Identification of γ-Secretase Inhibitor Potency Determinants on Presenilin*

Production of amyloid β peptides (Aβ), followed by their deposition in the brain as amyloid plaques, contributes to the hallmark pathology of Alzheimer disease. The enzymes responsible for production of Aβ, BACE1 and γ-secretase, are therapeutic targets for treatment of Alzheimer disease. Two presenilin (PS) homologues, referred to as PS1 and PS2, comprise the catalytic core of γ-secretase. In comparing presenilin selectivity of several classes of γ-secretase inhibitors, we observed that sulfonamides in general tend to be more selective for inhibition of PS1-comprising γ-secretase, as exemplified by ELN318463 and BMS299897. We employed a combination of chimeric constructs and point mutants to identify structural determinants for PS1-selective inhibition by ELN318463. Our studies identified amino acid residues Leu172, Thr281, and Leu282 in PS1 as necessary for PS1-selective inhibition by ELN318463. These residues also contributed in part to the PS1-selective inhibition by BMS299897. Alanine scanning mutagenesis of areas flanking Leu172, Thr281, and Leu282 identified additional amino acids that affect inhibitor potency of not only these sulfonamides but also nonsulfonamide inhibitors, without affecting Aβ production and presenilin endoproteolysis. Interestingly, many of these same residues have been identified previously to be important for γ-secretase function. These findings implicate TM3 and a second region near the carboxyl terminus of PS1 aminoterminal fragment in mediating the activity of γ-secretase inhibitors. Our observations demonstrate that PS-selective inhibitors of γ-secretase are feasible, and such inhibitors may allow differential inhibition of Aβ peptide production and Notch signaling.

The most common cause of dementia in the elderly is Alzheimer disease (AD), 3 which bears pathological hallmarks of amyloid plaques and neurofibrillary tangles. Extensive genetic, biochemical, molecular, and cellular studies over the last 2 decades have implicated the production and deposition of A␤ peptides as a critical initiating event leading to the pathogenesis of AD (reviewed in Ref. 1). A␤ peptides are produced by the action of two enzymes, ␤-secretase and ␥-secretase. The ␥-secretase enzyme is a large molecular complex composed of four principal subunits, the presenilins (PS1 or PS2), nicastrin, Aph-1, and Pen-2 (2)(3)(4)(5)(6). ␥-Secretase functions as an intramembranecleaving aspartyl protease (reviewed in Refs. [7][8][9]. In addition to processing of ␤-amyloid precursor protein (APP) and A␤ peptide production, ␥-secretase is also involved in the processing of many additional Type I transmembrane proteins, most notably Notch (reviewed in Refs. 9 and 10). ␥-Secretase cleavage of Notch leads to nuclear translocation of Notch intracellular domain and expression of downstream genes, a process referred to as Notch signaling. Notch signaling is critical for cellular differentiation during development, as well as tissue homeostasis in adults. Pharmacological inhibition of Notch signaling by nonselective ␥-secretase inhibitors in adult mice has been demonstrated to affect tissue differentiation and cell homeostasis in multiple systems (e.g. gastrointestinal tract, pancreas, skin, and lymphocytes) (11)(12)(13)(14), consistent with predictions from KO mouse models (15)(16)(17)(18)(19)(20)(21)(22). Inhibition of Notch signaling by nonselective ␥-secretase inhibitors is a potentially limiting issue for the clinical development of this class of therapeutics. Hence, a major objective of inhibitors targeting ␥-secretase for treatment of AD is the development of selective inhibitors that can reduce A␤ peptide production, without significantly interfering with the processing of other substrates of ␥-secretase, principally Notch.
Two presenilin homologues, referred to as PS1 and PS2, comprise the catalytic core of ␥-secretase. Observations from KO mice show that Presenilin 1 containing ␥-secretase contributes to ϳ80% of total A␤ production in brain, whereas PS2 containing ␥-secretase contributes to ϳ20% of total A␤ (16,17,23,24). Furthermore, the observation that PS1 KO mice exhibit perinatal lethality, whereas PS2 KO mice are viable suggests that PS2-selective inhibitors may spare Notch signaling while lowering A␤ appreciably to be viable therapeutics (16, 24 -27).
We employed transformed fibroblasts from presenilin double-KO mice transiently transfected with Swedish APP751 (APPsw) ϩ PS1 or APPsw ϩ PS2 to identify PS-selective inhibitors. Among the different classes of inhibitors we tested, we noted that although most were equipotent for inhibition of A␤ production from PS1 or PS2 ␥-secretase, sulfonamides exemplified by ELN318463 (this report) and BMS299897 (28 -30) displayed selectivity for inhibition of PS1 ␥-secretase. We exploited this observation to map determinants for PS1 selec-tivity of these sulfonamides using chimeric constructs and point mutants. The results of our studies identify 3 residues as necessary and sufficient for the observed PS-1 selectivity of sulfonamide ELN318463: Leu 172 , Thr 281 , and Leu 282 in PS1. These residues also contribute in part to the observed selectivity of BMS299897. We performed alanine scanning mutagenesis in regions flanking these residues to identify additional residues in PS1 that affect the inhibitor potency of the two sulfonamides as well as that of DAPT and L-685,458. Our findings provide insights into the possible binding sites of these inhibitors and, furthermore, demonstrate that selective targeting of presenilins is a feasible strategy for inhibiting ␥-secretase in an isoformspecific manner for treatment of AD.

EXPERIMENTAL PROCEDURES
Chemical Synthesis-Synthesis of (R)-3-(4-bromobenzylamino)azepan-2-one. D-␣-Amino-⑀-caprolactam (63) (0.04 M) in methanol (100 ml) was treated with 4-bromobenzaldehyde (0.046 m), and the resulting solution was stirred at room temperature for 24 h. Polymer-supported borohydride (40 g; Aldrich on Amberlite IRA-400) was added in one portion, and a slight exotherm and some bubbling was observed. The mixture was stirred at room temperature for 20 h, and the resin was then filtered off and thoroughly washed with methanol. The combined filtrates were evaporated to dryness. The residue was dissolved in ethyl acetate, and the organic phase was washed with water, saturated aqueous sodium bicarbonate, water, and brine. The organic phase was then dried with sodium sulfate. Solids were filtered off, filtrate was evaporated, and residue was recrystallized from hot ethyl acetate (20 ml) and hexane (150 ml). The crystalline product was filtered off, washed briefly with cold ethyl acetate/hexane (1:1) mixture, and air-dried to yield 7.8 g of (R)-3-(4-bromobenzylamino)azepan-2-one having m/z ϭ 299.0.
Molecular Cloning and Construction of Chimeras-Human PS1, PS2, and APPsw cDNA inserts were subcloned into pCF vector, which was modified from pcDNA3 (Invitrogen) by inserting the adenoviral tripartite leader sequence (31) 38 bp upstream of the starting ATG codon, between the cytomegalovirus promoter and the EcoRI site. Presenilin chimeras were constructed by blunt end ligation of PCR-amplified fragments (Pfu Turbo DNA polymerase kit; Stratagene), and sequence was verified prior to transfection into transformed fibroblasts derived from PS1/PS2 double knock-out mice. Alanine scanning mutagenesis of the C2 and C5 subregions was carried out with mutagenic primers and the QuikChange site-directed mutagenesis kit (Stratagene).
Cell Culture and Transfection-Mouse fibroblasts derived from the PS1 Ϫ/Ϫ /PS2 Ϫ/Ϫ double knock-out cells (dKO cells; gift from Dr. Bart De Strooper) (23) were grown at 37°C under 10% CO 2 in Dulbecco's modified Eagle's medium containing 2-10% fetal bovine serum and 100 g/ml penicillin/streptomycin (Invitrogen), and 2 M L-glutamine (Invitrogen). Transfections were performed using Nucleofector II (Amaxa GmbH, Germany) with about 1-10 million of PS1 Ϫ/Ϫ PS2 Ϫ/Ϫ cells, using transfection reagents supplied by the manufacturer. An RPMI-washed cell pellet was resuspended in 100 l of Solution R. To this cell suspension, 1-2 g of a DNA mixture composed of APPsw, (32), and PS1 (2) or PS2 (33) was added, and the cell-DNA mixture was electroporated immediately (program T-20). Electroporated cells were first suspended into 1 ml of warm RPMI and then transferred (within 2-5 min after the addition of RPMI) into 5-10 ml of Dulbecco's modified Eagle's medium with 10% fetal bovine serum, for plating into 96-well plates. For experiments using PS2, cells were plated at a density of ϳ50,000/well in 96-well plates; for experiments involving PS1, cells were plated at a density of ϳ15,000/well in 96-well plates. The cells were treated overnight (14 -20 h) with different concentrations of ␥-secretase inhibitors 1-3 h following plating. During the initial phases of this work, transfection efficiencies were monitored using green fluorescent protein reporter plasmid (ϳ90%) in conjunction with A␤ production from parallel transfections. The levels of A␤ produced were normalized for APP expression levels as well as PS expression levels, using Western blot analysis (see Figs. S1 and 9). Since the level of A␤ produced was highly consistent and reproducible across experiments with Nucleofector-mediated transfection under the experimental conditions described above (as illustrated in Fig. 8), in subsequent experiments with point mutant PS constructs shown in Figs. 5 and 6 and Tables 1 and 2, A␤ production from transfected cultures was used as a surrogate for transfection efficiency.
ELISAs for A␤40, A␤42, Total A␤, or A␤ 1-x-A␤ 1-x ELISA employed antibody 266 (recognizing A␤ 16 -23; Elan) for capture and antibody 3D6 (recognizing A␤ 1-5, with specificity of position 1 of A␤; Elan) for detection. A␤40 ELISA employed antibodies 266 as capture and 2G3 (specific for A␤40) as detection, respectively. A␤42 ELISA employed antibodies 266 as capture and 21F12 (specific for the carboxyl terminus of A␤ 42) as detection, respectively (34). A␤ peptide (California Peptide Research, Napa, CA) was used as concentration standard in the ELISA.
ELISAs were performed at room temperature using 50 l of diluted or original conditioned medium from overnight culture of transfected cells, treated with ␥-secretase inhibitors or Me 2 SO control. Following a 1-h incubation to capture A␤ peptide, the plates were incubated sequentially (45 min/incubation) with biotinylated detecting antibody (0.5 mg/ml), followed by horseradish peroxidase-strepavidin and horseradish peroxidase substrate (one-step slow TMB-Elisa; Pierce; catalogue number 34024). Plates were washed between incubations with Tris-buffered saline plus 0.05% Tween 20. Substrate reac-tions were terminated with 2 NH 2 SO 4 , for absorbance readings with a SpectraMax Plus (Molecular Devices, Sunnyvale, CA).
Determination of Inhibitor Potencies and Statistical Analyses-Dose-response curves and corresponding EC 50 values for the inhibitors described in this report were derived by curve fitting of secreted A␤ levels (determined by ELISA) from cells treated with a range of inhibitor concentrations using the ELfit program (IDBS, Alameda, CA). Due to experimental limitations involving transient transfection of a large number of constructs per experiment, the inhibitor potencies against a given construct or chimera were determined from individual experiments. Conclusions regarding the effect of a chimera or a mutation on inhibitor potency were based on average EC 50 values from replicate experiments. Drift in the absolute value of an inhibitor's potency between experiments over time is evident (e.g. compare ELN318463 potency in Fig. 1 with Fig. 4). However, despite this experimental drift in EC 50 values, the rank order of PS1 selectivity (ELN318463 Ͼ BMS299897) remained invariant. Furthermore, our conclusions from early studies with chimeric construct were corroborated by subsequent studies with point mutants as elaborated under "Results" and "Discussion." For the alanine scanning mutagenesis of the C2 and C5 subregions, the statistical significance of compound potency against a particular alanine mutant was determined using an unpaired t test (GraphPad Prism software package) by comparing the EC 50 of the compound against the mutant versus wild type PS1 (n ϭ 2-4 each).

RESULTS
We examined the inhibition potency of ␥-secretase inhibitors against PS1-or PS2-comprised ␥-secretase using transformed primary embryonic fibroblasts derived from PS1 Ϫ/Ϫ PS2 Ϫ/Ϫ knock-out mice (23) transfected with PS1 or PS2 expression constructs to reconstitute homogenous enzyme complexes in these cells, without the complication of endogenous presenilins. This experimental system allowed us to unambiguously compare functional properties of reconstituted wild type or mutant PS1 or PS2 ␥-secretases. In each experiment, expression constructs for either human PS1 or PS2, together with APPsw, were transiently transfected into PS1 Ϫ/Ϫ PS2 Ϫ/Ϫ dKO cells. The transfected cells were then incubated with different concentrations of inhibitors overnight. Total A␤ levels in conditioned medium from transfected cells were then measured by ELISA. Fig. 1 illustrates that among the classes of ␥-secretase inhibitors we tested, sulfonamides show preferential inhibition of PS1-␥-secretase, whereas nonsulfonamide inhibitors only have modest selectivity for PS1versus PS2-␥-secretase. The dose-response curves and EC 50 values from a representative experiment are shown in Fig. 1. The mean values from two independent experiments on PS1/PS2 selectivity of the inhibitors are shown in Fig. 3. ELN318463 is ϳ51-fold more selective for PS1, and BMS299897 is ϳ35-fold more selective for PS1, whereas L-685,458 is only ϳ3-fold more selective for PS1, and DAPT is actually 2-fold more selective for PS2. Additional sulfonamide inhibitors of the type represented by ELN318463 also displayed preferential PS1 selectivity (data not shown).
The observation of the differential inhibition of PS1 versus PS2, mainly by the sulfonamide series of inhibitors, prompted us to examine the structural basis for this differential inhibition. We employed chimeric PS1/PS2 molecules (illustrated in Fig.  2) to map the domain(s) in PS1 responsible for differences in inhibitor potencies. Evaluation of an initial set of chimeric presenilin molecules revealed that the middle third of PS1 (residues 128 -298) is both necessary and sufficient for its high potency inhibition by ELN318463 and BMS299897 (Fig. 3). For both ELN318463 and BMS299897, the EC 50 values of PS1/2B are similar to that of PS1, whereas EC 50 values of PS1/2A and PS1/2C are similar to those of PS2. More telling, inhibitor potencies against PS2/1C behaved just like PS1, in terms of its inhibition by ELN318463 and BMS299897, despite the fact that the majority of this construct is composed of PS2 sequence. As before (Fig. 1), nonsulfonamide inhibitors, such as DAPT and L-685,458, did not display Ͼ3-fold selectivity for PS1 nor PS2, and the chimeras did not reveal any consistent basis for this low level of selectivity.
We examined the region bounded by PS1 residues 128 -298 in more detail by studying inhibitor sensitivity of additional "small fragment chimeras" (PS1/2C1, PS1/2C2, PS1/2C3, PS1/ 2C4, and PS1/2C5) (Fig. 2) for the two sulfonamide inhibitors. Expression, A␤ production, and endoproteolytic processing of the chimeric PS molecules are indistinguishable from the wild type counterpart (supplemental Fig. S1, B and C). The nonsulfonamide inhibitors were not examined further due to their lack of appreciable PS selectivity (Fig. 3). As shown in Fig. 4, it is clear that for ELN318463 and BMS-299897, only the C2 (containing PS2 residues 171-200 substituted for PS1 residues 165-194) and C5 subregion (containing PS2 residues 275-304 sub-stituted for PS1 residues 269 -298) chimeras reported inhibitor EC 50 values appreciably different from that of parental PS1. Thus, the C2 and C5 subregions largely account for the inhibitor selectivity between PS1 and PS2. The PS2 C2 region in the context of PS1 reduces the potency of Elan G ϳ 3-fold relative to wild type PS1, whereas the PS2 C5 region in the context of PS1 reduces the potency of Elan G ϳ 18fold relative to PS1. It is noteworthy that the product of the -fold selectivity for ELN318463 reported by the C2 and C5 subregions (54-fold) is in close agreement with the average PS1 selectivity of ELN318463 from the experiments reported in Figs. 1, 3, and 4 (39-, 51-, and 65-fold, respectively).
Although both the C2 and C5 subregions contribute to the inhibition potency difference between PS1 and PS2 for BMS299897, the product of the -fold difference in EC 50 values between parental PS1 and the chimeras, PS1/2C2 (2-fold) and PS1/2C5 (4-fold), does not fully account for the ϳ46-fold difference in potency of BMS299897 between PS1 and PS2. One explanation for this observation with BMS299897 may be that both the C2 and C5 subregions from PS1 may need to be present in cis configuration to fully account for the potency difference between PS1 and PS2. The trans arrangement of the C2 or C5 subregions, as it occurs in the chimeras we tested, may mask the combined contributions of the two subregions in native PS1 molecule. The conversion mutants in the C2 and C5 subregions as well as the triple mutant constructs described below provide a test of this assumption. At this point, we can conclude that the C2 and C5 subregions probably make significant contributions to inhibitor selectivity of sulfonamide inhibitors.
Since the C5 subregion (encompassing the C-terminal segment of presenilin NTF) makes a greater contribution for PS1/ PS2 selectivity of ELN318463 (Fig. 4), we tested the five nonconserved residues within the C5 subregion ( Fig. 5A) first for their contribution to the observed selectivity of the sulfonamides. We mutated and assayed five conversion mutant con-    structs in the context of PS1/2C2 (Fig. 5, B and C). For convenience, PS1/2C2 was used as the reference construct in place of PS1. The results obtained using PS1/2C2 as a reference are val-idated by the triple mutation constructs described below as well as by the alanine scanning of the C2 region (see "Discussion").

34±2 430±24
As before, each mutant construct was assayed for inhibitor sensitivity in transiently transfected PSdKO cells treated with a range of inhibitor concentrations. As summarized in Fig. 5C, of the five conversion mutants tested in the PS1/2C2 backbone, T281P and L282I mutations significantly affected the potency of ELN318463 as evidenced by Ͼ2-fold, and Ͼ6-fold increase in EC 50 values, respectively, compared with parental PS1/2C2. The conversion mutations in the PS1 C5 region did not significantly affect the potency of BMS299897 as compared with the parental construct PS1/2C2. This was not surprising, because although the PS2 derived C5 subregion lowered the potency of BMS299897 ϳ4-fold in the context of PS1 alone (Fig. 4), the Ͻ2-fold difference between PS1/ 2C2 and PS1/2C5 (see Fig. 4) falls within the experimental variability of our assay; thus, it would not read out as a "significant" difference. This observation suggests that the 5 residues differing between PS1 and PS2 in the C5 subregion do not contribute more than 2-fold to the loss of potency observed with BMS299897 when tested in the context of PS1/2C2. The contributions of these amino acid residues may become noticeable if tested in the context of PS1 or in combination. This latter possibility was tested experimentally by the triple mutants described below.
We next queried the C2 subregion for dominant residues conferring PS1 selectivity. The C2 subregion in PS1 and PS2 encodes the majority of transmembrane domain 3 (TM3) as well as part of the loop connecting TM3 and TM4. Sequence alignment of PS1 and PS2 reveals seven nonconserved amino acid residues in the C2 subregion (Fig. 6A). Conversion mutants at each of these positions were generated in the context of PS1(L282I) (Fig. 6B) and assayed for the impact on inhibitor potency. We selected PS1(L282I), instead of wild type PS1, as  the backbone for this analysis in order to amplify the relative contribution of any conversion mutant in the PS1 C2 subregion, which alone contributes ϳ3-fold potency difference between PS1 and PS2. Inhibition potency of the sulfonamides against a particular conversion mutant in the C2 subregion of PS1 was referenced against the EC 50 value against PS1/ 2C2(L282I), a PS1/2C2 chimera molecule with the L282I mutation, as illustrated in Fig. 6C. We employed PS1/2C2(L282I) as the reference construct for assessment of PS1 C2 region conversion mutants, because it established the bottom of the range of inhibitor potency for the conversion mutants. Thus, we reasoned that if any one of the seven amino acid residues in the PS1 C2 subregion contributes to the higher inhibition potency of ELN318463 against PS1, a conversion mutant of that PS1 residue into a PS2 residue, in the context of PS1(L282I), should lead to a decrease in potency (i.e. an increase in EC 50 value) comparable with that observed against PS1/2C2(L282I). Indeed, this is what we observed for PS1(L282I,L172M). Inhibitor potencies of both ELN318463 and BMS299897 were lowered ϳ3-fold against this particular mutant (Fig. 6D), and the EC 50 values of both compounds are very similar to their potencies against PS1/2C2(L282I). In contrast, potencies of ELN318463 and BMS299897 are not affected by any of the other six conversion mutants in the C2 subregion.
The data strongly suggest that L172 in PS1 largely accounts for the contribution made by the C2 subregion to the high potency inhibition of PS1 for both ELN318463 and BMS299897. Alanine scanning mutagenesis of Leu 172 in the PS1 context lowered the potency of ELN318463 2.6-fold relative to wild type PS1 (see Table 1 and the alanine scanning results below), in close agreement with the 3-fold effect observed in the context of PS1/2C2(L281), validating use of this construct as a reference. Finally, in our experiments analyzing the C2 subregion, the potencies of ELN318463 and BMS299897 against PS1/2C2(L282I) are within 2-fold of the potencies determined during analysis of the C5 region (Fig. 5), further illustrating the limited range of interexperimental variability we observed during the course of this work.
Having identified Leu 172 , Thr 281 , and Leu 282 in PS1 as being primarily responsible for the high potency inhibition of PS1 by ELN318463 when tested individually and partly responsible for BMS299897, we next tested whether the 3 amino acid residues, when in combination, are both necessary and sufficient in determining PS1/PS2 selectivity for the two sulfonamides. We also included the two nonsulfonamide inhibitors, DAPT and L-685,458 as controls in this experiment to confirm that the effect is specific to sulfonamides. We prepared two composite mutants; triple-PS1 refers to PS1(L172M,T281P,L282I), and triple-PS2 refers to PS2 (M178L,P287T,I288L). The mutated amino acid residues in triple-PS2 correspond to the three identified amino acid residues in triple-PS1.
As revealed by the data shown in Fig. 7, the EC 50 for inhibition of triple-PS1 by ELN318463 is indistinguishable from wild type PS2. Thus, in terms of inhibitor sensitivity, triple-PS1 behaved like PS2, despite the fact that it is largely a PS1 molecule, with only 3 amino acid residues converted into the PS2 sequence (L172M,T281P,L282I). Similarly, the EC 50 for inhibition of triple-PS2 by ELN318463 is indistinguishable from wild type PS1, despite the fact that it is largely a PS2 molecule, with only 3 amino acid residues converted into PS1 sequence (M178L, P287T, and I288L). Furthermore, the ␥-secretase activity of each triple-PS construct is indistinguishable from

subregion alanine scan
Shown is the effect of alanine scanning mutagenesis of residues in the C2 subregion on inhibitor potencies. After deriving EC 50 values for each compound from every mutant construct, the EC 50 value for a given Ala mutation was normalized to the EC 50 value of the wild type PS1 (derived from the same experiment) to obtain a ratio. The effect of mutations can be analyzed by examining this ratio. Statistically significant differences in EC 50 ratios (see "Experimental Procedures") are indicated in solid black type, ratios Ͼ5-fold are denoted in boldface type, and nonsignificant ratios (p Ͼ 0.05) are denoted in light gray type. FAD-associated residues are indicated by arrowheads. Black arrows identify residues that significantly lowered potencies of all four inhibitors tested, and red arrows denote residues that significantly increased potency of the inhibitor(s) tested. The data in this figure are derived from n ϭ 3 independent transfections and EC 50 determinations. We included two PS1 transfections as a control within each experimental set (9 -12 constructs total/set) to enable comparing mutant PS constructs among different experiments. ns, p Ͼ 0.05. that of its wild type counterpart, as measured by A␤ production (supplemental Fig. S1D). The data strongly suggest that the 3 identified amino acid residues are both necessary and sufficient in determining PS1 Ͼ PS2 selectivity for ELN318463. For BMS299897, the triple mutations had a slightly less dramatic effect than that seen for ELN318463. Still, the effect was stronger than what one would have predicted based on the additive effects of single mutations in both the C2 and C5 regions (compare EC 50 values for BMS299897 in Fig. 6D and Fig. 5C with those in Fig. 7B for mutations L172M, T281P, and L282I). As expected, the triple mutations did not have a detectable effect on inhibition potencies of the nonsulfonamide inhibitors, DAPT and L-685,458, confirming that the effects of the three identified amino acid residues were specific to sulfonamide inhibitors.

Com
In order to test if additional residues flanking Leu 172 , Thr 281 , and Leu 282 of PS1 contribute to inhibitor potencies of ELN318463 and BMS299897, we conducted alanine scanning mutagenesis in the C2 and C5 subregions. We also tested the alanine scan mutants to identify any amino acid residues in these subregions for their effect on potencies of nonsulfonamide inhibitors, since this knowledge may shed light on the structural determinants for both types of inhibitors. We first performed alanine scanning around Leu 172 in the C2 subregion of PS1. Briefly, each of the 15 amino acid residues other than an alanine residue at positions 161-176 in human PS1 was individually mutated into an alanine residue. The mutants were then individually co-transfected with APPsw into the PS1 Ϫ/Ϫ PS2 Ϫ/Ϫ dKO cells, and dose response profiles were established with each mutant to determine EC 50

TABLE 2 C5 subregion alanine scan
Shown is the effect of alanine scanning mutagenesis of residues conserved between PS1 and PS2 in the C5 subregion on inhibitor potencies. Designations of residues and inhibitor potencies are same as in Table 1 residues that increased potency of certain inhibitor(s) (red arrow in Tables 1 and 2). The arrowheads in Tables 1 and 2 indicate residues identified as sites of familial Alzheimer disease (FAD) mutations. Three residues identified by alanine scanning to affect inhibitor potency in the C2 subregion have previously been described as sites of FAD mutations in PS1 (arrowheads in Table 1). L166A had the most dramatic effect on all four compounds tested, including sulfonamide and nonsulfonamide inhibitors. Relative to WT PS1, PS1(L166A) lowered the potency of the sulfonamides by ϳ10-fold, whereas the potency of the nonsulfonamides was lowered ϳ4-fold. This position is associated with a highly aggressive FAD mutation, L166P, which has been shown to lead to a 6-fold increase in A␤42 and to decrease both AICD generation and Notch cleavage (35). Consistent with earlier results (35), production of total A␤ relative to WT PS1 was not affected by this mutation (Fig. 8A), and L166A led to elevation of A␤42/A␤ total ratio (Fig. 8C). Inter-estingly, alanine mutation of the other two FAD-associated residues in the C2 subregion (Ser 169 and Leu 173 ) significantly affected the potency of sulfonamides and DAPT but not L-685,458. S169A increased inhibition potency of sulfonamides and DAPT slightly more than 2-fold, with no effect on the potency of L-685,458, whereas Leu 173 lowered potency of sulfonamides and DAPT but not L-685,458. Consistent with FAD mutations S169P and S169L, the S169A mutant exhibited a modest 1.5-2-fold elevation in A␤42/total ratio (Fig. 8C), whereas presenilin endoproteolysis was not affected (Fig. 9B). The S169A and L173A mutants did not affect production of total A␤ relative to wild type PS1 (Fig. 8A). In addition to L166A discussed above, I162A and W165A had a modest effect (3-4-fold reduction of potency) on both sulfonamide and nonsulfonamide inhibitors, whereas S170A primarily affected sulfonamide inhibitors (lowered potency 2.6 -4.5-fold). These mutants were indistinguishable from PS1 with respect to activity, as assessed by A␤ production (Fig. 8A). Although L172A mutation lowered potency of ELN318463 2.6fold. consistent with the 3-fold effect of the L172M conversion mutant (Fig. 6), the result did not reach statistical significance in this experiment (p ϭ 0.14).
Results from a similar alanine scanning mutagenesis of Leu 282 -flanking residues in the C5 subregion of PS1 are summarized in Table 2. Inhibitor potencies against 21 alanine mutants were determined as described above. Five residues in the C5 subregion affected potency of all four inhibitors tested: Leu 271 , Arg 278 , Leu 282 , Leu 286 , and Ile 287 (denoted by black arrows in Table 2). Of particular note, R278A had the most . Activity of alanine scanning mutants as assessed by A␤ production. The graphs above show the relative A␤ levels of the mutants, expressed as a ratio of A␤ produced from the mutant relative to wild type PS1. The activity of alanine scanning mutants in the C2 and C5 regions of PS was assessed by measuring total A␤ produced (1-x) from the mutant compared with total A␤ from wild type PS1 within the same experiment. The arrows denote the select mutations identified in Tables 1 and 2, which affected the potency of multiple compounds. The red arrows indicate mutants that increased potency Ͼ2-fold; black arrows indicate mutants that decreased potency Ͼ2-fold. The Y288A mutation is inactive. A, A␤ production from alanine scanning mutants in the C2 subregion relative to wild type PS1. B, A␤ production from alanine scanning mutants in the C5 subregion relative to wild type PS1. In A and B, the values for A␤ production are derived from n ϭ 3 independent experiments, and the error bars show S.E. C, assessment of A␤42/A␤ 1-x ratio from select alanine mutants of FAD residues (n ϭ 2). dramatic effect on all of the compounds tested. This mutation reduced the inhibition potency for L-685,458 by ϳ120-fold and about 10-fold for ELN318463, BMS299897, and DAPT. Arg 278 is situated near the amino terminus of cytosol-localized hydrophobic region 7 (36 -38). One study reported that R278T was associated with early onset FAD (39), but no further molecular or biochemical studies were reported for this PS1 FAD mutant. Two other FAD mutations, R278K and R278I, have been described at this site. A␤ production from this mutant was 50% of wild type PS1 (Fig. 8B). The mutant exhibited a 5-fold elevation in A␤ 42/total ratio over wild type PS1 (Fig. 8C), and a concomitant decrease in PS1 endoproteolysis (Fig. 9B). The FAD-associated L271A also affected both sulfonamide and nonsulfonamide inhibitor potencies. The Y288A mutation is also noteworthy, since it led to complete inactivation of PS1, despite normal endoproteolysis (Fig.  9B). Our findings corroborate the functional significance of Tyr 288 described previously (40). Interestingly, our mutational scan of two upstream residues, L286A and I287A, revealed an affect on the potency of all classes of inhibitors, without significantly affecting A␤ production (Fig. 8B). PS endoproteolysis was not affected by the L286A mutation (Fig. 9B). Consistent with FAD mutations at codon 286, L286A displayed elevation in A␤42/total ratio (Fig. 8C).
In contrast with the T281P conversion mutation identified above (Figs. 5 and 7), the T281A mutation did not affect potency of ELN318463 in the context of PS1. However, consistent with the 6.5-fold loss of potency effected by the conversion mutation L282I (Fig. 5), L282A lowered the potency of ELN318463 10-fold, with lesser effects on the other three inhibitors. Interestingly, V272A also reveals a strong effect on potency of ELN318463 as L282A, with no effect on BMS299897 inhibition, suggesting that Val 272 may also be involved in ELN318463 activity. A␤ production was not affected by V272A, T281A, and L282A mutants relative to wild type PS1 (Fig. 8B). Taken together, these observations support our studies with chimeric constructs, which mapped residues in the C5 subregion as contributing to the PS1 selectivity of ELN318463.
Although our alanine scanning analysis of the C2 and C5 subregions was driven by our observation of PS1-selective inhibition by sulfonamides, we identified certain residues that affected potencies of DAPT as well as L-685,458. As noted, many of these residues coincided with sites of FAD mutations in the C2 and C5 subregions (see "Discussion"). Interestingly, many of the non-FAD residues that affected inhibitor potency also elevated the A␤42/A␤ 1-x ratio (see Fig. S2 and "Discussion"). Furthermore, our study also identified certain residues that affect potencies of the nonsulfonamides exclusively. In this last category, P284A is noteworthy for affecting potency of DAPT and L-685,458, whereas alanine mutation of Arg 278flanking residues E277A and N279A, affected the potency of L-685,458 exclusively. Curiously, alanine mutation of Phe 283 , the residue upstream of Pro 284 , increased potency of BMS299897. Mutation of the FAD residue Glu 280 lowered potency of ELN318463, DAPT, and L-685,458, with a concomitant reduction in A␤ production (relative to WT PS1) (Fig. 8B).
In summary, alanine scanning mutagenesis of the C2 and C5 subregions in PS1 largely corroborates our studies mapping selectivity determinants for sulfonamides to these regions. In addition, the experiments reveal determinants in common with sulfonamides that affect the potency of additional classes of ␥-secretase inhibitors, namely DAPT and L-685,458.

DISCUSSION
We report the first observation of PS1-selective inhibition of ␥-secretase by sulfonamides. Using a series of chimeric constructs and point mutations, we identify the structural determinants for PS1-selective inhibition by ELN318463 and, to a lesser extent, BMS299897. Chimeric constructs offer a powerful and convenient solution for identifying domains in a protein that contribute to a particular phenotype when there is significant divergence in primary sequence between homologues. Our chimera studies identified two subregions in PS1 (designated C2 and C5, encompassing TM3 and hydrophobic region 7 following TM6, respectively), as comprising the principal determinants for PS1 selectivity of sulfonamide inhibitors. The divergence between PS1 and PS2 in the C2 and C5 subregions is among the lowest between the two presenilins. This implies that if other regions share a similar degree of divergence with the C2 or C5 subregions between PS1 and PS2 and make similar contributions to inhibition potency of sulfonamides, they would have been revealed in our chimera studies.
We identified 3 amino acid residues in PS1, Leu 172 , Thr 281 , and Leu 282 , that are largely responsible for the observed PS1 selectivity of sulfonamide inhibitor ELN318463 and, to a lesser extent, BMS299897. Since we did not exhaustively test all double mutant combinations of these residues, we cannot state with certainty that all 3 residues are necessary for completely affecting the potency of the sulfonamides we tested. However, the double mutant we did test, PS1(L282I,L172M) in Fig. 6D, had an intermediate effect on the potency of both sulfonamides tested when compared with the triple PS1 mutant (Fig. 7). This observation suggests that the other combinations of double mutant PS constructs would likewise have had an intermediate effect on inhibitor potency.
The difference in inhibition potency between PS1 and PS2 by sulfonamides could reflect either a difference in compound binding affinity or a difference in allosteric inhibition subsequent to compound binding. Our experimental approach does not distinguish between the two possibilities. However, the highly conservative L282I mutation in PS1 led to a 6.4-fold reduction in inhibition potency for ELN318463, whereas a nonconservative T281P mutation of the adjacent residue, a potentially much more disruptive change, only led to a ϳ2-fold decrease in potency for ELN318463 (Fig. 7C). This observation suggests that L282 in PS1 may make physical contact with ELN318463, whereas other amino acid residues (Leu 172 and Thr 281 in PS1) influence potency of ELN318463 indirectly. Verification of this assumption must await direct binding studies with cross-linkable compounds. Alternatively, the impact of the triple mutants on inhibitor potency could reflect a difference in the underlying biology of the presenilins as described by Mastrangelo et al. (41) (e.g. a difference in subcellular localization or in affinity for interacting proteins that indirectly affect sensitivity to inhibitors).
Our conclusions regarding selectivity determinants from chimeric constructs and conversion point mutants in the C2 and C5 subregions were confirmed by the subsequent results with the triple mutants (triple-PS1 and triple-PS2). Likewise, the data from conversion mutants in the C2 and C5 subregions was corroborated by the results of the alanine scanning experiments in these regions. Specifically, alanine scanning of every residue throughout the C2 and C5 regions did not reveal additional nonconserved residues that affected inhibitor potency. The additional residues revealed by alanine scanning to affect inhibitor potencies in the C2 and C5 subregions were all conserved between PS1 and PS2. Overall, the alanine mutations did not grossly affect function, as measured by total A␤ levels normalized relative to that of wild type PS1 (Fig. 8, A and B). Two noteworthy exceptions in the C5 subregion that affected A␤ production, Arg 278 and Tyr 288 , are discussed below. Alanine scanning mutagenesis of conserved residues in the C2 and C5 subregions revealed a preponderance of FAD-associated amino acids among the residues that affected potencies of both sulfonamide and nonsulfonamide inhibitors. A concordance between FAD residues and inhibitor sensitivity has been noted previously using peptidomimetic (42), active site (43), and most recently nonsteroidal anti-inflammatory drug (44) classes of ␥-secretase inhibitors. We therefore examined whether the FAD as well as non-FAD residues in the C2 and C5 subregions that affected inhibitor potencies in our alanine scanning studies impacted A␤42/A␤total ratio. Interestingly, many (but not all) alanine mutants of FAD as well as non-FAD residues elevated the A␤42/A␤ 1-x ratio (Figs. 8C and S2). The results from this analysis revealed that in addition to the known FAD residues noted in Tables 1 and 2, alanine mutation of non-FAD residues Trp 165 , Phe 283 , and Ile 287 also elevated A␤42/ A␤total ratios. In contrast, I162A and F283A did not result in elevation of the A␤42/A␤total ratio relative to WT PS1. Thus, there is a strong (but not strict), concordance between inhibitor sensitivity and elevation of A␤42/A␤total. We also observe that under our experimental conditions, alanine mutation of FAD residue Thr 274 caused neither an elevated A␤42/A␤total ratio nor a difference in inhibitor sensitivity relative to WT PS1 (Table 2 and Figs. S2 and S3).
Taken together, the observed concordance between inhibitor sensitivity and elevation of the A␤42/A␤ 1-x ratio suggests that the different classes of inhibitors not only share common structural determinants affecting potency but also that these common determinants coincide with residues in PS1 previously known to be functionally significant. The mutants L166A, R278A, and L286A had very large (Ͼ5-fold) effects on the inhibition potency of all classes of inhibitors we tested. The FAD residue Leu 166 has been demonstrated to differentially influence ⑀ (S3-like) and ␥ (S4-like) cleavage of substrates (35). The effects we observe upon alanine mutation of residue Arg 278 (A␤ production and reduced presenilin endoproteolysis) are similar to the observations of Nakaya et al. (45) from the FAD R278I mutation of PS1, whereas mutagenesis of Leu 286 has been documented to differentially affect A␤42 production and Notch cleavage (46). Alanine mutation of Leu 271 also affected sulfonamide and active site inhibitor potencies and, to a lesser extent, DAPT, and the FAD L271V has been associated with increased expression of exon 8-deleted transcripts, producing nonfunc-tional protein with regard to GSK 3␤ and -binding activities of PS1 (47).
The second noteworthy exception of alanine mutation that affected A␤ production in our studies is Y288A, which resulted in undetectable levels of A␤ (despite normal PS1 expression and processing) (Fig. 9B). Our findings on ␥-secretase activity and presenilinase processing of PS1 from this mutant are consistent with those of Laundon et al. (40), who reported that this mutation significantly lowered A␤ production from a APP-C99 substrate. Residues 286 -288 lie proximal to the presenilinase cleavage site (48,49), and in addition to the effects of Leu 286 mutation on inhibitor potency noted above, alanine mutation of Ile 287 also significantly affected inhibitor potencies. These residues fall within the PALIY motif identified by Laudon et al. (40) as important for assembly of PS1 NTF and C-terminal fragment into functional ␥-secretase high molecular weight complex. The preponderance of residues in the conserved cytosolic loop/hydrophobic region 7 of PS1 that affect the potency of all classes of inhibitors tested reveals a new role for this domain and suggests that this region is an important site for effecting inhibition of ␥-secretase by multiple classes of inhibitors.
The relationship among different types of inhibitors (sulfonamide, BMS299897; peptidic, DAPT; transition state isostere, L-685,458) revealed in our studies was not anticipated by earlier investigations into compound binding sites nor mechanism of action. For example, it was reported that BMS299897 did not compete with L-685,458, suggesting that sulfonamide inhibitors are allosteric inhibitors, in contrast to transition state isosteres, which directly bind to the enzyme's active site (50). Separately, it was suggested that DAPT mainly inhibits ␥-site cleavage, whereas L-685,458 mainly inhibitsand ⑀-site cleavage (51).
On the other hand, our alanine scanning experiments revealed several amino acid residues in TM3 as well as in the C terminus of PS NTF that affected the inhibition potency of all of the inhibitors tested (discussed above). This finding suggests that TM3 (the C2 subregion) and the C terminus of PS NTF (the C5 subregion) may be in close proximity, analogous to recent reports based on competition studies (52) and cysteine scanning mutagenesis (53), which revealed that (a) the binding sites for DAPT and transition state analogues partially overlap and (b) the proximity of inhibitor-interacting residues in TMD6 and TMD7. Hence, our data are not inconsistent with the recent finding on the binding site for DAPT (53,54). In fact, the two results may complement each other. Morohashi et al. (54) employed biochemical labeling to identify protein fragments that are in the vicinity of the DAPT-binding site and found that (a) DAPT mainly labels TMD7 of PS; (b) sulfonamide inhibitor can compete effectively with DAPT; and (c) L-685,458 attenuated DAPT labeling only at higher concentrations. In a follow-up study, Sato et al. (53) identified Leu 250 in TMD6 and Leu 383 in TMD7 as the primary residues for DAPT binding.
Hence, based on the information generated by this and the aforementioned studies, we can speculate that the binding pocket for ␥-secretase inhibitors may be composed of residues from several regions of presenilin. It is conceivable that the TMD6 and TMD7 binding sites (identified by Morohashi et al. (54) and Sato et al. (53)) and the TMD3 and the C terminus of PS NTF found by us are close to each other in three-dimensional space and form an extended binding site for different classes of inhibitors. This extended binding site may be spatially close to the active site or even overlap with the active site (composed of the catalytic aspartates in TMD6 and TMD7), because mutations in the C2 and C5 subregions led to changes in potencies for transition state analogue inhibitor. Biochemical labeling is a more direct approach for identifying binding site residues than the inhibition potency studies we performed. Since inhibition potency difference can be due to either a difference in compound binding or a difference in subsequent allosteric changes, we cannot definitively conclude that those identified residues are directly involved in compound binding. Hence, our findings provide an initial mechanistic understanding of how inhibitors interact with the ␥-secretase. More extensive structure-function studies (e.g. direct chemical labeling studies with different compounds) are needed to advance our understanding of inhibitor/␥-secretase interaction.