- Top of page
The advent of silicon chip based technologies for genome sequencing promises continuing exponential falls in the reagent costs of sequencing. When every patient has a full genome sequence as part of their medical records the science of drug discovery and drug design must adapt and improve to meet this challenge. This series covers computational, and experimental approaches for small molecules and biologics. From the virtual patient – a computational model of a complete human being, through in silico screening to RNA editing and antibody directed therapies.
In the world of modern technology, the extraordinary can become commonplace in a short time. Scarcely has the hoo-ha of the $1000 dollar genome died away when the $100 dollar is upon us and detailed planning for its impact is already ongoing by government agencies (1). An exponential decrease in reagent costs and corresponding increases in the rate and accuracy of screening have occurred in the last few years. The Archon genomics X prize is for $10 million dollars and is offered to the first team that can sequence a set of $100 genomes for $1000 dollars (http://genomics.xprize.org/). At the time of writing, two teams have entered the competition: Ion Torrent uses a semi-conductor CMOS chip-based technology that measures (as a tiny electrical current) a proton released from the attachment of a nucleotide base. A massively parallel architecture allows a complete genome to be sequenced in a few hours. The Wyss institute has also entered the competition but has not yet disclosed the technology to be used. Not only are the costs of sequencing being reduced but the accuracy is increasing allowing complete sequences to be read for the first time. Thus, the reality of cheap and essentially complete genome sequences is upon us.
In this themed issue, we deal not with genome sequencing per se, but with the future of drug discovery in its aftermath – the ideas, key technologies, and scientific drivers. Our distinguished authors have created some illuminating manuscripts which include radical and challenging ideas.
Simply knowing the genetic code alone may be insufficient. Agents interacting with a single target will affect a network of interacting proteins. Even this complexity is not enough as systems interact within a complete organism. Systems biology seeks to develop mathematical models of systems and even ambitiously, as Shublaq et al. (2) describe, for individual patients. Personalized medicine may be considered a step along this road by targeting specific genetic abnormalities driving a cancer. Genome-wide association (GWAS) studies point to specific genetic determinants of disease. The authors describe an HIV example, the ViroLab project that uses integrated HIV models of viral load, mutations, and CD4+ status. The model uses molecular dynamics and simulation of virus entry, drug binding, and immune response. This system is on clinical trial in Europe. Similar systems exist in the cancer field such as the p-medicine project. The ability to predict toxicology on the basis of a chemical fragment analysis is another approach likely to yield significant data with a direct application for patient treatment. The authors see the main challenge for systems biology as predicting physiology from the computational simulations.
RNA editing by the enzyme adenosine deaminase produces an adenine to inosine change in RNA in mammals (A–I editing). This can have a variety of effects including altering the sequence of RNA transcripts and thus changing the sequence of the protein relative to its genomically encoded version. In the review by Decher (3), he charts the extent of these changes and highlights the effect that they can have on the pharmacology of some major drug targets. These include glutamate receptors, 5HT receptors, and potassium channels (Kv1.1). RNA editing can be developmental or disease related as for 5HT2C in depression or Kv1.1 in epilepsy. Drug treatment can also alter RNA editing levels, and the levels can be different in a model organism or species. This is an alarming subject for a drug industry that has spent most of the last 20 years dealing with cloned and expressed human proteins. These proteins are generally faithful translations of the gene sequence. However, not only can a protein be subject to a myriad of post-translational modifications such as phosphorylation, glycosylation, acetylation, methylation, and nitrosation, but even the fundamental sequence of a drug target can be changed by a simple enzymic process acting on translated RNA. Any of these changes can profoundly affect the pharmacology of drug targets. Given these complexities, the surprise is not that drug discovery using cloned proteins has proved difficult, but that it has worked at all.
Makley and Gestwicki (4) explore the potential for expanding the range of small-molecule drug targets. They first deal with methods of identifying ligands for protein–protein interactions (PPIs), such as NMR-based screening and affinity selection mass spectrometry (AS-MS). Surface plasmon resonance (SPR) and the related, bilayer interferometry are techniques that allow increased throughput of up to 7000 samples per day. The nutlin class of MDM2-p53 inhibitors were discovered using SPR. More effective in silico methods are also being developed for identifying druggable sites by utilizing measures of the hydrophobicity of the solvent-accessible surface area combined with shape curvature. Such methods will allow tractable or druggable targets to be more clearly identified using comparatively cheap in silico methods. It is clear that larger compounds with more structural variations and greater lipophilicity are required for PPI inhibition.
Heikamp and Bajorath (5) discuss the future of virtual compound screening; a central problem here is the growth in the size of both compound databases, for example, ZINC, 20 million compounds, and in the amount of biological activity information available on each compound, for example, ChEMBL contains approximately 1.1 million compounds with activity against 8845 targets. Increases in computational efficiency, Moore’s law not withstanding, can only partly compensate for this increase. Structure-based virtual screening requires further development and improved methods for demonstrating relationships between compound similarity and compound activity. The explosion in the volume of data will make virtual screening more important than ever with chemogenomics and chemical biology highlighted as key areas.
In the review by Sinko et al. (6), the newer methods of accounting for protein flexibility in virtual screening are outlined. Proteins can change conformation upon binding, and docking and strategies that do not address this issue do not work well for such proteins. The cytochrome P450s CYP2B4 and CYP3A4 are examples of proteins that undergo relatively large conformational changes. A range of solutions to this problem have been developed including induced fit, Rosetta ligand, and ensemble-based methods. Computer hardware improvements, mainly through the advent of graphics processing units (GPUs), are allowing millisecond simulations without the use of supercomputer arrays. The ability to incorporate a single mutation into the calculation of a drug’s affinity will be necessary for determining the likely effectiveness of drugs in a patient-specific manner. Audie and Swanson (7) describe advances in the prediction of protein-peptide affinities and the relevance to peptide-based drug discovery. Given the importance of peptides in addressing some of the more difficult PPI targets, this intrinsically more challenging aspect of virtual screening still seems of great importance.
Swift and Amaro (8) highlight the importance of passive membrane permeability to the drug development process through its major influence on ADMET properties of molecules. The problems with the current quantitative structure permeability models (QSPR) are described and the potential for explicit solvent models. These models are currently computationally intensive but their promise is that they can be applied to diverse compound sets and utilize a deep understanding of the physical nature of permeability that QSPR models do not encompass. Perhaps this review highlights the challenges ahead if we cannot yet adequately predict a relatively simple process such as compound permeability.
Liu et al. (9) introduce thermodynamic guided alanine scanning mutagenesis as a way of discriminating binding hotspots in proteins from those that initiate signaling (allosteric hotspots). Traditionally, alanine scanning has been used to determine key binding residues and by implication the major contributors to binding energy; here, the technique is extended using microcalorimetry to identify binding molecules that do not induce an (agonist type) conformational change. In a fascinating account, the authors describe a ‘blueprint’ for discovering pure inhibitors as applied to the CD4/gp120 target of HIV. Critically large enthalpic contributions to binding predict restructuring, whereas small changes indicate binding contributions and the potential for the development of antagonist molecules.
Poltronieri et al. (10) report on the potential for anticancer therapy based on anti-miR-155-directed oligonucleotides. The micro-RNA miR-155 is overexpressed in many types of cancer, but especially glioblastoma. MiR-155 appears to target the GABA-A receptor which is linked to the proliferation of glioma cells. This intriguing biological insight opens the door for a new type of glioma therapy, although considerable hurdles for CNS delivery of oligonucleotides remain. Provost and Wallert (11) discuss NHE1, the sodium hydrogen exchanger isoform 1, as a potential cancer target. NHE1 can alter the intra- and extracellular pH and also act as a molecular scaffold. In all, the protein serves to direct signaling and cytoskeletal proteins to the leading edge of a migrating cell. The implications for metastasis of cancer cells are clear. A large number of interacting proteins and possible regulators of NHE1 have been identified, many of which are therapeutic targets. Giannattasio et al. (12) describe stress-related mitochondrial components and the mitochondrial genome as anticancer targets. This is clearly an emerging field with many technical difficulties to overcome in the delivery of therapeutic agents to mitochondria.
Antibody drug conjugates (ADCs) are not a new paradigm in drug design; in fact, the earliest example is from 1970 and utilized diphtheria toxin conjugated to an antibody (13). Flygare et al. (14) discuss the field with relevance to the >20 ADCs currently in development. The antibodies, linkers, and the cytotoxics all have to be exquisitely designed in order for efficacy to be obtained. Given this complexity, it is perhaps no surprise that it has taken some time to get effective drugs to market. The field has received a considerable boost with the recent approval of brentuximab vedotin (Adcetris) for two lymphoma indications. Brentuximab is a chimeric CD30-targeted antibody and is linked via a simple maleimide thiol coupling to a self-immolative spacer carrying the cytotoxic monomethyl auristatin E. CD30 is expressed on the surface of malignant cells. The authors identify spacer chemistry development as a key future area for exploitation allowing a much greater variety of cytotoxics to be attached to the targeting antibody. Biologics are not solely antibody based, and Chalker (15) discusses protein therapeutics in light of advances in chemoselective peptide ligations that can be used for their production. Recombinant production of proteins can lead to heterogeneous mixtures, or inappropriate glycosylation patterns. An advantage of chemically produced proteins would be the ability to precisely determine the final composition. Several recent examples including Kent’s homogeneous non-glycosylated EPO and Danishefsky’s human parathyroid hormone are given. These are convergent syntheses and involve ligations of intermediate-length peptides. Ligations are no longer restricted to cysteine-containing sequences and final mild desulfurization steps have been developed. Of course, chemical synthesis of proteins could be used to produce novel analogues with improved pharmacokinetics or selectivity.
The future of peptide-based drugs is discussed by Craik et al. (16). Peptides are defined as being of <50 amino acids. Several peptides are on the market including glatiramer acetate (Copaxone) (10 amino acids) used for the treatment of multiple sclerosis and the gonadotrophin antagonist leuprorelin (Lupron™) (nine amino acids). The authors see the potential for peptides especially in the PPI field where small-molecule drugs limited to <500 Daltons molecular weight are not easy to identify. The disadvantages of peptides are seen as low metabolic stability, high clearance, poor permeability, and high cost of manufacture. Despite these apparent shortcomings, over 100 peptides are marketed, and most are small <10 residues and delivered by injection. One of the most intriguing new peptides to reach the market is Exenatide (Byetta™) for type II diabetes. Originally isolated in the saliva from a lizard, the Gila monster, this injectable drug had sales of over $700 million dollars in 2010. Surprisingly when bioencapsulated with chitosan in an enteric-coated capsule, the 39-amino-acid peptide is orally bioavailable, which has stimulated much activity in the field. Chitosan is a cationic polysaccharide that can transiently open the tight junction between endothelial cells increasing absorption via the paracellular pathway. The authors also highlight the great potential for new drug leads in natural venoms of other organisms. The promise of cyclic peptides as new orally available drugs with high membrane permeability is outlined citing important studies by Lokey and others to show that intramolecular hydrogen bonding coupled with N-methylation is key to masking the H-bond donor groups on peptides. This masking results in high permeabilities and oral bioavailabilities in the 20–30% range. A rigid application of Lipinski rules is not appropriate for these peptides; rather, it is the more fundamental aspects of permeability and metabolic stability that are important. Jamieson et al. (17) provide a useful counterpoint to the article on thermodynamic alanine scanning with a broad review of peptide scanning techniques including the use of modern variants such as stapled peptides. These are peptides bridged across amino acids generally at the i, i + 4 or i, i + 7 positions and promote a helical conformation in a small peptide. The authors also highlight aza scanning as a technique with much promise in view of both the conformational restriction provided and the synthetic flexibility that allows the inclusion of many ‘unnatural’ side chains.
From these reviews, several themes come through the increasing importance of computational models and methods. Methods that might allow us to make sense of the data storm arriving from genomics, structural biology, and biological screening databases. The increasing ability to block PPI targets with small molecules or peptides and an increasingly sophisticated approach to biologics will enable many more effective therapies. Key problems such as ensuring sustainable health care are not addressed in this special issue but will be as important as the technological advances going forward. Small-molecule medicines are the cheapest to manufacture and provide economical health benefits once off patent, but even these are beyond the reach of many of the world’s poorest populations.