The AAOS-association landscape shows that highly associated SNPs at the SPI1/CELF1 locus span multiple genes (Fig. 1a). In the previous IGAP GWAS1, rs10838725 showed the strongest association at this locus (P=6.7×10−6, 1.1×10−8 vs. rs1057233: P=5.4×10−6, 5.9×10−7 in IGAP stage I, stage I and II combined, respectively). Rs10838725 is located in the intron of CELF1, which was assigned as the putative causal gene at this locus1 based on proximity to the index SNP, a criterion that has often proven to be erroneous14. In our survival analysis, however, rs10838725 showed weak association (P=0.12, HR=1.02, 95% CI=0.99–1.05) whereas rs1057233, located in the 3′UTR of a neighboring gene, SPI1, showed the strongest association (Table 1, P=5.4×10−6). The two SNPs exhibit only moderate linkage disequilibrium in the ADGC subset of the IGAP GWAS (R2=0.21, D’=0.96). Applying conditional logistic regression analysis of AD risk in the ADGC dataset, we found that rs1057233 remained significantly associated with AD after adjusting for rs10838725 (P=3.2×10−4), whereas rs10838725 showed no evidence of association after adjusting for rs1057233 (P=0.66). This suggests that rs1057233 is in stronger LD with the AD risk causal variant.
