Neither replication nor simulation supports a role for the axon guidance pathway in the genetics of Parkinson's disease

Abstract

Susceptibility to sporadic Parkinson's diseases (PD) is thought to be influence by both genetic and environmental factors and thier interaction with each other. Statistical models including multiples variants in axon guidance pathways genes have recenty been purported to be capable of predicting PD risk, survival free of disease and age at disease onset; however the specific models have not undergone independent validation. Here we tested the best proposed risk panel of 23 single nucleotide polymorphisms (SNPs)in two PD sample sets, with a total of 525 cases and 518 controls. By single marker analysis, only one marker was significantly associated with PD risk in one of samples sets (rs6692804: p=0.03). Multi-marker analysis using the reported model found a mild association in one sample sets (two sided P=0.049, odds ratio for each score change=1.07) but no significance in the other (two sided P=0.98, odds ratio=1), a stark conrast to the reported strong association with PD risk (P=4.64 × 10-38, odds ratio as high as 90.8). Following a procedure similar to that used to build the reported model, simulated muti-marker models containing SNPs from randomly chosen genes in a genome wide PD dataset produced P=values that were highly significant and indistinguishable from similar models where disease status was permuted (3.13 × 10-23 to 4.90 × 10-64), demontrating the potential for overfitting the model building process. Together, these result challenge the robustness of the reported panel of genetic a markers to predict PD risk in particular and a role of the axon guidance pathways in PD genetics in general. © 2008 Li et al.