Case subjects were Caucasian Americans with LOAD (n = 1009; mean age-at-onset [AAO] 72.8 ± 6.2 [SD] years; 67.7% female; 7.3% autopsy-confirmed) recruited by the University of Pittsburgh Alzheimer’s Disease Research Center. All cases were evaluated clinically and met criteria for probable or possible AD [11] or by autopsy and met neuropathological criteria for definite AD [13, 14]. Controls were Caucasian Americans of age 60 or above with no psychiatric or neurological disorders (n = 1009; mean age-at-baseline 74.1 ± 6.2 [SD] years; 59.9% female; 1.3% autopsy-confirmed). All experiments on human subjects were conducted in accordance with the Declaration of Helsinki, and all procedures were carried out with the adequate understanding and written consent of the subjects. The genetic study was approved by the University of Pittsburgh Institutional Review Board.

Genotypes for the six SORL1 SNPs were ascertained from genomic DNA using TaqMan SNP genotyping assays (Applied Biosystems, Foster City, California). Case and control samples were present on each 384-well plate used in genotyping. To estimate genotyping error rates, 10% of the samples were selected at random and repeated. Genotypes for APOE were determined either as previously described [5] or using TaqMan SNP genotyping assays.

Allele and genotype frequencies were calculated by the direct allele-counting method. Goodness of fit to Hardy–Weinberg equilibrium was tested using the χ² test. Differences between genotype and allele frequencies in cases and controls were tested with the χ² test. Differences between cases and controls stratified by APOE*4 carrier status were also tested with the χ² test. Haplotype frequencies were estimated in cases and controls, and the global difference in frequencies was tested. Linkage disequilibrium (LD) between the SNPs was estimated by calculating D′ and r² between each pair. Haplotype frequencies were then again compared after eliminating highly-correlated SNPs (those with r² > 0.8). These statistics were calculated using R 2.2.0 with the genetics and haplo.stats packages attached [16, 21, 24]. Power to detect associations was determined with PS 2.1.30 [3].

The allele and genotype frequencies for the SORL1 SNPs are shown in Table 1. Neither cases nor controls had genotype frequencies which differed significantly from those expected under Hardy–Weinberg equilibrium. The differences between cases and controls do not rise to statistical significance (range: p = 0.177–0.980).

Among, non–APOE*4 carriers, the differences were not statistically significant (range: p = 0.357–0.922). Among APOE*4 carriers, rs661057 was associated with AD risk (genotype frequency differences, p = 0.022; alleles, p = 0.008). None of the other SNPs were observed to be associated (range: p = 0.109–0.765).

Haplotype frequencies calculated for all six SNPs did not differ with statistical significance (p = 0.627) (data not shown). LD between the SNPs is shown in Table 2. Because the three SNPs in intron 6 (rs668387, rs689021 and rs641120) were highly correlated (r² = 0.988–0.996), haplotype frequencies were recalculated between just four of the SNPs (rs661057, rs668387, rs2070045 and rs3824968). The frequency differences were not significant (p = 0.858).

Given the minor allele frequencies at α = 0.05, we had 80% power to detect a risk odds-ratio of 1.23 with statistical significance for rs2070045 which had the lowest minor allele frequencies and 1.20 for rs641120 which had the highest.

Sortilin-related receptor 1 (also known as sorLA and LR11) is an excellent biological candidate gene due to its role in amyloid precursor protein (APP) processing and evidence that variation in SORL1 levels are associated with both APP [1, 15] and LOAD [19]. Association between variation in SORL1 and LOAD has been reported in Northern Europeans, Caucasian Americans, Caribbean Hispanics and Israeli Arabs [18]; Caribbean Hispanics and Caucasian Americans [6]; Han Chinese [23]; and Caucasian Americans with Down syndrome [7]. However, the same large study that reported association in multiple groups [18] found no association in Caucasian American or African American families. Two groups examined SORL1 SNPs in publicly available data from the same genomewide association studies of AD [17] and found marginal associations with some variants in specific regions of SORL1. Meng et al. [12] observed associated SNPs in the interval from exon 7 to exon 18;Webster et al. [25], from intron 25 to intron 39 (SORL1 comprises 48 exons). However, neither replicated any of Rogaeva et al.’s [18] significant SNPs in particular. Another high-density genomewide association study of AD failed to find any association between disease risk and any of 41 SNPs included from the SORL1 region [8].

Liu et al. [10] also failed to find evidence of association between SORL1 and LOAD in a large Dutch pedigree, although they did observe evidence of linkage nearby at 11q25 that they believe to be associated not with SORL1 but with OPCML and HNT. Li et al. [9] examined several SNPs and haplotypes in a population of size and demographics similar to ours, and observed only marginal association with AD risk for a single SNP (rs2070045) which they dismissed because it is not statistically significant after correcting for multiple testing.