Determinants of Affinity and Proteolytic Stability in Interactions of Kunitz Family Protease Inhibitors with Mesotrypsin*

An important functional property of protein protease inhibitors is their stability to proteolysis. Mesotrypsin is a human trypsin that has been implicated in the proteolytic inactivation of several protein protease inhibitors. We have found that bovine pancreatic trypsin inhibitor (BPTI), a Kunitz protease inhibitor, inhibits mesotrypsin very weakly and is slowly proteolyzed, whereas, despite close sequence and structural homology, the Kunitz protease inhibitor domain of the amyloid precursor protein (APPI) binds to mesotrypsin 100 times more tightly and is cleaved 300 times more rapidly. To define features responsible for these differences, we have assessed the binding and cleavage by mesotrypsin of APPI and BPTI reciprocally mutated at two nonidentical residues that make direct contact with the enzyme. We find that Arg at P1 (versus Lys) favors both tighter binding and more rapid cleavage, whereas Met (versus Arg) at P′2 favors tighter binding but has minimal effect on cleavage. Surprisingly, we find that the APPI scaffold greatly enhances proteolytic cleavage rates, independently of the binding loop. We draw thermodynamic additivity cycles analyzing the interdependence of P1 and P′2 substitutions and scaffold differences, finding multiple instances in which the contributions of these features are nonadditive. We also report the crystal structure of the mesotrypsin·APPI complex, in which we find that the binding loop of APPI displays evidence of increased mobility compared with BPTI. Our data suggest that the enhanced vulnerability of APPI to mesotrypsin cleavage may derive from sequence differences in the scaffold that propagate increased flexibility and mobility to the binding loop.

Because of the potential for proteolytic damage in living systems, elaborate mechanisms have evolved to regulate proteases. One mechanism for protease regulation involves complex formation with endogenous protein inhibitors. Maintenance of the protease/inhibitor balance is crucial, and the upset of this balance is in evidence in numerous pathological conditions, including cancer, arthritis, pancreatitis, and many vascular, pulmonary, and skin diseases (1)(2)(3)(4)(5). One important factor defining the balance between proteases and their inhibitors is the stability or vulnerability to proteolysis of protein protease inhibitors. Many inhibitors are exceptionally stable to proteolysis (6); however, some inhibitors can be gradually inactivated either by the very proteases that they inhibit (7) or by alternative endogenous proteases that they do not inhibit (8 -12). Infectious microorganisms can also produce proteases that perturb the protease-inhibitor balance in human tissue by selective cleavage and inactivation of human protease inhibitors (13)(14)(15)). An understanding of the fundamental features that render protein protease inhibitors stable inhibitors versus vulnerable substrates of the proteases that they encounter is critical for understanding the function of these molecules in tissue homeostasis.
The structurally diverse "canonical" inhibitors of serine proteases are named for a convex protease-binding loop of highly characteristic backbone conformation that is slightly extended from an inhibitor scaffold and serves as a simple recognition motif (16 -18). These inhibitors function via the "standard mechanism" or "Laskowski mechanism" (18 -20), in which the canonical loop binds very tightly in the active site of a cognate protease in a substrate-like manner (21), positioning a reactive site peptide bond for cleavage, but this bond is then cleaved only very slowly. One of the largest and best studied families of canonical inhibitors is the Kunitz-BPTI family, the archetypal member of which is bovine pancreatic trypsin inhibitor (BPTI), 2 also known as aprotinin (22). The Kunitz protease inhibitor fold is a small, very compact pear-shaped motif of ϳ60 amino acids, stabilized by a hydrophobic core and by three disulfide bonds; the canonical loop forms the narrow top of the pear. Kunitz inhibitors include both single domain proteins like BPTI and larger proteins in which the Kunitz domain comprises a functional module within a compound inhibitor or multidomain protein (23). For example, the amyloid precursor protein (APP) possesses a functional Kunitz protease inhibitor domain; the secreted ectodomain of APP, also known as protease nexin 2, inhibits coagulation factor XIa via the Kunitz domain, regulating hemostasis and thrombosis (24 -27). There are 36 human Kunitz domains and homologues, a number of which serve known key functions regulating serine proteases involved in coagulation, fibrinolysis, and tissue remodeling (MEROPS data base, available on the World Wide Web) (28).
Mesotrypsin, a human trypsin encoded by the PRSS3 gene, is produced and secreted as a digestive zymogen by the pancreas (29); a major splice isoform of mesotrypsinogen termed "trypsinogen 4" and lacking a classical secretion signal is highly expressed in brain tissue (30,31) and in some epithelial cell lines and tumors (32)(33)(34)(35)(36)(37). The splice isoforms differ only at the N terminus, and processing of either form by removal of the prodomain results in active mesotrypsin of identical amino acid sequence (38). Although mesotrypsin shows high sequence homology with other trypsins, its functional properties are very different. Mesotrypsin is highly resistant to inhibition by many polypeptide trypsin inhibitors (39 -41), although it is readily inhibited by serpin-type inhibitors with Arg in the reactive loop (42,43). Although it cleaves peptide anilide model substrates with kinetic constants similar to other trypsins (44 -46), it displays reduced or no activity toward several specific protein substrates of other trypsins (44,(47)(48)(49) and, unlike other trypsins, is incapable of autoactivation (44). By contrast, mesotrypsin is far more active than other trypsins in cleavage of canonical protease inhibitors at the reactive site peptide bond (44,46,50).
Although it seems very likely that a major physiological role of mesotrypsin lies in proteolytic cleavage and inactivation of canonical protein protease inhibitors (44,50), not all canonical inhibitors are equivalent substrates for mesotrypsin. We have found that BPTI behaves as a temporary inhibitor of mesotrypsin, being gradually cleaved with an enzymatic turnover time of 2.2 h (46), whereas the isolated Kunitz domain of APP (APPI) is cleaved by mesotrypsin with a kinetic profile more closely resembling that of a true substrate, with an enzymatic turnover time of 24 s (50). The two inhibitors display striking differences in mesotrypsin affinity as well, with APPI binding ϳ100-fold more tightly to the protease than does BPTI (45,46,50). These functionally contrasting but structurally similar inhibitors offer an ideal model system in which to explore the basis for resistance of Kunitz protease inhibitors to proteolysis by mesotrypsin and potentially offer insights into the fundamental basis of canonical inhibitor resistance to proteolysis more generally. Here, we have taken a mutagenesis approach to dissect the contributions of canonical loop residues and inhibitor scaffolds to the differential proteolytic stability and mesotrypsin affinity of APPI and BPTI.

EXPERIMENTAL PROCEDURES
Production of Recombinant Proteins-Recombinant human mesotrypsinogen and a catalytically inactive S195A mutant were expressed in Escherichia coli, isolated from inclusion bodies, refolded, purified, and activated with bovine enteropepti-dase as described previously (46). Kunitz domain inhibitors were expressed in the methylotrophic yeast Pichia pastoris under control of the alcohol oxidase (AOX1) promoter using the expression vector pPICZ␣A (Invitrogen); construction of the APPI-WT, BPTI-WT, and P 1 mutant expression constructs has been described previously (51,52). 3 Additional mutations were introduced using the QuikChange kit (Stratagene) according to manufacturer protocols; mutant plasmids were verified by sequencing. Expression constructs were linearized with SacI (New England BioLabs) and transformed via electroporation into P. pastoris X33 (Invitrogen). Chromosomal integration was verified by PCR amplification of genomic DNA using AOX1 forward and reverse primers. A number of verified transformants were screened for expression in small scale liquid cultures; those with media revealing the highest levels of trypsin inhibitory activity were selected for large (1 liter) scale expression and purification. Expression cultures grown in BMMY medium (buffered medium containing methanol and yeast nitrogen base) at 30°C were harvested after 48 h. The supernatant was subjected to ammonium sulfate precipitation (95% saturation) at room temperature. Protein pellets were resuspended in 20 mM Tris buffer, pH 7.8, and dialyzed overnight against 10 mM Tris buffer, pH 7.8, using 3.5-kDa cut-off dialysis tubing (ThermoScientific). Dialyzed APPI samples were chromatographed on Q-Sepharose FF (GE Healthcare) and eluted with a gradient of 0 -100% buffer B (50 mM Tris, pH 7.8, 1 M NaCl). Dialyzed BPTI samples were chromatographed on SP-Sepharose FF (GE Healthcare) and eluted with a gradient of 0 -100% buffer B (50 mM Tris, pH 7.8, 1 M NaCl). Both APPI and BPTI proteins were further purified on a trypsin affinity column, as described previously (50), and eluted with a gradient of 0 -100% buffer B (150 mM HCl). Protein purity was verified by HPLC using a reversed phase Jupiter 4 90-Å C 12 column (Phenomenex), to ensure that each inhibitor chromatographed as a single peak. BPTI-R17M and BPTI-K15R/R17M were further purified by HPLC using a reversed phase C 18 column (Luna 250 ϫ 10-mm 5) (Phenomenex) and a gradient of 0 -100% acetonitrile in 0.1% TFA.
Competitive Inhibition Studies-Mesotrypsin concentration was quantified by active site titration using 4-nitrophenyl 4-guanidinobenzoate (Sigma) (53). Concentrations of APPI and BPTI variants were determined by titration with bovine trypsin (Sigma) as described previously (46). Concentrations of the chromogenic substrate benzyloxycarbonyl-Gly-Pro-Arg-pnitroanalide (Sigma) were determined by an end point assay. Working stocks of enzyme, substrate, and inhibitors were prepared, and assays were conducted as described previously (46,50). Briefly, reactions carried out at 37°C in a Varian Cary-100 spectrophotometer were followed spectroscopically for 3-5 min, and initial rates were determined from the absorbance increase caused by the release of p-nitroaniline (⑀ 410 ϭ 8480 to Equation 1, the classic competitive inhibition equation, using Prism (GraphPad Software, San Diego CA). Reported inhibition constants are average values obtained from 2-4 independent experiments.
Inhibitor Hydrolysis Studies-The depletion of intact APPI and BPTI variants in time course incubations with active mesotrypsin was monitored by HPLC and 16% SDS-Tricine PAGE (55,56). Mesotrypsin concentrations were optimized in order to achieve a comparable degree of digestion within a similar time frame for different inhibitor variants. Incubations of mesotrypsin with BPTI mutants were carried out in 0.1 M Tris-HCl, pH 8.0, and 1 mM CaCl 2 at 37°C; BPTI concentration was 50 M, and mesotrypsin concentration was 5 M (for BPTI-WT and BPTI-R17M), 2.5 M (for BPTI-K15R), or 1.25 M (for BPTI-K15R/R17M). Aliquots for HPLC analysis were withdrawn at periodic intervals, adjusted to 6 M urea and 2 mM DTT, incubated for 10 min at 37°C, quenched by acidification to pH 1, and then frozen at Ϫ20°C until analyzed as previously reported (46). Incubations of mesotrypsin with APPI mutants were carried out similarly, except that APPI concentration was 50 M, and mesotrypsin concentration was 500 nM (for APPI-R15K and APPI-R15K/M17R) or 50 nM (for APPI-WT and APPI-M17R). APPI hydrolysis time point samples were not denatured or reduced; instead, samples were quenched immediately by acidification to pH 1 and then frozen at Ϫ20°C until analyzed. Enzyme, Kunitz inhibitors, and hydrolysis products were resolved on a 50 ϫ 2.0-mm Jupiter 4 90-Å C 12 column (Phenomenex) with a gradient of 0 -100% acetonitrile in 0.1% TFA at a flow rate of 0.6 ml/min over 50 min. Peak integration to quantify the disappearance of intact Kunitz inhibitors over time was carried out as described previously (46); initial rates were obtained by linear regression using a minimum of six data points within the initial linear phase of the reaction and not exceeding 50% conversion of intact inhibitor to hydrolysis products. Hydrolysis rates reported for each inhibitor represent the average of 2-4 independent experiments.
Crystallization of Mesotrypsin⅐APPI Complexes-Complexes of catalytically inactive mesotrypsin-S195A with APPI and APPI-R15K were crystallized by vapor diffusion. Mesotrypsin-S195A dissolved in 1 mM HCl and APPI dissolved in 10 mM NaOAc, pH 6.5, were mixed in a 1:1 stoichiometric molar ratio to achieve a total protein concentration of 3-6 mg/ml. Crystals were grown at 22°C in hanging drops over a reservoir of 4.5 M sodium formate. Drops (4 l) were prepared by mixing equal volumes of protein and reservoir solutions. Crystals (0.1 ϫ 0.2 ϫ 0.1 mm) appeared within 4 days and grew over the course of 3 weeks. Crystals were harvested, soaked in a cryoprotectant solution (4.5 M sodium formate and 17.5% glycerol), and flashfrozen in liquid N 2 .
X-ray Data Collection, Structure Solution, and Model Refinement-Synchrotron x-ray data were collected from crystals at 100 K using ADSC CCD detectors at beamlines X12-B, X12-C, and X25 at the National Synchrotron Light Source, Brookhaven National Laboratory. Mesotrypsin⅐APPI and mesotrypsin⅐APPI-R15K complex crystals belong to space group P22 1 2 1 with unit cell parameters of a ϭ 92.8, b ϭ 130.1, c ϭ 132.3 and a ϭ 92.9, b ϭ 131.4, c ϭ 131.9, respectively, and ␣ ϭ ␤ ϭ ␥ ϭ 90°. Diffraction data were measured to 2.48 and 2.38 Å resolution for mesotrypsin⅐APPI and mesotrypsin⅐APPI-R15K complexes, respectively. The automation package ELVES (57) was used to direct MOLFLM (58) for indexing and integration and SCALA for scaling and merging the reflections (59). The mesotrypsin⅐APPI structure was solved by molecular replacement using the program Phaser (60) operated by PHENIX (61) using mesotrypsin and APPI structures as pieces of a complete search model, allowing independent translation and rotation of each piece. The search model was derived from previous structures of mesotrypsin (PDB entry 2R9P, chain A) (46) and APPI (PDB entry 1ZJD, chain B) (51). The solved structure of the mesotrypsin⅐APPI complex was used as a model for the mesotrypsin⅐APPI-R15K complex structure. The successful solution contained four copies of each protein in the asymmetric unit, forming four canonical trypsin⅐APPI complexes. Cycles of manual rebuilding in COOT (62) were alternated with automated refinement using the refinement module of the PHENIX software suite (63). A test set composed of 10% of the total reflections was excluded from refinement to allow calculation of the free R factor. Non-crystallographic symmetry restraints were not used in refinement. TLS refinement was employed, with each protein chain assigned to a separate TLS group. Waters, ions, and alternative conformations of protein residues were added using COOT (62). The space group assignment was reconfirmed using the Zanuda Web server, hosted by the York Structural Biology Laboratory. All superpositions and structure figures were created using the graphics software PyMOL (64).

BPTI and APPI Show Striking Differences in Binding and
Susceptibility to Cleavage by Mesotrypsin-APPI binds to mesotrypsin ϳ100 times more tightly than BPTI (45,46,50) and is cleaved by mesotrypsin Ͼ300 times more rapidly than BPTI (46,50). To identify sequence or structural differences between these inhibitors potentially responsible for the differences in their interactions with mesotrypsin, we performed sequence and structural alignments (Fig. 1). BPTI and APPI share 44% sequence identity (Fig. 1A) and a highly conserved three-dimensional structure (Fig. 1B). Throughout this paper, we have adopted the conventional BPTI numbering system for designating analogous residues of APPI. Focusing on the canonical loop, which makes the majority of the inhibitor-enzyme close contacts, we identified two nonidentical residues: the P 1 position (Lys-15 in BPTI, Arg-15 in APPI) and the PЈ 2 position (Arg-17 in BPTI, Met-17 in APPI). To dissect the roles of the P 1 residue, the PЈ 2 residue, and the BPTI versus APPI scaffold in accounting for the striking differences in mesotrypsin binding affinities and cleavage rates, we generated a series of single and double mutants in which we interchanged P 1 and PЈ 2 residues, creating variants in which inhibitors differ only at the P 1 or the PЈ 2 position or have identical canonical loops in the context of different scaffolds.
As a measure of the binding affinity of mesotrypsin for BPTI and APPI variants, we obtained equilibrium inhibition constants (K i ) using classic competitive inhibition experiments in which we monitored cleavage of the chromogenic substrate benzyloxycarbonyl-Gly-Pro-Arg-p-nitroanalide by mesotrypsin in the presence of varying concentrations of inhibitor. All of the Kunitz variants tested exhibited kinetics of inhibition that were well described by a strictly competitive model, as we have reported previously for the inhibition of mesotrypsin by BPTI-WT (46) and APPI-WT (50). Values of K i for all of the Kunitz variants are summarized in Table 1. The competitive inhibition equation used to determine K i also describes the enzymatic reaction of one substrate in the presence of an alternative, competing substrate, with the added condition that the observed K i for the alternative substrate must be equivalent to its Michaelis constant K m (65). Thus, because we have previously found mesotrypsin to cleave both BPTI and APPI at the reactive site peptide bond (46,50), the measured K i values also represent K m values for cleavage of the Kunitz inhibitors as substrates of mesotrypsin.
We next used SDS-PAGE and HPLC assays to monitor inhibitor cleavage in time course incubations with mesotrypsin and to calculate rates of catalysis (k cat ). Qualitatively, the results of the two detection methods were highly consistent. HPLC peaks corresponding to BPTI and APPI cleavage products were clearly resolved from starting materials, and peak integration enabled quantification of cleavage rates (Fig. 2). Because hydrolysis studies were performed using Kunitz inhibitor concentrations that were more than 10-fold higher than the K m values, the enzyme binding capacity was saturated, and the measured rates of hydrolysis approximate true catalytic rate constants (k cat ). Values of k cat for all of the Kunitz variants are summarized in Table 1. We note that the k cat value reported here for BPTI-WT is ϳ40% higher than that which we reported previously (46); this updated value reflects the average of multiple additional experiments.
The Effect of P 1 Substitution on Binding Affinity and Hydrolysis-Because P 1 residue identity is the primary specificity determinant for serine protease substrates (66) as well as a major determinant influencing the strength and specificity of enzyme association with canonical inhibitors (18,67), we anticipated that the P 1 position might play a significant role in the differential affinity and proteolytic stability of BPTI and APPI. To dissect the contribution of the P 1 residue in BPTI and APPI interactions with mesotrypsin, we used the kinetic data in Table  1 to compare variants that differ in possessing Arg versus Lys at P 1 but are otherwise identical (Table 2). Among the four pairs of possible comparisons, it is apparent that Arg at P 1 (versus Lys) favors both tighter binding (by a factor of 2-6) and more rapid cleavage (by a factor of 5-16). Using measured K i values to represent K m (see above), we can calculate the specificity constant k cat /K m for each inhibitor considered as a substrate of mesotrypsin. We find that Arg at P 1 (versus Lys) gives a 17-35fold enhancement of k cat /K m ( Table 2). These data show that mesotrypsin has a moderate preference for binding of inhibitors with Arg versus Lys at the P 1 position and a more pronounced preference for cleavage of substrates with Arg versus Lys at the P 1 position.
The Effect of PЈ 2 Substitution on Binding Affinity and Hydrolysis-Mesotrypsin possesses an Arg residue at position 193, where nearly all serine proteases, including other trypsins, have a highly conserved Gly, and Arg-193 has been found to be a major determinant of mesotrypsin's atypical catalytic properties (44,45). In a previous study, we found that mesotrypsin Arg-193 undergoes conformational change upon binding to FIGURE 1. Comparison of BPTI and APPI sequences and structures. A, a sequence alignment of the two Kunitz inhibitors shows two nonidentical residues (indicated by asterisks) within the canonical binding loop (boxed region): a conservative difference (Lys versus Arg) at the P 1 position, and a nonconservative difference (Arg versus Met) at the PЈ 2 position. Identical residues are shown in red, and nonidentical residues are in black. Structural features including two ␤-strands and the C-terminal ␣-helix are indicated above the corresponding sequences. B, structural comparisons of BPTI (left) and APPI (right), from crystal structures solved in our laboratory of the two inhibitors in complexes with mesotrypsin, show the overall similarity of the Kunitz inhibitor fold and the conformational similarities within the canonical binding loops of the inhibitors (in color); these loops are responsible for the majority of the inhibitor-enzyme contacts. BPTI in order to ameliorate steric conflict between Arg-193 and BPTI Arg-17, the PЈ 2 residue (46). Because of the close contact between mesotrypsin Arg-193 and the inhibitor PЈ 2 residue, we anticipated that the identity of the PЈ 2 residue, in BPTI a basic charged Arg and in APPI a more hydrophobic Met, might be a major factor influencing both binding affinity and inhibitor cleavage rates. By comparing kinetic data extracted from Table  1 for the four pairs of Kunitz inhibitors that differ only at the PЈ 2 residue (Table 3), we find that Met (versus Arg) at the PЈ 2 position results in stronger binding by a factor of 4 -18. We were surprised to find that the effects on rates of catalysis were much more minimal; we found no significant effect for PЈ 2 Met versus Arg in the APPI variants and only a slight effect on BPTI cleavage rate, where Met (versus Arg) slowed k cat by ϳ2-fold ( Table  3). The combined contributions from k cat and K m result in a 4 -10-fold difference in the substrate specificity constant k cat / K m , favoring the Kunitz variants with Met at the PЈ 2 position. The Inhibitor Scaffold Is a Major Determinant of Inhibitor Susceptibility to Hydrolysis-The most surprising outcome from kinetic analyses of the Kunitz inhibitors comes from examining the contributions of the scaffolds of BPTI and APPI to differences in inhibitor binding and cleavage. Pairwise comparisons of BPTI and APPI variants sharing identical sequence throughout the canonical binding loop allow us to assess the role of the protein scaffolds, defined as the protein structure exclusive of the P 5 -PЈ 3 binding loop (Table 4). We find that the APPI scaffold confers minimal (2-fold) enhanced affinity but an impressive 18 -84-fold enhancement in k cat relative to the BPTI scaffold ( Table 4). The aggregate contributions from k cat and K m result in substrate specificity constants k cat /K m more than 2 orders of magnitude greater for the Kunitz inhibitors bearing the APPI scaffold, compared with the sequencematched BPTI variants (Table 4). Thus, the enhanced vulnerability of APPI to mesotrypsin hydrolysis (relative to BPTI) is not only attributable to the canonical binding loop sequence but is largely determined by elements of the scaffold, despite the facts that the cleavage site is embedded within the canonical binding loop and that this loop makes the majority of close contacts with the enzyme.
Additivity of Free Energy Contributions of the Scaffold, P 1 , and PЈ 2 Residues-We find that in the interactions of BPTI and APPI with mesotrypsin, the P 1 and PЈ 2 residues and the scaffold each impact the rate of hydrolysis, the strength of association, and the substrate specificity constant to various degrees. Our data also enable us to assess the extent to which the impacts of these features on the measured functional properties are independent and therefore energetically additive versus cooperative and therefore energetically nonadditive. As reviewed by Wells FIGURE 2. Hydrolysis of APPI and BPTI variants monitored by reversed phase HPLC. For each inhibitor, the first and last sample traces from the linear phase of a representative hydrolysis time course are shown; the x axis of the chromatogram shows retention time in minutes, and the y axis shows absorbance at 210 nm. The depletion of intact inhibitor over time was quantified by peak integration (plot shown below the chromatograms), allowing calculation of rates of hydrolysis. Catalytic turnover rates k cat were calculated from the linear rate of inhibitor depletion, taking into account the enzyme/inhibitor ratio in the reaction (indicated on each plot). A, separation of intact APPI (peak 1) from cleaved product(s). APPI concentration in reactions was 50 M, and mesotrypsin concentration was 50 nM (for APPI-WT and APPI-M17R) or 500 nM (for APPI-R15K and APPI-R15K/M17R). Because samples were not treated with a reducing agent to reduce internal disulfide bonds, the major cleaved product is eluted as a single species (peak 2). A peak was not observed for mesotrypsin in the digests of APPI variants because mesotrypsin concentration was below the level of detection in these reactions. For the APPI-R15K/ M17R variant, the intact inhibitor represented a mixture of two species that were hydrolyzed by mesotrypsin with identical rates. Mass spectrometry and N-terminal sequencing identified the major peak as APPI possessing the expected N-terminal sequence, whereas the minor peak represented APPI with an additional three amino acids at the N terminus, possibly due to differences in processing during secretion by P. pastoris. B, separation of intact BPTI (peak 3) from cleaved products. BPTI concentration in reactions was 50 M and mesotrypsin concentration was 5 M (for BPTI-WT and BPTI-R17M), 2.5 M (for BPTI-K15R), or 1.25 M (for BPTI-K15R/R17M). Because samples were reduced with DTT prior to analysis, the N-terminal fragment (peak 4) and C-terminal fragment (peak 5) are resolved, as is mesotrypsin (peak 6). Having evaluated all possible combinations of P 1 (Arg versus Lys), PЈ 2 (Met versus Arg), and scaffold (APPI versus BPTI), we place our variants at the corners of a "thermodynamic additivity cube" (Fig. 3), in which the x-dimension represents a Met 3 Arg mutation at PЈ 2 , the y-dimension represents an Arg 3 Lys mutation at P 1 , and the z-dimension represents the composite set of mutations required to convert the APPI scaffold to that of BPTI. The edge of the cube connecting each pair of variants is annotated with the calculated value for ⌬⌬G cat (Fig. 3A), ⌬⌬G a 0 (Fig. 3B), or ⌬⌬G T ‡ (Fig. 3C). Each face of each cube represents a thermodynamic additivity cycle, in which the principle of additivity is satisfied when the energetic effects of the two single mutations are independent and can be summed to yield the effect of the double mutant. In a nonadditive cycle, the free energy change of one edge can be subtracted from that of the opposite parallel edge to yield a non-zero free energy of interaction (⌬G I ).
Because the changes in free energy for each cycle are calculated using experimentally measured k cat and/or K i values from four different variants (with associated experimental errors), small non-zero values for ⌬G I may be explained by experimental error, whereas larger values for ⌬G I provide evidence for cooperative behavior between the individual mutations (67)(68)(69)(70). Previous studies of ⌬⌬G a 0 additivity for binding of canonical inhibitors to serine proteases have suggested S.E. margins for ⌬G I of Ϯ400 -800 cal/mol, calculated at the 2 level from 20 -50% errors in measurement (67,69). Here, we have propagated individual errors from S.D. values in multiple measurements of k cat and K i for our variants; a complete table of ⌬⌬G values with propagated error measurements is available as supplemental Table S1. At the 2 level of uncertainty, we calculate maximum errors of 470 cal/mol for ⌬G I in the ⌬⌬G cat cycles, 270 cal/mol for ⌬G I in the ⌬⌬G a 0 cycles, and 550 cal/mol for ⌬G I in the ⌬⌬G T ‡ cycles, although for some cycles the errors are much smaller.
Although none of the cycles reveal very large free energies of interaction (⌬G I ), a number of cycles do reveal small but significant nonadditivity. Furthermore, the opposite parallel faces of the cube provide redundant additivity cycles measuring interactions between the same pair of mutations, and when similar results are found in each, this bolsters confidence in the robustness of the measurement. For example, in analysis of ⌬⌬G cat (Fig. 3A), we find nearly perfect additivity between P 1 and PЈ 2 mutations (front face of cube, ⌬G I ϭ 100 cal/mol; rear face of cube, ⌬G I ϭ 20 cal/mol) and slightly greater nonadditivity still within the limits of experimental error for cycles involving PЈ 2 versus scaffold mutations (top face of cube, ⌬G I ϭ 310 cal/mol; bottom face of cube, ⌬G I ϭ 190 cal/mol). By contrast, we find significant nonadditivity beyond the margin of experimental error for interaction of P 1 mutations with the scaffold (left face of cube, ⌬G I ϭ 630 cal/mol; right face of cube, ⌬G I ϭ 750 cal/mol). For the ⌬⌬G a 0 comparisons (Fig. 3B), all cycles showed nonadditivity slightly in excess of error limits, except for the P 1 versus scaffold cycle represented by the left face of the cube, for which ⌬G I was just slightly below the 2 error threshold. For ⌬⌬G T ‡ cycles (Fig. 3C), interactions between mutations appear somewhat more complex because in several instances the individual nonadditivities in k cat and K m act in a compensatory fashion to give an apparent additivity of k cat /K m . The only  ⌬⌬G T ‡ cycles for which ⌬G I slightly exceeds the 2 error threshold are the P 1 versus scaffold cycle represented by the left face of the cube (⌬G I ϭ 440 cal/mol) and the PЈ 2 versus scaffold cycle represented by the top face of the cube (⌬G I ϭ 550 cal/mol).
Structural Insights into Differential Proteolytic Stability of BPTI and APPI-We have published previously the crystal structure of mesotrypsin bound to BPTI (PDB entry 2R9P) (50), and here we report two new crystal structures of mesotrypsin bound to APPI. Comparisons of these structures may offer insights into the observed differences in inhibitor binding and hydrolysis. Because we anticipated that APPI would be rapidly proteolyzed under crystallization conditions with active mesotrypsin, we used an essentially isostructural but inactive mutant of mesotrypsin featuring a Ser-195 to Ala mutation in the active site. We succeeded in growing cocrystals of mesotrypsin-S195A with both APPI-WT and APPI-R15K variants. The structures were solved by molecular replacement and refined against data extending to 2.4 -2.5 Å resolution; Table 5 summarizes the crystal, data collection, and refinement statistics. Both structures contain four highly similar complexes in the asymmetric unit, each featuring a molecule of APPI bound in the canonical fashion at the mesotrypsin active site. The four complexes show small differences when superimposed; for example, C ␣ positions in several mesotrypsin loops (residues 9, 10, 21, and 61) deviate by 1.5-2 Å, and there is a significant bending of the chain G APPI molecule such that C ␣ positions of residues at the end distal to mesotrypsin (residues 6, 26, 28, 53, and 54) are displaced by 3-3.5 Å.
The overall fold of APPI and mode of interaction with mesotrypsin are similar to those found previously in structures of APPI complexed with bovine trypsin (PDB entry 1TAW) (71) and rat anionic trypsin (PDB entry 1BRC) (72). The major difference between the mesotrypsin⅐APPI complex and these other trypsin⅐APPI complexes occurs in the vicinity of mesotrypsin Arg-193, where similarly to our previous findings for the mesotrypsin⅐BPTI structure (46), the presence of Arg-193 forces a displacement of the PЈ 2 residue to ameliorate a steric clash. Comparing the mesotrypsin⅐APPI structure with the structure of bovine trypsin complexed with APPI (PDB entry 1TAW) (71), we see that the S ␦ atom of Met-17 is displaced downward by Arg-193 by about 2.5 Å, in turn displacing Phe-34 downward by 1.5 Å (Fig. 4).
Several aspects of our data hint at an increased degree of mobility or flexibility of the primed side of the APPI canonical loop in complex with mesotrypsin, by comparison with BPTI in complex with mesotrypsin, and also by comparison with APPI bound to other trypsins. Although we do see electron density for the Met-17 side chain, it is somewhat diffuse; the position of S ␦ is fairly well determined, but density is weak for C ␥ and absent for C ⑀ (Fig. 5A). When the different copies of the molecule in the asymmetric unit are compared, conformations of Met-17 vary somewhat (Fig. 5B). This contrasts with the mesotrypsin⅐BPTI structure (PDB entry 2R9P), in which the Arg-17 PЈ 2 residue was more precisely oriented, forming a water-bridged hydrogen bond with Arg-193 of mesotrypsin (see Fig. 3B of Ref. 46).
Another indication of the dynamic properties of APPI is found in analysis of B-factors, also referred to as crystallographic temperature factors, which reflect the fluctuations of atoms about their average positions and provide qualitative information about the residual mobility of regions in a crystal structure. Our diffraction data were collected at 100 K, where motions are considerably reduced, but we would expect that flash cooling will generate a frozen static picture of protein dynamics under ambient conditions (73). The four copies of the mesotrypsin⅐APPI complex in the crystal differ considerably in their average B-factors, with the chain A⅐chain E complex possessing the lowest B-factors whereas the chain D⅐chain H complex possesses much higher B-factors throughout; however, the trends described here and plotted for the chain A⅐chain E complex in Fig. 5C are consistently observed across all complexes. The non-primed side residues of the canonical loop show B-factors among the lowest in the molecule, indicative of a fairly rigid structure at the enzyme-inhibitor interface, as is typical in structures of serine proteases inhibited by canonical inhibitors. By contrast, the backbone atoms of the primed side residues of the APPI canonical loop show significantly elevated B-factors, evident in both the coloration of these residues in Fig.  5B and in the plot shown in Fig. 5C; this trend is present but much more subtle in the mesotrypsin⅐BPTI structure and is not recapitulated in the bovine trypsin⅐APPI structure. Additionally, the side chain atoms of Arg-17 show highly elevated B-factors in the mesotrypsin⅐APPI structure. Although it is important to keep in mind that differences in B-factors between structures reflect differences in resolution and may also be influenced by packing arrangements in the different crystal forms, our observations may also reveal differential residual mobility at the enzyme-inhibitor interfaces that is intrinsic to the different complexes. We hypothesize that differences in flexibility and dynamics between APPI and BPTI contribute to the observed differences in scaffold-dependent vulnerability to proteolysis by mesotrypsin.
Next, we compared mesotrypsin⅐APPI and mesotrypsin⅐ APPI-R15K structures, focusing on the P 1 residue that differs between these variants. The two structures show differences in electrostatic interactions in the primary specificity pocket that may help to explain the impact of P 1 Arg versus Lys on binding and hydrolysis (Fig. 6). The P 1 Arg residue of APPI-WT appears to form a stronger salt bridge with mesotrypsin Asp-189, with both guanidine nitrogens interacting directly with the Asp-189 carboxylate oxygens with distances of 3.16 Å (Fig. 6A). By contrast, in the mesotrypsin⅐APPI-R15K structure, the N of Lys-15 forms a much longer interaction with Asp-189 O ␦1 (3.79 Å) and a water-bridged interaction with O ␦2 (Fig. 6B). Thus, we hypothesize that stronger electrostatic interactions in the S 1 site favor the binding of P 1 Arg variants. The potential impact of these differential interactions on catalysis is considered further under "Discussion."

DISCUSSION
Resistance to proteolysis is a defining feature of Kunitz and other canonical serine protease inhibitors. These inhibitors bind target enzymes in a substrate-like manner, forming stable noncovalent complexes in which a peptide bond is ideally oriented for proteolytic cleavage (21), yet this Michaelis complex persists, apparently unreactive, for weeks or months (46,74). Through examining an exception to this paradigm, the enzyme mesotrypsin, which cleaves a subset of canonical inhibitors with substrate-like kinetics, we seek insights into the physical basis for canonical inhibitor resistance to proteolysis as well as insights into substrate discrimination by mesotrypsin. Here, we have used mutagenesis to probe the source of functional differences between two Kunitz family inhibitors: BPTI, a slowly hydrolyzed weak inhibitor of mesotrypsin (46), and APPI, a mesotrypsin substrate (50).
A major advance of this study is the finding that binding affinity and resistance to proteolysis are separable phenomena that can be affected very differently by the same mutation or set of mutations. It is often assumed that the slow hydrolysis of canonical inhibitors is a direct consequence of their very tight association with target proteases, with the corollary that mutations decreasing affinity will increase rates of turnover and vice versa. Our prior work with another canonical inhibitor, chymotrypsin inhibitor 2, has been largely consistent with this view  (75-77). Here, however, we see that this relationship is far from universal. At the PЈ 2 position in the Kunitz inhibitors, substitution of Met for Arg significantly enhances affinity while slightly slowing k cat (Table 3), but by contrast, substitution of P 1 Arg for Lys enhances both binding affinity and catalytic rate ( Table 2). Our observations of moderately enhanced binding and greatly enhanced catalysis of inhibitors bearing the APPI scaffold (compared with the BPTI scaffold) ( Table 4) further undermine the assumption that stronger affinities dictate slower rates of cleavage. Our new data support the idea that whereas the canonical loop sequence largely determines binding affinity, resistance versus vulnerability to proteolysis is a more global property impacted greatly by sequence and structure determinants beyond the canonical binding loop. Examining the impact of mutation at the P 1 position, we find that Arg (relative to Lys) confers ϳ30-fold improved substrate specificity and that much of this preference is manifested through k cat (Table 2). Serine proteases of tryptic specificity vary substantially in their preferences for Arg versus Lys (78,79), and it is common for a large part of the substrate discrimination of a serine protease to be implemented through k cat rather than K m (66). However, mesotrypsin appears unique among trypsins in the degree of preference for Arg over Lys at the P 1 position because discrimination between these residues is minimal for both rat anionic trypsin (80,81) and bovine trypsin (78); in this regard, mesotrypsin resembles more highly selective serine proteases, such as tissue plasminogen activator (79,82).
Our crystal structures of mesotrypsin bound to APPI-WT and to the APPI-R15K mutant indicate that the P 1 Arg residue extends more deeply into the S 1 pocket and interacts much more extensively with mesotrypsin Asp-189 than does the P 1 Lys residue; we hypothesize that these electrostatic interactions contribute to the greater affinity of mesotrypsin  toward P 1 Arg Kunitz variants. We have previously solved very high resolution structures of bovine trypsin reaction intermediates and examined the positioning of P 1 Arg and Lys residues as a substrate progresses through the tetrahedral intermediate and on to the acyl-enzyme (83); in those structures, we observed that both Arg and Lys form closer contacts with Asp-189 in the intermediates than we see here in the mesotrypsin⅐APPI complexes. If mesotrypsin cleavage of APPI proceeds with similar P 1 positioning to that seen in our bovine trypsin acyl-enzyme structures (PDB entries 2AGE and 2AGG), then the P 1 side chain will be drawn more deeply into the specificity pocket as the reaction proceeds; the required adjustment will be greater for Lys (0.8 Å) than for Arg (0.3 Å). We thus speculate that stronger anchoring of Arg (versus Lys) in the mesotrypsin S 1 pocket produces a complex more closely resembling the structure of the transition state, resulting in a lower activation energy barrier and consequent favorable impact on k cat . By contrast, the more loosely bound P 1 Lys residue may be more permissive of misalignments in the scissile peptide bond, propagated from the flexible primed side of the canonical loop, that serve to depress the relative catalytic rate.
Examining the impact of mutation at the PЈ 2 position, we see a marked effect on binding but little impact on catalysis (Table 3). In an earlier report (46), we theorized that electrostatic repulsion between PЈ 2 Arg-17 of BPTI and mesotrypsin Arg-193 would diminish binding affinity and also accelerate catalysis, by actively promoting release of the primed side leaving group following acylation, a potentially rate-limiting step for cleavage of Kunitz and other canonical inhibitors (21,76,84,85). The present study, in which we have mutationally substituted PЈ 2 Arg with Met, a residue sterically similar but lacking the potential for electrostatic repulsion, allows us to evaluate that hypothesis. Superposition of mesotrypsin⅐BPTI and mesotrypsin⅐APPI structures confirms the steric similarity in Arg and Met binding modes and indicates that mesotrypsin Arg-193 undergoes an identical conformational change in binding a Kunitz domain with PЈ 2 Met versus Arg. Our kinetic analyses confirm the electrostatic role of PЈ 2 Arg in reducing binding affinity but reveal that electrostatic repulsion is not a significant factor in catalysis because PЈ 2 Met versus Arg variants are cleaved at similar rates. Future studies, employing a more diverse range of PЈ 2 variants, will be required to more comprehensively assess the PЈ 2 specificity of mesotrypsin and the roles of steric versus electrostatic complementarity in mesotrypsin substrate and inhibitor preference.
One of the most intriguing observations of the present study is the significant role of the inhibitor scaffold in susceptibility to cleavage by mesotrypsin. Our data show that catalytic rates are greatly accelerated for APPI variants relative to the corresponding BPTI variants (Table 4). In our crystal structures, we have identified higher B-factors for APPI, suggestive of increased protein dynamics, and we speculate that differential loop mobility, particularly on the primed side of the canonical loop, may be a contributing factor influencing the relative cleavage rates of BPTI and APPI. Our previous studies with chymotrypsin inhibitor 2 have supported a model in which the major barrier to inhibitor proteolysis occurs after acylation, when retention of the primed side leaving group amine in the active site blocks access to water and favors religation of the scissile peptide bond (21,76). Recent investigations by the Goldenberg laboratory have suggested that this phenomenon probably also contributes to the proteolytic stability of BPTI complexes (84,85). Perhaps, the comparatively greater residual mobility of the primed side of the APPI binding loop enables more rapid dissociation of the leaving group from the acyl-enzyme, accelerating hydrolytic attack and catalytic turnover. A further question remains as to whether the relative proteolytic vulnerability of APPI by comparison with BPTI is unique to hydrolysis by mesotrypsin or is instead a more general property of the scaffold. Our data here are limited, but we have also measured the much slower rates of proteolysis of both inhibitors by human cationic trypsin, which cleaves APPI ϳ50-fold faster than BPTI (46,50). Thus, it is plausible that intrinsic dynamical attributes of APPI render this scaffold generally more vulnerable to proteolysis.
We have assembled our data relating to the individual impacts of P 1 , PЈ 2 , and scaffold mutations to assess possible interactions between these parameters through the use of thermodynamic additivity cycles. For k cat , we find convincing evidence of cooperative effects between P 1 and the scaffold. As reviewed previously (68), there are two major causes for additivity of mutational effects; one is physical or electrostatic interactions between residues that render their behavior interdependent, and the second is when a mutation results in a change in the mechanism or rate-determining step of an enzyme. Because the Kunitz domain P 1 residue is fully exposed prior to enzyme association and makes no physical contacts with other inhibitor residues, the first explanation does not appear likely in this instance. Rather, our observations are completely consistent with a shift in the rate-determining step for proteolytic turnover. We hypothesize that for BPTI, departure of the leaving group following acylation represents a rate-determining step, as discussed above. Because the rate of this step is primarily impacted by interactions between the primed side residues of the canonical loop and the enzyme, the identity of the P 1 residue plays a less dominant role. We speculate that for APPI, greater mobility of primed side residues results in more rapid leaving group departure and subsequent deacylation, to the point that the acylation step becomes at least partially ratedetermining, conferring greater dependence of the overall catalytic rate upon the identity of the P 1 residue.
For K a , multiple thermodynamic cycles show evidence of nonadditivity. The cooperative nature of PЈ 2 versus scaffold mutations may be attributable to direct physical interactions, because the PЈ 2 residue interacts directly with nonconserved Kunitz domain residue 34 (Val in BPTI, Phe in APPI). However, for nonadditivity of P 1 mutations versus scaffold or PЈ 2 mutations, our data are less easily explained and contrast with previous studies. Prior investigations of the association of canonical inhibitors mutated at the P 1 position with various serine proteases have generally shown interscaffolding additivity or similar energetic effects of a P 1 mutation irrespective of scaffolding context (67, 69, 86). The primary exception occurs in rare cases where different scaffolds impose alternative binding conformations for the P 1 residue (69,86), a situation that we do not expect here. Because our calculated ⌬G I values are all fairly small, one possible explanation for the apparent nonadditivity could be underestimation of the margin of error in our K i determinations. Errors were based upon S.D. of multiple inhibition determinations but did not take into account errors in titrations of enzyme or inhibitor stocks that were then used for multiple experiments. It is unlikely that such errors would fully account for the nonadditivity, however, because our k cat determinations would also have been sensitive to such errors, but they were instead highly internally consistent. The remaining possibility is that our K a additivity cycles reveal complex yet subtle energetic cooperativity between P 1 , PЈ 2 , and scaffold mutations; the reason for such cooperativity here, in contrast with previous studies, might be attributable to differences in modes of molecular recognition. It has been suggested by Laskowski and coworkers (70) that one element responsible for the predictably additive behavior in serine protease-canonical inhibitor association is a lock and key rather than induced fit mode of molecular recognition. Dynamical properties also appear to play an important role in cooperativity between functional groups in protein-small molecule ligand binding (73). Thus, it is possible that the conformational adjustments required at the mesotrypsin⅐inhibitor interface, suggestive of an induced fit mechanism of recognition, lead to more complex energetic relationships between individual residues than are seen in most other canonical inhibitor complexes.
We, as well as the Sahin-Tóth laboratory, have found that mesotrypsin possesses greatly enhanced catalytic activity toward canonical inhibitors and have suggested that such protease inhibitors are among the natural substrates of mesotrypsin (44,46,50). The human proteome possesses 11 Kunitz inhibitor domains with Lys or Arg at the P 1 position, found within eight distinct proteins: APP, amyloid precursor-like protein 2 (APLP2), hepatocyte growth factor activator inhibitors 1 and 2 (HAI-1 and HAI-2), tissue factor pathway inhibitors 1 and 2 (TFPI-1 and TFPI-2), bikunin, and the compound inhibitor WAP, Kazal, immunoglobulin, Kunitz and NTR domain-containing protein-1 (WFIKKN1). We have found APP to be a good mesotrypsin substrate (50), and many of the other Kunitz inhibitors listed may represent physiological substrates of mesotrypsin as well. However, our findings here show that for mesotrypsin, being a good substrate is a complex property not easily minimized to a simple sequence motif flanking the cleavage site. Instead, prediction of Kunitz domains likely to be targeted by mesotrypsin will require the identification of critical sequence and structural determinants that lie beyond the canonical binding loop and may require incorporation of information about inhibitor dynamics in the context of the mesotrypsin⅐Kunitz inhibitor complex. Further investigations will require study of additional Kunitz domains and would benefit from assessment of protein dynamics in solution.
Along with a number of other serine proteases (87)(88)(89)(90), mesotrypsin has been implicated in tumor growth and progression. In a recent study, we found that expression of the mesotrypsinogen/trypsinogen IV gene PRSS3 is up-regulated with advancing malignancy in a cell series modeling breast cancer progression and that suppression of PRSS3 expression inhibits malignant growth of breast cancer cells, whereas treatment with recombinant mesotrypsin stimulates the malignant phenotype (37). PRSS3 is also transcriptionally up-regulated with progression in lung, colon, and prostate cancers (32,34,36) and has been correlated with increased metastasis and poorer survival in non-small cell lung cancer (32). Engineered Kunitz domain inhibitors have recently entered development as therapeutics designed to target serine proteases instrumental in cancer progression (91,92), and it is worth considering whether the possible presence of mesotrypsin in the tumor microenvironment may serve as a mechanism of inactivation or clearance of these drugs. If so, selection or modification of Kunitz domain therapeutics to optimize stability against mesotrypsin proteolysis may be desirable; the studies presented here toward elucidation of the determinants of proteolytic stability will aid such efforts. Furthermore, in the event that mesotrypsin itself is validated as a therapeutic target, our studies will offer guidance toward engineering the Kunitz scaffold to maximize both mesotrypsin affinity and proteolytic stability. The present study represents a first step in this direction because we have found that affinity and proteolytic stability are largely orthogonal properties that can be independently manipulated. In the BPTI-K15R/R17M double mutant, we effectively transplanted the higher binding affinity of APPI onto the BPTI scaffold through manipulation of the canonical binding loop but retained nearly all of the relative proteolytic stability of BPTI. Further studies into the physical basis for the proteolytic stability of the BPTI scaffold to mesotrypsin may ultimately facilitate the engineering of inhibitors with yet further improved affinity and stability, of potential therapeutic value.