Machine Learning Models to Predict Inhibition of the Bile Salt Export Pump.

Cholestatic liver injury is frequently associated with drug inhibition of bile salt transporters, such as the bile salt export pump (BSEP). Reliable in silico models to predict BSEP inhibition directly from chemical structures would significantly reduce costs during drug discovery and could help avoid injury to patients. We report our development of classification and regression models for BSEP inhibition with substantially improved performance over previously published models. We assessed the performance effects of different methods of chemical featurization, data set partitioning, and class labeling and identified the methods producing models that generalized best to novel chemical entities.

Rethinking drug design in the artificial intelligence era.

Artificial intelligence (AI) tools are increasingly being applied in drug discovery. While some protagonists point to vast opportunities potentially offered by such tools, others remain sceptical, waiting for a clear impact to be shown in drug discovery projects. The reality is probably somewhere in-between these extremes, yet it is clear that AI is providing new challenges not only for the scientists involved but also for the biopharma industry and its established processes for discovering and developing new medicines. This article presents the views of a diverse group of international experts on the 'grand challenges' in small-molecule drug discovery with AI and the approaches to address them.

Development of MK-8353, an orally administered ERK1/2 inhibitor, in patients with advanced solid tumors.

BACKGROUND: Constitutive activation of ERK1/2 occurs in various cancers, and its reactivation is a well-described resistance mechanism to MAPK inhibitors. ERK inhibitors may overcome the limitations of MAPK inhibitor blockade. The dual mechanism inhibitor SCH772984 has shown promising preclinical activity across various BRAFV600/RAS-mutant cancer cell lines and human cancer xenografts. METHODS: We have developed an orally bioavailable ERK inhibitor, MK-8353; conducted preclinical studies to demonstrate activity, pharmacodynamic endpoints, dosing, and schedule; completed a study in healthy volunteers (P07652); and subsequently performed a phase I clinical trial in patients with advanced solid tumors (MK-8353-001). In the P07652 study, MK-8353 was administered as a single dose in 10- to 400-mg dose cohorts, whereas in the MK-8353-001 study, MK-8353 was administered in 100- to 800-mg dose cohorts orally twice daily. Safety, tolerability, pharmacokinetics, pharmacodynamics, and antitumor activity were analyzed. RESULTS: MK-8353 exhibited comparable potency with SCH772984 across various preclinical cancer models. Forty-eight patients were enrolled in the P07652 study, and twenty-six patients were enrolled in the MK-8353-001 study. Adverse events included diarrhea (44%), fatigue (40%), nausea (32%), and rash (28%). Dose-limiting toxicity was observed in the 400-mg and 800-mg dose cohorts. Sufficient exposure to MK-8353 was noted that correlated with biological activity in preclinical data. Three of fifteen patients evaluable for treatment response in the MK-8353-001 study had partial response, all with BRAFV600-mutant melanomas. CONCLUSION: MK-8353 was well tolerated up to 400 mg twice daily and exhibited antitumor activity in patients with BRAFV600-mutant melanoma. However, antitumor activity was not particularly correlated with pharmacodynamic parameters. TRIAL REGISTRATION: ClinicalTrials.gov NCT01358331. FUNDING: Merck Sharp & Dohme Corp., a subsidiary of Merck & Co. Inc., and NIH (P01 CA168585 and R35 CA197633).

Efficient identification of novel leads by dynamic focused screening: PDK1 case stud.

A dynamic, focused screening strategy that utilized a limited but diversified set of target-specific compounds was explored as an efficient means for the identification of inhibitors of the protein kinase PDK1. Approximately 21,500 compounds, including a 19,000 molecule kinase-focused compound collection (KFCC), were screened at two concentrations to identify initial leads. The KFCC included several empirically-derived, general kinase libraries and molecules chosen by PDK1-specific virtual screens. As was expected, this initial screen mostly identified potent leads with limited novelty. In order to overcome this limitation, the data from the screen were used to drive several rounds of a customized iterative focused screening (IFS) campaign. A machine-learning technique was used to build a predictive model to identify compounds to be screened in subsequent rounds. Molecules deemed not to be novel were removed from the training set for the next round, which allowed this campaign to progressively walk away from the chemical space covered by the KFCC. This resulted in the identification of PDK1 inhibitors which are uniquely different from publicly known chemotypes after just three rounds of screenings. A retrospective analysis of this IFS approach against an ultra-high throughput screen (uHTS) indicated that while uHTS is still the most prolific paradigm for lead identification, this dynamic, focused screening approach was successful in discovering novel scaffolds for a medicinal chemistry effort. Finally, a theoretical optimization suggested the dynamic, focused screening approaches could provide either a complementary or alternative approach to uHTS for the efficient and rapid lead identification.

3,4-Disubstituted benzofuran P1' MMP-13 inhibitors: optimization of selectivity and reduction of protein binding.

Potent 3,4-disubstituted benzofuran P1' MMP-13 inhibitors have been prepared. Selectivity over MMP-2 was achieved through a substituent at the C4 position of the benzofuran P1' moiety of the molecule. By replacing a backbone benzene with a pyridine and valine with threonine, compounds (e.g., 44) with greatly reduced plasma protein binding were also obtained.

Molecular mechanism of action of pharmacoperone rescue of misrouted GPCR mutants: the GnRH receptor.

The human GnRH receptor (hGnRHR), a G protein-coupled receptor, is a useful model for studying pharmacological chaperones (pharmacoperones), drugs that rescue misfolded and misrouted protein mutants and restore them to function. This technique forms the basis of a therapeutic approach of rescuing mutants associated with human disease and restoring them to function. The present study relies on computational modeling, followed by site-directed mutagenesis, assessment of ligand binding, effector activation, and confocal microscopy. Our results show that two different chemical classes of pharmacoperones act to stabilize hGnRHR mutants by bridging residues D(98) and K(121). This ligand-mediated bridge serves as a surrogate for a naturally occurring and highly conserved salt bridge (E(90)-K(121)) that stabilizes the relation between transmembranes 2 and 3, which is required for passage of the receptor through the cellular quality control system and to the plasma membrane. Our model was used to reveal important pharmacophoric features, and then identify a novel chemical ligand, which was able to rescue a D(98) mutant of the hGnRHR that could not be rescued as effectively by previously known pharmacoperones.

Discovery of novel inhibitors of the ZipA/FtsZ complex by NMR fragment screening coupled with structure-based design.

ZipA is a membrane anchored protein in Escherichia coli that interacts with FtsZ, a homolog of eukaryotic tubulins, forming a septal ring structure that mediates bacterial cell division. Thus, the ZipA/FtsZ protein-protein interaction is a potential target for an antibacterial agent. We report here an NMR-based fragment screening approach which identified several hits that bind to the C-terminal region of ZipA. The screen was performed by 1H-15N HSQC experiments on a library of 825 fragments that are small, lead-like, and highly soluble. Seven hits were identified, and the binding mode of the best one was revealed in the X-ray crystal structure. Similar to the ZipA/FtsZ contacts, the driving force in the binding of the small molecule ligands to ZipA is achieved through hydrophobic interactions. Analogs of this hit were also evaluated by NMR and X-ray crystal structures of these analogs with ZipA were obtained, providing structural information to help guide the medicinal chemistry efforts.

Functional, biophysical, and structural bases for antibacterial activity of tigecycline.

Tigecycline is a novel glycylcycline antibiotic that possesses broad-spectrum activity against many clinically relevant species of bacterial pathogens. The mechanism of action of tigecycline was delineated using functional, biophysical, and molecular modeling experiments in this study. Functional assays showed that tigecycline specifically inhibits bacterial protein synthesis with potency 3- and 20-fold greater than that of minocycline and tetracycline, respectively. Biophysical analyses demonstrated that isolated ribosomes bind tigecycline, minocycline, and tetracycline with dissociation constant values of 10(-8), 10(-7), and >10(-6) M, respectively. A molecular model of tigecycline bound to the ribosome was generated with the aid of a 3.40-angstrom resolution X-ray diffraction structure of the 30S ribosomal subunit from Thermus thermophilus. This model places tigecycline in the A site of the 30S subunit and involves substantial interactions with residues of H34 of the ribosomal subunit. These interactions were not observed in a model of tetracycline binding. Modeling data were consistent with the biochemical and biophysical data generated in this and other recent studies and suggested that tigecycline binds to bacterial ribosomes in a novel way that allows it to overcome tetracycline resistance due to ribosomal protection.

Synthesis and biological evaluation of biphenylsulfonamide carboxylate aggrecanase-1 inhibitors.

Aggrecanases are recently discovered enzymes that cleave aggrecan, a key component of cartilage. Aggrecanase inhibitors may provide a unique means to halt the progression of cartilage destruction in osteoarthritis. The synthesis and evaluation of biphenylsulfonamidocarboxylic acid inhibitors of aggrecanase-1 are reported. Compound 24 demonstrated 89% inhibition of proteoglycan degradation at 10 microg/mL and has an oral bioavailability in rat of 35%.

Structure-based identification of the binding site for the hemiasterlin analogue HTI-286 on tubulin.

A binding mode of HTI-286, a synthetic analogue of the peptidic antimitotic agent hemiasterlin, to tubulin is proposed. The binding mode was derived from iterative docking experiments directed at regions of the tubulin interdimer interface that are believed to be consistent with all current experimental data regarding the HTI-286/tubulin interaction. These data include (1) competitive inhibition of the tubulin binding of the Vinca alkaloids and other antimitotic agents, (2) proximity to stretches of amino acid residues identified in two separate photoaffinity-labeling experiments, (3) structure-activity relationships for HTI-286 and its analogues, (4) saturation transfer difference nuclear magnetic resonance (NMR) experiments, and (5) NMR transfer nuclear Overhauser effect spectroscopy (NOESY) experiments that potentially identify the bioactive conformation. The predicted binding mode thus affords a means to understand the mode of action of hemiasterlin, HTI-286, and other closely related molecules.

Potent, selective, and orally bioavailable matrix metalloproteinase-13 inhibitors for the treatment of osteoarthritis.

Modification of alpha-biphenylsulfonamidocarboxylic acids led to potent and selective MMP-13 inhibitors. Compound 16 showed 100% oral bioavailability in rats and demonstrated >50% inhibition of bovine cartilage degradation at 10 ng/mL.

Synthesis and SAR of highly selective MMP-13 inhibitors.

The structure-based design and synthesis of a series of novel biphenyl sulfonamide carboxylic acids as potent MMP-13 inhibitors with selectivity over MMP-1, MMP-2, MMP-3, MMP-7, MMP-8, MMP-9, MMP-14, Aggrecanase 1, and TACE are described.

Identification of potent and selective MMP-13 inhibitors.

A potent, selective series of MMP-13 inhibitors has been derived from a weak (3.2 microM) inhibitor that did not bear a zinc chelator. Structure-based drug design strategies were employed to append a Zn-chelating group to one end of the molecule and functionality to enhance selectivity to the other. A compound from this series demonstrated rat oral bioavailability and efficacy in a bovine articular cartilage explant model.

A shape-based 3-D scaffold hopping method and its application to a bacterial protein-protein interaction.

In this paper, we describe the first prospective application of the shape-comparison program ROCS (Rapid Overlay of Chemical Structures) to find new scaffolds for small molecule inhibitors of the ZipA-FtsZ protein-protein interaction, a proposed antibacterial target. The shape comparisons are made relative to the crystallographically determined, bioactive conformation of a high-throughput screening (HTS) hit. The use of ROCS led to the identification of a set of novel, weakly binding inhibitors with scaffolds presenting synthetic opportunities to further optimize biological affinity and lacking development issues associated with the HTS lead. These ROCS-identified scaffolds would have been missed using other structural similarity approaches such as ISIS 2D fingerprints. X-ray crystallographic analysis of one of the new inhibitors bound to ZipA reveals that the shape comparison approach very accurately predicted the binding mode. These experimental results validate this use of ROCS for chemotype switching or "lead hopping" and suggest that it is of general interest for lead identification in drug discovery endeavors.

Combinatorial synthesis of substituted 3-(2-indolyl)piperidines and 2-phenyl indoles as inhibitors of ZipA-FtsZ interaction.

The ZipA-FtsZ protein-protein interaction is a potential target for antibacterial therapy. The design and parallel synthesis of a combinatorial library of small molecules, which target the FtsZ binding area on ZipA are described. Compounds were demonstrated to bind to the FtsZ binding domain of ZipA by HSQC NMR and to inhibit cell division in a cell elongation assay.

The application of x-ray, NMR, and molecular modeling in the design of MMP inhibitors.

The following review discusses the successful application of X-ray, NMR, and molecular modeling in the design of potent and selective inhibitors of matrix metalloproteinases (MMPs) and TNFalpha-converting enzyme (TACE) from Wyeth. The importance of protein and ligand mobility as it impacts structure-based design is also discussed. The MMPs are an active target for a variety of diseases, including cancer and arthritis.

Design and synthesis of indolo[2,3-a]quinolizin-7-one inhibitors of the ZipA-FtsZ interaction.

The binding of FtsZ to ZipA is a potential target for antibacterial therapy. Based on a small molecule inhibitor of the ZipA-FtsZ interaction, a parallel synthesis of small molecules was initiated which targeted a key region of ZipA involved in FtsZ binding. The X-ray crystal structure of one of these molecules complexed with ZipA was solved. The structure revealed an unexpected binding mode, facilitated by desolvation of a loosely bound surface water.

Structure-based design of carboxybiphenylindole inhibitors of the ZipA-FtsZ interaction.

Structural features of two weak inhibitors of the ZipA-FtsZ protein-protein interaction which were found to bind to overlapping but different areas of the key binding site were combined in one new series of carboxybiphenyl-indoles with improved inhibitory activity.

IR spectra of cytochrome c denatured with deuterated guanidine hydrochloride show increase in beta sheet.

Attenuated total reflectance Fourier transform IR (ATR-FTIR) spectra are obtained for horse heart ferricytochrome c in solutions of 0-7M guanidine hydrochloride and deuterated guanidine hydrochloride. Substitutions of deuterium for hydrogen in both the denaturant and protein provide resolvable amide I spectra over a wide range of denaturant concentrations. Deuteration enhances the ability to measure the true protein IR spectrum in the amide I region in which the secondary structure can be deduced, because spectra in D(2)O are less prone to spectral distortion upon background denaturant subtraction than spectra in H(2)O. Other investigators studying equilibrium unfolded cytochrome c were limited to guanidine concentrations below 3.0M because of detector saturation. Detector saturation is avoided with the use of ATR-FTIR spectroscopy, allowing one to obtain protein spectra at high denaturant concentrations. Second derivative spectra of samples show reductions in alpha helix and increases in beta sheet at high denaturant concentrations, contrary to expectations of finding primarily a random coil secondary structure. Using this new technique, the protein was estimated to consist of 51% beta sheet and only 15% random coil in the presence of 6.6M deuterated guanidine hydrochloride.

FeNO structure in distal pocket mutants of myoglobin based on resonance Raman spectroscopy.

FeNO vibrational frequencies were investigated for a series of myoglobin mutants using isotope-edited resonance Raman spectra of (15/14)NO adducts, which reveal the FeNO and NO stretching modes. The latter give rise to doublet bands, as a result of Fermi resonances with coincident porphyrin vibrations; these doublets were analyzed by curve-fitting to obtain the nuNO frequencies. Variations in nuNO among the mutants correlate with the reported nuCO variations for the CO adducts of the same mutants. The correlation has a slope near unity, indicating equal sensitivity of the NO and CO bonds to polar influences in the heme pocket. A few mutants deviate from the correlation, indicating that distal interactions differ for the NO and CO adducts, probably because of the differing distal residue geometries. In contrast to the strong and consistent nuFeC/nuCO correlation found for the CO adducts, nuFeN correlates only weakly with nuNO, and the slope of the correlation depends on which residue is being mutated. This variability is suggested to arise from steric interactions, which change the FeNO angle and therefore alter the Fe-NO and N-O bond orders. This effect is modeled with Density Functional Theory (DFT) and is rationalized on the basis of a valence isomer bonding model. The FeNO unit, which is naturally bent, is a more sensitive reporter of steric interactions than the FeCO unit, which is naturally linear. An important additional factor is the strength of the bond to the proximal ligand, which modulates the valence isomer equilibrium. The FeNO unit is bent more strongly in MbNO than in protein-free heme-NO complexes because of a combination of a strengthened proximal bond and distal interactions.