The power of structural information can be harnessed when greater chemical diversity is used in structure-based drug design.
For more than a decade, the use of protein structural information in lead validation, lead generation, and lead optimization has made an increasing impact in the discovery of new drugs reaching the market. Protein structural information is used in different ways to evaluate and, in some cases, identify the chemical starting points (leads) for drug discovery programs. This provides significant benefit.
The concept behind this article is that the value of structure-based drug design is maximized when there is an emphasis on chemical diversity for structural analysis. In reviewing how structural information is used to bring this chemical diversity into structure-based drug design programs, the different sources of chemical leads, fragment-based and high throughput screening, literature leads or “lead hopping”, and virtual ligand screening are discussed along with opportunities and limitations of each.
There has been growing interest in fragment leads in structure-based drug design over the past 10 years, with many companies focusing significant resources on fragment screening technologies.1 The use of protein NMR and X-ray crystallography to screen for fragment leads or to advance fragment leads identified in enzymatic- or affinity-based fragment screening is maturing and demonstrating value for drug discovery.2 Companies such as Abbott, Astex Pharmaceuticals, SGX Pharmaceuticals, and Astra-Zeneca have made fragment-based screening an important, if not critical part of their lead generation paradigm powered by either protein crystal structures or NMR structures for advancing those leads. The goal is to screen for small, relatively weak, but more efficient binders than a typical high throughput screen is designed to find. Fragment leads have been identified using a variety of screening methodologies, including: enzymatic assays at high compound concentrations with strict molecular weight cutoffs (generally under 350 Da); structural technologies such as protein NMR (SAR by NMR) or X-ray crystallography (CrystaLEAD); and, more recently, new technologies such as surface plasmon resonance (SPR).
No matter which fragment screening technology is employed, it makes intuitive sense and is well established that advancing fragment leads into suitable drug candidates typically requires detailed protein/inhibitor structural information (Figure 1). This is generally because fragment leads are typically small, weak, unelaborated compounds; therefore, binding modes and structure-activity relationships (SAR) can be very difficult to establish using traditional medicinal chemistry or molecular modeling. In addition, fragments are often relegated to the bottom of the priority list if far more potent HTS or literature leads are available to chemists. If structural information is available, however, the raw binding efficiency of fragment leads make them an ideal starting point for developing novel, potent, and efficient drug-like molecules.
There are, however, some significant challenges in obtaining structural information on fragment leads. Solubility issues and;weak activity of the fragment leads can be problematic in obtaining structures. This can be further hampered if the fragment adopts multiple binding modes. When crystallography is unable to obtain a fragment structure, protein NMR structures can be important for establishing a binding mode for initial optimization. Interestingly, each screening technology introduces a bias toward certain compound physical properties. Since enzymatic screening requires fairly high concentrations of compound, libraries tend to be biased toward more soluble fragments. Protein NMR-based screens that provide greater flexibility in compound physical properties are less sensitive to compound solubility and, therefore, have a tendency toward compounds with higher cLogP’s, unless close attention is paid to physical properties in library design. Surface plasmon resonance is in the early days of its application to fragment screening, but it is clear that chemical leads must be tested and filtered for non-specific interaction with the chip surface to avoid high false positive rates.
A very effective approach to fragment screening is “structure” or “pharmacaphore-directed” screening—that is, designing a library that exploits a known active site preference for binding functionalities such as acids, amines, aromatics, or non-aromatic ring systems, for example. A classic example of this was an enzymatic, fragment-based screen done at Abbott on the influenza target, neuraminidase.3 In this case, fragments were chosen to include an acid, acetyl, and a positive charge—functionalities required in all substrates and inhibitors known at the time—and a library of ß-amino acids was screened. From this screening exercise, a novel pyrrolidine-based molecule was identified as a new lead starting point. Crystal structures were ultimately required for advancing this lead into a drug candidate, but the activity screen was very effective at identifying a novel starting point for the chemistry program.
Another efficient approach to identifying and advancing fragment leads is through lead deconstruction by taking any compound of >350 Da molecular weight, typically an HTS lead or literature compound, systematically stripping it down to its essential binding elements (core fragment), and binding mode established in a crystal structure. This enables the chemist to understand the value of the individual components that made up the original lead and allows him to use a structure-based approach to re-design the core to facilitate re-optimization chemistry or to alter vectors for subsequent structure-based optimization. This approach is an extremely powerful and rapid way to learn about the properties of the binding site, and facilitate optimization by initiating chemistry at the most efficient starting point and only restoring molecular weight that provides real value.
The power of this approach is exemplified by the structure-based drug design work on DPP-4 (Figure 2). In this program, a 1µM HTS lead was systematically stripped down to a 25µM core and its crystal structure was determined.4 This enabled understanding of the detailed interactions with the protein and the relative value of those interactions. The core fragment was re-designed to provide an improved synthetic starting point for rapid structure-driven design, facilitating the progress of the project from a HTS hit to a viable drug candidate in less than one year.
High throughput screening
Crystallographic or protein NMR structural information are very powerful tools for HTS hit triage and lead validation, as well as the hit-to-lead and lead optimization processes. Although crystallography is often used to establish a binding mode for new HTS leads, at Abbott we have found it particularly efficient to use protein NMR to vet HTS screening leads for legitimate binders and to remove reactive compounds from the hit set via ALARM NMR.5 Removing false-positive HTS hits allows crystallographic resources to focus on the validated lead compounds for structure determination and improves efficiency.
When a binding mode of an HTS hit is determined in a crystal structure, the essential interactions, unproductive interactions, and available binding site space can be more accurately evaluated and significantly jump-start the hit-to-lead process. Furthermore, the value of structurally comparing multiple chemical series to identify common features and translating substituents (SAR) from one series to another is an excellent way to rapidly advance hit-to-lead chemistry. The impact of structure at the hit-to-lead process is so powerful that many companies focus the vast majority of their structural resources just on this step and do not expend significant, if any, resources on lead optimization.
Literature-based leads and “lead hopping”
When possible, it is very efficient to learn from the literature, especially, but not exclusively, for fast follower programs. Structurally and biochemically characterizing leads from literature sources can significantly jump-start SAR development for a new program. More information tends to be available for literature compounds, and often SAR development has already been invested into these compounds; therefore, this provides and excellent opportunity for learning. Crystal structures of literature compounds can enable chemists to see common features between HTS leads and literature leads7. Fragments can often be used in structure-based core replacement to rapidly translate a literature lead into a potent and proprietary lead for a program. This structural information can greatly facilitate both structure-based or cheminformatic, ligand-based “lead hopping.”7
Virtual ligand screening
Virtual ligand screening is a developing technology that makes use of a protein target structure. To find new leads, a crystal structure or NMR structure is typically needed, while a homology model can also be used with diminishing information content.8 This approach employs a variety of computational tools to virtually screen either existing compound libraries or even virtual libraries of compounds to identify hit lists that can then be tested in a biochemical assay. Although experimental approaches to ligand screening tend to be more robust for identifying chemical leads, virtual ligand screening can be inexpensive and efficient for the enrichment of leads when experimental technologies are either not available or are resource-limited. For smaller companies that do not have the resources for high throughput screening or have small compound libraries, this is an enabling technology. Even large Pharma uses this approach to jump-start a program quickly, especially where reagents and assays may be limiting or time-consuming. However, success of this approach is limited and if experimental screening approaches are available, the experimental approach is the most reliable way to find diverse sets of tractable leads.
Combination of multiple lead sources
Abbott has utilized structural information by X-ray crystallography or NMR to advance leads from all of these sources. Figure 4 shows the distribution of 271 lead compounds whose structures were determined by crystallography across 28 projects. These 271 leads are the first representatives of a new chemical series at Abbott:
- 13% are HTS hits;
- 31% are fragment-based;
- 14% are of novel, fragment-based kinase cores;
- 24% are structure-based, modified leads or “lead hops” that represent a new chemical series;
- 18% are literature-derived leads.
Furthermore, if we look at a measure of pair-wise similarity of compounds in the internal Abbott protein structure database, organized by target identity, we find a mean value of 0.18 for ECFP-6,9 which corresponds to a probability of 2% for any pair of compounds to be isoactive.7 In contrast, lead chemical series have an average ECFP-6 pair-wise similarity of 0.37, which corresponds to a probability of at least 20% for any compound pair to be isoactive.7 This illustrates the emphasis that is placed on chemical diversity when choosing compounds for our structure-based drug design programs.
Some of the most elegant examples of structure-based drug design at Abbott have resulted from combining structure-based fragment leads, HTS leads, and literature-derived information within the same chemical program, to produce a novel chemical series and rapidly progressing chemical optimization to advance high quality drug candidates. In a number of cases the hit-to-drug candidate process has been achieved within one year or less from obtaining initial lead structural information as exemplified by our work on DPP-4.4 The true power of structural information is harnessed when greater chemical diversity is used in structure-based drug design.
1. Everts, Sarah. “More and more companies are using fragment-based lead design as a drug discovery strategy”. C&E News. 2008;86,29:15-23.
2. Hajduk, Philip, J., Greer, Jonathan. “A decade of fragment-based drug design: strategic advances and lessons learned”. Nature Reviews/Drug Discovery. 2007;6:211-219.
3. Kati, Warren, M. et al. “Novel ?- and ß-Amino Acid Inhibitors of Influenza Virus Neuraminidase”. Antimicrobial Agents and Chemotherapy. 2001;45, 9;2563-2570.
4. Backes, Bradley J., et al. “Pyrrolidine-constrained phenethylamines: The design of potent selective and pharmacologically efficacious dipeptidyl peptidase IV (DPP4) inhibtors from a lead-like screening hit”. Bioorg. and Med. Chem. Lett.. 2007;17:2005-2012.
5. Huth R. J., et al. “Alarm NMR: A Rapid and Robust Method to Detect Reactive False Positives in Biochemical Screens”. J. Am. Chem Soc. 2005;127,217.
6. Stamper, Geoffrey, F., et al, “Structure-based Optimization of MurF Inhibitors”. Chem. Biol. Drug Des. 2006;58-65,67.
7. Muchmore, Steven, W., et al. “Application of Belief Theory to Similarity Data Fusion for Use in Analog Searching and Lead Hopping”. J. Chem. Inf. Model. 2008;48:941-948.
8. Klebe, Gehard, “Virtual Ligand Screening: Strategies, Perspectives and Limitations”. Drug Discovery Today. 2006;11,no.13/14:580-594.
9. Hassan, M., Brown, R.D., Varma -O’Brien, S., Rogers, D. Molecular Diversity. 2006;10:283.
About the Author
Vincent Stoll has been working in structure-based drug design at Abbott for over 11 years and is currently the project leader of Protein X-ray Crystallography. He is a co-inventor on over seven issued patents for drug molecules and has over 32 publications on structure-based drug design projects at Abbott.
This article was published in Drug Discovery & Development magazine: Vol. 11, No. 12, December, 2008, pp. 16-20.
Filed Under: Drug Discovery