Mechanism of Inhibition of β-Site Amyloid Precursor Protein-cleaving Enzyme (BACE) by a Statine-based Peptide*

Inhibition of β-site amyloid precursor protein-cleaving enzyme by a statine-based inhibitor has been studied using steady state and stopped-flow methods. A slow onset rate of inhibition has been observed under steady state conditions, and aK i of 22 nm has been derived using progress curves analysis. Simulation of stopped-flow protein fluorescence transients provided an estimate of theK d for initial inhibitor binding of 660 nm. A two-step inhibition mechanism is proposed, wherein slower “tightening up” of the initial encounter complex occurs. Two hypotheses have been proposed in the literature to address the nature of the slow step in the inhibition of aspartic proteases by peptidomimetic inhibitors: a conformational change related to the “flap” movement and displacement of a catalytic water. We compared substrate and inhibitor binding rates under pre-steady-state conditions. Both ligands are likely to cause flap movement, whereas no catalytic water replacement occurs during substrate binding. Our results suggest that both ligands bind to the enzyme at a rate significantly lower than the diffusion limit, but there are additional rate limitations involved in inhibitor binding, resulting in ak on of 3.5 × 104 m − 1 s− 1for the inhibitor compared with 3.5 × 105 m − 1 s− 1for the substrate. Even though specific intermediate formation steps might be different in the productive inhibitor and substrate binding to β-site amyloid precursor protein-cleaving enzyme, a similar final optimized conformation is achieved in both cases, as judged by the comparable free energy changes (ΔΔG of 2.01versus 1.97 kcal/mol) going from the initial to the final enzyme-inhibitor or enzyme-substrate complexes.

A substantial body of evidence indicates that accumulation of insoluble plaques in the brain is an important step in the pathogenesis of Alzheimer's disease (1). The extracellular amyloid plaques consist of aggregates of amyloid ␤-peptide isoforms, which are proteolytically derived from the amyloid precursor protein by two proteases, ␤and ␥-secretases. Until recently, the identities of these proteases have been elusive. Within the past 2 years the involvement of presenilins in ␥-cleavage (2) has been suggested. At the same time a novel transmembrane aspartic protease has been identified as the ␤-site amyloid precursor protein-cleaving enzyme (BACE) 1 (3)(4)(5). Even though the BACE polypeptide sequence appears to be most closely related to the pepsin aspartic protease family (5,6), the enzyme is not inhibited by pepstatin, suggesting significant differences at the active site level.
The proposed chemical mechanism for aspartic proteases involves activation of the attacking water molecule by the general base Asp-COO Ϫ with concomitant protonation of the substrate carbonyl by a general acid Asp-COOH, yielding a tetrahedral intermediate amide hydrate (7). The statine moiety of pepstatin constitutes a tetrahedral, hydroxymethylene-isosteric replacement for the scissile peptide bond, mimicking the putative reaction intermediate and resulting in potent inhibition of aspartic proteases. Several attempts to design ␤-secretase inhibitors, based on this chemical mechanism, have been made recently. Incorporation of the statine moiety into the P10-P4Ј peptide 2 representing the Swedish variant of the substrate sequence (Lys 3 Asn/Met 3 Leu at the P2-P1 positions) resulted in moderate inhibition of BACE (IC 50 ϳ 40 M (8)). Substitution of the P1Ј position Asp by Val in the same peptide resulted in a much more potent peptidomimetic inhibitor (IC 50 ϳ 30 nM (8)). Likewise, the replacement of the peptide bond between P1 and P1Ј by a hydroxyethylene isostere in a substrate octapeptide, together with a Asp 3 Ala substitution at the P1Ј position, yielded an inhibitor with an IC 50 of ϳ1 nM (9,10). The latter inhibitor was co-crystallized with a truncated version of BACE, and a significant number of protein-inhibitor interactions have thus been defined (11).
The interactions of eight residues of the inhibitor with BACE include four hydrogen bonds between two active site aspartates and the hydroxyl of the transition state isostere and 10 hydrogen bonds from different parts of the cleft and flap of the protein to the inhibitor backbone. Interestingly, binding of peptidomimetic inhibitors to the human immunodeficiency virus, type 1 protease involves substantial conformational changes, especially in the flap region, where backbone movements as large as 7 Å are observed (12). Considering the mechanistic similarity between BACE and human immunodeficiency virus protease, binding of peptidic inhibitors to BACE is likely accompanied by similar flap (residues VPYTQGKW) movement, and the tryptophan (Trp 137 ) in the flap region would be expected to provide a fluorescence probe for structural change studies. Additionally, pepstatin has been shown to be a slow and tight binding inhibitor of another aspartic protease, pepsin (13); hence, one might anticipate that the slow onset of inhibition is related to conformational changes accompanying inhibitor binding.
In this paper we report the kinetics of BACE inhibition by a statine-based inhibitor. These data are accompanied by stopped-flow studies of the intrinsic protein fluorescence change upon enzyme-inhibitor and enzyme-substrate complex formation, and the results are interpreted in light of the available three-dimensional structure.

EXPERIMENTAL PROCEDURES
Enzyme Expression and Purification-The cell line expressing the truncated (C-terminal His-tagged) version of human ␤-secretase was as described elsewhere (14). About 300 -400 ml of clarified media containing secreted enzyme was dialyzed overnight into phosphate-buffered saline (Life Technologies, Inc.) and loaded onto a 30-ml nickel-nitrilotriacetic acid (Qiagen) affinity column. The column was washed initially with 60 ml of 50 mM phosphate buffer, 0.3 M NaCl, 10 mM imidazole, pH 8.0, and then with 60 ml of the same buffer, but with the imidazole concentration raised to 20 mM. The protein was eluted with a 20 -250 mM linear imidazole gradient. Active fractions were pooled, dialyzed against 20 mM triethanolamine, loaded on a HR10 Mono Q column (Amersham Pharmacia Biotech) and eluted with a 0 -1 M NaCl linear gradient. Active fractions exhibiting a single band on SDS-polyacrylamide gel electrophoresis with Coomassie Blue staining were pooled and stored in 10% glycerol at Ϫ80°C. Protein concentration was estimated by the Bradford assay and by active site titration with the statinebased tight-binding inhibitor (Enzyme Systems Inc.); the results indicated that more than 95% of the total protein was catalytically active. These results were supported by matrix-assisted laser desorption ionization spectroscopy, suggesting that all of the protein was proteolytically activated with the prodomain being cleaved off during expression. 3 Enzyme Assays and Inhibition Studies-Enzyme activity was monitored following the increase in fluorescence at 400 nm (excitation at 328 nm) resulting from the cleavage of the peptide: 7-methoxycoumarin-4acetyl-EVNLDAEF(K-dnp)-COOH. Assays were performed in 50 mM acetate buffer with 0.25 mg/ml bovine serum albumin, pH 4.5 at 25°C with 2.5% Me 2 SO (substrate solvent) present in 96-well plates. Steadystate kinetic studies were performed on a Molecular Devices Spectra-MAX Gemini XS fluorescence plate reader. Data were processed using Softmax Pro 3.1.1.
Substrate concentration did not exceed 25 M, since above this concentration a decrease in signal was observed due to the inner filter effect, even though saturation had not been reached (14).
Inhibition studies were performed using a known statine-based inhibitor: Ac-KTEEISEVN(statine)VAEF-COOH (Enzyme Systems Inc.). Assays were performed by mixing the enzyme solution (final concentration: 20 nM) with the inhibitor (20 -80 nM) and immediately starting the reaction by addition of substrate (final concentration: 25 M). Progress curves were fit to the Morrison equation (15), modified for the depletion of the free enzyme and free inhibitor populations that occurs, , and [I] are the product, enzyme, and inhibitor concentrations, respectively, v i is the initial velocity, v s is the steady-state velocity, and k obs is the pseudo-first order rate constant for the approach to the steady state. Intrinsic Protein Fluorescence Measurements-Protein fluorescence spectra in the presence and absence of the inhibitor were obtained with a Hitachi F-2500 fluorescence spectrophotometer. An excitation wavelength of 280 nm was used, and emission spectra were recorded from 300 to 400 nm at room temperature.
Stopped-flow Experiments-Rapid kinetic studies were conducted using an Applied Photophysics SX. 18MV stopped-flow spectrofluorometer with excitation at 280 nm. A glass cutoff filter (WG-320) was installed in front of the photomultiplier. Experiments were performed by rapid mixing of equal volumes of enzyme (final concentration: 6 M) and inhibitor in sodium acetate buffer, pH 4.5 at 20°C. Stopped-flow spectrophotometric data for inhibitor binding were fit using KIMSIM and FITSIM simulation software. In an alternative experiment 3-5 M of the enzyme was mixed with 5-30 M substrate (Ac-EVNLDEEF(K-dnp)-OH). Typically, three traces were averaged to yield the final trace. These traces were fit to a double exponential equation and k obs was plotted versus substrate concentration.

Steady-state Inhibition of BACE-
The statine-containing peptide has been reported to be a potent inhibitor of human brain-derived BACE, displaying an IC 50 of 30 nM (8). We have reevaluated the inhibition of recombinant, truncated human BACE by this compound. Careful analysis of the product progress curves in the presence of this compound (Fig. 1A) revealed curvature consistent with slow onset inhibition (15). The biphasic nature of these progress curves was well modeled by Equation 1, allowing estimation of v s , v i , and k obs at each inhibitor concentration. A plot of k obs as a function of inhibitor concentration (Fig. 1B) is linear, suggesting either a single simple binding event, as in Scheme 1, or a two-step reaction mechanism (Scheme 2) for which K i * Ͻ Ͻ K i (see below).
The values of k 1 and k 2 were calculated from the slope and intercept of the linear fit in Fig. 1B and were thus determined to be (3.5 Ϯ 0.7) ϫ 10 4 M Ϫ1 s Ϫ1 and (7.8 Ϯ 4.0) ϫ 10 Ϫ4 s Ϫ1 , respectively. Hence, the value of K i obtained by the ratio k 1 /k 2 is 22 Ϯ 5 nM, in good agreement with the IC 50 reported earlier (8).
For all of the fits in Fig. 1A, the value of v s was non-zero, indicating that inhibition is reversible. To confirm this BACE (10 M) was incubated with inhibitor (10 M) for 15 min and then diluted ϳ2000-fold into the activity assay (final concentration of both enzyme and inhibitor was 5 nM). As seen in Fig.  2, this dilution resulted in a slow recovery of enzymatic activity that could be well fit by Equation 1, yielding an estimate of the reactivation rate constant k react of (9.4 Ϯ 0.3) ϫ 10 Ϫ4 s Ϫ1 . We note that the value of k react and k 2 are, within experimental error, the same; this result is consistent with fully reversible inhibition.
The statine-based inhibitor is composed of a peptide sequence similar to a known substrate of the enzyme, but the scissile bond is replaced to mimic the tetrahedral reaction intermediate of aspartyl proteases. We assume, therefore, that this inhibitor binds to the enzyme active site, with the statine isostere engaging the active site aspartate residues. We have not, however, been able to experimentally verify the competitive nature of this inhibitor because of the high K m of the substrate, which is beyond the solubility limits of this molecule (14). Hence we could not achieve sufficient substrate saturation to discern a substrate dependence on inhibition and thus confirm competitive inhibition (16).
Equilibrium and Pre-steady State Binding of Inhibitor to BACE-The amino acid sequence of the catalytic domain of human BACE contains 5 tryptophan residues. When the enzyme (1 M) is excited with 280 nm light, the fluorescence maximum is observed at 330 nm (data not shown), indicating that the majority of tryptophan residues in this protein experience a hydrophobic environment. The recent crystal structure of the human BACE catalytic domain complexed to a hydroxyethylene-containing peptide inhibitor reveals that there is a tryptophan residue in the "flap" region forming part of the binding pocket of the enzyme. It thus seemed likely that the fluorescence spectrum would be perturbed by ligand binding and subsequent flap movement, providing a convenient bio- physical marker of complex formation. Indeed, the fluorescence spectrum of BACE demonstrates a moderate shift of the emission wavelength and a 4.5% increase in fluorescence intensity when the inhibitor (5 M) is added. These spectral changes are reminiscent of those seen for inhibitor binding to the human immunodeficiency virus, type 1 aspartyl protease. In this case the fluorescence changes were associated with movement of the flap region of the enzyme that lays down on top of the inhibitor binding pocket and results in formation of high affinity enzyme-inhibitor complex (17). The spectral changes seen upon inhibitor binding to BACE are too modest to be reliably used to follow complex formation by equilibrium methods. They do, however, provide sufficient signal to follow binding interactions by transient spectroscopic methods. With stopped-flow instrumentation, we have used the changes in BACE tryptophan fluorescence to follow the time course of inhibitor binding.
When a 6 M (final concentration) solution of the enzyme was mixed with 5, 10, or 15 M solutions of the inhibitor under pre-steady-state conditions, an increase in fluorescence signal was observed (Fig. 3A). Higher inhibitor concentrations were not experimentally attainable due to the solubility limits of the inhibitory peptide. Hence, pseudo-first order conditions ([I] Ͼ Ͼ [E]) could not be achieved. Decreasing the enzyme concentrations resulted in insufficient signal amplitude. For these reasons in the concentration range of 5-15 M no dependence of k obs on inhibitor concentration was observed; hence we were not able to estimate k 1 and K d . Therefore, the same data were fit to a simple second order reaction (A ϩ B 3 C) using the program KINSIM, to obtain estimates of k 1 and k 2 . Initially, fitting was optimized by incremental adjustment of values of individual rate constants by visual inspection. Once reasonable fits of the experimental data were obtained, final parameter adjustment was made by use of the program FITSIM (Table I). The average value of the second order rate constant (k 1 ) thus obtained from the binding transients was (5.0 Ϯ 1.4) ϫ10 4 M Ϫ1 s Ϫ1 , which is in good agreement with the value of k 1 obtained from the measurements of steady-state inhibition (Fig. 1B, (3.5 Ϯ 0.7) ϫ 10 4 M Ϫ1 s Ϫ1 ). In contrast, however, the average value of k 2 from these experiments was (3.3 Ϯ 2.5) ϫ 10 Ϫ2 s Ϫ1 , some 43-fold faster than the corresponding rate constant for inhibitor dissociation obtained from steady-state measurements (7.8 Ϯ 4.0) ϫ 10 Ϫ4 s Ϫ1 . The K d estimated from these data by the ratio of k 2 /k 1 is thus 660 nM.
The discrepancy between the dissociation rate constants determined from pre-steady-state binding and steady-state inhibition measurements suggest a two-step inhibition mechanism, according to Scheme 2 (see above). The stopped-flow measurements essentially report on formation of the initial encounter complex, EI, while steady-state inhibition data reflect formation of the final complex, E*I. The similarity of the measured association rates for the two sets of experiments suggest that a common step is rate-limiting to both processes and must therefore be associated with initial binding of the inhibitor. The steady-state data do not directly reflect the two-step nature of the E*I formation, because the intermediate species EI does not accumulate under our experimental conditions ([I] Ͻ Ͻ K d ).
Pre-steady-state Substrate Binding to BACE-Considering the general ligand interactions with the active site of BACE, we wondered if substrate binding might also proceed through a two-step mechanism like that seen for the inhibitor. To address this we measured the binding of the substrate to the enzyme under pre-steady-state conditions, on a time scale where substrate cleavage and product dissociation is negligible. For these studies we used the 9-residue peptide substrate described under "Experimental Procedures." This peptide represents the region proximal to the scissile bond of the Swedish mutant amyloid precursor protein substrate and contains a dnp chromophore appended to the ⑀-amino group of the lysyl residue at P5Ј. The absorbance spectrum of dnp overlaps with the emission spectrum of the tryptophan residues in BACE. Hence, binding of the peptide to the enzyme should result in quenching of the proximal tryptophan fluorescence. Indeed, when the enzyme (3 M) and substrate (5-40 M) are mixed, a reduction in fluorescence signal is observed, as illustrated in Fig. 3B. No recovery of fluorescence intensity is observed over the time course studied (5 s), suggesting that cleavage and product dissociation occur at a rate less than 0.2 s Ϫ1 .
The diminution of fluorescence seen in Fig. 3B could not be adequately described by a single exponential process. Rather, the time course was best fit to a double exponential model. Similar data were obtained at a number of substrate concentrations ranging from 5 to 40 M. These two kinetic phases could correspond to a two-step substrate binding mechanism, similar to that proposed for the inhibitor, as represented by Scheme 3.
If so, one would expect that for the first phase, corresponding to ES formation, k obs1 would display a linear dependence on substrate concentration, with slope and y intercept values equal to k 1 and (k 2 ϩ k 3 ϩ k 4 ) (18). The second phase, corre- sponding to formation of E*S, would be expected to display a hyperbolic dependence of k obs2 on substrate concentration, according to Equation 3, where K d is the dissociation constant for the initial encounter complex, ES (i.e. K d ϭ k 2 /k 1 ). The dissociation constant for the E*S binary complex, K d *, is then given by Equation 4.
The first phase, corresponding to ES formation, could be described by the values of k 1 (i.e. the slope of the plot in Fig. 4A) of (3.52 Ϯ 0.31) ϫ 10 5 M Ϫ1 s Ϫ1 and (k 2 ϩ k 3 ϩ k 4 ) (i.e. the intercept of the same plot) of (3.9 Ϯ 0.4) s Ϫ1 . The second phase (Fig. 4B), corresponding to formation of E*S, could be described by Equation 3. Assuming that k 4 is very small, the K d and (k 3 ϩ k 4 ) values obtained from the data fit to Equation 3 are 21.1 Ϯ 4.9 M and 2.1 Ϯ 0.2 s Ϫ1 , respectively. Since calculation of the K d * value from the empirical data was limited by an unknown k 4 value, we simulated all four constants. Simulation of the binding progress curve data in Fig. 3B using a two-step substrate binding scheme (using the FITSIM program) yielded the following estimates of the kinetic parameters: k 1 ϭ (3.2 Ϯ 0.2) ϫ 10 5 M Ϫ1 s Ϫ1 , k 2 ϭ 6.5 Ϯ 1.2 s Ϫ1 , k 3 ϭ 2.7 Ϯ 0.3 s Ϫ1 , and k 4 ϭ 0.10 Ϯ 0.05 s Ϫ1 . The calculated values of K d and K d * were thus 20.3 M and 725 nM, respectively. The calculated values of K d and k 1 are in a good agreement with those determined from the k obs1 and k obs2 dependencies on substrate concentration (Table I), which confirms the validity of the proposed two-step substrate binding mechanism. DISCUSSION The data presented here suggest that both substrate and inhibitor binding to BACE induce structural rearrangement (most likely flap closing) of the initial enzyme-ligand complex, leading to a higher affinity binary complex. While the association rates for both ligands are significantly slower than the diffusion limit (19), our kinetic results demonstrate that the substrate binds to free enzyme about 10-fold and dissociates about 200-fold faster than does the inhibitor. These differences between the inhibitor and substrate binding/dissociation could be due to the different length of the peptides, or the Asp 3 Val substitution at the P1Ј position in the inhibitor, which has been reported to increase the affinity more than 1000-fold (8). Structural differences between the inhibitor and substrate at the scissile bond position (statine moiety with the 3(S)-hydroxyl group mimicking the reaction intermediate in the inhibitor molecule) could as well contribute to the faster substrate binding, since the free enzyme might be better designed to accommodate the ground state structure (substrate), even though the transition state analog would ultimately display much higher affinity. Finally, displacement of the catalytic water by the 3(S)-hydroxyl group of the statine could account for the slower association of the inhibitor compared with substrate as has been proposed for other aspartyl proteases (13,20).
Despite the differences in individual rate constants for the two ligands, the thermodynamic changes that accompany binding are remarkably similar. The difference in free energy between the initial encounter complex and the final enzymeligand binary complex can be measured from the ratio of dissociation constants for the two enzyme forms, ⌬⌬G ϭ RT ln͑K d */K d ) (Eq. 5) where R is the ideal gas constant and T is temperature (in Kelvin). Using this equation the ⌬⌬G calculated for the transition EI 3 E*I is 2.01 kcal/mol. When the same calculation is performed for the transition ES 3 E*S, a value of 1.97 kcal/mol is obtained. The close agreement between these values strongly implies that a common set of structural changes are induced to optimize substrate as well as inhibitor interactions with the enzyme.
In the absence of crystallographic data for the apoenzyme, it is speculative to correlate the slow step during inhibition with either a conformational change or displacement of a catalytic water. Our comparison of the kinetic data for the substrate and inhibitor indicate, however, that most likely both factors are contributing. Flap movement (conformational change) involved in the binding of both ligands could account for the much slower than diffusion-limited association kinetics, although additional rate limitations occur in the case of the inhibitor, as suggested by our data. This additional limitation could be attributed to the displacement of a catalytic water, since it does not occur during substrate binding; however, more detailed studies would be needed to address this hypothesis further.