Cost-effective molecular inversion probe-based ABCA4 sequencing reveals deep-intronic variants in Stargardt disease

2.1 Study cohort

We analyzed the ABCA4 gene for sequence variants employing STGD1 cases from Lille, France (n = 223 cases) and Regensburg, Germany (n = 188 cases). Samples were collected according to the tenets of the Declaration of Helsinki and written informed consent was obtained from all participants. STGD1 samples from the French cohort (originating from a cohort of 1,133 probands) were prescreened using the following techniques: high resolution melting (HRM) mutation detection analysis, denaturing high-performance liquid chromatography (dHPLC), or NGS of four IRD-associated genes, including ABCA4. The STGD1 cases from Germany (originating from a cohort of 335 probands) were previously sequenced using a custom-designed GeneChip CustomSeq Resequencing Array (RetChip; Affymetrix, Santa Clara, CA; Schulz et al., 2017) or NGS on an ION Torrent semiconductor personal sequencing machine (Thermo Fisher Scientific, Darmstadt, Germany) upon multiplex-polymerase chain reaction (PCR) amplification of the fragments with the STARGARDT MASTR Kit (Multiplicom, Niel, Belgium).

2.2 Design of smMIPs

The design of smMIPs was performed in two series. First, smMIPs were designed to sequence the 50 protein-coding exons of ABCA4 (NM_000350.2), including at least 20 base pair (bp) upstream and 20 bp downstream sequences encompassing the splice site consensus sequences. Also six previously reported deep-intronic variants including (c.4253+43G>A, c.5196+1137G>A, c.5196+1216C>A, c.5196+1056A>G, c.4539+2001G>A, c.4539+2028C>T) (Braun et al., 2013; Zernant et al., 2018) were captured covering small parts of introns 30 (182 bp) and 36 (226 bp). In the first sequence analysis of 22 test DNA samples, smMIPs targeting exons 7, 10, 13, 31, 33, 36, 37, 38 and 46 resulted in less than 10 reads whereas the average coverage of all smMIPs was >40 reads. To cover these gaps, 23 additional smMIPs were designed and added in the second sequence analysis to improve the coverage of these regions. All regions were covered well except for the last three nucleotides of exon 10 and the first 20 nucleotides of intron 10. Thereafter, the final rebalanced smMIP pool consisted of 309 smMIPs including 299 exonic and 10 intronic smMIPs.

In addition, eight recently discovered causal deep-intronic variants (c.769–784C>T, c.859–506G>C, c.859–540C>G, c.1937+435C>G, c.4539+1100A>G, c.4539+1106C>T, c.4539+2064C>T and c.5197–557G>T) (Bauwens et al., 2019; Sangermano et al., 2019) as well as 300 bp upstream and 300 bp downstream sequences were captured using 174 smMIPs (Table S1 and Figure S1). Larger segments were sequenced as they may carry novel causal deep-intronic variants that may affect the recognition of the same pseudoexons by the splicing machinery. Thereafter both pools were combined and consisted of 483 smMIPs.

To design smMIPs, the genomic positions were obtained from the UCSC genome browser (hg19; https://genome.ucsc.edu). smMIPs were designed using MIPgen pipeline (Boyle, O'Roak, Martin, Kumar, & Shendure, 2014). For each target region overlapping smMIPs were designed in a double-tiling fashion, that is, one on the plus strand and one on the minus strand. Details of the smMIPs and their distribution across ABCA4 are provided in Table S1 and Figure S1.

Each smMIP is 78 nucleotides long and targets 110 nt. The two annealing arms (denoted extension and ligation arm) together are 40 nt and are connected using a common linker sequence of 30 nt. In addition, an 8-nt random tag serves as “single molecule” identifier (Eijkelenboom et al., 2016). The smMIPs were synthesized by Integrated DNA Technologies (IDT, Leuven, Belgium). All smMIPs were phosphorylated after pooling, as described previously (Neveling et al., 2017).

2.3 Automated smMIPs library preparation and sequencing

ABCA4 sequencing libraries were prepared using a fully automated sequencing workflow based on smMIPs enrichment in combination with multiplex-PCR using barcoded PCR primers in two series, followed by sequencing with Illumina NextSeq 500 (by using Mid output kits, maximum of 130 million reads) as described previously (Neveling et al., 2017).

2.4 Variant calling and annotation

Data were analyzed using an in-house bioinformatics pipeline starting from the raw sequencing reads (FASTQ). In the first step unique molecular identifiers (UMIs) were trimmed from the sequencing reads and stored within the read identifier for later use. The paired-end sequencing reads were then directly mapped to the reference genome (human genome build GRCh37/hg19) using BWA mem (v0.7.12). The extension and ligation arms, as well as overlap between the read-pairs, were trimmed after the alignment of each read based on the smMIPs design file, to improve the initial alignment. Duplicated reads, located on the same position and having the same UMI, were removed while the remaining unique reads were written to a single BAM file per patient based on the barcode. Variants were called using both the UnifiedGenotyper as well as the HaplotypeCaller from GATK (v3.4–46) for each individual sample followed by merging of the two variant call sets using the GATK Combine Variants and joint genotyping functions. The two resulting VCF files were used as input for an in-house annotation pipeline where variants were annotated with effect predictions, gene information as well as frequency information from various population databases (Lelieveld et al., 2016).

2.5 Variant prioritization

First, variants were prioritized by using the criteria “Quality by Depth” >500, which represents the overall coverage of the region, an Allele frequency (AF) < 0.005 in the dbSNP database and in an in-house whole exome data set of 21,559 persons, and an AF < 0.01 in control population datasets, such as the Genome Aggregation Database (gnomAD; http://gnomad.broadinstitute.org/), and passing standard quality filters including gene components such as exons and canonical splice sites were prioritized. Previously reported pathogenic variants with high AFs such as c.2588G>C and c.5603A>T (AFs in non-Finnish European [nFE] in gnomAD 0.00784 and 0.06647, respectively) (Cornelis et al., 2017; F. P. Cremers et al., 1998; Schulz et al., 2017; Zernant et al., 2014, 2017), were selected separately. Known deep-intronic variants were selected based on prior knowledge from literature (Albert et al., 2018; Bauwens et al., 2015, 2019; Braun et al., 2013; Sangermano et al., 2019; Schulz et al., 2017; Zernant et al., 2014).

2.6 Coding variants classification

For novel coding variants, along with the AF filters, in silico pathogenicity assessment was performed including Sorting Intolerant From Tolerant (SIFT; http://sift.bii.a-star.edu.sg/) (Kumar, Henikoff, & Ng, 2009), Polymorphism Phenotyping v2 (PolyPhen-2) (Adzhubei et al., 2010) and MutationTaster (Schwarz, Cooper, Schuelke, & Seelow, 2014) by using Alamut Visual software version 2.7 (Interactive Biosoftware, Rouen, France; www.interactive-biosoftware.com). Variants were classified according to the guidelines of the American College of Medical Genetics (ACMG) (Richards et al., 2015).

2.7 Noncoding variants classification

For the selection of noncanonical splice site (NCSS) and deep-intronic variants, in silico predictions were performed by using five algorithms (SpliceSiteFinder-like, MaxEntScan, NNSPLICE, GeneSplicer, Human Splicing Finder) via Alamut Visual software version 2.7 (Biosoftware, 2014; Interactive Biosoftware, Rouen, France; www.interactive-biosoftware.com) (Cartegni, Wang, Zhu, Zhang, & Krainer, 2003; Desmet et al., 2009; Pertea, Lin, & Salzberg, 2001; Reese, Eeckman, Kulp, & Haussler, 1997; Yeo & Burge, 2004) by comparing splicing scores for wild-type (WT) and variant nucleotides (Table S2).

2.8 Segregation analysis

Segregation analysis was performed for 45 cases from Lille. PCR products were sequenced in sense and antisense directions using the BigDye^® Terminator v3.1 Cycle Sequencing Kit on a 3730 DNA analyzer (Applied Biosystems, Carlsbad, CA).

2.9 Midigene-based splice assay and reverse transcription (RT)-PCR assessment

The effect of seven NCSS and three deep-intronic variants was assessed by midigene-based splicing assays employing wild-type (WT) BA constructs described elsewhere (Sangermano et al., 2018) and a newly designed BA31 construct (Table S3). Details of mutagenesis primers and sequencing primers are given in Table S4. WT and mutant constructs were transfected in Human Embryonic Kidney 293T (HEK293T) cells and the extracted total RNA was subjected to reverse transcription (RT)-PCR, as previously described (Sangermano et al., 2018).

3.1 smMIPs performance

Sequence analysis of 16 STGD1 cases carrying known ABCA4 pathogenic variants and four control samples using 483 smMIPs and a NextSeq 500 mid-output kit was performed to assess the performance of each smMIP and coverage per patient. The average read for these 20 DNA samples ranged from 10 to 37,551 per smMIP, with an overall average coverage of 2,687×. A coverage plot of the smMIPs is provided in Figure S2. All previously known 34 variants present in the 32 alleles were identified robustly. All the targeted deep-intronic variants were found in ≥31% of the total reads ranging from 1,521 reads (c.859–506G>C) to 17,538 reads (c.5196+1137G>A) (Table S5). Thereafter, 411 STGD1 proband samples were sequenced in two subsequent runs, with 226 samples (average coverage 511×) and 185 samples (average coverage 655×).

3.2 Identification of ABCA4 variants

ABCA4 sequencing was performed using 483 smMIPs in 411 STGD1 persons, identifying 173 unique sequence variants. All variants and the respective cases were uploaded into the ABCA4 LOVD at www.lovd.nl/ABCA4. Of these, 34 (20%) were new variants, including coding, NCSS, and deep-intronic variants, as listed in Table 1. The 173 unique variants account for 534 alleles in STGD1 patients, details of which are given in Table S6. Of these 173 variants, 75% (n = 131) are coding variants, comprised of missense (n = 96), frameshift (n = 23), and nonsense mutations (n = 12). The other 25% (n = 42) of variants are located in noncoding regions including canonical splice site, NCSS, and deep-intronic variants.

The most common previously reported alleles were c.4253+43G>A (n = 28), c.[5461-10T>C;5603A>T] (n = 22), c.5882G>A (n = 22), c.[1622T>C;3113C>T] (n = 14), c.[2588G>C; 5603A>T] (n = 13) and c.5196+1137G>A (n = 13) (Table S7). The most frequent single variant allele was c.5603A>T (n = 144), but its penetrance, when present in trans with a severe ABCA4 variant, is incomplete in STGD1 families and in the Dutch population (see below).

3.3 Disease-associated ABCA4 alleles in 411 STGD1 persons

Two or more pathogenic or likely pathogenic variants in ABCA4 were found in 155 of 411 STGD1 cases, 97 (24%) of whom were considered solved as they carried two (likely) pathogenic variants. Another 58 (14%) cases were considered possibly solved as they carry c.5603A>T (p.Asn1868Ile), a frequent mild but low-penetrant variant (F. P. M. Cremers, Cornelis, Runhart, & Astuti, 2018; Runhart et al., 2018; Zernant et al., 2017) in trans with a moderate or severe coding or deep-intronic variant. Details of the identified alleles in each person are given in Table S7. Sixty-seven individuals carry complex alleles (Tables S7 and S8). In 133 (32%) STGD1 cases only one ABCA4 allele was identified whereas in 123 (30%) of the persons, no variants were identified in ABCA4.

3.4 Novel rare variants identification and classification

Thirty-four novel variants were identified in 534 identified alleles (Table 1). Among these, 13 variants were missense mutations, identified in 14 alleles, of which c.4128G>C (p.Gln1376His) and c.4633A>G (p.Ser1545Gly), were also situated in consensus splice site sequences of exons 28 and 31, respectively. Thirteen frameshift mutations were identified in 19 alleles (among which c.4706del, p.Val1569Alafs*12 in seven alleles), and three canonical splice site variants (c.442+1G>A, c.1760+1G>A, and c.5313–2del). One stop mutation, p.Gly1717* was identified in two alleles. In addition, a synonymous variant (c.4539G>A) was located in the splice donor site (SDS) sequence of exon 30 in two alleles, and variant c.3191–19G>A was situated close to exon 22 but not in consensus splice site sequences. Moreover, three novel deep-intronic variants (c.769–605T>C, c.4539+859C>T and c.4539+2065C>G) were identified in six alleles.

3.5 Splice defects due to selected ABCA4 noncanonical splice site variants

In vitro splice assays were performed in HEK293T cells to assess the impact of NCSS and novel near-exon at the transcript level by using eight WT constructs shown in Figure 1. Four of eight tested NCSS or near-exon splice site variants show splice defects: c.3608G>A, c.4128G>C, c.4539G>A and c.4849G>A. RT-PCR and Sanger sequencing results are provided in Figure 2. Variant c.3608G>A, reported previously by Cornelis et al. (2017) was predicted to reduce the strength of the SAS of exon 25 and the creation of a cryptic SAS 2 nt downstream of the canonical SAS (Table S2). It was tested in WT construct BA17 and RT-PCR was performed by using ABCA4 exonic primers located in exons 23 and 26. A fragment of 357 nt corresponding to the WT band and a 151-nt fragment in which exon 25 is skipped were observed (Figure 2a–c).

The novel missense variants residing in splice site consensus sequences, that is, c.4128G>C and c.4633A>G, were tested in WT constructs BA19 and BA21, respectively. Variant c.4128G>C, predicted to weaken the SDS of exon 27, resulted in a single band corresponding to a 12-nt elongation of exon 27 due to the presence of a cryptic SDS (Figure 2d–f). Variant c.4633A>G, predicted to significantly weaken the SDS of exon 31, showed no splice defect in HEK293T cells (Figure S3).

The novel synonymous variant c.4539G>A, predicted to impair the SDS of exon 30 (Table S2), was tested in midigene BA20. RT-PCR was performed by using primers residing in exons 28 and 31, and variant c.4539G>A showed three bands compared to the WT which upon sequencing revealed a band of 441 nt corresponding to WT, a band of 368 nt corresponding to a deletion of 73 nt of the 3′-end of exon 30 and the third band of 155 nt corresponding to the deletion of exons 29 and 30 (Figures 2g–i).

Variant c.4849G>A, also described previously (Downs et al., 2007) and predicted to reduce the SAS of exon 35, resulted in three fragments when tested in WT BA23. Sanger sequencing revealed a fragment of 578 nt corresponding to the correct transcript, a band of 408 nt corresponding to exon 35 skipping and the third band of 317 nt corresponding to the skipping of exon 35 and a 91-bp reduction of exon 36 due to the recognition of an internal SAS at position c.5110 (Figure 2k–o).

The other novel variant c.3191–19G>A, tested in BA16, did not show a splice defect (Figure S3). In addition to this novel splice variant, the previously reported variant c.6148G>C (Schulz et al., 2017) did not show a splice defect using BA31 (Figure S3).

3.6 Deep-intronic variants identification and in vitro assessment

To investigate the occurrence of deep-intronic variants among 411 STGD1 persons, 12 selected intronic regions carrying 14 previously identified variants were targeted by smMIPs. Among 534 alleles, 67 alleles carrying different deep-intronic variants were identified (Table S7). Ten already reported deep-intronic variants were identified in 61 (15%) alleles, that is, c.769–784C>T (four alleles), c.4253+43G>A (28 alleles), c.4539+1106C>T (one allele), c.4539+2001G>A (three alleles), c.4539+2028C>T (one allele), c.4539+2064C>T (seven alleles), c.5196+1056A>G (one allele), c.5196+1136C>A (one allele), c.5196+1137G>A (13 alleles) and c.5196+1159G>A (two alleles).

In addition, three novel deep-intronic variants, that is, c.769–605T>C (three alleles), c.4539+859C>T (two alleles) and c.4539+2065C>G (one allele), were found in the regions containing known causal variants. Variant c.769–605C>T, showing an allele frequency of 0.00613 in non-Finnish Europeans, was located 12 nt downstream of a pseudoexon (PE) generated by c.769–784C>T. It was tested in WT construct BA6 but did not show a splice defect (Figure S3). Similarly, variant c.4539+859C>T, showing an allele frequency of 0.01123 in non-Finnish Europeans in gnomAD, was located upstream of variant c.4539+1106C>T and increased the strength of the existing cryptic SAS in the intronic region. It was tested in WT construct BA20 but did not show any splice defect (Figure S3).

Conversely, variant c.4539+2065C>G (not found in 7713 non-Finnish Europeans), located next to c.4539+2064C>T, creates a new SD and alters the strength of ESEs in the region predicted by Alamut visual. It was also tested in WT construct BA20 and RT-PCR was performed by using ABCA4 exonic primers located in exons 28 and 31. Gel analysis and Sanger sequencing revealed a band of 441 nt corresponding to the WT and 611 nt corresponding to a 170-nt PE insertion located between c.4539+1891 and c.4539+2060, that was observed in most of the cDNA product and led to a frameshift p.[Arg1514Lysfs*35,=] (Figure 2g, h, j).

Therefore, a total of 11 causal deep-intronic variants were found in 62 STGD1 alleles. Most carried deep-intronic variants on one allele and a severe or a moderately severe ABCA4 variant on the other allele. Two patients carried causal deep-intronic variants on both alleles, that is, individual 067332 (c.4539+2064C>T, p.[=,Arg1514Leufs*36] and c.5196+1137G>A, p.[=,Met1733Glufs*78]) and individual 067241 (c.4539+2064C>T and c.4253+43G>A, p.[=,Ile1377Hisfs*3]). Two other individuals, 066666 and 066688, carried c.4539+859C>T, p.(?) and c.4253+43G>A in a compound heterozygous manner, but the former variant was not shown to result in a splice defect.