Structure-Function Analyses of Human Kallikrein-related Peptidase 2 Establish the 99-Loop as Master Regulator of Activity

Background: Serine proteases KLK2 and KLK3 clear the way for spermatozoa before impregnation. Results: Enzymatic assays and structures of KLK2 elucidate its catalytic action, especially when compared with conformations of similar proteases. Conclusion: Flexible loops around the active site of serine proteases open concertedly upon substrate binding. Significance: This mechanistic model will stimulate the design of pharmaceutical inhibitors.

Human kallikrein-related peptidases (KLKs) 2 comprise 15 serine proteases that display the chymotrypsin fold (MEROPS clan PA, family S1). The first member of this family (KLK1) was described almost a century ago (1) and subsequently named "kallikrein," as it was detected in the pancreas, or ␣⑀␣ (2). Together with KLK3 (prostate-specific antigen), whose discovery dates back to the 1960s (3), and KLK2, whose corresponding gene was isolated in the 1980s (4), KLK1 belongs to the classical kallikrein subfamily. These proteases are more closely related to each other than to the new kallikreins 4 -15 (called "new" because their gradual assignment to the KLK family started at the end of the 1990s (5)); KLK1-3 share a rather special surface loop that is 11 residues longer than the corresponding 99-loop of chymotrypsin (6), and therefore, this loop is also designated the "kallikrein loop" (7).
KLK2, which was formerly called hK2 or human glandular kallikrein 1 (Uniprot identifier P20191, MEROPS entry S01.161), is relatively restricted to prostatic tissue and seminal fluid in healthy individuals (8). Current knowledge pinpoints the main physiological role of KLK2 to sperm liquefaction. On the one hand, KLK2 may activate the zymogen form of KLK3 (9,10), although there is contradictory evidence (11). KLK3 in turn dissolves the sperm coagulum via degradation of semenogelins 1 and 2 and fibronectin (12). On the other hand, KLK2 itself is able to cleave the latter proteins at sites distinct from KLK3 (13). Maximum KLK2 activity in sperm appears immediately after ejaculation, and it decreases within 10 min due to complex formation with PCI. Because the time course of semenogelin/ fibronectin degradation and loss of KLK2 in vivo activity coincide, it is assumed that KLK2 complements KLK3 during sperm liquefaction (14).
However, KLK2 is aberrantly expressed in a range of human malignancies (15). Hence, elevated KLK2 levels in blood may constitute a valid marker for prostate cancer either alone or in combination with levels of various KLK3 isoforms (16). Due to its narrow tissue distribution, KLK2 has been regarded as a potential drug target in prostate cancer (17) or as a prodrug activator in targeted chemotherapy (18). In prostate carcinoma, KLK2 may promote growth or metastasis of tumor cells by interacting with the urokinase-type plasminogen activator system. KLK2 is able to activate the zymogen form of urokinasetype plasminogen activator (19), which may even initiate a positive feedback loop involving further activation of pro-KLK2 by urokinase-type plasminogen activator (11). Other cancer-related KLK2 targets include plasminogen activator inhibitor-1, an inhibitor of urokinase-type plasminogen activator (20), insulin growth factor-binding proteins 2-5 (21), and protease-activated receptor 2 (22).
Although KLK1 (23), KLK3 (24,25), and several new kallikreins (for review see Ref. 26) are well characterized on the structural level, the structure of KLK2 has remained elusive. To close this knowledge gap within the classical kallikreins, we present here two crystal structures of KLK2 obtained from Escherichia coli expression and refolding. Furthermore, we characterized a series of KLK2 mutants to elucidate its Zn 2ϩ inhibition and inactivation by proteolytic cleavage within the 99-loop. Kinetic properties of these mutants extend an in-depth comparison of KLK2 with related structures and investigate the diverse roles of the 99-loop in the regulation of KLK2 activity.
KLK2 was expressed as inclusion bodies and folded in vitro essentially as described for the catalytic domain of EK (28). In brief, E. coli M15[pREP4] cells (Qiagen) were transformed with the respective expression plasmid and grown in LB medium (supplemented with 100 g/ml ampicillin and 30 g/ml kanamycin) until the culture reached an A 600 of 1.2. Protein expression was induced with 0.5 mM isopropyl ␤-D-1-thiogalactopyranoside for 4 h at 37°C. Cells were disrupted by sonication, and insoluble matter was washed with Triton X-100-and EDTA-containing buffers. Washed inclusion bodies were solubilized 1:20 (w/v) in 7.5 M guanidine-HCl, pH 9, 50 mM Tris, 100 mM ␤-mercaptoethanol for 24 h, dialyzed against 5 mM citrate, pH 3.5-4.0, and resolubilized 1:10 (w/v) in 7.5 M guanidine-HCl, pH 4.0 -4.5, 50 mM Tris for several hours. Dropwise dilution of this solution into the 100-fold volume of 500 mM arginine, 50 mM Tris, pH 8.3, 20 mM NaCl, 1 mM EDTA, 5 mM cysteine-HCl, and 0.5 mM cystine yielded 5-10% folded protein after 3 days at 16°C.
Protein Purification-Wild type KLK2 and the mutants K95eM, K95eQ, and H101A were purified by (negative) ion exchange chromatography (IEC) and benzamidine (BEN) affinity chromatography (BENAC); purification of the mutants H25A, H91A, and H95fA comprised (positive) IEC, activation by EK, negative metal ion affinity chromatography, and BENAC ( Fig. 1). Chromatography resins were obtained from GE Healthcare. After tangential flow concentration, the refolding solution was loaded onto a Q-Sepharose column equilibrated in IEC buffer (50 mM Tris-HCl, pH 8.0). The ratio of load to resin was about 50:1 (v/v) in this and all following affinity chromatography steps; all buffers contained 3 mM sodium azide. In negative IEC, the flow-through contained mature KLK2. In positive IEC, pro-KLK2 was eluted from the column with 3 resin volumes of IEC buffer supplemented with 150 mM NaCl.
Zymogen forms of KLK2 from positive IEC were incubated with EK in a molar ratio of 1000:1 for 15 h at 20°C. EK was produced in-house as previously described (28). The digestion mixture was brought to 500 mM NaCl, 10 mM imidazole and loaded onto a Ni 2ϩ -or Co 2ϩ -Sepharose column equilibrated in 50 mM Tris-HCl, pH 8.0, 500 mM NaCl, 10 mM imidazole. The flow-through contained mature KLK2, whereas the resinbound residual pro-KLK2 cleaved propeptide and EK.
For BENAC, flow-through from negative IEC was brought to 500 mM NaCl and loaded onto a benzamidine-Sepharose column equilibrated in BENAC buffer (50 mM Tris-HCl, pH 8.0, 500 mM NaCl). Flow-through from metal ion affinity chromatography was directly loaded due to its proper sodium chloride concentration. After washing with 8 resin volumes of BENAC buffer, bound KLK2 was eluted with 3 ϫ 2.5 resin volumes of BENAC buffer supplemented with 25, 50, and 100 mM benzamidine.
As final polishing step, size exclusion chromatography was performed over a Superose 6 10/300 GL column connected to an ÄKTA FPLC system (GE Healthcare). To this end, the BENAC eluate was concentrated in Amicon Ultra-15 Centrifugal Filter Units, molecular weight cutoff 10 kDa (Millipore, Billerica, MA). Per run, 500 l of concentrate were loaded onto the column at 4°C (running buffer: 20 mM Tris-HCl, pH 8.0, 20 mM NaCl, 5 mM benzamidine). Fractions that corresponded to the monomeric KLK2 peak were combined and concentrated to 12 mg/ml. Chemicals of the highest purity available were either from AppliChem (Darmstadt, Germany), Carl Roth (Karlsruhe, Germany), Merck, or Sigma.
Enzyme Kinetics and Inhibitory Studies-Bz-PFR-pNA, H-GHR-AMC, H-PFR-AMC, H-Arg-AMC, and PPACK were obtained from Bachem (Weil am Rhein, Germany). Amidolytic activity was generally measured in 100 l of assay buffer (50 mM Tris-HCl, pH 7.5, 100 mM NaCl, 10% (v/v) DMSO, 0.1% (w/v) BSA) containing 400 ng (150 nM) of KLK2 and 250 M chromogenic or fluorogenic substrate. pH values of the reaction mixtures were routinely checked to exclude any effects of pH changes. Time-dependent substrate cleavage corresponded to changes in absorbance at 405 nm (for pNA substrates) or fluorescence at 460 nm (for AMC substrates; excitation wavelength: 380 nm) and was recorded on an Infinite M200 microplate reader (Tecan, Männedorf, Switzerland). Protein concentrations were determined by absorbance at 280 nm using computed extinction coefficients and molecular weights. For calculating k cat values, we performed active site titration of the respective KLK2 variant with PPACK and corrected the data accordingly.
The pH optimum was determined in 100 mM SPG buffer (12.5 mM succinate, 43.75 mM NaH 2 PO 4 , 43.75 mM glycine). Zn 2ϩ inhibition curves were measured without BSA, as its metal binding sites sequestered Zn 2ϩ ions from the reaction buffer. However, KLK2 adsorbed to the microplate walls in the absence of BSA, which interfered with the measurement of Michaelis-Menten kinetics at different Zn 2ϩ concentrations. Thus, we saturated all Zn 2ϩ binding sites in BSA by dialyzing BSA-containing assay buffer against the 200-fold volume of assay buffer with the desired Zn 2ϩ concentration. Reactivity toward the burst reagent 4-nitrophenyl-4-guanidinobenzoate (NPGB) (29) was determined by adding 75 l of 100 M KLK2 to 675 l of 50 mM HEPES, pH 7.0, 150 mM NaCl, 50 M NPGB and by detecting the concomitant change in absorbance at 405 nm. Substrate specificity was determined by positional scanning as previously described (30). Data were analyzed with nonlinear regression models as implemented in QtiPlot v0.9.8.8 (31).
Crystallization and Data Collection-Crystals of active wild type KLK2 were grown at 20°C by vapor diffusion in 500 nl of sitting drops of a 12 mg/ml protein solution in 20 mM Tris-HCl, pH 8.0, 20 mM NaCl, 10 mM benzamidine, 3 mM NaN 3 that were mixed with 500 nl of precipitant. Crystals of benzamidine-inhibited KLK2 appeared within 10 days in 500 mM (NH 4 ) 2 SO 4 , 1 M Li 2 SO 4 , 100 mM sodium citrate and were directly frozen in the nitrogen gas stream (100 K) without prior cryoprotection. Crystals of the KLK2-PPACK complex were obtained by soaking KLK2-benzamidine crystals for 1 h in mother liquor supplemented with 7 mM PPACK. Data sets were collected in-house (Bruker AXS microstar rotating anode, mar345 image plate) or at the beamline X12 (MarMosaic 225 charge-coupled device) at the DESY in Hamburg (see Table 1).
Structure Determination and Refinement-Diffraction data were integrated by iMosflm v1.0.5 (32) and scaled with Scala v3.3.20 (33) included in the CCP4 v6.2.0 program suite (34). Initial phases were obtained for the KLK2-BEN data set by molecular replacement with Phaser v2.3.0 (35) using KLK3 (2zch/chain P) as the search model in the resolution range of 1.9 -36.0 Å. The scores of the top solution were RFZ ϭ 17.5, TFZ ϭ 32.9, LLG ϭ ϩ1456, and R-factor ϭ 45.8. Neither the rotational nor the translational searches yielded a second unrelated peak. Phases for the KLK2-PPACK data were obtained by molecular replacement using Phaser with the KLK2-BEN polypeptide model after refinement. Essentially, the parameters were similar to the initial search, resulting in an LLG ϭ ϩ2688 and an R-factor ϭ 34.8. Both KLK2 structures contained one molecule in the asymmetric unit and had a Matthews coefficient of 2.33 and solvent content of 47.2%. Model building in Coot v0.6.2 (36) alternated with restrained maximum likelihood refinement in REFMAC v5.6.0117 (37) using standard target parameters (38). Global quality indicators of the final models were in the typical range for the obtained resolution (see Table 1). Geometric restraints for the two covalent bonds between PPACK and KLK2 were determined in JLigand v1.0.36 (39); two CIF files were generated, each describing one bond, and then manually merged (supplemental File S1). Correct Asn and Gln side chain rotamers were assigned by NQ-Flipper v2.7 (40). Side chains of the following surface residues lacked interpretable electron density in both models unless otherwise indi-  (41), these side chains were modeled in their most likely conformation with full occupancy, which resulted in high B values. Met-167-S␦ was modeled with two alternate conformations in the side chain beyond C␥. Structures were validated by MolProbity v3.19 (42), SFCHECK v7.03.16 (43) and phenix.model_vs_data (44), included in the Phenix v1.7.3-928 suite of programs (45).
Dendrograms that illustrate relationships between loop conformations in the KLK family were prepared as follows. Protease structures were aligned in PyMOL by the CEalign package (63). A PyMOL script (supplemental File S2) calculated mutual dissimilarities for a target loop and stored them in a square distance matrix (we define the mutual dissimilarity of a certain loop in two superposed structures as the r.m.s.d. between loop C␣ atoms with equal residue numbers in these structures). From this distance matrix the NEIGHBOR program in the PHYLIP v3.695 package (64) constructed a dendrogram according to the UPGMA algorithm.

Preparation of Wild Type and Mutant KLK2-A zymogen
form of wild type KLK2, whose propeptide contained the canonical EK recognition site (DDDDK), suffered from unspecific fragmentation during activation by EK, as already observed by Lövgren et al. (75). Because we expected these cuts to decrease protein homogeneity, we designed two alternative propeptides that differed in their recognition site; on the one hand, the sequence SGDR most likely constituted a more efficient EK site (76) and, therefore, would require significantly smaller amounts of EK, resulting in less unspecific cuts. On the other hand, we deduced from KLK2 specificity profiling (see below) that the sequence PSFR represented an autoactivation site and would eliminate the need for EK in the first place. Interestingly, both sequences ensured that autoactivation of refolded pro-KLK2 proceeded to completion within 48 h. Accordingly, we selected the SGDR propeptide for all expressions of the wild type peptidase, as it doubled refolding yields compared with the PSFR propeptide. Final yields of wild type KLK2 exceeded 10 mg per liter of bacterial culture. The protein was Ͼ95% pure and monomeric, as judged from SDS-PAGE and size exclusion chromatography (Fig. 1).
Proteolytic Activity and Specificity-Refolded KLK2 cleaved the chromogenic substrate Bz-PFR-pNA following Michaelis-Menten kinetics with K m and k cat values of 75 Ϯ 2 M and 1.23 Ϯ 0.02 s Ϫ1 , respectively ( Fig. 2A, see Table 2). A pH value around 8 was optimal for its proteolytic activity (Fig. 2B). The fluorogenic H-PFR-AMC turned out to be the best substrate for KLK2 so far, resulting in K m and k cat values of 69 Ϯ 3 M and 23.42 Ϯ 0.09 s Ϫ1 . Lövgren et al. (13) reported a comparable pH optimum and K m and k cat values of 40 M and 0.92 s Ϫ1 , respectively, for the cleavage of H-PFR-AMC by KLK2 expressed in eukaryotic cells. The lower K m value for the latter substrate may result from favorable interactions between its fluorogenic AMC group and the S1Ј site. By contrast, the significantly higher turnover number of E. coli KLK2 for H-PFR-AMC might depend on a more accessible active site. A glycosylated Asn-95 in KLK2 from eukaryotic expression could favor a closed Positional scanning with fluorogenic peptide substrates from a synthetic combinatorial library (30) confirmed that KLK2 is a protease with predominantly tryptic specificity (Fig. 2D). Accordingly, P1 residues (applying the Schechter and Berger (77) nomenclature) are restricted to basic side chains, with Arg preferred ϳ2-fold over Lys and only a minor appearance of His. Surprisingly, KLK2 accepts proline as third best residue in P1. The S2 pocket strongly favors the aromatic residues Phe, Tyr, and Trp, whereas Asp is similarly preferred; Leu and Met are also allowed. By contrast, basic (Lys, Arg, His) and small hydrophobic side chains (Gly, Ala, but also Pro) are nearly excluded in P2. Substrate specificity is less pronounced for the S3 and S4 sites. In accordance with the solvent-exposed S3 subsite, charged, polar, and small side chains are accepted (Ser, Glu, Ala, Asp, Gln, Lys) with the notable exception of Arg, whereas all hydrophobic residues are hardly tolerated at this position. In contrast, S4 accepts hydrophobic and to a lesser extent polar side chains, with Trp being strongly preferred over Pro, Met, and several other residues.
Crystallization and Overall Structure-One batch of purified KLK2 yielded single crystals of the benzamidine-inhibited peptidase (KLK2-BEN) that diffracted in-house up to 1.9 Å. These crystals could also be soaked with PPACK (KLK2-PPACK), which left the unit cell parameters essentially unchanged (see Table 1). Only two interface areas between symmetry-related protein molecules exceed 400 Å 2 (707 Å 2 and 533 Å 2 , respectively). Moreover, the PISA server classified these interfaces as a result of crystal packing. In agreement with these findings, size exclusion chromatography and dynamic light scattering detected KLK2 monomers exclusively independent of refolding protocol and buffer conditions. The KLK2 molecule resembles an oblate ellipsoid with diameters of 35 and 50 Å, respectively (Fig. 3, A and B). In addition to the two six-stranded ␤-barrels of trypsin-like serine proteases, residues 20 -21 form a short ␤-strand adjacent to strand B of the C-terminal barrel (chymotrypsinogen numbering scheme) (78) adapted for kallikreins (7). In KLK2-PPACK, a 3 10 (Fig. 3C). The benzamidine-and PPACK-containing KLK2 structures superpose with a C␣ r.m.s.d. of 0.88 Å, which is mainly explained by backbone deviations around the 99-loop, Glu-174, and Trp-215 in KLK2-PPACK.
Overall, the substrate binding site of KLK2 has a slightly negative potential, which is prominent at the S1 pocket (Fig. 3B). Two positive surface potential regions exist: first, at the back of the 37-loop with Trp-67 and Arg-82; second, at His-48 and the C-terminal Arg-235, Lys-236, and Lys-239. This extended, positively charged patch at the back of the protein corresponds partially to the anion binding exosite II of thrombin (79). It may bind allosteric effectors like heparin, which accelerates the association of KLK2 and PCI 4-fold (13).
Active Site Cleft-Adjacent to the extended catalytic triad (Asp-102, His-57, Ser-195, and Ser-214), the backbone amide NH groups of Gly-193 and Ser-195 constitute the oxyanion hole, which is occupied by a sulfate ion in the KLK2-BEN complex structure. The amidine group of benzamidine (BEN 301) and the carboxylate group of Asp-189 at the bottom of the S1 pocket interact via a symmetrically bidentate salt bridge. BEN-N2, which points toward the surface of the protein, also forms a hydrogen bond to the amide oxygen of Pro-217, similarly as in the KLK6-BEN complex (48). Also, BEN-N1, which points toward the core of the protein, binds to Thr-190-O␥1, Tyr-228-O, a water molecule (w412), and Asp-189-O␦1, sitting at the corners of a pentagonal hydrogen bond network. The aromatic ring of BEN lies between the peptide bonds Trp-215-Gly-216 and Cys-191-Gly-192.
Thr-190 explains why KLK2 prefers Arg merely 2-fold over Lys, in contrast to other trypsin-like peptidases (80); its O␥ atom provides a hydrogen bonding partner for P1-Arg-N. Consequently, KLK2 binds P1-Arg substrates too tight for efficient catalysis. Ala-226 might further decrease the preference for Arg in P1, as its methyl side chain is in close contact with the P1-Arg guanidinium group (81).
To avoid bias regarding the existence of PPACK in the KLK2-PPACK model, we also calculated electron density in the absence of the ligand (82). This approach yielded well defined positive difference electron density in the nonprimed substrate binding sites, which we could clearly identify as PPACK (Fig.  4A). However, electron density around His-57 indicated that the KLK2-PPACK crystal consisted of two species of proteaseinhibitor complexes. One species contained only one covalent bond between PPACK (0G6-C2) and Ser-195-O␥, characteristic for the tetrahedral epoxy ether intermediate of the chloromethyl ketone inhibition reaction (83). To avoid clashes with the chlorine atom, His-57 bends away from Ser-195 in this intermediate complex (Fig. 4B). The other species contained two covalent bonds between PPACK and KLK2 (0G6-C2 to Ser-195-O␥ and 0G6-C3 to His-57-N⑀2) and, therefore, represents the final protease-inhibitor complex (Fig. 4C). Despite the heterogeneous composition of the crystal, the deposited KLK2-PPACK structure represents the covalent adduct with both His-57 and Ser-195. We believe that the short soaking time (30 min) and the initial presence of competing benzamidine in the S1 site delayed the reaction of KLK2 with PPACK.
Backbone atoms of PPACK and KLK2 form a set of five "canonical" hydrogen bonds (84) in the KLK2-PPACK structure. On the one hand, two hydrogen bonds connect P1-Arg-O  The Pro-217-Glu-218 -Pro-219 motif (Fig. 3C), which is unique among the human kallikreins, most likely rigidifies the entrance frame of the S1 pocket. Notably, it also tilts the backbone carbonyl group of Gly-216 by at least 30°compared with typical serine protease geometries, resulting in an orientation of Gly-216-O that requires sufficient backbone flexibility of the P2 residue to allow a favorable hydrogen bond geometry between Gly-216-O and P3-N. The conformation of Gly-216 explains why specificity profiling disallows proline in P2 (Fig. 2D); P2-Pro will induce a P3 main chain geometry that impedes the antiparallel ␤-sheet interactions between the backbone of the P3 residue and Gly-216, which serves as S3 residue in KLK2.
In the KLK2-BEN structure, a second benzamidine molecule (BEN 302) is located at the lower back of KLK2. Its amidine group forms hydrogen bonds to Trp-20-N⑀1, Glu-23-N, Tyr-137-O, and two water molecules from a symmetry-related protein molecule. To check whether this secondary benzamidine binding site only existed in this interface, we scored the interactions between KLK2 and BEN 302 with DSX. This program evaluates protein-ligand complexes by using a knowledge-based scoring function where increasingly negative scores indicate more favorable interactions. If DSX considered both symmetry-related molecules, it reported a score of Ϫ62.3. This value is comparable to the score of benzamidine bound to the S1 site (Ϫ65.4) and less negative than the score of the strongly binding ligand PPACK (Ϫ130.9). However, if DSX neglected contributions of the symmetry-related molecule, the score of BEN 302 increased to Ϫ26.6. Thus, the secondary benzamidine binding site presumably requires formation of the crystal lattice, whereas it is absent in solution.
Loops around the Active Site Cleft-Although calcium ions are well known allosteric modulators of trypsin (85) or factor IXa, they did not stimulate the proteolytic activity of KLK2 (data not shown). The 75-loop of trypsin provides the Glu-70 and Glu-80 side chains as well as the Asn-72 and Val-75 carbonyl-O atoms as Ca 2ϩ ligands with two water molecules, completing the octahedral calcium binding site (86), which is comparable to the site found in factor IXa (87). By contrast, the 75-loop of KLK2 exhibits Arg-70 and Gly-80, respectively (Fig.  5A). Furthermore, the guanidinium group of Arg-70 essentially The tetrahedral intermediate is present in the crystal with low occupancy. C, the final KLK2-PPACK complex, which is characterized by a second covalent bond between the inhibitor and its target (0G6-C3 to His-57-N⑀2). Note that the deposited KLK2-PPACK coordinates (PDB ID 4nff) model the latter situation only. Maximum likelihood-based difference electron densities are drawn in blue (2mF obs Ϫ DF calc map contoured at 1.5), green, and red (mF obs Ϫ DF calc map contoured at ϩ3 and -3, respectively). Structure factors and electron densities were calculated by REFMAC (37) and fft (67), respectively. Because crystallographic programs do not contain parameters for the two covalent bonds between PPACK and KLK2, they were generated with the software JLigand (39) and written to a crystallographic information file (CIF), which was employed in crystallographic refinement (supplemental File S1).  In contrast, the more flexible and ill-defined side chain of Arg-153 protrudes into the solvent, which may explain why proteolysis after Arg-153, but not after Arg-151 has been observed (88,89).
As pointed out above, residues of the 99-or kallikrein loop contribute to nonprimed substrate binding sites. Because this loop greatly varies in length among the members of the KLK family, it is a distinguishing structural feature. To compare the 99-loop of KLK2 with other 99-loops whose conformation is already known, we aligned all available KLK structures: human KLKs 1-8, porcine KLK1 (pKLK1), rat tonin (rKlk1c2, formerly called rat kallikrein 2), equine KLK3 (eKLK3), mouse kallikreins (mKlks) 1b4, 1b26 (formerly called mouse glandular kallikrein 13), and 8 as well as bovine trypsin and chymotrypsin A. As evident from Fig. 6A, 99-loops may be generally classified as short, intermediate, or long. Short 99-loops lack any insertion with respect to chymotrypsin (KLKs 4 -7, 14, and 15). Intermediate 99-loops contain 2-8 additional amino acids after residue 95 (KLKs 8 -13, mKlk8). Long 99-loops with 11 additional amino acids (residues 95a to 95k) characterize the so-called classical kallikreins KLK1-3, pKLK1, rKlk1c2, mKlk1b4, mKlk1b26, and eKLK3. In the respective crystal structures most of the long 99-loops lack electron density for at least two central residues. Likewise, the KLK2-BEN and KLK2-PPACK structures lack any interpretable electron density for Leu-95d to Glu-97 and Lys-95e to Asp-96, respectively. Apparently, the notable insertion confers extra flexibility to this loop. Nevertheless, the 99-loop is entirely well defined in mKlk1b26, KLK3, and eKLK3. In these structures the kallikrein loop is rigidified by interactions with a symmetry-related molecule (eKLK3), with another molecule in a noncrystallographic dimer (mKlk1b26), or with a stabilizing antibody and N-linked glycans (KLK3).
An exhaustive analysis of their secondary structure (Fig. 6, B and C) revealed that all 99-loops contain two conserved hydrogen-bonded 3-turns, one at their N terminus (91-O to 94-N) and one at their C terminus (99-O to 102-N). The latter presumably stabilizes the catalytically efficient orientation of Asp-102. Furthermore, all long 99-loops that are sufficiently well resolved contain a short 3 10 -helix, which starts at residue 95a. However, we sought to extend these rather qualitative observations by a quantitative approach. To this end we performed a multiple alignment of KLK structures from which we calculated a symmetrical distance matrix. Matrix elements represented the pairwise 99-loop dissimilarities, i.e. r.m.s.d. between equivalent 99-loop C␣ atoms. Using this distance matrix, hierarchical clustering based on the UPGMA algorithm yielded a 99-loop dendrogram (Fig. 6C). Accordingly, the 99-loops belong to six structurally similar types or clusters (Fig. 6D). Cluster I comprises the long 99-loops of free pKLK1 (PDB ID 2pka) and of pKLK1 in complex with bovine pancreatic trypsin inhibitor (PDB ID 2kai); cluster II only contains the short 99-loop of KLK4 (PDB ID 2bdh). Although the pKLK1 and KLK4 99-loops differ in length, they both bend away from the active site and form a roof above the substrate binding site. All other 99-loops (clusters III to V*) bend toward the active site cleft or even form a lid over it. Loops in cluster III (KLK5 (PDB ID 2psx), KLK6 (PDB IDs 1l2e and 3vfe), pro-KLK6 (PDB ID 1gvl), and KLK7 (PDB IDs 2qxg and 3bsq) resemble the short loops of trypsin (PDB ID 1ce5) and chymotrypsin (PDB ID 1yph). The 99-loops of human KLK8 and mouse Klk8 (PDB ID 1npm) in cluster IV exhibit a comparable conformation, but a three-residue elongation induces an additional bend at its midpoint. Cluster V comprises the long 99-loops of human KLKs 1 (PDB ID 1spj), 2 (PDB IDs 4nfe and 4nff), and 3 (PDB ID 2zch and 3qum/chains P and Q), pKLK1 in complex with hirustasin (PDB ID 1hia), mKlk1b26 (PDB ID, 1ao5), and mKlk1b4 (PDB ID 1sgf/chain A). Finally, cluster V* only contains the kallikrein loop of eKLK3 (PDB ID 1gvz). Type V and V* loops represent interconvertible open and closed conformations, respectively. (24). The active site cleft of human KLK3 (type V) is accessible for a substrate, whereas the 99-, 148-, and 220-loop block the substrate binding site in eKLK3 (type V*) (the nomenclature of clusters V and V* emphasizes the connection between open and closed 99-loop conformations; it also refers to the E-E* equilibrium of conformational selection, which is discussed below). The central eight residues of the rKlk1c2 99-loop (PDB ID 1ton) are disordered, and the remainder of the loop seems markedly distorted, as two of its residues participate in Zn 2ϩ binding. As a consequence, we did not assign this loop any cluster.
Irreversible Inhibition by 99-Loop Autolysis-The close proximity of the 99-loop to the active site cleft prompted us to investigate its influence on the inactivation of KLK2 by autolysis and on its inhibition by divalent zinc cations. Mature KLK2 underwent autolysis in solution at a single site within the 99-loop (Lys-95e2His-95f), which was confirmed by N-terminal sequencing. To confirm that autolysis depended on Lys at position 95e, we mutated this residue to Met or Gln. Indeed, the KLK2 mutants K95eM and K95eQ were nearly as active as the wild type (Table 2) but completely resisted autolytic 99-loop cleavage. Besides, they displayed less unspecific cuts during EKmediated activation, most likely because EK was unable to cleave after Met95e or Gln95e and due to a more compact conformation of intact KLK2. Interestingly, Lys-95e resides next to the single putative N-glycosylation site of KLK2 at Asn-95 (90). Glycosylation at the latter residue might limit access to the cleavage site, which would explain why 99-loop cleavage has never been reported for KLK2 isolated from natural sources.
Intriguingly, autolysis had a major impact on proteolytic activity; KLK2 with a cleaved 99-loop was completely inactive toward proteinaceous (itself) and chromogenic substrates (Bz-PFR-pNA). The time dependence of activity loss coincided with progressive 99-loop cleavage (Fig. 7A and its inset) and indicated a second-order reaction mechanism, which agrees well with autolysis in trans, i.e. KLK2 ϩ KLK2 3 KLK2 ϩ cut KLK2. Also, the mutants K95eM and K95eQ became gradually inactive during storage at room temperature over several days (not

Kinetic constants of KLK2 variants and IC 50 /K i values for their inhibition by Zn 2؉ ions
Proteolytic activity was measured in 100 l of assay buffer (50 mM Tris-HCl, pH 7.5, 100 mM NaCl, 10% (v/v) DMSO, 0.1% (w/v) BSA) containing 400 ng (150 nM) of KLK2 and 250 M chromogenic substrate (Bz-Pro-Phe-Arg-pNA unless otherwise indicated). k cat values were normalized according to active site titration. The error of the catalytic efficiency k cat /K m was calculated according to the Fenner (61) formula from the respective standard errors. ND, not determined. Bz-PFR, Bz-Pro-Phe-Arg-pNA; H-GHR, H-Gly-His-Arg-AMC; H-PFR, H-Pro-Phe-Arg-pNA. shown). Contrary to wild type KLK2, the slow activity loss of these mutants corresponded to a decrease in their concentration instead of 99-loop cleavage. Apparently, even the substitution of a single amino acid destabilized KLK2 and facilitated its precipitation, which is in line with a noticeable reduction in refolding yields for all KLK2 mutants. We further addressed possible mechanisms for the unexpected enzyme inactivation triggered by the 99-loop cleavage. Obvious possibilities are (i) an increase in affinity for Zn 2ϩ , which itself inhibits KLK2 activity, (ii) blockage of the substrate recognition sites (K m -type inhibition), or (iii) disruption of the catalytic triad (k cat -type inhibition). We could exclude option (i), which is 99-loop autolysis did not inactivate KLK2 by increasing its affinity for Zn 2ϩ ; neither 10 mM EDTA nor 1 mM 1,10-phenanthroline-5-amine was able to reactivate clipped KLK2 (Fig. 7D). Because the 99-loop defines the S2 and S4 sites, but not the S1 site, we expected that 99-loop-cleaved KLK2 would still bind to immobilized benzamidine. The observed binding affinity exceeded that of wild type KLK2, indicating the accessibility of the S1 site with possible subtle changes in the binding geometry. Next, we tested if cleaved KLK2 would still react with H-Arg-AMC and the active site titrant 4-nitrophenyl-4-guanidinobenzoate, which require only an intact S1 site for binding. Importantly, this was not the case. Together, these findings would speak against option (ii) and rather for option (iii), i.e. k cat -type inhibition. Thus, 99-loop cleavage primarily affects the catalytic triad and possibly the presentation of the bound P1 residue to Ser-195. Indeed, 99-loop cleavage inactivates KLK2 in a way that is analogous to noncompetitive inhibition (Fig. 7, B and C). To further investigate the degree of distortion of the catalytic residues, we employed PPACK as a competitor to benzamidine binding. PPACK harbors an extraordinarily electrophilic carbonyl car-bon, compensating for a partially disrupted triad. Indeed, both intact and cleaved KLK2 were able to react with PPACK at a 10-fold molar excess (i.e. 100 M PPACK added to 8 M KLK2), as evidenced by the lack of binding to a benzamidine-coupled resin. These observations support the notion of a distorted, but not completely disrupted, catalytic triad and oxyanion pocket. Proteolysis within the 99-loop may perturb the triad in the following way; cleavage between Lys-95e and His-95f generates two novel, flexible termini. Movement of these termini dislocates Asp-102 and concomitantly distorts the catalytic triad.

KLK2/mutant
Reversible Inhibition by Zinc-Zn 2ϩ ions inhibit wild type KLK2 in the micromolar range (Table 2). After incubation with 200 M Zn 2ϩ for 16 h followed by the addition of 10 mM EDTA, 99% of the initial activity was recovered. Lineweaver-Burk and Eadie-Hofstee plots illustrate that the inhibition type is almost perfectly competitive (Fig. 7, E and F). KLK2 shares this susceptibility to Zn 2ϩ with the prostatic KLKs 3 and 4 and with the epidermal KLKs 5,7,and 14 (91). In both tissues high Zn 2ϩ concentrations presumably modulate KLK activity (92,93). The lack of structural evidence for a Zn 2ϩ binding site prompted us to investigate Zn 2ϩ inhibition by a panel of KLK2 mutants ( Table 2). In each mutant we substituted alanine for one residue, which contributes to known inhibitory Zn 2ϩ sites in other KLKs. Substitutions of residues Lys-95e and His-95f reduced the inhibitory effect not Ͼ3-fold, which suggests that other residues are the critical Zn 2ϩ ligands.

DISCUSSION
Substrate specificity as determined by positional scanning somewhat disagrees with specificity inferred from phage display (94). These discrepancies may result from the different proteases that were used in these experiments. Although we Shown are Lineweaver-Burk (b) and Eadie-Hofstee plot (c) of autolytic inactivation. Substrate concentration-dependent reaction velocities were measured at three time points during autolysis when 100, 40, and 15% of KLK2 were active, respectively. These plots confirm that autolysis inactivates the protease in a way that is analogous to noncompetitive inhibition, as the regression lines intersect at the x axis (b) or run in parallel (c). AU, absorbance units. d, Zn 2ϩ chelators did not restore proteolytic activity of KLK2 with a cut 99-loop (Phen is the Zn 2ϩ -specific chelator 1,10-phenanthroline-5-amine). Lineweaver-Burk (e) and Eadie-Hofstee (f) plots indicate that inhibition of KLK2 by Zn 2ϩ is competitive, as the regression lines almost intersect at the y axis in both plots. All measurements involved 150 nM wild type KLK2 and 250 M Bz-PFR-pNA. Relative velocities v rel are relative to the highest velocity in the respective panel.
carried out positional scanning with KLK2 expressed in prokaryotes, phage display employed KLK2 from human samples, which most likely was glycosylated. Furthermore, subsite cooperativity may influence the phage display results. Investigation of cooperativity between subsites requires other methods, such as enzyme kinetic measurements of synthetic substrates with systematically varying residues in one position. Despite a Lys: Arg ratio of about 1:2 in positional scanning, phage displayderived peptides contained P1-Arg in 40 of 41 cases. A similar, almost exclusive occurrence of Arg in P1 has been reported for physiological KLK2 substrates (14) and for complexes of KLK2 with proteinaceous inhibitors (95,96). The atypical acceptance of Pro in P1, but not in P2, hints toward a kinked substrate binding mode preceding the scissile peptide bond. In phage display, aromatic side chains only appeared four times (twice in P2 and once in P1Ј and P4Ј) but never in the remaining positions. Instead, small or uncharged residues were preferred in P3, P2, P1Ј, and P2Ј (65, 55, 60, and 70% of the recovered peptides, respectively) followed by hydrophobic residues. Interestingly, P1-Arg and P1Ј-Ser surrounded the scissile bond in one third of all cases. These findings agree well with the reactive center loop sequence of PCI (P4-FTFR2SAR-P3Ј), a highly potent KLK2 inhibitor (k ass ϭ 2 ϫ 10 5 M Ϫ1 s Ϫ1 ) (13). PCI contains P1-Arg and aromatic side chains in P2 and P4 as well as the Arg2Ser scissile bond. However, KLK2 also forms complexes with serpins whose reactive center loop sequences do not reflect the results from specificity profiling; reactivity toward plasminogen activator inhibitor 1 (P4-VSAR2MAP-P3Ј) (20), protease inhibitor 6 (P4-MMMR2CAR-P3Ј) (97), and even ␣ 1 -antichymotrypsin (P4-ITLL2SAL-P3Ј) (98) demonstrates that substrate recognition depends on exosite properties aside from the active site recognition sequence.
So far, three Zn 2ϩ inhibition mechanisms have been described for KLKs. First, Zn 2ϩ binding to Glu-77 and His-25 of KLK4 disturbs the Ile-16 -Asp-194 salt bridge, which destabilizes the active site (46). Second, Zn 2ϩ binding to His-91, His-101, and His-233 of KLK3 has been proposed to pull Asp-102 away from the catalytic triad (99). Such a binding mode has been confirmed in the equine KLK3 homologue (56). Third, Zn 2ϩ binding to residues in the 99-loop (e.g. His-97 and His-99 in rKlk1c2, His-96 and His-99 in KLK5, Thr-96-O and His-99 in KLK7), and to the catalytic His-57 destroys the active site (47,51,55).
Essentially unaltered Zn 2ϩ inhibition of the KLK2 mutants H25A, H91A, and H101A (Table 2) allowed us to exclude a KLK4-or KLK3-like Zn 2ϩ inhibition mechanism for KLK2. Instead, we believe that Zn 2ϩ binds to His-57 and at least one residue in the C-terminal region of the 99-loop. This region contains three potential Zn 2ϩ ligands (Asp-96, Glu-97, and Asp-98). Asp and Glu side chains represent 15 and 12% of all reported Zn 2ϩ ligands (His, 38%; Cys, 29%) (100). In the KLK2-PPACK structure with its open 99-loop conformation (type V), the side chains of Glu-97 and His-57 are separated by about 12 Å. However, a counterclockwise rotational movement of the 99-loop would yield a closed, eKLK3-loop conformation (type V*) and might also create a Zn 2ϩ binding site by moving Glu-97 closer to His-57 (Fig. 8A). Such a rearrangement would be based on the observed plasticity of the 99-loop, as corroborated by its highly flexible region that lacks electron density in both KLK2 structures. By contrast, only residues 95g and h are disordered in KLK1, which indicates that its 99-loop is relatively rigid and might explain why Zn 2ϩ ions do not inhibit KLK1. N-Glycosylation at both Asn-95 and Asn-95f and O-glycosylation at Ser-95b (101) may additionally decrease the flexibility of the KLK1 99-loop. Subtle variations in the K i value of KLK2 mutants ( Table 2) provide further evidence for a Zn 2ϩ -induced 99-loop shift. The H91A and K95eM mutations exhibit a decreased K i value that may correspond to an increase in 99-loop flexibility. Conversely, mutants with an increased K i value (K95eQ, H95fA) may contain a more rigid 99-loop. Notably, the K i value of the H25A mutant (residue 25 is far away from the 99-loop) is almost equal to the one of wild type KLK2.
If Zn 2ϩ binding alters the conformation of the KLK2 99-loop as described above, it may also promote further structural changes that inactivate the protein. Several KLK structures (KLK3 (PDB ID 3qum/chains P and Q), pro-KLK6, rKlk1c2, eKLK3, and mKlk1b4) have been proposed to represent such inactive conformations. When we compared KLK2 to eKLK3 (56), we observed several prominent changes. On the one hand, eKLK3 lacks an oxyanion hole due to a 170°change of the dihedral angle between Gly-193-N and -C␣ (KLK2, 100°; eKLK3, Ϫ70°). Similarly, the oxyanion hole is absent in pro-KLK6 and KLK3 (PDB ID 3qum/chains P and Q) as these structures miss the activating salt bridge between the N terminus and Asp-194 (25,50). On the other hand, three loops next to the active site cleft change their conformation, namely the 99-, 148-, and 220loop. The latter loop is notable for its role in the conformational selection mechanism exhibited by trypsin-like proteases; it exists in two different conformations that either occlude the active site (known as the E* form) or render it accessible to a substrate (the E form) (102,103). As we confirmed by visual inspection and calculation of UPGMA dendrograms based on C␣ r.m.s.d. (Fig. 9), not only the 220-loop but also the 148-loop assumes the E* form in all KLK structures that are presumably inactive. FIGURE 9. Open and closed 220-and 148-loop conformations. A and C, UPGMA dendrograms derived from pairwise 220-or 148-loop C␣ r.m.s.d., which were calculated after structural superposition as for the 99-loop with a PyMOL script (Fig. 6, supplemental File S2); PDB codes are given in brackets. B and D, alignments of all KLK 220-or 148-loops whose structures are known. According to these superpositions, both loops may assume open (green, semitransparent) or closed (red) conformations, which correspond to the E and E* form of the protease, respectively.
These observations suggest a compelling possibility for Zn 2ϩ ions to inactivate KLK2 by modulating its E-E* equilibrium. According to this model, structural changes of the 99-, 148-, and 220-loop accompany the E-E* transition ( Fig. 8B and supplemental Movie S3). An open (type V) 99-loop stabilizes the E form of the 220-loop. By contrast, Zn 2ϩ binding promotes the closed (type V*) 99-loop conformation, which favors the E* form of the 220-loop. Indeed, Arg-95g approaches Glu-218 in an inactive KLK3 structure (PDB ID 3qum/chain Q) (25). Together with the movement of the 148-loop, these relocations block the substrate binding site. Existence of a Zn 2ϩ -induced E* form that comprises a closed 99-loop conformation agrees with several experimental results that are otherwise hard to explain. First, occlusion of the active site cleft upon Zn 2ϩ binding accounts for the apparently competitive inhibition by Zn 2ϩ (Fig. 7E, F) even if the displacement of His-57 rather suggests a noncompetitive inhibition mechanism. Conformational selection implies that KLK2 has to assume its E form before it can bind to a substrate. This transition expels the Zn 2ϩ from the 99-loop and simultaneously restores the catalytic triad. Second, we may explain why Zn 2ϩ remarkably increases the affinity of KLK2 for benzamidine-coupled resin. Both Zn 2ϩ -free and Zn 2ϩ -inhibited KLK2 bind to this resin during chromatography. However, although benzamidine concentrations beyond 25 mM elute Zn 2ϩ -free KLK2, Zn 2ϩ -inhibited protein remains bound even in the presence of 100 mM benzamidine is due to a strongly reduced k off . By overlapping the substrate binding site, the 99-loop and the spacer arm that couples benzamidine to the resin completely block access to the S1 site for benzamidine in the elution buffer. Third, it becomes clear why soaking with Zn 2ϩ ions invariably destroyed KLK2 crystals grown in the absence of Zn 2ϩ ; the concomitant rearrangement of the 99-, 148-, and 220-loop severed the crystal contacts that involve these loops.

CONCLUSIONS
In summary, we were able to complement the structures of the classical KLKs 1 and 3 by solving the structure of KLK2. Although both KLK2 structures represent Zn 2ϩ -free forms of the protease, mutational analysis located a Zn 2ϩ binding site in the 99-loop. Intriguingly, this loop also contains an autolysis site, whose cleavage inactivated KLK2. Hence, our novel structural and enzymatic data are in line with earlier findings, which have established KLK2 as a key player of the semen liquefaction cascade. Accordingly, male fertility demands that at least four factors tightly regulate KLK2, namely its N-terminal propeptide, inhibition by Zn 2ϩ ions, permanent inactivation by cleavage of surface loops, and complex formation with PCI. Whereas the 99-loop limits KLK2 to certain substrates by accommodating their P2 and P4 residues, it may also confine proteolytic activity to a narrow post-ejaculatory time frame; this loop contains Zn 2ϩ binding residues for reversible inhibition and might even conduct the concerted movements of other surface loops (148-, 220-loop), which mark the transition between active and inactive states of trypsin-like proteases. Thus, we consider the 99-loop as master regulator of KLK2 activity and possibly in related KLKs.