Definition and Redesign of the Extended Substrate Specificity of Granzyme B*

Granzyme B is a protease involved in the induction of rapid target cell death by cytotoxic lymphocytes. Definition of the substrate specificity of granzyme B allows for the identification of in vivo substrates in this process. By using the combinatorial methods of synthetic substrate libraries and substrate-phage display, an optimal substrate for granzyme B that spans over six subsites was determined to be Ile-Glu-Xaa-(Asp↓Xaa)-Gly, with cleavage of the Asp↓Xaa peptide bond. Granzyme B proteolysis was shown to be highly dependent on the length and sequence of the substrate, supporting the role of granzyme B as a regulatory protease. Arginine 192 was identified as a determinant of P3-Glu and P1-Asp substrate specificity. Mutagenesis of arginine 192 to glutamate reversed the preference for negatively charged amino acids at P3 to positively charged amino acids. The preferred substrate sequence matches the activation sites of caspase 3 and caspase 7 and thus is consistent with the role of granzyme B in activation of these proteases during apoptosis. The caspase substrate poly(ADP)-ribose polymerase is cleaved by granzyme B in a cell-free assay at two sites that resemble the granzyme B specificity determined by the combinatorial methods. Many caspase substrates contain granzyme B cleavage sites and are proposed as potential granzyme B targets, suggesting a redundant function with certain caspases.

and natural killer cells, are one of the most potent host defenses against tumor-and virus-infected cells. Upon recognition by the CLs, apoptosis of the target cell is initiated by the granule-exocytosis mechanism and the Fas-FasL mechanism (1). Both pathways result in the activation of intracellular proteolysis that mediates apoptotic death of the target cell. The granule-exocytosis model proposes that granules are released from the CLs after recognition of the target cell. The major components of the granules are perforin, a putative pore-forming protein, and the granzymes, a subclass of serine proteases displaying the chymotrypsin fold (2). Perforin is believed to facilitate entry and localization of the granzymes within the target cell to effect death (3,4). Multiple granzymes have been identified and cloned from the granules of cytotoxic lymphocytes. Although granzymes are strongly implicated as key mediators of cell death (5,6), the mechanisms by which individual granzymes carry out their functions have not been fully elucidated. For example, it is unclear why multiple granzymes are used and if the granzymes mediate cell death through unrestricted proteolysis of the target cell or through a more selective cleavage of target cell proteins (7).
Recent studies have focused on granzyme B because of its unusual preference for cleaving after aspartate residues (8). Although granzyme B is the only known mammalian serine protease to have this P1-proteolytic 2 specificity, it is shared with the caspases, a family of cysteine proteases that are also activated during apoptosis (10). The link between granzyme B and the caspases has been strengthened by studies indicating that granzyme B can cleave and activate certain members of the caspases (11), and it has been suggested that this is one of the mechanisms by which granzyme B mediates apoptosis in vivo (4).
Participation of granzyme B in apoptosis may also involve a complementary activity with the caspases through the direct cleavage of proteins that are the proteolytic targets of caspases (12). Poly(ADP-ribose) polymerase (PARP) is a nuclear enzyme that is thought to function in numerous nuclear processes such as DNA repair and transcription (13). Cleavage of PARP by caspase 3 occurs rapidly upon induction of apoptosis (14). In vitro cleavage of PARP by granzyme B suggests a cleavage site distinct from that of caspase 3 (15).
To understand the mechanism by which granzyme B mediates apoptosis of target cells, we have undertaken the identification of its extended substrate specificity. Although individual granzymes have been isolated from different species, the possibility of contamination from homologous granzymes also present in the granules has hindered substrate specificity studies. We therefore developed a recombinant expression system in the yeast Pichia pastoris to produce reagent quantities of pure, catalytically active rat granzyme B. In this study, we have used two combinatorial methods to extend the definition of the substrate specificity of granzyme B to six subsites, from P4 to P2Ј. Individual amino acids responsible for determining the stringent substrate specificity of granzyme B were identified through the construction of a three-dimensional model of granzyme B complexed to substrate. Variant granzyme B enzymes with altered P1 and P3 substrate recognition properties were created to define the molecular determinants of specificity. The elucidated substrate specificity was shown to be relevant within a macromolecular context by locating cleavage sites in defined molecular targets.

EXPERIMENTAL PROCEDURES
Materials-Ac-IEPD-AMC, Ac-IKPD-AMC, Ac-IEPD-pNA, Ac-EPD-pNA, Ac-PD-pNA, Ac-D-pNA, Ac-IKPD-pNA, Ac-IEPD-SBzl, IKPD-SBzl, Ac-IEPDWGA-NH 2 , and Ac-IEPDWNA-NH 2 were purchased from SynPep (Dublin, CA). Suc-AAPX-pNA (X ϭ A, D, E, F, L, M, R) substrates and the Z-DEVD-FMK inhibitor were purchased from Bachem (Torrance, CA). Boc-AAD-SBzl was purchased from Enzyme Systems Products (Livermore, CA). N-Glycosidase F was purchased from Boehringer Mannheim. All DNA modifying enzymes were purchased from New England Biolabs (Beverly, MA) or Stratagene (La Jolla, CA) and were used according to manufacturer's guidelines. The coupled transcription-translation reticulocyte lysate system was purchased from Promega (Madison, WI) and used according to manufacturer's guidelines. Protein assay Bradford reagent was purchased from Bio-Rad and used according to manufacturer's guidelines. Substrates in the positional scanning synthetic combinatorial library (PS-SCL) were prepared as described previously (16). Oligonucleotides were synthesized with an Applied Biosystems 391 DNA synthesizer (Foster City, CA). The P. pastoris expression system was purchased from Invitrogen (San Diego, CA). Recombinant rat granzyme B antiserum was produced and purchased from Berkeley Antibody (Richmond, CA).
Heterologous Expression of Rat Granzyme B in Yeast-The XhoI and NotI DNA recognition sites were introduced by polymerase chain reaction onto the 5Ј and 3Ј ends, respectively, of the 681-base pair cDNA encoding mature rat granzyme B, amino acids 16 -245 (17). The resulting fragment was subcloned into the XhoI and NotI sites of the yeast vector, pPICz␣A (Invitrogen, San Diego, CA). This construct permitted the fusion of the mature granzyme B sequence to immediately follow the Kex2 signal cleavage site of the Saccharomyces cerevisiae ␣-factor secretion signal (18). The vector was linearized with SacI and transformed into the X33 strain of P. pastoris. Clones with the integrated rat granzyme B cDNA were selected for by resistance to Zeocin TM (19). A granzyme B-expressing clone was isolated and used for large scale protein expression in a 12-liter B. Braun Biostat-E fermentor (Allentown, PA). Yeast growth and protein expression were maintained at pH 6.0, 30°C, and dissolved oxygen Ն20%. Ten-milliliter aliquots were harvested at 24, 36, 48, 60, and 72 h after induction with methanol, and the cell density was determined by wet cell separation and weight determination. The total protein concentration in the culture supernatant was determined by Bradford analysis (20) using the Bio-Rad protein assay reagent (Livermore, CA) and bovine serum albumin as the protein standard. The concentration of granzyme B in the culture supernatant was determined at the time points by both active site titration and V max measurements of granzyme B.
Purification of Recombinant Granzyme B-After 48 h of induction with methanol, the supernatant from the granzyme B expressing culture was harvested. The supernatant was adjusted to 50 mM NaCl and loaded onto a 100-ml SP-Sepharose cation exchange column (Pharmacia Biotech, Uppsala Sweden). The column was washed with 5 column volumes of 50 mM MES, pH 6.0, and 50 mM NaCl and eluted with a linear salt gradient of 50 -1000 mM NaCl. Active granzyme B eluted at 600 mM NaCl and was approximately 10% pure based on SDS-PAGE visualization with Coomassie Brilliant Blue, Bradford analysis of total protein, and determination of granzyme B concentration by V max measurement. The fractions from the SP-Sepharose column were pooled and dialyzed against 50 mM MES, pH 6.0, 100 mM NaCl and loaded onto a 1-mL Mono-S cation exchange column (Pharmacia Biotech, Uppsala, Sweden). The Mono-S column was washed with 8 column volumes of buffer containing 50 mM MES, pH 6.0, 100 mM NaCl. The column was then treated with a salt gradient from 100 to 800 mM to elute active granzyme B at a salt concentration of 580 mM NaCl. The final product was Ն98% pure as judged by SDS-PAGE Coomassie Brilliant Blue staining as described by Shä gger and von Jagow (21), Bradford analysis of total protein, and V max measurement of granzyme B.
The concentration of granzyme B protein was determined by absorbance measured at 280 nm and based on an extinction coefficient of 13,000 M Ϫ1 cm Ϫ1 (22). The proportion of catalytically active protein was quantitated as follows: active sites of trypsin solutions were titrated with 4-methylumbelliferyl p-guanidinobenzoate (23). Assuming a 1:1 stoichiometry (24), ecotin solutions were then quantitated with active site-titrated trypsin. Concentration of granzyme B was then quantitated with the active site-titrated ecotin, again assuming a 1:1 stoichiometry, using Ac-IEPD-pNA as the substrate. The percentage of catalytically active protein was Ͼ95%.
To determine the glycosylation state of recombinant granzyme B, 20 g of granzyme B was denatured by boiling for 10 min in the presence of 0.5% SDS and 1% ␤-mercaptoethanol. The denatured granzyme B was separated into 2 aliquots, and 1 aliquot was incubated at 37°C for 3 h with 10 units of N-glycosidase F in 50 mM Tris, pH 7.5, and 1% Triton X-100. The products were analyzed by SDS-PAGE and Coomassie Brilliant Blue staining. Expression of Macromolecular Inhibitors of Granzyme B-Conditions of expression and purification of ecotin and ecotin IEPD were identical to those previously described (26). Site-directed mutagenesis was performed by the method of Kunkel (25) using the single-stranded template of the coding sequence of ecotin. The oligonucleotide used to incorporate the four-amino acid replacement of the ecotin active site loop, from VSTM to IEPD, was as follows (mismatched are underlined): (ecotin IEPD ) GTC AGT TCC CCG ATT GAA CCG GAT ATG GCA TGC CCG GAT GGC.
Positional Scanning Synthetic Combinatorial Library-Preparation and screening of the positional scanning synthetic combinatorial library (PS-SCL) was carried out as described previously (16,27). The concentration of substrates was 0.25 M, making the activity directly proportional to the specificity constant, k cat /K m . Enzyme activity of the PS-SCL was assayed in 100 mM HEPES, pH 7.5, 10 mM dithiothreitol at 25°C in a Tecan Fluostar (Research Triangle Park, NC) at excitation and emission wavelengths of 380 and 460 nm, respectively.
Single Substrate Kinetic Assays-Enzyme activity was monitored at 25°C in assay buffer containing 50 mM Tris, pH 7.4, and 100 mM NaCl. Substrate stock solutions were prepared in Me 2 SO. The final concentration of substrate ranged from 0.005 to 4 mM; the concentration of Me 2 SO in the assay was less than 5%. Enzyme concentrations ranged from 5 to 50 nM. Hydrolysis of pNA substrates was monitored spectrophotometrically at 410 nm on a UVIKON 860 spectrophotometer. Hydrolysis of SBzl substrates was monitored spectrophotometrically at 324 nm in the presence of 4,4Ј-dithiodipyridine. Hydrolysis of AMC substrates was monitored fluorometrically with an excitation wavelength of 380 nm and emission wavelength of 460 nm on a Fluoromax-2 spectrofluorimeter.
Hydrolysis of Ac-IEPDWGA-NH 2 and Ac-IEPDWNA-NH 2 by granzyme B was monitored at 220 nm by reverse phase HPLC using a C18 column (Vydac, 5 m, 4.6 ϫ 250 mm) with a 0 -70% gradient of 0.1% trifluoroacetic acid and 0.08% trifluoroacetic acid, 95% acetonitrile. Substrate concentrations ranged from 0.01 to 2 mM and enzyme concentrations from 9 to 90 nM. The enzymatic hydrolysis was quenched by the addition of trifluoroacetic acid to 0.4%. Extent of hydrolysis was determined from the calculated area of the product peak.
Creation of P3, P1Ј, P2Ј His-tagged Substrate Phage Library-The phagemid pHisX3P3 was constructed based on the phagemid pBSeco-gIII (26). The vector contains the following amino acid sequence inserted between the ecotin secretion signal and residues 198 -406 of pIII coat protein of M13 bacteriophage: AESVQPLGPGHHHHHHHGH-AGIXPDXXAGPGGG. The inserted sequence produces a histidine tag (underlined) linked to a substrate sequence (boldface) randomized in the P3, P1Ј, and P2Ј positions (boldface and italicized) followed by a GPGGG linker to pIII. The degenerate oligonucleotides synthesized to create the library consisted of the following sequence (where N indi-cates equimolar concentrations of G, C, A, and T; S indicates equimolar concentrations of G and C): CAT GGG CAT GCA GGA ATT NNS CCA GAC NNS NNS GCA GGG CCC GGA GGC GGT CCA TTC GTT. This oligonucleotide was used in combination with a primer that annealed to the HindIII sequence 3Ј to the pIII coding sequence for polymerase chain reaction. Digestion of the 680-base pair polymerase chain reaction product with SphI and HindIII was followed by ligation into the SphI/HindIII cut pHisX3P3 vector. The library of substrate phage has 32,768 possible DNA sequences that translates into 8000 possible protein sequences.
Phage particles expressing the engineered pIII-substrate fusion protein were prepared as described previously (26). Briefly, the library plasmid DNA was transformed into Epicurian Coli® XL2-Blue MRFЈ cells (Stratagene, La Jolla, CA). The transformation efficiency was determined by plating a portion of the transformed cells onto LB plates containing 60 g/ml ampicillin (LB-AMP). The transformation efficiency was 1 ϫ 10 6 individual clones, which allowed for Ͼ99.9% completeness of the library. The rest of the transformed cells were allowed to grow at 37°C with shaking in 2YT containing 60 g/ml ampicillin to an A 600 of 0.25 and were then infected with VCSM13 helper phage at a multiplicity of infection of 100 phage/cell for the production of recombinant phage. The culture was allowed to grow for 6 h. Phage particles were precipitated by addition of polyethylene glycol-8000 to 5% and NaCl to 500 mM. Phage were resuspended in 50 mM Tris, pH 7.4, 100 mM NaCl.
His-tagged Substrate Phage Cleavage Assay-Two hundred microliters of nickel(II)-nitrilotriacetic acid resin (Qiagen, Santa Clarita, CA) was washed with 10 ml of activity buffer (50 mM Tris, pH 7.4, 100 mM NaCl, 0.1% Tween 20). Phage particles (10 9 ) were added to the washed Ni(II) resin and allowed to bind with gentle agitation for 3 h. The Ni(II) resin was then washed with 40 ml of activity buffer to remove unbound phage. Recombinant granzyme B was added to a final concentration of 10 nM. After an incubation period of 6 h, the cleaved phage were separated from the resin in a total volume of 5 ml of buffer. The cleaved phage were amplified by the infection of male strain JM101 (FЈ) cells with the addition of VCSM13 helper phage to form recombinant phage which were then used for the next round of cleavage selection. After four rounds of cleavage selection, JM101 (FЈ) cells were infected with the eluted phage and plated onto LB-AMP. Twenty individual colonies were selected and grown in 3 ml of LB-AMP, and plasmid DNA was isolated and sequenced in the region of the cleavage site.
Molecular Modeling of Granzyme B-Substrate Complex-Rat mast cell protease-2 (RMCP-2) (28) was used as the framework to model rat granzyme B since they share 49% sequence identity. The coordinates for RMCP-2 were obtained from the Brookhaven Protein Data Bank (code 3RP2). The Biopolymer module of the Biosym (San Diego, CA) molecular modeling package was used to replace the amino acids of RMCP-2 with those of granzyme B. The structure was minimized with the Discover module of the Biosym molecular modeling package. The substrate structure is based on the trypsin-ecotin inhibitor complex (24) and docked onto the granzyme B structure.
Poly(ADP-ribose) Polymerase Mutagenesis and Cleavage Assay-The 3045-base pair cDNA for human PARP was cloned downstream of the T7 RNA polymerase promoter into the SacI and XbaI sites of pBluescript II KS(ϩ) vector (Stratagene). Site-directed mutagenesis was performed by the method of Kunkel (25) to introduce an alanine substitution at aspartate 536. The oligonucleotides used were as follows (mismatches are underlined): (D536A) GCA GCT GTG GAT CCT GCA TCC GGA CTG GAA CAC TC and (D644A) CCC CTG GAG ATT GCG TAC GGC CAG GAT GAA GAG. The coupled transcription-translation reticulocyte lysate system (Promega) was used to transcribe and translate the PARP, PARP (D536A), and PARP (644A) genes in the presence of [ 35 S] methionine. The translation products were then incubated with 100 nM granzyme B in the reticulocyte lysate at 25°C in 50 mM Tris, pH 7.4, and 100 mM NaCl.

Expression and Purification of Rat Granzyme B-
The active form of rat granzyme B was expressed and secreted from the methylotropic yeast P. pastoris following induction with methanol. Cleavage activity of a P1-Asp substrate was detected 24 h into induction and increased to a plateau at 48 h (Fig. 1A). A 150-fold purification of recombinant granzyme B was achieved by cation exchange chromatography using SP-Sepharose and Mono-S column resins (Fig. 1B, lanes 1 and 2, and Table I). The final yield of purified (Ͼ98%) active granzyme B was 0.7 mg/ liter of initial culture. Recently, a similar expression system for mouse granzyme B was designed utilizing his4 rather than zeocin selection (29).
Two species of granzyme B are detected by Coomassie Brilliant Blue staining (Fig. 1B). The apparent molecular mass of the major species is ϳ30 kDa, which exceeds that of 25 kDa based on the amino acid sequence of granzyme B. This difference between the actual and the expected molecular mass suggests glycosylation of granzyme B at the N-glycosylation consensus site, Asn-66. Treatment of recombinant granzyme B with N-glycosidase F resulted in a decrease of apparent molecular mass to that of the minor species, approximately 27 kDa (Fig. 1C, lanes 1 and 2). Removal of the N-linked glycosylation site by site-directed mutagenesis (N66Q) resulted in a catalytically active enzyme. However, the yield of secreted protein was decreased 100-fold (data not shown).
The direct detection of granzyme B hydrolytic activity in the culture supernatant indicates that the ␣-factor leader sequence is correctly processed immediately before Ile-16 of granzyme B. Like most enzymes of this class, granzyme B is synthesized as an inactive zymogen with an N-terminal propeptide. Another protease is required for cleavage of the propeptide for mature processing of granzyme B. Our expression system was designed  to bypass the addition of another protease to activate granzyme B thereby eliminating the potential of complicating subsequent substrate specificity studies. Primary (P1) Substrate Specificity-Purified recombinant granzyme B was tested for hydrolytic activity against a panel of tetrapeptide substrates of the form Suc-AAPX-pNA which contained various amino acids at the P1 position (P1-Ala, Glu, Phe, Leu, Met, or Arg). The only substrate with detectable activity was that with P1-Asp, with a k cat /K m of 62 M Ϫ1 s Ϫ1 . Cleavage of the thiobenzyl ester substrate, Boc-AAD-SBzl, by recombinant rat granzyme B demonstrates that the enzyme is kinetically indistinguishable from natively purified human granzyme B (k cat (recombinant) ϭ 11.4 s Ϫ1 , K m (recombinant) ϭ 140 M; k cat (native) ϭ 11 s Ϫ1 , K m (native) ϭ 145 M (13)).
Extended (P4 -P2) Substrate Specificity-A positional scanning combinatorial substrate library (PS-SCL) was used to elucidate the specificity of purified, recombinant, rat granzyme B. This library, of the general structure Ac-Xaa P4 -Xaa P3 -Xaa P2 -Asp-AMC, has been previously used to identify the amino acid preferences of the caspases and human granzyme B purified from cultured natural killer leukemia YT cells (27). The PS-SCL is composed of three libraries, each of which consist of 20 sublibraries, for a total of 8000 compounds. In each sublibrary, one position (P4, P3, or P2) contains a defined amino acid and the other two positions contain an equimolar mixture of amino acids (two unnatural amino acids, D-alanine (D-A) and nor-leucine (n), are included; cysteine and methionine are excluded). Thus analysis of the three libraries affords a complete understanding of enzyme preferences for amino acids at P4, P3, and P2. This approach has been previously validated as providing an accurate measure of protease specificity using caspase-1 (16).
By using this method we determined that the preferred tetrapeptide substrate recognition sequence for recombinant, rat granzyme B is (Ile Ͼ Val)(Glu Ͼ Gln ϭ n)Xaa-Asp ( Fig. 2A). Granzyme B prefers the ␤-branched aliphatic amino acids, isoleucine and valine, in the P4 position. Glutamate, glutamine, and norleucine are the preferred amino acids in the P3 position. The PS-SCL indicates that granzyme B can accept a broad range of amino acids in the P2 position, although proline is the preferred amino acid.
Kinetic Consequences of Extended (P4 -P2) Substrate Specificity-The positional scanning synthetic combinatorial library suggests that granzyme B exhibits unique and extended substrate specificity. To quantitate dependence on extended interactions and to explore the mechanism by which granzyme B utilizes extended binding energy to achieve catalysis, kinetic parameters were determined for various substrates (Table II). Whereas catalysis by the familiar pancreatic serine proteases is influenced by the length of the peptide substrate, the majority of substrate specificity is observed in P1. Granzyme B, on the other hand, is absolutely dependent on extended binding of substrate for efficient catalysis. Granzyme B is not capable of cleaving the dipeptide substrate, Ac-PD-pNA or the single residue substrate, Ac-D-pNA, at concentrations as high as 4 mM. Reducing the optimal tetrapeptide, Ac-IEPD-pNA, to the tripeptide, Ac-EPD-pNA, results in a Ͼ100-fold decrease in activity, from 666.0 ϫ 10 2 M Ϫ1 s Ϫ1 in k cat /K m to 6.5 ϫ 10 2 M Ϫ1 s Ϫ1 . Primary sequence recognition is also an important requirement for catalysis. Hydrolysis of the suboptimal P4-P3 substrate Ac-AAPD-pNA is disfavored by greater than 1000-fold as compared with the optimal substrate Ac-IEPD-pNA, decreasing from 666.0 ϫ 10 2 M Ϫ1 s Ϫ1 for the latter substrate to 0.62 ϫ 10 2 M Ϫ1 s Ϫ1 for the former. The suboptimal P3 substrate Ac-IKPD-pNA is disfavored by 10-fold when compared with the optimal substrate (Table II).
The dependence of granzyme B catalysis on recognition of a particular tetrapeptide sequence becomes more manifest when the scissile bond of the substrate contains a poorer leaving group (i.e. more peptide-like). The suboptimal P3 substrate Ac-IKPD-AMC results in a 140-fold decrease in catalysis, from 33.3 ϫ 10 2 M Ϫ1 s Ϫ1 for Ac-IEPD-AMC to 0.23 ϫ 10 2 M Ϫ1 s Ϫ1 for Ac-IKPD-AMC (Table II). Whereas under conditions of ester hydrolysis, Ac-IEPD-SBzl and Ac-IKPD-SBzl are almost indistinguishable, k cat /K m values of 3233 ϫ 10 2 to 2683 ϫ 10 2 M Ϫ1 s Ϫ1 , respectively (Table II).
The demonstration that granzyme B does not display substrate specificity in ester hydrolysis is consistent with what is known about other enzymes of this class (30). Serine proteases, including granzyme B, catalyze the cleavage of substrates through Mechanism 1 as follows.
Equations 1 and 2 used to define the steady-state macroscopic constants in terms of microscopic constants for the above mechanism are as follows. k cat ϭ k 3 k 5 k 3 ϩ k 5 (Eq. 1) Typically, the deacylation step is completely rate-limiting for ester hydrolysis by serine proteases (i.e. k 3 Ͼ Ͼ k 5 and therefore k cat (ester) ϭ k 5 ) (31). Extending this assumption to recombinant rat granzyme B, the data obtained from ester hydrolysis of Ac-IEPD-SBzl and Ac-IKPD-SBzl can be used to derive the mechanistic constants for the amide hydrolysis of Ac-IEPD-AMC and Ac-IKPD-AMC from Equations 3 and 4.
The observation that K d increases approximately 5-fold, from 167 M for Ac-IEPD-AMC to 805 M for Ac-IKPD-AMC (Table  III), demonstrates that granzyme B utilizes extended substrate-binding sites to enhance formation of the ground state Michaelis complex. However, while ground state stabilization is observed, an even greater stabilization occurs in the acylation transition state, k 3 decreases 24-fold from 0.558 s Ϫ1 for Ac-IEPD-AMC to 0.023 s Ϫ1 for Ac-IKPD-AMC (Table III).
Extended (P3, P1Ј, and P2Ј) Substrate Specificity Determined by Substrate Phage Display and Single Peptide Kinetics-To elucidate the granzyme B substrate preference C-terminal to the scissile bond (prime side), the method of monovalent "substrate phage" display was used (32,33). One million individual clones representing 32,768 nucleotide sequences and 8000 protein substrate sequences were displayed on phage particles between a histidine tag affinity anchor and the M13 phage coat protein, pIII. The histidine tag allows the "substrate" displaying phage to be immobilized on Ni(II) resin at an approximately Assuming that for the hydrolysis of the thiobenzyl ester substrates, Ac-IEPD-SBzl and Ac-IKPD-SBzl, the deacylation step is completely rate-determining, the microscopic constants for hydrolysis of the corresponding amide substrates can be determined (see "Results").  nanomolar concentration (Fig. 3A). Upon incubation with granzyme B, the phage that are displaying substrate sequences susceptible to cleavage by granzyme B lose their histidine tags; the phage displaying sequences refractory to cleavage remain bound to the Ni(II) resin (Fig. 3B). The eluted phage are subsequently used to infect E. coli cells (Fig. 3C) and are amplified by the addition of helper phage for another round of cleavage selection or are isolated for plasmid isolation and DNA sequencing (Fig. 3D). After four rounds of cleavage selection, 20 of the phagemid DNA plasmids were sequenced and analyzed. Translation of the resulting DNA sequences (Table IV) from the substrate phage confirms the PS-SCL result for preference of glutamate at the P3 position. Interestingly, methionine, which was not included in the positional scanning synthetic combinatorial library, is also represented at this position. While the amino acid preference of rat granzyme B at the P1Ј position is not strict, there is a distinct absence of charged amino acids.

Ac-IEPD-AMC
The results of the substrate phage library also indicate that rat granzyme B has specificity for glycine at the P2Ј position. The specificity for glycine at P2Ј was tested through cleavage of two P4 -P3Ј-extended peptides, one with a glycine and the other with an asparagine in the P2Ј position (Table II). The resulting kinetic data indicated that in the context of the peptides tested, P2Ј-glycine is 5-fold more efficiently cleaved than asparagine at the same position. Structural Determinants of Granzyme B Substrate Specificity-To understand the structural determinants of the extended substrate specificity of granzyme B, a homology-built threedimensional model of the enzyme-substrate complex was made (Fig. 4). This model can provide a structural framework to rationalize our experimental results as well as guide future studies. In this model a hydrophobic pocket is formed from Ile-99, Tyr-215, and Tyr-175 around the P4 position of the substrate. Acidic P3 and P1 specificity appears to be a result of the positive electrostatic surface formed by Arg-192 and Arg-226. Although the overall sequence identity between human and rat granzyme B is greater than 70%, there is a conservative amino acid change at position 192 from an arginine in the rat to a lysine in the human granzyme B. The subtle differences in P3 substrate specificity between the human (Glu Ͼ Ͼ Gly Ͼ Ser Ͼ Asp) and the rat (Glu Ͼ Ͼ Gln ϭ n Ͼ Ala Ͼ Ser Ͼ Asp) granzyme B may be dictated by the difference at position 192 (27). A hydrophobic pocket is formed by loop A which may allow for accommodation of large hydrophobic amino acids. This pocket is formed by the extended conformation of loop A, which contains a two-amino acid insertion when aligned with other chymotrypsin-like proteases. A tryptophan modeled at the P1Ј position of substrate can make favorable hydrophobic interactions with Ile-35, the Cys-42-Cys-58 disulfide bond, and the aliphatic portion of Lys-41. The specificity for glycine at P2Ј may be a result of glycine's unique absence of a side chain, allowing for the maximization of backbone hydrogen bonding between P2Ј-Gly and Lys-41, while not sterically clashing with the Arg-192 side chain.
To explore the mechanism of substrate discrimination by granzyme B, mutations of the arginine 192 side chain were   (Table V). The smaller differences between these two substrates for the (R192A) mutant versus the wild-type enzyme indicate that the decreased activity is not due solely to P3 interactions but may also be affecting P1 interactions. Indeed, the P4 -P2 substrate specificity profile for GrB(R192A), as screened by the PS-SCL, is identical to the wild-type enzyme, indicating that there are additional determinants for P3-Glu specificity (Fig. 2B). Support for the role of arginine 192 as a determinant for P3 specificity occurs when arginine 192 is mutated to glutamate. The specificity profile indicates that GrB(R192E) has decreased activity for acidic amino acids at P3, whereas the activity against basic amino acids has significantly increased (Fig. 2C). This mutant, GrB(R192E), retains only 0.3% of the wild-type activity for the optimal substrate, Ac-IEPD-AMC, while rescuing the majority, 71%, of the wild-type activity for the complementary, although non-optimal, substrate Ac-IKPD-AMC (Table V). Analysis of the mechanistic rate constants for the arginine 192 mutants provides insight into how substrate selectivity is attained by granzyme B. The wild-type enzyme is able to utilize extended P3 interactions to preferentially stabilize the acylation transition state versus the Michaelis complex ground state as indicated by the 5-fold increase in K d for Ac-IEPD-AMC versus Ac-IKPD-AMC and the 24-fold decrease in k 3 for these same substrates (Table III). For the optimal substrate, Ac-IEPD-AMC, there is a sequential decrease in both the affinity and acylation rate upon mutation of arginine 192, indicated by increased K d and decreased k 3 , respectively (Table III). For the less-optimal substrate, Ac-IKPD-AMC, there are smaller differences in K d upon mutation. Whereas approximately the same decrease in acylation exists between the two substrates for the (R192A) variant, acylation is completely rescued in the (R192E) variant for the Ac-IKPD-AMC substrate (Table III).
Cleavage of Poly(ADP-ribose) Polymerase by Granzyme B-Based on the current specificity studies with granzyme B, two potential granzyme B cleavage sites were identified in poly(ADP-ribose) polymerase (PARP), VDP(D2S)G at position 536 in PARP and LEI(D2Y)G at position 644. While neither of these sites is absolutely ideal for granzyme B, they contain a majority of the defined specificity determinants for granzyme B. Cleavage at Asp-536 would yield 59-and 54-kDa fragments (Fig. 5B). Cleavage at Asp-644 would yield 72-and 41-kDa fragments (Fig. 5B). To establish if these two sites were processed by granzyme B, wild-type PARP and the cleavage site mutants (D536A) and (D644A) were constructed and produced in rabbit reticulocyte lysate. Upon incubation of PARP with granzyme B in the lysate, bands at 89, 59, 54, 41, 35, and 24 kDa were observed (Fig. 5C, lane 2). The 89-and 24-kDa fragments were reminiscent of the cleavage of PARP by caspase-3 at the site DEVD2EV. Cleavage at this caspase site would also explain the appearance of the 35-kDa fragment (Fig.  5B). Additional experiments were undertaken to determine if the caspase-like cleavage of PARP was due directly to granzyme B or if the reticulocyte lysate contained a latent caspase that was activated upon incubation with granzyme B. We observed that in the presence of a macromolecular inhibitor that is specific for granzyme B, ecotin IEPD (K i ϭ 1 nM), 3 cleavage of PARP was abolished (Fig. 5C, lane 3). However, in the presence of the specific caspase-3 inhibitor, Z-DEVD-FMK, the predominant cleavage bands at 59 and 54 kDa were observed (Fig. 5C,  lane 4). Incubation of reticulocyte lysate alone with granzyme B followed by inhibition of the granzyme B with ecotin IEPD resulted in an activity in the lysate that was capable of cleaving PARP to the 89-and 24-kDa fragments (Fig. 5C, lane 5). Indeed, incubation of purified PARP with granzyme B resulted in the 59-and 54-kDa and, to a lesser extent, 41-kDa cleavage products (data not shown). Confirmation that VDPD2SG was the major cleavage site in PARP was obtained upon removal of the cleavage site through mutation of aspartate 536 to alanine to give the altered sequence VDPA2SG. Cleavage to the 59and 54-kDa fragments was not observed with PARP(D536A) in the reticulocyte lysate (Fig.5C, lanes 6 -10). However, 72-and 41-kDa fragments appeared, suggesting that the LEID2YG is a second cleavage site for granzyme B. Upon removal of the secondary cleavage site by mutation of the aspartate 644 to alanine to give the sequence LEIA2YG, the 41-kDa fragment was no longer observed (Fig. 5C, lanes 11-15). DISCUSSION Cytotoxic lymphocytes express multiple serine proteases that appear to be essential to targeted cell death administered through the granule exocytosis pathway. One of the most abundant serine proteases in the granules is granzyme B. Whereas cytotoxic lymphocytes derived from mice with a disruption in the granzyme B gene are impaired in inducing rapid target cell death (6), the mechanism of involvement for granzyme B in this process has not been fully elucidated. A current model is that granzyme B enters the cell and mediates proteolytic activation of the apoptotic class of caspases (4), although it may also act directly on cellular substrates. Recent studies have indicated that granzyme B is transported to the nucleus in the presence of perforin (3,34). This suggests that granzyme B may have nuclear substrate targets that it cleaves directly, rather than indirectly through caspase activation. Indeed, granzyme B has recently been shown to efficiently cleave several caspase substrates (12). Definition of the substrate specificity of granzyme B may help uncover the role it plays in CL-mediated cell death.
Multiple granzymes are stored in the granules of CLs and each may have an individual or redundant function. We have focused the present study on granzyme B since it may serve as an important structural model for the related granzymes. To avoid contamination by other granzymes, a recombinant expression system in yeast was developed. In addition to producing reagent quantities of granzyme B, the heterologous expression system has the advantage of making variant versions of the enzyme including the mature enzyme, circumventing the need for an additional activating protease. Expression in yeast also allows for correct post-translational modification, such as disulfide bond formation and glycosylation. We believe that glycosylation may aid in stabilizing or solubilizing granzyme B, and this may explain why our attempts at expression in prokaryotes failed (data not shown).
Like native granzyme B, the recombinant protease is rigorously selective for cleavage after aspartate residues. Previously reported studies with ester substrates indicated that granzyme B has the ability to cleave after Asn, Met, and Ser (29,35). Although we have observed the esterolytic activity of granzyme B on these substrates, cleavage of cognate amide substrates was not seen. These observations are a direct result of the catalytic mechanism of granzyme B. Under conditions of ester hydrolysis, deacylation of the acyl-enzyme intermediate is ratelimiting (i.e. k cat ϭ k 5 ). Since the enzyme does not distinguish between multiple ester substrates, substrate specificity is not manifested in deacylation. Granzyme B is an efficient, but not very specific, esterase. Presumably, once the acyl-enzyme intermediate is formed, the covalent nature of this interaction may be sufficient for correct positioning of the substrate in relation to the catalytic groups for the deacylation reaction to occur.
In contrast to ester hydrolysis, amide hydrolysis by granzyme B is very specific and dependent on extended enzymesubstrate binding interactions. For amide hydrolysis, granzyme B activity increases with the ability of the leaving group to delocalize the developing negative charge in the acylation transition state (i.e. pNA Ͼ AMC Ͼ peptide). This trend in leaving group dependence is consistent with acyl-enzyme formation being rate-determining for amide hydrolysis. Using a synthetic combinatorial approach in which the hydrolysis of a fluorogenic amide was monitored, the extended non-prime specificity of rat granzyme B was defined from P4 to P2. The optimal P4 -P1 substrate sequence for granzyme B was determined to be Ile-Glu-Pro-Asp. Enzymological characterization of FIG. 5. Biological significance of granzyme B P4-P2 substrate specificity. A, the optimal substrate specificity as determined by this study is shown. B, poly(ADP-ribose) polymerase (PARP) cleavage patterns resulting in caspase and granzyme B incubation (see "Results"). C, cleavage of wild-type (WT) PARP, PARP (D536A), and PARP (D644A) by granzyme B in rabbit reticulocyte lysate. Ecotin IEPD , a specific granzyme B inhibitor, and Z-DEVD-FMK, a specific caspase inhibitor, was used to differentiate the various protease activities in the lysate (see "Results"). D, the activation cleavage sequence of caspases 1-10 is shown. The correspondence of the activation sequence of several of the caspases to the optimal substrate specificity of granzyme B provides evidence that caspase activation may be a crucial role for granzyme B. extended substrates showed that granzyme B is critically dependent on favorable extended interactions and is not capable of hydrolyzing amide substrates less than three amino acids in length. Truncating the substrate from a tetrapeptide to a tripeptide resulted in a dramatic 100-fold decrease in catalysis. One possible explanation for these data is that extensive substrate interactions are necessary to stabilize the enzyme active site in a catalytically productive orientation.
Specificity of an enzyme for a substrate is dependent on the ability of an enzyme to utilize substrate binding energy to promote catalysis. Implicit in this definition is the recognition that specificity requires the stabilization of the rate-determining transition state complex. In the case of the pancreatic serine proteases, the importance of maximizing k cat /K m is observed by the preferential stabilization of the transition state over stabilization of the ground state Michaelis complex. Loss of catalytic efficiency through ground state stabilization is not desirable in proteases whose main biological function is digestion. In the case of the granzymes, whose biological role may be more regulatory in nature, the pressure to maximize catalytic rate may be secondary to the pressure to maximize control. Therefore, stabilization of the transition state would still be necessary for the manifestation of specificity, but binding energy could also be used to stabilize ground state interactions. The large increases in K m (and K d where determined) for suboptimal substrates indicate instability of the ground state Michaelis complexes. In addition to the decreased ground state stabilization, transition state stabilization is significantly decreased for the amide hydrolysis of suboptimal substrates as observed by k cat (and k 3 where determined).
To examine the substrate specificity on the prime side of the substrate (C-terminal to the scissile bond) and to verify the P3 specificity results from the synthetic library in a protein context, the combinatorial method of substrate phage was used. This method has been used successfully by other groups to determine the substrate specificities of multiple proteases (33,36,37). Strict selectivity by granzyme B was not observed in the P1Ј position, but there is a general preference for large hydrophobic amino acids at this position, demonstrated by the presence of Trp, Leu, Phe, and Ile. There also appears to be a general aversion toward charged residues at the P1Ј position as demonstrated by the absence of Arg, Lys, Asp, and Glu. The presence of serine at this position may be reflective of the broad P1Ј specificity, coupled with the increased occurrence of serine in the initial library (serine is represented by 3 of the 32 possible codons at this position). Another explanation is that rat granzyme B has multiple binding modes and can accommodate both serine and tryptophan at the S1Ј subsite. In contrast to the broad specificity at the P1Ј position, results from the substrate-phage cleavage assay reflect a strong preference for glycine at the P2Ј position. We have shown that enzyme-substrate interactions C-terminal to the scissile bond are catalytically significant and play a role in determining substrate specificity. The presence of prime side interactions are consistent with the mechanism we have described. If deacylation determined specificity, non-primed interactions would be of little consequence.
An advantage of the substrate-phage assay is that cleavage of the substrate occurs in a protein context. This allows us to evaluate how the results obtained from cleavage of the tetrapeptide substrates in the positional scanning combinatorial library reflect cleavage of protein substrates. The agreement between the two libraries for the preferential cleavage of substrates that contain glutamate at the P3 position supports the synthetic approach in elucidating protein substrate specificity. The correspondence of P3-Glu also confirms that granzyme B is predominantly cleaving the substrate-phage within the designed substrate sequence and not elsewhere in the pIII-His fusion protein. The appearance of methionine at the P3 position in the substrate phage library indicates that rat granzyme B has the ability to cleave substrates with P3-Met. Although methionine and cysteine are excluded from the PS-SCL for synthetic reasons, granzyme B has significant activity against norleucine, an unnatural amino acid that is approximately isosteric with methionine. Hence, these combinatorial approaches complement each other in the determination of substrate specificity.
The homology-built three-dimensional molecular model of granzyme B bound to a hexapeptide substrate with the optimal sequence provides additional insight into the structural determinants of specificity. One of the most striking structural differences between granzyme B and the pancreatic serine proteases is the lack of a disulfide bond at Cys-191-Cys-220. This may result in granzyme B having a less rigid binding site and therefore requiring multiple substrate-enzyme interactions to overcome this entropic factor for efficient catalysis. In addition, the loss of the disulfide bond may allow the residue at position 192 to have more interactions with substrate. We postulate that arginine 192 plays a synergistic role with arginine 226 in determining the P1-aspartate and P3-glutamate substrate specificity. Qualitative evidence of this model was observed when arginine 226 was replaced with glycine, which resulted in decreased hydrolysis of a P1-Asp thiobenzyl ester substrate in the crude cell lysates in which the enzyme was expressed (38). We have demonstrated and quantitated the interaction of arginine 192 in determining the P3 and P1 specificity for amide substrates. Arginine 192 appears to be important in translating the extended substrate binding interactions into specific catalysis. We hypothesize that this position in the enzyme is important for determining the substrate specificity of the other granzymes. Indeed, granzyme C contains a glutamate at position 192. Loops A and B are capable of interacting directly with substrate and are likely to provide substrate specificity with residues C-terminal to the scissile bond. Recently, it has been shown that these loops act together to determine the S1Ј specificity in trypsin (39).
The recognition that granzyme B is a protease with extended specificity allows speculation on its biological role during cytotoxic lymphocyte-mediated cell death. The amino acid preference of granzyme B has been defined for six subsites, P4 to P2Ј, as (Ile Ͼ Val)(Glu Ͼ Gln ϭ Met)Xaa-Asp2Xaa-Gly (Fig. 5A). In vitro, the substrate specificity of granzyme B can be kinetically defined; in vivo, other factors such as substrate availability, cellular localization, and potential cofactors must be considered. However, there is a functional relationship between the preferential substrate sequence of granzyme B and the activation site of members of the caspases (Fig. 5D). Indeed, studies have shown that granzyme B cleaves and activates several caspases involved in apoptosis (11, 40 -44). However, granzyme B does not activate caspase 1 (ICE) or caspase 4 (45), both caspases that are involved in inflammation (46). Although these studies demonstrate that granzyme B has the ability to cleave and activate multiple caspases, the kinetic efficiencies have not yet been determined. Our data on the substrate specificity of granzyme B suggest that caspase 3 and caspase 7 are preferentially activated during apoptosis.
Knowledge of the extended substrate specificity of granzyme B allows for the proposal of additional targets of granzyme B during apoptosis. The substrate specificity of caspase 6 matches that of granzyme B (27), suggesting that both enzymes cleave the same substrates. Several proteins known to be cleaved during apoptosis, such as nuclear lamin A (40) and nuclear poly (ADP)-ribose polymerase (14), contain the potential granzyme B cleavage sites VEID2NG and VDPD2SG, respectively (Fig. 5B). In support of this, PARP has been shown to be cleaved directly by granzyme B and indirectly through the granzyme B activation of caspases. Mutation of the major and minor granzyme B cleavage sites in PARP abolishes cleavage at those sites. This demonstrates that the kinetically determined specificity of granzyme B is relevant in the context of a protein. Furthermore, other proteins that contain the preferred substrate sequence, but have not yet been shown to be cleaved during cytotoxic lymphocyte-mediated cell death, may be targets for granzyme B.
In conclusion, we have demonstrated that granzyme B displays extended substrate specificity and that there is a significant dependence on these extended interactions for catalysis. Structural determinants of substrate specificity have been identified through construction of a structural model of granzyme B. This model has been tested through the redesign of the substrate specificity of the enzyme. Definition of the preferred substrate cleavage sequence has led us to propose a model in which granzyme B can activate members of the caspases as well as cleave other intracellular proteins. Although we have limited this study to granzyme B, we expect that other members of the granzymes will display extended substrate specificities. The identification of their specificity will further expand our knowledge of the role that granzymes play in cytotoxic lymphocyte-mediated cell death.