Genome-Wide Association and Pharmacological Profiling of 29 Anticancer Agents Using Lymphoblastoid Cell Lines

Abstract

Aims Association mapping with lymphoblastoid cell lines (LCLs) is a promising approach in pharmacogenomics research, and in the current study we utilize this model to perform association mapping for 29 chemotherapy drugs. Materials and Methods Currently, we use LCLs to perform genome-wide association mapping of the cytotoxic response of 520 European Americans to 29 different anticancer drugs, the largest LCL study to date. A novel association approach using a multivariate analysis of covariance design was employed with the software program MAGWAS, testing for differences in the dose-response profiles between genotypes without making assumptions about the response curve or the biological mode of association. Additionally, by classifying 25 of the 29 drugs into 8 families according to structural and mechanistic relationships, MAGWAS was used to test for associations that were shared across each drug family. Finally, a unique algorithm using multivariate responses and multiple linear regressions across pairs of response curves was used for unsupervised clustering of drugs. Results Among the single drug studies, suggestive associations were obtained for 18 loci, 12 within/near genes. Three of these, MED12L, CHN2 and MGMT, have been previously implicated in cancer pharmacogenomics. The drug family associations resulted in 4 additional suggestive loci (3 contained within/near genes). One of these genes, HDAC4, associated with the DNA alkylating agents, shows possible clinical interactions with temozolomide. For the drug clustering analysis, 18 of 25 drugs clustered into the appropriate family. Conclusions This study demonstrates the utility of LCLs for identifying genes having clinical importance in drug response, for assigning unclassified agents to specific drug families, and proposes new candidate genes for follow-up in a large number of chemotherapy drugs.

Evaluating the Role of Admixture in Cancer Therapy Via in Vitro Drug Response and Multivariate Genome-Wide Associations

“…Genotype data and drug response phenotypes were collected from two sources: (i) non-Hispanic Caucasian patients from the Pharmacogenomics and Risk of Cardiovascular Disease clinical trial study at the Children's Hospital of Oakland Research Institute (CHORI) at Oak Ridge [17] and (ii) Hispanic Mexican-American individuals from the Human Variation Panel (HVP) of the International HapMap consortium [23]. For (i), genotype data were collected for each of the 500 patients using one of two technologies -314,621 or 620,901 markers using HumanHap300 BeadChip or HumanQuad610 BeadChip platforms, respectively, as previously described in detail in [17].…”

“…For (i), genotype data were collected for each of the 500 patients using one of two technologies -314,621 or 620,901 markers using HumanHap300 BeadChip or HumanQuad610 BeadChip platforms, respectively, as previously described in detail in [17]. The phenotypic data on dose-dependent, drug-induced cytotoxicity was measured in LCLs derived from each subject.…”

“…Genotyping data collection and quality control (QC) from the CHORI dataset was described previously elsewhere [17]. Identical QC methods were applied to the HVP in order to facilitate comparison of the two distinct cohorts as well as subsequent combined analyses for association mapping.…”

“…Dose-dependent cytotoxicity data were collected across 28 drugs for each cell line. The study design and QC pipeline for the CHORI phenotypic data were described elsewhere in detail [17]. We will summarize the phenotyping and QC methods for the 93 HVP samples, and refer the reader to [17] for specific details on CHORI.…”

“…As recently reviewed in [4], Epstein-Barr virus (EBV) immortalized lymphoblastoid cell lines (LCLs) have been used to demonstrate the heritability of dose response [15] and performing association mapping for pharmacogenomics [16,17] and toxicogenomics [18]. Additionally, the association mapping results can be functionally tested using knock-down experiments [19,20] or candidate gene analyses in clinical outcomes [21,22].…”