Amyloid-β Pathology and APOE Genotype Modulate Retinoid X Receptor Agonist Activity in Vivo*

Background: Human APOE effects on RXR agonist activity are unclear. Results: In RXR agonist-treated EFAD mice, beneficial effects in APOE4 hippocampus include ABCA1/ABCG1-induced apoE lipoprotein association/lipidation. Detrimental effects in APOE3 and APOE4 cortex might be from systemic hepatomegaly. Conclusion: Aβ pathology and APOE genotype modulate RXR agonist effects. Significance: Although promising for later stage Alzheimer disease, detrimental side effects limit translational application of RXR agonists. Previous data demonstrate that bexarotene (Bex), retinoid X receptor (RXR) agonist, reduces soluble and insoluble amyloid-β (Aβ) in Alzheimer disease (AD)-transgenic mice either by increasing the levels of mouse apolipoprotein E (apoE) or increasing ABCA1/ABCG1-induced apoE lipoprotein association/lipidation. However, although the mechanism of action of RXR agonists remains unclear, a major concern for their use is human (h)-APOE4, the greatest AD genetic risk factor. If APOE4 imparts a toxic gain-of-function, then increasing apoE4 may increase soluble Aβ, likely the proximal AD neurotoxin. If the APOE4 loss-of-function is lipidation of apoE4, then induction of ABCA1/ABCG1 may be beneficial. In novel EFAD-Tg mice (overexpressing h-Aβ42 with h-APOE), levels of soluble Aβ (Aβ42 and oligomeric Aβ) are highest in E4FAD hippocampus (HP) > E3FAD-HP > E4FAD cortex (CX) > E3FAD-CX, whereas levels of lipoprotein-associated/lipidated apoE have the opposite pattern (6 months). In E4FAD-HP, short-term RXR agonist treatment (Bex or LG100268; 5.75–6 months) increased ABCA1, apoE4 lipoprotein-association/lipidation, and apoE4/Aβ complex, decreased soluble Aβ, and increased PSD95. In addition, hydrogel delivery, which mimics low sustained release, was equally effective as gavage for Bex and LG100268. RXR agonists induced no beneficial effects in the E4FAD-HP in a prevention protocol (5–6 months) and actually increased soluble Aβ levels in E3FAD-CX and E4FAD-CX with the short-term protocol, possibly the result of systemic hepatomegaly. Thus, RXR agonists address the loss-of-function associated with APOE4 and exacerbated by Aβ pathology, i.e. low levels of apoE4 lipoprotein association/lipidation. Further studies are vital to address whether RXR agonists are an APOE4-specific AD therapeutic and the systemic side effects that limit translational application.

Alzheimer disease (AD), 2 the most common form of dementia, has reached epidemic proportions and is a serious economic and social burden worldwide (1). Soluble forms of the amyloid-␤ peptide (A␤42) (2)(3)(4) are considered a proximal cause of impaired synaptic function in AD, particularly soluble oligomeric A␤ (oA␤) (5)(6)(7)(8). Unfortunately, therapeutic strategies targeting A␤42 production or clearance have not yet proved to be viable due to both efficacy and safety issues (9,10). Thus, there is a drive to develop novel therapeutic approaches targeting A␤. Apolipoprotein E (apoE)-containing lipoproteins, produced primarily by glia, represent a promising physiological target for modulating A␤ levels in the CNS. Cramer et al. (11) identified bexarotene (Bex), a retinoid X receptor (RXR) agonist developed as an anti-cancer drug, as a potential AD therapeutic candidate to increase apoE levels. Indeed, Bex increased mouse (m)-apoE levels, decreased soluble A␤ within hours, and reduced insoluble A␤ after 3 days in transgenic mouse models expressing familial AD mutations (FAD-Tg) (11). However, recent data in multiple FAD-Tg models, although variable (12), demonstrate that Bex has no effect on insoluble A␤, but may lower soluble A␤42 and soluble oA␤ levels (13)(14)(15)(16). In addition, these recent studies indicate that rather than increasing APOE (gene coding for apolipoprotein E) expression, Bex may lower soluble A␤ through increasing the levels of other RXR target proteins, specifically, the ATP-binding cassette transporters (ABC) ABCA1 and ABCG1 (13)(14)(15)(16). As ABCA1 and ABCG1 function to transport lipids to apoE-containing lipoproteins (17), increased apoE lipoprotein association/lipidation may be the mechanism for reducing soluble A␤ levels (18 -24). Thus, there is a critical need to address confounding factors that may interact with the activity and mechanism of action of RXR agonists leading to inconsistent data in FAD-Tg models, particularly as Bex is currently under consideration for clinical trials.
The major concern for the use of RXR agonists in AD is human (h)-APOE4. APOE4 is the greatest genetic risk factor for sporadic AD, increasing risk up to 15-fold compared with APOE3 (25,26). Despite representing more than half of all AD patients, APOE4 carriers are often omitted from clinical trials because they fail to respond or respond anomalously (27)(28)(29)(30). Although the impact of h-APOE on CNS function is multifactorial, compared with APOE3, APOE4 increases insoluble and soluble A␤ levels in FAD-Tg mice (31) and in AD patients (32)(33)(34). However, the effects of RXR agonists on A␤ levels in the presence of h-APOE are unclear. The critical unresolved question is whether RXR agonists act via increasing apoE levels or increasing the ABCA1/G1-dependent lipidation of apoE. Overall, concern centers on whether APOE4 imparts a toxic gain-offunction, and thus Bex-induced increased apoE4 levels may actually increase A␤ levels or, if the APOE4 loss-of-function is lipidation of apoE4, induction of ABCA1/ABCG1 may then be beneficial. However, the majority of published data tests Bex in FAD-Tg mice expressing m-apoE (11, 12, 14 -16), which is structurally and functionally distinct from any of the human isoforms. In the context of h-APOE, limited data in one FAD-Tg model (APOE/APP/PS1) demonstrate that Bex lowers soluble A␤ levels for APOE3 and APOE4 as assessed in the interstitial fluid by microdialysis, but not when measured using biochemical extractions (13). Furthermore, APOE/APP/PS1 mice display minimal A␤ accumulation (13), and the levels of A␤ pathology at the time of Bex treatment may determine the overall response. In APOE-TR mice, which do not express h-A␤, Bex reverses the lipidation deficiency of APOE4 (35). Given the conflicting data on Bex treatment in FAD-Tg mice expressing m-apoE (12), it is vital, from a translational perspective, to further dissect the interactive effects of h-APOE and RXR agonists on A␤ accumulation in an in vivo model that expresses h-APOE.
We recently developed novel EFAD mice, which express 5ϫFAD mutations and h-APOE3 or h-APOE4 and are a tractable model of APOE genotype-specific A␤ accumulation (31). In EFAD mice, total brain apoE levels are lower with APOE4. However, using a three-step sequential protein extraction protocol, TBS, TBSX (TBS containing Triton X-100) and formic acid (FA), the reduced levels of apoE4 are seen only in the TBSX fraction, indicating that apoE4 is less lipoprotein-associated and/or less lipidated (31). Additional data from our group and others support the pathway that the reduced lipoprotein association/lipidation of apoE4 (31,35) results in lower levels of apoE4/A␤ complex and higher levels of oA␤ (31,33), compromising synaptic viability. Thus, EFAD mice allow the preclinical analysis of RXR agonists in the presence of h-APOE.
There are further issues related to RXR agonists such as Bex as an AD therapeutic. Bex is a synthetic retinoic acid analogue, classed as a pan-RXR agonist, and is currently approved for treat-ment of cutaneous T-cell lymphoma. Adverse events, including severe but reversible hypertriglyceridemia (36 -39), have prompted the design of improved Bex analogues, and LG100268 (LG) was first reported in 1995 as delivering improved potency and selectivity for RXR over other nuclear receptors (NR), particularly the retinoic acid receptors (RAR) (40), although further improvement is needed (41). In the livers of rodents and humans, Bex modulates metabolic enzymes associated with drug-drug interactions and induces hepatomegaly (14,42), which may outweigh any beneficial effects in the CNS, particularly in an elderly population. Comparison of Bex and LG is essential to confirm that any actions of Bex represent an effect of the RXR agonist drug class, rather than any off-target effect; furthermore, LG and later generation RXR agonists may prove safer and potentially more effective than Bex for AD therapy.
In this study, EFAD mice (31) were treated with the RXR agonists Bex and LG and indirect target engagement, mechanistic pharmacodynamic (PD), PD, and efficacy markers were measured. Overall, the data demonstrate that A␤ levels and APOE genotype determine the effect of RXR agonists on PD and efficacy readouts. In untreated EFAD mice, levels of soluble A␤ (A␤42 and oA␤) are highest in hippocampus (HP) of E4FAD mice and follow the pattern E4FAD-HP Ͼ E3FAD-HP Ͼ E4FAD-cortex (CX) Ͼ E3FAD-CX, although the levels of lipoprotein-associated/lipidated apoE have the opposite pattern. In the E4FAD-HP, short-term (5.75-6M) Bex and LG treatment support the hypothesis that RXR agonists increase ABCA1/G1 levels, apoE4 lipoprotein association/lipidation, and apoE/A␤ complex levels, which results in decreased soluble A␤ and increased PSD95 levels. In addition, hydrogel delivery, which mimics low sustained release, was equally effective as gavage for Bex and LG in lowering soluble A␤ with APOE4. However, in the 30-day prevention protocol, no beneficial effects were observed in the E4FAD-HP (5-6M). Furthermore, in E3FAD-CX and E4FAD-CX, brain regions with the lowest A␤ levels at the time of short-term treatment, RXR agonists actually increased soluble A␤ levels in the absence of any changes in ABCA1 levels or apoE lipoprotein association/lipidation. The lack of a response with RXR agonists in E4FAD-HP during prevention and the induction of detrimental CNS effects in the CX in short-term treatment may be the result of systemic side effects, particularly hepatomegaly, which have been observed in models of acute and chronic liver disease (43)(44)(45)(46). Collectively, these data suggest that Bex and LG can address a loss-of-function associated with APOE4 and exacerbated by A␤, i.e. lower apoE lipoprotein association/lipidation. However, future studies are vital to determine whether this pathway is relevant for APOE3 carriers or whether RXR agonists are an APOE4-specific AD therapeutic. In addition, addressing the systemic side effects of RXR agonists is important for their translational application in longer term AD prevention and treatment trials.
Pharmacokinetics-Drug delivery was identical to that used for EFAD mouse studies. 3-Month-old male C57BL/6 mice (Charles River, Wilmington, MA) were treated with Bex (100 mg/kg) or equimolar LG (104.3 mg/kg) by gavage or hydrogel. At various time points between 0 and 960 min for gavage ( Fig. 1, A and B) and at 1, 4, and 24 h for hydrogel treatment (Fig. 1C), mice were sacrificed using CO 2 ; plasma was immediately drawn from the dorsal aorta, and mice were intracardially perfused, decapitated, and brain hemispheres flash-frozen. Levels of Bex and LG in the plasma and brain tissues of were measured by LC-MS/MS analysis. Working internal standard solutions were prepared by serially diluting DMSO stock solutions of Bex or LG (10 mM) in MeCN to 10 -2000 nM. Working internal standard solutions were freshly prepared in MeCN from DMSO stocks to a working concentration of 50 nM Bex and 12.5 nM LG. Bex was used as the internal standard for LG and vice versa. Analysis was carried out on commercial C57BL/6 blank plasma and perfused brain (BioreclamationIVT, New York) for calibration and quality control. Blank plasma samples (45 l) were transferred into low retention Eppendorf tubes stored on ice, mixed with 5 l of standard solution, and 200 l of ice-cold internal standard solution in MeCN. Blank perfused brains were homogenized in 3 volumes of 30% MeOH using a disposable pestle. Homogenized samples (145 l) were mixed with 5 l of standard solution and treated with ice-cold internal standard solution in MeCN (600 l). All samples were centrifuged at 13,000 ϫ g for 10 min at 4°C. The supernatant was removed, dried with N 2 , and reconstituted with 30% MeOH (100 l). After centrifuging at 13,000 ϫ g for 10 min at 4°C, an aliquot of 25 l was analyzed by LC-MS/MS. Analysis of plasma (50 l) and brain samples (150 l) from treated animals followed the same procedures. 30-day PK hydrogel treatment measurements were made on E3FAD and E4FAD mice using plasma samples obtained at sacrifice. Samples were analyzed on a QTRAP 5500 system equipped with an Agilent 1260 UHPLC, using an Acquity BEH C18 column (3.0 ϫ 100 mm, 1.7 m, at 40°C) with a flow rate of 0.45 ml/min (mobile phase 50% B to 100% B over 8.5 min: A ϭ 30% MeOH containing 5 mM NH 4 HCO 2 ; B ϭ 95% MeCN containing 5 mM NH 4 HCO 2 ). Measurements were made in the negative ion multiple reaction-monitoring mode at m/z 348.48 for Bex and 363.49 for LG.
Tissue Harvest-Brain and liver tissue was harvested as described previously (31). EFAD mice were anesthetized with sodium pentobarbital (50 mg/kg) and transcardially perfused with ice-cold PBS containing protease inhibitors (Calbiochem, set 3). Right hemi-brains were dissected on ice into CX, HP, and cerebellum, snap-frozen in liquid nitrogen, and stored at Ϫ80°C until use.
Protein Extraction-Serial extraction of soluble detergent and insoluble proteins was performed as described previously (31,52). Briefly, the dissected brains were homogenized in 15 volumes (w/v) of Tris-buffered saline (TBS) and centrifuged (100,000 ϫ g, 1 h at 4°C), and the TBS-soluble extract was frozen in liquid nitrogen and stored at Ϫ80°C. The pellet was washed in TBS, resuspended in 15 volumes of TBS containing 1% Triton X-100 (TBSX), mixed by rotation at 4°C for 30 min, and centrifuged (100,000 ϫ g, 1 h at 4°C), and the TBSX-soluble fraction was stored as for TBS. Pellets were washed with TBSX, weighed, resuspended to 150 mg/ml in FA, and rotated end over end for 2 h at room temperature. Finally, samples were centrifuged (100,000 ϫ g, 1 h at 4°C), and the supernatant was neutralized using 20 volumes of Tris base and stored as for TBS/TBSX. Total protein was measured in the TBS and FA (QuickStart TM Bradford assay, Bio-Rad) and TBSX (Micro BCA protein assay kit, Pierce) fractions.
ELISAs-ApoE and A␤42 levels were measured in the TBS, TBSX, and FA extracts, and oA␤ and apoE/A␤ complex in the TBS extracts by ELISA. ApoE ELISA utilizes ␣-apoE (1:2000, Millipore) capture and ␣-apoE biotin-gt (1:5000, Meridian) detection antibodies as described previously (52). A␤42 levels were measured using ELISA kits from Invitrogen (KHB3441). oA␤ levels were measured using MOAB-2 based ELISA kits from Biosensis (BEK-2215-1P). ApoE/A␤ complex levels were measured based on a modified protocol described in (52). The new protocol followed the oA␤ ELISA kit protocol (Biosensis) but used the ␣-apoE biotin-gt (1:5000, Meridian) detection antibody in place of the biotin-MOAB-2 antibody, with standards as prepared in Ref. 52. All data were normalized to total protein in each extract.
mRNA Analysis-1 ml of RNAlater solution (Ambion) was added to ϳ20 mg of brain tissue, incubated at 4°C overnight, and stored at Ϫ80°C to ensure minimal mRNA degradation. Quantitative analysis of mRNAs was performed by the Quanti-Gene 2.0 Plex assay system (Affymetrix) for LXR␣/␤ and PPAR␥ expression. The raw output is "mean fluorescence intensity" for each assay. Mean fluorescence intensity values are adjusted by blank value subtraction, and standardized with appropriate endogenous control gene expression value. Data are analyzed with MasterPlexEX software (Hitachi-Solution America).
All samples from control and RXR-treated mice within a brain region, for example E4FAD-HP, were run on the same Western blot to allow for appropriate comparison. In addition, to enable a comparison across brain regions, all the samples for the VC groups within a treatment protocol were run on a single blot, for example the VC groups for E3FAD-CX, E4FAD-CX, E3FAD-HP, and E4FAD-HP of T1. For the representative blots presented in the figures, all the lanes originate from the same gel but from different areas, i.e. the lanes were not immediately adjacent to each other. This re-organization corresponds to the specific results of a given figure.
Statistical Analysis-Data were analyzed by one-way analysis of variance (ANOVA) followed by Tukey's post hoc analysis using GraphPad Prism version 5 for Macintosh. p Ͻ 0.05 was considered significant. Correlation analysis was conducted using Pearson's correlation.

Study Design and Pharmacokinetics
RXR Agonist Treatment Protocols of EFAD Mice-To determine the interactions between h-APOE and RXR agonists, we used three treatment protocols (see Table 1). In all studies, male E3FAD and E4FAD mice (31) were orally delivered Bex (100 mg/kg/day) or the RXR agonist LG (104.3 mg/kg/day, equimolar to Bex). T1 short-term (7 day) treatment was from 5.75 to 6 months by oral gavage. For T1, the dosing, method of oral delivery, and 7-day treatment protocol match that previously reported for Bex in FAD-Tg mice (13)(14)(15)(16). Because Bex is almost insoluble in water, gavage must be delivered as a sesame oil slurry. For this reason and the reported toxicity of Bex in C57BL/6 mice (maximum tolerated dose ϳ80 mg/kg/day) (54), we also chose an alternative mode of oral drug delivery. T2 short-term (7 days) treatment was from 5.75 to 6 months by hydrogel. For water-insoluble drugs, dissolution in DMSO and delivery in hydrogel is a viable alternative, with hydrogel replacing drinking water (53,55). This short-term study was first completed to establish comparable results to drug delivery by gavage, thus informing protocol three. T3 medium-term (30 days) prevention protocol was from 5 to 6 months by hydrogel. The goal of T3 was to dissect the effect of RXR agonists in a prevention protocol.
Pharmacokinetics (PK)-Bex and LG have brain bioavailability. Initially, the plasma PK of Bex and LG was determined. In male C57BL/6 mice, after gavage delivery of Bex (100 mg/kg), the plasma values we observed were C max ϳ3000 ng/ml (8.6 M) and T max ϳ2 h (Fig. 1A). Similar values were reported for rats after bolus oral delivery of Bex, specifically T max ϳ2 h at 100 mg/kg/day p.o. (56). In a female Tg mouse strain that is susceptible to mammary cancer (C3(1)-SV40 Tag mice), gavage delivery in sesame oil, similar to our procedure, gave reported values for plasma of C max ϳ20,000 ng/ml (58 M) and T max ϳ0.5 h (57). After equimolar gavage delivery, the half-life for LG in plasma was shorter compared with Bex but with a similar C max (Fig. 1, A and B). Importantly, both were brain bioavailable with an area under the curve (brain/plasma) of 1.46 and 0.41 and a brain C max of ϳ6700 ng/ml (  Bex delivery to rodents has been primarily via feed an intermittent, continuous drug delivery similar to hydrogel. To estimate PK after hydrogel delivery, drug levels in plasma and brain were measured at 1, 4, and 24 h, and brain concentrations for Bex and LG were ϳ50 ng/ml (140 nM) and ϳ45 ng/ml (125 nM), respectively (Fig. 1C). After a 30-day delivery of Bex to EFAD mice by hydrogel, plasma concentrations were not significantly different by 7-day hydrogel (Fig. 1C). Thus, with both gavage and hydrogel, the concentrations of Bex and LG in the brain were at least 3-fold higher than the in vitro IC 50 (40,41), suggesting full agonism.

Short-term RXR Agonist Treatment Induces Beneficial Effects in E4FAD-HP (T1 and T2)
Initially, the effect of short-term (5.75-6M) gavage (T1) or hydrogel (T2) Bex and LG treatment was determined in the E4FAD-HP, the brain region with the highest A␤ pathology at the time of treatment compared with E3FAD/E4FAD-CX or E3FAD-HP (31). Indeed, the E4FAD-HP has the highest levels of insoluble A␤42 and soluble A␤ (A␤42 and oA␤), the lowest levels of apoE/A␤ complex, and the lowest lipoprotein-associated/lipidated apoE compared with other brain regions in E4FAD and E3FAD mice at 6 months (see other brain regions under "Results" and Refs. 31,52). Focusing on the E4FAD-HP allowed testing of the hypothesis that RXR agonists will address a loss-of-function associated with apoE4 in treatment protocols (Fig. 2).
RXR Agonists Increase ABCA1/ABCG1 but Not Total ApoE Levels in E4FAD-HP-Although Bex and LG were brain-penetrant with gavage and hydrogel delivery, it is important to demonstrate target engagement to interpret PD and efficacy data. Data are conflicted whether RXR agonists function through increasing the levels of ABCA1/ABCG1 or total apoE (13)(14)(15)(16). Therefore, ABCA1 and ABCG1 and total apoE levels were assessed as markers of indirect target engagement after RXR agonist treatment. Both RXR agonists increased ABCA1 levels by gavage (T1) and hydrogel (T2) treatment ( Fig. 2A). In T1, compared with VC, Bex and LG increased ABCA1 levels by ϳ200 and 300%, respectively, as assessed by Western blot (WB) analysis (TBSX extract). Similar increases were also observed in T2 with increases in ABCA1 levels of ϳ300% with Bex and 100% with LG compared with VC. Both RXR agonists also increased ABCG1 levels in T1, but not in T2 (Fig. 2B). There were no changes in total apoE levels for T1 or T2 by Bex or LG (Fig. 2C) when assessed by ELISA (total apoE ϭ sum of apoE levels in the TBS, TBSX, and FA extracts divided by total protein).
RXR Agonists Increase Detergent-soluble/Lipoprotein-associated ApoE in E4FAD-HP-A previously developed three-step sequential protein extraction protocol (TBS, TBSX, and FA) was used to determine the effect of APOE genotype on the levels of lipoprotein-associated apoE (31,52) and was used to homogenize the EFAD mice tissue in this study. TBSX employs a nonionic detergent to release apoE from lipoproteins without inducing the formation of new micelles, as can occur with ionic detergents such as SDS (31,58), and thus it indicates the levels of lipoprotein-associated apoE. In EFAD mouse brain homogenates, total brain apoE levels are lower with APOE4, but this difference is seen only in the TBSX fraction, indicating that apoE4-containing lipoproteins are less lipoprotein-associated and/or less lipidated (31). As RXR agonists increased ABCA1 and ABCG1 levels, apoE levels were plotted as extraction profiles (Fig. 2, D and E). Bex and LG increased TBSX-apoE4 levels in the E4FAD-HP by ϳ80% by gavage (T1) (Fig. 2D) and by ϳ140% using hydrogel (T2) (Fig. 2E) compared with the VC. In contrast, TBS-soluble apoE4 levels were decreased by RXR treatment by ϳ40% in both T1 and T2. Thus, although total apoE levels were not affected by RXR agonist treatment, soluble (TBS) apoE levels were decreased, and detergent-extracted apoE4 levels were increased. These data support that RXR agonists increase the lipoprotein association of apoE4.
RXR Agonists Increase ApoE/A␤ Complex Levels in E4FAD-HP-A potential consequence of RXR agonist-induced increased apoE4 lipoprotein association is increased apoE/A␤ complex levels (33,52). To determine the effect of APOE genotype on apoE/A␤ complex levels, we previously developed an apoE/A␤ complex ELISA and demonstrated that apoE/A␤ complex levels are lower with AD and APOE4 in human CSF and with APOE4 in the TBS extracts of EFAD mice (33,52). In this study, both RXR agonists increased apoE/A␤ complex levels by gavage (T1) and hydrogel (T2) by ϳ70 and 100%, respectively, compared with VC when assessed by ELISA (Fig. 2F). Thus, shortterm RXR agonist treatment increased ABCA1 levels, apoE4 lipoprotein association, and apoE/A␤ complex levels in the E4FAD-HP.
RXR Agonists Do Not Change Total A␤42 Levels but Decrease Soluble A␤ (A␤42 and oA␤) in E4FAD-HP-Currently, there is debate on the effect of RXR agonists on the levels of total A␤42, measured by immunohistochemistry or biochemically. Initial data demonstrated that Bex reduced total A␤ by ϳ60% in APP/ PS1 mice after a 7-day gavage treatment (11), although subsequent studies failed to replicate these changes in a number of mouse models (13)(14)(15)(16). In this study, total A␤42 (TBS ϩ TBSX ϩ FA) levels were unchanged by RXR agonist treatment in both short-term treatment protocols (Fig. 2G), consistent with the latter publications with Bex in FAD-Tg mice (13)(14)(15)(16).
RXR Agonists Increase PSD95 Levels in E4FAD-HP-An important read-out for the preclinical analysis of a potential AD therapeutic is a measure of neuronal viability. Increased A␤ levels are linked to reduced postsynaptic density protein 95 (PSD95) levels in humans and FAD transgenic mouse models (59 -62). Thus, in this study we utilized PSD95 levels as a marker of drug efficacy. RXR agonist treatment increased PSD95 levels in E4FAD-HP by ϳ160% with oral gavage (T1) and ϳ60% by hydrogel treatment (T2) (Fig. 2J).
Therefore, both RXR agonists in short-term treatment protocols exhibit beneficial effects in the E4FAD-HP through increasing ABCA1/G1 levels, apoE4 lipoprotein-association, and apoE/A␤ complex levels, which result in decreased soluble A␤ (A␤42 and oA␤) and increased PSD95 levels.

Medium-term RXR Agonist Prevention Protocol Induces No Effects on ApoE or A␤ in E4FAD-HP (T3)
To determine whether RXR agonists could prevent increases in A␤ pathology and decreases in PSD95 levels in the E4FAD-HP, a medium-term prevention study was conducted ending at 6 months (Fig. 3). E4FAD mice were treated with both Bex and LG from 5 to 6 months by hydrogel rather than gavage, as hydrogel represents less stress to the animal than repeated daily gavage and induced comparable beneficial PD and efficacy effects in short-term treatment in the E4FAD-HP. Bex and LG both increased ABCA1 (but not ABCG1) levels by 70% (Fig.  3A). However, despite demonstrating indirect target engagement, neither Bex nor LG exerted any changes on markers for mechanistic PD (TBSX-apoE and apoE/A␤ complex), PD (insoluble and soluble A␤), or efficacy (PSD95) (Fig. 3, B-H). Thus, the beneficial effects of RXR agonists observed in E4FAD-HP in short-term treatment protocols (from 5.75 to 6 months) are not replicated in a prevention protocol when E4FAD mice are treated at an earlier stage of A␤ pathology, but still ending at 6 months.

Short-term RXR Agonist Treatment Increases Soluble A␤ (A␤42 and oA␤) Levels in E3FAD-CX and E4FAD-CX (T1 and T2)
As data were conflicted between the treatment and prevention protocols in E4FAD-HP, the effect of short-term RXR agonist treatment in brain regions with lower levels of A␤ pathology at the time of treatment was dissected (i.e. E3FAD-CX, E4FAD-CX, E3FAD-HP). Readouts of indirect target engagement, mechanistic PD, PD for A␤, and efficacy in the VC group are presented first, as these are an indication of the levels at the time of treatment. The effects of RXR agonists are plotted for a comparison (Figs. 4 and 5 represent T1 and, as data were similar, T2 data is not shown).

RXR Agonists Elicit No Effects on ABCA1 Levels, ApoE Lipoprotein Association, or ApoE/A␤ Complex Levels in E3FAD-CX and E4FAD-CX
ABCA1-Generally, in the VC groups, ABCA1 (Fig. 4A) and ABCG1 (data not shown) levels are lower in E4FAD-HP compared with other brain regions. For example, in the VC group
ApoE/A␤ Complex-Consistent with published data, apoE/A␤ complex levels in the VC groups of both treatments were lower in E4FAD compared with E3FAD mice in both the CX and HP (Fig. 4E) (52). LG and Bex did not change apoE/A␤ complex levels in the E3FAD-CX or E4FAD-CX. In E3FAD-HP (Fig. 4F), LG decreased apoE/A␤ complex levels in T1, and Bex decreased apoE/A␤ complex levels in T2 (data not shown).

RXR Agonists Induce an Increase in Soluble A␤ (A␤42 and oA␤) in the E3FAD-CX and E4FAD-CX
Total A␤42-Consistent with previous data, total A␤42 levels in the VC followed the order E4FAD-HP Ͼ E3FAD-HP Ͼ E4FAD-CX Ն E3FAD-CX (Fig. 5A). Gavage (T1) and hydrogel (T2) treatment with Bex did not affect total A␤42 levels in the CX or HP of EFAD mice (Fig. 5B).
LG increased total A␤42 levels in the E3FAD-HP of T1-treated mice and in E3FAD-CX of T2-treated mice.
Soluble A␤42-As demonstrated previously in the VC groups, soluble A␤42 levels followed the order E4FAD-HP Ͼ E3FAD-HP Ͼ E4FAD-CX Ն E3FAD-CX (Fig. 5C). RXR agonist treatment increased soluble A␤42 levels by ϳ50% in the E3FAD-CX and E4FAD-CX (Fig. 5D). These data were significant for Bex and LG in the E4FAD-CX in T1, and LG in E3FAD-CX for T1 and T2. In E3FAD-HP, Bex increased soluble A␤42 by 100% following both short-term treatments.
Collectively, these data demonstrate that short-term RXR agonist treatment in the CX of E3FAD and E4FAD mice does not significantly change ABCA1, TBSX-apoE, or apoE/A␤ complex levels, and in the E3FAD-HP it increases ABCA1 levels without significant increases in apoE lipoprotein association. Importantly, in the E3FAD-CX and E4FAD-CX, short-term RXR agonist treatment increases soluble A␤ levels (A␤42 and oA␤), which tends to be associated with decreased PSD95 levels in the E4FAD-CX.

Medium-term RXR Agonist Prevention Protocol Induces No Effects on ApoE or A␤ in E3FAD-CX, E4FAD-CX, and E3FAD-HP (T3)
As RXR agonists increased oA␤ levels in E3FAD-CX and E4FAD-CX and had no effect in the E3FAD-HP, as for the E4FAD-HP, the effect of a medium-term prevention protocol was dissected in these regions (5-6 months, Fig. 6).
LG and Bex increased levels of ABCA1 (Fig. 6A), but not ABCG1 (Fig. 6B), in all brain regions (HP and CX) for all genotypes (APOE3 and APOE4) in EFAD mice (with exception of Bex treated E4FAD-CX). However, in general, there were no changes in TBSX-apoE levels or apoE/A␤ complex levels for any of the brain regions (Fig. 6, C and D, exceptions include Bex-induced decrease in TBSX-apoE in E3FAD-CX and LG-dependent decreased apoE/A␤ complex in E3FAD-HP). In addition, there were no effects of RXR agonist treatment on total A␤ or soluble A␤ (A␤42 and oA␤) levels in the medium-term prevention protocol (Fig. 6, E-G). For PSD95, RXR agonists decreased levels by ϳ15% in the E4FAD-CX but elicited no other effects (Fig. 6H). Thus, in a medium-term prevention study, in which treatment was initiated in brain regions with the lowest A␤ pathology at the time of treatment (lower than the short-term studies described above), despite increasing ABCA1 levels, no beneficial effects on any readouts for PD or efficacy were observed.

Collective Data Analysis and Interpretation, Beneficial Effects of RXR Agonists
Increasing A␤ Levels Are Associated with Decreased Levels of ApoE Lipidation-related Genes and Proteins, Potentiated by APOE4 -Our data raised the following important question. Why did Bex and LG induce beneficial effects on apoE lipoprotein association only in short-term studies in the E4FAD-HP? Critical to understanding this issue is that in 6-month EFAD mice, A␤ accumulation, including insoluble A␤, soluble A␤42, and soluble oA␤, are from highest to lowest as follows: E4FAD-HP Ͼ E3FAD-HP Ͼ E4FAD-CX Ͼ E3FAD-CX (Fig. 5) (31, 52). In contrast, ABCA1, ABCG1 levels, and mRNA levels of the permissive nuclear receptor heterodimer targets of RXR, namely LXR␣,␤ and PPAR␥, followed the order E3FAD-CX Ͼ E4FAD-CX Ͼ E3FAD-HP Ͼ E4FAD-HP (Fig. 7A, data from VC groups T1 and T2). Thus, the levels of apoE lipidation-related pathway genes and proteins are all decreased with increased A␤ pathology, an effect that is potentiated by APOE4. These effects may underlie the progressive decrease in lipoprotein-associated apoE with increased A␤ pathology, which is therefore the lowest in the E4FAD-HP at 6 months (Fig. 7A).

RXR Agonists May Increase Lipoprotein-associated ApoE
Only at a Critically Low Threshold-Through analyzing the effect of RXR agonists in the context of the continuum of A␤ accumulation, ABCA1 levels are increased in the E3FAD-HP FIGURE 6. Medium-term RXR agonist prevention protocol induces no effects on apoE or A␤ in E3FAD-CX, E4FAD-CX, and E3FAD-HP. EFAD mice were treated with Bex (B) (100 mg/kg/day), LG (104 mg/kg/day), or VC from 5 to 6 months by hydrogel (T3). In E3FAD-CX, E4FAD-CX, and E3FAD-HP, markers of indirect target engagement include the following: A, ABCA1 (WB and TBSX) and B, ABCG1 (WB and TBSX). The mechanistic pharmacodynamics include the following: C, TBSX-apoE (ELISA) and D, soluble apoE/A␤ complex (ELISA, TBS). The pharmacodynamics for A␤ include the following: E, total A␤42 (ELISA, TBS ϩ TBSX ϩ FA); F, soluble A␤42 (ELISA and TBS), and G, soluble oA␤ (ELISA and TBS); and the efficacy, H, PSD95 (WB was TBSX) was measured. Data were analyzed by one-way ANOVA followed by Tukey's multiple comparison post hoc analysis. n ϭ 6. *, p Ͻ 0.05 versus VC. and E4FAD-HP; however, TBSX-apoE levels are only increased in the E4FAD-HP (Fig. 7B). These data are important as follows: 1) ABCA1 levels may only be increased once the RXR targets that increase ABCA1 levels reach a critical low threshold, equivalent to E3FAD-HP; and 2) a similar "critically low threshold" of apoE lipoprotein association may also exist, equivalent to the E4FAD-HP. Once this critically low threshold is attained, RXR agonists can increase ABCA1-dependent apoE lipoprotein association and lower levels of soluble A␤. Indeed, it is striking that RXR agonist treatment lowers soluble oA␤ levels in the E4FAD-HP to those of E3FAD-HP (Fig. 7C).

Collective Data Analysis and Interpretation, Detrimental Effects of RXR Agonists
RXR Agonist Increased oA␤ Levels in the Absence of Changes in ABCA1 Levels-A further issue raised by the data is that in the brain regions with the lowest A␤ pathology at the time of short-treatment (E3FAD-CX and E4FAD-CX), RXR agonists increased oA␤ levels (Fig. 7C). Additional analysis demonstrated no significant correlation between ABCA1 and oA␤ (Fig. 8A, Pearson's r ϭ ϩ0.05) in the CX of both E3FAD and E4FAD mice (T1 and T2), indicating that the increased oA␤ levels in the CX of EFAD mice are not a result of changes in ABCA1 levels and, more broadly, CNS indirect target engagement. Thus, there is likely an alternative mechanism of action for the RXR agonist-induced increases in oA␤ levels in the CX of EFAD mice.
RXR Agonists Induced Hepatomegaly in EFAD Mice-Evidence indicates that Bex can increase liver weight (hepatomegaly), potentially via hepatic steatosis (14). Indeed, hepatomegaly is touted as an indirect indication of bioavailability and target engagement for Bex (14). Furthermore, it is unclear whether agonists that are more selective than Bex for RXR compared with RAR (i.e. LG) will negate the hepatomegaly. In this study, short-term treatment by gavage (T1) with Bex and LG increased liver weight in E3FAD and E4FAD mice by ϳ80% (Fig. 8B). The 7-day RXR agonist treatment with hydrogel (T2) also increased liver weight in E3FAD mice; LG (ϳ140%) induced a greater increase than Bex (60%) (Fig. 8C). Interestingly, liver weight was not affected by RXR agonist hydrogel treatment in E4FAD mice (although LG increased liver weight nonsignificantly by ϳ25%). However, medium-term prevention (T3) with either RXR agonist resulted in an ϳ140 and ϳ70% increase in liver weight for E3FAD and E4FAD mice, respectively (Fig. 8D). Thus, both short-term treatment and medium-term prevention protocols with RXR agonists result in an enlarged liver (hepatomegaly), a potentially negative systemic side effect. One potential interpretation is that for RXR agonists there is balance between the beneficial effects for increasing apoE lipoprotein association in the CNS versus negative effects in the liver, which may also impact CNS function (see "Discussion" below). In the absence of beneficial affects in the CNS (i.e. no critical low threshold of apoE lipoprotein association), systemic liver effects may induce a stress in the CNS that is observed here as increased oA␤ levels, and in some cases with decreased PSD95 levels (e.g. E4FAD-CX in T1/2 and T3). These systemic effects may also confound interpretation of longer term RXR agonist treatment strategies, such as the lack of beneficial effects in the E4FAD-HP observed in our mediumterm prevention study.

DISCUSSION
Developing therapeutics targeting processes that underlie AD progression is a major challenge of the 21st century. As APOE4 is the greatest AD genetic risk factor, and either decreased apoE4 levels or apoE4 lipoprotein association may result in increased A␤ levels, targeting apoE-containing lipoproteins, particularly apoE4, is an exciting therapeutic prospect. Furthermore, the repurposing of drugs to target AD-relevant pathways is considered an important approach to accelerate drug development in AD. Thus, the provocative paper by Cramer et al. (11) raised great hope in the field, as the anti-cancer drug Bex (an RXR agonist) was demonstrated to FIGURE 7. Collective analysis, beneficial effects of RXR agonist treatment. A, protein levels of ABCA1, ABCG1, and TBSX-apoE, as well as the expression (mRNA) of LXR␣, LXR␤, and PPAR␥ plotted across brain region as % E3FAD-CX for the VC control groups of T1 and T2. B, levels of ABCA1 and TBSX-apoE in the VC and RXR agonist treatment groups, Bex and LG combined (RXR), plotted across brain region as % E3FAD-CX. C, levels of oA␤ (ng/mg protein) in the VC and RXR agonist treatment groups, Bex and LG combined (RXR), plotted across brain region. n ϭ 12. *, p Ͻ 0.05 versus VC.
lower A␤ levels and improve behavior through m-apoE-dependent effects in multiple FAD-Tg mouse models. However, there is concern about the effects of Bex, and therefore RXR agonists, in the presence of h-APOE4, with subsequent reports producing conflicting results on both the activity and mechanism of action of Bex (11, 12, 14 -16). Thus, key factors that may determine the translatability of RXR agonists for AD, including the effects of h-APOE and the level of A␤ pathology at time of treatment, were assessed in vivo.
The first important finding of this study is that both RXR agonists induced beneficial effects in short-term treatment protocols in the E4FAD-HP, the brain region with the highest A␤ levels and lowest apoE lipoprotein association at the time of treatment. In the E4FAD-HP, Bex and LG increased ABCA1/G1 levels, apoE4 lipoprotein association, and apoE4/A␤ complex levels, decreased soluble oA␤, and improved synaptic viability (Fig. 9). These data begin to directly address important issues for the use of RXR agonists as AD therapeutics.
ApoE Lipoprotein Association/Lipidation, ApoE/A␤ Complex Levels, and Soluble A␤ Clearance-Through increasing apoE lipoprotein association, RXR agonists increased apoE/A␤ complex levels, as they are positively correlated in the FIGURE 8. RXR agonists induce hepatomegaly in EFAD mice. A, correlation analysis between ABCA1 and oA␤ levels in EFAD-CX. Liver weights were measured in E3FAD and E4FAD mice treated with Bex, LG, or VC by 7-day gavage (5.75-6 months) (B), 7-day hydrogel (5.75-6 months) (C), and 30-day hydrogel (5-6 months) (D). Data are expressed as the mean Ϯ S.E. and were analyzed by one-way ANOVA followed by Tukey's multiple comparison post hoc analysis. n ϭ 6. *, p Ͻ 0.05 versus VC.
E4FAD-HP (Fig. 9A, parts 2 and 3, Pearson's r value ϭ ϩ0.42, p Ͻ 0.05). Our novel data begin to address the mechanism by which RXR agonist-dependent increased apoE4 lipoprotein association/lipidation may increase soluble A␤ clearance, i.e. the apoE/A␤ complex. In vivo, apoE isoform specifically modulates A␤ clearance (63)(64)(65)(66)(67)(68)(69), with soluble A␤ clearance lower with apoE4 compared with apoE3. Furthermore, NR agonists increase A␤ degradation in vitro in an ABCA1 and apoE-dependent manner (63). However, the exact process of how apoE4 decreases and how NR increases A␤ clearance is unclear. The levels of the soluble apoE/A␤ complex are lower and oA␤ higher with AD and APOE4 in EFAD mice and human tissue (32). Furthermore, in this study, the apoE/A␤ complex and oA␤ levels are inversely correlated in the E4FAD-HP (Fig. 9A, parts 3 and 4, Pearson's r value ϭ Ϫ0.53, p Ͻ 0.001). Thus, increased levels of the apoE/A␤ complex may contribute to the increased clearance rate of soluble A␤ after RXR agonist treatment.
Soluble Versus Insoluble A␤-The RXR agonist-induced beneficial effects on synaptic viability, as measured by increased PSD95 levels, are likely the result of decreased levels of soluble A␤42 and oA␤ levels, rather than total/insoluble A␤. An important caution is that using PSD95 as a measure of efficacy FIGURE 9. Summary hypothesis, RXR agonists induce beneficial effects in the E4FAD-HP. A, data from this study support the hypothesis that in the E4FAD-HP, the short-term RXR agonist treatment (T1 and T2) is as follows: 1, increases ABCA1/G1 levels; 2, increases apoE4 lipoprotein association/lipidation (as ABCA1 and TBSX-apoE levels are positively correlated); 3, increases apoE4/A␤ complex levels (as TBSX-apoE and apoE4/A␤ complex levels are positively correlated); 4, decreases soluble A␤ levels (as oA␤ and apoE4/A␤ complex levels are inversely correlated); and 5, improves synaptic viability (as oA␤ and PSD95 levels are inversely correlated). Thus, RXR agonists address a loss-of-function associated with APOE4 and exacerbated by A␤, i.e. lower apoE4 lipoprotein association/lipidation. B, schematic representation of the proposed hypothesis.
for synaptic viability requires future validation that RXR agonist treatment correlates with improvements in functional and cognitive readouts. Further analysis of the E4FAD-HP demonstrated an inverse correlation between levels of oA␤ and levels of PSD95 (Fig. 9A, parts 4 and 5, Pearson's r value ϭ Ϫ0.61, p Ͻ 0.001). This is in agreement with the correlation of soluble A␤ and oA␤ with cognitive decline and disease severity in humans (70), and the association of oA␤ levels with memory decline in FAD-Tg mice (70). Furthermore, recent data demonstrate that transient soluble oA␤ levels may underlie neuronal deficits in FAD-Tg mice using a genetic approach (71). However, as a dynamic equilibrium likely exists between soluble and insoluble A␤ levels (i.e. soluble A␤ may derive from insoluble A␤ and vice versa (72,73)), further research is required to rule out the contribution of a particular A␤ form impairing neuronal viability.
A second important finding is that the levels of soluble A␤ (A␤42 and oA␤) at the time of treatment influence the effect of RXR agonists on PD and efficacy readouts. In EFAD mice, increasing A␤ levels in the presence of APOE4 leads to decreased levels of apoE lipidation-related pathway genes and proteins, with the lowest levels in E4FAD-HP, the only brain region examined where RXR agonists improved PD and efficacy markers. Thus, RXR agonists may elicit beneficial effects only when apoE lipoprotein association/lipidation falls below some critical pathological threshold. The mechanism through which A␤ down-regulates the expression of NRs may involve decreased production of oxysterol ligands, positive modulators of NR expression. Overall, activation of the sterol regulatory element-binding protein 2 (SREBP2) results in the production of oxysterol ligands from cholesterol, activating NRs, including LXR and RXR, to increase ABCA1 expression (74). Recent data demonstrate that A␤ down-regulates SREBP-2 cleavage (75), supporting disruptions in lipid metabolism as a potential pathogenic process in AD (76). Thus, A␤ may lower RXR expression through modulation of lipid homeostasis.
As the proposed critically low pathological threshold of apoE lipoprotein association/lipidation is not reached in the E3FAD-CX or E4FAD-CX, RXR agonists would be expected to exert no effects on soluble A␤. However, soluble A␤ levels actually increased with both Bex and LG in the CX of E3FAD and E4FAD mice, the brain regions with the lowest A␤ levels at the time of treatment, and the effects that may be the result of hepatomegaly, as described below. These results may facilitate interpretation of the variable results previously reported for Bex (11, 12, 14 -16). Several FAD-Tg mouse models expressing m-apoE were used in previous studies, with widely variable A␤ levels at the time of treatment, and the level of m-apoE associated with lipoprotein-association was not measured and was difficult to predict.
Additional AD risk factors may also modulate the activity of the RXR agonists. Recent data demonstrate that in APOE-TR mice, Bex increases apoE4 lipidation to apoE3 levels (35). These data raise the intriguing possibility that apoE4-containing lipoproteins may be less able to modulate their lipoprotein-associated/lipidated state either at baseline or in response to stressors both A␤-dependent and A␤-independent, specifically during aging. An additional interaction that modifies AD risk is sex and APOE genotype. Clinical data indicate that the APOE4-induced risk for AD is significantly greater, perhaps exclusive to, females compared with males (77)(78)(79). Mechanistically, APOE4 may increase AD risk in women via multiple pathways involving the cardiovascular system (80), cellular aging (81), and importantly, A␤ accumulation. Indeed, we previously demonstrated that insoluble A␤ is increased in female E4FAD mice compared with female E3FAD mice (82). Thus, although male mice were the focus of this study, one could speculate that RXR agonists may be particularly efficacious in E4FAD female mice.
Although repurposing drugs for AD is an attractive alternative, there are important considerations when adapting drugs for use in the CNS that were initially developed for peripheral indications, particularly anti-cancer drugs and specifically Bex. The major side effects associated with Bex include severe but reversible hypertriglyceridemia, central hypothyroidism, pancreatitis, and importantly hepatomegaly (83)(84)(85). In this study, only 1-week Bex treatment induced an increase in liver size up to 150%, a side effect that may have impacted CNS function, as measured by increased levels of soluble oA␤. Adverse CNS effects are observed in models of both acute and chronic liver disease (43-46, 86, 87). Indeed, hepatic encephalopathy, alcoholic liver disease and nonalcoholic fatty liver disease are associated with significant impairments in cognitive function (43-46, 86, 87). In the liver, hepatomegaly, including steatosis (fatty liver), induces neuroinflammation, vascular dysfunction, lipid peroxidation, a decreased capacity for plasma detoxification, and insulin resistance (43-46, 86, 87). Evidence indicates that these peripheral pathways, particularly inflammation, modulate neuronal function either directly or by the induction of a toxic neuroinflammatory response (43-46, 86, 87). The potential for the liver to modulate cognitive function is highlighted by encephalopathy induced by liver failure, characterized by increased plasma toxin levels, brain edema, astrocyte swelling, and impaired neuronal function (45).
As LG and Bex both showed similar effects on apoE-A␤ biomarkers and induced hepatomegaly, these actions would appear common to RXR agonists. There are at least three alternative approaches that could be pursued. The first approach is to identify an optimal dose of Bex or a similar RXR agonist to induce beneficial effects in the CNS without detrimental effects in the periphery. However, our data do not support this approach, as the lower concentrations of Bex and LG delivered via hydrogel did not attenuate hepatomegaly. The second approach would focus on identifying an RXR/NR heterodimer optimal for AD symptoms. Indeed, novel RXR agonists selective for specific NR heterodimers, or tissues, have been described (83) that may minimize liver toxicity and other unwanted side effects. The third approach would determine whether agonists for alternative NRs, including LXR, PPAR, or RAR, are more beneficial than RXR.
In conclusion, our data demonstrate that in the E4FAD-HP, RXR agonists increase apoE4 lipoprotein association/lipidation and apoE4/A␤ complex levels, which result in decreased oA␤ levels and improved synaptic viability. Thus, RXR agonists address a loss-of-function associated with apoE4 and exacerbated by A␤, i.e. lower apoE4 lipoprotein association/lipidation. However, identifying whether this pathway is relevant for APOE3 carriers and addressing the detrimental effects of RXR agonists, including hepatomegaly, may be critical for providing viable long term prevention and treatment options for AD patients.