Impact of APOE on cerebrovascular lipid profile in Alzheimer’s disease

Abstract

Disturbances within the cerebrovascular system substantially contribute to the pathogenesis of age-related cognitive impairment and Alzheimer’s disease (AD). Cerebral amyloid angiopathy (CAA) is characterized by the deposition of amyloid-β (Aβ) in the leptomeningeal and cortical arteries and is highly prevalent in AD, affecting over 90% of cases. While the ε4 allele of apolipoprotein E (APOE) represents the strongest genetic risk factor for AD, it is also associated with cerebrovascular dysregulations. APOE plays a crucial role in brain lipid transport, particularly in the trafficking of cholesterol and phospholipids. Lipid metabolism is increasingly recognized as a critical factor in AD pathogenesis. However, the precise mechanism by which APOE influences cerebrovascular lipid signatures in AD brains remains unclear. In this study, we conducted non-targeted lipidomics on cerebral vessels isolated from the middle temporal cortex of 89 postmortem human AD brains, representing varying degrees of CAA and different APOE genotypes: APOE ε2/ε3 (N = 9), APOE ε2/ε4 (N = 14), APOE ε3/ε3 (N = 21), APOE ε3/ε4 (N = 23), and APOE ε4/ε4 (N = 22). Lipidomics detected 10 major lipid classes with phosphatidylcholine (PC) and phosphatidylethanolamine (PE) being the most abundant lipid species. While we observed a positive association between age and total acyl-carnitine (CAR) levels (p = 0.0008), the levels of specific CAR subclasses were influenced by the APOE ε4 allele. Notably, APOE ε4 was associated with increased PE (p = 0.049) and decreased sphingomyelin (SM) levels (p = 0.028) in the cerebrovasculature. Furthermore, cerebrovascular Aβ40 and Aβ42 levels showed associations with sphingolipid levels including SM (p = 0.0079) and ceramide (CER) (p = 0.024). Weighted correlation network analysis revealed correlations between total tau and phosphorylated tau and lipid clusters enriched for PE plasmalogen and lysoglycerophospholipids. Taken together, our results suggest that cerebrovascular lipidomic profiles offer novel insights into the pathogenic mechanisms of AD, with specific lipid alterations potentially serving as biomarkers or therapeutic targets for AD.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00401-025-02949-5.

Introduction

The brain is the second-most lipid-dense organ after adipose tissue, with lipids comprising approximately 50–60% of its dry weight, primarily in the form of phospholipids including glycerophospholipids and sphingolipids [8, 24, 47, 48]. Previous studies demonstrated lipid dysregulations in Alzheimer’s disease (AD) brain parenchyma characterized by reduced levels of phosphatidylinositols (PI), phosphatidylcholine (PC), phosphatidylethanolamines (PE), and plasmalogens compared to healthy controls [8, 26, 69]. Genome-wide association studies (GWAS) underscore the role of lipid metabolism in AD, revealing that approximately one-third of AD risk genes are implicated in lipid metabolism, transport, or binding [37]. Of these, the ε4 allele of the APOE gene is the strongest genetic risk factor for late-onset AD, while the ε2 allele is protective [71].

Apolipoprotein E (APOE) is a major component of brain-derived lipoproteins with HDL-like properties, functioning as a transporter of cholesterol and phospholipids [42]. Studies have shown that APOE4 exhibits markedly reduced efficiency in facilitating lipid efflux from cells, due to its impaired lipidation status [54, 71]. Indeed, PI, PC, and PE levels in AD brains were significantly reduced in individuals carrying the APOE ε4 allele compared to those with the APOE ε3/ε3 genotype, suggesting that disruptions in APOE-mediated lipid transport may underlie alterations in parenchymal lipid composition within AD brains [40]. Furthermore, as the cerebrovascular system plays an essential role in maintaining brain homeostasis, cerebral small vessel diseases mainly caused by arteriolosclerosis and cerebral amyloid angiopathy (CAA) substantially influence the development of cognitive impairment and AD [35]. CAA occurs in more than 90% of patients with AD [36]. CAA characterized by amyloid-β (Aβ) deposits in the leptomeningeal and cortical arteries causes cerebral lobar hemorrhages and dementia in elderly individuals [66, 70]. Despite the well-recognized alterations in lipid composition in AD brains, the cerebrovascular lipid composition, particularly in relation to APOE genotypes, remains to be fully elucidated. Given the critical role of lipids in maintaining cellular membrane integrity and signaling [64], alterations in cerebrovascular lipid composition may be causally or consequently related to vascular pathologies in AD brains. Therefore, we isolated blood vessels from postmortem human brain tissues derived from AD patients with different APOE genotypes and performed non-targeted lipidomics. Our study identified cerebrovascular lipids associated with age, APOE genotype, and AD-related vascular pathologies. These cerebrovascular-specific lipid signatures provide insights into APOE genotype-related molecular mechanisms underlying vascular contributions to AD and the development of targeted therapeutics.

Materials and methods

Human AD brain samples

Middle temporal cortex samples of pathologically confirmed AD cases were obtained from the Mayo Clinic Brain Bank for neurodegenerative diseases. At the outset of this study, the focus was to investigate the impact of APOE genotypes on lipid signatures in the cerebrovasculature. To this end, we initially enrolled 23 cases carrying the APOE ε2 allele (ε2/ε3, N = 9; ε2/ε4, N = 14). Subsequently, we enrolled age-, sex-, CAA-score matched cases from the APOE ε3/ε3 (N = 21), ε3/ε4 (N = 23), and ε4/ε4 (N = 22) genotypes. Participants were required to meet inclusion criteria: (1) age greater than 60 years, (2) Thal phase III or higher [62], and (3) Braak stage IV or higher [7]. Individuals treated with antibody-based AD immunotherapies against Aβ or tau were not included. This resulted in a total of 89 cases, all of whom were non-Hispanic and White. We also collected demographic data (age of death and sex) as well as pathological information, including average CAA scores, Braak stage, Thal phase, and arteriolosclerosis pathology grading. To represent the diverse spectrum of an autopsied AD population, cases with comorbid neuropathologies were included in this study (Supplementary Table 1). The demographic and pathological characteristics of the cohort are summarized in Table 1. No significant differences in age, sex, pathological scores, or cerebrovascular Aβ and tau levels were observed across APOE genotypes. Genomic DNA was extracted from frozen brain tissue using the standard protocols. The APOE single nucleotide variants (rs429358 C/T and rs7412 C/T), which define the APOE ε2, ε3, and ε4 alleles, were genotyped using custom TaqMan Allelic Discrimination Assays on a QuantStudio 7 Flex Real-Time PCR system (Applied Biosystems, Foster City, CA, USA). Genotype calls were generated using TaqMan Genotyper Software v1.3 (Applied Bio-Systems). Genotype call rates were 100% for each variant, and no deviations from the Hardy–Weinberg equilibrium were observed (all p > 0.01). Experimental procedures were conducted in compliance with protocols approved by the Mayo Clinic Institutional Review Board.

Table 1.

Summary of participant characteristics overall and separately according to APOE group

	Median (minimum, maximum) or No. (%) of subjects
	Total (N = 89)	APOE ε2/ε3 (N = 9)	APOE ε2/ε4 (N = 14)	APOE ε3/ε3 (N = 21)	APOE ε3/ε4 (N = 23)	APOE ε4/ε4 (N = 22)	P value
Age at death (years)	81.0 (62.0, 93.0)	81.0 (62.0, 93.0)	81.5 (66.0, 91.0)	82.0 (62.0, 92.0)	83.0 (63.0, 93.0)	80.5 (64.0, 92.0)	0.93
Female sex (%)	29 (32.6%)	3 (33.3%)	4 (28.6%)	7 (33.3%)	8 (34.8%)	7 (31.8%)	1.00
Braak stage							0.30
Ⅳ	16 (18.0%)	2 (22.2%)	3 (21.4%)	4 (19.0%)	4 (17.4%)	3 (13.6%)
Ⅴ	24 (27.0%)	2 (22.2%)	3 (21.4%)	9 (42.9%)	8 (34.8%)	2 (9.1%)
Ⅵ	49 (55.1%)	5 (55.6%)	8 (57.1%)	8 (38.1%)	11 (47.8%)	17 (77.3%)
Thal phase							0.24
3	3 (3.4%)	0 (0.0%)	1 (7.1%)	2 (9.5%)	0 (0.0%)	0 (0.0%)
4	9 (10.1%)	2 (22.2%)	0 (0.0%)	3 (14.3%)	1 (4.3%)	3 (13.6%)
5	77 (86.5%)	7 (77.8%)	13 (92.9%)	16 (76.2%)	22 (95.7%)	19 (86.4%)
Average CAA score	1.0 (0.0, 3.0)	1.1 (0.0, 2.2)	0.9 (0.0, 2.8)	1.0 (0.0, 2.2)	0.9 (0.0, 3.0)	1.0 (0.1, 3.0)	0.98
Arteriolosclerosis score	1.0 (0.0, 3.0)	0.0 (0.0, 2.0)	0.0 (0.0, 3.0)	1.0 (0.0, 3.0)	1.0 (0.0, 2.0)	1.0 (0.0, 3.0)	0.42
AD/CAA related molecules Aβ40 (fmol/mg protein)
Large vessel	3845 (109, 269,438)	13,614 (281, 176,152)	23,667 (254, 134,901)	3518 (109, 269,438)	2886 (464, 190,864)	3824 (388, 172,079)	0.89
Small vessel	3417 (48, 645,764)	7798 (288, 103,568)	17,259 (323, 154,052)	1059 (132, 645,764)	2578 (48, 260,889)	4182 (207, 138,389)	0.80
Aβ42 (fmol/mg protein)
Large vessel	557 (57, 21,001)	717 (174, 2815)	1311 (151, 2959)	1266 (62, 8763)	521 (69, 11,065)	366 (57, 21,001)	0.22
Small vessel	1394 (150, 18,821)	1770 (406, 4957)	1891 (256, 13,519)	2690 (185, 5603)	1327 (266, 8474)	963 (150, 18,821)	0.39
t-Tau (ng/mg protein)
Large vessel	75.0 (1.4, 684.0)	67.3 (17.2, 166.9)	67.3 (24.7, 270.6)	82.3 (1.4, 684.0)	46.7 (15.4, 350.9)	87.9 (25.7, 260.9)	0.56
Small vessel	89.6 (0.5, 446.6)	79.6 (0.5, 198.9)	99.1 (8.5, 440.7)	87.3 (0.5, 446.6)	80.1 (41.2, 361.3)	119.6 (15.2, 397.4)	0.73
p-Tau231 (Unit/mg protein)
Large vessel	438.6 (56.7, 6024.7)	386.7 (147.4, 956.9)	697.1 (56.7, 2388.7)	385.8 (118.3, 6024.7)	438.1 (84.2, 1758.1)	518.4 (129.5, 2127.4)	0.86
Small vessel	879.0 (48.8, 5814.8)	879.0 (122.7, 2276.2)	1313.7 (48.8, 5814.8)	878.4 (61.0, 5473.7)	623.9 (133.0, 4658.8)	971.9 (174.3, 5076.9)	0.85

P values result from a Kruskal–Wallis rank sum test (continuous and ordinal variables) or Fisher’s exact test (categorical variables)

Histopathologic assessment for cerebrovascular disease

All cases were subjected to standardized neuropathological sampling and evaluation as previously described [56]. We assessed AD neuropathologic changes, including Braak tangle stage and Thal amyloid phase, as described previously [7, 46, 62]. CAA staging of parenchymal vessels was performed using the method previously described [57]. Thioflavin-S staining was used to evaluate the severity of CAA and was scored in 5 cortical regions: superior temporal cortex, inferior parietal cortex, middle frontal cortex, motor cortex, and visual cortex. A semi-quantitative approach was employed with the following scoring system: 0, the absence of amyloid-positive vessels; 0.5, amyloid deposition limited to the leptomeninges; 1, mild amyloid deposition in both leptomeninges and parenchymal vessels; 2, moderate circumferential amyloid deposition in some vessels; 3, widespread severe amyloid deposition in leptomeninges and parenchymal vessels; 4, reflecting more severe CAA with dysphoric changes (Fig. 1a-d). The average CAA scores were defined as the mean of region-specific CAA scores from five cortical regions. Hematoxylin and eosin staining sections obtained from superior temporal cortex were assessed for arteriolosclerosis according to the vascular cognitive impairment neuropathology guidelines (VCING) [16, 58]. Severities were scored using a semi-quantitative method as follows: Arteriolosclerosis: 0, normal; 1, mild fibrosis with mild medial thickening; 2, moderate fibrosis; 3, severe fibrosis, based on overall impression (Fig. 1e-g).

Fig. 1.

Morphologies of arteriolosclerosis and Cerebral amyloid angiopathy (CAA) with different scores in AD brains. CAA score was assessed using thioflavin-S fluorescence (green). a Sparse amyloid presence in both leptomeningeal and cortical vessels was rated as score 1. b Strong, circumferential amyloid deposition in some, but not all vessels were rated as 2. c Widespread, strong amyloid deposition in both leptomeningeal and cortical vessels were rated as score 3. d The most severe cases, exceeding the criteria for score 3 were rated as 4. Arteriolosclerosis score was assessed using hematoxylin and eosin-staining. e Cases with mild arteriolosclerosis, characterized by mild fibrosis and slight medial thickening were rated as 1. f Cases with moderate arteriolosclerosis, with evident fibrosis and moderate medial thickening were rated as 2. g Cases with severe arteriolosclerosis, marked by extensive fibrosis and pronounced medial thickening were rated as 3. Scale bars: 200 µm (A, B); 100 µm (C, D); 40 µm (E); 50 µm (F, G)

Isolation of human cerebrovasculatures

We isolated vessels from middle temporal cortex samples dissected from AD patients, following a previously described protocol with minor modification [6] (Supplementary Fig. 1a). All procedures were conducted on ice. After the removal of meninges with tweezers, we thawed each temporal cortex sample (400 mg) on ice in vascular isolation buffer (10 μM HEPES (Thermo Fisher Scientific, Cat No. 15630080, Waltham, MA, USA) in HBSS (Thermo Fisher Scientific, Cat No. 14025092)). We manually minced the brains with a scalpel in the vascular isolation buffer to obtain small pieces of brain tissues and transferred them to a 7 ml tissue grinder. We homogenized the samples, transferred to 15 ml conical tubes, and centrifuged at 1000 g for 10 min at 4 °C. We removed the supernatants, suspended the pellet with 5 ml of vascular isolation buffer containing 18% dextran (from leuconostoc mesenteroides, M.W. 70,000; Sigma-Aldrich, Cat No. 31390, St. Louis, MO, USA), and centrifuged at 4000 g for 20 min at 4 °C. We then discarded the myelin layer with the supernatant, carefully wiped the inside wall of the conical tube with absorbent paper and resuspended the residual pellet with 1 ml of vascular isolation buffer. Next, we sequentially filtered the resuspended homogenates using 100 μm and 20 μm nylon filters in succession (Pluriselect, Cat No. 43–50,100-51 and 43–50,020-03, Leipzig, Germany). We defined the homogenates retained on the 100 μm filter as large vessel fractions, comprising large-sized vessels, such as leptomeninges, penetrating arteries and arterioles. Homogenates retained on the 20 μm filter were defined as small vessel fractions, primarily consisting of capillaries. We also defined the filtrates mainly containing vascular-depleted parenchymal tissues as parenchymal fractions. Each retained vascular fraction was washed off from filters with 500 μl of lysis buffer (150 mM NaCl, 10 mM NaH₂PO₄, 1% Triton X-100 (Sigma-Aldrich, Cat No. T9284), 0.5% SDS (Invitrogen, Cat No. AM9820, Carlsbad, CA, USA) and 0.5% sodium deoxycholate (Thermo Fisher Scientific, Cat No. 89904) containing phosphatase inhibitors (Roche Diagnostics, Cat No. 4906845001, Indianapolis, IN, USA) and protease inhibitors (Roche Diagnostics, Cat No. 11697498001), and 1 mM EDTA. Each fraction was placed on ice for 30 min, sonicated, and centrifuged at 100,000 g for 20 min at 4 °C in an ultracentrifuge tube (Beckman Coulter, Cat No. 357448, Indianapolis, IN, USA). The vascular-depleted parenchymal tissues were centrifuged at 16,000 g for 20 min at 4 °C, and pellets were homogenized in 500 μl of lysis buffer, sonicated and centrifuged at 100,000 g for 20 min at 4 °C. The supernatants derived from each fraction were collected and stored at −80 °C (referred to as soluble fraction). We further homogenized the pellets with 125 μl of formic acid (Sigma-Aldrich, Cat No. 33015), sonicated, and centrifuged at 16,000 g for 20 min at 4 °C. After evaporating formic acid, the pellets were resuspended in 350 μl of 5 M guanidium solution in Tris–HCl 50 mM and stored at -80 °C (referred to as the insoluble fraction). We determined protein concentrations of all fractions using the bicinchoninic acid (BCA) assay (Thermo Fisher Scientific, Cat No. 23225). We also isolated vessels for lipidomic analyses using the same set of AD brain samples as separate single filtration step. The resuspended vascular homogenates were filtered through a 20 μm nylon filter (Pluriselect, Cat No. 43–50,020-03). The retained vascular fraction was washed off from the filter with 500 μl of vascular isolation buffer, sonicated, and centrifuged at 100,000 g for 20 min at 4 °C in ultracentrifuge tubes. The supernatant was aspirated, and the pellet was stored at −80 °C for subsequent lipidomic analysis.

Preparation of homogenized whole brain tissues

Brain tissue from the middle temporal cortex (50 mg) was mechanically homogenized in lysis buffer (150 mM NaCl, 10 mM NaH₂PO₄, 1% Triton X-100, 0.5% SDS, and 0.5% sodium deoxycholate) containing phosphatase and protease inhibitors. The brain homogenates were subsequently ultracentrifuged at 100,000 × g for 1 h at 4 °C. The supernatant was then collected as whole brain fractions (W) and stored at −80 °C for western blot analysis to compare with the vascular-enriched and vascular-depleted parenchymal fractions.

Enzyme-linked immunosorbent assay for Aβ and tau

We determined Aβ40 and Aβ42 levels of vascular-enriched fractions using highly sensitive ELISA kits (Fujifilm Wako Pure Chemical Corporation, Cat No. 292–62,301 and 296–6440, Osaka, Japan) according to the manufacturer’s instructions, as previously described [63]. We also determined phospho-tau (Thr231) and total tau levels using a kit (Meso Scale Diagnostics, Cat No. K15121D, Rockville, MD, USA) according to the manufacturer’s instructions. Signal detection was performed using MESO QuickPlex SQ 120MM and data analysis was carried out with Methodical Mind software (Meso Scale Diagnostics). All measurements were normalized against total protein amounts.

Lipidomics

Lipid species were analyzed using multidimensional mass spectrometry-based shotgun lipidomic analysis [25]. In brief, each sample homogenate containing 0.5 mg of protein which was determined with a Pierce BCA assay was accurately transferred to a disposable glass culture test tube. A pre-mixture of lipid internal standards (IS) was added prior to lipid extraction for quantification of the targeted lipid species. Lipid extraction was performed using a modified Bligh and Dyer procedure [67], and each lipid extract was reconstituted in chloroform: methanol (1:1, v:v) at a volume of 400 µL/mg protein. For shotgun lipidomics, lipid extract was further diluted to a final concentration of ~ 500 fmol total lipids per µL. Mass spectrometric analysis was performed on a triple quadrupole mass spectrometer (TSQ Altis, Thermo Fisher Scientific) and a Q Exactive mass spectrometer (Thermo Scientific, San Jose, CA), both of which were equipped with an automated nanospray device (TriVersa NanoMate, Advion Bioscience Ltd., Ithaca, NY) as described [28]. Identification and quantification of lipid species were performed using an automated software program [68, 73]. Data processing (e.g., ion peak selection, baseline correction, data transfer, peak intensity comparison, and quantitation) was performed as described [73]. The results were normalized to protein content (nmol lipid/mg protein).

Weighted correlation network analysis

Weighted gene correlation network analysis (WGCNA) [38] was performed using quantile-normalized and log₂-transformed lipid abundance measurements. Based on the relationship between power and scale independence, a power of 6 was chosen to build scale-free topology using signed hybrid network. We set the minimum modules size as 5, and merged modules whose correlation coefficients were greater than 0.7 (mergeCutHeight = 0.3). Lipid modules were represented in colors, and the eigengene (ME) of each module was tested for correlation with a series of genetic and clinical traits, including age, sex (male was coded as 1 and female was coded as 0), APOE ε2 allelic numbers, APOE ε4 allelic numbers, Braak stage, Thal phase, average CAA score, arteriosclerosis score, Aβ40, Aβ42, total tau, and p-tau231 values. Modules correlated with these traits were visualized using VisANT [33] and annotated using lipid ontology (LION) enrichment analysis [45].

Assessment of acyl chain compositions in lipids

Lipid molecular species levels were converted to lipid sum composition level using LIPID MAPS, defined by chain length (total number of carbon atoms) and double bonds (DB). Abundances at the sum composition level within each lipid class were aggregated by chain length and DB number. (Supplementary material). Spearman rank correlations were then calculated between the aggregated abundances (grouped by the carbon number and DB number), APOE genotype, and cerebrovascular Aβ and tau levels. Heatmaps showed the correlations between chain length and DB number of lipid species with colors representing the correlation coefficients.

Immunostaining of isolated cerebrovasculature

The vascular-enriched pellets were resuspended in 100 μl of 1% bovine serum albumin (BSA) (Roche, Cat No. 10735078001, Mannheim, Germany) in PBS. In parallel, the vascular-depleted parenchymal fraction was resuspended in 200 μl of 1% BSA in PBS. Vascular-enriched and vascular-depleted parenchymal fractions were deposited on glass slides (10 μl/ per slide) and left at room temperature for 30 min. Samples were fixed with ice-cold methanol, then permeabilized with PBS containing 0.3% Triton X-100 (Sigma-Aldrich, Cat No. T9284) and blocked with Protein Block Serum-Free Ready-To-Use (Agilent Technologies, Cat No. X0909, Santa Clara, CA, USA). They were then incubated overnight at 4 °C with anti-alpha smooth muscle actin (α-SMA) antibody (1:100 dilution; Abcam, Cat No. ab5694, Cambridge, MA, USA), anti-platelet-derived growth factor receptor β (PDGFRβ) (1:100 dilution; Abcam, Cat No. ab32570), anti-Type IV collagen (1:500 dilution; MilliporeSigma, Cat No. AB769, Burlington, MA, USA), and anti-NeuN (1:100 dilution; Abcam, Cat No. ab177487) diluted in background-reducing dilution buffer (Agilent Technologies, Cat No. S3022), subsequently incubated with Alexa Fluor 488 conjugated donkey anti-rabbit secondary antibody and Alexa Fluor 594 conjugated donkey anti-goat secondary antibody (1:500 dilution; Invitrogen, Cat No. A21206 and A11058) for 2 h at room temperature. The samples were mounted with DAPI hard set mounting medium (Vector Laboratories, Cat No. H1500, Burlingame, CA, USA). The images were captured using a Keyence BZ-X800 fluorescence microscope (Keyence, Osaka, Japan).

Western blotting

The soluble fractions derived from vascular-enriched, vascular-depleted parenchymal fractions, and whole brain fractions were mixed with Laemmli’s loading sample buffer and heated 5 min at 95 °C. Equal amounts of proteins per sample (10 μg) were electrophoresed on 4–20% Tris–glycine sodium dodecyl sulfate–polyacrylamide gels (Bio-Rad, Cat No. 5671094, Hercules, CA, USA). Proteins were electroblotted on polyvinylidene fluoride (PVDF) membranes (MilliporeSigma, Cat No. IPVH00010), which were then blocked for 2 h at room temperature with a PBS containing 5% non-fat dry milk. The membranes were incubated overnight at 4 °C with the primary antibodies: anti-occludin (1:1000 dilution; Invitrogen, Cat No. 71–1500), anti-α-SMA antibody (1:1000 dilution; Abcam, Cat No. ab5694), anti-PDGFRβ (1:2000 dilution; Abcam, Cat No. ab32570), anti-NeuN (1:1000 dilution; Abcam, Cat No. ab177487). The membrane was then probed with horseradish peroxidase (HRP)-conjugated secondary antibodies (Abcam, Cat No. ab6721 and ab6789). The blots were detected using Pierce™ Western Blot Signal Enhancer (Thermo Fisher Scientific, Cat No. 21050) and the ChemiDoc Touch Imaging System (Bio-Rad).

Generation of human iPSC-derived vascular mural-like cells

We used human induced pluripotent stem cell (iPSC) lines from the male-derived KOLF2.1 J line obtained from the iPSC Neurodegenerative Disease Initiative (iNDI), consisting of isogenic iPSC lines with APOE ε3/ε3 or ε4/ε4 [52]. The characterization of the parental KOLF2.1 J line has been reported previously [52]. Human iPSC-derived vascular mural-like cells (iVMLCs) were differentiated as previously described with modifications [19]. Briefly, iPSCs were dissociated into single cells using Accutase (Stem Cell Technologies, Cat No. 07920, Vancouver, Canada) and reseeded at 25,000 cells /cm² on Matrigel-coated plates in mTeSR1 medium supplemented with 10 μM Y27632 (Stem Cell Technologies, Cat No. 72302) for the first 24 h. To initiate the differentiation, medium was changed to STEMdiff™ Mesoderm Induction Medium (Stem Cell Technologies, Cat No. 05221). The medium was exchanged every 24 h. On day 6, cells were dissociated using Accutase and re-seeded at 35,000 cells/ cm on Matrigel-coated wells in pericyte medium (ScienCell Research Laboratories, Cat No. 1201, Carlsbad, CA, USA) and cultured for additional 8 days. All cultured cells were maintained at 37 °C in a humidified incubator with 5% CO₂.

Immunostaining of iPSC-derived vascular mural-like cells

For immunostaining, human iVMLCs were fixed with 4% paraformaldehyde for 15 min and subsequently washed three times with PBS. Following fixation, cells were permeabilized using PBS containing 0.3% Triton X-100 (Sigma-Aldrich, Cat No. T9284) and blocked with Protein Block Serum-Free Ready-To-Use solution (Agilent Technologies, Cat No. X0909). Primary antibody incubation was performed overnight at 4 °C in a background-reducing dilution buffer (Agilent Technologies, Cat No. S3022) using mouse monoclonal anti-αSMA antibodies (Sigma-Aldrich, Cat No. A2547). After washing, cells were incubated with Alexa Fluor 488 conjugated donkey anti-mouse secondary antibody (Invitrogen, Cat No. A21202) for 2 h at room temperature. Nuclei were visualized by counterstaining with DAPI (Thermo Fisher Scientific, Cat. 62,248) at a 1:5000 dilution in PBS. Fluorescence images were acquired using a Keyence BZ-X800 microscope (Keyence, Osaka, Japan).

RNA isolation and real-time PCR analysis

For comparative gene expression studies, we extracted total RNA from human iVMLCs, human primary brain vascular smooth muscle cells (HBSMC; ScienCell Research Laboratories, Cat. No. 1100), and immortalized human cerebral microvascular endothelial cells, clone D3 (hCMEC/D3; MilliporeSigma, Cat. No. SCC066) using the RNeasy Mini Kit (QIAGEN, Cat No. 74104, Valencia, CA, USA) according to the manufacturer’s instructions. Reverse transcription was performed with the iScript™ cDNA Synthesis Kit (Bio-Rad, Cat No. 1708890) to synthesize complementary DNA (cDNA) from total RNA. For quantitative real-time PCR analysis, cDNA samples were combined with gene-specific primers and SsoAdvanced™ Universal SYBR® Green Supermix (Bio-Rad, Cat No. 1725271). Amplification and detection were performed using the QuantStudio 7 Flex Real-Time PCR System (Applied Biosystems, Foster City, CA, USA). Data were processed using the ΔΔCT method with QuantStudio 7 Flex system software. Gene expression was assessed using predesigned primer sets (Integrated DNA Technologies, Coralville, CA, USA) for ACTA2 (Hs.PT.56a.2542642), PDGFRB (Hs.PT.58.22892761), OCLN (Hs.PT.58.15235048), PECAM1 (Hs.PT.58.19487865), and ACTB (Hs.PT.39a.22214847). The relative gene expression was normalized to ACTB.

Cell culture and Aβ treatment

Lyophilized Aβ40 and Aβ42 peptides (Anaspec, Cat No. AS-24235 and AS-20276, Fremont, CA, USA) were dissolved in dimethyl sulfoxide (DMSO) (BioWorld, Cat No. 40470005, Dublin, OH, USA) and aliquoted for storage at −80 °C as stock solutions. On day 8, medium was replaced with non-FBS containing pericyte medium, and then human iVMLCs were treated with Aβ40 or Aβ42 for 48-h incubation. DMSO diluted in PBS (0.2% DMSO, corresponding to the amount contained in Aβ) was used for vehicle control. After treatment, cells were rinsed with PBS and were subsequently lysed using Mammalian Cell Lysis Buffer 5X (Abcam, Cat No. ab179835) supplemented with phosphatase inhibitors and protease inhibitors. The lysates were centrifuged at 1500 rpm for 5 min at 4 °C. The resulting supernatants were collected and stored at −80 °C until further analysis.

Assay of sphingomyelinase activity and quantification of sphingolipids

Neutral sphingomyelinase (nSMase) activity was measured using the Sphingomyelinase Assay Kit (Abcam, Cat No. ab138876). Quantitative analyses of sphingomyelin and ceramide were performed with the Sphingomyelin Assay Kit (Abcam, Cat No. ab133118) and Human Ceramide Elisa Kit (AFG Scientific, Cat No. EK710698, IL, USA), following the manufacturers’ instructions.

MTT assay

Cell viability was assessed using the 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyl-2H-tetrazolium bromide (MTT) assay (Roche Diagnostics, Cat No. 11465007001) following the manufacturer’s instructions. After 48 h of Aβ treatment, 10 μL of MTT reagent were added to each well and incubated for 4 h at 37 °C. Then, 100 μL of solubilization buffer was added, and the plates were incubated overnight at 37 °C. Absorbance was then measured at 550 and 600 nm using a microplate reader, and the difference between the two wavelength values was used to determine cell viability.

BrdU assay

Cell proliferation was evaluated using 5-bromo-2′-deoxy-uridine (BrdU) colorimetric assay (Abcam, Cat No. ab126556) following the manufacturer’s instructions. After 48 h of Aβ treatment, BrdU was added for 18 h, followed by incubation with anti-BrdU antibody and peroxidase-conjugated secondary antibody. The reaction was visualized using TMB substrate, stopped with stop solution, and absorbance was measured at 450 nm using a microplate reader.

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 characteristics according to APOE group were conducted using a Kruskal–Wallis rank sum test (continuous and ordinal variables) or Fisher’s exact test (categorical variables). Due to distributional skewness, in the subsequently described regression analyses, a base-2 logarithm transformation was utilized for AD-related molecules including cerebrovascular Aβ40, Aβ42, t-tau, and p-tau231. A square root transformation was utilized for average CAA score, and a rank transformation was used for lipid analytes and lipid class sums. Associations of average CAA score, arteriolosclerosis score, Thal phase, Braak stage, age, sex, presence of the APOE ε2 allele, number of APOE ε4 alleles, and AD-related molecules with lipid analytes and lipid class sums were examined using unadjusted and multivariable linear regression models. Regression coefficients (denoted as β) and 95% CIs were estimated and were interpreted as the change in the mean lipid analyte or lipid class sum rank corresponding to a specified increase (continuous variables) or presence of the given characteristic (categorical variables). Multivariable models were adjusted for age, sex, number of APOE ε4 alleles (except in analysis involving presence of the APOE ε2 allele), Braak stage, and Thal phase. We utilized a Bonferroni correction for multiple testing separately for analysis assessing associations with lipid class sums (10 tests for each characteristic, p < 0.005 considered as significant) and for analysis examining associations with lipid analytes (122 tests for each characteristic, p < 0.00041 considered as significant). All statistical tests were two-sided. Statistical analyses were performed using SAS (version 9.4; SAS Institute, Inc., Cary, North Carolina). Results through iVMLCs were analyzed by one-way ANOVA followed by Tukey’s post hoc test using GraphPad Prism (Version 10; GraphPad Software Inc., San Diego, CA, USA) and p < 0.05 was considered as significant. Lipid acyl chain compositions were analyzed in R (v4.4.0). Spearman rank correlation analyses were performed by the rstatix package (v0.7.2), and heatmaps were visualized using the ComplexHeatmap package (v2.20.0) [23].

Results

Isolation of vascular-enriched fraction from frozen human brain tissue

Microvessels were isolated from frozen human brain tissues (Supplementary Fig. 1a), and isolation efficiency was validated by assessing vascular and neuronal markers using western blotting and immunofluorescent staining. The endothelial marker occludin was enriched in the large and small vessel fractions (Supplementary Fig. 1b). Vascular mural cell markers, PDGFRβ and α-SMA were also detected in the large and small vessel fractions. In contrast, the neuronal marker NeuN was predominantly detected in the vascular-depleted parenchymal fraction, indicating successful separation of vascular and neuronal components. Immunofluorescent staining also revealed that the large vessel fraction contains abundant type IV collagen-positive vessels co-localized with α-SMA (Supplementary Fig. 1c). The small vessel fraction was enriched for PDGFRβ-positive vessels (Supplementary Fig. 1d). These findings indicate that the isolated vascular fractions retained characteristics of vascular mural cell markers, whereas the vascular-depleted parenchymal fraction contained numerous NeuN-positive neurons and relatively few type IV collagen-positive vessels (Supplementary Fig. 1e).

Lipid alterations in isolated cerebral vessels from AD brains

We performed lipidomics in the isolated cerebral vessels from AD brains using mass spectrometry-based shotgun platform and identified 10 major lipid classes: phosphatidylcholine (PC), phosphatidylethanolamine (PE), phosphatidylinositol (PI), phosphatidylserine (PS), Phosphatidylglycerol (PG)/ Bis (monoacylglycerol) phosphate (BMP), lyso-phosphatidylcholine (LPC), lyso-phosphatidylethanolamine (LPE), sphingomyelin (SM), ceramide (CER), and acyl-carnitine (CAR), encompassing a total of 122 lipid analytes. The mean summed lipid abundances across lipid classes demonstrated that PC was the most abundant lipid subclass, comprising 39.4% of the total lipid content, followed by PE (36.2%), PS (9.6%), SM (6.4%), PI (3.7%), LPE (2.2%), CER (1.5%), LPC (0.6%), PG/ BMP (0.5%), and CAR (0.02%). We found that each APOE genotype group exhibited similar trends in the order of mean summed lipid abundances (Fig. 2a).

Fig. 2.

Associations of APOE and demographic/pathological variables with lipid class sums. a Pie chart showing the relative abundances of each lipid class sum. Pie charts showing the relative abundances of each lipid class sum across APOE genotypes. b Associations of demographic factors, APOE genotype, and vascular pathology with alterations in lipid class sums. The regression coefficient (β) and the corresponding p-value shown in parentheses were derived from multivariable linear regression models. p-values < 0.05 are displayed. CAA cerebral amyloid angiopathy; ARS arteriolosclerosis; LV large vessels; SV small vessels

Multivariable linear regression analyses revealed associations between the total sum of each lipid class and age, sex, APOE, pathological scores, and cerebrovascular Aβ and tau levels after adjusting for potential confounders (Fig. 2b). Significant (p < 0.005 considered as significant in this part of the analysis) positive correlation was observed between age and the total sum of CAR levels (β = 15.33, p = 0.0008), while male sex was negatively correlated with the total sum of LPE levels (β = -16.02, p = 0.012) (Supplementary Table 2). Allelic number of APOE ε4 was negatively correlated with the total sum of SM levels (β = -8.23, p = 0.028) but positively correlated with the total sum of PE levels (β = 7.15, p = 0.049). APOE ε2 allele did not show evident effects on the levels of major lipid classes (Supplementary Table 3). Pathological assessment demonstrated that the average CAA score was positively correlated with the total sum of PI levels (β = 7.27, p = 0.020), while Braak stage was negatively correlated with the total sum of PC levels (β = -10.21, p = 0.012). There was no significant correlation between major lipid class sum and arteriolosclerosis (ARS) score or Thal phase (Supplementary Table 4). Cerebrovascular Aβ40 levels were negatively correlated with the total sum of SM (β = -7.68, p = 0.0079 for large vessels and β = -5.72, p = 0.046 for small vessels). In contrast, cerebrovascular Aβ42 levels exhibited positive correlation with the total sum of CER (β = 7.10, p = 0.024 for large vessels and β = 6.64, p = 0.036 for small vessels) and PC (β = 7.67, p = 0.013 for small vessels), while showing negative correlation with the total sum of LPE (β = -5.98, p = 0.048 for small vessels). No significant correlations were observed between cerebrovascular t-tau or p-tau231 levels and lipid class sums (Supplementary Table 5 and 6).

Subsequently, we focused on alterations of cerebrovascular lipid subclasses. After multivariable linear regression analyses adjusting for potential confounders and with p < 0.00041 considered as significant, we found that age was positively correlated with the levels of total sum of CAR subspecies, CAR(20:1), CAR(18:0), CAR(16:0), and CAR(18:1) levels (Table 2). The male sex was negatively correlated with the levels of total sum of LPE subspecies, LPE(P16:0), and LPE (20:4) levels (Table 2). APOE ε2 allele was negatively correlated with PE(D18:1–22:6) and PE(D18:2–22:6), whereas APOE ε4 alleles were positively correlated with CAR(20:0) but negatively correlated with CAR(18:1) and CAR(16:0) levels (Table 3). In addition, there were negative associations of Aβ40 levels in large vessels with the levels of total sum of SM subspecies, SM(N16:0), SM(N24:1), SM(N24:2), SM(N24:0), SM(N22:1), and SM(N16:1). We also found that Aβ40 levels in both large and small vessels were positively correlated with PG/BMP(18:1–18:1) levels. CER(N24:1) exhibited positive correlations with Aβ42 levels in large and small vessels (Table 4).

Table 2.

Associations of age, sex, and lipid analytes

	Unadjusted analysis		Adjusting for sex, number of APOE ε4 alleles, Braak stage, and Thal phase
Lipid analytes	β (95% CI)	P value	β (95% CI)	P value
Age
CAR(20:1)	11.91 (4.46, 19.35)	0.0021	14.10 (5.32, 22.88)	0.0020
CAR(18:0)	10.15 (2.74, 17.55)	0.0078	13.95 (5.24, 22.67)	0.0020
CAR(16:0)	9.49 (1.89, 17.09)	0.015	12.58 (3.79, 21.37)	0.0056
PE(D18:0–20:4/D16:0–22:4)	−9.41 (−16.99, −.83)	0.016	−12.04 (−20.90, −3.18)	0.0083
PC(D16:0–16:0)	−8.01 (−15.60, −0.42)	0.039	−11.33 (−20.32, −2.34)	0.014
PC(D18:2–18:2/D16:0–20:4)	−3.52 (−11.31, 4.27)	0.37	−10.71 (−19.48, −1.95)	0.017
PE(P18:0–22:4/P20:0–20:4/P18:1–22:3)	−11.01 (−18.52, −3.49)	0.0046	−10.22 (−19.22, −1.22)	0.027
PG/BMP(16:0–18:1)	−8.70 (−16.35, −1.04)	0.027	−10.27 (−19.40, −1.14)	0.028
CAR(18:1)	6.63 (−1.11, 14.37)	0.092	9.73 (0.90, 18.55)	0.031
PI(16:0–20:4)	−9.27 (−16.90, −1.65)	0.018	−9.22 (−18.18, −0.27)	0.044
LPC(18:1)	11.10 (3.58, 18.61)	0.0043	9.06 (0.19, 17.93)	0.045
SM(N18:0)	−9.58 (−17.16, −2.01)	0.014	−8.86 (−17.64, −0.08)	0.048

	Unadjusted analysis		Adjusting for age, number of APOE ε4 alleles, Braak stage, and Thal phase
	β (95% CI)	P value	β (95% CI)	P value
Sex
PC(A16:0–16:0)	−21.61 (−32.34, −10.88)	0.0001	−19.89 (−31.90, −7.87)	0.0015
LPE(P16:0)	−19.62 (−30.52, −8.72)	0.0006	−18.89 (−31.17, −6.61)	0.0030
PC(D16:0–18:2)	−20.08 (−30.92, −9.24)	0.0004	−17.67 (−29.72, −5.61)	0.0046
PE(P16:0–20:4)	−14.12 (−25.40, −2.84)	0.015	−14.49 (−27.00, −1.98)	0.024
PG/BMP(18:1–18:2)	−13.53 (−24.41, −2.65)	0.015	−13.93 (−26.21, −1.65)	0.027
PI(16:0–18:2)	−11.94 (−23.30, −0.58)	0.040	−14.26 (−26.92, −1.60)	0.028
PI(18:0–18:1)	12.97 (1.64, 24.29)	0.025	13.77 (1.08, 26.46)	0.034
PC(P16:0–18:1/P18:1–16:0)	−11.71 (−23.12, −0.31)	0.044	−13.96 (−26.86, −1.05)	0.034
LPE(20:4)	−11.84 (−23.24, −0.44)	0.042	−13.43 (−26.16, −0.70)	0.039
PE(P20:1–18:1/P18:1–20:1)	−13.94 (−25.23, −2.65)	0.016	−12.86 (−25.25, −0.47)	0.042

β  regression coefficient; CI  confidence interval. β values, 95% CIs, and p values result from linear regression models. β values are interpreted as the change in mean lipid analytes corresponding to each 10-year increase in age, or for males compared to females. P values < 0.00041 are considered as statistically significant after applying a Bonferroni correction for multiple testing and are given in bold. P values < 0.05 in adjusted analysis are displayed

Table 3.

Associations of APOE genotype and lipid analytes

	Unadjusted analysis		Adjusting for age sex, Braak stage, and Thal phase
Lipid analytes	β (95% CI)	P value	β (95% CI)	P value
Presence of APOE ε2 allele
PE(D18:1–22:6)	−17.30 (−29.24, −5.35)	0.0050	−17.28 (−29.32, −5.24)	0.0054
PE(D18:2–22:6)	−14.75 (−26.85, −2.64)	0.018	−14.80 (−26.82, −2.79)	0.016
PE(P18:0–20:3)	−14.66 (−26.76, −2.56)	0.018	−14.58 (−26.58, −2.57)	0.018
PE(D18:0–18:2/D18:1–18:1/D16:0–20:2)	−14.10 (−26.23, −1.97)	0.023	−13.99 (−26.38, −1.61)	0.027
LPC(18:0)	13.51 (1.35, 25.68)	0.030	13.66 (1.28, 26.03)	0.031
LPE(22:4)	−12.87 (−25.07, −0.67)	0.039	−12.81 (−25.00, −0.61)	0.040
Number of APOE ε4 alleles
CAR(20:0)	8.42 (1.46, 15.39)	0.018	9.38 (2.35, 16.42)	0.0095
CAR(18:1)	−7.27 (−14.30, −0.25)	0.043	−8.69 (−15.74, −1.63)	0.016
PE(A20:0–20:4/P18:0–22:3)	9.05 (2.11, 15.99)	0.011	8.79 (1.58, 16.01)	0.018
PE(D18:0–20:3/D18:1–20:2/D16:0–22:3)	8.49 (1.51, 15.47)	0.018	8.82 (1.54, 16.10)	0.018
PS(18:0–20:4)	−8.45 (−15.43, −1.47)	0.018	−8.64 (−15.95, −1.34)	0.021
PS(20:0–20:4/18:0–22:4)	−8.94 (−15.89, −1.99)	0.012	−8.38 (−15.47, −1.29)	0.021
PE(P20:1–18:1/P18:1–20:1)	7.58 (0.55, 14.60)	0.035	8.30 (1.26, 15.34)	0.021
SM(N18:1)	−7.33 (−14.37, −0.29)	0.041	−7.96 (−15.24, −0.67)	0.033
SM(N18:0)	−6.48 (−13.53, 0.56)	0.071	−7.64 (−14.66, −0.62)	0.033
CAR(16:0)	−6.35 (−13.42, 0.72)	0.078	−7.45 (−14.47, −0.42)	0.038

β  regression coefficient; CI  confidence interval. β values, 95% CIs, and p values result from linear regression models. β values are interpreted as the change in mean lipid analytes corresponding to presence of the APOE ε2 allele, or to each additional APOE ε4 allele. P values < 0.00041 are considered as statistically significant after applying a Bonferroni correction for multiple testing. P values < 0.05 in adjusted analysis are displayed

Table 4.

Associations of AD-related molecules and lipid analytes

	Unadjusted analysis		Adjusting for age, sex, number of APOE ε4 alleles, Braak stage, and Thal phase
Lipid analytes	β (95% CI)	P value	β (95% CI)	P value
Large vessels
Aβ40
PG/BMP(18:1–18:1)	2.45 (0.70, 4.20)	0.0067	2.76 (0.92, 4.60)	0.0037
PE(P16:0–22:6/D18:0–18:0/P18:2–20:4)	2.70 (0.96, 4.44)	0.0027	2.57 (0.74, 4.41)	0.0065
SM(N16:0)	−2.73 (−4.46, −0.99)	0.0024	−2.55 (−4.42, −0.68)	0.0081
SM(N24:1)	−1.97 (−3.75, −0.19)	0.030	−2.25 (−4.15, −0.36)	0.021
Aβ42
CER(N24:1)	4.60 (1.50, 7.70)	0.0041	4.41 (0.83, 8.00)	0.016
PC(D16:1–16:0/D14:1–18:0)	3.25 (0.08, 6.43)	0.045	4.27 (0.62, 7.91)	0.022
SM(N24:2)	−3.44 (−6.61, −0.27)	0.033	−4.17 (−7.84, −0.51)	0.026
PC(D16:0–18:1)	1.19 (0.05, 2.33)	0.040	1.45 (0.15, 2.74)	0.029
Small vessels
Aβ40
PE(D20:6–22:6)	1.93 (0.34, 3.53)	0.018	1.93 (0.23, 3.63)	0.026
PG/BMP(18:1–18:1)	1.56 (−0.05, 3.17)	0.058	1.76 (0.09, 3.43)	0.039
PE(P18:0–22:2)	1.38 (−0.23, 3.00)	0.093	1.72 (0.03, 3.40)	0.046
Aβ42
PC(D16:1–16:0/D14:1–18:0)	5.46 (2.11, 8.82)	0.0017	7.24 (3.44, 11.05)	0.0003
PI(18:0–22:6)	6.26 (2.96, 9.55)	0.0003	6.13 (2.26, 10.00)	0.0023
PC(D16:0–18:1)	1.82 (0.61, 3.04)	0.0037	2.15 (0.77, 3.53)	0.0027
SM(N24:2)	−4.78 (−8.19, −1.38)	0.0065	−5.47 (−9.43, −1.50)	0.0075
CAR(18:0)	−4.23 (−7.60, −0.87)	0.014	−5.14 (−8.87, −1.41)	0.0075
Large vessels
t-Tau
LPE(P16:0)	4.79 (0.65, 8.93)	0.024	4.97 (0.83, 9.10)	0.019
PC(D16:0–16:0)	5.00 (0.93, 9.08)	0.017	5.06 (0.79, 9.33)	0.021
p-Tau231
LPE(22:6)	−5.25 (−9.33, −1.17)	0.012	−5.53 (−10.35, −0.71)	0.025
PC(D16:0–18:1)	1.34 (-0.15, 2.84)	0.077	1.91 (0.20, 3.63)	0.029
PE(P18:0–20:4/P16:0–22:4/P18:1–20:3)	3.91 (−0.24, 8.06)	0.065	5.30 (0.36, 10.24)	0.036
PE(P18:1–20:4/P16:0–22:5)	3.85 (-0.31, 8.00)	0.069	5.23 (0.29, 10.18)	0.038
PE(D18:0–18:1/D16:0–20:1)	−4.32 (−8.45, −0.19)	0.041	−5.06 (−9.93, −0.20)	0.042
Small vessels
t-Tau
PE(P20:0–18:0/P18:0–20:0)	−2.77 (−6.10, 0.56)	0.10	−5.64 (−9.29, −1.99)	0.0029
PE(D18:2–22:6)	−2.38 (−5.73, 0.97)	0.16	−4.95 (−8.62, −1.29)	0.0087
PE(D16:1–22:6)	−2.34 (−5.68, 1.01)	0.17	−4.39 (−8.11, −0.67)	0.021
PC(D18:0–22:4/D20:0–20:4/D20:2–20:2)	2.91 (−0.42, 6.23)	0.086	4.32 (0.58, 8.07)	0.024
PG/BMP(16:0–18:1)	4.34 (1.09, 7.60)	0.0095	3.92 (0.18, 7.65)	0.040
p-Tau231
PC(D16:1–16:0/D14:1–18:0)	3.10 (−0.35, 6.55)	0.077	5.76 (1.85, 9.68)	0.0044
PC(D16:0–18:1)	1.30 (0.06, 2.53)	0.040	1.78 (0.37, 3.18)	0.014
PI(16:0–20:4)	4.19 (0.79, 7.59)	0.016	4.53 (0.65, 8.40)	0.022
PC(D16:0–16:0)	3.48 (0.09, 6.87)	0.044	4.20 (0.30, 8.10)	0.035

β  regression coefficient; CI  confidence interval. β values, 95% CIs, and p values result from linear regression models. β values are interpreted as the change in mean lipid analytes corresponding to each doubling of Aβ40, Aβ42, total-tau level, or phospho-tau231 levels. P values < 0.00041 are considered as statistically significant after applying a Bonferroni correction for multiple testing and are given in bold. Items that meet both p values < 0.05 in adjusted analysis and top five-ranked lipid analytes are shown

WGCNA of cerebrovascular lipid analytes in AD brains

To characterize the alterations in cerebrovascular lipid composition in AD brains, we performed WGCNA on the lipidomics data and identified six distinct lipid modules based on co-expression patterns (Fig. 3a). Module brown was positively correlated with age and negatively correlated with average CAA score, which was enriched for SM subspecies (Fig. 3b). The Module blue exhibited a positive correlation with the allelic number of APOE ε4 (Fig. 3c). In contrast, the module yellow exhibited a negative correlation with the presence of the APOE ε2 allele (Fig. 3d). The top-ranked pathways in the module yellow included “fatty acids with 22–24 carbon atoms” and “fatty acids with more than three double bonds,” suggesting its distinct enrichment in highly unsaturated long-chain lipids. PE(D16:0–22:6), PE(D20:6–22:6), and PE(D18:1–22:6) were identified as the top-ranked lipids in the module. Module red was positively correlated with t-tau and p-tau231 levels in large vessels, enriched for monounsaturated PE subspecies, with PE(P18:0–20:4/P16:0–22:4/P18:1–20:3) identified as the top-ranked lipid. In the module red, “plasmalogen” and “negative intrinsic curvature” were top‐ranked pathways (Fig. 3e). Module green was negatively correlated with male sex, t-tau, and p-tau231 levels in small vessels, which was enriched for LPE and LPC subclasses (Fig. 3f). Module turquoise composed of PI and PC subclasses was positively correlated with Aβ42 levels in small vessels (Fig. 3g).

Fig. 3.

Lipid modules linked to APOE and demographic/pathological variables. a Module-trait relationships between module eigengenes (MEs) and sample characteristics are shown. The correlation coefficient (r) and the corresponding p-value in the parentheses are indicated in each module. p-values < 0.05 are displayed. WGCNA results from the lipidomics dataset, highlighting modules associated with APOE genotype and demographic factors (b-g). Top‐ranked hub lipids and GO terms through LION with the highest intramodular connectivity in module brown (b), blue (c), yellow (d), red (e), green and (f), turquoise (g) are shown. CAA cerebral amyloid angiopathy; ARS arteriolosclerosis; LV large vessels; SV small vessels; WGCNA weighted correlation network analysis; GO gene ontology; LION lipid ontology

Associations of APOE, Aβ, and tau levels with lipid acyl chain compositions in cerebral vessels

We mapped the acyl chain composition of each lipid subspecies, focusing on numbers of chain length and double bonds to visualize the relationship between cerebrovascular lipid properties and APOE genotype, Aβ, and tau levels. The presence of APOE ε2 allele tended to correlate negatively across PE subspecies regardless of acyl chain compositions. Increasing number of APOE ε4 alleles was associated with a compositional shift toward fewer carbon chain numbers and double bonds, although these associations did not reach statistical significance. We also found that APOE ε4 allelic number was negatively associated with SM subspecies, particularly those with two or three double bonds (Fig. 4). Most PE subspecies with different acyl chain compositions exhibited significant positive correlations with Aβ40 in large and small vessels, while various PC subspecies positively correlated with Aβ40 in large vessels and Aβ42 in small vessels. Positive associations of some PS and PI subspecies with Aβ42 in small vessels were also detected. (Supplementary Figs. 2 and 3). In addition, t-tau in small vessels were positively correlated with PE plasmalogen (PE-P) subspecies, with significant correlations observed in specific structural PE-P subsets characterized by longer chain length and greater number of double bonds (Supplementary Fig. 4 and 5).

Fig. 4.

Associations of APOE with lipid acyl chain composition. Two-dimensional lipid class plot showing the associations of APOE2 or APOE4 allelic numbers with lipid acyl chain composition. The x-axis represents the total number of carbon atoms in acyl chain (Total C), and the y-axis represents the total number of double bonds (Total DB). Correlation strength and direction are depicted by the color scale within each grid cell. *p < 0.05. a PE Phosphatidylethanolamine, b PE-P Phosphatidylethanolamine-plasmalogens, c PC Phosphatidylcholine, d PS Phosphatidylserine, e SM Sphingomyelin, f LPE Lysophosphatidylethanolamine, g PI Phosphatidylinositol, h PG Phosphatidylglycerol, BMP bis(monoacylglycero)phosphate, i CER Ceramide, j LPC Lysophosphatidylcholine, k CAR Acyl-carnitine

Impact of Aβ on sphingolipids metabolism in human iPSC-derived vascular mural-like cells

Since our results revealed the associations of APOE ε4 and Aβ with sphingolipids in cerebral vessels, we examined the phenotypes of vascular mural-like cells differentiated from isogenic human iPSCs with APOE ε3/ε3 (APOE3) or ε4/ε4 (APOE4) (Fig. 5a). We subsequently compared gene expression profiles across vascular cells. We confirmed that iVMLCs and primary human brain vascular smooth muscle cells (HBSMCs) share a similar gene expression signature, characterized by robust expression of ACTA2 and PDGFRB. While endothelial-specific genes, including OCLN and PECAM1, were predominantly expressed in hCMEC/D3 endothelial cell lines, they were expressed at low levels in the iVMLCs (Fig. 5b). Next, the iVMLCs were cultured with Aβ40 (10 μM) or Aβ42 (5 μM) for 48 h, followed by the assessments of cell viability and proliferation using MTT and BrdU incorporation assays, respectively. Aβ42 exposure led to a more pronounced reduction in cell viability and proliferative capacity in APOE4 iVMLCs compared to APOE3 counterparts. Aβ40 reduced cell viability only in APOE4 iVMLCs (Fig. 5c, d). In addition, APOE4 iVMLC displayed higher nSMase activity after Aβ42 administration (Fig. 5e). Consistently, we found that Aβ42 decreased SM levels (Fig. 5f) and increased CER levels (Fig. 5g) in APOE4 iVMLCs, but not APOE3 iVMLCs. Aβ40 also significantly reduced SM levels in APOE4 iVMLCs. These results suggest that CAA disturbs sphingolipid metabolism predominantly in vascular mural cells in the presence of APOE ε4.

Fig. 5.

Aβ modulates sphingolipid metabolism in human iPSC-derived vascular mural-like cells. a Representative immunostaining images of α-SMA in isogenic human iPSC-derived vascular mural-like cells (iVMLCs) with APOE ε3/ε3 (APOE3) or ε4/ε4 (APOE4). b The mRNA expressions of vascular mural cell markers (ACTA2 and PDGFRβ) and endothelial cell markers (PECAM1 and OCLN) were quantified by RT-qPCR in iVMLCs (ε3/ε3 or ε4/ε4), human brain vascular smooth muscle cells (HBSMC, ε3/ε4), and human cerebral microvascular endothelial cell line (hCMEC/D3, ε3/ε3), followed by normalization to ACTB mRNA expression. b, d The iVMLCs were treated with Aβ40 (10 µM), Aβ42 (5 µM), or vehicle (0.25% DMSO) for 48 h. Cell viability and proliferation were assessed using MTT (c) and BrdU (d) assays, respectively. e The cell lysates were analyzed for neutral sphingomyelinase (nSMase) activity. f, g Amounts of sphingomyelin (f; SM) and ceramide (g; CER) in the iVMLCs were measured by ELISA and normalized to protein concentrations. Data are presented as mean ± SEM (n = 3–4 technical replicate/each). Scale bars = 100 µm. *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001 by one-way ANOVA followed by Tukey’s post hoc test

Discussion

In this study, we conducted non-targeted lipidomics of isolated cerebral vessels derived from 89 human AD brains and identified 10 major lipid classes, among which PC was the most abundant, followed by PE, PS, and SM. To the best of our knowledge, this is the first study to investigate cerebrovascular lipid composition in AD cases. A lipidomics study on the entorhinal cortex from human AD cases has revealed that PC is the most abundant lipid subclass, followed by PE, plasmalogen PE, and SM [11]. Another report on the frontal cortex demonstrated a similar lipid composition, with PC being the most abundant, followed by PE, PS, and SM [60]. Interestingly, we observed a positive correlation between age and levels of total CAR subspecies. CAR facilitates the transport of long-chain fatty acids into mitochondria for β-oxidation, thereby generating ATP as a primary energy source for the brain [9, 14]. Elevated CAR levels may indicate mitochondrial deficiency in fatty acid oxidation and reflect an adaptive compensatory mechanism to support energy metabolism during aging [55]. Indeed, cerebrovascular endothelial cells, especially those in capillaries, demonstrate a greater dependence on mitochondrial respiration than peripheral endothelial cells and muscles [49, 50]. In addition, CAR levels in the brain parenchyma and cerebrovasculature have been shown to increase with age in APOE2- and APOE3-targeted replacement (TR) mice but not in APOE4-TR mice [34]. Notably, we observed that CAR(16:0) and CAR(18:1) levels were negatively correlated with APOE ε4 allele. The CAR(16:0) and CAR(18:1) are synthesized from fatty acid by carnitine palmitoyltransferase 1 (CPT1), located in the outer mitochondrial membrane, and subsequently transported into the mitochondrial matrix [2, 44]. Thus, the reductions in CAR(16:0) and CAR(18:1) levels may be attributable to the impairment of CPT1 function, indicating impairment of mitochondrial fatty acid oxidation in APOE ε4 carriers, which either contributes to or results from cerebrovascular dysfunction [1, 18, 20, 53].

Furthermore, we found that APOE ε4 allelic number was negatively correlated with total levels of SM subspecies in cerebrovasculature. These observations align with prior lipidomics studies of human plasma and brain parenchyma showing that APOE ε4 carriers exhibit decreased SM levels compared to non-APOE ε4 carriers in AD [4, 27]. Given the role of SM in maintaining membrane structural integrity and function, such reduction may compromise cerebrovascular function [32, 59]. Indeed, our WGCNA of lipidomics demonstrated a negative correlation between average CAA scores and SM, APOE ε4 may modulate sphingolipid metabolism within CAA-affected vessels in AD. We also observed a negative correlation between cerebrovascular Aβ40 levels and the total sum of SM. In contrast, cerebrovascular Aβ42 levels showed a positive correlation with the total sum of CER. Consistent with our findings, previous studies have demonstrated altered sphingolipid metabolism in AD brain parenchyma, characterized by decreased SM and increased CER levels [15, 29, 30]. Indeed, it has been reported that CER modulates γ-secretase activity, leading to increases in Aβ42 production [61]. CER also induces excessive reactive oxygen species (ROS) formation, subsequently triggering apoptosis in various cell types [17]. Conversely, Aβ42 has been shown to promote CER accumulation in astrocytes by activating through the activation of nSMase [13, 21], an enzyme that hydrolyzes SM to CER. Aβ accumulation promotes CER generation, which further induces a vicious pathological circle in exacerbating Aβ production. Indeed, we found that Aβ42 administration activates nSMase in the iVMCLs with APOE ε4/ε4, resulting in the reduced SM and increased CER. The nSMase has been co-localized with senile plaques [51], and nSMase has also been detected in cerebral microvessels [10]. Since selective inhibition [39] or gene knockdown [72] of nSMase has been shown to attenuate Aβ-induced cell death, highlighting the role of the nSMase-ceramide cascade in Aβ-related pathogenesis, inhibiting this pathway may offer a potential therapeutic strategy for mitigating CAA. We also observed a positive association of cerebrovascular Aβ40 with PG/BMP(18:1–18:1). As BMP is a lipid associated with endosomal-lysosomal trafficking and known to reflect late endosomal expansion [22], this result raises the possibility of a link between Aβ40 and disrupted vascular lysosomal functions in CAA.

WGCNA of lipidomics demonstrated a negative association between APOE ε2 allele and levels of PE subspecies, including those possessing the acyl chain with 22 carbons and 6 double bonds. This result suggests the potential involvement of APOE ε2 in docosahexaenoic acid (DHA) metabolism. In AD brains, reduction of DHA levels in lipid rafts has been reported [43]. Protective effects of DHA on endothelial cells and pericytes have also been demonstrated in vitro, where DHA administration improves barrier integrity and suppresses ROS production [65]. Collectively, dysregulated DHA metabolism in the cerebrovasculature due to APOE ε2 may contribute to the increased vascular vulnerability, although further studies are needed. In contrast, APOE ε4 was positively associated with PE subspecies composed of more saturated acyl chains with fewer double bonds, which may reduce membrane fluidity and increase membrane rigidification [3, 5]. As rigid lipid membrane enhances Aβ accumulates and aggregation [12], APOE ε4 may exacerbate CAA severity by modulating PE distributions in the lipid membranes of cerebral vessels. WGCNA of lipidomics also found positive associations of PE plasmalogens with t-tau and p-tau231 levels in large cerebral vessels. Plasmalogens, a subclass of glycerophospholipids characterized by a vinyl ether-linked aliphatic chain at the sn-1 position of the glycerol backbone, play a crucial role in membrane integrity and protection against oxidative damage [41]. The increased plasmalogen levels in the cerebrovasculature may represent an adaptive compensatory response to oxidative stress and inflammation, possibly triggered by tau accumulation in perivascular lesions [31].

Several limitations of our study should be mentioned. First, the small sample size limits the generalizability of our findings. Second, we did not measure non-polar lipids such as cholesterol, cholesteryl esters, diacylglycerol (DAG), and triacylglycerol (TAG) due to the limited volumes of isolated cerebral vessels available for lipidomics. Additionally, the quantification of low-abundance lipids through non-targeted shotgun mass spectrometry was also challenging. Our future studies should aim to validate and expand upon these results in larger cohorts by employing sensitive and/or targeted lipidomic approaches, which will help overcome these limitations and provide a more comprehensive understanding of cerebrovascular lipid alterations in AD.

In summary, our results demonstrated that APOE and AD-related molecules influence cerebrovascular lipid profiles in AD brains. These findings collectively underscore the critical role of lipid homeostasis in cerebrovascular function and imply potential therapeutic avenues targeting lipid metabolism to mitigate AD-related vasculopathy.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

Aβ

Amyloid beta

AD

Alzheimer’s disease

ARS

Arteriolosclerosis

CAA

Cerebral amyloid angiopathy

APOE

Apolipoprotein E

BBB

Blood–brain barrier

GWAS

Genome-wide association studies

CAR

Acyl-carnitine

CER

Ceramide

LPC

Lysophosphatidylcholine

LPE

Lysophosphatidylethanolamine

PC

Phosphatidylcholine

PE

Phosphatidylethanolamine

PG

Phosphatidylglycerol

PI

Phosphatidylinositol

PS

Phosphatidylserine

SM

Sphingomyelin

WGCNA

Weighted gene co-expression network analysis

DHA

Docosahexaenoic acid

ROS

Reactive oxygen species

CPT1

Carnitine palmitoyltransferase 1

nSMase

Neutral Sphingomyelinase

iPSC

Induced pluripotent stem cell

iVMLCs

IPSC-derived vascular mural-like cells

HBSMC

Human primary brain vascular smooth muscle cells,

hCMEC/D3

Immortalized human cerebral microvascular endothelial cells/clone D3

BrdU

5-Bromo-2′-deoxy-uridine

DMSO

Dimethyl sulfoxide

BSA

Bovine serum albumin

α-SMA

Alpha smooth muscle actin

PDGFRβ

Platelet-derived growth factor receptor β

PVDF

Polyvinylidene fluoride

HRP

Horseradish peroxidase

MTT

3-(4,5-Dimethylthiazol-2-yl)-2,5-Diphenyltetrazolium Bromide

DB

Double bond

Author contributions

Y.I., and T.K. conceived the study; Y.I., and H.W. were responsible for data acquisition; Y.I., and A.A were involved in tissue processing for ELISA and MSD measurement Y.I., M.H., L.J.W., and T.K. were responsible for statistical analysis; Y.I., Y.R., and T.K. were responsible for bioinformatics and figure generation; Y.I., P.W., J.TCW., and T.K. were responsible for double bond-chain length analysis and figure generation; D.W.D. provided human postmortem brain samples; M.D. dissected the frozen human postmortem brain tissues; S.K., M.E.M., and D.W.D provided pathological evaluations and made substantial contributions in sample stratification; The Neuropathology Core, from which the brain samples were obtained, was operated under the supervision of R.C.P.; Y.I., W.L. executed the cell culture-related experiments and analyzed the data; G.B., X.H., and T.K. provided expert advice on the interpretation of the data; Y.I., and T.K. wrote the manuscript with input from co-authors; T.K. supervised the project; All authors read and approved the final manuscript.

Funding

This work was supported by NIH grants U19AG069701 (to T.K., X.H., and J.TCW.), R01AG071226, R01AG068034, R01AG083981, and RF1 AG081203 (to T.K.), R01AG082362, R01AG083941 (to J.TCW.) and P30AG062677 (to R.C.P), a Cure Alzheimer’s Fund grant (to T.K.), Florida Department of Health Ed and Ethel Moore Alzheimer’s Disease Research Program 22A08 (to Y.I.), Carol and Gene Ludwig Family Foundation, and The Edward N. and Della L. Thome Memorial Foundation Awards Program in AD Research, Health Resources in Action (to J.TCW).

Data availability

Supplementary materials were available as the online version.

Declarations

Conflict of interest

The authors declare no competing interests.