A comparison of combinatorial partitioning and linear regression for the detection of epistatic effects of theACE I/DandPAI‐1 4G/5Gpolymorphisms on plasma PAI‐1 levels

The detection and characterization of epistasis or non-additive gene-gene interactions remains a statistical challenge in genetic epidemiology. The recently developed combinatorial partitioning method (CPM) may overcome some of the limitations of linear regression for the exploratory analysis of non-additive epistatic effects. The goal of this study was to compare CPM with linear regression analysis for the exploratory analysis of non-additive interactive effects of the angiotensin converting enzyme (ACE) insertion/deletion (I/D) and plasminogen activator inhibitor 1 (PAI-1) 4G/5G polymorphisms on plasma PAI-1 levels in a sample of 50 unrelated African Americans and 117 unrelated Caucasians. Using linear regression, we documented the additive effects of the ACE and PAI-1 genes on plasma PAI-1 levels in African American females (R(2) = 0.10), African American males (R(2) = 0.16), Caucasian females (R(2) = 0.11), and Caucasian males (R2 = 0.09). Using CPM, we found evidence for non-additive effects of the ACE and PAI-1 genes in both African American females (R(2) = 0.22) and African American males (R(2) = 0.24) but not in Caucasian females (R(2) = 0.10) or Caucasian males (R(2) = 0.11). The results of this exploratory data analysis support previous experimental, clinical, and epidemiological studies that have proposed as a working hypothesis that the ACE gene mediates interaction effects of the fibrinolytic and renin-angiotensin systems on plasma levels of PAI-1.