Advanced in silico Approaches in Antiviral Research

Structure-based approaches

Molecular docking is the most commonly used computational technique when the 3D structure of the desired target is available. This methodology tries to predict the geometry of the complex of two molecular entities (usually a small molecule and its target) starting from their individual structures. This is a fast technique and it allows the screening in silico of large compound libraries in a relatively short time. The use of such virtual screening (VS) approaches is increasing constantly, often associated with pharmacophore filtering and other refinement methodologies. Several examples from the antiviral field on the use of VS are available in the literature and they cover a variety of protocols. A few recent examples include screening of the freely available ZINC library against the H5N1 haemagglutinin [10]; screening NCI libraries against the dengue virus E protein [11]; and ensemble-based VS on avian flu neuraminidase [12].

Another notable example is represented in the work reported by Pérez-Nueno et al. [13] on the HIV entry inhibitors. In this study, different VS approaches were compared and it should be noted that the structure of the cellular receptors targeted by this class of inhibitor has not been solved to date. However, the authors were able to build a model for these receptors in silico, by homology modelling using bovine rhodopsin as structural template.

An interesting evolution of traditional VS techniques is the so called ‘steered’ VS. This is an attempt to take into account information about the protein and the ligands in the simulation. For example, the software ProPose [14] implements a general method for performing simultaneous protein–ligand docking, ligand–ligand alignment and pharmacophore queries with the aim to incorporate all a priori knowledge into screening protocol. With this particular procedure of steered VS it is possible to take into account the flexibility of the receptor, using more than one 3D structure, and the ligand data, using different alignment rules and pharmacophore queries, thus improving the enrichment factors and overcoming the problem of weak-binding or non-binding false-positive ligands. The application of steered VS to herpes simplex virus thymidine kinase with the possibility to consider chemical, biological and structural factors in the early steps of VS, has given the opportunity to identify a much better starting point for lead discovery and optimization [15].

Another exciting field of application of VS might be found in the inhibition of protein–protein interactions [16]. Protein–protein binding involves the interaction of a small number of residues on the exposed surface of both binding partners, called hot-spots, the identification of which can be obtained by alanine scanning or other mutagenesis studies. These kinds of interactions are at the base of most cellular pathways and enzymatic control mechanisms, and are involved in virtually every physiological and pathological process. By interacting with known hot-spots on a protein surface, a small molecule could inhabit the protein–protein interactions by preventing the binding of the two protein partners. Therefore, protein–protein interaction inhibition has a great therapeutic potential. This method is producing significant results in anti-cancer research [17,18], and its application to the antiviral area could achieve the same important results. In this regard, the work of Betzi et al. [19] on the SH3 binding surface of the HIV type-1 Nef protein is interesting. In that paper, a virtual screening on 1,420 compounds filtered by a docking procedure and then by a pharmacophore model built using the hot-spot knowledge is reported [19]. Remarkably, all the screened compounds were also evaluated by a biological assay, and the selected leads from VS and highthroughput screening were the same. The results of the use of the VS method in the identification of the small molecules as protein–protein interaction inhibitors led to the discovery of the first two compounds that bind the HIV type-1 Nef SH3 in the micromolar range.

Protein–protein interactions were also at the core of the work reported by De Luca et al. [20]. However, in this case, the interaction targeted was one between a viral protein (HIV integrase) and a cellular cofactor (LEDGF/p75). Interestingly, the VS simulation was performed using a structure-based pharmacophore.

De novo drug design is another interesting structure-based methodology. This approach can be described as the process of producing novel molecular structures with desired pharmacological properties, using the structural information of the biological target [21]. It is a fascinating field, but its use has been somehow restricted by some inherent limitations of the technique, in particular the inability of computers to accurately assess the chemical feasibility of the proposed structures. For this reason, an extensive user intervention in the design process is often required, as is noted in the discovery of novel HCV helicase inhibitors that we have recently reported [22].

It is now common practice to combine different in silico approaches instead of relying on a single technique to generate more accurate results. Indeed, strong positive results have been obtained where a combination of different experimental and in silico techniques were used in conjunction. The discovery of HIV protease inhibitors is an excellent example of this combined approach: crystallography, molecular docking and de novo simulations allowed a very rapid development of a class of compounds that are now extremely important in the HIV therapy [23–25].

Molecular dynamics (MD) simulations are a further example of structure-based approaches. The aim of MD is the prediction of the dynamic evolution of a molecular system over time. This is of key importance in the study of processes that involve conformational changes or in the evaluation of molecular binding interactions and energies. This is a computationally expensive technique, but constantly improving computer speed gives the possibility to use more time-consuming approaches such as meta- and quantum mechanics molecular dynamics (QM-MD) on ligand–receptor complexes and also on ternary complexes (for example ligand receptor nucleic acids or ligand receptor cofactors) [26]. These MD methods provide the opportunity to analyse conformational modifications over time and the interactions between atoms and the stability of the targeted complex (ΔG and K_eq calculation). MD simulations find application in the elucidation of several aspects of drug activity. For instance, MD can be helpful in understanding, from a structural point of view, the reasons why the mutation of a particular residue in the binding site of a targeted receptor gives resistance to an inhibitor. Furthermore, QM-MD could be useful to evaluate the polarization energy of all the residues of the binding site of a targeted inhibitor–enzyme complex. This, in turn, means the possibility to know, with a high rate of accuracy, the contributions of individual binding site residues to the enzyme–inhibitor interactions. A recent paper reports the application of a QM-MD analysis on three high affinity HIV protease inhibitors (nelfinavir, mozenavir and tipranavir) [27]. Hensen et al. [27] showed, by using electrostatic energy analysis on the inhibitors and on the binding site residues, that the two specific amino acids (Asp25 and Ile50) of both polypeptide chains play a crucial role in the inhibitor binding and that the 4-hydroxy-dyhydropyrone of tipranavir gives the strongest polarization effects as a result of a stronger binding force of this moiety. It is easy to imagine the importance of this kind of information in the design process of novel biologically active molecules. Although QM-MD analysis and the computation of the polarization energy can not be applied to a large number of compounds, it could be very useful, (for example, after a VS, when the number of scaffolds is limited) to identify the core moiety with the higher probability to become a lead.

Ligand-based approaches

Quantitative structure–activity relationship (QSAR) predictions are possibly the most significant example of ligand-based drug design. This approach involves establishing a mathematical function that correlates biological data (such as 50% inhibitory concentration values) of a series of compounds with the properties of those molecules (descriptors). More than 50 years ago, Corwin Hansch, created the first QSAR model quantifying a relationship between three physical–chemical properties and the biological activities of a molecular set and, since then, QSAR calculations have been applied to a wide range of projects [28]. The number and identity of the descriptors used in QSAR can be very different and have varied significantly over the years. The understanding of the important role played by the spatial geometry of the compounds and the adoption of 3D descriptors was the key to the development of the more advanced 3D QSAR, which allows the analysis of both ligand binding conformations and directional forces (for example, hydrogen bonds). Although 3D QSAR has scored several notable successes and many exciting discoveries that have been obtained by it exist, the determination of the correct bioactive conformation, the protonation state and the alignment rules are not free of bias [29]. Four-dimensional and five-dimensional QSAR methods are a recent development, and could offer a convincing answer to these limitations: four-dimensional QSAR models include a sampling of the different conformations accessible to a compound, as well as its protonation states, whereas five-dimensional QSAR allows the model to include induced fit effects on the receptor as well. The application of these new QSAR methods can offer a new lead optimization strategy and some interesting results in antiviral research have been reported. Their application to HIV reverse transcriptase [30,31] and protease [32] inhibitors provided information that was not obtainable by other methods, confirming the crucial role of these tools and of CADD methods in general as a powerful way to identify a lead compound in a quicker and more cost-effective way than physically synthesizing and testing it.

Alignment and superposition of small molecules is also at the core of another ligand-based set of methodologies based on the concept of shape recognition [33]. The rationale is that molecules that interact with the same target might have a certain degree of shape similarity. Furthermore, in addition to the simple structure superposition, a series of molecular descriptors, such as cLogP or pKa, can be considered, generating a rich set of information that can be used for drug design. Some examples of applications of these approaches in the antiviral field available in the literature include a work on HIV type-1 non-nucleoside reverse transcriptase inhibitors [34] and a study on HIV protease inhibitors [35]. Recently, a novel non-superposition-based shape recognition methodology, which has been validated also using a set of neuraminidase inhibitors, has been reported [36]. This approach proved to be considerably faster than other shape-based methods and it could be extremely useful in ligand-based VS projects.