A rapid NGS strategy for comprehensive molecular diagnosis of Birt-Hogg-Dubé syndrome in patients with primary spontaneous pneumothorax

Forty unrelated patients with PSP who had given written informed consent participated in the genetic screen and were divided into two groups.

The study was approved by the Ethical Committees of Taizhou Hospital of Zhejiang Province (TZYY-2011106) and the Ethical Committees of Nanjing Chest Hospital (NCH-2011062). Informed consent was obtained from all individual participants included in the study.

Genomic DNA was extracted from peripheral blood samples of collected family members and controls using a Qiagen DNA Mini blood Kit (Qiagen, Hilden, Germany) according to manufacturer instruction. DNA purity and concentration were assessed by the NanoDrop2000 spectrophotometer (Thermo Fisher Scientific, Florida, USA), A260/A280 ratios of 1.8 to 2.0 were accepted. DNA fragmentation was assessed by agarose gel electrophoresis.

A custom Haloplex panel was designed using Agilent’s online SureDesign tool (https://​earray.​chem.​agilent.​com/​suredesign/​index.​htm). In all, 78 target regions including the whole gene sequences of FLCN, TSC1, TSC2, CBS, COL3A1 and all coding exons, intron-exon boundaries including 50 intronic nucleotides and 5′ UTR (Untranslated Regions), 3′ UTR of FBN1, SERPINA1 were captured for enrichment (Table 1). According to previous reports, intragenic large deletions have been identified in FLCN, TSC1, TSC2, CBS and COL3A1 genes [8, 31‐33], most breakpoints of the deletions locate in intronic sequences. In order to assess the performance of breakpoint analysis using NGS system, we captured both exonic and intronic sequences. However, large deletions in SERPINA1 gene were rarely reported. Blyth et al. (2008) suggested that Marfan patients with large deletions in FBN1 gene have a more severe phenotype, and barely manifest as isolated pneumothorax [34]. Therefore, we only targeted coding exons, intron-exon boundaries, 5′- and 3′- UTRs of SERPINA1 and FBN1genes. The predicted target coverage was 98.72 %. The 214 kb target regions were captured using the Agilent HaloPlex Target Enrichment System Kits for Illumina Sequencing (Custom Panel Tier 1, ILM, 48 reactions; Agilent Technologies, Inc. Santa Clare, CA) following Agilent protocols. In brief, 225 ng of genomic DNA was fragmented using eight different restriction enzymes and denatured. Hybridization was performed with a biotinylated custom probe library and sample indexes, and then the target DNA fragments were captured with magnetic streptavidin beads and circularized by ligation. The captured target libraries were amplified by PCR, quality controlled and quantified using the BioAnalyzer 2100 (Agilent Technologies, Inc. Santa Clare, CA). Equimolar amounts of differentially indexed samples were pooled before pair-ended sequencing 300 bp on the Illumina MiSeq platform (Illumina Inc., San Diego, CA, USA) and the mean read depth within the regions of interest was ~300 reads per base.

The authors implicated in the bioinformatics analysis had no information on the patient data. Post-run sequencing quality was assessed by FastQC (Babraham Bioinformatics, Cambridge, UK). Sequence reads were aligned to the hg19 human reference genome (February 2009 assembly), and analyzed with Agilent software Surecall v2.1.1.13 using the default Haloplex parameters. The mutation reports (*.vcf files) were then annotated using ANNOVAR [35]. After removing all SNVs/indels occurring within non-coding regions, the variations were filtered against dbSNP137 NonFlagged, 1000 genomes and ESP 6500 databases. Rare, non-synonymous, exonic variants were subjected to bioinformatics tool PolyPhen [36] and SIFT [37] to evaluate their possible impact on protein structure and function. To perform extensive analysis, we further analyzed the intron sequences to identify potential splice and branch-site mutations using splice prediction tool ESEfinder [38]. If any potentially harmful mutation identified, validation was performed by Sanger sequencing.

CNVs were identified using a depth-based method. Coverage statistics were generated using GATK Depth of Coverage tool version 3.1 [39, 40]. We normalized coverage in each sample by dividing the average coverage of each exon by the total number of on-target mapped reads for that sample \( \left(\mathrm{normalized},\mathrm{coverage},=,\frac{\mathrm{the}\ \mathrm{average}\ \mathrm{coverage}\ \mathrm{of}\ \mathrm{each}\ \mathrm{exon}}{\mathrm{the}\kern0.22em \mathrm{total}\;\mathrm{number}\;\mathrm{of}\;\mathrm{on}-\mathrm{target}\;\mathrm{mapped}\;\mathrm{reads}\;for\;\mathrm{that}\;\mathrm{sample}}\right) \). To identify copy number variations at individual exons, we compared the normalized coverage of each exon of testing patients with that of controls. Coverage data and bars/plots graph were produced using Prism 6. Exons with a normalized coverage ratio below/above 0.7/1.3 of the mean in controls were classified as heterozygously deleted/duplicated and further analyzed by MLPA [8]. A 20-bp interval was used for precise breakpoint determination.

Sanger sequencing was performed to confirm all detected variants. The primers Prism 3130 were designed using the online software Primer3. The amplification reaction mixture (50 μl) was subjected to denaturation at 95 °C for 2 min followed by 30 cycles at 94 °C for 1 min, annealing temperature 60 °C for 1 min, 72 °C for 1 min and by a final extension at 72 °C for 15 min. Bi-directional sequencing was performed using BigDye Terminator v3.1 Cycle Sequencing Kit, version 3.1 (Applied Biosystems, Foster, CA, USA) and analyzed on an ABI genetic analyzer (Applied Biosystems, Foster City, CA, USA).

The obtained sequencing data was analyzed by FastQC software, which indicated good-quality sequencing metrics, with an average 80.53 % of the sequenced bases above Q30. On average, a mean coverage of 364× was obtained per sample (ranged 210 ×–699×) (Fig. 1a). Different average coverage were obtained for the seven analyzed genes, ranging from 243× (COL3A1) to 592× (SERPINA1) (Fig. 1b). All genes were covered at more than 30× for at least 96 % of the target regions, promising the CNV analysis. The percentage of sequencing on target was 72.35 %.

All 15 different pathogenic sequence changes were correctly identified in the 19 positive control samples (Table 2; patients F01-F19), resulting in a diagnostic rate of 100 % for the detection of pathogenic mutations in our system. These encompassed a variety of mutations, including 2 SNVs (1 stop gain and 1 splice site mutation), 4 insertions/6 deletions ranging from 1 to 23 bases and 3 large intragenic deletions. The false-negative rate was 0 %. All samples were analyzed blind to their mutations.

We observed that the coverage varied significantly between positions from the same sample, however, the samples captured with an identical protocol had very similar distribution of sequencing depth. In this context, we normalized coverage of each exon for on-target mapped bases for that sample and performed the CNV analysis using a depth-based method. This analysis allowed us to identify 4 patients with heterozygous large intragenic deletions of FLCN gene in our study cohort: 1 deletion in FLCN comprised exons 1–3 (average normalized depth ratio = 0.356). One deletion in FLCN comprised exon 14 (normalized depth ratio = 0.601) and 2 deletions in FLCN comprised exons 9–14 (average normalized depth ratio = 0.533 and 0.539, respectively) (Fig. 2a-d).

We subsequently managed to identify exact breakpoints of the deletions by calculating the normalized depth of a set of 20 bp-interval contigs covered the entire FLCN gene (Fig. 2e-h, Table 3). For patient F16, by using MLPA, of which probes are designed in the exons of FLCN gene, we detected a deletion of exons 1–3 [8]. The estimated deletion size is between 5346–18666 bp in length, while with the NGS strategy, we were able to restrict it to about 7650 bp, closed to its actual deletion size (Fig. 2i). Likewise, in patient F17 and F18/F19, we reduced the estimated size of deletion from 1645–8671 bp and 5555–13782 bp down to about 1740 bp and 6117–12017 bp, respectively (Fig. 2i).

The breakpoints for four large intragenic deletions in FLCN had been identified by genomic amplification of the DNA regions across the deletion junctions, followed by Sanger sequencing. Each large deletion is flanked by Alu repeat sequences that mediate the mutation. Our NGS system accurately identified exon deletions and determined breakpoints on the targeted sequence within about 100 bp, largely narrowed the boundaries of the deletions, which made it much easier to design primers for genomic amplification of the deletion junction.

Our study shown there were an average of 194 variants per sample within the 214 kb targeted region in our blinded PSP cohort of PSP. After removing all SNVs/indels occurring within non-coding regions, the variations were filtered against dbSNP137 NonFlagged, 1000 genomes and ESP 6500 databases, and then submitted to PolyPhen and SIFT to evaluate possible impact of the variations on protein structure and function. Using this filtering criteria, there was 0–1 novel, predicted harmful variant remained for each sample. Of the 21 patients, we detected clearly pathogenic mutations in 3 cases (2 in FLCN, 1 in FBN1) and likely pathogenic mutations in the other 3 cases (Table 4). All of the variants were confirmed by Sanger sequencing.

Two small frameshift deletions were identified in FLCN gene in patients F29 and F37, both resulting in a truncated protein of FLCN. And one of them, a c.1648_1658 deletion in exon 14 of the FLCN gene had been detected in patient F09. A nonsense mutation were detected in FBN1 gene in patient F32, it was a C to T transition (c.6169 C > T) in exon 51, predicting an early stop codon at position 2057 (p.Arg2057Ter) which causes a premature termination of the protein. This mutation has not been recorded in any database including FBN1 mutation database [41]. In this unscreened cohort of PSP patients, we also identified two missense mutations predicted to be damaging by SIFT and Polyphen2. One of them was a c.2269G > C transversion in exon 19 of the FBN1 gene, which converted the codon GAT for Aspartic at position 757 to CAT, a codon for Histidine. This variant is predicted by ScanProsite to situate in the EGF-like calcium-binding domain of FBN1 [42]. The other was a c.1631G > A transition in exon 15 of the TSC1 gene, which was predicted to substitute a polar uncharged glycine residue for a negatively charged glutamic residue (p.Gly544Glu), presenting a distinct change in amino acid properties. We also detected a G to A point mutation in intron 10 (c.1177–21G > A) of FLCN gene (Fig. 3a). This variant located within the branch-site sequence, by using the ESEfinder, we obtained a matrix score for the c.1177–21G > A substitution (2.2919) of the branch-site sequence, relatively less than a score for the wild type (2.7018) (Fig. 3b), indicating that it may disrupt the branch-site in this gene. However, because of unavailability of the patient’s tissue, we were unable to verify it at transcript level.

In general, NGS method not only successfully identified missense, nonsense, splice site, intronic branch site, small indel, and large deletion changes in positive and blinded samples, but also determined approximate deletion junction within about 100 bp deviation by using a depth-based method.