Polypharmacology studies are critical in drug discovery and development. However, the exhaustive coverage of all targets by experimental methods is still an industry-wide challenge. Apparently the most competent approach would be to utilize the -omics (proteomics, cheminformatics, etc.) technologies [39
]. The enormous molecular data generated in the post-genomic era has significantly accelerated the polypharmacology research. Systems biology approaches integrated with pharmacology are being frequently used to identify new off-targets [16
]. There are a large number of public and private molecular databases available and are continuously growing in both size and number; some of them are listed in . They integrate diverse information of molecular pathways, crystal structures, binding experiments, side-effects, and drug targets. There are also number of other small molecule databases including: ZINC [42
], PubChem [43
], Ligand Expo [44
], KEGG DRUG [45
], etc., with enormous information about their disease relevance, chemical properties, and biological activities. These databases can be used not only to predict the protein targets of a small molecule, but also to obtain insights of designing polypharmacological agents in a prospective manner.
Some of the representative databases for polypharmacology studies.
With the increasing availability of the above databases, various methods have been applied to predict molecular polypharmacology. Recently Keiser and colleagues used the similarity ensemble approach (SEA) [54
] in a large scale to predict the activity of 656 marketed drugs on 73 unintended ‘side-effect’ targets, and confirmed half of the predictions with the IC50
activity values ranging from 1nM to 30μM [55
]. Among these new associations was the finding that the abdominal pain side-effect of chlorotrianisene was due to its newly discovered inhibition of cyclooxygenase-1 (COX-1). Oprea et al. also reported results from their text mining of 7,684 approved drug and mapped the “adverse reactions” of 988 unique drugs onto 174 side-effects[15
]. They were then clustered with principal component analysis into a self-organizing map and integrated it into a Cytoscape network. They expected that this type of data could streamline drug repurposing [20
]. Barabasi et al. used a polypharmacology approach to build a bipartite graph composed of US Food and Drug Administration-approved drugs and proteins linked by drug-target binary associations [56
]. Zhang et al. demonstrated that polypharmacology can not only speed the wide identification of drug targets but also find new applications for the existing drugs [57
]. Additionally Apsel et al. reported systematic discovery of molecules that potently inhibit both tyrosine kinases and phosphatidylinositol-3-OH kinases [9
]. Crystal structures revealed that the dual selectivity of these molecules is controlled by a hydrophobic pocket conserved in both enzyme classes and accessible through a rotatable bond in the drug skeleton [9
]. In another polypharmacology prediction study, Cheng et al. proposed two different weighted network-based inference methods for chemical-protein interaction prediction [58
]. Bork and colleagues used phenotypic side-effect similarities to infer whether two drugs share a target. They applied to 746 marketed drugs with a network of 1,018 side-effect followed up with experimental validations, and found that 11 out of 13 implied drug-target interactions reveal inhibition constant equal or less than 10um [59
]. Mattingly and coworkers with the aid of text mining tools developed CTD [51
] (comparative toxicogenomics database) which includes curated data describing cross-species chemical–gene/protein interactions and chemical– and gene–disease associations. The database is intended to provide insights into complex chemical–gene and protein interaction networks.
Structure-based methods including inverse docking [60
] are also widely used to predict protein targets of small molecules. A panel of tractable targets involved in a disease network are screened against the approved drug molecules with docking. The targets with available 3D structure are normally used. The top ranked targets (excluding the original known targets) can be treated as the lead off-targets for further testing. For instance, Zou and coworkers [60
] used inverse docking approach to identify the potential direct target oxidosqualene cyclase (part of the cholesterol synthetic pathway) of PRIMA-1 (known for its ability to restore mutant p53 tumor suppressor functions). Experimentally they have shown that both PRIMA-1 and Ro 48-8071 (a known potent OSC inhibitor) could significantly reduce the viability of breast cancer cells relative to normal mammary cells. Similarly Tawa et al [63
] tested 3D-shape/chemical similarity analysis program ROCS (Rapid Overlay of Chemical Structures) to generate off-target profiles of drugs from DrugBank[24
] and KEGG[45
]. A systems pharmacology approach was used by Dar and co-workers to indentify polypharmacological agents with an optimal balance of activity and less toxicity against the kinases Ret, Raf, Src, Tor and S6K [23
]. These examples demonstrate the importance of this highly promising field, and it is clear that developing multi-target therapeutics agents represents an important avenue to advance drug discovery. Some of the methodologies are summarized in .
Broad classification of polypharmacological methods.
Recently, our group, among others [10
], have embarked on the development of novel methods to explore the ligand-target interactions using molecular networks [70
]. These approaches are to deduce polypharmacological relations among drugs and the protein crystal structures reported in PDB [73
]. The drug molecules were obtained from the available public databases such as DrugBank[24
], and the corresponding target information was obtained from PDB [73
]. After data curation with our in-house programs, databases were constructed for both drugs and targets, and networks were created to evaluate their relationships. The structural similarity among the drug molecules that are likely to bind to the same targets has also been evaluated using JChem software [74
]. Using our method, the polypharmacology network created for a query compound Raltitrexed (brand name Tomudex; an anti-metabolite and inhibitor of thymidylate synthase and folylpolyglutamate synthase) is depicted in . The complexity of the network is reduced by hiding the targets and only showing the ligand molecules connected to Raltitrexed in three levels. The network shows multiple clusters of compounds that have close relationships with each other based on their shared targets. Ralititrexed is connected to multiple molecules such as Floxuridine and Leucovorin which share common target (thymidylate synthase) in the first level of network, meaning they might have similar biological activities or they bind to similar targets. The network is further extended to more levels by ligand-target pairing.
Figure 2 Graphical depiction of the drug network for Raltitrexed, derived based on their shared targets. The complexity of the network is reduced by hiding the targets and only showing the drug molecules that are directly or in directly related to Raltitrexed (more ...)