The Alzheimer’s disease GWAS risk alleles in the ABCA7 promoter and 5’ region reduce ABCA7 expression

Recently Liu et al [2] have presented evidence from publicly available sources that the risk allele in the Alzheimer’s disease (AD)-associated variant rs4147929 at the adenosine triphosphate-binding cassette transporter subfamily A member 7 (ABCA7) locus is associated with increased ABCA7 expression. We noticed that rs4147929 is located at the 3’ end of the gene where there are few cis-acting regulatory elements and is in a perfect linkage disequilibrium (LD) with rs3752246 (risk allele frequency [RAF]=0.19 for both, R²=0.97), a missense variant (Val1527Gly) that may affect ABCA7 post-translational modification [1]. The promoter and 5’ end of ABCA7 also contain genome-significant AD-associated variants that have not been inspected by Liu et al (Supplementary Figure 1 and 2, online resource).

We investigated 16 AD-associated variants that had the lowest p values in the recent AD GWAS and were distributed along the entire ABCA7 locus for the association with ABCA7 expression in brain tissues (Supplementary Table 1 and 2, online resource). An inspection of the summary statistics from GTEx (n=165–205), ROSMAP (n=534), Mayo RNAseq (n≈260), CommonMind (n=598) PsychENCODE (n=1866), BrainSeq (phase 1 n=412, phase 2 n=551), Lopes et al (microglia only, n=100) and the ROSMAP/Mayo RNAseq/CommonMind (n=1433) and ROSMAP/GTEx v.8/PsychENCODE (n=2119) meta-analyses revealed that 15 of the variants were associated with ABCA7 expression in certain brain tissues at a nominal or adjusted level of significance and neatly segregated into two non-overlapping groups (Fig 1a; Supplementary Notes, Supplementary References, Supplementary Table 3–11, online resource). The risk alleles in the variants comprising the first group were correlated with lower ABCA7 expression or had no association with ABCA7 transcript levels, while the risk alleles in the variants in the second group were associated with higher ABCA7 expression or had no association with ABCA7 transcript levels. For example, the risk allele at rs111278892, a variant in the first group, was associated with reduced ABCA7 expression in the hippocampus (GTEx p=3.4×10⁻³, p threshold [p_th]=1.1×10⁻⁵) and cerebral cortex (ROSMAP p=0.04, p_th=8×10⁻⁸; Mayo RNAseq p=0.03, false discovery rate [FDR]=0.6; CommonMind p=6.9×10⁻⁰⁵, FDR=0.005; ROSMAP/Mayo RNAseq/CommonMind p=7.2 × 10⁻⁶, FDR=4.3×10⁻⁴; ROSMAP/GTEx v.8/PsychENCODE 8.0×10⁻⁷, p_th=1×10⁻⁶) but not in the cerebellum (GTEx p=0.6–0.15; Mayo RNAseq p=0.17). The risk allele at the coding variant rs3752246, which fell in the second group, was associated with increased ABCA7 expression in the cerebellum (GTEx p=1.7×10⁻⁷-2.9×10⁻¹², p_th=2.8×10⁻⁵-3.1×10⁻⁵; Mayo RNAseq p=1.9×10⁻⁶, FDR=4.7×10⁻⁴) and cerebral cortex (ROSMAP/Mayo RNAseq/CommonMind p=3.7×10⁻⁶, FDR=2.3×10⁻⁴; ROSMAP/GTEx v.8/PsychENCODE p=1.7×10⁻¹⁷, p_th=1×10⁻⁶) but not in the hippocampus (GTEx p=0.09). The second group included another missense variant rs3764645 (Glu188Gly). Mutations in ABCA1, a transporter closely related to ABCA7, in the regions corresponding to the locations of rs3764645 and rs3752246 in ABCA7 impair ABCA1 activity [4]. The variants in the first group tended to reside in the promoter or 5’ region of ABCA7, while those in the second group tended to locate at the 3’ area of the gene.

Fig. 1.

The selected AD-associated variants at the ABCA7 locus. a Variant location at the locus and direction of the risk allele effect on ABCA7 expression for the investigated AD-associated variants. The direction of the effect data are from GTEx. Histone methylation marks and DNase I hypersensitive sites are from ENCODE. Abbreviations: Acc - anterior cingulate cortex, NucAcc - nucleus accumbens, SubNi - substantia nigra, N/I - not imputed. b The proposed hypothesis for the mechanisms of variants’ effects on ABCA7 expression and interaction with one another. The risk alleles at the variants located toward the 5’ end of the gene likely act by disrupting the promoter, while the risk alleles at the variants located toward the 3’ end likely act by compromising ABCA7 activity. Reduced ABCA7 activity then causes accumulation of the transported substrate and feed-forward up-regulation of ABCA7 transcription, which is futile when the promoter is dysfunctional. Thus, the 5’ variants inhibit the 3’ variants with respect to ABCA7 transcription.

When any of the 13 variants were in LD, the risk alleles were always correlated (Supplementary Table 12 and 13, online resource). The risk alleles at rs72973584 (RAF=0.11) and rs3752246 (RAF=0.19) were in a tight LD (D’=0.96) but were significantly associated with lower and higher, respectively, ABCA7 expression in the cerebral cortex (Supplementary Table 10 and 11, online resource). This can occur if the less-frequent causative variant tagged by rs72973584 reverses the effect of the more-frequent causative variant tagged by rs3752246 in the same cells or if the negative effect tagged by rs72973584 in one cell type overwhelms the positive effect tagged by rs3752246 in another cell type within the cortex. Expression of ABC transporters is frequently up-regulated by the transported substrate [3]. We therefore favor the first interpretation and suggest that the coding risk alleles at rs3764645 and rs3752246 (D’=0.93) impair ABCA7 activity and cause accumulation of the transported substrate and sustained compensatory elevation of ABCA7 expression, while the risk alleles in the 5’-resident AD-associated variants both disrupt ABCA7 transcription and block the compensatory mechanism (Fig. 1b). The transcription-based otiose compensatory mechanism for impaired ABCA7 activity can explain the apparently elevated ABCA7 expression in AD [2]. Thus, the presently available eQTL data intimate complex regulatory architecture of ABCA7 expression and are consistent with the hypothesis that loss in ABCA7 transcription or activity is neurodegenerative.