Introduction

Aortic aneurysm and/or dissection (AAD) is a life-threatening condition. As significant symptoms do not usually appear before the rupture of the AAD, which can be lethal, it is often difficult to prevent death from AAD. Timely cardiovascular surgery may prevent AAD rupture and save the patient’s life. Approximately 20% of patients with thoracic aortic disease have a family history of the disease, which is typically inherited in an autosomal dominant manner with decreased penetrance and variable expressivity (Wang et al. 2010). Therefore, if a causative mutation is detected in a patient, it is worth checking for the mutation in their asymptomatic family members to prevent future aortic events by medical and/or surgical intervention. Several genes are known to be associated with syndromes presenting with hereditary AAD and vascular disruption: FBN1 (Dietz et al. 1991; Lee et al. 1991a), TGFBR2 (Mizuguchi et al. 2004), TGFBR1 (Loeys et al. 2005), MYH11 (Zhu et al. 2006), ACTA2 (Guo et al. 2007), COL3A1 (Superti-Furga et al. 1988), PLOD1 (Hautala et al. 1993), and SLC2A10 (Coucke et al. 2006) (Table 1). Most AAD patients who have been surgically treated are not affected by these syndromes. However, the contribution of these genes to non-syndromic AAD has not been thoroughly investigated. A comprehensive study of these genes by conventional Sanger sequencing is a huge and expensive undertaking. Even high-resolution melting methods and denaturing high performance liquid chromatography require the amplification of at least 210 exons from these eight genes (Table 1). Therefore, it has been unrealistic for most laboratories to analyze these genes in multiple samples.

Table 1 Overview of genes associated with AAD analyzed in this study
Full size table

Resequencing array technology (ResAT) enables the investigation of multiple genes on one chip. This technology has been used for multiple-gene analysis in childhood hearing loss (Kothiyal et al. 2010), breast-ovarian cancer syndrome (Schroeder et al. 2010), dilated cardiomyopathy (Zimmerman et al. 2010), X-linked intellectual disability (Jensen et al. 2011), familial hypercholesterolemia (Chiou et al. 2011), and hypertrophic cardiomyopathy (Fokstuen et al. 2011). Different research groups have shown ResAT to be a highly efficient, relatively accurate, cost-effective, and rapid method. However, several drawbacks have been pointed out, including its insensitivity in detecting nucleotide insertions/deletions (indels) and nucleotide changes in GC-rich regions and repeat sequences.

Next-generation sequencing (NGS) is now regarded as the most powerful technology for detecting mutations (Ng et al. 2010; Tsurusaki et al. 2011). This platform is advantageous in finding almost all types of mutations including small indel mutations. The high throughput and multiplexing of NGS allows multiple genes to be sequenced in many samples in a single run (Farias-Hesson et al. 2010; Gabriel et al. 2009).

In this study, we analyzed the eight AAD-associated genes (FBN1, TGFBR2, TGFBR1, COL3A1, PLOD1, MYH11, SLC2A10, and ACTA2) in 70 patients with non-syndromic AAD by two methods: ResAT (all eight genes on one chip) and multiplex NGS. We describe here a comparison of the results.

Materials and methods

Patients

Seventy Japanese patients, who had surgery for AAD, were recruited from Yokohama City University Hospital and Yokohama City University Medical Center. The patients’ clinical information is summarized in Table 2. Thoracic AAD involves the aorta above the diaphragm and abdominal AAD is located along the portion of the aorta passing through the abdomen. None of the patients in this study had any clinical test results supporting a diagnosis of syndromic AAD. Experimental protocols were approved by the Institutional Review Board of Yokohama City University School of Medicine. Informed consent for genetic analysis was obtained from the patients. DNA was extracted from peripheral blood leukocytes using a QuickGene-610L kit (Fujifilm, Tokyo, Japan).

Table 2 Clinical information of AAD patients
Full size table

Array design

Eight genes (FBN1, TGFBR2, TGFBR1, COL3A1, PLOD1, MYH11, SLC2A10 and ACTA2) (Table 1) associated with AAD were selected for one custom chip (Affymetrix, Santa Clara, CA). All coding exons as well as 29 bp of sequence from each intron (21 bp on the 5′-side and 8 bp on the 3′-side of each exon) were analyzed. Repetitive sequences and intragenic low complexity regions larger than 25 bp were excluded from the chip. A total of 33,116 bp from the eight genes could be sequenced using this chip.

PCR amplification, purification, hybridization, scanning, and data analysis

The targeted regions were amplified as 124 fragments by PCR (ranging from 965 to 2,999 bp) using Blend Taq Plus (TOYOBO, Osaka, Japan) or KOD FX (TOYOBO) and genomic DNA as a template in a 20 μL volume. The PCR conditions were: denaturing at 94°C, 35 cycles of 94°C for 30 s, 62°C for 30 s, and 72°C for 3 min, and a final extension at 72°C for 7 min. The DNA concentration of the amplicons was determined using a Quant-iT PicoGreen dsDNA Assay Kit (Invitrogen, Carlsbad, CA, USA) with a Spectra Fluor F129003 (Tecan, Männedorf, Switzerland). The PCR amplicons were pooled in equimolar quantities (110 fmol). The mixed samples were purified and the volume was reduced using a Microcon YM-100 filter (Millipore, Brussels, Belgium). Fragmentation of the products, labeling with biotin, hybridization, washing, and scanning procedures were carried out based on the CustomSeq resequencing array protocol version 2.1 (Affymetrix). An FS450 fluidics station (Affymetrix) was used for washing and staining and a GCS3000 7G scanner (Affymetrix) was used for scanning. To test the efficiency of mutation detection, PCR products containing 20 known heterozygous mutations (Table 3) from three genes (FBN1, TGFBR2, and TGFBR1), as well as another 104 PCR products amplified from normal control DNA, covering all the other exons, were analyzed using the chip. Affymetrix GCOS and GSEQ software were used to process the raw data and analyze the nucleotide sequences, respectively. The default settings of GSEQ were adopted.

Table 3 Known mutations used as positive controls for testing ResAT
Full size table

Multiplex next-generation sequencing

The PCR amplicons from one patient were mixed and processed using a multiplexing sequencing primers and PhiX control kit (Illumina, San Diego, CA, USA) according to the manufacturer’s instructions but with minor changes. In brief, amplicons were fragmented with Covaris S1 (Covaris, Woburn, MA, USA), and purified using Agencourt AMPure (Beckman Coulter, Brea, CA, USA) instead of gel extraction. DNA quality was checked with an Agilent 2100 bioanalyzer (Agilent Technologies, Santa Clara, CA, USA) and a bar code DNA tag (Illumina) was ligated on. The bar code DNA tags contain unique 6 bp sequences and allow the processing of up to 96 DNA fragments in a single run using an Illumina GAIIx (Illumina). Twelve processed DNA fragments, each with a different tag, were mixed and analyzed with single 76 bp reads in one lane of the flow cell. Six lanes were necessary for the analysis of 70 samples. Image analysis and base calling were performed by sequence control software real-time analysis (Illumina) and offline Basecaller software v1.8.0 (Illumina). The reads were aligned to the human reference genome sequence (UCSC hg19, GRCh37) using the ELAND v2 algorithm in CASAVA software v1.7.0 (Illumina).

Mapping strategy and variant annotation

An average of 2.4 million reads (ranging from 1.7 to 4.0 million reads) for each sample passed quality control (Path Filter) and were mapped to the human reference genome using mapping and assembly with qualities (MAQ) (Li et al. 2008), NextGENe software v2.00 (SoftGenetics, State College, PA, USA), and Burrows-Wheeler Aligner (BWA)/sequence alignment/map tools (SAMtools) (Li and Durbin 2010; Li et al. 2009). Single nucleotide polymorphisms (SNPs) and indels were extracted from the alignment data using an original script created by BITS, Tokyo, Japan along with information on the registered SNPs (dbSNP131). A consensus quality score of 40 or more was used for the SNP analysis in MAQ. SNPs in MAQ-passed reads were annotated using the SeattleSeq website (http://gvs.gs.washington.edu/SeattleSeqAnnotation/). A minimum base quality of 13, a minimum root mean square mapping quality for SNPs of 10, and a minimum read depth of 2 were used in BWA/SAMtools (Li and Durbin 2010; Li et al. 2009). NextGENe (SoftGenetics) was also used to analyze the reads, employing default settings apart from using the no-condensation mode. For base substitutions, we focused on variants detected in common by both MAQ and NextGENe. Small indel variants were classified as positive if found by both BWA and NextGENe.

Validation of novel variants

Novel variants (not in dbSNP131, the 1,000 genomes dataset or our in-house database) identified by ResAT and NGS were validated by Sanger sequencing. Surplus PCR products were treated with ExoSAP IT (GE Healthcare, Piscataway, NJ) and sequenced using a standard protocol using BigDye terminators (Applied Biosystems, Foster City, CA, USA) on an ABI PRISM 3100 genetic analyzer (Applied Biosystems). Furthermore, novel variants were screened in 94 Japanese controls by high-resolution melt curve analysis (LightCycler 480; Roche Diagnostics, Basel, Switzerland) and subsequent Sanger sequencing. Novel variants were evaluated using web-based programs including PolyPhen (http://genetics.bwh.harvard.edu/pph/), PolyPhen2 (http://genetics.bwh.harvard.edu/pph2/), Mutation Taster (http://www.mutationtaster.org/), and ESEfinder (http://rulai.cshl.edu/cgi-bin/tools/ESE3/esefinder.cgi?process=home).

Results

Array performance

Across all 70 samples, the mean nucleotide call rate was 95.7% (range 87.3–97.6%) using the default settings of GSEQ. We observed an improvement of the call rate as the number of samples increased. For example, the call rate by GCOS for the first two samples was 90.1 and 90.6% and was 93.3 and 93.9% when 10 samples were analyzed, and was 94.9 and 95.5% when 33 samples were analyzed. However, between 34 and 70 samples, the call rate did not greatly improve (only by 1%). We had constant difficulty in reading approximately 4% of the sequences per array (i.e., no sequence called), mostly in regions of high GC and CC content.

Detection of known mutations by ResAT

To validate the quality of mutation detection in our resequencing array, we analyzed amplicons containing 16 known nucleotide substitutions, three small deletions (1–2 bp), and one 1 bp insertion, plus all the other normal exons (Sakai et al. 2006; Togashi et al. 2007) (Table 3). Fourteen out of 16 nucleotide substitutions were detected (87.5%) by GSEQ in the automated mode. Two mutations (c.772C > T in FBN1 and c.1142G > C in TGFBR2) were not detected. The former was insensitive, and the latter was indicated as a no-call. Visual inspection in the manual mode enabled easy detection of the TGFBR2 mutation. The mutation detection rate was 93.8% (15/16) using both the automated and manual modes. None of the small indels were detected by our array in either the automated or manual modes.

Variant detection by ResAT

We detected 70 nucleotide substitutions in the automated mode in the 70 patients analyzed (0–3 variants per sample). Fifty-one variants were already registered in dbSNP131 and/or in our in-house database (Supplementary table). The remaining 19 novel variants were validated by Sanger sequencing (Table 4). One variant (c.976–16C > T in PLOD1) was homozygous and the others were heterozygous. No indel mutations were detected.

Table 4 Novel variants detected by ResAT and/or NGS
Full size table

Variant detection by NGS

The target regions were completely covered by NGS reads (100%). The average read depth (coverage of sequence reads) was approximately 600 for each gene (Table 5). The NextGENe software detected a mean of 876 variants in the 70 patients with mutation scores of 10 or more (ranging from 581 to 1209 with SD = 131). MAQ and SeattelSeq detected a mean of 271 variants (ranging from 111 to 384 with SD = 52). Semi-automatic exclusion of variants that were out of the target regions (22 bp or more away from the 5′-end of exons and 9 bp or more away from the 3′-end of exons) or were known variants in dbSNP131 was performed using Excel 2008 for Mac (Microsoft, Redmond, WA, USA), narrowing the data down to 0–6 variants per sample. Twenty novel variants were detected by both MAQ and NextGENe, which were further validated by Sanger sequencing. No indel mutations were detected by MAQ, NextGENe, or BWA/SAMtools.

Table 5 Gene-based read depth in NGS
Full size table

Comparison of ResAT and NGS variants

Eighteen novel variants were detected by ResAT and NGS. One was detected by ResAT only and two by NGS only. The two variants undetected by ResAT were c.1388G > A (p.Arg463Gln) in PLOD1 and c.136A > C (p.Ser46Arg) in TGFBR2. The former was indicated as a no-call, but was detected later in the manual mode. The latter was within a repetitive sequence. One variant (c.1815 + 5G > A in COL3A1) was undetected by NGS due to our set criteria (the variant was detected by MAQ, but not by NextGENe or BWA/SAMtools).

Pathological significance of the variants

We realized that none of the known pathogenic mutations were identified. The pathological impact of the variants was considered if none of the healthy controls showed the same change, if the variants altered evolutionarily conserved amino acids in functional repeats/domains, or if they were predicted to cause abnormal splicing resulting in protein truncation or degradation. Moreover, homozygous and compound heterozygous changes that were found in PLOD1 and SLC2A10 may confer autosomal recessive effects. At least three heterozygous variants were considered as putative pathogenic gene alterations (Table 6):

Table 6 Pathogenic variants found in the patients
Full size table
  1. 1.

    c.1815 + 5G > A in COL3A1 (patient 29). A similar mutation, c.1815 + 5G > T, associated with the skipping of exon 25, was reported in a patient with Ehlers–Danlos syndrome type IV (EDS IV) (Lee et al. 1991b). ESEfinder suggested that the binding position of the splice donor matrix was changed similarly by c.1815 + 5G > A and c.1815 + 5G > T. Thus, C.1815 + 5G > A is highly likely to be pathogenic.

  2. 2.

    c.4963C > T (p.Arg1655Cys) in MYH11 (patient 16). In addition to this mutation, the patient had two novel heterozygous variants: c.4625G > A (p.Arg1542Gln) in MYH11 and c.1220T > G (p.Leu407Arg) in SLC2A10. Mutations in SLC2A10 cause autosomal recessive arterial tortuosity syndrome (MIM #208050) (Coucke et al. 2006), although it is unknown whether the heterozygous variant we identified would be related to this, assuming a second-hit model of recessive disease. Both p.Arg1542Gln and p.Arg1655Cys in MYH11 were similarly predicted to be pathogenic by three programs (PolyPhen, PolyPhen2, and Mutation Taster). These residues are located in the coiled-coil region, and both are evolutionarily conserved amino acids (Fig. 1). Paircoil2 (http://groups.csail.mit.edu/cb/paircoil2/) was used to predict the effect of variants on the parallel coiled coil fold using pairwise residue probabilities (McDonnell et al. 2006). Paircoil2 indicated that p.Arg1655Cys altered the p score from 0.00096 (wild type) to 0.00579 (mutation), while p.Arg1542Gln did not alter the p score, 0.00016 (mutation) and 0.00018 (wild type) (Fig. 1). Thus, p.Arg1655Cys was more likely than p.Arg1542Gln to be pathogenic.

    Fig. 1
    figure 1

    Double mutations in MYH11. a Schematic representation of the MYH11 protein. Three functional domains are indicated: the myosin N-terminal SH3-like domain, the myosin motor domain for type II myosin, and the myosin tail. Both the mutations are located in the myosin tail. b, c Paircoil2 analysis showing a significantly decreased probability of coiled-coil formation for p.Arg1655Cys relative to the wild-type sequence, but no change for p.Arg1542Gln

    Full size image
  3. 3.

    c.482T > C (p.Val161Ala) in ACTA2 (patient 27). The patient was found retrospectively to suffer from familial thoracic AAD. The patient has an affected brother, but his DNA was unavailable. Valine at amino acid 161 is evolutionarily conserved and located within the actin domain. However, as we could not analyze the DNA of the affected brother, it may be more appropriate to call this variant ‘of unknown significance’.

Discussion

Exon-by-exon Sanger sequencing is the gold standard for genetic analysis, but multiple-gene analysis in many patients is a huge task in terms of time and cost. In this study, we applied two emerging technologies providing rapid and efficient analysis of eight genes in 70 AAD patients. We also compared the results of the two technologies.

The overall mean call rate of our custom array by GSEQ software was 95.7%, which is comparable with previous studies (Bruce et al. 2010; Chiou et al. 2011; Jensen et al. 2011; Schroeder et al. 2010). The call rates became higher as the number of patients increased. Approximately 33 samples were necessary to attain the maximum call-rate in GSEQ. A similar observation was described previously (Fokstuen et al. 2011). No-call regions are one of the problems of ResAT. Other groups have previously suggested that most of the no-call regions are GC- and CC-rich (Bruce et al. 2010; Chiou et al. 2011; Fokstuen et al. 2011). In our custom array, approximately 4% of the target sequences were difficult to obtain (no-calls) in most of the samples.

The mean detection rate of known variants using our custom array and GSEQ with the default settings (automated analysis) was 87.5%. This rate increased to 93.8% after manual inspection. For our ResAT data, the detection rate of nucleotide substitutions in the automated mode was higher, and that in the manual mode was slightly lower, compared with detection rates in previous studies (82.1 vs. 81%, respectively, in automated mode, and 97.4 vs. 100%, respectively, in manual mode) (Bruce et al. 2010; Chiou et al. 2011). Our ResAT analysis was unable to detect any small indel mutations; this is similar to other studies (Hartmann et al. 2009; Kothiyal et al. 2010). In the human gene mutation database (HGMD; http://www.hgmd.cf.ac.uk/ac/index.php), insertions/deletions account for a substantial proportion of the total registered mutations in our genes of interest: FBN1 23.6%, TGFBR2 6.4%, TGFBR1 10%, COL3A1 12.8%, PLOD1 46.2%, MYH11 20%, SLC2A10 21.1%, and ACTA 20%. Thus, the incapability of ResAT to detect indel mutations is one of its most significant drawbacks.

Our NGS analysis missed one of 21 variants (c.1815 + 5G > A in COL3A1). Our protocol focused on variants identified by two different informatics methods, to increase the true-positive rate. For example, MAQ (single-end reads) can detect nucleotide substitutions well, but is not good at detecting small indels (Li et al. 2008). BWA is more sensitive at detecting small indels because it can align gapped sequence (Krawitz et al. 2010). NextGENe is based on the Burrows-Wheeler transform algorithm, which is good at detecting small indels. NGS needs more efficient informatics methods to extract all the nucleotide changes correctly with lower error rates.

In this study, concomitant variants in two genes were detected in four patients (Table 4): c.4625G > A and c.4963C > T in MYH11, and c.1220T > G in SLC2A10 (patient 16); c.136A > C in TGFBR2 and c.130–18T > C in ACTA2 (patient 24); c.84T > C in COL3A1 and c.692C > T in TGFBR2 (patient 28); c.119C > T in COL3A1 and c.330C > T in SLC2A10 (patient 89) (Table 4). It may be quite difficult to detect variants in two or more genes by conventional methods. ResAT and NGS permitted us to find multiple variants in multiple genes easily and rapidly. Double or triple mutations in unusual clinical cases will also be found using such technologies.

Three different putative pathological mutations in a heterozygous state in three of 70 patients were found in this study (4.3%). Interestingly, all the three patients suffered from thoracic AAD. Considering only those patients with thoracic AAD (n = 35), the rate increased to 8.6%. Thus, non-syndromic AAD (especially thoracic AAD) can be explained to some extent by aberrations of genes related to Mendelian disorders, although our sample size was small. Interestingly, among these three patients, only patient 29 showed hyperlipidemia and the other two (patients 16 and 27) did not, which supports the genetic origin of thoracic AAD.

In this study, we compared ResAT and NGS. Considering the drawbacks of ResAT, including its inability to detect small indels and its no-call regions, we believe that NGS is the better technology for comprehensive analysis of multiple genes, especially with improved informatics methods, as it can detect all types of mutations with no bias. Another advantage of NGS is its flexibility. Resequencing array technology requires a custom-made sequencing array. It is not easy or practical to update arrays frequently. However, NGS is currently quite expensive for most laboratories. Next-generation sequencing combined with the pooled genomic DNA method with indexing may improve its cost-effectiveness (Calvo et al. 2010; Druley et al. 2009).

In conclusion, we found that 4.3% of non-syndromic AAD patients (8.5% of thoracic AAD patients) have abnormalities in genes that cause Mendelian disorders. ResAT and NGS enabled multiple genes to be analyzed efficiently. In addition to the 70 AAD patients, a patient with familial Marfan syndrome and a patient with Loeys–Dietz syndrome were initially included before their diagnosis was known. We detected c.6793T > G (p.Cys2265Gly) in FBN1 in the Marfan syndrome patient [by ResAT (NGS was not done)] and c.797A > G (p.Asp266Gly) in TGFBR1 in the Loeys–Dietz patient (by ResAT and NGS). We excluded these two patients from this study because they are syndromic AAD patients, but the efficient detection of their mutations highlights the validity of our approach. Finally, high throughput technologies have the potential to routinely identify novel variants of known or unknown significance in clinical settings. Therefore, more sophisticated methods to evaluate gene variants as well as databases containing normal (rare) variants are needed.