In Silico Design and Evaluation of Carboxylesterase Inhibitors

Abstract

Carboxylesterases (CEs) are important enzymes that catalyze biological detoxification, hydrolysis of certain pesticides, and metabolism of many esterified drugs. The development of inhibitors for CE has many potential uses, including increasing drug lifetime and altering biodistrubution; reducing or abrogating toxicity of metabolized drugs; and reducing pest resistance to insecticides. In this review, we discuss the major classes of known mammalian CE inhibitors and describe our computational efforts to design new scaffolds for development of novel, selective inhibitors. We discuss several strategies for in silico inhibitor development, including structure docking, database searching, multidimensional quantitative structure activity analysis (QSAR), and a newly-used approach that uses QSAR combined with de novo drug design. While our research is focused on design of specific inhibitors for human intestinal carboxylesterase (hiCE), the methods described are generally applicable to inhibitors of other enzymes, including CE from other tissues and organisms.

INTRODUCTION

Carboxylesterases are enzymes that convert esters into the corresponding alcohol and carboxylic acid. Carboxylesterases are members of the larger α/β-hydrolase fold family that includes a wide variety of enzymes such as lipases, cholinesterases, haloalkane dehydrogenases, and epoxide hydrolases^1–3). The catalytic machinery of the CE is an amino acid triad consisting of the residues serine, histidine, and glutamate that sit in a nucleophilic elbow, making these enzymes members of the even larger family of serine hydrolases.

Carboxylesterases are ubiquitous in nature. In mammals, they are expressed in numerous tissues^{4, 5)}. The hydrolytic activity will vary based on tissue type, with the liver isozyme being one of the most active of all isozymes studied. The hydrolytic activity of these enzymes are important in drug metabolism, protection against xenobiotics, and detoxifying pesticides such as pyrethoids⁶⁾. CE’s are also responsible for hydrolyzing clinically useful drug such as capecitabine and irinotecan^{7, 8)}, as well as illicit drugs such as heroin⁹⁾.

CEs have potential as therapy for pesticide overexposure. Organophosphate and carbamate insecticides have the potential to poison humans by acting on acetylcholinesterase (AChE). CEs can bind organophosphate and carbamate insecticides and reduce toxicity by two pathways: metabolism of the insecticide by CEs and irreversible binding by the insecticides^{10, 11)}.

Carboxylesterases are also important in the activation of prodrugs. One such clinically important prodrug is irinotecan (7-ethyl-10-[4-(1-piperidino)-1-piperidino]carbonyloxycamptothecin), an anticancer agent used as front-line therapy for colorectal cancer (Figure 1). In humans, several CEs have been well characterized, particularly liver carboxylesterase 1 (hCE1) and human intestinal carboxylesterase (hiCE), the latter of which is localized to the small intestinal epithelia. Both hCE1 and hiCE are known to convert irinotecan into its active form, SN-38 (7-ethyl-10-hydroxycamptothecin; Fig. 1). However, hiCE is expressed at high levels in the small intestine and is much more efficient at converting irinotecan than hCE1^{8, 12)}. Thus, overproduction of SN-38 occurs in the small intestine during irinotecan therapy, and tissue damage to this organ contributes to delayed diarrhea in patients, a side-effect that often requires hospitalization¹³⁾. This is the dose-limiting toxicity of irinotecan. Development of hiCE-selective CE inhibitors to reduce or eliminate unwanted toxicity of irinotecan has been the recent goal of our laboratories. Selective hiCE inhibitors are envisioned as adjuvant therapy for the modulation of diarrhea, potentially allowing for higher dosing of the drug and more effective treatment of colorectal cancer.

Figure 1.

Irinotecan (left) conversion to SN-38 (right) is one of many reactions catalyzed by carboxylesterases.

Several molecular structural scaffolds of CE inhibitors exist, which include, sulfonamides, benzils, benzoins, carbamates, isatins, organophosphates, oxysterols, pyrethoids, acridines, trazines, trifluromethylketones, piperidines, serine specific agents, and inorganic compounds. In general, most of these inhibitors show limited selectivity or specificity among isozymes or across species.

Over the last several years, we have developed a number of specific or semi-specific hiCE inhibitors. However, one difficulty that has persisted among these inhibitors is poor water solubility. This review gives a brief overview of previously described molecular scaffolds of CE inhibitors, followed by a description of recent developments in the in silico methodologies we are using to design new selective and specific inhibitors with greater water solubility and overall “drugability” of their chemical properties. Other, more detailed reviews of CE inhibitors have appeared recently^{14, 15)}. Here, our focus is on computer-based development of selective inhibitors of hiCE with the specific purpose of modulating irinotecan activation in the small intestine. However, the methodology used is general enough that it could be applied to the development of inhibitors of any enzyme.

Carboxylesterase Inhibitors: Scaffolds, Selectivity, Specificity and Mechanism of Inhibition

1. Acetylcholinesterase Inhibitors and their CE inhibition

We begin this review by looking at selectivity for CEs by several known groups of AChE inhibitors, starting with the alkylphosphonic esters, also known as organophosphate (OPs), which may be the most well known class of CE inhibitors. OPs are commonly used as insecticides (e.g., malathion (diethyl(dimethoxythiophosphorylthio) succinate) and paraxon (diethyl 4-nitrophenyl phosphate). Organophosphates are toxic due to their inhibition of AChE activity, which allows the muscarinic receptors to be continually activated by acetylcholine. The result of this AChE inhibition is the cholinergic effect, with symptoms of salivation, lacrimation, urination, defecation, gastrointestinal distress, vomiting, muscle spasms and ultimately death if the reaction is severe enough. Hence, sarin (O-isopropylmethylphosphonofluoridate), soman (O-pinacolylmethylphosphonofluoridate) and tabun (ethyl dimethylamidocyanophosphate), which demonstrate high affinity for human AChE, have been used as chemical warfare agents.

Carboxylesterases can also be irreversibly inhibited by organophosphates. The mechanism of inhibition involves the attack from the enzyme’s serine Oγ on the phosphate atom of the organophosphate. This creates a covalent bond ending the mechanism at an acyl–enzyme intermediate. Strategies to develop CEs to combat chemical warfare compounds have been pursued by the military. Essentially, CEs could be used as “bioscavengers”, which work by sequestering or hydrolyzing a toxic substrate¹⁶⁾. Recent work shows that hCE1 shows a preference for binding the P_R enatiomers of soman and cyclosarin analogues (1700-, 2900-fold respectively) and a slight preference for the P_S enantiomer of sarin analogues (5-fold)¹⁷⁾. However, only one group of OPs has been found to be semi-selective for CEs: the benzodioxaphosphorines (Bomins; attributed to Patent USSR 06.22.1985. No. 1187444). Their selectivity however is only about 10-fold greater than for AChE, and hence the problematic toxicity of OPs make them an unlikely scaffold for development of clinically useful compounds.

The carbamates (CBs) contain a central amide ester group and two alkyl or aryl substituents located on the nitrogen atom. CBs are not as toxic as the OPs but they still induce the same cholinergic effect. The CBs are reversible inhibitors of AChE and, like the OPs, have been used as insecticides (e.g., Carbaryl (1-naphthyl methylcarbamate)). The carbamates are also used in many clinical applications. For example rivastigmine (S)-N-Ethyl- N-methyl- 3-[1-(dimethylamino)ethyl]-phenyl carbamate is used to treat Alzheimer’s disease Rivastigmine is a selective, reversible brain AChE inhibitor¹⁸⁾.

Irinotecan also contains a carbamate moiety (Figure 1), which is primarily responsible for its initial cholinergic activity^{19, 20)}. Carbamates for CE inhibition have been explored by the Potter and Danks groups, who showed that four nitrophenyl derivatives with carbamate linked side chains could selectively inhibit hiCE and rCE²¹. These derivatives were the first attempt at developing specific hiCE inhibitors. However, inhibition by these compounds is also confounded by their AChE inhibition, and hence has the potential problems associated with all cholinergic drugs.

Trifluoromethylketones (TFKs), like the OPs and CBs will also inhibit AChE; the carbonyl carbon of the TFK can form a covalent bond to the serine residue of the catalytic triad. Thus they are non-specific inhibitors of CEs. Although these CE inhibitors are some of the most potent known, they exhibit both partially competitive and partially noncompetitive modes of inhibition. Their mode of binding is somewhat unique compared to the rest of the CE inhibitors, in that they take time to reach equilibrium before maximum inhibitor potency can be reached²²⁾. TFKs inhibit a range of mammalian CEs. Hence, selectivity toward hiCE is problematic.

Acridine inhibitors of AChE include tacrine, which is used to treat Alzheimer’s disease. Tacrine itself was not an inhibitor of hCE1, but 6,9-diamino-2-ethoxyacridine and 9-amino-6-chloro-2-methoxyacridine did inhibit hCE1 selectively over hiCE. These derivatives are low μM inhibitors of hCE1²³⁾.

2. More specific CE inhibitors

The AChE/CE cross inhibitors listed above came primarily from studies that emphasized the similar catalytic activity of the members of the α/β-hydrolase fold family. Indeed, the very first collaboration between the Potter and Wadkins laboratories was on molecular modeling of the interaction of irinotecan with AChE and the related butyrylcholinesterase²⁴⁾. In the subsequent decade, the quest for isozyme-specific CE inhibitors has produced several molecular scaffolds that do not possess anti-AChE activity. The design of many of the analogs was a combination of chemical library screens to isolate reasonably selective CE inhibitors, followed by chemical intuition and plausible synthetic schemes.

Telik’s Target Related Affinity Profiling (TRAP) method was originally employed in the search for novel CE inhibitors. In this method, compounds were screened for their binding affinities toward a panel of protein targets. The binding affinities were subsequently used to create an “affinity fingerprint”, which was subsequently used to identify novel inhibitors. Telik’s methodology has the advantage of needing few initial compounds to generate several new scaffolds of inhibitors. The TRAP analysis had been utilized in the identification of novel inhibitors for cyclooxygenase-1 (COX-1)²⁵⁾. Hsu and coworkers used 19 known COX-1 inhibitors, all reversible competitive inhibitors of COX-1, comprising 8 structurally dissimilar classes, to create an affinity fingerprint for COX-1 inhibition. Using TRAP analysis, they derived 3 new COX-1 inhibitors. We used this method to identify novel CE inhibitors. The sulfonamides and benzils were both discovered through this methodology^{26, 27)}.

3. Sulfonamides

The initial screening of a library of compounds isolated 9 sulfonamide derivatives (Fig. 2) as specific hiCE inhibitors vs. other human hydrolases²⁸⁾. Using a limited QSAR analysis of these 9 compounds, a number of other sulfonamides were synthesized or obtained from commercial sources. Examination of this larger group of sulfonamides determined that the ring at the central core moiety could accommodate both benzene and fluorene^{27, 29)}. The fluorene sulfonamides, in general, are more potent than the benzene sulfonamides with inhibition constants ranging from 41 nM to 3240 nM for benzene sulfonamides and from 14 nM to 91 nM for the fluorene sulfonamides. The majority of sulfonamides do not inhibit AChE or butyrylcholinesterase (BChE), and hence are highly selective CE inhibitors.

Figure 2.

Chemical structure of 4 distinct classes of inhibitors that show selectivity for carboxylesterase isoforms.

The mechanism of inhibition for the sulfonamides is partially competitive. However, unlike the majority of the CE inhibitors, there is no carbonyl carbon atom or electron deficient atom that would be susceptible to attack from the serine Oγ residue. Attack on the sulfur atom of the sulfonamide is an energetically disfavored process due to the stability of the sulfonamide. The sulfonamides could potentially hydrogen bond to the enzyme and lock it into a stable complex, or they may bind to the opening of the active site, blocking the entrance to the gorge.

The extended conformation of the sulfonamides matches the shape of the active site gorge of a homology model of hiCE, so it is entirely possible that they occupy the binding site of the enzyme, giving rise to isoform specificity. However, the sulfonamides as a whole suffer from very poor water solubility. Evaluation of over 50 sulfonamides²⁹⁾ revealed an inverse correlation of clogP and log K_i. Hence, the least soluble compounds were the best inhibitors of the enzyme. This is the major difficulty with using sulfonamides as scaffolds for a potential drug application.

4. Benzils

The diphenylethane-1,2-dione (benzil; Fig. 2) analogues are another class of isotype-selective mammalian CE inhibitors that we have examined after the initial drug screening. Benzil itself is not a new compound (discovered in the late 1800s). However, it was not until 2005 that is was discovered to be a selective inhibitor for human carboxylesterases²⁶⁾. Inhibition constants (K_i) ranged from 4 nM to 18 μM for 31 benzil analogues that were evaluated, with no inhibition of AChE. The analogs were found to be a competitive reversible inhibitor of the CEs The proposed mechanism of inhibition involves the inability of the enzyme to release the aldehyde as a leaving group. With no appropriate leaving group after the formation of the tetrahedral intermediate, the initial carbonyl is reformed. The crystal structure of hCE1 has been solved with benzil in the active site, suggesting that cleavage of the dione moiety could occur, consequently generating a benzoic acid or benzaldehyde³⁰⁾. In contrast to the sulfonamides, there is no correlation between clogP values for these compounds and their K_i values, making them a much more interesting platform for drug development. Their selectivity appears to arise from the dihedral angle of the dione moiety. When the carbonyl oxygen atoms are cis-coplanar, greater selectivity for hCE1 occurs, while non-planarity results in selectivity for hiCE³¹⁾.

5. Benzoins and Fluorobenzoins

During the exploration of the benzil scaffold, compounds of similar structure were also tested for activity against the mammalian CEs. The compound 1,2-diphenyl-2-hydroxy-ethanone (benzoin; Fig. 2) was found to be a weak but selective inhibitor of CEs, having a K_i of 2.7 μM for hiCE and 7.2 μM for hCE1. Subsequent addition of electron withdrawing groups to the benzene rings in both benzils and benzoins produced more potent, highly selective inhibitors In particular, addition of fluorine significantly increased their ability to inhibit mammalian CEs without resulting in inhibition of AChE or BcCHE. The inhibition constants for the fluorinated analogues ranged from 8 nM to 1.3 μM³²⁾.

in silico Methods and Development of Isatins

1. Computational approaches

One conceivable method of computationally deriving selective inhibitors for CEs might be to dock small molecule libraries into the active site of CEs in order to predict new inhibitors. However, in practice there are complications with that strategy that limit its effectiveness. The substrate specificity of CEs is dependent on two structural features: the dimensions of their active site gorge and the external opening to the gorge. We have gathered preliminary data on these important parameters using molecular dynamics calculations of CEs from rabbit, human liver, human intestine, and a bacterial CE (rCE, hCE1, hiCE, and pNB, respectively). In addition, we have used normal mode calculations to examine low-frequency motions of the CEs (large conformational changes). Both the active site gorge diameter and the opening to the active site fluctuate significantly with time (Fig. 3), so determining which structure the known inhibitors are binding to is complicated. The basic rationale for the need to include molecular dynamics in the development of enzyme inhibitors for the related AChE has been recently reviewed³³⁾. This rationale also applies to CEs. Briefly, enzymes are in constant motion at temperatures near 37°C and the understanding of the fluctuation is crucial for in silico docking or assembly of inhibitors. In the case of the α/β hydrolase fold family, the active site residues are at the bottom of a ~22 Å gorge, the walls of which are also fluctuating. The crystal structure of hiCE has not been determined, and so by necessity a homology model would need to be used for this enzyme (although, it should be noted that earlier homology models of rCE²⁰⁾ based on the folding of AChE were remarkably similar to the subsequently-determined crystal structure³⁴⁾; α-carbon RMSD ~ 2Å). The crystal structures for the other 3 enzymes (rCE, hCE1, pNB) have been determined and hence MD calculations are easily accomplished. Even though the crystal structures exist, there is an additional reason MD is an important tool for docking ligands into these molecules. As with other enzymes, the crystal structures containing known substrates or inhibitors in the active site have an active site that is too small to accommodate other known, larger substrates. For example, the structure of hCE1 containing a product of benzil hydrolysis would be too close-packed to allow placement of CPT-11 in the same locale³⁰⁾. This can be resolved by allowing the enzyme structure to fluctuate. An example of this is shown in Fig. 3, where the active site gorge diameter size of the pNB CE is calculated throughout a lengthy 10 ns MD simulation. Note that access to the active site can fluctuate from as little as 3.0 Å to as wide as 7.5 Å over a relatively short period of 1 ns. Other regions of structural fluctuation in CEs are the loops that form a putative lid over the entrance to the active site gorge. Normal mode analysis (NMA) of hCE1 and a homology model of hiCE (modeled using hCE1 as a template) was performed with the ElNémo web server³⁵⁾, which is a web interface to the Elastic Network Model that implements the ‘rotation-translation-block’ (RTB) approximation. The lowest non-zero frequency mode for the enzymes having a high degree-of-collectivity (mode 7) are shown in Fig. 4. It is this fluctuation that makes MD a critical component of inhibitor design by computer for the family of α/β hydrolases. However, these fluctuations are also a drawback to this type of computational design since computer docking to multiple enzyme structures must be performed.

Figure 3.

Fluctuation in the diameter of the active site gorge of a carboxylesterase from B. subtilis. The analysis was performed with a procedure modified from the work of McCammon and colleagues⁶⁴⁾. The solvent-accessible surface was calculated with an increasing probe radius until it no longer made contact with the active site Ser and His residues. The structure was taken from PDB code 1QE3⁶⁵⁾.

Figure 4.

Fluctuation of the loop forming part of the “lid” over the active site in carboxylesterases (yellow residues) determined from normal mode calculations for (A) hCE1 and (B) hiCE homology model. The “lid” is oriented toward the bottom of each molecule. Arrows indicate the extent of motion for the entire enzyme for the lowest frequency mode (mode 7). The view is into the active site gorge. The active site residues Ser, Glu and His are shown in green. While the effect on catalysis of these loops is unknown, they may serve to either bind substrate and guide it toward the active site gorge or to cover the active to prevent diffusion of substrate out of the gorge before hydrolysis.

The combination of MD and docking has been used for development of specific inhibitors of AChE vs. other serine hydrolase enzymes, and hence we expect this approach has the potential to work for CEs as well. Generic docking of known AChE inhibitors to a static AChE are generally not predictive for the relative magnitudes of the inhibition constants³³⁾. This has been attributed to the inability of static models to correctly calculate the entropy change upon inhibitor binding. A more successful approach taken by the McCammon laboratory involves using not just one static structure of CE for docking potential inhibitors, but rather numerous CE structures taken from a molecular dynamics simulation³⁶⁾. However, given the complications inherent in this approach, we developed a computationally-simpler multi-dimensional QSAR model as a reduced representation of the inhibitor binding site. This has the advantage over crystal structure docking approaches in that it does not require us to know the molecular details of the inhibitor binding site, just the chemical properties and K_i values for the inhibitors.

The idea for the indole-2,3-diones (isatin compounds; Fig. 2) was developed from this QSAR approach using biochemical data on the sulfonamide and benzil inhibitors. Our data suggested the presence of aromatic moieties were important for inhibition^{31, 37)}. Other earlier work found that there was a size constraint on the entrance to the active site gorge²⁰⁾. Using these two considerations as a parameter guide, a database search of commercially available compounds related to benzil was initiated, resulting in isatins as potential inhibitors of CEs. Simultaneously, a combination of QSAR and computerized model building using the sulfonamide data led to the prediction that indole-containing compounds would be CE inhibitors, and would also lead to selectivity for hiCE (Fig. 5). This was ultimately borne out by analysis of 74 compounds related to isatin, and the discovery of several that were selective for hiCE³⁸⁾, with inhibition constants as low as 6 nM. Below, we describe in detail the computational analysis that led to the predicted structures containing indole. Further, we describe the QSAR “grand model” that combines all compounds that we have evaluated for inhibition of CEs.

Figure 5.

Selected indole-containing compounds originally predicted by the de novo design software LigBuilder to be good inhibitors of hiCE using a QSAR model based entirely on sulfonamides. The prediction led to discovery and evaluation of isatins³⁸⁾.

2. Quantitative Structure Activity Relationships (QSAR)

QSAR correlates the molecular structure and properties of a set of compounds with their activity. The parameters used for QSAR are molecular descriptors that can range from a simple count of atoms to HOMO and LUMO calculations; electronegativity; and other quantum molecular features. A QSAR model can also be modeled as a three-dimensional structure, which will show visually the nature of the relationship between the inhibitors and the receptor. QSAR has long been popular in assisting in the determination of essential interactions between the receptor and inhibitor. In our previous studies, almost all of the CE inhibitors have been investigated through QSAR. These studies generated suggestions for improving the activities of inhibitor compounds (e.g., the addition of halogens, and inclusion of a larger aromatic moiety in the core of the inhibitor structure^{28, 39, 40)}.

QSAR reveals the important interactions between the ligands and the receptor site by using ligands that are known to be active inhibitors of the receptor of interest. The de novo design is an approach to structure-based drug discovery that utilizes a binding site pocket to build ligands specifically for the receptor^{41, 42)}. Common de novo design programs that are used today are GRID⁴³⁾, SPROUT^{44, 45)}, CONCERTS⁴⁶⁾, SYNOPSIS⁴⁷⁾, LeapFrog⁴⁸⁾, and LigBuilder⁴²⁾.

Using QSAR models as pocket sites for de novo design is an approach that can be utilized when no experimental receptor structure is available. Combining QSAR and de novo design has been used by several investigators^49–53). For example, Gueto and coworkers⁴⁹⁾ developed a CoMFA (Comparative Molecular Fields Analysis) model for aromatase inhibitors with 45 training set compounds and used a test set of 10 sulfonanilide compounds. This CoMFA model was used to generate new inhibitors in silico with the NEW module of LeapFrog, followed by calculation of their predicted activity. As another example, Kapou and coworkers⁵⁰⁾combined CoMFA and CoMSIA with LeapFrog to study and design steroidal mustard esters using a training set of 26 compounds and test set of 12 compounds. This model was then subsequently used to design new inhibitors using the OPTIMIZE module of LeapFrog.

CoMFA uses a grid-based approximation to develop 3D-QSAR models, and includes parameters such as steric and electrostatic effects by using both a Lennard-Jones and Coulombic potential. CoMSIA and LeapFrog also use a grid-based approximations to develop new compounds with good binding energies. In our laboratory, we have used the Quasar program^{54, 55)} to develop a 6D-QSAR model for hiCE inhibitors, and the de novo design program LigBuilder to generate new scaffolds of hiCE inhibitors (Table 1). The Quasar program is a grid-based technique as well, and it employs a genetic algorithm for model generation and induced fit. In addition of the 3 spatial dimensions of Quasar, the 4^th dimension allows for analysis of different conformations of ligands (4D-QSAR), while the 5^th dimension takes into account the potential for induced fit (5D-QSAR)⁵⁴⁾. The 6^th dimension of Quasar added the ability to investigate different solvation effects (6D-QSAR)⁵⁵⁾. The following equation is used for calculation of binding affinity in Quasar.

Table 1.

Potential selective hiCE inhibitors output by LigBuilder using the QSAR structure shown in Fig. 6. While many of these compounds are synthetically unfeasible, they do present a series of scaffolds that, combined with chemical intuition, may produce selective inhibitors in much the way the isatins were discovered.

$${\text{E}}_{\text{binding}}={\text{E}}_{\text{ligand}-\text{receptor}}-{\text{E}}_{\text{ligand}\text{desolvation}}-{\text{E}}_{\text{ligand}\text{strain}}-\text{T}\mathrm{\Delta }\text{S}-{\text{E}}_{\text{induced}\text{fit}}$$

The term E_{ligand–receptor} is calculated by the following equation,

$$E_{\text{electrostatic}}+E_{\text{van}\text{der}\text{Waals}}+E_{\text{hydrogen}\text{bonding}}+E_{\text{polarization}}.$$

Spreafico and coworkers have successfully used Quasar to develop models that include several classes of compounds⁵⁶⁾. Their data indicate that Quasar is effective when combining steroids, quinoline derivatives fluorophenylindazole derivatives, and spirocyclic derivatives as inhibitors of glucocorticoid receptor.

The QSAR model developed in our study included sulfonamide, benzil, fluorobenzoin, and isatin inhibitors (Fig. 2). The TFKs were omitted because their mechanism of binding is different from the rest of these inhibitors, and requires a pre-incubation time to be most effective⁵⁷⁾. A more thorough investigation of TFK inhibition has recently appeared⁴⁰⁾. Our previous studies have used QSAR for development of pseudo-receptor site models in an effort to delineate pertinent information about the active site that could be useful in the design of new inhibitors^{26, 27, 29, 31, 57–59)}. However, these studies only somewhat used the pseudo-receptor site models for de novo design of potential inhibitors and were derived from limited data sets. Here we report our results on predicted structures using a “grand model” of CE inhibition based on data from 210 CE inhibitors comprising four inhibitor families. Our expectations were that using this pseudo-receptor site as a hypothetical binding site might yield inhibitors other than indoles with improved drug-like properties (e.g., better water solubility).

The 6D-QSAR modeling of hiCE inhibitors was performed using Quasar 6.2 software running under Mac OS X. The 210 structures used included 4 diverse scaffolds; benzils, fluorobenzoins, isatins, and sulfonamides. Each analog had been previously drawn using Chemdraw, imported into Chem3D, and minimized first by using the MM2 molecular mechanics force field, followed by the PM3 semi-empirical method within MOPAC⁶⁰⁾. The solvation energies and charges were calculated using AMSOL and the SM5.42R solvation method⁶¹. SM5.42R is a rigid solvation model, which optimizes a compound’s structure in the gas phase but optimizes only its electronic structure in the solvent phase. Gas phase charges were used for all QSAR analysis.

A training set of 133 compounds and a test set of 77 compounds were used to build the QSAR model. The K_i values that were used for these computations were those derived from the inhibition of o-nitrophenylacetate (oNPA) hydrolysis. As with any QSAR model that uses structurally divergent ligands, the complication in this process is the choice of how to align the structures with one another. We chose the central benzene ring of the sulfonamide inhibitors as the template to which all the other compounds would be aligned. The resulting 3D-pseudo-receptor site model describing the interactions between the active site and the ligands is shown in Fig 6, and is referred to as the “grand model” for hiCE inhibitors. The r² (squared correlation coefficient of model) and q² (predicted squared correlation coefficient, which describes the predictive quality of the model) generated were both 0.835, and the predicted vs. experimental ΔG° for inhibition constant values are shown in Fig. 7. This QSAR model was fairly robust in prediction of inhibitor binding, and therefore we used it as a template upon which to build new molecules.

Figure 6.

Stereo view of he 3-dimensional “grand model” for hiCE inhibitors is depicted as colored spheres on a hydrophobic gray grid. Areas that are hydrophobic are indicated in gray, whereas dark blue spheres represent areas that are positively charged (+0.25e) and light blue spheres correspond to lesser charge (+0.1e). Dark red spheres represent areas that are negatively charged (−0.25e), light red spheres indicate less negative charge (−0.1e). In all cases, e is the charge of the proton. The structure of representative sulfonamides (black), isatins (blue), benzils (green), and benzoins (pink) are shown. The figure was generated using Molscript⁶⁶⁾ and Raster3D⁶⁷⁾.

Figure 7.

Predicted vs. experimental ΔG° values (in kcal/mol) for inhibition constants of 4 classes of molecules that inhibit the hiCE hydrolysis of o-nitrophenylacetate. K_i values were determined using a partially-competitive inhibitor model. The training set is indicated by the filled ovals, while the test set is represented by open boxes. The compound labeled “a” is a disulfide and is the lone member having this moiety. The compound labeled “b” is a 2,5-bis(trifluoromethyl) benzil analog. Its surprising lack of activity is under investigation. For clarity, compounds with K_i values > 100 μM are not included on this plot.

3. de novo ligand design

The next step in the process was to have LigBuilder synthesize potential inhibitors based on the 3D-QSAR structure. LigBuilder begins with a user-defined seed structure that is the starting point for building new ligands by adding molecular fragments to the seed until the binding site is fully occupied. Each compound derived in this manner is evaluated via the Lipinski “rules”⁶²⁾, and compounds violating these rules are rejected. LigBuilder uses an empirical scoring function to evaluate binding affinity, which is much different than that used by Quasar. LigBuilder alone has been combined with docking and pharmacophore modeling to design inhibitors for tyrosine kinase 2⁶³⁾, but to our knowledge has not been combined with QSAR models for lead compound development. To account for the other 3-dimensions of 6-D QSAR, the vertices of the model shown in Fig. 6 were assigned pseudo-atomic properties within LigBuilder based on the corresponding property indicated in the legend of Fig. 6. With this model as a hypothetical active site, potential inhibitors for hiCE were constructed from the seed structures 1,2-dione, p-aminoaniline, and benzene. For benzene, growth was allowed from all carbons in the ring. For p-aminoaniline, two seeds were used. The first allowed growth only off the amino group nitrogens, while the second allowed growth off both the amino groups and all unsubstituted ring carbons.

The results of this process is shown in Table 1 for the different starting seeds. Once the structures were generated, they were evaluated in the 6D-QSAR shown in Fig. 6, and the top 5 ranked structures (lowest predicted K_i values) are shown. All of these computationally-derived structures were predicted to have good activities against hiCE, with free energy scores in the range of −10 to −13 kcal/mol (K_i from 46.7 nM to 0.3 nM). While only a select few compounds are shown in Table 1, over 40 new structural classes of inhibitors emerged from this in silico process. These classes show new features such as a large aromatic or planar groups attached to a long alkyl chain, the inclusion of a large part of the polar groups within cyclic ring systems, two aromatic group join together by a central alkyl group, and hybrid compounds that are a cross between two or more groups of inhibitors.

Table 1 also illustrates a problem with computer-generated ligands. Clearly, some of the predicted compounds are not feasible as biochemically active molecules, particularly the anhydrides that would be highly reactive with water. However, ignoring such unfeasible compounds, water solubility of most of the predicted structures using ALOGPS 2.1 indicated they have greater solubility than the sulfonamides, and many have solubility comparable to the anticancer drug irinotecan. This suggests, that the solubility can be enhanced while maintaining or even improving the activity when a combined approach is used. Synthesis of a subset of these new classes of inhibitors is underway to verify the predicted results.

Conclusions

Over the last decade, QSAR has allowed a detailed analysis of known CE inhibitors and helped to generate CE inhibitors with greater specificity. As an example, tetrafluorine substituted sulfonamide analogs were synthesized when QSAR analysis of nine sulfonamides suggested the addition of halogens would increase potency²⁷⁾. The observed results concluded that this new inhibitor had greater selectivity for hiCE versus hCE1 than the previous nine. Another example is the fluorene sulfonamides²⁹⁾, which were pursued because results from pseudo-receptor sites suggested that the active site of hiCE could accommodate a larger aromatic moiety. The fluorene sulfonamides proved to be much more potent than the structurally-similar benzene sulfonamide. The specificity was also not affected by inclusion of the fluorene moiety in the core. Hence, QSAR (particularly, the 6D-QSAR methodology used by Quasar) is an invaluable tool for developing specific hiCE inhibitors.

While QSAR analysis has aided in the design to-date of more potent and specific inhibitors, it has not effectively addressed the issue that plagues most of the CE inhibitors designed theoretically: low aqueous solubility. The analysis and design described in this article suggest features of potential inhibitors that will increase their water solubility and yet retain or enhance potency. Further, our QSAR models are beneficial in understanding how the inhibitors are interacting within their binding site, which is presumed to be the enzyme active site. Such studies are useful in pointing toward optimization of lead compounds for structural scaffolds that may be discovered through means other than computer-generated structures.