Exploration of Plasma Lipids in Mild Cognitive Impairment due to Alzheimer’s Disease

Abstract

Background:

Lipids have important structural roles in cell membranes and changes to these membrane lipids may influence β- and γ-secretase activities and thus contribute to Alzheimer’s disease (AD) pathology.

Objective:

To explore baseline plasma lipid profiling in participants with mild cognitive impairment (MCI) with and without AD pathology.

Methods:

We identified 261 plasma lipids using reversed-phase liquid chromatography/mass spectrometry in cerebrospinal fluid amyloid positive (Aβ+) or negative (Aβ–) participants with MCI as compared to controls. Additionally, we analyzed the potential associations of plasma lipid profiles with performance on neuropsychological tests at baseline and after two years.

Results:

Sphingomyelin (SM) concentrations, particularly, SM(d43:2), were lower in MCI Aβ+ individuals compared to controls. Further, SM(d43:2) was also nominally reduced in MCI Aβ+ individuals compared to MCI Aβ–. No plasma lipids were associated with performance on primary neuropsychological tests at baseline or between the two time points after correction for multiple testing.

Conclusion:

Reduced plasma concentrations of SM were associated with AD.

Keywords

Alzheimer’s disease lipid mild cognitive impairment sphingomyelin

INTRODUCTION

Alzheimer’s disease (AD) is a heterogeneous disorder where both amyloid and non-amyloid centric mechanisms could play different causative roles for the manifestations of the disease, i.e., familial versus sporadic AD [1, 2]. The hallmarks of AD are extracellular plaques mainly containing amyloid-β (Aβ) peptides and intracellular neurofibrillary tangles consisting of hyper- and abnormally phosphorylated tau protein, both of which are reflected in the concentrations of the cerebrospinal fluid (CSF) biomarkers Aβ₄₂ and CSF phosphorylated tau (p-tau), respectively [1]. Despite current clinical trials focusing on altering amyloid metabolism [3], and reports of some positive results [4], no effective disease-modifying treatment is currently available [5]. It is therefore crucial to explore other disease mechanisms as potential novel treatment targets, and to that end, novel non-amyloid markers as low-cost and feasible diagnostic and prognostic blood-based biomarkers are warranted [6].

The human brain is a lipid-rich organ, and neuronal membranes are highly abundant of cholesterol, glycerophospholipids, and sphingolipids [7]. Lipids serve important structural and regulatory roles in cellular membrane formation, including cellular transport, protein stabilization and modulation, cell signalling, and regulation of gene expression [6].

The cellular membrane exhibits lipid rafts, liquid-ordered domains rich in cholesterol, sphingolipids, including sphingomyelin (SM), and glycerophospholipids [7].

Amyloid-β protein precursor (AβPP), β- and γ-secretases are all transmembrane proteins, hence changes to the lipid raft and its composition and function might contribute to changes in β- and γ-secretase activities and consequently affect the production of Aβ₄₂ in AD [6, 8].

Interestingly, several studies report altered blood lipid levels in sporadic AD pathology [9 –12]. Lipids measured in blood could serve as a potential blood-based biomarker of AD with great advantage over CSF with regard to feasibility, invasiveness, and cost [6]. However, as the relationships between systemic abnormalities in metabolism and the pathogenesis of AD are poorly understood, Varma et al. undertook parallel metabolomics analyses in blood and postmortem brain tissue, and concluded that perturbations in sphingolipid metabolism were associated with AD pathology [13].

Investigating lipid homeostasis alterations during AD pathogenesis will complement the proteomic approaches channeled toward the development of early diagnosis of AD and possibly also AD progression [6]. These studies should be done in well characterized longitudinal cohorts aiming to link blood-based lipidomic changes with neuropathology and to integrate findings with known genomic and proteomic alterations in AD, and might make way for novel disease-modifying treatments [6, 11] although this has not always been the case in most previous studies.

The objective of this study was to explore as many lipids as possible and identify plasma lipid profiles in patients with mild cognitive impairment (MCI) due to AD. We hypothesized that patients with MCI-AD have a specific plasma lipid profile, and that this profile is associated with cognitive impairment.

METHODS

Materials

Participants were drawn from the Norwegian multicenter longitudinal cohort study “Dementia Disease Initiation” (DDI), which from 2013 onwards included participants with cognitive impairment and normal controls who completed neuropsychological testing and a comprehensive biomarker program at baseline and two years later as described previously [3]. The cognitive examination battery included the Mini-Mental State Examination (MMSE) [14], verbal learning and memory (CERAD word list test) [15], visuoperceptual ability (VOSP silhouettes) [16], psychomotor speed (Trail Making Test A: TMT-A), divided attention (TMT-B) [17], and verbal fluency (COWAT) [18]. Except for the MMSE, standardized T-scores (M = 50, SD = 10) were calculated for the tests based on demographically adjusted norms [16, 19, 20].

We identified 50 participants with MCI due to AD based on CSF measurements, MCI Aβ+ (as defined below), who were sex and age matched by manual matching with 50 Aβ–MCI and 50 healthy controls. One participant was later excluded due to being reclassified as not having MCI. There were a few (1–5) missings on cognitive tests, and 138 had complete follow-up cognitive assessment after an average of 24.5 months.

Cognitive test battery and standardized classification of MCI diagnosis

The NIA-AA criteria were used to diagnose MCI, requiring reporting of subjective cognitive impairment or decline, verified objectively by low performance on clinical cognitive tests in one or more cognitive domains [21, 22]. The cutoff value for MCI (defined as normal versus abnormal cognition) was results ≤1.5 standard deviations below the age, sex, and education adjusted normative mean on either CERAD word list (delayed recall) [19], TMT-B, COWAT [20], or VOSP silhouettes (this test was adjusted for age only) [16].

CERAD memory composite score

In order to provide a robust measure of memory function, we constructed a memory composite score comprising subtests from The Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) word list memory test (WLT). The composite included CERAD subtest total learning, recall, and recognition and was constructed using an established method for cognitive composites [23, 24]. In order to provide normative adjustment for pertinent demographics, a regression-based norming procedure [19, 25] was employed using n = 146 healthy controls from the DDI cohort [3]. Standardized T-scores were then calculated for the participants in the present study. (See the Supplementary Material, including Supplementary Tables 1–3 for a full description).

Blood and cerebrospinal fluid

Blood samples were drawn, collected, and handled according to standardized procedures. Total cholesterol, high-density lipoprotein (HDL) cholesterol, low-density lipoprotein (LDL) cholesterol, and triglycerides measurements were performed in serum locally at each center according to local procedures. At two centers, LDL cholesterol measurements were not done. For lipid profiling analyses EDTA blood samples were drawn, centrifuged at 1200 g for 13 min before plasma was aliquoted in polypropylene tubes and stored at –80°C. Time from venipuncture until aliquoted plasma was frozen was below 2 h. Plasma was kept at –80°C until analysis.

Plasma lipid profiling was performed at King’s College London, UK, using methods as described elsewhere [26, 27]. Briefly, 20μL of plasma sample was added to a 2 mL Eppendorf tube. 20μL of 0.9% w/v NaCL (aq), 56μL of Chloroform/Methanol (2:1) containing 12 internal standards (10 ug/mL for all), and 184μL of Chloroform/methanol (2:1) were added to the Eppendorf tube containing samples. The mixture was then vortexed and centrifuged at 1000 g for 10 min under 4°C. Another 5μl of the plasma from each sample was taken to form the pooled QC samples. Same procedures were followed for QCs extraction. QC sample was running in between every 10 samples. The standard solution contained the following compounds: Trioctanoin-1,1,1–13C3 (TG(8:0/8:0/8:0)–13C3), 1-heptadecanoyl-2-hydroxy-sn-glycero-3-phosphocholine (LPC(17:0)), N-hepta-decanoyl-D-erythro-sphingosine (Cer(d18:1/17:0)), 1,2-dimyristoyl-sn-glycero-3-phospho(cholin-e-d13) (PC(14:0/d13)), N-heptadecanoyl-D-erythro-sphingosylphosphorylcholine (SM(d18:1/1-7:0)), 1,2-diheptadecanoyl-sn-glycero-3-phosphoethanolamine (PE(17:0/17:0)), 1,2-diheptadec-anoyl-sn-glycero-3-phosphocholine (PC(17:0/17:0)), 1-palmitoyl-d31-2-oleoyl-sn-glycero-3-pho-sphocholine (PC(16:0/d31/18:1)), 1-palmitoyl-d30-2-oleoyl-sn-glycero-3-phosphocholine (PC(1–6:0/d30/18:1)), Tripalmitin-1,1,1–13C3 (TG(16:0/16:0/16:0)–13C3), PG(17:0/17:0), PS(17:0/17:–0). The lipids containing lower chloroform layer was extracted for reverse phase analysis using Waters ACQUITY ultra-high performance chromatography coupled with Xevo® quadrupole time-of-flight mass spectrometer (UHPLC-QTOFMS). Separations were performed on an ACQUITY UPLC BEH C18 column (2.1 mm×100 mm, particle size 1.7μm) by Waters (Milford, CT, USA). The flow rate was 0.4 ml min–1 and the injection volume was 5μl. H2O + 1% NH4Ac (1M) + 0.1% HCOOH (A) and ACN:IPA (1:1, v/v) + 1% NH4Ac + 0.1% HCOOH (B) were used as the mobile phases for the gradient elution. The gradient was as follows: from 0 to 2 min 35–80% B, from 2 to 7 min 80–100% B, and from 7 to 14 min 100% B. Each run was followed by a 7 min re-equilibration period under initial conditions (35% B). In terms of the mass spectrometric condition, for positive mode, a capillary voltage of 3.6 kV and a cone voltage of 35 V were used. The desolvation gas flow was 840 L/h and the source temperature was 120°C. All analyses were acquired using the lock spray to ensure accuracy and reproducibility; leucine enkephalin was used as lock mass (m/z 556.2771 and 278.1141) at a concentration of 200 ng/mL and a flow rate of 10μL/min. Data were collected in the centroid mode over the mass range m/z 100–1700 with an acquisition time of 0.5 s a scan. For negative mode, a capillary voltage of –4.3 kV and a cone voltage of 45 V were used. Desolvation gas flow and desolvation temperature were fixed at 800 L/h and 350°C, respectively. The reference solution (leucine enkephalin, m/z 554.2639) was infused with the same conditions described in the positive mode. Lipidomic data were pre-processed with Skyline (v19.1) and peaks were identified based on the internal peak library. Peak areas were normalized by lipid class-specific internal standards.

The lumbar puncture procedure and CSF analyses, including Aβ₄₂ and total tau (t-tau) and phosphorylated tau (p-tau) analyzed at the Department of Interdisciplinary Laboratory Medicine and Medical Biochemistry at Akershus University Hospital have been described previously [3, 28]. The CSF Aβ₄₂ measurements were dichotomized using a cutoff of ≤708 ng/L previously determined in a DDI PET [18F]-Flutemetamol uptake study [29].

Ethics

All participants signed a written informed consent and the study was approved by the Regional Ethics Committee (2013/150). The entire study conduct was in line with the guidelines provided by the Helsinki declaration of 1964 (revised 2013) and the Norwegian Health and Research Act.

Statistical analyses

Demographic and serum lipids statistical analyses

SPSS version 26 was used for statistical analyses for demographical baseline data, serum lipids, and lipoproteins. Normality was assessed by inspection of QQ-plots, histograms, and the Shapiro-Wilk test of normality. For continuous variables with normal distributions, between-group comparisons were carried out using one-way ANOVA. For the continuous variables of non-normal distributions, between-group comparisons were performed with the Kruskal-Wallis tests. For statistically significant ANOVA, post-hoc Bonferroni (equal variances assumed) or Thamhane’s T2 (equal variances not assumed) were applied. For Kruskal-Wallis tests Bonferroni adjusted Dunn’s pairwise comparisons were performed. For dichotomous variables, between group comparisons were performed using Chi-square tests.

Plasma lipid analyses

RStudio (1.2.1335) was used for statistical analyses of the plasma lipids. All plasma lipids were normalized using inverse normal transformation (INT). Two participants with extremely high body mass index (BMI) were removed as outliers.

The primary analyses were the associations of plasma lipids with a) diagnosis, i.e., MCI Aβ+, MCI Aβ–and controls, and b) memory function as measured by the baseline CERAD composite T-score and the change between baseline and follow-up as residualized change in CERAD composite T-score as described below.

Secondary analyses were the associations of plasma lipids with the baseline of the other cognitive tests, i.e., CERAD learning T-score, CERAD recall T-score, TMT A T-score, TMT B T-score, COWAT T-score, and VOSP T-score, and the change between baseline and follow-up for each cognitive test.

In preliminary analyses, linear regression analyses were used in order to investigate the association of each plasma lipid with each potential covariate (age, sex, education, BMI, HDL, LDL, triglycerides (TG), diabetes mellitus, hypercholesterolemia (HC), hypertension (HT), lipid lowering medication, smoking status, and APOE). In the main univariate analyses both logistic and linear regression analyses were performed with diagnosis and cognition as the respective outcomes. Briefly, logistic regression was used to investigate the association of lipids with diagnosis at baseline and linear regression analyses were used to investigate the association of lipids with CERAD composite T-score at baseline and the change in CERAD composite T-score between baseline and follow-up. Linear regression analyses were also run using all secondary outcomes. All logistic and linear regression analyses were adjusted for BMI, HC, HT, and smoking status. As a next step the analyses were adjusted for APOE status. To calculate the change in all cognitive outcomes between baseline and follow-up, each cognitive test at follow-up was regressed against the baseline and the residuals were used (further adjusted for months of follow-up). A Bonferroni threshold of p < 0.0007 (0.05/70) was used where 70 is the number of lipid principal components explaining >95% of variation in lipids following principal component analysis. For the binary outcomes, the odds ratios (OR) represent the odds ratio for being MCI Aβ+ per 1–SD (INT transformed) compared to controls and compared to MCI Aβ-metabolite concentration and for the continuous outcomes the β-regression coefficients (beta) represent the change in the respective T-score between baseline and follow-up per 1–SD (INT transformed) metabolite concentration.

As lipids are highly correlated and the number of variables exceed that of the observations (p > n), multivariate analysis was also performed on the main outcomes to observe whether associations between the lipids and the tested outcomes remained when taking into account lipids’ intercorrelation, and to identify which lipids and combinations of lipids are strong contributors to the outcomes. Two types of multivariate analyses, PLS-DA and Random Forests (RA), were run on the main outcomes. All lipids were regressed against all covariates and the lipid residuals were used for downstream multivariate analyses. Internal cross-validation was used (data was internally split into 75-25 train-test and 1000 bootstraps took place and average results are presented).

RESULTS

The baseline characteristics of the 149 participants are presented in Table 1. There were no significant differences between the groups regarding age, sex, education, medical history, smoking status, BMI, or serum lipids and lipoproteins. The three groups differed regarding cognitive test scores (Table 1). A total of 261 lipids were identified, and annotated as ceramides (Cer), diacylglycerols (DG), phosphatidylcholines (PC), Lysophosphatidylcholines (LPC), phosphatidylethanolamines (PE), phosphatidylinositols (PI), sphingomyelins (SM), or triglycerides (TG). Most lipids were found to be inter-correlated (Supplementary Figure 1) and associated with many of the covariates (Supplementary Figure 2).

Table 1

Baseline demographic and clinical characteristics

Variable	Groups			F/χ², (p)	ANOVA post-hoc (Bonferroni/Tamhane’s)/Dunn’s pairwise comparisons (p)
	1. Controls n = 50	2. MCI Aβ₄₂+ n = 50	3. MCI Aβ₄₂– n = 49		1 vs 2	1 vs 3	2 vs 3
Age Mean (SD)	65.0 (6.8)	65.3 (8.0)	64.9 (9.5)	F = 0.03^#, (p = 0.97)
Female n (%)	25 (50)	25 (50)	24 (49)	χ² = 0.01, (p = 1.0)
Years of education Mean (SD)	13.7 (3.0)	13.9 (3.4)	13.4 (3.3)	F = 0.28, (p = 0.75)
Smoking count (%)		n = 49
Current smoker	5 (10.0)	8 (16.3)	12 (24.5)	χ² = 6.0 (p = 0.2)
Previous smoker	18 (36.0)	22 (44.9)	20 (40.8)
Non-smoker	27 (54.0)	19 (38.8)	17 (34.7)
BMI Mean (SD)	26.3 (3.6)	25.1 (3.8)	26.6 (4.7)	F = 1.83, (p = 0.17)
Medical history, count (%)
Hypercholesterolemia	16 (32)	19 (38)	18 (37)	χ² = 0.4, (p = 0.8)
Hypertension	19 (38)	12 (24)	19 (39)	χ² = 3.1, (p = 0.21)
Diabetes Mellitus	3 (6)	3 (6)	2 (4)	χ² = 0.2, (p = 0.89)
Serum lipids Mean (SD)
Total Cholesterol	5.3 (1.2)	5.7 (1.4)	5.3 (1.1)	F = 1.67, (p = 0.19)
Triglycerides median (IQR)	1.2 (0.82)	1.2 (0.82)^n = 49	1.1 (0.77)	χ² = 0.90, (p = 0.64)
Serum lipoproteins Mean (SD)
HDL Cholesterol	1.5 (0.4)	1.6 (0.4)^n = 49	1.5 (0.4)	F = 0.02, (p = 0.99)
LDL Cholesterol	3.4 (1.2)^n = 47	3.9 (1.6)^n = 24	3.1 (0.9)^n = 30	F = 2.4^#, (p = 0.05)
CERAD composite T-score Mean (SD)	52.2 (9.6)	32.3 (9.4)^n = 48	40.1 (9.1)^n = 46	F = 57.7, ( p < 0.001)	p < 0.001	p < 0.001	p < 0.001
CERAD Learning T-score Mean (SD)	51.5 (10.2)	34.8 (8.4)	40.8 (10.4)	F = 37.9, ( p < 0.001)	p < 0.001	p < 0.001	p = 0.01
CERAD Recall T-score Mean (SD)	52.3 (9.7)	31.4 (9.6)	38.9 (10.2)	F = 57.7, ( p < 0.001)	p < 0.001	p =< 0.001	p = 0.001
TMT-A T-score Mean (SD)	47.3 (9.6)^n = 49	41.0 (10.6)	42.1 (9.4)	F = 5.6, ( p = 0.004)	p = 0.006	p = 0.032	p = 1.0
TMT-B T-score Mean (SD)	49.0 (7.4)^n = 49	37.2 (13.3)	43.1 (11.1)	F = 16.4^#, ( p < 0.001)	p < 0.001	p = 0.007	p = 0.06
COWAT T-score Mean (SD)	49.8 (8.6)^n = 49	46.2 (9.7)	43.7 (7.2)^n = 48	F = 6.1, ( p = 0.003)	p = 0.12	p = 0.002	p = 0.47
VOSP T-score Mean (SD)	2.9 (10.4)^n = 49	44.0 (11.1) ^n = 47	45.8 (11.7)^n = 48	F = 8.7, ( p < 0.001)	p < 0.001	p = 0.006	p = 1.0
CSF Aβ_1 - 42 Mean (SD)	1010 (248)	546 (100)	1046 (196)	F = 172.5^#, ( p < 0.001)	p < 0.001	p = 0.81	p < 0.001
CSF t-tau Mean (SD)	334 (165)	577 (305)	373 (258)	F = 12.3^#, ( p < 0.001)	p < 0.001	p = 0.75	p = 0.002
CSF p-tau Mean (SD)	56 (20)	82 (38)	58 (32)	F = 9.2^#, ( p < 0.001)	p < 0.001	p = 1.0	p = 0.004
APOE-ɛ4 count (%)			n = 48
Heterozygote	22 (44)	23 (46)	19 (40)	χ² = 27.5, ( p < 0.001)
Homozygote	1 (2)	14 (28)	1 (2)
Non-carriers	27 (64)	13 (26)	28 (58)

n, sample size; F, F-statistic; #, Welsh correction; χ², chi-square or Kruskal-Wallis statistic; p, p-value, SD=standard deviation, IQR=interquartile range.

Plasma lipid profile and MCI-AD versus MCI non-AD

A sphingomyelin (SM(d43:2) was the only lipid associated with MCI Aβ+ compared to controls after passing correction for multiple testing, being decreased in MCI Aβ+ (OR = 0.29, 95% CI 0.14–0.56, p = 6.2×10^–4) (Fig. 1A). An additional 17 lipids were associated with MCI Aβ+ compared to controls at p < 0.05. Further, 11 of these lipids were also associated with MCI Aβ+ compared to MCI Aβ–but no association passed correction for multiple testing (Fig. 1B), with the strongest association at p < 0.05 being with a TG (TG(60:2) (OR = 2.32, 95% CI 1.30–4.41, p = 6.4×10^–3). It was also observed that SM(d43:2) was associated with MCI Aβ+ compared to MCI Aβ–at p < 0.05 (OR = 0.52, 95% CI 0.29–0.89, p = 2.1×10^–2) (Fig. 1C).

Fig. 1

Volcano plots depicting the association of the 261 lipids with the main diagnostic outcomes: Controls versus MCI Aβ+ (A) and MCI Aβ–versus MCI Aβ+ (B) following logistic regression analyses. The black lines in the volcano plots represent the p < 0.05 threshold and the red lines represent the multiple correction threshold at p < 0.0007. X-axis represents the OR (MCI Aβ+ versus Controls and MCI Aβ–, respectively) per 1–SD increase in Inverse-variance transformed lipid levels and y-axis represents –log10 transformed p-value from the logistic regression. C) Boxplot of SM.d43.2. levels (Inverse-variance transformed) in the three groups (Controls, MCI Aβ–, and MCI Aβ+).

Plasma lipid profile and cognitive impairment

No associations with CERAD composite T-score at baseline or the change in CERAD composite T-scores between baseline and follow-up passed correction for multiple testing (Fig. 2A, B). Altogether 8 lipids were associated with CERAD composite T-scores at baseline at p < 0.05, the strongest association being with a Phosphatidylinositol (PI), PI(36:4) (beta = –2.82, 95% CI –4.9 - –0.74, p = 8.2×10^–3), and no lipids were associated with the change in CERAD composite T-scores between baseline and follow-up at p < 0.05 (the strongest association being with PC(O–36:0) (beta = –0.182, 95% CI –0.37 - 0.001, p = 0.05).

Fig. 2

Volcano plot depicting the association of the 261 lipids with Baseline CERAD composite T-score (A) and change in CERAD composite T-score between baseline and follow-up (B) following linear regression analyses. The black lines in the volcano plots represent the p < 0.05 threshold and the red lines represent the multiple correction threshold at p < 0.0007. X-axis represents the beta coefficient (change in T-score per 1–SD increase in Inverse-variance transformed lipid levels) and y-axis represents –log10 transformed p-value.

Regarding the secondary outcomes, two associations passed correction for multiple testing: two PIs (PI(38:3) and PI(38:4)) were associated with baseline VOSP T-score (beta = –3.98, 95% CI –6.0 ––2.00 p = 1.12×10^–4 and beta = –3.65, 95% CI –5.59 ––1.71, p = 2.89×10^–4 respectively) (Supplementary Figure 3A-C).

The associations of all lipids with all main and secondary outcomes are presented in Supplementary Figure 4. There was modest overlap between the lipids associated with each outcome. Most of the overlap across the different outcomes was for PIs. For example, PI(36:4), was associated with MCI Aβ+ diagnosis, with CERAD composite, learning and recall, and with VOSP T-score at p < 0.05.

Multivariate data analysis highlighted that the lipids with the highest variable importance (VIP) in most models were the same lipids highlighted by univariate analysis (Supplementary Figure 5). This was most evident for the diagnosis models (MCI Aβ+ compared to controls), where the top lipids based on their VIP using PLS-DA(SM(d42:3) and SM(d43:2)) and RF (SM(d43:2)) were also the top molecules in univariate associations. The PLS-DA model predicted MCI Aβ+ with 0.624 accuracy (Sensitivity = 0.623, Specificity = 0.629 and AUC = 0.638; the top model included 5 components) and the RF model with 0.662 accuracy (Sensitivity = 0.691, Specificity = 0.692 and AUC = 0.662).

DISCUSSION

The main finding in the current study was that a number of plasma SM concentrations, and particularly, SM(d43:2), were lower in MCI Aβ+ individuals compared to controls. Further, SM(d43:2) was also nominally reduced in MCI Aβ+individuals compared to MCI Aβ–. Regarding the secondary outcomes, two PIs were found to be negatively associated with VOSP T score at baseline, i.e., increase in PI was associated with decrease in VOSP at baseline after correction for multiple testing.

Previous metabolomics studies have shown alterations in SM pathways in AD [13], although results are not always in agreement. A small cross-sectional study reported lower levels of plasma SM in AD patients compared to controls [30], while Toledo et al. found serum SM to be increased in AD and to be associated with worse cognitive outcomes [31]. Another study found SM changes in CSF with abnormal AD biomarkers [32]. Interlaboratory variability and methodology have been observed regarding the use of serum versus plasma and also marked differences in how they are processed. This might affect the results, further underlining the importance of consistency across laboratories [6].

Conflicting results could also be due to the stage of the disease [30], as it has been reported that the levels of serum SM vary according to the timing of the onset of memory impairment, a deficit observed early in AD pathogenesis [33]. Similar results have been found concerning CSF, as Kosicek et al. report significantly increased SM levels in CSF from individuals with prodromal AD compared to normal controls, however no change between mild and moderate AD groups and normal controls [34].

Interestingly, SM(d43:2) was detected as one of the key lipids that was altered in a study investigating the CSF lipidomic signature of amyotrophic lateral sclerosis (ALS) patients by mass spectrometry, similar to our approach, suggesting an involvement of the sphingolipid pathway in neurodegeneration in both AD and ALS [35].

The provenance of SM(d43:2) is not clear; however, in previous lipidomics validation studies using MS-MS [27], we and other laboratories have detected this lipid in plasma. Of note, in humans, SM levels across bio-fluids are largely regulated by multiple signaling pathways and are dependent on the contribution from diet but also de novo synthesis, recycling, and intestinal uptake. Moreover, gut bacteria also constitute an endogenous source of SMs that can supply SMs to the host and influence its pools [36].

PI have been found to be present in tau aggregates [37]. However, despite previous recommendations of linking longitudinal changes in lipids not only to Aβ levels, but also tau pathology [6], associations of PI or any plasma lipid with tau pathology was not within the scope of this study. While Mapstone et al. [9] found a PI to be one in ten serum lipids that can accurately predict memory loss in up to 90% of cases 2 years before the onset of dementia [9], these results could not be replicated in a later study [38]. Thus, the role of PI in cognitive impairment remains unclear.

In the current study we present norms for a CERAD composite measure. The use of a CERAD composite measure as a primary outcome could be a limitation, as this could mask domain specific cognitive functions such as learning, recall, and recognition, which are qualitatively different aspects of learning and memory. However, a composite measure may also offer a more robust and reliable index of learning and memory function. A composite score capitalizes on regression toward the mean, i.e., the participant is less likely to obtain two or more low scores on several measures of learning and memory function and may be more robust against chance low performance on one measure not related to neurodegeneration or cerebral dysfunction (e.g., low motivation or inattention during a particular test). We also investigated associations with lipids and two of the subdomain measures of the CERAD word list test (learning and delayed recall) as well as other cognitive domains (psychomotor speed, executive functions, verbal fluency/language, and visual cognition) in the secondary analyses. We did not find any significant associations between lipids and the composite score or specific subdomain measures of verbal learning and memory. The utility of this measure needs to be further explored with regard to sensitivity and specificity for AD in longitudinal follow-up cohorts.

Other limitations of this study include the rather small size of the samples and potentially the non-fasting design [39]. We have done plasma lipid analyses, which have also been done by other researchers [9, 10], while others have used serum [33]. Future interlaboratory agreement regarding methodology is of importance in order to be able to replicate the findings.

The strengths of this study include the randomized and longitudinal design and the age- and sex-matched samples. In addition, this study holds information regarding proteomics and genetics in relation to lipidomics, which has previously been identified as a potential focus for further studies [6].

Furthermore, we used logistic and linear regression analyses controlling for a number of variables that are associated with lipids and which are also known to be associated with AD and cognitive decline. These included age, sex, BMI, lipid lowering medication, smoking status, and history of hypercholesterolemia and hypertension. As the number of lipid variables is high and many lipids are inter-correlated we also employed a machine learning approach using PLS-DA and RF for the main outcomes. Results from the two approaches were in agreement, especially for the diagnosis that showed the strongest associations with lipids.

In conclusion, we found that plasma sphingomyelin concentrations, and particularly, SM(d43:2), were lower in MCI Aβ+ individuals compared to controls and also nominally reduced in MCI Aβ+ individuals compared to MCI Aβ–.

Future randomized studies with a longitudinal design, possibly with longer observational times are warranted in order to obtain additional knowledge and understanding of the lipid contribution to AD pathology.

Lipid alterations associated with AD pathology could possibly complement the proteomic approach channeled toward development of a low-cost and safe method to identify early AD pathology, progression, and potential novel treatment modalities.

Footnotes

ACKNOWLEDGMENTS

The study was financed by a grant from the Norwegian Health Association, grant number 7330. The authors would like to thank all participants and researchers involved in the DDI study.

This paper represents independent research partly funded by the National Institute for Health Research (NIHR) Biomedical Research Centre at South London and Maudsley NHS Foundation Trust and King’s College London. The views expressed are those of the author and not necessarily those of the NHS, the NIHR or the Department of Health and Social Care.

Petroula Proitsi is an Alzheimer’s Research UK Senior Research Fellow.

Especially we would like to thank Marianne Wettergreen and Berglind Gísladóttir for logistic, and practical help with sample collection, preparation, and sending, and Sandra Tecelao for technical support when generating the data file. We would like to thank statistician Ingvild Dalen for statistical guidance regarding analysis of the demographic and serum data.

Authors’ disclosures available online ().

The supplementary material is available in the electronic version of this article: .

References

Ballard

, Gauthier

, Corbett

, Brayne

, Aarsland

, Jones

(2011) Alzheimer’s disease. Lancet 377, 1019–1031.

Long

, Holtzman

(2019) Alzheimer disease: An update on pathobiology and treatment strategies. Cell 179, 312–339.

Fladby

, Pålhaugen

, Selnes

, Waterloo

, Bråthen

, Hessen

, Almdahl

, Arntzen

K-A

, Auning

, Eliassen

(2017) Detecting at-risk Alzheimer’s disease cases. J Alzheimers Dis 60, 97–105.

Schneider

(2020) A resurrection of aducanumab for Alzheimer’s disease. Lancet Neurol 19, 111–112.

Fink

, Jutkowitz

, McCarten

, Hemmy

, Butler

, Davila

, Ratner

, Calvert

, Barclay

, Brasure

, Nelson

, Kane

(2018) Pharmacologic interventions to prevent cognitive decline, mild cognitive impairment, and clinical Alzheimer-type dementia: A systematic review. Ann Intern Med 168, 39–51.

Wong

, Braidy

, Poljak

, Pickford

, Thambisetty

, Sachdev

(2017) Dysregulation of lipids in Alzheimer’s disease and their role as potential biomarkers. Alzheimers Dement 13, 810–827.

Mesa-Herrera

, Taoro-Gonzalez

, Valdes-Baizabal

, Diaz

, Marin

(2019) Lipid and lipid raft alteration in aging and neurodegenerative diseases: A window for the development of new biomarkers. Int J Mol Sci 20, 3810.

Xiang

, Lam

, Shui

(2015) What can lipidomics tell us about the pathogenesis of Alzheimer disease?. Biol Chem 396, 1281–1291.

Mapstone

, Cheema

, Fiandaca

, Zhong

, Mhyre

, MacArthur

, Hall

, Fisher

, Peterson

, Haley

, Nazar

, Rich

, Berlau

, Peltz

, Tan

, Kawas

, Federoff

(2014) Plasma phospholipids identify antecedent memory impairment in older adults. Nat Med 20, 415–418.

10.

Fiandaca

, Zhong

, Cheema

, Orquiza

, Chidambaram

, Tan

, Gresenz

, FitzGerald

, Nalls

, Singleton

, Mapstone

, Federoff

(2015) Plasma 24-metabolite panel predicts preclinical transition to clinical stages of Alzheimer’s disease. Front Neurol 6, 237.

11.

Proitsi

, Kim

, Whiley

, Simmons

, Sattlecker

, Velayudhan

, Lupton

, Soininen

, Kloszewska

, Mecocci

, Tsolaki

, Vellas

, Lovestone

, Powell

, Dobson

, Legido-Quigley

(2017) Association of blood lipids with Alzheimer’s disease: A comprehensive lipidomics analysis. Alzheimers Dement 13, 140–151.

12.

Klavins

, Koal

, Dallmann

, Marksteiner

, Kemmler

, Humpel

(2015) The ratio of phosphatidylcholines to lysophosphatidylcholines in plasma differentiates healthy controls from patients with Alzheimer’s disease and mild cognitive impairment. Alzheimers Dement (Amst) 1, 295–302.

13.

Varma

, Oommen

, Varma

, Casanova

, An

, Andrews

, O’Brien

, Pletnikova

, Troncoso

, Toledo

, Baillie

, Arnold

, Kastenmueller

, Nho

, Doraiswamy

, Saykin

, Kaddurah-Daouk

, Legido-Quigley

, Thambisetty

(2018) Brain and blood metabolite signatures of pathology and progression in Alzheimer disease: A targeted metabolomics study. PLoS Med 15, e1002482.

14.

Folstein

, Folstein

, McHugh

(1975) “Mini-mental state”: A practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res 12, 189–198.

15.

Fillenbaum

, van Belle

, Morris

, Mohs

, Mirra

, Davis

, Tariot

, Silverman

, Clark

, Welsh-Bohmer

, Heyman

(2008) Consortium to Establish a Registry for Alzheimer’s Disease (CERAD): The first twenty years. Alzheimers Dement 4, 96–109.

16.

Warrington

, James

(1991) The Visual Object and Space Perception Battery, Thames Valley Test Company, Bury St Edmunds, England.

17.

Reitan

, Wolfson

(1985) The Halstead-Reitan neuropsychological test battery: Therapy and clinical assessment. Neuropsychological Press, Tucson, AZ.

18.

Benton

, Hamsher

(1989) Multilingual aphasia examination. AJA Associates, Iowa City.

19.

Kirsebom

, Espenes

, Hessen

, Waterloo

, Harald Johnsen

, Gundersen

, Botne Sando

, Rolfseng Grontvedt

, Timon

, Fladby

(2019) Demographically adjusted CERAD wordlist test norms in a Norwegian sample from 40 to 80 years. Clin Neuropsychol 33(sup1), 27–39.

20.

Heaton

, Miller

, Taylor

, Grant

(2004) Revised Comprehensive Norms for an Expanded Halsted-Reitan Battery: Demographically Adjusted Neuropsychological Norms for African American and Caucasian Adults. Psychological Assessment Resources, Odessa.

21.

Albert

, DeKosky

, Dickson

, Dubois

, Feldman

, Fox

, Gamst

, Holtzman

, Jagust

, Petersen

, Snyder

, Carrillo

, Thies

, Phelps

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

22.

McKhann

, Knopman

, Chertkow

, Hyman

, Jack

CR Jr

, Kawas

, Klunk

, Koroshetz

, Manly

, Mayeux

, Mohs

, Morris

, Rossor

, Scheltens

, Carrillo

, Thies

, Weintraub

, Phelps

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

23.

Malek-Ahmadi

, Chen

, Perez

, He

, Mufson

(2018) Cognitive composite score association with Alzheimer’s disease plaque and tangle pathology. Alzheimers Res Ther 10, 90–90.

24.

Langbaum

, Hendrix

, Ayutyanont

, Chen

, Fleisher

, Shah

, Barnes

, Bennett

, Tariot

, Reiman

(2014) An empirically derived composite cognitive test score with improved power to track and evaluate treatments for preclinical Alzheimer’s disease. Alzheimers Dement 10, 666–674.

25.

Testa

, Winicki

, Pearlson

, Gordon

, Schretlen

(2009) Accounting for estimated IQ in neuropsychological test performance with regression-based techniques. J Int Neuropsychol Soc 15, 1012–1022.

26.

O’Gorman

, Suvitaival

, Ahonen

, Cannon

, Zammit

, Lewis

, Roche

, Mattila

, Hyotylainen

, Oresic

, Brennan

, Cotter

(2017) Identification of a plasma signature of psychotic disorder in children and adolescents from the Avon Longitudinal Study of Parents and Children (ALSPAC) cohort.e. Transl Psychiatry 7, 1240.

27.

Bowden

, Heckert

, Ulmer

, Jones

, Koelmel

, Abdullah

, Ahonen

, Alnouti

, Armando

, Asara

, Bamba

, Barr

, Bergquist

, Borchers

, Brandsma

, Breitkopf

, Cajka

, Cazenave-Gassiot

, Checa

, Cinel

, Colas

, Cremers

, Dennis

, Evans

, Fauland

, Fiehn

, Gardner

, Garrett

, Gotlinger

, Han

, Huang

, Neo

, Hyötyläinen

, Izumi

, Jiang

, Kachman

, Kiyonami

, Klavins

, Klose

, Köfeler

, Kolmert

, Koal

, Koster

, Kuklenyik

, Kurland

, Leadley

, Lin

, Maddipati

, McDougall

, Meikle

, Mellett

, Monnin

, Moseley

, Nandakumar

, Oresic

, Patterson

, Peake

, Pierce

, Post

, Postle

, Pugh

, Qiu

, Quehenberger

, Ramrup

, Rees

, Rembiesa

, Reynaud

, Roth

, Sales

, Schuhmann

, Schwartzman

, Serhan

, Shevchenko

, Somerville

, St John-Williams

, Surma

, Takeda

, Thakare

, Thompson

, Torta

, Triebl

, Trötzmüller

, Ubhayasekera

SJK

, Vuckovic

, Weir

, Welti

, Wenk

, Wheelock

, Yao

, Yuan

, Zhao

, Zhou

(2017) Harmonizing lipidomics: NIST interlaboratory comparison exercise for lipidomics using SRM 1950–metabolites in frozen human plasma. J Lipid Res 58, 2275–2288.

28.

Kirsebom

, Nordengen

, Selnes

, Waterloo

, Torsetnes

, Gisladottir

, Brix

, Vanmechelen

, Brathen

, Hessen

, Aarsland

, Fladby

(2018) Cerebrospinal fluid neurogranin/beta-site APP-cleaving enzyme 1 predicts cognitive decline in preclinical Alzheimer’s disease. Alzheimers Dement (N Y) 4, 617–627.

29.

Kalheim

, Fladby

, Coello

, Bjørnerud

, Selnes

(2018) [18F]-flutemetamol uptake in cortex and white matter: Comparison with cerebrospinal fluid biomarkers and [18F]-fludeoxyglucose. J Alzheimers Dis 62, 1595–1607.

30.

Han

, Rozen

, Boyle

, Hellegers

, Cheng

, Burke

, Welsh-Bohmer

, Doraiswamy

, Kaddurah-Daouk

(2011) Metabolomics in early Alzheimer’s disease: Identification of altered plasma sphingolipidome using shotgun lipidomics. PLoS One 6, e21643.

31.

Toledo

, Arnold

, Kastenmuller

, Chang

, Baillie

, Han

, Thambisetty

, Tenenbaum

, Suhre

, Thompson

, John-Williams

, MahmoudianDehkordi

, Rotroff

, Jack

, Motsinger-Reif

, Risacher

, Blach

, Lucas

, Massaro

, Louie

, Zhu

, Dallmann

, Klavins

, Koal

, Kim

, Nho

, Shen

, Casanova

, Varma

, Legido-Quigley

, Moseley

, Zhu

, Henrion

MYR

, van der Lee

, Harms

, Demirkan

, Hankemeier

, van Duijn

, Trojanowski

, Shaw

, Saykin

, Weiner

, Doraiswamy

, Kaddurah-Daouk

, Alzheimer’s Disease Neuroimaging Initiative, the Alzheimer Disease Metabolomics Consortium (2017) Metabolic network failures in Alzheimer’s disease: A biochemical road map. Alzheimers Dement 13, 965–984.

32.

Koal

, Klavins

, Seppi

, Kemmler

, Humpel

(2015) Sphingomyelin SM (d18:1/18:0) is significantly enhanced in cerebrospinal fluid samples dichotomized by pathological amyloid-β 42, tau, and phospho-tau-181 levels. J Alzheimers Dis 44, 1193–1201.

33.

Mielke

, Bandaru

, Haughey

, Rabins

, Lyketsos

, Carlson

(2010) Serum sphingomyelins and ceramides are early predictors of memory impairment. Neurobiol Aging 31, 17–24.

34.

Kosicek

, Zetterberg

, Andreasen

, Peter-Katalinic

, Hecimovic

(2012) Elevated cerebrospinal fluid sphingomyelin levels in prodromal Alzheimer’s disease. Neurosci Lett 516, 302–305.

35.

Blasco

, Veyrat-Durebex

, Bocca

, Patin

, Vourc’h

, Kouassi Nzoughet

, Lenaers

, Andres

, Simard

, Corcia

, Reynier

(2017) Lipidomics reveals cerebrospinal-fluid signatures of ALS. Sci Rep 7, 17652.

36.

Johnson

, Heaver

, Waters

, Kim

, Bretin

, Goodman

, Gewirtz

, Worgall

, Ley

(2020) Sphingolipids produced by gut bacteria enter host metabolic pathways impacting ceramide levels. Nat Commun 11, 2471.

37.

Talaga

, Smeralda

, Lescos

, Hunel

, Lepejova-Caudy

, Cullin

, Bonhommeau

, Lecomte

(2018) PIP2 phospholipid-induced aggregation of tau filaments probed by tip-enhanced Raman spectroscopy. Angew Chem Int Ed Engl 57, 15738–15742.

38.

Casanova

, Varma

, Simpson

, Kim

, An

, Saldana

, Riveros

, Moscato

, Griswold

, Sonntag

, Wahrheit

, Klavins

, Jonsson

, Eiriksdottir

, Aspelund

, Launer

, Gudnason

, Legido Quigley

, Thambisetty

(2016) Blood metabolite markers of preclinical Alzheimer’s disease in two longitudinally followed cohorts of older individuals. Alzheimers Dement 12, 815–822.

39.

Nordestgaard

(2017) A test in context: Lipid profile, fasting versus nonfasting. J Am Coll Cardiol 70, 1637–1646.

40.

Oosterhuis

, van der Ark

, Sijtsma

(2016) Sample size requirements for traditional and regression-based norms. Assessment 23, 191–202.