BioPrint meets the AI age: development of artificial intelligence-based ADMET models for the drug-discovery platform SAFIRE

Background: To prioritize compounds with a higher likelihood of success, artificial intelligence models can be used to predict absorption, distribution, metabolism, excretion and toxicity (ADMET) properties of molecules quickly and efficiently. Methods: Models were trained with BioPrint database proprietary data along with public datasets to predict various ADMET end points for the SAFIRE platform. Results: SAFIRE models performed at or above 75% accuracy and 0.4 Matthew's correlation coefficient with validation sets. Training with both proprietary and public data improved model performance and expanded the chemical space on which the models were trained. The platform features scoring functionality to guide user decision-making. Conclusion: High-quality datasets along with chemical space considerations yielded ADMET models performing favorably with utility in the drug discovery process.

Discovery of 2-(((1r,4r)-4-(((4-Chlorophenyl)(phenyl)carbamoyl)oxy)methyl)cyclohexyl)methoxy)acetate (Ralinepag): An Orally Active Prostacyclin Receptor Agonist for the Treatment of Pulmonary Arterial Hypertension

The design and synthesis of a new series of potent non-prostanoid IP receptor agonists that showed oral efficacy in the rat monocrotaline model of pulmonary arterial hypertension (PAH) are described. Detailed profiling of a number of analogues resulted in the identification of 5c (ralinepag) that has good selectivity in both binding and functional assays with respect to most members of the prostanoid receptor family and a more modest 30- to 50-fold selectivity over the EP3 receptor. In our hands, its potency and efficacy are comparable or superior to MRE269 (the active metabolite of the clinical compound NS-304) with respect to in vitro IP receptor dependent cAMP accumulation assays. 5c had an excellent PK profile across species. Enterohepatic recirculation most probably contributes to a concentration-time profile after oral administration in the cynomolgus monkey that showed a very low peak-to-trough ratio. Following the identification of an acceptable solid form, 5c was selected for further development for the treatment of PAH.

(1aR,5aR)1a,3,5,5a-Tetrahydro-1H-2,3-diaza-cyclopropa[a]pentalene-4-carboxylic acid (MK-1903): a potent GPR109a agonist that lowers free fatty acids in humans

G-protein coupled receptor (GPCR) GPR109a is a molecular target for nicotinic acid and is expressed in adipocytes, spleen, and immune cells. Nicotinic acid has long been used for the treatment of dyslipidemia due to its capacity to positively affect serum lipids to a greater extent than other currently marketed drugs. We report a series of tricyclic pyrazole carboxylic acids that are potent and selective agonists of GPR109a. Compound R,R-19a (MK-1903) was advanced through preclinical studies, was well tolerated, and presented no apparent safety concerns. Compound R,R-19a was advanced into a phase 1 clinical trial and produced a robust decrease in plasma free fatty acids. On the basis of these results, R,R-19a was evaluated in a phase 2 study in humans. Because R,R-19a produced only a weak effect on serum lipids as compared with niacin, we conclude that the beneficial effects of niacin are most likely the result of an undefined GPR109a independent pathway.

Potent tricyclic pyrazole tetrazole agonists of the nicotinic acid receptor (GPR109a)

Tricyclic pyrazole tetrazoles which are potent partial agonists of the high affinity niacin receptor, GPR109a, have been discovered and optimized. One of these compounds has proven to be effective at lowering free fatty acids in vitro and in vivo.

3-(1H-tetrazol-5-yl)-1,4,5,6-tetrahydro-cyclopentapyrazole (MK-0354): a partial agonist of the nicotinic acid receptor, G-protein coupled receptor 109a, with antilipolytic but no vasodilatory activity in mice

The discovery and profiling of 3-(1H-tetrazol-5-yl)-1,4,5,6-tetrahydro-cyclopentapyrazole (5a, MK-0354), a partial agonist of GPR109a, is described. Compound 5a retained the plasma free fatty acid lowering effects in mice associated with GPR109a agonism, but did not induce vasodilation at the maximum feasible dose. Moreover, preadministration of 5a blocked the flushing effect induced by nicotinic acid but not that induced by PGD2. This profile made 5a a suitable candidate for further study for the treatment of dyslipidemia.

3-Nitro-4-amino benzoic acids and 6-amino nicotinic acids are highly selective agonists of GPR109b

A series of 3-nitro-4-substituted-aminobenzoic acids were prepared and found to act as potent and highly selective agonists of the orphan human GPCR GPR109b, a low affinity receptor for niacin. No activity was observed at the closely homologous high affinity niacin receptor, GPR109a. A second series, comprising 6-amino-substituted nicotinic acids was, also prepared and several analogues showed comparable activity to the nitroaryl series.

Agonist lead identification for the high affinity niacin receptor GPR109a

A strategy for lead identification of new agonists of GPR109a, starting from known compounds shown to activate the receptor, is described. Early compound triage led to the formulation of a binding hypothesis and eventually to our focus on a series of pyrazole acid derivatives. Further elaboration of these compounds provided a series of 5,5-fused pyrazoles to be used as lead compounds for further optimization.

Analogues of acifran: agonists of the high and low affinity niacin receptors, GPR109a and GPR109b

Recently identified GPCRs, GPR109a and GPR109b, the high and low affinity receptors for niacin, may represent good targets for the development of HDL elevating drugs for the treatment of atherosclerosis. Acifran, an agonist of both receptors, has been tested in human subjects, yet until recently very few analogs had been reported. We describe a series of acifran analogs prepared using newly developed synthetic pathways and evaluated as agonists for GPR109a and GPR109b, resulting in identification of compounds with improved activity at these receptors.

1-Alkyl-benzotriazole-5-carboxylic acids are highly selective agonists of the human orphan G-protein-coupled receptor GPR109b

1-Substituted benzotriazole carboxylic acids have been identified as the first reported examples of selective small-molecule agonists of the human orphan G-protein-coupled receptor GPR109b (HM74), a low-affinity receptor for the HDL-raising drug niacin. No activity was observed at the highly homologous high-affinity niacin receptor GPR109a (HM74A). The high degree of selectivity was attributed to a difference in the amino acid sequence adjacent to a key arginine-ligand interaction allowing somewhat larger ligands to be tolerated by GPR109b.

Mutation in the beta-tubulin signature motif suppresses microtubule GTPase activity and dynamics, and slows mitosis

We introduced a threonine-to-glycine point mutation at position 143 in the "tubulin signature motif" 140Gly-Gly-Gly-Thr-Gly-Ser-Gly146 of Saccharomyces cerevisiae beta-tubulin. In an electron diffraction model of the tubulin dimer, this sequence comes close to the phosphates of a guanine nucleotide bound in the beta-tubulin exchangeable E site. Both the GTP-binding affinity and the microtubule (MT)-dependent GTPase activity of tubulin isolated from haploid tub2-T143G mutant cells were reduced by at least 15-fold, compared to tubulin isolated from control wild-type cells. The growing and shortening dynamics of MTs assembled from alphabeta:Thr143Gly-mutated dimers were also strongly suppressed, compared to control MTs. The in vitro properties of the mutated MTs (slower growing and more stable) are consistent with the effects of the tub2-T143G mutation in haploid cells. The average length of MT spindles in large-budded mutant cells was only 3.7 +/- 0.2 microm, approximately half of the size of MT arrays in large-budded wild-type cells (average length = 7.1 +/- 0.4 microm), suggesting that there is a delay in mitosis in the mutant cells. There was also a higher proportion of large-budded cells with unsegregated nuclei in mutant cultures (30% versus 12% for wild-type cells), again suggesting such a delay. The results show that beta:Thr143 of the tubulin signature motif plays an important role in GTP binding and hydrolysis by the beta-tubulin E site and support the idea that tubulins belong to a family of proteins within the GTPase superfamily that are structurally distinct from the classic GTPases, such as EF-Tu and p21(ras). The data also suggest that MT dynamics are critical for MT function in yeast cells and that spindle MT assembly and disassembly could be coordinated with other cell-cycle events by regulating beta-tubulin GTPase activity.

Molecular hashkeys: a novel method for molecular characterization and its application for predicting important pharmaceutical properties of molecules

We define a novel numerical molecular representation, called the molecular hashkey, that captures sufficient information about a molecule to predict pharmaceutically interesting properties directly from three-dimensional molecular structure. The molecular hashkey represents molecular surface properties as a linear array of pairwise surface-based comparisons of the target molecule against a common 'basis-set' of molecules. Hashkey-measured molecular similarity correlates well with direct methods of measuring molecular surface similarity. Using a simple machine-learning technique with the molecular hashkeys, we show that it is possible to accurately predict the octanol-water partition coefficient, log P. Using more sophisticated learning techniques, we show that an accurate model of intestinal absorption for a set of drugs can be constructed using the same hashkeys used in the aforementioned experiments. Once a set of molecular hashkeys is calculated, its use in the training and testing of property-based models is very fast. Further, the required amount of data for model construction is very small. Neural network-based hashkey models trained on data sets as small as 30 molecules yield statistically significant prediction of molecular properties. The lack of a requirement for large data sets lends itself well to the prediction of pharmaceutically relevant molecular parameters for which data generation is expensive and slow. Molecular hashkeys coupled with machine-learning techniques can yield models that predict key pharmacological aspects of biologically important molecules and should therefore be important in the design of effective therapeutics.

Inactivity of N229A thymidylate synthase due to water-mediated effects: isolating a late stage in methyl transfer

Mutation of thymidylate synthase N229(177) to alanine results in an essentially inactive enzyme, yet it leads to formation of a stable ternary complex. The kinetics of N229(177)A show that kcat for Escherichia coli is reduced by 200-fold while the Km for dUMP is increased 200-fold and the Km for folate increased by tenfold versus the wild-type enzyme. The crystal structures of N229(177)A in complex with dUMP and CB3717, and in complex with dUMP alone are determined at 2.4 A, and 2.5 A resolution. These structures identify the covalently bound ternary complex and show how N229(177)A traps an intermediate, and so becomes inactive in a later step of the reaction. Since the smaller alanine side-chain at N229(177)A does not directly sterically impair binding of ligands, the structures implicate, and place quantitative limits on the involvement of the structured water network in the active site of thymidylate synthase in both catalysis and in determining the binding affinity for dUMP (in contrast, the N229(177)V mutation in Lactobacillus casei has minimal effect on activity).

D221 in thymidylate synthase controls conformation change, and thereby opening of the imidazolidine

In thymidylate synthase (TS), the invariant residue Asp-221 provides the only side chain that hydrogen bonds to the pterin ring of the cofactor, 5,10-methylene-5,6,7,8-tetrahydrofolate. All mutants of D221 except cysteine abolish activity. We have determined the crystal structures of two ternary complexes of the Escherichia coli mutant D221N. In a complex with dUMP and the antifolate 10-propargyl-5,8-dideazafolate (CB3717), dUMP is covalently bound to the active site cysteine, as usual. CB3717, which has no imidazolidine ring, is also bound in the usual productive orientation, but is less ordered than in wild-type complexes. The side chain of Asn-221 still hydrogen bonds to N3 of the quinazoline ring of CB3717, which must be in the enol form. In contrast, the structure of D221N with 5-fluoro-dUMP and 5,10-methylene-5,6,7, 8-tetrahydrofolate shows the cofactor bound in two partially occupied, nonproductive binding sites. In both binding modes, the cofactor has a closed imidazolidine ring and adopts the solution conformation of the unbound cofactor. In one of the binding sites, the pterin ring is turned around such that Asn-221 hydrogen bonds to the unprotonated N1 instead of the protonated N3 of the cofactor. This orientation blocks the conformational change required for forming covalent ternary complexes. Taken together, the two crystal structures suggest that the hydrogen bond between the side chain of Asp-221 and N3 of the cofactor is most critical during the early steps of cofactor binding, where it enforces the correct orientation of the pterin ring. Proper orientation of the cofactor appears to be a prerequisite for opening the imidazolidine ring prior to formation of the covalent steady-state intermediate in catalysis.

The additivity of substrate fragments in enzyme-ligand binding

BACKGROUND

Enzymes have evolved to recognise their target substrates with exquisite selectivity and specificity. Whether fragments of the substrate--perhaps never available to the evolving enzyme--are bound in the same manner as the parent substrate addresses the fundamental basis of specificity. An understanding of the relative contributions of individual portions of ligand molecules to the enzyme-binding interaction may offer considerable insight into the principles of substrate recognition.

RESULTS

We report 12 crystal structures of Escherichia coli thymidylate synthase in complexes with available fragments of the substrate (dUMP), both with and without the presence of a cofactor analogue. The structures display considerable fidelity of binding mode and interactions. These complexes reveal several interesting features: the cofactor analogue enhances the localisation of substrate and substrate fragments near the reactive thiol; the ribose moiety reduces local disorder through additional specific enzyme-ligand interactions; the pyrimidine has multiple roles, ranging from stereospecificity to mechanistic competence; and the glycosidic linkage has an important role in the formation of a covalent attachment between substrate and enzyme.

CONCLUSIONS

The requirements of ligand-protein binding can be understood in terms of the binding of separate fragments of the ligand. Fragments which are subsystems of the natural substrate for the enzyme confer specific contributions to the binding affinity, orientation or electrostatics of the enzymatic mechanism. This ligand-binding analysis provides a complementary method to the more prevalent approaches utilising site-directed mutagenesis. In addition, these observations suggest a modular approach for rational drug design utilising chemical fragments.

An essential role for water in an enzyme reaction mechanism: the crystal structure of the thymidylate synthase mutant E58Q

A water-mediated hydrogen bond network coordinated by glutamate 60(58) appears to play an important role in the thymidylate synthase (TS) reaction mechanism. We have addressed the role of glutamate 60(58) in the TS reaction by cocrystalizing the Escherichia coli TS mutant E60(58)Q with dUMP and the cofactor analog CB3717 and have determined the X-ray crystal structure to 2.5 A resolution with a final R factor of 15.2% (Rfree = 24.0%). Using difference Fourier analysis, we analyzed directly the changes that occur between wild-type and mutant structures. The structure of the mutant enzyme suggests that E60(58) is not required to properly position the ligands in the active site and that the coordinated hydrogen bond network has been disrupted in the mutant, providing an atomic resolution explanation for the impairment of the TS reaction by the E60(58)Q mutant and confirming the proposal that E60(58) coordinates this conserved hydrogen bond network. The structure also provides insight into the role of specific waters in the active site which have been suggested to be important in the TS reaction. Finally, the structure shows a unique conformation for the cofactor analog, CB3717, which has implications for structure-based drug design and sheds light on the controversy surrounding the previously observed enzymatic nonidentity between the chemically identical monomers of the TS dimer.

Site-directed mutagenesis of putative GTP-binding sites of yeast beta-tubulin: evidence that alpha-, beta-, and gamma-tubulins are atypical GTPases

The exchangeable GTP-binding site on beta-tubulin has been extensively studied, but the primary sequence elements which form the binding site on beta-tubulin remain unknown. We have used site-directed mutagenesis of the single beta-tubulin gene of Saccharomyces cerevisiae to test a model for the GTP-binding site on beta-tubulin, which was based on sequence comparisons with members of the GTPase superfamily [Sternlicht, H., Yaffe, M.B., & Farr, G. W. (1987) FEBS Lett. 214, 226-235]. We analyzed the effects of D295N, N298K, and N298Q mutations in a proposed base-binding motif, 295DAKN298, on tubulin-GTP binding and on nucleotide-binding specificity. We also examined the effects of a D203S mutation in a putative phosphate-binding region, 203DNEA206, on nucleotide binding affinity, on the assembly-dependent tubulin GTPase activity in vitro, and on the dynamic properties of individual "mutant" microtubules in vitro. The effects of the mutations on cell phenotype and on microtubule polymerization in cells were also measured. The results do not support the proposal that the 203DNEA206 and 295DAKN298 [corrected] motifs are cognate to motifs found in GTPase superfamily members. Instead, the data argue that the primary sequence elements of beta-tubulins that interact with bound nucleotide, and presumably also those of the alpha- and gamma-tubulin family members, are different from those of "typical" GTPase superfamily members, such as p21ras. The GTPase superfamily should thus be broadened to include not just the typical GTPases that show strong conservation of primary sequence consensus motifs (GxxxxGK, T, DxxG, NKxD) [corrected] but also "atypical" GTPases, exemplified by the tubulins and other recently identified GTPases, that do not show the consensus motifs of typical GTPases and which also show no obvious primary sequence relationships between themselves. The tubulins and other atypical GTPases thus appear to represent convergent solutions to the GTP-binding and hydrolysis problem.

beta-Tubulin mutation suppresses microtubule dynamics in vitro and slows mitosis in vivo

Microtubule (MT) dynamics vary both spatially and temporally within cells and are thought to be important for proper MT cellular function. Because MT dynamics appear to be closely tied to the guanosine triphosphatase (GTPase) activity of beta-tubulin subunits, we examined the importance of MT dynamics in the budding yeast S. cerevisiae by introducing a T107K point mutation into a region of the single beta-tubulin gene, TUB2, known to affect the assembly-dependent GTPase activity of MTs in vitro. Analysis of MT dynamic behavior by video-enhanced differential interference contrast microscopy, revealed that T107K subunits slowed both the growth rates and catastrophic disassembly rates of individual MTs in vitro. In haploid cells tub2-T107K is lethal; but in tub2-T107K/tub2-590 heterozygotes the mutation is viable, dominant, and slows cell-cycle progression through mitosis, without causing wholesale disruption of cellular MTs. The correlation between the slower growing and shortening rates of MTs in vitro, and the slower mitosis in vivo suggests that MT dynamics are important in budding yeast and may regulate the rate of nuclear movement and segregation. The slower mitosis in mutant cells did not result in premature cytokinesis and cell death, further suggesting that cell-cycle control mechanisms "sense" the mitotic slowdown, possibly by monitoring MT dynamics directly.

Microtubule dynamics modulated by guanosine triphosphate hydrolysis activity of beta-tubulin

Microtubule dynamic instability underlies many cellular functions, including spindle morphogenesis and chromosome movement. The role of guanosine triphosphate (GTP) hydrolysis in dynamic instability was investigated by introduction of four mutations into yeast beta-tubulin at amino acids 103 to 109, a site thought to participate in GTP hydrolysis. Three of the mutations increased both the assembly-dependent rate of GTP hydrolysis and the average length of steady-state microtubules over time, a measure of dynamic instability. The fourth mutation did not substantially affect the rate of GTP hydrolysis or the steady-state microtubule lengths. These results demonstrate that the rate of GTP hydrolysis can modulate microtubule length and hence dynamic instability.

Purification and biochemical characterization of tubulin from the budding yeast Saccharomyces cerevisiae

We describe a method for isolating milligram quantities of assembly-competent tubulin from the budding yeast Saccharomyces cerevisiae. The tubulin is > 95% purified and free of contaminating enzyme activities. As a result, it has been possible to determine the yeast tubulin equilibrium-binding constant for Mg-GTP and the tubulin GTPase activity under nonassembling and assembling conditions. We also determined the critical concentration for yeast tubulin polymerization and found it to be significantly lower than that for bovine brain tubulin under identical conditions. Similarly, the dynamic properties both of individual yeast microtubules and of bulk microtubule suspensions were significantly different from those of bovine brain microtubules free of microtubule-associated proteins. The data suggest that the properties of the yeast tubulin may reflect the particular functional requirements of the yeast cell. With this method, it is now possible to introduce any desired tubulin gene mutation into the budding yeast and correlate the phenotypic effects of the mutation in cells with the effects of the mutation on the biochemical and polymerization properties of the tubulin.