Study populations
The current study was performed in cohorts participating in the Alzheimer’s Disease Apolipoprotein Pathology for Treatment Elucidation and Development (ADAPTED) consortium including the Barcelona-based memory clinic Fundació ACE (142 CSF-plasma paired samples) and the Department of Geriatric Psychiatry at the Medical Faculty Mannheim, University of Heidelberg (40 CSF samples). Both participating studies are approved by the medical ethical committee of their respective institutes and informed consents were collected from all participants, which allow the use of phenotype and biomarker information for research purpose. From both participating cohorts, we selected MCI patients for which complete information was available on age at blood collection, sex, body mass index (BMI), and lipid-lowering medication use, as well as AD biomarkers in CSF (i.e., Aβ-42, p-tau, and total tau).
Fundació ACE cohort
All the MCI patients from the Fundació ACE (ACE) cohort were recruited and assessed between 2016 to 2017 at the Memory Disorders Unit from Fundació ACE, Institut Català de Neurociènces Aplicades, Barcelona, Spain [27]. Each patient was assigned a diagnosis after consensus at a case conference attended by neurologists, neuropsychologists, and social workers. MCI patients fulfilled MCI Petersen’s diagnostic criteria [28, 29] including subjective memory complaints, decline from normal general cognition, preserved performance in activities of daily living, absence of dementia, and a measurable impairment in one or more cognitive functions, with or without deficit in other cognitive domains (amnestic MCI: single domain or amnestic MCI: multiple domain). At follow-up, dementia was defined according to the DSM-V criteria [30]. The underlying etiologies of the cognitive deficits within the dementia group were classified according to the following criteria: the 2011 National Institute of Aging-Alzheimer’s Association (NIA-AA) [31] for Alzheimer’s disease and the National Institute of Neurological Disorder and Stroke and Association Internationale pour la Recherche et l’Enseignement in Neurosciences criteria(NINDS-AIREN) [32] for vascular dementia, frontotemporal dementia [33], and Lewy body dementia [34].
Paired samples of CSF and plasma were collected from patients under fasted conditions. CSF was obtained by lumbar puncture following the established consensus recommendations [35]. Briefly, the lumbar puncture (LP) was performed by an experienced neurologist with the patients in a sitting position. After local anesthesia (1% mepivacaine) was injected subcutaneously, CSF was obtained by LP in the intervertebral space of L3-L4. The fluid was collected passively in two 10-ml polypropylene tubes (Sarstedt ref. 62610018). The first tube of CSF was analyzed for basic biochemistry (glucose, total proteins, proteinogram, and cell type and cell number). The second tube was centrifuged (2000×g 10 min at 4 °C), and the fluid was aliquoted into polypropylene tubes (Sarstedt ref. 72694007) and stored at − 80 °C until analysis. The time delay between CSF collection and storage was less than 2 h. On the same day as the AD biomarker analysis (Aβ-42, p-tau, and total tau), an aliquot was thawed at room temperature and vortexed for 5–10 s. CSF Aβ1–42, total tau, and p-tau levels were measured using commercially available enzyme-linked immunosorbent assays, namely Innotest Aβ1–42, Innotest hTAU Ag, and Innotest PHOSPHO-TAU (181P) (Innotest, Fujirebio Europe) [35, 36].
For APOE genotyping in the ACE cohort, genomic DNA was obtained from whole blood collected in BD Vacutainer tubes (K2-EDTA). DNA extraction was performed using DNA Chemagen technology (Perkin Elmer). Afterward, the APOE genotype was determined by TaqMan probes analysis in a system of Real-Time PCR QuantStudio3 (Thermofisher).
Heidelberg/Mannheim memory clinic sample
Forty MCI patients were recruited and assessed between 2012 to 2016 at the Memory Clinic of the Central Institute of Mental Health (Mannheim, Germany). Neuropsychiatric or general medical causes of impaired cognition were excluded by detailed medical history, physical and neuropsychiatric examination, and standard serum laboratory assessment. Thus, all MCI patients met the MCI Petersen’s diagnostic criteria [28, 29] including subjective memory complaints, normal general cognition, only minimally impaired performance in instrumental activities of daily living, absence of dementia, and a measurable impairment in one or more cognitive domains. Cognitive impairment was defined as performance below 1.2 standard deviation in one or more cognitive domains in standard neuropsychological test battery [37] (test battery of the Consortium to Establish a Registry for Alzheimer Disease (CERAD) [38] plus the Wechsler memory scale – logical memory (WMS) immediate and delayed recall [39] and the trail making test A (TMT-A) and B (TMT-B) [40]. For biomarker assessments, lumbar puncture was performed to determine amyloid pathology in CSF following the NIA/AA criteria for the diagnosis of MCI due to AD [41]. The results of the clinical assessment for each patient were discussed at a case conference attended by geriatric psychiatrists and neuropsychologists. The diagnosis of MCI due to AD or prodromal AD [42] was assigned by consensus using all clinical and biomarker data (CSF Aβ-42, t-tau, and p-tau). Paired samples of CSF and plasma were collected from patients according to the established consensus recommendations [35]. Aliquots were stored in polypropylene tubes at − 80 °C. Aβ1–42, p-tau, and t-tau were performed in the Neurochemistry Laboratory at the Department of Neurology, University Medical School, Göttingen, using established protocols. P-tau levels in CSF were quantified with a commercially available ELISA kit [INNOTEST® PHOSPHO-TAU(181P), Innogenetics]. Aβ1–42 was detected with a commercially available ELISA kit [INNOTEST®β- AMYLOID (1–42) Innogenetics] for quantitative analysis.
APOE genotyping in Heidelberg/Mannheim memory clinic sample was performed on an Illumina GSA1.0 SharedCustom Content bead array according to the manufacturer’s instructions. GenomeStudio 2.0 software was used to determine APOE genotypes and results were exported in PLINK format.
Metabolomics profiling
All CSF and plasma samples of both cohorts were profiled for the same set of metabolites using a UHPLC-MS/MS approach targeting signaling lipid mediators including LPAs, alkyl-lysophosphatidic acid (aLPAs), and cyclic-lysophosphatidic acids (cLPAs) ranging from C14 to C22 acyl chain length [43].
Samples were stored at − 80 °C, thawed at room temperature, and randomized prior to analysis. Quality control (QC) samples, consisting of a pool of all samples, and blanks were also analyzed to ensure the quality of the obtained data. For CSF samples, 350 μL of samples were evaporated to dryness, spiked with isotopically labeled internal standards and antioxidant (BHT:EDTA 1:1, 0.2 mg/mL), and reconstituted in two aliquots using a mixture of methanol to water (70:30, v/v). Plasma samples were first acidified through the addition of 0.2 M citric acid and 0.1 M disodium hydrogen phosphate buffer at pH 4.5. Metabolites were extracted using liquid-liquid extraction with a mixture of 1-butanol:ethyl acetate (1:1, v/v) prior to mixing, centrifugation, collection of the supernatant, evaporation, and reconstitution into two aliquots with a mixture of ice-cold methanol to water (70:30, v/v).
Samples were measured using a Shimadzu LC-30 AD system coupled to a LCMS-8050 Triple Quadrupole system (Shimadzu, Japan).
For both plasma and CSF samples, the first aliquot (high pH injection) was analyzed using a Kromasil EternityXT-1.8 C18 column, 2.1 × 50 mm, 1.8 μm (Akzo Nobel, Netherlands) with a mobile phase composed of (A) water with 5 mM ammonium acetate and 0.0625% ammonium hydroxide and (B) 80% acetonitrile with 20% isopropanol and 0.1% ammonium hydroxide. For both matrices, the second aliquot (low pH injection) was analyzed using an Acquity BEH C18 column, 2.1 × 50 mm, 1.7 μm (Waters) with a mobile phase composed of (A) water with 0.1% acetic acid, (B) 75% acetonitrile with 25% methanol and 0.1% acetic acid, and (C) 100% isopropanol. For both pH injections, polarity switching and dynamic multiple reaction monitoring (dMRM) mode were used for MS acquisition.
To perform the QC, metabolites showing a relative standard deviation (RSD) higher than 30% on corrected peak areas in QC samples were excluded. After QC correction, 19 and 17 LPAs in CSF and plasma, respectively, were used for further data analysis (Supplementary Table 1). Common metabolites detected in both CSF and plasma included LPAs (C14:0, C14:1, C14:2, C16:0, C18:0, C18:1, C18:2, C20:1, C20:3, C20:5, C22:4, C22:5) and three cyclic-LPAs (C16:0, C18:0, C18:1). Metabolites only detected in CSF samples included some LPAs (C20:4, C22:6, C22:5) and an alkyl-LPA C16:1. LPA C18:3 and two cyclic-LPAs (C18:2, C20:4) were detected only in plasma samples. The inverse rank transformation was used to normalize the distribution of metabolites in both cohorts.
Statistical analysis
Association of LPAs with Aβ-42, p-tau, and t-tau
We performed linear regression analysis to test the association of Aβ-42, p-tau, and t-tau with the profiled metabolites in paired CSF and plasma samples from the ACE cohort and CSF samples from Heidelberg-Manheim memory clinic. Levels of Aβ-42, p-tau, and t-tau in CSF were used as the outcome variable in the regression model. Analyses were adjusted for age, sex, body mass index (BMI), and lipid-lowering medications. The inverse rank transformation was applied to normalize the distribution of both CSF AD biomarkers (Aβ-42, p-tau, and t-tau) and LPA metabolite levels in CSF and plasma. A meta-analysis of the regression analysis results of the two cohorts was performed using METAL software [44] using the inverse-variance fixed-effect model. Meta-analysis results of associations were also corrected for multiple testing separately for each AD biomarker using false discovery rate (FDR) by the Benjamini and Hochberg method [45] and findings with FDR < 0.05 were considered significant in the overall analysis. All analyses were performed in R (https://www.r-project.org/). To test whether conversion from LPA to another was relevant, we have tested all ratios between LPAs.
APOE-stratified regression analysis
To identify APOE-specific associations of metabolites with AD biomarkers, APOE-stratified analyses were performed in both participating cohorts based on three APOE strata including APOE 44/34/24, APOE 33, and APOE 22/23. In the stratified analyses, subjects with APOE 24 genotype were pooled with patients having APOE 44/34 genotypes based on their similar risk profiles, as reported in an earlier study [46]. APOE-stratified analyses results were reported as a combined meta-analysis of both datasets included in the current study (ACE cohort and Heidelberg/Manheim cohort). Due to the smaller number of APOE 22/23 carriers in the two datasets, a combined regression analysis was performed, aggregating all APOE 22/23 carriers from two cohorts while adjusting for cohort effects. Multiple testing correction was performed using the false-discovery rate (FDR < 0.05) based on Benjamin and Hochberg [45].
To assess the association of the APOE genotype with LPAs, we compared levels of LPAs in CSF of APOE ε4 (APOE 44/34/24) and APOE ε2 (22/23) carriers versus APOE ε33 carriers using regression analysis adjusting for the age, sex, BMI, and lipid-lowering medications. This regression analysis was conducted for each cohort and their combined meta-analysis.
MCI to AD dementia progression analysis
In the ACE cohort, follow-up information was available for 138 out of 142 MCI patients including 17 non-amnestic and 121 amnestic MCIs. A total of 43 MCI patients progressed into AD dementia (31%) during follow-up, while 95 MCI patients did not convert to AD dementia. The mean follow-up time in converters was 1.42 years (SD = 0.53) and 1.44 years (SD = 0.70) in non-converters. The rate of MCI to AD dementia progression in our sample is similar to other clinic-based studies [47]. We analyzed the association of LPAs with MCI to AD dementia progression using the cox proportional hazard model adjusting for age at blood collection, sex, BMI, and lipid-lowering medication use. In the ACE cohort, 11 MCI patients also progressed to other types of dementia including vascular dementia (n = 6), semantic dementia (n = 1), Parkinson dementia (n = 1), Lewy Body dementia (n = 2), and frontal temporal dementia (n = 1). We repeated the conversion analysis in the Heidelberg/Mannheim cohort. Among the 40 MCIs, 23 converted to AD dementia. The mean follow-up time in the Heidelberg/Mannheim cohort was 1.80 years (SD = 1.06). Three MCI patients also progressed to frontal temporal dementia in this sample.
Association of cognitive measures with LPA levels
We also assessed the association of cognitive measures, MMSE, and CDR with LPAs levels in CSF of both ACE and Heidelberg/Mannheim cohort. We used linear regression analysis adjusted for age, sex, BMI, and lipid-lowering medication. Results were meta-analyzed using METAL software [44] using the inverse-variance fixed-effect model and multiple testing was performed using false discovery rate (FDR) by the Benjamini and Hochberg method [45].