Genotype/Phenotype Interpretation of DNA Microarray Analysis

By Dr. L. Scott Clark, Chief Scientific Officer, Gentris Corporation

Microarray technology has advanced the field of pharmacogenomics immensely. In the past ten years alone, microarray analysis has allowed researchers to increase the number of human genetic polymorphisms analyzed from several hundreds to thousands. As the number of genes and alleles has increased on microarray chips, data analysis has become more cumbersome. While software has been developed to aid in the interpretation of these data, traditional scientific expertise is still required for accurate analysis of the results. From the service provider industry perspective, it is imperative to understand the intricacies of human genetic testing. As will be discussed, pharmacogenomic expertise is needed to select the correct genes and alleles for drug metabolizing enzymes in DNA arrays so that the most useful data are obtained.

While microarray technology has increased the amount of data that can be collected simultaneously, testing for the correct combination of alleles is critical in accurately determining genotypes and ultimately accurately predicting phenotypes. In addition to clinically relevant alleles, many arrays include genetic polymorphisms of unknown clinical relevance. Furthermore, some of the tested alleles have an incidence of less than 0.001%. These alleles would only be pertinent in very large studies (>10,000 subjects). Understanding these matters is what distinguishes a typical service provider from one with expertise in pharmacogenomics.

Another example on how scientific expertise is desirable is illustrated by the drug metabolizing enzyme, CYP2D6, and two branded microarray chips: the Roche Amplichip and the Affymetrix DMET chip. Both chips test for multiple alleles of CYP2D6; however the Amplichip targets the drug metabolizing enzymes CYP2D6 and CYP2C19 while the DMET is designed for assessing drug efficacy in preclinical and early Phase I and II trials. Both chips are useful clinical research tools. If the protocol and informed consent states that testing for CYP2D6 is required, selection of one chip over the other may result in a different genotype and predicted phenotype. The chips have a similar number of alleles for CYP2D6 but there is only a 70-80% overlap of the same alleles on both chips. Of these overlapping alleles, many have such a low allelic frequency in the general population that they may not be clinically relevant. Conversely, other alleles relevant to the protocol may not get tested. For example, CYP2D6*16, a hybrid polymorphism of CYP2D6 and CYP2D7, is not represented on either chip. This allele conveys a poor metabolizer phenotype and has an allelic frequency equivalent to that of the normally tested CYP2D6*7 allele (1%).

An additional consideration when choosing a chip is the population base. For example, CYP2D6*10 is normally associated with Asians while the CYP2D6*17 allele is usually associated with an African American population. Both of these alleles are standard on both chips, however, if a clinical trial is based solely in a Caucasian population, testing for these alleles may not be necessary.

Alleles with varying frequencies based on ethnicity, such as those described above, have raised controversy in the public domain. Some clinicians and researchers state that this type of testing is viewed as racist. Ethnic groups feel that by testing these alleles companies are discriminating based on race as the companies do not suggest these populations be given the drug. In reality, the drugs are usually not recommended for some ethnicities as they may be ineffective or harmful to them. Knowing about the presence of such alleles is quite important during clinical trial design, recruiting, and reporting of results as efforts can be made to minimize risk to subjects and/or enhance our understanding of outcomes. Having expertise with these alleles is a tremendous asset to clinicians, CRO's, and pharmaceutical companies.

A tremendous amount of data can be obtained from a microarray chip, but the resulting data are reported in a raw format (i.e., the presence or absence of the allele). To compile a final genotype which in turn can predict a phenotype, data needs to be reviewed, not just reported. Returning to the CYP2D6 example, certain alleles have a great affect over other alleles. For example, a chip could reveal the presence of both CYP2D6*10 and CYP2D6*4 alleles. Independently each allele predicts a different phenotype: CYP2D6*10 conveys an intermediate metabolizer phenotype whereas CYP2D6*4 conveys a poor metabolizer phenotype. Clearly an individual can not be both an intermediate metabolizer and a poor metabolizer for the same compound. Without pharmacogenomics expertise, it may not be known that these alleles appear as a haplotype, or linked, in Caucasians. Because they are linked and they have different enzymatic affects, one allele has a greater weight than the other - CYP2D6 *4 in this case.

Simply receiving a list of allelic variants is not enough for clinicians who are not educated in genomics. What is more important is knowing the final cumulative genotype so that decisions can be made about whether or not to enroll a subject in the trial.

This question brings another concern into play. How does the final genotype become a predicted phenotype?  Existing software can list that variant alleles are present and that these variants determine a certain phenotype. However, alleles do not "act" independently. A compilation of alleles and an understanding of their linkages and associations with one another are required to accurately predict a phenotype. For example, if a subject is tested for CYP2D6 and the microarray data indicate the presence of CYP2D6*5, *4, and *10 alleles, what is the final genotype and predicted phenotype of this subject?  The presence of a CYP2D6*5 indicates that one allele of CYP2D6 has been deleted but the there is still one functional allele. The presence of CYP2D6 *4 indicates a variant that conveys a poor phenotype, but the presence of the CPY2D6 *10 variant conveys an intermediate phenotype. We have already learned the *4 and *10 alleles are linked, giving weight to *4. In this case the final genotype would be listed as CYP2D6 *4/*5 and the final predicted phenotype would be considered a poor metabolizer. While this may seem simple, more complex results such as the presence of a CYP2D6 gene duplication,*4, *10, *2, and *41 make determining a predicted phenotype more challenging. *4 has more weight than a *10 and a *41 has more weight than a *2, therefore, the initial predicted phenotype is *4/*41. What happens with the gene duplication?

What happens when the genotype is the result of CYP2D6 gene duplication?  An advantage of the DNA microarray chip is that it does determine what allele is duplicated. *4XN and *41XN are present on the Amplichip but not on the DMET chip. This difference is also noteworthy when choosing a chip for a study, so an accurate phenotype can be determined. Based on which allele is duplicated determines the predicted phenotype. If the *4XN is present, a functional but reduced *41 allele is present and the subject would be considered an intermediate metabolizer. However, if the *41 is duplicated, then the phenotype is dependent on the number of copies present. The more copies of *41, the more enzymatic activity present, and the subject could present as an extensive metabolizer. The copy number or 'N" is not determined by any chip, so it is up to the scientific expert to decide to perform further testing or to make a predicted phenotype call based on the existing data. So, while there is software that can reveal the variants and can associate the variants with a predicted phenotype, the final call for genotype and predicted phenotype requires scientific expertise in the association of genotype to phenotype. This expertise will aid clinicians or project coordinators to make appropriate decisions during their trials.

The DMET or drug metabolizing enzyme and transporter chip by Affymetrix is a cost effective microarray chip that analyzes over 225 genes in 1936 drug metabolizing markers. While this chip is mainly used in preclinical trials and Phase I or II trials, Gentris has had some inquires about using this chip in general as it's more cost effective than testing for individual assays. Again, the informed consent may only state that CYP2D6 or another DME is to be tested. The complication with using chips that have more than the desired gene and/or alleles is that data from other genes are obtained. While the data can be masked after analysis, it still has been collected. Most pharmaceutical companies do not inform the subject of their genotype or phenotype. With more and more genes/alleles being tested simultaneously, it becomes even more difficult to determine what the subject should be told. If the DMET chip was used to examine the cytochrome P450s in a subject, the whole chip is tested not just the cytochrome P450s. What happens if the chip reveals that the subject possesses UGT1A1*28 and should not be placed on Irinotecan therapy as severe neutropenia may ensue. Isn't the scientific community responsible to inform the patients/subjects so that they are aware of this? This example is another area where a scientific expert in pharmacogenomics can help. By being involved in clinical study design, these questions can be addressed and the correct decisions be made up front.

Microarray analysis is a powerful tool to investigate the probability of drug response and disease state. With such a tool, comes responsibility of ensuring the right microarray test is being performed and interpreted to answer the hypothesis correctly. As discussed with DNA arrays that contain drug metabolizing enzymes, it is important to know the differences between arrays. Having more genes and alleles on a DNA array is not always beneficial, and with so many genes and alleles being analyzed, an ethical question of whether to share the data with the subject arises. Recommendation by scientific experts in the field will always be necessary. Without the correct recommendations and interpretations, incorrect or unnecessary tests may be performed.