Clinical trials in pharmacogenetics and pharmacogenomics - CiteSeerX

PRIMER Pharmacogenetics and pharmacogenomics

PRIMER

Clinical trials in pharmacogenetics and pharmacogenomics: Methods and applications PATRICE P. RIOUX

P

harmacogenetics is the study of the impact of heritable traits on pharmacology and toxicology. Characterization of individual or population traits uses the principles of pharmacology and genetics to gain an understanding of the mechanism of human susceptibility to exogenous chemicals.1 An extension of pharmacogenetics is the discovery that genetic polymorphisms have the potential to affect a drug’s mechanism of action. The interplay of genotype and drug efficacy has been defined as pharmacogenomics. Therefore, one potential application of pharmacogenomics is drug development.2 Ideally, specific genetic polymorphisms can be rapidly identified in parallel with the development of diagnostic procedures to select patients for whom a pharmaceutical agent may be safe and effective or, conversely, contraindicated. Incorporation of these data into clinical trial designs is a recent application of pharmacogenetic and pharmacogenomic principles. This article reviews the principles of pharmacogenetics and pharmacogenomics, discusses published clinical trials of the impact of pharmacogenetics and pharmacoge-

Abstract. Clinical and other aspects of pharmacogenetics and pharmacogenomics are discussed. Pharmacogenetics is the study of the impact of heritable traits on pharmacology and toxicology. An extension of pharmacogenetics is the discovery that genetic polymorphisms have the potential to affect a drug’s action. The interplay of genotype and drug efficacy has been defined as pharmacogenomics. For most drugs, variations in patient response have until recently been considered a result of pharmacokinetic rather than pharmacodynamic differences. However, it now seems that pharmacodynamic variability in humans is large, reproducible, and usually more pronounced than pharmacokinetic variability. Some examples of the impact of pharmacogenomics on pharmacokinetics involve cytochrome P-450 isoenzymes, dihydropyrimidine dehydrogenase, and thiopurine methyltransferase; some examples of the impact on pharmacodynamics involve cholesteryl ester transfer protein, angiotensin-converting enzyme, and serotonin transporter. There are no specific statistical techniques for analyzing data from pharmacogenomic clini-

nomics on pharmacokinetics and pharmacodynamics, and introduces technical and strategic issues for designing clinical trials that take human

PATRICE P. RIOUX, M.D., PH.D., is Consultant, Clinical Pharmacogenomics and Pharmacoepidemiology, PMB 214, 405 Waltham Street, Lexington, MA 02421-7934 ([email protected]). The contributions of Marilyn Gosse and James Basilion (for review and comment) and of Lynn Sutherland (for editing) are gratefully acknowledged. This is article 204-000-00-008-H04 in the ASHP Continuing Edu-

cal trials. However, a tabulated relationship for the determination of the maximum possible gain in response rate for the highestresponding genotypic subgroup of patients is provided as an aid to determining whether it is worth having a pharmacogenomic strategy for a given drug. Ethical issues in pharmacogenomics tend to be based on the general concern that the ability to diagnose a genetic disorder before any treatment is available does more harm than good to the patient. Pharmacogenomic approaches to drug discovery and delivery have been recognized by FDA. Pharmacogenomics cannot improve the efficacy of a given drug, but it helps in selecting patients who are likely to respond well. Pharmacogenomics provides a view of drug behavior and sensitivity useful to improving the efficacy of drug development and utilization. Index terms: Clinical studies; Drug administration; Drugs; Ethics; Food and Drug Administration (U.S.); Pharmacogenetics; Product development; Regulations; Toxicity Am J Health-Syst Pharm. 2000; 57:887-901

genetic variations into account. Genetics and genomics Definitions. The phenotype is the

cation System; it qualifies for 2.0 hours of continuing-education credit. See page 899 or http://ce.ashp.org for the learning objectives, test questions, and answer sheet. Copyright © 2000, American Society of Health-System Pharmacists, Inc. All rights reserved. 1079-2082/00/0501-0887$06.00.

Am J Health-Syst Pharm—Vol 57 May 1, 2000

887

PRIMER

Pharmacogenetics and pharmacogenomics

The Primer section covers basic information in various fields of knowledge of interest to pharmacists who practice in health systems. Within the scope of the section are reviews of fundamental concepts in, for example, pharmacy, pharmaceutics, pharmacology, physiology, therapeutics, and health care technology. Also covered are topics somewhat out of the mainstream of pharmacy (e.g., advances in nondrug health care technology) but nevertheless of interest to practitioners.

outward, physical manifestation of a trait that can be observed in an organism. In contrast, the genotype is the internally coded, heritable information present in all living organisms. The complete set of instructions for making an organism is called its genome. The human genome consists of tightly coiled threads of deoxyribonucleic acid (DNA) and associated protein molecules that are organized into 23 physically distinct microscopic units called chromosomes. A DNA molecule consists of two strands, each strand being a linear arrangement of repeating units called nucleotides, each of which is composed of one sugar, one phosphate, and one nitrogenous base. Four nitrogenous bases are present in DNA: adenine (A), thymine (T), cytosine (C), and guanine (G). The particular order of the bases along the sugar– phosphate backbone is called the DNA sequence. The two DNA strands are held together by weak bonds between the bases on each strand; these paired bases are termed base pairs. The human genome contains roughly 3 billion base pairs (bp). During cell division, the DNA molecule unwinds and the bonds between the base pairs break, allowing the strands to separate. Each strand then directs the synthesis of a complementary new strand according to strict base-pairing rules: Adenine pairs only with thymine (forming an A–T pair), and cytosine pairs only with guanine (a C–G pair). Each DNA molecule contains many genes, or defined stretches of DNA with a specific sequence of nucleotide bases containing instructions for the synthesis of a protein. The human ge-

888

nome is estimated to comprise 80,000–100,000 genes, 3,000–10,000 of which may be potential targets for drug discovery. Human genes vary widely in length, often extending over thousands of bases, but only about 5% of the genome is known to include the protein-coding sequences—exons— of genes. For a gene’s instructions to be expressed in the cell, they must be decoded. Interspersed within many genes are intron sequences, which have no protein-coding function. The rest of the genome is thought to consist of other noncoding regions whose functions to date are unknown. Within the gene, a specific sequence of three DNA bases—a codon—directs the cell’s proteinsynthesizing machinery to add a specific amino acid. Because the chromosomes are in the cell nucleus and the sites of protein synthesis, the ribosomes, are in the cytoplasm, the decoding of the information in DNA to synthesize proteins involves two stages, transcription and translation. Transcription is the process in which a segment of one of the two strands of genomic DNA is used as a template to synthesize a singlestranded ribonucleic acid (RNA) molecule. The first step of transcription is the binding of a dedicated enzyme, an RNA polymerase, to specific short sequences within a DNA segment called promoter regions. Promoters (and enhancers and silencers in eukaryotes, including humans) are sequences responsible for modulating gene expression. The stage after transcription is called translation. However, before translation can take place, RNA transcripts are processed by the addition of chemicals at either end (caps and


tails) and the deletion of noncoding intron sequences (splicing). The processed RNA chains, called messenger ribonucleic acid (mRNA), are transported to ribosomes where the cell’s protein-synthesizing machinery translates the codons into a string of amino acids that will constitute the protein molecule. In the laboratory, mRNA can be isolated and used as a template to synthesize a complementary DNA (cDNA) strand, which can then be used to locate the corresponding gene on a chromosome map. Differences in DNA sequences. DNA sequences can be altered at many levels, from major alterations in the genome to minor changes in the sequence of nucleotides in a particular gene. A few types of major DNA abnormalities, including missing or extra copies of a chromosome and gross breaks and rejoinings (translocations), can be detected by microscopic examination. Most changes in DNA, however, are too subtle to be detected by this technique and require specific molecular analysis. Molecular DNA abnormalities, or mutations, are responsible for many inherited diseases, such as cystic fibrosis, sickle cell anemia, and Huntington’s disease, and may predispose an individual to cancer, major psychiatric illness, and other complex diseases. Yet most of the abnormalities in DNA sequences that are passed from generation to generation have no visible clinical impact and are detected only by generalized analysis of sequencing. A polymorphism is a stable difference in DNA sequence at the same locus (a specific position in the genome) among individuals. The differing DNA sequences at a locus are known as alleles. For a given gene, there may be sequence differences in the promoter, coding, noncoding, or untranslated region. For example, a known DNA sequence in one individual may differ in others by one or


more base pairs. These differences in base pairs may result in different amino acids (if the base-pair differences are in the coding region), intron and exon splice sites (if in the borders of introns and exons), or stability of an mRNA (if in a 5′ or 3′ untranslated region). Single-nucleotide polymorphisms (SNPs) are the most frequent type of variation in the human genome; as markers, they provide powerful tools for a variety of medical genetics studies, such as genetic linkage studies. Linkage is a measure of proximity of two or more markers on a chromosome. The closer the markers, the lower the probability that they will be separated during DNA repair or replication (mitosis or meiosis in eukaryotes) and hence the greater the probability that they will be inherited together. A genetic variation occurring in more than 1% of a population would be considered a useful marker for genetic linkage analysis. Association of a polymorphism with expression of a particular phenotypic trait can help in identifying the location of the gene responsible for that trait. If a mutation event introduces a new allele, the mutation will tend to be inherited along with other markers (e.g., SNPs) at tightly linked loci. The tendency for markers to be coinherited, called linkage disequilibrium, declines over time through a natural shuffling of the genome during cell division. Linkage disequilibrium has been used to map disease genes and can provide information on the history of a population. In general, the distribution of an allele in a population varies around the world. Linkage disequilibrium varies among populations, being absent in some and highly prevalent in others.3 Homozygosity and heterozygosity. An individual has two genes for each phenotype, and these genes are defined as his or her genotype for the phenotype. When the two genes are identical (both genes have the same allele, AA or aa for example), then the

genotype is said to be homozygous; otherwise, it is said to be heterozygous (e.g., Aa). If individuals with genotype Aa are phenotypically the same as individuals with genotype AA but different from individuals with genotype aa, then allele A is said to be dominant to allele a, or, equivalently, allele a is said to be recessive to allele A. In addition, the phenotype associated with AA is said to be dominant, and the phenotype associated with aa is said to be recessive. If Aa yields a phenotype that is different from the phenotypes of both AA and aa, then alleles A and a are said to be codominant. The Hardy-Weinberg equilibrium equation. The biological sciences now generally define evolution as the sum total of the genetically inherited changes in the individuals who are the members of a population’s gene pool. Although evolution affects individuals, it is the population as a whole that actually evolves. Evolution is simply a change in the distribution of alleles in the gene pool of a population. This definition of evolution was developed largely as a result of independent work in the early twentieth century by Godfrey Hardy, an English mathematician, and Wilhelm Weinberg, a German physician. Through mathematical modeling, they concluded that allele frequencies in a gene pool are inherently stable but that evolution should be expected in all populations virtually all of the time.4 Hardy and Weinberg went on to develop a simple equation that can be used to relate genotype distribution to allele frequency in a population and to track changes from one generation to another. Consider a trait controlled by a pair of alleles, A and a. Let p be the allele frequency for A and q the allele frequency for a. Because there are only two alleles, the frequency of one plus the frequency of the other must equal 100%, which is to say p + q = 1. The probability of all possible combinations of alleles

occurring randomly in a population can be described by the Hardy-Weinberg equilibrium equation: (p + q)2 = 1 OR p2 + 2pq + q2 = 1

In this equation, p2 is the frequency of homozygous AA people in a population, 2pq is the frequency of heterozygous Aa people, and q2 is the frequency of homozygous aa ones. So, the Hardy-Weinberg equation describes the distribution of all three genotypes for a selected trait within the population (Figure 1). Clinical data Pharmacokinetics is the study of the conditions that affect how the body acts on drugs, and pharmacodynamics is the study of the conditions that affect how drugs act on the body. Drugs are altered in the liver to increase the rate at which they are eliminated from the body, and metabolism is frequently the rate-limiting step in drug clearance. The source of large interindividual metabolic variation seems to be differential expression of several allelic forms of hepatic drug-metabolizing enzymes. Many drug-metabolizing enzymes exhibit polymorphic expression (e.g., acetyltransferases and sulfotransferases), but the principal enzymes causing hereditary variations in drug metabolism found in the liver belong to the cytochrome P-450 (CYP) isoenzyme system. Polymorphic expression of at least three of these isoenzymes divides the human population into two groups: “poor” metabolizers, whose genes express dysfunctional or inactive enzymes, and “extensive” metabolizers, whose genes express enzymes with normal activity. Poor metabolizers account for 3–7% of whites1; the distribution of these polymorphisms can be very different in other ethnic groups. Furthermore, the magnitude of variation in drug metabolism between poor metabolizers and extensive metabo-


889


lizers is normally not trivial; the rate at which many drugs are metabolized may vary 10-fold to 100-fold between these two groups of people. For a majority of drugs, variations in patient response have until recently been considered a result of pharmacokinetic rather than pharmacodynamic differences; that is why most publications in pharmacogenetics or pharmacogenomics still address drug metabolism. However, it now seems that pharmacodynamic variability in humans is large, reproducible, and usually more pronounced than pharmacokinetic variability. Thus, after reviewing some typical examples of the impact of pharmacogenomics on pharmacokinetics (CYP isoenzymes, dihydropyrimidine dehydrogenase, and thiopurine methyltransferase), this article will focus on the limited examples of pharmacogenomics affecting pharmacodynamics (cholesteryl ester transfer protein, angiotensin-converting enzyme, and serotonin transporter). CYP isoenzymes. Drugs and other xenobiotics are usually eliminated from the body after metabolic conversion in two steps. The first step involves enzymes, often CYP isoenzymes, that modify functional groups on the drug molecules. In the second step, conjugation enzymes make the modified drugs more water soluble for efficient elimination via the kidneys. In some cases, CYP isoenzymes activate the parent compound, which may then become toxic or carcinogenic. The CYP isoenzyme system is adaptable to the environment, and specific isoenzymes are inducible in response to increased plasma levels of a drug. To date, nearly 50 human CYP genes have been identified participating in drug and xenobiotic metabolism (36–39 sequenced CYP genes and 10 pseudogenes); however, only about 8 of them are quantitatively important in drug metabolisms.1 CYP2D6 and CYP3A4 metabolize a majority of clinically used drugs. Some SNPs render a CYP isoen-

890

Figure 1. Distribution of genotypes AA, Aa, and aa in a population, according to the HardyWeinberg equilibrium equation.

1

Frequency of Genotype

PRIMER

aa

0.8

0.6

0.4

0.2

0

Aa AA 0

0.2

0.4

0.6

0.8

1

Frequency of Allelle A

zyme inactive, and people with the variant CYP gene can have impaired drug metabolism. This causes an increased plasma level of the drug at ordinary dosages, increasing the risk of adverse effects. Similarly, some SNPs increase the activity of a CYP isoenzyme, resulting in very rapid metabolism of drugs and failure to reach therapeutic levels at ordinary dosages—a problem that might be misinterpreted as poor compliance. Polymorphisms in genes encoding CYP isoenzymes that metabolically activate chemicals may explain the differences seen in individual susceptibility to carcinogenicity and toxicity from several chemical agents. Traditionally, the determination of polymorphisms in CYP genes was done phenotypically by determining the metabolism of a drug (e.g., debrisoquin and sparteine) known to be specifically metabolized by one isoenzyme. More recently, genomic determination of CYP alleles has allowed for identification of poor metabolizers and extensive metabolizers as effectively as phenotypic determination. For example, CYP2D6 genotyping and debrisoquin testing of


1053 Caucasian and African-American patients with lung cancer and control individuals were done to assess phenotype–genotype relationships.5 The clinically adjusted data showed positive identification of 88% of phenotypic poor metabolizers by genotyping. This sensitivity was comparable to that of phenotyping, which identified as poor metabolizers 90% of the patients with two variant alleles. While there are difficulties in detecting through genotyping all people whose CYP2D6 functions at a poor rate, phenotypic identification of poor metabolizers also has limitations, including interference by other drugs (e.g., cimetidine, warfarin, omeprazole, and antidepressants) and the need to ensure compliance. Therefore, it is likely that genotyping will be the method of choice in clinical trials. Dihydropyrimidine dehydrogenase. Dihydropyrimidine dehydrogenase (DPD) is the initial and rate-limiting enzyme in the catabolism of fluorouracil, a reaction that occurs mainly in the liver. DPD deficiency causes a toxic buildup of fluorouracil,


along with pyrimidinemia and pyridinuria. Diasio et al.6 conducted a familial study that suggested an autosomal recessive pattern of inheritance of DPD deficiency. However, it seems that a complete absence of DPD activity is extremely rare and that even partial enzyme inactivity might result in severe toxicity from fluorouracil. The impact of DPD deficiency is hard to delineate precisely because it depends on determination of the DPD activity of peripheral blood mononuclear cells. Milano et al.7 recently showed, in a group of 19 new cases of fluorouracil-induced toxicity associated with DPD deficiency, that women are particularly prone to this deficiency. Moreover, the researchers found that lymphocytic DPD activity was, on average, 15% lower in women than in men and that this difference in DPD activity was of the same order of magnitude as that observed for fluorouracil clearance between men and women. Van Kuilenburg et al.8 found the highest activity of DPD in monocytes, followed by lymphocytes, granulocytes, and platelets. A strong positive correlation was observed between the DPD activity of peripheral blood mononuclear cells and the percentage of monocytes, thus introducing a large interpatient and intrapatient variability in the activity of DPD and hindering the detection of patients with partial DPD deficiency. A molecular basis for DPD deficiency has been suggested: A G-to-A mutation in the 5′ splicing site on the gene encoding DPD causes the exon preceding the mutation to be skipped rather than transcribed, leading to an inactive enzyme.9 However, it is not certain that genotyping will in this case constitute a better solution than the current phenotyping approach, since different subjects genotyped as heterozygous for the mutation may exhibit highly variable lymphocytic DPD activity among themselves. Polymorphism of the gene encod-

ing DPD is a typical example of what could be a useful pharmacogenomic approach to preventing toxicity from a very effective drug that normally has a high level of toxicity. Thiopurine methyltransferase. Thiopurine methyltransferase (TPMT) is a cytosolic enzyme whose precise physiological role is unknown. It catalyzes the methylation of immunosuppressive or cytotoxic thiopurine drugs, such as thioguanine, mercaptopurine, and azathioprine.10 The in vivo activity of this enzyme is characterized by interindividual and interethnic variability caused by polymorphism of the TPMT gene. Through the use of pharmacogenetic techniques, three major TPMT phenotypes have been identified: high, intermediate, and deficient methylators. As a result, individuals differ greatly in detoxification of thiopurine drugs to 6-methylmercaptopurine, which is correlated with adverse effects and therapeutic efficacy. Spire-Vayron de la Moureyre et al.11 used pharmacogenomic techniques to define the mutational and allelic spectra of the TPMT gene in 191 Europeans. The samples were screened for mutations in the entire cDNA and the exon–intron boundaries, promoter region, and 3′-flanking region of the gene. Six mutations were detected in 10 exons, and 7 TPMT alleles were characterized. Within the promoter region, six alleles corresponding to a variable number of tandem repeats (short sequences of DNA repeated several times, corresponding to highly variable regions of human DNA) were identified. The TPMT phenotype was correctly predicted by genotyping for 87% of individuals. A clear negative correlation between the total number of tandem repeats in the promoter region and the TPMT activity level was observed, indicating that a variable number of tandem repeats contributes to interindividual variations in TPMT activity. The results of the TPMT study are

good examples of the difference between pharmacogenetics, which demonstrates the relationship between genetic polymorphisms and toxicity, and pharmacogenomics, which uses new molecular techniques to explain the impact of the whole genome on toxicity and efficacy. This example also emphasizes that analyzing cDNA is not sufficient by itself for predicting phenotype, as the major genotypic differences resulting in altered phenotypes were manifest in the gene’s promoter region (which is not part of the cDNA). This study clearly points out that pharmacogenomic analysis, as opposed to cDNA analysis, is necessary for looking at introns and the promoter region. Also, pharmacogenomics shows that drug toxicity is generally linked to a change in an amino acid related to a polymorphism in exons, whereas variations in drug efficacy—more difficult to see because of more continuous changes—are mostly linked to polymorphisms in the 5′-flanking promoter region. The promoter region is suspected to contribute to the regulation of gene transcription. Cholesteryl ester transfer protein. Phenotype analysis has shown that high-density-lipoprotein (HDL) cholesterol concentration is inversely related to the risk of coronary artery disease. Cholesteryl ester transfer protein (CETP) has a central role in the metabolism of HDL cholesterol and may therefore be one of the underlying genetic factors in susceptibility to atherosclerosis. The DNA of 807 men with angiographically documented coronary atherosclerosis who had participated in a clinical trial was analyzed for a polymorphism in the gene encoding CETP.12 This DNA variation was referred to as B1 and its absence as B2. Patients had been randomly assigned to pravastatin treatment or placebo for two years. From DNA samples collected before treatment, researchers later found that patients with genotype B1B1 had higher plasma CETP


891


concentrations and lower HDL cholesterol concentrations than patients with genotype B2B2. Among patients in the placebo group, those with the B1 allele had greater decreases in mean luminal diameter, which meant greater progression of diffuse atherosclerosis (Figure 2). Specifically, patients with genotype B1B1 had the greatest disease progression, those with genotype B2B2 had the least progression, and those with genotype B1B2 had an intermediate degree of progression. In the pravastatin group, the B1 allele was associated with less progression of diffuse atherosclerosis (i.e., smaller decreases in luminal diameter). Patients with two B1 alleles benefited most from pravastatin. They had significantly less progression of coronary atherosclerosis than their B1B1 counterparts in the placebo group. Furthermore, patients with the B1B2 genotype who were receiving pravastatin had significantly less focal atherosclerosis than their placebo-group counterparts. Both the association of the CETP genetic polymorphism with the decrease in mean luminal diameter or minimal luminal diameter in the placebo group and the interaction between the genotype and pravastatin treatment remained significant after adjusting for luminal diameter at baseline, HDL cholesterol concentration at baseline, changes in HDL cholesterol concentration, and activities of hepatic lipase and lipoprotein lipase. The molecular mechanism that underlies the relationship between the CETP gene variant and the angiographic response to pravastatin treatment cannot be deduced from this study. However, the mechanism may be related to plasma concentrations of CETP.12 Although pravastatin significantly reduced CETP concentrations, CETP concentrations at baseline differed among the genotype subgroups, and this may account for the effects observed. The CETP gene variant provides

892

Figure 2. Changes in the average of the mean luminal diameter among men with coronary atherosclerosis who were randomly assigned to receive pravastatin (circles) or placebo (squares) and who were later identified by genotyping for a polymorphism in the gene encoding the cholesteryl ester transfer protein. B1 is the variant allele, and B2 is the normal allele. The error bars show the 95% confidence intervals for the changes. Reprinted from reference 12, with permission.

0.20 Decrease in Mean Luminal Diameter (mm)

PRIMER

0.15

0.10

0.05

0.00

–0.05

B1B1

an example of a recessive SNP in the noncoding region with a negative impact on drug efficacy and activity, as well as a negative impact on the disease itself. Angiotensin-converting enzyme. Angiotensin-converting-enzyme (ACE) inhibitors have been shown to reduce mortality in patients with both symptomatic and asymptomatic left ventricular dysfunction and after acute myocardial infarction. An insertion/deletion polymorphism (consisting of a 287-bp alu repeat sequence) in intron 16 of the DCP1 gene, which encodes ACE, has been shown to predict approximately half of the statistical variance in serum ACE levels among individuals.13 People homozygous for the deletion allele (DD) have serum ACE levels twice as high on average as those homozygous for the insertion allele (II), whereas heterozygous (ID) people have intermediate levels. Todd et al.14 demonstrated that, in men, genotype continued to predict residual ACE activity after inhibition of the enzyme with a single oral dose


B1B2 Genotype

B2B2

of enalapril. In what could be considered a typical pharmacogenomic Phase I trial, Ueda et al.15 compared two groups of homozygotic volunteers, 12 with genotype DD and 11 with genotype II, identified from genotyping 200 healthy normotensive men. The researchers found that the effect of an i.v. dose of enalaprilat was significantly greater and lasted longer in the group with genotype II (Figure 3). However, in a recent study of the effect of ACE genotype on the longterm (six-week) response to ACE inhibitor therapy in patients with heart failure, the DD genotype did not contribute significantly to the inability of about 30% of the patients to improve with the ACE inhibitors.16 Serotonin transporter. Serotonin transporter (5-HTT) plays a critical role in the termination of serotonin neurotransmission and represents the prime target for selective serotonin-reuptake inhibitors (SSRIs). A functional polymorphism in the transcriptional control region upstream of the 5-HTT-coding se-


Figure 3. Extent of inhibition of angiotensin-converting enzyme among normotensive men with genotypes II and DD. Dose ratio for angiotensin I was defined as the dose of angiotensin I required to raise a person’s mean blood pressure by 20 mm Hg after an i.v. infusion of enalaprilat compared with the dose of angiotensin I required to raise the person’s mean blood pressure by 20 mm Hg after a dose of placebo. Compared with men with genotype DD, men with genotype II had greater dose ratios for angiotensin I at 1 and 10 hours (p = 0.003 and p = 0.001, respectively, Mann-Whitney U test) after the dose of enalaprilat or placebo. The horizontal lines show the means for the dose ratios. Reprinted from reference 15, with permission.

Dose Ratio for Angiotensin I

20

10 hr

1 hr

15

10

5

1 0 II

DD

II

DD

Genotype

quence has been reported.17 It consists of a 44-bp insertion (long variant) or deletion (short variant). The long (l) and short (s) variants of this polymorphism affect transcriptional efficiency. In vitro studies showed that the difference in synthesis of 5HTT mRNA affected expression of 5HTT and uptake of serotonin in lymphoblasts. Smeraldi et al.18 recently reported that fluvoxamine’s efficacy in the treatment of major depression with psychotic features seems to be related to allelic variation in the promoter region of the gene coding for 5-HTT. Both patients homozygous for the long variant (ll) of the 5-HTT promoter region and heterozygous patients (ls) had a better response to fluvoxamine than patients homozygous for the short variant (ss). Adding pindolol (a mixed β-adrenergic receptor and serotonin type 1A receptor [5-HT1A] antagonist) has been proposed as an augmentation therapy for nonresponders and partial responders to SSRIs19; it appears that,

in the group treated with fluvoxamine plus pindolol, all the genotypic subgroups acted like ll patients treated with fluvoxamine alone. This supports the hypothesis that the effect of pindolol is related to its ability to block 5-HT 1A autoreceptors, thus preventing a negative feedback of serotonin at the somatodendritic level and suggesting that activation of 5HT1A autoreceptors could modulate the clinical effect of the SSRI-induced blockade of 5-HTT. This 5-HTT polymorphism is another example of a recessive SNP polymorphism in the noncoding region with a negative impact on drug efficacy. Drug toxicity and efficacy It is important to distinguish between pharmacogenomic approaches that attempt to link polymorphisms to toxicity and those that seek to identify a relationship with efficacy. In the case of toxicity, a polymorphism in cDNA that leads to amino acid changes in a drug-metabolizing

enzyme results in patients, generally with a homozygous genotype, who cannot metabolize the drug at all. In the case of efficacy, the objective is to define a dose–response relationship (dose is the presence of the variant allele) that takes into account the impact of different levels of enzyme activity. This implies that the comparison of gene expression profiles for DNA polymorphisms (including those involving the promoter region, introns, exons, and 3′ and 5′ untranslated regions) is a screening tool that should be used before looking at the clinical impact of drugs on patients. To precisely evaluate the clinical impact of a SNP involved in drug efficacy or toxicity, it is necessary to take into account whether the polymorphism is recessive or dominant. Recessive means that this genotypic variation has a phenotypic impact only when the polymorphism is present in a homozygous form. Dominant means that this genotypic variation has a phenotypic impact as soon as the polymorphism is present on one of the two parental copies of a given chromosome. Also, because of the strict rules of base pairing of nucleotides, a SNP is a biallelic variation. Therefore, a recessive SNP linked to inefficacy can be seen as a dominant SNP linked to efficacy, and a dominant SNP linked to inefficacy can be seen as a recessive SNP linked to efficacy. Theoretically, there should be four distinct possibilities: recessive efficacy, dominant efficacy, recessive toxicity, and dominant toxicity. In the early years of pharmacogenetics, DPD deficiency was considered a typical example of an autosomal recessive trait related to toxicity. More recently, however, the discovery of several polymorphisms on the gene encoding DPD indicates that there is no clear phenotypic cutoff for toxicity. It is important to review simple models addressing the three meaningful pharmacogenomic possibilities with respect to clinical trials: tox-


893

PRIMER


icity, dominant efficacy, and recessive efficacy. Toxicity. Warfarin toxicity. A strong association has been found between polymorphisms in the gene coding for CYP2C9 and warfarin sensitivity. 20 CYP2C9 genotyping may identify a subgroup of patients who are poor metabolizers of warfarin and thus will have difficulty at induction of anticoagulant therapy and are potentially at higher risk of bleeding. Thiopurine toxicity. Evaluating the phenotypic status of patients by measuring the TPMT activity of red blood cells has become a clinical test performed routinely before thiopurine therapy is begun and in the monitoring of treatment. The protocols for measuring TPMT activity and problems with correctly evaluating the test results when patients have received a transfusion of red blood cells limit the usefulness of phenotyping procedures. However, the large interindividual variability in enzyme activity has been mainly explained by genetic factors, and genotyping tests have been recently proposed to avoid these limitations. Recent publications demonstrate the accuracy of the genotyping method for predicting TPMT phenotype to identify patients at highest risk of thiopurine toxicity.21,22 Fluorouracil toxicity. Identifying patients with polymorphisms in the gene encoding DPD is another example of a useful pharmacogenomic approach for preventing toxicity with fluorouracil. Although prospective studies of random populations indicate that complete or partial DPD deficiency is a rare event, and although no one has identified a completely DPD-deficient subgroup, more than 10 important SNPs that could have a metabolic impact have been described in the literature. SNPs responsible for changes in the amino acid sequence of the enzyme and other SNPs not related to protein changes had a clinical impact on toxicity.

894

Although this type of pharmacogenomic approach will present challenges for the pharmaceutical industry and the Food and Drug Administration (FDA) in deciding on definitive labeling for adverse effects, it is clear that identification and avoidance of a population with a tendency toward drug toxicity could help maximize therapeutic benefit. Dominant efficacy. Evaluation of the clinical impact on drug efficacy of polymorphisms in the genes encoding CETP, ACE, and 5-HTT led to my identification of two groups of drug responders: (1) homozygous patients for whom the drug was no better than placebo, corresponding to a trait of “recessive inefficacy,” and (2) heterozygous patients for whom the drug was better than placebo but still less effective than in the wild-type group of homozygous patients, corresponding to a trait of “dominant efficacy.” These pharmacogenomic approaches seem useful for clinicians and public health professionals trying to identify and therefore not treat groups of nonresponders. The results of pharmacogenomic studies could help avoid unnecessary health care expenditures, undesirable or potentially deleterious drug effects, and drug interactions. Furthermore, identification of these polymorphisms could help clinicians adjust dosages when there is a clear clinical difference between heterozygous and homozygous individuals for a given SNP (e.g., level of expression of TPMT linked to a polymorphism in the gene’s promoter region). These studies can lead to enhancement of the quality of medical therapy. Recessive efficacy. The results of a Phase II trial23 and the negative findings of a Phase III trial announced one year later provide an example of the benefit that can be derived from identifying a SNP linked to recessive efficacy with a high allele frequency. Substance P is a neurotransmitter believed to be involved mainly in pain. The localization of substance P


in brain regions that coordinate stress responses and receive convergent monoaminergic innervation suggests that substance P antagonists might have psychotherapeutic properties.24 Substance P is highly metabolized by ACE. The insertion/deletion polymorphism in intron 16 of the DCP1 gene, which encodes ACE, provides an excellent pharmacogenomic model: The polymorphism has a high allele frequency in healthy individuals (D, 34%; I, 66%), and there are clinical studies clearly demonstrating its impact on ACE inhibitors.13,14 Moreover, it has been found that DD homozygous patients (11% of patients) have a higher brain concentration of substance P than II homozygous patients (43%); the concentration in heterozygous patients (46%) is intermediate.25 Given the results of the initial Phase II trial, we can reasonably hypothesize that a substance P antagonist will have more impact on patients with high brain levels of substance P and specifically the DD patients who are more at risk for affective disorders. In the Phase II trial, 54% of patients treated with the substance P antagonist responded (≥50% change from baseline to week 6 in total Hamilton Depression Rating Scale score), and 28% of the patients in the placebo group responded. Now, consider the results of the failed Phase III trial of the substance P antagonist. There was a high placebo group response rate: 54% of the treatment group patients responded, versus 48% of the placebo group patients; that is, a 6 percentage-point, nonsignificant difference in response rates. Let us hypothesize that the ACE polymorphism (allele I) is a dominant SNP for substance P metabolism. With these assumptions, we calculate that the patients with genotype DD must have had a response rate of about 100% and that the patients with genotypes II and DI had response rates similar to those of the placebo group (appendix). This pharmacogenomic approach


can be useful, in drug development, for targeting subgroups of patients genetically predisposed for high drug efficacy. However, to achieve maximum benefit, this approach has to be based on the frequency of the desirable allele and the difference between treatment and placebo already obtained without any pharmacogenomic hypothesis. For instance, if the initial difference between treatment and placebo is low (5 percentage points), the allele frequency potentially linked to a recessive efficacy to test has to be less than 60% (9 percentage-point gain) to be useful but could not have been less than 30% to achieve the initial 5 percentage-point difference between treatment and placebo (Table 1). When the global difference is 20 percentage points, this allele frequency has to be less than 80% (11 percentage-point gain) to be useful but could not have been less than 50% to achieve the initial 20 percentagepoint difference between treatment and placebo. Moreover, for various reasons (e.g., clinicians focusing on toxicity and inefficacy, pharmaceutical companies interested in developing drugs working well for everybody instead of very well for subgroups of patients), none of the pharmacogenomic examples already published corresponds to these kinds of poly-

morphisms linked to recessive efficacy. Statistical issues Unlike genetic epidemiology and routine clinical trials, pharmacogenomic clinical trials use no specific statistical techniques for analyzing data. Therefore, it is important to discuss an analytical strategy for pharmacogenomic studies. Safety. The primary objective of a pharmacogenomics-based safety study is to compare the frequency distribution of polymorphisms in candidate genes between two groups of patients: one with high toxicity (cases) and the other with low or minimal toxicity (controls). To make a sample-size calculation, data are generally treated as independent 2 × 2 contingency tables, one table for each polymorphism in the cases and controls. The statistical significance of the difference between polymorphism frequencies can be assessed by a Pearson chi-square test of homogeneity of proportions, with n – 1 degrees of freedom (where n is the number of polymorphisms in the gene), an a priori level of significance of 0.05, and a power of 80%. Then, to determine which polymorphism is responsible for a significant finding, each polymorphism is compared

with the rest. This results in chisquare tests that are individually valid but that in totality constitute a form of multiple testing. Bonferroni’s procedure (α/n) is used to adjust the selected level of significance for multiple testing. Efficacy. A general procedure to use in a clinical trial examining whether drug efficacy is related to a pharmacogenetic effect is as follows: (1) determine a set of candidate genes (look first at the drug-related pathophysiological pathway and eventually at the disease or some other criterion), (2) select the genetic polymorphisms with allele frequency higher than 20%, and (3) when possible, look at already completed Phase III and Phase IV trials to determine if these polymorphisms could affect drug efficacy. The same statistical techniques as described for toxicity can be used. The main issue to resolve is multiple testing: If there is no correction (e.g., Bonferroni’s procedure), false-positives may be problematic; if multiple corrections are made, there is a risk of missing valid hypotheses. To confirm that a polymorphism has a pharmacologic effect, first set up a Phase I or Phase IIa trial comparing two homozygous groups by using a surrogate biological marker

Table 1.

Maximum Possible Percentage-Point Gain in Treatment Response Rate for Subgroup Homozygous for a Polymorphism Defined by a Biallelic Locus Not Linked to Disease under Treatmenta Percentage Point Difference in Response Rates between Treatment and Placebo Groups 5 10 15 20 25 30 40

10/90 1 2 4 5 6 7 9

20/80 3 6 8 11 14 17 23

Allele Frequency (%)b 30/70 50/50 40/60 5 10 16 21 26 31 42

9 18 27 36 44 53 71

15 30 45 60 75 90

60/40

70/30

26 53 79

51

a

See appendix. The first number in the pair is the allele frequency for a polymorphism associated with dominant inefficacy; the second number is the allele frequency for recessive efficacy. b


895


(like enzyme activity). The main issue here is the number of individuals to genotype (healthy volunteers for Phase I trials, patients for Phase IIa trials) in order to get enough homozygous individuals to show a significant result; this number is related to the polymorphism’s allele frequency. In some cases, it will not be difficult to recruit a large number of healthy volunteers. In other cases, it may be impractical (Figure 4). Second, set up a sequential Phase II trial analyzing the impact of all homozygous and heterozygous groups on drug efficacy. Finally, perform a Phase III trial with patient genotyping for a very limited number of polymorphisms as inclusion criteria in order to validate the pharmacogenomic approach for the drug, as well as the diagnostic test. Sample-size issues in Phase II and Phase III trials are discussed by Elston et al.26 This is mainly a proactive approach, whereby a pharmaceutical company can establish a subgroup in which the drug is most effective in case it is not effective in the larger population. Ethics Ethical issues in pharmacogenomics tend to involve the general concern that the ability to diagnose a genetic disorder before any treatment is available does more harm than good to the patient. It may create anxiety and fear and could have a negative impact on insurance policies and employment. Nonetheless, geneticists have isolated several disease-causing gene mutations and have studied them in great detail without a treatment being available. However, in pharmacogenomics, we are normally only looking at genes in relation to a drug’s metabolism or mechanism of action. The accepted strategy in clinical trials to date is to include an informed-consent form that is either an open statement or, more realistically, a finite list of genes to test.27

896

Figure 4. Sample size for genotype testing. The number of people needed for genotyping is a function of the number of people in the smaller of the two groups of homozygous patients and the frequency of the corresponding allele (q).

No. People Required for Genotype Testing

PRIMER

1000 800

600 400 50 40 30 20 10

200

0 0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5

Allele Frequency (q)

Increasingly, drugs are designed to take into account the molecular basis of diseases. Therefore, polymorphisms related to drugs could also be related to diseases. For example, a person’s apolipoprotein E genotype could have an impact on drug treatment by acetylcholinesterase inhibitors but also seems to be a risk factor for early onset of Alzheimer’s disease.28 The ethics of genetics is too vast a subject to discuss in detail in this article. For more information, the reader is referred to the “Proposed International Guidelines on Ethical Issues in Medical Genetics and Genetic Services” (www.who.int/ncd/hgn/ hgnethic.htm), the report of a 1997 World Health Organization meeting. Regulations The recent approval by FDA of trastuzumab for marketing is an indication that pharmacogenomic approaches to drug discovery and delivery are being recognized. Trastuzumab, a monoclonal antibody for the treatment of late-stage breast cancer, inhibits the action of the protein encoded by the HER2 gene. Some 10–40% of all breast cancer patients overexpress HER2. 29


These are the women who will benefit from treatment with the drug. The drug’s manufacturer has already announced a deal with a diagnostics company to develop a diagnostic screening test for HER2; competitors are close behind. Pushed forward by public advocacy, the trastuzumab story is a striking example of how identifying patientpopulation subsets can bring a new measure of safety and efficacy: It takes us out of the days of poisoning the patient and hoping that the tumor dies first and moves us into a time when tumor targeting could be effected without debilitating, or even life-threatening, adverse effects. Strategies Which disease indications are the most appropriate candidates for a pharmacogenomic approach? Which stages of drug development? Indications. A tremendous amount of effort is required to submit regulatory paperwork for a drug, along with a genetic test, to FDA. Therefore, this effort should be undertaken only for drugs for life-threatening or chronic diseases. Such justification is particularly important when the impact of a drug cannot be evaluated


early in the course of the disease it is intended to treat (e.g., Alzheimer’s disease, schizophrenia, multiple sclerosis, diabetes mellitus, obesity). Pharmacogenomics cannot improve the efficacy of a given drug; it simply helps in selecting patients who are likely to respond well. The main interest is in identifying patients for whom we can predict drug efficacy and in sparing patients from avoidable adverse effects. To get this information in the case of antineoplastic therapy, we set up a trial with a tumor chemosensitivity assay (which is based on tumor genotyping) in parallel with genotyping of patients for the assessment of toxicity.30 Treating tumors and patients differently in this manner is a strategy that seems very promising.31 Pharmacogenomics should help with compliance: Patients who are told that genetic testing has identified them as ideal candidates for a drug are likely to be compliant with that therapy, especially when, as predicted, they respond to it. Society benefits too, when the use of expensive drugs is avoided in patients who clearly would never have responded to them. Development strategy. It is important to resolve whether a drug should be designed for everyone (independent of potential known pharmacogenomic interactions) or if a clinical form of the disease defined by genetic criteria should be targeted. From an economics standpoint, if the pharmaceutical company could benefit from something like orphan drug status for its product, this would help encourage stratification of populations on the basis of pharmacogenomics, since the reduction in the size of a population to be treated could be offset by preference for the drug. Otherwise, pharmaceutical companies will generally not be encouraged to develop drugs for niche markets. If the initial target is a drug that works for the whole population, it is

useful to start pharmacogenomic analysis at the Phase II level in order to reach a decision, as early as possible in the development process, on whether to proceed. It also helps to identify a group of true responders for later phases of drug development. If the target is a drug that works in a genetic subset of the population starting with Phase I studies, finding homozygous patients is important before going further. Conclusion Pharmacogenomics provides a view of drug behavior and sensitivity useful to improving the efficiency of drug development and utilization.

12.

13.

14.

15.

16.

References 1. Weber WW. Pharmacogenetics. New York: Oxford Univ. Press; 1997:344. 2. Houseman D, Ledley F. Why pharmacogenomics? Why now? Nature Biotech. 1998; 16(suppl):2-3. Editorial. 3. Gelernter J, Cubells JF, Kidd JR et al. Population studies of polymorphisms of the serotonin transporter protein gene. Am J Med Genet. 1999; 88(1):61-6. 4. Hardy GH. Mendelian proportions in a mixed population. Science. 1908; 28:49-50. 5. Leathart JBS, London SJ, Steward A et al. CYP2D6 phenotype-genotype relationships in African-Americans and Caucasians in Los Angeles. Pharmacogenetics. 1998; 8:529-41. 6. Diasio RB, Beavers TL, Carpenter JT. Familial deficiency of dihydropyrimidine dehydrogenase. Biochemical basis for familial pyrimidinemia and severe 5-fluorouracil induced toxicity. J Clin Invest. 1988; 81:47-51. 7. Milano G, Etienne MC, Pierrefite V et al. Dihydropyrimidine dehydrogenase deficiency and fluorouracil-related toxicity. Br J Cancer. 1999; 79:627-30. 8. Van Kuilenburg ABP, van Lenthe H, Blom MJ et al. Profound variation in dihydropyrimidine dehydrogenase activity in human blood cells: major implications for the detection of partly deficient patients. Br J Cancer. 1999; 79:620-6. 9. Wei X, McLeod HL, McMurrough J et al. Molecular basis of the human dihydropyrimidine dehydrogenase deficiency and 5fluorouracil toxicity. J Clin Invest. 1996; 98:610-5. 10. Weinshilboum RM, Sladek SL. Mercaptopurine pharmacogenetics: monogenic inheritance of erythrocyte thiopurine methyltransferase activity. Am J Hum Genet. 1980; 32:651-62. 11. Spire-Vayron de la Moureyre C, Debuysere H, Mastain B et al. Genotypic and phenotypic analysis of the polymorphic thiopurine Smethyltransferase gene (TPMT) in a Euro-

17.

18.

19. 20.

21.

22.

23.

24.

25.

26.

pean population. Br J Pharmacol. 1998; 125:879-87. Kuivenhoven JA, Jukema JW, Zwinderman AH et al., for the Regression Growth Evaluation Statin Study Group. The role of a common variant of the cholesteryl ester transfer protein gene in the progression of coronary atherosclerosis. N Engl J Med. 1998; 338:8693. Rigat B, Hubert C, Alhenc-Gelas F et al. An insertion/deletion polymorphism in the angiotensin I-converting enzyme gene accounting for half the variance of serum enzyme levels. J Clin Invest. 1990; 86:1343-6. Todd GP, Chadwick IG, Higgins KS et al. Relation between changes in blood pressure and serum ACE activity after a single dose of enalapril and ACE genotype in healthy subjects. Br J Clin Pharmacol. 1995; 39:131-4. Ueda S, Meredith PA, Morton JJ et al. ACE (I/D) genotype as a predictor of the magnitude and duration of the response to an ACE inhibitor drug (enalaprilat) in humans. Circulation. 1998; 98:2148-53. O’Toole L, Stewart M, Padfield P et al. Effect of the insertion/deletion polymorphism of the angiotensin-converting enzyme gene on response to angiotensin-converting enzyme inhibitors in patients with heart failure. J Cardiovasc Pharmacol. 1998; 32:988-94. Lesch KP, Bengel D, Heils A et al. Association of anxiety-related traits with a polymorphism in the serotonin transporter gene regulatory region. Science. 1996; 274:1527-31. Smeraldi E, Zanardi R, Benedetti F et al. Polymorphism within the promoter of the serotonin transporter gene and antidepressant efficacy of fluvoxamine. Mol Psychiatry. 1998; 3:508-11. Blier P, Bergeron R. The use of pindolol to potentiate antidepressant medication. J Clin Psychiatry. 1998; 59(suppl 5):16-23. Aithal GP, Day CP, Kesteven PJL et al. Association of polymorphisms in the cytochrome P450 CYP2C9 with warfarin dose requirement and risk of bleeding complications. Lancet. 1998; 353:717-9. McLeod HL, Pritchard SC, Githang’a J et al. Ethnic differences in thiopurine methyltransferase pharmacogenetics: evidence for allele speificity in Caucasian and Kenyan individuals. Pharmacogenetics. 1999; 9:773-6. Relling MV, Hancock ML, Rivera GK et al. Mercaptopurine therapy intolerance and heterozygosity at the thiopurine S-methyltransferase gene locus. J Natl Cancer Inst. 1999; 91:2001-8. Kramer MS, Cutler N, Feighner J et al. Distinct mechanism for antidepressant activity by blockade of central substance P receptors. Science. 1998; 281:1640-5. Snider RM, Constantine JW, Lowe JA III et al. A potent nonpeptide antagonist of the substance P (NK1) receptor. Science. 1991; 251:435-7. Arinami T, Li L, Mitsushio H et al. An insertion/deletion polymorphism in the angiotensin converting enzyme gene is associated with both brain substance P contents and affective disorders. Biol Psychiatry. 1996; 40:1122-7. Elston RC, Idury RM, Cardon LR et al. The


897

PRIMER

27. 28.

29. 30. 31.


study of candidate genes in drug trials: sample size considerations. Stat Med. 1999; 18:741-51. Merz JF, Leonard DGB, Miller ER. IRB review and consent in human tissue research. Science. 1999; 283:1647-8. Farlow MR, Lahiri DK, Poirier J et al. Treatment outcome of tacrine therapy depends on apolipoprotein genotype and gender of the subjects with Alzheimer’s disease. Neurology. 1998; 50:669-77. Menard S, Tagliabue E, Campiglio M et al. Role of HER2 gene overexpression in breast carcinoma. J Cell Physiol. 2000; 182:150-62. Support for non-competitive trials. Lancet. 1999; 353:855. Editorial. Basilion J, Schievella A, Burns E et al. Selective killing of cancer cells based on loss of heterozygosity and normal variation in the human genome: a new paradigm for anticancer drug therapy. Mol Pharmacol. 1999; 56:359-69.

Appendix—Determination of the maximum possible treatment response rate for the highest-responding genotypic subgroup of patients based on the response rate in the placebo group Let PR be the placebo response rate and TR be the treatment response rate in a classic placebo-controlled clinical trial of a drug. The observational relative risk, ObsRR, is defined as the ratio between TR and PR: ObsRR = TR/PR

(1)

Suppose that there is a genetic polymorphism in one of the drug-metabolizing enzymes, with each genotypic subgroup of patients having a distinct rate of response to the drug. The polymorphism involves a single nucleotide. Let q be the frequency of allele a and p = 1 – q be the frequen-

898

cy of allele A. According to the Hardy-Weinberg equilibrium, there are p2 patients with genotype AA, 2pq patients with genotype Aa, and q2 patients with genotype aa, with p2 + 2pq + q2 = 1. The treatment response rate for each of these subgroups of patients is as follows: AAR, patients with genotype AA; AaR, patients with genotype Aa; and aaR, patients with genotype aa. Thus, the treatment response rate for all patients can be described by the following equation: TR = AAR · p2 + AaR · 2pq + aaR · q2

(2)

Now, assume that, of the three genotypic groups of patients, the subgroup with genotype aa has the poorest response to the treatment— the response rate equals the placebo response rate (which means that the polymorphism has no impact on natural disease evolution but only on drug action) and the subgroup with genotype AA responds best. The greatest expected relative risk, ExpRR, is defined as the ratio between AAR and aaR: ExpRR = AAR/aaR. Depending on the response of the heterozygous patients (Aa), allele A can be considered linked to a recessive efficacy if the heterozygous patients are not responding to the treatment (i.e., AaR = PR) like the homozygous patients (aa); alternatively, A can be considered linked to a dominant efficacy if the heterozygous patients are responding to the treatment (i.e., AaR = AAR) like the homozygous patients (AA). From the previous equations, we have the following relationships: ObsRR ≤ ExpRR ≤ 1/PR (3) ObsRR = TR/PR = (AAR · p2 + AaR · 2pq + aaR · q2)/PR The maximum value for expected relative risk, max[ExpRR], is reached if the subgroup with genotype Aa has the same response rate as the placebo group: AaR = aaR = PR. Replacing


these rates in Equation 3 leads to ObsRR = ExpRR · p2 + 2pq + q2 ExpRR = (ObsRR – 2pq – q2)/p2

(4)

The minimum value for expected relative risk, min[ExpRR], is reached if the heterozygous patients have the same response rate as the homozygous patients (AA): AaR = AAR. Replacing these rates in Equation 3 leads to ObsRR = ExpRR · (p2 + 2pq) + q2 ExpRR = (ObsRR – q2)/(p2 + 2pq)

(5)

For example, if p = 0.6, PR = 40%, and TR = 60%, then the expected relative risk is at least 1.6 but not more than 2.4. Under these conditions, the highest response rate we can expect for the subgroup with genotype AA is 95.6%. Another useful way to present this information is in terms of maximum possible gain (G) from the observed difference in response rate (TR – PR), without any pharmacogenomic hypothesis for the maximum expected difference in response rates (max[ExpRR] · PR – PR) with a strong pharmacogenomic hypothesis, that is, recessive efficacy (Table 1). G = (max[ExpRR] · PR – TR) = [(ObsRR – (6) 2pq – q2)/p2] · PR – TR G = [TR – PR · 2pq + q2) – TR · p2]/p2 G = [TR · (1 – p2) – PR · (2pq + q2)]/p2 G = [(1 – p2)/p2] · (TR – PR) For p = 0.6, PR = 40%, and TR = 60%, this means that the greatest improvement in response to treatment we can expect from a potential interaction with a polymorphism corresponding to a recessive efficacy is G = [(1 – 0.62)/0.62] · (60% – 40%) = 35.6% (~36% in Table 1) (7)