Peter W. Kenny

An Integrated Approach to Fragment-Based Lead Generation:Philosophy, Strategy and Case Studies from AstraZenecas Drug Discovery Programmes

Fragment-based lead generation (FBLG) has recently emerged as an alternative to traditional high throughput screening (HTS) to identify initial chemistry starting points for drug discovery programs. In comparison to HTS screening libraries, the screening sets for FBLG tend to contain orders of magnitude fewer compounds, and the compounds themselves are less structurally complex and have lower molecular weight. This report summarises the advent of FBLG within the industry and then describes the FBLG experience at AstraZeneca. We discuss (1) optimising the design of screening libraries, (2) hit detection methodologies, (3) evaluation of hit quality and use of ligand efficiency calculations, and (4) approaches to evolve fragment-based, low complexity hits towards drug-like leads. Furthermore, we exemplify our use of FBLG with case studies in the following drug discovery areas: antibacterial enzyme targets, GPCRs (melanocortin 4 receptor modulators), prostaglandin D2 synthase inhibitors, phosphatase inhibitors (protein tyrosine phosphotase 1B), and protease inhibitors (b-secretase).

(1R,2R)-N-(1-cyanocyclopropyl)-2-(6-methoxy-1,3,4,5-tetrahydropyrido[4,3-b]indole-2-carbonyl)cyclohexanecarboxamide (AZD4996): a potent and highly selective cathepsin K inhibitor for the treatment of osteoarthritis.

Directed screening of nitrile compounds revealed 3 as a highly potent cathepsin K inhibitor but with cathepsin S activity and very poor stability to microsomes. Synthesis of compounds with reduced molecular complexity, such as 7, revealed key SAR and demonstrated that baseline physical properties and in vitro stability were in fact excellent for this series. The tricycle carboline P3 unit was discovered by hypothesis-based design using existing structural information. Optimization using small substituents, knowledge from matched molecular pairs, and control of lipophilicity yielded compounds very close to the desired profile, of which 34 (AZD4996) was selected on the basis of pharmacokinetic profile.

Dipeptidyl nitrile inhibitors of Cathepsin L.

A series of potent Cathepsin L inhibitors with good selectivity with respect to other cysteine Cathepsins is described and SAR is discussed with reference to the crystal structure of a protein-ligand complex.

Matched molecular pair analysis of activity and properties of glycogen phosphorylase inhibitors

Matched molecular pair analysis has been used in design of inhibitors of glycogen phosphorylase.

Discovery of novel imidazo[1,2-a]pyridines as inhibitors of the insulin-like growth factor-1 receptor tyrosine kinase

We disclose a novel series of insulin-like growth factor-1 receptor kinase inhibitors based on the 3-(pyrimidin-4-yl)-imidazo[1,2-a]pyridine scaffold. The influence on the inhibitory activity of substitution on the imidazopyridine and at the C5 position of the pyrimidine is discussed. In the course of this optimization, we discovered a potent and selective inhibitor with suitable pharmacokinetics for oral administration.

High throughput solubility determination with application to selection of compounds for fragment screening.

The development and application of a high throughput aqueous solubility assay is reported. Measurements for up to 637 compounds can be made in a fully automated experiment. Results from this assay were used to quantify risk of unacceptable solubility as a function of lipophilicity for neutral fragment-like compounds. Assessment of risk of unacceptable solubility was combined with experimental solubility measurement to select compounds for inclusion in a fragment-screening library.

Design of selective Cathepsin inhibitors

A number of molecular recognition features have been exploited in structure-based design of selective Cathepsin inhibitors.

5-Aminopyrimidin-2-ylnitriles as cathepsin K inhibitors.

A series of pyrimidine nitrile inhibitors of Cathepsin K with reduced glutathione reactivity has been identified and Molecular Core Matching (MoCoM) has been used to quantify the effect of an amino substituent at C5.

Pharmacokinetic benefits of 3,4-dimethoxy substitution of a phenyl ring and design of isosteres yielding orally available cathepsin K inhibitors.

Rational structure-based design has yielded highly potent inhibitors of cathepsin K (Cat K) with excellent physical properties, selectivity profiles, and pharmacokinetics. Compounds with a 3,4-(CH₃O)₂Ph motif, such as 31, were found to have excellent metabolic stability and absorption profiles. Through metabolite identification studies, a reactive metabolite risk was identified with this motif. Subsequent structure-based design of isoteres culminated in the discovery of an optimized and balanced inhibitor (indazole, 38).

Computation, experiment and molecular design.

Perhaps the best way to start a Perspective like this is to say what we mean by ‘Molecular Design’ and one definition is control of properties and behaviour of compounds and materials through manipulation of molecular properties [1]. Molecular interactions [2] occupy a special place in the framework of Molecular Design because the functional behaviour of a compound is determined by the strengths of the interactions of its molecules with the different environments (e.g. solution; crystal lattice; bound to protein) in which they exist [1]. Molecular Design can be hypothesisdriven or prediction-driven and the objectives of Computer-Aided Molecular Design (CAMD) are to provide physicochemical insight for framing design hypotheses and to build predictive models with which to evaluate compounds. Energy, geometry and connectivity form the basis of the tools used in CAMD. For example, a method for predicting crystal structure might use geometry to pack molecules into prospective unit cells and energy to rank the resulting crystal lattices. The molecular connection table is the defining cheminformatic data structure and we can use connectivity to infer physicochemical characteristics such as interaction potential (e.g. strong hydrogen bond acceptor) and likelihood of being ionised in solution. From the CAMD perspective, Drug Discovery is a process in two steps. Lead Identification is a search for active compounds that can be optimised in the second step to compounds with properties and behaviour consistent with dosing in humans as therapeutic agents. Although the second step is frequently described as a process of multiobjective optimisation, it can be argued that lead optimisation actually has minimisation of therapeutic dose as its principal, if not sole, objective. The last 25 years have seen adoption of a number of technologies by pharmaceutical and biotechnology industries and much CAMD activity is a reaction to these developments. Typically, each new technology is introduced with the promise that it will revolutionise Drug Discovery and much of the hyperbole spills over into the CAMD arena. Some of the difficulty that CAMD scientists face in gaining acceptance for their approaches can be traced to extravagant claims made earlier by other CAMD scientists. Assessing the value added by a new technology is not always easy. When a pharmaceutical company spends a large amount of money to acquire new capability, it is in the interests of both vendor and customer that purchases and collaborations are seen in the most favourable light. Over-selling of technologies leads to panacea-centric thinking, which is especially dangerous in CAMD because success frequently depends on bringing together diverse computational tools to both define and solve problems. One important lesson from the last 25 years of Drug Discovery is that technology is a good servant but a poor master. The software tools of CAMD are in many cases interconnected and the CAMD scientist must recognise these connections in order to exploit them. I will attempt to illustrate this point with reference to computational methods for identifying compounds with interesting biological activity, which has become known as virtual screening. Pharmacophore searching, molecular shape-matching and docking are essentially geometry-based approaches to virtual screening. The first two of these are ligand-based and complement each other in virtual screening campaigns. Despite algorithmic differences, pharmacophore searching and shape-matching are actually more similar than first appearances might suggest in that each method aligns hit molecules with the respective query. We can rank P. W. Kenny (&) Macclesfield, UK e-mail: pwk.pub.2008@gmail.com