Small Molecule Structure Correctors Abolish Detrimental Effects of Apolipoprotein E4 in Cultured Neurons*

Background: Apolipoprotein E4 (apoE4), the major gene involved in Alzheimer disease, has a unique structure, intramolecular domain interaction, that is associated with neuropathology. Results: Potent small molecule structure correctors block apoE4 domain interaction and reverse apoE4 detrimental effects in cultured neurons. Conclusion: Structure correctors negate the detrimental effects of apoE4 in neurons. Significance: ApoE4 structure correctors could represent a therapeutic approach for treating apoE4-associated neuropathology. Apolipoprotein E4 (apoE4), the major genetic risk factor for late onset Alzheimer disease, assumes a pathological conformation, intramolecular domain interaction. ApoE4 domain interaction mediates the detrimental effects of apoE4, including decreased mitochondrial cytochrome c oxidase subunit 1 levels, reduced mitochondrial motility, and reduced neurite outgrowth in vitro. Mutant apoE4 (apoE4-R61T) lacks domain interaction, behaves like apoE3, and does not cause detrimental effects. To identify small molecules that inhibit domain interaction (i.e. structure correctors) and reverse the apoE4 detrimental effects, we established a high throughput cell-based FRET primary assay that determines apoE4 domain interaction and secondary cell- and function-based assays. Screening a ChemBridge library with the FRET assay identified CB9032258 (a phthalazinone derivative), which inhibits domain interaction in neuronal cells. In secondary functional assays, CB9032258 restored mitochondrial cytochrome c oxidase subunit 1 levels and rescued impairments of mitochondrial motility and neurite outgrowth in apoE4-expressing neuronal cells. These benefits were apoE4-specific and dose-dependent. Modifying CB9032258 yielded well defined structure-activity relationships and more active compounds with enhanced potencies in the FRET assay (IC50 of 23 and 116 nm, respectively). These compounds efficiently restored functional activities of apoE4-expressing cells in secondary assays. An EPR binding assay showed that the apoE4 structure correction resulted from direct interaction of a phthalazinone. With these data, a six-feature pharmacophore model was constructed for future drug design. Our results serve as a proof of concept that pharmacological intervention with apoE4 structure correctors negates apoE4 detrimental effects in neuronal cells and could be further developed as an Alzheimer disease therapeutic.

Apolipoprotein E4 (apoE4), the major genetic risk factor for late onset Alzheimer disease, assumes a pathological conformation, intramolecular domain interaction. ApoE4 domain interaction mediates the detrimental effects of apoE4, including decreased mitochondrial cytochrome c oxidase subunit 1 levels, reduced mitochondrial motility, and reduced neurite outgrowth in vitro. Mutant apoE4 (apoE4-R61T) lacks domain interaction, behaves like apoE3, and does not cause detrimental effects. To identify small molecules that inhibit domain interaction (i.e. structure correctors) and reverse the apoE4 detrimental effects, we established a high throughput cell-based FRET primary assay that determines apoE4 domain interaction and secondary celland function-based assays. Screening a ChemBridge library with the FRET assay identified CB9032258 (a phthalazinone derivative), which inhibits domain interaction in neuronal cells. In secondary functional assays, CB9032258 restored mitochondrial cytochrome c oxidase subunit 1 levels and rescued impairments of mitochondrial motility and neurite outgrowth in apoE4-expressing neuronal cells. These benefits were apoE4specific and dose-dependent. Modifying CB9032258 yielded well defined structure-activity relationships and more active compounds with enhanced potencies in the FRET assay (IC 50 of 23 and 116 nM, respectively). These compounds efficiently restored functional activities of apoE4-expressing cells in secondary assays. An EPR binding assay showed that the apoE4 structure correction resulted from direct interaction of a phthalazinone. With these data, a six-feature pharmacophore model was constructed for future drug design. Our results serve as a proof of concept that pharmacological intervention with apoE4 structure correctors negates apoE4 detrimental effects in neuronal cells and could be further developed as an Alzheimer disease therapeutic.
Apolipoprotein E4 (apoE4) 2 is the major genetic risk factor for late onset Alzheimer disease (AD), increasing the risk of developing AD at earlier ages of onset (1)(2)(3). In fact, 65-80% of all AD patients are carriers of at least one apoE4 allele (4). ApoE functions as a major carrier of lipids and cholesterol in both the peripheral and central nervous systems (5,6). Human apoE has 299 amino acids with three common isoforms (apoE2, apoE3, and apoE4) that differ at residue 112 or 158. ApoE3 contains a cysteine at residue 112 and an arginine at residue 158, whereas apoE4 has arginine at both positions, and apoE2 has cysteine (5,6). These minor variations give rise to profound differences in the tertiary protein structure and function (6 -11). ApoE4 displays an intramolecular domain interaction between its aminoand carboxyl-terminal domains, leading to a compact structure (12,13). Domain interaction in apoE4 is induced by Arg-112, which facilitates the formation of a salt bridge between Arg-61 in the amino-terminal domain and Glu-255 in the carboxylterminal domain (12,13). The Cys-112 residue in apoE2 and apoE3 weakens the domain interaction, resulting in more open structures (13).
ApoE4 domain interaction was demonstrated by FRET in living neuronal cells (14). FRET is a well established technique to detect changes in protein conformation or protein-protein interactions that relies on a distance-dependent transfer of photon energy between a donor and an acceptor fluorophore (15). The apoE4-FRET reporter showed a much stronger signal than the apoE3-FRET reporter, indicating that apoE4 has stronger domain interaction (14). The apoE-FRET specifically measures the interaction between the amino-and the carboxylterminal domains of apoE. Importantly, when threonine is substituted for arginine at residue 61 (R61T), this disrupts domain interaction in apoE4 and converts apoE4 to an "apoE3-like" molecule and decreases the FRET signal of apoE4 to the level of apoE3 in neuronal cells (14).
Evidence has suggested that domain interaction mediates a broad spectrum of apoE4 isoform-specific effects (8 -11). In plasma, apoE4 binds preferentially to very low density lipoproteins (12,13). Mutant apoE4-R61T lacking domain interaction, like apoE3, binds preferentially to high density lipoproteins (12,13). In neurons, apoE4 reduces expression of the protein subunits of mitochondrial respiratory complexes, such as subunit 1 of complex IV (mtCOX1) and subunit ␣ of complex V, resulting in a reduction in mitochondrial respiratory function (16). Unlike apoE4, apoE3 and apoE4-R61T did not induce mitochondrial dysfunction (16), suggesting that domain interaction mediates apoE4-specific detrimental effects on mitochondria. Similarly, apoE4 perturbs neuronal function by reducing mitochondrial motility, decreasing neurite outgrowth (17)(18)(19), and inhibiting synaptogenesis (20). In addition, amyloid ␤ production in rat neuroblastoma cells is enhanced by apoE4 to a greater extent than apoE3 (21), whereas apoE4-R61T has no such effect (21). Unlike human apoE4, mouse apoE has threonine at residue 61, which prevents domain interaction (7). Introducing an apoE4-like domain interaction in mouse apoE by substituting arginine for threonine at the residue corresponding to amino acid 61 induces neurodegenerative pathology and memory deficits in the mice (22,23). These detrimental effects of apoE4, which are mediated by domain interaction, probably contribute to AD pathogenesis (8 -11). How apoE4 may be causative in AD has been reviewed previously (8).
Disrupting apoE4 domain interaction with small molecules, so-called apoE4 structure correctors, converts apoE4 into an apoE3-like conformation and reverses the apoE4-specific detrimental effects. We previously identified a prototype structure corrector, GIND-25, from an in silico screen (21). Treatment of neuronal cells with GIND-25 reduced the apoE4-enhanced amyloid ␤ production (21) and restored mtCOX1 levels in apoE4-expressing cells (16). In addition, GIND-25 and a small molecule, PH-002, restored endoplasmic reticulum and Golgi apparatus transit of apoE4 in cultured neurons to levels equivalent to apoE3 and apoE4-R61T and rescued apoE4-caused impairment of neurite outgrowth and synaptogenesis (24). In the present study, we identified additional potent small molecules that inhibit domain interaction and thus reverse the detrimental effects of apoE4, using a high throughput, cell-based FRET primary screening assay that quantifies the change in apoE4 domain interaction and a panel of secondary cell-and function-based assays in neuronal cells. Screening a small molecule library led to the identification of phthalazinone analogs as a new class of apoE4 structure correctors that abolish apoE4 domain interaction-related detrimental effects in neuronal cells in culture.

EXPERIMENTAL PROCEDURES
Materials-Minimum essential medium (MEM) with Glutamax, Opti-MEM I, Dulbecco's PBS (DPBS) with or without calcium and magnesium, nonessential amino acids, sodium pyruvate, and G-418 were purchased from Invitrogen. FBS was purchased from HyClone, poly-L-lysine (PLL; 0.01% solution) from Sigma-Aldrich, and paraformaldehyde (16%) from Electron Microscopy Sciences. All chemicals were of analytical grade and were obtained from Sigma unless stated otherwise.
FRET Assay-Neuro-2a cells stably expressing the GFP/ eDHFR FRET reporters were seeded at a density of 300,000 cells/ well in 96-well plates and grown in riboflavin-free MEM with 10% FBS. Riboflavin was omitted to reduce background fluorescence. Two h after seeding, cells were loaded with a trimethoprim-hexachlorofluorescein conjugate (TMP-HEX, (LigandLink TM Label, Active Motif) at 2 M in culture medium and incubated for an additional 4 h. TMP is a cell-permeable and high affinity eDHFR ligand. After washing with PBSϩ (PBS supplemented with 10 mM glucose, 0.9 mM CaCl 2 , and 0.49 mM MgCl 2 ) three times, cells were incubated with various concentrations of compounds in riboflavin-free MEM without serum for 16 h. After washing once with PBSϩ, 200 l of PBSϩ was added to each well. Signals for both GFP (excitation, 485 nm; emission, 525 nm; cut-off, 515 nm) and FRET (excitation, 485 nm; emission, 565 nm; cut-off, 550 nm) were recorded on a Flexstation III (Molecular Devices, Sunnyvale, CA). The ZЈ is Ͼ0.5, S/B is 1.6 with a CV Ͻ10% and is tolerant of 1% DMSO or less.
Clones of Neuro-2a cells stably expressing GFP-apoE4-eDHFR at different levels were used to demonstrate that the expression level did not affect the FRET ratio. Five Neuro-2a clones (supplemental Fig. S1A) with increasing levels of GFP-apoE4-eDHFR were labeled with HEX, the GFP fluorescence and the FRET signal were measured, and the FRET ratio was calculated (supplemental Fig. S1B).
MtCOX1 In-cell Western Assay-Neuro-2a cells stably expressing apoE were seeded in 96-well black wall/clear bottom plates (Costar) at a density of 40,000 cells/well and incubated overnight in complete culture medium. At 70% confluence the following day, cells were carefully washed with PBS and changed to fresh Opti-MEM supplemented with MEM-NEAA and sodium pyruvate. For EC 50 determination, each compound was tested in triplicate at seven concentrations (a 3-fold dilution series of 10,000 to 13.7 nM) plus mock control wells. The concentration of DMSO was kept at 0.5% in all of the wells. After a 24-h incubation, cells were fixed immediately with 4% paraformaldehyde in PBS for 20 min at room temperature and then washed three times with PBS containing 0.1% Triton X-100. An in-cell Western assay was performed using an in-cell Western kit (LI-COR Biosciences, Lincoln, NE), according to the manufacturer's protocol with minor modifications. Briefly, cells were incubated in blocking buffer for 1.5 h, followed by overnight incubation with 100 l of blocking buffer without (for background control) or with mouse anti-mtCOX1 antibody (1:1000, MitoSciences, Eugene, OR; lots D0804 and E0594 displayed high specificity to purified COX1 and for COX1 in lysate from Neuro-2a cells expressing apoE3 and apoE4). Then the wells were washed three times in PBS containing 0.1% Tween 20 (Tween washing solution). Cells were then incubated in 50 l of blocking buffer containing 0.2% Tween 20, IRDye 800CW-labeled anti-mouse secondary antibodies (1:800), Sap-phire700 (1:1000, a nonspecific cell stain), and DRAQ5 (1:2000, a DNA dye). Fifty l of the same staining solution were added to four empty wells (without cells) to determine background signals for Sapphire700/DRAQ5. We used Sapphire700 and DRAQ5 in combination for cell number normalization as suggested in the manufacturer's protocol. For background control (with cells), 50 l of blocking buffer containing secondary antibodies alone (without Sapphire700 or DRAQ5) were added to the wells. After 1 h of incubation with the secondary antibodies, cells were washed three times with Tween washing solution and then once with PBS without Tween. After the final wash, the solution was removed completely from the wells. Signals from the 700-and 800-nm fluorophores were detected by scanning the plates on an Odyssey infrared imaging scanner immediately. Integrated intensity of the mtCOX1 immunoreactivity from each well was first normalized against the integrated intensity of Sapphire700/DRAQ5. Normalized mtCOX1 immunoreactivity from each compound-treated sample was then divided by the mean value from all mock control wells on the same plate (base line) to determine the percentage of changes in mtCOX1 immunoreactivity with drug. Activities of com-pounds were compared by their abilities to increase mtCOX1 immunoreactivity (percentage increase over the mock controls).
Mitochondrial Motility-PC12 pheochromocytoma cells stably expressing comparable levels of apoE (0.2 ng/mg protein) together with the mitochondrial marker mRFP were maintained in DMEM supplemented with 10% horse serum, 5% FCS, and 1 mM L-glutamine (all from Invitrogen). PC12 cells (7.5 ϫ 10 3 ) were plated in 8-well chambered slides (Lab-Teq, Nunc) coated with PLL; differentiated in regular growth medium supplemented with 0.75% horse serum, 1 mM L-glutamine, and 40 ng/ml NGF; and used for experiments after 14 days of differentiation.
To determine if compounds that block apoE4 domain interaction rescue mitochondrial motility, differentiated PC12 cells were incubated with the indicated dose for 24 h. Digital images of mRFP fluorescence in neurites of PC12 cells differentiated with NGF for 14 -16 days were captured at 10 frames/s for 10 s with a water-cooled EM-CCD camera (Hamamatsu, Bridgewater, NJ) mounted to an Axio Observer microscope (Zeiss) equipped with a ϫ60 oil objective. During recordings, cells were kept in CO 2 -independent medium (Invitrogen) supplemented with 1 mM L-glutamine, 2.5% FCS, and 2.5% horse serum at 37°C. Digital images were analyzed with NIH ImageJ software. The manual tracker plug-in was used to track and analyze mitochondrial movement. The percentage of motile mitochondria was (number of images in which the observed mitochondria are moving/number of recorded images) ϫ 100.
To determine the effect on exogenous apoE, PC12 cells were cultured in DMEM supplemented with 5% horse serum, 2.5% FCS, and 1 mM L-glutamine. PC12 cells (25 ϫ 10 4 ) were plated on 30-mm glass coverslips, coated with PLL, and differentiated in medium supplemented with 0.75% horse serum, 1 mM L-glutamine, and 40 ng/ml NGF. Differentiated PC12 cells were incubated for 24 h at 37°C with 7.5 g/ml of various forms of apoE, and mitochondrial dynamics were determined during a 15-min recording (12 frames/min) at room temperature. Recombinant human apoE, including the apoE4 variant carrying the R61T mutation, was prepared as described (12).
Neurite Outgrowth Assay-N2a-apoE3, N2a-apoE4, or N2a-apoE4-R61T cells grown to 80% confluence were seeded at a density of 8000 cells (N2a-apoE3) and 7500 cells (N2a-apoE4) per well into PLL-coated 24-well plates containing Opti-MEM with either 0.03% DMSO for control or various doses of compounds in quadruplicate. Cells were grown for 72 h at 37°C to induce neurite outgrowth. After fixation with 4% paraformaldehyde for 30 min, cells were washed with DPBS and then stained by SimplyBlue TM Safe Stain (Invitrogen). Two h later, cells were washed with DPBS and kept in DPBS in the dark at 4°C. Phase-contrast images were taken on an Axio Observer Z1 microscope equipped with a motorized stage using the Volocity software program. For each well, eight images were automatically taken from the same eight image fields. Images were analyzed using NIH ImageJ for total cell count per image and cells that had grown neurites longer than the cell body diameter.
Primary Neurons Treated with Structure Corrector-Primary neurons were obtained from the cortex and hippocampus of neuron-specific enolase (NSE)-apoE4 transgenic mice. The NSE mice have been fully characterized (26,27). The cells were cultured for 10 days (28) and then treated with 200 nM PH-002 in DMSO for 4 days. Fresh compound was added on days 1 and 3, and then the cells were lysed on day 4 for Western blot analysis with anti-COX1 antibody. Voltage-dependent anion channel isoform 1 (VDAC1, mitochondrial porin) served as an internal control to monitor loading of sample on gels (Calbiochem, anti-mouse VDAC1).
ApoE4 Cys-76 Mutant Construct, Expression, and Purification-A cysteine mutation was introduced at position 76 into a cDNA encoding the apoE4 amino-terminal domain (residues 1-191) with a QuikChange mutagenesis kit and a modified pET32a construct as the template (9). The mutated sequence was verified by DNA sequencing. The apoE4 aminoterminal domain was expressed in E. coli. The cells were harvested and lysed, and cellular debris was pelleted as described (29). The apoE4 Cys-76 amino-terminal domain was purified as described (30) and spin-labeled as described for intact apoE4 cysteine mutants with (1-oxyl-2,2,5,5-tetramethylpyrrolidin-3yl)methyl methanethiosulfonate (31).
EPR Experiments-EPR spectra were collected on a JEOL X-band spectrometer equipped with a loop-gap resonator. Spin-labeled apoE4 Cys-76 amino-terminal domain (25 l, 57 M) in TBS was incubated with an equal volume of PH-002 in TBS, prepared from a stock solution (500 M in 10% DMSO/ TBS) at two concentrations, 114 and 228 M, respectively. After a 1-h incubation, ϳ4 l of the mixture was placed in a sealed quartz capillary tube. Spectra from triplicate incubations were acquired at 20 -22°C with a single 60-s scan over 100 G at a microwave power of 2 milliwatts and a modulation amplitude optimized to the natural line width of the attached nitroxide. The effect of PH-002 was determined by comparing the effects on the central peak of the EPR spectra (h1/h0) of the compound and a control without compound.
Pharmacophore Modeling-Pharmacophore modeling was performed with the Pharmacophore Protocol within Discovery Studio version 2.1. Low energy conformations of all nine phthalazinone compounds in this report were determined with the CHARMm force field using the default potentials ( Fig. 1). Multifeatured pharmacophores were constructed by generating common feature pharmacophores from the two most active compounds (PH-001 and PH-002). This procedure evaluates all possible hydrophobic, hydrogen bond acceptor, and donor features that the two compounds have in common, based on shared conformations. Up to 255 low energy conformations of each compound were included in the analysis. The algorithm starts with matching a small set of features and searches for more shared features until no further common conformation of the input compounds is found. Multiple hypotheses involving the spatial orientation of shared chemical features between the input compounds were generated and scored. All hypotheses were evaluated with the ligand pharmacophore mapping function and scoring module in the pharmacophore protocol and used to fit the remaining compounds.
Structure-based Pharmacophore Analysis-The high resolution crystal structure of the 22-kDa domain of human apoE4 (receptor binding domain, residues 1-165; Protein Data Bank entry 1GS9) was prepared for analysis using the Receptor-Ligand Interaction module within Discovery Studio version 2.1. Heteroatom molecules were removed from the structure, and a continuous dielectric was used in place of the water molecules. Missing hydrogens were added, valences were fixed where necessary, and atom types and

Structural Correctors of ApoE4 Domain Interaction
partial charges were assigned. The protein model was adjusted so that the optimal positions of polar hydrogens were identified. Steric clashes were removed by an energy minimization procedure. A binding site was built within 20 Å of Arg-61 in 1GS9, and an interaction map was generated in which all pharmacophore features (acceptors, donors, and hydrophobes) were created in the defined binding site. This resulted in ϳ1800 features. The features within 1GS9 served as a reference pharmacophore, and a pharmacophore comparison was conducted with the six-feature pharmacophore to find the closest match to the reference.
Statistical Analysis-Statistical analyses were performed by two-tailed t test or one-way analysis of variance (ANOVA) followed by Tukey's post hoc comparisons. p Ͻ 0.05 was considered statistically significant.

RESULTS
Establishing High Throughput Cell-based GFP-apoE-eDHFR FRET Assay for ApoE4 Structure Corrector Screening-Previously, we reported a CFP-apoE-YFP FRET system that was successfully used to demonstrate apoE4 domain interaction in living neuronal cells (14). To improve the assay performance in a high throughput format, we replaced the fluorophores by fusing a GFP (as a donor) to the amino terminus of apoE4 and an eDHFR to the carboxyl terminus of apoE4 (GFP-apoE4-eDHFR) ( Fig. 2A). TMP-HEX served as the acceptor fluorophore. After loading with TMP-HEX, Neuro-2a cells expressing GFP-apoE4-eDHFR had a 36.3 Ϯ 6.6% higher FRET signal (defined as the ratio of FRET to GFP) than those expressing GFP-apoE3-eDHFR (p Ͻ 0.0001, n ϭ 6; Fig. 2B). The higher FRET signal of GFP-apoE4-eDHFR, compared with GFP-apoE3-eDHFR, reflects the closer proximity of the amino-and the carboxyl-terminal domains in apoE4. The GFP-apoE3-eDHFR signal was comparable with that in the cells expressing GFP-apoE4 and eDHFR from two separate constructs (GFP-apoE4/eDHFR, n ϭ 5; Fig. 2B). Thus, the GFP-apoE-eDHFR FRET system sensitively detects conformational changes between the two intramolecular domains of apoE. Moreover, this assay allowed the development of a high throughput format (96 or FIGURE 2. FRET assay used to monitor intracellular apoE4 domain interaction in Neuro-2a cells and to identify apoE4 structure correctors. A, apoE4 domain interaction allows FRET to occur between the donor and acceptor fluorophores. Inhibiting domain interaction by structure correctors reduces FRET from the apoE4 reporter. SC, structure corrector. B, FRET intensity (ratio of FRET signal divided by GFP signal) from GFP-apoE4-eDHFR is 36.3% higher than GFP-apoE3-eDHFR. *, p Ͻ 0.0001 (n ϭ 6). FRET intensity from GFP-apoE4/eDHFR (expressing GFP-apoE4 and eDHFR by separate constructs, n ϭ 5) is comparable with that of GFP-apoE3-eDHFR. C, CB9032258 inhibits FRET intensity from GFP-apoE4-eDHFR but not GFP-apoE3-eDHFR. *, p Ͻ 0.05, by ANOVA with Tukey's post hoc test. D, dose-response curves show relative potencies of PH-001, PH-002, CB9032258, and PH-008 to inhibit FRET intensity. Values are mean Ϯ S.D. (error bars). FEBRUARY 17, 2012 • VOLUME 287 • NUMBER 8 JOURNAL OF BIOLOGICAL CHEMISTRY 5257 384 wells) for screening for small molecules that inhibit apoE4 domain interaction ( Fig. 2A).

Structural Correctors of ApoE4 Domain Interaction
Establishing Cell-and Function-based Secondary Assays for ApoE4 Structure Corrector Validation-To identify small molecules that inhibit apoE4 domain interaction and also reverse the adverse effects caused by domain interaction, we established three cell-and function-based secondary assays for apoE4 structure corrector validation. First, we developed a 96-well format in-cell Western assay to monitor changes in mtCOX1 levels in neuronal cells. Mitochondria have emerged as a prime target for AD pathology (32) and the detrimental effects of apoE4 (16,(33)(34)(35)(36). Steady-state levels of mtCOX1, an essential component of mitochondrial cytochrome c oxidase (complex IV), were significantly lower in Neuro-2a cells expressing apoE4 (N2a-apoE4) than in those expressing apoE3 (N2a-apoE3) or no apoE (control) (Fig. 3A, left). Unlike apoE4, mutant apoE4-R61T had no such effect (Fig. 3A, left), suggesting that the reduction in mtCOX1 is a biological consequence of apoE4 domain interaction (16). The cell numbers in each well were comparable, as revealed by counterstaining for total protein (Sapphire700) and DNA (DRAQ5) (Fig. 3A, middle). ApoE4-specific reduction in mtCOX1 levels is readily detectable in the merged image of anti-mtCOX1 and Sapphire700/ DRAQ5 (Fig. 3A, right) or by quantification of signal intensities from both channels on an Odyssey scanner (LI-COR) (Fig. 3B). After cell number normalization, the relative mtCOX1 levels in N2a-apoE4 cells were ϳ60 Ϯ 4.8% (n ϭ 3) of those in control cells, which were also significantly lower than those in N2a-apoE3 or N2a-apoE4-R61T cells (Fig. 3B, p Ͻ 0.001 by ANOVA). Therefore, the mtCOX1 in-cell Western assay can provide a biological readout for apoE4 domain interaction and thus for the validation of small molecules that inhibit domain interaction in neuronal cells.
Third, a neurite outgrowth assay of Neuro-2a cells was also developed by modifying our published protocol (17,18). As reported, apoE4-expressing Neuro-2a cells had fewer cells with long neurites than those expressing apoE3 (Fig. 5A). Importantly, Neuro-2a cells expressing apoE4-R61T had neurite length similar to those expressing apoE3, which were significantly longer than those expressing apoE4. Again, these data indicate that apoE4 domain interaction is responsible for the apoE4-related neurite outgrowth impairment and thus can be used as an additional secondary assay to validate functionally the effect of apoE4 structure correctors.
Identification and Validation of CB9032258 as an ApoE4 Structure Corrector-A set of compounds from the Chem-Bridge library was selected for screening in our assays. CB9032258, a phthalazinone derivative, was found to reduce the FRET signal by 23.6 Ϯ 1.7% (n ϭ 3 at 10,000 nM) in GFP-apoE4-eDHFR but not in GFP-apoE3-eDHFR cells (Fig. 2C), suggesting that CB9032258 is a specific inhibitor of apoE4 domain interaction. The apoE4-specific inhibitory effect of CB9032258 on FRET is dose-dependent (IC 50 ϭ 4245 Ϯ 2149 nM, n ϭ 3) (Fig. 2D and Table 1).
CB9032258 was then functionally validated as a structure corrector in the cell-based secondary assays. Treatment of Neuro-2a cells expressing apoE4 with CB9032258 restored mtCOX1 levels to those seen in Neuro-2a cells expressing apoE3 (Fig. 3C). CB9032258-induced elevation of mtCOX1 immunoreactivity is dose-dependent, as shown by the percent- . Mitochondrial motility assay used to monitor apoE4-specific reduction in percentage of motile mitochondria and the reversibility of this impairment by apoE4 structure correctors. A, differentiated PC12 cells were incubated for 24 h at 37°C without (control) or with exogenous apoE3, apoE4, or apoE4-R61T (7.5 g/ml). Mitochondrial dynamics, analyzed as a percentage of motile mitochondria, was measured during a 15-min recording (12 frames/min) (control, n ϭ 218 mitochondria from 12 cells; apoE3, n ϭ 209 mitochondria from 14 cells; apoE4, n ϭ 212 mitochondria from nine cells; apoE4-R61T, n ϭ 164 mitochondria from 12 cells). Values are mean Ϯ S.E. (error bars). ***, p Ͻ 0.001 versus control (two-tailed t test). B, mitochondrial motility analyzed in apoE3-and apoE4-expressing PC12 cells at 37°C without the addition of small molecules served as a control. After the addition of increasing doses of inactive PH-008 incubated for 24 h, the percentage of motile mitochondria was determined in the differentiated PC12 cells expressing apoE4 (n ϭ 100 mitochondria from 10 cells in two separate experiments). Values are mean Ϯ S.E. C, mitochondrial motility analyzed in apoE4-expressing PC12 cells at 37°C with increasing concentrations of the active phthalazinone PH-001. Values are mean Ϯ S.E. EC 50 , Hill slope, and R 2 reported for the positive dose response. FEBRUARY 17, 2012 • VOLUME 287 • NUMBER 8 age increase over the mock-treated cells (up to 52.3 Ϯ 3.6% at 10,000 nM, EC 50 ϭ 2094 Ϯ 544 nM, n ϭ 3) (Fig. 3C, top right, and Table 1). CB9032258 treatment also abolished the apoE4 domain interaction-dependent detrimental effects on mitochondrial motility (EC 50 ϭ 30 nM) ( Table 1) and neurite outgrowth ( Fig. 5B and Table 1). Importantly, the beneficial effects of CB9032258 on mtCOX1 levels (Fig. 3D), mitochondrial motility, and neurite outgrowth (data not shown) were observed in apoE4-expressing cells but not in control (wild type) and/or apoE3-expressing cells. Thus, CB9032258 is functionally validated as an apoE4 structure corrector that can reverse the apoE4 domain interaction-dependent detrimental effects in neuronal cells in culture. It was chosen as a potential lead for further investigation.

Structural Correctors of ApoE4 Domain Interaction
Analysis of Structure-Activity Relationships of Phthalazinone Analogs-To study the structure-activity relationships, 118 phthalazinone derivatives based on CB9032258 were obtained from Merck Research Laboratories (data not shown). Chemical modification of CB9032258 resulted in well defined structureactivity relationships, exemplified by the nine representative compounds shown in Fig. 1. Interestingly, the potencies of these compounds to inhibit domain interaction, as determined by the GFP-apoE4-eDHFR FRET assay, correlated well with their abilities to restore mtCOX1 levels in Neuro-2a cells expressing apoE4 (Table 1). For example, PH-001 (FRET IC 50 ϭ 23 nM) and PH-002 (FRET IC 50 ϭ 116 nM) were the most potent phthalazinones, with 185-fold and 37-fold enhanced FRET potencies over CB9032258, respectively ( Fig. 2D and Table 1). They were also the most potent compounds in the secondary mtCOX1, mitochondrial motility, and neurite outgrowth assays (Figs. 3C, 4C, and 5B and Table 1). In fact, for the 47 phthalazinone analogs that were tested by both the FRET and the mtCOX1 assays, a strong positive correlation between FRET-IC 50 and mtCOX1-EC 50 was observed (Pearson's r ϭ 0.8030, p Ͻ 0.0001) (Fig. 3E). In contrast, the inactive compounds identified in the FRET assay, PH-007 and PH-008, were also inactive in the mtCOX1 assay at the highest dose (10,000 nM) tested (Table 1), and PH-008 was also inactive in the mitochondrial motility (Fig. 4B) and neurite outgrowth (Fig. 5B) assays. The combination of the primary FRET assay and functional secondary assays identified potent phthalazinone analogs as a new class of apoE4 structure correctors. PH-002 also was shown to increase COX1 levels in primary neurons from NSE-apoE4 transgenic mouse cortex and hippocampus. After 4 days of treatment with PH-002 (200 nM), COX1 levels were increased by ϳ60% (Fig. 6). We have previously demonstrated that PH-002 (100 nM) increased dendritic spine development in primary neurons from NSE-apoE4 trans-  genic mice to levels comparable with those in NSE-apoE3 primary neurons (apoE3-expressing primary neurons treated with PH-002 gave results identical to untreated primary neurons (24)). EPR Demonstrating Direct Target Engagement-To demonstrate that PH-002 affected apoE4 structure through a direct binding mechanism, the apoE4 amino terminus spin-labeled at position 76 was incubated with PH-002 at two molar ratios, 2:1 and 4:1 (PH-002/apoE), and the effect on the motility of the Cys-76 spin label was determined. Position 76 was chosen because it is in the vicinity of the interface between the aminoand carboxyl-terminal domains of apoE4. A concentration-dependent effect of PH-002 on the mobility of the spin label was observed, demonstrating direct binding of PH-002 (Fig. 7).
Building Pharmacophore Model of Phthalazinones-The nine phthalazinone analogs (Fig. 1) display a good degree of structural diversity and well defined structure-activity relationships. These data allow us to build a pharmacophore model of phthalazinone for drug design. The best six-feature pharmacophore (Fig. 8A) is composed of three hydrophobic spheres (blue; the spheres are labeled "A", "E", and "F") and three hydrogen bond acceptors (green arrows with arrowheads pointing toward the donor (labeled "B", "C", and "D"). This model was based on the common features of the two most active compounds, PH-001 and PH-002.
Multiple hypotheses were derived from this analysis, and the model that best fit with all nine phthalazinones was selected. Fig. 8B shows the fit of PH-002 to the six-feature pharmacophore. This compound fits all six features, as expected. Fig. 8C shows a two-dimensional depiction of PH-002 annotated with the moieties making contact with these features. On the other hand, PH-008, an inactive compound, fits five of the six features, the exception being hydrophobe "F" (Fig. 8D), suggesting that hydrophobe "F" is necessary for activity. The other inactive compound, PH-007, only has five of the six features (Fig. 8E). This compound makes good contact with hydrophobe "F"; however, hydrogen bond acceptor "C" is not possible. The core ring structure is not a typical phthalazinone, but a 2-methyl-2H-isoquinolin-1-one that lacks acceptor functionality at this position. This result suggests that the hydrogen bond acceptor "C" is also required for activity.
Structure-based pharmacophore analysis is an independent way of confirming the results of the ligand-based methodology described above. Access to a high resolution crystal structure of the amino-terminal region of human apoE4 (12) allows the potential docking of the six-feature pharmacophore (Fig. 8A) into this binding domain. This type of analysis will confirm the importance of the two features ("C" and "F") identified in the above analysis. It is known that mutation of Arg-61 (R61T) (8 -12) abolishes activity, so any docking model must include interaction with Arg-61. This was utilized as a starting point in this analysis. A binding site was built within 20 Å of Arg-61, and all possible pharmacophore features were identified within this binding site (all hydrophobes, acceptors, and donors). This resulted in ϳ1800 features. These features were used as a reference to find the best match for the six-feature pharmacophore described above. The results of this analysis are shown in Fig. 9A. PH-002 is docked into the 22-kDa amino-terminal region of human apoE4. This figure has the side chains of the 22-kDa domain removed except for Arg-61 (shown as purple CPK spheres), Arg-119 (shown as yellow CPK spheres), and Glu-50 (shown as turquoise CPK spheres). The structure of PH-002 is shown along with the six features described above (Fig. 8). Arg-61 shows a strong interaction with hydrophobe "F", which is consistent with the ligand-based findings described above. Glu-50 shows an interaction with hydrogen bond acceptor "C" that is consistent with the findings described. Arg-119 shows a tight interaction with hydrogen bond acceptor "B". Fig. 9B shows a close-up of PH-002 in the binding pocket. This structure-based analysis, in parallel with target engagement by EPR, confirms the results of the independent ligand-based approach.
This study was undertaken to identify a new potent (low nanomolar) class of structure corrector molecules that could be the basis for developing structure-activity relationships in a chemical series and to establish proof of concept that small molecules identified by their ability to alter the structure of apoE4 (i.e. block domain interaction) reduce its detrimental effects on neurons in culture. Previously, we identified several compounds of low potency (millimolar) that corrected apoE4 activity (21). However, these compounds lacked druglike properties.
The present study relied on small molecule library screening and chemical synthesis of lead compounds to identify struc-FIGURE 8. Pharmacophore modeling for phthalazinone analogs using a ligand-based methodology. A, phthalazinone pharmacophore. The six-feature pharmacophore is composed of three hydrophobic spheres (blue; spheres are labeled "A", "E", and "F") and three hydrogen bond acceptors (green arrows with arrowheads pointing toward donor labeled "B", "C", and "D"). B, pharmacophore fitted with PH-002. This potent compound fits all six features. C, twodimensional structure of PH-002 annotated with moieties making contact with features of the pharmacophore. D, pharmacophore fitted with PH-008. This inactive compound fits five of the six features. Hydrophobe "F" is not satisfied, suggesting the importance of this feature for activity. E, pharmacophore fitted with PH-007. This inactive compound fits five of the six features. Hydrogen bond acceptor "C" is not possible with this compound. The core ring structure is not a phthalazinone but a 2-methyl-2H-isoquinolin-1-one, which lacks acceptor functionality at this position. Note that unlike PH-008, hydrophobe "F" is satisfied. The importance of the hydrogen bond acceptor "C" is suggested by the lack of activity of this compound.
ture-activity relationships. The phthalazinones represent such a series, demonstrating that the active compounds enter the cultured neurons and alter domain interaction, specifically in apoE4-expressing cells, without affecting apoE3-expressing cells. The active phthalazinones enhance neuronal mtCOX1 expression, restore mitochondrial motility, and promote neurite outgrowth to levels equivalent to apoE3. The critical importance of blocking domain interaction for restoring these functional activities was further established by showing that mutant apoE4-R61T, which lacks domain interaction, behaved like apoE3.
Previously, we demonstrated that the intracellular trafficking of apoE4 is impaired in the endoplasmic reticulum and Golgi apparatus of Neuro-2a cells expressing apoE4 compared with apoE3 trafficking, as determined by fluorescence recovery after photobleaching (24). ApoE4-R61T, which lacks domain interaction, gave results comparable with those for apoE3. Likewise, small molecule structure correctors GIND-25 (21) and PH-002 rescued the apoE4-impaired trafficking. In concert with the present data and the fluorescence recovery after photobleaching studies, two important points are made demonstrating where and how blocking domain interaction alters both the structure and function of apoE4. (a) small molecule structure correctors modulate domain interaction as soon as apoE4 is synthesized and/or enters the endoplasmic reticulum. Domain interaction has been shown to occur in poorly lipidated apoE4; however, when apoE complexes with lipids, the helical structure is significantly reorganized (9). (b) structure correctors, which rescue apoE4 intracellular trafficking, restore neurite outgrowth in apoE4-expressing Neuro-2a cells and dendritic spine development in primary hippocampal neurons from NSE-apoE4-expressing mice (no effects of structure correctors on apoE3-expressing Neuro-2a cells or apoE3-expressing primary neurons were observed).
The GFP-apoE4-eDHFR FRET system served as a relatively high throughput cell-based assay providing a signal that approximates the distance between the amino-terminal and carboxyl-terminal domains of apoE and identifying active small molecules that reduce the FRET signal intensity in apoE4. The FRET signal in the Neuro-2a cells expressing the GFP-apoE4-eDHFR was 36.3% higher than seen with GFP-apoE3-eDHFR (Fig. 2). After treatment of the cultured cells with potent small molecules (i.e. PH-001 and PH-002), the GFP-apoE4-eDHFR signal was inhibited by about 25-30% without significantly affecting the GFP-apoE3-eDHFR signal. Thus, this assay identifies apoE4-specific structure correctors.
The FRET data, which measure a biophysical structural property of apoE, correlate well with the mitochondrial COX1 in-cell Western data, which reflect a function property of apoE (Fig. 3E). Likewise, mitochondrial motility and neurite outgrowth represent downstream functional properties of apoE4 that are restored after treatment of the cells with the active small molecules. Furthermore, the inactive phthalazinones lacked potency across all assays, and the active phthalazinones modulated the activities only in the apoE4-expressing cells and not in apoE3-expressing cells.
In agreement with the apoE4-R61T mutant and the PH-002 studies indicating that disruption of apoE4 domain interaction reverses the effects of apoE4 in the FRET, mtCOX1, mitochondrial motility, and neurite outgrowth assays, PH-002 binds directly to the apoE4 amino-terminal domain. This provides direct evidence that PH-002 disrupts domain interaction by binding in the vicinity of the interface between the amino-terminal and carboxyl-terminal domains of apoE4. Structure-activity relationships established by the nine phthalazinone analogs allowed construction of a six-feature pharmacophore model. Important hydrophobic regions and hydrogen bond acceptors were identified in modeling the critical elements of the active phthalazinones. Docking of PH-002 onto the apoE4 22-kDa amino-terminal region suggested how the reactive molecules interact to form a binding pocket (Fig. 9). Thus, in agreement with EPR direct binding, pharmacophore modeling indicates how the phthalazinones interact with apoE4 and disrupt domain interaction, confirming the importance of Arg-61. These approaches can serve as a starting point for further optimization of apoE4 structure correctors.
Pharmacophore modeling has been proven to be successful for identifying various CNS drug targets (48 -51), including catechol-O-methyl transferase (52), neuronal nicotinic acetylcholine receptor agonists (53), -opioid receptor agonists (54), GABA A (55), CNS ion channel modulators (56), and PDE4 (57). Results from the present study are also consistent. The sixfeature pharmacophore model was built on the data in Table 1 and indicates that at least two of the features are essential for activity. This was confirmed by the structure-based modeling and is consistent with the EPR data establishing that residues in the vicinity of Arg-61 were essential for activity. The ligandbased approach can serve as a basis for further development of this important class of compounds.
The concept of structure correctors as a new therapeutic approach for certain targets and diseases has been gaining support since it arose about a decade ago (58 -60). Protein structure and misfolding that affect intracellular transport and function occur commonly in human disease. The cystic fibrosis transmembrane conductance regulator cell surface chloride  FEBRUARY 17, 2012 • VOLUME 287 • NUMBER 8 channel is commonly mutated in cystic fibrosis (61). Approximately 80 -90% of patients have a common deletion (F508 deletion) in the cystic fibrosis transmembrane conductance regulator protein that results in abnormal folding, retention in the endoplasmic reticulum, and impaired Cl Ϫ secretory capacity. Several groups have discovered small molecules that bind and stabilize the protein, enhancing transit to the cell surface (structure correctors) and improving the ion channel function of receptors that reach the cell surface (potentiators) (62)(63)(64)(65). The correctors increase cystic fibrosis transmembrane conductance regulator activity at low millimolar concentrations in cultured cells. Vertex Pharmaceuticals has completed a Phase IIa clinical trial on a structure corrector (VX-809) and a Phase III trial on a potentiator (VX-770) (see the United States Clinical Trials and Vertex Pharmaceuticals Web sites).

Structural Correctors of ApoE4 Domain Interaction
Fabry disease, a lipid storage disorder caused by mutations in ␣-galactosidase A, is also the target of structure corrector or pharmacological chaperone approaches to treatment. Mutated ␣-galactosidase A is retained in the endoplasmic reticulum and prematurely degraded before it reaches the lysosome; the normal ␣-galactosidase A lysosomal enzyme catabolizes the complex lipid globotriaoxylceramide (GL-3). The abnormal accumulation of this ceramide is responsible for an array of problems, including kidney failure, risk of myocardial infarction, and stroke (66 -69). A small molecule has been identified (1-deoxygalactonojirimycin; migalastat) that binds and stabilizes the mutant enzyme, facilitating its transport to lysosomes (58,70) and decreasing the accumulation of the ceramide by increasing the enzyme levels in cultured cells (71) and in transgenic mice (72). Phase II trials (Amicus Therapeutics; AT1001) showed long term safety and tolerability and a reduction in kidney GL-3 (73), and Phase III trials have commenced (74) (see the United States Clinical Trials Web site).
A structure corrector approach to restore normal function to p53 has also been actively investigated. Mutations in p53 result in an instability in the protein, preventing normal target interaction and usually causing accumulation to high concentrations in tumor cells. The mutant p53 prevents apoptosis and sustains tumor cell growth by impairing the ability of the p53 to bind to DNA and regulate gene transcription (75)(76)(77). Small molecules that restore p53 activity and conformation have been identified (77)(78)(79)(80)(81)(82). One interesting compound, CP31398, a styrylquinazoline, restores the structure of mutant p53 to the normal conformation and rescues the p53 function in tumor cells (78,82). Another class of compounds, called PRIMA-1 and APR-246, are pro-drugs that give rise to MQ, a reactive compound that binds to mutant p53 and enhances its apoptotic activity (83,84). The active compound appears to modify thiol groups in the mutant, stabilizing the folding and restoring DNA binding (84). Clinical trials are beginning.
Thus, evidence is accumulating that small molecule structure correctors may be an effective approach to modulate biophysical properties and function of abnormal proteins. As shown in this paper, we identified structure correctors with high potency that modulate the structure of apoE4 by blocking or impairing intramolecular domain interaction. These same small molecules restore the functional intracellular activities of apoE4 to properties similar to those of apoE3 in cultured cells.
Future studies will examine the in vivo effects of small molecule structure correctors and allow the design of active molecules that enter the CNS and block the detrimental effects of apoE4. Further structure-activity relationships will allow the identification of molecules that can be used to prevent or correct the detrimental effects of apoE4 in neuropathology.