Pharmacogenomic Biomarkers of Susceptibility to Adverse Drug Reactions: Just Around the Corner or Pie in the Sky?

Mark I. Avigan


Personalized Medicine. 2009;6(1):67-78. 

In This Article

Challenges for Future Discovery of Genomic ADR Biomarkers

An expanding catalogue of different pharmacogenetic variants and polymorphisms at loci of interest will emerge over time, as more treatment populations for each particular drug are studied. In some cases, subsets of these will be associated with risk for different degrees of ADR severity. With this potential for complexity, there are significant but not insurmountable challenges that must be faced squarely in order to successfully translate the discovery of such pharmacogenomic variants into clinically useful predictive biomarkers for ADRs. What are these?

Specificity of a Genomic Marker(s) to Predict an ADR in an Individual

Idiosyncratic ADRs are governed both by processes associated with primary mechanism(s) of drug-induced toxicity and defects in biochemical and cellular adaptation. The frequency of each of these perturbations may be strongly affected by demographics, reflecting differences in the prevalence of genetic and/or nongenetic susceptibility factors. Correlation of a rare ADR (e.g., incidence <1/1000) with a minor variant allele whose frequency is relatively high (e.g., population frequency >1/100) suggests that the ADR risk is the result of a combinatorial effect of the variant allele with another allele(s) that segregates independently, and/or other nongenetic risk factors.[22] In this case, identification of the variant allele in an individual would likely have limited positive predictive value or specificity for development of the ADR prior to treatment with the drug. Conversely, a rare genetic variant or mutation at a specific genetic locus that strongly correlates with an ADR may have a relatively high positive predictive value and may locate a gene that plays a crucial role in modulating the potential for drug toxicity.[23]

Number Needed to Screen for a Specific Genomic Marker to Prevent One ADR

Prevalence of ADR susceptibility genomic variants as ADR susceptibility biomarkers may vary tremendously in different populations, impacting on their application(s) as predictive biomarkers for risk management. This can have a direct impact on the number needed to screen (NNS) to prevent one clinically serious ADR. For example, in the case of SJS/TEN linked to carbamazepine exposure, prevalence of HLAB*1502 is relatively high in certain Asian demographic groups (5%), and significantly lower in many European and other populations.[103] Assuming a uniform effect of the HLA B*1502 allele on risk for SJS/TEN across all populations, the NNS in a high prevalence Asian population would be approximately 1/400 to prevent one case of a carbamazepine-induced serious skin reaction, whereas the NNS would be significantly higher in populations with a lower prevalence of this risk-associated allele. Because of the potential for a clinically severe and life-threatening outcome, testing in appropriate high-prevalence populations prior to deciding on a treatment course with the anti-epileptic agent has been recommended.[103]

Limitations of Specific Methods Used to Detect a Real Association Between an ADR and a Genomic Marker

Using known genetic polymorphisms/variants as allelic and/or locus-specific linkage markers, two general approaches have been undertaken to measure genetically determined associations with ADR risk. The candidate gene approach or targeted gene analysis (TGA) screens for genetic markers contained in a small subset of genetic loci or alleles within the genome that have previously been implicated to play a role in a particular category of drug-induced toxicity. Using case-control methods comparing genomic samples from ADR cases with population-matched controls, modestly sized sample sets (<100 ADR cases) will often be sufficiently powered to screen for an association between a genetic marker and ADR risk. There are two important limitations to using this approach. First, an association(s) of a genetic variant with an ADR would be missed if it is not contained within the targeted regions of the genome that are analyzed. Second, if the prevalence of a high-risk allele(s) in the population is low, a larger number of genomic samples from patients experiencing ADRs would be required to power the study adequately. For illustrative purposes, in a case-control study when the allelic frequency is less than 10%, access to more than 1000 ADR case genomic samples will be required to determine a relative risk of only 1.5 at 95% power.[22] Because of typically limited availability of genomic samples from clinical trial patients with serious forms of a rare ADR, such real associations may not be detected. By contrast, to detect a fourfold risk for an ADR associated with a variant allele whose prevalence in a demographic group is 10%, compared with individuals without this variant, it is necessary to test genomic samples from only 105 ADR cases to achieve 95% power. In this case, the bar set to detect a higher relative risk (larger effect size) requires testing of samples from a smaller number of cases, since a statistically significant association can be more easily demonstrated. However, the absolute effect size of the highly prevalent variant in the absence of other genetic and/or nongenetic risk factors could be small. Thus, if the ADR of concern is rare (<1/1000), the positive predictive value and specificity of this marker for the ADR will be low. In some instances, limited specificity does not preclude clinical utility of the genomic variant as a biomarker. For example, even though the positive predictive value of HLAB*1502 in carbamazepine-treated Asians is only 1/20, with ample choices of other anti-epileptic agents for managing patients, testing provides an important clinical tool to avoid exposure to this agent in high-risk individuals.[103]

As they are not hypothesis-driven, genome-wide association studies (GWAS) survey the whole genome using a select library of SNPs that are distributed throughout the genome and serve as tags of virtually all allelic variants within a population (e.g., 'HapMAP').[24,25,26] Allelic variants that increase risk for ADRs possess altered DNA sequences that are responsible for altered gene product activity. These include DNA base-pair mutations, insertions, deletions, transpositions and microsatellite lengthening. ADR-associated SNPs may either play a direct causal role in ADR risk or simply serve as markers that are in linkage disequilibrium with nearby variant genes that are responsible for this effect.

Establishing biomarker links between specific SNPs and particular ADR phenotypes is challenging. Because of random variation owing to the large number of genomic markers that are tested, there would be a high false discovery rate of ADR associations that must be discounted in measurements of their statistical significance.[27] Moreover, the power requirement for numbers of ADR cases needed to achieve significance substantially rise as the prevalence of the ADR falls in a treatment population. Therefore, GWAS will only detect ADR associations with highly prevalent minor variant alleles in a test population, whereas real associations with low prevalence variants will be missed.