that a conservative approach is appropriate. We selected only the top 100 SNPs from the GWAS for subsequent replication study and a larger number of significant associations may have been uncovered if more of the most promising SNPs had been followed up. A negative association in the replication studies may not rule out a true association, since the power to detect association in the replication populations may be limited. The primary replication cohort (ICGN) is moderately powered to detect the replicated associations. Though the sample sizes of the NETT/NAS and BEOCOPD studies are relatively low, these studies include a large percentage of severely affected individuals, who may be enriched for COPD susceptibility genes. This likely account for the high rate of replication in these populations. COPD is a heterogeneous disease and we used a spirometry-based definition for COPD in all of the populations. Differences in smoking exposure, current smoking status, entry criteria and geographic origin of the cohorts might contribute to phenotypic heterogeneity and may lead to lack of replication. The fact that the replicated associations holds-up strongly and consistently in all the populations shows that phenotypic heterogeneity likely has little effect on the most significant results.
