The modern drug discovery process in operation today is frequently described in terms of the sequential elements of target identification and target validation, lead identification and lead optimization, followed by the selection and development of a clinical candidate. This process is illustrated in Fig. 2. Also shown are the technological advances that are being implemented to improve efficiency, as well as the strategic impact of these advances. The discovery of new genes is being accelerated enormously by the new science of genomics (9,13,14). Through the use of robotic high throughput sequencing methods all the potential 100,000 genes or more in the human genome will be identified and sequenced in a working draft version by spring 2000 and in fully accurate, completed form by 2003 (15). As gene discovery gets faster and is eventually completed, the major
bottlenecks become the assignment of function to the encoded proteins, their positioning within cellular complexes and pathways, and their validation as feasible and attractive targets for pharmacological intervention. This process is being greatly facilitated by the high throughput, global biology methodologies of nucleic acid microarrays (16) and proteomics (17).
There is currently a proliferation of "omics" research (Fig. 3) from genomics (DNA level), transcriptomics (mRNA), and proteomics (proteins), through to the more complex, higher organizational level omics of cellomics (cells), phenomics (phenotypes), metabolomics (metabolic capabilities), and finally pharmacogenomics/ phenomics in humans. The determination of gene expression patterns in normal and disease tissues, using expressed sequence tag (EST) libraries or gene microarrays, is proving to be exceptionally useful. As a case in point, cathepsin K was identified as a novel cysteine protease in a database of EST sequences derived from an osteoclast library (13). This led to the rapid development of cathepsin K inhibitors, which should block bone resorption by osteo-clasts and thereby prevent osteoporosis.
In cancer, target validation is helped considerably by studying the genetics and gene expression patterns of tumors. Cancer is now frequently referred to as a genetic disease. It is, in fact, a series of around 100 diseases in which sets of genes undergo germline or, more commonly, somatic mutation, or are subject to aberrant expression (18,19). With regard to the validation of potential therapeutic targets and the selection of those likely to be more promising for therapeutic intervention, it seems logical to propose that those genes and pathways that are most commonly subject to mutation or deregulated expression are likely to be the most fruitful to pursue (12,20). The receptor tyrosine kinase ^ Ras ^ Raf ^ MEK ^ MAP kinase pathway regulating proliferation and the control of the cell cycle by the cyclin-dependent kinase (CDK)-retinoblastoma gene product axis
(21) are excellent examples. A further recent example of genetic validation is that of PI3 kinase. The PIK3CA gene that encodes the p110a catalytic subunit of PI3 kinase is amplified and overexpressed in ovarian cancer (22). Activation of p110a PI3 kinase signal transduction may also be achieved by genetic deletion of the PTEN tumor suppressor gene; the product of the PTEN gene acts as a lipid phosphatase that reverses the reaction catalysed by p110a PI3 kinase (23,24). Enzymes, such as kinases, are very good examples of pharmacologically tractable targets, because the technical feasibility of discovering small molecule enzyme inhibitors is very high as a result of the presence of a small-molecule binding site. This is not the case for blocking protein-protein interactions, e.g., as exemplified by the search for nonpeptidic, "drug-like" small molecule SH2 domain inhibitors which has not been successful so far.
Data are now beginning to emerge on the differences in global gene expression between normal and cancer cells. For example, in one of the first papers in this area, more than 300,000 transcripts derived from at least 45,000 different genes were analyzed in gastrointestinal tumors and corresponding normal cells, using the technique of serial analysis of gene expression (SAGE; 25). Although there was extensive similarity, 548 transcripts (1.5%) were differentially expressed. Perhaps surprisingly, the main differences were seen in differentiation markers, genes associated with protein synthesis, ribosomal proteins, elongation factors, and glycolysis rather than oncogenes.
Following identification, validation, and selection of a target, the next phase is lead identification. This phase is now mainly carried out by high throughput screening of large chemical libraries against recombinant protein targets (12,26,27). Compound libraries must be chemically diverse. Computational chemistry methods can be used to maximize diversity in an efficient way (28). There are advantages to removing chemicals that are generally poor starting points for a medicinal chemistry program, are unlikely from past experience to result in drugs, or are highly toxic. Such nondrug-friendly chemical types include highly charged compounds, heavy metals, alkylating agents, and Michael acceptors. Chemically reactive compounds cause considerable problems (29). The screening approach is complemented by molecular design, often aided by the use of peptide mimetic chemistry and the structural biology techniques of X-ray crystallography and NMR (30,31).
A novel strategy is to combine organic synthesis, screening of libraries of natural product-like substances, site-directed mutagenesis, and X-ray crystallography in a creative approach known as "chemical genetics" or "chemical biology" (32). This approach uses chemical compounds to probe protein function, in an analogous way to the genetic mutation method, and also to develop synthetic derivatives of natural products as potential drug candidates. This approach has been applied to signal transduction targets. Examples include the use of rapamycin and trapoxin to study FRAP and histone deacetylase, respectively, and the design of potent peptide ligands for SH3 domains (32,33).
Leads discovered by screening or design are then optimized by iterative cycles of medicinal chemistry refinement and rapid feedback from biological evaluation, building up an understanding of structure-activity relationships for the desirable and undesirable features of the lead molecule (34). This process is aided today by the revolutionary method of combinatorial chemistry (35,36), which also generates chemical diversity for primary screening. By these means, identification of leads and their optimization in terms of potency, selectivity, and activity in intact cells by the desired mechanism has become much more readily achievable.
A major hurdle is, however, the transition from activity at the level of in vitro cell culture to activity in the intact animal. Poor pharmacokinetics is a major bottleneck (12,37). Problems can arise in all aspects of pharmacokinetic behavior, i.e., absorption, distribution, metabolism, and excretion (ADME). Oral absorption (usually required for chronic administration schedules) can be problematic, as can tissue uptake. Elimination from the body may be too fast because of overly rapid metabolism or renal excretion. As a result of such problems, pharmacologically and therapeutically active drug levels of lead compounds may not be achieved. At this stage of the project some degree of target potency and selectivity may have to be sacrificed in order to gain the necessary improvement in ADME properties. Progress can be made by modifying the physicochemical properties of compounds, e.g., introducing solubilizing functions, and there are valuable guidelines, such as Lipinski's "rule of five", to help optimize bioavailability (38). Although useful rules of thumb can be employed to improve ADME within a particular lead series (e.g., 39,40), the development of structure-pharmacokinetic relationships is fairly primitive (41,42) and it is extremely difficult to predict the pharmacokinetic properties of a given chemical compound.
In modern drug discovery projects, this important issue is currently being addressed in a pragmatic way by increasing the throughput of pharmacokinetic and metabolic analysis, for example using "cassettes" or mixtures of compounds dosed together in low amounts with very sensitive HPLC-MS-MS detection (43,44). The most promising structures can then be selected for further rounds of optimization. Nevertheless, the development of chemoinformatics algorithms to predict structures having good pharmacokinetic properties is an important future objective. However, the rational design of robust, druglike character looks set to remain a major bottleneck for the next several years.
With sufficient bioavailability achieved, the next hurdle is to demonstrate some mechanism-based pharmacodynamic activity in the animal. Better animal models are needed to give a faster readout of mechanism-related pharmacodynamic activity. These may involve the use of reporter genes in transgenic animals or other genetically engineered models. A relatively high throughput model for the effects of anticancer agents in vivo is the hollow fiber assay (45). This has some advantages in terms of speed and cost over solid tumor xenografts, but has not been validated for the new generation of signal trans-duction inhibitors and it is unsuitable for antiangiogenics.
Current issues surrounding late preclinical and clinical development of cancer drugs have been discussed recently (12,46). Of particular importance in cancer and most therapeutic areas is the need for pharmacodynamic endpoints that will enable us to judge whether the molecular target is being modulated in the intact animal and patient. Early clinical trials of agents affecting novel molecular targets must contain a strong component of hypothesis testing. Is the intended molecular target being affected (e.g., kinase or farnesyltransferase inhibition)? Is the biochemical pathway being modulated (e.g., MAP kinase activation)? And is the desired biological effect being achieved (e.g., inhibition of proliferation, cell cycle transit, survival, or angiogenesis)? If the answer to those questions is yes this provides confidence to move forward to the more expensive phases of clinical development. A structured, logically based approach to clinical development can be a major aid to decision making. If problems are seen at any level then these can be addressed or resources reallocated to other more promising projects. For example, if the target is failing to be appropriately modulated, this might suggest a limitation with the drug candidate, and indicate that an appropriately designed back-up compound could be more effective. If the target is being modulated to the required degree but the desired biological effect is not seen, this suggests that the target is not valid but modulation of other targets in the pathway to achieve the biological effect may be worth pursuing. If, however, both the target and the biochemical pathway or biological effect are being suitably affected (i.e., to an extent defined in a preclinical model) but there is no impact on the disease process, then this would indicate that the pathway or biological effect is not linked to the disease in humans, and further approaches to the whole biochemical pathway and biological effect may not be worthwhile.
In terms of the development of pharmacodynamic endpoints, the use of modern molecular techniques will be crucial, and it seems likely that nucleic acid microarray and proteomics technology in particular are poised to play a major role. Noninvasive imaging technologies can provide valuable information, especially positron emission tomography and magnetic resonance imaging (47).
Creative trial design will be important (46). Trial designs aimed at demonstrating a slowing of disease progression, as in Alzheimer's disease (with tacrine), amyotrophic lateral sclerosis (with riluzole), and rheumatoid arthritis (with prednisolone) find parallels in other therapeutic areas, including cancer, where prolonged disease control, rather than cure, would have significant value.
Toxicology is essential to ensure acceptable safety in humans. However, since excessive toxicology can cause delay and in oncology it is often poorly predictive with respect to the qualitative nature of particular organ toxicities, nonprofit organizations in Europe (the Cancer Research Campaign [CRC] and European Organization for Research and Treatment of Cancer [EORTC]) utilize a system with a relatively simple program of rodent-only toxicology (48). This has proved safe and effective (49,50).
Regulatory review is also speeding up. FDA approval for Herceptin (triatuzumab), a humanized monoclonal antibody against the tyrosine kinase receptor erbB2, was obtained in a record 4.5 mo (51). Arguably the first molecular target therapy based on cancer genom-ics, this agent shows promising activity in breast cancer.
4. THE CONTEMPORARY PARADIGM: HOW MANY NEW TARGETS?
The contemporary paradigm of new drug discovery can be summarized as: New Genes ^ Novel Targets ^ Innovative Medicines
This paradigm is based on the premise that the discovery of innovative agents having a high degree of selectivity for a given molecular target involved in disease causation and progression will lead to drugs that have markedly improved efficacy and tolerability in humans. Thus, much will depend on the correct identification and selection of the disease target. What is our expectation of the likely numbers of targets arising now that we have entered the genomic era of drug discovery? It has been estimated that genomics has the potential to deliver 3000-10,000 interesting new targets for therapeutic intervention out of approx 100,000 genes in the human genome (9). How is this figure computed?
The calculation is based on the proposal that there are likely to be 5-10 disease-related genes for each of the 100 or so, at a conservative estimate, really important human diseases with major unmet medical need: hence there could be 500-1000 key disease-related genes. Each gene product interacts in biochemical pathways with, say, 3-10
Proliferation Differentiation Apoptosis
Fig. 4. Cell fate is controlled by signal transduction pathways, e.g., the decision to proliferate, differentiate, or undergo programmed cell death.
Fig. 4. Cell fate is controlled by signal transduction pathways, e.g., the decision to proliferate, differentiate, or undergo programmed cell death.
upstream or downstream partners, which when multiplied up gives a total of 150010,000 potentially interesting drug targets. This number of targets represents a very significant increase over today. As shown in a 1996 survey (9), today's drugs, across all therapeutic areas, act on only 417 or so targets (enzymes, receptors, ion channels, and so forth, excluding anti-infectives). These drugs were mainly discovered by classical, empirical methods, usually without detailed knowledge of the molecular target. Not all of the potential 10,000 new targets will prove pharmacologically tractable (e.g., some will be structural proteins that are difficult to modulate), but it is clear that there should be a major opportunity to increase the number of therapeutic targets. Moreover, these targets will be genetically and biochemically validated and ideal for highly focused, mechanism-based drug discovery and development.
Was this article helpful?