Int J Pept Res Ther DOI 10.1007/s10989-013-9382-8
Evolution of High-Affinity Peptide Probes to Detect the SH3 Domain of Cancer Biomarker BCR–ABL Hui-Min Liu • Li-Juan Li • Juan Guo • Zhan-Jia Yang • Xiao Yang • Run-Peng Qi Wei Cao
•
Accepted: 20 November 2013 Ó Springer Science+Business Media New York 2013
Abstract The BCR–ABL fusion protein is closely associated with the pathological progression of chronic myelogenous leukaemia and some other myeloproliferative diseases, which has long been recognized as one of the most important cancer biomarkers in the tumor diagnosis community. The SH3 domain of BCR–ABL is a small, conserved protein module that specifically recognizes and binds proline-rich peptide fragments. In the current study, we used a synthetic strategy to discover new peptide probes with high affinity binding to the BCR–ABL SH3 domain. In the procedure, a sequence-based machine learning predictor was developed based on a set of affinity-known SH3 binders, and the predictor was then used to guide the evolutional optimization of numerous virtual peptides to enrich high binding potency for the SH3 domain. Subsequently, a evolved peptide population was generated, from which ten peptides with the highest affinity scores were selected and their interaction free energies with SH3 domain were characterized systematically using a combination of molecular dynamics simulation and binding free
Electronic supplementary material The online version of this article (doi:10.1007/s10989-013-9382-8) contains supplementary material, which is available to authorized users. H.-M. Liu (&) L.-J. Li J. Guo Z.-J. Yang X. Yang R.-P. Qi Department of Clinical Laboratory, Zhengzhou People’s Hospital Attached to Southern Medical University, Zhengzhou 450003, People’s Republic of China e-mail:
[email protected] W. Cao (&) Clinical Research Center, Zhengzhou People’s Hospital Attached to Southern Medical University, Zhengzhou 450003, People’s Republic of China e-mail:
[email protected]
energy analysis. Consequently, four peptides were suggested as promising candidates and their affinities toward SH3 domain were assayed; two peptides, APTYTPPPPP and APTYAPPPPP, were identified to have potent binding capability with dissociation constants Kd of 3 and 8 lM, respectively. Further, the structural basis and energetic property of SH3 domain in complex with APTYTPPPPP were examined in detail, revealing a non-specific interaction in SH3–peptide recognition that should render a broad ligand spectrum for the domain. Keywords Cancer biomarker Tumor diagnosis BCR–ABL SH3 domain Peptide
Introduction The BCR–ABL chimeric protein is thought to play a central role in the pathogenesis of myeloproliferative diseases (Clark et al. 1988). This abnormality is due to the reciprocal translocation between chromosomes 9 and 22 [t(9;22)], which brings the 50 end of the BCR gene into juxtaposition with the tyrosine kinase domain of the ABL gene to produce a hybrid gene retaining tyrosine kinase activity, which encodes a 210 kDa BCR–ABL fusion protein (Nashed et al. 2003). Clinical and laboratory studies indicate that the BCR–ABL is essential for initiation, maintenance and progression of chronic myelogenous leukaemia (CML), yet the transformation of CML from chronic phase to blast phase requires additional genetic and/or epigenetic abnormalities (Ren 2005). Over the past decades, the BCR–ABL is one of the most well-established cancer biomarkers, which has long been simply used to detect CML and some other subtypes of leukemia. However, drug developers were eventually able to discover
123
Int J Pept Res Ther
imatinib, a powerful agent that effectively inhibits this fusion protein and significantly decreases production of cells containing the Philadelphia chromosome (Druker et al. 2001). In addition to imatinib, nowadays a number of BCR–ABL tyrosine kinase inhibitors such as dilotinib and dasatinib have been approved or are under clinical/preclinical development (Mughal et al. 2013). Very recently, Lambert et al. (2013) addressed a comprehensive review on the background, discovery and clinical development of second-generation small-molecule BCR–ABL inhibitors to combat the acquired drug resistance of CML. Due to the importance of BCR–ABL in the pathological progress of CML as well as its targeting therapy, rational design of molecular probes to recognize and bind this protein with high affinity is fundamentally significant for detecting and treating the CML. The BCR–ABL oncoprotein is consisted of N-terminal oligomerization domain, bcr region, tyrosine kinase domain, SH3, SH2 and C-terminal sequence (Sawyers 1993), in which the SH3 domain (Src homology 3 domain) is a small protein module of about 60 amino acid residues that recognizes and binds polyproline helix (PPII helix) peptides on the surface of partner proteins to orchestrate complicated protein–protein interaction networks (Li 2005). In the current study, we aimed at the rational design of high-affinity peptide probes to potently target the SH3 domain of BCR–ABL by integrating a number of strategies. With this approach we were able to derive a series of peptide fragments that exhibited both high potency and selectivity for the SH3 domain. These peptide probes can be further conjugated with functional molecular entities such as kinase inhibitors and fluorescent labels for practical applications. We also gave a detailed analysis regarding the molecular basis and biological implication underlying the interaction between the designed peptides and SH3 domain. This work would help to establish a systematic framework for rational peptide probe development.
Fig. 1 The flowchart representation of the rational evolution of SH3binding peptide probes
peptide samples with high scores in the evolved population were selected to perform atomistic molecular dynamics (MD) simulations on the basis of their complex structures with SH3 domain, and the interaction free energies between the SH3 and peptides were computed using a post molecular mechanics/Poisson Boltzmann (MM–PB/SA) analysis. In the third step (step 3): few peptides that were identified as promising binders by the MM–PB/SA analysis were extracted and their affinities to the SH3 domain were assayed using fluorescence. Consequently, two high-affinity peptide probes were obtained and their interactions with SH3 domain were examined in detail at molecular level. Peptide Population Evolution Development of Machine Learning Predictors
Materials and Methods Overview of the Integrated Protocol The protocol proposed in this study to rationally evolve peptide probes with high affinity to BCR–ABL SH3 domain is schematically shown in Fig. 1, which can be divided into three steps; In the first step (step 1): a set of sequence-based SH3–peptide affinity data was collected from a number of literatures, which was utilized to train a sequence-level SH3-binding peptide affinity predictor. The predictor was then used as the scoring function in a computational evaluation procedure to optimize a population of virtual peptide probes. In the second step (step 2): several
123
A total of 395 ABL SH3-binding peptides were compiled from a number of previous publications (see Supporting Information Table S1) (Hou et al. 2009; Xu et al. 2012). These peptides were labeled with Cy-3 fluorescence and their binding strengths toward SH3 domain were measured quantitatively on peptide microarray; the fluorescent intensity of a microarray spot was defined as its own intensity minus the background intensity around it, as determined from the scanned image (Xu et al. 2012). Here, we used logarithmic intensity values to perform modeling analysis. The 395 peptide samples were randomly split into a training set consisted of 300 peptides and a test set containing 95 samples; the former was employed to
Int J Pept Res Ther
X represents any residue) was randomly generated and their affinities to SH3 were scored using the optimal machine learning predictor developed above. The score of each peptide i was converted to a probability pi using P following formula: pi ¼ si = Ni ðsi sm Þ, where N is the population size 1,000, and si and sm are the predicted affinities of ith peptide and the lowest-affinity peptide in the population, respectively. Then, a new population with the same size was derived from the old using a roulette wheel selection strategy, that is, an individual was copied from the old population with probability pi; the copy was repeated for 1,000 times to obtain the 1,000 individuals in the new population. Subsequently, *5 % residues of peptides in the new population were randomly mutated to other amino acid types. The mutated population was then imposed with a one-point crossover operation to randomly hybridize any two peptides in the population. MD Simulation and MM–PB/SA Analysis
Fig. 2 Schematic representation of peptide population evolution
statistically develop machine learning models, and the latter would be used to blindly validate the developed models. Five widely used amino acid descriptors, including z-scale (Hellberg et al. 1987), t-scale (Tian et al. 2007), WHIM (Zaliani and Gancia 1999), VHSE (Mei et al. 2005) and DPPS (Tian et al. 2009), were adopted here to characterize the amino acid composition and sequence pattern of a peptide. These descriptors were generated by principal component analysis of numerous physicochemical and/or topological properties of 20 amino acids. Subsequently, the statistical correlation between the characterized variables and the logarithmic intensity values of the 395 peptides was established by using linear partial least squares (PLS) (Boulesteix and Strimmer 2007) and nonlinear support vector machine (SVM) (Cortes and Vapnik 1995). The two regression methods are sophisticated machine learning techniques that have been widely used in peptide activity prediction (Zhou et al. 2013). Here, the PLS and SVM regressions were implemented with the programs ChemoAC and LIBSVM, respectively. Evolution of peptide population A evolutionary algorithm modified from previous work (Jing et al. 2013) was utilized to optimize a peptide population, aiming at improving SH3-binding affinity for these peptides. The evaluation procedure is illustrated in Fig. 2. The initial population consisted of 1,000 decapeptides with consensus motif XXXXXXPXXP (where P is proline and
The crystal structure of ABL SH3 domain in complex with a cognate decapeptide APSYSPPPPP (PDB: 1bbz) was used as template to generate the complex structure model of the domain with other peptide ligands. In the procedure, the decapeptide in template was virtually mutated to a target peptide ligand using the SCWRL program (Krivov et al. 2009), which predicted optimal rotamer combination for the side chains of the target based on template peptide backbone. The mutated SH3–peptide complex structure was submitted to a molecular dynamics (MD) simulation with AMBER03 force field (Duan et al. 2003) implemented in the AMBER9.0 package (Case et al. 2005). A scheme modified from (Hou et al. 2006a, b) was used here: a TIP3P ˚ away from any solute atom and a water box extended 9 A ? number of Na counter ions were added to the complex system. First, all hydrogen atoms, water molecules and counter ions were minimized, followed by a minimization to relax all atoms without constraints. Then, the MD simulation was consisted of a gradual temperature increase from 0 to 300 K over 100 ps and a 5 ns equilibration for data collection. The resulting trajectory was analyzed using the MM–PB/SA method (Kollman et al. 2000) to calculate the interaction free energy between the SH3 domain and its peptide ligand. Very recently, Gumbart et al. ( 2013) have successfully reproduced the experimental binding free energy of peptides to ABL SH3 domain with potentials of mean force (PMF) theory on the basis of a MD framework. Peptide Affinity Analysis The binding affinity between the domain and peptide was determined using fluorescence spectroscopy as described
123
Int J Pept Res Ther Table 1 Statistics of regression models with different combinations of machine learning methods and amino acid descriptors Machine learning
Descriptor
Training set (300 samples)
Test set (95 samples)
r2fit
r2cv
r2prd 0.313
PLS
z-scale
0.454
0.396
PLS
t-scale
0.389
0.320
0.294
PLS
WHIM
0.354
0.236
0.221
PLS
VHSE
0.430
0.328
0.307
PLS
DPPS
0.456
0.337
0.325
SVM
z-scale
0.432
0.329
0.310
SVM
t-scale
0.529
0.414
0.497
SVM
WHIM
0.446
0.354
0.332
SVM
VHSE
0.586
0.489
0.520
SVM
DPPS
0.687
0.594
0.578
previously (Pisabarro and Serrano 1996). Briefly, the fluorescence emission spectra of the tryptophan residues in the SH3 domain were used to monitor any changes in their environment upon peptide binding. Fluorescence was measured at an excitation wavelength of 298 nm with a 2 nm slit width and an emission wavelength with a 4 nm slit. The protein concentration was kept at 1 lM. All experiments were performed at room temperature. Upon addition of the peptide solution, changes in fluorescence were measured. The experimental data were fitted to the equation F ¼ Fmax ½p=ðKd þ ½pÞ (Schweimer et al. 2002), where [p] is the final peptide concentration at each measurement point, F is the measured protein fluorescence intensity at the particular peptide concentration, and Fmax is the observed maximal fluorescence intensity of the protein when saturated with the peptide.
Results and Discussion Optimization of Machine Learning Predictors Here, five amino acid descriptors (i.e. z-scale, t-scale, WHIM, VHSE and DPPS) and two machine learning methods (i.e. PLS and SVM) were available to develop statistical regression models. In order to build the optimal predictor, we have systematically explored all the ten possible combinations between the descriptors and the methods in predicting SH3–peptide affinity. Prior to this, considering that blind validation was highlighted previously as the only way to develop predictive models (Golbraikh and Tropsha 2002), the 395 peptide samples tabulated in Table S1 were randomly split into a training set consisted of 300 peptides and a test set containing 95
123
peptides; the regression models of different descriptormethod combinations were then developed based on the training set, and their statistics are listed in Table 1, where the significant principal number of PLS was detected by 10-fold cross-validation and the optimal parameter combination (including insensitive loss function e, penalty C and radial width c) for SVM was determined by grid search. As can be seen, the SVM performed much well as compared to PLS in fitting through the 300 training samples, with coefficients of determination r2fit of resulting models ranging from 0.432 (z-scale) to 0.687 (DPPS). This is reasonable if considering that the binding of highly flexible peptides to SH3 domain is a very complicated behavior that may not be described accurately using the simple linear PLS approach; instead, the nonlinear SVM adopted RBF kernel to sufficiently explore peptide structure–affinity relationship in a high-dimensional space, which was expected to obtain a better result than PLS. To verify the external predictive power of these built models, they were further used to perform extrapolation over the 95 test peptides. The obtained statistics of different descriptor-method combinations are summarized in Table 1 and illustrated in Fig. 3a. As might be anticipated, the SVM method also showed a much better performance than PLS; the optimal prediction was achieved by the combination of DPPS descriptor and SVM method, with predictive coefficient r2prd = 0.578. The DPPS descriptor was originally developed by Tian et al. (2009) by extracting ten informative components from 23 electronic properties, 37 steric properties, 54 hydrophobic properties and 5 hydrogen bond properties of 20 amino acids, which has been successfully applied to quantitative structure– activity relationship (QSAR) modeling of diverse peptide bioactivities. The experimentally measured affinities against calculated values for the 300 training and 95 test samples derived from the optimal model are plotted in Fig. 3b. It is seen that the sample points distributed evenly across the plot but most low-affinity peptides appear to be generally overestimated by the model. This is acceptable because many factors such as peptide flexibility and the interactive effect among different peptide residues were not considered in the modeling that may cause the systematic bias. Virtual Evolution of Peptide Population A initial population consisted of 1,000 decapeptides with the consensus motif XXXXXXPXXP (where P is proline and X represents any residue) was generated randomly; this population was then evolved according to the scheme shown in Fig. 2, and the affinity of each peptide in the population was scored by the SVM–DPPS predictor developed above. After hundreds of cycle the evaluation
Int J Pept Res Ther
Fig. 3 a The predictive powers (r2prd) of different combinations of machine learning methods and amino acid descriptors. b Plot of experimental affinities (logINT) against calculated values using the optimal SVM–DPPS combination for the 300 training samples and 95 test samples
Fig. 4 The affinity distributions of a initial and b evolved populations
procedure achieved convergence and the final population was outputted. In order to avoid randomness, the evaluation was repeated for 10 times. The optimal evolved population was considered here as an optimized candidate pool from which high-affinity peptide probes were selected. We compared the theoretical affinity distributions of the 1,000 peptides in initial and evolved populations, as shown in Fig. 4. It is evident that the two distribution profiles are distinct; the initial population exhibits a monotone decreasing with peptide affinity increasing, that is, most members are fallen into low- or moderate-affinity interval pKd = 2.5–4.0 (where pKd : –logKd), and only very few peptides possess high affinity pKd [ 4.5. By contrast, the evolved population shows an unimodal pattern with peak position located at pKd = *4.25. The comparison of peptide affinity distributions clearly suggests that the evolution procedure can effectively improve the SH3-binding strength of peptide members in the population, and thus the
evolved peptides could be regarded as a promising candidate pool to discover new high-affinity probes. Here, we selected the top-ten peptide members with highest theoretical affinities in the evolved population, and their sequences as well as predicted pKd values are listed in Table 2. As can be seen, the ten candidates share a similar sequence feature; their C-terminuses are rich of proline residues that define moderate hydrophobicity and type II polyproline (PPII) helix for the peptides, and their N-terminuses are occupied by small, amphiphilic amino acids that have large variability and confer specificity for peptide recognition. In particular, the sequence pattern [A/P/S]P[T/ S]Y[T/S]P[P/L/A/V]PPP is conserved among these peptides; the pattern has been observed to be shared by a number of known ABL SH3 ligands, for example, APSYSPPPPP (Pisabarro et al. 1998) and APTYSPPLPP (Hou et al. 2006b). Furthermore, we performed an exhaustive literature survey and found two (APTYSPPPPP
123
Int J Pept Res Ther
Measure
be 0.4 (Pisabarro and Serrano 1996) and 7 lM (Xu et al. 2012), respectively.
DGttl (kcal/ mol)
Kd (lM)
Binding Affinity Analysis
87.6
-44.4
3
-147.4
97.3
-50.1
0.4b
4.93
-154.2
116.4
-37.8
19
SPTYPPPLPP
4.92
-146.4
100.8
-45.6
7c
APSYTPPLPP
4.90
-132.7
86.7
-46.0
27
APTYAPPPPP
4.90
-146.9
95.1
-51.8
8
PPAFPPPVPP
4.88
-124.0
105.9
-18.1
ND
PPSFSPPAPP
4.87
-115.1
82.5
-32.6
ND
GPTFSPPPVP
4.86
-125.9
88.0
-37.9
ND
LPSYPPPLPP
4.86
-128.3
103.4
-24.9
ND
Table 2 The top-10 peptides in the evolved population Peptide
Predictor
MM–PB/SA
pKad
DGnbd (kcal/ mol)
DGslv (kcal/ mol)
APTYTPPPPP
4.96
-132.0
APTYSPPPPP
4.95
VPSYSPPPPP
ND Not determined a
pKd : –log Kd
b
Taken from Pisabarro and Serrano (1996)
c
Taken from Xu et al. (2012)
and SPTYPPPLPP) of the ten selected peptides have already been identified previously as effective SH3 binders, and their affinity values were measured experimentally to
In order to obtain a comprehensive picture about the structural basis and energetic property of the ten promising peptide candidates in interaction with SH3 domain, we herein performed MD simulation and MM–PB/SA analysis to characterize SH3 complexes with these peptides. First, 5 ns MD equilibration was addressed to generate a trajectory file for each of SH3–peptide complexes and, subsequently, the file was analyzed using a post MM–PB/SA scheme to determine total binding free energy and its decomposed components for the complex. The calculated results are tabulated in Table 2. As expected, all the ten peptides display favorable affinity toward SH3 domain; their total free energies DGttl range from -18.1 to -51.8 kcal/mol, albeit these values might be overestimated if considering that the entropic penalty due to the binding of highly flexible peptides to SH3 domain was not taken into account. However, entropic effect could be regarded as a constant over these peptides since they share a similar binding mode and have the same sequence length. Moreover, the total free energy DGttl can be decomposed into
Fig. 5 a Stereoview of BCR–ABL SH3 domain–APTYTPPPPP complex structure. b Schematic representation of nonbonded interactions across the complex interface. This plot was produced using LIGPLOT program (Wallace et al. 1995)
123
Int J Pept Res Ther
nonbonded contribution DGnbd and solvent effect DGslv; the former characterizes the direct nonbonded interactions such as hydrogen bonding and van der Waals force at the complex interface and the latter describes indirect desolvation penalty associated with the complex formation. As can be seen from Table 2, the solvent effect is always unfavorable to peptide binding with associated energetic penalty of as much as DGslv = *100 kcal/mol, which, however, can be completely compensated from the favorable nonbonded contribution ranging from -115 to -150 kcal/mol. Consequently, the total free energy DGttl is preferable for these peptide binding. Furthermore, four out of the ten peptide candidates, i.e. APTYTPPPPP, VPSYSPPPPP, APSYTPPLPP and APTYAPPPPP, which were predicted by SVM–DPPS to have high affinities (pKd C 4.90) and characterized by MM–PB/SA to possess strong binding potencies (DGttl \ -30 kcal/ml), were assayed to determine their affinities toward SH3 domain, resulting in Kd values of 3, 19, 27 and 8 lM, respectively. It is seen that the measured affinities (pKd = 5.52, 4.72, 4.56 and 5.10, respectively) are basically consistent with the predicted values, in which two peptides APTYTPPPPP and APTYAPPPPP possess particularly high binding capability to SH3 domain (Kd \ 10 lM). Here, the complex structure model of SH3 domain with APTYTPPPPP is shown in Fig. 5a, and the nonbonded interactions across the complex interface is depicted in Fig. 5b. Evidently, the interface can be found with intensive non-specific hydrophobic interactions, whereas only one specific hydrogen bond is identified, suggesting a high stability but low specificity of the complex. This is consistent with the fact that SH3 family members exhibit strong ligand cross-reactivity (Cesareni et al. 2002).
Conclusions Peptides have long been used as molecular probes to detect cancer cells and tumor-related proteins. In this study, we attempted to design new peptide entities that possess high affinity to BCR–ABL SH3 domain. The peptides may also be used as potential candidates to disrupt the pathological function of BCR–ABL and to develop peptide agents to treat CML, although they do not target the functional kinase domain of BCR–ABL. To achieve this, we proposed a synthetic protocol to discover potent SH3 binders. With this protocol, we were able to derive a number of promising peptide candidates, from which two peptides were identified as high-affinity binders of SH3 domain. A further theoretical analysis revealed that the binding of peptide ligands to the SH3 receptor would incur significant desolvation penalty, which, however, could be completely
compensated from the favorable nonbonded interactions across SH3–peptide complex interface, and the lack of specific chemical forces at the interface may result in a board ligand spectrum for the SH3 domain. Acknowledgments This work was supported by the National Natural Science Foundation of China (No. 81171992). Conflict of interest
None.
References Boulesteix AL, Strimmer K (2007) Partial least squares: a versatile tool for the analysis of high-dimensional genomic data. Brief Bioinform 8:32–44 Case DA, Cheatham TE, Darden T, Gohlke H, Luo R, Merz KM (2005) The amber biomolecular simulation programs. J Comput Chem 26:1668–1688 Cesareni G, Pann S, Nardelli G, Castagnoli L (2002) Can we infer peptide recognition specificity mediated by SH3 domains? FEBS Lett 513:38–44 Clark SS, McLaughlin J, Timmons M, Pendergast AM, Ben-Neriah Y, Dow LW, Crist W, Rovera G, Smith SD, Witte ON (1988) Expression of a distinctive BCR–ABL oncogene in Ph1-positive acute lymphocytic leukemia (ALL). Science 239:775–777 Cortes C, Vapnik VN (1995) Support-vector networks. Mach Learn 20:273–297 Druker BJ, Sawyers CL, Kantarjian H, Resta DJ, Reese SF, Ford JM, Capdeville R, Talpaz M (2001) Activity of a specific inhibitor of the BCR–ABL tyrosine kinase in the blast crisis of chronic myeloid leukemia and acute lymphoblastic leukemia with the Philadelphia chromosome. N Engl J Med 344:1038–1042 Duan Y, Wu C, Chowdhury S, Lee MC, Xiong GM, Zhang W (2003) A point-charge force field for molecular mechanics simulations of proteins based on condensed-phase quantum mechanical calculations. J Comput Chem 24:1999–2012 Golbraikh A, Tropsha A (2002) Beware of q2! J Mol Graph Model 20:269–276 Gumbart JC, Roux B, Chipot C (2013) Standard binding free energies from computer simulations: what is the best strategy? J Chem Theory Comput 9:794–802 Hellberg S, Sjo¨stro¨m M, Skagerberg B, Wold S (1987) Peptide quantitative structure-activity relationships, a multivariate approach. J Med Chem 30:1126–1135 Hou T, McLaughlin W, Lu B, Chen K, Wang W (2006a) Prediction of binding affinities between the human amphiphysin-1 SH3 domain and its peptide ligands using homology modeling, molecular dynamics and molecular field analysis. J Proteome Res 5:32–43 Hou T, Chen K, McLaughlin WA, Lu B, Wang W (2006b) Computational analysis and prediction of the binding motif and protein interacting partners of the Abl SH3 domain. PLoS Comput Biol 2:e1 Hou T, Xu Z, Zhang W, McLaughlin WA, Case DA, Xu Y, Wang W (2009) Characterization of domain-peptide interaction interface: a generic structure-based model to decipher the binding specificity of SH3 domains. Mol Cell Proteomics 8:639–649 Jing T, Feng J, Li D, Liu J, He G (2013) Rational design of angiotensin-I-converting enzyme inhibitory peptides by integrating in silico modeling and an in vitro assay. Chem Med Chem 8:1057–1066
123
Int J Pept Res Ther Kollman PA, Massova I, Reyes C, Kuhn B, Huo SH (2000) Calculating structures and free energies of complex molecules: combining molecular mechanics and continuum models. Acc Chem Res 33:889–897 Krivov GG, Shapovalov MV, Dunbrack RL Jr (2009) Improved prediction of protein side-chain conformations with SCWRL4. Proteins 77:778–795 Lambert GK, Duhme-Klair AK, Morgan T, Ramjee MK (2013) The background, discovery and clinical development of BCR–ABL inhibitors. Drug Discov Today 18:992–1000 Li SS (2005) Specificity and versatility of SH3 and other prolinerecognition domains: structural basis and implications for cellular signal transduction. Biochem J 390:641–653 Mei H, Liao ZH, Zhou Y, Li SZ (2005) A new set of amino acid descriptors and its application in peptide QSARs. Biopolymers (Pept Sci) 80:775–786 Mughal A, Aslam HM, Khan AM, Saleem S, Umah R, Saleem M (2013) BCR–ABL tyrosine kinase inhibitors—current status. Infect Agent Cancer 8:23 Nashed AL, Rao KW, Gulley ML (2003) Clinical applications of BCR–ABL molecular testing in acute leukemia. J Mol Diagn 5:63–72 Pisabarro MT, Serrano L (1996) Rational design of specific highaffinity peptide ligands for the Abl-SH3 domain. Biochemistry 35:10634–10640 Pisabarro MT, Serrano L, Wilmanns M (1998) Crystal structure of the abl-SH3 domain complexed with a designed high-affinity peptide ligand: implications for SH3-ligand interactions. J Mol Biol 281:513–521
123
Ren R (2005) Mechanisms of BCR–ABL in the pathogenesis of chronic myelogenous leukaemia. Nat Rev Cancer 5:172–183 Sawyers CL (1993) Molecular consequences of the BCR–ABL translocation in chronic myelogenous leukemia. Leuk Lymphoma 11:S101–S103 Schweimer K, Hoffmann S, Bauer F, Friedrich U, Kardinal C, Feller SM, Biesinger B, Sticht H (2002) Structural investigation of the binding of a herpesviral protein to the SH3 domain of tyrosine kinase Lck. Biochemistry 41:5120–5130 Tian F, Zhou P, Li Z (2007) T-scale as a novel vector of topological descriptors for amino acids and its application in QSARs of peptides. J Mol Struct 830:106–115 Tian F, Yang L, Lv F, Yang Q, Zhou P (2009) In silico quantitative prediction of peptides binding affinity to human MHC molecule: an intuitive quantitative structure-activity relationship approach. Amino Acids 36:535–554 Wallace AC, Laskowski RA, Thornton JM (1995) LIGPLOT: a program to generate schematic diagrams of protein-ligand interactions. Protein Eng 8:127–134 Xu Z, Hou T, Li N, Xu Y, Wang W (2012) Proteome-wide detection of Abl1 SH3-binding peptides by integrating computational prediction and peptide microarray. Mol Cell Proteomics 11:O111.010389 Zaliani A, Gancia E (1999) MS-WHIM scores for amino acids: a new 3D-description for peptide QSAR and QSPR studies. J Chem Inf Comput Sci 39:525–533 Zhou P, Wang C, Ren Y, Yang C, Tian F (2013) Computational peptidology: a new and promising approach to therapeutic peptide design. Curr Med Chem 20:1985–1996