Plasma high‐density lipoprotein cargo is altered in Alzheimer's disease and is associated with regional brain volume

Abstract

Cholesterol levels have been repeatedly linked to Alzheimer's Disease (AD), suggesting that high levels could be detrimental, but this effect is likely attributed to Low‐Density Lipoprotein (LDL) cholesterol. On the other hand, High‐Density Lipoproteins (HDL) cholesterol levels have been associated with reduced brain amyloidosis and improved cognitive function. However, recent findings have suggested that HDL‐functionality, which depends upon the HDL‐cargo proteins associated with HDL, rather than HDL levels, appears to be the key factor, suggesting a quality over quantity status. In this report, we have assessed the HDL‐cargo (Cholesterol, ApoA‐I, ApoA‐II, ApoC‐I, ApoC‐III, ApoD, ApoE, ApoH, ApoJ, CRP, and SAA) in stable healthy control (HC), healthy controls who will convert to MCI/AD (HC‐Conv) and AD patients (AD). Compared to HC we observed an increased cholesterol/ApoA‐I ratio in AD and HC‐Conv, as well as an increased ApoD/ApoA‐I ratio and a decreased ApoA‐II/ApoA‐I ratio in AD. Higher cholesterol/ApoA‐I ratio was also associated with lower cortical grey matter volume and higher ventricular volume, while higher ApoA‐II/ApoA‐I and ApoJ/ApoA‐I ratios were associated with greater cortical grey matter volume (and for ApoA‐II also with greater hippocampal volume) and smaller ventricular volume. Additionally, in a clinical status‐independent manner, the ApoE/ApoA‐I ratio was significantly lower in APOE ε4 carriers and lowest in APOE ε4 homozygous. Together, these data indicate that in AD patients the composition of HDL is altered, which may affect HDL functionality, and such changes are associated with altered regional brain volumetric data.

Remodeling of high‐density lipoprotein (HDL) during disease onset and progression, in which cholesterol and Apolipoprotein D levels are increased while Apolipoprotein A‐II levels are decreased on HDL (left). Additionally, increased cholesterol levels on HDL are associated with increased ventricular volume and decreased grey matter volume, while increased levels of Apolipoprotein A‐II and Apolipoprotein J are associated with increased hippocampal and grey matter volume and reduced ventricular volume (right).

Abbreviations

AD

Alzheimer's Disease

Aβ

amyloid‐β

BBB

blood‐brain barrier

CAA

cerebral amyloid angiopathy

CRP

C‐reactive protein

CSF

cerebrospinal fluid

GM

grey matter

HC

healthy control

HC‐Conv

healthy control converter

HDL

high‐density lipoprotein

HL

hippocampus left

HR

hippocampus right

LD

lipid droplets

LDL

low‐density lipoprotein

LTP

long‐term potentiation

MMSE

mini‐mental state examination

SAA

serum amyloid A

SUV

standardized uptake values

SUVR

standardized uptake value ratio

VLDL

very low‐density lipoproteins

WM

white matter

1. INTRODUCTION

Alzheimer's disease (AD) is a neurodegenerative disease which is characterized by the extracellular deposition of amyloid β (Aβ) in the brain to form amyloid plaques and by the intracellular accumulation of hyperphosphorylated tau filaments (Masters et al., 2015). These combined events eventually are responsible for neuroinflammation, neuronal death, and reduction of brain volume, ultimately leading to the onset of disease symptoms. Several reports have indicated that diet, high cholesterol, high low‐density lipoproteins (LDL) cholesterol, and low high‐density lipoproteins (HDL) cholesterol levels are possible AD risk factors. Low cholesterol levels have been linked to Aβ precursor protein (AβPP) processing through the non‐amyloidogenic pathway (Buxbaum et al., 2001; Fassbender et al., 2001; Kojro et al., 2001; Simons et al., 1998), while high levels of intracellular cholesterol have been linked to an increase in Aβ deposition in the brain (Burns et al., 2003; Refolo et al., 2000; Wahrle et al., 2002). These results were also supported by studies in rodents indicating that high‐fat diets affect brain amyloid levels and brain mass (Levin‐Allerhand et al., 2002; Pedrini et al., 2009; Refolo et al., 2000). Many studies have therefore suggested a protective role of HDL for AD, indicating that higher HDL cholesterol levels have been associated with better cognitive outcomes, higher MMSE, a reduced risk for AD (but not to MCI), and increased brain grey matter and hippocampal volume (Atzmon et al., 2002; Bates et al., 2017; Reitz et al., 2008; Reitz et al., 2010; Ward et al., 2010; Wolf et al., 2004). Furthermore, HDL cholesterol levels have been shown to inversely correlate with brain Aβ deposits (Reed et al., 2014). Conversely, low HDL cholesterol levels were associated with white matter changes, a higher probability of memory deficits, or a greater risk of dementia (Crisby et al., 2010; Singh‐Manoux et al., 2008; Wolf et al., 2004). However, in spite of several studies indicating a protective role for HDL, other reports failed to confirm such an association (den Heijer et al., 2005; van Velsen et al., 2013). Additionally, discrepancies were also reported when assessing the overall levels of HDL cholesterol, with some studies indicating altered levels of HDL cholesterol between controls and AD, while other studies did not (den Heijer et al., 2005; Isbir et al., 2001; Reitz et al., 2004; Xiao et al., 2012; Zhang et al., 2004). These findings may indicate that HDL particles provide actions that go far beyond the classical role of excess cholesterol elimination by the liver. Many of these effects, such as anti‐inflammatory and anti‐oxidant, are mediated through HDL‐protein cargo, which comprises many apolipoproteins (and several other proteins) that are associated with and form the HDL particles, extensively described in several studies (Ronsein & Vaisar, 2019; Shah et al., 2013; Yassine et al., 2014). For example, ApoA‐I, the major constituent of plasma HDL, has shown protective effects in AD by binding to Aβ and its precursor protein (AβPP), reducing Aβ aggregation and Aβ‐induced toxicity (Koldamova et al., 2001; Paula‐Lima et al., 2009). In animal studies, overexpression of ApoA‐I reduced cognitive deficits and cerebral amyloid angiopathy (CAA), despite unaltered Aβ deposition, likely through reduced neuroinflammation (Lewis et al., 2010). Conversely, depletion of ApoA‐I increased CAA and Aβ aggregation with consequent worsening of cognitive function without altering APP processing and Aβ deposition (Lefterov et al., 2010). Altogether, it appears that ApoA‐I modulates Aβ toxicity through mechanisms other than altering APP processing and Aβ deposition. Whilst high levels of ApoA‐I have been associated with a reduced risk of dementia (Saczynski et al., 2007), reduced levels of ApoA‐I have been described in AD and correlate with the severity of the disease (Kawano et al., 1995; Liu et al., 2006; Merched et al., 2000). On the other hand, ApoE is the major constituent of CSF HDL and the ε4 isoform is considered the major risk factor for sporadic AD (Corder et al., 1993; Strittmatter et al., 1993). ApoE ε4 isoform has been associated with increased Aβ production, accumulation and oligomerization (Bales et al., 2009; Hashimoto et al., 2012; Koffie et al., 2012; Ye et al., 2005; Youmans et al., 2012) and with reduced Aβ clearance (Castellano et al., 2011; Cook et al., 2003; Deane et al., 2008; Du et al., 2009). With regard to Aβ degradation mediated by binding to ApoE, ApoE ε3 binds to Aβ more efficiently than ApoE ε4 and astrocytes from ApoE ε4 transgenic mice clear amyloid plaques less efficiently than astrocytes from ApoE ε3 transgenic mice (Simonovitch et al., 2016; Tokuda et al., 2000). Additionally, in an AD mouse models, CAA is induced more prominently by ApoE ε4 compared to ApoE ε3 (Fryer et al., 2005) and Aβ stimulation affected Long Term Potentiation (LTP) more prominently in ApoE ε4 transgenic mice rather than in ApoE ε2 or ApoE ε3 transgenic mice (Trommer et al., 2005). ApoE ε4 has also been strongly associated with increased atrophy and reduced hippocampal volume (Agosta et al., 2009; Hostage et al., 2013; Manning et al., 2014; Moffat et al., 2000; Tang et al., 2015). Finally, since low levels of ApoE and ApoE ε4 levels were found significantly reduced in AD patients (Beffert et al., 1999; Gupta et al., 2011), it is fitting that lower plasma ApoE levels are linked with the smaller hippocampus (Teng et al., 2015), while higher plasma ApoE levels are linked with reduced brain amyloidosis (Koch et al., 2018). ApoA‐II is capable to form complexes with ApoE ε2 and ApoE ε3 but not with ApoE ε4 and these complexes (ApoE ε2/ApoA‐II and ApoE ε3/ApoA‐II) increased cell viability by binding to Aβ and reducing its internalization. Such protective features were absent in ApoE ε4‐treated cells because of the absence of ApoEε4/ApoA‐II complexes (Yamauchi et al., 2000). An AD transgenic mouse model lacking ApoA‐IV displayed increased cerebral Aβ and learning deficits, likely through mechanisms that are associated with Aβ clearance (Cui et al., 2011), while the absence of ApoD was associated with an increased number of amyloid plaque in the hippocampus and, at the opposite, overexpression of ApoD was associated with a reduced number of them (Li et al., 2015). ApoJ forms complexes with Aβ (ApoJ:Aβ) for their interaction with low‐density lipoprotein receptor‐related protein 2 (Megalin), which mediates the transport and the clearance of Aβ from and to the brain (Bell et al., 2007; Hammad et al., 1997; Zlokovic et al., 1994; Zlokovic et al., 1996). Levels of ApoJ are increased in plasma and brain regions of AD patients, specifically in areas that are loaded with amyloid plaques (Bertrand et al., 1995; Gupta et al., 2016; Gupta et al., 2017; Howlett et al., 2013; Lidstrom et al., 1998; May et al., 1990; Miners et al., 2017). However, it is important to point out that the composition of plasma HDL is different from CSF HDL as not all apolipoproteins associated with HDL are produced by brain‐resident cells. In the CSF ApoE and ApoJ are the main apolipoproteins, while ApoA‐I, the main constituent of plasma HDL is not produced in the central nervous system. However, damage to the blood‐brain barrier (BBB), common in AD, allows ApoC‐III, not usually produced in the brain, to be detected in the CSF (Picard et al., 2022), while ApoA‐I can be transported across the BBB at the choroid plexus (Stukas et al., 2014). In general, lipid dysregulation in the brain affects AD like many other neurodegenerative diseases. In this regard, lipid droplets (LD) which are cytoplasmatic organelles containing cholesteryl esters which provide energy for cell metabolism and membrane synthesis (Walther & Farese Jr., 2012) were first described by Alois Alzheimer (Alzheimer et al., 1995). In AD, lipid droplet accumulation is an event that takes place before amyloid accumulation, emphasizing the importance of lipid metabolism in the disease (Hamilton et al., 2015). The importance of LD in AD and many other neurodegenerative diseases was recently assessed (Farmer et al., 2020; Ralhan et al., 2021).

Modulation of HDL cargo has been observed and reported in many diseases, such as Coronary Heart Disease, Acute Coronary Syndrome, End Stage Renal Disease and Liver Disease (Alwaili et al., 2012; Trieb et al., 2016; Vaisar et al., 2007; Weichhart et al., 2012; Yan et al., 2014). Aging has also been reported to be a factor in modulating HDL protein cargo (Holzer et al., 2013). In light of these tight connections between HDL‐protein cargo and AD and the fact that modulation of HDL protein cargo is a common factor in many other diseases, our goal in this study is to determine if HDL‐protein cargo is altered in AD and if specific changes can be directly associated with the extent of brain amyloidosis. If so, alteration of HDL‐protein cargo could serve as a direct measurement of amyloid deposits in the brain, providing a diagnostic tool that could indicate healthy controls at risk for AD.

2. MATERIAL AND METHODS

2.1. Participants

The AIBL study was conducted in accordance with the Declaration of Helsinki and was approved by the ethics committees of St. Vincent's Health and Austin Health in Melbourne and Hollywood Private Hospital and Edith Cowan University in Perth (Australia; Protocol number HPH215). All volunteers gave written and informed consent before participating in our study. A total of 213 participants were divided into stable healthy controls (HC, n = 87, HC for at least 36 months (t1 = 18 months and t2 = 36 months after HDL‐cargo analysis (t0))), healthy controls converters (HC‐Conv, n = 38, HC at the time of HDL‐cargo analysis (t0) but converted to MCI/AD within the following 36 months (either at t1 or t2)) and Alzheimer's patients (AD, n = 88, from t0 through t2) from the AIBL cohort were used in this study (Figure S1). Exclusion criteria included a history of non‐AD dementia, schizophrenia, bipolar disorder, current depression (GDS score above 5/15), Parkinson's disease, uncontrolled hypertension (systolic BP > 170 or diastolic BP > 100), cancer (other than basal cell skin carcinoma) within the last two years, symptomatic stroke, uncontrolled diabetes, or current regular alcohol use exceeding two standard drinks per day for women or four per day for men (Ellis et al., 2009). The AIBL Study clinical panel meets on a monthly basis to discuss baseline classification for each new patient and to ensure that diagnoses were made in accordance with the NINCDS‐ARDA criteria (McKhann et al., 1984; Winblad et al., 2004). Body parameters such as weight, height, blood pressure, and pulse rate were evaluated during the examinations. Blood was drawn from overnight fasting participants and collected to obtain plasma for our analysis. APOE status was determined by genotyping cells from whole blood as previously described (Gupta et al., 2011). Total cholesterol, LDL, HDL, and TG levels in plasma were also assessed.

2.2. HDL fractionation

HDL fraction was obtained by precipitating VLDL and LDL fractions using a solution identical to the one reported in the Quantolip® protocol (Reagent A: Na₂HPO₄x2H₂O 8 g/L, NaH₂PO₄xH₂O 11 g/L, Na₂EDTA 1 g/L, Sodium Azide 0.9 g/L, Polyethylenglycol 20,000 95 g/L). Briefly, 30 μl of plasma were mixed with 60 μl of Reagent A and sat for 10 min at RT, followed by 15 min of centrifugation at 2500 g. The supernatant containing the HDL fraction was then collected, aliquoted, and stored at −80°C until analysis. HDL isolation by Polyethylenglycol precipitation has already been used in the past (Kostner et al., 1985).

2.3. HDL and HDL‐associated protein measurement in HDL fractions

HDL‐cholesterol in HDL fractions (dilution 1:75) was assessed using HDL‐Cholesterol Assay Kit (Cell Biolabs) using the manufacturer's instructions.

HDL‐cargo proteins in HDL fractions were assessed by Bioplex analysis. Serum Amyloid A (SAA) levels associated with HDL fractions (dilution 1:200) were evaluated using SAA bioplex kit (Merck Millipore) using the manufacturer's instructions. A broad panel of Apolipoprotein (ApoA‐I, ApoA‐II, ApoC‐I, ApoC‐III, ApoD, ApoE, ApoH, ApoJ, and CRP) levels associated with HDL fractions (dilution 1:10,000) were evaluated using Pro Human Apolipoprotein Panel bioplex kit (Biorad) using the manufacturer's instructions (Jóźwiak et al., 2022; Mlambo et al., 2020).

2.4. Calculations of HDL and HDL‐associated proteins

For references and calculation, in this paragraph, we will refer to cholesterol‐HDL_PL as the cholesterol concentration on HDL in plasma (PL) and to cholesterol‐HDL_HF as the cholesterol concentration on HDL in the HDL fraction (HF). We will also refer to Apolipoproteins/SAA/CRP as Protein_HF, for formula convenience, as all proteins were measured in the HDL fraction.

To calculate the correct amount of each Protein (in ng) or cholesterol (in μg) per μg of ApoA‐I in the HDL fraction, we used the following formulas: Protein _HF (ng/ml)/ApoA‐I _HF (μg/ml) and Cholesterol‐HDL _HF (μg/ml)/ApoA‐I _HF (μg/ml). This calculation indicates the HDL‐cargo composition with regard to ApoA‐I which is the major and more stable apolipoprotein in HDL particles.

To calculate the correct amount of each Protein (in μg) (associated to HDL) in 1 ml plasma, we used the following formula: Protein _HF (μg/ml)/Cholesterol‐HDL _HF (mg/ml)*Cholesterol‐HDL _PL (mg/ml).

For general purposes, we will refer to HDL‐cargo to the 10 analytes assessed (Cholesterol, ApoA‐II, ApoC‐I, ApoC‐III, ApoD, ApoE, ApoH, ApoJ, CRP, SAA) compared to ApoA‐I levels.

2.5. PET scan

PET scans consisting of 30 min acquisitions were performed 40 min after injection of 370 MBq ¹¹C‐PiB. PET images were processed using a semi‐automatic region‐of‐interested method as previously described (Villemagne et al., 2011). Standardized uptake values (SUV) for ¹¹C‐PiB were calculated for all brain regions examined. The SUV ratio (SUVR) was calculated by dividing all regional SUV by the cerebellar cortex SUV. However, the centiloid scale was recently proposed to provide a standard quantification of Aβ‐PET images. In the centiloid scale, the Aβ burden can be expressed with values ranging from 0 (the typical Aβ burden in young controls) to 100 (the typical Aβ burden in mild AD patients) (Klunk et al., 2015). Centiloid values were generated using CapAIBL as described elsewhere (Bourgeat et al., 2018).

2.6. MRI imaging

Scanning centers either in Melbourne or Perth were used to acquire images using Siemens 3T Trio and Siemens 3T Skyra scanners (Melbourne) or Siemens 3T Verio and Siemens 1.5T Avanto scanners (Perth). The scans also included a 3D MPRAGE (Magnetization Prepared Rapid Acquisition Gradient Echo) image (voxel size 1.2 × 1 × 1 mm3, repetition time/echo time = 2300/2.98, flip angle = 9°). A 3D T2‐weighted Fluid‐attenuation inversion recovery (FLAIR) sequence, included in the image acquisition protocol, was obtained using two different sets of parameters. Gradient Recalled Echo (GRE) images used for SWI (Susceptibility‐Weighted Imaging) and QSM (Quantitative Susceptibility Mapping) were also acquired. Full details of protocols and parameters are described elsewhere (Fowler et al., 2021).

2.7. Statistical analysis

The operator was unaware of the experimental groups during the experiments and data collection. The sample size was determined using G‐Power, assuming a Cohen's D of 0.5 (medium‐size effect). For such analysis, to obtain a significance of p < 0.05, n = 64 samples were required in both HC and AD groups. Data were assessed for normality using Skewness and Kurtosis analysis and were log‐transformed when values were outside of the −2/+2 range. An interquartile range test was used to assess for outliers. Participant demographic and clinical characteristics were compared using either ANOVA or Wilcoxon Signed Ranks test for age, HDL, MMSE, and Brain Aβ deposition, and the Chi‐squared test for Gender, Site, APOE ε4 allele status, and brain Aβ deposition status. Statistical comparison of biomarker means in different groups was performed using Generalized linear models, unadjusted or adjusted for age, gender, and APOE ε4 allele status. When necessary, data were log‐transformed to better approximate normal distribution. p‐values less than 0.05 were regarded as nominally significant, with values less than 0.005 (Bonferroni adjusted value [0.05/10]) regarded as statistically significant. Partial correlation analysis was run upon adjustment for adjustment for covariates such as age, gender, and APOE ε4 allele status. Analyses were carried out using the R Statistical Environment (R Core Team, www.r‐project.org, v4.0.2) and SPSS version 27. For the Generalized linear model analyses, outliers were replaced with the median value for each marker. For the correlation analyses, outliers were not replaced.

We applied recursive feature elimination (RFE) to rank the HDL‐cargo proteins in differentiating different groups of individuals (AD, HC‐Conv, and HC) based on random forest (RF) classifier. RF is among the most robust to noise and missing data machine learning methods and RF is robust to overfitting and redundancy in predictors (Grinberg et al., 2020). RFE first fit the RF model using all the proteins as predictors, where each predictor was then ranked using its importance to the classification performance. At each iteration of the RFE process, top‐ranked predictors were retained, the model was refitted and performance was reassessed. Consequently, the proteins and factors that did not improve the classification accuracy were removed, and the top predictors were used to fit the final model. 1000 trees were used to build the RF model and root mean square error (RMSE) was used as the criterion to select the optimal model. Auto tuning was then applied to optimize the parameters of the RF. We implemented RFE with RF classification using caret package v6.0‐86 (Kuhn, 2008). This was performed by ten times five‐fold cross‐validation. This variable selection method provided insight into which variables and/or factors have the most distinguishing power in classifying Alzheimer's patients, thus those proteins and factors can serve as biomarkers for better diagnosis and prognosis. We evaluated all HDL‐cargo proteins as potential biomarkers and assessed the 5 most important with the goal to determine the best set of proteins relevant to other factors as diagnostic and prognostic biomarkers.

3. RESULTS

The basic demographics of the study participants are summarized in Table 1. In total, 87 stable healthy controls (HC), 38 HC‐Converters (HC‐Conv), and 88 Alzheimer's patients (AD) were studied (all participants were 65 years old and older). All 213 samples were evaluated for total HDL cholesterol levels and Mini‐Mental State Exam (MMSE) scores, while a smaller subset was assessed for brain amyloid deposition (assessed by PET scan) (Table 1).

TABLE 1.

Demographic characteristics, HDL levels, cognitive score, and amyloid brain deposition in HC, HC‐Conv, and AD participants

	HC	HC‐Conv	AD	p
N	87	38	88
Age	73 ± 6	75 ± 7	77 ± 7	0.27
Gender (M/F)	43/44	19/19	39/49	0.75
Site (Melbourne/Perth)	45/42	19/19	58/30	0.10
APOE ε4 (no/yes)	57/30	19/19	26/62	<0.001
HDL Cholesterol (mg/dl)	64 ± 15	68 ± 21	65 ± 17	1.00
MMSE	29 ± 1	28 ± 1	20 ± 4	<0.001
Brain Aβ deposition (n/y)	60/23	7/4	1/18	<0.001

Note: Values are presented as mean ± SD or as frequency. Analysis was considered significant with p < 0.05 (bold).

The levels of HDL‐cargo proteins, along with HDL cholesterol (which refers to the cholesterol associated with HDL particles), were assessed compared to the levels of ApoA‐I, which is the main apolipoprotein expressed on HDL particles. Generalized linear model (HC compared to AD) corrected for age, gender, and APOE ε4‐carrier status are shown in Table 2. Of the 10 HDL‐cargo analytes assessed among clinical groups, analyses showed statistically significant increases in cholesterol/ApoA‐I (p < 0.001 and p = 0.001, unadjusted and adjusted, respectively) and ApoD/ApoA‐I (p < 0.001 for both unadjusted and adjusted) and a significant decrease in ApoA‐II/ApoA‐I (p < 0.001 for both unadjusted and adjusted) in AD participants as compared with controls (Table 2).

TABLE 2.

Comparison of HDL‐cholesterol and HDL‐cargo (ApoA‐II, ApoC‐I, ApoC‐III, ApoD, ApoE, ApoH, ApoJ, CRP, and SAA) expressed as a ratio to ApoA‐I in stable HC vs AD

	HC (n)	AD (n)	p	p a
μgCholesterol/μgApoA‐I	1.34 ± 0.33 (87)	1.54 ± 0.38 (88)	0.000183	0.00139
ngApoA‐II/μgApoA‐I	293 ± 87 (77)	246 ± 53 (88)	6.11e‐05	0.000403
ngApoC‐I/μgApoA‐I	298 ± 73 (75)	273 ± 66 (85)	0.0213	0.274
ngApoC‐III/μgApoA‐I	58.4 ± 20.5 (87)	56.9 ± 19.8 (88)	0.617	0.531
ngApoD/μgApoA‐I	44.3 ± 8.1 (87)	49.8 ± 9.2 (88)	4.36e‐05	0.000383
ngApoE/μgApoA‐I	13.7 ± 5.3 (87)	10.4 ± 4.5 (88)	2.25e‐05	0.0243
ngApoH/μgApoA‐I	524 ± 155 (87)	492 ± 141 (88)	0.149	0.126
ngApoJ/μgApoA‐I	46.4 ± 10.4 (87)	45.1 ± 9.7 (88)	0.382	0.751
ngCRP/μgApoA‐I	5.39 ± 3.36 (87)	4.63 ± 2.37 (88)	0.0867	0.117
ngSAA/μgApoA‐I	12.9 ± 14.8 (87)	18.9 ± 20.1 (88)	0.0284	0.107

Note: Values are presented as mean ± SD. Generalized Linear Model analyses were performed unadjusted (p) and adjusted for age, gender, and ApoE ε4‐carrier status (p ^a). Data are presented as raw values, but statistical analysis was performed on log‐transformed data to better approximate normal distribution. Analysis was considered significant with p < 0.05 (bold).

^a Data were adjusted for age, gender, and ApoE ε4‐carrier status.

We then assessed if such changes in HDL cholesterol and HDL‐cargo (compared to ApoA‐I) occurred before the conversion from HC to AD or if such changes were mainly associated with the ongoing disease. Generalized linear model corrected for age, gender, and APOE ε4‐carrier status for HC compared to HC‐Conv are shown in Table 3. Unlike the HC vs AD comparisons, only the levels of cholesterol/ApoA‐I were nominally significantly (increased in HC‐Conv as compared with HC; p = 0.019 and p = 0.031, unadjusted and adjusted, respectively). All other analytes in the HDL‐cargo did not statistically differ between HC and HC‐Conv, suggesting that some HDL‐cargo remodeling (such as ApoA‐II/ApoA‐I and ApoD/ApoA‐I) takes place once after the clinical onset of the disease (Table 3).

TABLE 3.

Comparison of HDL‐cholesterol and HDL‐cargo (ApoA‐II, ApoC‐I, ApoC‐III, ApoD, ApoE, ApoH, ApoJ, CRP, and SAA) expressed as a ratio to ApoA‐I in stable HC vs HC‐Converters

	HC (n)	HC‐Conv (n)	p	p a
μgCholesterol/μgApoA‐I	1.34 ± 0.33 (87)	1.51 ± 0.38 (38)	0.0188	0.0307
ngApoA‐II/μgApoA‐I	293 ± 87 (77)	272 ± 69 (36)	0.156	0.195
ngApoC‐I/μgApoA‐I	298 ± 73 (75)	291 ± 72 (35)	0.611	0.928
ngApoC‐III/μgApoA‐I	58.4 ± 20.5 (87)	56.9 ± 17.7 (38)	0.687	0.709
ngApoD/μgApoA‐I	44.3 ± 8.1 (87)	43.8 ± 6.1 (38)	0.711	0.613
ngApoE/μgApoA‐I	13.7 ± 5.3 (87)	12.6 ± 4.1 (38)	0.238	0.733
ngApoH/μgApoA‐I	524 ± 155 (87)	516 ± 118 (38)	0.759	0.742
ngApoJ/μgApoA‐I	46.4 ± 10.4 (87)	44.1 ± 8.8 (38)	0.211	0.43
ngCRP/μgApoA‐I	5.39 ± 3.36 (87)	4.76 ± 1.18 (38)	0.128	0.334
ngSAA/μgApoA‐I	12.9 ± 14.8 (87)	9.4 ± 9.4 (38)	0.113	0.152

Note: Values are presented as mean ± SD. Generalized Linear Model analyses were performed unadjusted (p) and adjusted for age, gender, and ApoE ε4‐carrier status (p ^a). Data are presented as raw values, but statistical analysis was performed on log‐transformed data to better approximate normal distribution. Analysis was considered significant with p < 0.05 (bold).

^a Data were adjusted for age, gender, and ApoE ε4‐carrier status.

As the cholesterol/ApoA‐I was the only analyte that differed between HC and HC‐Conv, we investigated if such change took place in the proximity of the clinical onset or years before. As the HC‐Conv group includes individuals who converted to MCI/AD at 18 and 36 months, we assessed if the increased ratio of cholesterol/ApoA‐I in HC‐Conv was affected by a similar magnitude in both HC‐Conv groups (i.e. @18 & 36 months, Figure 1). Interestingly, individuals whose conversion to MCI/AD will take place within 18 months and were, therefore, closer to conversion (HC‐Conv 18m) had a cholesterol/ApoA‐I ratio significantly higher than in HC (p < 0.001) and similar to the AD group (Figure 1). Conversely, individuals whose conversion to MCI/AD will take place in 36 months and therefore were farther from conversion (HC‐Conv 36m) had levels similar to HC (Figure 1), suggesting that cholesterol overload on HDL takes place shortly before conversion to AD (within 18 months before conversion).

FIGURE 1.

Graphical representation of cholesterol levels on HDL (to ApoA‐I levels) in healthy controls (HC), healthy controls who will convert to MCI/AD in 36 months (HC‐C36), healthy controls who will convert to MCI/AD in 18 months (HC‐C18), and AD participants (AD) (ANOVA df = 2, F = 5.633, 0 < 0.01). p‐values from pairwise comparisons unadjusted for confounders. Statistical analysis was performed using log‐transformed data to better approximate normal distribution. *p < 0.05; ***p < 0.001.

Additionally, we have analyzed if the levels of HDL‐cargo proteins were altered when assessed as levels of HDL‐associated protein per ml of plasma (Table S1). However, as many of the HDL‐protein cargo are not exclusively expressed on HDL (with the sole exception of ApoA‐I), these values do not represent the total amount of circulating protein in plasma, but only the fraction that it is directly associated with HDL particles.

To determine the effect of APOE genotype on HDL‐cargo, we then assessed if any HDL‐cargo protein was influenced by APOE genotype and clinical classification, or by APOE genotype alone. ApoE/ApoA‐I ratio was significantly affected by APOE genotype (p < 0.001) alone. However, there was no significant interaction between clinical classification and APOE genotype, indicating that differences in ApoE levels were solely affected by APOE genotype, while ApoE levels within each APOE genotype were not affected by the clinical classification (Figure 2a). When assessed by APOE genotype alone, ApoE/ApoA‐I ratio showed the highest levels in APOE ε2/3 (p < 0.001 vs ε3/3, ε3/4, and ε4/4), while APOE ε4/4 showed the lowest (p < 0.001 vs ε3/3, ε2/4 and ε3/4) (Figure 2b). Of the remaining 9 HDL‐cargo analytes, no HDL‐cargo protein (compared to ApoA‐I) was affected either by APOE genotype alone or by the interaction between clinical classification and APOE genotype, indicating that their levels were independent of APOE genotype or the clinical classification within each APOE genotype (data not shown).

FIGURE 2.

Graphic representation of ApoE levels (to ApoA‐I levels) in different APOE genotypes (ε2/3. ε3/3, ε2/4, ε3/4, and ε4/4) in different clinical groups (healthy controls (HC), healthy controls converters (HC‐Conv), and AD patients (AD)) (Figure 2a) (ANOVA df = 8, F = 1.131, p = 0.344). Graphical representation of ApoE levels (to ApoA‐I levels) in different APOE genotypes (ε2/3. ε3/3, ε2/4, ε3/4, and ε4/4). p‐values from pairwise comparisons unadjusted for confounders. Statistical analysis was performed using log‐transformed data to better approximate normal distribution. p < 0.05; **p < 0.01, ***p < 0.01 (Figure 2b) (ANOVA df = 4, F = 29.236, p < 0.001).

To determine if any of the HDL‐cargo analytes correlated with brain amyloid levels in individuals with ongoing brain amyloidosis (HC, HC‐Conv, and AD), correlation analyses were carried out (Table 4). In unadjusted correlations, only the levels of cholesterol/ApoA‐I significantly correlated with the levels of brain amyloid (p = 0.036). However, upon correction for age, gender, and APOE ε4‐carrier status, the correlation analysis only approached nominal significance (p = 0.061). None of the other HDL‐cargo analytes were significantly correlated with brain amyloid levels, regardless of whether the analysis was carried out unadjusted or adjusted for age, gender, and APOE ε4‐carrier status (Table 4).

TABLE 4.

Correlation and partial correlation of HDL‐cholesterol and HDL‐cargo (ApoA‐II, ApoC‐I, ApoC‐III, ApoD, ApoE, ApoH, ApoJ, CRP, and SAA) expressed as a ratio to ApoA‐I with brain amyloid deposition and brain amyloid deposition longitudinal changes in individuals with ongoing brain amyloidosis (PiB+)

	Centiloid t0 (0m)
	r	p	r a	p a
μgCholesterol/μgApoA‐I	0.314	0.036	0.292	0.061
ngApoA‐II/μgApoA‐I	−0.041	0.790	−0.030	0.850
ngApoC‐I/μgApoA‐I	0.228	0.142	0.163	0.314
ngApoC‐III/μgApoA‐I	0.065	0.685	0.020	0.900
ngApoD/μgApoA‐I	−0.041	0.791	−0.021	0.893
ngApoE/μgApoA‐I	0.009	0.952	−0.077	0.629
ngApoH/μgApoA‐I	−0.086	0.575	−0.060	0.707
ngApoJ/μgApoA‐I	−0.074	0.629	−0.089	0.574
ngCRP/μgApoA‐I	−0.186	0.222	−0.189	0.229
ngSAA/μgApoA‐I	0.026	0.867	−0.028	0.858

Note: Pearson's correlation unadjusted and adjusted for age, gender ApoE ε4‐carrier status was performed between HDL‐cargo and brain amyloid data. Analysis was considered significant with p < 0.05 (bold).

^a Data were adjusted for age, gender, and ApoE ε4‐carrier status. For all analytes n = 45 except for ngApoC‐I/μgApoA‐I where n = 43.

Subsequently, we carried out a correlation analysis (adjusted for age, gender, and APOE ε4‐carrier status) to determine if any HDL‐cargo protein levels were associated with brain volumetric parameters such as grey matter volume (GM), white matter volume (WM), ventricle volume (Vent), left and right hippocampal volume (HL and HR, respectively) in individuals with ongoing brain amyloidosis (Table 5). Cholesterol/ApoA‐I ratio positively and significantly correlated with ventricular volume (p = 0.043), while negatively correlated with grey matter volume, albeit this was not quite significant (p = 0.063). ApoA‐II/ApoA‐I ratio positively correlated with grey matter and hippocampal volume (p < 0.001, p = 0.038 and p = 0.035 for GM, HL, and HR, respectively), while negatively correlated with ventricular volume (p = 0.024). ApoJ/ApoA‐I ratio positively correlated with grey matter and left hippocampal volume (p = 0.029 and p = 0.033, respectively), while right hippocampal volume only trended toward significance (p = 0.080). None of the other HDL‐cargo analytes were significantly correlated with brain volumetric parameters (Table 5).

TABLE 5.

Partial correlation of HDL‐cholesterol and HDL‐cargo (ApoA‐II, ApoC‐I, ApoC‐III, ApoD, ApoE, ApoH, ApoJ, CRP, and SAA) expressed as a ratio to ApoA‐I with brain volumetric parameters in individuals with ongoing brain amyloidosis (PiB+)

	GM t0		WM t0		Vent t0		HL t0		HR t0
	r a	p a	r a	p a	r a	p a	r a	p a	r a	p a
μgCholesterol/μgApoA‐I	−0.327	0.063	−0.105	0.562	0.355	0.043	−0.201	0.263	−0.263	0.140
ngApoA‐II/μgApoA‐I	0.543	<0.001	0.183	0.308	−391	0.024	0.363	0.038	0.368	0.035
ngApoC‐I/μgApoA‐I	0.002	0.993	−0.077	0.680	−0.140	0.452	−0.021	0.912	0.011	0.953
ngApoC‐III/μgApoA‐I	0.190	0.290	−0.027	0.883	−0.036	0.844	0.109	0.548	0.185	0.304
ngApoD/μgApoA‐I	0.140	0.438	0.097	0.592	−0.174	0.333	0.076	0.673	0.104	0.564
ngApoE/μgApoA‐I	−0.067	0.712	−0.184	0.306	−0.005	0.977	0.229	0.200	0.093	0.605
ngApoH/μgApoA‐I	0.301	0.089	0.055	0.760	−0.129	0.475	0.163	0.365	0.125	0.487
ngApoJ/μgApoA‐I	0.380	0.029	0.033	0.855	−0.324	0.066	0.373	0.033	0.309	0.080
ngCRP/μgApoA‐I	0.291	0.101	0.105	0.561	−0.256	0.150	0.284	0.109	0.231	0.196
ngSAA/μgApoA‐I	0.204	0.255	0.127	0.482	0.029	0.872	0.007	0.969	0.003	0.986

Note: Pearson's correlation adjusted for age, gender ApoE ε4‐carrier status was performed between HDL‐cargo and brain volumetric parameters at t0. Analysis was considered significant with p < 0.05 (bold).

Abbreviations: GM, grey matter volume; HL, left hippocampal volume; HR, right hippocampal volume; Vent, ventricular volume; WM, white matter volume.

^a Data were adjusted for age, gender ApoE ε4‐carrier status. For all analytes n = 36 except for ngApoC‐I/μgApoA‐I where n = 34.

On assessing whether MMSE scores were associated with any HDL‐cargo analytes in AD patients, MMSE scores were found to decrease with the disease progression. No HDL‐cargo analytes were significantly associated with MMSE score or MMSE longitudinal changes, suggesting that HDL‐cargo is not involved with altered cognitive impairment changes (Table S2).

RFE variable selection indicated that among all the HDL‐cargo protein variables ApoE/ApoA‐I and ApoD/ApoA‐I have the best distinguishing powers in classifying AD and HC, followed by ApoA‐II/ApoA‐I, CRP/ApoA‐I, and cholesterol/ApoA‐I (data not shown).

4. DISCUSSION

Over the past decades, the importance of cholesterol, LDL, and HDL in AD have been extensively studied, and in this regard, it is widely accepted that high cholesterol levels and high LDL levels are considered risk factor for the disease, albeit it appears that such increased risk is present at mid‐life and disappeared with increased age (Mielke et al., 2005). As such, a clear distinction between the roles of LDL and HDL has indicated that while LDL cholesterol levels were associated with increased brain amyloid deposition (Reed et al., 2014), high HDL cholesterol levels were instead considered protective. This latest notion came from several studies in which HDL cholesterol levels were associated with lower brain amyloid deposition, a reduced risk for AD, better cognitive functions, and higher MMSE scores (Atzmon et al., 2002; Bates et al., 2017; Reitz et al., 2010). It is, however, important to note that recent evidence suggested that HDL functionality, rather than HDL overall levels, determine HDL functions (Rosenson et al., 2016).

In this report, we, therefore, assessed the HDL‐cargo composition (Cholesterol, ApoA‐II, ApoC‐I, ApoC‐III, ApoD, ApoE, ApoH, ApoJ, CRP, and SAA, all measured as ratio to ApoA‐I) in stable healthy control, healthy controls who will convert to MCI/AD within 36 months and AD to determine if any of the HDL‐cargo analytes were associated to brain amyloid deposition, brain volumetric parameters (cortical grey matter, cortical white matter, ventricular volume, hippocampal volume), and cognitive functions scores (MMSE). We found an increased amount of cholesterol/ApoA‐I ratio (in both HC‐Conv and AD, compared to HC), increased ApoD/ApoA‐I (in AD), and decreased ApoA‐II/ApoA‐I ratios (in AD). While the cholesterol/ApoA‐I ratio varies in the general population and as a function of HDL size and HDL maturation, the increased cholesterol/ApoA‐I ratio on HDL particles in our study was unexpected, as the overall plasma levels of HDL‐Cholesterol were unchanged in all clinical groups. Since ApoA‐I is solely expressed on HDL, our data also indicated that levels of HDL‐associated ApoA‐I per ml of plasma were significantly lower in AD. Such a decrease can be considered an ApoA‐I decrease in plasma and is in accordance with other reports which have indicated lower levels of ApoA‐I in AD (Kawano et al., 1995; Liu et al., 2006; Merched et al., 2000). As ApoA‐I is also the main (and more stable in number on HDL particles) Apolipoprotein on HDL, and in consideration of its reduced levels in AD, our data suggest that in AD (and in HC‐Converters, but only in those who are within 18 months from conversion to MCI/AD), there is a reduced number of HDL particles which are overloaded with cholesterol. This would explain the lower amount of ApoA‐I in plasma, but the increased amount of cholesterol on HDL particles (increased cholesterol/ApoA‐I ratio) would counterbalance the reduced numbers of HDL particles and explain the absence of different Cholesterol‐HDL plasma levels across clinical groups. Accordingly, most of the other HDL‐cargo proteins (evaluated in the same assay with ApoA‐I), which have shown reduced levels in AD compared to HC per ml of plasma (HDL‐associated protein cargo/ml plasma), do not display a different ratio to ApoA‐I, suggesting that there is no altered composition of HDL particle, but rather a reduced number.

Overall, the only exception in the composition, along with the Cholesterol/ApoA‐I ratio which occurs before the clinical conversion from HC to MCI/AD, are ApoA‐II/ApoA‐I ratio, which is decreased in AD compared to HC and ApoD/ApoA‐I ratio, which is increased in AD compared to controls. However, since both ApoA‐II/ApoA‐I and ApoD/ApoA‐I ratios are not altered in HC‐Conv compared to stable HC, it is possible that such changes take place once the disease has started, unlike the cholesterol overload which may happen before the onset of the disease (Figure 3). It has to be noted that in this study we focused on HDL composition assessed as apolipoprotein ratio to ApoA‐I on HDL particles, which may represent a better marker for conversion to AD and/or pathological changes in AD, rather than individual plasma apolipoprotein levels, which could provide confounding results as many apolipoproteins are shared among several lipoprotein particles (LDL, VLDL).

FIGURE 3.

Schematic representation of HDL remodeling in which cholesterol on HDL particles increases before the conversion to AD, while the decrease in ApoA‐II and the increase in ApoD take place once the disease has started.

Interestingly, ApoE/ApoA‐I ratio was significantly lower in carriers of the allele APOE ε4 and lowest in the homozygous carrier for the allele APOE ε4. Such ApoE/ApoA‐I ratio, which defines HDL composition, was not affected by clinical classification but was solely affected by APOE genotype. As the presence of the allele APOE ε4 is widely considered the biggest risk factor for sporadic AD, such reduced ApoE presence on HDL may reduce HDL functionality, therefore providing an additional explanation for the increased risk associated with APOE genotype. This reduced binding of ApoE ε4 to HDL (with a preferential binding to chylomicrons and VLDL) is also in accordance with other studies (Poirier et al., 2014). In this regard, several studies which evaluated the effects of ApoE inducers, such as Liver‐X‐receptor (LXR) agonists, have reported protective effects in transgenic AD mice (Fitz et al., 2010; Lefterov et al., 2007; Riddell et al., 2007), supporting the notion that high levels of ApoE are beneficial in the disease. However, one limitation of this study is that it was not powered to determine the effects of ApoE genotype on disease progression.

When the influence of HDL‐cargo on regional brain volumes was assessed, we selected individuals with ongoing brain amyloidosis, in whom regional changes may be more prominent. Higher cholesterol/ApoA‐I ratio on HDL resulted to be marginally associated with lower grey matter volume. In parallel, higher cholesterol levels were also associated with greater ventricular volume. While this seems to be counterintuitive at first, as higher HDL cholesterol levels have been associated with protective features, the cholesterol increase here mentioned is per HDL particle (overall plasma HDL levels were unaffected in this study). In this report we suggest that cholesterol overload on HDL is detrimental as it affected brain volumetric parameters, suggesting that HDL quality may be more important than quantity itself. Such a hypothesis would also be in accordance with another report indicating that cholesterol overload on HDL has a negative effect on HDL anti‐atherogenic functions (Qi et al., 2015), while an increased cholesterol/ApoA‐I ratio on HDL can be a predictor of cardiovascular disease and associated mortality (Rhee et al., 2017). In accordance with these findings, a lower cholesterol/ApoA‐I ratio was linked to protective features, as higher levels of small HDL particles (known to have lower cholesterol/ApoA‐I ratio) in CSF have been associated with better cognitive performances (Martinez et al., 2022).

Conversely, ApoA‐II/ApoA‐I, and ApoJ/ApoA‐I ratio displayed the opposite effect on grey matter and ventricular volume, with higher levels of both proteins being associated with higher grey matter volume and smaller ventricular volume. In addition, a higher ApoA‐II/ApoA‐I ratio was also significantly associated with higher hippocampal volume (both left and right). It is, therefore, fitting that ApoA‐II and ApoJ on HDL particles displayed protective features, as both proteins have previously been linked to AD for their capacity to bind to Aβ reducing its effects (Bell et al., 2007; Hammad et al., 1997; Yamauchi et al., 2000). Ultimately, we observed a trend between higher cholesterol levels on HDL and higher levels of brain amyloid deposition, further indicating that cholesterol overload on HDL may result in non‐functional HDL particles with consequent negative impact.

Taken together, our study suggested that in AD (a) there is a reconfiguration of HDL particles with significant increased cholesterol/ApoA‐I ratio (which may take place before the onset of the disease), increased ApoD/ApoA‐I and reduced ApoA‐II/ApoA‐I ratios on HDL which may affect HDL functionality itself; (b) such reconfiguration with the consequent increase of cholesterol/ApoA‐I ratio on HDL is associated in individuals with ongoing brain amyloidosis with lower cortical grey matter volume and greater ventricular volume; while (c) other apolipoproteins on HDL, such as ApoA‐II and ApoJ, displayed protective features and higher levels of both were associated in individuals with ongoing brain amyloidosis with higher cortical grey matter volume and smaller ventricular volume; and (d) that ApoE/ApoA‐I ratio on HDL are solely a function of APOE genotype and reduced levels of ApoE on HDL in APOE ε4 carrier may further alter HDL functionality, reduce its protective features and provide another reason for the increased risk associated with APOE genotype.

Altogether, these data are supporting the notion that the functionality of HDL is related to its protein cargo and it is independent of its absolute levels, and that HDL cargo and HDL functionality may be altered in AD as they are in many other diseases. Further studies will, however, be necessary to better define the extent of HDL functionality in relationship with HDL‐cargo in the disease.

AUTHOR CONTRIBUTIONS

Steve Pedrini: Conceptualization, Methodology, Investigation, Data analysis, Visualization, Writing ‐ original draft, Writing ‐ review/editing. James D. Doecke: Data analysis, Writing ‐ review/editing. Eugene Hone: Investigation, Writing ‐ review/editing. Penghao Wang: Data analysis, Writing ‐ review/editing. Rohith Thota: Conceptualization, Writing ‐ review/editing. Ashley I. Bush: Writing ‐ review/editing. Christopher C. Rowe: Data curation, Writing ‐ review/editing. Vincent Dore: Data curation, Writing ‐ review/editing. Victor L. Villemagne: Data curation, Writing ‐ review/editing. David Ames: Data curation, Writing ‐ review/editing. Stephanie Rainey‐Smith: Data curation, Writing ‐ review/editing. Giuseppe Verdile: Conceptualization, Writing ‐ review/editing. Hamid R. Sohrabi: Data curation, Writing ‐ review/editing. Manfred R. Raida: Conceptualization, Methodology, Investigation, Writing ‐ review/editing. Kevin Taddei: Writing ‐ review/editing, Project administration. Sam Gandy: Conceptualization, Writing ‐ review/editing. Colin L. Masters: Data curation, Writing ‐ review/editing. Pratishtha Chatterjee: Conceptualization, Writing ‐ original draft, Writing ‐ review/editing. Ralph N. Martins: Conceptualization, Writing ‐ original draft, Writing ‐ review/editing, Supervision, Project administration.

CONFLICT OF INTEREST

Ashley Bush is a shareholder in Alterity Ltd, Cogstate Ltd and Mesoblast Ltd. He is a paid consultant for, and has a profit share interest in, Collaborative Medicinal Development Pty Ltd. He has received lecture fees from Biogen and Merck Sharp & Dohme P/L. Dr. Gandy serves as a consultant for Ritrova Therapeutics and as a founder of Recuerdo Pharmaceuticals (inactive). He has served as a consultant in the past for Diagenic, and he has received research support in the past from Warner‐Lambert, Pfizer, Baxter, and Avid. He currently receives research support from the NIH. Dr. Victor Villemagne is a Senior Editor for the Journal of Neurochemistry.

Supporting information

Table S1

Table S2

Figure S1

Pedrini, S. , Doecke, J. D. , Hone, E. , Wang, P. , Thota, R. , Bush, A. I. , Rowe, C. C. , Dore, V. , Villemagne, V. L. , Ames, D. , Rainey‐Smith, S. , Verdile, G. , Sohrabi, H. R. , Raida, M. R. , Taddei, K. , Gandy, S. , Masters, C. L. , Chatterjee, P. , Martins, R. N. , & the AIBL Research Group (2022). Plasma high‐density lipoprotein cargo is altered in Alzheimer's disease and is associated with regional brain volume. Journal of Neurochemistry, 163, 53–67. 10.1111/jnc.15681

DATA AVAILABILITY STATEMENT

AIBL data have been provided to the Global Alzheimer's Association Interactive Network (GAAIN), and software installed, thereby enabling GAAIN users to interrogate metadata and receive cohort summaries, whereupon users can request further information (and biofluid samples; blood and CSF) if needed by submitting an Expression of Interest (EoI). This mechanism ensures worldwide sharing and utilization of AIBL data (and biofluid samples) and complements the existing framework whereby AIBL imaging scans and demographic data are available on the ADNI LONI (Laboratory of Neuro Imaging, University of Southern California) website for free download and use by researchers worldwide. Up to 3 August 2019, LONI/ GAAIN data applications had been received from over 120 companies, more than 2000 organizations/institutes/departments, and over 3500 individuals, spanning 81 countries. Moreover, hundreds of EoIs from academic and industry‐based individuals have been approved resulting in further data and biofluid sample sharing. Data will also be provided to the National Institute on Aging‐funded ADOPIC (Alzheimer's Dementia Onset and Progression in International Cohorts) project which will harmonize data from the AIBL, ADNI, Washington University, St. Louis, and University of Washington, Seattle, cohorts to determine factors which influence cognitive decline in AD.