Structures of human calpain-3 protease core with and without bound inhibitor reveal mechanisms of calpain activation

Limb-girdle muscular dystrophy type 2a arises from mutations in the Ca2+-activated intracellular cysteine protease calpain-3. This calpain isoform is abundant in skeletal muscle and differs from the main isoforms, calpain-1 and -2, in being a homodimer and having two short insertion sequences. The first of these, IS1, interrupts the protease core and must be cleaved for activation and substrate binding. Here, to learn how calpain-3 can be regulated and inhibited, we determined the structures of the calpain-3 protease core with IS1 present or proteolytically excised. To prevent intramolecular IS1 autoproteolysis, we converted the active-site Cys to Ala. Small-angle X-ray scattering (SAXS) analysis of the C129A mutant suggested that IS1 is disordered and mobile enough to occupy several locations. Surprisingly, this was also true for the apo version of this mutant. We therefore concluded that IS1 might have a binding partner in the sarcomere and is unstructured in its absence. After autoproteolytic IS1 removal from the active Cys129 calpain-3 protease core, we could solve its crystal structures with and without the cysteine protease inhibitors E-64 and leupeptin covalently bound to the active-site cysteine. In each structure, the active state of the protease core was assembled by the cooperative binding of two Ca2+ ions to the equivalent sites used in calpain-1 and -2. These structures of the calpain-3 active site with residual IS1 and with bound E-64 and leupeptin may help guide the design of calpain-3–specific inhibitors.

Limb-girdle muscular dystrophy type 2a arises from mutations in the Ca 2؉ -activated intracellular cysteine protease calpain-3. This calpain isoform is abundant in skeletal muscle and differs from the main isoforms, calpain-1 and -2, in being a homodimer and having two short insertion sequences. The first of these, IS1, interrupts the protease core and must be cleaved for activation and substrate binding. Here, to learn how calpain-3 can be regulated and inhibited, we determined the structures of the calpain-3 protease core with IS1 present or proteolytically excised. To prevent intramolecular IS1 autoproteolysis, we converted the active-site Cys to Ala. Small-angle X-ray scattering (SAXS) analysis of the C129A mutant suggested that IS1 is disordered and mobile enough to occupy several locations. Surprisingly, this was also true for the apo version of this mutant. We therefore concluded that IS1 might have a binding partner in the sarcomere and is unstructured in its absence. After autoproteolytic IS1 removal from the active Cys 129 calpain-3 protease core, we could solve its crystal structures with and without the cysteine protease inhibitors E-64 and leupeptin covalently bound to the active-site cysteine. In each structure, the active state of the protease core was assembled by the cooperative binding of two Ca 2؉ ions to the equivalent sites used in calpain-1 and -2. These structures of the calpain-3 active site with residual IS1 and with bound E-64 and leupeptin may help guide the design of calpain-3-specific inhibitors.
Calpains are intracellular cysteine proteases transiently activated by calcium signaling (1)(2)(3). Triggering of these complex multidomain proteases causes them to make transformative cuts in their protein targets to accomplish cellular events like reorganization of the cytoskeleton needed for cell movement (4), membrane repair (5), and vesicle transport (6). In mammals, calpains exist as a family of over a dozen different isoforms, all but one of which share a two-domain protease core (7). Many of the isoforms have a C-terminal penta-EF-hand (PEF) 2 domain, which serves as a dimerization module (8,9). The structural basis for this dimerization is that calcium-binding EF-hands are typically paired. Thus, the fifth EF-hand that is not paired within the PEF domain is available to hetero-or homodimerize with another unpaired fifth EF-hand from a second PEF domain.
In the ubiquitously expressed main calpain-1 and -2, the PEF domain is heterodimerized to a two-domain small subunit through its C-terminal PEF domain (10,11). In contrast, calpain-3 uses the fifth EF-hand for homodimerization (Fig. 1). Evidence for this association came initially from the absence of the small subunit in calpain-3 preparations (12) and from the apparent molecular mass of the enzyme (ϳ180,000 Da) being twice that expected for the monomer (94,000 Da) when autolysis was suppressed by mutation of the catalytic Cys to Ser (13). Subsequently, we showed direct evidence of homodimerization of the calpain-3 PEF domain (14,15) and have subsequently solved the crystal structure of the Ca 2ϩ -bound PEF homodimer (16). Other structural differences from the large subunit of calpain-1 and -2 are the presence of two insertion sequences (ISs) that are not seen in any of the other calpain isoforms. IS1 is 48 residues long and interrupts the second domain (PC2) of the protease core after Asp 267 . The other, IS2, is a similar-sized insertion present in the linker region between the calpain-type ␤-sandwich domain and the C-terminal PEF(L) domain (17), which is reported to bind titin in muscle (18). The N-terminal leader sequence (NS) is also different from all others in the calpains and has no homologs in the database.
Aside from these structural differences, calpain-3 is unique in being especially abundant in skeletal muscle (19,20). It is encoded by a single gene, and loss-of-function mutations cause limb-girdle muscular dystrophy (LGMD) type 2a (21,22). This progressive wasting disease affects a characteristic set of muscles with onset in early adulthood. Almost 500 distinct LGMD type 2a mutations have been mapped in different families across the world according to the Leiden muscular dystrophy database (http://www.dmd.nl/). 3 Some of these introduce stop codons that lead to truncated, non-functional calpain-3; others change an obviously essential amino acid, like one of the catalytic triad residues. However, many mutations affect residues somewhat removed from key sites, and it is not clear from models of apocalpain-3 (23) or the Ca 2ϩ -bound active form (24) how these changes affect the enzyme. In an attempt to answer this question, we previously transferred several unexplained mutations into rat calpain-2 and found that they typically predisposed the enzyme to rapid autoproteolytic turnover, which could translate into accelerated loss of function of calpain-3 in the sarcomere (24).
Autoproteolysis of isolated or recombinant calpain-3 is a characteristic of this isoform and has proven difficult to prevent in vitro (13). Cleavage initially occurs in the IS1 region. Modeling studies suggest that the N-terminal part of IS1 lies in the catalytic cleft within range of the catalytic Cys (Cys 129 ) (25). Thus, the instant the enzyme is activated, it can cut the N-terminal region of IS1 in an intramolecular reaction (26). When this autolysis was studied on a recombinant version of the calpain-3 protease core, the rate of the first IS1 cleavage was independent of enzyme concentration (26). This zero order reaction is consistent with intramolecular cleavage. Furthermore, when an attenuated version (C129S) of the calpain-3 core was incubated with active wildtype core (Cys 129 ), the latter was unable to cut the C129S version, although it was able to cut itself. This confirmed that the activating first cut in IS1 is the result of an intramolecular reaction, whereas the subsequent cut at the other end of the insertion sequence could occur by intermolecular proteolysis. Although these cleavages have been regarded as autoproteolytic inactivation of the enzyme (13), we deem at least the first cleavage as an essential step to activation through removal of an internal propeptide that otherwise blocks substrate access to the catalytic cleft (25).
Here, as part of our ongoing structural analysis of the whole enzyme and its binding partners, we have determined structures of the calpain-3 protease core with IS1 present and with it proteolytically excised. The insertion sequence appears to be largely unstructured, both in the absence and presence of calcium. The predicted explanation for the intramolecular cleavage of IS1, that the N-terminal region lies in the catalytic cleft in range of the active site Cys, is here confirmed by X-ray crystallography. In addition, we have solved the first inhibitor-bound structures of the calpain-3 core, which is a step toward making inhibitors more specific for this calpain isoform. Selective inhibition of calpain-3 activity may have a sparing effect on those LGMD2A mutants that show accelerated autoproteolytic loss (24).

Most of IS1 was not seen in the crystal structure of human calpain-3 C129S protease core
The calpain-3 protease core immediately follows a unique N-terminal sequence of 47 amino acids (NS), which is predicted by PSIPRED (27,28) to be highly disordered (Fig. 2). Because this region is rapidly autolyzed (26), the first 45 residues of the NS sequence were deleted from the expression construct used here. The 48-residue IS1, which interrupts the protease core domain PC2 and acts as an internal autoinhibitory peptide (25), is also rapidly autolyzed. To slow down this internal autolysis A, conventional calpain-1 and -2 form heterodimers through the penta-EF-hand domains of the large PEF(L) and small PEF(S) subunits. B, calpain-3 forms a homodimer through PEF(L) domain connections and has unique sequences of NS, IS1, and IS2 shown in gray. The protease core domains (PC1 and PC2) are colored orange and yellow, respectively. C, H, and N, the catalytic triad residues, Cys, His, and Asn. The red section preceding PC1 is the anchor helix that contacts the small subunit. Other domains are the calpain ␤-sandwich domain (CBSW) in green; the PEF(L) domain in light blue; the glycine-rich domain in pink; and PEF(S) in purple. Numbers, residues at the domain boundaries. Arrows, autolysis sites (1b and 2b) in IS1.

Internal propeptide of calpain-3 is unstructured
and thereby provide an opportunity to determine its structure and function within the core, we replaced the catalytically active residue Cys 129 with Ser (26). This mutation produced a high yield of soluble protein that was amenable to crystallization.
The crystal of calpain-3 protease core ⌬NS C129S formed in the presence of Ca 2ϩ diffracted to 2.3-Å resolution and contained four molecules in the asymmetric unit (A-D). No electron density was seen for the N-terminal residues Ile 46 -Ile 54 of any molecule or even up to Glu 58 in some molecules (Fig. 3). Whereas this N-terminal end of the core is flexible, the C terminus is more rigid and is only missing the last residue, Asp 419 , in one of the four molecules. Molecule B is described in the following unless otherwise noted.
The overall structure of the Ca 2ϩ -bound calpain-3 protease core is similar to the structures of other active protease cores of the calpain family. Superimposing the protease core structure of calpain-3 onto the human calpain-1 protease core (PDB code 2ARY) and rat calpain-2 protease core (PDB code 1MDW) gave root mean square deviation (r.m.s.d.) values of 1.37 and 1.86 Å for alignment of 318 C␣ atoms, respectively. The catalytic triad residues of PC1 and PC2 are suitably spaced for proteolysis. Specifically, the distance of Ser 129 O␥ to His 334 N␦ of the imidazole ring is 3.6 Å, which is a functional distance for nucleophilic attack. Two Ca 2ϩ ions reside in the core at the conserved Ca 2ϩ -binding sites of PC1 and PC2. Ca 2ϩ coordination at both sites matches that seen in other calpain protease cores (29 -31). In PC1, the Ca 2ϩ is coordinated by the side chains of Glu 199 and Asp 120 , the main-chain carbonyl groups of Ile 113 and Gly 115 , and forms electron pairs with two water molecules. In PC2, the Ca 2ϩ is coordinated by the side chains of Glu 364 , Asp 371 , and Asp 394 and the main-chain carbonyl groups of Thr 392 and Glu 396 and forms an electron pair with one water molecule. The PC1 and PC2 domains are connected by a well-defined double salt bridge (Arg 118 -Glu 396 ) adjacent to the Ca 2ϩ -binding sites.
Unexpectedly, most of the unique IS1 region of the calpain-3 protease core was not seen in the structure, because it was either disordered or missing. Following the clear continuing electron density from the N terminus of IS1 (Asp 268 ), eight residues were manually built in molecule B, and seven residues in molecules C and D, but only three residues could be traced in molecule A due to insignificant electron density thereafter. The C-terminal end of IS1 is missing, and electron density is not traceable until partway into PC2 at Pro 319 in molecule B, Val 320 in molecule C, Gln 321 in molecule D, and Glu 323 in molecule A. The missing portion of IS1 lies between the 1b and 2b autolysis sites (25,26).

Replacement of the catalytic Cys by Ala prevents autolysis
Because the region of IS1 lying between the two autolysis sites in the calpain-3 protease core was not seen in the C129S crystal structure, we hypothesized that the C129S mutant might have had enough protease activity to excise IS1 during crystallization over many weeks in high CaCl 2 concentration (0.1 M). To examine this possibility, we washed some C129S crystals with buffer, dissolved those in water, and analyzed the protein by SDS-PAGE (Fig. 4A) and MALDI-TOF mass spectrometry (data not shown). Both methods showed only 26-and 13-kDa species present. The former matched in size the ⌬NS core up to the 1b IS1 cleavage site in PC2, and the latter matched the remainder of PC2 from the 2b IS1 cleavage site onward. These results from the C129S crystals are in agreement with a previous experiment using the Cys 129 protease core (25) over a much shorter time frame. It suggests that IS1 of the C129S crystals was indeed autoproteolyzed during crystallization. To investigate the source of proteolysis, we did a 4-week incubation of the calpain-3 core with a variety of reagents that would distinguish between autoproteolysis and cleavage by exogenous proteases. Calpain-3 core C129S at 48 M was mixed with crystallization solution except for PEG 8000, which affected visualization of protein bands on SDS-PAGE. Individual assays were done with the following additives: 0.2% NaN 3 , 10 mM EDTA, 20 M E-64, 0.5 mM PMSF, 100 mM NaCl, and 10% glycerol. To stop protease digestion on aliquots removed during the time course, the samples were mixed with 3ϫ SDS-PAGE loading buffer and boiled for 5 min. A trace amount of autoproteolysis of the protease core of calpain-3 C129S was detectable after 24 h (Fig. 4B). SDS-PAGE analysis exhibited three bands: a prominent one at 43 kDa and faint ones at 26.6 and 17 kDa. After 2 weeks, a 13 kDa band was noticeable and gradually increased in intensity in parallel with the 26.6 kDa band at the expense of the 43 kDa starting material. Neither the protease inhibitors (EDTA, E-64, or PMSF), the bactericidal The calpain-3 protease core sequence, including NS, was analyzed for the probability of disorder using a web-based program (http://bioinf.cs. ucl.ac.uk/psipred) 3 (27,32). The likelihood of a region being disordered is plotted as a confidence score against the amino acid sequence. The blue tracing represents sequence in a disordered state, and the orange tracing corresponds to disordered residues likely to be involved in protein-protein interactions. agent (NaN 3 ), nor the protein stabilizers (glycerol and NaCl) were able to prevent enzyme autolysis (Fig. 4, D-I). Therefore, it is confirmed that the protease core of calpain-3 C129S retains weak protease activity, presumably that of a serine protease, because Ser 129 is equally well aligned with His 334 for catalysis, as is Cys-129 in the native protein.
To completely abolish protease activity, we mutated Cys 129 to Ala 129 and assayed its activity as we did for C129S. No autolysis bands were generated during the 4-week incubation with this mutant (Fig. 4C). Crystallization of this enzymatically inactive calpain-3 construct was undertaken, and C129A crystals were obtained in half the time (2 weeks) it took to form C129S crystals.
The C129A crystal diffracted to 2.75 Å, and, once again, four protease cores were present in the asymmetric unit. Although there was no autolysis of C129A, most of IS1 still could not be built due to poor quality or missing electron density. Only the N-terminal seven residues of IS1 could be traced based on clear electron density. Therefore, we contend that IS1 is highly flexible in the Ca 2ϩ -bound core. Despite the difference in autolysis between C129S and C129A, the structures are superimposable, with two Ca 2ϩ bound and the PC1 and PC2 domains rotated to form the catalytic cleft. IS1 is exposed at the surface of the protein, and, therefore, its absence from the crystal structure due to autolysis or flexibility has no effect on the core structure.

IS1 of the calpain-3 core is highly mobile in solution as revealed by small-angle X-ray scattering (SAXS)
To better understand how IS1 relates to the enzyme, we looked at the conformation of the insertion sequence in solution using SAXS. To avoid any autolysis, even during the shorter time frame of the SAXS experiments, we used only the C129A version of the calpain-3 core. SAXS data were collected on calpain-3 core C129A in the Ca 2ϩ -free state, with 5 mM EDTA present, and in the Ca 2ϩ -bound state in 2 mM CaCl 2 ( Table 1). The scattering patterns of both samples were similar, and based on the Guinier plot at low q values, there were no signs of aggregation or intermolecular repulsion for the Ca 2ϩfree core and the Ca 2ϩ -bound core (Fig. 5, top). The Kratky plots showed bell-shaped curves for both samples (Figs. S1 and S2, D-F). They indicated that the calpain-3 protease cores are globular molecules and were well folded. Molecular masses of 42.02 and 42.8 kDa were calculated from the Porod volume ϫ 0.625 for the Ca 2ϩ -free and -bound samples, respectively.

Internal propeptide of calpain-3 is unstructured
These values are close to the 43.3 kDa value calculated from the amino acid sequence of the monomer.
Structural predictions based on sequence analysis (32) indicate that IS1 lacks a well-defined tertiary structure (Fig. 2). Therefore, the missing IS1 region was built in using the ensemble optimization method (EOM) (33). A large conformational pool (10,000 possibilities) was generated using sequence information and the crystal structure of Ca 2ϩ -bound calpain-3 C129S core as a template but with PC1 and PC2 unfixed and, therefore, free to rotate. However, only those conformers that were consistent with the SAXS experimental scattering data were retained. The final ensemble of conformers of the calpain-3 core with the IS1 structural model inserted was in excel-lent agreement with both sets of experimental scattering data, 2 values of 1.304 and 0.497 corresponding to the Ca 2ϩ -free calpain-3 core and the Ca 2ϩ -bound core, respectively. The calculated flexibility of the ensemble conformers by EOM showed that the Ca 2ϩ -free core value was relatively high in solution (Rflex 63.2/R 1.2), whereas the Ca 2ϩ -bound core was more compact (Rflex 58.0/R 0.34). Both structural models from EOM were then used to calculate a CRYSOL scattering curve (34) to compare against experimental scattering. The theoretical and experimental curves are well matched, with 2 values of 1.94 and 0.62 for the Ca 2ϩ -free core and Ca 2ϩ -bound cores, respectively. Also, these ensemble models aligned quite closely with the ab initio envelope profiles developed from DAMAVER Internal propeptide of calpain-3 is unstructured calculations (Fig. 6). However, the Ca 2ϩ -free core model shows a higher 2 value than that of the Ca 2ϩ -bound form, which indicates that domains PC1 and PC2 are more dynamic in the apo state than with Ca 2ϩ bound.
In the structural model of the Ca 2ϩ -free core, IS1 lies in the vicinity of the active site groove between PC1 and PC2 (Fig. 7, A  and B). The N-terminal end of IS1 up to residue Thr 270 enters the active site and is close to PC1. A little beyond the C terminus of IS1, structure is regained at Gln 321 of PC2. Between these points, the IS1 region forms an elongated random structure that protrudes out of the enzyme cleft and is exposed to the solvent. The profile of the modeled IS1 is different in the Ca 2ϩbound core, being more extended and flexible in the solvent space around the front of active site than in the Ca 2ϩ -free sample (Fig. 7, C and D). A change in IS1 disposition is to be expected because when Ca 2ϩ binds to the core, there is substantial rotation of the PC1 and PC2 domains relative to each other, and the ends of IS1 are shifted in their locations between the Ca 2ϩ -bound and -free forms. Using the SAXS data, the distances between the IS1 ends in the Ca 2ϩ -free and -bound forms are 33.4 and 16.5 Å, respectively. The BILBOMD program (35) was applied to monitor IS1 dynamic motions at a high temperature (1500 K). PC1 and PC2 were treated as rigid bodies while allowing IS1 to remain flexible. radius of gyration (R g was limited to the range between 20 and 26 Å, and 400 conformations per R g value were sampled. The Ca 2ϩ -free and Ca 2ϩ -bound structures resulted in large average r.m.s.d. values of 9.0 and 12.3 Å, respectively, when compared with the initial IS1 position, indicating that IS1 is flexible and can adopt diverse conformations. Comparison of the P(r) functions (pair-wise distance distribution function) for both samples (Fig. 5, bottom) shows that the calpain-3 cores have different distance distributions in solution although the Ca 2ϩ -free and Ca 2ϩ -bound calpain-3 cores are globular molecules. The Ca 2ϩ -free protein has a more extended shape (D max ϭ 84.7 Å and R g ϭ 25.2 Ϯ 0.04 Å) than the Ca 2ϩ -bound version (D max ϭ 83.7 Å and R g ϭ 24.9 Ϯ 0.04 Å). This suggests that there is some flexibility between the PC1 and PC2 domains in both Ca 2ϩ -free and Ca 2ϩ -bound states in solution. In the absence of supporting domains in the whole enzyme structure, it seems that the PC1 and PC2 can pivot around a hinge region (29), such that the two halves of the core are not always correctly aligned as they would be in the native state. Similar conclusions were reached with the isolated calpain-1 core. Flexibility was also observed by superposition of the crystal structure of Ca 2ϩ -bound calpain-3 core on the solution structures with Ca 2ϩ absent/present. These comparisons yielded r.m.s.d. values of 13.7 Å for the Ca 2ϩ -free solution structure and 7.6 Å for the Ca 2ϩ -bound solution structure (data not shown).

Role of IS1 of calpain-3 protease core in autoinhibition
The basis for the inactivity of apocalpains was immediately clear when the first crystal structures of calpain-2 were solved (10,11). The catalytic cleft is not properly formed and remains this way until two Ca 2ϩ bind cooperatively to sites in the core, which causes a rotation of the PC1 and PC2 domains such that the catalytic triad residues are properly aligned for catalysis (29). Based on modeling, the same calcium activation process would seem to be necessary to form the catalytic cleft of calpain-3. But in addition to the metal ion requirement, it is known that calpain-3 has an additional inhibitory feature, which is IS1 acting as an internal autoinhibitory propeptide to block the enzyme's active site (25). Although in the Ca 2ϩ -bound crystal structure, most of IS1 is not visible, what can be seen of IS1 in the active site cleft supports the autoinhibition role. First, the N-terminal segment of IS1 (Met 272 -Thr 273 -Tyr 274 -Gly 275 ) protrudes into the active-site cleft through the S3, S2, S1, and S1Ј subsites and makes extensive contacts with residues on either  (Fig. 8A). These contacts restrict access to the catalytic cleft and are not seen in the calpain-1 and calpain-2 cores. This region of IS1 that occupies the active-site cleft is perfectly conserved in the comparison of the IS1 sequences from 22 different mammals, which reinforces its functional importance (Fig. S3).

Autolytic removal of IS1 paves the way for solving inhibitor-bound calpain-3 core structures
Previous studies by our group proposed that the loss of IS1 was required for calpain-3 activation. Preincubation of the cal-pain-3 core with Ca 2ϩ present increased its proteolytic activity over a 24-h period as IS1 was autolyzed away (25). At this time point, calpain activity could be inhibited by adding cysteine protease inhibitors E-64 and leupeptin. To obtain inhibitorbound structures for the calpain-3 core (C129), the autolysis reaction was allowed to proceed for even longer (5 days) before reacting the Ca 2ϩ -bound core with inhibitors E-64 and leupeptin before crystallization.
Crystals of the core bound to E-64 and leupeptin diffracted to 2.8 and 3.2 Å resolution, respectively. Both crystal types had four molecules in the asymmetric unit and had the same space group (P2 1 2 1 2 1 ) as the inhibitor-free structures (C129S-core-Ca 2ϩ and C129A-core-Ca 2ϩ ). Thus, the molecular replacement method was performed using the C129S-core-Ca 2ϩ structure as a search model. The initial 2F o Ϫ F c electron density maps for both co-crystals showed continuous density in the active site; subsequently, the E-64 molecule and leupeptin molecule were unambiguously placed there and were well-defined (Table 2). Electron density was missing for the N-terminal 8 -13 residues of the construct in all four molecules of both Ca 2ϩ -free and Ca 2ϩ -bound samples are in green and red, respectively. Bottom, pair distribution functions P(r) of Ca 2ϩ -free (green) and Ca 2ϩ -bound (red) samples. Both samples give similar maximum dimensions (D max ) for the particles (84.7 Å for Ca 2ϩ -free and 83.7 Å for Ca 2ϩ -bound), but the P(r)-distribution profiles are different.

Internal propeptide of calpain-3 is unstructured
complex structures. Almost the entire IS1 (residues 268 -324) was missing except in molecule A of the calpain-3 core-Ca 2ϩ ⅐E-64 complex structure (missing residues 270 -321) and in molecule D (missing residues 269 -321) from the structure of the calpain-3 core-Ca 2ϩ ⅐leupeptin complex. Molecules A and D are described in the following unless otherwise noted. Both complex structures display similar overall architecture to the inhibitor-free structure of C129S-core-Ca 2ϩ with r.m.s.d. values of 0.65 Å (E-64) and 0.41 Å (leupeptin), using 314 and 306 C␣ atoms for alignment, respectively.
The E-64 molecule is accommodated in the active-site cleft of the enzyme core by occupying the S1, S2, and S3 subsites in place of IS1, after the latter has been autolyzed away (Fig. 9). Note that the N-terminal region of IS1 is not observed because it has been displaced from the active site by the inhibitor and is too disordered to have stable electron density. The N-terminal epoxy and central leucyl moieties of E-64 fit well into the electron density, whereas the C-terminal 4-guanidinobutane moiety of E-64 has weak electron density that suggests this region is flexible (Fig. 10A). There is strong continuous electron density between the C2 of E-64's epoxy group and the catalytic Cys 129 S␥, which directly confirms covalent bond formation (1.64 Å) between E-64 and calpain-3. One of the oxygen atoms from the epoxysuccinic motif of E-64 is pointing to the oxyanion hole and forms hydrogen bonds with Gln 123 NE2 (2.51 Å) and the main-chain amide of Cys 129 (3.0 Å) (Fig. 9A). In addition, it makes an electrostatic contact with the main-chain amide of Asp 128 (3.4 Å). The second oxygen atom of the epoxysuccinic motif produces a hydrogen bond with catalytic His 334 ND1 (2.9 Å) at the P1-S1 site. Also, the main-chain amide of the leucyl group of E-64 makes electrostatic contacts with the main-chain carbonyl group of the conserved Gly 222 (3.3 Å) from PC1 and with the conserved Gly 333 (3.5 Å) from PC2 at the P2-S2 subsite. The 4-guanidinobutane moiety of E-64 is pointing to the solvent, and no P-S contact is observed in the S3 subsite.
The leupeptin molecule is also positioned in the active site pocket occupying the S1, S2, and S3 subsites with a good fit to the electron density map (Fig. 10B). Leupeptin adopts the same orientation as natural substrates, whereas E-64 is in the reversed orientation. The sequence Leu-Leu-arginal binds to Figure 6. Agreement between structural models and SAXS experimental scattering patterns for the calpain-3 core. A and B, theoretical scattering patterns calculated by CRYSOL from structural models based on the ensemble optimization method for the Ca 2ϩ -free and Ca 2ϩ -bound calpain-3 cores, respectively. C and D, show solution structural models fit to envelopes, which correspond to Ca 2ϩ -free and Ca 2ϩ -bound sample, respectively. Green ribbon, PC1 and PC2 domains of the calpain-3 core. Red spheres represent IS1 residues. N and C termini of the core are indicated by N and C, respectively.

Internal propeptide of calpain-3 is unstructured
the S3-S2-S1 sites, respectively. Continuous electron density is seen between the aldehyde carbonyl group of leupeptin and the catalytic Cys 129 S␥ (Fig. 10B). The density indicates that a 2.0-Å hemithioacetal bond is formed between leupeptin and calpain-3. The interactions between leupeptin and the calpain-3 protease core are weaker than those observed in the E-64bound calpain-3 protease core (Fig. 9B). The hemiacetal oxygen atom of leupeptin faces the oxyanion hole and forms three electrostatic contacts with the main-chain amide of catalytic Cys 129 (3.6 Å), Gln 123 NE2 (3.5 Å), and catalytic His 334 (3.9 Å), respec-

Internal propeptide of calpain-3 is unstructured
tively. Elsewhere in the active-site pocket, P1 Arg 3 interacts with the carbonyl oxygen of Gly 333 (3.2 Å) of PC2. Additionally, the backbone of the P2 Leu 2 residue makes a parallel electrostatic contact with the main-chain amide and carbonyl oxygen of Gly 222 (3.0 and 3.4 Å) of PC1 at the P2-S2 subsite. However, the N-terminal acetyl-leucine moiety of leupeptin makes no contact with neighboring residues in the S3 subsite, much the same way the C-terminal 4-guanidinobutane moiety of E-64 fails to contact the S3 subsite in its complex (Fig. 9B).
Solving the inhibitor-bound structures has also given us a chance to see whether and how the residual N-terminal IS1 section might interfere with substrate or inhibitor binding to the cleft of the calpain-3 core. Comparing the structures of the residual IS1 in the Ca 2ϩ -bound C129S core with the E-64/leupeptin-bound Cys 129 core structures, the backbone of Thr 273 -Tyr 274 from IS1 overlaps well on the P1 to P2 moieties of E-64 and leupeptin, but their side chains are in different orientations and make different contacts with S1Ј to S2 subsite residues (Fig.  8). The carbonyl oxygen of Gly 275 of IS1 projects into the S1Ј site and forms polar contacts with Gln 123 NE2 (3.2 Å) and Trp 360 NE1 (3.4 Å), whereas the Gly 275 ␣-carbon also makes van der Waals contacts with catalytic His 334 CE1 (3.2 Å). The

Internal propeptide of calpain-3 is unstructured
latter two interactions were not seen in the inhibitor complex structures (Fig. 9).  (Fig. 8B). The amide of Tyr 274 from IS1 also makes a hydrogen bond with the carbonyl oxygen of Gly 333 (3.2 Å) from PC2 at the S1 subsite. These bonds between IS1 and residues of the S1 subsite clearly show the residual IS1 fragment blocking the catalytic center of the enzyme. In further keeping with this stability, Thr 273 of IS1 imitates the role of the P2 leucine residue of the inhibitors in making a double electrostatic contact with conserved Gly 222 (3.3 and 3.6 Å) from PC1 and van der Waals contacts with the conserved Gly 333 from PC2 at the S2 site. Moreover, additional contacts are seen between IS1 and residues of the S3 subsite. These are Asn 271 OD1 of IS1 with Lys 410 NZ (3.1 Å) and with Asn 223 ND2 (4.0 Å), respectively. By comparing these results, it seems that IS1 is potentially more stable in the active site as an internal autoinhibitory propeptide than are the exogenous inhibitors E-64 and leupeptin. What gives these inhibitors the edge in occupying the catalytic cleft is that they can form covalent bonds with Cys 129 . Although the residual portion of IS1 makes a network of interactions in the active site, these connections can be broken when the PC1 and PC2 domains rotate around each other. However, this N-terminal IS1 peptide can obviously compete with substrates and inhibitors for access to the catalytic cleft whenever it forms.

Activating cleavage of the internal IS1 propeptide in calpain-3 is regulated by calcium signaling
In the crystal structure of Ca 2ϩ -bound calpain-3 C129S protease core, the IS1 region, with the exception of the eight N-ter-minal residues, cannot be observed. According to SDS-PAGE analysis of Ca 2ϩ -bound calpain-3 C129S protease core crystals sampled and the time course of autoproteolysis of the enzyme in solution, the calpain-3 C129S core with Ca 2ϩ present shows weak protease activity. The explanation for this activity is that the serine replacement for cysteine in the catalytic triad can be deprotonated and act as a nucleophile to hydrolyze IS1. This activity came to light during the weeks required for crystallization and could potentially occur within the crystal because it is a zero-order reaction. Cys 129 -catalyzed cleavage of IS1 is observable in a matter of minutes and is complete within hours (25,26). This explanation is supported by the C129A mutation completely abolishing the protease activity of the calpain-3 core. In all four active structures presented here, Ca 2ϩ ions are located in the expected sites of PC1 and PC2. The core structures have clearly undergone the same cooperative realignment of the two core domains seen in the calpain-1 core when Ca 2ϩ ions bind that brings the catalytic triad residues into alignment for catalysis. At least in this in vitro artificial system, where the calpain-3 core is examined in isolation, the driving force for activation is Ca 2ϩ signaling, and the IS1 peptide can only delay access to the catalytic cleft but not prevent it.
SAXS solution structural models of C129A, with and without Ca 2ϩ bound, show that most of IS1 is a long extended flexible loop exposed to the solvent. Only a short N-terminal portion of IS1 is visible and is located in the active site cleft, as shown by the crystal structures. We had earlier anticipated that IS1 would be structured in at least the calcium-free form, because there is a central region of IS1 with high ␣-helix propensity (25). Also, the protein sequence alignment of IS1 from humans and 21 other mammals shows that most of the insertion sequence has been well-conserved during the evolution with 87.7% identity and 95.2% similarity (Fig. S3). Such conservation suggests that IS1 performs an important biological function rather than simply being a disordered protein loop. In general, molecular recognition can induce unstructured proteins to fold in the cell. The folding transition can occur when the unstructured protein region binds to its target, such as other proteins or a nucleic acid  (36). Indeed, this is how calpastatin recognizes and inhibits calpain-2. Based on SAXS solution structural models with/without Ca 2ϩ bound and the Ca 2ϩ -bound crystal structures, we propose that IS1 is intrinsically unstructured, as predicted by PSIPRED (27,28) (Fig. 2), and has affinity for an unknown protein. In a calcium-free environment, IS1 of the protease core could associate with the unknown partner and form a stable complex that is enzymatically inactive. When there is an influx of calcium ions to the sarcomere and two Ca 2ϩ bind to the calpain-3 core, they could activate the enzyme through reorientation of PC1 and PC2. In this model, the IS1-binding protein could potentially be released from the Ca 2ϩ -activated enzyme, allowing IS1 to be cut to help open up the cleft to calpain-3 substrates. Another possibility is that the binding partner serves to hold the IS1 loop away from the cleft, making it fully accessible to substrates. In this model, there would be no need for proteolysis. Activation would be fully reversible without a physical change to the enzyme. Another advantage of this model is that the N-terminal end of IS1 would not be lying in the catalytic cleft as it does after cleavage, needing to be displaced by substrates. It used to be thought that autoproteolysis was needed for activation of calpain-1 and -2, but this was disproven by the structure of the activated calpain-2 in complex with calpastatin, showing that the anchor helix is released on calcium binding, rather than being proteolyzed away (37,38). Again, the advantage of a non-proteolytic activation is a return to the inactive state after calcium signaling without damage or loss to the calpain, so that the system is ready to respond to the next signaling event.

IS1 as a guide for designing inhibitors of calpain-3
Crystal structures show the N-terminal portion of IS1 is located in the active site cleft after cleavage, with the residues closest to the cleavage site making extensive hydrophilic interactions with residues of PC1 and PC2 from the S3 to S1Ј subsites, as detailed under "Results." Although E-64 and leupeptin bind in the active site of calpain-3 core and form a covalent bond to the catalytic Cys 129 , both exogenous inhibitors have fewer contacts with residues of the active site compared with the N terminus of IS1. Thus, this section of IS1 can serve as a model for the design of a peptidomimetic inhibitor of calpain-3. For exogenous calcium-dependent cysteine-protease inhibitors, the irreversible/reversible covalent bond between inhibitor and catalytic cysteine thiol is crucial for effectiveness (39). IS1 forms a weak electrostatic contact with O␥ group of Ser 129 (catalytic Cys 129 ), which suggests the possible location of a warhead to turn the IS1 peptide from a competitor peptide into a covalent inhibitor.
E-64 is an epoxy succinic inhibitor with broad effectiveness against cysteine proteases. It forms an irreversible covalent bond with the active-site cysteine thiol. Leupeptin is a peptide aldehyde inhibitor against serine, cysteine, and threonine proteases and makes a transient reversible covalent bond with the cysteine thiol. Both inhibitors had no effect on calpain-3 autolysis, presumably because they could not displace the uncleaved IS1 from the catalytic cleft (12). The calpain-3 splice variant Lp82, which lacks IS1, could be inhibited by E-64 (40,41), and the calpain-3 isoform from human peripheral blood mononu-clear cells, which lacks IS1 and IS2, could also be inhibited by leupeptin (42). Although E-64 and leupeptin bind in the active site of calpain-3 core after IS1 cleavage, and form a covalent bond to the catalytic Cys 129 , both exogenous inhibitors have many fewer contacts with residues of the active site than does the N terminus of IS1. The electrostatic interactions made by E-64 and leupeptin in the calpain-3 core are also fewer than those made in the calpain-1 core crystal structures (43). This further suggests that E-64 and leupeptin are not optimal activesite inhibitors of the calpain-3 protease core. Thus, based on IS1, there are options for designing superior inhibitors of calpain-3 that might compete with IS1 and block activity even before the initial autolytic event.

Site-directed mutagenesis, expression, and purification of human calpain-3 protease core
The protease core (PC1 and PC2) of human calpain-3 active form Cys 129 with its 45-residue N-terminal extension sequence deleted (⌬NS) and the same construct with its catalytic Cys 129 mutated to Ser (C129S) have been described previously (25,26). These two constructs include residues Ile 46 -Asp 419 with a C-terminal His 6 tag. In a third construct, the catalytic Cys 129 was mutated to Ala using the QuikChange site-directed mutagenesis method (Stratagene) according to the manufacturer's protocol. The three calpain-3 protease core constructs ⌬NS Cys 129 , C129S, and C129A were expressed and purified as described previously (26).

Derivatization of calpain-3 protease core ⌬NS C129 with inhibitors E-64 and leupeptin
Calpain-3 protease core ⌬NS C129 was purified through the DEAE anion-exchange, nickel-nitrilotriacetic acid, and Sephacryl S-200 size-exclusion chromatography steps. Fractions containing purified calpain-3 protease core were combined and analyzed by 12% SDS-PAGE and then incubated in 10 mM CaCl 2 , 25 mM Tris-HCl, pH 7.6, and 150 mM NaCl for 5 days at room temperature until IS1 autolysis reached its maximum extent as judged by SDS-PAGE. Subsequently, a 10-fold molar excess of inhibitor E-64 was mixed with autolyzed protein solution and incubated for 3 h at room temperature. The derivatized protein was further purified by Q-Sepharose chromatography (GE Healthcare). All purified proteins were bufferexchanged into 10 mM NaHEPES (pH 7.6), 10 mM DTT. For making the complex with leupeptin, after 5 days of IS1 autolysis, the protein was mixed with a 10-fold molar excess of leupeptin and incubated for 1 h. Crystallization trials were done without further purification of the leupeptin-modified core.
Protein solutions of C129A for SAXS experiments were dialyzed exhaustively against 10 mM NaHEPES (pH 7.6), 3% glycerol, 5 mM DTT, and 2 mM CaCl 2 , (or without CaCl 2 but with 5 mMEDTAinstead).Proteinmolaritiesweredeterminedbyabsorbance at 280 nm using a Nano-drop spectrometer with a molecular mass of 43,318.08 Da and a theoretical molar extinction coefficient of 80,370 M Ϫ1 cm Ϫ1 .

Internal propeptide of calpain-3 is unstructured Molecular weight determination of calpain-3 core from protein crystals
Calpain-3 protease core C129S crystals were transferred to a new drop containing crystallization well solution devoid of calpain-3 core C129S solution. The crystals were then washed three more times with well solution. Subsequently, these crystals were dissolved in Millipore water and subjected to both SDS-PAGE and MALDI-TOF mass spectrometry analysis.

Autoproteolysis assays
The degree of autoproteolysis of calpain-3 protease core constructs was assessed by SDS-PAGE. Calpain-3 protease core C129S and C129A (48 M final concentration) were tested for autolysis in 0.1 M MES (pH 6.5), 0.1 M CaCl 2 . This buffer mimicked crystallization conditions except for the absence of PEG 8000, which affected protein band visualization on SDS-PAGE. The various additives assessed for their effect on protease core C129S autolysis included the cysteine protease inhibitor E-64 (0.02 mM), serine protease inhibitor PMSF (0.2 mM), chelating agent EDTA (10 mM), and chemicals that could potentially protect the protein from degradation, namely 100 mM NaCl, 0.2% NaN 3 , and 10% glycerol. Samples were incubated at 22°C in a final volume of 60 l. Aliquots were removed at various times (0, 1, 3, 6, 14, 21, and 28 days), and the reaction was stopped by the addition of 3ϫ SDS-PAGE loading buffer, which contained 187.5 mM Tris-HCl (pH 6.8), 6% SDS, 30% glycerol, and 0.03% bromphenol blue. Protein bands were visualized after 12% SDS-PAGE by Coomassie Blue staining. The band intensity was measured by densitometry. The control assay consisted of calpain-3 core C129S or C129A incubated with 5 mM EDTA in place of CaCl 2 as described above.

Crystallization and data collection
Purified proteins were concentrated to 9 -13 mg/ml for crystallization trials. The hanging-drop vapor diffusion method was used at 22°C. Precipitant conditions were 8 -11% (w/v) PEG 8000, 0.1 M MES (pH 6.5), and 0.1 M CaCl 2 . Irregular, soft crystals of protease core ⌬NS C129S were initially obtained after 4 weeks. Streak seeding was carried out immediately after new drops were set up. Through several rounds of seeding, rectangular-shaped crystals grew over a 4-week period to reach diffraction size. For protease core ⌬NS C129A crystallization, a cross-seeding procedure was performed. Initial seeds were from the C129S crystals, and afterward, the seeds were all from C129A crystals. Clusters of long needle-shaped crystals of Cys 129 -E-64 derivative and crystals of co-crystallized Cys 129 -leupeptin appeared within 1 week and reached diffraction size in 3 weeks. Because all of the crystals were fragile and diffracted poorly, dehydration was used. Crystals were transferred into stabilizing solutions that contained 18% (w/v) PEG 8000, 0.1 M MES (pH 6.5), 0.1 M CaCl 2 , and sorbitol, with the latter increasing in 5% increments from 5 to 30% over a period of 3 days, before crystals were flashfrozen in liquid nitrogen.

Structural determination and refinement
For crystals of calpain-3 protease core ⌬NS C129S and C129A, X-ray diffraction data were collected on the X6A beam line at the Brookhaven National Synchrotron Light Source. For crystals of calpain-3 protease core ⌬NS Cys 129 -E-64 derivative and -leupeptin derivative, X-ray diffraction data were collected on the 23ID-B beam line at the Argonne National Laboratory. All crystals belonged to the P2 1 2 1 2 1 space group with four molecules per asymmetric unit. The data sets were indexed, integrated, and scaled using XDS (44). Molecular replacement calculations were carried out with PHASER (45) through the CCP4 GUI (46). The calpain-2 protease core structure (PDB code 1MDW), but with Ca 2ϩ omitted, was used as a search model. The structure was then refined with REFMAC5 (47) and PHENIX (48). All inspections and manual manipulations were completed with COOT (49,50). Figures were generated in PyMOL. Crystallographic data collection and refinement statistics are summarized in Table 2.

SAXS data collection
SAXS experiments were performed on the Bio-SAXS beamline BM29 at the European Synchrotron Radiation Facility (ESRF) in France and Bio-SAXS beamline G1 at Cornell Mac-CHESS. Calpain-3 protease core ⌬NS C129A solutions at concentrations of 1.0, 3.0, and 5.0 mg/ml were prepared. Data were collected using a dual Pilatus detector with sample-to-detector distances of 2.8 and 1.5 m and beam energies of 0.9918 and 1.055 Å, respectively. SAXS data were collected separately for protease core ⌬NS C129A in dialysis buffer of 10 mM NaHEPES (pH 7.6), 5 mM DTT, and 3% glycerol with 2 mM CaCl 2 or with 5 mM EDTA. For each of the proteins, measurements were made at three different concentrations to detect any concentration-dependent intermolecular interactions. A series of 10 1-s exposures and 20 1-s exposures were taken at ESRF BM29 and CHESS G1, respectively. The data scaling, azimuthal integration, curve average, background subtraction, and ab initio reconstruction were automatically performed on a data-processing pipeline driven by the EDNA framework with ATSAS software (51) at ESRF BM29. The BioXTAS RAW software (52) was used for data processing and preliminary analysis at CHESS G1. Radiation damage was not observed in any of the samples (Figs. S1 and S2, A-C). The data for 1-5 mg/ml concentrations were merged to give a single SAXS profile for calpain-3 protease core ⌬NS C129A with and without Ca 2ϩ . The pair distance distribution function P(r) was calculated with GNOM (53). Estimated maximum dimension (D max ) values of 84.7 Å for Ca 2ϩ -free and of 83.7 Å for Ca 2ϩ -bound were obtained from SAXS data sets. The real-space R g values were in good agreement with Guinier R g . DAMMIN (54) was used for ab initio shape determination from which a set of 10 models were superimposed and averaged by DAMAVER (55). The NSD (normalized spatial discrepancy) values of the models were very similar, with mean values of 0.669 and of 0.542 calculated by DAMAVER that corresponded to Ca 2ϩ -bound and Ca 2ϩ -free C129A cores, respectively. A summary of the SAXS parameters is given in Table 1.