Elsevier

Methods

Volume 71, 1 January 2015, Pages 113-134
Methods

Methods for generating and applying pharmacophore models as virtual screening filters and for bioactivity profiling

https://doi.org/10.1016/j.ymeth.2014.10.013Get rights and content

Highlights

  • Pharmacophore models represent the 3D-arrangement of the chemical functionalities that make a molecule active towards its target.

  • Pharmacophore models are constructed either the ligand-based or the structure-based way.

  • Pharmacophore models are used for virtual screening, parallel screening, target fishing, and for SAR-studies.

Abstract

Biological effects of small molecules in an organism result from favorable interactions between the molecules and their target proteins. These interactions depend on chemical functionalities, bonds, and their 3D-orientations towards each other. These 3D-arrangements of chemical functionalities that make a small molecule active towards its target can be described by pharmacophore models. In these models, chemical functionalities are represented as so-called features. Commonly, they are obtained either from a set of active compounds or directly from the observed protein–ligand interactions as present in X-ray crystal structures, NMR structures, or docking poses. In this review, we explain the basics of pharmacophore modeling including dataset generation, 3D-representations and conformational analysis of small molecules, pharmacophore model construction, model validation, and its benefits to virtual screening and other applications.

Introduction

When a small molecule enters an organism, such as the human body, it has thousands of proteins to potentially interact with. Which of these proteins (typically enzymes, receptors, and transporters) it chooses as targets, is defined by chemical interactions: the small molecule binds to the proteins, where the ligand–protein interactions are energetically favorable. Already in 1890, Emil Fischer described the lock-key hypothesis on enzyme-activity: only the key with the right size and shape opens a specific lock, and similarly an enzyme chooses its substrate [1]. The same principle is also suitable for receptors. In the chemistry world, this translates to the theory that only the small molecules with the right size and correct complementary chemical functionalities can bind to the target protein and cause a biological effect (Fig. 1).

The chemical functionalities of the amino acid residues in the binding site, the binding site size, and its shape determine which small molecules it tolerates. Therefore, all molecules binding to the same binding site share similar chemical functionalities, size, and shape restrictions. The chemical functionalities that are needed for a small molecule to block or activate its target protein can be represented as pharmacophore models [2]. The concept of pharmacophore has evolved since the early 1900s: First, it was describing actual chemical functionalities called haptophore and toxophore by Paul Ehrlich. Since the 1960s, the modern term pharmacophore describing the chemical functionalities as abstract features was introduced by Schuler [3], [4], [5]. The basic theory behind pharmacophore modeling is that common chemical functionalities in similar 3D arrangements lead to a biological activity on the same target. Pharmacophore models consist of a defined 3D arrangement of so-called features that represent the chemical functionalities of active small molecules: hydrogen bond acceptors (HBAs), hydrogen bond donors (HBDs), hydrophobic areas (H), positively and negatively ionizable groups (PI/NI), and metal coordinating areas (M). Additional size restrictions in the form of a shape or exclusion volumes (XVOL) – forbidden areas – can be added to represent the size and the shape of the binding pocket. Since the models themselves do not focus on actual atoms, but chemical functionalities, they are good tools in recognizing similarities between molecules. For example, a simple hydrogen bond donor could be an NH2-group or an OH-group.

In relation to Fischer’s theory, pharmacophore models work like soap prints of the keys that fit to a lock: before trying all the available keys to the lock, they are first fitted to the soap print, and the ones that definitely do not fit can be excluded from the trial already. Thus, the aim of pharmacophore modeling and pharmacophore-based virtual screening is to predict activities by sorting the compounds into actives (compounds that match the model) and inactives (compounds not fitting to the model). The output of such a screen is a list of compounds (hit list) that are proposed to be active. Therefore, the advantage of pharmacophore-based virtual screening in the drug discovery process is that most of the compounds with low probability to be active can be excluded from further studies already in a very early stage of a project. Thereby, a lot of resources in the further drug discovery process, especially in in vitro experiments, can be saved: To find 50 new active compounds for a specific target, thousands of molecules need to be tested if one uses in vitro high-throughput screening methods, but with the help of pharmacophore modeling, one only has to evaluate a few hundred compounds experimentally [6], [7].

Pharmacophore models can be constructed via two ways: ligand-based and structure-based. The ligand-based method is used when no 3D structure of the target protein is available, but there is at least information on active molecules for this target. The structure-based method is applicable in cases where the 3D structure of the protein is known, e.g. as an X-ray crystal structure (ideally in complex with an active small molecule ligand), NMR structure, or homology model. In addition, Klabunde et al. and Sanders et al. [8], [9] introduced a way to generate pharmacophore models from a target protein sequence without any information of the crystal structure or ligands. In their method, they derived structure-based pharmacophore models from G-protein coupled receptor crystal structures and their homology models. Each of these pharmacophore models was analyzed and the pharmacophore feature-interacting residues were identified. In case the interaction was present in multiple cases, it was marked and stored as a residue-feature-pair. These kinds of pairs can be then applied to any G-protein coupled receptor sequence: in case a specific residue is found, the corresponding feature with its coordinates will be added to a pharmacophore model. This method enables therefore pharmacophore model generation based on a protein sequence only. However, it requires pre- calculated data on a protein family with high sequence- and structural identity.

The development of a high quality pharmacophore model is a multi-step procedure. Independent from the generation method, a pharmacophore model should be first theoretically validated before applying it to prospective virtual screening. After virtual screening, compounds from the hit list can be experimentally validated. Depending on the results, the pharmacophore model can be improved using the newly generated activity data. Later, additional pharmacophore model refinement should be done if the model did not perform well in experimental validations or if there is new information on active compounds that do not support the old model hypothesis [10].

In this publication, we comprehensively review the principles of pharmacophore modeling: model construction, its theoretical validation, use as virtual screening filter, application to structure–activity relationship-predictions, and as a bioactivity profiling tool. We guide the reader through the model generation and validation process. Finally, we also outline the limits and future challenges for the pharmacophore modeling field.

Section snippets

Pharmacophore modeling

Pharmacophore modeling, as every virtual screening study, begins with a thorough literature survey: What is already known about the target? How is the binding site composed? Are there already known small molecules that bind to the target? Is there already a 3D structure (crystal structure or homology model) of the target? For ligand-based pharmacophore generation, at least two active molecules or one rigid, highly active compound are needed. In case of structure-based modeling, a 3D-protein

Virtual screening

In virtual screening, the pharmacophore models are used as a filter. All the compounds that fit to the model are called hits, and they are proposed to be active against the modeled target. Since the pharmacophore features represent functionalities, not exact chemical groups, pharmacophore models are excellent tools for scaffold hopping and identifying structurally diverse compounds as hits [94].

Pharmacophore model refinement

The first in vitro experiments with newly constructed pharmacophore models will confirm or negate the predictive power of the pharmacophore model. In case the experimental validation did not yield satisfying results, it is advisable to either reject the model from further virtual screening studies, or refine it for improved performance. Additionally, pharmacophore model quality needs to be checked regularly, especially in fields where new highly active compounds are reported [10]. The goal of

Combinatorial use of pharmacophore models

As described above, the use of multiple pharmacophore models has advantages over using only one model for predictions. Similarly, the use of multiple pharmacophore modeling programs results in a higher probability to find active compounds. It has been shown that the use of different pharmacophore-based screening programs can yield hit lists with low overlap, even when the models are identical in their feature compositions and locations [87]. However, in the reported study, each hit list

SAR modeling with pharmacophore models

Since pharmacophore models are tools for distinguishing active compounds from inactive ones, they are useful in the development of structure–activity-relationship rules (SAR rules). Pharmacophore models can not only categorize structurally diverse compounds, but in their best, a good enrichment of active compounds can also be seen within one scaffold [130]. This is of course important for SAR-rule development.

Since pharmacophore models represent functionalities, they are good tools in

Pharmacophore-based bioactivity profiling

When a small molecule (a drug or an environmental chemical) enters a human body, it can interact with thousands of the over 10,000 proteins in the human body [132]. Because each of these interactions may lead to a biological response, it is crucial to examine compounds for potential (side) effects arising from these so-called off-target effects. To test all possible interacting targets in in vitro assays would be a time-consuming and cost-intensive task. Therefore, pharmacophore-based

Limits and future challenges

Pharmacophore modeling and pharmacophore based virtual screening are powerful methods in the search for new active compounds, in establishing SAR-rules, and in target-fishing. However, there are some limits and challenges that need to be overcome for the optimal performance.

The users should be aware that any virtual screening method including pharmacophore-based screening is currently not able to correctly find all active compounds for a target. As has been exemplified for acetylcholinesterase

Conclusions

In this review, we described the basic principles and algorithms of pharmacophore modeling starting from dataset generation and ending with widely used applications of pharmacophore models. Pharmacophore models are excellent tools for activity predictions, especially if they are constructed based on high quality data on the active and inactive training and test set molecules. Additionally, there are multiple software packages and tools for pharmacophore generation and application, many of them

Acknowledgements

A. Vuorinen thanks the Austrian Academy of Sciences (ÖAW) DOC-scholarship for financial support. We also thank Katalin Nadassy from Accelrys for kind help with the documentation. D. Schuster is grateful for her position in the Erika Cremer Habilitation Program by the University of Innsbruck and a Young Talents Grant provided by the University.

References (141)

  • M. Hein et al.

    Drug Discov. Today Technol.

    (2010)
  • D. Kirkland et al.

    Mutat. Res. Gene. Toxicol. Environ. Mutagene.

    (2007)
  • L. Di et al.

    Drug Discov. Today

    (2006)
  • Y. Hu et al.

    Drug Discov. Today

    (2014)
  • A.J. Williams et al.

    Drug Discov. Today

    (2012)
  • Daniel A. Keedy et al.

    Structure

    (2014)
  • M. Chen et al.

    Bioorg. Med. Chem. Lett.

    (2012)
  • S. Fox et al.

    J. Biomol. Screen.

    (2006)
  • C.H. Schwab

    Drug Discov. Today Technol.

    (2010)
  • H. Wang et al.

    J. Biol. Chem.

    (2006)
  • T. Seidel et al.

    Drug Discov. Today Technol.

    (2010)
  • G. Hessler et al.

    Drug Discov. Today Technol.

    (2010)
  • V. Schnecke et al.

    Drug Discovery Today

    (2006)
  • F.W. Lichtenthaler

    Angew. Chem. Int. Ed. Engl.

    (1995)
  • C.G. Wermuth et al.

    Pure Appl. Chem.

    (1998)
  • O.F. Güner et al.

    J. Chem. Inf. Model.

    (2014)
  • F.W. Schueler

    Chemobiodynamics Drug Design

    (1960)
  • P. Ehrlich

    Ber. Dtsch. Chem. Ges.

    (1909)
  • T.N. Doman et al.

    J. Med. Chem.

    (2002)
  • T. Klabunde et al.

    J. Med. Chem.

    (2009)
  • M.P.A. Sanders et al.

    J. Chem. Inf. Model.

    (2011)
  • A. Vuorinen et al.

    Mol. Inform.

    (2014)
  • D. Fourches et al.

    J. Chem. Inf. Model.

    (2010)
  • A. Cherkasov et al.

    J. Med. Chem.

    (2013)
  • J.B. Baell et al.

    J. Med. Chem.

    (2010)
  • B.Y. Feng et al.

    Nat. Chem. Biol.

    (2005)
  • D.J. Diller et al.

    J. Med. Chem.

    (2004)
  • B.A. Posner et al.

    J. Chem. Inf. Model.

    (2009)
  • A. Gaulton et al.

    Nucl. Acids Res.

    (2012)
  • E. Bolton et al.

    Annual Reports in Computational Chemistry

    (2008)
  • Y. Wang et al.

    Nucl. Acids Res.

    (2014)
  • P. Tiikkainen et al.

    J. Chem. Inf. Model.

    (2013)
  • H.M. Berman et al.

    Nucl. Acids Res.

    (2000)
  • T. Castrignanò et al.

    Nucl. Acids Res.

    (2006)
  • U. Pieper et al.

    Nucl. Acids Res.

    (2004)
  • R.A. Laskowski
  • M. Red Brewer et al.

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

    (2013)
  • A.R. Leach et al.

    J. Med. Chem.

    (2010)
  • R.J. Kavlock et al.

    Risk Anal.

    (2009)
  • N. Huang et al.

    J. Med. Chem.

    (2006)
  • A.M. Virshup et al.

    J. Am. Chem. Soc.

    (2013)
  • M.M. Mysinger et al.

    J. Med. Chem.

    (2012)
  • K. Heikamp et al.

    J. Chem. Inf. Model.

    (2013)
  • J. Sadowski et al.

    J. Chem. Inf. Comput. Sci.

    (1994)
  • J. Sadowski et al.

    Chem. Rev.

    (1993)
  • B.-J. Sung et al.

    Nature

    (2003)
  • P.C.D. Hawkins et al.

    J. Chem. Inf. Model.

    (2010)
  • B.R. Brooks et al.

    J. Comput. Chem.

    (1983)
  • T.A. Halgren

    J. Comput. Chem.

    (1996)
  • S. Lifson et al.

    J. Chem. Phys.

    (1968)
  • Cited by (0)

    View full text