Deleterious ABCA7 mutations and transcript rescue mechanisms in early onset Alzheimer’s disease

The EOAD patients and control individuals were recruited within the European Early Onset Dementia (EU EOD) consortium. Details about recruitment, in- and exclusion criteria, and demographic and patient characteristics were previously described [51]. In summary, patients [n = 928, 60.2% (558/927) female, 51.2% (391/763) APOE ε4+ (MIM: 107741)] were diagnosed according to NINCDS-ADRDA [34] and/or NIA-AA [19, 35] diagnostic criteria, and had a mean onset age of 57.4 ± 5.6 years (30–65). In 47.7% (274/575) of patients, a positive familial history for dementia was reported. Patients were ascertained in neurological centers with an expertise in memory disorders, and originated from Spain (n = 403), Italy (n = 159), Sweden (n = 160), Germany (n = 83), Portugal (n = 66), Greece (n = 52), and the Czech Republic (n = 5) (Table S1). Seventeen EOAD patients carried a known pathogenic mutation in PSEN1, PSEN2, or APP (MIM: 104311, 600759, and 104760). Control individuals (n = 980, 64.9% (607/936) female) had a mean inclusion age of 63.9 ± 7.7 years (34-89), and were recruited in Spain (n = 223), Italy (n = 304), Sweden (n = 295), Portugal (n = 120), Greece (n = 35), and the Czech Republic (n = 3). The study was approved by the respective ethics committees, and all participants and/or their legal guardian provided written informed consent before inclusion.

Genomic DNA (gDNA) was amplified with Illustra GenomiPhi v2 (Thermo Fisher, Waltham, MA, USA). Enrichment of 43 of the 47 canonical ABCA7 (RefSeq NM_019112.3) exons and splice sites was based on a custom multiplex PCR assay (primer sequences available upon request) generated with mpcr software (Multiplicom, Niel, Belgium). Regions of interest were amplified using flanking primers with universal adapter sequences [5′-TCGTCGGCAGCGTCAGATGTGTATAAGAGACAG-(sequence-specific forward primer) and 5′-GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAG-(sequence-specific reverse primer)]. Amplicons were PCR barcoded with Nextera XT sequences (Illumina, USA) targeting previously incorporated adapters. Samples were 2 × 300 bp paired-end sequenced with MiSeq Reagent Kit v3 (Illumina, San Diego, CA, USA). Adapter clipping of demultiplexed FASTQ output was performed with fastq-mcf [3], aligned with Burrows-Wheeler Aligner MEMv0.7.5a [30] and variants were called with GATKv2.4 UnifiedGenotyper and GATKv3.5 HaplotypeCaller. Variants were annotated with GenomeComb [42] and SnpEff [9].

For four exons (12, 17, 19, and 23), no compatible primers were found. Exons 12, 17, and 19 were analyzed with Sanger sequencing: exons were PCR amplified, subsequently dideoxy-terminated with BigDye termination cycle sequencing kit v3.1 (Thermo Fisher), and sequenced with ABI3730 DNA Analyzer (Thermo Fisher). Sanger sequences were analyzed using Seqman (DNAstar, Madison, WI, USA) and NovoSNP software [52]. Exon 23 (73 bp + 4 bp splice sites) was not duly screened. Of note, this exon contains no known PTC mutations [Exome Aggregation Consortium (ExAC) (accessed April 2017)]. In addition, gaps are present (less than 50% of individuals sequenced at 20× coverage) in exon 9 (29 bp), 16 (23 bp), 18 (4 bp), and 21 (87 bp), as depicted in Fig. 1, mainly due to repetitive regions flanking ABCA7 exons, limiting efficient PCR primer design.

The sequencing read-depth statistics for all coding and splice sites regions are represented in Fig. 1 based on Rsamtools [36] and Circos visualization [25]. All PTC variants and c.5570+5G>C were validated on gDNA with Sanger sequencing as described above. In addition, we validated rare [minor allele frequency (MAF) <0.01%] missense variants with at least 20× read depth, a less than threefold difference between reference and alternative allele counts for heterozygous variants, and a Phred-scaled CADD score [24] above 20, corresponding to the 1% most deleterious variants in the human genome. This cutoff was chosen given the relatively high mutational tolerance of ABCA7 and incomplete disease penetrance for very deleterious PTC mutations.

Rare variant (MAF < 0.01) association meta-analysis by ethnicity was performed separately for PTC and predicted deleterious missense mutations using heterogeneous genetic effect SKAT-O statistics within the MetaSKAT framework [29]. APOE dosage (number of ε4 alleles) was tested as a covariate. Summary odds ratios (OR_MH) were calculated with a Cochran–Mantel–Haenszel test in PLINK [40], stratified by country of origin, on individuals without missing genotyping information.

For common variants (MAF ≥ 0.01), a fixed-effects (Cochran–Mantel–Haenszel) meta-analysis was used to calculate single variant allelic association. We evaluated all SNPs located within the CDS and 15 bp exon flanking regions passing Hardy–Weinberg equilibrium quality control (p > 0.001). Odds ratios are reported for the minor allele with 95% confidence intervals. Pairwise linkage disequilibrium (LD) was calculated in PLINK [40]. To account for LD between variants, multiple testing thresholds were based on spectral decomposition [38], leading to a study-wide multiple testing p value cutoff of 0.0033. Carriers of a known pathogenic mutation in APP, PSEN1, or PSEN2 were excluded from association analyses.

Fresh frozen brain was available for patient EOD-P1 (c.67-1G > A mutation). RNA was extracted from anterior cingulate cortex tissue with Ambion RiboPure™ kit (Life Technologies, Carlsbad, CA, USA). Reagent volumes were adapted to support 25 mg of input brain material. In addition, RNA from blood was extracted with Tempus™ blood RNA tube (Thermo Fisher) for carriers EOD-P7 (p.Met370fs) and EOD-P19 (c.3577+1G>C). For a subset of ABCA7 PTC mutations, EBV-transformed lymphocytes (p.Glu709fs, p.Trp1336*, and c.5570+5G>C) or fresh frozen brain tissue (p.Leu1403fs) from AD patients was available through the BELNEU consortium [11]. RNA was extracted with Ambion RiboPure™ kit (Life Technologies). All RNA was treated with Ambion TURBODNase kit (Life Technologies) to degrade potential gDNA contamination. First-strand cDNA was synthesized with SuperScript^®III First-Strand Synthesis System (Life Technologies).

We performed cDNA sequencing on a MinION platform (Oxford Nanopore Technologies, Oxford, UK). Exonic primers were designed with Primer3Plus to generate amplicons containing exons of interest, flanked by at least one splicing event on each side (Table S2). All PCR amplifications were performed with 35 cycles to have sufficient amplification of the lowly expressed ABCA7 in all tissues. Titanium Taq (Clontech Laboratories, Mountain View, CA, USA) or Platinum Taq (Thermo Fisher) enzymes were used with or without supplementation of betain, depending on the GC content of the amplicon. In case of overlapping PCR amplicons, a 5′ barcoding adapter sequence was added to the primers and amplicons were barcoded with the PCR 96 barcoding genomic DNA (R9) kit (Oxford Nanopore Technologies) using 15 amplification cycles on a 1/125 diluted template. DNA was purified with Agencourt AMPure XP SPRI beads (Beckman Coulter, Brea, CA, USA) and concentrations were measured with Qubit (Thermo Fisher). Amplicons were then pooled equimolar to a total quantity of 0.2 pmol. The resulting PCR library was then further prepared according to the manufacturer’s protocol. Briefly, DNA was treated with NEBNext Ultra II End Repair/dA-Tailing Module (New England BioLabs, Ipswich, MA, USA). The product was then purified with AMPure XP beads, after which the ONT sequencing adapter and hairpin were ligated with the NEB Blunt/TA Ligase Master Mix (New England BioLabs). Next, biotin containing hairpin tethers were added and washed MyOne Streptavidin C1 Dynabeads (Thermo Fisher) were used for pull-down. Finally, the sequencing library was eluted from the beads and supplemented with Running Buffer with Fuel Mix. SQK-MAP006/NSK007 chemistry and FLO-MAP103/MIN104 flow cells were used for sequencing. Base calling was done with the 2D plus barcoding protocol (Metrichor, Oxford, UK) after which FASTQ sequences were extracted with poretools v0.6.0 [33] according to the “best” protocol. Sequencing reads were aligned to the human genome (hg19) with GMAP v2016-06-30 [53] to account for splicing events. Wild-type (WT), nonsense, and rescue alleles were quantified with Rsamtools v1.26.0 for aligned sequences [36], and with GMAP splicesites output for splicing events. The PTC expression fraction was calculated as (expression_Nonsense/expression_Nonsense+WT); 0% corresponds to complete degradation of PTC bearing transcripts and 50% to equal observance of PTC and WT transcripts. Results were validated with Sanger sequencing of cDNA amplicons as described above and compared with RNA sequencing (RNAseq) data from a Belgian unrelated AD-control transcriptome study [51]. Briefly, total RNA was isolated from Epstein–Barr virus immortalized lymphoblasts derived from whole blood lymphocytes. RNA isolation of lymphoblast cells was performed with the RNeasy mini kit (Qiagen Inc., Valencia, CA, USA) according to the manufacturer’s protocol. Sequence libraries were constructed using the Truseq stranded mRNA Library Prep Kit v2 (Illumina) using 1 mg total RNA for each sample. Sequencing of prepared libraries was performed using an Illumina HiSeq 2000 sequencer generating an average of 72 × 106 ± 6 × 106 paired-end sequence reads/sample. Subsequent data processing consisted out of removal of read adapters and trimming of read ends with Trimmomatic [6]. Reads were then aligned to hg19 using the Bowtie short read aligner integrated in Tophat2 [22].

Sequencing of the ABCA7 CDS in 928 EOAD patients and 980 control individuals revealed 17 different PTC mutations (six frameshift indels, six nonsense mutations, and five PTC-introducing splice site mutations) (Table 1), which were more frequent in EOAD patients (3.02%; n = 28) than controls (0.61%; n = 6) [p value = 0.0004, OR_MH = 5.01 (95% CI = 1.59–15.72)]. This association remained significant after correction for APOE (p = 0.001). Most PTC mutations (n = 10) were not reported before and—together with c.67-1G > A, c.3577 + 1G > C, and p.Arg1489*—were absent from control individuals (Fig. 1; Table S3). Patient EOD-P5 carried two PTC mutations (p.Trp69*and c.579+1G>T) segregating on the same haplotype (Figure S1). Two previously reported variants (p.Trp1336*, c.4416+2T>G) were only observed in control individuals. The two most frequent PTC mutations (p.Glu709fs and p.Leu1403fs) were observed more in patients than control individuals (Fig. 1; Table 1). All carriers of p.Glu709fs and p.Leu1403fs shared the same respective haplotype (Table S4). Carriers of less frequent multiple occurring variants (c.67-1G>A, p.Thr849fs, c.4416+2T>G, and p.Arg1489) were geographically confined (respectively, originating from Spain, Portugal, Italy, and the Iberian Peninsula), though no familial relatedness was known (Table S3).

Patients (n = 28, 75.0% female, 54.2% (13/24) APOE ε4 +) carrying PTC mutations had a mean onset age of 56.7 ± 5.5 years (42-65, upper limit determined by inclusion criteria for EOAD), and a mean disease duration of 8.6 ± 1.8 years (Table S3). In comparison, carriers of established pathogenic mutations in PSEN1 (n = 12), PSEN2 (n = 3), or APP (n = 2) had a lower mean onset age of, respectively, 46.0 ± 9.0, 53.0 ± 5.1, or 49.5 ± 1.5 years (35-63, p = 0.0002) (Figure S2). Patient EOD-P6.1 (ABCA7 c.302 + 1G>C) also carried a pathogenic PSEN1 mutation (p.His163Arg) and had the earliest onset age (42 years) of all ABCA7 PTC carriers. Two affected relatives of EOD-P6.1 also carried PSEN1 p.His163Arg, but not ABCA7 c.302+1G>C, and had a slightly older onset age of 46 years.

A positive familial history for dementia was reported in 61.5% (16/26) of patient carriers. Of note, for patient EOD-P21.1 carrying p.Leu1403fs, DNA was available of two affected relatives with onset ages of 68 and 70 years, who both also carried the mutation. In addition, EOD-P7 (p.Met370fs) and EOD-P20 (p.Leu1403fs) had a negative familial history for dementia, but both had a first-degree relative with Parkinson’s disease. Information on clinical presentation was available for 23 patients, of whom 82.6% (19/23) had a predominant amnestic presentation. In two patients, the onset of memory dysfunction was accompanied by language dysfunction. EOD-P6.1 (carrying PSEN1 p.His163Arg) had prominent behavioral symptoms (aggressiveness). Only one patient presented with a clear nonamnestic phenotype (logopenic progressive aphasia). Neuropathology was available for EOD-P1, confirming the clinical AD diagnosis (Fig. 2, Neuropathological description S1).

We examined ABCA7 expression with MinION sequencing for three frameshift mutations (p.Met370fs, p.Glu709fs, and p.Leu1403fs), one nonsense (p.Trp1336*), one splice donor (c.3577 + 1G > C), and one splice acceptor mutation (c.67-1G>A). We observed varying degrees of PTC bearing transcripts in all cDNA libraries (Fig. 3; Table 1), indicative of incomplete NMD. The most N-terminal mutation (c.67−1G>A) had the highest NMD efficiency with only 5% of sequencing reads showing out-of-frame exon 3 skipping. All other mutations presented higher PTC abundancy (21–41%), approaching expression equal to the WT allele (50%). On top of apparent NMD escape, we observed alternative splicing events, absent from public databases (e.g., Ensembl, GENCODE, and UCSC genes), in mutated regions of interest. Some have the ability to restore reading frameshifts caused by PTC mutations (Fig. 3; Table 1). For three mutations (p.Met370fs, p.Trp1336*, and p.Leu1403fs), in-frame skipping of the respective PTC bearing exon [exon 11 (168 bp), exon 30 (255 bp), and exon 31 (33 bp)] was observed in patient carriers (Figures S3, S4, and S5). Overall, potential PTC rescue transcripts had a modest abundance (2–8%); however, in brain cDNA of one individual, we observed skipping of exon 30 in 30% of all sequencing reads, in this case unrelated to a PTC (Figure S5). These in-frame exon skipping transcripts were validated with RNAseq, confirming their presence in individuals not carrying an ABCA7 mutation (Figure S6 and S7). Furthermore, for p.Glu709fs (exon 16), we observed usage of a cryptic splice donor site (10%) in the same exon (Figure S8), which can negate the frameshift effect of p.Glu709fs. Validation of this splicing isoform with RNAseq is presented in Figure S9, along with confirmation of its presence in a physiological context. Western blotting was performed on brain of c.67−1G>A and p.Leu1403fs (Method S1), confirming a reduction of expression by PTC mutations (Figure S11), as well as an overall variability in expression. One splice mutation (c.5570+5G>C)—3 bp downstream of the canonical splice donor site—was previously reported to cause out-of-frame intron retention [47]. Here, this variant showed no association: we observed the minor C-allele in six EOAD patients (MAF = 0.38%), of which one was homozygous, and seven control individuals (MAF = 0.38%) [OR_MH = 1.28 (0.44–3.74), p value = 0.86; Table S5]. RNAseq and MinION sequencing confirmed out-of-frame partial intron 41 retaining capability of c.5570+5G>C, but additional splicing events in the locus were observed as well (i.e., exon 41 skipping, and varying intron 41 retention lengths) (Figure S10).

We identified 19 predicted deleterious missense mutations (CADD score >20). There was no enrichment of predicted deleterious missense variants in patients (1.7%; n = 16) compared to controls (0.92%; n = 9; SKAT-O p value = 0.67) (Fig. 1; Table S6). Of note, p.Ala676Thr and p.Ser1723Leu segregated together in two controls and one patient; for which the patient, in addition, carried a third deleterious in-frame deletion (c.4922_4924delTCT). Clinical characteristics of missense mutation carriers did not differ substantially from PTC mutation carriers [mean onset age 57.0 ± 5.5 years (range = 46–55); positive familial history in 60.0% (6/10)].

Twenty-two common coding variants (MAF > 1%), were assessed for association with EOAD. No variants passed study-wide multiple testing correction (p = 0.0033), but nominal significance was observed for three SNPs (Table S7). The strongest effect [OR = 0.60 (95% CI = 0.42–0.87), p = 0.006] was observed for p.Gly215Ser (rs72973581, MAF = 4.1%), a missense SNP previously suggested to be protective [44]. For silent variant p.Asn1829Asn (rs78320196, MAF = 4.7%), we observed an OR of 0.65 (95% CI = 0.47–0.90; p value = 0.009). The potential protective effect of p.Asn1829Asn seems independent of p.Gly215Ser, given low pairwise LD (R ² = 0.001 and D′ = 0.596). The third variant (p.Glu188Gly, rs3764645, and MAF = 43.4%) also showed a suggestive protective effect (OR = 0.86 95% CI = 0.75–0.99), p value = 0.030), which was previously reported in a Belgian cohort of AD [11]. We further observed that p.Glu188Gly and p.Gly215Ser were in strong LD (D′ = 0.961).