An additional consideration when viewing these results is the difference in ascertainment method across COGA, SAGE and OZ-ALC. The discovery sample (COGA) consisted of extended pedigrees ascertained for a dense family history of alcoholism and it was not expected that all variants associated with alcohol-related measures in such densely affected pedigrees would generalize to other cohorts. SAGE cases were selected for DSM-IV alcohol dependence from among several studies focused on alcohol, tobacco and cocaine and thus, as expected, we note a stronger degree of replication in this sample. In contrast, OZ-ALC comprises of samples ascertained for heavy smoking, discordance of heavy alcohol consumption measures and also for large sibship size (without any oversampling for substance-related phenotypes). Therefore, it is not surprising that replication in OZ-ALC, a less severely affected sample, is weaker for AO-AD, but occurs for ages of onset for earlier drinking milestones (for example, AO-I) and for measures that are quantitative indices of problem drinking (for example, AD-SX). In OZ-ALC, AO-I and AO-R, as well as AD-SX and Maxdrinks may serve as proxies for genetic liability to problematic
