APOE ε4 and Alzheimer’s disease diagnosis associated differences in L-carnitine, GBB, TMAO and acylcarnitines in blood and brain

Summary

Background:

The apolipoprotein E (APOE) ε4 allele, involved in fatty acid (FA) metabolism, is a major genetic risk factor for Alzheimer’s disease (AD). This study examined the influence of APOE genotypes on blood and brain markers of the L-carnitine system, necessary for fatty acid oxidation (FAO), and their collective influence on the clinical and pathological outcomes of AD.

Methods:

L-carnitine, its metabolites γ-butyrobetaine (GBB) and trimethylamine-n-oxide (TMAO), and its esters (acylcarnitines) were analyzed in blood from predominantly White community/clinic-based individuals (n = 372) and plasma and brain (n = 79) from the Religious Order Study (ROS) using liquid chromatography tandem mass spectrometry (LC-MS/MS).

Findings:

Relative to total blood acylcarnitines, levels of short chain acylcarnitines (SCAs) were higher whereas long chain acylcarnitines (LCAs) were lower in AD, which was observed preclinically in APOE ε4s. Plasma medium chain acylcarnitines (MCAs) were higher amongst cognitively healthy APOE ε2 carriers relative to other genotypes. Compared to their respective controls, elevated TMAO and lower L-carnitine and GBB were associated with AD clinical diagnosis and these differences were detected at pre- clinically among APOE ε4 carriers. Plasma and brain GBB, TMAO, and acylcarnitines were also associated with post-mortem brain amyloid, tau, and cerebrovascular pathologies.

Interpretation:

Alterations of blood L-carnitine, GBB, TMAO and acylcarnitines occur early in clinical AD progression and are influenced by APOE genotype. These changes correlate with post-mortem brain AD and cerebrovascular pathologies. Additional studies are required to better understand the role of the FAO disturbances in AD.

Introduction

Late onset AD is a progressive degenerative brain disease of aging, and the most common cause of dementia in persons older than 60 years [1]. The APOE ε4 allele is a major genetic risk factor for late onset AD and has a frequency of 38% in White AD patients compared with 14% in cognitively healthy White individuals [2], thereby increasing the risk of AD by approximately 15 times in APOE ε4 homozygotes compared to APOE ε3 homozygotes along with reducing the age of onset by 12 years [3]. On the other hand, APOE ε2 is independently associated with longevity, protection against cognitive decline and AD risk, particularly among APOE ε2/ε2 homozygous individuals [4,5]. Evaluation of a 5000 cases of neuropathologically confirmed AD and controls suggested that the risk of AD was 34% lower among ε2/ ε3 and 13% lower among ε2 homozygous [6]. Pathologically, AD is defined by an accumulation of amyloid plaques and tau neurofibrillary tangles (NFTs) in the brain and clinically presents with a progressive cognitive decline [7]. Although, brain amyloid plaques and NFTs are considered key pathological hallmarks of AD, cerebrovascular dysfunction, neuroinflammation and metabolic disorders are also considered major contributors to the pathobiology of AD, all of which are influenced by the APOE ε4 allele [8].

Low cerebrospinal fluid (CSF) Aβ₄₂ is an early AD biomarker, followed by elevated levels of brain amyloid through positive emission tomography (PET) imaging (15 years before symptomatic onset), and then elevated CSF tau (both total and phosphorylated at residue 181), decreased ¹⁸fluorodeoxyglucose (FDG) on PET in the tempo-parietal cortex and lastly atrophy on magnetic resonance imaging in the temporal lobe regions and medial parietal cortex [9,10]. Specifically, decreased FDG is further suggested to be modulated by APOE genotypes, as in cognitively healthy middle-aged APOE ε4 homozygotes lower brain glycolytic rates are observed, compared to non-ε4s, in the same regions as those later affected by AD [8]. Further, brain glucose hypometabolism is detected in individuals with mild cognitive impairment (MCI) (in all APOE genotypes) before progression to AD but not in individuals with stable MCI [12]. The evolving pathology of AD varies between individuals in terms of severity which can be affected by genetic factors such as the APOE ε4 allele [6]. Functionally, the apolipoprotein E (apoE) protein associates with lipoprotein particles that facilitate lipid binding and transport between tissues for energy metabolism, steroid synthesis, and the regulation of innate immunity [13]. Although, glucose metabolism in the brain is one of the better studied bioenergetic pathways, the understanding of the use of alternative fuels in aging and with AD is an expanding area of research. In aging, cerebrovascular dysfunction and brain glucose hypometabolism may require a shift in substrate use for energy production, such as FA, as their oxidation by-product (acetyl-CoA) is important for ketogenesis [8,14]. Studies suggest that the APOE ε4 allele leads to an increased reliance on brain FAO compared to other genotypes [13,15], which could represent an earlier shift in the use of alternative energy substrates in APOE ε4 carriers due to earlier impairments in glucose use, possibly making them more vulnerable to subsequent AD pathology. Animal studies also show that compared to the ε4 allele, the presence of the ε2 allele corresponds with a robust glucose uptake and metabolism capacity [16]. However, relatively little is known about FAO capacity in APOE ε2.

Markers of FAO have previously been studied in the context of cardiovascular diseases and metabolic disorders, such as diabetes [17,18]. Acylcarnitines are also used clinically as part of the diagnostic work-up of FAO and organic acid disorders [19]. As such, blood biomarkers of FAO may also provide important information about this shift in bioenergetics with aging and its impairment in APOE ε4 carriers. The L-carnitine transport system regulates mitochondrial FAO by controlling the entrance of long chain FAs (LCFAs) into the mitochondrion, as LCFAs first need to be esterified to L-carnitine to form acylcarnitines, before being oxidized to produce acetyl-CoA which can enter the tricarboxylic acid cycle or be used for ketone production [20]. L-carnitine is also involved in the export of FAs from the mitochondria as SCAs, MCAs, and LCAs when there is a build-up of these acyl-groups due to their incomplete FAO (Figure 1). Additionally, MCAs can be produced by peroxisomes after chain shortening of very long chain fatty acids (VLCFAs), where like in the mitochondrial L-carnitine shuttle carnitine palmitoyl-transferases trans-acetylate medium chain acyl-CoAs to acylcarnitines for export [19]. Odd chain acylcarnitines (OCAs) can also be produced from branched chain amino acid catabolism [24].

Figure 1. Schematic illustration of mitochondrial acylcarnitine metabolism.

First, LCFAs are activated with a -CoA group on the outer mitochondrial membrane before their trans-esterification to LCAs by CPT1, which releases a free CoA and uses an L-carnitine. These LCAs are then able to enter the mitochondrion through CACT in exchange for L-carnitine. Once in the mitochondrial matrix LCAs are trans-esterified back to LCFA-CoAs by CPT2 using one CoA and releasing one L-carnitine. These LCAs can then undergo β-oxidation in the mitochondrion. Shorter chain FAs can directly enter the mitochondrion and be activated with a CoA before undergoing β-oxidation. Each round of β-oxidation yields an acetyl-CoA. Incomplete β-oxidation leads to intra-mitochondrial build-up of acyl-CoA groups. This results in reverse transport of excess acyl- groups to the cytosol and blood. This happens by the trans-esterification of excess intra-mitochondrial acyl-CoAs to their corresponding acylcarnitines by either CPT2 or CAT (depending on chain length). Acylcarnitines can then exit the mitochondrion through CACT. Abbreviations: FABP: fatty acid binding protein, CD36: fatty acid translocase, LCFA: long chain fatty acid (C13–21), LACS: long chain fatty acyl-CoA ligase, LCFA-CoA: long chain fatty acyl-CoA, CPT1/2: carnitine palmitoyltransferase 1/2, CACT: Carnitine acylcarnitine translocase, SCFA: short chain fatty acid (C2–5), MCFA: medium chain fatty acid (C6–12), SCAS: short chain fatty acyl-CoA ligase, MCAS: medium chain fatty acyl-CoA ligase, SCFA-CoA: short chain fatty acyl-CoA, MCFA-CoA: medium chain fatty acyl-CoA, CAT: carnitine acetyltransferase, SCA: short chain acylcarnitine, MCA: medium chain acylcarnitine, LCA: long chain acylcarnitine, PDH: pyruvate dehydrogenase.

While brain acylcarnitine levels can inform on FAO in the brain, blood levels are believed to reflect the removal of incompletely oxidized FAs from the mitochondria and peroxisomes [21]. Blood acylcarnitines reflect contributions of many tissue sources. The blood-brain-barrier (BBB) expresses required transporters for L-carnitine, its precursor γ-butyrobetaine (GBB), and SCAs [25–27], suggesting an exchange of these lipids and metabolites between the periphery and the brain. Taken together, changes in their blood levels may be informative about the L-carnitine mediated FAO status within the brain. Altered levels of plasma L-carnitine and TMAO in blood and CSF have also been associated with cognitive decline and AD diagnosis, however, the influence of APOE on these metabolites in AD have not been examined [28,29]. Further, blood TMAO, L-carnitine, GBB and acylcarnitine levels have also been independently associated with fasting insulin levels and cardiovascular disease, factors that are thought to contribute to AD risk [30,31]. We therefore hypothesized that alterations in the FAO markers derived from the L-carnitine system in combination with APOE genotype will provide an indication of altered bioenergetics associated with AD onset and progression. Using a targeted LC-MS/MS assay, we investigated blood and brain L-carnitine, L-carnitine metabolites (GBB and TMAO) and acylcarnitines over the continuum of AD clinical diagnosis and pathology to establish a timeline of changes in these compounds and examined the influence of APOE genotype on their levels.

Methods

Control and preclinical MCI/AD (ADAPT subsample 1)

Our first dataset of preclinical MCI/AD and cognitively healthy control individuals comes from a subset of samples (n= 160) from a single participating site of the Alzheimer’s Disease Anti-inflammatory Prevention Trial (ADAPT). The ADAPT trial was a randomized multicenter double-blinded and placebo-controlled phase 4 trial with three intervention arms (one placebo and two treatment groups) aimed at investigating the utility of preventing or delaying AD using either of the non-steroidal anti-inflammatory drugs (NSAIDs) celecoxib (Celebrex^®, Pfizer) or naproxen sodium (Aleve^®, Bayer) [32]. This cohort was enriched for individuals at risk of developing AD by enrolling participants who were 70 years or older with a family history of AD-like dementia. Exclusion criteria included cognitive impairment or dementia (at the time of enrollment), clinically significant hypertension, anemia, liver disease, kidney disease and plasma creatinine ≥ 1.5 mg/dL, current alcohol dependence or abuse, use of anticoagulants or use of 4 doses/week (14 days before study start) of histamine H2 antagonists, non-aspirin NSAIDs, aspirin (>81mg/day), or corticosteroids (see ref. 29 for full list of inclusion/exclusion criteria). Participants were followed for up to five years, during which a subset progressed to either MCI or AD; these were defined as the pre-clinical MCI/AD group, while those who remained cognitively healthy throughout the follow-ups were considered controls. No effect of either treatment was observed on the prevention of dementia [33], therefore all intervention arms were used for this study. A separate consent form was obtained from individuals who participated in this sub-study. For this study serum samples from year 1 of the cohort (up to two years before symptoms appeared in participants who went on to develop MCI or AD) were used for lipidomic analysis. In brief, non-fasting blood draws were obtained by trained phlebotomists. Venous blood was collected in serum separator tubes and a whole blood aliquot was taken for APOE genotyping. Samples were aliquoted and stored at −80 °C. Whole blood aliquots or peripheral blood mononuclear cells (PBMCs) were used for APOE genotyping in our laboratory, purifying the DNA from whole blood using the Gentra Puregene Blood Kit (Gentra Systems) and then using the Amplification Refractory Mutation System as described by Wenham et al. [34]. A consensus team provided a diagnosis of probable AD in accordance with the National Institute of Neurological and Communicative Disorders and Stroke- Alzheimer’s Disease and Related Disorders Association (NINCDS-ADRDA) [35]. For MCI diagnosis participants needed to have an impairment in at least one cognitive domain including memory, executive function, attention, language and visuospatial skill while having preserved independent living activities consistent with the Petersen criteria [36]. Efforts were made to rule out other causes of cognitive impairment such as vascular, traumatic and other medical causes of cognitive decline as per Albert et al., (34). For 3 samples APOE genotype was undetermined; these were still used for analyses of other factors.

Control, MCI, and AD (ADAPT subsample 2)

Plasma from another subset of the ADAPT cohort described above was used. This second subset of ADAPT samples (n= 158) is constituted of participants from multiple participating research centers. The same exclusion/inclusion and diagnostic criteria as described above were used. Plasma samples were collected in ethylenediaminetetraacetic acid (EDTA)-containing vacutainers and processed before being aliquoted and stored at −80°C until sample analysis. Whole blood was used for APOE genotyping as described above.

Control, MCI, and AD (RNC)

The Roskamp Neurology Clinic (RNC) cohort (n= 54) is constituted of community-dwelling individuals who were recruited from the RNC and agreed to blood donation. In this cohort participants were assigned diagnoses of AD or MCI, according to the NINCDS-ADRDA criteria [7]. Healthy controls were individuals from the community who denied any history of a cognitive disorder and did not have a diagnosis of MCI or AD or reported any memory concerns. Non-fasting blood draws were conducted by trained phlebotomists, and plasma samples were then collected in EDTA-containing vacutainers and processed as described in [38] before being aliquoted and stored at −80°C until analysis. Whole blood was used for APOE genotyping using the EZWay direct APOE genotyping Kit (Cosmo Bio, USA) as per manufacturer’s instructions.

Control, MCI, and AD (ROS)

Brain tissues from the inferior orbital cortex and (fasting as well as non-fasting) plasma samples from the same individuals were obtained from the Religious Order Study (ROS) (n = 79), an ongoing longitudinal clinical-pathological study of aging and AD which started in 1994 [39,40]. This study enrolls Catholic nuns, priests, and brothers, who agreed to annual clinical evaluation and brain donation. The ROS was approved by the Institutional Review Board of Rush University Medical Center. Each participant signed an informed consent, Anatomic Gift Act, and Repository Consent to allow their data to be repurposed. In this cohort a self-report of medical history was obtained by trained clinicians and clinical diagnosis of AD was made using NINCDS-ADRDA criteria, as previously reported [41]. Plasma samples examined here were collected on average three years before death (3 ± 1.7 years, mean ± SD). As part of basic metabolic and lipid panels, blood glucose, total cholesterol, high density lipoprotein (HDL) cholesterol, low density lipoprotein (LDL) cholesterol, and triglycerides were measured from a commercial CLIA certified laboratory. Brain autopsies were performed on both cerebral hemispheres and the brainstem, first by macroscopic then microscopic evaluation. Systematic evaluations were conducted in the following brain regions: mid-frontal, superior temporal, inferior parietal and entorhinal cortices, hippocampus, basal ganglia, thalamus and substantia nigra, where neuritic plaques (NPs), diffuse plaques (DPs) and NFTs were visualized and counted [41]. Based on detailed evaluations, a neuropathological diagnosis was reached by a board-certified neuropathologist without access to clinical information based on the Consortium to Establish a Registry for Alzheimer’s Disease [42] and a second diagnosis based on the National Institute on Aging-Reagan criteria [43]. Braak stage was also assessed [44]. Additionally, macroscopic, and microscopic infarcts were evaluated and recorded. Due to the low number of APOE ε4 homozygotes (n = 3) these were paired with APOE ε3/ε4 individuals into the APOE ε3/ε4 + APOE ε4/ε4 group for genotype stratification. The brain samples were homogenized in a glass dounce homogenizer (Cole Parmer) adding 500μL of lysis buffer mix containing 10mL Mammalian Protein Extraction Reagent (M-PER, Thermo Fisher), 100μL Protease & Phosphatase inhibitor (Thermo Scientific) and 100μL of 0.5M EDTA (Thermo Scientific). The homogenates were then aliquoted and stored at −80°C. The glass vessel, plunger and spatula were thoroughly cleaned between each sample. The Thermo Scientific^™ Pierce^™ bicinchoninic acid (BCA) Protein Assay Kit was used to quantify proteins in the ROS brain tissue homogenates, which were diluted 50x for optimal reading.

Acylcarnitine assay

Acylcarnitines were analyzed using LC-MS/MS, extracting acylcarnitines from 50μl of plasma or 50μl brain homogenate. Samples were spiked with 5μl of acylcarnitine internal standard (IS) mix containing 12μM TMAO-d9 (Cambridge Isotope Laboratories, Inc., CIL), as well as CIL’s L-carnitine and acylcarnitine reference standards mixes (NSK-B and NSK-B-G). Protein precipitation was performed by adding 10 volumes of 25% methanol (MeOH) in acetonitrile (ACN). This was followed by vortex mixing for approximately 1 min before centrifuging samples at 10,000 relative centrifugal force (RCF) for 20 mins at 4°C. Then 80% of the supernatant was taken and dried down before reconstitution in 100μl of HPLC mobile phase A (90:5:5 ACN: H₂O:100mM ammonium formate (AmFm) in H₂O) and vortexed for approximately 10 seconds. Samples were then transferred to 0.2μm centrifugal filters (ThermoFisher Scientific) and centrifuged again at 10,000 RCF for 5 mins at 4°C and the filtrate transferred to glass auto-sampler vials for injection.

A Shimadzu Prominence Ultra-Fast LC interfaced to a Thermo Scientific Q Exactive MS with a Thermo Scientific^™ heated electrospray ionization (HESI-II) probe was used for LC-MS/MS. Chromatography was on a Kinetex 2.6μm HILIC 100Å, 100 × 2.1mm ID LC column (Phenomenex), with a constant flow rate of 250μl/min with a solvent gradient from 20% Mobile Phase B (50:45:5, ACN:H₂O:100mM AmFm in H₂O) to 22% B at 5 mins, 40% B at 10mins, 60% B at 13 mins, 80% B at 15 mins, 99% B 15 mins. The column was re-equilibrated at 20% B for 5 mins before the next sample injection. All samples were kept at 5°C in the auto-sampler tray for the duration of the analysis. Data was acquired in positive mode using parallel reaction monitoring with an inclusion list of acylcarnitines of interest (Table 1S). Acylcarnitine species were abbreviated as follow; Cx:y-CAR; acylcarnitines, Cx:y-OH-CAR; hydroxy acylcarnitines.

Data processing and statistical analysis

Peak areas were integrated using the Tracefinder^™ software with a target compound list of acylcarnitines of interest. A mass window of 5 ppm was used for all ion plots. Concentrations were calculated based on relative IS amount added, with the data normalized across runs using a quality control sample that was run along with each batch. Each sample was injected in triplicate, and all triplicate runs with a CV above 20% were excluded from further analysis. The SPSS software was used for statistical analysis. Normal distribution of data were assessed by examining mean skewness and kurtosis for each group with values −2 ≤ or ≥2 (where applicable) considered reflective of non-normal distribution. Non-normally distributed data were transformed using the natural log (ln) function and tested for normality again. Data that could not be normalized using ln were analyzed using non-parametric tests. For normally distributed data a mixed linear model (MLM) was used. Individual IDs were chosen as subject variable. Sex, diagnosis, and APOE genotype were chosen as factors and age as a covariate (post-mortem interval, PMI, was an additional covariate for the ROS subset) and participant IDs as random effects. When comparing more than 2 groups if the F tests of fixed effects was statistically significant (p<0.05) a pair-wise comparison with least significant difference (LSD) was performed. For non-normally distributed data independent-samples non-parametric tests were used (either Mann-Whitney U test for two group analysis or the Kruskal-Wallis 1-way ANOVA for more than two groups). To minimize the risk of false positives with multiple comparisons a Benjamini–Hochberg (B–H) correction was performed, as described in Huguenard et al. [38]. For the combined datasets analysis, pre-clinical MCI/AD and MCI diagnostic groups were combined since there were no significant differences in MMSE between these groups. Due to the low prevalence of certain APOE allelic combinations in the general population, participants of the combined datasets were grouped into APOE ε4− and APOE ε4+ groups where the APOE ε4− group comprised of participants who did not have any APOE ε4 alleles while the APOE ε4+ group consisted of participants who had at least one APOE ε4 allele. For hierarchical clustering, data were normalized, scaled, and analyzed with Ward’s clustering method. The top analytes were selected by ANOVA. For the pathology and acylcarnitine correlation analyses Kendall’s tau-b was used with a two-tailed significance threshold p<0.05.

For total, acylcarnitines of all chain lengths were summed. These composite variables excluded hydroxy- species, TMAO, GBB and L-carnitine.

Ratios of individual species to total levels of acylcarnitines were calculated as follows:

$$\frac{analyte(nM)}{sumoftotalacylcarnitines(nM)}$$

For calculation of SCAs, acylcarnitines of chain lengths C2-C5 were summed, for medium chain MCAs chain lengths C6-C12 were summed, for LCAs chain lengths C13-C20 were summed and for very long chain acylcarnitines (VLCAs) chain lengths ≥C21 were summed. Molar chain length ratios were calculated:

$$\frac{chainlengthgroup(nM)}{sumofallacylcarnitines(nM)}$$

Graphs were built using the GraphPad Prism 9.2.0 software and MetaboAnalyst 5.0 (https://www.metaboanalyst.ca/). The diagram figure was drawn using BioRender (https://biorender.com/).

Results

Blood levels of L-carnitine, TMAO, GBB and acylcarnitines are associated with AD clinical progression and APOE genotypes in combined datasets.

A subset of samples from two different cohorts were combined to assess the impact of APOE genotypes on blood acylcarnitines in individuals at different stages of AD (Table 1). Basic demographics for each of these cohorts can be found in supplementary tables 2–4. In the combined datasets there were no significant differences in age, sex, and race between AD diagnostic groups and controls stratified by the APOE ε4 carrier status.

Table 1.

Basic demographics of the combined datasets.

Combined datasets (n= 372)
		Control		Pre-clinical MCI/AD + MCI		AD
APOE ε4 *		−	+	−	+	−	+
Numbers		n= 183	n= 112	n= 30	n= 26	n= 8	n= 13
Age, mean ± SD		77±4	76±4	78±4	77±4	77±4	76±4
Sex, n	Male	89	63	16	16	5	5
	Female	94	49	14	10	3	8
Race, n	Asian/Pacific Islander	2	0	0	0	0	0
	Black	1	1	0	1	1	0
	Hispanic/Latino	1	0	0	0	0	0
	White	179	110	30	25	7	13
	Other	0	1	0	0	0	0
APOE, n	ε2/ε2	2	0	0	0	0	0
	ε2/ε3	24	0	5	0	1	0
	ε2/ε4	0	7	0	2	0	1
	ε3/ε3	157	0	25	0	7	0
	ε3/ε4	0	99	0	19	0	10
	ε4/ε4	0	6	0	5	0	2
MMSE, mean ± SD		29±1	29±2	28±2	27±2	21±9	18±7

The MMSE scores were not significantly different between APOE ε4− and APOE ε4+ controls, but were significantly lower in the APOE ε4+ Pre-clinical MCI/AD + MCI, APOE ε4− AD and APOE ε4+ AD groups compared to both the APOE ε4− Control and the APOE ε4+ Control group (bolded values, all p< 0.001).

Abbreviations; MCI: mild cognitive impairment, AD: Alzheimer’s Disease, MMSE: Mini-mental state exam.

Statistics: independent samples Kruskal-Wallis test with multiple comparison and B-H correction.

^*The APOE ε4− group includes individuals who do not carry any ε4 alleles, while the APOE ε4+ group includes individuals who carry one or more ε4 allele.

The effect of diagnosis stratified by APOE ε4 carrier status on both acylcarnitine species ratios and chain length ratios to total acylcarnitines were further investigated (Figure 2A–C). These ratios are reflective of differences in the relative abundance of acylcarnitine species (and chain length groups) within total acylcarnitine levels. Levels of acylcarnitine were compared across diagnostic groups stratified by APOE ε4 carrier status in the combined cohort using cohort as a covariate to account for any differences between the datasets used. Raw concentration data can be found in Supplementary table 5S.

Figure 2. Decreases in plasma and serum LCAs, L-carnitine, GBB, and increase in SCAs and TMAO with AD cognitive progression and APOE ε4 carrier status.

A Heatmaps showing species ratios most strongly associated with cognitive diagnoses stratified by APOE ε4 carrier status. B Z-score heatmap showing acylcarnitine chain length ratios with cognitive diagnoses stratified by APOE ε4 carrier status. C Dot plot showing the mean ratio of TMAO to L-carnitine ± 95% CI with cognitive diagnoses stratified by ε4 carrier status.

Numbers per group: E4− Control n= 183, E4+ Control n= 112, E4− Pre-MCI/AD + MCI n= 30, E4+ Pre-MCI/AD + MCI n= 26, E4− AD n= 8, E4+ AD n= 13. Abbreviations; AD: Alzheimer’s disease, MCI: mild cognitive impairment, TMAO: trimethylamine-n-oxide, GBB: γ-butyrobetaine, Cx:y-CAR: acylcarnitines, Cx:y-OH-CAR: hydroxy acylcarnitines, SCA: short chain acylcarnitines, MCA: medium chain acylcarnitines, LCA: long chain acylcarnitines, VLCA: very long chain acylcarnitines. Statistics: For the clustering heatmap data was normalized and scaled before being analyzed with the Ward’s clustering method, top analytes were selected by ANOVA.

Cluster analysis showed that profiles of individual species (as ratios relative to total acylcarnitines) were most similar between the APOE ε4− pre-clinical MCI/AD and APOE ε4+ and APOE ε4− control groups with no significant changes in APOE ε4− pre-clinical MCI/AD compared to APOE ε4− controls (Figure 2A and Supplementary Figure 1SA). Acylcarnitines were similar between APOE ε4+ pre-clinical MCI/AD + MCI and APOE ε4+ and APOE ε4− AD groups and several species were significantly altered compared to APOE ε4+ controls (Figure 2A and Supplementary Figure 1SA). Differences in blood acylcarnitine ratios were most pronounced in the APOE ε4+ AD group where the odd chain OCA C5:0-CAR as well LCA species were significantly lower compared to the APOE ε4+ control group (Figure 2A and Supplementary Figure 1SA). Blood L-carnitine, GBB, OCA, LCA and VLCA species ratios were also lower in women compared to men while the SCA C2:0-CAR ratio was significantly greater in women (Supplementary Figure 1SA).

Ratios of SCAs relative to total acylcarnitines were highest among the APOE ε4+ AD group (Figure 2B). The LCA ratios were lower with greater AD severity and lowest among APOE ε4+ individuals with AD with similar trends seen in the VLCA ratios (Figure 2B). However, interestingly there were no significant differences in total acylcarnitine levels between clinical diagnoses with APOE ε4 carrier status groups (Supplementary Figure 2SA).

L-carnitine metabolites differed by clinical diagnosis and APOE genotypes where significant increases in TMAO and decreases in L-carnitine (relative to total acylcarnitine levels) were detected in the APOE ε4+ pre-clinical MCI/AD + MCI group and both APOE ε4− and APOE ε4+ AD groups compared to their respective controls (Figure 2A and Supplementary Figure 1SA). Analyses of ratios of TMAO to L-carnitine showed that while APOE ε4+ and APOE ε4− control had similar ratios compared to other groups, the APOE ε4+ pre-clinical MCI/AD + MCI and APOE ε4+ AD groups had higher ratios than the APOE ε4− pre-clinical MCI/AD + MCI and ε4− AD groups and controls (Figure 2C). Further, GBB ratios were lower in APOE ε4+ pre-clinical MCI/AD + MCI, APOE ε4− AD and APOE ε4+ AD groups compared to their respective controls (Figure 2A and Supplementary Figure 1SA). Conversely, GBB and L-carnitine were significantly higher in APOE ε4+ controls compared to APOE ε4− controls (Figure 2A and Supplementary Figure 1SA).

Characterization of the ROS cohort demonstrates expected effects of APOE genotype on brain pathologies.

Samples from a subset of the ROS cohort was used to investigate the association of clinical diagnosis, brain pathology and APOE genotypes with plasma and brain L-carnitine, GBB, TMAO and acylcarnitines. In this subset of samples, plasma (drawn approximately 3 years before death) and brain samples from the inferior orbital cortex were available from deceased individuals with a clinical diagnosis of MCI or AD. This cohort was also enriched for the APOE ε2 allele among controls and AD cases, allowing for the investigation of diagnosis by APOE ε2 and APOE ε4 carrier status (of note: the MCI group was only constituted of APOE ε3/ε3 individuals). No significant differences were found in age, sex, race, fasting status or PMI between groups by diagnosis stratified by APOE ε2 or APOE ε4 carrier status (Table 2). Participants in this sample subset were approximately 9 years older than in the combined clinical datasets previously analyzed at the time of blood sampling. The APOE ε2 AD group had significantly higher total cholesterol to HDL cholesterol ratios compared to their controls (Table 2). As expected, there were significant differences between AD diagnosis and APOE genotypes in relation to tau, amyloid and cerebral amyloid angiopathy (CAA) pathologies (Supplementary Figure 3SA and B). Carriers of the APOE ε4 allele also had significantly more CAA than APOE ε3 homozygotes but not APOE ε2 carriers (Supplementary Figure 3SC). Additional data on sex and age influences can be found in supplementary Figure 4SA and B.

Table 2.

Basic demographics of the ROS subset.

ROS (n=79)
		Control			MCI	AD
APOE genotype		ε2/ε3	ε3/ε3	ε3/ε4+ ε4/ε4*	ε3/ε3	ε2/ε3	ε3/ε3	ε3/ε4+ ε4/ε4*
Numbers		n= 10	n= 12	n= 11	n= 12	n= 10	n= 12	n= 12
Age at death (years), mean ± SD		89±6	87±6	87±5	88±6	93±7	92±6	92±7
Sex, n	Male	3	6	5	4	2	7	4
	Female	7	6	6	8	8	5	8
Race, n	Black	0	0	1	0	0	0	0
	White	10	12	10	11	10	12	12
	Other	0	0	0	1	0	0	0
Years of education, mean ± SD		18±3	21±4	19±3	18±2	17±2	17±4	17±5
MMSE, average ± SD		29±1	29±3	29±1	27±2	23±7	22±8	13±9
Glucose (mg/dL), mean ± SD		94±11	100±23	121±61	137±61	134±77	105±32	136±37
Total cholesterol (mg/dL), mean ± SD		168±35	171±40	167±21	160±43	181±35	181±40	174±44
HDL cholesterol (mg/dL), mean ± SD		65±29	58±15	53±9	54±17	47±14	57±13	45±8
Total cholesterol:HDL cholesterol, mean ± SD		3±1.5	3±0.7	3±0.7	3±0.9	4±0.8	3±1	4±0.8
LDL cholesterol (mg/dL), mean ± SD		79±17	90±29	88±21	77±33	102±30	102±37	98±42
Triglycerides (mg/dL), mean ± SD		124±68	112±58	129±67	144±68	161±53	113±52	154±59
PMI, average ± SD		9±4	11±7	9±4	8±4	6±1	11±6	8±5
CERAD score (NPs), % (n)	No AD	70 (7)	58 (7)	46 (5)	25 (3)	30 (3)	17 (2)	0 (0)
	Possible AD	10 (1)	8 (1)	18 (2)	33 (4)	20 (2)	0 (0)	0 (0)
	Probable AD	20 (2)	25 (3)	27 (3)	25 (3)	40 (4)	50 (6)	17 (2)
	Definite AD	0 (0)	8 (1)	9 (1)	17 (2)	10 (1)	33 (4)	83 (10)
Braak stage (NFTs), % (n)	I	30 (3)	0 (0)	0 (0)	0 (0)	0 (0)	8 (1)	0 (0)
	II	10 (1)	17 (2)	27 (3)	25 (3)	10 (1)	17 (2)	0 (0)
	III	10 (1)	33 (4)	27 (3)	33 (4)	50 (5)	0 (0)	0 (0)
	IV	50 (5)	42 (5)	46 (5)	33 (4)	40 (4)	25 (3)	0 (0)
	V	0 (0)	8 (1)	0 (0)	8 (1)	0 (0)	50 (6)	92 (11)
	VI	0 (0)	0 (0)	0 (0)	0 (0)	0 (0)	0 (0)	8 (1)
NIA-Reagan diagnosis, % (n)	Low	80 (8)	83 (10)	64 (7)	67 (8)	50 (5)	33 (4)	0 (0)
	Intermediate	20 (2)	8 (1)	36 (4)	25 (3)	50 (5)	50 (6)	17 (2)
	High	0 (0)	8 (1)	0 (0)	8 (1)	0 (0)	17 (2)	83 (10)
Lewy body pathology, % (n)	Not present	90 (9)	100 (12)	91 (10)	75 (9)	50 (5)	58 (7)	58 (7)
	Nigral-predominant	0 (0)	0 (0)	0 (0)	8 (1)	10 (1)	0 (0)	0 (0)
	Limbic type	0 (0)	0 (0)	0 (0)	8 (1)	20 (2)	8 (1)	25 (3)
	Neocortical type	10 (1)	0 (0)	0 (0)	0 (0)	20 (2)	25 (3)	17 (2)
	Unknown	0 (0)	0 (0)	9 (1)	8 (1)	0 (0)	8 (1)	0 (0)
Hippocampal sclerosis, % (n)	Not present	100 (10)	100 (12)	91 (10)	100 (12)	80 (8)	83 (10)	75 (9)
	Present (CA1 affected)	0 (0)	0 (0)	9 (1)	0 (0)	20 (2)	17 (2)	25 (3)
TDP-43 pathology, % (n)	None	60 (6)	83 (10)	73 (8)	67 (8)	40 (4)	17 (2)	17 (2)
	Amygdala	10 (1)	17 (2)	9 (1)	25 (3)	20 (2)	25 (3)	25 (3)
	Amygdala and limbic	10 (1)	0 (0)	9 (1)	8 (1)	20 (2)	8 (1)	25 (3)
	Amygdala, limbic and neocortical	20 (2)	0 (0)	9 (1)	0 (0)	20 (2)	50 (6)	33 (4)
Arteriolosclerosis, % (n)	None	30 (3)	25 (3)	55 (6)	58 (7)	30 (3)	17 (2)	50 (6)
	Mild	50 (5)	58 (7)	27 (3)	25 (3)	10 (1)	50 (6)	25 (3)
	Moderate	20 (2)	17 (2)	18 (2)	17 (2)	50 (5)	17 (2)	17 (2)
	Severe	0 (0)	0 (0)	0 (0)	0 (0)	10 (1)	17 (2)	8 (1)
Cerebral atherosclerosis, % (n)	None/possible	40 (4)	33 (4)	46 (5)	25 (3)	0 (0)	33 (4)	17 (2)
	Mild	40 (4)	58 (7)	36 (4)	42 (5)	40 (4)	50 (6)	58 (7)
	Moderate	20 (2)	8 (1)	18 (2)	33 (4)	40 (4)	17 (2)	25 (3)
	Severe	0 (0)	0 (0)	0 (0)	0 (0)	20 (2)	0 (0)	0 (0)
Cerebral amyloid angiopathy, % (n)	None	40 (4)	42 (5)	18 (2)	50 (6)	20 (2)	33 (4)	0 (0)
	Mild	10 (1)	42 (5)	55 (6)	17 (2)	60 (6)	17 (2)	33 (4)
	Moderate	40 (4)	17 (2)	18 (2)	8 (1)	30 (3)	8 (1)	67 (8)
	Severe	10 (1)	0 (0)	9 (1)	25 (3)	0 (0)	8 (1)	17 (2)
Cerebral infarcts - Gross Chronic Any Location, % (n)	None	90 (9)	75 (9)	9 (9)	75 (9)	60 (6)	75 (9)	58 (7)
	One or more	10 (1)	25 (3)	18 (2)	25 (3)	40 (4)	25 (3)	42 (5)
Cerebral infarcts - Micro Chronic Any Location, % (n)	None	90 (9)	58 (7)	55 (6)	75 (9)	50 (5)	58 (7)	67 (8)
	One or more	10 (1)	42 (5)	46 (5)	25 (3)	50 (5)	42 (5)	33 (4)

The MMSE scores were lower in the MCI (p< 0.001), APOEε2, APOEε3/ε3 and APOEε4 AD groups (p= 0.005, p= 0.002, and p< 0.001, respectively) compared to the APOEε3/ε3 Control group (all bolded values). Total cholesterol to HDL cholesterol were also higher in APOEε2/ε3 AD compared to APOEε2/ε3 controls (bolded value, p= 0.006). Abbreviations; MCI: mild cognitive impairments, AD: Alzheimer’s disease, HDL: high density lipoprotein, LDL: low density lipoprotein, MMSE: Mini-mental state exam, PMI: post-mortem interval, NPs: neuritic plaques, NFTs: neurofibrillary tangles. Statistics: independent samples Kruskal-Wallis test with multiple comparison and B-H correction.

^*Due to the low number of APOE ε4 homozygotes in this subset (ε4/ε4 control n = 1, ε4/ε4 AD n =2) these were paired with APOE ε3/ε4 individuals into APOE ε3/ε4 + APOE ε4/ε4 groups.

Plasma levels of acylcarnitines are altered with diagnoses of MCI/AD and APOEε2 and APOE ε4 carrier status in the ROS subset.

The effect of MCI and AD diagnoses stratified by APOE genotypes on acylcarnitine species ratios and chain length ratios as well as sex for main and interaction effects were investigated by MLM analysis with age as covariate. Raw concentration data can be found in Supplementary table 6S. When investigating plasma acylcarnitine, similarly to what was observed in the combined datasets, a trend could be seen for a decrease in LCA ratios and an increase in SCA ratios in APOE ε2, APOE ε3/ε3, and APOE ε4 AD groups, which further clustered together based on their acylcarnitine ratio profiles (Figure 3A and B). The APOE ε4 carrier control group also had a trend for greater SCA ratios, and its acylcarnitine species ratio profile was similar to that of the APOE ε3/ε3 AD group. Interestingly, APOE ε2 carrier controls had an elevated plasma MCA ratio compared to other groups (Figure 3A and B). There were no significant effects of diagnosis and APOE genotype on L-carnitine, GBB and TMAO ratios in the ROS subset (Supplementary Figure 1SB). Individual acylcarnitine species ratios were not significantly different between groups, whereas there was a significant effect of the age as a covariate on OCA, LCA and VLCA ratios, which were higher with increasing age (Supplementary Figure 1SB). There were no significant differences between diagnosis with APOE genotype groups in plasma total acylcarnitines raw concentration levels (Supplementary Figure 2SB). However, additional MLM analysis of the raw concentrations of individual species showed that the levels of GBB, OCAs and the VLCA C26:0-CAR were significantly lower in women compared to males (Supplementary Figure 2SD).

Figure 3. Decreases in plasma LCAs and increases in SCAs with AD cognitive progression and APOE genotype.

A Heatmaps showing species ratios most strongly associated with cognitive diagnoses stratified by APOE genotype. B Z-score heatmap showing acylcarnitine chain length ratios with cognitive diagnoses stratified by APOE genotype. C Scatter plot with linear regression showing blood glucose plotted against plasma SCA levels. D Scatter plot with linear regression showing blood HDL cholesterol plotted against plasma OCA levels. E Scatter plot with linear regression showing blood total cholesterol to HDL cholesterol ratio plotted against plasma L-carnitine levels. Numbers per group: E2/E3 Control n= 10, E3/E3 Control n=12, E3/E4+E4/E4 Control n= 11, E3/E3 MCI n= 12, E2/E3 AD n= 10, E3/E3 AD n= 10, E3/E3 AD n= 12, E3/E4+E4/E4 AD n=12. Abbreviations; AD: Alzheimer’s disease, MCI: mild cognitive impairment, TMAO: trimethylamine-n-oxide, GBB: γ-butyrobetaine, Cx:y-CAR: acylcarnitines, Cx:y-OH-CAR: hydroxy acylcarnitines, SCA: short chain acylcarnitines, OCA: odd chain acylcarnitines, MCA: medium chain acylcarnitines, LCA: long chain acylcarnitines, VLCA: very long chain acylcarnitines, HDL: high density lipoprotein, Corr. coef.: correlation coefficient, Sig.: significance. Statistics: For the clustering heatmap data was normalized and scaled before being analyzed with the Ward’s clustering method, top analytes were selected by ANOVA. Correlations analyses were performed using Kendall’s tau-b.

As blood glucose and lipid levels have previously been associated with blood acylcarnitine profiles, we also investigated the associations of plasma acylcarnitines raw concentrations and glucose, total cholesterol, HDL cholesterol, total cholesterol to HDL ratio, LDL cholesterol, as well as triglyceride blood levels. Glucose was positively correlated with blood levels of raw SCAs concentrations (Figure 3C), while HDL cholesterol was negatively correlated with plasma raw OCAs concentrations (Figure 3D). Lastly, total cholesterol to HDL cholesterol ratios were positively correlated with plasma L-carnitine levels (Figure 3E). Since the APOE ε2 AD group had higher total to HDL cholesterol ratios compared to their control group, correlations were further explored in those groups. L-carnitine levels and total to HDL cholesterol ratios stayed positively correlated in each group with an increase of the y intercept in the APOE ε2 AD group compared to APOE ε2 controls (Supplementary Figure 2E).

Plasma L-carnitine, TMAO and acylcarnitine levels are differentially associated with AD and cerebrovascular pathologies in the ROS subset.

The predictive value of plasma acylcarnitine levels on brain pathology was examined in each APOE genotype in plasma samples obtained three years before brain pathological examination. When examining the relationship between plasma acylcarnitine levels and brain pathology in APOE ε2 carriers, plasma TMAO was found negatively correlated with overall amyloid (i.e., percent brain area occupied by amyloid beta protein) across multiple brain areas, while a LCA species was positively correlated with amyloid, and other acylcarnitines of different chain lengths showed either negative or positive correlations with tau (NFT) pathology. In APOE ε3 homozygotes, plasma levels of an OCA were negatively correlated with amyloid in four different brain areas, while two LCAs were negatively correlated with tau. Lastly, in APOE ε4 carriers there were negative correlations between OCAs and both amyloid and tau pathologies. Additionally, in APOE ε4s VLCAs were also negatively correlated with DPs and tau and unlike in APOE ε2 carriers TMAO (and GBB) were each positively correlated with DPs (Figure 4A).

Figure 4. Examination of plasma L-carnitine, TMAO, GBB and acylcarnitines associations with brain pathology in different APOE genotypes.

A Amyloid load was measured in nine brain regions, DPs in four brain regions, NPs in five brain regions, and tau (NFTs) in eight brain regions. Significant correlations are shown between plasma acylcarnitines and AD pathology in different brain regions.The number of brain regions found correlated with each plasma species is indicated in parentheses.

The arrowheads “▲” indicates a positive and “▼” a negative correlation. B The CAA pathology was evaluated in four neocortical regions, for cerebral atherosclerosis large vessels of the circle of Willis (and their proximal branches) were inspected, for arteriolosclerosis small vessels of the basal ganglia were examined. Significant correlations are shown between plasma acylcarnitines and cerebrovascular pathologies. The direction of correlations are shown by arrows (as described above).

Numbers per group: ε2/ε3 n= 20, ε3/ε3 n= 36, ε3/ε4+ε4/ε4 n= 23. Abbreviations; TMAO: trimethylamine-n-oxide, Cx:y-CAR: acylcarnitines, Cx:y-OH-CAR: hydroxy acylcarnitines, SCA: short chain acylcarnitines, MCA: medium chain acylcarnitines, LCA: long chain acylcarnitines, VLCA: very long chain acylcarnitines, DPs: diffuse plaques, NPs: neuritic plaques, NFTs: neurofibrillary tangles, CAA: cerebral amyloid angiopathy. Statistics: correlations analyses were performed using Kendall’s tau-b.

We next interrogated whether APOE genotypes influenced associations between plasma acylcarnitines and cerebrovascular pathology. Among APOE ε2 carriers, several MCAs and LCAs were positively correlated with CAA, and there were negative correlations between MCAs, the LCA C14:0-CAR and arteriolosclerosis. In APOE ε3 homozygotes, different species of VLCAs were negatively correlated with CAA, cerebral atherosclerosis, and arteriolosclerosis. Lastly, in APOE ε4 carriers, TMAO was negatively correlated with CAA, and the LCA C20:3-CAR was positively correlated with cerebral atherosclerosis. Further in APOE ε4 carriers, an OCA was also negatively correlated with arteriolosclerosis while MCAs as well as one LCA were positively correlated with arteriolosclerosis (Figure 4B).

We examined possible independent influences of amyloid, tau and cerebrovascular pathologies on blood profiles while adjusting for their potential confounding effects on each other. It was observed that plasma TMAO to carnitine ratios were independently associated with CERAD scores (p < 0.001) and there was a significant interaction between the presence of CAA and APOE genotypes on these ratios (p = 0.054). Moreover, LCA to total acylcarnitine ratios were independently associated with CAA (p = 0.010) and plasma VLCA ratios were independently associated with Braak stages (p = 0.021) but no significant interactions with APOE genotypes were detected.

Brain GBB, TMAO and acylcarnitine levels are differentially associated with AD and cerebrovascular pathologies in the ROS subset.

In the APOE ε3/ε3 MCI group, MCAs and LCAs ratios to total acylcarnitines within the inferior orbital cortex were lower and SCA ratios were higher compared to other diagnostic groups (Figure 5A and B). Further, SCAs and MCAs ratios were higher among APOE ε2 carrier controls compared to other groups (Figure 5A and B). Lastly, MCAs ratios were higher in APOE ε2 carriers with AD but, unlike controls, this was not accompanied by higher SCA levels (Figure 5A and B). Raw concentrations of multiple LCA species (C16:1-CAR, C18:1-CAR, C18:2-CA, C18:3-CAR, C20:0-CAR, and C20:1-CAR) as well as a VLCA (C22:0-CAR) were significantly lower in the APOE ε3/ε3 MCI group compared to APOE ε3/ε3 controls (all p < 0.05, Supplementary Table 7S). Analysis of overall acylcarnitine raw concentrations also showed that total brain acylcarnitine levels were lower in the APOE ε3/ε3 MCI group compared to all control groups regardless of APOE genotype, as well as APOE ε2 and APOE ε4 AD groups but not APOE ε3/ε3 AD groups (Supplementary Figure 2SC). Similarly, in the APOE ε3/ε3 AD group there was an overall decrease in acylcarnitine concentrations compared to APOE ε2 and APOE ε4 AD groups but not controls (Supplementary Figure 2SC).

Figure 5. Increases in brain SCAs accompanied by decreases in brain LCAs in MCI.

A Heatmaps of analyte ratios most strongly associated with cognitive diagnoses stratified by APOE genotype. B Z-score heatmap showing acylcarnitine chain length ratios with cognitive diagnoses stratified by APOE genotype.

Numbers per group: E2/E3 Control n= 10, E3/E3 Control n=12, E3/E4+E4/E4 Control n= 11, E3/E3 MCI n= 12, E2/E3 AD n= 10, E3/E3 AD n= 10, E3/E3 AD n= 12, E3/E4+E4/E4 AD n=12. Abbreviations; AD: Alzheimer’s disease, MCI: mild cognitive impairment, GBB: γ-butyrobetaine, Cx:y-CAR: acylcarnitines, Cx:y-OH-CAR: hydroxy acylcarnitines, SCA: short chain acylcarnitines, MCA: medium chain acylcarnitines, LCA: long chain acylcarnitines, VLCA: very long chain acylcarnitines. Statistics: For the clustering heatmap data was normalized and scaled before being analyzed with the Ward’s clustering method, top analytes were selected by ANOVA.

When examining the association between brain L-carnitine, GBB, TMAO, acylcarnitine levels and pathology, in APOE ε2 carriers, overall amyloid pathology was negatively correlated with L-carnitine, several OCAs, one MCA and LCAs. Similarly, several SCAs, one MCA and multiple species LCAs were negatively correlated with tau pathology. The negative correlations with OCAs were observed with amyloid pathology in multiple brain regions and were significant for overall amyloid pathology_, NPs, and DPs. In APOE ε3 homozygotes, GBB, SCAs, and LCAs were negatively correlated with several amyloid pathologies and MCA was positively correlated with amyloid pathology. Two SCAs (C2:0-CAR and C5:0-OH-CAR), GBB and TMAO were negatively correlated with tau pathology. In APOEε3 homozygotes, correlations between GBB and amyloid and tau pathologies occurred for multiple brain regions. In APOE ε4 carriers, relatively fewer correlations were seen with amyloid pathologies compared to other genotypes, with OCAs and a LCA negatively correlated with amyloid pathology as well as TMAO positively correlated with DPs in one brain region. Lastly, one OCA, one LCA and one VLCA were negatively correlated with tau pathology (Figure 6A). In APOE ε2 carriers, GBB, and two OCAs (C5:0-CAR and C5:0-OH-CAR) were negatively correlated with CAA while in APOE ε3/ε3s a LCA was positively correlated with cerebral atherosclerosis. There were no correlations between cerebrovascular pathology and brain acylcarnitines in APOE ε4 carriers (Figure 6B).

Figure 6. Examination of inferior orbital cortex L-carnitine, TMAO, GBB, and acylcarnitines associations with brain pathology in different APOE genotypes.

Amyloid load was measured in nine brain regions, DPs in four brain regions, NPs in five brain regions, and tau (NFTs) in eight brain regions. This table shows significant correlations between acylcarnitine levels in the inferior orbital cortex and AD pathology in different brain regions.The number of brain regions found correlated with each acylcarnitine species is indicated in parentheses. The arrowheads “▲” indicates a positive and “▼” a negative correlation. B The CAA pathology was evaluated in four neocortical regions, for cerebral atherosclerosis large vessels of the circle of Willis (and their proximal branches) were inspected, for arteriolosclerosis small vessels of the basal ganglia were examined. This table shows significant correlations between acylcarnitine levels in the inferior orbital cortex and cerebrovascular pathologies. The direction of correlations are shown by arrows (as described above).

Numbers per group: ε2/ε3 n= 20, ε3/ε3 n= 36, ε3/ε4+ε4/ε4 n= 23. Abbreviations; TMAO: trimethylamine-n-oxide, GBB: γ-butyrobetaine, Cx:y-CAR: acylcarnitines, Cx:y-OH-CAR: hydroxy acylcarnitines, SCA: short chain acylcarnitines, MCA: medium chain acylcarnitines, LCA: long chain acylcarnitines, VLCA: very long chain acylcarnitines, DPs: diffuse plaques, NPs: neuritic plaques, NFTs: neurofibrillary tangles, CAA: cerebral amyloid angiopathy. Statistics: correlations analyses were performed using Kendall’s tau-b.

When examining the influences of AD and cerebrovascular pathologies on possible association between APOE genotype and brain L-carnitine and its metabolites as well as acylcarnitine profiles, significant associations were observed. For brain TMAO to carnitine ratios, there was an independent influence of APOE genotypes (p =0.042) and arteriosclerosis (p = 0.047), as well as an interaction between APOE genotypes and the presence of CAA (p =0.039). For SCA ratios, there was a significant influence of APOE genotype and Braak stages (p = 0.023) and between APOE carrier status and cerebral atherosclerosis (p =0.018). Similarly, for LCA ratios, there was an interaction between APOE genotype and Braak stage (p = 0.027) as well as between APOE genotype and cerebral atherosclerosis (p = 0.010). A summary of the results can be found in Figure 7.

Figure 7. Summary of findings.

Aging-related bioenergetics disturbances, such as brain glucose hypometabolism, are exacerbated by the APOE ε4 allele and AD. These bioenergetics disturbances lead to alterations in the L-carnitine system involved in FAO. Here, we found a pattern of changes in blood acylcarnitine ratios with AD clinical progression which was further mediated by the APOE ε4 allele and female sex. Specifically, decreases in L-carnitine, OCA, LCA, VLCA and increases in SCAs and TMAO ratios (to total acylcarnitines) in the periphery. These suggest changes in fuel use with disturbances in FAO in cognitively impaired individuals, particularly APOE ε4 carriers and females. Plasma SCA levels were also positively correlated with blood glucose levels, indicating a possible link between glucose impairments and FAO dysfunction. Plasma TMAO and acylcarnitine levels were also predictive of cerebrovascular pathology, whereas brain GBB and acylcarnitine levels were correlated with amyloid and tau pathologies in an APOE genotype-dependent manner. Positive correlations are indicated in red and negative correlations in blue in the summary of results tables.

Abbreviations; AD: Alzheimer’s Disease, TMAO: trimethylamine-n-oxide, GBB: γ-butyrobetaine, C2:0-CoA: acetyl-CoA, SCFA-CoA: short chain fatty acyl-CoA, SCA: short chain acylcarnitines, OCFA-CoA: odd chain fatty acyl-CoA, OCA: odd chain acylcarnitines, MCFA-CoA: medium chain fatty acyl-CoA, MCA: medium chain acylcarnitines, LCFA-CoA: long chain fatty acyl-CoA, LCA: long chain acylcarnitines, VLCFA-CoA: very long chain fatty acyl-CoA, VLCA: very long chain acylcarnitines, FAO: fatty acid oxidation, AA: amino acid, CAA: cerebral amyloid angiopathy, AtSc: Arteriolosclerosis, CA: cerebral atherosclerosis, Aβ: amyloid beta.

Discussion

The central role of the L-carnitine system in FAO is well established and blood levels of these metabolites are known to be altered in metabolic disorders [18,19]. Given the fact that the ApoE protein plays a major role in lipid homeostasis and is a risk factor for metabolic disorders and AD [13], we examined L-carnitine, its metabolites GBB and TMAO, as well as acylcarnitines over the progression of AD symptoms and pathologies within different APOE genotypes to gain insight into altered bioenergetic pathways with aging, AD and APOE genotypes. In the context of the current knowledge about the role of APOE genotypes on brain and blood markers of bioenergetics, our data suggest that blood acylcarnitines reflect alterations in the catabolism of FAs and amino acids in aging and AD which are influenced by APOE genotypes.

Data from the combined datasets showed that plasma L-carnitine, GBB, TMAO, and acylcarnitine ratios in the APOE ε4− pre-clinical MCI/AD + MCI group were similar to those of controls, while profiles of the APOE ε4+ pre-clinical MCI/AD + MCI group were similar to AD groups, suggesting that peripheral profiles of these metabolites were altered earlier in the disease process among APOE ε4s compared to non-ε4s. Since cognitively healthy APOE ε4 carriers exhibit brain glucose hypometabolism prior to presenting with clinical symptoms of AD [8], changes in blood TMAO, L-carnitine, GBB and acylcarnitine levels in APOE ε4s could be indicative of compensatory efforts to supplement energy needs in the brain.

In the APOE ε4+ AD group, plasma SCA ratios to total acylcarnitines were higher whereas LCA ratios were lower in the ADAPT and the RNC combined datasets analyzed in this study. This is in agreement with previously reported decreases in plasma MCAs and LCAs in AD [45]. In the ROS subset, control APOE ε2 carriers also had a trend towards higher plasma MCA and LCA ratios. Peripheral medium chain fatty acids and MCAs from the liver could be efficient ketone precursors since they can directly enter the mitochondria without using the L-carnitine shuttle yielding acetyl-CoA for ketone synthesis, the main alternative fuel to glucose [4, 43]. Alternatively, these MCAs could be a product of incomplete mitochondrial FAO of LCFAs or the peroxisomal FAO of VLCFAs [22].

The VLCA, C26:0-CAR, was lower peripherally in APOE ε4 carriers with pre-clinical MCI/AD or MCI, and also in AD patients in the combined datasets with similar trends in the AD groups in the ROS subset. Since VLCFAs are shown to be higher in AD brains [48], changes in VLCAs could be reflective of altered peroxisomal metabolism of VLCFAs in AD. In women, VLCAs were also consistently lower in plasma in all datasets studied. Therefore, abnormal peroxisomal metabolism of VLFAs could be a contributing factor to the increased risk of AD observed in women.

Increases in blood SCA ratios were associated with AD in both APOE ε4 and non-ε4 carriers in the combined datasets. Similarly, this was also noted in the ROS subset where plasma SCAs were elevated among APOE ε2-carriers, APOE ε3 homozygous and APOE ε4 carriers with AD. The SCA levels were also positively correlated with blood glucose, suggesting a link between disturbances in glucose metabolism and FAO in AD. Interestingly, APOE ε4 controls also had elevated plasma SCA ratios whereas APOE ε2 and APOE ε3 controls and APOE ε3 MCI (made up of ε3 homozygotes) had relatively lower plasma SCA ratios. Hence, despite the differences in age by nearly a decade between these clinical and the ROS groups, cognitively healthy APOE ε4 carriers as well as AD cases present with similar SCA profiles, with similar trends in brain ratios in APOE ε3/ε3 MCI. However, reverse trends were seen in the brain for AD cases with lower SCA and higher LCA ratios. Although a direct correlation of plasma and brain was not possible given the time differences in plasma and brain tissue collection, this may point to an inverse association and perhaps indicate a defective transport process for SCA into the brain with the onset of clinical symptoms. This may also reflect a failure of the FAO pathway in the brain which could lead to an accumulation of long chain FAO intermediates (i.e., LCAs) and decreases in their end products (i.e., SCAs). In APOE ε4 controls, brain and blood SCA and LCA profiles already resembled that of AD cases, suggesting that SCA and LCA changes may correspond with an imminent onset of AD. Clearly, this warrants further investigation and strategies for restoring brain SCA and LCA balance could be helpful in improving brain bioenergetics. Acetyl-carnitine has been suggested to exert a beneficial role in brain health by increasing mitochondrial acyl-CoA levels, enhancing acetylcholine synthesis, stimulating protein and phospholipid synthesis [49]. It is also able to cross the BBB and very well tolerated by patients [49]. A recent review on the use of acetyl-carnitine in improving cognition in MCI/AD which included 19 trials found that overall trials point to an improvement in cognition [49]. However, none of these trials examined the interaction of APOE genotype and treatment effects.

In the current study, the OCA C5:0-CAR concentrations were elevated in the blood of cognitively healthy APOE ε4+ controls in the combined datasets with a similar trend observed in the ROS subset. Others have found serum levels of C5:0-CAR to be significantly lower in APOE ε4 carriers with AD compared to APOE ε4 carriers with MCI (47) and conversely greater levels of plasma C5:0-CAR in older adults with superior memory performance compared to controls [51]. Here, plasma C5:0-CAR was also higher in APOE ε2 carriers with AD (which had the lowest pathological burden). Short OCAs such as C5:0-CAR can come from amino acid catabolism, which has been proposed as a potential alternative fuel in aging (49), as such changes in blood OCA levels could be reflective of altered amino acid metabolism over the course of AD. There was also a trend for lower brain concentrations of the OCA C3:0-CAR with AD in all APOE genotypes. Decreases in brain C3:0-CAR have also been previously described in the brain in AD in at least two different studies [53,54].

Changes in L-carnitine and its metabolites were also detected with AD progression. Specifically, higher TMAO was accompanied by lower L-carnitine and GBB, this inverse association continued to persist among AD patients in more severe stages of AD. While the exact mechanism of this remains to be investigated, collectively these findings suggest an increased gut degradation of trimethylamines (such as GBB and L-carnitine) followed by their subsequent liver conversion to TMAO with increasing disease severity. Lower plasma L-carnitine has previously been shown to predict cognitive decline in healthy middle-aged individuals [28]; while TMAO has not been measured in the blood over the continuum of AD along with APOE genotype, studies have suggested that TMAO can cause brain aging and neurodegeneration, increase inflammation, potentiate amyloid and tau aggregation and contribute to mitochondrial dysfunction [29,55]. Further, TMAO has also been associated with the development of insulin resistance, which is a risk factor for AD [55]. Of note, there were no significant differences in plasma levels of these metabolites in the samples from the ROS study. This could be because in the ROS subset individuals were approximately 9 years older than in the other datasets. However, we observed that cerebrovascular pathology significantly affected blood levels of TMAO in the ROS subset, since this subset is composed of older participants, they are more likely to have a greater pathological burden which could also have altered TMAO levels. The ROS subset is also constituted of priests, brothers, and nuns whereas the combined datasets is composed of community-dwelling individuals; therefore, lifestyle differences, including differences in diets, may have also been responsible for these discrepancies. Relatively few studies have examined L-carnitine, its metabolites (GBB and TMAO) as well as acylcarnitines in the brain in AD with APOE genotype. A study examining plasma and CSF L-carnitine and acylcarnitine found no difference between AD and controls, however, in that study total acylcarnitine levels were measured, and therefore any species or chain length specific effects may have been obfuscated and further APOE genotype was not examined [56]. Another study found lower L-carnitine CSF levels in non-ε4s with mild AD [57]. In yet another study, L-carnitine and total acylcarnitines declined with age in the superior frontal gyrus of individuals with AD but not in controls [58] indicating a possible age/disease interaction on these lipid profiles in AD.

When examining the association between blood and brain L-carnitine, GBB, TMAO and acylcarnitines with brain pathologies, APOE genotypes modulated the relationships between these lipid levels and brain pathologies, suggesting differential associations of amyloid pathology and these lipids in the presence of APOE ε2 or APOE ε4 genotypes. Few studies have investigated correlations between peripheral acylcarnitines and AD pathology although one found serum LCAs to be positively correlated with Aβ_1–42 only in cognitively healthy individuals and serum MCAs and LCAs were positively correlated with total tau [59]. Brain TMAO to L-carnitine ratios and acylcarnitine ratios were also independently associated with APOE genotypes and with arteriolosclerosis and both brain SCA and LCA ratios were associated with Braak stage and interacted with APOE genotypes and with cerebral atherosclerosis also in interaction with APOE genotypes. This suggests a modulation of the L-carnitine system by cerebrovascular pathology irrespective of amyloid and tau pathologies which is further mediated by APOE genotypes.

Sex effects were also observed in plasma across all groups, specifically a significant decrease in raw concentrations of GBB, C3:0-CAR and C26:0-CAR in women relative to men, while no effect of sex was observed in the brain. Since decreases in GBB, OCAs and VLCAs were also associated with AD clinical progression and pathology, therefore, women have profiles more similar than men to that observed in AD regardless of diagnosis. Lower plasma levels of L-carnitine and acylcarnitines have previously been described in women [60]. In young individuals no sex differences in FAO are observed between men and women, however with age there is a decrease in resting energy expenditure in both sexes and specifically a decrease in resting fat oxidation in men but not in women [58, 59]. The specific mechanisms behind these sex differences remain to be elucidated but suggest that in old age there are sex differences in FAO which could impact AD risk.

Study limitations

Some of the considerations and limitations of the studies presented here are that blood acylcarnitine levels come from multi-organ contributions including the heart, muscle and liver tissues and are known to fluctuate depending on fasted or fed states, adding to both intra- and inter-subject variability as well as being affected by metabolic disorders such as diabetes [60, 61]. Nevertheless, we anticipate that the within subject normalization process used here should help limit such variability. Additionally, while clinical study samples were non-fasting, in the ROS samples, fasting status did not differ by diagnostic groups. Further, due to the cross-sectional design of these studies, and as changes in acylcarnitine levels reflect current metabolic states, these cannot be expected to remain static, as such fluctuating levels of these analytes may obfuscate data. Although we examined acylcarnitine classes in relation to total acylcarnitines to account for possible differences in absolute acylcarnitine levels, further longitudinal studies of the L-carnitine system are needed. Diet is also known to have an important impact on blood L-carnitine, GBB and TMAO levels [65] which may explain some of the variations seen between our community-based datasets and the ROS subset. It will therefore be important to obtain diet information for future studies examining these metabolites. However, the ratios examined herein are of added value to this study since they can partially normalize for inter-individual differences in acylcarnitine concentrations, thereby allowing us to, in addition to the investigation of absolute levels, to examine changes in the proportion of acylcarnitines in relation to each other. Acylcarnitine ratios have also experimentally been demonstrated to be more sensitive to detecting FAO related disorders than raw concentration values [66].

Since these study populations were largely composed of White individuals, generalization of these findings to other races is unknown. Despite the fact that African-Americans and Hispanics are disadvantaged by having a significantly worse risk of AD compared to Whites, the APOE ε4 risk of AD seem to be weaker in African Americans compared to Whites [67,68] and lead to a lower age of onset for dementia [69] irrespective of race [70]. Additionally, among non-ε4 carriers, risk of AD is 4 times higher in African Americans and 2 times higher in Hispanics compared to Whites [69]. Interestingly, African Americans women with diabetes display alterations in acylcarnitine profiles when diagnosed with type 2 diabetes [71]. This suggest that among other racial groups, association between acylcarnitines and metabolic syndromes may be mediated by other genetic and/or environmental factors, which clearly warrants further investigation.

Another limitation of our study is that the brain homogenate is a combination of multiple cell types; therefore, specific contributions of different cell types cannot be clarified. Further in vitro studies are needed to elucidate this. Nevertheless, acylcarnitine plasma levels have been associated with AD in multiple studies suggesting a dysregulation of FAO in AD warranting further investigation [45,50,51,59,72–75]. Our study builds on this previous research by offering a comprehensive investigation of L-carnitine, its metabolites (GBB and TMAO), and acylcarnitines in blood and brain tissue with AD clinical-pathological data in different APOE genotypes. Future longitudinal studies examining the relationship between blood acylcarnitines and AD brain biomarkers through imaging would be useful to correlate these changes with glucose hypometabolism and amyloid and tau imaging as well as exploring the association of cerebrovascular pathologies with FAO defects. Further, studies on the influence of diet and perhaps how dietary intervention, in different APOE genotypes and sex, can influence the L-carnitine system within the context of AD would also be useful.

Summary

In summary, this study of acylcarnitines suggests that lower LCA ratios in the brain and plasma with concurrently higher SCA ratios are associated with clinical disease progression of AD, amyloid, tau, and cerebrovascular pathologies. Our results implicate dysfunctional peripheral and brain L-carnitine metabolism and oxidation of LCFAs, as well as impairments in amino acid and VLCFA metabolism with disease progression amyloid, tau, and cerebrovascular pathologies. There is also a dose-response between APOE genotypes and brain acylcarnitine correlations with pathology, with more correlations in APOE ε2 > APOE ε3 > APOE ε4 which also coincides with pathological severity.

Research in context.

Evidence before this study

Brain bioenergetic disturbances are considered important aspects of AD, this is supported by the findings of disturbances in brain glucose metabolism and FAO, particularly in the presence of the APOE ε4 allele. The L-carnitine system is involved in FAO and is known to be dysregulated in metabolic disorders. L-carnitine and its metabolites (GBB and TMAO) are also linked to cardiovascular and cerebrovascular dysfunction, suggesting that these metabolites may be related to both metabolic and vascular dysfunctions observed in AD. However, disturbances in the L-carnitine system with clinical and pathological features of AD and the influence of APOE genotypes on this system are still incompletely understood.

Added value of this study

This study builds on previous research by providing a comprehensive profiling of L-carnitine, GBB, TMAO and acylcarnitines in samples from individuals at various clinical and pathological stages of AD with different APOE genotypes. Studying the L-carnitine system could be informative in expanding our understanding of the link between different APOE genotypes and AD risk as it relates to bioenergetic disturbances.

Implications of all the available evidence

Altogether these results suggest that the L-carnitine system reflects metabolic disturbances associated with APOE ε4 and could play a role in AD clinical and pathological progression. Further research is needed to understand how disturbances in this system with APOE ε4 carrier status may increase the vulnerability to AD.

Data sharing and statement

All raw concentration data can be found in Supplementary tables 5S–7S. The authors declare that the data supporting the findings of this study are available from the corresponding author upon request. Additionally, ROS resources can be requested at https://www.radc.rush.edu.