Human beings are complex gene machines that rely on the intricate interplay of ~25,000 genes to impart biological function. Genomic function is affected by genetic variability and environmental factors, giving rise to considerable functional heterogeneity in the human population. Drug responsiveness and side-effects can vary from person to person, as seen in the recent observation of COX-2 inhibitor toxicity in a subset of prescribed users.
Currently available microarray assays for genotyping and genome-wide expression analysis offer prospects for reducing the incidence of COX-2 inhibitor toxicity by providing diagnostic information regarding susceptibility to cardiovascular complications.
The same technologies may be used in an attempt to select patients most likely to benefit from novel therapies (for example, selecting people more likely to be at risk of gastrointestinal complications from NSAID use, the driver for development of COX-2 selective inhibitors in the first place). The extent to which the pharmaceutical industry embraces these recent events as an opportunity to improve drug safety hinges on whether COX-2 inhibitor toxicity is seen as the ‘warning shot’ that it is.
Human beings can be viewed as ‘gene machines’ that rely on the genes encoded by the genome to mediate and direct biological function. This biological model is analogous to the manner in which mechanical devices (eg computers, robots, wristwatches) derive proper function from the concerted action of semiconductors, resistors, gears, springs and other mechanical parts (Figure 1). Mechanical and electronic parts in the machines that human ingenuity has invented interact with each other and the environment in the performance of their roles. While built to withstand a variety of conditions, mechanical wear, changes in weather, gravity and even the amount of cosmic irradiation of the earth have been known to affect performance in complex ways. The same is true for humans. The ‘human as gene machine’ model is conceptually useful in that it emphasises the complexity of biological systems and the interrelatedness of the parts (genes), explains how malfunctioning can occur if one or more of the parts is defective (gene variant) or becomes inoperative (mutation), provides for key links between the environment (diet, stress, infectious agent) and the proper functioning of the machine, and allows for remediation through the use of pharmaceutical compounds that improve the system by altering the output or function of the genes. The gene machines model in the context of drug therapy illustrates how biological systems, genetics and environmental factors collaborate to determine therapeutic outcomes (Figure 2).
The most recent data indicate that the human genome contains approximately 25,000 genes1, with major gene families encoding proteins of regulatory and signalling activity. Popular drug targets such as kinases, phosphatases, G-protein coupled receptors, nuclear hormone receptors and oxygenases, which mediate exquisitely complex signally hierarchies, comprise more than 1,000 of the 25,000 genes in the human genome. The enormous genetic and biochemical complexity of a human being can be illustrated using simple combinatorial mathematics. In simplified form, one can model a gene as having two expressed states (‘on’ and ‘off’). This binary ‘on/off’ model can also be applied to proteins that assume regulatory or non-regulatory states and so forth. Using a binary model, a single human cell expressing only 100 genes could assume 2100 (1 x 1030) different physiological states. With a repertoire of 25,000 genes operating in more than 1013 cells, it should be obvious that the human ‘gene machine’ possesses an essentially infinite number of possible physiological states (although a very limited number of such states are optimised at any one time for the state of growth and development of the organism). With respect to drug development and toxicity, the enormous complexity and diversity of human beings mandate approaches that can identify important variants and states. Deployment of a technology such as microarrays that has sufficient ‘bandwidth’ to analyse this biological complexity on a genomic scale has enormous potential to reduce what is essentially incomprehensible complexity, to a set of manageable data useful for making scientific, clinical and business decisions.
The genomic biological view is perhaps somewhat difficult to assimilate in a pharmaceutical environment, where so much emphasis is placed on single targets. Certainly a complete understanding of drug target structure and function is a pre-requisite for successful drug development. But the ‘target is king’ mentality greatly oversimplifies and perhaps blinkers one to the fact that drugs work at the genomic (proteomic) level in the context of complex genetic variability and environmental influence. It might actually be more efficient and effective to initiate the drug discovery process at the genomic level and then work backwards to studying the primary target once promising compounds have been identified. In the long run, only a holistic biological view affords the proper perspective in terms of designing and administering safer and more effective drugs. The pharmaceutical and biotechnology industries are obviously embracing this concept, but perhaps not aggressively enough.
Current estimates based on genomic sequencing and genotyping indicate that human beings share approximately 99.9% sequence identity at the nucleotide level1. At first glance, this would seem to suggest that all humans are essentially identical genetically, except for the fact that the genome comprises approximately 2.9 x 109 nucleotides. A 0.1% polymorphism rate across 2.9 billion nucleotides would therefore on average give rise to 30 million sequence variants from one person to another, the most important class of which is the single nucleotide polymorphisms (SNP). SNPs that occur in coding regions can significantly alter protein function. Polymorphisms that alter the function of protein involved in drug binding (targets), side reactions (‘off-targets’), or drug metabolism, can exert a major influence on drug efficacy and toxicity.
A good example of the importance of understanding patient genotype as a pre-requisite for optimal drug treatment can be seen in recent advances in cytochrome P450 (CYP) genetics and diagnostics. The complete sequence of the human genome confirms that the CYP genes comprise a family of approximately 50 genes, important members of which encode iron-based monooxygenases expressed in the liver. CYP liver enzymes use atmospheric oxygen to catalyse the breakdown of foreign compounds such as caffeine, codeine, steroid hormones, organic solvents, antibiotics, antidepressants, non-steroidal anti-inflammatory drugs (NSAIDs and other exogenous compounds and medicines. Patients with fully active CYP enzymes metabolise drugs much more quickly than patients bearing SNPs that impair CYP enzyme function (‘slow metabolisers’). Recent estimates suggest that as much as 40% of all P450-dependent drug metabolism is carried out by enzymes encoded by polymorphic genes2. The CYP2C19 gene, for example, contains eight common alleles, seven of which encode inactive P450 enzymes. The CYP2C9 gene, another important CYP family member that catalyses the turnover of the NSAID Cyclooxygenase-2 (COX-2) inhibitor known as Celecoxib, is also polymorphic. It stands to reason that the 100mg and 200mg recommended daily dosages of Celecoxib (Celebrex®) prescribed to treat osteoarthritis (OA) and rheumatoid arthritis (RA), respectively, may result in very different serum concentrations from patient to patient depending on CYP2C9 patient genotype. These individual differences in metabolism, rarely appreciated in standard pharmacokinetic studies on a small number of selected humans during the standard drug development process, can have enormous implications when a drug is launched and used by a wide variety of people with different metabolic activities. The recent launch of P450 genotyping microarrays by several biotechnology companies offers a means of affordable P450 genotyping in advance of the administration of COX-2 inhibitors and other drugs.
But the COX-2 story vis-à-vis patient genetics is more complex than simply accounting for genetic variability at the P450 loci. A second important gene family encoding the UDP-glucuronosyltransferase (UGT) enzymes appears to provide the major biotransformation function of another widely used NSAID known as Rofecoxib (Vioxx®). Similar to the CYP genes, significant genetic variability has been identified in UGT2B7 and UGT2B15, and these polymorphisms may underlie different rates of Vioxx® metabolism in humans3. The message here is that patient genetics appear to not only dictate different rates of COX-2 inhibitor turnover in patients, but polymorphisms in the CYP and UGT genes probably lead to differential rates of metabolism of Celebrex® and Vioxx® depending on whether the patients bear SNPs at CYP450, UGT or both. Insomuch as patients sometimes ‘switch’ from one COX-2 inhibitor to another, it should be clear that these two important NSAIDs are likely to work differently in patients expressing polymorphic CYP and UGT enzymes. The COX-2 inhibitors all share the same primary target (COX-2 enzyme), but the body appears to clear Celebrex® and Vioxx® using rather different biochemical pathways. While not emphasised in the scientific literature on COX-2s and NSAIDS in general, it has been widely known among practicing physicians that an individual patient may seem not to ‘respond to’ (gain clinical benefit from) one COX-2, but have excellent effect with another. This observation is likely due in some part, and maybe in very large part, to these individual differences in metabolism.
There is increasing pressure to guide usage and dosage recommendations for Celebrex®, Vioxx® and other important drugs based on patient genetics. To the extent that microarray technology offers an affordable means of consumer-scale genotyping, microarrays appear to be a platform of choice for determining ‘drug genotype’. Recent approval of Roche’s AmpliChip Cytochrome P450 test by the Food and Drug Administration (FDA) and the publication of the FDA guidance for drug metabolising genotyping systems4 are likely to pave the way for a steady stream of genotyping microarray tests for in vitro diagnostic use. An expanding repertoire of microarray genotyping tests for cytochrome P450, UGT and other genes involved in drug binding and turnover, as prepackaged chips and testing services by reference laboratories, are likely to be available in the near future. The use of electronic databases and Internet-based models of data transfer will provide a much-needed real-time link between genotyping information and clinical trials.
Except for relatively rare somatic mutational events, patient gene sequences are permanently ‘hardwired’ into the genome as either wild type (‘normal’) or altered alleles (SNPs). The same is not the case for the expression of genes, which is a highly dynamic process whereby messenger RNA (mRNA) and protein molecules are synthesised from their gene templates. Genes are activated and repressed in response to a plethora of intracellular and environmental stimuli including hormone levels, nutritional status, body temperature, salt balance, stress, bacterial and viral infection, drug therapy and others. Changes in gene expression are mediated by transcription factors, sequence-specific DNA binding proteins that bind in the vicinity of regulated genes and modulate their expression.
The Cox-2 gene is represented once in the human genome on chromosome 1 and its promoter region contains putative regulatory sites for nuclear factor KB, interleukin-6, cyclic AMP responsiveness, glucocorticoid receptor, and other factors that regulate Cox-2 expression5,6. The expression of Cox-2 is regulated through these DNA elements by a complex set of physiological signals including high salt and sugar levels (hypertonicity), dehydration (hypertonicity), nitric oxide (NO), lipopolysaccharides (LPS), arachidonic acid (AA) and other stimuli7. Inflammation-induced activation of the Cox-2 gene leads to elevated levels of the COX-2 enzyme and a concomitant increase in prostaglandin biosynthesis, the biochemical product of COX-2 catalysis. Higher levels of prostaglandins produce the pain and swelling associated with osteoarthritis and rheumatoid arthritis. NSAIDs such as Celebrex® and Vioxx® inhibit the COX-2 enzyme which reduces prostaglandin levels, conferring the anti-inflammatory, analgesic and antipyretic effects of these drugs. Given that multiple cellular factors modulate Cox-2 expression, the effectiveness of COX-2 inhibitors is expected vary depending on the idiosyncrasies of patient physiology. The complex role of genetic and environmental determinants in COX-2 inhibition (Figure 3) calls into question the current practice of prescribing fixed dosages of these agents to a heterogeneous patient population.
COX-2 inhibitor toxicity
NSAIDs are the most widely prescribed drugs in the world, with an estimated 100 million users of Celebrex® and Vioxx® since the drugs were introduced on the market in 1999. In the United States alone, sales of Celebrex® and Vioxx® were approximately $5 billion and $1.8 billion, respectively, representing nearly 3% of the $235 billion US prescription drug market. After a recent threeyear colon cancer study showed a slight increase in the risk of heart attack and stroke in patients taking Vioxx®, Merck & Company, Inc voluntarily pulled Vioxx® off the market in the fall of 2004, causing an immediate loss of $28 billion in the company’s market value (Figure 4). The Merck stock price is recovering slowly, after the company received FDA permission to place Vioxx® back on the market with a strict ‘black box’ drug safety warning. Pfizer’s widely used COX-2 inhibitors Celebrex® and Bextra® remain on the market, though the safety of both compounds is being studied intensively in view of the fact that Vioxx®, Celebrex® and Bextra® share nearly identical chemical structures. How dangerous are COX-2 inhibitors?
The seriousness of COX-2 inhibitor cardiovascular toxicity, coupled with the widespread use of these drugs and the pervasiveness (and persuasiveness) of an aggressive news media focused on controversy, portray a rather negative portrait of these important therapeutics, and certainly it is important to embrace safety concerns with candour and compassion. But the numbers and the clinical data deserve rigorous analysis. According to one study, the incidence of heart attack and stroke was 0.45% higher in colon cancer patients taking Vioxx® for longer than 18 months compared to a control group taking a placebo8. At the very worst, this means that Vioxx® is probably safe for >99.5% of the population and the drug may actually be quite a bit safer than this. For example, it may be that patients predisposed to colon cancer are also predisposed to cardiovascular complications when taking Vioxx®. The connection between cancer and cardiovascular complication is tenuous (at least with what we know today), but the human ‘gene machine’ is very complex and the interconnectedness of the cellular signalling pathways is relatively poorly understood at the genomic level. It may also be that Vioxx® slightly increases the rate of cardiovascular disease, but only in patients taking the drug for prolonged periods or at elevated dosages. It may well be that Vioxx® is entirely safe if taken at recommended dosages either periodically or for less than one year. What is needed is a diagnostic test sufficiently powerful to predict COX-2 inhibitor toxicity in advance of heart attack or stroke. More generally, a diagnostic that could be used to help physicians make more evidence-based decisions on risk/benefit ratios for individual patients would be transformative in medicine.
Whole genome microarrays
Nearly a decade after the first microarray publications emerged9-14, the technology has advanced to the point where single ‘chips’ can be used to quantitatively monitor expression of all 25,000 genes in the human genome (Figures 5 and 6). The basic design of whole genome microarrays is similar to the more modest original assays, though the technical details have changed. Whole genome microarrays are manufactured using either in situ synthesis or post-synthesis deposition (Table 1) to produce microscopic collections of oligonucleotides arranged in an orderly fashion on glass. Oligonucleotide microarrays containing gene-specific targets are hybridised with labelled probes derived from human messenger RNA, washed to remove unbound probe, and scanned to detect the fluorescent signals produced by the sequence-specific hybridisation events. The images are quantified to ‘extract’ the gene expression values and expression data are assembled into databases for data mining and modelling. Examination of a large number of human samples typically provides a clear ‘expression signature’, several dozen to several hundred genes that are activated or repressed in a physiological state. The success sub-genomic microarray experiments using celecoxib (Celebrex®) and rofecoxib (Vioxx®) as drug treatments15-18 indicate that whole genome microarray analysis of a large number of COX-2 inhibitor patients is likely to furnish extremely valuable data.
The appeal of whole genome microarray expression analysis is that the assay is likely to provide a functional diagnostic of COX-2 inhibitor toxicity that is inclusive of many if not all of the factors that affect toxicity including patient genotype (eg CYP450 and UGT), environmental factors (eg diet, stress), patient behaviour (eg dosage nonadherence), and other complexities. The whole genome microarray provides the ultimate ‘end point’ measurement of the entire human genome in a single test. A COX-2 inhibitor toxicology (risk) profile is expected to emerge in patients at risk for stroke and myocardial infarction in advance of these serious cardiovascular complications. A microarray-based expression profiling test could be offered to patients taking physician-prescribed Vioxx®, Celebrex®, and Bextra® as prolonged treatments for OA and RA.
In the wake of revelations over COX-2 inhibitor toxicity, good science and medicine suggest that we see this as an opportunity to fine tune the drug discovery process. Genotyping and gene expression microarray assays will likely provide SNP and expression signatures that will be useful in determining patient compatibility with COX-2 inhibitors and other drugs. But the extent to which lemons can be made into lemonade in this case hinges on whether the pharmaceutical industry sees this as the shot over the bow that it is. Vioxx® is back on the market and the dust seems to be settling, but Merck is hardly out of the woods and Pfizer needs to be (and likely is) sweating. The liability associated with toxic drugs is real, and it is in the interest of nobody to see the pharmaceutical enterprise take a direct hit. COX-2 inhibitors are good medicines and they should be made as safe as possible, if for no other reason than for the least defensible reason of all – that making them such would be very good business. More generally, a medical-industrial enterprise that is continually punished post hoc for innovation and risk-taking will simply cease to add value. New technologies such as whole-genome analysis provide the industry with both a tool to refine drug discovery and development, and a weapon with which to fight back.
The authors wish to thank their colleagues for helpful suggestions and comments. Mark Schena is a Visiting Scholar in the ArrayIt Life Sciences Division at TeleChem International, Inc. This work was supported with funding courtesy of TeleChem International, Inc.
Dr Mark Schena received his BA in Biochemistry from UC Berkeley (CA, USA), his PhD from UCSF, and conducted postdoctoral research with Dr Ronald Davis in the Stanford Biochemistry Department. Dr Schena and his Stanford colleagues published the first paper on microarrays in Science magazine in 1995, and since then Dr Schena has lectured and published extensively on the subject.
Dr William Greene is a General Partner at MPM Capital, the world’s largest dedicated healthcare venture capital firm. Prior to joining MPM, Dr Greene was a Senior Clinical Scientist and Epidemiologist at Genentech with development, regulatory and drug safety responsibilities. He received his MD from UCSF, clinical training and research at Yale, and did further research work at the NIH. He is currently an Assistant Professor of Medicine at UCSF and serves on the boards of several private companies.
1 International Human Genome Sequencing Consortium. Finishing the euchromatic sequence of the human genome. Nature 431: 931-945, 2004.
2 Ingelman-Sundberg, M, Oscarson, M and McLellan, RA. Polymorphic human cytochrome P450 enzymes: an opportunity for individualized drug treatment.Trends Pharmacol Sci 20: 342-349, 1999.
3 Zhang, JY, Zhan, J, Cook, CS, Ings, RM, Breau,AP. Involvement of human UGT2B7 and 2B15 in rofecoxib metabolism. Drug Metab Dispos. 31: 652-658, 2003.
4 Guidance for Industry and FDA Staff. Class II Special Controls Guidance Document: Drug Metabolizing Enzyme Genotyping System. March 10, 2005.
5 Tazawa, R, Xu, XM,Wu, KK, Wang, LH. Characterization of the genomic structure, chromosomal location and promoter of human prostaglandin H synthase-2 gene. Biochem Biophys Res Commun. 203: 190–199, 1994.
6 Nie, M, Pang, L, Inoue, H, Knox, AJ.Transcriptional regulation of cyclooxygenase 2 by bradykinin and interleukin- 1beta in human airway smooth muscle cells: involvement of different promoter elements, transcription factors, and histone h4 acetylation. Mol Cell Biol. 23:9233-9244, 2003.
7 Harris, RC, Breyer MD. Physiological regulation of cyclooxygenase-2 in the kidney.Am J Physiol Renal Physiol. 281:F1-11, 2001.
8 Bresalier, RS, Sandler, RS, Quan, H, Bolognese, JA, Oxenius, B, Horgan, K, Lines, C, Riddell, R, Morton, D, Lanas, A, Konstam, MA, Baron, JA. Cardiovascular Events Associated with Rofecoxib in a Colorectal Adenoma Chemoprevention Trial. N Engl J Med, 2005.
9 Schena, M, Shalon, D, Davis, RW, Brown, PO. Quantitative monitoring of gene expression patterns with a complementary DNA microarray. Science 270:467- 470, 1995.
10 Schena, M. Genome analysis with gene expression microarrays. Bioassays. 18: 427- 31, 1996.
11 Schena, M, Shalon, D, Heller, R, Chai,A, Brown, PO, Davis, RW. Parallel human genome analysis: microarray-based expression monitoring of 1000 genes. Proc Natl Acad Sci U S A. 93: 10614-10619, 1996.
12 DeRisi, J, Penland, L, Brown, PO, Bittner, ML, Meltzer, PS, Ray, M, Chen,Y, Su,YA,Trent, JM. Use of a cDNA microarray to analyse gene expression patterns in human cancer. Nat Genet. 14:457-460, 1996.
13 Nelson, N. Microarrays pave the way to 21st century medicine. J Natl Cancer Inst. 88:1803-1805, 1996.
14 Lockhart, DJ, Dong, H, Byrne, MC, Follettie, MT, Gallo, MV, Chee, MS, Mittmann, M, Wang, C, Kobayashi, M, Horton, H, Brown, EL. Expression monitoring by hybridization to high-density oligonucleotide arrays. Nat Biotechnol. 14: 1675-1680, 1996.
15 Reilly,TP, Brady, JN, Marchick, MR, Bourdi, M, George, JW, Radonovich, MF, Pise-Masison, CA, Pohl, LR.A Protective Role for Cyclooxygenase-2 in Drug- Induced Liver Injury in Mice. Chem Res Toxicol 14: 1620- 1628, 2001.
16 Tseng,WW, Deganutti,A, Chen, MN, Saxton, RE, Liu, CD. Selective cyclooxygenase-2 inhibitor rofecoxib (Vioxx) induces expression of cell cycle arrest genes and slows tumor growth in human pancreatic cancer. J Gastrointest Surg 6: 838-843, 2002.
17 Levitt, RJ, Buckley, J, Blouin, MJ, Schaub, B,Triche,TJ, Pollak, M. Growth inhibition of breast epithelial cells by celecoxib is associated with upregulation of insulin-like growth factor binding protein-3 expression. Biochem Biophys Res Commun. 316: 421-428, 2004.
18 Mima, S,Tsutsumi, S, Ushijima, H,Takeda, M, Fukuda, I,Yokomizo, K, Suzuki, K, Sano, K, Nakanishi,T,Tomisato,W, Tsuchiya,T, Mizushima,T. Induction of claudin-4 by nonsteroidal anti-inflammatory drugs and its contribution to their chemopreventive effect. Cancer Res. 65: 1868-1876, 2005.