Understanding in a comprehensive manner the genetic risk factors contributing to variation in drug response is a crucial component of the realization of personalized medicine. The drugs evaluated in our study are widely used in the treatment of many cancer types, including ovarian, colorectal, testicular, and lung; all are associated with particular toxicities and resistance. Although SNPs have long been used in association studies to elucidate the effect of genetic polymorphisms on drug response, CNVs have been relatively understudied. Recent genome-wide surveys of CNVs have now established that these structural variants are a common phenomenon in the human genome [5
]. With rapid advances in methods that facilitate their assay and analysis, variation in copy number for genes encoding drug metabolizing enzymes has been increasingly implicated for their dramatic consequences on responsiveness to drugs. Such CNVs have been observed to alter gene dosage and are thus likely to play an important role in determining drug efficacy or toxicity.
In this study, we set out to utilize recent developments in the assay of CNVs in recent population-scale projects, including an extensive comparative genomic hybridization-based catalog of CNVs [4
] and a map of structural variants based on whole genome DNA sequencing data (the 1000 Genomes Project) [14
], in order to evaluate the role of CNVs in cellular sensitivity to chemotherapeutic agents. The cell lines for the samples express a sizable part of the genome [28
], thus enabling the investigation of genes represented in biologically relevant pathways. While the cancer genome is clearly necessary for understanding chemotherapeutic response, the importance of germline genetic variation in drug sensitivity has also been consistently demonstrated [15
For each drug included in our study, we found that the top associated CNVs are more likely to act as eQTLs and predict transcript levels than minor allele frequency (MAF)-matched SNPs. The overlap of the drug susceptibility-associated CNVs with expression-associated CNVs (eCNVs) is greater than is expected, based on simulation studies. Consistent with a previous report [6
], CNVs associated with cellular sensitivity to drug treatment are not likely to overlap exons, suggesting that they act not to disrupt coding sequence but to regulate gene expression. The high proportion of eQTLs among the CNVs associated with cellular sensitivity to each of the drugs further supports the hypothesis that these CNVs mediate their phenotypic consequences through their effect on the transcriptome. Genome-wide studies of pharmacologic phenotypes, such as response to antineoplastic agents, may benefit from studies of CNVs as eQTLs.
This study, to our knowledge, is the first comprehensive genome-wide study of the effect of CNVs, from the most extensive array-based and sequencing-based surveys of these structural variants, on pharmacologic phenotypes. In contrast to a recent disease susceptibility study that concluded that most CNVs that are well-typed have been indirectly explored by SNP studies [6
], we found a number of CNVs associated with drug sensitivity that are independent of SNPs. These CNVs therefore constitute novel genetic variations that have not been previously interrogated by SNP-based GWAS of pharmacologic phenotypes. Our discovery of drug susceptibility-associated variations, in the form of CNVs, that are independent of previous SNP findings and that show evidence for altering gene expression as eQTLs, suggests that CNVs should be included in comprehensive pharmacogenomic studies.
Candidate pharmacogenetic studies on drug metabolism-related genes, namely CYP2D6, CYP2A6, SULT1A1 and GSTM1, have documented the effect of CNVs on gene activity. Our results strongly support the necessity of integrating both SNP and CNV data to tighten the genotype-phenotype gap in pharmacogenetic studies. While the functional validation we conducted in this study may not allow robust predictions, the functional characterization of the effect of CCND1 mRNA level on cellular sensitivity to etoposide underscores the importance of considering the role of the transcripts that are the targets of drug susceptibility-associated CNVs (acting as eQTLs) in conferring drug susceptibility.
We found a significant overlap (P
= 7.7 × 10-10
) between the CNVs associated with cisplatin and carboplatin. Platinating agents share a similar mechanism of therapeutic action and interact with DNA to form interstrand and intrastrand cross-links, leading to cytotoxic DNA lesions and eventually apoptosis-induced cell death. Our findings strongly support the hypothesis that CNV-based mechanisms play a crucial role in determining platinum sensitivity. Particularly, we identified a duplication that is associated with cellular sensitivity to both carboplatin and cisplatin. Furthermore, the CNV predicts the expression of glutathione reductase (GSR
), a gene that has been the subject of several studies on cisplatin sensitivity [26
]. The glutathione pathway is involved in the metabolism of platinum compounds, which are subject to inactivation by glutathione conjugation [27
A significant level of overlap is also observed with the topoisomerase II inhibitors. Daunorubicin is a DNA intercalator that indirectly interacts with topo II while etoposide binds directly to the enzyme. We identified 14 CNVs associated with both etoposide and daunorubicin at P < 0.05. The extent of overlap between the platinating agents (as well as between the topoisomerase II inhibitors) is significantly higher than the level of overlap across drug classes (7%).
There is a general caveat to our findings concerning the set of CNVs included in this analysis. The CNVs tested for association with cellular sensitivity to drugs may be biased towards genotypeable variants; consequently, many highly complex regions may have been excluded. Furthermore, our study makes no assertions about low frequency variants. Nevertheless, our findings represent the most comprehensive study of the effect of common CNVs, from the most extensive map of these variants available, on chemotherapeutic susceptibility to a wide array of drugs.
Finally, we provide the results of our genome-wide study of CNVs and sensitivity to chemotherapeutic agents in a publicly available online database, PACdb. Analysis results on our cell-based model are easy to query, which should allow investigators to utilize the resource as a discovery platform or as a validation tool for clinical observations.