Trends in Drug Discovery and Drug Targets

Currently, many of approved drugs have been developed without knowing much about the molecular mechanism by which they work. In this article, Yildirim et al. were interested in (1) understanding the roles of drug targets in the context of a cellular network, (2) determining if there have been any trends in drug discovery, and (3) determining the relationships between drug targets and the products of genes related to various diseases.

Yildirim et. al analyzed over 4,200 drug entries in the DrugBank database as a network between drugs and their targets. DrugBank contains detailed drug data such as chemical composition and prescribed uses, as well as information about the protein targets of the drug (i.e. sequence, structure and pathway). DrugBank has two major sets of drug entries: the first category is US Food and Drug Administration (FDA)-approved drugs, which accounts for approximately 1,200 drugs, and a second set composed of more than 3,000 "experimental" drugs under investigation. Out of the set of FDA-approved medications there are nearly 400 distinct human proteins being targeted. The inclusion of experimental drugs increased the number of drug targets to about 1,000.

Yildirim et al. created a bipartite graph made up of two distinct sets of nodes: drugs and the protein drug targets (shown below). In this type of network graph, drugs can have connections to drug targets, but never to another drug. Out of the 890 FDA-approved drugs, 788 were found to share a drug target with another drug. There analysis showed that several drugs had over 10 target proteins including propiomazine (Largon), promazine (Sparine), olanzapine (Zyprexa, Zydis), and ziprasidone (Geodon); all of these drugs are prescribed as anti-psychotics. The most-targeted proteins included the histamine receptor H1 (HRH1) and the muscarinic 1 cholinergic receptor (CHRM1) with about 50 drugs targeting each protein. Two networks were then created from the bipartite graph: a drug network with drugs sharing common targets being connected and a target-protein (TP) network was created that includes only drug targets and connections are drawn between two proteins, if a drug targets both. In the TP network, it was shown that 305 out of the 394 drug targets shared a connection.The giant component, the largest connected part of the network, for the drug network was 476 and 122 for the TP network, both smaller than would be expected using randomly generated networks with the same number of nodes and edges. This leads to the conclusion that highly connected nodes in the drug and TP networks are connected to other similarly connected nodes. By looking at the anti-psychotics with many target proteins listed above, we see this trend in that they all share a common function.

From the experimental drug set, an additional 617 protein targets were identified from the approximately 800 experimental drugs with known targets. The giant component was determined to be 596 for the network of proteins targeted with the inclusion of experimental drugs. This is larger than would be expected for random networks, which was calculated as 551 (+/- 10). This differs from the result of using only the targets of FDA-approved drug targets, leading to the conclusion that experimental drug targets are more diversified than currently approved FDA-medications indicating a trend towards polypharmacology (drugs with multiple targets). A large segment, 62%, of FDA-approved drugs target proteins on cell membranes. This contrasts with many experimental drugs attempting to derive a treatment by targeting cellular components inside the cell; only about 40% of experimental drugs target membrane proteins. But looking at the drugs approved from 1996 to 2006, 69% of them target membrane proteins showing that while drug-discovery research is expanding targets, it has not been overly fruitful.

For the next part of their analysis, Yildirim et al. attempted to determine the essentiality of targets by identifying target proteins in a protein-protein interaction (PPI) map. They found that 262 targets were present and on average drug targets had more connections than other proteins in the network. Essential proteins are stated as proteins whose orthologous encoding genes in mice were necessary to produce a viable mice in gene knockout experiments; the quantitative measure is the number of connections of the essential predicted proteins. Using data from Online Mendelian Inheritance in Man (OMIM) Morbid Map, which contains disorder to disease gene associations, a human disease-gene (HDG) network was produced drawing connections between two genes which share a common disease. Only 166 out of 1,777 disease-related genes encode for drug targets; 43% are associated with multiple diseases. Including the target proteins of the experimental drugs, the percentage associated with multiple diseases dropped to 26%, which shows a trend to towards targeting disease genes more specifically.

The authors suggest that most drugs are palliative in nature (relieve symptoms, but do not address the specific genetic cause) due to the distance between the disease-gene and target proteins closely matching the distances for randomized pairings of drug target and disease-gene pairings (shown below). Since 1996, the fraction of disease-gene and drug-target with shorter distances has increased (also shown below). Diseases having drugs that are more directly targeted included cancer, endocrine, psychiatric, and respiratory disease classes. Developmental, muscular, and ophthalmologic diseases had longer distances from target protein. For medications treating advanced cases of cancer, the path from drug target to disease-gene is longer indicating that the drugs are more palliative. One thing lacking from this paper information of the efficacy of the drugs. Overington et al. stated in 2006 that the number of targeted genes would increase to over 600 genes, if you assume that drugs would have some effect on proteins showing 50% identity similarity. Another measure of efficacy that could have been used is binding-affinities to give strengths to the protein-protein interactions. Another interesting piece of information that Yildirim et al. do not highlight is the categories of target proteins. In the same 2006 perspectives paper by Overington et al., drug targets are classified by gene family and they show that over 50% of drugs fell into one of four categories: class I G-protein coupled receptors (GPCRs), nuclear receptors, ligand-gated ion channels, and voltage-gated ion channels.

Yildirim, M.A., Goh, K., Cusick, M.E., Barabasi, A., Vidal, M. (2007). Drug-target network. Nature Biotechnology, 25(10), 1119-1126. DOI: 10.1038/nbt1338

Overington, J.P, et al. How many drug targets are there? Nature Reviews Drug Discovery, Vol. 5, No. 12., pp. 993-996.