The debrisoquine metabolic phenotype and DNA-based assays: implications of misclassification for the association of lung cancer and the debrisoquine metabolic phenotype.

Debrisoquine is an antihypertensive drug that is metabolized by cytochrome P4502D6. Deficient metabolism is inherited as an autosomal recessive condition. We previously reported in a case-control study that extensive metabolizers of debrisoquine were at greater risk of lung cancer compared to poor and intermediate metabolizers. Cloning of the gene that encodes P4502D6 (CYP2D6) led to the identification of both wild-type and mutant forms of the gene. Subsequently, a DNA-restriction fragment length polymorphism (RFLP) was identified, and a Southern hybridization-based test was developed in an attempt to define the genotype. When the DNA-RFLP test was applied to stored DNA from our study subjects there was neither a significant association with the metabolic phenotype nor an association with lung cancer. Further work has demonstrated that the wild-type gene, which was characterized by a 29-kb allele, can also contain mutations that result in nonfunctional or absent proteins. When these mutations are present, individuals exhibit the poor or intermediate metabolizer phenotype in spite of the presence of the 29-kb putative wild-type allele. Sequence determination of the mutants led to the development of techniques to exploit the polymerase chain reaction, which, together with Southern analysis, have been reported to detect as many as 95% of poor metabolizers. This technique is being used to examine the association of the extensive metabolizer genotype with lung cancer in the subjects from the case-control study. Preliminary results indicate a weak association between the homozygous wild-type genotype and lung cancer; in contrast, the extensive metabolizer phenotype is strongly associated with lung cancer in this subset.(ABSTRACT TRUNCATED AT 250 WORDS)


Introduction
While exposure to tobacco smoke is widely accepted as the major etiologic factor in lung cancer, there are clear differences in individual susceptibility consistent with a heritable component to risk. The metabolism of the anti-hypertensive drug debrisoquine (DBR) is under autosomal genetic control (1)(2)(3), and inheritance of the trait conferring ability to "extensively" metabolize the drug has been suggested as a host susceptibility factor for lung cancer. Studies consistent with a genetic component to lung cancer susceptibility (4)(5)(6)(7) and studies to test the hypothesis of an association between DBR metabolism and lung cancer (8)(9)(10)(11)(12)(13)(14) are reported.
In a case-control study, we tested the hypothesis that the ability to metabolize DBR is related to lung cancer risk. Overall, extensive metabolizers of DBR, as determined by the metabolic ratio after administration of DBR, were at significantly elevated risk oflung cancer compared to poor or intermediate metabolizers (odds ratio = 7.5 [95% confidence interval, 2. 5-22.8]). Controls were individuals with either chronic obstructive pulmonary disease or other cancers; results were adjusted for age, race, asbestos exposure, family history of lung cancer, and smoking. While the results of the six case-control studies (7)(8)(9)(11)(12)(13)15) generally support varying degrees of association of susceptibility to lung cancer with this phenotype, in our first attempt to evaluate a genotype-lung cancer association in the same case-control study, using a restriction fragment length polymorphism (RFLP) marker, we found no evidence for an association (16). While certain XbaI allelic fragments (11.5 and 44 kb) were associated with the poor metabolizer phenotype, overall the ability of the various RFLP patterns to predict the phenotype was poor. The high degree of misclassification (low sensitivity for detecting mutations that result in deficient ability to metabolize DBR) of this marker rendered it unacceptable for epidemiologic inference, and the question of a genotype/lung cancer association remained indeterminate. Continued progress in the description of the gene and pseudogene sequences, along with the identification of new mutations that account for the majority ofpoor metabolizers, have permitted application of a polymerase chain reaction (PCR) approach to further characterization of the genotype (17,18). In this presententation, we compare the degree of association based on the metabolic phenotype, the earlier reported RFLP marker, and a preliminary application of a new PCR-based marker using the method of Gough et al. (18) to a subset of 133 subjects from the case-control study and consider the relative merits of phenotype and genotype determination in future studies.

Case-Control Study
The design and conduct of the NCI-Maryland casecontrol study were described in the original study report (15). Briefly, cases were patients with histologically confirmed, untreated lung cancer identified at the University of Maryland and Baltimore Veterans Administration Hospitals between 1985 and 1989. Controls consisted of patients with chronic obstructive pulmonary disease and subjects with other malignancies, including cancers of the colon, esophagus, stomach, and breast, and melanoma, but excluding bladder cancer [because of the proposed association between bladder cancer and DBR metabolism (19)]. A personal, structured interview of approximately 45 min was conducted by a trained interviewer/phlebotomist. Subjects were excluded if they had hypotension, inability to take oral medications or to be interviewed, general anesthesia within the previous 5 days, severe renal, liver, or medical illness, or previous diagnosis of separate primary malignancy.
Although 200 subjects were accrued in the original study, 192 had previously undergone DBR phenotyping; samples from 92 had previously undergone Southern hybridization with a cDNA clone of the human CYP2D6 gene after digestion by XbaI; 133 have undergone PCR analysis.

Laboratory Methods
Phenotype. DBR and its chief metabolite, 4-hydroxy-DBR, were determined in an aliquot of urine using the method of Idle et al. (20). DBR (Declinax, Roche) is an adrenergic blocker used as an antihypertensive drug in Canada and Europe. Following an overnight fast, 10 mg (tracer dose) of DBR were administered orally. After the initial voiding had been discarded, urine was collected over the next 8 hr. Nonessential medications were not given on the morning of DBR administration; fluids and a light breakfast were permitted 1 hr after the dose. No significant hypotensive or other adverse reaction was noted in the study (21).
The DBR metabolic phenotype was determined by calculating the metabolic ratio, i.e., percent dose excreted as unchanged DBR divided by percent dose excreted as 4-hydroxy-DBR. This ratio can be used to classify individuals into one of three categories: extensive, intermediate, and poor metabolizers of DBR. The method used for cutpoint determination involves a mixture model to fit three normal distributions to the frequency distribution of metabolic ratios observed in controls. Cutpoints for the determination were derived in blacks ( (22).
Statistical analyses were performed using the SAS statistical analysis package (23).
Genotype. The method of Gough et al. (18), employing PCR to characterize the 29-kb allele, identifies a base deletion at the junction ofintron 3 and exon 4, which results in a splice-site defect. This mutation, designated the CYP2D6 "B", is the most common one among poor metabolizers and accounts for 75% of the alleles in this group. Next most common is CYP2D6 "D, " a complete deletion of gene, associated with the 11.5 kb XbaI haplotype and accounting for slightly greater than 10% of alleles in poor metabolizers. CYP2D6 'A" (deletion in exon 5) and "C" (single base pair deletion in exon 5) are less common and at least 5% of mutations are as yet uncharacterized (24). When these point mutations are present, individuals exhibit the poor or intermediate metabolizer phenotype in spite of the presence of the putative 29-kb wild-type allele (17,25,26). The PCR is used to amplify a 298-bp fragment, using primers from an area that is not homologous with CYP2D7 and CYP2D8. The product is then enzymatically digested with BstNl (New England Biolabs, MA), according to the manufacturers' instructions. Samples were analyzed by electrophoresis on agarose (2.2%) (17).

Results and Discussion
In the original case-control study, 13 poor metabolizers were identified from among 181 subjects who had undergone phenotyping [89 lung cancer cases, 1 poor metabolizer; 92 pooled controls, 12 poor metabolizers (15)]. Genotyping using the XbaI restriction fragment size in an overlapping subset of controls and normal volunteers (n = 80) revealed 11 poor metabolizers with the following distribution of haplotypes: five were 29/29 homozygous, three were 29/44, two were 29/11.5, and one was 44/16/9 [see Table 1 in Sugimura et al. (16)]. These data failed to demonstrate that the 11.5 kb and 44 kb allele fragments were of value in identifying the deficient metabolizer phenotype. Finally, in a recent preliminary application of a modification of Gough's (18) method for detection of the intron 3/exon 4 mutation in eight poor metabolizers, 3/3 were homozygous mutant subjects, but four were wildtype homozygous and two were heterozygotes. Misclassification is apparently possible with either assay, but the PCR method is clearly an improvement over the earlier RFLP approach.
The extensive metabolizer phenotype is strongly associated with lung cancer in the original study [data not shown, see Caporaso et al. (15)]. There was no association between the earlier RFLP marker and lung cancer risk (16), and there was only a weak association with the PCRdetermined genotype. An explanation of the differing degrees of association with the different tests requires discussion.
From both a technical and an epidemiologic perspective, accounting for the subjects in whom the genotype and the phenotype do not correspond is of central importance. Two general explanations for a lack of correspondence are: an influence of the disease state or other distorting factors on the determination of phenotype (unlikely, but difficult to exclude owing to the case-control study design) and misclassification error in the genotyping assay.
With regard to the first possibility, the non-correspondence of phenotype and genotype may be due to an effectcause relationship; that is, in theory, the tumor, tumor products, or tumor treatment (although subjects in this study were untreated) could modulate expression or measurement (i.e., the phenotyping procedure) of the ability to metabolize DBR. While studies have generally found no effect of chemotherapy on the phenotype (27), and the phenotype of subjects after cancer therapy has not been altered (7,11), the possibility is difficult to totally rule out. Declining role but still important in selected Steady increase in use likely in the future settings aSensitivity = true positives/(true positives + false negatives); in this case a positive is a deficient metabolizer. Here sensitivity does not refer to how well the test is able to identify subjects with the disease, but only how well the test identifies deficient metabolizers. The "gold standard" is phenotyping performed in healthy, fasting subjects, receiving no medications, an administration of debrisoquine, followed by 8-hr urine collection. bSpecificity = true negatives/(true negatives + false positives) or the ability of test to identify nondiseased, where diseased or positives are deficient metabolizers, and negatives are the more common extensive metabolizers. By this definition, the specificity of the genotype test is quite good because all subjects identified with two mutant alleles are phenotypic poor metabolizers. This problem will be addressed directly in a study currently in progress in which early-stage lung cancer patients are phenotyped prior to treatment and again following surgical treatment of lung cancer. Preliminary results from this study indicate no change in the phenotype following treatment (28). Alternatively, unrecognized medications may have rendered phenotyping inaccurate. Individuals who are true extensive metabolizers have been misclassified in the phenotyping assay because of failure to recognize that these subjects were receiving a medication or dietary item capable of distorting phenotyping or owing to laboratory error. We have carefully reviewed data from questionnaires and medical abstracts used to specifically capture this information, and found that an unrecognized dietary agent (29) or medication [e.g., quinidine (30,31)] might in theory account for the findings, but the probability appears small.
The second possibility is that the genotype assay does not yet recognize all the possible mutations that may result in the poor metabolizer phenotype. While perfect correspondence between the genotype and phenotype in extensive and intermediate metabolizers is not expected because family studies have demonstrated incomplete dominance (that is, obligate heterozygotes may have either extensive or intermediate metabolizer phenotypes), the failure to detect mutant alleles consistently in phenotypic poor metabolizers is problematic. Of the nine poor metabolizer subjects, three are homozygous for the mutant genotype, two are heterozygous, and four are wild-type homozygous. The presence of the latter group indicates the shortcomings of the current assay. Further mutations may account for the deficient metabolism phenotype in these subjects. It is likely that elucidation of further mutant alleles will be required before the assay fulfills its promise as the final arbiter of the question of an association with lung cancer.
In conclusion, this preliminary examination of a new approach to genotyping the CYP2D6 locus (DBR phenotype) allows certain conclusions. The genotype and phenotype show a significant association, although complete correspondence is not present. The question of the degree of association with lung cancer is the subject of ongoing study. It would be premature to draw conclusions from the results of the "B" mutation alone, however, it is of interest to note that the odds ratio for risk in extensive metabolizers (pct genotype) is similar to that which is derived from the published case-control studies in the aggregate [odds ratio for EM is approximately 2 (32)]. If nondifferential misclassification is assumed, the characterization of further mutations (improved sensitivity) should adjust the point estimate upward while improvements in phenotyping (i.e., recognizing and eliminating some currently unappreciated medication which inhibits CYP2D6) will improve specificity and would likely reduce published point estimates, derived from this study. Finally, we list the relative merits of the phenotyping and genotyping approaches as applied to population studies in Table 1. With certain important exceptions, we anticipate that the advantages of genotyping will mandate increasing reliance on this approach. One further possibility must be mentioned. Implicit in the previous discussion is the assumption that the "real association" must be with the genotype rather than the phenotype. In fact, it may be that the DBR metabolic ratio, ultimately determined in the individual as a complex of genetic and environmental factors, most accurately reflects lung cancer susceptibility itself the consequence of tobacco and other carcinogen exposures, in concert with hereditary predisposition. Table 1 summarizes characteristics of the phenotype and genotype approaches.