Virtual screening filters for the design of type II p38 MAP kinase inhibitors: A fragment based library generation approach

https://doi.org/10.1016/j.jmgm.2011.12.009Get rights and content

Abstract

In this work, we introduce the development and application of a three-step scoring and filtering procedure for the design of type II p38 MAP kinase leads using allosteric fragments extracted from virtual screening hits. The design of the virtual screening filters is based on a thorough evaluation of docking methods, DFG-loop conformation, binding interactions and chemotype specificity of the 138 p38 MAP kinase inhibitors from Protein Data Bank bound to DFG-in and DFG-out conformations using Glide, GOLD and CDOCKER. A 40 ns molecular dynamics simulation with the apo, type I with DFG-in and type II with DFG-out forms was carried out to delineate the effects of structural variations on inhibitor binding. The designed docking-score and sub-structure filters were first tested on a dataset of 249 potent p38 MAP kinase inhibitors from seven diverse series and 18,842 kinase inhibitors from PDB, to gauge their capacity to discriminate between kinase and non-kinase inhibitors and likewise to selectively filter-in target-specific inhibitors. The designed filters were then applied in the virtual screening of a database of ten million (107) compounds resulting in the identification of 100 hits. Based on their binding modes, 98 allosteric fragments were extracted from the hits and a fragment library was generated. New type II p38 MAP kinase leads were designed by tailoring the existing type I ATP site binders with allosteric fragments using a common urea linker. Target specific virtual screening filters can thus be easily developed for other kinases based on this strategy to retrieve target selective compounds.

Highlights

► Generation and application of target-specific filters for virtual screening. ► Screening filters based on DFG conformation, interactions, chemotype specificity. ► Design of type II leads from existing type I and identified allosteric fragments. ► MD simulations to explore effect of structural variations on inhibitor binding.

Introduction

The involvement of kinases in a wide range of disease conditions from inflammation to oncogenesis makes them pharmaceutically important druggable targets for therapeutic intervention [1], [2]. Majority of kinase drugs are type I inhibitors which target the catalytic ATP binding site but due to the high level of similarity within the ATP binding sites of protein kinases, it is often difficult to achieve the required pharmacological selectivity. However, each kinase has unique structural motifs, protein–protein interaction segments, specific dynamic properties, and allosteric binding site which distinguish them from each other [3], [4], [5]. This uniqueness of a protein kinase can be exploited pharmaceutically to create specificity. A number of strategies targeting different aspects of kinases like size of gatekeeper residue [6], metabolic properties of the target involved in signal transduction [7], protonation switch in drug binding [8], network analysis and fingerprint of structure and sequence of kinome [9], recognition properties affecting selectivity [10], fragment based drug design (FBDD) [11] have been employed. Most of the kinase inhibitors developed to date bind to the ATP binding site when the kinase is in its active conformation, but targeting the inactive conformations is currently being explored as a strategy for the design of more selective kinase inhibitors. Design of type II inhibitors which exploit the conserved ATP site as well as the non-conserved allosteric site of the inactive kinase conformation is a desirable option. Type II inhibitors bind to the inactive DFG-out conformation of the kinase caused due to the conformational transition of the DFG-loop. This opens up the second hydrophobic pocket referred to as the allosteric site in the current study. The DFG-loop formed by the catalytic amino acid triad Asp, Phe and Glu is found at the anterior end of activation loop and is a common feature in kinases. A number of studies have demonstrated the advantages of targeting the DFG-out binding mode of kinases in general and p38 MAP kinase (p38 MAPK) in particular [12]. The kinase inhibitors are governed by an integrated interaction-network with major motifs like hinge, conserved residues, DFG-loop, water, aromatic residues, etc. The patterns of interactions are conserved for a particular class of inhibitors and play a key role in rendering specificity [13].

Previous studies construe that a given fragment has a tendency to occupy the same part of the binding site irrespective of its position in the parent scaffold. Thus, the final inhibitor adopts the same binding mode as the original fragment hit and reiterates similar pattern of interactions [14]. The abundant structural information obtained from crystal structures of proteins and their various inhibitors facilitate fragment-based lead discovery leads [15]. Considering the availability of a large number of type I ATP site inhibitors with good efficacy but less specificity, the redesigning of these existing type I inhibitors into type II to bring in specificity is a promising approach. In the current study, we put forth the development and application of target specific virtual screening filters to identify allosteric fragments for the design of type II p38 MAP kinase leads. Allosteric site inhibitors are less in number as compared to the ATP site binders therefore virtual screening of a database of 107 compounds was undertaken to identify more number of diverse allosteric fragments. Virtual screening is an important and favorable tool for the identification of leads [16], [17]. However, considering the increase in chemical space and availability of a spectrum of screening techniques, choice of the right screening tool and filter is necessary to identify potential leads in minimum time and with maximum precision [18]. Many reports have been published which describe different docking algorithms and scoring functions for virtual screening [19], compare and contrast two or more programs [20] or probe into deeper issues like unbiased construction of benchmarks [21], [22], ensemble docking [23], induced-fit effects [24], multi-step strategies [25], etc. However, none of them have tried to decipher the influence of kinase DFG-loop transition and chemotype specificity on docking. Therefore, an evaluation of the performance of docking protocols for the twin DFG-conformations was carried out as a prelude to the design of filters. A 40 ns molecular dynamics simulation with different conformations of p38 MAPK was carried out to delineate the effects of structural variations on inhibitor binding.

Filters were designed based on the DFG-loop conformation, binding interactions and chemotype specificity. The first filter is based on the score components of the two docking protocols used in the study and the other on the sub-structure interactions. Both the filters were tested on a dataset of 249 potent p38 MAP kinase inhibitors from seven diverse series and 18,842 kinase inhibitors from PDB, to measure their capacity to discriminate between kinase and non-kinase inhibitors and also to selectively filter-in target-specific inhibitors. These filters were then used in virtual screening to identify potential hits from which a library of p38 MAPK specific allosteric fragments was extracted. New type II p38 MAPK leads were designed by merging the existing type I ATP site binders and the identified allosteric fragments with a common linker. Modelling and lead design studies are an important step in the screening and design of leads [26], [27], [28], [29], [30], [31], [32]. Target specific virtual screening filters can thus be easily developed for other kinases based on this strategy and applied to retrieve target selective compounds for the fragment based design of type II kinase leads.

Section snippets

Dataset

Four major datasets have been used in the current study at different stages. The first dataset comprising of p38 MAPK inhibitors and drugs from Protein Data Bank (PDB) was used for the evaluation of docking protocols and design of filters. The second dataset containing the diverse p38 MAPK and the third dataset containing kinase inhibitors from PDB were used for testing the designed virtual screening filters and the fourth dataset consisting 107 compounds was used for virtual screening. To

Evaluation of different docking protocols for p38 Map kinase

The design of leads for a particular drug target invokes meticulous study of the target, existing inhibitors and the tools used [19], [20], [21]. Evaluation of docking methods is therefore necessary to measure the capacity of most of these protocols to identify the correct binding pose and appropriately rank the pose having lowest RMSD difference with the co-crystal pose as the best scoring pose. Docking plays a pivotal role in the current study as it is the basic protocol used to decipher the

DFG-loop conformation and chemotype specificity

The 98 p38 MAPK crystal structures with 138 bound inhibitors comprise of seven different sequences which belong to different origin and sub-families (Table S1a). 86 of these are from Homo sapiens (p38α, p38β, p38δ, MAPK2) while 11 are from Mus musculus (Mapk14, MEF2A, MKK3b) and one chimera containing chains from both H. sapiens and M. musculus (Table S4). The activation loop in majority of the crystal structures is not phosphorylated. The DFG-loop in these crystal structures is observed to be

Virtual screening filters

Virtual screening filters are designed based on evaluation of docking protocols, DFG-loop conformation and chemotype specificity studies carried out in the previous sections. They have been developed to filter out docked compounds on basis of their binding affinity and interactions. The logic behind designing such filters is that existing p38 MAPK co-crystals represent a diverse series of DFG-in and DFG-out binders. In addition, with the observation that a chemotype binding to a particular

Virtual screening

Virtual screening was done using the fourth dataset of 107 compounds to identify potential allosteric fragments to form the tail part of the designed leads. The compounds of the dataset with a Tanimoto similarity co-efficient of >0.5 to more than two of the existing sub-structures were removed in order to ensure identification of sub-structures different from the ones comprising the existing sub-structure set. The 8 million compounds thus retained were further clustered using maximum

Fragment library generation and lead design

The binding modes and interactions of the 100 virtual screening hits were used to mark out fragments with a potential to be a part of the type II lead. The hundred virtual screening hits were pruned into fragments based on their interactions with different regions of active site in order to extract the allosteric site fragments (Fig. S7). The 98 prioritized, allosteric site interacting fragments extracted from the 100 virtual screening hits were used to formulate the fragment library (Fig. 7).

Conclusions

Achieving specificity is a problem of outstanding importance for kinases in general and p38 MAPK in particular owing to its key therapeutic potential. The drug–receptor interactions play pivotal role in inhibition as evidenced through molecular dynamics simulations deciphering the role of interaction-network in inhibitor binding. In addition, the chemotype in kinase inhibitors was observed to display specificity for a particular region of binding site. This emphasizes the use of target specific

Acknowledgement

Department of Science and Technology, New Delhi is thanked for Swarnajayanti Fellowship to GNS and Women Scientist Fellowship to PB. DBT and CSIR, New Delhi are also thanked for financial assistance.

References (69)

  • S.F. Altschul et al.

    Basic local alignment search tool

    J. Mol. Biol.

    (1990)
  • F. Filomia et al.

    Insights into MAPK p38a DFG flip mechanism by accelerated molecular dynamics

    Bioorg. Med. Chem.

    (2010)
  • R.G. Kulkarni et al.

    Novel targets for anti-inflammatory and anti-arthritic agents

    Curr. Pharm. Des.

    (2006)
  • M.E.M. Noble et al.

    Protein kinase inhibitors: insights into drug design from structure

    Science

    (2004)
  • L.A. Smyth et al.

    Measuring and interpreting the selectivity of protein kinase inhibitors

    J. Chem. Biol.

    (2009)
  • R. Morphy

    Selectively nonselective kinase inhibition: striking the right balance

    J. Med. Chem.

    (2010)
  • F. Zuccotto et al.

    Through the “Gatekeeper door”: exploiting the active kinase conformation

    J. Med. Chem.

    (2010)
  • I. Collins et al.

    Design and development of signal transduction inhibitors for cancer treatment: experience and challenges with kinase targets

    Curr. Signal Transduct. Ther.

    (2006)
  • Y. Shana et al.

    A conserved protonation-dependent switch controls drug binding in the Abl kinase

    Proc. Natl. Acad. Sci. U.S.A.

    (2009)
  • D. Huang et al.

    Kinase selectivity potential for inhibitors targeting the ATP binding site: a network analysis

    Bioinformatics

    (2010)
  • R. Soliva et al.

    Dissection of the recognition properties of p38 MAP kinase determination of the binding mode of a new pyridinyl-heterocycle inhibitor family

    J. Med. Chem.

    (2007)
  • J.J. Sutherland et al.

    Chemical fragments as foundations for understanding target space and activity prediction

    J. Med. Chem.

    (2008)
  • C. Pargellis et al.

    Inhibition of p38 MAP kinase by utilizing a novel allosteric binding site

    Nat. Struct. Biol.

    (2002)
  • A.K. Ghose et al.

    Knowledge based prediction of ligand binding modes and rational inhibitor design for kinase drug discovery

    J. Med. Chem.

    (2008)
  • P. Badrinarayan et al.

    Sequence structure, and active site analyses of p38 MAP kinase: exploiting DFG-out conformation as a strategy to design new type II leads

    J. Chem. Inf. Model.

    (2011)
  • S.V. Shelke et al.

    A fragment-based in situ combinatorial approach to identify high-affinity ligands for unknown binding sites

    Angew. Chem. Int. Ed.

    (2010)
  • A.S. Reddy et al.

    Virtual screening in drug discovery—a computational perspective

    Curr. Protein Pept. Sci.

    (2007)
  • P. Badrinarayan et al.

    Virtual high-throughput screening in new lead identification

    Comb. Chem. High Throughput Screen.

    (2011)
  • V. Zoete et al.

    Docking virtual high throughput screening and in silico fragment-based drug design

    J. Cell Mol. Med.

    (2009)
  • M. Stahl et al.

    Detailed analysis of scoring functions for virtual screening

    J. Med. Chem.

    (2001)
  • E. Perola et al.

    A detailed comparison of current docking and scoring methods on systems of pharmaceutical relevance

    Proteins Struct. Funct. Bioinf.

    (2004)
  • S.M. Vogel et al.

    DEKOIS: demanding evaluation kits for objective in silico screening—a versatile tool for benchmarking docking programs and scoring functions

    J. Chem. Inf. Model.

    (2011)
  • S.G. Rohrer et al.

    Maximum unbiased validation (MUV) data sets for virtual screening based on PubChem bioactivity data

    J. Chem. Inf. Model.

    (2009)
  • B. Musafia et al.

    Bioactive conformational biasing: a new method for focusing conformational ensembles on bioactive-like conformers

    J. Chem. Inf. Model.

    (2009)
  • Cited by (0)

    View full text