Discovery and Early Clinical Development of LY3202626, a Low-Dose, CNS-Penetrant BACE Inhibitor.

The beta-site APP cleaving enzyme 1, known as BACE1, has been a widely pursued Alzheimer's disease drug target owing to its critical role in the production of amyloid-beta. We have previously reported the clinical development of LY2811376 and LY2886721. LY2811376 advanced to Phase I before development was terminated due to nonclinical retinal toxicity. LY2886721 advanced to Phase II, but development was halted due to abnormally elevated liver enzymes. Herein, we report the discovery and clinical development of LY3202626, a highly potent, CNS-penetrant, and low-dose BACE inhibitor, which successfully addressed these key development challenges.

Preparation and biological evaluation of BACE1 inhibitors: Leveraging trans-cyclopropyl moieties as ligand efficient conformational constraints.

Inhibition of BACE1 has become an important strategy in the quest for disease modifying agents to slow the progression of Alzheimer's disease. We previously reported the fragment-based discovery of LY2811376, the first BACE1 inhibitor reported to demonstrate robust reduction of human CSF Aß in a Phase I clinical trial. We also reported on the discovery of LY2886721, a potent BACE1 inhibitor that reached phase 2 clinical trials. Herein we describe the preparation and structure activity relationships (SAR) of a series of BACE1 inhibitors utilizing trans-cyclopropyl moieties as conformational constraints. The design, details of the stereochemically complex organic synthesis, and biological activity of these BACE1 inhibitors is described.

Idea2Data: Toward a New Paradigm for Drug Discovery.

Increasing the success rate and throughput of drug discovery will require efficiency improvements throughout the process that is currently used in the pharmaceutical community, including the crucial step of identifying hit compounds to act as drivers for subsequent optimization. Hit identification can be carried out through large compound collection screening and often involves the generation and testing of many hypotheses based on available knowledge. In practice, hypothesis generation can involve the selection of promising chemical structures from compound collections using predictive models built from previous screening/assay results. Available physical collections, typically used during hit identification, are of the order of 106 compounds but represent only a small fraction of the small molecule drug-like chemical space. In an effort to survey a larger portion of chemical space and eliminate inefficiencies during hit identification, we introduce a new process, termed Idea2Data (I2D) that tightly integrates computational and experimental components of the drug discovery process. I2D provides the ability to connect a vast virtual collection of compounds readily synthesizable on automated synthesis systems with computational predictive models for the identification of promising structures. This new paradigm enables researchers to process billions of virtual molecules and select structures that can be prepared on automated systems and made available for biological testing, allowing for timely hypothesis testing and follow-up. Since its introduction, I2D has positively impacted several portfolio efforts through identification of new chemical scaffolds and functionalization of existing scaffolds. In this Innovations paper, we describe the I2D process and present an application for the discovery of new ULK inhibitors.

Discovery of (1S,2R,3S,4S,5R,6R)-2-Amino-3-[(3,4-difluorophenyl)sulfanylmethyl]-4-hydroxy-bicyclo[3.1.0]hexane-2,6-dicarboxylic Acid Hydrochloride (LY3020371·HCl): A Potent, Metabotropic Glutamate 2/3 Receptor Antagonist with Antidepressant-Like Activity.

As part of our ongoing efforts to identify novel ligands for the metabotropic glutamate 2 and 3 (mGlu2/3) receptors, we have incorporated substitution at the C3 and C4 positions of the (1S,2R,5R,6R)-2-amino-bicyclo[3.1.0]hexane-2,6-dicarboxylic acid scaffold to generate mGlu2/3 antagonists. Exploration of this structure-activity relationship (SAR) led to the identification of (1S,2R,3S,4S,5R,6R)-2-amino-3-[(3,4-difluorophenyl)sulfanylmethyl]-4-hydroxy-bicyclo[3.1.0]hexane-2,6-dicarboxylic acid hydrochloride (LY3020371·HCl, 19f), a potent, selective, and maximally efficacious mGlu2/3 antagonist. Further characterization of compound 19f binding to the human metabotropic 2 glutamate (hmGlu2) site was established by cocrystallization of this molecule with the amino terminal domain (ATD) of the hmGlu2 receptor protein. The resulting cocrystal structure revealed the specific ligand-protein interactions, which likely explain the high affinity of 19f for this site and support its functional mGlu2 antagonist pharmacology. Further characterization of 19f in vivo demonstrated an antidepressant-like signature in the mouse forced-swim test (mFST) assay when brain levels of this compound exceeded the cellular mGlu2 IC50 value.

Preparation and biological evaluation of conformationally constrained BACE1 inhibitors.

The BACE1 enzyme is a key target for Alzheimer's disease. During our BACE1 research efforts, fragment screening revealed that bicyclic thiazine 3 had low millimolar activity against BACE1. Analysis of the co-crystal structure of 3 suggested that potency could be increased through extension toward the S3 pocket and through conformational constraint of the thiazine core. Pursuit of S3-binding groups produced low micromolar inhibitor 6, which informed the S3-design for constrained analogs 7 and 8, themselves prepared via independent, multi-step synthetic routes. Biological characterization of BACE inhibitors 6-8 is described.

Fragment-based design of kinase inhibitors: a practical guide.

Fragment-based drug design has become an important strategy for drug design and development over the last decade. It has been used with particular success in the development of kinase inhibitors, which are one of the most widely explored classes of drug targets today. The application of fragment-based methods to discovering and optimizing kinase inhibitors can be a complicated and daunting task; however, a general process has emerged that has been highly fruitful. Here a practical outline of the fragment process used in kinase inhibitor design and development is laid out with specific examples. A guide to the overall process from initial discovery through fragment screening, including the difficulties in detection, to the computational methods available for use in optimization of the discovered fragments is reported.

The potent BACE1 inhibitor LY2886721 elicits robust central Aß pharmacodynamic responses in mice, dogs, and humans.

BACE1 is a key protease controlling the formation of amyloid ß, a peptide hypothesized to play a significant role in the pathogenesis of Alzheimer's disease (AD). Therefore, the development of potent and selective inhibitors of BACE1 has been a focus of many drug discovery efforts in academia and industry. Herein, we report the nonclinical and early clinical development of LY2886721, a BACE1 active site inhibitor that reached phase 2 clinical trials in AD. LY2886721 has high selectivity against key off-target proteases, which efficiently translates in vitro activity into robust in vivo amyloid ß lowering in nonclinical animal models. Similar potent and persistent amyloid ß lowering was observed in plasma and lumbar CSF when single and multiple doses of LY2886721 were administered to healthy human subjects. Collectively, these data add support for BACE1 inhibition as an effective means of amyloid lowering and as an attractive target for potential disease modification therapy in AD.

Structure-guided expansion of kinase fragment libraries driven by support vector machine models.

This work outlines a new de novo design process for the creation of novel kinase inhibitor libraries. It relies on a profiling paradigm that generates a substantial amount of kinase inhibitor data from which highly predictive QSAR models can be constructed. In addition, a broad diversity of X-ray structure information is needed for binding mode prediction. This is important for scaffold and substituent site selection. Borrowing from FBDD, the process involves fragmentation of known actives, proposition of binding mode hypotheses for the fragments, and model-driven recombination using a pharmacophore derived from known kinase inhibitor structures. The support vector machine method, using Merck atom pair derived fingerprint descriptors, was used to build models from activity from 6 kinase assays. These models were qualified prospectively by selecting and testing compounds from the internal compound collection. Overall hit and enrichment rates of 82% and 2.5%, respectively, qualified the models for use in library design. Using the process, 7 novel libraries were designed, synthesized and tested against these same 6 kinases. The results showed excellent results, yielding a 92% hit rate for the 179 compounds that made up the 7 libraries. The results of one library designed to include known literature compounds, as well as an analysis of overall substituent frequency, are discussed.

Predicting the accuracy of ligand overlay methods with Random Forest models.

The accuracy of binding mode prediction using standard molecular overlay methods (ROCS, FlexS, Phase, and FieldCompare) is studied. Previous work has shown that simple decision tree modeling can be used to improve accuracy by selection of the best overlay template. This concept is extended to the use of Random Forest (RF) modeling for template and algorithm selection. An extensive data set of 815 ligand-bound X-ray structures representing 5 gene families was used for generating ca. 70,000 overlays using four programs. RF models, trained using standard measures of ligand and protein similarity and Lipinski-related descriptors, are used for automatically selecting the reference ligand and overlay method maximizing the probability of reproducing the overlay deduced from X-ray structures (i.e., using rmsd < or = 2 A as the criteria for success). RF model scores are highly predictive of overlay accuracy, and their use in template and method selection produces correct overlays in 57% of cases for 349 overlay ligands not used for training RF models. The inclusion in the models of protein sequence similarity enables the use of templates bound to related protein structures, yielding useful results even for proteins having no available X-ray structures.

Lessons in molecular recognition. 2. Assessing and improving cross-docking accuracy.

Docking methods are used to predict the manner in which a ligand binds to a protein receptor. Many studies have assessed the success rate of programs in self-docking tests, whereby a ligand is docked into the protein structure from which it was extracted. Cross-docking, or using a protein structure from a complex containing a different ligand, provides a more realistic assessment of a docking program's ability to reproduce X-ray results. In this work, cross-docking was performed with CDocker, Fred, and Rocs using multiple X-ray structures for eight proteins (two kinases, one nuclear hormone receptor, one serine protease, two metalloproteases, and two phosphodiesterases). While average cross-docking accuracy is not encouraging, it is shown that using the protein structure from the complex that contains the bound ligand most similar to the docked ligand increases docking accuracy for all methods ("similarity selection"). Identifying the most successful protein conformer ("best selection") and similarity selection substantially reduce the difference between self-docking and average cross-docking accuracy. We identify universal predictors of docking accuracy (i.e., showing consistent behavior across most protein-method combinations), and show that models for predicting docking accuracy built using these parameters can be used to select the most appropriate docking method.

Solid-phase synthesis of novel trimers containing a phenylstatine core and analysis by high-resolution magic angle spinning.

Here we describe a multistep solid-phase synthetic approach for the addition of amino acid residues to both the C- and N-termini of a phenylstatine core, yielding a library aimed at the development of structure-activity relationships in the S2 and S2' regions of the aspartyl proteases. Optimization of the synthetic strategy was performed on the basis of the in situ analysis of the compounds bound to the solid support through high-resolution magic angle spinning NMR Spectroscopy (HR-MAS NMR).

Biochemical and kinetic characterization of BACE1: investigation into the putative species-specificity for beta- and beta'-cleavage sites by human and murine BACE1.

Beta-amyloid peptides (Abeta) are produced by a sequential cleavage of amyloid precursor protein (APP) by beta- and gamma-secretases. The lack of Abeta production in beta-APP cleaving enzyme (BACE1)(-/-) mice suggests that BACE1 is the principal beta-secretase in mammalian neurons. Transfection of human APP and BACE1 into neurons derived from wild-type and BACE1(-/-) mice supports cleavage of APP at the canonical beta-secretase site. However, these studies also revealed an alternative BACE1 cleavage site in APP, designated as beta', resulting in Abeta peptides starting at Glu11. The apparent inability of human BACE1 to make this beta'-cleavage in murine APP, and vice versa, led to the hypothesis that this alternative cleavage was species-specific. In contrast, the results from human BACE1 transgenic mice demonstrated that the human BACE1 is able to cleave the endogenous murine APP at the beta'-cleavage site. To address this discrepancy, we designed fluorescent resonance energy transfer peptide substrates containing the beta- and beta'-cleavage sites within human and murine APP to compare: (i) the enzymatic efficiency; (ii) binding kinetics of a BACE1 active site inhibitor LY2039911; and (iii) the pharmacological profiles for human and murine recombinant BACE1. Both BACE1 orthologs were able to cleave APP at the beta- and beta'-sites, although with different efficiencies. Moreover, the inhibitory potency of LY2039911 toward recombinant human and native BACE1 from mouse or guinea pig was indistinguishable. In summary, we have demonstrated, for the first time, that recombinant BACE1 can recognize and cleave APP peptide substrates at the postulated beta'-cleavage site. It does not appear to be a significant species specificity to this cleavage.

Lessons in molecular recognition: the effects of ligand and protein flexibility on molecular docking accuracy.

The key to success for computational tools used in structure-based drug design is the ability to accurately place or "dock" a ligand in the binding pocket of the target of interest. In this report we examine the effect of several factors on docking accuracy, including ligand and protein flexibility. To examine ligand flexibility in an unbiased fashion, a test set of 41 ligand-protein cocomplex X-ray structures were assembled that represent a diversity of size, flexibility, and polarity with respect to the ligands. Four docking algorithms, DOCK, FlexX, GOLD, and CDOCKER, were applied to the test set, and the results were examined in terms of the ability to reproduce X-ray ligand positions within 2.0A heavy atom root-mean-square deviation. Overall, each method performed well (>50% accuracy) but for all methods it was found that docking accuracy decreased substantially for ligands with eight or more rotatable bonds. Only CDOCKER was able to accurately dock most of those ligands with eight or more rotatable bonds (71% accuracy rate). A second test set of structures was gathered to examine how protein flexibility influences docking accuracy. CDOCKER was applied to X-ray structures of trypsin, thrombin, and HIV-1-protease, using protein structures bound to several ligands and also the unbound (apo) form. Docking experiments of each ligand to one "average" structure and to the apo form were carried out, and the results were compared to docking each ligand back to its originating structure. The results show that docking accuracy falls off dramatically if one uses an average or apo structure. In fact, it is shown that the drop in docking accuracy mirrors the degree to which the protein moves upon ligand binding.

Design and synthesis of statine-containing BACE inhibitors.

Utilizing structure-based techniques and solid-phase synthesis, statine-based tetrapeptide BACE inhibitors were designed and synthesized using a heptapeptide BACE transition-state mimetic, 1, as the starting point. Structure-activity relationship studies at the P(3), P(2), and P(2)' positions as well as the N-terminal capping group on scaffold 5 led to the discovery of potent inhibitors 27, 32, and 34 (IC(50) <100 nM). In addition, computational analysis and the X-ray structure of BACE-inhibitor 38 are discussed.

Novel potent 5-HT(1F) receptor agonists: structure-activity studies of a series of substituted N-[3-(1-methyl-4-piperidinyl)-1H-pyrrolo[3,2-b]pyridin-5-yl]amides.

Compound 1a (LY334370), a selective 5-HT(1F) receptor agonist (SSOFRA), inhibited dural inflammation in the neurogenic plasma protein extravasation model of migraine and demonstrated clinical efficacy for the acute treatment of migraine. Although 1a was greater than 100-fold selective over both the 5-HT(1B) and 5-HT(1D) receptors, it exhibited appreciable 5-HT(1A) receptor affinity. Described here is the synthesis and evaluation of a series of pyrrolo[2,3-c]pyridine and pyrrolo[3,2-b]pyridine (2a and 3a) as well as pyrrolo[3,2-d]pyrimidine (4a) analogues of 1a, compounds prepared in an effort to identify SSOFRAs with improved selectivity over other 5-HT(1) receptor subtypes. The pyrrolo[3,2-b]pyridine analogue 3a showed high 5-HT(1F) receptor affinity but offered no improvement in selectivity compared to 1a. However, the C-5 acetamide derivative, 3b, was greater than 100-fold selective over the 5-HT(1A), 5-HT(1B), and 5-HT(1D) receptors. SAR studies of this series determined that alkylamides in particular exhibited high selectivity for the 5-HT(1F) receptor. Replacement at C-5 with other substituents decreased affinity or selectivity. These SAR studies identified SSOFRAs that demonstrated oral activity in the neurogenic plasma protein extravasation model, a model indicative of antimigraine activity.

A pharmacophore for human pregnane X receptor ligands.

The pregnane X receptor (PXR) is involved in transcriptional regulation of multiple cytochromes P450 and multidrug resistance-associated protein (MDR1), which encodes for the drug transporter P-glycoprotein. Crystal structure analyses suggest that the ligand binding domain is highly hydrophobic and flexible, allowing molecules of differing sizes to bind in multiple orientations. Using literature data for EC(50) (half-maximal inhibitory concentration) values for PXR activation derived for 12 human PXR ligands, a pharmacophore was developed. This pharmacophore supports the hydrophobic nature of the ligand binding domain recently deduced from the X-ray crystal structure because it contains four hydrophobic regions and one hydrogen bond acceptor. These features are consistent with at least one of the three experimentally determined orientations in which SR12813 binds to PXR, as determined by overlay studies. SR12813 fulfills all of the five pharmacophore features, as does the potent ligand hyperforin. The pharmacophore was also used to predict the binding affinity for 28 molecules not in the model but known to be PXR ligands of differing potencies. The pharmacophore distinguished the most potent activators of PXR (that display >5-fold activation/deactivation), like ecteinascidin, troglitazone, nifedipine, and dexamethasone-t-butylacetate, from poor activators, such as scopoletin and kaempferol. The model could be useful in drug development, potentially acting as a high-throughput filter for identifying compounds that may bind to PXR before in vitro determination. Ultimately, this will aid in the selection of molecules with a lesser capacity to be potent PXR ligands and thus avoid induction of numerous drug-metabolizing enzymes and MDR1.