Determination of Peptide Substrate Specificity for μ-Calpain by a Peptide Library-based Approach

Calpains are proteases that catalyze the limited cleavage of target proteins in response to Ca2+ signaling. Because of their involvement in pathological conditions such as post-ischemic injury and Alzheimer and Parkinson disease, calpains form a class of pharmacologically significant targets for inhibition. We have determined the sequence preference for the hydrolysis of peptide substrates of the ubiquitous μ-calpain isoform by a peptide library-based approach using the proteolytic core of μ-calpain (μI-II). The approach, first described by Turk et al. (Turk, B. E., Huang, L. L., Piro, E. T., and Cantley, L. C. (2001) Nat. Biotechnol. 19, 661–667), involved the digestion of an N-terminally acetylated degenerate peptide library in conjunction with Edman sequencing to determine the specificity for residues found at primed positions. The cleavage consensus for these positions was then used to design a second, partially degenerate library, to determine specificity at unprimed positions. We have improved upon the original methodology by using a degenerate peptide dendrimer for determination of specificity at unprimed positions. By using this modified approach, the complete cleavage specificity profile for μI-II was determined for all positions flanking the cleaved peptide. A previously known preference of calpains for hydrophobic amino acids at unprimed positions was confirmed. In addition, a novel residue specificity for primed positions was revealed to highlight the importance of these sites for substrate recognition. The optimal primed site motif (MER) was shown to be capable of directing cleavage to a specific peptide bond. Accordingly, we designed a fluorescent resonance energy transfer-based substrate with optimal cleavage motifs on the primed and non-primed sides (PLFAER). The μ-calpain core shows a far greater turnover rate for our substrate than for those based on the cleavage site of α-spectrin or the proteolytic sequence consensus compiled from substrate alignments.

nals (3) into a proteolytic signal by catalyzing the limited cleavage of target proteins (4,5). Among the known cellular substrates of calpain are numerous cytoskeletal proteins, as well as some receptors and integral membrane proteins like the Na ϩ /Ca 2ϩ exchanger, NCX-3 (6).
Calpains must be strictly regulated because they catalyze irreversible processing in the cell. In one scenario, calpains localize to the plasma membrane under activating conditions (7,8). This placement may act to position calpains where they can respond to brief calcium influxes from the opening of calcium channels, resulting in very localized and transient activity. Deactivation of calpain can come about in several ways: binding to the endogenous calpain inhibitor calpastatin (9); autoproteolytic inactivation; or simply the dissipation of local high calcium levels. Unregulated intracellular calcium influx leads to hyperactivation of calpain and is associated with a large numbers of pathologies such as postischemic injury (10). Calpains are thus of great biomedical and therapeutic interest, and there is a need for calpain-specific inhibitors that will not affect other cysteine proteases.
The active site of any enzyme serves the dual purpose of binding substrates and performing catalysis. For proteases, improvements in the selectivity constant (k cat /K m ) for small substrates are mainly driven though improvements in K m . Optimized cleavage sequences are therefore strongly linked to an optimized binding and are valuable for the design of sensitive and specific peptidomimetic substrates, as well for the design of lead compounds in drug and inhibitor design (12,13). Because of the broad substrate specificity of calpains, the determination of their optimal cleavage sequence has been elusive. This is reflected in the paucity of sensitive substrates and highly specific, active site-directed inhibitors.
The wide range of residues that flank known calpain cleavage sites suggests that the conventional calpains can accommodate a variety of side chains within their active site binding pocket (14,15). Efforts have been made to determine the cleavage specificity of calpains (14, 16 -22). Early kinetic studies of synthetic peptidyl compounds possessing a variety of unprimed side 4 residues revealed some preferences at the P1 and P2 positions (13,16). The high cost and effort of synthesis limited the early studies to certain residues that reflected observed cleavages in natural peptides, and this created a bias in the observed specificity. Other attempts to deduce a sequence motif from observed cleavage sites in proteins and naturally occurring peptides revealed no consistent sequence preference, other than that observed from synthetic substrates (17, 19 -22). This suggests that higher order structural features, * This work was funded by the Heart and Stroke Foundation of Ontario, the Canadian Institutes of Health Research, and the Government of Canada's Network of Centres of Excellence program supported by the Canadian Institutes of Health Research and the Natural Sciences and Engineering Research Council through the Protein Engineering Network of Centres of Excellence. 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 with 18  such as peptide bond accessibility, backbone conformation, or threedimensional structure, are key factors in cleavage site determination. A recent compilation of 106 known calpain cleavage sites in protein substrates has revealed propensities for certain residues at positions P4 to P7Ј (14). However, because such proteome studies ignore the kinetics of cleavage, the ability to extrapolate these results to small peptidyl or peptidomimetic substrates and inhibitors is limited. Moreover, because the data were compiled from protein substrates, the primary sequences of the cleavage sites may have a minimal role in directing the cleavage, which would be expected if higher order structural factors were involved in the recognition. It is also plausible that substrates of calpain may have evolved to possess a suboptimal primary sequence, surrounding the presumably exposed scissile peptide bond, to control its susceptibility to the protease. In this case optimal sequence data would not be observed from a natural substrate.
Results from a number of other recent studies suggest that, contrary to the dogma of papain-like cysteine proteases, calpains possess significant structural determinants at primed positions (14,23). Substrates designed to include primed side interactions are kinetically superior to other commercially available substrates that use only unprimed positions. A systematic kinetic analysis of residue preference remains to be done for primed positions.
Library-based approaches for the determination of substrate specificity allow for fast and exhaustive surveys of a large number of residuesubsite combinations (24). Also, mixture-based libraries are relatively inexpensive to generate. Short peptides of random sequence are unlikely to possess extensive secondary structure, confining cleavage determinants to the primary structure, and control of the digestion time allows for selection of kinetically favored cleavages. Recently, a methodology was published to allow for the quick, systematic determination of the cleavage specificity of proteases using a limited number of mixturebased peptide libraries. This method, first described by Turk et al. (25), involves a minimum of two rounds of digestion of different peptide library mixtures, followed by N-terminal sequencing to determine the relative abundance of each residue (see Fig. 1). Each sequencing cycle corresponds to a specific position in the substrate. The first round reveals primed side residue specificity (see Fig. 1A), and this is applied in the design of a second, partially degenerate library, used to reveal unprimed side residue specificities (see Fig. 1B) (25,26). We have applied and improved upon this methodology (see Fig. 1C) to determine the complete cleavage specificity profile as well as the optimal cleavage sequence for the protease core of -calpain.
Members of the calpain family share a calcium-dependent papainlike protease core that can be divided into two domains (I and II) or subdomains (IIa and IIb) (27,28). Each protease core domain possesses a calcium-binding site that upon occupation cooperatively drives the catalytic triad into an active configuration (29). Recombinant expression of the proteolytic core of the -calpain isoform (I-II), as an isolated construct, results in a soluble calpain, which when compared with the full-length calpain has a similar specificity for cleavage of protein substrates and a similar response to active site-directed inhibitors (30). Although this mini-calpain has a much lower catalytic efficiency, unlike the native enzyme, it resists autolysis. Also, it maintains its solubility in the presence of calcium and so presents a useful model for determining the specificity of -calpain.
The results obtained in this study reveal that I-II possesses a general preference for hydrophobic amino acids at unprimed positions, which is consistent with previous literature. Also, a novel residue specificity for primed positions is revealed, one that is not apparent from studies of protein substrate cleavage sites. Specifically it shows a preference for methionine or alanine at P1Ј, glutamic acid at P2Ј, and basic amino acids at P3Ј. The optimal (MER) sequence derived from primed side specificity was validated by its ability to successfully direct cleavage to a specific bond within a partially degenerate peptide. The incorporation of the optimal cleavage sequence (PLFAER) in an internally quenched fluorescent substrate resulted in a significantly faster turnover of the substrate compared with analogous compounds incorporating other known calpain cleavage sequences.

EXPERIMENTAL PROCEDURES
Materials-Peptide libraries (see TABLE ONE) were obtained from the Tufts University Core Facility (Boston, MA). Typical yields were 100 -200 mg of crude peptide. The FRET 5 -based substrates were synthesized in the Peptide Synthesis Laboratory of the Protein Function Discovery Facility at Queen's University (Kingston, Canada). The reagents were obtained as follows: fluorescamine (Sigma), endoprotease Asp-N (Roche Applied Science), and Sephadex G25 (medium) (Amersham Biosciences). ␣-Cyano-hydroxycinnamic acid for use as a matrix in MALDI-TOF mass spectrometry was recrystallized from crude product obtained from Sigma. All other chemicals were obtained from Sigma-Aldrich. The recombinant protease core of -calpain, I-II, was purified from Escherichia coli as previously described (30).
Proteolytic Digestion of Libraries-The crude peptide libraries (see TABLE ONE) were dissolved in Me 2 SO prior to a 20-fold dilution to stock concentrations of 5-10 mg/ml with 20 mM MOPS-NaOH (pH 7.6). Digestion of the degenerate library by the protease Asp-N was performed at 37°C in 20 mM MOPS-NaOH (pH 7.6), using 0.4 mg/ml peptide library and 1.3 g/ml of Asp-N. All digestions by I-II were carried out at 37°C in 20 mM MOPS (pH 7.6), 1 mM EDTA, 6 mM CaCl 2 , 50 -100 mM NaCl, and 0.1% (v/v) 2-mercaptoethanol, using 50 g/ml final concentration of the protease I-II. The digestion was initiated by the addition of 0.4 mg/ml of the N-acetylated library or 1.0 mg/ml of the MAP-conjugated library. Control digests had either the peptide library or the calcium omitted. Digestion was stopped after a specific time period by adding 0.5 M EDTA (pH 8.0) to 11 mM final concentration. Digested samples were stored frozen when necessary.
Monitoring the Extent of Digestion with Fluorescamine-An aliquot (10 l) of 0.2% (w/v) fluorescamine in acetone was added to 100 l of digest and allowed to react for at least 5 min at room temperature. Fluorescence was read in a PerkinElmer Life Sciences LS-50-B fluorimeter using an excitation wavelength of 388 nm and an emission wavelength of 475 nm.
Determination of the Cleavage Profile at Primed Positions-The degenerate N-acetylated 12-mer library (see TABLE ONE) was digested by I-II as described above. After 0.5 h of digestion (corresponding to ϳ15% completeness, where 100% digestion refers to the fluorescence  C). A, a completely degenerate, N-acetylated, dodecameric peptide library is partially digested by the protease. Because the N-terminal fragments of cleaved peptides have blocked N termini, and those of the C-terminal fragments are free, sequential Edman degradation divulges the proportion of each residue found at each of the primed positions, with cycle y showing the proportion of residues found at position y-primed. B, the preferred consensus sequence for primed positions determined in A (MER) is included as fixed positions in the synthesis of a partially degenerate library possessing a C-terminal biotin tag. The fixed positions are used to direct cleavage to the Xaa-Met bond, so that partial digestion of the library results in N-terminal fragments that are primarily six residues long. After digestion, tagged C-terminal fragments and uncleaved peptides are removed using immobilized avidin. Unbound N-terminal fragments are then sequenced, yielding the abundance of each residue found at each unprimed position, with cycle y showing the proportion of residues found at position (7 Ϫ y)-unprimed. C, as an alternative method to that described in B, the partially degenerate peptide containing the preferred primed position motif (MER) is synthesized such that its C terminus is conjugated onto a MAP scaffold, yielding a peptide dendrimer. Partial digestion of this library results in N-terminal fragments primarily six residues long. Because these N-terminal fragments are of much smaller size than the dendrimer, they are isolated using size exclusion chromatography and are sequenced as described in B. The MAP scaffold, composed of oligomeric lysines linked by their backbone and side chain amino groups is shown. X refers to all 20 natural amino acids minus cysteine, Z refers to norleucine, and AHA refers to aminohexanoic acid. value at the asymptote for the digestion curve for I-II (see Fig. 2)), a 400-l aliquot was stopped by the addition of EDTA and lyophilized prior to N-terminal sequencing.
Determination of the Cleavage Profile at Unprimed Positions-A partially degenerate library conjugated to a MAP scaffold (see TABLE ONE) was partially digested by I-II as described above. A 0.5-h digest (0.5 ml) was quenched with EDTA and applied onto a 30-cm ϫ 0.6-cm column containing Sephadex G-25(M) gel permeation resin previously equilibrated in 50 mM ammonium bicarbonate. Fractions (0.4 ml) were collected and analyzed for peptide content using fluorescamine as described above. This gel permeation step was repeated a total of eight times to recover enough N-terminal fragments for Edman degradation. The fractions containing the N-terminal fragments (fractions 12-15 inclusive) from all eight runs were then pooled, concentrated by lyophilization, and reapplied onto the column to remove any contaminating scaffold. The pool of N-terminal fragments was then lyophilized to dryness, with three cycles of resuspension in water to remove traces of residual ammonium bicarbonate that could interfere with the peptide sequencing.
Peptide Analysis-Edman degradation of samples was performed either at the Alberta Peptide Institute (Edmonton, Canada) or the Advanced Protein Technology Center (Hospital for Sick Children, Toronto, Canada). The HPLC profiles resulting from phenylthiohydantoin-derivative analysis were calibrated using 10 pmol of each phenylthiohydantoin-derivative standard. For the N-acetylated library, the background resulting from a small percentage (ϳ5%) of unacetylated peptides was obtained by sequencing an undigested aliquot and subtracting the values obtained from the sequencing results of the digest. No background subtraction was necessary for the MAP-conjugated library. The selectivity of a residue at any one position is defined as the ratio of the abundance of the residue observed at the position to the abundance of the residue expected in the absence of any specificity. The latter was determined by amino acid analysis (Advanced Protein Technology Center) for the N-acetylated library or from the N-terminal sequencing of an aliquot of undigested library for the MAP-conjugated library. The values were not calculated for tryptophan because of its instability during amino acid analysis and sequencing conditions. Digestion analysis by MALDI-TOF mass spectrometry was performed at the Protein Function Discovery Facility (Queen's University, Kingston, Canada) on a M@LDI (Micromass) instrument. For sample preparation, a 1.0-l volume of freshly prepared and recrystallized ␣-cyanohydroxycinnamic acid matrix was overlaid with 1.0 l of a digestion sample. The spectra shown are a sum of 100 shots, with peak signals of 3-95% intensity collected.
Kinetic Analysis of FRET-based Substrates-The crude FRET-based peptides were purified by reversed-phase HPLC on a C18 column. The

IEF Correction factor
The reference contained a low concentration of substrate that had been verified to show an insignificant inner filter effect (IFE) (usually 10 M). The I o value was determined by extrapolation of the linear portion of the reaction curve back to the time of initiation of the reaction. Under these conditions, a typical correction factor for 50 M of a FRET-based substrate was ϳ1.4. The fluorescence coefficients to convert changes in fluorescence to changes in concentrations of cleaved substrate were determined by allowing the hydrolysis of a known amount of substrate to go to completion using a vast excess of I-II. The turnover rate was obtained by dividing the molar rate of substrate hydrolysis (M/s) observed at each substrate concentration by the enzyme concentration used.

RESULTS
Specificity Profile at Primed Positions-To obtain the specificity profile at primed positions, an N-acetylated, completely degenerate peptide library (TABLE ONE) was digested and subjected to N-terminal Edman degradation to reveal the proportion of residues found at each primed position (Fig. 1A). This method was validated and calibrated using a protease with a strict specificity at primed positions. Digestion of the library with the endoprotease Asp-N yielded almost exclusively Asp residues in the first cycle of sequencing (results not shown), agreeing with the known specificity of the protease (31).
Digestion of the library was monitored by observing the production of N termini using fluorescamine, which reacts with primary amino groups (including those of the ⑀-amino group from lysine side chains) to form a fluorescent product (32). The calcium dependence of this diges-tion, demonstrated by the production of new N termini in the presence of CaCl 2 but not in its absence, establishes that cleavage results from the action of I-II and not a co-purifying E. coli protease (Fig. 2). Because Asp residues should occur at a frequency of approximately one residue for every two degenerate peptides, the comparison of the fluorescence signal arising from the complete digestion of the peptide library by Asp-N with the signal from digestion by I-II allows an estimate of Ն0.5 cuts/peptide catalyzed by I-II within the first 12 h.
To select for kinetically favored cleavages, an early time point from the digestion (0.5 h, equivalent to 15-20% completeness) was subjected to N-terminal sequencing (Fig. 3A). The sequencing results show a strong selection for Ala (selectivity, 2.8) and Met (4.1) in the first sequencing cycle, indicating a preference for these residues at position P1Ј of a peptide substrate. The second cycle, representing position P2Ј, shows a preference for the residues Glu (selectivity, 2.1), Met (selectivity, 2.3), and Ala (selectivity, 2.0). Incomplete cleavage in the first cycle may partially explain the apparent selection for Ala and Met in the second cycle, whereas Glu shows a distinct 3-fold increase in selectivity from the first to the second cycle. This indicates a genuine preference for Glu at position P2Ј. The third sequencing cycle features a selectivity for Arg (selectivity, 2.9), with Lys (selectivity, 2.1) also ranking high, suggesting that the residue at P3Ј may interact with the protease-binding subsite through ionic interactions with acidic residues. The fourth and fifth cycles show no distinct preference for a residue (specificity shown as the selectivity ratio. The selectivity ratio is defined as the ratio of the abundance of the residue observed at the position to the abundance of the residue expected in the absence of any specificity. The latter was obtained as described under "Experimental Procedures." A, primed side specificity obtained from the N-acetylated library as described in Fig. 1A. The error bars show the standard deviation of the results from two separate digests. B, unprimed side specificity obtained from the MAP-conjugated library as described for Fig. 1C. ratio Ͻ 1.8 for all residues), apart from a general overabundance of residues bearing hydrophilic side chains relative to those bearing hydrophobic ones (data not shown), indicating no specificity for positions more distal than P3Ј.
Specificity Profile at Unprimed Positions-To obtain the specificity profile for unprimed positions, the consensus motif for the first three primed positions, Met-Glu-Arg, was used to direct cleavage in a second library to the scissile Xaa-Met bond to sample the unprimed residues that help direct cleavage to this point. If successful, this directed cleavage would result in N-terminal hydrolysis fragments that are primarily six residues long, thereby linking the cycles of N-terminal sequencing of these fragments to unprimed positions in the substrate (Fig. 1, B and C).
To minimize the formation of secondary structure within the peptides and ensure that the peptides are in an extended conformation, a Pro-Gly (Fig. 1B) or a Pro-Gly-Pro (Fig. 1C) motif C-terminal to the fixed positions was included in the library. This motif also served the purpose of preventing cleavage from occurring C-terminal to the fixed Arg, because proline residues have a restrained backbone conformation such that the Xaa-Pro peptide bond is not cleaved.
The original methodology for determining sequence specificity at unprimed positions used a biotinylated library to allow for the isolation of the N-terminal fragments by the removal of C-terminal fragments and uncleaved peptides on immobilized avidin (Fig. 1B) (25,26). Because the biotin moiety may be damaged during or post synthesis, the peptides possessing an "active" biotin required preselection on monomeric avidin prior to digestion by I-II. However, even after this preselection, poor efficiency in the removal of the biotinylated peptides from the digest was observed. This resulted in strong background signal upon sequencing of the flow-through from the immobilized avidin (results not shown). This portion of the methodology was modified to circumvent the avidin-biotin affinity chromatography step by synthesizing the peptide moiety onto a MAP scaffold to generate a peptide dendrimer (Fig. 1C) (33). In this approach, the N-terminal fragments (average mass, 700 Da) are of significantly smaller size than the C-terminal fragments conjugated to the MAP scaffold (average mass, 12 000 Da), allowing for their isolation by gel permeation chromatography (Fig. 4). To ensure that the protease would have access to the scissile bond within the context of the dendrimer, an aminohexanoic acid spacer was inserted between the scaffolding and the peptide moiety (Fig. 1C).
Gel permeation chromatography of the MAP-conjugated library resulted in early eluting, high molecular mass fragments corresponding to peptides conjugated to the MAP scaffold (Fig. 4, squares). Upon partial (0.5 h) digestion of the library, a second, late eluting peak became apparent (Fig. 4, diamonds). MALDI-TOF analysis confirmed that these smaller molecular mass fragments correspond to degenerate hexameric N-terminal fragments (average molecular mass, ϳ750 Da) (Fig. 5). The importance of selecting an early time point in this methodology is illustrated by the profile arising from a more extensive digestion (12 h), whereupon the fragments are shifted to a lower apparent molecular mass, suggesting further processing of these hexamers to undesired smaller fragments (Fig. 4, circles).
A number of observations confirm that cleavage is efficiently directed to the Xaa-Met peptide bond. First, the peptide corresponding to the C-terminal fragment resulting from hydrolysis at the Xaa-Met bond of the biotinylated library (MERPGK-biotin) was shown by MALDI-TOF mass spectrometry to be enriched relative to the other C-terminal fragments (ERPGK-biotin and RPGK-biotin) after both 0.25 and 2 h of digestion, supporting cleavage occurring primarily at this bond relative to cleavage within the other fixed positions (Fig. 6). Second, N-terminal sequencing of an unseparated digest (prior to avidin treatment) of the same library yielded primarily the residues Met, Glu, and Arg for its first, second, and third sequencing cycles, respectively (results not shown).   Third, the N-terminal fragments isolated from digestion of the MAPconjugated library show an m/z envelope centered at ϳ750 Da, corresponding to the presence of primarily degenerate hexamers (Fig. 5).
The sequencing results for these fragments feature a general propensity for hydrophobic residues at the unprimed positions (Fig. 3B). A strong preference for Leu is observed in the fourth, fifth, and sixth sequencing cycles, corresponding to positions P3, P2, and P1, respectively. Superimposed on these results is a selectivity for the bulky aromatic residues Phe (selectivity, 3.0) and Tyr (selectivity, 1.6) at P1 and ␤-branched residues Val (selectivity, 2.4), Ile (selectivity, 1.9), and Thr (selectivity, 1.4) at P2. The second, third, and fourth sequencing cycles reveal a propensity at positions P5, P4, and P3 for Pro (selectivity, 2.2, 1.6, and 1.7, respectively) as well as Phe (selectivity, 1.6, 2.1, and 1.9, respectively). The possibility that the observed preference for bulky amino acids at P1 includes tryptophan is not excluded, because tryptophan could not be observed due to its instability during Edman degradation sequencing.
Overall specificity, as defined by the highest observed selectivity at each position, was observed to extend farther from the cleaved bond (up to P5) for the unprimed positions relative to the primed positions (up to P3Ј). A summary of the preferred residues observed for each position P5-P3Ј is presented in TABLE TWO.
Cleavage Kinetics of the FRET-based Substrates-To validate the optimal cleavage sequence obtained from the library-based approach and to compare it with previously reported calpain substrates in terms of cleavage kinetics, a series of peptidyl substrates based on FRET (34) were synthesized. These peptides were synthesized such that an N-terminal FRET donor (EDANS conjugated to an N-terminal Glu side chain) and quencher (DABCYL conjugated to a C-terminal Lys side chain) are separated by a hexapeptide sequence (Fig. 7). Hydrolysis of these internally quenched peptides results in the relief of fluorescence quenching by DABCYL and a consequent 40 -60-fold increase in fluorescence of EDANS (data not shown) that allows for real time monitoring of the hydrolysis.
FRET-based peptides can be sensitive protease substrates when used at low concentrations but are limited in their use for the determination of kinetic parameters at high micromolar concentrations. This limitation is due to the IFE, which is amplified in FRET-based compounds by the presence of two chromophores with overlapping spectra. The IFE causes an exponential decrease in the observed fluorescence intensity with increasing substrate concentrations and can therefore be misinterpreted as saturation of the protease if not taken into account. IFE corrections were applied to the fluorescent substrates in our assays but only hold at low M concentrations. We therefore restricted our kinetic analysis of the FRET-based substrates to the determination of I-II turnover (the rate at which one enzyme molecule hydrolyzes one substrate molecules) using substrates at lower concentration.
The hexapeptide sequence for the substrate incorporating the preferred cleavage motif (PLFAER) was designed to minimize secondary cleavage while maximizing solubility. Proline was selected over phenylalanine and leucine at P3 because it improves solubility and likely maintains the peptide in an unstructured conformation. Alanine was selected over methionine at P1Ј because the latter is susceptible to oxidation and decreases solubility.
On monitoring substrate cleavage by fluorimetry, this optimized substrate was found to be turned over more than eight times more rapidly (Fig. 8) than an analogous substrate incorporating the central hexapeptide of the consensus sequences (PLKSPP) described by Tompa et al. (14), and more than 18 times more rapidly than the substrate incorporating a cleavage site of a naturally occurring calpain substrate (␣-spectrin, EVYGMM) (23) when present at a concentration of 10 M. The commercially available amino methyl-coumarin-based substrate SLY-MCA was cleaved at an even lower rate than the three FRET-based substrates, (Ͼ310-fold slower than PLFAER at 10 M) (16). The use of SLY-MCA as an analytical substrate is partially vindicated by the   10 -12-fold higher fluorescence yield of the amido-methylcoumarin group of SLY-MCA compared with the EDANS fluorophore used in the FRET-based substrates (data not shown). All three FRET-based substrates demonstrated a linear increase in turnover rate by I-II with a concentration up to at least 50 M (Fig. 8), indicating subsaturating conditions. SLY-MCA, which shows comparatively little IFE, showed a linear relationship up to at least 2 mM (not shown).

DISCUSSION
The cleavage specificity of calpains appears to be only weakly dictated by the primary sequence of its protein substrates. Rather, other structural factors within the substrate, such as the necessity for an unstructured backbone region or possibly a specific three-dimensional conformation, appear to be more important in defining the precise cleavage sites (18,21). However, if the cleavage is restricted to within an unstructured region, a situation that is representative of the hydrolysis of small synthetic substrates, calpains will presumably cleave preferentially at positions where the sequences flanking the scissile bond contain the most favorable substrate-subsite interactions.
The broad specificity of calpains for the hydrolysis of peptides is reflected in the representation of all residues to various extents at every substrate position investigated (with the possible exception of proline at P1Ј) (Fig. 3, A and B), as well as in the extent of cleavage of the fully degenerate library by I-II (0.5-1.0 cuts/peptide after 12 h). Note that digestion of the library by Asp-N, a protease of strict specificity, rapidly reaches an end point, whereas with I-II the extent of digestion approaches an asymptote much more slowly, as though a wide range of cleavage rates contribute to the digestion (Fig. 2).
The unprimed side specificity obtained from our approach agrees fairly well with previous reports using small synthetic inhibitors and substrates. Consistency is maintained with the observation by Sasaki and colleagues (13) that synthetic chloromethyl ketone inhibitors show an inactivation potency with a residue preference at P1 in the order of Phe Ͼ Tyr Ͼ Lys and that small MCA-derived substrates are cleaved preferentially with Leu Ͼ Val at P2 and Tyr Ͼ Met at P1 (16). Current commercially available peptidomimetic calpain inhibitors that use unprimed subsite interactions to dock the inhibitor, such as calpain inhibitors I-VI, contain residue-subsite combinations that reflect our results (Fig. 3B), with large hydrophobic residues Met (calpain inhibitor II), Phe (calpain inhibitor III), Tyr (calpain inhibitor IV), or Leu (calpain inhibitor VI) at P1; Val (calpain inhibitors III, V, and VI) or Leu (calpain inhibitors II and IV) at P2; and Leu (calpain inhibitors I, II, and IV) at P3 (structures available on the Calbiochem website). These peptidomimetic inhibitors are therefore close to optimal with respect to the recognition of natural amino acids by the unprimed subsites of calpain.
However, the cleavage specificity obtained here using peptidyl substrates is different from the consensus motif obtained from the tabulation of known cleavage sites of protein substrates (14). Most notably, within the cleaved sequences of identified protein substrates there is a large proportion of basic residues at P1 and small hydrophilic residues at P1Ј and a paucity of acidic or basic residues at positions P2Ј or P3Ј (14). A possible explanation for the discrepancies at unprimed positions, particularly for positions closest to the cleaved bond, is that the fixed primed sequence (Met-Glu-Arg) influences the selectivity at unprimed positions. However, a more global explanation is that the primary sequence surrounding the cleavage site of a protein substrate does not need to be similar to an optimized cleavage sequence because higher order structure recognition may orient, or favor transient substrate binding, to an extent sufficient to direct cleavage to a specific bond. This would be independent of whether the primary sequence surrounding the cleavage site is optimized for cleavage by calpain (18). A simple scenario illustrating this point is where a small, exposed, and unstructured loop in a protein is cleaved by active calpain. Cleavage could occur within this loop more or less independently of the specific residues present in the loop as it fulfills the requirements of being unstructured and exposed, whereas the kinetics of the cleavage might be affected by the sequence. Indeed, modifying the primary structure surrounding the scissile bond would be an evolutionarily simple way of regulating the relative susceptibility of specific substrates to cleavage by calpain by affecting the kinetics of cleavage without necessarily changing the occurrence or site of cleavage. An apparent exception to this "suboptimal cleavage" hypothesis appears at position P2, where Leu, Val, and Thr are selected, both kinetically within peptide substrates (Fig. 3B) and within the consensus obtained from protein substrate sequences (14), which suggests that this position may play an important role in dictating the site of cleavage in proteins. Consistent with this central role of the residue at position P2, mutation of the hydrophobic residue at position P2 of the major cleavage site of both ␣-spectrin and talin to a glycine blocks their susceptibility to calpain (18,35).
A moderate preference for proline is found at positions distal to P2(Ј) (Fig. 3). Structural analyses of protease-substrate analogue complexes have revealed that most protease substrates bind to the active site in an extended (␤-strand) conformation (36). Because prolines facilitate extended peptide conformations, we hypothesize that the predilection toward proline at these distal positions is due to this conformational effect rather than to primary sequence recognition by the protease. This effect may be of particular importance in calpains because of the presence of a deep active site cleft (37) that substrates have to enter prior to encountering the active site.
The lower selectivity observed at positions more distal to the cleaved bond (P3(Ј)-P5(Ј)) may be partially interpreted as an artifact arising from the use of a proteolytic construct that is of smaller size than the native protease. However, the similarity of the residue preference obtained using mini-calpain at positions P1 and P2 with that of previous kinetic studies using small substrates and inhibitors obtained using native calpain (13,16) help establish that the subsites in the mini-calpain, especially those close to the active site, are not significantly altered from those in the whole enzyme.
Small, synthetic substrates have been reported to generally be inferior substrates for calpains in comparison with proteins/ polypeptides (16,38) and natural peptides. This observation is corroborated with the popular calpain substrate SLY-MCA, demonstrating a Ͼ17-fold lower turnover rate than the FRET-based substrate with the lowest turnover (EVYGMM) (Fig. 8). Presumably, this large difference is due to the lack of primed site interactions in the former compound, designed to interact only at the P1 and P2 positions (16), whereas our FRET-based substrates, similarly to natural proteins and peptides, use both primed and unprimed positions within the enzyme recognition region of the substrate. Because most synthetic methylcoumarin-based substrates use unprimed positions only, by extension they should be expected to be kinetically inferior substrates compared with FRET-based substrates.
By revealing an optimal cleavage sequence Met/Ala-Glu-Arg for P-primed positions (Fig. 3A), our results confirm the general observation of other groups (14,23) that the primed subsites of calpains are significantly involved in peptide recognition. The importance of primed side residues is further highlighted by the FRET-based substrates containing the optimal sequence obtained (PLFAER) and the consensus sequence from known cleavage sites (PLKSPP). These substrates, despite having identical P3 and P2 residues, show a 6 -8-fold difference in their turnover rate at 10 M substrate (Fig. 8), strongly supporting the important influence of P-primed residues in the kinetics of cleavage by calpains.
The crystal structure of calcium-bound I-II (Molecular Modeling Database accession number 1KXR) lends some support to the residue preferences obtained. On the unprimed side of the substrate-binding pocket, a relatively narrow cleft forms the S1 and S2 subsites, whereas subsites S3-S5 are more open, consistent with the lower specificity observed at positions P3-P5 relative to P2 and P1 (Fig. 3). A tightly fitting binding pocket at the S2 subsite, shown to accommodate the Leu 2 moiety of leupeptin in the inhibitor-bound structure (37), supports the accommodation of other ␤-branched side chains such as Val, Thr, and Ile, as well as the ␥-branched side chain of Leu. The narrow cleft found at the S1 subsite can accommodate relatively planar residues such as Phe and Tyr (Fig. 3B). The subsites on the primed side of the substrate-binding pocket are more exposed, consistent with a preference for polar and charged residues on this side of the cleaved bond (Fig. 3A). In accordance with our results, residue Lys 79 , which is within one of the flexible loops gating the active site (37), could easily be placed in a position to interact with a Glu at P2Ј. In addition, the S3Ј subsite is in proximity to the carboxylate group of Glu 300 that could favorably interact with Arg or Lys at P3Ј.
In humans, the calpain family includes two major, ubiquitous isoforms, m-and -calpain, that have 63% sequence identity and a dozen other "tissue-specific" isoforms, some of which show distinct expression patterns and roles within the cell (39). The library-based approach to studying cleavage specificity used here presents a potentially useful method to identify differences in the specificity of the many isoforms. Any such differences could be applied in the design of isoform-specific substrates and inhibitors.
In summary, we have improved upon the peptide library-based approach described by Turk and Cantley (26) for determining the cleavage specificity of proteases at P-unprimed positions by synthesizing the degenerate library onto a peptide dendrimer scaffold to simplify the separation of N-terminal and C-terminal peptidyl fragments. This innovative approach was used to identify the kinetically preferred cleavage sequence of -calpain at positions flanking the scissile bond. This sequence was applied in the design of a FRETbased substrate that is significantly more sensitive to hydrolysis by -calpain than both the consensus sequence identified from the literature (14) and a sequence obtained from an individual natural substrate (␣-spectrin). This sequence differs significantly from sequences observed in natural protein substrate cleavage sites at almost all residue positions except for P2, suggesting that calpain substrates in the cell have sequences around the cleavage sites that have been designed to be suboptimal, perhaps as a mechanism to control substrate susceptibility.