- © 2006 by American Society of Clinical Oncology
Interpreting Disparate Responses to Cancer Therapy: The Role of Human Population Genetics
- From the Section of Hematology/Oncology, Department of Medicine; Committee on Clinical Pharmacology and Pharmacogenomics; Department of Human Genetics; Cancer Research Center; University of Chicago, Chicago, IL
- Address reprint requests to Mark Ratain, MD, University of Chicago Medical Center, Clinical Pharmacology & Pharmacogenomics, 5841 S Maryland Ave, MC2115, Chicago, IL 60637-1470; e-mail: mratain{at}medicine.bsd.uchicago.edu
Abstract
Increasingly, investigators are recognizing differences in tumor biology, drug metabolism, toxicity, and therapeutic response among different patient populations receiving anticancer agents. These observations provide exciting opportunities to identify the factors most important for predicting individual variability in pharmacologically relevant phenotypes and consequently for personalizing the delivery of cancer therapy. Although pharmacogenomic differences may explain some of these disparities, rigorous investigation of both genetic and nongenetic differences is important to identify the variables most important for optimal selection and dosing of treatment for an individual patient. For example, pharmacogenetic tests currently used in cancer therapy, such as genotyping UGT1A1 to reduce the incidence of severe toxicity of irinotecan and sequencing epidermal growth factor receptor from tumors to identify somatic mutations conferring sensitivity to tyrosine kinase inhibitors, were developed without initial identification of interpopulation differences. Although interpopulation variability in toxicity and efficacy of these agents has been observed, the basis for these population differences remains only partially explained. Here, we review concepts of human population genetics to inform interpretations of disparate drug effects of cancer therapy across patient populations. Understanding these principles will help investigators better design clinical trials to identify the variables most relevant to subsequent individualization of a cancer therapy.
IDENTIFYING DISPARITIES IN TREATMENT EFFECTS IN DIFFERENT POPULATIONS
During the last 15 years, heritable variation in drug-metabolizing enzymes has been demonstrated to explain some of the interindividual variability in patient outcomes1,2 from medical therapies for cancer. Since completion of the Human Genome Project, more articles on “cancer” and “pharmacogenetics/pharmacogenomics” have been published every year. Contemporaneously, US governmental funding and regulatory agencies have encouraged recruitment of patients from previously under-represented populations3,4 (here and throughout this article we will use the term “population” to refer to any group of humans that may be broadly categorized by shared ancestry or self-described ethnic affiliation), and international efforts at cancer drug development have become more common.5 With this confluence of events, cancer researchers are expected to report with increasing frequency disparities in toxic and therapeutic drug effects between selected patient populations, as they have for the tyrosine kinase inhibitor gefitinib and the topoisomerase I inhibitor irinotecan.6,7
These reports occur at a time when new genotyping technologies and large public databases of human genetic variation have become readily available to cancer researchers,8,9 population geneticists have refined concepts of the genetic structure of human populations,10-12 and prominent biomedical journals have discussed the role of ethnicity/race in categorization of patients in biomedical research.13-20 In this setting, it is tempting for cancer investigators to categorize patients in a clinical trial according to race, ethnicity, or national origin, and if a difference in toxicity or benefit associates with a particular population category to suggest possible “pharmacogenomic differences.”
A goal of cancer drug development is to maximize the therapeutic benefit and minimize the toxicity of treatment for patients receiving a given drug (ie, maximize the drug’s therapeutic index). When reproducible and clinically significant, differences in treatment effects across patient populations provide opportunities to determine the factors causing interindividual variability in the therapeutic index. Identification of these factors enables clinicians to select either alternative treatments or to modify the dose to improve outcomes for their individual patients. Both intrinsic and extrinsic factors (Table 1) will account for interpopulation variability, and the factors with the greatest proportional and most reproducible effect on therapeutic index will be the most clinically useful for individualizing patient therapy.
With newly recognized importance for some cancer therapies and increasing ease of detection, genetic polymorphisms are appealing intrinsic factors to test for association with disparate treatment effects between populations in clinical trials. To design such association studies and interpret the results effectively requires recognition of some principles of human population genetics. In this article, we review the basic concepts about the evolutionary forces shaping human genetic variation, recent work describing the geographic distribution of this variation, and the implications of the biogeography of human populations for the pharmacogenetics of anticancer agents.
SOURCES OF HUMAN GENETIC VARIATION
Humans share more than 99.5% sequence identity across the approximately 3 billion–nucleotide genome,21,22 with one common polymorphism occurring at every 500 to 2,000 nucleotides. This nucleotide diversity is low relative to other primate species,23,24 and an indication of our recent ancestry. Various analyses of our genetic diversity, well reviewed elsewhere,12,25-28 support a model for the Recent African Origin (RAO) of our species (Fig 1). According to this model, anatomically modern humans evolved in Africa more than 100 thousand years ago (kya). It has also been proposed that migrations out of Africa to the Middle East originated from an East African gene pool approximately 100 kya and were followed by waves of migration and subsequent population expansions into Australia and Melanesia (approximately [approx] 60 kya), Europe (approx 40 kya), Asia (approx 35 kya), the Western Hemisphere (approx 30 kya), and the Pacific Islands (approx 3 kya).
During and since the course of these migrations and population expansions, as for any other species, our genetic diversity was shaped by two primary evolutionary forces: random drift acting on neutral mutations29,30 and natural selection acting on mutations affecting the individual’s fitness.31,32 Neutral mutations can be defined as genetic variants that do not cause a change in gene function resulting in either an increase or decrease in fitness. For example, variants that do not alter the encoded amino acid sequence or an enhancer element might be neutral. Polymorphisms within intergenic sequences and even those causing changes in gene function that do not result in a difference in survival or reproduction might also be neutral mutations. Because neutral variants are inconsequential to the survival and reproductive success of the individuals carrying them, their fate is affected only by random chance (ie, the random sampling of gametes from generation to generation) which in turn is influenced by the demographic history of human populations. In this sense, fluctuations in population size, migration of a subset of members of one population to new locations (ie, founder effect), and the movement of genes between populations as a result of migration and mating of individuals from different populations, (ie, gene flow), all affect patterns of neutral variation. Because the demographic history of a population is obviously the same for all loci in the genome, its effects on neutral variation occur on a genome-wide scale. In contrast, natural selection, the differential contribution of genetic variants to future generations,31 eliminates or favors specific alleles at specific loci based on their differential effects on survival and reproductive fitness.
PATTERNS OF HUMAN GENETIC VARIATION
The properties of gene sequence variation among individuals and populations today provide clues to the evolutionary history of our genes. During the course of recent human history, some selective forces such as malaria endemia affected humans in specific geographic locations and drove adaptive changes. For example, in populations from Sub-Saharan Africa, alleles conferring resistance to malaria infection, such as glucose-6-phosphate dehydrogenase (G6PD) deficiency,33 hemoglobin S, and loss of expression of the Duffy blood group glycoprotein,34 reached high frequency. For G6PD deficiency, different deficiency alleles with the same phenotype have also been identified in populations outside Africa exposed to malaria, whereas the absence of Duffy glycoprotein expression is rarely observed in humans of different geographic ancestry. Because of the correlation in evolutionary history imposed by tight linkage, the effect of natural selection will extend to the genomic region surrounding the selected site and will generate patterns of variation different from those expected of neutral evolution. These patterns are often referred to as the “signature” for natural selection.35 Genes with such a signature are relatively uncommon36; most genes display a pattern of variation consistent with neutrality.37
When genetic diversity has been assessed in samples collected from many human populations, the pattern of diversity follows geographic “clines” (ie, a gradual change of allele frequency going from one geographic location to another; Fig 2).10,38,39 Most of human sequence diversity is among individuals rather than populations,40 and interpopulation genetic variability is found in gradients probably resulting from the history of gene flow between nearby human populations.11,12,26 The Human Genome Project spurred identification of myriad genetic polymorphisms across the genome, making it possible to characterize genetic diversity with greater resolution than classical methods. However, identifying and genotyping polymorphic variants in samples from the opposite ends of the geographic clines (eg, Northwestern Europe, Central Africa, and East Asia), may accentuate the differences between human populations. Analyzing these data with computerized clustering algorithms may tend to generate distinct population groups14,26 according to the five main population groups depicted in the branches of Figure 1. One such algorithm, with results diagrammed in Figure 3, inferred individual ancestry from the particular alleles at 100 genetic loci.42 The proportional fraction of an individual’s ancestry shared with Africans, Asians, or Europeans is represented by the distance between a circle and each side of the triangle (Fig 3A). For many individuals the proportion of ancestry is 100%. When these same markers are typed in individuals from South India (Fig 3B), a large population in between the continental extremes, the proportion of shared ancestry with the initially typed populations varies widely.10,11,41,42 The descendants of migrants of the original populations (essentially all people of European, African, and Asian descent born in the United States), may also be classified according to shared ancestry—though not to the same extent as descendants of nonmigrating populations (Fig 3C). For example, one self-described African American (highlighted by the arrow) is estimated to share 60% ancestry with other African-Americans and approx 40% with European Americans. These findings confirm that although the geographic distribution of human genetic variation is clinal, sampling from populations at the continental extremes and using the large number of markers now available may allow grouping individuals that share the same continental ancestry.
A generalization from these data is that gene variants affecting medically relevant phenotypes may fall into three main categories: (1) common variants (ie, variants found at readily measurable frequencies; at least 5% to 10% depending on the study) across all or most human populations, (2) rare variants (ie, variants found at frequencies less than 1% and typically within only one population), and (3) variants at intermediate frequencies more than 1% within single populations but absent in others. The latter may have reached intermediate frequency in a single population as a result of either natural selection or various stochastic processes bringing a neutral variant to higher than usual frequency. Typically, common variants—as reflected by their presence among all human populations—are likely to have predated the human population expansion out of Africa, whereas rare variants tend to have arisen after the establishment of geographically defined populations. Whether common, complex, chronic diseases of humans today are primarily associated with common or rare variants is the subject of an ongoing debate43-46that is far from settled. This summary is an oversimplification, but sufficient for understanding the implications of patterns of human genetic variation for identifying clinically useful genetic markers for drug-related phenotypes.
VARIABLE RESPONSE TO PHARMACOLOGIC AGENTS AND THE STRUCTURE OF HUMAN GENETIC DIVERSITY
These patterns of human genetic diversity have important ramifications for the development and application of pharmacogenetic tests. Because most human genetic variation is within rather than between populations,10,40 only a fraction of gene variants will both display large differences in allele frequency between the populations of interest and have sufficiently strong effects on drug response phenotype to be the intrinsic factor primarily accounting for interpopulation differences in treatment effects.47,48 In the pharmacogenetics literature, there are several examples of disparities in frequencies of pharmacologically relevant genetic variants among populations.17,47,49,50 These discrepancies lead to the hypothesis that genes encoding and regulating expression of drug-metabolizing enzymes might have been subjected to natural selection49 as a result of changes in diet or other environmental stresses that arose since humans first exited Africa, perhaps even more recently. That alleles for the cytochrome p450 gene CYP1A2,51 the alcohol-metabolizing genes aldehyde dehydrogenase (ALDH2), alcohol dehydrogenase (ADH),52,53 and persistence of lactase (LCT)54 all reveal signatures of natural selection supports this hypothesis. Although the allele frequency of polymorphisms affecting pharmacokinetic and pharmacodynamic variability might differ across populations, how much of the clinically significant variability in toxic and therapeutic effects of a given drug among populations and more importantly, among individual patients, is attributable to these gene variants must be determined empirically.48
Without any other basis for suspecting genetic polymorphism(s) to be among the most important of the various intrinsic and extrinsic factors contributing to disparities in treatment outcomes between two populations, an association mapping approach to find pharmacogenomic differences may lead to spurious results. For example, in a study of more than 2,000 European American subjects designed to identify the genetic determinants of height55 the LCT allele conferring lactase persistence was tightly associated with tall stature (P = 3.6 × 10−7), and standard genotyping controls suggested that this was a true association. Only after testing the association between this allele and height in independent studies of 1,000 Polish subjects, and more than 60 trios of parents and offspring from Scandinavia were these investigators able to demonstrate that the association between the LCT allele and height in the original sample was a false-positive finding. Because studies of cancer therapies tend to enroll fewer subjects and may not have independent samples for confirmation of initial association findings, investigators should be skeptical of a significant genotype-phenotype association if there is no prior basis for the relationship. Furthermore, without independent pharmacologic confirmation for the gene-drug-effect relationship, one may be more likely to map genetic markers associated with different ancestral origins than the specific causative allele (if it exists) itself.
The geographic structure of human populations complicates not only efforts to identify gene variants underlying differences in drug outcomes, but also the development of new anticancer agents and effective use of testing for a causative variant in the clinical setting. The consequences for development of safe, effective anticancer agents potentially useful in one population but not another has been addressed by Tate and Goldstein.50 More imminently relevant for oncologists is the challenging situation facing implementation of the changes to the label for irinotecan approved by the US Food and Drug Administration in 2005. This label change warned prescribers that patients homozygous for the UGT1A1*28 allele are at increased risk for grade 4 neutropenia when irinotecan is administered at the standard dose, and recommended that such patients initiate treatment at a reduced dose. Approximately 10% of the North American population is homozygous for this allele, and in a prospective study of patients receiving irinotecan, 50% of these individuals experienced grade 4 neutropenia,56 compared with none of the subjects homozygous for the more common, fully functional allele UGT1A1*1. However, the optimal usage of this test has not yet been established.
The predictive value of this test will depend, in part, on the predominant ancestral origins of a physician’s patient population. The allele frequency for UGT1A1*28 varies across populations: 35% in Singaporean patients reporting Indian heritage,57 and the same in persons reporting Italian heritage, but only 13% among Japanese.58 Meanwhile, Japanese populations carry at intermediate frequencies (also approximately 15%) another polymorphism affecting UGT1A1-mediated metabolism of irinotecan known as UGT1A1*6.59,60 This allele was not detected in a survey of the UGT1A1 sequence of both European and African-Americans.61 Additional UGT1A1 polymorphisms possibly affecting irinotecan metabolism and toxicity have been identified at different frequencies in different populations.60,62 Some analyses of the diversity of human genetic variation for pharmacologically relevant genes have predicted that the ancestral geographic origins of a patient should not significantly affect the genotype-phenotype relationship.17,63 As the first clinical pharmacogenetic test relevant to therapy for a sizeable adult oncology population, UGT1A1*28 genotyping will serve as an important test of this concept.
NONGENETIC FACTORS ASSOCIATED WITH RACIAL/ETHNIC AFFILIATION
Categorization of individuals according to self-reported race or ethnicity is inherently imprecise. Especially in United States populations, individuals may have membership in more than one biogeographical cluster, an individual’s perception of this affiliation may change over time, and study sampling strategies influence this assignment.13,64 Short of deeper understandings of the biologic basis for interpopulation differences in drug-related phenotypes, categorization by “race, ethnicity, and ancestry” may serve as a useful proxy14,19,65,66 for the “complex web of biologic and social connections that link individuals and groups to each other.”13 But these categories are only proxies for a wide range of heterogeneous factors that includes both genetic and environmental components.
The various factors that may account for population disparities in cancer therapy outcomes listed in Table 1 may be interdependent. In any specific case, one factor may predominate, and identifying that factor will lead to the best individualization of therapy for the broadest set of patients, but dissecting the predominant factor is not straightforward. For example, in industrialized countries, the prevalence of tobacco smoking is greater in lower socioeconomic strata,67 and the same is true for alcohol use.68 As highlighted in this issue of the Journal of Clinical Oncology,69 in the United States, self-reported race/ethnicity is associated with socioeconomic status and diminished access to care. These factors may be controlled in the setting of a clinical trial, but even if a genetic polymorphism affecting toxicity or efficacy is identified within a clinical trial, the relative importance of that polymorphism in clinical practice may be diminished by these extrinsic factors.
Diet can also directly affect drug metabolism,70,71 and dietary preferences may vary with self-reported ethnicity72,73 and region of current residence.74 Epidemiologic studies to identify the effects of many neutral variants on the relationship between diet and risk for cancer incidence focus on specific populations with shared ancestry living in different nations.74 As the same enzymes that lead to inactivation or elimination of pharmacologic agents may also detoxify dietary and environmental carcinogens, the genetic polymorphisms that affect the incidence and natural history of cancer among individuals may affect the toxicity and efficacy of therapy as well.48,75 Consequently, assuming population differences in cancer therapy outcomes to be explained by pharmacogenomic differences may lead investigators into the complex web of social and biologic interactions that link individuals with groups rather than elucidate how best to personalize therapy for patients who may benefit from a given treatment.
DISPARITIES IN TREATMENT EFFECTS IN DIFFERENT POPULATIONS: AN OPPORTUNITY TO IDENTIFY FACTORS CONTROLLING INDIVIDUAL TREATMENT OUTCOMES
A panoply of intrinsic and extrinsic factors determine an individual’s response to cancer therapy. Because some of these factors will tend to be more frequently associated with one population than another, identifying a disparity in the outcomes of therapy between populations provides the opportunity to generate hypotheses about the most important factors and test these hypotheses in subsequent studies. As for most phenotypes, drug-related phenotypes entail the complex interaction between genes and environment.15,66 The advantage of studying pharmacogenetics over broader common, complex diseases, is that, in many cases, it is easier to identify the proximal relationships between a gene variant and drug response by in vitro and ex vivo assays in the laboratory and detailed evaluation of clinical subjects. For example, investigators can confirm the importance of an association between a gene variant and severe toxicity from standard-dose therapy by identifying the effects of a specific drug-metabolizing enzyme allele on drug clearance and exposure. Alternatively, somatic mutations within tumors of a particular population, as in the case of therapeutic response to gefitinib therapy, may also be more readily identified.
The increasing availability of DNA specimens from patients in clinical trials and gene sequencing technology will enable investigators to better elucidate the many complex interactions between gene variants and environment leading to treatment outcomes. At this time, pharmacogenetics/genomics is probably most useful when wide variability in the efficacy and/or toxicity for an active anticancer agent is observed. With the many confounding factors described in this review, it is clear that the observation of disparate outcomes of a given therapy does not necessarily signify the importance of pharmacogenomic differences to the variability in therapy responses. Important financial and technical resources may be wasted in these pursuits if the relevance of more readily measurable variables is not excluded first. Optimal individualization of cancer therapy will come from understanding the interaction of genetic and nongenetic factors affecting the outcome of treatment with a specific drug. Disparate results of treatment provide the first clue to the basis for this variability and an opportunity to identify the specific factors that will apply to each patient regardless of ethnic affiliation.
More detailed information on drugs and related genes, including annotated diagrams of drug metabolism pathways, can be found on the Pharmacogenetics Knowledge Base (PharmGKB) Web site, http://www.pharmgkb.org.
Authors' Disclosures of Potential Conflicts of Interest
Although all authors completed the disclosure declaration, the following authors or their immediate family members indicated a financial interest. No conflict exists for drugs or devices used in a study if they are not being evaluated as part of the investigation. For a detailed description of the disclosure categories, or for more information about ASCO’s conflict of interest policy, please refer to the Author Disclosure Declaration and the Disclosures of Potential Conflicts of Interest section in Information for Contributors.
| Authors | Employment | Leadership | Consultant | Stock | Honoraria | Research Funds | Testimony | Other |
|---|---|---|---|---|---|---|---|---|
| Michael L. Maitland | Astra Zeneca (A) | |||||||
| Anna DiRienzo | Royalties on UGT1A1 testing from Mayo Medical Laboratories (A) | |||||||
| Mark J. Ratain | Prometheus Laboratories, Inc. (A); Genzyme Corporation (A); Genentech (A); Neopharm (B) | Solexa (A); Applera (B) | Royalties for UGT1A1 genotyping from Mayo Medical Laboratories (A) |
Dollar Amount Codes (A) < $10,000 (B) $10,000-99,999 (C) ≥ $100,000 (N/R) Not Required
Author Contributions
Conception and design: Michael L. Maitland, Anna DiRienzo, Mark J. Ratain
Financial support: Mark J. Ratain
Manuscript writing: Michael L. Maitland, Anna DiRienzo, Mark J. Ratain
Final approval of manuscript: Michael L. Maitland, Anna DiRienzo, Mark J. Ratain
A model for the human “family tree.” Multiple sources of evidence as described by Tishkoff and Verelli27 support the Recent African Origin model of modern human history. The “branching” of modern non-African populations leads to differences in genetic variation across the genome, and the opportunity to develop population differences in allele frequencies. Substantial gene flow between each of the populations is represented by horizontal lines. Kya, thousand years ago. Reprinted with permission from Tishkoff and Verelli.27
Human genetic diversity is distributed in gradients among and within continents. Geographic plot of variation across 82 genes reveals gradual changes in the proportion of shared genetic structure among populations sampled across the continents. However, this international map incompletely represents global genetic variation, and emphasizes intercontinental variation. The inset figure of Africa highlights finer gradations over shorter geographic spans, emphasizing the “clinality” of human genetic variation. Reprinted with permission from Cavalli-Sforza, Menozzi, Piazza.38
Inferring individual ancestry proportions from genetic data. (A) Distribution of inferred ancestry for 255 individuals; (B) overlay of 263 individuals from South India over part A; and (C) similar plot for 60 individuals from the United States. (D) Line length is proportional to measured genetic distance. The differences between any two individuals (black lines) of the same continental ancestry is larger than the distance between populations (dotted lines) of shared continental ancestry. Individuals with high proportion ancestry from more than one population (arrow in 3C) are connected by branches directly to the branches between clusters. Reprinted with permission from Bamshad et al.42
Examples of Intrinsic and Extrinsic Factors Relevant to Different Treatment Outcomes in Different Patient Populations5
Footnotes
-
Supported in part by the Pharmacogenetics of Anticancer Agents Research (PAAR) Group and by National Institute of General Medical Sciences (National Institutes of Health, Bethesda, MD) and Grant No. U01GM61393.
Authors' disclosures of potential conflicts of interest and author contributions are found at the end of this article.
- Received December 6, 2005.
- Accepted February 3, 2006.
