Genome-wide association (GWA) studies lead to long-awaited breakthroughs in the genetics of late-onset Alzheimer disease (AD) [MIM 104300] [1] by providing conclusive genetic association evidence for novel AD risk genes [2–5]. Notably, GWA significance with similar effect sizes was reached for the top single nucleotide polymorphism (SNP) rs11136000 in the clusterin gene (CLU) [MIM 185430] [1]. The CLU protein (also known as apolipoprotein J) is a multifunctional protein showing functional similarities with the major apolipoprotein of the brain, apolipoprotein E (APOE) [6]. In relation to AD, CLU expression is increased in pyramidal neurons and astrocytes of the hippocampus and entorhinal cortex, the most severely affected brain regions in AD [7]. CLU is present in senile plaques [8], binds to Aβ and is involved in Aβ₄₂ clearance across the blood brain barrier [9]. Moreover, CLU enhances endocytosis of Aβ aggregates to brain phagocytes [10]. Taking its variety of physiological functions, CLU could be a guardian or enemy in AD [11].

The CLU transcriptional unit is located in the chromosomal region 8p21-p12 and comprises 9 exons in the longest transcript that translates in the main CLU protein isoform of a 449 amino-acid residues. The CLU precursor peptide is internally cleaved to produce an α- and β-subunit, held together by disulphide bridges and is subsequently secreted from the cell (Figure 1)[12].

In this follow-up study aiming at identifying the genetic variant underlying the CLU association with increased AD risk, we examined the CLU genetic variability using a resequencing approach including all coding exons and regulatory regions in a well-documented Flanders-Belgian patient/control cohort (n = 1930 subjects) [2, 3]. Significant genetic findings were verified by targeted resequencing in two independently ascertained French [3] and Canadian [13] AD cohorts and by meta-analyses of the genetic data sets generated in this study with published data sets obtained of previous genetic screenings of CLU in AD cohorts [14, 15].

We used an extensive resequencing approach to follow-up on the significant association of common SNPs in the CLU locus with increased risk for AD in 2 independent GWA studies. In the discovery stage, we used unbiased resequencing of all coding and regulatory regions of CLU in a Flanders-Belgian AD cohort followed by targeted sequencing of 2 independent replication cohorts of French and Canadian origin. In the Flanders-Belgian AD cohort, we obtained significant evidence for clustering of rare coding variants in exons 5 to 8, predicting 3 possible (p.T345M, p.N369H, p.T440M) and one probable (p.R338W) damaging non-synonymous substitutions and a 9-bp insertion/deletion p.T445_D447del affecting the CLU β-chain domain. Interestingly, 4 AD patients carried multiple CLU variations: a patient with a positive family history and onset age of 68 years harbored 2 predicted pathogenic substitutions (p.R338W, p.T345M) and three patients carried the deletion p.T445_D447del together with the benign variant p.A309T. Targeted resequencing in the Lille and Toronto AD cohorts identified predicted pathogenic substitutions (p.R338W and p.N369H) in French AD patients and the p.T445_D447del deletion in French and Canadian AD patients. Noteworthy, we observed haplotype sharing with 7 markers around the insertion/deletion site in 3 Belgian and 2 French AD patient carriers which is suggestive for a common ancestor. Partial haplotype sharing was further observed with another French and Canadian patient.

Moreover, we identified a 1-bp insertion/deletion variation p.I303NfsX13, predicting a premature stop codon after 315 amino acid residues (24 of which in β-chain), in a French patient (onset age 68 years). This frame-shift mutation would most likely produce an unstable transcript that is degraded by the nonsense mediated mRNA decay control system or an unstable C-truncated protein that is degraded within the cell, however, no patient material was available for further testing. Substitutions p.R338W and p.T345M are positioned inside the disulphide rich region which is involved in the folding of the α and β-domains into a heterodimer complex. Variant p.N369H deletes a CLU N-glycosylation signal [19], and since CLU is highly glycosylated (20-25% of its total mass comprises carbohydrates), this substitution might affect CLU processing or functioning in ligand interaction. Although absent from all Belgian and French control individuals (> 2500 control chromosomes), p.N369H was present in one Canadian control person aged 58 years, which is suggestive of an intermediate penetrance for this substitution. We also observed 3 non-synonymous substitutions with possible damaging effects (α-chain and β-chain) in both Flanders-Belgian patients and control individuals. As these variants were found in 1 to 4 control individuals only and the control group had on average a younger inclusion age than the patient group, these substitutions might represent risk factors of intermediate disease penetrance and might still contribute to AD risk. In contrast, all non-synonymous substitutions observed in control individuals were predicted to be benign and to unlikely affect CLU protein functioning, strengthening our observation that rare non-synonymous substitutions might contribute to AD risk.

Meta-analysis including all rare coding variants affecting the CLU β-chain domain observed in the discovery and replication cohorts plus those reported in 3 additional cohorts [14, 15], totaling 5772 patients and controls, lent further support to an increased occurrence of rare coding variants predicted to affect the CLU β-chain in AD patients compared to healthy individuals (OR_MH 1.96 [95% CI 1.18-3.25]; p = 0.009). Assessing the prevalence of variations affecting the β-chain domain by targeted resequencing of CLU exons 5 to 8 in a Caribbean Hispanic AD cohort, detected novel benign variants while predicted pathogenic variants found in stage I and II AD cohorts were absent. Of note, predicted pathogenic variant p.N369H was more frequently observed in Hispanic patient and control individuals then in stage I and II Caucasian cohorts, suggesting this variant might be a polymorphism which is in agreement with previous reports in African-Americans [15]. Excluding p.N369H from our meta-analysis on rare variants did not alter the significance of our observation that rare coding variants affecting the CLU β-chain are more frequently observed in AD patients than control individuals (p = 0.001). It is possible that rare CLU variants contribute in a different way to disease risk in diverse populations and alternatively, that part of the genetic CLU variability is population-specific (e.g. Hispanic origin). Screening other and larger AD cohorts of different ethnicity and conducting functional analyses on putative pathogenic variants will help unraveling the role of these rare coding variants to AD risk. In this context it is relevant to mention that we have primarily used protein prediction algorithms that have their limitations as well. While our meta-analysis suggested that the CLU β-chain is involved in AD risk, to adequately discriminate between rare or low frequent pathogenic variants and benign polymorphisms, their pathogenicity needs to be experimentally evaluated. These experiments will have to be performed along with comparable efforts aiming at understanding the role of CLU in AD pathogenesis.

Although we identified an association of AD risk with clustering of rare coding variants affecting the CLU β chain, this association was independent from the association with rs11136000, the top GWA SNP [2, 3]. Genotyping rs11136000 in the stage I and II cohorts followed by a meta- analysis showed significant association (p = 0.013). This association remained significant after exclusion of carriers with predicted pathogenic β-chain variants. This is in line with recent LRRK2 observations in which independent common risk associations were found after exclusion of carriers with known pathogenic mutations [20]. Further, high-density association fine-mapping by genotyping common SNPs throughout the CLU locus in the stage I cohort, identified 4 additional SNP associations, that all except one (rs9331908) were in high LD with rs11136000 (r² > 0.89) and located in the 5' CLU region. This association region ranges from CLU intron 1 to exon 5, which is only marginally overlapping with the rare variant association region in the β-chain (exon 5 to 8). Combined with the two independent association signals for common and rare CLU variants, this could imply two distinct mechanisms underlying AD risk.

Further genotyping of the 4 significant SNPs and combined meta-analyses on all 3 AD cohorts showed, after Bonferroni correction, the strongest evidence of association with rs1532278 (p = 0.005). This variant is located in an intron 3 sequence that strongly resembles a regulatory element based on sequence alignment of seven species. Also, association evidence was found for rs867230 in CLU intron 1, which potentially affects a MEF2 transcription factor binding site. These two novel associated common variants may represent variants of functional relevance underlying the common association signal with rs11136000 in various populations [21–23]. In agreement with other reports [14, 21], we did not observe any association with common coding variants. Though we did not find association with the recently described splice site variant rs9331888 [24] in our Flanders-Belgian population (OR 1.11 [95% CI 0.94-1.31]; p = 0.2), our association findings do point to common variants with possible regulatory effects (rs1532278, rs867230). It is likely that a number of common and rare CLU variants contribute to an increased risk for AD. In addition, different mechanisms may contribute to disease: our common CLU associations might regulate CLU transcription while our rare CLU association signal results from non-synonymous variants affecting protein functioning. Whether these rare CLU variants represent loss or gain of function variants remains at present unclear and requires further studies.

In conclusion, in-depth CLU resequencing showed significant clustering and association of rare coding variants with AD risk in a combined meta-analysis of 3 Caucasian AD cohorts analyzed in this study and cohorts of previous published studies. The rare coding variants are non-synonymous substitutions and insertion/deletion mutations that affect the CLU β-chain domain. While the physiological properties of the β-chain domain remain unclear, our data suggests that this protein subunit may be of interest in AD pathogenesis, and merits follow-up with detailed functional analyses. Also, in our combined AD data set, the rare coding variant association was independent of the common AD risk association signal that we fine-mapped to a more 5' region of CLU. The strongest association was not obtained with GWA SNP rs11136000 but with an intron 3 SNP (rs1532278) located in a sequence with high regulatory potential. Altogether, our data suggest that CLU may be a risk gene in which multiple rare and common variants have independent effects on AD disease susceptibility.