Pharmacogenomics research is becoming more prevalent in both academia and the pharmaceutical industry. While some discoveries have been integrated into practice and are benefiting patient care, these successes have been limited given the vast amount of research undertaken. However, the advances in high-throughput genomic technologies, better study designs and improved understanding of complexity, means that pharmacogenomic determinants of drug response will continue to be identified. It is important to develop an understanding of the basis of pharmacogenomics in clinical teams to allow accurate interpretation of the findings, and facilitate their implementation into clinical care (if appropriate). This article explains the science behind pharmacogenomics, and describes some of the challenges that have been encountered in the field, with a specific focus on paediatrics.
- Basic Science
Statistics from Altmetric.com
The goal of medical therapy is to select the right drug, at the right dose, for the right person. Within medicine, and especially paediatrics, some ‘personalisation’ of medicines already occurs, with clinicians considering the patient's age, weight, renal/hepatic function and coexisting medications, as well as the most appropriate formulation, when prescribing. Despite these considerations, the patient may not gain any benefit from the medication, the estimated efficacy of medications varies from 25% to 80%, (most 50–75%).1 In addition, patients using any medication risk harm through adverse drug reactions (ADRs). In adult practice, ADRs are responsible for 6.5% of admissions, costing the National Health Service (NHS) an estimated £466 million.2 ADRs in children also represent a significant cause of hospital admissions (2–4%),3–6 and complications in inpatients (9–15%).5 ,7
Pharmacogenomic research provides a method of generating additional information, potentially identifying patients who are likely to respond to medication, or those in whom there is a high probability of a severe ADR. As well as benefit to the individual patient, there are potential benefits for the healthcare service and pharmaceutical industry that could accrue from pharmacogenomic research and application in clinical practice (figure 1).
This review article will detail some of the ways in which genetic variation can occur, the types of pharmacogenomic research and analysis that are undertaken, the difficulties that can occur in application into clinical practice, and where it has already been integrated clinically and is affecting adult and paediatric practice.
Definition of pharmacogenomics
The terms ‘pharmacogenomics’ and ‘pharmacogenetics’ are often used interchangeably to describe how genetic determinants affect an individual's response to a medication. Pharmacogenetics, the narrower term, is defined as ‘the study of interindividual variations in DNA sequence related to drug response’.8 However, it has become clear that the sequence of individual genes is not the only factor involved, but many genes interact with each other and, thereby, affect the functioning of the cell, organ and individual. Pharmacogenomics is a broader term, covering these additional factors, defined as ‘the study of the variability of the expression of individual genes relevant to the disease susceptibility as well as drug response at cellular, individual or population level’.8 Pharmacogenomics is the term that will be used in the remainder of this article.
Pharmacogenomics, therefore, affects both the pharmacokinetics (PK) (absorption, distribution, metabolism, excretion) and pharmacodynamics (PD) of a drug (figure 2).
Causes of variability in gene expression
Genetic variation can occur in a number of different ways. The most common inherited sequence variations in the human genome are single nucleotide polymorphisms (SNP, pronounced snip), where one base pair in the DNA sequence is replaced with another (eg, C to T). A SNP is distinguished from a mutation solely by how frequently it occurs. By convention, if the frequency of the base pair substitution occurs in ≥1% of the population it is a SNP, <1% denotes it is a mutation.9 It has been estimated that there are approximately 10 million SNPs in the human genome, but only approximately 60 000 are located in the coding regions of genes, and only half of these affect the amino acid sequence of the protein produced.10 ,11 However, the SNP does not have to be located in the coding sequence to have an effect on disease; a SNP in a promoter region could increase or decrease the expression of a gene without affecting the quality of the resulting protein.
In addition to SNPs, there are many other ways in which DNA varies, genes are expressed and, subsequently, the resulting protein expression can be altered (table 1).
The most common study design within pharmacogenomics is the case-control association study—patients receiving a drug are divided into those with a positive response (or ADR) (the cases) and those who do not exhibit this positive response (or ADR) (the controls). These two groups are then genotyped, using one of the methodologies shown below, and the frequency of mutations or SNPs is compared between the cases and controls.
A ‘candidate gene’ approach is the simplest technique, and is still frequently employed. Identification of the candidate gene(s) can be guided by biological plausibility; searching the literature to establish if particular genes, for instance, drug metabolising enzymes (DMEs) or receptors, are relevant to the PK or PD of the drug. Genes thus identified are then genotyped looking for variations in sequence, and if these variations are associated with the outcome under investigation (response, or ADR). Pharmacogenomic effects on warfarin dose requirements were found using a candidate gene approach. Studies have shown that polymorphisms in two genes, vitamin K epoxide reductase complex, subunit 1 (VKORC1), and Cytochrome P450 2C9 (CYP2C9), in combination with environmental factors, exert a significant effect on warfarin dosing in adults,29–31 and adult dosing algorithms incorporating these data have been developed.30 ,32 Paediatric studies into warfarin using a candidate gene approach have shown similar effects of VKORC1 and CYP2C9 polymorphisms,33–35 but larger prospective studies are required before this is incorporated into dosing algorithms for children.
However, the candidate gene approach has drawbacks. If there is no a priori knowledge of a gene affecting the pathway of a drug, it may be overlooked. Assuming the correct gene/s is/are selected, it can be technically difficult to identify the functionally important polymorphisms, which can include promoter or enhancer polymorphisms, gene duplications, synonymous coding SNPs that affect transcript stability, or intronic SNPs that cause splice variants that create early stop codons.36 Even after the analysis, it is more common for a large proportion of the variability to remain unexplained. This may reflect the polygenic nature of a drug's journey, from absorption to excretion, as well as external factors, such as adherence, environmental factors or effects of the underlying disease. In addition, the candidate gene approach cannot account for the post-transcriptional, epigenetic and post-translational modifications detailed in table 1.
Technological improvements, and reductions in cost, have allowed researchers to analyse much greater quantities of genetic data. A genome-wide association study (GWAS) is ‘a case-control study in which genetic variation, often measured as SNPs that form haplotypes across the entire genome, is compared between people with a particular condition and unaffected individuals.’37 A GWAS is capable of detecting over 1 million SNPs in a DNA sample,38 with accuracy approaching 100%.39 The frequency of the SNPs found in the samples taken from the cases are then tested for association against controls, or if controls are not available (usually due to the cost of undertaking a GWAS on further participants), against a population from one of the published databases. These published databases include the HapMap project, a publicly accessible database of over 3 million SNPs from multiple ethnic backgrounds.40 Specific to children, the Children's Hospital of Philadelphia and its Centre for Applied Genomics has GWAS data on over 100 000 children. A GWAS generally requires large sample sizes to achieve statistical significance and avoid false positives. Indeed, for a GWAS, it is usual for an association only to be considered significant if p<0.00000001.
The pharmacogenomic relationship between carbamazepine use and onset of Stevens Johnson Syndrome (SJS) was explored using GWAS techniques. Susceptibility to SJS in Han Chinese patients (adult and child) using carbamazepine and carrying the HLA-B*1502 allele was discovered using a candidate gene approach.41 However, in adult epilepsy sufferers of northern European descent, this relationship was not found, and it was only when a GWAS was undertaken that a relationship between patients who possessed the human leukocyte antigen (HLA) allele HLA-A*3101, the risk of SJS was noted.42 This showed an increased risk of SJS from 5.0% to 26.0%.
Greater genomic detail can still be gained by sequencing either the exome (all of the DNA, ie, translated into protein, omitting the introns, which is about 1% of the human genome) or even the entire genome. This produces a huge volume of data, and requires a plan for the management of these data, and experienced statistical input, to ensure interpretable results are generated.
Regardless of the method used, a pharmacogenomic association between a drug and outcome is only valid if it can be replicated in a separate cohort (with similar drug history, disease burden and ethnic origin). Ideally, a replication cohort should be included in the publication describing the association, but this is not always the case. Failure to replicate findings in subsequent publications leads to contradictory data in the literature. However, it may not always be possible to identify a replication cohort, for instance, when investigating rare events. In such cases, functional analysis of the implicated SNP(s) may provide data on the biological plausibility of the association.
There are also other concerns about the general quality of the publications relating to pharmacogenomic studies, with deficits noted in all areas, but particularly in regard to testing for multiple associations.43 There are some well-recognised factors that can predispose to false associations (box 1).44
Considerations when setting up a pharmacogenomic study to guard against false results
Matching of genetic background for cases and controls
Ensures that any genetic difference is related to the disease and not biased-sampling.
Certainly ethnicity and preferably geographical areas (or grandparents’ place of birth) should be considered.
Multiple unlinked markers can be used to assess this confounding variable.
A sufficient sample size
Must be powered to detect variants that are common, but have low relatives risks, or rare, but have high relative risks.
Rare variants with low relative risk are currently beyond the reach of genetic epidemiology because of the massive sample sizes required for such results.
Adequate statistical methods to analyse data
Hardy–Weinberg equilibrium to screen control group.
Using up-to-date statistical methods in this evolving field of statistics.
Can be performed with either a second case-control association study or family-based study.
Consistency in phenotype definition within and between studies
The reporting of randomised controlled trials is guided by the Consolidated Standards of Reporting Trials (CONSORT) statement, detailing what information should be presented,45 and this has improved reporting.46 There are now discussions in the literature about how modifications of the CONSORT guidelines could help with the presentation of pharmacogenomic data.47 There are also guidelines on the publication of genetic association studies.48
Paediatric-specific issues and pharmacogenomics
Additional factors need to be considered when undertaking pharmacogenomic research in a paediatric population, including consent, sample collection techniques and volumes of blood available. With regards to consent, it might be expected that parents would be reluctant to participate in such research, but our own unpublished experience is that participation in research is not hindered by including a request for collection of a DNA sample, provided there is a clear explanation of both the purpose and limitations of the DNA analysis that will be undertaken. Collecting a DNA sample can be more problematical in a paediatric population. DNA has previously been collected using blood sampling (EDTA samples), but if there is no prospect of direct participant benefit, then this is ethically problematic (venepuncture is an invasive, painful procedure), unless the blood was already being collected because of the underlying illness. Volume of blood that can be taken is also limited in a research setting.49 However, it is now possible to collect DNA using saliva samples,50 and although the quantity of DNA recovered is reduced compared with blood, the quality is the same.50 Advances in the saliva collection kits mean that samples can now be collected from children too young to spit or with learning difficulties as well.
Although the DNA sequence remains constant from birth, the expression of genes is not constant. There are a number of examples of genes which have greater expression in early life than in adulthood, or vice versa. This is well exemplified by drug-metabolising enzymes, including the CYP450 enzymes and UDP glucuronyltransferase (UGT). CYP3A7 has detectable expression at 50–60 days postconceptual age in fetal liver, but declines after the first week of postnatal life.51 As CYP3A7 expression declines, CYP3A4/3A5 expression begins to dramatically increase at 1 week of age until it reaches 30% of adult levels by 1 month.51 This maintains a level of CYP3A protein expression, but function may vary as CYP3A4 and CYP3A7 exhibit different substrate specificities, catalytic efficiency and, consequently, metabolic capacity.51 UGT has only 1% of adult activity at birth52 before rapidly increasing to adult levels by 14 weeks.53 Decreased glucuronidation of chloramphenicol in newborn infants is believed to be a factor in the susceptibility of babies to Grey Baby syndrome.54 ,55
Variations in expression such as these, combined with the different disease states children experience, and the general paucity of information on developmental PK during early childhood and puberty, means extrapolation of the results of adult pharmacogenomic data to the paediatric population is problematical.
Areas of current research and application
There are numerous paediatric pharmacogenomic studies published in a wide variety of populations, drugs and diseases, from attention deficit hyperactivity disorder (ADHD) medication and asthma (steroids, β2 agonists and leukotriene modifiers), to chemotherapy and anticoagulation (warfarin). It is beyond the scope of a single review to cover all the published research in this area, especially as there is little scientific consensus on the results in many areas. However, that is not to say that consensus and application into clinical practice is not possible. Indeed, there is already application of pharmacogenomic research in paediatric oncology practice in the UK, with treatment of acute lymphoblastic leukaemia using the standard treatment protocol (UKALL-2003) incorporating a pharmacogenetic test to determine if patients are poor metabolisers of thiopurine methyltransferase, and then varying the dose of 6-mercaptopurine by up to 90%.56 By detecting those who are poor metabolisers, unwanted accumulation of 6-mercaptopurine can be avoided leading to a reduced incidence of bone marrow toxicity. Another example that is also routine in practice includes genotyping HIV positive patients prior to using abacavir, to ensure that a severe hypersensitivity reaction is avoided.57 Although largely used in adults because of the prevalence of HIV, genotyping for HLA-B*5701 before prescribing abacavir to children is now advised in current guidelines.58
Another area of research particularly relevant to paediatrics is aminoglycoside-induced ototoxicity. Multiple studies conducted in many different ethnicities have shown that mutations within the mitochondrial 12S ribosomal RNA (mt 12S-rRNA) gene to be more prevalent in congenital/prelingual hearing impairment, especially when associated with aminoglycoside use,59 evidence that is further supported by phenotypic consistency in matrilineal relatives.60 Given the high incidence of aminoglycoside-induced otoxicity, and the ever increasing availability of rapid and inexpensive screening methods, it seems likely that preadministrative screening has the potential to substantially reduce the number of affected children.
The proliferation of pharmacogenomic studies is likely to continue (figure 3), and it would be hoped that the results from these will continue to filter into clinical practice. Other scientific fields are also developing that will work synergistically with pharmacogenomics, including transcriptomics (the study of gene transcripts, analysing complementary DNA), proteomics (study of the expressed protein complement at a particular time) and metabolomics (study of metabolite profiling).44 ,61
Both metabolomics and proteomics provide a ‘snap-shot’ of evolving cellular processes that will be predominantly predetermined by the genome, but will also be specific to disease, age/development, drugs, environment, microbiota and many other factors. They will enhance our understanding at a molecular level alongside pharmacogenomics, and will, conceivably, identify profiles, protein or metabolite, that can be used as phenotypes for drug response, or predict an adverse drug reaction susceptibility. They do, however, both share the common problem of obtaining the relevant source of sampling, as results will be specific to the analysed cells.
These developments mean we have the increasing ability to look at the whole pathway of cellular events, from DNA, through the selective expression of genes and the post-transcriptional modifications that occur, to the protein produced and the post-translational modifications that can affect structure and function, and the metabolite changes that occur (as a result of gene function, and also of environmental factors). Techniques to integrate data from the different -omics technologies are being developed through integrative or systems biology approaches, and this is going to be essential in understanding how drugs affect the function of individual cells, and of the cell-cell interaction of whole organs and, ultimately, of the whole body. Understanding these pathways will, hopefully, lead to a greater understanding of disease and treatment in both populations and individuals, but the volume and complexity of the data may be daunting. It is important that clinicians are kept up to date on these technologies to allow integration into clinical care (where appropriate) to maximise patient benefit.
Pharmacogenomic research in paediatric populations is ongoing in the UK and around the world. Although there are currently few clinical applications in paediatrics, the potential to improve the personalisation of medicines, improving the efficacy and safety profile of medicines used in children is considerable. The use of comprehensive approaches which span the whole spectrum from biomarker discovery to the demonstration of clinical utility and implementation in clinical practice will be essential in the successful translation into practice.
Contributors DBH and BT are joint first authors. DBH: devised original concept, contributed to initial draft article and revisions in later drafts for important intellectual content. BT: involved in designing article, researching subject matter, and contributed to initial manuscript. Involved in later revisions. RLS and MP: provided supervisory support for initial stages of concept and article drafting, then critically revised for important intellectual content. All have approved the final version to be published.
Funding BT is supported by a fellowship grant from Pfizer. We acknowledge the support of the NHS Chair of Pharmacogenetics from the UK Department of Health.
Competing interests None.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.