Levels of Soluble Apolipoprotein E/Amyloid-β (Aβ) Complex Are Reduced and Oligomeric Aβ Increased with APOE4 and Alzheimer Disease in a Transgenic Mouse Model and Human Samples*♦

Background: An ELISA was developed to determine the role of apoE/Aβ on soluble Aβ accumulation. Results: In AD transgenic mouse brain and human synaptosomes and CSF, levels of soluble apoE/Aβ are lower and oligomeric Aβ levels are higher with APOE4 and AD. Conclusion: Isoform-specific apoE/Aβ levels modulate soluble oligomeric Aβ levels. Significance: ApoE/Aβ and oligomeric Aβ represent a mechanistic approach to AD biomarkers. Human apolipoprotein E (apoE) isoforms may differentially modulate amyloid-β (Aβ) levels. Evidence suggests physical interactions between apoE and Aβ are partially responsible for these functional effects. However, the apoE/Aβ complex is not a single static structure; rather, it is defined by detection methods. Thus, literature results are inconsistent and difficult to interpret. An ELISA was developed to measure soluble apoE/Aβ in a single, quantitative method and was used to address the hypothesis that reduced levels of soluble apoE/Aβ and an increase in soluble Aβ, specifically oligomeric Aβ (oAβ), are associated with APOE4 and AD. Previously, soluble Aβ42 and oAβ levels were greater with APOE4 compared with APOE2/APOE3 in hippocampal homogenates from EFAD transgenic mice (expressing five familial AD mutations and human apoE isoforms). In this study, soluble apoE/Aβ levels were lower in E4FAD mice compared with E2FAD and E3FAD mice, thus providing evidence that apoE/Aβ levels isoform-specifically modulate soluble oAβ clearance. Similar results were observed in soluble preparations of human cortical synaptosomes; apoE/Aβ levels were lower in AD patients compared with controls and lower with APOE4 in the AD cohort. In human CSF, apoE/Aβ levels were also lower in AD patients and with APOE4 in the AD cohort. Importantly, although total Aβ42 levels decreased in AD patients compared with controls, oAβ levels increased and were greater with APOE4 in the AD cohort. Overall, apoE isoform-specific formation of soluble apoE/Aβ modulates oAβ levels, suggesting a basis for APOE4-induced AD risk and a mechanistic approach to AD biomarkers.

gates of A␤ (oA␤). For this study, an apoE/A␤ ELISA was developed to determine the effect of the APOE genotype on the levels of soluble apoE/A␤ and A␤.
The amyloid hypothesis posits that deposition of extracellular amyloid is central for producing the neurodegenerative processes characteristic of AD (3). In the landmark 1992 paper, Wisniewski and Frangione (4) proposed that apoE was a "pathological chaperone," based on the co-localization of apoE with A␤ in amyloid plaques as detected via immunohistochemistry (IHC). Thus, apoE was thought to facilitate the process of A␤ deposition as amyloid. Biochemical analyses validate IHC measures, as the levels of apoE and A␤ are equivalent in the insoluble extraction fraction from brains of transgenic mice expressing familial AD (FAD) mutations (FAD-Tg), specifically the 5ϫFAD-Tg mice (5). The association of APOE4 with AD risk was first described in 1993 (6,7), leading to research efforts focused on the effects of the APOE genotype on plaque burden and the structural relationship between apoE and amyloid. IHC analysis demonstrates that plaque deposition is greater with APOE4 compared with APOE3 in AD and nondemented controls (8,9) and that a higher proportion of A␤ within a plaque is associated with apoE4 than with apoE3 (10). Biochemical analysis confirms that the levels of apoE and A␤ are also higher with APOE4 compared with APOE3 in the insoluble extraction fraction from brains of FAD-Tg mice (11). Thus, APOE4 not only facilitates amyloid deposition but also forms a greater amount and/or more stable form of apoE4/amyloid than apoE3/amyloid.
The amyloid hypothesis has been revised, as plaque burden does not correlate with the dementia that is characteristic of AD (12,13). However, soluble A␤ and oA␤ do correlate with cognitive decline and disease severity in humans (14). oA␤ is also detected in FAD-Tg mice and is associated with memory decline (14). Thus, the structure-function relationship of soluble A␤ and oA␤ is an area of intense research. However, unlike amyloid, which refers to a specific parallel ␤-sheet structure, oA␤ refers to a number of assemblies defined by a variety of detection methods (14). This makes interpretation and comparison of results problematic, particularly with in vivo data. We recently developed an oA␤ ELISA and demonstrated that in EFAD-Tg mice soluble A␤42 and oA␤ are greater in E4FAD mice, compared with E2FAD and E3FAD (15). A␤ clearance also appears to be decreased with APOE4 (16), suggesting that soluble apoE/A␤ may modulate soluble A␤ and oA␤ levels.
Research efforts to determine apoE/A␤ levels, particularly soluble complex levels, have been hindered by a lack of quantitative detection methods. A variety of techniques have produced results that can be inconsistent and difficult to interpret (7,(17)(18)(19)(20)(21)(22)(23)(24)(25)(26)(27). Even the initial biochemical characterizations of the molecular interactions between apoE and A␤ were problematic, primarily because of two parameters. The first variable was the lipidation state of apoE. Using purified protein, apoE4 bound A␤ with a higher affinity than apoE3 (28,29). However, this result is reversed using physiologically relevant, lipidated apoE; levels of the apoE3/A␤ complex are significantly greater than the apoE4/A␤ complex (21,28,29). Second, the definition of an apoE/A␤ is primarily operational, with assay stringency the primary variable (7,(17)(18)(19)(20)(21)(22)(23)(24)(25)(26)(27)30). For example, apoE3/A␤ levels are greater than apoE4/A␤ as determined by Western analysis of SDS-PAGE (21), but by nondenaturing gel electrophoresis, the levels of the apoE3/A␤ complex are comparable with apoE4/A␤ (31). Although these data are consistent with an SDS-stable apoE3/A␤ complex (32), and an apoE4/A␤ complex that is disrupted by SDS, the total amount of apoE/A␤ cannot be quantified by Western analysis of SDS-PAGE. The first goal of this study was to define biochemically generated apoE/A␤ in the context of a single quantitative and potentially high throughput method that would also define both total and detergent (SDS)-stable apoE/A␤, providing a platform for comparison among apoE isoforms and across methods. Thus, a new apoE/A␤ ELISA was developed and optimized biochemically. In vitro, total complex levels were equivalent among the apoE isoforms, although the apoE3/A␤ complex was more SDS-stable than the apoE4/A␤ complex but was less SDS-stable than apoE2/A␤. These results are consistent with previous results utilizing several methods that suggest the levels of apoE3/A␤ and apoE4/A␤ complex are comparable in the absence of SDS but that SDS-stable apoE3/A␤ complex levels are greater than apoE4/A␤ (18,21,22,27).
In contrast to biochemical analysis, the number of in vivo reports on soluble apoE/A␤ is limited. ApoE/A␤ complex has been detected in the soluble fraction of human brain (33) and in human cerebrospinal fluid (CSF) (30,34), although the data were primarily produced using Western analysis of SDS-PAGE. By IHC, apoE also co-localizes with A␤ at the synapse (35), and insoluble apoE/A␤ complexes appear to form preferentially with apoE4 compared with apoE3 (36). However, the effect of the APOE genotype on soluble synaptic apoE/A␤ levels remains unclear (37)(38)(39)(40). Thus, the new apoE/A␤ ELISA was used in vivo to determine the levels of soluble apoE/A␤ and the effect of the APOE genotype. In EFAD transgenic mice, previous data demonstrated that with APOE4 the soluble A␤42 and oA␤ levels were greater (15), and in the data presented herein, soluble apoE4/A␤ complex levels were lower and less stable compared with apoE3/A␤ and apoE2/A␤ levels. In human synaptosome preparations and CSF, apoE/A␤ levels were lower in AD compared with controls and with APOE4 compared with APOE3 in the AD cohort. Importantly, in human CSF, although total A␤42 levels decreased in AD patients compared with controls, oA␤ levels increased and were greater with APOE4 in the AD cohort. Taken together, the low levels of the soluble apoE4/A␤ complex and high levels of the soluble oA␤ suggest an impaired clearance mechanism for the soluble forms of A␤ and a potential basis for APOE4-induced AD risk, as well as a mechanistic approach to CSF biomarkers for AD. MI). Goat ␣-apoE antibodies were from Calbiochem, Meridian (Memphis, TN), and Millipore (Billerica, MA). Recombinant apoE3 was from BioVision (Milpitas, CA), and synthetic A␤ peptides were from California Peptide (Napa, CA).

ApoE/A␤ Complex Standard Development and Biochemical Characterization
ApoE/A␤ Complex Formation-HEK-apoE or recombinant apoE and A␤ were incubated at the indicated concentrations for 2 h at room temperature, pH 7.4, with SDS (Sigma) or vehicle at the indicated concentrations. The pH profile for apoE/A␤ levels was conducted as described (21).
ELISA Curve Fitting-In the absence of SDS, the EC 50 value for A␤ and apoE was calculated using the four-parameter logistic Equation 1, apoE/A␤ complex levels ϭ bottom ϩ ͑top Ϫ bottom͒/ ͑1 ϩ 10 ∧ ͑͑logEC 50 Ϫ X͒ ⅐ Hill slope͒͒ (Eq. 1) Top and bottom represents the apoE/A␤ levels at the plateaus. EC 50 is the concentration of A␤ or apoE that produces 50% maximal response. X is the concentration of the variable i.e. A␤ or apoE.
In the presence of SDS, the IC 50 value for SDS was calculated according to Equation 2, apoE/A␤ complex levels ͑% of control͒ ϭ 100/͑1 ϩ 10 ∧ ͑X Ϫ logIC 50 ͒͒ (Eq. 2) IC 50 is the effective concentration of SDS that produces 50% response. X corresponds to the concentration of SDS.
Analysis was conducted for each individual experiment, and data were analyzed using one-way analysis of variance (ANOVA) followed by Tukey's post hoc analysis GraphPad Prism Version 5 for Macintosh was used for all curve-fitting analyses.

ApoE/A␤ Complex ELISA
Biochemical ELISA Development-To accurately quantify total and SDS-stable levels of apoE/A␤, a specific ELISA was developed. The apoE/A␤ complex formed between HEK-apoE and unaggregated A␤42 was utilized for ELISA development.
To minimize nonspecific binding of apoE and A␤ and to maximize apoE/A␤ detection, a number of antibody combinations were screened as capture or detection antibodies on NUNC Maxisorp TM (high bind) or Microwell TM (low bind) plates (supplemental Fig. 1). Results demonstrated the following. 1) Nonspecific binding of A␤ to high and low bind plates precludes the use of ␣-A␤ antibodies for detection. 2) Nonspecific binding of HEK-apoE prevents the use of high bind plates (supplemental Fig. 1A). 3) ApoE/A␤ complex, but not apoE or A␤, is detected on low bind plates using ␣-apoE capture and ␣-A␤ detection antibodies (supplemental Fig. 1B). 4) ␣-A␤ (MOAB-2) capture and ␣-apoE (Calbiochem) detection antibodies produce the highest signal/background ratio compared with other antibodies tested (supplemental Fig. 1, C and D). Thus, the optimal reagents/conditions for specific HEK-apoE/A␤ detection by ELISA were low bind 96-well plates with ␣-A␤ (MOAB-2) capture and ␣-apoE (Calbiochem) detection antibodies.
ApoE/A␤ ELISA-For protocol 1, low bind plates were coated with MOAB-2 at 6.25 g/ml in carbonate coating buffer overnight at 4°C. Plates were washed (three times in PBS), blocked (4% BSA, 1.5 h, 37°C), washed again (three times in PBS), and incubated with samples overnight. The plates were then washed (three times in PBS), incubated with a 200-fold dilution of ␣-apoE (Calbiochem) (1.5 h, 37°C), washed, and incubated with HRP-conjugated antibodies (1.5 h, RT, 1:5000 dilution, Jackson ImmunoResearch, West Grove, PA). Following a final wash step (three times in PBS), 3,3Ј,5,5Ј-tetramethylbenzidine liquid substrate Superslow (Sigma) was added, and absorbance was measured at A 620 . For protocol 2, all steps were identical to protocol 1, with the exception that high bind plates were utilized (see ELISA analysis of human CSF).

Soluble ApoE/A␤ Complex Detection in EFAD Mice
EFAD Transgenic Mice-Experiments follow the UIC Institutional Animal Care and Use Committee protocols. EFAD mice (15) are the result of crossing 5ϫFAD mice, which co-express five FAD mutations (APP K670N/M671L, I716V, and V717I and PS1, M146L and L286V) under the control of the Thy-1 promoter with apoE-targeted replacement mice. Details on the production, breeding, genotyping, and genetic background of these mice are described in Ref. 15.
Tissue Preparation-Brain tissue isolation and serial protein extraction were conducted as described previously (5,15). Briefly, 6-month-old male EFAD mice were anesthetized with sodium pentobarbital (50 mg/kg) and transcardially perfused (PBS plus protease inhibitors (Calbiochem, set 3)), and brains were removed and dissected at the midline. Right hemi-brains were dissected on ice into cortex, hippocampus, and cerebellum, immediately snap-frozen in liquid nitrogen, and stored at Ϫ80°C until use. The dissected tissue was homogenized in 15 volumes (w/v) of TBS; samples were centrifuged (100,000 ϫ g, 1 h at 4°C), and the TBS (soluble) fraction was aliquoted prior to freezing in liquid nitrogen and storage at Ϫ80°C.
ApoE/A␤ Complex-The apoE/A␤ levels were measured using a 4-fold sample dilution of the TBS extraction fraction from the hippocampus of EFAD mice according to apoE/A␤ ELISA protocol 1. The standard curve used a fixed HEK-apoE concentration of 140 nM (apoE concentration in the TBS extraction of EFAD mice at 6 months for all APOE genotypes) and varied A␤ concentrations. Data were normalized to protein concentration in each sample.

ApoE/A␤ Detection in Human Synaptosomes
Brain samples of parietal cortex (A7, A39, and A40) were obtained at autopsy for cases followed by the Alzheimer disease research centers at UCLA, University of California at Irvine, and University of Southern California (supplemental Table 1); the last clinical diagnosis and full neuropathological report and diagnosis were available for all cases. Control samples included normal cases and pathological controls. Immediately upon receipt, samples (ϳ0.3-5 g) were minced in 0.32 M sucrose with protease inhibitors (2 mM EDTA, 2 mM EGTA, 0.2 mM PMSF, 1 mM sodium pyrophosphate, 5 mM NaF, 10 mM Tris) and then stored at Ϫ70°C until homogenization. The P-2 (crude synaptosome; synaptosome-enriched fraction) was prepared as described previously (42); briefly, tissue was homogenized in ice-cold buffer (0.32 M sucrose, 10 mM Tris, pH 7.5, plus protease inhibitors: pepstatin (4 mg/ml), aprotinin (5 mg/ml), trypsin inhibitor (20 mg/ml), EDTA (2 mM), EGTA (2 mM), PMSF (0.2 mM), leupeptin (4 mg/ml)). The homogenate was first centrifuged at 1000 ϫ g for 10 min; the resulting supernatant was centrifuged at 10,000 ϫ g for 20 min to obtain the crude synaptosomal pellet. Aliquots of P-2 were routinely cryopreserved in 0.32 M sucrose and banked at Ϫ70°C until the day of the experiment. On the day of the experiment, cryopreserved human P-2 aliquots were defrosted at 37°C, resuspended in PBS with protease inhibitors, sonicated, and centrifuged for 4 min at 6000 rpm. Supernatant was collected, and total protein concentration was defined using BCA protein assay (Pierce). ApoE/A␤ complex levels were measured using a 5-fold sample dilution according to apoE/A␤ ELISA protocol 1. The standard curve used a fixed HEK-apoE concentration of 14 nM (apoE concentration in the synaptosomes) and varied A␤ concentrations. Human data were normalized according to total protein concentration in each sample.

ELISA Analysis of Human CSF
CSF samples were obtained at autopsy at the Alzheimer Disease Center at the University of Kentucky (supplemental Table  2). Diagnoses of AD and non-AD were performed at a consensus conference of the AD Center Neuropathology and Clinical Cores and were based upon evaluation of both cognitive status, i.e. Clinical Dementia rating and Mini-Mental State Examination (MMSE) scores, as well as neuropathology, i.e. Braak stages that rate the extent of neurofibrillary pathology into the neocortex and the NIAReagan Institute neuropathology classification, which includes counts of both neuritic senile plaques and neurofibrillary tangles and provides a likelihood staging of AD neuropathological diagnosis (47,48). For ELISA analysis, A␤42, total Tau (T-Tau), and phosphorylated Tau 181 (p-Tau-181) levels were measured using Innotest ELISA kits (Innogenetics, Gent, Belgium) according to the manufacturer's protocol; apoE levels were measured using ␣-apoE (Millipore) as capture and ␣-apoE (Meridian) as detection as described (11). oA␤ levels were measured using MOAB-2 capture (5) and biotinylated MOAB-2 as detection antibody as described previously (15). ApoE/A␤ complex levels were measured using a 2-fold sample dilution according to apoE/A␤ ELISA protocol 2, with a standard curve of 5 g/ml recombinant apoE (reported CSF apoE concentration) and varied A␤ concentrations.

Statistical Analysis
Data were analyzed by one-way ANOVA followed by Tukey's post hoc analysis (Figs. 2, A and B, and 3-5) or by two-way ANOVA followed by Bonferroni post hoc analysis (Fig. 2C). Correlation analysis was conducted using Spearman's correlation (Fig. 5, E and F). All data were analyzed using GraphPad Prism version 5 for Macintosh, and p Ͻ 0.05 was considered significant. Receiver operating characteristic (ROC) curves ( Fig. 6) were constructed for each marker using the pROC package in R (49,50). Areas under the curves were compared by the method of DeLong et al. (51).

Biochemical Development of ApoE/A␤ Complex ELISA-Ini-
tially, biochemical analysis using HEK-apoE and synthetic A␤ preparations ( Fig. 1) (45) was conducted to validate the apoE/A␤ ELISA and to determine the effect of apoE isoform on soluble apoE/A␤ levels and stability.
Total ApoE/A␤ Complex Levels Are Not Affected by ApoE Isoform-Total complex levels were measured in samples containing a fixed apoE concentration (30 nM) and a varied concentration of unaggregated A␤42 (0.15-150 nM) (Fig. 1A) or using varied apoE concentration (0 -1500 nM) and a fixed A␤ concentration (3 nM) (Fig. 1B). Overall, apoE/A␤ levels were saturable and dependent on apoE and A␤ concentrations but not apoE isoform. Indeed, when these data were analyzed using a fourparameter logistic equation, which is appropriate for analyzing ELISA saturation curves (52), there were no differences between the calculated EC 50 values among the apoE isoforms (ϳ3 nM for A␤ in Fig. 3A and ϳ30 nM for apoE in Fig. 1B). Total apoE/A␤ levels were also equivalent for apoE2, apoE3, and apoE4 with unaggregated A␤40, oA␤42, and fibrillar A␤42 (data not shown). Thus, apoE isoform does not determine total apoE/A␤ levels biochemically.
ApoE2/A␤ and ApoE3/A␤ Complex Exhibits Greater Stability than ApoE4/A␤ Complex-As apoE isoform did not affect total complex levels when assessed by ELISA, SDS was added to samples as a measure of stability (Fig. 1C). ApoE and A␤ were incubated for 2 h at concentrations that correspond to the EC 50 values identified for total apoE/A␤ levels, specifically 3 nM A␤42 and 30 nM apoE, and then SDS was added over a range of concentrations (up to 2%). Complex stability from highest to lowest was apoE2/A␤ Ͼ apoE3/A␤ Ͼ apoE4/A␤. This was evident as the SDS IC 50 value was 1.5-fold higher for apoE2/A␤ complex and 3-fold lower for the apoE4/A␤ complex compared with the apoE3/A␤ complex. In addition to SDS, the apoE4/A␤ complex was less stable at mildly acidic pH (5) than apoE2/A␤ and apoE3/A␤ complex (Fig. 1D). Therefore, apoE/A␤ levels were not determined by the apoE isoform; however, the apoE4/A␤ complex is less stable, and the apoE2/A␤ is more stable than the apoE3/A␤ complex.
Soluble ApoE/A␤ Complex Levels in EFAD Mice-To determine the effect of APOE genotype on soluble apoE/A␤ levels, the tractable EFAD mouse model was utilized. For this study, apoE/A␤ levels were measured in the soluble hippocampal homogenates from EFAD mice at 6 months (Fig. 2), an age where soluble oA␤ levels are greater in E4FAD (APOE4) compared with E2FAD (APOE2) and E3FAD (APOE3) mice (15).

ApoE/A␤ Complex ELISA Optimization in EFAD Mice-Ini-
tially, soluble apoE/A␤ detection by ELISA was validated using E3FAD mice at 6 months ( Fig. 2A). For a quantitative standard to enable cross-plate comparisons, the complex formed between HEK-apoE3 at a fixed concentration of 140 nM, which corresponds to soluble apoE levels in EFAD mice at 6 months, and a varied concentration of unaggregated A␤42 was utilized. Specific soluble apoE/A␤ levels were detected by ELISA as follows. 1) Soluble hippocampal apoE/A␤ was only detected using  MOAB-2 as a capture antibody, as no signal was seen when using a nonspecific IgG 2b isotype-matched capture antibody. 2) Complex levels decreased with increased sample dilution. 3) There were no detectable soluble complex levels in the cerebellum, a region spared of A␤ pathology in EFAD mice. These data validate soluble apoE/A␤ detection in vivo by ELISA.
Soluble ApoE/A␤ Complex Levels Are Lower and Less Stable with APOE4-Next, the effect of APOE genotype on soluble apoE/A␤ levels and stability was determined. ApoE/A␤ complex levels were 50% lower in E4FAD mice compared with E2FAD and E3FAD mice (Fig. 2B). For apoE/A␤ stability (Fig.  2C), complex levels were measured from the same sample in the presence of 0, 0.02, or 0.2% SDS. Complex levels were normalized to the 0% SDS control for each paired samples set. The addition of SDS reduced apoE/A␤ levels, in an APOE genotypespecific manner. With 0.02% SDS, apoE4/A␤ complex levels were reduced by ϳ60%, apoE3/A␤ complex levels by ϳ50%, and apoE2/A␤ complex levels by ϳ30%. The addition of 0.2% SDS further lowered complex levels in E3FAD and E4FAD mice but not E2FAD mice. These data demonstrate that soluble hippocampal apoE/A␤ levels are lower and less SDS-stable in E4FAD mice compared with E3FAD and E2FAD mice and that the apoE2/A␤ complex is more stable than the apoE3/A␤ complex.
Soluble ApoE/A␤ Complex in Synaptosomes-To determine the effect of AD and APOE genotype on soluble synaptic apoE/A␤ levels, cortical synaptosomes were isolated from control (APOE3/3 and APOE4/X) and AD patients (APOE3/3 and APOE4/X) (Fig. 3).
ApoE/A␤ Complex Levels Were Lower in AD Patients Compared with Controls and with APOE4 in the AD Cohort-In the absence of SDS (Fig. 3A), the data are normalized to APOE3/3 controls. In the control individuals, there was no significant difference between apoE/A␤ levels in the APOE3/3 and APOE4/X. In the AD patients, apoE/A␤ levels were significantly lower, i.e. 70% lower for APOE3/3 AD patients compared with APOE3/3 controls and 90% lower in APOE4/X AD patients compared with APOE4/X controls (Fig. 3A). In addition, within the AD cohort, total apoE/A␤ levels were 66% lower with APOE4/X compared with APOE3/3.
To address the SDS stability of the apoE/A␤ (Fig. 3B), complex levels were measured from the same sample in the presence of 0 or 0.02% SDS. ApoE/A␤ complex levels were then normalized to the 0% SDS control for each paired sample set. In the control individuals, the addition of SDS results in a significant decrease in apoE/A␤ levels with APOE4/X compared with APOE3/3. ApoE/A␤ complex stability was not significantly different in the APOE4/X AD patients compared with APOE3/3 AD patients. This is primarily due to the very low levels of complex present in APOE4/X AD samples in the absence of SDS (Fig. 3A); thus, after the addition of SDS, any further reduction results in values for the complex that are at the limit of detection for this ELISA. Again, this results from the pairwise comparison between apoE/A␤ levels in APOE4/X AD patients in the absence of SDS (for example Fig. 3A), where the levels of apoE/A␤ are already low, and apoE/A␤ levels in the presence of SDS (Fig. 3B).
ApoE/A␤ Complex in Human CSF-The CSF is as an indication of the concentration of soluble proteins in the brain parenchyma. Therefore, the hypothesis that reduced levels of soluble apoE/A␤ and an increase in soluble oA␤ levels are associated with AD and APOE4 was tested in post-mortem CSF samples from control (APOE3/3) and AD patients (APOE3/3 and APOE4/4).
This ELISA was used to determine whether oA␤ levels were influenced by AD diagnosis or APOE genotype. oA␤ levels were significantly increased in AD patients compared with controls, and importantly, oA␤ levels were significantly greater in APOE4/4 AD patients compared with APOE3/3 AD patients (Fig. 4B). For comparative purposes, the established AD biomarkers A␤42 (Fig. 4C), total Tau (T-Tau) (Fig. 4D), and phosphorylated Tau 181 (p-Tau-181) (Fig. 4E) levels were measured by ELISAs (54) in the same samples as oA␤. As expected, A␤42 levels were significantly lower, and p-Tau-181 levels significantly greater in both the AD groups (APOE3/3 and APOE4/4) compared with age-matched controls (APOE3/3). T-Tau levels were significantly greater in the APOE4/4 AD patients but not the APOE3/3 AD patients compared with the APOE3/3 controls. Of particular interest, in the AD group, APOE genotype did not affect the levels of A␤42, T-Tau, or p-Tau-181, all established AD biomarkers. Thus, oA␤ levels may play a role in AD progression, in an APOE genotype-specific manner.
ApoE/A␤ Complex ELISA Optimization for Human CSF-An important consideration for measuring specific proteins in the CSF by ELISA is a standard to allow quantification of samples used on different microplates, across studies, so as to allow future retrospective analysis. Although HEK-apoE is an important apoE source for biochemical studies, the relatively long sample preparation time, potential intra-laboratory differences in production quality, long term stability issues, and lack of commercialization hinders routine use as an apoE/A␤ standard. Therefore, apoE/A␤ formed between recombinant apoE3, at concentrations corresponding to those in human CSF (5 g/ml), and varied unaggregated A␤42 concentrations was used as a standard curve for the CSF samples (Fig. 5A). Specific apoE/A␤ detection by ELISA in human CSF (APOE3/3 control) was demonstrated by a high signal with capture antibody (MOAB-2) that decreased proportionately to sample dilution and no observed signal without a capture antibody, all using high bind plates to increase sensitivity (Fig. 5B).
ApoE/A␤ Complex Levels Were Lower in CSF from AD Patients Compared With Controls and, Importantly, Significantly Lower with APOE4 within the AD Cohort-ApoE/A␤ complex levels may modulate soluble oA␤ levels in the CNS. The increased oA␤ levels by AD and APOE4 raised the important question of what was the effect on apoE/A␤ levels. ApoE/A␤ complex levels were significantly lower in both AD cohorts compared with the control group (Fig. 5C). Importantly, complex levels were significantly lower in APOE4/4 AD patients compared with the APOE3/3 AD patients. Thus, APOE4 did affect the levels of oA␤ (increased) and apoE/A␤ (decreased) in the AD cohort and did not affect the levels of A␤42, T-Tau, or p-Tau-181, suggesting oA␤ and apoE/A␤ levels may play a role in AD progression. Although apoE levels were lower in AD patients compared with controls (Fig. 5D), there was no correlation between apoE/A␤ levels and either apoE (Fig. 5E, Spearman's r value 0.27, p ϭ 0.25) or A␤42 (Fig.  5F, Spearman's r value 0.04, p ϭ 0.87) levels in the AD patient sample set. This suggests that the levels of apoE/A␤ are independent of the values of its two components. Thus, apoE/A␤ levels are affected by both AD and APOE genotype. oA␤ and ApoE/A␤ Complex as AD Biomarkers-In addition to a potential mechanistic interpretation for AD progression, oA␤ and apoE/A␤ levels may act as AD biomarkers. As APOE genotype affects oA␤ and apoE/A␤ levels, the optimal method for assessing the diagnostic potential of these markers is analysis in control and AD patients with the APOE3/3 genotype. ROC curves were utilized to determine the predictive accuracy of each marker (Fig. 6). The ROC curves are constructed by varying the threshold to classify predicted AD cases and controls. Predicted probabilities of being an AD case are calculated from marginal logistic regression models, and sensitivity (the proportion of AD cases correctly predicted, y axis) and specificity (the proportion of AD controls correctly predicted, i.e. true negative rate, x axis) are calculated based on each subject's predicted case probability being above or below the varying threshold, respectively. The area under the curve (AUC) of the ROC curves is calculated, and an AUC of 0.5 demonstrates no information or diagnostic ability, whereas the higher the AUC is above 0.5, the greater the diagnostic accuracy of the biomarker. ROC analysis demonstrated the potential use of both oA␤ and apoE/A␤ as AD biomarkers, with AUCs of 0.7 and 0.875, respectively. Next, the ROC AUCs of oA␤ and apoE/A␤ were compared with the traditional AD biomarkers as follows: A␤42, p-Tau-181, and T-Tau. A␤42 was significantly more predictive of AD (p Ͻ 0.05) compared with the other markers except apoE/A␤ (p ϭ 0.14). Both oA␤ (p ϭ 0.70) and apoE/A␤ (p ϭ 0.41) complexes were as predictive for AD as p-Tau-181, and apoE/A␤ was more predictive than T-Tau (p Ͻ 0.012). Overall, in control and AD patients with the APOE3/3 genotype, the estimated predictive ability for AD based on the AUC values for each marker was A␤42 (AUC ϭ 0.98) Ն apoE/A␤ (0.875) Ͼ p-Tau-181 (0.775) Ն oA␤ (0.7) Ͼ T-Tau (0.59).

DISCUSSION
For this study, a quantitative apoE/A␤ ELISA was developed and characterized biochemically and then applied in vivo to determine the effect of the APOE genotype and AD on soluble levels of apoE/A␤ complex and oA␤. Soluble levels of oA␤ are higher, and apoE/A␤ are lower with AD and specifically APOE4.
Biochemical data using HEK-apoE demonstrate that the apoE isoform does not affect total levels of apoE/A␤ but that the apoE4/A␤ complex is less stable than the apoE3/A␤ complex. By measuring total and SDS-stable apoE/A␤ levels, these results resolve previous contradictory in vitro findings (21,28,30,43). HEK-apoE has been utilized in numerous studies for apoE/A␤ formation (21,28). Previous data have demonstrated that HEK-apoE3 but not HEK-apoE4 forms an SDS-stable apoE/A␤ as measured using Western analysis of SDS-PAGE (21,28,30,43). However, a corresponding value for total complex levels was not possible by Western analysis. With this new ELISA, the apoE isoforms exhibit a comparable affinity for A␤ in the absence of SDS, defined here as total apoE/A␤, consistent with previous reports using nonstringent conditions to measure the complex (31). The mechanism by which the apoE4/A␤ complex is less stable is unclear. ApoE4/A␤ disruption can occur by global effects on protein structure, disrupting the binding sites on the individual proteins, as well as local effects at the complex interface. ApoE4 has a lower stability and increased propensity to populate an intermediate molten globule conformation compared with the other isoforms (55,56). The apoE4/A␤ complex exhibits the lowest stability under all denaturing conditions, potentially due to the greater susceptibility of the apoE4 tertiary structure to disruption. Therefore, specific effects on the complex interface cannot be separated from effects on the stability/structure of the individual components apoE and A␤. An additional consideration for the stability of the apoE/A␤ is the effect of apoE isoform on lipoprotein lipidation. Increased lipoprotein lipidation increases the levels of SDS-stable apoE/A␤, as determined by Western analysis of SDS-PAGE (21,22,30,43,57,58). If glial cell-derived apoE4 is less lipidated than apoE3, the apoE4/A␤ complex would be less stable (59,60). Thus, the biochemical development and characterization of the ELISA have resolved some of the inconsistencies in the apoE/A␤ literature. Having validated the ELISA in vitro, we used it in vivo to address the hypothesis that the levels of soluble apoE/A␤ isoform specially modulate oA␤ levels.
Soluble oA␤ levels are thought to include the proximal neurotoxic A␤ assemblies in AD (61). Soluble oA␤ assemblies are neurotoxic in vitro and in vivo (61), and both soluble A␤ and oA␤ correlate with disease progression in AD patients (62)(63)(64)(65). We previously demonstrated that soluble levels of total A␤42 and oA␤ were increased in E4FAD transgenic mice compared with E2FAD and E3FAD, although the levels of apoE were comparable, suggesting a functional difference between the isoforms. In this study, soluble apoE4/A␤ complex levels were lower than apoE2/A␤ and apoE3/A␤ complex levels in EFAD mice. These data indicate an inverse association between apoE/A␤ and oA␤ levels and are consistent with previous publications that suggest that apoE/A␤ levels isoform-specifically modulate soluble A␤ (11,15). Synapse degeneration is considered a proximal cause of cognitive deficits in AD. ApoE/A␤ complex levels may affect synaptic A␤ levels and function. At the synapse, as with the whole brain, apoE/A␤ appears to be present as an insoluble and soluble form. IHC co-localization of apoE and A␤ at the synapse is a measure of primarily insoluble apoE/A␤ (35,36), and data demonstrate insoluble apoE/A␤ appears to form preferentially with apoE4 compared with apoE3 (36). Previous results have shown that the detergent and guanidine extraction pattern of mouse apoE parallels that of A␤42 in 5ϫFAD mice (5), and the proportion of apoE/A␤ in insoluble fractions was increased in AD synaptosomes compared with controls. 6 Importantly, insoluble apoE4/A␤ complex may accumulate in autophagic structures within synaptic terminals (66,67). However, the effect of APOE genotype on soluble synaptic apoE/A␤ levels is less clear. Soluble A␤, soluble oA␤, p-Tau, and SDS-stable p-Tau oligomers (37)(38)(39)68) are detected in AD synaptosomes, and data presented herein also demonstrate the presence of soluble apoE/A␤ in AD synaptosomes, with levels reduced compared with controls. Soluble apoE/A␤ levels were also lower in synaptosomes from AD patients with APOE4 compared with APOE3. These data indicate a difference in apoE/A␤ solubility during disease progression, which may lead to alterations in synaptic A␤ trafficking or clearance.
Although the cellular process by which the apoE isoform modulates soluble A␤ pathology is unclear, a number of apoE/ 6 K. H. Gylys, unpublished observations. FIGURE 6. AD prediction by A␤42, oA␤, apoE/A␤, T-Tau, and p-Tau-181 in human CSF using ROC curves. ROC curves for A␤42, oA␤, apoE/A␤, T-Tau, and p-Tau-181 in control and AD patients with the APOE3/3 genotype in human CSF are shown. ROC curves represent the predicted probabilities of being an AD case using marginal logistic regression models. Specificity (true negative rate, the proportion of AD controls correctly predicted) is plotted on the x axis and sensitivity (the proportion of AD cases correctly predicted) is plotted on the y axis, as calculated based on each subject's predicted case probability being above or below the varying threshold, respectively.
A␤-based mechanisms have been proposed. Examples include effects on the following: 1) A␤ oligomerization (69,70); 2) A␤ clearance via glia (71)(72)(73)(74), neurons (75,76), and/or the bloodbrain barrier (77,78); 3) enzymatic degradation (57); and 4) drainage via the interstitial fluid (ISF) (16) or perivasculature (79). Furthermore, the dynamic compartmentalization of A␤ in the CNS has been identified as an important factor in regulating the level of soluble A␤ (80), which may be affected by and/or affect apoE/A␤. Importantly, the levels of A␤ in each compartment affect the equilibrium between compartments, again with further modulation by the apoE isoforms. To understand this process, new techniques have been developed to determine apoE and A␤ turnover via stable isotope-labeling kinetics (81) and microdialysis of ISF (80). In addition, Hong et al. (80) have recently identified A␤ in biochemically distinct compartments in the brain, including an ISF pool, a TBS-extractable pool, an SDS-extractable pool, and an insoluble or plaque pool. In FAD-Tg mice with a high plaque burden, ISF A␤ appears to be rapidly sequestered in a TBS-soluble pool (80). Overall, the majority of A␤ in the ISF originates from a less soluble parenchymal A␤ pool rather than from production (80). ApoE isoform-specific apoE/A␤ levels could affect the dynamic compartmentalization of A␤ through the mechanisms discussed above. For example, with APOE3, high levels of soluble apoE/A␤ may reduce soluble and oA␤ levels via clearance. With APOE4, low levels of soluble apoE/A␤ may result in increased soluble A␤ levels, particularly oA␤. Alternatively, if apoE is acting as a pathological chaperone for soluble A␤, reducing the level or stability of apoE/A␤ may decrease oA␤ levels (24,82). As described herein, the ability to detect apoE isoform-specific differences in the levels of soluble oA␤ and apoE/A␤ levels in vivo is a critical step in identifying the mechanism by which the apoE isoforms modulate soluble A␤ pathology.
As with human synaptosomes, in human CSF levels of soluble oA␤ were greater and apoE/A␤ lower with APOE4 compared with APOE3 in the AD cohort. The ability of both oA␤ and apoE/A␤ to distinguish between APOE3/3 and APOE4/4 AD patients is consistent with the increased risk and earlier age of disease onset with APOE4, highlighting the potential for these markers to track disease progression. In addition, apoE/A␤ and oA␤ may represent novel CSF biomarkers, an important focus for AD research (54). In control and AD patients with the APOE3/3 genotype, both oA␤ and apoE/A␤ diagnosed AD with the same accuracy as p-Tau-181, a currently accepted AD biomarker. Furthermore, as oA␤ and apoE/A␤ are based on potential mechanisms of AD progression, both represent biomarkers to assess therapeutic efficacy in vivo and in clinical trials. Currently, drug trials targeting oA␤ and apoE/A␤ are either in the preclinical phase or underway. For A␤, therapies include both passive and active A␤ immunotherapy, ␤-secretase inhibitors, ␥-secretase inhibitors, and ␥-secretase modulators. Aside from measures of cognition, neuroimaging for amyloid with Pittsburgh compound B and CSF biomarkers such as A␤42 and p-Tau levels are the only biomarkers available to determine drug efficacy (83). Given that amyloid plaques appear not to correlate with dementia and may not represent the ideal target, and it is unclear whether low CSF A␤42 levels will be reversible with long term treatment, the relevance of these biomarkers for therapeutic trials is unclear. oA␤ levels represent a novel biomarker to monitor drug efficacy. For apoE/A␤, therapies are in development that disrupt (24,82) or increase (84) apoE/A␤ levels. Examples include retinoid X receptor, liver X receptor, and peroxisome proliferator-activated receptor-␥ agonists, which increase the levels and lipidation state of apoE (57, 84 -86), apoE structural correctors (87), and A␤12-28P that blocks apoE/A␤ interactions (24,82). However, there are no data on whether these drugs will affect apoE/A␤ levels in vivo, and more importantly, the effects of these therapeutic interventions on the human apoE isoforms are unknown. The data presented here indicate that increasing soluble levels of apoE/A␤ is a therapeutic target, as it will reduce oA␤ levels. Importantly the efficacy of therapeutic treatments targeting soluble levels of oA␤ and apoE/A␤ can now be determined using the ELISAs described herein.