Feasibility of whole genome association study to find SNPs predictive of colorectal cancer outcome

B29

Purpose
 Genome wide association studies (GWAS) have successfully identified low-penetrance loci associated with colorectal (CRC), breast, lung, and prostate cancer. Here we describe the use of high-density SNP arrays to perform a GWAS aimed at identifying candidate germline loci which can predict outcome and might influence success of chemotherapy in stage 2 and 3 CRC.
 Methods
 We typed 318,237 tagSNPs on Illumina HumanHap300duo arrays (Illumina Inc, San Diego, CA) in 933 patients enrolled in the Victor trial of AJCC stage 2 or 3 CRC treated with surgery and chemo- or radiotherapy as appropriate and randomized to rofecoxib or placebo. Two patients were excluded from the analysis due to call rates <95%.
 Survival analysis per SNP was based on a minimum of three years follow-up. Hazard ratios (HR) were generated by log-rank test for time-to-event data using a genotypic, recessive, and dominant model. Allelic comparison was performed using Fisher’s exact test. All statistical analysis was performed using Stata 9.2 (StataCorp, College Station, TX). Overall significance levels were based on minimum p-value. The p-value cut-off was set at p<0.0001 to avoid missing potentially associated SNPs.
 Results
 Age and sex were well matched between relapsed (775) and non-relapsed patients (156) included in the analysis. As expected, AJCC stage 3 was more common in the relapse group. SNP typing was highly accurate (>99.97% concordance between repeats) and reliable (median call rate 99.95%). Testing each model with a q-q plot did not reveal any substructure in the population tested, with good alignment of observed and expected p-value distribution.
 Of the 309200 autosomal SNPs, 180 failed and 56 were non-informative (monomorphic). 565 SNPs had p<0.0001. Of these SNPs, 457 had a minor allele homozygous frequency <10 in the control group, and 11 SNPs were not in Hardy-Weinberg equilibrium (HWE) with p<0.05 and therefore excluded. The remaining 97 SNPs were analysed for their positions relative to known genes. SNPs in intergenic regions were examined for their location in LD blocks tagging nearby genes.
 Based on this strategy, 33 SNPs in or near candidate genes were identified. These genes are predominantly involved in cell adhesion and motility, cell signaling or are kinases or transcription factors, but none of the SNPs have previously been implicated in CRC prognosis. 19 SNPs are in or near genes with limited plausibility as prognostic marker, and 45 SNPs do not appear plausible as prognostic marker, either because the closest gene is an unlikely candidate (30 SNPs) or there are no known genes within 150Mb (15 SNPs).
 The top 20 SNPs based p-value (including ‘implausible’ SNPs), and 20 SNPs based on known function will be verified in 1500 patients with stage 3 CRC in the second phase of this project.
 Conclusion
 WGA studies are feasible for testing germline genetic variation as a marker of prognosis in stage 2 and 3 CRC.