Drug repurposing a reality: from computers to the clinic

“Computer-aided methods have [therefore] attempted to tackle drug repurposing from a variety of perspectives, ranging from isolated … to more integrative approaches…”

Modern drug discovery has reached a roadblock, suffering a diminished pipeline with escalating costs, development times and safety concerns coupled to a very low chance of success [1–3]. As seren dipitous discoveries dwindle, there is a need for a shift from traditional drug discovery to the concept of drug repositioning, where currently approved drugs are repurposed for new indications. The technology to evaluate or re-evaluate new diseases, targets, pathways and functions continues to evolve so that research-led repositioning rather than random screening is now a viable strategic model for rapid drug development. Indeed, the number of repositioning publications is growing rapidly with the promise of reduced drug development costs and timelines. Using drug repositioning, pharmaceutical companies and academic institutions have achieved a number of successes, and the rate of new indication approval is growing every year [4]. A major advantage of utilizing approved drugs, given their previously successful clinical trials, is the potential for fast entry into Phase II trials for new indications. Therefore, the benefits of increased success rate and decreased costs, resources and development time make drug repurposing an ideal process to kick start productivity in drug development.

Drug repurposing: a ‘virtual’ reality?

Most drug repurposing efforts have focused on high-throughput chemical screening in vitro to discover new drug–target signatures. While chemical screening can serve well and can uncover unanticipated sites of action, financial and technological limitations preclude its use as a large-scale process. In silico methods that take advantage of modern high-performance computing for efficient virtual screening represent an additional approach. The appeal of virtual screening is the potential for large-scale automation and the processing of datasets several orders of magnitude larger than previously realizable. However, the limiting factor is the validity and accuracy of methods implemented for the prediction of drug–protein or drug–disease signatures.

When drug development was first adapted for computational platforms in the early 1990s, the emphasis was on molecular docking – the prediction of a molecule’s orientation and functional contacts within a protein target’s designated binding site, as well as the biological viability of the ligand–protein complex [5]. Since that time, there have been many strides in the development of computational algorithms for predicting ligand–protein binding affinities, and the literature is filled with comparison studies that objectively assess the validity of different docking software across a variety of benchmarks, as exemplified by the recent study from Bohari and Sastry [6]. However, while docking strategies have undoubtedly become more sophisticated, they still suffer from high false-positive rates, which beg the question of whether our understanding of ligand–protein binding is comprehensive enough and if the focus on ligand–protein signatures is sufficient for accurate pharmacodynamics and clinical outcomes.

Given the context of drug repurposing, there is a wealth of pharmacological and clinical information available for drugs currently being prescribed that may provide a different avenue for accurate predictions, such as known off-target effects. Computer-aided methods have therefore attempted to tackle drug repurposing from a variety of new perspectives, ranging from isolated protein- and chemo-centric approaches to more integrative approaches utilizing multiple modalities as well as biological markers. With respect to the former, these isolated methodologies are driven by the assertion that similar binding sites bind similar molecules. Thus, protein-centric and chemo-centric approaches that focus on modeling target binding sites and ligands, respectively, utilize similarity measurements for their drug repurposing predictions. For example, the anti-Parkinsonian compound entecapone was repurposed as a potential antimicrobial against multiple drug-resistant Mycobacterium tuberculosis based on the high degree of binding site similarity between human COMT and the bacterial enol-acyl carrier protein reductase, an enzyme essential for fatty acid synthesis [7]. On a larger scale, Keiser et al. modeled chemical similarity across ligands into a polypharmacological network such that the repurposing potential of ligands can be analyzed within the entire space of all known human protein targets for all of the small molecule pharmacological agents currently available on the market [8]. Using this method, called the similarity ensemble approach, unexpected drug–protein associations were successfully validated, such as methadone, which targets both μ-opioid and NMDA receptors while exhibiting antimuscarinic activity via binding to M3 receptors. To further enrich the accuracy rate of in silico repurposing predictions, increasingly integrative methods are also being established, for it is likely that drug pharmacology is a multifaceted process depending on multiple modalities. A recent example is the approach devised by Dakshanamurthy et al. called ‘train, match, fit, streamline’, which combines proteo- and chemo-centric modalities with descriptors of complementarity between drug and protein, such as shape [9]. The viability of ‘train, match, fit, streamline’ is evident in the successful experimental validation of the anti-inflammatory celecoxib as a novel cadherin-11 inhibitor and mebendazole, an anti-hookworm medication, as an inhibitor of VEGFR-2.

“…simply establishing a valid drug–protein interaction does not necessarily provide a valid drug–disease association, which is the endgame of clinical pharmacology.”

While computer-based methods such as these are continuously fine-tuned as new biochemical and pharmacological knowledge is published, what has yet to be emphasized in the process is the patient’s clinical scenario. Ultimately, it is the translation of the patient’s biochemistry to his or her clinical presentation that dictates the success of any drug repurposing effort. In other words, simply establishing a valid drug–protein interaction does not necessarily provide a valid drug–disease association, which is the endgame of clinical pharmacology. To make this association and to do so accurately, some drug repurposing initiatives have implemented systems and network biology/pharmacology into their strategies, encompassing markers ranging from gene expression profiles and protein functional pathways to clinical symptoms as well as drug adverse effects.

Systems pharmacology: the new frontier

Systems biology has offered new approaches to our views of drug discovery. We have transitioned from thinking about isolated biological mechanisms to the idea of systems having interacting components and thus being able to form a network showing the interactions between the components. The first indication that networks and systems biology had a role in pharmacology was with the construction of the ‘Human Disease Network’ or the ‘Diseasesome’ and the ‘Drug Target Network’ [10,11]. The Human Disease Network, which depicts disease–gene associations, shows that closely related diseases with common genes would be positioned more closely together. Alternatively, the Drug-Target Network is a bipartite network where drugs are connected to all their protein targets. These networks lay the foundation for network pharmacology.

“…drug repurposing offers multiple opportunities to revitalize the currently dismal state of the drug-discovery pipeline.”

One of the advantages of network pharmacology is the ability to better examine the mechanisms of complex diseases. Many of these approaches use the ‘guilt by association’ principle, in that if a drug targets a gene product associated with one disease, then another disease associated with the same gene product also has the potential to be treated by the same drug [12]. One approach is to integrate disease and drug–target networks and analyze the relationships between drugs and diseases [11,13,14]. These networks predict how drugs may affect a disease based on gene-expression data. Thus, using gene-expression alterations as a novel end point allows for not just a unique perspective for the discovery of compounds, but also for the finding of non-obvious similarities between clinically disparate diseases.

In addition, there are other systems approaches that can be taken to better explore the complex nature of disease. One approach is incorporating metabolic pathway information that has been obtained from databases, such as the Kyoto Encyclopedia of Genes and Genomes, BioCarta or Gene Ontology. Network analysis using metabolic pathways is used because many diseases involve multiple genes and these gene products can be involved in multiple pathways [15,16]. The combination of disease–pathway with drug–target networks attempts to answer how a drug may be targeting a particular disease, thus allowing for inferences based on similarities with other diseases. Another approach used to examine how a drug acts on a system involves using an interactome network [11,17]. This type of network uses an interactome to show the protein–protein interactions for drug targets. Yildirim et al. constructed a network in which the distance or the number of interactions between a drug target and the gene product associated with a disorder was analyzed as an indication of the mechanism of action of the drug [11].

As systems biology is becoming an increasingly implemented perspective in research endeavors, drug repurposing stands to benefit much from the application of network pharmacology. A prominent study by Lamb et al. illustrates the utilization of the networks approach to create the Connectivity Map or CMap, an mRNA expression profiling for physiological, diseased and drug-induced states [18]. Approved drugs can be queried in CMap to establish a drug–gene signature that can then be used to find disease states that are closely related, therefore implying repurposing possibilities. For example, CMap designated the mTOR inhibitor rapamycin as a potential therapeutic for dexamethasone-resistant acute lymphoblastic leukemia in children. A clinical trial is currently underway at the Dana–Farber Cancer Institute (MA, USA) assessing this possible new indication (ClinicalTrials.gov identifier: NCT00874562 [101]).

A new beginning

Moving forward, drug repurposing offers multiple opportunities to revitalize the currently dismal state of the drug discovery pipeline. The advancement of computational tools for more accurate pharmacological predictions along with the development of network systems biology has redefined the conventional understanding of disease and the underlying biochemical components associated with those clinical states. Recently, more integrative approaches encompassing these modalities are appearing, which provide further insight into pharmacodynamics and mechanisms of action that previously went unnoticed. For example, Xie et al. proposed that nelfinavir, a potent HIV-protease inhibitor, is able to exhibit antitumorigenic effects via the inhibition of multiple kinases using a chemical systems biology approach superimposed upon in silico predictions derived from docking and molecular dynamics simulations [19].

Conclusion

The computational approaches discussed here for drug repurposing have their own limitations in terms of predictive power and it is clear that there are obstacles to overcome in terms of accuracy and quality assessment. One way to address these limitations is to integrate ligand, target, phenotype and biological network based approaches, which would likely multiply the predictive power. Further development of more sophisticated techniques that can address the shortcomings of existing computational approaches will be required to efficiently turn shelved compounds into new medicines and predict new indications for existing drugs.

Biographies

Naiem T Issa 

Jordan Kruger 

Stephen W Byers 

Sivanesan Dakshanamurthy