Self-organizing neural networks for modeling robust 3D and 4D QSAR: application to dihydrofolate reductase inhibitors

We have used SOM and grid 3D and 4D QSAR schemes for modeling the activity of a series of dihydrofolate reductase inhibitors. Careful analysis of the performance and external predictivities proves that this method can provide an efficient inhibition model.     

Brominated trimetrexate analogues as inhibitors of pneumocystis carinii and toxoplasma gondii dihydrofolate reductase

Five previously undescribed trimetrexate analogues with bulky 2′-bromo substitution on the phenyl ring were synthesized in order to assess the effect of this structure modification on dihydrofolate reductase inhibition. Condensation of 2-[2-(2-bromo-3,4,5-trimethoxyphenyl)ethyl]-1,l-dicyanopropene with sulfur in the presence of N,N-diethylamine afforded 2-amino-5-(2′-bromo-3′,4′,5′-trimethoxybenzyl)-4-methyl-thiophene-3-carbonitrile ( 15 ) and 2-amino-4-[2-(2′-bromo-3′,4′,5′-trimethoxyphenyl)ethyl]thiophene-3-car-bonitrile ( 16 ). Further reaction with chloroformamidine hydrochloride converted 15 and 16 into 2,4-diamino-5-(2′-bromo-3′,4′,5′-trimethoxybenzyl)-4-methylthieno[2,3-d]pyrimidine ( 8a ) and 2,4-diamino-4-[2-(2′-bromo-3′,4′,5′-trimethoxyphenyl)ethylthieno[2,3-d]pyrimidine ( 12 ) respectively. Other analogues, obtained by reductive coupling of the appropriate 2,4-diaminoquinazoline-6(or 5)-carbonitriles with 2-bromo-3,4,5-trimethoxyaniline, were 2,4-diamino-6-(2′-bromo-3′,4′,5′-trimethoxyanilinomethyl)-5-chloro-quinazoline ( 9a ), 2,4-diamino-5-(2′-bromo-3′,4′,5′-trimethoxyanilinomethyl)quinazoline ( 10 ), and 2,4-diamino-6-(2′-bromo-3′,4′,5′-trimethoxyanilinomethyl)quinazoline ( 11 ). Enzyme inhibition assays revealed that space-filling 2′-bromo substitution in this limited series of dicyclic 2,4-diaminopyrimidines with a 3′,4′,5′-trimethoxyphenyl side chain and a CH₂, CH₂CH₂, or CH₂NH bridge failed to improve species selectivity against either P. carinii or T. gondii dihydrofolate reductase relative to rat liver dihydrofolate reductase.     

Carbon-deuterium bonds as probes of dihydrofolate reductase

Much effort has been directed toward understanding the contributions of electrostatics and dynamics to protein function and especially to enzyme catalysis. Unfortunately, these studies have been limited by the absence of direct experimental probes. We have been developing the use of carbon-deuterium bonds as probes of proteins and now report the application of the technique to the enzyme dihydrofolate reductase, which catalyzes a hydride transfer and has served as a paradigm for biological catalysis. We observe that the stretching absorption frequency of (methyl- d 3) methionine carbon-deuterium bonds shows an approximately linear dependence on solvent dielectric. Solvent and computational studies support the empirical interpretation of the stretching frequency in terms of local polarity. To begin to explore the use of this technique to study enzyme function and mechanism, we report a preliminary analysis of (methyl- d 3) methionine residues within dihydrofolate reductase. Specifically, we characterize the IR absorptions at Met16 and Met20, within the catalytically important Met20 loop, and Met42, which is located within the hydrophobic core of the enzyme. The results confirm the sensitivity of the carbon-deuterium bonds to their local protein environment, demonstrate that dihydrofolate reductase is electrostatically and dynamically heterogeneous, and lay the foundation for the direct characterization protein electrostatics and dynamics and, potentially, their contribution to catalysis.     

Three-dimensional quantitative structure-activity and structure-selectivity relationships of dihydrofolate reductase inhibitors

Three-dimensional quantitative structure-activity relationship (3D-QSAR) modelling using comparative molecular similarity indices analysis (CoMSIA) was applied to a series of 406 structurally diverse dihydrofolate reductase (DHFR) inhibitors from Pneumocystis carinii (pc) and rat liver (rl). X-ray crystal structures of three inhibitors bound to pcDHFR were used for defining the alignment rule. For pcDHFR, a QSAR model containing 6 components was selected using leave-10%-out cross-validation (n= 240, q2 = 0.65), while a 4-component model was selected for rlDHFR (n= 237, q2 = 0.63); both include steric, electrostatic and hydrophobic contributions. The models were validated using a large test set, designed to maximise its diversity and to verify the predictive accuracy of models for extrapolation. The pcDHFR model has r2 = 0.60 and mean absolute error (MAE) = 0.57 for the test set after removing 4 outliers, and the rlDHFR model has r2 = 0.60 and MAE = 0.69 after removing 4 test set outliers. In addition, classification models predicting selectivity for pcDHFR over rlDHFR were developed using soft independent modelling by class analogy (SIMCA), with a selectivity ratio of 2 (IC50,rlDHFR/ IC50,pcDHFR) used for delimiting classes. A 5-component model including steric and electrostatic contributions has cross-validated and test set classification rates of 0.67 and 0.68 for selective inhibitors, and 0.85 and 0.72 for unselective inhibitors. The predictive accuracy of models, together with the identification of important contributions in QSAR and classification models, offer the possibility of designing potent selective inhibitors and estimating their activity prior to synthesis.     

How dihydrofolate reductase facilitates protonation of dihydrofolate

Dihydrofolate Reductase (DHFR) catalyzes the reduction of dihydrofolate (H2F) to tetrahydrofolate. On the basis of 10-12.5 ns molecular dynamics simulations of two conformations (closed and occluded) of the ternary DHFR/NADPH/H2F complex from Escherichia coli and a free energy perturbation approach, we have calculated the pKa value for the N5 atom in H2F. Our results suggest that the N5 atom in H2F is responsible for the pH dependency of the catalyzed reaction, meaning that DHFR facilitates protonation of H2F by approximately 4 pKa units. The mechanism behind this increase is due to favorable electrostatic interactions between the Asp27 residue and a proton at the N5 atom. The electrostatic interactions are enhanced by a hydrophobic active site, which to a large extent is made hydrophobic by the M20 loop in DHFR. Moreover, we find that the conformation imposed on H2F by DHFR to some extent also favors protonation of the N5 atom. Our results add support to previous findings and suggestions by Callender and co-workers [e.g., Deng, J.; Callender, R. J. Am. Chem. Soc. 1998, 120, 7730-7737] and explain why mutation of Asp27 may lead to severely reduced activity at neutral pH.     

Structure-based rational quest for potential novel inhibitors of human HMG-CoA reductase by combining CoMFA 3D QSAR modeling and virtual screening

3-Hydroxy-3-methylglutaryl-coenzyme A reductase (HMGR) catalyzes the formation of mevalonate. In many classes of organisms, this is the committed step leading to the synthesis of essential compounds, such as cholesterol. However, a high level of cholesterol is an important risk factor for coronary heart disease, for which an effective clinical treatment is to block HMGR using inhibitors like statins. Recently the structures of catalytic portion of human HMGR complexed with six different statins have been determined by a delicate crystallography study (Istvan and Deisenhofer Science 2001, 292, 1160-1164), which established a solid basis of structure and mechanism for the rational design, optimization, and development of even better HMGR inhibitors. In this study, three-dimensional quantitative structure-activity relationship (3D QSAR) with comparative molecular field analysis (CoMFA) was performed on a training set of up to 35 statins and statin-like compounds. Predictive models were established by using two different ways: (1) Models-fit, obtained by SYBYL conventional fit-atom molecular alignment rule, has cross-validated coefficients (q2) up to 0.652 and regression coefficients (r2) up to 0.977. (2) Models-dock, obtained by FlexE by docking compounds into the HMGR active site, has cross-validated coefficients (q2) up to 0.731 and regression coefficients (r2) up to 0.947. These models were further validated by an external testing set of 12 statins and statin-like compounds. Integrated with CoMFA 3D QSAR predictive models, molecular surface property (electrostatic and steric) mapping and structure-based (both ligand and receptor) virtual screening have been employed to explore potential novel hits for the HMGR inhibitors. A representative set of eight new compounds of non-statin-like structures but with high pIC(50) values were sorted out in the present study.     

Pyrimethamine analogs as strong inhibitors of double and quadruple mutants of dihydrofolate reductase in human malaria parasites

Pyrimethamine acts against malarial parasites by selectively inhibiting their dihydrofolate reductase-thymidylate synthase. Resistance to pyrimethamine in Plasmodium falciparum is due to point mutations in the DHFR domain, initially at residue 108 (S108N), with additional mutations imparting much greater resistance. Our previous work, the development of a simple rational drug design strategy to overcome such resistance, used suitable meta-substituents in the pyrimethamine framework to avoid the unfavorable steric clash with mutant side chains at position 108. Interestingly, the meta-chloro analog of pyrimethamine not only overcame the resistance due to S108N, but also that contributed by the more remote mutation, C59R. The present work improves on this by means of other meta-substituents. Against wild type DHFR, double mutant types A16V + S108T and C59R + S108T, and the highly pyrimethamine/cycloguanil-resistant quadruple-mutant form N51I + C59R + S108N + I164L, pyrimethamine itself gave Ki values of 1.5, 2.4, 72.3 and 859 nM, respectively. The meta-substituted analogs, especially the meta-bromo analog, were much more powerful inhibitors of these DHFRs, including the quadruple-mutant form (meta-bromo analog, Ki 5.1 nM). For comparison, the dihydropyrazine antifolate, WR99210, gave Ki values of 0.9, 3.2, 0.8 and 0.9 nM, respectively. Ki values were also measured against recombinant human DHFR, as were their activities against the growth of Plasmodium falciparum cultures bearing the double mutations (FCB and K1 strains) and quadruple mutation (V1/S) and the wild type (3D7). The meta-analogs were highly active against all of these, with the meta-bromo again being the strongest, having an IC50 of 37 nM against V1/S, compared to > 5000 nM for pyrimethamine itself and 1.1 nM for WR99210.     

Synthesis and biological evaluation of 5-arylfuro[2,3-d]pyrimidines as novel dihydrofolate reductase inhibitors

A series of about fifty novel 5-arylfuro[2,3-d]pyrimidine derivatives were synthesized as potential inhibitors of dihydrofolate reductase (DHFR) arising from different species. Weak enzyme inhibition was observed for most of the compounds, with only a few reaching IC50 values less than 30 microM. With regards to antibacterial and anti-malarial potency, only seven compounds showed a modest in vitro activity against some bacteria strains and only three products proved significantly active against P. falciparum.     

Modified ant colony optimization algorithm for variable selection in QSAR modeling: QSAR studies of cyclooxygenase inhibitors

A new version of an ant colony optimization (ACO) algorithm has been proposed. A modified ACO algorithm is proposed to select variables in QSAR modeling and to predict inhibiting action of some diarylimidazole derivatives on cyclooxygenase (COX) enzyme. As a comparison to this method, the evolution algorithm (EA) was also tested. Experimental results have demonstrated that the modified ACO is a useful tool for variable selection that needs few parameters to be adjusted and converges quickly toward the optimal position.     

Molecular orbital study of the ring nitrogen basicity of various dihydrofolate reductase inhibitors

We present molecular orbital (CNDO /2) calculations on the key fragments of different dihydrofolate reductase inhibitors. Distance geometry analysis, physicochemical parameter dependent QSAR , and molecular shape analysis raised some questions regarding the basicity of the ring nitrogen (N1) in these inhibitors and the effect of the various substituents on the basicity. We show that the ring nitrogen N1 of methotrexate has a considerably higher tendency to be protonated compared to that of folic acid. However, not all 2,4-diamino inhibitors are equally basic. Even 2-amino-4-hydroxyquinazoline is sufficiently basic to be protonated, but not the 2,4-diamino-5-sulfonyl derivatives. The pyrimidinium ion seems to be highly solvated, since in spite of its high protonation energy it is strongly basic. Triazines were found to be the most basic of all the classes studied.     

Effect of the structural characteristics of dihydrofolate reductase inhibitors on their metabolic properties

The orientation of dihydropyrimidine derivatives containing podand chains is determined in the model receptor using the CiS algorithm. The orientation of compounds with podand chains is compared with the location of compounds without podand chains in the model receptor. The pharmacophore and antipharmacophore parts of the compounds are analyzed. Amino acid residues responsible for the effective interactions of the compounds with podand chains with the binding site of dihydrofolate reductase (DHFR) are identified using the X-ray diffraction data. The mechanism of the action of tuberculostatic compounds with podand chains in their composition is proposed, which assumes the studied molecules to undergo metabolism by cytochrome P450 isoform 3A4 forming metabolites. The most active are the metabolites without the fragments of podand chains, which interact with DHFR. Therefore, the compounds containing podand chains are prodrugs.     

Drug design of human dihydrofolate reductase (hDHFR) inhibitors with deep learning

Human dihydrofolate reductase (hDHFR) inhibitors have been a popular research object designed as anti-cancer, anti-malarial, and antibacterial drugs for decades. Besides quantitative structure-activity relationship (QSAR), artificial intelligence (AI) has recently been introduced in numerous professional biological researches, such as molecular drug design and biological activity prediction. In this study, we construct a deep-learning workflow for designing novel hDHFR inhibitors. This workflow mainly includes two networks, as described in the following: The first one is the artificial neural network trained by the molecules selected from the ChEMBL database with experimental hDHFR inhibitions as the label to evaluate the bioactivity of the designed molecular structures constructed from the second network. The second network utilizes conditional generative and adversarial networks (cGAN) to generate candidate molecules with the desired properties. Finally, the obtained candidate molecules with high hDHFR inhibition are subjected to a molecular docking process to verify their binding patterns and affinity strengths inside the active site of hDHFR. In the end, we have successfully identified several novel drug-like compounds with hDHFR inhibition comparable to those currently used in clinics. We present a new tool to effectively design new drug-like compounds through an AI approach.     

Designing specific inhibitors against dihydrofolate reductase of W. bancrofti towards drug discovery for lymphatic filariasis

Structural Chemistry - Lymphatic filariasis (LF) is one among the leading neglected diseases caused by mosquitoe-borne parasite Wuchereria bancrofti to humans. Though drugs are available for the...     

Synthesis of 5-(4-substituted benzyl)-2,4-diaminoquinazolines as inhibitors of candida albicans dihydrofolate reductase

Several 5-(4-substituted benzyl)-2,4-diaminoquinazolines were prepared as potentially selective inhibitors of Candida albicans dihydrofolate reductase. These compounds were synthesized by a novel route, which included as a key step the displacement of a fluoro group in 2,6-difluorobenzonitrile by the anions of ethyl or methyl 4-substituted phenylacetates. The resultant diarylacetates were saponified and decarboxylated to the 2-fluoro-6-(4-substituted phenyl)benzonitriles. Ring closure of these benzonitriles with guanidine carbonate gave the 5-(4-substituted benzyl)-2,4-diaminoquinazolines.     

Freezing a single distal motion in dihydrofolate reductase

Constraining a single motion between distal residues separated by approximately 28 A in hybrid quantum/classical molecular dynamics simulations is found to increase the free energy barrier for hydride transfer in dihydrofolate reductase by approximately 3 kcal/mol. Our analysis indicates that a single distal constraint alters equilibrium motions throughout the enzyme on a wide range of time scales. This alteration of the conformational sampling of the entire system is sufficient to significantly increase the free energy barrier and decrease the rate of hydride transfer. Despite the changes in conformational sampling introduced by the constraint, the system assumes a similar transition state conformation with a donor-acceptor distance of approximately 2.72 A to enable the hydride transfer reaction. The modified thermal sampling leads to a substantial increase in the average donor-acceptor distance for the reactant state, however, thereby decreasing the probability of sampling the transition state conformations with the shorter distances required for hydride transfer. These simulations indicate that fast thermal fluctuations of the enzyme, substrate, and cofactor lead to conformational sampling of configurations that facilitate hydride transfer. The fast thermal motions are in equilibrium as the reaction progresses along the collective reaction coordinate, and the overall average equilibrium conformational changes occur on the slower time scale measured experimentally. Recent single molecule experiments suggest that at least some of these thermally averaged equilibrium conformational changes occur on the millisecond time scale of the hydride transfer reaction. Thus, introducing a constraint that modifies the conformational sampling of an enzyme could significantly impact its catalytic activity.