Successful GWASs of cancer pharmacogenomic phenotypes are possible (), but replication of germline variant associations has been difficult, often because of challenges associated with large, clinical trials and a lack of well-defined replication populations in oncology. Germline DNA collection and consent for genetic studies from as many participants in future cancer drug clinical trials as possible will allow genome-wide pharmacogenomic association studies of cohorts with standardized dosing and phenotype collection. Another approach that can be considered is pathway-based analysis (BOX 3
); like methods that combine rare variants within a gene into a single association test, variants within a pathway can also be combined. Pathway-based approaches provide a more powerful analysis of GWAS data sets41,89
than do analyses of single variants or genes. Such approaches may be particularly useful for pharmacogenomic analysis of oncology clinical trials, which are often underpowered to uncover variants with small effect sizes.
Box 3. Pathway-based association approaches in cancer pharmacogenomics
Pathway-based association analysis combines variants in genes in a known molecular pathway to test whether the pathway is associated with the phenotype. Genes do not work in isolation; instead, complex molecular networks and pathways are often involved in biological processes. Thus, it is feasible that variation in different genes from the same pathway may lead to similar phenotypic outcomes. The pathway-based approach is useful because an implicated pathway is readily biologically interpretable. For example, the interleukin 12 (IL-12)–IL-23 cytokine pathway has been found to associate with susceptibility to the autoimmune disorder Crohn’s disease in multiple populations119
, and this is plausible given the role of cytokines in immune responses. It may not be possible to uncover variants conferring modest phenotypic risk in multiple underpowered genome-wide association studies (GWASs), but these variants can sometimes be readily identified by a pathway-based approach in a single study119
. Therefore, such approaches may be particularly useful in cancer pharmacogenomics. Importantly, as the most associated gene in a pathway might not be the best candidate for therapeutic intervention, knowledge of potential targets within a pathway may have clinical implications for finding new drugs that either decrease toxicity or increase tumour response.
Multiple statistical methods have been developed to combine variants within a pathway into an association test and have been reviewed elsewhere41,89
. Key considerations are which pathways to test and how to assign variants to genes. Genome-wide approaches often define pathways according to the Kyoto Encyclopedia of Genes and Genomes (KEGG)120
and the Gene Ontology121
. Variants can be assigned to genes on the basis of either a predefined base pair distance or putative variant function (for example, amino acid change or regulatory activity). Candidate pathway approaches may also be useful in cancer pharmacogenomics. The Pharmacogenomics Knowledgebase (PharmGKB)122
manually curates pharmacokinetic and pharmacodynamic pathways for well-studied drugs, including many anticancer agents. The pathway for a particular drug could be used to determine whether variation in included genes associates with the variation in response to that drug. Additionally, in the case of a lesser-studied drug, multiple PharmGKB pathways could be tested to determine whether any known pathways also associate with phenotypes induced by the lesser-studied drug. Such an analysis could reveal related mechanisms of action between drugs.
Cancer pharmacogenomic studies have demonstrated the potential to make therapy safer and more effective for patients. Although most current recommendations are for somatic variants (BOX 1
), the FDA has included information in the labels of at least seven cancer drugs for which germline variants predict toxicity90
. Because of phenotypic heterogeneity (for example, some heterozygotes for reduced TPMT
activity tolerate full mercaptopurine doses, but others do not), the FDA will often recommend rather than require a particular pharmacogenetic test (for example, see these FDA summary minutes). The Pharmacogenomics Research Network routinely publishes gene-based drug-dosing guidelines for well-established associations, such as TPMT
and mercaptopurine, through the Clinical Pharmacogenetics Implementation Consortium (CPIC)7,91
. For these guidelines to improve patient care, full clinical implementation will require widespread physician education, acceptance and automated decision support.
As studies move beyond known drug targets and drug metabolism enzymes, the common variants associated with cancer pharmacogenomic traits may have smaller effect sizes so that they are able to predict a response only when combined. Until discoveries are made and validated to high confidence, clinical utility cannot be assessed. Recently, two polygenic modelling methods have been developed to detect the contribution of larger numbers of common SNPs to complex phenotypes in GWAS data: polygenic risk score analysis92
and mixed linear modelling93,94
. In polygenic risk score analysis, an additive polygenic risk score based on SNPs below a predetermined P
value threshold in a discovery set of samples is then tested in an independent set of samples. The mixed linear modelling method estimates additive genetic variance under a mixed linear model with a random effect representing the polygenic component of trait variation. Applying similar models to the analysis of cancer pharmacogenomics may implicate new biological factors that influence such traits and inform the types of genetic variants that should be examined in future studies.
Clinical translation will be more challenging when results move beyond individual genes of strong effect and into such polygenic models. However, advances in sequencing technologies, statistical genetics analysis methods and clinical trial designs have shown promise for additional cancer pharmacogenomic discovery. In the future, every patient’s catalogue of drug-related germline variants may be readily available, and algorithms that combine well-validated genetic variants of small effect to explain a large proportion of the variance in treatment toxicity or response could be applied to a patient’s data to provide clinicians with immediate treatment recommendations95
. Until then, with the goal of reducing toxicity and improving patient outcomes in mind, the next wave of cancer pharmacogenomic discovery will inform researchers about the underlying genetic architecture of variable drug response and may potentially reveal genes and pathways that can be used as targets for new drugs.