This study, assigning case or controls status on the basis of CSF biomarkers, provides further confirmatory evidence that CR1, PICALM, TOMM40, and APOE E4 are risk factors for the development of AD pathology. This was possible using just over 300 subjects, an order of magnitude fewer than used in traditional GWAS studies. These findings suggest that confirmatory or exploratory genetic analyses based on biomarker evidence of AD pathology may have increased power to detect case/control differences, and may therefore be possible using smaller sample sizes.
Whilst due to the small sample size confidence intervals were large, the minor alleles of CR1, PICALM, TOMM40
E4 were associated with greater odds ratios than have previously been suggested in many other GWAS, significantly so in the case of APOE
E4. Thus odds ratios were for CR1
(rs1408077) 1.59, PICALM
(rs541458) 0.68, TOMM40
(rs2075650) 4.29 and APOE E4
vs. E3 8.32, with meta-analyses of previous studies reporting odds ratios of 1.13, 0.88, 2.79 and 3.68 respectively (Bertram et al., 2007
). A previous confirmatory GWAS study using 740 of the ADNI cohort and employing a logistic regression model across clinical diagnosis groups reported significant, but smaller effects of APOE
E4 [OR=2.07] and CR1
(rs1408077) [OR=1.27], and no effect of PICALM
(Biffi et al., 2010
). These differences are likely to reflect the difficulties of relying on clinical diagnosis: in keeping with previous reports (De Meyer et al., 2010
; Shaw et al., 2009
) of all the controls available for analysis, ~20% would have been classified as CSF-positive; and ~19% of the MCI group and ~4% of the AD group as CSF-negative. Basing the analysis on patients with a CSF AD profile and those without, independent of clinical diagnosis, might explain the larger odds ratios; and whilst considerable caution is required given the small numbers in the study and the wide confidence intervals, this suggests that these haplotypes may confer larger risk of developing AD pathology than has previously been described.
Compared to results from formal GWAS, there was a directionally similar but non-significant association for CLU. This is likely to an issue of insufficient power. Based on case/control minor allele frequencies from the Alzgene meta-analysis, 232 cases and 94 controls would have 99% and 85% power (5% level) to detect differences in APOE ε4 and TOMM40 respectively, but only 5-7% power for CLU, CR1, BIN1 or PICALM. Based on these estimates, the chance of detecting significance for CR1 and PICALM in this sample is <1/400, providing further support for the hypothesis that better group separation may be achievable by basing diagnosis on disease biomarkers than clinical diagnosis.
There are a number of important caveats that need to be considered in relation to this study. Assigning case/control status neither on the basis of cognition nor on evidence of neurodegeneration means that the genetic risks identified can only truly be associated with the development of CSF signatures of AD and not of AD itself. Nonetheless, these findings which accord closely with previous literature, suggest that employing endo-phenotypic traits may be a useful means of providing confirmatory and exploratory GWAS studies in neurodegenerative diseases. The use of any CSF cut-off is inevitably associated with a degree of inaccuracy, and standardisation of CSF measurement is important if similar, pre-defined cut-offs are to be used in other studies. This study is not a formal GWAS, but was designed as to replicate known genetic risk factors as a proof-of-concept for the use of an enrichment strategy. As such, and to allow comparisons with other such studies and the Alzgene meta-analytic data, uncorrected p-values are presented. Applying a strict Bonferonni correction results in an adjusted statistical significance level of p=0.00625, at which level only the TOMM40 and APOE genes remain significant. This is likely to reflect the much higher risk factor conferred by these two genes. Determination of genes with relatively small influences may however also aid in our understanding of the pathogenesis of neurodegenerative diseases, and whilst use of endophenotypes to enrich case/control studies may increase power to determine genetic associations, this does not negate the fact that large sample sizes will be required to determine small effects.
There is increasing realisation that a substantial proportion of apparently normal older individuals may be in the prodromal stage of AD (Schott et al., 2010
). Presuming these individuals are also likely to harbour risk variants, GWAS studies assuming that do not take this into account risk missing potential genetic associations, or underestimating the effects of identified genes. Using biomarkers to define cases and controls, or as quantitative traits, may increase the power of studies to detect genetic influences: indeed during the revision of this paper, a formal GWAS study based on the CSF data from the ADNI cohort was published (Kim et al., 2011
). The findings reported here require replication in larger cohorts of patients with CSF; and in subjects stratified on the basis of other biomarkers including amyloid PET imaging.