A Meta-Analysis of Genome-Wide Association Studies in Electrocardiographic QT Interval Duration (QTGEN)

QT interval duration reflecting myocardial repolarization on the electrocardiogram is a heritable risk factor for sudden cardiac death and drug-induced arrhythmias.

We conducted a meta-analysis of three genome-wide association studies in 13,685 individuals of European ancestry from the Framingham Heart Study (FHS; genotyped on the Affy 500K platform plus an add-on chip), the Rotterdam Study (ERGO; Illumina 550) and the Cardiovascular Health Study (CHS; Illumina 370cnv) as part of the CHARGE Consortium. In these samples, we imputed non-genotyped SNPs using HapMap CEU from Phase II as the reference panel. Using inverse variance weights, we combined genomic-controlled association results for 2.5 million SNPs under an additive model from the three cohorts in fixed effects meta-analysis.

You can download here the scripts that we developed for the meta-analysis. To keep it light, we have packaged the association results for only the SNP with the strongest association (lowest p-value): rs12143842 in the NOS1AP gene on chromosome 1, with 0.21 s.d. QT interval duration increase per copy of the minor allele (T on the fwd/+ strand of build 36) and an overall p-value = 8e-46.

The tarball contains the following files:

  • rs12143842_nos1ap.dat -- the data (rsID, chromosome, position, BETA, SE, p-value, coded and non-coded alleles, coded allele frequency and imputation quality) for the top SNP for the three studies in QTGEN (see the documentation in the Perl script for details)
  • meta_qtgen.params -- study-specific parameters (lambda, sample size)
  • hapmap_ceu_r22_b36_fwd.bim -- PLINK BIM file of HapMap CEU release 22 (build 36)
  • hapmap_ceu_r22_b36_fwd.freq.frq -- PLINK frequency output file
  • all.genes.b36.txt -- genomic coordinates (build 36) for all known RefSeq genes
  • meta_qt_022409.pl -- the main script written in Perl (with extensive documentation inside)
  • meta.R -- a very simple R script (3 lines) to convert z-scores to p-values
  • run_this.csh -- a csh script to run this example
  • example.out -- the output from the Perl script
  • example_pvalues.out -- the output from the R script (only 2 extra columns -- PFIXED for the p-value based on the inverse variance weighted meta-analysis, and PSQRTN for the p-value based on the effective sample size weighted z-score based meta-analysis)

The results for each study are first compared to the two observed HapMap alleles and alleles reverse complemented (i.e. A->T, C->G, G->C, T->A) if they do not match. (This will catch non-A/T and non-C/G SNPs.) The minor allele from HapMap CEU genotypes (data provided by the pre-made PLINK files) is used to define the coded allele for the meta-analysis, regardless of observed allele frequency in individual studies. If the coded allele is not the minor allele in HapMap, the sign of BETA (effect estimate) is flipped. To implement genomic control, the lambda value is used to correct the standard error as follows, SE(corrected) = SE * sqrt(lambda), where lambda is specified in the parameter file (provided). Each effect estimate (BETA) is standardized to the standard deviation of the study-specific adjusted residuals to put all results on the s.d. scale (specified in the parameter file). The ratio of the observed to the expected variance of the imputed allele frequency (dosage) is used as the quality metric for the imputation of a given SNP (see de Bakker et al. HMG 2008). To account for the difference in sample size and power, we compute N(effective) which discounts the total sample size as follows: N(effective) = N * (observed/expected variance). The N(effective) for the meta-analysis result is the sum of all cohort-specific N(effective) values. The script calculates two types of p-values (z-scores). We note that the correlation of the -log(p-value) using inverse variance weights and effective sample size weights for genome-wide QTGEN results with p<0.01 was high (>99%).

Questions?

You can contact Paul de Bakker here.

References

P.I.W. de Bakker, M.A.R. Ferreira, X. Jia, B.M. Neale, S. Raychaudhuri, B.F. Voight. Practical aspects of imputation-driven meta-analysis of genome-wide association studies. Human Molecular Genetics. 2008 Oct 15;17(R2):R122-8.

C. Newton-Cheh, M. Eijgelsheim, K. Rice, P.I.W. de Bakker, X. Yin, K. Estrada, J. Bis, K. Marciante, F. Rivadeneira, P.A. Noseworthy, N. Sotoodehnia, N.L. Smith, J.I. Rotter, J.A. Kors, J.C.M. Witteman, A. Hofman, S.R. Heckbert, C.J. O'Donnell, A.G. Uitterlinden, B.M. Psaty, T. Lumley, M.G. Larson, B.H.Ch. Stricker. Common variants at ten loci influence QT interval duration in the QTGEN study. Nature Genetics. In press.