Predicting Monoamine Oxidase Inhibitory Activity through Ligand-Based Models

Abstract

The evolution of bio- and cheminformatics associated with the development of specialized software and increasing computer power has produced a great interest in theoretical in silico methods applied in drug rational design. These techniques apply the concept that “similar molecules have similar biological properties” that has been exploited in Medicinal Chemistry for years to design new molecules with desirable pharmacological profiles. Ligand-based methods are not dependent on receptor structural data and take into account two and three-dimensional molecular properties to assess similarity of new compounds in regards to the set of molecules with the biological property under study. Depending on the complexity of the calculation, there are different types of ligand-based methods, such as QSAR (Quantitative Structure-Activity Relationship) with 2D and 3D descriptors, CoMFA (Comparative Molecular Field Analysis) or pharmacophoric approaches. This work provides a description of a series of ligand-based models applied in the prediction of the inhibitory activity of monoamine oxidase (MAO) enzymes. The controlled regulation of the enzymes’ function through the use of MAO inhibitors is used as a treatment in many psychiatric and neurological disorders, such as depression, anxiety, Alzheimer’s and Parkinson’s disease. For this reason, multiple scaffolds, such as substituted coumarins, indolylmethylamine or pyridazine derivatives were synthesized and assayed toward MAO-A and MAO-B inhibition. Our intention is to focus on the description of ligand-based models to provide new insights in the relationship between the MAO inhibitory activity and the molecular structure of the different inhibitors, and further study enzyme selectivity and possible mechanisms of action.

INTRODUCTION

Monoamine oxidases (MAO) are a family of flavin adenine dinucleotide (FAD)-containing enzymes located in the mitochondrial outer membrane [1–3]. MAO enzymes catalyze the oxidative deamination of endogenous and exogenous monoamines to the corresponding aldehyde, hydrogen peroxide and ammonia (from primary amines) or substituted amine (in the case of secondary amines) [1–3]. The enzyme inhibition allows the monoamine neurotransmitters to remain active in the brain for longer periods. For this reason, MAO enzymes play a crucial role in the inactivation and regulation of intracellular levels of monoamine neurotransmitters [1–3].

There have been identified two isoforms of MAO enzymes in humans: MAO-A and MAO-B [4]. Both isoenzymes have 70% identity/similarity in their primary amino acid sequence and they are coded by different genes with similar structure [5, 6]. They differ in cell and tissue distribution, inhibitor sensitivity and specificity in regards to the neurotransmitter type. Serotonin (5-HT), epinephrine and norepinephrine are mostly deaminated by MAO-A. Other neurotransmitters, such as phenylethylamine (PEA) are more specific to MAO-B. However, the tryptamine and dopamine enzymatic breakdown is carried out by both MAO [3, 7]. Due to the diverse inhibitor sensitivity and substrate specificity, both isoenzymes are considered pharmacological targets of different clinical disorders. In fact, MAO-A is of particular importance in the treatment of psychiatric disorders, such as depression and anxiety [8–10], whereas MAO-B is an important target in neurological disorders, such as Alzheimer’s and Parkinson’s disease [11–13].

Since MAO enzymes are implicated in many biological processes of clinical relevance, they are an attractive field in pharmaceutical research [14–20]. This potential clinical importance have led to an intense research to discover new compounds with MAO inhibitory activity without the adverse effects showed by the earliest irreversible inhibitors, such as hepatotoxicity or the ‘cheese effect’ characterized by hypertensive crisis [21, 22]. The disclosure of the crystal structure of the two MAO isoforms provided to the scientific community of structural characterization to elucidate the molecular mechanisms implicated in the ligand-protein interaction [23–26]. This information is very useful in the rational design of new MAO inhibitors. However, ligand-based models are still very effective in the design of new compounds with improved activity (lead discovery and lead optimization), the establishment of structure-activity hypothesis that help to understand the interaction with MAO enzymes, and the explanation of possible mechanisms of action. Ligand-based approaches do not take into account protein structure information and are developed on the concept of molecular similarity (structural similar molecules are likely to have similar biological properties).

There are different types of ligand-based models, such as QSAR with 2D and 3D descriptors (Quantitative Structure-Activity Relationship) [27–30], 3D-CoMFA (Comparative Molecular Field Analysis) [31, 32], 3D-pharmacophores [33] or ligand-network models [34, 35]. QSAR studies have become one of the most popular ligand-based approaches in modern chemistry [27, 36–40]. The main steps implicated in the development of a QSAR model are (see Fig. 1):

Figure 1.

Flowchart showing the Main Steps involved in the Development of a QSAR Model.

1. Collection of a database (molecular structures and biological activity values). The set of molecules used to develop the QSAR equation should be representative of the problem under investigation.

2. Calculation of molecular descriptors. Depending on the nature of the molecular descriptors it is possible to distinguish between 2D and 3D QSAR. Topological [41] or physicochemical descriptors [42], in which the 2D structure of the molecules is taken into account for their calculation, are used in the 2D QSAR. However, the development of QSAR with 3D descriptors, such as topographic [43] or quantum-chemical descriptors [44], implies the calculation of the most stable 3D conformation for the molecules as a previous step.

3. Data analysis to establish the model that relates the descriptors with the biological activity. Some techniques, such as Multiple Linear Regression (MLR) [45], Linear Discriminant Analysis (LDA) [46, 47] or Partial Least Squares (PLS) regression [46] have been widely applied. However, when no linear patterns can be found, the structure-activity relationship can be explained through non-linear methods, such as Artificial Neural Networks (ANN) [48, 49] or Support Vector Machines (SVM) [50].

4. Model validation. Cross-validation series (sub-sampling test), leave-one-out cross-validation (jackknife test) or the evaluation of an independent external prediction series are useful procedures to validate the final model [51].

Once the QSAR model has been developed, biological activity prediction of new molecules and interpretation of the results focused on the molecular mechanism of action could be carried out.

However, since the 3D structure is an important component for the molecular recognition of a ligand by the biological receptors, there have been developed different methodologies to compare the distribution of different molecular properties in the tridimensional space. In 3D QSAR methods, like CoMFA or pharmacophore modeling, the calculation of 3D molecular conformations are complemented by the alignment of the given structures to develop electronic and steric fields (CoMFA) or pharmacophoric chemical features (pharmacophore modeling) that can explain the biological/chemical complementarity with the receptor. These type of models better explain how structurally different ligands can interact with a common receptor. However, the final model construction could be challenging when large structural differences exist in the compounds taken into account [31, 52]. The main steps implicated in the development of CoMFA models are shown in Figure 2.

Figure 2.

Main Steps implicated in the Construction of CoMFA Models.

Other types of models that provide a pharmacological general strategy are based on network analysis. Computational biology network modeling provides a useful tool to infer relationships between ligands and pharmacological targets with application in drug design [34, 35, 53]. As an example, Keiser et al. [35] developed a statistical model that relates protein targets based on the chemical similarity of their ligands. The analysis of ligand-protein networks can provide an understanding of the relevance of some biological targets to improve the efficacy in the process of drug design and discovery.

In this work we reviewed a series of ligand-based models to provide more insights in the relationship between MAO activity and different structural scaffolds. Our intention is to unify different criteria to help in the development of potential agents with MAO activity understanding the reasons of the selectivity of both isoforms. In the following sections we will describe with more details the different types of ligand-based models applied to MAO activity available in the chemical literature. However, it is worth noting that some results should be carefully considered when small datasets were analyzed. It is also important to highlight the differences between distinct chemical scaffolds that can show correlations between different molecular descriptors or pharmacophoric features and the MAO inhibitory activity. Even within the same scaffold, different models can be reported depending on the level of the structural complexity of the compounds belonging to the same chemical group.

PREDICTION OF MAO ACTIVITY THROUGH 2D QSAR

Some studies showed that topological and physicochemical descriptors offer a good potential in the prediction of the MAO inhibitory activity. Different statistical analyses have been used in the development of the QSAR function that relates the molecular structure and the biological activity.

2D QSAR using Multiple Linear Regression (MLR)

Xanthone derivatives

Núñez et al. [54] studied a series of 42 xanthones (see Fig. 3, structure 1) which had been previously reported to be potent MAO-A inhibitors [55]. MAO activities spanning values from 0.04 to 65 μM are reported as the effective concentration of the compound to achieve the 50% of enzyme inhibition (IC₅₀). 36 descriptors were calculated for the set of compounds and the best descriptors were selected according to their predictive ability. The authors related the MAO-A inhibitory activity with 12 descriptors, such as the E-state indices (S_i), molecular connectivity (χ) and shape (k) index. The E-state descriptors [56] were calculated for all the atoms in the molecules providing structural information at atom level. The connectivity index (χ) is defined as the sum of weighted edges in the molecular graph [57, 58]. The shape index is intended to capture aspects of the molecular shape [58]. Multiple linear regression was developed using Statgraphics Plus package [59] to establish the final model obtaining r²=0.847 (squared correlation coefficient) and s=8.069 (standard deviation) taking into account a set of 34 compounds (8 compounds were considered outliers). Cross-validation was carried out using the method LOO (leave-one-out). In the equation established for xanthone derivatives, the connectivity index (χ₅), the shape (k₄) index and the E-states indices in C₂, C₃, C₅, C₉, and R₁ (see Fig. 3, structure 1) presented negative coefficients whereas E-state in O₁₀, C₄, R₄, R₆ and R₇ showed positive coefficients.

Figure 3.

Structure of Some Compounds studied with Ligand-Based Methods.

Pyrrole derivatives

Another example of 2D QSAR in the study of MAO-A activity using topological descriptors is provided by Kumar and Bansal [60]. A database of 32 pyrrole derivatives and analogues was collected from Regina et al. [61] (see Fig. 3, structure 2) and a set of 28 topological descriptors were calculated. A model with a squared correlation coefficient of 0.9 was found through multiple linear regression (MLR) using SPSS software [62]. Seven topological descriptors were selected by stepwise regression to be part of the final model: total structure connectivity index (X_t), mean square distance index (MSD), all-path Wiener index (WAP), eccentric index (DECC), Kier flexibility index (PHI), superpendentic index (SPI) and mean Wiener index (WA) [58, 63–69]. Cross-validation was performed through leave-one out and leave-many out methods. The positive coefficients of the indices DECC, MSD, PHI and SPI shows that an increase of their values produces higher values for the inhibitory constant (K_i). On the other hand, an increase of WA, WAP and X_t (negative coefficients) will decrease the K_i value.

MAO-B activity (K_i values) in a series of 1-methyl-3-phenylpyrroles was related by Ogunrombi et al. [70] with Taft steric parameter (E_s) and Swain-Lupton electronic constant (F) [71] of the substituents at C₄ and C₃ of the phenyl ring. Enhancement in the inhibitory potency was found with electron-withdrawing and bulky substituents. As an example, the potency of the analogue with a trifluoromethyl group substitution at C₄ is 90-fold compared to the unsubstituted analogue. A possible explanation for the electron-withdrawing substituents contribution could be related to the promotion of planarity between the phenyl and pyrrolyl rings. Bibliography supports the idea that planar heterocyclic structures could act as MAO-B inhibitors [72, 73].

Pyridazine and pyrimidine derivatives

Lipophilicity effects were investigated by Altomare et al. [74] in a set of pyridazine and pyrimidine derivatives by measuring partition coefficients, thermodynamic and physicochemical parameters of RP-HPLC retention. The best equation extracted through MLR for a set of 14 pyridazine derivatives (see Fig3, structure 3) yielded an r²=0.821 and q²=0.704 (cross-validation) showing the importance of lipophilic, electronic and steric properties in the explanation of MAO-B inhibition. The results reported that lipophilicity can modulate MAO-B inhibition but it has no effect in the A isoenzyme. The selective influence of lipophilicity to increase MAO-B activity was reported previously [75]. In previous modeling studies using 2D and 3D QSAR, Kneubuhler et al. [76] also related MAO-B inhibitory activity with pyridazines through lipophilic, electronic and steric properties. On the other hand, it has been previously shown that electrostatic interactions and charge-transfer bonding are important parameters in the interaction between inhibitors and the FAD cofactor of the MAO-A [77, 78]. The results described by Altomare et al. [74] revealed that the majority of the condensed pyridazines were selective towards MAO-B whereas the condensed pyrimidines were more active against MAO-A enzyme.

Caffeine analogues

Seven 8-benzyloxycaffeine analogues showed human MAO-A and MAO-B inhibitory activity with enzyme-inhibitor dissociation constant (K_i) values from 0.14 to 1.30 μM and from 0.023 to 0.59 μM respectively [79]. A QSAR model was developed to indicate that the MAO-B activity for this type of compounds could be dependent on the Hansch lipophilicity (π) and Hammet electronic (σ) constants [71] of the substituents at C₃ of the benzyloxy ring. The potency of these analogues is increased with electron-withdrawing substitutions with high degree of lipophilicity. Molecular docking calculations were performed to obtain more insights about the binding modes of the analogues. A similar set of styrylcaffeine analogues was studied by Vlok et al. [80] that established through different QSAR equations that MAO-B activity was dependent on van der Waals (V_W), lipophilicity (π)and the Hammett constant (σ_m) of the substituents at C₃ and C₄ of the phenyl ring of the styryl moiety. An analysis of larger databases could help to confirm these results in this type of analogues.

2D QSAR using Linear Discriminant Analysis (LDA)

Heterogeneous data

A Markov model was used by Santana et al. [81] to calculate a set of topological descriptors to describe a heterogeneous database made up of 1,406 active/non-active compounds. 674 compounds belonging to different representative scaffolds with MAO-A inhibitory activity defined with IC₅₀ values ≤25 μM (some cases were considered showing an inhibition percentage of ≥50% at inhibitor concentrations ≤25 μM) were collected. The non-active compounds included cases belonging to the above mentioned families but with IC₅₀ >25 μM and other scaffolds with different structural patterns. Forward stepwise linear discriminant analysis (LDA) was carried out to derive from the STATISTICA package [82] the predictive equation. The statistical parameters that defined the quality of the developed model were Wilk’s lambda (λ=0.36), the Fisher ratio (F=273.93) and the significance level (p-level<0.01). Nine descriptors providing information about molecular electron delocalization, polarizability, refractivity and n-octanol/water partition coefficients, were found important to describe the MAO-A activity with an overall correct classification of 92.8%. Model validation took into account a resubstitution approach along with the evaluation of a prediction series of 15 novel MAO-A inhibitors (9 compounds with IC₅₀ ≤25 μM and 6 compounds with IC₅₀ >25 μM).

2D QSAR using Partial Least Squares (PLS) regression

Phenyl alkylamines

Norinder et al. [83] analyzed the structural properties for a series of 29 phenyl alkylamines responsible for MAO in vitro and in vivo inhibition (see Fig. 3, structure 4). A set of 56 physicochemical descriptors, such as Hammett constant [84], Swain-Lupton descriptors [85], Hansch aromatic fragment constant [84], molecular refractivity [84] and Verloop Sterimol parameters [86], were reduced to 3–5 significant components to generate different PLS models with squared correlation coefficients (r²) greater than 0.85. The structure-activity relationship examined by the authors concluded that: (S)-stereochemistry and no substitution on the aliphatic chain is important for high in vitro and in vivo activities; electronic descriptors are the most important variables in the developed QSAR; small, electron-withdrawing and hydrophilic substituents in ortho and meta positions increase in vivo activities whereas symmetrical, electron-withdrawing and lipophilic substituents in ortho position favor in vitro activity. In vivo and in vitro potency of the compounds decreases in para positions following the order NHMe>NMe₂>NH₂≫CHMe₂.

A similar set of phenyl alkylamines was studied by Hasegawa et al. [87]. The authors developed different models through nonlinear partial least squares where the in vitro and in vivo MAO inhibitory activities were analyzed. A detailed description of the method has been published previously [88]. The structural descriptors codifying the chemical entities are similar to the descriptors mentioned in a previous study [83]. Structural requirements for MAO activity were estimated according to the PLS loadings that show the contribution of the descriptors to the components of the model. Structural descriptors at ortho positions of the phenyl ring are important for the in vitro activity that is favored by large, electron-withdrawing and hydrophobic substituents. Descriptors in meta positions are no significant whereas electron-donating substitutions favor the activity at para positions. S-stereochemistry in α position is also an important requirement for the in vitro activity. In vivo activity is favored by electron-withdrawing substituents at ortho and electron-donating and large substituents at para positions.

2D binary QSAR

Heterogeneous data

binary QSAR correlates structural descriptors with a binary biological activity measurement (active and inactive) [89]. Gao et al. [90] applied a genetic algorithm method to select the most important molecular descriptors to explain the MAO activity in a heterogeneous series using binary QSAR analysis. The predictive accuracy obtained in this study was 85% and 84% for the training and test set respectively.

PREDICTION OF MAO ACTIVITY THROUGH 3D LIGAND-BASED MODELS

3D ligand-based methods are useful tools to further study structure-activity relationships that imply the use of the 3D conformation for the molecules under study. Different types of studies will be described in the next sections, such as the development of QSAR models with 3D descriptors, 3D QSAR CoMFA (Comparative Molecular Field Analysis) models and 3D pharmacophores. Although CoMFA and pharmacophoric approaches can propose some leads about the nature and shape of the protein binding site, some results should be taken with care since the chemical features that make up the final model are highly dependent on the chemical structure of the studied inhibitors.

QSAR with 3D descriptors

Xanthone derivatives

Deeb et al. [91] included information about nodal orientation to improve the xanthones MAO-A QSAR (see Fig 3, structure 1) previously published by Núñez et al. [54]. They exploited the idea that orientations of the nodes in π-like orbitals of aromatic molecules are crucial in their activity [92, 93]. A set of 10 descriptors involving information about the energy of different π orbitals and nodal angles were introduced in a final model obtained by MLR analysis. The quality model parameters for the set of 42 xanthones were r²=0.825 (squared correlation coefficient), S=0.393 (standard error of estimate) and q²=0.632 (squared correlation coefficient based on the leave-one-out residuals).

The same dataset reported by Nuñez et al. [54] and Deeb et al. [91] including diverse xanthone derivatives was studied by Masand et al. [94]. The structures of the compounds were optimized geometrically as a previous descriptors calculation step. Dragon software [58, 95] was used to calculate all the available 3D descriptors (MoRSE, RDF, WHIM, GATEAWAY, etc). Different 3D QSAR models were developed through stepwise MLR analysis. Ten 3D descriptors were included in the best model with an r²=0.92 and q²(leave-one-out)=0.85.

Phenylisopropylamines

Charge-transfer interactions implicated in the inhibition of MAO-A were analyzed from the point of view of QSAR methodology for a set of 33 phenylisopropylamines (see Fig. 3, structure 4) [96]. All molecules were optimized geometrically and HOMO energies and charges on the aromatic carbons were calculated using AM1 method [97]. Since the initial database can be divided in two different sets with opposite trends, different QSAR models relating pIC₅₀ values [83, 98] with the molecular descriptors were obtained through MLR. For a set of molecules, a QSAR equation was found where E_HOMO coefficient was positive and the charge-descriptors coefficients were negative. The model pointed out an enhancement of the activity for electron-rich phenyl rings in agreement with the fact that electron donation favors charge transfers to the flaving ring. The second QSAR in the other set of molecules showed opposite results and predicted a decrease of the potency with increased electron donation to the phenyl ring. Two different equations were extracted:

$$\begin{array}{l}\mathit{pIC}{50}_{\mathit{MAO}-A}=0.43\times \text{E}\_\text{HOMO}-26.86\times \text{Q}1-5.04\times \text{Q}2-6.00\times \text{Q}3-1.87\times \text{Q}4+6.42\\ N(\text{number}\text{of}\text{data}\text{points})=18;r^2=0.96;r^2(\text{cross}-\text{validation})=0.89;s^2(\text{standard}\text{error}\text{of}\text{estimate})=0.05;F_{\text{value}}=56.89\end{array}$$ (1)

$$\begin{array}{l}\mathit{pIC}{50}_{\mathit{MAO}-A}=-1.51\times \text{E}\_\text{HOMO}+25.39\times \text{Q}1+29.88\times \text{Q}4+9.44\times \text{Q}6-5.35\\ N(\text{number}\text{of}\text{data}\text{points})=15;r^2=0.77;r^2(\text{cross}-\text{validation})=0.63;s^2(\text{standard}\text{error}\text{of}\text{estimate})=0.32;F_{\text{value}}=8.33\end{array}$$ (2)

The main difference in both sets was the presence or absence of an amino group at para position of the phenyl ring. It was hypothesized that the amino substituent could be protonated by hydrogen bond donor interactions with the enzyme that could change the HOMO levels and charges on the ring. The final conclusion suggests that electron-rich rings systems and higher HOMO levels increased the potency of these derivatives against MAO-A enzyme. The results are in agreement with some studies suggesting the existence of charge-transfer interactions between inhibitors and the FAD cofactor of MAO enzymes [77, 99, 100]. A previous study carried out in phenylalkylamines also revealed that electronic descriptors had an important contribution in the development of the QSAR model [83]. Deeb et al. [101] further studied a set of phenylisopropylamines through semiempirical (AM1) and density functional theory (DFT) calculations [102]. They showed that the orientation of nodes and the energies of the occupied π orbitals have a powerful explanatory contribution to the variance in the MAO-A activity. MAO polar nucleophilic mechanisms were also studied by Erdem et al. [103] using quantum chemical calculations.

Hydrazides

QSAR models in a set of aryloxyacetohydrazides were developed by Hall et al. [104]. Comparison of the models developed with molecular orbital parameters computed through the AM1 Hamiltonian and with electrotopological state (E-state) indexes was carried out. The model with 2D descriptors (E-state) showed in this case a better performance (r2=0.90; s(standard deviation)=0.19) than the model with 3D descriptors (r²=0.80; s(standard deviation)=0.27), supporting the idea that topological descriptors are also suitable representations.

Heterogeneous data

1,650 MAO inhibitors following different mechanisms of action were analyzed using 3D descriptors and Recursive Partitioning (RP) techniques [105] by Chen et al. [106]. Single low energy conformation and multiple conformation databases were generated with the help of CONCORD [107] and SYBYL software [108]. 3D atom pair descriptors were calculated containing information about atom types or the center of some chemical feature and the Euclidean distance between them. 17 atom types were defined, such as the explicit atoms carbon, nitrogen, oxygen, etc., or pharmacophoric chemical features, such as positive and negative charge center, aromatic ring center, hydrogen-bond acceptor and donor, etc. The descriptors were incorporated to a binary file that contained a bit string indicating the presence (value 1) or absence (value 0) of the 3D atom pair descriptors in any of the conformations generated for the compounds. The MAO activity was codified where 0 indicates no activity and 1, 2, and 3 values indicate increasing MAO activities. All the RP analyses were carried out with the help of SCAM software [109]. SAR trees derived from the single conformation and multiple conformation databases were developed sequentially splitting the data into two subsets according to whether the atom pair type is in the compound or not. This approach provides a strategy to generate 3D pharmacophoric models in heterogeneous and large datasets by grouping the active compounds with similar key structural features. Similar approaches were carried out using different types of 2D molecular descriptors [110–112].

3D QSAR CoMFA models

Xanthone derivatives

Gnerre et al. [55] studied a set of 59 natural and synthetic xanthones as MAO inhibitors using 3D-QSAR (see Fig. 3, structure 1). The compounds showed selectivity towards MAO-A, the most active inhibitor having a potency of IC₅₀=40 nM. Although the molecular mechanism by which these molecules present MAO-A activity is not completely understood, a possible hypothesis takes into account charge-transfer interactions with the FAD cofactor [77]. The initial SAR pointed out the importance of -OH substituent at position 1 or 5 rather than -OMe. However, the opposite effect is shown in position 3, where -OMe substituents yield more active MAO-A compounds. The authors further studied the structure-activity relationship through a combination of techniques, involving CoMFA studies and ALMOND procedure [113].

In order to develop the CoMFA model, a training set was selected. The training set was made up of 34 xanthones with three types of substituents (-H, -OH, -OMe) in some of the 8 positions available in the scaffold (Fig. 3, structure 1). The superposition of the molecules was made taking into account the xanthone ring system. The calculated molecular fields in CoMFA were exported to GOLPE [114]. A principal component analysis (PCA) [115] was applied to reduce the number of components. The final PLS model yielded an r² of 0.87 (squared correlation coefficient) and q² of 0.74 (cross-validated squared correlation coefficient). The best model was obtained with steric and electrostatic parameters. Lipophilicity was not an important variable in the activity explanation. Electrostatic analyses showed two important zones: a favorable influence of high electron density between positions 4 and 5, and a negative influence of high electron density at position 7, such as -OMe and -OH substituent groups. CoMFA graphics also revealed a favorable steric area near the position 3 and unfavorable steric areas in position 5 and 7 (see Fig. 3, structure 1). As an example, -OMe substituent is positive at C₃ but its contribution is unfavorable for MAO-A activity at C₅ and C₇.

Since the alignment is a critical step that can condition the CoMFA results, ALMOND procedure [113] was carried out to complete the evaluation. Molecular Interaction Fields (MIF) generated with GRID program [116] were transformed into the alignment independent GRid INdependent Descriptors (GRIND) and analyzed by means of ALMOND software. Three probes, such as methyl, carbonyl and amide, were used to extract H-bond acceptor/donor and steric interaction regions that define the virtual receptor site. The final model was constructed with four latent variables and yielded an r²=0.86 and q²=0.66. This study showed that NN interaction energies have a positive effect in the MAO-A activity in short distances (9 Å) and negative effect at longer distances (14.5 Å). A similar effect was found for CN and ON interaction energies. However, CC interaction energies at short distances resulted unfavorable for the molecular activity.

Indole and isatin analogues

A series of indole and isatin analogues has been studied by Medvedev et al. [117] as MAO-A and MAO-B inhibitors (see Fig. 4, structure 5). This study pointed out that selective MAO-A or B inhibitors presented different molecular size [117, 118]. It also described that MAO-A inhibition required presence of hydroxy substituent at C₅ of the isatin and coplanar structure of the substitutions at C₂ and C₃ in the indole. Moreover, MAO-B inhibition is related to the electron density distribution. In a posterior article, Medvedev et al. have further studied the quantitative structure-activity relationship in the set of indole and isatin analogues (Fig. 4, structure 5) through Comparative Molecular Field Analysis (CoMFA) [119]. CoMFA analysis was developed using SYBYL 6.1 software [108]. Conformations with the lowest energy were calculated as a previous step of the molecular alignment. Both steric and electrostatic fields were taken into account in the CoMFA analysis. Optimal number of components was determined though PLS. Cross-validation r² values were 0.743 for MAO-A and 0.603 for MAO-B. The pharmacophore model included four features: two hydrophobic (aromatic rings), one donor atom (nitrogen in the pyrrole) and one acceptor atom (substituent in the nitrogen of the pyrrole). This data is consistent with previous publications where aromatic rings were important elements in the MAO inhibitors structure [120, 121]. According to the QSAR model for MAO-A inhibitors, steric fields near the aromatic substituents in C₂ of the indole could increase the activity (Fig. 4, structure 5). Six unfavorable regions are close to C₅, C₆ and surrounding the favorable region. Negative charges areas that could increase the activity are located in the aromatic substituents. There are also some positive charges areas surrounding the molecules. In the case of MAO-B inhibitors, there are four favorable steric regions near C₅ in the indole and surrounding substituents at C₂. Negative charges favorable areas are near C₄-C₇ of the indole and close to the aromatic substituents at C₂. Positive charges that would increase the MAO-B activity are located close to the indole nitrogen and at para phenyl substituents at C₂. Although the comparison between steric and electrostatic areas in both MAO-A and MAO-B inhibitors showed some common regions, the analysis also revealed different patterns in the favorable and unfavorable steric and electrostatic regions. These differences could explain the different behavior of both enzymes in inhibitor selectivity.

Figure 4.

Structure of Some Compounds studied with CoMFA Analysis.

Pirlindole analogues

Inhibitory activity (IC₅₀) for pirlindole analogues with different substituents at C₈ was studied by Medvedev et al. [122] through CoMFA analysis (see Fig. 4, structure 6). The molecules that formed part of the study were optimized geometrically and aligned by fitting the indole heterocyclic ring atom by atom. Steric and electrostatic fields were generated using a sp3 carbon probe and a +1 charge. The optimal number of components were calculated in the PLS analysis to develop the final model. The models yielded an r²_cv (squared of the correlation coefficient with cross-validation) of 0.444 and 0.525 for MAO-A and MAO-B respectively. However, the combination in the case of MAO-B with logP yielded a model with r²_cv=0.625. The fact that the model with MAO-B took into consideration hidrophobicity is in accordance with previous publications where MAO-B inhibitors were considered more hydrophobic [123, 124]. Analysis of the molecular size showed that the rigid pirlindole analogues and with certain size limits (X, Y, Z; 13.0×7.0×4.4 Å) were more potent against MAO-A. However, the flexible analogues with independence of the size showed good potency in both MAO. The comparison of the location of favorable/unfavorable steric and electrostatic regions using diverse substituents at C₈ showed a different profile between MAO-A and MAO-B providing new evidences of the diversity of the active sites of both enzymes. A steric obstacle at C₈ of the inhibitors in MAO-B explains the inactivity of long rigid pirlindole analogues. The location of favorable areas of positive charges in MAO-B is coincident with the placement of favorable negative charges in MAO-A.

Indolylmethylamine derivatives

A set of indolylmethylamine derivatives were studied by Morón et al. [125] (see Fig. 4, structure 7). K_i values spanned a range from 0.8 to >10⁶ nM in the case of MAO-A and a range from 0.75 to 476,000 nM in MAO-B. Full geometry optimization was carried out for the compounds using AM1 method [126]. Molecular alignment was made by superimposing the heavy atoms of the indole ring. CoMFA was developed using default parameters of the SYBYL software [108]. Cross-validated squared correlation coefficients (q²) were 0.895 and 0.859 for MAO-A and MAO-B respectively. Contributions of steric, electrostatic and solvation terms are similar in both enzyme models. Electrostatic maps in the side chain moiety are very similar in both systems. However, at C₅ position of the indole, high electron density substituents could increase the binding affinity in MAO-A, while it could decrease the activity in MAO-B. Electrostatic map in MAO-B also showed a high electron density area that extends throughout the indole ring and explains the MAO-B selectivity for some compounds. Analysis of the steric fields showed that bulky substituents at N₁ position are not tolerated in either isoforms. Favorable steric regions are present near the unsaturated moiety in both MAO, although is more extensive in MAO-A. This area is surrounded in both maps by unfavorable steric areas. Nevertheless, the bigger difference in the steric maps is located in the substituents at C₅ of the indole ring. Bulky substituents at this position are beneficial in the interaction with the MAO-B receptor. MAO-A steric maps do not present favorable areas at this position. CoMFA analysis showed that differences in steric and electrostatic fields at the 5 position of the indole ring could be crucial for enzyme recognition. Mutagenesis studies identified some enzyme residues responsible for MAO selectivity. More specifically, Phe-208 is important in MAO-A and the corresponding Ile-199 is a key amino acid in MAO-B substrate selectivity [127]. Computational simulations were carried out in the same study to further investigate the possible aromatic interactions between substituents at C₅ and Phe-208 of MAO-A and possible hydrophobic van der Waals interactions between inhibitors and MAO-B (Ile-199) [127].

Phenethylamine derivatives

CoMFA analysis was developed to study the MAO inhibitory activity (IC₅₀) in a series of 38 phenethylamine derivatives (see Fig. 4, structure 8) [128]. This scaffold is present in many catecholamine neurotransmitters and small variations in the structure can yield different biogenic amine target inhibitors [98, 129–131]. The best CoMFA model was obtained with four components and yielded an r²=0.92 and q²=0.72. The analysis showed that in this type of inhibitors the steric properties of the substituents play a more important role than electrostatic properties.

The inhibitory potency was dependent on the length of the chain attached to the sulfur at position 4 of the phenyl ring [98]. Long chain or branched substituents could decrease the activity. The compounds with substitutions at 2,4,5-positions of the phenyl ring had generally lower potencies [98, 132]. Electronegative groups at position 6 of the phenyl inhance the inhibitory activity [133]. The authors discussed with more details the structural insights that differentiate the interaction between this type of inhibitors and different biogenic amine target proteins. The possible interactions with the active site enzyme were also analyzed through molecular modeling of the crystal structure of the inhibitor clorgyline bound to MAO-A (1O5W) [133].

Coumarin derivatives

A series of 3-, 4-, 7-polysubstituted coumarins and their MAO inhibitory potency determined on rat brain mitochondria have been studied by Catto et al. (see Fig. 4, structure 9) [134]. 3D-QSAR CoMFA-GOLPE studies revealed the key physicochemical nature responsible for the interactions with both isoenzymes. Molecular alignment was carried out following the same criteria than a previous publication relating coumarins activity with CoMFA [73]. Different interaction fields were calculated with SYBYL [108] and CLIP software [75], including steric, electrostatic and lipophilic fields. The PLS analysis were carried out in GOLPE [114]. The results of the different CoMFA analysis for both MAO are shown in Table 1. According to CoMFA, both MAO activities are modulated by steric, electrostatic and lipophilic fields. In a previous analysis, only electrostatic and lipophilic interactions were determined important to model the MAO activity [73]. However, the new CoMFA models provide complementary information since the different results can be due to a diverse coumarin structures set used to develop the analysis.

Table 1.

CoMFA-GOLPE Statistical Parameters for 38 Coumarin Derivatives studied by Catto et al. [134].

Fields	Squared correlation coefficient (r2)	Leave-one-out squared cross-validated correlation coefficient (q2)
MAO-A
Ste	0.819	0.727
Ele	0.811	0.673
Lipo	0.802	0.708
Ele+Ste	0.859	0.751
Lipo+Ste	0.823	0.738
Ele+Lipo	0.819	0.659
Ste+Ele+Lipo	0.881	0.789
MAO-B
Ste	0.821	0.692
Ele	0.885	0.805
Lipo	0.785	0.674
Ele+Ste	0.899	0.830
Lipo+Ste	0.787	0.677
Ele+Lipo	0.895	0.834
Ste+Ele+Lipo	0.906	0.837
MAO-B/MAO-A selectivity
Ste	0.867	0.751
Ele	0.941	0.897
Lipo	0.881	0.829
Ele+Ste	0.932	0.883
Lipo+Ste	0.834	0.832
Ele+Lipo	0.944	0.908
Ste+Ele+Lipo	0.926	0.889

Halogen atoms at meta position of the 7-benzyloxy group favor the MAO activity. In fact, there are favorable regions at this position in steric, electrostatic and lipophilic maps. Sulfonate groups in R₇ (see Fig. 4, structure 9) have negative effects in MAO-B activity. This fact was also pointed out by the three isocontour maps. Increased MAO-B activity in derivatives with monomethyl and dimethyl substitutions at positions 3 and 4 was corroborated by favor areas in the steric contour. Sterically unfavored areas are located close to phenyl rings in R₇ of some long inhibitors with low MAO-A affinity. A more detailed description of the CoMFA contour maps is given by the authors [134].

Besides classical CoMFA for MAO-A and B, an additional 3D-QSAR taking into account the difference between pIC₅₀ in MAO-B and MAO-A was developed to study the isoenzyme selectivity. Electrostatic fields played an important role in the enzyme selectivity whereas lipophilic and steric fields were not so important to explain the selectivity in this kind of inhibitors (see Table 1). The different electron density localized on the α and β positions of the bridge that links the coumarin core with a phenyl ring was found important for MAO selectivity. Molecular docking experiments were carried out by the authors to further study the interactions between coumarin derivatives and MAO enzymes.

Hydrazothiazoles

A training set of 36 hydrazothiazoles along with MAO activity data (IC₅₀) was used for CoMFA model generation (see Fig. 5, structure 10) [135]. After the calculation of the 3D conformation of the molecules, a structure alignment was carried out and steric and electrostatic fields were computed. PLS was derived using cross-validation leave-one-out (LOO) method. Different CoMFA models were presented using diverse alignment approaches (substructure, pharmacophore and docking receptor-based alignment). However, the best results are obtained when the compounds are aligned through molecular docking (see Table 2). The predictive capability of the generated models was assessed with good results in a test set.

Figure 5.

Some Scaffolds studied with CoMFA.

Table 2.

Comparative CoMFA Results for Hydrazothiazoles studied by Chimenti et al. [135] using Different Molecular Alignment Approaches (r² is the Squared Correlation Coefficient and q² is the Leave-One-Out Squared Cross-Validated Correlation Coefficient).

Model (different alignment)	r2	q2	Number of components
MAO-A
Substructure-based alignment	0.752	0.449	5
Pharmacophore-based alignment	0.768	0.352	4
Receptor-based alignment (docking)	0.961	0.811	4
MAO-B
Substructure-based alignment	0.678	0.356	4
Pharmacophore-based alignment	0.790	0.444	4
Receptor-based alignment (docking)	0.948	0.831	4

The contour maps for MAO-B suggested that an increase in steric bulk around the alicyclic portion and ortho positions of the phenyl ring enhanced the activity. However, there is also a steric unfavorable area surrounding the phenyl ring. As an example, substitution of fluorine in the phenyl ring by bulkier methoxy groups resulted in a decrease of the activity. Electrostatic fields showed an area of desirable negative electrostatic interactions around the phenyl ring para position. A similar region is placed in correspondence of the hydrazo group, while a desirable positive electrostatic interactions area is surrounding the NH-position of this moiety. Possible hydrogen-bond interactions could be established between the protein and the inhibitors in this region. The contour maps for MAO-A revealed the importance of the hydrazo group for MAO recognition. Unlike MAO-B, substitution with steric bulky groups around the alicyclic portion could decrease the MAO-A activity. Appropiate modifications of the scaffold in this area could lead towards selective MAO inhibitors.

Pyrazole derivatives

A series of 72 pyrazole derivatives [136–138], were studied according to 3D CoMFA models (see Fig. 5, structure 11) [139]. Molecular alignment was carried out through docking calculations against the crystallographic structures of hMAO-A (2BXR) and hMAO-B (1GOS) [23, 24, 26]. PLS analysis of the steric and electrostatic fields identified the best configuration alignment selecting the best enantiomer for each compound. Different CoMFA results were extracted for MAO-A (r² = 0.978; q²= 0.671; PLS components=10) and MAO-B (r² = 0.912; q²= 0.555; PLS components=8). The contour maps for MAO-A were analyzed using a compound with different substituents as an example (R=COCH₃, R′=2-OH, R″=4-Cl-C₆H₅). The detected pharmacophoric regions are: a hydrogen-bond acceptor region close to the diazo moiety (N₂ and the oxygen of the acetyl group in N₁, see structure 11) showing a possible interaction with Tyr444 and Ser209; a hydrophobic region near the aromatic ring R″ (possible π-π stacking interaction with aromatic residues such as Tyr407 and Tyr444); an electrostatic region close the 2-OH of the 3-phenyl ring (possible Tyr69 hydroxyl group interaction). Another compound was used to show graphically the MAO-B contour maps (R=COCH₃, R′=2-OH, R″=3-CH₃-C₆H₅). The main pharmacophoric areas for MAO-B are: two hydrogen-bond acceptors located close to the –OH in the 3-phenyl ring and near N₂ and the oxygen of the acetyl group attached to N₁ (possible hydrogen bond interactions with Cys172, Tyr435 and Gln206); a hydrophobic region near the 3-methyl group in R″ (possible interaction with aromatic residues Tyr60, Phe343 and Tyr398). The results were consistent with molecular docking experiments carried out by the authors [139, 140].

Heterogeneous data

A database of 130 reversible MAO-B inhibitors, (see Fig. 5, structure 12) including tetrazole, oxadiazolone, oxadiazinone derivatives and N-acylhydrazones (all of them contained the diazo N-N moiety), were studied by Carrieri et al. [141] using CoMFA and GOLPE procedures. The results after calculating steric, electrostatic and lipophilic fields are provided in Table 3. The steric and lipophilic interactions are more important than electrostatic in the explanation of the potency for the set of compounds taken into account. The analysis of the contour maps showed that hydrophobic interactions of the benzyloxy moiety could increase the activity. On the other hand, unfavorable hydrophobic interactions are close to the diazoheterocyclic moiety. The length of the lateral chain in the diazo N-N moiety is also important for the activity pointed out by unfavorable steric areas. Molecular docking calculations carried out by the authors agreed with the 3D-QSAR results. CoMFA also provided support to the SAR published by Wouters et al. [142]. The influence of lipophilicity to model the MAO-B activity was also reported in previous studies [73, 76].

Table 3.

CoMFA-GOLPE Statistical Parameters for 130 MAO-B Inhibitors studied by Carrieri et al. [141].

Fields	Squared correlation coefficient (r2)	Leave-one-out squared cross-validated correlation coefficient (q2)
MAO-B
Ste	0.79	0.72
Ele	0.39	0.30
Lipo	0.74	0.67
Ele+Ste	0.79	0.72
Lipo+Ste	0.79	0.73
Ele+Lipo	0.74	0.68
Ste+Ele+Lipo	0.80	0.73

3D Pharmacophoric models

Thiazole and thiosemicarbazide derivatives

A pharmacophore generation approach was developed by Gritsch [143] using the Catalyst 4.5 software package [144] to study the MAO-B inhibitory activity (IC₅₀) in a set of 100 thiazole and thiosemicarbazide derivatives, both having in common a diazo N-N pharmacophoric substructure (see Fig. 6, structures 13 and 14). Once the structures of the different compounds were built, a minimum energy conformational model was calculated for the molecules using the default parameters. Pharmacophore selection was made analyzing different features: H-bond acceptor, H-bond donor, hydrophobic substructures, positive ionizable (atoms that can be protonated), and ring aromatic. Different pharmacophoric models were developed looking for common features for thiazole derivatives, common features for thiosemicarbazides, and features existing for both scaffolds together. The comparison of the generated pharmacophores showed that four features model with two or three H-bond acceptors and one hydrophobic and/or aromatic ring could be representative to explain the interaction for both thiosemicarbazide and thiazole scaffolds. The above mentioned models were used to search for molecules in a 3D database (Dement World Drug index database) that fit all the features of the query established in the developed pharmacophores. These findings are useful in the design of new MAO-B inhibitors.

Figure 6.

Compounds used in the Development of Pharmacophoric Models.

Coumarin analogues

A Hypogen three-dimensional pharmacophore for a set of 64 coumarin analogues (see Fig. 6, structure 15) using CATALYST software [144] was described by Sairam et al. [145]. The compounds showed MAO-A inhibitory activity with IC₅₀ values from 2.0×10⁻⁸ to 1.0×10⁻⁴ M. The HypoGen algorithm allows the establishment of hypothesis using structure and activity data to generate chemical features (hydrophobic, hydrogen-bond donor, hydrogen-bond acceptor, positive and negative ionizable sites) that are present in the active molecules and not present in the inactive ones. The best obtained model consisted of five features: two hydrogen-bond acceptors and three hydrophobic groups. The correlation coefficient between the experimental and predicted activity was 0.95.

Tricyclic derivatives

In order to establish the structural requirements of a series of tricyclic derivatives as MAO-A inhibitors (see Fig. 7, structure 16), Suryawanshi et al. [146] developed ligan-based pharmacophores with four chemical features consisted of three hydrogen-bond acceptors and one aromatic ring (see Fig. 8). The dataset of 65 compounds previously published [121, 147, 148] was divided in training set (52 compounds were used to generate the pharmacophoric model) and test set (13 compounds). Pharmacophore modeling was developed with the help of Maestro software [149] through Phase module [150] that combines conformational sampling with different scoring methods to identify pharmacophoric hypothesis. Atom-based 3D QSAR models were constructed for all pharmacophoric hypothesis generated previously. Van der Waals models of the aligned molecules were placed in a grid that lead to binary-valued occupation variables used in the QSAR models. The best model yielded a good predictability (r²=0.9595; q²=0.6229).

Figure 7.

Structure of Some Compounds studied with Pharmacophoric Models.

Figure 8.

Pharmacophore Model developed by Suryawanshi et al. [146] for Tricyclic Derivatives and Distance (Å) between the Chemical Features (3 Hydrogen-Bond Acceptors and 1 Aromatic Ring).

The study highlighted some binding features of tricyclic derivatives with MAO-A receptor, which can be useful in the guidance of rational design of new inhibitors. Substitutions with aliphatic chains at C₇ aromatic (see Fig. 7, structure 16B) are favorable for the activity. The presence of N-methyl amide group at C₃ aromatic is crucial for the activity (structure 16B). The presence of polar substituents at C₆ in the structure 16A can decrease the activity. Anilides at C₃ favors the activity whereas substitutions at C₂ decrease the binding with the receptor (structure 16A).

Indole and pyrrole derivatives

A similar approach was described by Shelke et al. [151] that combined ligand-based pharmacophore models with atom-based 3D QSAR analysis to study a series of 82 indole and pyrrole derivatives as MAO-A inhibitors (see Fig. 7, structure 17). A database was collected from previous publications along with pK_i values [61, 152–154] and divided in raining (67 molecules) and test set (15 compounds). The necessary steps to develop the pharmacophore models, including the conformation analysis, alignment and pharmacophoric hypothesis generation, were carried out with the help of Phase module [150]. The best pharmacophore was developed with four chemical features: a hydrogen-bond donor, two hydrophobic groups and an aromatic ring. The different features along with the distances are represented in Fig. 9. Information about the interaction with the MAO-A receptor is analyzed though molecular docking (see Fig. 9). The top ranking pharmacophores were subjected to atom-based 3D QSAR analysis. The best statistically significant model yielded good predictability (r²=0.979; q²=0.699). The obtained pharmacophoric-QSAR models were used to explore potential novel scaffolds in the ZINC database [155, 156].

Figure 9.

Pharmacophore Model generated by Shelke et al. [151] for Pyrrole Derivatives: a Hydrogen-Bond Donor, two Hydrophobic Groups and an Aromatic Ring. Interactions between Pharmacophoric Features and the Active Site of MAO-A are represented.

PREDICTION OF MAO ACTIVITY THROUGH MULTI-QSAR AND NETWORK MODELS

Santana et al. [53] combined complex network approaches with multi-QSAR methods in a unified model to predict MAO activity on a heterogeneous and large database that includes propargyl derivatives, benzamides, indoles, coumarins, and thioxanthenes among others. The analyzed dataset included in vitro and ex vivo pharmacological activity measured in cellular lines from different organisms. Molecular descriptors were calculated using the MARCH-INSIDE approach [157]. Activity assay conditions were also taken into account through dummy variables assigning two values (1 and 0) to define the presence or absence of a certain condition. Each case in the study is defined by 159 variables (150 descriptors and 9 assay conditions). Reduction of the dimensionality to 10 factors was performed by principal component analysis (PCA). Stepwise linear discriminant analysis (LDA) led the establishment of an equation to explain MAO-A and MAO-B activity with three variables (factors) with Rc=0.79 (canonical regression coefficient), λ=0.38 (Wilk’s lambda) and p-level<0.001 (significance level). The model also classified correctly 94.5% of the compounds in the database (3,222 out of 3,408 compounds were correctly classified). PCA scores for each case weighted with the QSAR coefficients were used to generate a matrix containing similarity/dissimilarity information between the different compounds in different pharmacological assays. A Boolean matrix with 0 and 1 values was calculated establishing a threshold. This matrix was represented through CentiBiN [158] in a complex network. The MAO inhibitors network topology is compared to the free scale network topology. On the basis of the developed model, different coumarins have been synthesized and evaluated against MAO-A and B with excellent results.

Multiple drug-protein interactions were investigated through multi-target QSAR models in a recent publication [159]. 2D molecular descriptors in DRAGON software [58, 95] and 3D structural parameters in MARCH-INSIDE [157] were calculated and introduced as an input of different Artificial Neural Networks (ANN) [48]. The best non-linear multi QSAR model was generated through Multi-Layer Perceptron (MLP) obtaining a sensitivity of 89% and specificity of 94%. The model led to the reconstruction of a large drug-target complex network. Different numerical parameters quantify the relevance of the nodes (drugs or proteins) in the graph and detect drugs that bind different proteins along with the most important therapeutic targets. The prediction and the experimental MAO-A and MAO-B inhibitory activity was reported for a series of 10 oxoisoaporphines.

Multi-target QSAR methods were also used by Molina et al. [19] to predict MAO-A inhibitory activity. A database of 2,246 heterogeneous compounds (1,725 in the training set and 521 in the cross-validation set) were studied in different species (Bos taurus, Mus musculus, Rattus norvegicus and Homo sapiens). Descriptors from spectral moments of the bond adjacency matrix [160] and atom-centered fragment and functional group count descriptors were calculated [58, 95]. Forward stepwise linear discriminant analysis (LDA) was performed to derive the multi-species MAO-A QSAR model. The final model was used to study the MAO-A activity of new oxoisoaporphine derivatives. Multi-QSAR techniques could be useful in the prediction of molecular activities with great application to study possible mechanisms of action.

CONCLUSION

Different types of ligand-based models, with diverse levels of complexity and taking into account a high variability in the structural scaffold, were applied in the literature to study the relationship between MAO activity and molecular structure. Ligand-based approaches provided significant structure-activity relationship (SAR) information useful in the rational design of new MAO inhibitors. Ligand selectivity showed by both isoenzymes could also be explained through this type of models. Ligand-based methods can be combined with protein-structure models to further study the interactions between ligands and MAO enzymes in a more realistic way.