Using In silico LD clumping and meta-analysis of genome-wide datasets as a complementary tool to investigate and validate new candidate biomarkers in Alzheimer's disease.

Despite the recent wealth of genome-wide association studies, insufficient power may explain why much of the heritable contribution to common diseases remains hidden. As different SNP panels are genotyped by commercial chips, increasing study power through meta-analysis is made problematic. To address these power issues we suggest an approach which permits meta-analysis of candidate SNPs from multiple GWAS. By identifying correlated SNPs from different platforms (r(2)=1), using PLINK's 'clumping' method, we generated combined p-values (using Fisher's combined and random effects meta-analysis) for each clump. P-values were corrected for the number of clumps (representing the number of independent tests). We also explored to what extent commercial platforms tag HapMap SNPs within these candidate genes. To illustrate this approach, and to serve as 'proof-of-principle', we used 3 late-onset Alzheimer's disease GWAS datasets to explore SNP-disease associations in 4 new candidate genes encoding cerebro-spinal fluid biomarkers for Alzheimer's disease; Fibrinogen γ-chain (FGG), SPARC-like1 (SPARCL1), Contactin-1 (CNTN1) and Contactin-2 (CNTN2). Genes encoding current Alzheimer's biomarkers; APP (Aβ), MAPT (Tau) and APOE were also included. This method identified two SNP 'clumps'; one clump in APOE (rs4420638) and one downstream of CNTN2 (which harboured rs7523477 and rs4951168) which were significant following random effects meta-analysis (P < 0.05). The latter was linked to three conserved SNPs in the 3'-UTR of CNTN2. We cannot rule out that this result is a false positive due to the large number of statistical tests applied; nevertheless this approach is easily applied and might well have utility in future '-omics' studies.