Genetic characterization of dilated cardiomyopathy patients undergoing heart transplantation in the Chinese population by whole-exome sequencing

A total of 208 unrelated DCM patients who underwent heart transplantation (HTx) at Fuwai Hospital (Fuwai DCM HTx cohort) from June 2004 to June 2017 were selected. All participants were diagnosed according to the following clinical criteria: (1) left ventricular end-diastolic dimension (LVEDD) > 117% of the predicted value corrected for body surface area and age; (2) left ventricular ejection fraction (LVEF) < 45% in the absence of abnormal loading conditions (hypertension, primary valve disease) or coronary artery disease sufficient to cause global systolic impairment; and (3) hypertension and primary valve disease were excluded by medical history or by cardiac magnetic resonance imaging and echocardiography, and patients with coronary artery disease with stenosis > 50% of at least one main vessel were excluded by coronary angiography relying on experienced clinicians. Familial DCM was defined as at least one additional family member with DCM or in the presence of one relative with sudden cardiac death before 35 years of age [11].

A group of 187 obese children without cardiac defects and 381 patients with transposition of the great arteries was used as the reference, which was reported in an article published previously [12].

WES was performed at Novogene Bioinformatics Technology Co., Ltd. (Beijing, China). Genomic DNA was extracted from transplanted heart tissues using a Magnetic Universal Genomic DNA kit (TIANGEN Biotech, Beijing, China) according to the manufacturer’s protocol. DNA samples with an optical density 260/280 ratio ranging from 1.8 to 2.0 and a content above 1.0 μg were used for library preparation. The Agilent SureSelect Human All Exon V5 kit (Agilent, Santa Clara, CA, USA) was used to capture the exome regions according to the manufacturer’s protocol. First, qualified genomic DNA was randomly fragmented to the 180–280 bp size range by Covaris LE220R-plus (Covaris, USA). Second, DNA fragments were end repaired and phosphorylated, followed by A-tailing and ligation at the 3’ ends with paired-end adapters. PCR was conducted to selectively enrich DNA fragments with ligated adapter molecules on both ends. After PCR, libraries were hybridized in liquid phase with a biotin-labelled probe, and then magnetic beads with streptomycin were used to capture the target exons of genes. Index tags were added through PCR with captured libraries. PCR products were purified using the AMPure XP system (Beckman Coulter, Beverly, USA) and analysed for size distribution by the Agilent 5400 system (AATI) (Agilent, USA). The qualified libraries were sequenced on an Illumina HiSeq X-ten platform (Illumina Inc., San Diego, CA, USA) with the PE150 strategy.

The FASTQ format files, which contain sequence information and corresponding sequencing quality information, were obtained from the sequencing platform. For the raw data, the percentage of bases with a Phred score greater than 20 to the total bases (Q20) was required to be above 90%. The percentage of bases with a Phred score greater than 30 to the total bases (Q30) was required to be above 80%, and the average error rate of all bases was required to be below 0.1% (Additional file 3: Table S1). The average amount, number of reads and depth of raw data were 12.59 G, 83,943,430, and 250x, respectively (Additional file 3: Table S1). Adapter trimming and quality filtering of the raw data were performed using Trimmomatic v0.93 [13]. The average clean read number was 82,437,280 (Additional file 3: Table S1). The clean data were mapped to the human reference genome (UCSC hg19) using BWA MEM (0.7.17-r1188) under default settings [14]. The average mapped reads were 82,271,411, and the average mapped rate was 99.80% (Table S1). Duplicate reads were marked using MarkDuplicates tools in the Picard toolkit (2.21.1, http://​broadinstitute.​github.​io/​picard/​). Base quality recalibration was performed using the BaseRecalibrator tool in the Genome Analysis Toolkit (GATK, v4.1.4.0) [15]. The average sequencing depth on target was 145x (Additional file 3: Table S1).

Variant discovery and quality filtering followed the GATK best practice pipeline for germline SNPs and indels. In brief, variant calling was performed on samples of cohorts using the HaplotypeCaller algorithm in Genomic Variant Call Format mode, and only the variants located within the capture regions of the Agilent SureSelect Human All Exon V5 kit were retained. Next, variants called from the VCF files of all samples, including DCM and reference samples, were subjected to joint genotyping analysis. Finally, variant quality filtering was performed based on variant quality score recalibration. After completing these processes, 1,507,591 variants were obtained in 208 DCM patients and the reference. The number of variants carried by each individual in DCM patients is shown in Additional file 3: Table S2. The VCF files of 208 DCM patients were deposited in the Genome Variation Map (GVM) [16] in the National Genomics Data Center, Beijing Institute of Genomics, Chinese Academy of Sciences and China National Center for Bioinformation [17], under accession number GVM000540.

To examine the population structure of the DCM cases and references, we performed principal component analysis (PCA) based on the genotypes of the called variants in all samples. We visualized the first three components in two-dimensional space. Using PLINK [18], we checked the relatedness of all samples. Specifically, Yoruba in Ibadan, Nigeria, Utah residents with Northern and Western European ancestry, Chinese Dai in Xishuangbanna, China, Han Chinese in Beijing, China, and Southern Han Chinese from the HapMap phase3 dataset were used. As shown in Additional file 1: Fig. S1, we first retained variants with two alleles, a call rate greater than 90%, and a minor allele frequency (MAF) greater than 0.05 in all samples and the HapMap phase3 dataset. Then, the 180,463 overlapping variants were pruned using PLINK, considering window sizes of 100 variants, a step size of 25 and a pairwise r² threshold of 0.05 (–indep-pairwise 100 25 0.05), and a total of 15,286 independent SNPs for PCA were retained. The identity-by-decent was calculated with PLINK within the case and reference groups, and no sample was excluded.

ANNOVAR [19] was used to annotate the variants. Variants predicted to alter the coding sequence of the gene product were classified as truncating or nontruncating variants. Truncating variants include those resulting in frameshifts, premature stop-gain or canonical splice sites, and nontruncating variants include damage missense variants (missense variants with Rare Exome Variant Ensemble Learner > 0.5) [20], stop-loss, and nonframeshift variants. To filter for rare variants, we used allele frequencies of all population exome data documented in a public database: the exome dataset in the Genome Aggregation Database. A variant was considered rare if the allele frequency in this public database was lower than 0.0001 or not documented.

We followed the steps shown in Additional file 2: Fig. S2 to filter the obtained 1,507,591 variants in 8226 protein-coding genes for gene burden analysis. The number of variants before and after filtering is shown in Additional file 3: Table S2. We ran burden tests with four models, including the CMC, Fp, SkatO, and Zeggini models implemented in the toolkit RVTESTS (v20171009): variable-threshold burden tests with MAF cut-offs of ≤ 5%. The top 3 principal components and sex were used as covariates. We used a significance threshold of P ≤ 6.1 × 10^–6, corresponding to a Bonferroni correction for 8226 protein-coding genes. A quantile‒quantile plot was generated by the R package qqman [21].

All variants of the DCM high evidence genes were classified into five categories, including pathogenic (P), likely pathogenic (LP), variant of unknown significance (VUS), likely benign (LB), or benign (B), based on the American College of Medical Genetics and Genomics and the Association for Molecular Pathology (ACMG/AMP) 2015 guidelines.

Continuous variables are expressed as the mean ± SD values, and all categorical variables are depicted using relative frequency distributions. Differences between means were compared using Student’s t test. Comparisons of categorical variables between different groups, such as patients with or without P/LP, were performed using a χ² test where appropriate; otherwise, continuity correction or Fisher’s exact test was used. The statistical analysis was performed in SPSS 20.0 (IBM, USA) or R (https://​www.​r-project.​org/​), and a two-sided P value < 0.05 indicated statistical significance.

The Fuwai DCM HTx cohort comprised 208 unrelated patients with DCM who had undergone heart transplantation. The clinical characteristics are shown in Table 1. One hundred sixty-seven (80.3%) patients were male, and 41 (19.7%) patients were female. All individuals had end-stage heart failure in New York Heart Association (NYHA) classes III–IV. The mean age of onset and heart transplantation was 36.5 ± 11.8 years and 42.4 ± 13.2 years, respectively, and the mean disease course between the diagnosis of DCM and heart transplantation was 71.6 ± 65.6 months. A family history of DCM was identified in 18 (8.7%) individuals by follow-up investigation. The mean value of LVEDD was 75.7 ± 10.8 mm, which is categorized as a severely abnormal left ventricle dimension, and the mean value of LVEF was 23.9 ± 6.4%, indicating severe left ventricle dysfunction.

The workflow of the study design is shown in Fig. 1. We performed WES for 208 patients to investigate the genetic variants in coding regions of the genome that are associated with DCM. WES data from 187 obese children without heart defects and 381 patients with transposition of the great arteries were used as the reference, which was reported in a previously published article [12]. PCA suggested that the two datasets are comparable in terms of population structure (Fig. 2).

After variant discovery and quality filtering, we used the deleteriousness prediction and allele frequency in the public database to screen for rare protein-altering variants (see Methods) and obtained 22,622 SNPs and 5545 indels in the exons and canonical splicing regions of 8226 genes for gene burden analysis. Figure 3 shows the quantile‒quantile plot (Q–Q plot) of four different models for gene-burden tests, and Table 2 shows the genes that reached or approached the exome-wide significance level.

A combined analysis of truncating and nontruncating variants showed that in DCM patients, aggregated protein-altering variants in TTN (case, 31.7%, 66/208 vs. reference, 14.8%, 84/568), filamin C (FLNC) (case, 15.4%, 32/208 vs. reference, 2.1%, 12/568), and LMNA (case, 4.3%, 9/208 vs. reference, 0.5%, 3/568) were significantly enriched (Fig. 3A, Table 2). Protein-truncating variants in TTN and FLNC were significantly enriched in DCM patients (TTN: case, 18.8%, 39/208 vs. reference, 1.9%, 11/568; FLNC: case, 8.7%, 18/208 vs. reference, 0%, 0/568) (Fig. 3B, Table 2). For nontruncating variants, we did not identify any genes that reached exome-wide significance (Fig. 3C, Table 2). In addition, BAG cochaperone 3 (BAG3), a DCM-associated gene, also showed relatively strong associations with truncating variants through the combined analysis, but this has not yet reached exome-wide significance (Fig. 3A, B).

Burden analysis of WES strongly suggested that TTN, FLNC, LMNA, and BAG3 harbour pathogenic rare variants but do not directly inform the interpretation of any single variant in our cohort. To interpret the pathogenicity of a single variant, we conducted an online (http://​wintervar.​wglab.​org/​) pathogenicity assessment of these rare variants according to ACMG/AMP 2015 guidelines. We focused on the rare variants among the high evidence genes (ACTC1, ACTN2, BAG3, DES, DSP, FLNC, JPH2, LMNA, MYH7, NEXN, PLN, RBM20, SCN5A, TNNC1, TNNI3, TNNT2, TPM1, TTN, and VCL) summarized in the previous review [22]. Among these 19 genes, 16 genes with 165 rare variants were identified in our cohort (Additional file 3: Table S3); these variants included 87 missense, 30 stop-gain, 23 frameshift deletion, 12 frameshift insertion, 7 nonframeshift deletion, and 6 splicing variants. Among these 165 rare variants, 81 have been reported in the Clinvar database (https://​www.​ncbi.​nlm.​nih.​gov/​clinvar/​), 37 have been reported in the published papers, and only 4 have been reported in papers which focused on the HTx patients (Additional file 3: Table S3).

According to the ACMG/AMP 2015 guidelines, we found that 27 (16.4%), 59 (35.8%), 73 (44.2%), and 6 (3.6%) variants were interpreted as P, LP, VUS, and LB, respectively (Additional file 3: Table S3). Among these 86 P/LP variants, 41 (47.7%), 16 (18.6%), 8 (9.3%), and 4 (4.7%) were in TTN, FLNC, LMNA, and BAG3, respectively. In addition, 32 of these 86 P/LP variants have been reported in the Clinvar database, of which 18 variants are interpreted as P/LP (Table 3).

We first compared clinical characteristics between P/LP-positive and P/LP-negative patients, and no significant difference was observed (Additional file 3: Table S4). Then, we compared clinical characteristics between the P/LP-negative group and different variant groups (TTN, FLNC, LMNA, and BAG3 groups) (Additional file 3: Table S4). Compared with the P/LP-negative group, the TTN group exhibited a longer left atrial diameter (LAD) (TTN, 52.1 ± 9.2 mm vs. P/LP-negative, 48.9 ± 8.7 mm, P = 0.048) (Additional file 3: Table S4). The FLNC group contained more patients with NYHA class IV than the P/LP-negative group (FLNC, 16/18 vs. P/LP-negative, 81/123, P = 0.049) (Additional file 3: Table S4). The LMNA group exhibited more frequent pacemaker implantation (LMNA, 5/8 vs. P/LP-negative, 26/123, P = 0.025) and smaller LVEDD (LMNA, 65.0 ± 6.6 mm vs. P/LP-negative, 76.7 ± 11.3 mm, P = 0.005) than the P/LP-negative group (Additional file 3: Table S4). In addition, we compared the clinical characteristics of individuals carrying one P/LP variant with those carrying more than one P/LP variant, and no significant differences were observed (Additional file 3: Table S5).