-, d18:2/22:0-, d18:2/24:0-, and d18:2/24:-MHC—that was significantly decreased in AD (Fig. 5A). MHC refers to ceramide with one hexose including glucose or galactose. MHC can be synthesized from ceramide by glucosylceramide (GluCer) or galactosyl synthase (Fig. 5B). Significantly decreased activity of GluCer synthase has been reported in a study of AD brains, especially in the frontal cortex ²² and we speculate that GluCer synthase in CSF is also possibly downregulated, which can contribute to decreased CSF MHC (Fig. 5B). Axonal growth regulates the activity of GluCer synthase, and decreased activity of GluCer synthase indicates a lower rate of axonal and dendrite growth, which relates to the neuronal decay observed in MCI and AD. To our knowledge, this is the first study to show a decrease of MHC in the CSF of patients with AD dementia.

Fig. 5. Decrease in cerebrospinal fluid (CSF) monohexosylceramides (MHC) in patients with mild cognitive impairment (MCI) and Alzheimer disease (AD).

(A) Heatmap of MHC species in controls, MCI and AD. Each square represents average peak area of each species in each group indicated at the top of the heatmap. (B) Pathway of ceramide metabolism. Glucosylceramide (GluCer) and galactosylceramide (GalCer), which are representative forms of MHC, can be synthesized from ceramide by GluCer synthase and GalCer synthase, respectively.

A correlation between CSF Aβ42 and CSF SM has been reported in cognitively normal individuals with parental history of AD ¹⁴. CSF Aβ42 is decreased in AD compared to controls (Table 1) and the significantly decreased level of d18:1/18:0-SM in AD suggests that positive correlation between CSF Aβ and SM is also seen in patients with AD (Table 2). In addition to changes in MHC and SM, several phospholipid species, including PC, PE, and PI, were decreased significantly in AD (Fig. 4B–D), which is similar to those reported in postmortem AD brains ^23–25. Amyloid deposition in AD is known to inhibit production of new synapses ^{26, 27}, and phospholipids are increased in synaptogenesis because enhanced synaptic activities result in increased neurotransmitters and synthesis of phospholipids ^{28, 29}. In our study, a majority of phospholipid species were decreased in the participants with AD, and this indicates a decrease in synaptic activity..

Phospholipase A₂ (PLA₂) converts PC to lysoPC and higher activity of PLA₂ is reportedly associated with an increased risk of dementia and AD ³⁰. PC species have been reported to increase with age, and decreased levels of CSF PC species were seen in patients with MCI and AD (Fig. 4B), which indicates that altered PC may have a role in AD pathologic progression. PLA₂ has important roles in synaptic signaling, learning, and memory; thus, decreased levels of several PC species in our study suggest decline in neuronal and synaptic activity. Decreased levels of PC species in our study are consistent with other studies that have reported decreased levels of plasma PC in patients with AD by increased PLA₂ activity and amyloid accumulation ^{31, 32}.

In addition to PC, PE species were downregulated in both the MCI and AD groups (Table 2 and Fig. 4C). Functions of PE lipids are related to mitochondrial biogenesis activity, oxidative phosphorylation, and autophagy ^33–36. Ether lipids contain an ether-linked alkenyl or alkyl chain at the sn-1 position of fatty acids instead of an ester linkage, which makes ether lipids unique with different properties including hydrophilicity. Lack of a carbonyl oxygen at the sn-1 enables stronger intermolecular hydrogen bonding between lipids in membranes. In many studies, altered metabolism of ether lipids with alkenyl chain has been associated with AD pathology as a major biochemical change of lipids occurring inside the brain tissue of AD ^37–39. Decreases in ether PE are known to be correlated with severity of cognitive decline, and significant decreases in ether PE have been documented in postmortem brain tissues of decedents with AD, although it is unknown whether depletion of ether PE is the cause or effect of the pathologic processes in AD ³⁸.

Similar results of depleted ether PE species, including 18:0e/22:4, 18:0e/22:6, 18:0p/22:4, and 18:0p/22:6-PE, were observed in our study (Table 2). Peroxisomes are essential in ether PE synthesis and metabolism of long-chain fatty acids ⁴⁰. Because ether PE–containing long-chain fatty acids showed marked decreases in patients with AD (Fig. 4C), a deficiency of peroxisomes is suspected in AD. Considering that depletion of ether PE is known to be specific to AD brain—its decrease was not observed in other neurodegenerative diseases including Parkinson disease—decreased levels of the aforementioned species in CSF could be effective biomarkers of AD ⁴¹. Impaired synthesis of ether lipids has been linked to defects in myelination and myelin sheaths in AD ^{42, 43}, which could explain the decreased levels of ether PE in our results. Decreases in ether lipids are known to reduce membrane fluidity, enzyme activity, diffusion of signal transduction molecules, and vesicular fusion ⁴¹. By the downregulation of ether PE seen in our study, we can assume alterations in these aforementioned activities in CSF of patients with AD.

There is increasing evidence linking diabetes to AD, because patients with diabetes are known to be at greater risk for AD or other types of dementia ⁴⁴. Diabetes is known to cause impaired insulin action and cognitive function, along with oxidative stress in the brain, resulting in hyperlipidemia and hyperinsulinemia ^45–49. TG is elevated in diabetes and increased levels of CSF TG species in AD were observed in our study. Some studies have reported increased plasma TG levels in participants with AD ^{50, 51}. As CSF TG levels were increased in our study, it can be deduced that increased TG is a shared trait of plasma and CSF in persons with AD. Increases in TG have been reported to precede the onset of amyloid deposition in mouse models of AD ⁵², and we observed the same phenomenon of increased TG levels in the CSF. TG species were also increased in the participants with MCI compared with controls, although greater increases were noted in AD.

We discovered that 18:0p/22:6-PE, which contains docosahexaenoic acid (DHA, 22:6) at sn-2 of the glycerol backbone, was significantly decreased in AD. DHA is an ω-3 fatty acid, which has neuroprotective roles and is enriched in brain ⁵³. Essential for normal neuronal functions, ω-3 fatty acids have been observed to be decreased in older persons and patients with AD. Administration of DHA has been reported to alleviate neuronal dysfunction and correct abnormal neurotransmitter release ^{54, 55}. Lower levels of DHA from postmortem AD brain tissue have been previously documented ⁵⁵. In our study, in addition to 18:0p/22:6-PE, other CSF lipids containing DHA in the fatty acyl chain were also found in lower levels in subjects with AD dementia and also in MCI, although to a lesser extent (Fig. 6).

Fig. 6. Changes in cerebrospinal fluid (CSF) lipids with docosahexaenoic acid (DHA).

Relative peak areas of lipid species containing DHA at sn2 and (A) ether linked fatty acid, (B) 18:0 and (C) 16:0 at sn1 of the glycerol backbone are shown in controls (green), mild cognitive impairment (MCI, purple) and Alzheimer’s disease (AD, red) groups. The structures of lipids are shown in each panel with R representing fatty acyl chain and X the different head groups of phospholipids. Black diamond represents average peak area in each group, top and bottom whiskers indicate maximum and minimum value, respectively. The bottom of the box represents the first quartile while the top of the box represents the third quartile of relative peak area in each group.

CSF lipid biomarkers discovered in the study were then assessed with paired area under the receiver operating characteristic curve (AUC-ROC) analysis. Lipid species that showed significant alterations in the AD group were selected for individual and paired AUC-ROC analysis. Although several species exhibited relatively high AUC as individual markers, the combinational AUC-ROC analysis resulted in improved AUC (Table 3). Among them, the AUC from the combinations of 16:0/18:3-PC to 18:0/18:1-PI and 16:0/18:3-PC to d18:2/24:1-MHC were extremely high—0.977 and 0.958, respectively (Fig. 7). The results of paired AUC-ROC analysis demonstrate that selected lipid pairs may be more effective at discriminating patients with AD dementia from cognitively normal individuals.

Table 3.

Area Under the Receiver Operating Characteristic Curve for Pairs of Cerebrospinal Fluid Lipid Biomarkers

Lipid	16:0/18:3-PC	18:0/18:1-PI	d18:2/24:1-MHC	16:0/18:1-PI	18:0/22:5-PI	18:0/22:4-PE
16:0/18:3-PC	0.927	0.977	0.958	0.954	0.954	0.948
18:0/18:1-PI		0.810	0.814	0.928	0.954	0.954
d18:2/24:1-MHC			0.797	0.928	0.935	0.941
16:0/18:1-PI				0.922	0.935	0.944
18:0/22:5-PI					0.930	0.944
18:0/22:4-PE						0.895

Abbreviations: MHC, monohexosylceramide; PC, phosphatidylcholine; PE, phosphatidylethanolamine; PI, phosphatidylinositol.

Fig. 7. Area under the receiver operating characteristic curve (AUC-ROC) analysis for assessment of cerebrospinal fluid lipid biomarkers of Alzheimer’s disease.

On the top and bottom left, overlapped ROC curves of individual lipids with the calculated AUC values are shown. On the top and bottom right, ROC curves of paired lipids from the left are illustrated with corresponding AUC values. MHC indicates monohexosylceramide; PC, phosphatidylcholine; PI, phosphatidylinositol.

Sex-Specific Lipid Biomarkers of AD

Sex differences in the development and prevalence of AD dementia have been reported, with women at greater (≈2-fold) risk for AD and men at greater risk for vascular dementia ⁵⁶. Men also have also been reported to have a more rapid cognitive decline than women, which might be due to biological differences such as lipid profiles between sexes ⁵⁶. Few studies have explored sex-specific alterations of lipids in AD and, in this study, we examined lipids that are differentially expressed in each sex. In the control group, there were 7 males and 11 females while AD group was consisted of 10 males and 7 females. Many lipid species showed a similar degree of changes in both sexes compared with sex-matched controls. However, some of the species were significantly altered only in women or men with AD (Table 4). Selected sphingolipids, especially, showed changes in women they were unchanged in men (Fig. 8). Although the mechanisms underlying differential lipidomic profiles between sexes are unknown and the number of samples for each sex in the study was relatively small, our study showed that sex-specific lipid markers of AD may exist. In order to robustly discovery and validate any sex-specific lipid markers, larger studies will be required.

Table 4.

Significant Sex-Specific Lipid Alterations^a Associated With Alzheimer Disease

Molecular species	m/z	Women		Men
		Fold-change	P value	Fold-change	P value
16:0e/22:6-PC	792.59	0.60	.02	0.90	.68
16:0p/18:1-PC	744.59	0.54	.04	0.94	.82
18:0e/18:2-PC	772.62	0.60	.02	0.84	.39
18:2/18:2-PC	782.57	0.65	.04	0.83	.53
16:0/20:4-dMePE	766.54	0.62	.02	0.77	.37
16:0p/20:4-PE	722.51	0.55	.04	0.99	.96
18:0e/20:4-PE	752.56	0.60	.03	0.83	.39
18:0p/20:4-PE	750.54	0.50	.03	0.83	.39
d18:1/24:1-Cer	648.63	0.55	<.01	1.01	.97
d18:0/24:0-MHC	814.71	0.49	<.01	0.92	.64
d18:1/16:0-MHC	700.57	0.58	.01	0.94	.60
d18:1/24:0-SM	815.70	0.64	.02	0.89	.50
d18:1/24:1-SM	813.68	0.59	<.01	0.90	.52
d18:1/24:2-SM	811.67	0.64	.01	0.93	.68
d18:2/18:1-SM	727.57	0.58	<.01	0.93	.68
18:1-LdMePE	506.33	0.95	.90	0.57	.05
16:1/20:4-dMePE	764.52	1.16	.62	0.62	.03
16:1/22:6-dMePE	788.52	1.11	.73	0.60	.04
16:0e/22:4-PE	752.56	0.87	.55	0.51	.03
18:0/22:5-PE	792.55	0.83	.39	0.53	.04
40:0-TG	712.64	1.03	.90	0.62	<.01
54:0-TG	908.86	0.87	.73	0.61	.02

Abbreviations: Cer, ceramide; dMePE, dimethylphosphatidylethanolamine; LdMePE, lysodimethylphosphatidylethanolamine, MHC, monohexosylceramide; m/z, mass/charge ratio; PC, phosphatidylcholine; PE, phosphatidylethanolamine; PG, phosphatidylglycerol; SM, sphingomyelin; TG, triacylglycerol.

^aFold-change >1.5 or <0.67 and P<.05 vs sex-matched controls.

Fig. 8. Sex-specific lipids showing significant alterations in male or female patients with Alzheimer disease (AD).

Relative peak area of (A) d18:1/24:1-SM, (B) d18:0/24:0-MHC and (C) 18:0/22:5-PE in female (closed) and male (open) controls (green) and AD patients (red) are shown with horizontal line representing the average peak area observed in each group. MHC indicates monohexosylceramide; PE, phosphatidylethanolamine; SM, sphingomyelin.

Study limitation

One critical limitation of our study is the lack of apolipoprotein E (ApoE) genotyping as it relates to lipids in this study, given the known risk of ApoE in sporadic late-onset AD. The AD dementia participants did not have blood samples collected for DNA extraction. We are planning to address this limitation in ongoing studies.

Conclusion

A subset of CSF lipids showed similar changes in subjects with MCI and AD dementia compared to controls. We will validate the lipid biomarkers identified in this study in the future to determine whether these changes are predictive of progression. Also, persons with the ApoE ε4 allele are known to be at higher risk for AD than those with other alleles ^{57, 58}. In future studies, we will investigate changes in lipidomic profiles among patients with different ApoE genotypes to understand how different genotypes can affect lipid metabolism.