Design of inhibitors of Ras–Raf interaction using a computational ...

Protein Engineering vol.14 no.1 pp.39–45, 2001

Design of inhibitors of Ras–Raf interaction using a computational combinatorial algorithm

Jun Zeng1,2,3, Thao Nheu1, Anna Zorzet1,4,5, Bruno Catimel1,4, Ed Nice1,4, Hiroshi Maruta1, Antony W.Burgess1,4 and Herbert R.Treutlein1,2,4 1Ludwig

Institute for Cancer Research, PO Box 2008, Royal Melbourne Hospital, Parkville, VIC 3050, 2Cytopia Pty Ltd, 7th Floor, Daly Wing, St. Vincent’s Hospital, 41 Victoria Parade, Fitzroy, VIC 3065, Australia and 4Cooperative Research Centre for Cellular Growth Factors, Royal Melbourne Hospital, Parkville, VIC 3050, Australia 5Present

address: Uppsala School of Engineering, Uppsala University, Uppsala, Sweden

3To

whom correspondence should be addressed

Drugs that inhibit important protein–protein interactions are hard to find either by screening or rational design, at least so far. Most drugs on the market that target proteins today are therefore aimed at well-defined binding pockets in proteins. While computer-aided design is widely used to facilitate the drug discovery process for binding pockets, its application to the design of inhibitors that target the protein surface initially seems to be limited because of the increased complexity of the task. Previously, we had started to develop a computational combinatorial design approach based on the well-known ‘multiple copy simultaneous search’ (MCSS) procedure to tackle this problem. In order to identify sequence patterns of potential inhibitor peptides, a three-step procedure is employed: first, using MCSS, the locations of specific functional groups on the protein surface are identified; second, after constructing the peptide main chain based on the location of favorite locations of N-methylacetamide groups, functional groups corresponding to amino acid side chains are selected and connected to the main chain Cα atoms; finally, the peptides generated in the second step are aligned and probabilities of amino acids at each position are calculated from the alignment scheme. Sequence patterns of potential inhibitors are determined based on the propensities of amino acids at each Cα position. Here we report the optimization of inhibitor peptides using the sequence patterns determined by our method. Several short peptides derived from our prediction inhibit the Ras–Raf association in vitro in ELISA competition assays, radioassays and biosensor-based assays, demonstrating the feasibility of our approach. Consequently, our method provides an important step towards the development of novel anti-Ras agents and the structure-based design of inhibitors of protein–protein interactions. Keywords: computational combinatorial chemistry/inhibitor design/ras protein Introduction Structure-based ligand design has become an important tool in drug discovery and pharmaceutical research (Amzel, 1998). It aims to identify chemical compounds or peptides that bind strongly to key regions of biologically relevant molecules, e.g. © Oxford University Press

enzymes or receptors, for which three-dimensional structures are known. Consequently, these compounds should be able to inhibit or stimulate the biological activity of these target molecules. Several drugs discovered using this approach have been tested clinically (Greer et al., 1994; Hilpert et al., 1994; Von Itzstein et al., 1996; Bohacek and McMartin, 1997; Varghese et al., 1998) and its application has been further facilitated by the recent advances in molecular simulation methods and structural determination technologies, as well as ever-increasing computer power. Up to now, structure-based drug design has in most cases only been applied to targets (e.g. enzymes) with a clearly defined binding pocket for small molecules. The potential to take advantage of protein surfaces as drug targets has not been fully explored so far. Computational combinatorial inhibitor design is one of the newly developed structure-based de novo design methods (Caflisch et al., 1993; Caflisch and Karplus, 1995, 1996; Caflisch, 1996; 1996; Joseph-McCarthy et al., 1997). It consists of three steps: first, the positions of functional chemical groups are identified on the key regions of a target protein using an exhaustive ‘multiple copy simultaneous search’ (MCSS) approach (Caflisch et al., 1993); second, the MCSS minima of a specific functional group are used to construct links to the other functional groups; finally, potential inhibitors are selected according to their calculated or estimated binding affinities to the target protein. Previously, we had developed a novel scheme that allows the application of computational combinatorial inhibitor design to identify peptide inhibitors that target the protein surface where the details of the peptideprotein binding site are not clearly defined (Zeng and Treutlein, 1999; Zeng, 2000). In our scheme, the third step involves the sequence alignment of designed peptides and evaluation of the probability of a specific amino acid type at each position of the peptides. Peptide sequences of potential inhibitors were based on the amino acid preferences at each position, rather than on approximately calculated binding affinities (Caflisch and Karplus, 1995, 1996; Caflisch, 1996). Our method is therefore also suitable to design combinatorial libraries for specific drug targets and to identify and analyze amino acid sequence patterns of potential inhibitor peptides. Here we describe the successful application of our approach to the optimization of peptides that inhibit the association between Ras and its downstream target protein Raf. Ras acts as a molecular switch between its active state where guananine triphosphate (GTP) is bound and its inactive form where a guananine diphosphate (GDP) is bound (Barbacid, 1987). Only the GTP-bound form of Ras binds to the downstream targets, one of which is the threonine/serine kinase Raf (Barbacid, 1987). Once Ras binds to Raf, it stimulates the activation of the MEK–MAPKK signaling pathway, transmitting signaling events into the nucleus to control cell growth and differentiation. An oncogenic form of Ras exists only in the active form, resulting in a constitutive activation of Ras-mediating signaling events and promoting aberrant 39

J.Zeng et al.

growth in more than 30% of all human tumors (Barbacid, 1987). Antagonists of the Ras–Raf interactions that are likely to inhibit the Ras-stimulated signal transduction pathway are thus of great potential value to anti-cancer therapy (Maruta and Burgess, 1994). Ras–Raf interactions have been studied extensively by sitedirected mutagenesis [see Campbell et al. (1998) and references therein]. The important regions of Raf that bind to Ras protein have been identified as a 78 amino acid N-terminal region, the so-called ‘Ras binding domain’ (RBD) (Chuang et al., 1994; Scheffler et al., 1994) and a zinc-finger domain, the socalled ‘cysteine-rich domain’ (CRD) (Drugan et al., 1996). Whereas the binding mechanism between the CRD and Ras remains unknown, details of the interaction between RBD and Ras have been revealed from the crystal structure of the complex between the RBD and Rap1A (Nassar et al., 1995), a protein with a ⬎80% sequence homology to Ras. Previously, we have carried out extensive computational and experimental studies on the structural and binding properties of the Ras–Raf interactions (Zeng et al., 1998, 1999a,b). Based on the available structural information (Zeng et al., 1999b), potential peptide inhibitors have been designed using the protocol we have described previously (Zeng and Treutlein, 1999). Here we describe the application of our previous results to optimize a previously published inhibitor and its experimental characterization by in vitro radioassay, enzymelinked immunosorbent assay (ELISA) and BIAcore techniques. Our optimized peptides should act as lead compounds for the development of novel pharmaceuticals (Bures and Martin, 1998; Schneider et al., 1999). Materials and methods Computational combinatorial peptide design The details of our computational combinatorial inhibitor design approach have been described elsewhere (Zeng and Treutlein, 1999; Zeng, 2000). Our procedure consists of three steps. In the first step, the MCSS method is used to determine energetically favorable positions and orientations of functional groups on the surface of the complex between Ras and a Ras-binding helix (RBH). The RBH helix (residues 84–89) contains two critical residues (Lys84 and Arg89) for Ras–Raf interaction, as mutants at these positions have been shown to abolish the Ras–Raf binding in vivo completely (Nassar et al., 1996; Zeng et al., 1999a). The structure of the Ras–RBD complex was taken from our 2 ns molecular dynamics simulation (Zeng et al., 1999b) and shown in Figure 1. In the complex, residues Lys84 and Arg89 of RBH form direct salt bridges to Asp33 and Asp38 of the Ras effector loop (residues 30–40), respectively, consistent with mutagenesis studies (Nassar et al., 1996; Zeng et al., 1999a). Our basic design strategy was to extend the RBH in order to allow interactions with all the important regions such as the effector loop, switch II (residues 60–76) (Wittinghofer and Nassar, 1996) and an ‘interaction surface’ (residues 5–8, 52–56) that was identified previously (Zeng and Treutlein, 1999). Inclusion of the RBH into our design scheme significantly reduces the number of peptide conformations that need to be considered on the surface of Ras. The functional groups chosen for the MCSS calculations were N-methylacetamide (NMA), benzene, propane, phenol, methanol, acetate ion, methylammonium, methylguanidinium and water, representing functional moieties of the peptide backbone, hydrophobic residues, polar uncharged residues, charged residues 40

Fig. 1. The complex structure of Ras–RBD obtained from a 2 ns molecular dynamics simulation (Zeng et al., 1999b). In Ras, the effector loop, the interaction surface and the switch II regions are colored green, pink and red, respectively; these are the important regions with which inhibitor will be designed to interact. In Raf, Ras-binding helix (RBH) of Raf is colored brown and a fragment of residues 94–101 and sequence CCAVFRL light blue, respectively. This short peptide was identified previously as a Ras–Raf inhibitor in vitro (Barnard et al., 1998). Nucleotide substrate GTP is shown in CPK.

and solvent. All parameters for the functional groups were taken from the CHARMM22 all-hydrogen atom force field (MacKerell et al., 1998). Amongst the MCSS minima on the surface of Ras, only those within a 4.0 Å distance from Ras were selected for further analysis. In the second step, peptide main chains were defined by the energetically favorite positions of NMA replicas on the Ras surface. A random algorithm was developed to connect the carbon atoms of NMA replica to form a Cα trace (Zeng and Treutlein, 1999; Zeng, 2000). With the Cα atoms fixed, backbone carbonyl and amide groups were inserted or attached to transform a poly-NMA chain into a polyglycine peptide. Ten polyglycine peptides were constructed and the one with the lowest interaction energy to Ras was chosen as a representative conformation for the construction of the inhibitor peptides. In the third step, subsets of functional groups were selected for each Cα position of the polyglycine peptide according to the distance from each replica of functional groups to the Cα atom of the polyglycine representative. The amino acid type was randomly chosen from the available subsets and the missing side chain atoms were grown to connect to the Cα atoms by inserting CH2 groups. All completely built side chains were energy minimized to insure feasible side chain geometry. The sequences of functional groups were thus transformed into a peptide. All the procedures were performed using the program XPLOR (Brunger, 1992). The peptides were then connected to the RBH and those with significant geometric distortion (i.e. large bond and angle energies) were ignored. Our procedure resulted in 100 peptides on the surface of Ras. We aligned the sequences of these peptides using the program CLUSTALX (Thompson et al., 1994). The multiple sequence alignment was used to determine the probability of an amino acid type at each position of the peptide. The optimal sequences of the Ras binding peptides

Design of inhibitors of Ras–Raf interaction

were constructed by placing amino acid types with high probabilities at each position of the peptide. Inclusion of RBH for construction of peptide inhibitors The designed peptides that are expected to inhibit Ras–Raf interaction contains both RBH and the peptides designed using our computational combinatorial inhibitor design approach. However, the RBH itself hardly affects Ras–Raf binding (Barnard et al., 1995, 1998) as it might not adopt its wildtype helical conformation. Therefore, our design strategy for Ras–Raf inhibitor is to identify the lead peptide using first our design method and then to include specific residues of the RBH which directly form salt bridges or hydrogen bonds in the crystal structure of the Ras–RBD complex (Nassar et al., 1995; Zeng et al., 1999b). The resulting inhibitors can therefore maintain the important binding determinants between the RBH and the effector loop. Reversed sequences of designed peptides Our method described above primarily identifies favorite interaction sites of amino acids. Those sites are then linked together to form a peptide. The locations of the N- and C-termini are initially determined by our design strategy to extend the RBH towards the ‘interaction surface’: the N-terminus would be located on an area that coincides with the RBH binding site, the C-terminus would be located on the ‘interaction surface’. However, the linkage algorithm which links the amino acids is not precise enough to exclude the possibility that a peptide with its amino acid sequence reversed would bind to the surface patch with significantly higher affinity, although it would bind in a reversed way on to the surface, i.e. with its C-terminus at the position of the RBH and its N-terminus at the ‘interaction surface’. This imprecision of our algorithm makes it necessary also to take into account reversed peptide sequences and sequence patterns when searching for an optimized inhibitor peptide. Expression and preparation of Ras, GST–RBD–Raf and designed peptides V-Ha-Ras was purified from Escherichia coli according to the procedure of Gibbs et al. (1984). The concentration of Ras was determined with the Bio-Rad protein assay. The RBD was produced in E.coli as a GST fusion protein, which was affinity-purified on GSH beads (Fridman et al., 1994). The concentrations of purified GST–RBD constructs were determined by scanning Coomassie Brilliant Blue-stained SDS–PAGE gels using densitometry and measuring the band densities relative to BSA standards with ImageQuant 3.3. The peptides were purchased from Chiron Technology (Melbourne, Australia) purified to apparent homogeneity by micropreparation RP-HPLC using a Brownlee RP3000 (300⫻ 201 mm i.d.) column eluted at a flow-rate of 100 µl/min with a linear 60 min gradient between a primary solvent of 0.15% (v/v) TFA and a secondary solvent of 60% aqueous CH3CN– 0.125% (v/v) TFA. The column temperature was 45°C and detection was effected by measuring the absorbance at 215 nm. Enzyme-linked immunosorbent assay (ELISA) Purified Ras–GTP complex (100 µl, 10 µg/ml) was plated overnight on to 96-well microtiter plates (Nunc) coated with a buffer solution containing 10 mM Na2CO3 and 30 mM NaHCO3 (pH 9.6). The plates were washed twice with a ‘blocking’ buffer solution of 0.05% Tween, 20 mM Tris, 2 mM MgCl2, 150 mM NaCl, 1 mM DTT and 0.5% BSA. The wells were then blocked in the blocking buffer at 37°C for 1 h.

GST–RBD–Raf (10 µg/ml) was diluted into PBS containing 0.05% Tween, 20 mM Tris, 2 mM MgCl2, 150 mM NaCl and 1 mM DTT. The peptides (100 µM) were incubated with GST–RBD–Raf (10 µg/mol) at 37°C for 15 min prior to addition to the wells. The plates were washed three times with PBS, then incubated at 37°C for 2 h with a goat anti-GST antibody (Amersham Pharmacia Biotech) diluted 2000-fold in PBS. The washing step was repeated and the plates were incubated with a 1:1000 dilution of rabbit anti-goat antibody conjugated to horseradish peroxidase (HRP) (Amersham Pharmacia Biotech) for 1 h. The wells were washed three times, followed by development with ABST substrate (ZYMED, San Francisco, CA) diluted 100-fold in a buffer solution containing 0.1 M sodium citrate buffer and 0.03% hydrogen peroxide. Optical densities (416 nm) were read after 15 min in a microtiter plate reader. The amount bound was quantified by comparison with an appropriate calibration curve (30–10 µg/ml). Radioassay for Ras–GTP/RBD binding Binding between Ras and RBD–GST was carried out as described previously (Fridman et al., 1994; Zeng et al., 1999a). Briefly, v-Ha-Ras was labelled with [γ32P]GTP (Du Pont). A 2.5 µg amount of the radioactively labelled Ras (100 µl) was incubated with GST–RBD bound to GSH beads in presence of 100 µM of peptides for 15 min at room temperature on a rotator. The mixture was washed four times with 500 µl of the binding buffer [50 mM Tris–HCl (pH 7.5), 1 mM DTT, 10 mM MgCl2, 0.5 mg/ml BSA and 1 mM ATP]. The GSH beads were recovered by centrifugation at 104 g for 5 min. The radioactivity bound to the beads in the pellet was measured with a scintillation counter. Binding was expressed as the percentage of the radioactivity associated with RBD. Nonspecific binding of Ras to the beads was determined by incubation with GST alone. Biosensor experiments The binding of GST–Ras to the peptides was also analyzed on the BIAcore optical biosensor (Pharmacia Biosensor, Sweden) using peptide-derivatized CM 5 sensor chips. The immobilization of peptides on the sensor chip surface was performed essentially as described previously (Catimel et al., 1999). The peptides at a concentration of 150 µg/ml in 20 mM sodium acetate (pH 4.5) was coupled at 1 µl/min on to sensor chips that activated by a pulse of 0.2 M N-ethyl-N⬘(dimethylaminopropyl)carbodiimide and 0.05 M N-hydroxysuccinimide (45 µl, 1 µl/min), in order to yield an increase in the response level of 800–1500 response units. Known Ras/Raf inhibitors To date, only three inhibitors of Ras–Raf interaction have been identified. However, even these bind to Ras with only low affinity and are unsuitable for practical use. While sulindac sulfide has been shown to inhibit Ras–Raf interaction with an IC50 of 400 µM (Hermann et al., 1998), two peptide inhibitors identified from the Ras effectors (i.e. GAP, c-raf-1, etc.) indicate only ~20% inhibition at a concentration of 100 µM (Clark et al., 1996). More recently, a peptide sequence CCAVFRL derived from the Ras-binding domain of c-raf-1 has been shown to interfere with Ras–Raf association (Barnard et al., 1998). This peptide was determined by a random screening approach without any consideration of the Ras–Raf complex structure. In fact, this short peptide is located ~15 Å away from the binding interface in the crystal structure of the 41

J.Zeng et al.

Ras–RBD complex (Nassar et al., 1995; Zeng et al., 1999b), as shown in Figure 1. As we will demonstrate later, our approach can identify this peptide and its derivatives, providing a way towards de novo design of inhibitor that cannot be revealed from the complex structure. Results Determination of sequence patterns for inhibitor peptides In our approach, the RBH was intended to interact with the effector loop of Ras protein as observed in the X-ray structure of the RBD–Rap1A complex (Nassar et al., 1995). The extended part of the inhibitor was designed to bind to the ‘interaction surface’ and the switch II region. Based on the minimum energy positions of functional groups obtained from the MCSS calculations (Zeng and Treutlein, 1999), 300 sequences were generated using the protocol described in Materials and methods. After growing the side chains and merging to the Ras-binding helix (RBH), 196 peptides with significant bond distortion (i.e. bond stretch energy ⬎30 kcal/ mol) were excluded. The remaining 104 peptides were kept for further investigation in the study. Sequence alignment using the CLUSTALX program shows distinct regions for RBH and extended regions, consistent with our design strategy. The probabilities of different amino acids at each position of the 104 peptides were derived from the alignment of the peptide sequences. Two additional linker residues, namely glycine and phenylalanine, were inserted between the RBH and extended region such that the bulky side chain of phenylalanine could prevent cross-interaction between the regions. Figure 2 shows the probabilities of amino acids at each position of the extended part. Only the amino acids with probabilities 艌0.30 were considered for the design of the inhibitor and labeled in the figure. While position P1 is clearly defined by Tyr by its hydrophobic interaction to the aliphatic part of Ras–Arg41, position P2 is dominated by Asp/Glu owing to its internal electrostatic interaction with Arg89 and Lys84 of RBH. The residues at positions P3, P4 and P5 interact with Glu37, Asp38, Ser39 and Tyr40 within the Ras effector loop and consequently are controlled by amino acids Arg/Lys or Phe, Arg/Lys and Ser/Thr, respectively. Positions P6–P8 are located on the ‘interaction surface’, demanding hydrophobic Phe or hydrophobic/polar residue Tyr to interact residues Leu56, Lys5 and Val7 of Ras, respectively. The last residue (position P9) is located between Lys5 of the interaction surface and Thr74 of the switch II, resulting in uniform distribution amongst Phe, Tyr and Asp/Glu. This position is therefore not clearly defined and can be occupied by any amino acid, labeled X in Figure 2. By joining the RBH and extended parts with two linker residues (glycine and phenylalanine), the motif of inhibitor sequences can thus be defined as KALKVRGFYDRRTFFFX. Figure 3 shows the comparison of the designed peptides to the known inhibitors (Clark et al., 1996). The sequence of –RRTFF within the extended part resembles hydrophobic and charged residues at the first five positions in the fragment (sequence RKTFLKLA) derived from cysteine-rich domain of c-Raf-1. This fragment has been shown to block Ras-mediated activation of mitogen-activated protein kinase in vivo (Clark et al., 1996). Moreover, since our method cannot distinguish between N- and C-terminal orientation, when the extended part (–RRTFF) is read backwards, it also resemble the hydrophobic and charged motif of the last five residues of a short peptide with the sequence CCAVFRL derived from the Ras42

binding domain of c-Raf-1 (Barnard et al., 1998,; see also Materials and methods). Construction of peptides that interfere with Ras–Raf association Table I lists the peptide inhibitors derived from our designed peptides. We first identify the lead peptide based on the extended part of the designed inhibitor sequence. Three modifications are introduced: first, we replace the first two residues (Y and D) by a Leu as these residues mainly interact with the RBH; second, we attach two cysteines to the C-terminus, considering their importance demonstrated previously (Barnard et al., 1998); third, we substitute the hydrophobic ‘cluster’ XFFF into AVFL in order to dilute the peptide in water. The resulting lead peptide has a sequence of CCAVFLTRRL (peptide 1), altering Ras–Raf interaction by 23.4% from ELISA and 12.5% from radioassay. The lead peptide is subsequently optimized according to the designed peptide as given in Table I: first, an arginine is inserted at position P3; second, Thr at the position P5 is deleted as this residue competes with arginines at positions P3 and P4 by interacting the Ras effector loop; third, Leu at the next position is substituted into Phe; fourth, the N-terminal cysteines are replaced by Ala, which corresponds to the undefined residue X in the designed peptide. The resulting peptide (peptide 2; sequence AAVFFRRL) inhibits Ras–Raf binding by 13.0% from radioassay. Increasing the hydrophobicity of the peptide 2 by substituting three N-terminal residues into CCF, peptide 3 (sequence CCFFFRRRL) inhibits Ras–Raf interaction by 15.0% from radioassay and 8.9% from ELISA. Deletion of one Arg in peptide 3 has only a minor effect on the inhibition (peptide 4), indicating a maturation of the inhibition based on the extended part of our designed peptides. The inhibitors are further optimized by extending the lead peptides (i.e. peptides 1 and 4) to inclusion of specific residues of RBH. These residues are defined in Materials and methods and highlighted in bold in Table I. In addition, a fragment of –DYFG– between the RBH and extended parts is ignored because it mainly interacts with the RBH and precludes the cross-interaction between the RBH and extended regions. Extending peptide 4, peptide 5 increases inhibition up to 39.0% from ELISA, but decreases it to 8.3% from radioassay. Although peptides 1–4 have similar magnitudes of inhibition from radioassay, peptide 1 inhibits Ras–Raf binding as twice as much of peptide 4 from the ELISA assay. By elongating peptide 1 to the RBH, however, peptide 6 decreases the inhibition to 10.4% from ELISA and to 4.2% from radioassay. Binding of GST–Ras protein to sensor chip-immobilized peptides As both ELISA and radioassay experiments showed that peptides 1 and 3 inhibit Ras–Raf interaction, BIAcore analyses were also carried out to detect the direct binding of GST– Ras protein to the peptides. The peptides were sensor chipimmobilized at levels of ~800 and ~1500 response units, respectively. Figure 4 shows the binding curves for these interactions. The relative response units clearly indicate that both peptides bind to Ras, in agreement with ELISA and radioassay results (see Table I). However, the biosensor response for peptide 1 is shown to be significantly greater than the response of peptide 4, consistent with the ELISA experiment where peptide 1 inhibits Ras–Raf binding twice as much as peptide 4.

Design of inhibitors of Ras–Raf interaction

Fig. 2. Amino acid profiles at each position of the extended part within the inhibitor peptides. Amino acids with probability 艌0.30 are considered for design of inhibitor and labeled. The last residue cannot be defined and is labeled X.

43

J.Zeng et al.

Fig. 3. Comparison of designed peptides to the consensus peptides identified from the Ras effectors (Clark et al., 1996; Barnard et al., 1998). The Ras binding helix and extended components are highlighted. Sequence patterns of the designed peptide that is in agreement with the known inhibitor peptides are colored gray.

Table I. Inhibition of peptides to RBD–Raf binding to Ras–GTP at a concentration of 100 µM Peptides

Designed 1c 2 3 4c 5 6

Sequences

987654321 .YY....a c-XFFFTRRDYFGRVKLAK-nb n-CCAVFLTRL-c n-AAVFFRRL-c n-CCFFFRRL-c n-CCFFFRL-c n-CCFFFRLRARL-c n-CCAVFLTRRLRA-c

Inhibition (%) ELISA

Radioassay

23.4 ndd 8.9 12.6 39.0 10.4

12.5 13.0 15.0 12.0 8.3 4.2

Peptide was generated using our computational combinatorial design approach for identification of Ras inhibitors. n- and c- indicate N-terminus and C-terminus, respectively. The charged residues homologous to the template sequences are underlined and the extension residues for inclusion of the binding moiety between the RBH (sequence RVKLAK) and Ras are highlighted in bold. The position numbers (1–7) correspond to Figure 2. aYY indicate other possible amino acids at positions 7 and 8, determined from the probabilities shown in Figure 2. bX means the residue which cannot be defined from the calculated probabilities (Figure 2; see text for details). cBIAcore experiments were performed to detect the binding between these peptides and the Ras–GTP. See text for details. dNot detected.

Discussion While the computational combinatorial design approach has been playing an important role in the interdisciplinary process of drug discovery (Caflisch et al., 1993; Caflisch and Karplus, 1995, 1996; Caflisch, 1996; Joseph-McCarthy et al., 1997), its application for the design of inhibitor on the surface of a protein is believed to be limited owing to the flexibility of a protein surface and the lack of a distinctive binding pocket. Previously, we have developed a novel scheme to tackle this complex problem (Zeng and Treutlein, 1999; Zeng, 2000). Our method used sequence alignment to identify the amino acids that are most likely to occur at each position within a peptide inhibitor. The method was first applied to design inhibitors that could potentially block Ras–Raf binding. The constructed peptides resembled consensus peptides identified from Ras effectors (Zeng and Treutlein, 1999). The amino acid sequence patterns derived from this initial study were used to identify and improve the binding affinity of inhibitors of Ras–Raf interaction. Our designed peptides have been experimentally validated, providing lead compounds for future development of potent Ras inhibitors. 44

Fig. 4. Binding curves for the interaction of GST–Ras proteins with (A) immobilized peptide 4 and (B) peptide 1. GST–Ras protein was injected across peptide-derivatized or blank sensor surfaces, highlighted with solid and dashed lines, respectively. See text for details.

In total, six peptides were evaluated using ELISA and radioassay, as well as in BIAcore assay. The sequences correspond to the RBH and extended components in the designed inhibitors. The peptides from the extended part were shown to inhibit Ras–Raf interaction up to 12.0–15.0 and 8.9– 23.4% from radioassay and ELISA, respectively. BIAcore analysis suggested that the inhibition mechanism is mainly due to the direct binding of peptides to Ras. Relative biosensor responses can be correlated with data from competition experiments (ELISA). Extension of these peptides to include the RBH improved the inhibition up to 39.0% in ELISA, but reduced the inhibition to 8% in radioassay. This difference might be due to different assay protocols and conditions used in ELISA and radioassay (see Materials and methods). As indicated from the results for peptides 1–4 in Table I, the magnitude of inhibition obtained from the ELISA experiment is more sensitive to the condition than those from radioassay. Regardless of the uncertainty in the assays, our results demon-

Design of inhibitors of Ras–Raf interaction

strate that relatively small molecules can disrupt the formation of the Ras–Raf signalling complex in vitro. It is also interesting that extension of peptides by charged residues, i.e. peptide 1 (sequence CCAVFLTRL) vs peptide 6 (sequence CCAVFLTRRLRA), indeed decreases the inhibition in both ELISA and radioassay. In principle, the difference of binding affinities between two peptides is the difference of the free energy change of two peptides in the complexes with the target protein and the desolvation free energies of the two peptides in solution (Zeng et al., 1999a). When charged residues are properly introduced into a peptide to interact strongly with the target protein, it could result in a larger desolvation free energy in solution. If the desolvation free energy decreases more than the binding free energy, the charged residues could indeed decrease the binding affinity of the initial peptide. It seems likely that a proper increase in hydrophobicity should be considered when charged residues are inserted/substituted within the peptides, in order to maintain the desolvation free energy of the peptides in solution. Previous studies have demonstrated that the multiple conformations in the effector loop and switch II regions of Ras are important for target recognition (i.e. Raf–RBD) (Sydor et al., 1998; Terada et al., 1999), so that the dynamics of Ras and peptides should be an important aspect in the design of Ras inhibitors. Therefore, molecular dynamics simulation on the complex between Ras and each designed peptide solvated in the explicit solvent molecules and considering the protein flexibility in the design algorithm (Stultz and Karplus, 1999) are expected to improve the design results. From the results of the bioassays we conclude that our computational combinatorial design approach combined with experimental measurements (ELISA or radioassay) as a guide in sequence pattern identification and optimization of Ras inhibitors shows promising results. With the rapid development of computer technology, our method could be used for the economical design of combinatorial libraries, which could be screened for more optimal inhibitor peptides.

Fridman,M., Tikoo,A., Varga,M. Murphy,A., Nur-El-Kamal,M. and Maruta,H. (1994) J. Biol. Chem., 269, 30105–30108. Gibbs,J., Sigal,I., Poe,M. and Scolnic,E. (1984) Proc. Natl Acad. Sci. USA, 81, 5704–5708. Greer,J., Erickson,J., Baldwin,J. and Varney,M. (1994) J. Med. Chem., 37, 1035. Herrmann,C., Block,C., Geisen,C., Haas,K., Weber,C., Winde,G., Moroy,T. and Muller,O. (1998) Oncogene, 17, 1769–1776. Hilpert,K. et al. (1994) J. Med. Chem., 37, 3889. Josephy-McCarthy,D., Hogle,J. and Karplus,M. (1997) Proteins: Struct. Funct. Genet., 29, 32–58. MacKerell,A. et al. (1998) J. Phys. Chem. B, 102, 3586–3616. Maruta,H. and Burgess,A. (1994) Bioessays, 16, 489–496. Nassar,N., Horn,G., Herrmann,C., Scherer,A., McCormick,F. and Wittinghofer,A. (1995) Nature, 375, 554–560. Nassar,N., Horn,G., Herrmann,C., Block,C., Janknecht,R. and Wittinghofer,A. (1996) Nature Struct. Biol., 3, 723–729. Scheffler,J. et al. (1994) J. Biol. Chem., 269, 22340–22346. Schneider,G., Schrodl,W., Gerd,W., Muller,J., Nissen,E., Ronspeck,W., Wrede,P. and Kunze,R. (1999) Proc. Natl Acad. Sci. USA, 95, 12179–12184. Stultz,C. and Karplus,M. (1999) Proteins, 37, 512–529. Sydor,J., Engelhard,M., Wittinghofer,A., Goody,R. and Herrmann C. (1998) Biochemistry, 37, 14292–14299. Terada,T. et al. (1999) J. Mol. Biol., 286, 219–232. Thompson,J., Higgins,D. and Gibson,T. (1994) Nucleic Acids Res., 22, 4673–4680. Varghese,J., Smith,P., Sollis,S., Blick,T., Sahasrabudhe,A., McKimmBreschkin,J. and Colman,P. (1998) Structure, 6, 735–746. Von Itzstein,M., Dyson,J., Oliver,S., White,H., Wu,W., Kok,G. and Pegg,M. (1996) J. Med. Chem., 39, 388–391. Wittinghofer,A. and Nassar,N. (1996) Trends Biochem. Sci., 21, 488–491. Zeng,J. (2000) Comb. Chem. High Throughput Screen., 3, 355–363. Zeng,J. and Treutlein,H.R. (1999) Protein Eng., 12, 457–468. Zeng,J., Treutlein,H.R. and Simonson,T. (1998) Proteins: Struct. Funct. Genet., 31, 186–200. Zeng,J., Fridman,M., Maruta,H., Treutlein,H.R. and Simonson,T. (1999a) Protein Sci., 8, 50–61. Zeng,J., Treutlein,H.R. and Simonson,T. (1999b) Proteins: Struct. Funct. Genet., 35, 89–100. Received October 3, 2000; revised October 30, 2000; accepted October 31, 2000

Acknowledgement J.Z. acknowledges a C.J.Martin Fellowship awarded from the Australian National Health and Medical Research Council (No. 967362).

References Amzel,L. (1998) Curr. Opin. Biotechnol., 9, 366–369. Barbacid,M. (1987) Annu. Rev. Biochem., 56, 779–827. Barnard,D., Diaz,B., Hettich,L., Chuang,E., Zhang,X.F., Avruch,K. and Marshall,M. (1995) Oncogene, 10, 1283 Barnard,D., Sun,H., Baker,L. and Marshall,M. (1998) Biochem. Biophys. Res. Commun., 247, 176–180. Bohacek,R. and McMartin,C. (1997) Curr. Opin. Chem. Biol., 1, 157–161. Brunger,A.T. (1992) X-PLOR Version 3.1, A System for X-Ray Crystallography and NMR. Yale University Press, New Haven, CT. Bures,M. and Martin,Y. (1998) Curr. Opin. Chem. Biol., 2, 376–380. Caflisch,A. (1996) J. Comput.-Aided Mol. Des. 10, 372–396. Caflisch,A. and Karplus,M. (1995) Perspect. Drug. Discov. Des., 3, 51. Caflisch,A. and Karplus,M. (1996) J. Comput.-Aided Mol. Des., 10, 372–396. Caflisch,A., Miranker,A. and Karplus,M. (1993) J. Med. Chem., 36, 2142. Campbell,S., Khosravi-Far,R., Rossman,K., Clark,G. and Der,C. (1998) Oncogene, 17, 1395–1413. Catimel,B., Domagala,T., Nerrie,M., Weinstock,J., White,S., Abud,H., Heath,J. and Nice,E. (1999) Protein Pept. Lett., 6, 319–340. Chuang,E., Barnard,D., Hettich,L., Zhang,X.-F., Avruch,J. and Marshall,M.S. (1994) Mol. Cell. Biol., 14, 5318–5325. Clark,G., Drugan,J., Terrell,R., Bradham,C., Der,C., Bell,R. and Campbell,S. (1996) Proc. Natl Acad. Sci. USA, 93, 1577–1581. Drugan J., Khosravi-Far,R., White,M., Der,C., Sung,Y., Hwang,Y. and Campbell,S. (1996) J. Biol. Chem., 271, 233.

45