Kinetic Studies on β-Site Amyloid Precursor Protein-cleaving Enzyme (BACE)

The steady-state kinetic mechanism of β-amyloid precursor protein-cleaving enzyme (BACE)-catalyzed proteolytic cleavage was evaluated using product and statine- (Stat(V)) or hydroxyethylene-containing (OM99-2) peptide inhibition data, solvent kinetic isotope effects, and proton NMR spectroscopy. The noncompetitive inhibition pattern observed for both cleavage products, together with the independence of Stat(V) inhibition on substrate concentration, suggests a uni-bi-iso kinetic mechanism. According to this mechanism, the enzyme undergoes multiple conformation changes during the catalytic cycle. If any of these steps are rate-limiting to turnover, an enzyme form preceding the rate-limiting conformational change should accumulate. An insignificant solvent kinetic isotope effect (SKIE) on k cat/K m , a large inverse solvent kinetic isotope effect onk cat, and the absence of any SKIE on the inhibition onset by Stat(V) during catalysis together indicate that the rate-limiting iso-step occurs after formation of a tetrahedral intermediate. A moderately short and strong hydrogen bond (at δ 13.0 ppm and φ of 0.6) has been observed by NMR spectroscopy in the enzyme-hydroxyethylene peptide (OM99-2) complex that presumably mimics the tetrahedral intermediate of catalysis. Collapse of this intermediate, involving multiple steps and interconversion of enzyme forms, has been suggested to impose a rate limitation, which is manifested in a significant SKIE on k cat. Multiple enzyme forms and their distribution during catalysis were evaluated by measuring the SKIE on the noncompetitive (mixed) inhibition constants for the C-terminal reaction product. Large, normal SKIE values were observed for these inhibition constants, suggesting that both kinetic and thermodynamic components contribute to theK ii and K is expressions, as has been suggested for other iso-mechanism featuring enzymes. We propose that a conformational change related to the reprotonation of aspartates during or after the bond-breaking event is the rate-limiting segment in the catalytic reaction of β-amyloid precursor protein-cleaving enzyme, and ligands binding to other than the ground-state forms of the enzyme might provide inhibitors of greater pharmacological relevance.

Extracellular amyloid deposits in brain, a characteristic feature of Alzheimer's disease, is a result of proteolytic cleavage of membrane-bound amyloid precursor protein by two enzymes, ␤-secretase and ␥-secretase. The second cleavage activity (␥secretase) is strongly associated with the presenilin multisubunit complexes (1), whereas ␤-secretase (BACE) 1 has been identified as a novel transmembrane aspartyl protease (2)(3)(4).
Although aspartyl proteases have been studied for more than 4 decades, new aspects of catalysis and inhibition continue to emerge. A substantial number of these enzymes have been identified as useful targets for chemotherapeutic intervention in human diseases (5)(6)(7)(8), yet there has been limited success in identifying clinically relevant inhibitors; hence, it is important to explore alternative drug design approaches. An understanding of the catalytic mechanism of the target enzyme is a powerful tool in the search for new inhibitors, and this has motivated us to study BACE more carefully in an attempt to find determinants that could lead the way to more successful drug design.
The proposed chemical mechanism for aspartyl 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 (9). Kinetic mechanism studies suggest that pepsin and HIV protease undergo isomerization steps during catalysis (10,11), and some evidence was obtained that this iso-step might be rate-limiting (12). Inhibitors binding to the form of the enzyme that accumulates because of the slow interconversion between enzyme isomers would be expected to display higher affinity and greater selectivity for the target enzyme. 2 A recent proposal for the mechanism suggests that aspartyl proteases undergo multiple conformation states, which differ in geometry, protonation state, and proton localization (13). The coplanar geometry of the aspartyl groups in the protease active site (14) and the pK values of the catalytic aspartates allow for the possibility that aspartyl proteases hydrolyze peptide bonds using cyclic proton transfers and reactant state hydrogen tunneling, featuring a low barrier hydrogen bond (LBHB) (13). If so, ligands, attracted and preserving this particular feature, would yield very tight and specific inhibitors (15). With this in mind, we performed product and statine-based peptide inhibition analysis, proton NMR of the free and inhibitor-bound enzyme, as well as solvent isotope effect studies to address the kinetic mechanism catalyzed by BACE and to dissect possible rate-limiting segments and dominating enzyme forms.

EXPERIMENTAL PROCEDURES
Cloning, Expression, and Purification of Human ␤-Secretase-The cDNA for the catalytic domain of human BACE (residues 1-460) was PCR-amplified and subcloned into the mammalian expression vector pTV1.6, upstream of a thrombin cleavage site linked to cDNA encoding human IgG 1 heavy chain. This construct, pTV1.6-BACE-T-IgG, was used to produce stably transfected dihydrofolate reductase-deficient Chinese hamster ovary DG44 cells, which were then scaled up using methotrexate for selection. Multiple rounds of selection were used, with a final expression level of ϳ10 mg/liter based on quantitation by enzyme-linked immunosorbent assay. BACE-T-IgG was scaled up to 35-80 liters of bioreactor production runs, and the secreted fusion protein was purified as described below.
The clarified growth media harvested from a bioreactor run was concentrated using tangential flow filtration fitted with a 30,000 molecular weight cut-off unit to a smaller volume, typically from 35-80 down to 4 liters to facilitate purification. The fusion protein was captured on rProtein A-Sepharose TM column (5 ϫ 20 cm, Amersham Biosciences) at 4°C. The column was washed with PBS, pH 7.1, until base-line absorbance was observed, and BACE-T-IgG was eluted with 0.10 M citrate, pH 3.0, into tubes containing 0.5 volumes of 4 M Tris, pH 8. Fractions containing the fusion protein were dialyzed extensively against PBS, pH 7.1, at 4°C. Analysis by SDS-PAGE and N-terminal sequencing (not shown) showed the presence of a mixture of two forms, consistent with published results from Amgen. The protein was sterilefiltered (0.2 m) and stored at 4°C for long term storage with no significant loss of activity observed even after ϳ1 year. BACE-T-IgG was treated with human ␣-thrombin (Enzyme Research Laboratories) at a ratio of 1:500 in PBS, pH 7.1, at 37°C for 2 h. Thrombin was removed by passing the sample over benzamidine-Sephadex column (Amersham Biosciences), and IgG was captured with rProtein A-Sepharose TM . Finally, the protein sample was further purified by applying onto Superdex 200 (26/60) gel filtration column washed with PBS, pH 7.1, at room temperature. Fractions containing BACE were combined, sterile-filtered (0.22 m), and stored at 4°C. The protein was characterized by SDS-PAGE, and a variety of biophysical techniques, including isothermal titration calorimetry, to demonstrate that it was glycosylated, monomeric, catalytically active, and fully competent to bind BACE active site inhibitor OM99-2 (16).
Substrates and Inhibitors-Two BACE substrates were employed in this study: Ac-EIDL2MVLDWHDK-DNP-OH (synthesized in-house) and MCA-EVNL2DAEF(K-DNP)-COOH (Biosource International), where MCA is 7-methoxycoumarin. The first substrate (low K m ) was used in experiments where substrate saturation was needed. The second substrate (high K m ) was used in fluorescence assays to perform experiments at substrate concentrations well below the K m value. Product inhibition studies were performed using Ac-EIDL-OH (corresponding to the N-terminal part of the cleaved substrate, referred to as product P) and Ac-MVLDWHDK-OH (corresponding to the C-terminal portion of the substrate, referred to as product Q). The peptidomimetic inhibitor OM-99 and statine-valine inhibitor (Stat(V)) ( Fig. 1) were purchased from Bachem and Enzyme Systems Inc., respectively. All substrate, product, and inhibitor stock solutions were prepared in Me 2 SO and stored at 4°C for about a month. The concentration of the low K m substrate was calculated using the extinction coefficient for DNP (⑀ 360 ϭ 17,700 M Ϫ1 ⅐cm Ϫ1 at neutral pH).
Enzyme Assays-Assays with the low K m substrate were performed in 200-l reaction mixtures containing the substrate (1-40 M) in 50 mM acetate buffer with 0.25 mg/ml BSA, pH 4.5, at 25°C. Me 2 SO content was adjusted in all assays to be the same and not to exceed 6% (no loss of enzyme activity was observed at this Me 2 SO concentration). Reactions were initiated by addition of 10 or 20 nM BACE and quenched at various time points (depending on the experiment) by either boiling for 2 min or by addition of 200 l of 3% (v/v) trifluoroacetic acid. The quenched reaction mixtures were subject to HPLC separation (Waters Associates) monitoring absorbance at 360 nm. Product and residual substrate peaks were well resolved with an increasing gradient of 0 -80% acetonitrile containing 0.1% trifluoroacetic acid, allowing precise integration of the peak areas. The area units were converted into molar concentration using a standard curve of known substrate concentrations. Because of nonspecific adsorption of product during the separation step, the calibration curves were non-linear. Therefore, sums of product and substrate peak areas after the reaction were plotted against the known substrate concentrations at the beginning of the reaction, and non-linear calibration curves were constructed. These calibration curves showed best fitting to the quadratic Equation 1, where y is sum of the product and substrate peak areas after the reaction; x is substrate concentration in M before the reaction; and b and c are coefficients derived from these data (a is assumed to be zero, at zero substrate concentration). Then the coefficients b and c were used to calculate product concentration from the product peak area. All measurements with the fluorescently labeled, high K m substrate were performed as described in detail elsewhere (17). Inhibition Studies-The IC 50 for steady-state inhibition of BACE by Stat(V) was determined at several fixed substrate concentrations in a reaction mixture (200 l) containing buffer, low K m substrate, and varying amounts of Stat(V) (0 -5 M). As in all assays, Me 2 SO content was constant in all mixtures and did not exceed 6% of the final volume. The reaction was initiated by addition of 10 nM BACE and stopped after 15 min by boiling the reaction mixtures for 2 min. All experiments were performed at 25°C. The HPLC method was applied to separate product and residual substrate peaks and to calculate their molar concentrations as described above. IC 50 values were determined by fitting the data to the Cheng-Prusoff equation (18) as shown in Equation 2, where Y is percent of inhibition at x concentration of the inhibitor, and s is the Hill slope. Product inhibition was studied in reaction mixtures (200 l) containing fixed concentrations of one of the products, varying amounts of substrate (1-30 M), Me 2 SO (to adjust its content to 6% final), and standard buffer (50 mM acetate with 0.25 mg/ml BSA, pH 4.5). After mixing the assay components, BACE (20 nM final) was added to initiate the reaction, and after 15-min 200 l of 3% trifluoroacetic acid was used to stop it. By this method the reaction velocities were determined as a function of substrate and inhibitor concentration. These data were analyzed by fitting them globally to the equations for competitive (Equation 3), non-competitive (Equation 4), or mixed type (Equation 5) inhibition, respectively, where [S] and K m are the concentration and Michaelis constant of the low K m substrate; V max and are the values for maximal and initial velocities; [I] corresponds to the inhibitor (one of the products) concentration; K is is the dissociation constant for the enzyme-inhibitor complex; and K ii is the dissociation constant for ternary enzyme-substrateinhibitor complex. The goodness of fit was inspected visually and then evaluated statistically by the F-test (GraFit, Erithacus) as shown in Equation 6, where n 2 is the smallest possible sum of squares deviations of the experimental values from the calculated ones (n denotes the number of paired fits being compared), and is the number of degrees of freedom (defined as ϭ N Ϫ n Ϫ 1, where N is the number of data points, and n is the number of variables in the equation).
Solvent Kinetic Isotope Effects-Solvent kinetic isotope effects were investigated by parallel measurements of initial velocities of substrate cleavage in 50 mM acetate with 0.25 mg/ml BSA prepared in H 2 O and D 2 O. To derive pH/pD profiles for BACE activity, 10-fold concentrated buffers (without BSA) were first prepared in H 2 O; the pH was adjusted to the desired values, and these solutions were diluted in either H 2 O (pH buffer) or in D 2 O (pD buffer). Lyophilized BSA was added after reading the pH of the diluted buffers with a pH meter and calculating pD as pD ϭ pH measured ϩ 0.33 for 88% D 2 O (19). The enzyme stock solutions were then prepared in the same buffers to match the desired pH/pD. Substrate and inhibitor stock solutions were prepared in deuterated Me 2 SO, which yielded 90% of the label in the final reaction mixtures. Reactions were initiated by BACE addition (20 nM) into mixtures of specific pH (or pD) buffer and low K m substrate (1-30 M in 200 l of final volume). After 5 min the reactions were quenched by adding 200 l of 3% trifluoroacetic acid, and the mixtures were analyzed by HPLC as described above.
Solvent kinetic isotope effects on the inhibition by cleavage products (P and Q) were performed using the same strategy. Buffer (prepared by 10-fold dilution into H 2 O or D 2 O) containing variable concentrations of products P or Q and substrate were mixed with the enzyme (prepared as well either in H 2 O or D 2 O) and quenched (by adding trifluoroacetic acid) after 15 min of reaction. Product formation and substrate contents were analyzed by HPLC as described earlier. Data for the k cat /K m dependence on the pH(D) were fitted to the Equation 7, where Y is the experimentally derived k cat /K m value; Y max refers to the observed maximum of that experimental value; and K a and K b are the lower and higher K a values for the two titratable groups. Kinetic solvent isotope effects on k cat ( D2O k cat ) and k cat /K m ( D2O k cat / K m ) were calculated using Cleland's program (20) as shown in Equation 8, where Y is the initial velocity; [S] is the substrate concentration; k cat and K m refer to the maximal turnover number and Michaelis constant, respectively; and F i is fraction of heavy atom label. E k cat /K m ϭ (( D2O k cat / K m ) Ϫ 1) and E k catx ϭ (( D2O k cat ) Ϫ 1). Solvent kinetic isotope effects on the inhibition by cleavage products (P and Q) were studied by using the same strategy. Buffer (prepared by 10-fold dilution into H 2 O or D 2 O) containing variable concentrations of products P or Q and substrate were mixed with the enzyme (prepared as well either in H 2 O or D 2 O) and quenched (by adding trifluoroacetic acid) after 15 min of reaction. Product formation and substrate contents were measured by the HPLC method described above, and the data were analyzed by fitting to Equations 3-5.
Solvent Kinetic Isotope Effects on the Onset of Inhibition by Stat(V)-Stat(V) exhibited a time-dependent onset of the inhibition during the catalysis, and we attempted to measure solvent kinetic isotope effects on the k on of this inhibition. The fluorescence intensity change upon cleavage of the high K m substrate was monitored in parallel experiments in 50 mM acetate with 0.25 mg/ml BSA at pH 4.5 and pD 5.05 (buffers were prepared from the same 10-fold concentrated buffer with pH 4.5 as described above) using the same concentrations of substrate (25 M), enzyme (20 nM), and different fixed concentrations of the inhibitor (0 -90 nM). Progress curves analysis was performed as described (17) to derive k obs values for the approach to the steady-state. Data were fit to Equation 9, where [P], [E], and [I] are the product, enzyme, and inhibitor concentrations, respectively; i is the initial velocity; s is the steady-state velocity, and k obs is the pseudo-first order rate constant for the approach to the steady state.
Proton NMR Experiments-NMR spectra were recorded on a Varian Inova 600 spectrometer equipped with a triple resonance 5-mm probe with triax gradients. Spectra were recorded using the 1-1 sequence to suppress the water signal with minimal saturation (21 (22). The fractionation factor () was determined by fitting the data to the Equation 10, where y is the signal intensity; C is a constant, and X is the mole fraction of H 2 O (23).

RESULTS
Inhibition Studies-Mechanistic studies of BACE using the fluorogenic substrate described previously (24) were severely limited because of the inner filter effects and the high K m of this substrate. Another peptidic substrate with a much lower K m has been reported recently (25), and we were able to advance mechanistic studies using this substrate and HPLCbased detection method. Under our assay conditions, the latter substrate was cleaved by BACE with k cat of 0.32 Ϯ 0.02 s Ϫ1 and K m of 5.2 Ϯ 0.7 M.
A number of peptides, based on the amino acid sequence around the amyloid precursor protein ␤-cleavage site and containing a hydroxyethylene (including OM99-2 (16)) or statine (including Stat(V) (26)) isostere (Fig. 1), were tested as inhibitors in the assay using this low K m substrate. Both of the above-mentioned inhibitors showed tight binding behavior with K i , calculated using Morrison's quadratic equation, of 2 and 20 nM for OM99-2 and Stat(V), respectively. The observed K i (IC 50 ) for competitive inhibitors, which bind exclusively to the free enzyme, should be linearly dependent on substrate concentration (27). However, upon varying Stat(V) (0 -5 M) and substrate concentrations (5-25 M), we found that the IC 50 was independent of substrate concentration (Fig. 2). Similar inhibition pattern was observed for a number of hydroxyethylene-containing peptides (data not shown). This type of pattern has traditionally been characteristic of non-competitive inhibitors that bind to both the free enzyme and the enzyme-substrate complex.
To follow up on this unexpected finding we conducted prod- uct inhibition studies. Initial velocity was measured varying substrate and one of the two cleavage products (Ac-EIDL or Ac-MVLDWHDR). Data collected using Ac-EIDL as an inhibitor were best fit globally (reconfirmed by F-test, Grafit, Erithacus) to a noncompetitive inhibition pattern (Fig. 3A) with K is ϭ K ii ϭ 3.2 Ϯ 0.4 mM. The second product, Ac-MVLDWHDR, likewise displayed a noncompetitive inhibition pattern (Fig.  3B) with the values of K is ϭ K ii ϭ 1.48 Ϯ 0.08 mM.
Solvent Kinetic Isotope Studies-Kinetic parameters for the BACE-catalyzed reaction were measured over the pH range of 3.5-6.0. Saturating kinetics in the substrate concentration range of 1-33 M were observed over the pH range of 3.5-5.0. Increasing pH resulted in an increased K m to the extent that no saturation could be observed above pH 5.0; therefore, k cat /K m was estimated from the slope of the velocity Ϫ substrate concentration plot. For this reason we were not able to evaluate the k cat dependence on pH. The pK values for k cat /K m calculated from fitting the data to Equation 7 were 3.54 Ϯ 0.13 and 5.21 Ϯ 0.11, confirming the expectation of general acid and general base catalysis. Repeating the entire pH profile in D 2 O-based buffer (78% 2 H) gave an increase in the pK value of the general base (4.43 Ϯ 0.05), whereas no shift of the general acid value (5.21 Ϯ 0.06) was observed (Fig. 4A). An equilibrium solvent deuterium isotope effect usually is accountable for an increase in pK values by ϳ0.4 Ϫ0.6 units (28). The larger changes in pK seen here suggest that additional factors contribute to the almost unit increase of the lower pK, whereas the higher pK is invariant. From the data in Fig. 4A we chose values of pH ϭ 4.5 and pD ϭ 5.03 (as isotopically equivalent in a pH-independent region) to analyze further the solvent kinetic isotope effect on k cat . An increased reaction velocity was observed in the D 2 O solution (Fig. 4B), and these data revealed an inverse solvent kinetic isotope effect on k cat (0.67 Ϯ 0.11), with very modest, if any, effect on k cat /K m (0.84 Ϯ 0.31).
Solvent Isotope Effect on Product Inhibition-Even though solvent kinetic isotope effects on substrate or inhibitor binding are very rare and usually small (28), we decided to explore the effect of D 2 O on product inhibition parameters. The reactions were run in parallel in H 2 O and 82% D 2 O, pH 4.5 and pD 5.17, varying substrate and product Ac-MVLDWHDR concentrations. Data for the reaction in H 2 O solution fit well to noncompetitive inhibition, yielding K is ϭ K ii ϭ 1.87 Ϯ 0.11 mM. Inhibition by Ac-MVLDWHDR in deuterated solution appeared to be stronger, and these data could be fit successfully to either noncompetitive inhibition (K is ϭ K ii ϭ 0.60 Ϯ 0.03 mM) or mixed type inhibition equations (K is ϭ 0.35 Ϯ 0.09 mM, K ii ϭ 0.80 Ϯ 0.02 mM). The calculated D2O K is (as ratio of K is in water and D 2 O) was 5.1 Ϯ 1.1 and D2O K ii ϭ 2.3 Ϯ 0.1. (17) and could be monitored by the following reaction progress curves of fluorogenic substrate cleavage at S Ͻ K m . Inhibition onset kinetics were compared in H 2 O and D 2 O solutions at different concentrations of inhibitor, and the data were fit to Equation 9 to calculate k obs . The dependence of the first order rate constant (k obs ) on inhibitor concentration was linear and very similar in H 2 O and D 2 O solutions (Fig. 5); the calculated values of k on ((1.9 Ϯ 0.5) ϫ 10 5 M Ϫ1 s Ϫ1 in H 2 O and (1.7 Ϯ 0.5) ϫ 10 5 M Ϫ1 s Ϫ1 in D 2 O), k off ((5 Ϯ 1) ϫ 10 Ϫ4 s Ϫ1 in both solvents) and subsequently K i * (26.4 versus 28.1 nM) were solvent-independent. Therefore, no obvious solvent isotope effects were observed on BACE inhibition by Stat(V) during the catalytic turnover.

Solvent Isotope Effect on Inhibition by Stat(V)-Slow onset inhibition of BACE by Stat(V) has been observed previously
Protein Proton NMR Studies-Low barrier hydrogen bonds display distinctive physicochemical properties that are readily amenable to detection by NMR. Two such properties are unusually downfield proton chemical shifts (typically ␦ Ͼ15.0 ppm) and low deuterium fractionation factors ( Ͻ 0.8). To determine whether the formation of an LBHB may be participating in the mechanism of action of BACE, proton NMR studies were initiated. Fig. 6 shows the downfield region of the proton NMR spectrum of BACE. In the free protein, the furthest downfield signal resonates at ␦ ϭ 11.8 ppm. No other downfield resonance is observed. When complexed with OM99-2, a potent inhibitor of BACE, an additional resonance appears at ␦ ϭ 13.0 ppm. The presence and resonance fre- , and 2 (f) mM product P (N-terminal product). The lines are global fits of the data to the non-competitive inhibition pattern revealed to be the best fit after comparing with other inhibition modes by using Equations 3-5. B, inhibition of BACE (20 nM) by 0 (E), 0.5 (q), 1 (Ⅺ), and 2 (f) mM of product P2 (C-terminal portion). The graph represents the best global fit of the data to the non-competitive inhibition mode after analyzing the data by using Equations 3-5. quency of this peak was unchanged across a range of pH values (7.0, 5.3, and 4.5) and temperatures (5, 10, 15, and 20°C).
To measure the strength of the newly formed hydrogen bond, the D/H fractionation factors at pH 4.5 were determined from the slope of the lines obtained in plots of the integrated signal intensity of the downfield resonances against the D 2 O/H 2 O composition (Fig. 7). In the BACE-OM99-2 complex, the value for the proton resonating at ␦ ϭ 13.0 ppm is 0.6, indicating it is participating in a relatively strong hydrogen bond. For comparison, the value for the proton resonating at ␦ ϭ 11.8 ppm is 2.2, which is comparable with the typically observed for a backbone amide proton in free exchange with water. The ␦ and values observed for the downfield proton in the BACE-OM99-2 complex suggest the formation of a strong hydrogen bond coinciding with the formation of the state close to transition and that the hydrogen bond may not be as strong as the LBHBs observed previously in other enzyme-inhibitor complexes. Possible reasons for this may include sub-optimal hydrogen bonding geometry and imperfectly matched pK values of the hydrogen bond donor and acceptor in the BACE-OM99-2 complex.

DISCUSSION
For more than a decade, significant attention has been paid to the concept of structure-based ligand design, in which the crystal-or NMR-based structure of a target protein is utilized for the de novo design and/or optimization of the ligands intended to fit well into a specific pocket (typically the catalytic active site of an enzyme). Nevertheless, the structural information obtained by x-ray crystallography and NMR spectroscopy represent structural "snapshots" of the enzyme active site, and do not always account for the conformational dynamics of enzymes and other target proteins. Both the solvated ligand (e.g. substrate molecules) and the solvated enzyme can exist in multiple conformational states prior to complex formation. Interconversions among such conformational states are often associated with specific steps in catalysis, and in some cases these requisite conformational changes can impose rate limitations on turnover. Ligands that specifically stabilize the conformational state preceding the rate-determining step in catal- Initial velocity values at pH 4.5 (E) and pD 5.03 (q) were transformed into Lineweaver-Burk coordinates, and solvent kinetic isotope effects on k cat ( D2O k cat ) and k cat /K m ( D2O k cat /K m ) were calculated by fitting the data to Equation 8. ysis can offer great advantages with respect to increased binding affinity and target selectivity. Thus, the combination of kinetic, chemical, and structural information provides the most complete opportunity for design and optimization of ligands for specific targets.
Aspartyl proteases have been shown to contain unusually extended and mobile active sites that bind inhibitors in unpredictable ways due to multiple conformations of the active site (29, 30). The best studied conformers of the aspartyl proteases are the free enzyme and the enzyme-statine or enzyme-hydroxyethylene complexes. These two forms can differ in the co-planarity (from crystal structure (31,32)) and/or ionization state of the catalytic aspartates. Although the crystal structure resolution of the free enzyme and aspartyl protease-ligand complexes are insufficient to localize protons, the generally accepted chemical mechanism predicts that if the enzyme starts its catalytic cycle with Asp-32 as a general acid and Asp-228 as a general base (BACE amino acid numbering, species EH in Scheme 1), the amide hydrate tetrahedral intermediate (species FHA in Scheme 1) would be in the opposite protonation state. Asp-228 is protonated after abstracting a proton from the catalytic water molecule, and Asp-32 is ionized as a result of substrate carbonyl protonation. It should be noted that direct evidence for such a monoprotonated form of a tetrahedral intermediate-mimicking enzyme-hydroxyethylene complex was obtained for endothiapepsin using neutron diffraction methods (33). After protonation of the leaving amine and collapse of the tetrahedral intermediate (step 4 in Scheme 1), the enzyme returns to its original protonation state (species GH in the Scheme 1) but still lacks the catalytic water molecule. It has been proposed that reprotonation or rehydration of the enzyme active site (step 7 in Scheme 1) may be a rate-limiting step during pepsin (12) and HIV protease catalysis (10).
Our product inhibition data (both cleavage products showed noncompetitive inhibition patterns) are consistent with a unibi-iso-mechanism (34,35). This mechanism predicts that one substrate yields two products and that there is a kinetically significant conformational change required during catalysis. Isoenzyme mechanisms related to the protonation/deprotonation have been well described for proline racemase (36), fumarase, and carbonic anhydrase (37) and a number of aspartyl proteases (13). It is very possible that similar mechanism functions during ␥-secretase catalyzed cleavage, because a noncompetitive inhibition by intermediate analogs was observed recently (38).
If the change of conformation involves rate-limiting proton transfer(s), this should be manifested in a solvent kinetic iso-tope effect. A very small and statistically insignificant inverse SKIE was observed on the k cat /K m of BACE ( Fig. 4A and Table  I), suggesting either that there were no rate-limiting proton transfer steps up to the first irreversible step (step 4 in Scheme 1) or that there were multiple proton transfers that offset each other. Loop closing subsequent to substrate or inhibitor binding could be accompanied by multiple water displacements, yielding an inverse solvent isotope effect, although both normal and inverse effects have been reported to accompany conformational changes in other systems (39,40). Alternatively, this effect could be offset by a normal isotope effect originating from amide hydrate intermediate formation (two protons). These possibilities could be further distinguished by studying proton inventory effects on k cat /K m . Unfortunately, we were not able to perform such experiments, because by changing the H 2 O/D 2 O ratio, the pH/pD is altered in such a way that parameters change not because of the label ratio but rather because of the effective pH change. The absence of a solvent kinetic isotope effect on k on for BACE inhibition by Stat(V) (Fig. 5 and Table I) suggests that this inhibitor binds equally well to all isoforms of the enzyme. Thus rate-limiting and proton transfer-associated interconversions of these enzyme forms are not reflected in onset of the inhibition.
Turnover number (k cat ), on the other hand, was significantly affected by the solvent nature, showing an inverse kinetic isotope effect of 0.67 ( Fig. 4B and Table I). This is an unprecedented finding for aspartyl proteases, because modest and normal kinetic isotope effects on k cat have been reported for HIV protease (10), renin (41), and pepsin (12). The slow step in these cases was associated with the "recharging" of the enzyme after bond breaking, most likely reflecting enzyme reprotonation to the initial state and/or regaining of the catalytic water. The inverse solvent isotope effect on k cat seen for BACE could arise either because of a low fractionation factor of proton(s) in the reactant state or a high fractionation factor in the transition or product state (28). The reactant state reflected in k cat /K m is the free enzyme (EH in Scheme 1), whereas the reactant state reflected in k cat is the tetrahedral intermediate (FHA in Scheme 1) (19). The transition state for k cat /K m would be FHA and that for k cat could be either GH or EH in Scheme 1. Because the SKIE on k cat /K m is ambiguous, we will not discuss it further. We suggest that an inverse SKIE on k cat is a result of a short, strong hydrogen bond in the FHA complex (reactant state for k cat ) that fractionates with a low factor. LBHB have been documented for the transition states of serine proteases (42), triose-P-isomerase, citrate synthase, and other enzymes (43), and for these enzymes the existence of a LBHB was reconfirmed by NMR measurements. Aspartyl proteases have also been suggested to feature LBHB in the free enzyme state (13), but no direct evidence of this feature has yet been demonstrated. Nevertheless a short and strong hydrogen bond was recently reported for endothiapepsin-statine, gem-diol, and other inhibitor complexes (44). We were not able to obtain any data confirming or dismissing this hypothesis for free BACE (i.e. no downfield NMR signal in free enzyme), but we propose the possibility of formation of a short, strong hydrogen bond in one of the intermediates. This hypothesis is supported by proton NMR studies, assuming that the enzyme-OM99-2 complex mimics the tetrahedral (FHA) or other intermediate complex. A downfield NMR signal at ␦ 13.0 ppm was observed (Fig. 6) when BACE was mixed with the inhibitor, and this proton showed a fractionation factor () of 0.6 (Fig. 7). It should be noted that interpreting the ␦ and values in terms of absolute hydrogen bond strength is an inexact task. A theoretical study of the relationship between the proton chemical shift and hydrogen bond strength found an excellent linear correlation FIG. 7. Determination of the D/H fractionation factor for the downfield signals in BACE. The relative intensities of the signals at 13.0 and 11.8 ppm (Fig. 6) are plotted as a function of the mole fraction H 2 O (X) according to Loh and Markley (23). The slope of the line is the D/H fractionation factor ().
between the hydrogen bond strength and predicted chemical shift in model LBHB complexes (45). This correlation held only for compounds within the same class, however, and the authors cautioned about comparing the LBHB proton chemical shifts of structurally unrelated compounds. The correlation between and hydrogen bond strength may be more appropriate than ␦. Factor is reduced by as much as a 4-fold for LBHBs relative to exchangeable protons in proteins (46). Whether sufficient data exist to allow direct comparisons of and hydrogen bond strengths for LBHBs in chemically different environments remains unclear. For example, in chymotrypsinogen A, two protons shown to be involved in LBHBs were found to have significantly different values for ␦ and : for Asp-His␦ 1 , ␦ ϭ 18.1 and ϭ 0.4, whereas for His ⑀2 -Ser, ␦ ϭ 13.2 and ϭ 0.69 (22). In the present study, the values of ␦ and measured for the downfield proton in the BACE-OM99-2 complex (␦ ϭ 13.0, ϭ 0.6) are comparable with those measured for the LBHB between His ⑀2 -Ser in chymotrypsinogen A.
Formation of this tetrahedral intermediate (featuring a strong short hydrogen bond) during catalysis is unlikely to be rate-limiting (no SKIE on k cat /K m ). On the other hand, either the collapse (protonation of the leaving group) or enzyme recharging (correct enzyme reprotonation or rehydration) most likely does impose a rate limitation, and this is what is manifested in the SKIE on k cat . We suggest that there are at least three enzyme forms that can bind peptidomimetic inhibitors as follows: EH (correct protonation, hydrated), FH (alternative protonation, dehydrated), and GH (correct protonation, dehydrated). Our product inhibition data in D 2 O (Table I) suggest that there is a kinetic component to the noncompetitive inhi-SCHEME 1. Proposed catalytic mechanism of BACE. EH is a free monoprotonated form of the enzyme, which after substrate binding (step 1) undergoes loop closure (step 2) to form a tightened enzyme-substrate complex FHS. Activation of the catalytic water by the active site base and substrate carbonyl protonation by the general acid leads to formation of a tetrahedral intermediate FHA (step 3). Collapse of FHA is followed by (or concomitant with) peptide bond breaking, protonation of the leaving amine, and restoration of the enzyme to the initial state, albeit without the active site water (step 4, GHPQ). Free enzyme (GH) is restored to a catalytically relevant form by water incorporation (step 7). We chose to picture step 4 as a single bond breaking event but note that we cannot exclude the possibility that some sequential steps are occurring and other intermediate transitions (F Ϫ or FH 2 ) may exist.

TABLE I Solvent kinetic isotope effects (SKIE) on the reaction catalyzed by BACE and inhibition
bition constant expression (47), which is potentially indicative of a rate-limiting interconversion between enzyme forms during catalysis. The species featuring a strong hydrogen bond (most likely complex with FH) is favored in D 2 O where its accumulation results in more potent inhibition and a large SKIE on K ii (13). A rate-limiting proton transfer accompanying FH collapse could be slowed in D 2 O resulting in accumulation of FH and increased binding as well. The inhibition constant K is measures inhibitor dissociation from all free enzyme forms (mostly EH and GH) that are in equilibrium; any solvent isotope effect on K is must have a thermodynamic origin. GH will accumulate in D 2 O, resulting in tighter inhibitor binding and a large SKIE on inhibition (Table I).
Taken together, our data suggest that BACE catalyzes proteolysis involving a rate-limiting enzyme isomerization step(s), which occurs after tetrahedral intermediate formation. Solvent kinetic isotope effects and proton NMR results point toward collapse of the amide hydrate and restoration of the initial enzyme protonation state as being the slow step. We do not have any strong evidence that reprotonation of the enzyme occurs after product release. A prevailing hypothesis relates restoration of the general base by protonation of the leaving product amine and restoration of the general acid by proton abstraction to yield the carboxyl product carbonyl. It is quite possible that these two steps occur sequentially along the reaction pathway resulting in multiple transitions. Our results reinforce (i.e. the presence of a short, strong hydrogen bond) the inference that hydroxyethylene inhibitors bind more potently to the FH form of the enzyme (oppositely monoprotonated and dehydrated). Nevertheless, it is difficult to define the extent to which hydroxyethylene complexes resemble the true tetrahedral intermediate of catalysis, because only one hydroxyl is present in the enzyme-inhibitor complexes. This could be a reason for the observed distortion of the aspartate coplanarity (32) in these complexes, and thus the potential of the inhibitor is not maximized. Hence, compounds that bind the monoprotonated form of the enzyme (the same FH), but engage the aspartates in a more coplanar interaction, could be even more attractive as inhibitors and possibly feature an even stronger and shorter hydrogen bond (48). On the other hand, depending upon the sequence of reprotonation of the catalytic aspartates, the transitions can emerge as dianionic (F Ϫ if the amine protonation is first and not rate-limiting) or diprotonated (FH 2 , if Asp-32-H regeneration is first and not rate-limiting). Either of these forms would be stabilized by distinct classes of inhibitors. Given the fact that the BACE active site is a large and flexible cavity, designing an optimal inhibitor is a subtle interplay between identifying the right isostere and optimizing adjacent parts of the molecule. The ability to take advantage of the short and strong hydrogen bonds that engage the catalytic aspartates depends on finding optimal interactions between the enzyme and the rest of the inhibitor molecule, allowing proper positioning of the isostere.