Structural Basis for Proteolytic Specificity of the Human -Inducing M Lianfeng Wu, Li Wang, Guoqiang Hua, Kan Liu, Xuan Yang, Yujia Zhai, Mark Bartlam, Fei Sun and Zusen Fan This information is current as of September 29, 2021. J Immunol 2009; 183:421-429; ; doi: 10.4049/jimmunol.0803088 http://www.jimmunol.org/content/183/1/421 Downloaded from

References This article cites 42 articles, 16 of which you can access for free at: http://www.jimmunol.org/content/183/1/421.full#ref-list-1

Why The JI? Submit online. http://www.jimmunol.org/

• Rapid Reviews! 30 days* from submission to initial decision

• No Triage! Every submission reviewed by practicing scientists

• Fast Publication! 4 weeks from acceptance to publication

*average by guest on September 29, 2021

Subscription Information about subscribing to The Journal of Immunology is online at: http://jimmunol.org/subscription Permissions Submit copyright permission requests at: http://www.aai.org/About/Publications/JI/copyright.html Email Alerts Receive free email-alerts when new articles cite this article. Sign up at: http://jimmunol.org/alerts

The Journal of Immunology is published twice each month by The American Association of Immunologists, Inc., 1451 Rockville Pike, Suite 650, Rockville, MD 20852 Copyright © 2009 by The American Association of Immunologists, Inc. All rights reserved. Print ISSN: 0022-1767 Online ISSN: 1550-6606. The Journal of Immunology

Structural Basis for Proteolytic Specificity of the Human Apoptosis-Inducing Granzyme M1

Lianfeng Wu,*† Li Wang,*† Guoqiang Hua,*† Kan Liu,*† Xuan Yang,*† Yujia Zhai,† Mark Bartlam,‡ Fei Sun,2,3† and Zusen Fan2,3*†

Granzyme M (GzmM), a unique protease constitutively expressed in NK cells, is important for granule-mediated cytolysis and can induce rapid caspase-dependent apoptosis of tumor cells. However, few substrates of GzmM have been reported to date, and the mechanism by which this recognizes and hydrolyzes substrates is unknown. To provide structural insights into the proteolytic specificity of human GzmM (hGzmM), crystal structures of wild-type hGzmM, the inactive D86N-GzmM mutant with bound peptide substrate, and the complexes with a catalytic product and with a tetrapeptide chloromethylketone inhibitor were solved to 1.96 Å, 2.30 Å, 2.17 Å and 2.70 Å, respectively. Structure-based mutagenesis revealed that the N terminus and of hGzmM are most essential for proteolytic function. In particular, D86N-GzmM was found to be an ideal inactive Downloaded from enzyme for functional studies. Structural comparisons indicated a large conformational change of the L3 loop upon substrate binding, and suggest this loop mediates the substrate specificity of hGzmM. Based on the complex structure of GzmM with its catalytic product, a tetrapeptide chloromethylketone inhibitor was designed and found to specifically block the catalytic activity of hGzmM. The Journal of Immunology, 2009, 183: 421–429.

4 ranzyme (Gzm) -induced cell death is a major pathway plays very important roles in granule-mediated cytolysis and can http://www.jimmunol.org/ used by CTL and NK cells to eliminate -infected or induce rapid cell death via an as yet undefined mechanism (13, 14). G transformed tumor cells (1, 2). Gzms are normally ex- Several substrates have been identified for GzmM so far, including pressed in an inactive prostate called Pro-Gzm, and the N terminus the GzmB PI9 (15), the inhibitor of caspase-activated of Pro-Gzm is subsequently cleaved to release the active form with DNase (14), the antagonist TRAP1 (16), the constitutive N-terminal sequence IIGG. Five types of human the component of cytoskeleton ␣-tubulin (17), and the abundant Gzms (A, B, H, K, and M) have been identified to date. GzmA and nucleolar phosphoprotein nucleophosmin (18). B are the most abundant Gzms in CTLs and lymphokine-activated Human Gzms share high and similar struc- killer cells and have been extensively studied (3–7). tures. However, positional screening techniques determined that they GzmM, a -like , preferentially cleaves possess distinctive substrate specificities (8). High-resolution struc- by guest on September 29, 2021 its substrate after Met or Leu (8, 9). Human Gzm (hGzm)M is en- tures of the Gzms are indispensable for further investigation of their coded in a distinctive cluster on 19 and colocalizes specific substrate binding sites. To date, the three-dimensional struc- with a family of (10). GzmM is constitutively tures of human GzmA, B, and Pro-GzmK are known and have re- and highly expressed in activated NK cells, but is never detected vealed the structural basis for their substrate recognition (19–22), yet in CD4ϩ or CD8ϩ T cells even after activation (11, 12). GzmM no structures are available for hGzmM and hGzmH. To elucidate the substrate-binding specificity and catalytic mech- anisms of GzmM, we have determined the crystal structures of wild- *Center for Infection and Immunity, and †National Laboratory of Biomacromol- type hGzmM, the inactive D86N-GzmM mutant bound with a peptide ecules, Institute of Biophysics, Chinese Academy of Sciences, Beijing, China; and substrate (D86N-Sub), hGzmM in complex with a catalytic product ‡ College of Life Sciences, Nankai University, Tianjin, China (aM-Prod), and hGzmM in complex with a tetrapeptide CMK inhib- Received for publication September 17, 2008. Accepted for publication April itor (aM-Inhibitor) to 1.96 Å, 2.30 Å, 2.17 Å, and 2.70Å, respectively. 25, 2009. Based on our structural analysis, we generated a series of mutants to The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance characterize the determinants of hGzmM enzymatic activity and with 18 U.S.C. Section 1734 solely to indicate this fact. found that Asp86 and His41 in the catalytic triad contribute more to 1 This work was supported by the National Science Foundation of China (30525005, proteolytic activity than the attack residue Ser182. We also found the 30830030, 30623005, and 30772496), 863 program (2006AA02Z4C9, 2006AA02Z173), D86N-GzmM mutant is an ideal catalytically inactive (dead) enzyme 973 programs (2006CB504303, 2006CB806506 and 2006CB910901), the Chinese Acad- emy of Sciences (KSCX2-YW-R-42 and the Hundred Talents Program), and the support for functional studies. From structural comparisons, we observed a of K.C. Wong Education Foundation (Hong Kong) to Z.F. large conformational change of the L3 loop upon substrate binding 2 These authors contributed equally to this work. and found this loop could endow the substrate with recognition and 3 Address correspondence and reprint requests to: Dr. Fei Sun or Dr. Zuse Fan, In- specificity. Based on the complex structure with a catalytic product, stitute of Biophysics, Chinese Academy of Sciences, 15 Datun Road, Beijing 100101, we designed a tetrapeptide CMK inhibitor and found it can specifi- China. E-mail address: [email protected] or [email protected] cally block the catalytic activity of hGzmM. 4 Abbreviations used in this paper: Gzm, granzyme; aM, active hGzmM; aM-Inhib- itor, active GzmM bound to its inhibitor; aM-Prod, active GzmM in complex with its product; CMK, chloromethylketone; CPEP, cleaved peptide; ⌬II-Gzm, truncated hGzmM; G3P-GzmM, Gly3 mutated to Pro; hGzm, human Gzm; I:E, molar inhibitor Materials and Methods to enzyme; OPEP, octa-peptide; P, substrate ; PI, propidium iodide; S, Plasmid construction for hGzmM and its variants substrate . The cDNA fragments of Pro-hGzmM, active hGzmM, and the truncated Copyright © 2009 by The American Association of Immunologists, Inc. 0022-1767/09/$2.00 hGzmM (⌬II-GzmM) were amplified from the full length cDNA of www.jimmunol.org/cgi/doi/10.4049/jimmunol.0803088 422 CRYSTAL STRUCTURE OF HUMAN GRANZYME M

Table I. Data collection, phasing, and refinement statisticsa

aM D86N-Subb aM-Prod aM-Inhibitor

Data collection

Space group P3121 P3121 P3121 P3121 Unit cell: a, b, c (Å), ␣, ␤, ␥ (°) 74.4, 74.4, 113.2, 74.5, 74.5, 112.7, 74.6, 74.6, 112.9, 74.7, 74.7, 113.6, 90, 90, 120 90, 90, 120 90, 90, 120 90, 90, 120 Molecule in ASU 1111 Resolution (Å)c 50-1.96 (2.01-1.96) 50-2.30 (2.36-2.30) 50-2.17 (2.25-2.17) 57-2.70 (2.77-2.70) Completeness (%)c 98.5 (92.1) 99.4 (99.4) 98.9 (98.2) 99.8 (99.7) c,d Rmerge (%) 6.5 (34) 5.1 (39.9) 4.9 (44.8) 9.4 (40.8) I/␴ϽIϾb 28.0 (2.1) 36.1 (3.4) 20.5 (2.0) 7.3 (1.9) Refinement statistics d Rwork/Rfree (%) 20.9/25.7 24.0/28.8 21.2/25.0 22.3/29.3 r.m.s.d. Bond length (Å) 0.023 0.019 0.017 0.016 Bond angle (°) 1.726 1.859 1.641 1.880 Ramachandran plot (%) Most favored (%) 86.5 84.1 85.1 80.8 Allowed (%) 13.0 15.9 14.4 18.7 Generous allowed (%) 0.5 0 0.5 0.5 Downloaded from a Corresponding parameters for the highest-resolution shell are shown in parentheses. b D86N-Sub, D86N-GzmM bound to substrate (OPEP). c ϭ⌺⌺ ͉ ϪϽ Ͼ͉ ⌺ ⌺ Ͻ Ͼ Ͻ Ͼ Rmerge h i Iih Ih / h i Ih , where Ih is the mean of the observations Iih of reflection h. d ϭ⌺࿣ ͉ Ϫ ͉ ࿣ ⌺͉ ͉ ϭ Rwork ( Fp(obs) Fp(calc) )/ Fp(obs) ; Rfree R factor for a selected subset (5%) of the reflections that was not included in prior refinement calculations.

hGzmM (RZPD German Resource Center for Genome Research) by stan- ␭ ϭ 1.5418 Å) at 93 K. With the exception of diffraction data for the http://www.jimmunol.org/ dard PCR cloning strategies. The other mutations were generated by site- aM-Inhibitor complex, which were processed by Mosflm 7.0.3 (24), all directed mutagenesis using the Phusion DNA polymerase kit (New En- other diffraction data were processed by HKL2000 (25). Data processing gland Biolabs). The enterokinase cleavage sequence (DDDDK) was statistics are summarized in Table I. introduced directly before the N terminus of each target and the exact N terminus (IIGG) was released after enterokinase cleavage (the Structure determination and refinement construction strategies are shown in Fig. 2B). All segments were subcloned The molecular replacement method was used to determine the structure of into the pET-26b vector (Novagen). active hGzmM. An initial search model was generated from the coordinates Expression and purification of recombinant of human complement ( entry 1DST) with the program Modeler7 (26). Phaser (27) was used to find the unique top so- All GzmM variant proteins were expressed in Escherichia coli and re- lution in the rotation and translation function and to calculate the optimal by guest on September 29, 2021 folded from inclusion bodies according to a previous protocol (21). The phases. ARP/wARP (28) was then used for automatic model building with Rosetta (DE3) cell strain was used to express GzmM and its variants. The 99% completeness. COOT (29) was used to refine the model manually, harvested pellets were lysed in a lysis buffer (50 mM Tris, 500 mM NaCl, combined with interactive restrained refinement by Refmac5.0 (30). The ϭ 0.25 mg/ml lysozyme, 10 ␮g/ml RNase A, 5 ␮g/ml DNase I, 2 mM MgCl , final model was refined to 1.96 Å resolution with Rwork 20.9% and 2 ϭ 0.1% Triton X-100, pH 7.9). The inclusion bodies were prepared and dis- Rfree 25.7%. All complex structures were solved by the molecular re- solved overnight in a buffer (6 M guanidinium chloride, 100 mM Tris-HCl, placement method using the active hGzmM structure and refined with the 20 mM EDTA, 150 mM GSH, and 15 mM GSSG, pH 8.0). They were translation/libration/screw motion (TLS) and restrained refinement method refolded in a refolding buffer (0.5 M Tris-HCl, 0.5 M L-, 20 mM using Refmac5.0. All final models were judged to have good stereochem- istry from Ramachandran plots calculated by PROCHECK (31). Structure CaCl2, 0.1 M NaCl, and 0.5 mM L-cysteine, pH 8.5) at 4°C according to the reported protocol. After adequate refolding, the proteins were dialyzed into refinement statistics are shown in Table I. All figures for surfaces, ribbons, the binding buffer (50 mM Tris-HCl, 500 mM NaCl, 5 mM imidazole 10% balls, and sticks were generated with PYMOL (http://pymol.sourseforge. glycerol, pH 8.0) and purified by Ni-NTA columns. Subsequently, the en- net) and BobScript 2.6b (32). riched proteins were treated by enterokinase (Novagen) in 50 mM Tris- assays HCl, 2 mM CaCl2, pH 8.0 at 18°C. Finally, the treated proteins were purified with a Resource S column (GE Healthcare) to exclude enteroki- The enzymatic activity of hGzmM was detected by cleavage of a synthetic nase. rTRAP1, rSET, and GST-Bid were expressed in E. coli and purified substrate, Suc-AAPL-pNA, labeled with fluorescence groups (pNA) at the as described previously (16, 23). C terminus, 0.5 ␮M active GzmM or its variants were incubated with the substrate (300 ␮M) at 37°C for 2 h and the fluorescence signal was mon- Crystallization and data collection itored at 405 nm with a Multilabel Counter (Wallac 1420 Victor; PerkinElmer). For the rTRAP1 cleavage assay, 1.5 ␮M rTRAP1 was in- The purified proteins were concentrated to 15 mg/ml in a buffer containing ␮ 20 mM Tris-HCl, pH 7.9 for crystallization screening experiments. Crys- cubated with 1.5- M active GzmM or its variants at 37°C for 2 h. The products were analyzed by SDS-PAGE. All data were measured from at tals of hGzmM were grown in 0.2 M Li2SO4, 0.1 M Tris, and 25% w/v PEG 3350, pH 8.5. The D86N-GzmM mutant enzyme was mixed with an least three separate experiments. octapeptide substrate (SSGKVPLS) in a molar ratio (1:5) in a buffer con- Loading assay taining 20 mM Tris-HCl, pH 8.0 and crystallized in the same condition as active hGzmM. The octapeptide substrate or inhibitor was soaked into the To further investigate the physiological relevance of hGzmM and its vari- active GzmM crystals to achieve the two complex crystals (aM-Prod and ants, they were loaded into Jurkat cells with an optimal dose of adenovirus aM-Inhibitor). All crystallization experiments were performed using the at 37°C for4haspreviously described (14). Briefly, Jurkat cells (2 ϫ 105) hanging drop vapor diffusion method at 16°C. Before data collection, crys- were washed three times in HBSS and resuspended in loading buffer

tals were soaked in cryo-protectant solution containing 0.1 M Bicine, 0.2 (HBSS with 0.5 mg of BSA per ml, 1 mM CaCl2, 1 mM MgCl2). The ␮ MLi2SO4, 30% PEG 3350, pH 8.0 for 30 s, followed by direct flash- resuspended cells were then treated with 1.5 M hGzmM or its variants cooling in a liquid nitrogen cryostream. Diffraction data for the active plus an optimal concentration of adenovirus at 37°C for 4 h. Treated cells hGzmM and aM-Inhibitor were collected on beam line 3W1A of the Bei- were double-stained with Annexin V-Fluos (recombinant human Annexin jing Synchrotron Radiation Facility using x-rays of wavelength 0.9794 Å V conjugated with FITC; Bender MedSystems) and propidium iodide (PI) at 95 K. Data for the D86N-Sub and aM-Prod complexes were collected in and followed by flow cytometry (FACSCalibur; BD Biosciences). The data house (Rigaku FR-E x-ray generator, R-Axis IVϩϩ image plate detector, were analyzed by CellQuest software. The Journal of Immunology 423

FIGURE 1. Stereo view of the ac- tive hGzmM structure. ␤-strands (␤1to ␤13) are labeled and presented in cyan for domain N and slate blue for domain C. ␣-helices (␣1to␣4) are labeled and colored in magenta. Three loops are de- picted as L1 (1–15, dark blue), L2 (99– 124, orange) and L3 (199–212, red). Four pairs of disulfide bonds (DS1, DS2, DS3, and DS4) are colored with carbon atoms in yellow and sulfur at- oms in green. The catalytic triad is dis- played as ball-and-stick and labeled. The N and C terminus are indicated with characters (N and C). In this text, if not declared, all oxygen atoms are col- ored by red and nitrogen atoms by blue.

Results by two loops, L1 (residues 1–15) and L2 (residues 99–124), and Downloaded from Overall structure of human GzmM the catalytic triad (Asp86, His41, and Ser182) is located in the cleft Recombinant hGzmM was expressed as inclusion bodies in E. coli, between domains akin to GzmA and B (19, 20, 22). Loop L2 joins but could be successfully refolded with strong enzymatic activity. the two domains and is usually referred to as the “autolysis loop” (21). Loop L1 effectively hooks domain N by inserting its con- Purified hGzmM was crystallized in the P3121 space group with one molecule per asymmetric unit and its crystal structure was served N-terminal tail into the hydrophobic pocket of domain C. solved to 1.96 Å by molecular replacement. The overall structure, The nonconserved loop L3 (residues 199–212), close to the cata- http://www.jimmunol.org/ containing four ␣-helices and 13 ␤-strands, is separated into two lytic triad, protrudes into the molecular surface and makes an im- domains. The first domain (domain N) is comprised of seven portant contribution to the specificity of substrate recognition, as ␤-strands (␤1to␤7) and three ␣-helices (␣1, ␣2, and ␣4). The described below. Four disulfide bonds, DS1 (C26–C42), DS2 second domain (domain C) contains six ␤-strands (␤8to␤13) and (C120–C188), DS3 (C151–C167), and DS4 (C178–C205), help to one ␣-helix (␣3) (Fig. 1). Domains N and C are largely connected maintain the overall stability of the structure. by guest on September 29, 2021

FIGURE 2. The N-terminal insertion is critical for catalytic activity of hGzmM. A, Ribbon plot and surface representations illustrating the structure of the N-terminal insertion. Hydrophobic, hydrophilic, negative, and positive surfaces are indicated by white, green, red, and blue, respectively. Four N-terminal residues are labeled and displayed as ball-and-stick. B, Scheme for the expression strategies of GzmM variants. All GzmM variants were initially expressed with enterokinase (EK) recognition sites, which were removed from their N terminus after purification with Ni-NTA column. Expression strategies for Pro-GzmM, active GzmM (aM), and I1,2L- and I1,2A-GzmM are demonstrated in the scheme. Moreover, other mutations involved in this ,in the plot. C, Substrate hydrolysis activity for hGzmM and its variants detected by the fluorescence peptide substrate. ProM ء study are indicated with Pro-GzmM; I1,2L, I1,2L-GzmM; I1,2A, I1,2A-GzmM; D181N, D181N-GzmM. D, Proteolytic activity assay for hGzmM and its variants by cleavage of rTRAP1. M, molecular marker. E, Location relationship between the N terminus and catalytic pocket of hGzmM. All related residues are labeled and shown in ball-and-stick representation, and their surrounding environment is displayed in ribbon representation. N-terminal residues are colored with green carbon atoms and catalytic residues with cyan carbon atoms. Hydrogen bonds are shown with gray dashed lines and the Ile1-Asp181 hydrogen bond with a blue dashed line. Those regions interacting with the N terminus are colored purple and the substrate binding sites are colored green. 424 CRYSTAL STRUCTURE OF HUMAN GRANZYME M

The N-terminal hook is important for GzmM stability and activity Active Gzm proteins, with a few exceptions, possess a highly conserved hydrophobic N-terminal IIGG sequence that turns in toward the interior of the structure after activation (19, 33, 34). In the case of hGzmM, the N-terminal tail twists at Gly3 and forms strong interactions with the surrounding pocket (Fig. 2A), including hydrophobic interactions among Ile1, Ile2, and the pocket; a hydrogen bond between the nitrogen of Ile1 and car- boxyl oxygen of Asp181; a hydrogen bond between the car- bonyl oxygen of Ile2 and the amide nitrogen of Ala176; and several bridges mediated by water molecules. The importance of the N-terminal conformation has previously been addressed in chymotrypsin (35), but more systematic data for hGzmM are provided here. Based on structural analysis, several variants were constructed with an enterokinase cleavage sequence to exclude redundant residues from the N terminus (Fig. 2B). The Gly3 mutated to Pro (G3P-GzmM) mutant and ⌬II-GzmM (Ile1 and Ile2 deleted), in which the N terminus insertion structure Downloaded from has been abolished, were highly unstable and quickly degraded. Compared with active wild-type hGzmM, the mutants I1,2L- GzmM (Ile1 and Ile2 mutated to Leu), I1,2A-GzmM (Ile1 and Ile2 mutated to Ala), and the proenzyme Pro-hGzmM (with an additional propeptide SSFGTQ before the active N terminus of hGzmM) showed greatly decreased activity when cleaving the http://www.jimmunol.org/ peptide substrate or the rTRAP1 protein substrate (Fig. 2, C and D). In particular, I1,2A-GzmM underwent a complete loss of catalytic activity. It is evident that mutation of Ile to Leu will negatively influence the hydrophobic interactions between the N terminus and its surrounding pocket, while mutation of Ile to Ala will greatly decrease this interaction. These data indicate that any modification to the N terminus of active hGzmM could abolish its stability or catalytic activity. by guest on September 29, 2021 FIGURE 3. Proteolytic determinants of the catalytic triad. A, Substrate hydrolysis activity for hGzmM and its variants detected by the fluorescence The catalytic triad and the catalytically inactive mutant peptide substrate. H41N, H41N-GzmM; D86N, D86N-GzmM; S182A, D86N-GzmM S182A-GzmM. B, Proteolytic activity for hGzmM and its variants by The catalytic triad of hGzmM is composed by Asp86, His41, and cleavage of rTRAP1. M, molecular marker. C, pH dependent substrate Ser182 (Fig. 2E), similar to many classical serine proteases. hydrolysis activity for hGzmM. D, Proteolytic activity for hGzmA and its Ser182, whose acidity is strengthened by Asp86 via His41 through variants by cleavage of rSET. aA, active GzmA; S195A, S195A-GzmA; hydrogen bonds, will attack and cleave the substrate directly. As in D102N, D102N-GzmA. previous reports, all three residues are clearly important for the proteolytic activity of hGzmM, which we further confirmed by site-directed mutagenesis. The H41N-GzmM, D86N-GzmM, and greatly strengthen the cleavage of the Ser182-OϪ scissile bond. S182A-GzmM mutants were expressed in E. coli and refolded Taken together, Asp86 is the most dominant residue in the cat- from inclusion bodies as previously described. From in vitro pro- alytic triad. Mutation of Asp86 to Asn86 will result in a loss of teolytic activity assays involving cleavage of the hGzmM peptide the catalytic driving power and completely abolish the enzy- substrate or the rTRAP1 protein substrate, S182A-GzmM ap- matic activity of hGzmM. peared to retain only 5–10% of the wild-type enzymatic activity, When performing functional studies of Gzm , re- while H41N-GzmM showed little activity and D86N-GzmM com- searchers usually create a catalytically inactive enzyme by mu- pletely lost its proteolytic activity (Fig. 3, A and B). Similar results tating the attacking Ser residue to Ala. However, it is difficult to were observed in the kinetic parameter measurement assays; the completely abolish the proteolytic activities for these mutants. activities of H41N- and D86N-GzmM could not be detected within For example, the S182A-GzmM mutant retained 5–10% enzy- the 40 min time limit, whereas S182A-GzmM exhibited little ac- matic activity and could still exert its roles in cleaving sub- tivity (data not shown). strates (Fig. 3, A and B) and inducing cell death (unpublished Proteolytic activity assays suggest that Asp86 makes a greater data). Similar results were also observed for GzmA (Fig. 3D) contribution to the enzymatic activity than does Ser182, indi- and reported for GzmH (36). cating that the acidity of Asp86 provides the major driving Despite being catalytically inactive, the structure of the D86N- power for catalysis. We further confirmed this by a pH-depen- GzmM mutant (data not shown) is identical with that of the wild- dent enzyme activity assay (Fig. 3C), from which the enzymatic type enzyme. This indicates that the D86N-GzmM mutant can be activity of hGzmM gradually increased as the pH was increased used as a catalytically dead enzyme for functional studies. To ver- from 5.0 to 9.0 and reached its maximum at pH 8.0. These ify whether the Asp mutation results in a catalytically inactive results indicate that Asp86 can readily lose its carboxyl proton enzyme in other Gzms, D102N-GzmA and S195A-GzmA mutants under basic conditions to promote formation of His41●Hϩ and were generated and their proteolytic activities were assayed as The Journal of Immunology 425 Downloaded from

FIGURE 4. Determinants of sub- strate recognition and specificity. Ste- reo views of D86N-Sub (A) and aM- http://www.jimmunol.org/ Prod (B). OPEP/CPEP (green carbon) and substrate binding sites (yellow carbon) are labeled and shown in ball- and-stick representation. 2Fo-Fc omit maps (0.6␴) are colored in gray. Schematic diagram for the interaction between D86N-GzmM and substrate (D86N-Sub) (C) or between active hGzmM and its product (aM-Prod) by guest on September 29, 2021 (D). Initial plots were generated by Ligplot v4.4.2 (43) and slightly mod- ified manually. W, water.

described (3). Compared with active GzmA in degrading rSET all (Fig. 3D). These results indicate that GzmM and GzmA mu- experiments, S195A-GzmA appeared to possess clear proteolytic tants, in which Asp is changed to Asn, are catalytically dead en- activity, whereas D102N-GzmA showed no enzymatic activity at zymes. This may also hold true for other Gzm enzymes. 426 CRYSTAL STRUCTURE OF HUMAN GRANZYME M

Determinants of substrate recognition and substrate specificity To investigate the detailed interaction between hGzmM and its substrate, the octa-peptide (OPEP) SSGKVPLS was synthesized according to previous work (15), and the complex structure (D86N-Sub) between D86N-GzmM and OPEP was solved to 2.30 Å. At the same time, OPEP was used to soak crystals of wild-type hGzmM. The resulting complex structure (aM-Prod) between wild-type hGzmM and the cleaved peptide (CPEP) SSGKVPL was solved to 2.17 Å. The bound OPEP and CPEP were clearly iden- tified and traced in 2Fo-Fc omit maps (Fig. 4, A and B). In the D86N-Sub structure, OPEP binds to D86N-GzmM without cleav- age. The electron density around Ser at the C terminus of OPEP is unclear (Fig. 4A), which might suggest terminal flexibility and weak binding affinity at this site. In the aM-Prod structure, how- ever, no electron density at this site can be observed, indicating the OPEP peptide has indeed been cleaved into CPEP (Fig. 4B). The substrate binding sites (S) and the corresponding substrate amino acids (P) for hGzmM could be easily defined from the D86N-Sub and aM-Prod structures. The S1 pocket is located in the Downloaded from catalytic site of GzmM and is composed of Pro177-Cys178, Gly180-Ser182, and Ser199. The S1 site determines the specificity for Leu in the P1 position of the substrate (Fig. 4). Due to the size and hydrophobic properties of the S1 pocket, it can only accom- modate long, narrow hydrophobic amino acids such as leucine, methionine, or norleucine, which is consistent with previous re- http://www.jimmunol.org/ ports (15, 37). The S2 pocket is formed by His41, Val80, Leu83, and Phe200, and interacts with the P2 residue of the substrate (Pro) FIGURE 5. Conformational change of GzmM upon substrate binding. via hydrophobic interactions (Fig. 4). Phe200 and Val83, forming A, Substrate binding site superpositions between active hGzmM (aM) and a small valley, limit the space of the S2 pocket and ensure that it D86N-Sub (left panel) or between D86N-Sub and aM-Prod (right panel). can only accommodate Pro or Ala in the P2 position, as reported aM is colored in orange, D86N-Sub in cyan and aM-Prod in purple. Res- previously (8, 15). The S3 pocket comprises of the loop Ser201– idues involved in substrate recognition sites and the catalytic triad are shown in ball-and-stick representation. For clarity, only the residues of Arg203 and solvent molecule W4 (Fig. 4). Ser201 and Ser202 active hGzmM are labeled. B, Representation of substrate binding pockets form hydrogen bonds with the main chain of the P3 residue (Val) of aM (left) and aM-Prod (right) by their electrostatic potential surfaces. by guest on September 29, 2021 either directly or via W4. The hydrophobic contribution by the side The position of Phe200 is indicated. Substrate binding pockets are marked chain of Arg203 is critical for the specificity of the S3 site. The S4 with S1-S4 respectively. pocket, defined mainly by the groove formed by Pro81, Ala82, Glu84, and Ser160, interacts with the P4 residue (Lys) via hydro- Phe200 changes in the opposite direction, like a lid, to expose the gen bonds (Fig. 4). In addition, Leu83 and Phe200 in the S2 pocket attacking Ser182 residue in the catalytic pocket (Fig. 5A, left and supply partial hydrophobic contributions to the interaction with the B). Phe200 is thus a pivotal residue to confer substrate specificity P4 residue. The calculated electrostatic potential of the S4 pocket of hGzmM, especially for the P2 site where the only preferred is predominantly negative due to the contribution of Glu84. As a residue is Pro or Ala (8, 15). With the exception of the conforma- result, only residues such as Lys, His, or Arg with a hydrophobic tional change upon substrate binding, comparing the structures of neck and basic head could be accommodated by the S4 pocket. D86N-Sub and aM-Prod revealed no obvious structural changes as Although the SЈ pocket could not be defined clearly in the D86N- a result of substrate cleavage (Fig. 5A, right). Sub or aM-Prod structures, Lys179 is reported to play an important role in substrate recognition (37). It could be involved in formation A specific inhibitor of GzmM of the catalytic pocket and might limit the binding specificity at PЈ sites. Based on structural analysis, the P1 to P4 sites were used to design a peptide inhibitor of hGzmM. The tetrapeptide inhibitor has the sequence KVPL and includes an acetylated N terminus Conformational change of the L3 loop upon substrate binding and chloromethylketone (CMK) covalently linked to the C ter- The D86N-Sub structure was superimposed onto the structure of minus. The CMK group was added to ensure the formation of hGzmM (Fig. 5A, left), from which a conformational change of the covalent bonds with both Ser182 and His41 when the inhibitor L3 loop was clearly observed. The most remarkable change occurs (Ac-KVPL-CMK) binds to the enzyme, thus irreversibly abol- in the region of the L3 loop from Phe200 to Arg203 (Fig. 5A, left), ishing the enzyme activity. The crystal structure of hGzmM while another region of the L3 loop is locked tightly by the disul- bound with this inhibitor (aM-Inhibitor) was solved to 2.7 Å fide bond DS4 (Figs. 1 and 5A). Although the overall root mean resolution. A 2Fo-Fc omit map clearly defined the bound in- square deviation between the D86N-Sub and hGzmM structures is hibitor in the catalytic site of the enzyme (Fig. 6A). Covalent only 0.35 Å for all C␣ atoms, the root mean square deviation for bonds formed between Leu-CMK and Ser182/His41 effectively the region of L3 that forms part of the S2 and S3 pockets is ϳ2.7 lock the inhibitor into the enzyme. Å. We propose that the movement of these residues should allow When assaying the proteolytic cleavage activity of hGzmM for the entrance of the optimal substrate into the catalytic site (Fig. a peptide substrate or the protein substrate rTRAP1, the inhibitor 5B). In the substrate-bound D86N-Sub structure, Ser201 switches was found to block the enzymatic activity of hGzmM in a dose- from the interior to the protein surface, while the phenyl ring of dependent manner (Fig. 6, B and C). Approximately 90% of the The Journal of Immunology 427 Downloaded from http://www.jimmunol.org/

FIGURE 6. An efficient and specific inhibitor for hGzmM. A, Inhibitor Ac-KVPL-CMK binds to hGzmM at the substrate binding pocket. The inhibitor is shown in ball-and-stick representation with carbon atoms in green; hGzmM is represented by electrostatic surface potential. The inhibitor is cross-linked to Ser182 and His41. 2Fo-Fc omit map (1.2␴) around the inhibitor is displayed as cyan mesh. Negative and positive potentials are colored in red and blue respectively. ACE, acetyl. B, The designed inhibitor Ac-KVPL-CMK can efficiently eliminate the hydrolytic activity of hGzmM by substrate hydrolysis by guest on September 29, 2021 activity assay. I, Inhibitor; E, enzyme. C, GzmM-mediated rTRAP1 cleavage was inhibited by Ac-KVPL-CMK. aM, active GzmM; I/E, inhibitor/enzyme. D, The inhibitor can block GzmM-induced cell death. Jurkat cells treated as described in Materials and Methods were stained with Annexin V and PI, then analyzed by flow cytometry using a FACSCalibur (BD Biosciences). Dead cells were calculated as double Annexin V and PI stained, and shown as mean Ϯ SD% of total analyzed cells. aM, active GzmM; Ad, adenovirus; I, inhibitor. E, The inhibitor cannot inhibit the substrate hydrolysis activity of hGzmA or hGzmB. aA, active hGzmA; aB, active hGzmB; M-I, the inhibitor. hGzmM enzymatic activity was lost when the molar inhibitor to sin might not be applicable to all serine proteases. In contrast, enzyme ratio (I:E) equalled 1:1. The enzymatic activity was com- the acidic Asp86 is indispensable for catalysis, which was fur- pletely lost at an I:E ratio of 4:1. The inhibitor can also abolish ther confirmed by pH-dependent enzyme activity assays. Be- hGzmM-induced target cell death when the I:E ratio is larger than sides the lack of catalytic activity, D86N-GzmM showed very 3:1 (Fig. 6D). To further confirm the specificity of this inhibitor good substrate binding affinity using a BiaCore surface plasmon among Gzms, we tested its inhibitory activity against GzmA or resonance assay (data not shown) and adopted the same con- GzmB. However, no obvious inhibitory effects were observed dur- formation as the wild-type enzyme. This implies that, for Gzm ing the GzmA-mediated rSET degradation or GzmB-induced enzymes, a mutant with Asp substituted to Asn would be an GST-Bid cleavage experiments at an I:E ratio of 4:1 (Fig. 6E)or ideal catalytically inactive enzyme that could be used as a neg- even up to 50:1 (data not shown). In summary, the CMK-linked ative control in functional studies. Alternatively, an Asp-to-Asn tetrapeptide (Ac-KVPL-CMK) is a specific and efficient inhibitor mutant could also be used to fish out physiological substrates by for human GzmM. affinity chromatography. In this study, Asp181 was mutated to Asn (D181N-GzmM) to Discussion weaken the hydrogen bond between Ile1 and Asp181, resulting From the structure of hGzmM, we observed strong interactions in dramatically decreased activity when cleaving the peptide between Asp86 and His41 via a hydrogen bond with a distance substrate or the protein substrate rTRAP1 (data not shown). of 2.79 Å, and between His41 and Ser182 with a distance of This suggests the Ile1-Asp181 hydrogen bond makes an impor- 2.70 Å. The role of Asp in catalytically active Gzm enzymes tant contribution to the catalytic activity of hGzmM. Any ad- was therefore re-investigated in this study. Interestingly, we justment to the N terminus of hGzmM, such as truncation of the found that a mutation from Asp86 to Asn completely abolished first two Iles (⌬II-GzmM), G3P-GzmM, mutation from Ile1,2 to enzyme activity, whereas a mutation from Ser182 to Ala still Leu or to Ala, and adding more residues (SSFGTQ) before the retained 5–10% of catalytic activity. Similar results were also N terminus, would decrease the catalytic activity of hGzmM. observed for GzmA and GzmH (unpublished data). These data Deletion of the first two residues (⌬II-GzmM) or mutation of suggest that classical knowledge derived from the study of tryp- G3P-GzmM yielded extremely unstable proteins, which might 428 CRYSTAL STRUCTURE OF HUMAN GRANZYME M result from the loss of the conformation of the N-terminal 5. Fan, Z., P. J. Beresford, D. Zhang, and J. Lieberman. 2002. HMG2 interacts with insertion. Taken together, these mutation studies suggest that the nucleosome assembly protein SET and is a target of the cytotoxic T-lym- phocyte protease granzyme A. Mol. Cell Biol. 22: 2810–2820. the N-terminal insertion not only aids the stability of the whole 6. Trapani, J. A., and V. R. Sutton. 2003. : pro-apoptotic, antiviral and structure, but also directly affects the catalytic activity of antitumor functions. Curr. Opin. Immunol. 15: 533–543. hGzmM. 7. Lieberman, J., and Z. Fan. 2003. Nuclear war: the granzyme A-bomb. Curr. Opin. Immunol. 15: 553–559. Phe200 in the L3 loop is proposed as a major factor in deter- 8. Mahrus, S., and C. S. Craik. 2005. Selective chemical functional probes of gran- mining the range of GzmM substrates. The flexibility of the L3 zymes A and B reveal granzyme B is a major effector of - loop also enables the possibility that hGzmM might have adaptable mediated lysis of target cells. Chem. Biol. 12: 567–577. 9. Smyth, M. J., T. Wiltrout, J. A. Trapani, K. S. Ottaway, R. Sowder, substrate specificity and allow the binding of different substrates. L. E. Henderson, C. M. Kam, J. C. Powers, H. A. Young, and T. J. Sayers. 1992. This could explain why the preference of the P1 site explored in Purification and cloning of a novel serine protease, RNK-Met-1, from the gran- this study is different to our previously reported cleavage site (Ser ules of a rat natural killer cell leukemia. J. Biol. Chem. 267: 24418–24425. 10. Pilat, D., T. Fink, B. Obermaier-Skrobanek, M. Zimmer, H. Wekerle, P. Lichter, in inhibitor of caspase-activated DNase) (14), and why the P2 site and D. E. Jenne. 1994. The human met-ase (GZMM): structure, sequence, is different from the recent report by Cullen et al. (Lys in nucleo- and close physical linkage to the serine protease gene cluster on 19p13.3. Genom- phosmin) (18). ics 24: 445–450. 11. Krenacs, L., M. J. Smyth, E. Bagdi, T. Krenacs, L. Kopper, T. Rudiger, A. Zettl, In this study, we provided one efficient and specific inhibitor H. K. Muller-Hermelink, E. S. Jaffe, and M. Raffeld. 2003. The serine protease for hGzmM and determined its bound structure. This inhibitor granzyme M is preferentially expressed in NK-cell, ␥␦ T-cell, and intestinal T-cell lymphomas: evidence of origin from lymphocytes involved in innate im- (Ac-KVPL-CMK) can completely block the catalytic activity of munity. Blood 101: 3590–3593. hGzmM with just a 3- or 4-fold excess in vitro or in cell-loaded 12. Sayers, T. J., A. D. Brooks, J. M. Ward, T. Hoshino, W. E. Bere, G. W. Wiegand, assays. Surprisingly, it did not suppress the activities of GzmA J. M. Kelly, and M. J. Smyth. 2001. The restricted expression of granzyme M in Downloaded from or GzmB, even with much more excessive doses. GzmM knock- human lymphocytes. J. Immunol. 166: 765–771. 13. Kelly, J. M., N. J. Waterhouse, E. Cretney, K. A. Browne, S. Ellis, J. A. Trapani, out mice demonstrated increased susceptibility to some viral and M. J. Smyth. 2004. Granzyme M mediates a novel form of -depen- infections that are reminiscent of those for GzmA- or GzmB- dent cell death. J. Biol. Chem. 279: 22236–22242. deficient mice (38–40), indicating the functional redundancy of 14. Lu, H., Q. Hou, T. Zhao, H. Zhang, Q. Zhang, L. Wu, and Z. Fan. 2006. Gran- zyme M directly cleaves inhibitor of caspase-activated DNase (CAD) to unleash the Gzm enzymes. Therefore, specific inhibitors are required for CAD leading to DNA fragmentation. J. Immunol. 177: 1171–1178. the functional investigation of a single member of the Gzm 15. Mahrus, S., W. Kisiel, and C. S. Craik. 2004. Granzyme M is a regulatory pro- http://www.jimmunol.org/ family. GzmM is constitutively and abundantly expressed in tease that inactivates proteinase inhibitor 9, an endogenous inhibitor of granzyme B. J. Biol. Chem. 279: 54275–54282. innate effector NK cells. Previous reports by Kelly et al. and Lu 16. Hua, G., Q. Zhang, and Z. Fan. 2007. Heat shock protein 75 (TRAP1) antago- et al. demonstrated that GzmM-induced rapid cell death is con- nizes reactive oxygen species generation and protects cells from granzyme M- sistent with the kinetics of cytolysis by NK cells (13, 14). NK mediated apoptosis. J. Biol. Chem. 282: 20553–20560. 17. Bovenschen, N., P. J. de Koning, R. Quadir, R. Broekhuizen, J. M. Damen, cells are effectors of innate immunity to trigger adaptive im- C. J. Froelich, M. Slijper, and J. A. Kummer. 2008. NK cell protease granzyme munity by secretion of IFN-␥, acting as early initiators of im- M targets ␣-tubulin and disorganizes the microtubule network. J. Immunol. 180: mune response (41). Granule-mediated cytotoxicity has been 8184–8191. 18. Cullen, S. P., I. S. Afonina, R. Donadini, A. U. Luthi, J. P. Medema, P. I. Bird, linked to allograft rejection and autoimmune diseases (42). Spe- and S. J. Martin. 2008. Nucleophosmin is cleaved and inactivated by the cyto- cific inhibitors to block NK cell-induced cytolysis may, therefore, toxic granule protease granzyme M during NK cell-mediated killing. J. Biol. by guest on September 29, 2021 be a potential approach for organ transplantation or autoimmune Chem. 284: 5137–5147. 19. Bell, J. K., D. H. Goetz, S. Mahrus, J. L. Harris, R. J. Fletterick, and C. S. Craik. diseases. The inhibitory efficiency of the specific inhibitor for 2003. The oligomeric structure of human granzyme A is a determinant of its GzmM in animal models requires further investigation. extended substrate specificity. Nat. Struct. Mol. Biol. 10: 527–534. 20. Hink-Schauer, C., E. Estebanez-Perpina, F. C. Kurschus, W. Bode, and Accession codes D. E. Jenne. 2003. Crystal structure of the apoptosis-inducing human granzyme A dimer. Nat. Struct. Mol. Biol. 10: 535–540. Coordinates of active hGzmM, aM-Prod, D86N-Sub, and aM-In- 21. Hink-Schauer, C., E. Estebanez-Perpina, E. Wilharm, P. Fuentes-Prior, W. Klinkert, W. Bode, and D. E. Jenne. 2002. The 2.2-A crystal structure of hibitor have been deposited in the Protein Data Bank (http:// human pro-granzyme K reveals a rigid with unusual features. J. Biol. www.rcsb.org) with accession codes 2ZGC, 2ZGH, 2ZGJ, and Chem. 277: 50923–50933. 2ZKS, respectively. 22. Waugh, S. M., J. L. Harris, R. Fletterick, and C. S. Craik. 2000. The structure of the pro-apoptotic protease granzyme B reveals the molecular determinants of its specificity. Nat. Struct. Mol. Biol. 7: 762–765. Acknowledgments 23. Zhao, T., H. Zhang, Y. Guo, Q. Zhang, G. Hua, H. Lu, Q. Hou, H. Liu, and We thank Dr. Yuhui Dong (Beijing Synchrotron Radiation Facility) for Z. Fan. 2006. Granzyme K cleaves the nucleosome assembly protein SET to induce single-stranded DNA nicks of target cells. Cell Death Differ. 14: help with data collection and processing; Dr. Hongxia Lu for helpful in- 489–499. struction and plasmid construction; Xia Xu and Xudong Zhao of the IBP 24. Powell, H. R. 1999. The Rossmann Fourier autoindexing algorithm in MOSFLM. core facilities centre for technical assistance; Yuanyuan Chen, Peng Xue, Acta. Crystallogr. D Biol. Crystallogr. 55: 1690–1695. and Chunchun Liu for their technical help; and Shuo Wang and Haidong 25. Otwinowski, Z., W. Minor, and Charles W. Carter, Jr. 1997. Processing of X-ray Tang for recombinant protein donations. diffraction data collected in oscillation mode. In Methods in Enzymology.C.W. Carter, Jr., and R. M. Sweet, eds. Academic Press. New York, pp. 307–326. 26. Marti-Renom, M. A., A. C. Stuart, A. Fiser, R. Sanchez, F. Melo, and A. Sali. Disclosures 2000. Comparative protein structure modeling of and genomes. Annu. Rev. Biophys. Biomol. Struct. 29: 291–325. The authors have no financial conflict of interest. 27. Read, R. J. 2001. Pushing the boundaries of molecular replacement with maxi- mum likelihood. Acta. Crystallogr. D Biol. Crystallogr. 57: 1373–1382. References 28. Lamzin, V. S., and K. S. Wilson. 1993. Automated refinement of protein models. 1. Chowdhury, D., and J. Lieberman. 2008. Death by a thousand cuts: granzyme Acta. Crystallogr. D Biol. Crystallogr. 49: 129–147. pathways of programmed cell death. Annu. Rev. Immunol. 26: 389–420. 29. Emsley, P., and K. Cowtan. 2004. Coot: model-building tools for molecular 2. Fan, Z. S., and Q. X. Zhang. 2005. Molecular mechanisms of lymphocyte-me- graphics. Acta. Crystallogr. D Biol. Crystallogr. 60: 2126–2132. diated cytotoxicity. Cell Mol. Immunol. 2: 259–264. 30. Murshudov, G. N., A. A. Vagin, and E. J. Dodson. 1997. Refinement of macro- 3. Fan, Z., P. J. Beresford, D. Zhang, Z. Xu, C. D. Novina, A. Yoshida, Y. Pommier, molecular structures by the maximum-likelihood method. Acta. Crystallogr. D and J. Lieberman. 2003. Cleaving the oxidative repair protein Ape1 enhances cell Biol. Crystallogr. 53: 240–255. death mediated by granzyme A. Nat. Immunol. 4: 145–153. 31. Laskowski, R. A., J. A. Rullmannn, M. W. MacArthur, R. Kaptein, and 4. Fan, Z., P. J. Beresford, D. Y. Oh, D. Zhang, and J. Lieberman. 2003. Tumor J. M. Thornton. 1996. AQUA and PROCHECK-NMR: programs for checking the suppressor NM23-H1 is a granzyme A-activated DNase during CTL-mediated quality of protein structures solved by NMR. J. Biomol. NMR 8: 477–486. apoptosis, and the nucleosome assembly protein SET is its inhibitor. Cell 112: 32. Esnouf, R. M. 1997. An extensively modified version of MolScript that includes 659–672. greatly enhanced coloring capabilities. J. Mol. Graph Model 15: 132–134. The Journal of Immunology 429

33. Birktoft, J. J., and D. M. Blow. 1972. Structure of crystalline ␣-chymotrypsin. V. Functional analysis of granzyme M and its role in immunity to infection. J. Im- The atomic structure of tosyl-␣-chymotrypsin at 2 A resolution. J. Mol. Biol. 68: munol. 175: 3235–3243. 187–240. 39. Smyth, M. J., S. E. Street, and J. A. Trapani. 2003. Cutting edge: A 34. Jing, H., Y. S. Babu, D. Moore, J. M. Kilpatrick, X. Y. Liu, J. E. Volanakis, and and B are not essential for perforin-mediated tumor rejection. J. Immunol. 171: S. V. Narayana. 1998. Structures of native and complexed complement factor D: 515–518. implications of the atypical His57 conformation and self-inhibitory loop in the 40. Waterhouse, N. J., V. R. Sutton, K. A. Sedelies, A. Ciccone, M. Jenkins, regulation of specific serine protease activity. J. Mol. Biol. 282: 1061–1081. S. J. Turner, P. I. Bird, and J. A. Trapani. 2006. Cytotoxic T lymphocyte-induced 35. Matrai, J., G. Verheyden, P. Kruger, and Y. Engelborghs. 2004. Simulation of the killing in the absence of granzymes A and B is unique and distinct from both activation of ␣-chymotrypsin: analysis of the pathway and role of the propeptide. apoptosis and perforin-dependent lysis. J. Cell Biol. 173: 133–144. Protein Sci. 13: 3139–3150. 41. Fan, Z., P. Yu, Y. Wang, Y. Wang, M. L. Fu, W. Liu, Y. Sun, and Y. X. Fu. 2006. 36. Fellows, E., S. Gil-Parrado, D. E. Jenne, and F. C. Kurschus. 2007. Natural killer NK-cell activation by LIGHT triggers tumor-specific CD8ϩ T-cell immunity to cell-derived human granzyme H induces an alternative, caspase-independent cell- reject established tumors. Blood 107: 1342–1351. death program. Blood 110: 544–552. 42. Li, B., C. Hartono, R. Ding, V. K. Sharma, R. Ramaswamy, B. Qian, D. Serur, 37. Smyth, M. J., M. D. O’Connor, J. A. Trapani, M. H. Kershaw, and J. Mouradian, J. E. Schwartz, and M. Suthanthiran. 2001. Noninvasive diagnosis R. I. Brinkworth. 1996. A novel substrate-binding pocket interaction restricts the of renal-allograft rejection by measurement of messenger RNA for perforin and specificity of the human NK cell-specific serine protease, Met-ase-1. J. Immunol. granzyme B in urine. N. Engl. J. Med. 344: 947–954. 156: 4174–4181. 43. Wallace, A. C., R. A. Laskowski, and J. M. Thornton. 1995. LIGPLOT: a pro- 38. Pao, L. I., N. Sumaria, J. M. Kelly, S. van Dommelen, E. Cretney, M. E. Wallace, gram to generate schematic diagrams of protein-ligand interactions. Protein Eng. D. A. Anthony, A. P. Uldrich, D. I. Godfrey, J. M. Papadimitriou, et al. 2005. 8: 127–134. Downloaded from http://www.jimmunol.org/ by guest on September 29, 2021