Our thoughts on control sample augmentation were initially motivated by the emergence of multisample GWA study designs, such as that employed by the WTCCC, in which there are multiple disease samples drawn from the same, largely homogeneous population [The Wellcome Trust Case Control Consortium, 2007]. Given this internal “genetic matching” and assuming no unmeasured con-founders, the greatest theoretical power to detect genetic effects would be gained by forming an expanded reference group for each disease cohort by combining the primary controls with cases of all other diseases. However, in this type of multisample design, some of the diseases may have overlapping genetic aetiologies, thus inducing genotypic correlations between the different phenotypes due to pleiotropic effects. For example, the main WTCCC experiment includes cases of multiple autoimmune diseases and metabolic disorders. For pleiotropic genes influencing correlated traits, such as PTPN22 for T1D and RA, making use of cases of one disease as controls for the other can reduce power to detect disease variants, a “dilution” of the genetic effect due to bias in the control group. The inclusion of disease samples
