De Novo Design of Peptides Targeted to the EF Hands of Calmodulin*

This report describes the use of the concept of inversion of hydropathy patterns to the de novo design of peptides targeted to a predetermined site on a protein. Eight- and 12-residue peptides were constructed with the EF hands or Ca2+-coordinating sites of calmodulin as their anticipated points of interaction. These peptides, but not unrelated peptides nor those with the same amino acid composition but a scrambled sequence, interacted with the two carboxyl-terminal Ca2+-binding sites of calmodulin as well as the EF hands of troponin C. The interactions resulted in a conformational change whereby the 8-mer peptide-calmodulin complex could activate phosphodiesterase in the absence of Ca2+. In contrast, the 12-mer peptide-calmodulin complex did not activate phosphodiesterase but rather inhibited activation by Ca2+. This inhibition could be overcome by high levels of Ca2+. Thus, it would appear that the aforementioned concept can be used to make peptide agonists and antagonists that are targeted to predetermined sites on proteins such as calmodulin.

Accumulating evidence suggests that a simple binary code of polar and nonpolar amino acids arranged in the appropriate order is sufficient to build helical bundle structures and artificial peptides with rudimentary function (for review see Ref. 1). Therefore, only the sequence location, not the identity, of the polar and nonpolar amino acids must be explicitly specified for the formation of a stable helical structure or biologically active peptide. Such binary coding has been successfully employed to produce biologically active analogs of corticotrophin and growth hormone-releasing hormone (2,3), to design proteins that fold into compact ␣-helical bundles (4), and to develop computer programs that simulate or predict some aspects of protein folding (5). Considering that 20 different amino acids are encompassed by the binary code, one would expect a marked degree of sequence degeneracy for a given shape, since any one of a number of specific polar or nonpolar amino acids could occupy a given position in the sequence. Indeed, experi-mental evidence has confirmed that gross shape is degenerate with regard to sequence in that any number of different primary amino acid sequences with the same binary code can fold into compact ␣-helical structures (4).
More recent studies have shown that, unlike simple helical structures, a five-letter amino acid alphabet is minimally required to build a well ordered, ␤-sheet containing protein architecture (6). In this particular instance, a small ␤-sheet protein, the SH3 domain, could be constructed with 95% of the residues being Ile, Lys, Glu, Ala, and Gly. Interestingly, the pattern of hydropathy of the wild type SH3 domain was largely maintained in the two sequence-simplified structures. 1 This would seem once again to point to the importance of the pattern of hydropathy in building more complex structures as well as simple helical ones.
If, as described above, the gross architecture of a peptide or protein is in part determined by its pattern of hydropathy, then exactly inverting a particular pattern or code may result in a second peptide or protein with a complementary surface contour to the first since the hydrophobic effect is involved yet in reversed orientation (for review see Ref. 1). Inversion of the hydropathic pattern of one sequence relative to another can be achieved by computer programs designed for this task (7)(8)(9) or by simple reliance on an interesting characteristic of the genetic code (10). In the later instance, since A and U are complementary and in the second codon position specify hydrophilic and hydrophobic R groups, respectively, and considering that second base G and C generally encode slightly hydrophilic R groups, amino acid sequences deciphered from noncoding strands of DNA will have exactly inverted patterns of hydropathy relative to those of coding strands (for review see Ref. 1).
Such peptides specified by complementary nucleotide sequences (11) or designed by simply inverting the hydropathic pattern (8) are termed complementary peptides and have characteristics suggestive of complementary structure. For instance, we and many others (for review see Refs. 1 and 12) have shown that in almost 40 different systems complementary peptides bind one another with specificity and moderate affinity. Additional evidence of complementary structure includes the following: the ability to locate the interactive sites of ligands and receptors by identification of complementary sequences (13)(14)(15)(16)(17); to generate interacting pairs of monoclonal idiotypic and anti-idiotypic antibodies with complementary combining sites by immunization with pairs of complementary peptides (18 -27); to produce antibodies to receptor-binding sites by immunization with complementary peptides for the receptor's ligand (11,18,(27)(28)(29)(30)(31)(32)(33)(34). Most recently, a novel hormone receptor was cloned, and its binding site was localized using this principle (17). Not unexpectedly, the aforementioned degeneracy that is observed for the relationship between sequence and structure applies to the complementary sequence since the same codes are operative. Thus a number of complementary peptides that are properly patterned are predicted to bind the appropriate target sequence. This has been previously observed (35)(36)(37).
The above results suggested that one could specifically target a complementary peptide to interact with a particular site(s) on a given protein. This was initially tested with an eight-residue complementary peptide (termed calcium-like peptide, CALP1) designed to interact with an EF hand motif based on the primordial Ca 2ϩ -binding site of the troponin C superfamily (38). Micromolar concentrations of CALP1 together with CaM, 2 but neither CALP1 nor CaM alone, were found to activate phosphodiesterase in the absence of Ca 2ϩ . This demonstrated that the combination of CALP1 and CaM functioned similarly to a combination of Ca 2ϩ and CaM. These results implied but did not prove that CALP1 directly interacted with CaM. Thus in the present studies we have sought evidence for direct binding of CALP1 to CaM, and we tested whether this occurs at or near the EF hands of CaM. Also, as a consequence of the CALP1 design being based on the primordial EF hand motif (38,39), the hydropathic profile of CALP1 is not perfectly inverted relative to the EF hand motifs of CaM. In the present study, we have designed a complementary peptide to invert more perfectly the pattern of hydropathy and to extend the sequence length to determine the role of these two factors on the interaction with CaM. Specifically, we synthesized a 12mer peptide (CALP2; Table I) with an optimized and extended inverted hydropathic profile to EF hand motif 4 of human CaM using AMINOMAT® (Tecnogen, Italy) (40), a computer program able to determine the optimal complementary peptide sequence to a target. In order to distinguish if any increased affinity was due to the optimal hydropathic pattern or to the increased length or both, we also synthesized CALP3, which represents the first eight residues of CALP2, and CALP4, a 12-mer peptide identical to CALP1 in the first eight residues and CALP2 in the last four (Table I). We tested the binding characteristics of these peptides for CaM using three different approaches as follows: inhibition of complex formation between CaM and the EF hand-binding dye, Stains-all; changes in the emission spectra of dansylated CaM to monitor conformational changes; and real time measurement of the interaction with a biosensor based on surface plasmon resonance detection. We also characterized certain of the interactions with nuclear magnetic resonance spectroscopy and tested for CaM activation with a CaM-stimulated phosphodiesterase (PDE) assay. Finally, we tested the generality of the interaction by assessing the ability of CALP2 and -3 to bind the EF hands of another Ca 2ϩ -binding protein, troponin C.

EXPERIMENTAL PROCEDURES
Precautions to Eliminate Ca 2ϩ Contamination-In all the experiments, buffers were treated with the chelating resin Chelex 100. Stock solution of the peptides were checked for Ca 2ϩ contamination by atomic absorption. Levels were always below 2 M Ca 2ϩ /mM CALP. Thus, Ca 2ϩ contamination was at most 200 nM and was usually 2-20 nM. These levels are below those required to affect any of the assays.
Reagents-Preloaded polyethylene glycol graft polystyrene, O-pentafluorophenyl ester amino acid, and 1-hydroxy-7-azabenzotriazole were purchased from Perspective Biosystems (Framingham, MA). High purity CaM from bovine brain and biotin-labeled calmodulin were purchased from Calbiochem. Its purity was checked by SDS-polyacrylamide gel electrophoresis and the identity verified by electrospray ionization-mass spectroscopy (University of Alabama at Birmingham Core Facility). Neutravidin was from Pierce. D-CaM, PDE, MOPS, Chelex 100, and Stains-all were from Sigma. N,N-Dimethylformamide, DCM, and acetonitrile were purchased from Mallinckrodt (Paris, KY). Mant-cGMP was from Molecular Probes (Eugene, OR). All other reagents were high purity grade obtained from Fisher or Sigma.
Design of the Hydropathically Complementary Peptides-Selection of the complementary peptide targeted to CaM EF hand 4 motif was carried out as described previously (40) using the computer program AMINOMAT® (Tecnogen ScpA, Italy), with an averaging window r ϭ 9, a range of inverted hydropathy of 0.8 and considering also the eight amino acids of the flanking regions. The program generated 1,417,176 possible sequences, and we chose the one with the lowest Q value (0.0068). This process resulted in a 12-residue peptide defined CALP2. CALP3 was generated by eliminating the four carboxyl-terminal amino acids of CALP2. CALP4 is a chimeric sequence carrying the eight amino acids of CALP1 with the addition at the carboxyl terminus of the four amino acids from the corresponding region of CALP2 (Table I).
A scrambled version of CALP2 and peptides with sequences not related to the CALP series were used as negative controls.
Peptide Synthesis-All the peptides were synthesized in our laboratory using continuous flow solid phase peptide synthesis with Fmoc (N-(9-fluorenyl)methoxycarbonyl) chemistry, on a PerSeptive Biosystems 9050 Peptide synthesizer. Pre-activated O-pentafluorophenyl ester amino acids with 1-hydroxy-7-azabenzotriazole and preloaded polyethylene glycol graft polystyrene resin was used. All the peptides were purified by preparative RP-HPLC. The purity of the product was checked by analytical RP-HPLC. The identity of the peptides was con-   Surface Plasmon Resonance Detection Studies-Direct measure of the affinities of the peptides for CaM were obtained using a surface plasmon resonance detector (IAsys, Affinity Sensor, Cambridge, UK). A carboxymethylated dextran cuvette was prepared immobilizing Neutravidin using an NHS/EDC activation procedure according to the manufacturer's protocols (IAsys protocol 1.1) This was followed by the attachment of biotin-labeled CaM to the Neutravidin on the cuvette. Binding buffer was 20 mM MOPS, 150 mM NaCl, pH 7.4, containing 0.05% Tween 20 treated with Chelex 100. K d values were calculated using the software FASTfit, supplied by the manufacturers for the analysis of the binding curves. For the association, the curves were fitted to a single rate constant. For the dissociation we also used a single rate constant.
Fluorescent Troponin C-The rabbit slow muscle troponin C mutants (C84S/Y111W, C84/R147W, and C84S/Y111W/R147W) were a gift of Dr. Herbert Cheung, and their preparation and characteristics are discussed elsewhere. 3 Steady state fluorescence for peptide-protein interaction were conducted using a ISS PC1 photo-counting fluorometer using a 0.1-mm slit. Protein was solubilized at 1 M in 50 mM MOPS, 100 mM KCl, 50 M EDTA, pH 7.2. Peptides were solubilized at a 3 mM stock concentration in this protein containing solution. By using 295 nm as the excitation wavelength, we measured the emission spectra between 300 and 400 nm with increasing concentration of the peptides. Spectra were recorded 5 min after addition of the peptide stocks. Acrylamide quenching of CALP2⅐TnC complex (75 and 1 M, respectively) and of TnC (1 M) were conducted in the same buffer, by the addition of aliquots of an 8 M acrylamide solution with ex ϭ 295 nm and em at the peak of the emission spectrum. In the range of acrylamide concentrations used (0 -400 mM), we observed a linear relationship between tryptophan emission decay and quencher concentration. The quenching data were fitted to a modified Stern-Volmer equation, where F 0 and F are the fluorescence intensities in the absence of and presence of quencher, respectively; [Q] is the molar quencher concentration; and K SV is the Stern-Volmer dynamic quenching constant.
Time-resolved fluorescence measurements were conducted as previously reported (42). Transient kinetic measurements were performed on a Hi-Tech Scientific PQ/SF-53 stopped-flow spectrometer equipped as described (43). In a typical binding experiment, one syringe contained 10 M C84S/R147W TnC or C84S/R147W TnC together with 50 M CALP2 in 30 mM MOPS, 100 mM KCl, and 50 M EGTA, pH 7.2. The protein was dialyzed against Chelex 100 resin before use. The other syringe contained 10 mM Ca 2ϩ in the same buffer.
Circular Dichroism-CD studies were conducted on a 62 DS VCD spectrometer AVIV (Lakewood, NJ) using a 0.01-cm quartz cuvette, with a scanning range of 195-310 nm, a time constant of 2 s, and a scanning rate of 0.5 nm/s. The buffer used was 20 mM phosphate, pH 7.0 CaM-stimulated Phosphodiesterase (PDE) Assay-Measurements were obtained using a continuous fluorimetric assay based on Mant-cGMP. Mant-cGMP, a fluorescent derivative of cGMP, shows maximal fluorescence emission at 450 nm. Hydrolysis of Mant-cGMP by PDE in the presence of CaM and Ca 2ϩ causes a decrease in the emission spectra of the compound. The enzymatic reaction can be followed in real time using 280 nm as excitation wavelength and monitoring the emitted light at 450 nm. Experiments were conducted using an excess of the substrate at fixed concentrations of CaM and PDE with constant mixing. By using these conditions the V max of the reaction can be calculated by the slope of the rectilinear part of the enzymatic reaction curve. Mant-cGMP was dissolved in MeOH (1.6 mM) and kept on ice. The reaction was conducted in 0.01 M MOPS, 0.09 M KCl, 5 mM MgCl 2 , 0.1 mM EGTA, pH 7, at 37 Ϯ 2°C. Mant-cGMP was diluted in the reaction buffer (8 M final concentration) with CaM (5 nM) and the different peptides or free Ca 2ϩ . The system was equilibrated for 5 min. The reaction was followed using a PTI fluorescence instrument. The rate of the reaction was calculated with a linear regression of 150 data points after addition of PDE. Excitation wavelength was 280 nm, and emitted light was monitored at 450 nm, collecting 30 data points/min. The reaction was started by addition of PDE (0.05 units/ml). For Fig. 10, the percent inhibition was calculated using the formula: 100 is the rate of degradation of cGMP for a given Ca 2ϩ concentration in presence of the peptides, V max(no peptide) is the rate of degradation for the same Ca 2ϩ concentration without the peptides, and V max(no CaM) is the basal rate without CaM, defined as the basal activity of PDE.
Production of 15 N CaM and 15 N TRC1 and TRC2-Escherichia coli strain BL21 (DE3) harboring plasmid pRK7a for Drosophila melanogaster CaM was obtained from Dr. K. Beckingham (Rice University). Unlabeled CaM was expressed in cells grown in Terrific Broth, and 15 N-labeled CaM was purified from cells grown in M9 medium using 15 NH 4 Cl as the sole nitrogen source. CaM (labeled and unlabeled) was purified by repeated chromatography on phenyl-Sepharose CL4B followed by ion exchange chromatography on DEAE-cellulose using a linear gradient (0 -0.5 M) of sodium chloride in 0.05 M sodium phosphate, pH 5.7. Under these conditions, CaM is obtained as a single peak, eluting at ϳ0.35 M NaCl. 15 N-Labeled CaM tryptic fragments, TRC1 (amino-terminal domain) and TRC2 (carboxyl-terminal domain), were prepared as described previously (44,45) except the purification of the two fragments was done by RP-HPLC on a Waters Delta Pack C18 300 A (300 ϫ 39 mm inner diameter) column. The fragment identity was check by time of flight matrix-assisted laser desorption ionization mass spectrometry. NMR chemicals shift were consistent with the literature (46).
NMR Experiments-Apocalmodulin samples were prepared by dissolving the sample in 3 ml of water and dialyzed against 2 changes of 1 M EDTA, 1 mM urea, pH 7.8, (2 times) followed by 0.5 M KCl (2 times), 0.25 M KCl (2 times), and deionized, distilled water (3 times). NMR samples were prepared by dissolving the protein sample in either 100% D 2 O (Cambridge Isotopes, MA) or 90% H 2 O, 10% D 2 O. The pH was adjusted using 0.1 mM DCl or 0.1 mM NaOD solutions. Concentrated peptide sample (2.0 mM CALP1 or 4.0 mM CALP3) solutions were prepared, the pH was adjusted and then titrated into the protein samples.
One-dimensional 1 H NMR spectra were acquired using a 12 ppm spectral width and presaturation of the HOD resonance. These spectra were acquired on either a Bruker WH400 or a Bruker AM600 instrument. The spectra were processing by Felix (MSI, San Diego, CA). Two-dimensional 1 H-15 N HSQC (47) data were acquired on the Bruker AM600 instrument. The spectra were acquired with 12 ppm spectral width, 2048 complex t 2 points, and 512 t 1 points. All measurements were made at 298 K. Spectra were processed using Felix 95.0 (MSI, San Diego, CA), by zero filling to 2 by 2K data sets. Solvent suppression and first point correction were used for each spectra. A 45°s hifted sinebell apodization function was applied in each dimension.

RESULTS
Surface Plasmon Resonance Detection-By using a biosensor based on surface plasmon resonance detection, we were able to measure the real time interaction of CALP1 and CALP2 with CaM ( Fig. 1). This binding was specific in that free CaM (25 M) in the binding buffer completely inhibited the interaction (data not shown). When the binding curves were analyzed with the computer program Fast-Fit, using a first order kinetic equation, CALP1 and CALP2 were found to have dissociation con-stant (K d ) for CaM of 88 and 7.9 M, respectively (Fig. 1). Thus, optimizing the pattern of inverted hydropathy and increasing the complementary peptide length from 8 to 12 residues resulted in a 11-fold increase in affinity.
CALP Does Not Structurally Resemble a Common CaM-binding Motif-The binding that was observed by surface plasmon resonance detection could be due to CALP interaction with the EF hands of CaM or its hydrophobic core or both. Based on the following, it is unlikely that CALP is simply a binding peptide for the hydrophobic pocket of CaM. First, our CALP peptides don't match any of the Block consensus patterns for CaMbinding motifs (48,49). Second, CaM-binding peptides show a high helical structural propensity. In contrast, the CD spectra for all of the CALP peptides are consistent with a ␤-turn (data not shown). In addition, the experimental data that follow clearly demonstrate an EF hand interaction.
Inhibition of Stains-all-Calmodulin Complex Formation-To establish experimentally that CALP bound the EF hands of CaM, we used the cationic carbocyanine dye, Stains-all, which is one of the most sensitive and useful probes for studying the calcium-binding sites of proteins such as CaM (50). This dye has been shown to bind to Ca 2ϩ -binding sites or EF hand motifs and to generate characteristic spectral changes. One of the important features of Stains-all binding is the ability of Ca 2ϩ to displace the dye from its binding site. Thus Stains-all binding was used to test whether the CALP peptides were interacting with the Ca 2ϩ -binding sites of CaM. Fig. 2 shows that as expected Ca 2ϩ can inhibit the interaction of the dye with CaM with a 50% inhibitory concentration (IC 50 ) of 2 to 3 M. Interestingly, both CALP1 and CALP2, but not a scrambled version of CALP2, were able to block Stains-all⅐CaM complex formation (IC 50 of 125 and 9 M, respectively) (Fig. 2). Once again CALP2 had a 14-fold higher affinity than CALP1. Perhaps more importantly these results strongly suggest that the complementary peptides were interacting with the sites to which they were targeted. Conformational Changes in CaM-When Ca 2ϩ binds to CaM, the protein undergoes a major change in conformation that can be monitored by a shift in the emission spectrum of dansylated CaM from a maximum at 530 to 494 nm (51) Fig. 3 shows that, like Ca 2ϩ , the CALP peptides, but not control peptides, were able to induce the expected change in the emission spectrum of dansylated CaM. EGTA was added at the end of the titrations to ensure the effects were not due to Ca 2ϩ contamination. EGTA addition (0.5 mM) at the end of the Ca 2ϩ titration reversed the spectrum to a maximum at 530 nM. In contrast, the chelating agent caused no change in the spectra of the CALP peptides complexed with CaM (Fig. 3). Thus Ca 2ϩ contamination was not responsible for the spectral changes caused by the CALP peptides. This is also in marked contrast to a typical CaM-binding compound, Mastoparan (Mas) 7, which is an amphipathic basic peptide that doesn't bind apoCaM, but interacts with the hydrophobic core of CaM after the four EF hands are occupied by Ca 2ϩ (Fig. 3). This interaction causes an additional change in conformation that leads to a further increase in the emission of dansylated CaM at 494 nM (52, 53) that is completely reversible by the addition of EGTA (Fig. 3). The dansylated CaM system, because of its sensitivity, was also used to evaluate the relative contribution of optimal inversion of the pattern of hydropathy and complementary peptide length on the affinity for CaM. In this assay system (Fig.  3), CALP2 had an affinity (2.4 M) that approached that of Ca 2ϩ (2.2 M) and was 50 times higher than CALP1 (120 M). CALP3 which represents the first eight amino acids of CALP2 and is equal in length to CALP1 was about 2-fold (66 M) more active than CALP1. CALP4 which is the same length as CALP2 and is composed of CALP1 of four additional amino acids from the carboxyl-terminal region of CALP2 was 2.5 times (48 M) more active than CALP1. Thus, although optimizing hydropathy or increasing length can improve the affinity, the combination of the two leads to an exponential increase.
Interaction with TnC-In order to test for the generality of CALP binding to EF hands as well as to evaluate interaction with a Ca 2ϩ -binding protein that lacks a hydrophobic core, we studied the interaction of the peptides with TnC. Tyr-111 and Arg-147 of smooth muscle troponin C, corresponding to position 7 of EF hand motif III and IV, respectively, were mutated to generate either single tryptophan mutants (C84S/Y111W/TnC and C84S/R147W TnC) or double tryptophan mutants (C84S/ Y111W/R147W TnC) (42). 3 One characteristic of these mutants, as seen also with similar mutants of CaM (54 -56), is a quenching effect in the presence of Ca 2ϩ . In the single tryptophan mutants, this is most likely due to a change in the environment of the tryptophan due to the conformational change induced by binding Ca 2ϩ . In the double tryptophan mutant, Ca 2ϩ co-ordination by the adjacent EF hands III and IV, results in a reduced distance between the two tryptophans leading to fluorescence self-quenching. 3 We used this system to test for an interaction of CALP with the EF hands of TnC, observing different effects with CALP2 and CALP3.
Interaction of CALP2 with C84S/Y111W/R147W TnC resulted in a strong increase (Ͼ50%) in the tryptophan emission (Fig. 4a) and a shift in the maximal emission from 347 to 337 nm. The concentration of CALP2 resulting in half-maximal effect is 33 M. Neither scrambled CALP2 nor Mas7 had a significant effect on the tryptophan emission (Fig. 4a). Binding of Ca 2ϩ to this TnC mutant causes the opposite effect, that is a quenching of the tryptophan emission. 3 CALP2 had no effect on the emission of free tryptophan (2 M), indicating that the effect of this peptide on the emission from the TnC mutant is not due to a change in the viscosity of the solution. The same effect of CALP2 was seen when complexed with the single tryptophan mutants, although the increase in emission was lower (approximately 18% for C84S/Y111W TnC and 25% for C84S/R147W TnC) (data not shown).
Stop flow analysis was done with C84S/R147W TnC to determine the kinetics of quenching due to binding of Ca 2ϩ either in the absence (Fig. 4b) or presence (Fig. 4c) of CALP2. Single exponential fitting of the association curves at saturating concentrations of Ca 2ϩ gave a decay rate constant (k 1 ) for TnC alone of 284 s Ϫ1 , and for the TnC⅐peptide complex this was 0.38 s Ϫ1 . The approximately 1,000-fold slower kinetics for this quenching in the presence of the peptide indicates that CALP2 is binding in the same region, and thus competing with Ca 2ϩ , analogous to the results from the Stains-all experiments. The slow dissociation of the peptide from such a site, as shown also with surface plasmon resonance detection, is the rate-limiting step for the binding of Ca 2ϩ resulting in the slower kinetics.
To verify further that the effect seen with CALP2 is due to the direct interaction of the peptide with the EF hands of C84S/Y111W/R147W TnC, we measured the fluorescence parameters in the presence of a quenching agent (acrylamide, Table II). The complex of CALP2 with C84S/Y111W/R147W TnC exhibited a decrease in quenching constant (k q ) as well as an increase in the fluorescence lifetime (ϽϾ) indicating that the tryptophan residues are buried in a more hydrophobic environment. Based on a model of the double tryptophan mutant TnC, this hydrophobic environment results from a direct binding of the peptide to the EF hands involved.
CALP3 had the opposite effect on the tryptophan emission of the double tryptophan mutant TnC, causing an 18% decrease with an IC 50 around 400 M (Fig. 4d). This is similar to the effect seen with Ca 2ϩ , which results in a change in the loop conformation bringing the two tryptophan residues closer in proximity. Mas7 and another control, cylicCALP3 (a head to tail cyclic version of CALP3), had no effect on tryptophan emission (Fig. 4d). Thus it would appear that as with CaM (see below), CALP2 and -3 are causing changes in TnC that are antagonistic and agonistic, respectively, to the effects of Ca 2ϩ . In addition, the effects are due to interactions occurring at or near the EF hands.
NMR Studies of the CALP⅐CaM Complex-One-and twodimensional NMR techniques were used to test further for CALP-induced conformational changes in CaM as well as to determine whether the complementary peptides preferentially bind to the Ca 2ϩ -coordinating sites in the amino-or carboxylterminal domains of CaM. At a 2 to 1 mole ratio of CALP1 to recombinant 15 N-labeled Drosophila CaM, there are at least 28 new peaks observed in the two-dimensional 1 H-15 N HSQC spectrum. Of these 28 peaks, 11 overlay with peaks in the Ca 2ϩ -bound CaM HSQC spectrum (Fig. 5). The finding that the apoCaM spectra (Fig. 5a) is that expected of the Ca 2ϩ -free form of the protein would seem to unequivocally demonstrate that the preparation is devoid of Ca 2ϩ . Furthermore, the obvious differences between the Ca 2ϩ and CALP1 bound CaM (Fig. 5, b and c) clearly show that the interaction is with CALP1 as opposed to Ca 2ϩ contamination of the peptide. Collectively, these results suggest that CALP1 causes conformational changes that in part overlap and yet in other ways are different than those elicited by Ca 2ϩ . This idea is confirmed by the To confirm the possibility that CALP1 and -3 preferentially interact with the carboxyl-terminal region of CaM, the protein was fragmented with trypsin into its amino-and carboxylterminal domains. For the amino-terminal domain of apoCaM, there was no evidence by NMR of an interaction with either CALP1 (Fig. 6) or CALP3 (data not shown). In contrast, when the carboxyl-terminal fragments of apo-Drosophila (Fig. 7) and bovine (data not shown), CaMs were titrated with CALP1; new peaks arise in the aromatic and methyl regions of the onedimensional 1 H NMR spectrum (Fig. 7). These peaks show excellent correspondence with peaks that are seen in the Ca 2ϩbound form of this fragment. Virtually identical results were seen when the carboxyl-terminal fragment of Drosophila apoCaM was titrated with CALP3 (data not shown).
The fact that CALP1 and -3 cause Ca 2ϩ -like changes in the NMR spectra of the carboxyl-but not amino-terminal fragment of CaM clearly shows that Ca 2ϩ contamination is not respon-

TABLE II Fluorescence parameter of TnC and CALP2/TnC complex
Where is em ϭ emission maximum with an excitation of 295 nm, K SV is the Stern-Volmer dynamic quenching constant calculated from the acrylamide quenching experiments. The dynamic bimolecular quenching constant (k q ) was calculated from k q ϭ K SV /͗͘, where ͗͘ is the intensity-weighted mean fluorescence lifetime calculated from ͗͘ ϭ ⌺ sible for the effects since Ca 2ϩ would alter the spectra of both domains. Despite this, two other methods were used to rule out Ca 2ϩ contamination. First, a sample of the carboxyl-terminal CaM fragment was prepared in D 2 O, and a one-dimensional 1 H spectrum was acquired. One-dimensional spectra were then acquired on this sample 24 and 72 h later. Over this period, no change was seen in the spectra indicating that there was not a problem of Ca 2ϩ leaching from the NMR tube. There was also no slow conformational change in the free carboxyl-terminal fragment during this period. These results are significant as all CALP/CaM experiments were completed within 24 h. Second, EGTA (5 mM) addition reversed the spectral changes due to Ca 2ϩ but not CALP (data not shown). Collectively, the NMR results confirm and extend those with dansylated CaM. They clearly show that CALP1 and -3 interact with CaM and cause a conformational change. The binding and conformational change, while Ca 2ϩ -like, is not identical to that seen with Ca 2ϩ since binding preferentially seems to occur in the carboxyl-terminal domain, and the CALP⅐CaM complexes have peaks in common with, as well as distinct from, those seen in the Ca 2ϩ -bound form of CaM.
Activation of Phosphodiesterase (PDE)-The observations that the CALP peptides bind the carboxyl-terminal EF hands of CaM and cause a conformational change raised the question of whether the CALP⅐CaM complex is in an active or "productive" conformation. This was explored with a Ca 2ϩ /CaM-dependent PDE assay employing a fluorescent form of cGMP as the substrate (57,58). At saturating levels of substrate and low levels of CaM, changes in Ca 2ϩ concentration are reflected as a change in the V max of the enzymatic reaction due to the stimulating effect of the active Ca 2ϩ -bound form of CaM. By using an EGTA concentration that reduces PDE activity to its basal level (the enzyme activity in the absence of CaM), it was possible to titrate the Ca 2ϩ activation of CaM. As had been observed before in a different type of PDE assay (38), CALP1 in the presence, but not the absence, of CaM was able to increase PDE activity (Fig. 8). In contrast, CALP2 and -4 were without effect (Fig. 8). Under the assay conditions (0.1 mM EGTA), any Ca 2ϩ contamination of the CALP peptides (see "Experimental Procedures") at the highest peptide dose would have led to only 2.5 nM free Ca 2ϩ which is far below that required to activate PDE (Fig. 8).
To determine whether CALP2 and -4 acted as CaM antagonists rather than agonists, PDE activity was measured in the presence of fixed concentrations of Ca 2ϩ and CaM and increasing concentrations of the complementary peptides. Both CALP2 and -4 inhibited the Ca 2ϩ /CaM-dependent PDE activity (Fig. 9). Since Mastoporan 7 causes a similar effect (Fig. 9), it was not possible in this experimental design to attribute con- clusively the inhibition of these two peptides to competition for the Ca 2ϩ -binding sites as opposed to the PDE-binding site of CaM.
To seek experimental evidence that the mechanism of inhibition of CALP2 was different from that of Mastoporan 7, we investigated the ability of Ca 2ϩ to reverse this inhibition. Mastoporan 7 inhibition, based on the binding of the peptide in the hydrophobic core of CaM, cannot be reversed by increasing Ca 2ϩ since it requires the divalent cation to bind CaM (Fig. 10). The inhibition due to CALP2, however, could be reversed by Ca 2ϩ . This provides strong experimental support for CALP peptides not binding CaM in the hydrophobic core but rather interfering with the calcium-binding site. Additional evidence of the difference between the CALP peptides and peptides binding the hydrophobic core of CaM comes from the analysis of the CD spectra of the CALP peptides. CALP2 and -4 CD spectra are characterized by a negative minimum at 230 nm and by a strong positive maximum between 220 and 200. This is characteristic of a ␤-sheet or ␤-turn structure (59). In contrast, other peptides interacting with the CaM hydrophobic core are characterized by an amphiphatic ␣-helix structure, as is the case for Mas7. DISCUSSION One of the major challenges for biological chemistry is to rationally design molecules to interact specifically with given sites of interest on biologically important proteins. The evolving recognition of the role of patterning of hydropathy in protein folding and shape suggested a means to design proteins or peptides with complementary contours by merely inverting hydropathic codes (1). This study reports evidence that de novo design by reciprocal patterning of hydropathy can be used to target specifically a peptide to a predetermined site on a protein. In this particular instance, complementary peptides, generically referred to as CALP, were designed with EF hand motifs as the anticipated sites of interaction. The following data are entirely consistent with CALP binding with the expected target on CaM. First, there is a direct interaction measured in real time between CALP and CaM, and this occurs at the Stains-all-binding sites that are known to be EF hands. Second, CALP bound to the EF hands of TnC which precludes an interaction with a hydrophobic core such as that found in CaM. Third, conformational changes like those induced by Ca 2ϩ interacting with its coordinating sites are observed in the CALP⅐CAM complex by NMR spectroscopy as well as by the use of dansylated CaM. Fourth, the conformation of the CALP⅐CAM complex, like the Ca 2ϩ ⅐CAM complex, is compati- ble with a PDE interaction leading to activation of the enzyme. Throughout these studies, constant attention was taken to ensure that the observed effects were not due to Ca 2ϩ contamination. For instance, any Ca 2ϩ contamination would have been below the limits of detection by atomic mass spectroscopy and consequently below the amounts required to affect any of the assays. In addition, unlike CALP, which only affected the carboxyl-terminal domain, contaminating Ca 2ϩ would have affected the NMR spectra of the amino-terminal as well as carboxyl-terminal fragments of CaM. Finally, EGTA was able to reverse spectral changes in CaM induced by Ca 2ϩ but not CALP. Thus it is extremely unlikely that the effects are due to Ca 2ϩ contamination.
Although these peptides are termed CALP to indicate Ca 2ϩ -like effects or antagonism, NMR determination of a solution structure for CALP1 4 coupled with computer-assisted docking to a known CaM structure (EF hand 4 region) suggest, not unexpectedly, that the interaction with CaM is unlike that of Ca 2ϩ (Fig. 11). CALP1 has a complementary contour to a large part of EF hand 4, and only the positively charged amino group on the amino-terminal valine protrudes into the Ca 2ϩ -binding pocket. Despite the different binding modalities and obviously different structure, Ca 2ϩ and CALP1 interactions lead to similar biochemical and biological effects. There are, of course, examples of similar situations wherein different substances can elicit similar effects by interaction with the same site on a given protein. These include morphine and ␤-endorphin binding to opiate receptors, acetylcholine and snake neurotoxins interaction with the acetylcholine receptor, and thaumatin and monellin as well as sucrose detection by "sweet" taste receptors (60 -62).
In addition to targeting specifically a peptide to a given site on a protein, the feasibility of improving the affinity of interaction as well as changing functional activity was also demonstrated. Two factors that markedly governed affinity were an optimization of reciprocity of the pattern of hydropathy and an increase in peptide length. Interestingly when the two were combined into a single construct, there was an exponential increase in affinity. Increasing the length by four amino acids also changed a given peptide from an agonist for CaM activation of PDE to an antagonist and caused opposing effect on the EF hands of TnC. Although conclusive proof must await the determination of CALP/CAM structures, it is tempting to speculate that the antagonist action occurs as a result of the longer peptides extending out of the Ca 2ϩ -binding pockets in CaM. In particular, the three residues of the EF hand loops in CaM are involved in an anti-parallel ␤-sheet structure between two adjacent Ca 2ϩ -binding loops of CaM (44,45,53,57,58,63,64). An interaction of the four amino acid extension with this area of  (41,71). The protein structure used was Calmodulin Tr2c-domain (PDB entry 1CMG). The file was elaborated to eliminate the Ca 2ϩ atoms, and the amino acids were assigned the theoretical charges at neutral pH using the software Sybyl (Tripos). A surface with a radius of 15 Å centered on tyrosine 99 covering both EF hand 3 and 4 of the model was selected as center of the binding cavity. As ligand we used a model of CALP1 base on NMR data of the peptide in Me 2 SO solution. We applied no constraints to the peptide. The model presented is the solution ranked first from the program. On the left we present the structure of Tr2C, using a Connolly solvent surface. The dark blue area corresponds to EF hand loop IV, and the calcium atom is shown in red. The structure on the right is the docked model. CALP1 is the transparent yellow molecule. Indicated in red are the amino terminus and carboxyl terminus of the peptide. the loop might be expected to interfere with overall conformational changes signaled by alterations in the ␤-sheet structure.
One unexpected finding in this study was the observation by NMR spectroscopy that CALP interacts with EF hands in the carboxyl-terminal but not the amino-terminal domain. Whereas a molecular understanding of this also awaits threedimensional structural determinations, interaction in this domain may result from the fact that it contains the two high affinity Ca 2ϩ -coordinating sites, and only these may have the requisite structure for CALP binding. Another intriguing question is how an interaction with only one of the two domains leads to a CaM conformation that is capable of activating PDE. This may be related to intermediary conformations of CaM having some enzyme-activating ability (65).
Although this study raises a number of interesting specific questions about CaM, its general conclusion seems clear as follows: hydropathic patterning can be used to tailor-make peptide ligands of varying affinity and opposing (agonist or antagonist) activity that are targeted to predetermined sites on proteins. It would appear that the design principles herein reported may well be generally applicable to other proteins whose classic ligands are not proteinaceous (66). These include "digitalis"-like peptides for the Na ϩ ,K ϩ -ATPase (67), sapid peptides for taste receptors (68), a "dopamine"-like peptide for the D2 dopamine receptor (69), and an "acetylcholine"-like peptide for the nicotinic acetylcholine receptor (70). We view these peptide ligands as first generation or lead compounds. Once the three-dimensional structure of the peptide ligand-protein target complex is solved, the active conformation of the complementary peptide can then serve as a template for the design of constrained organic analogs with perhaps higher affinities and better pharmacologic profiles.