Clinical study cohorts and classification criteria

We included 540 privately owned Dobermanns between 1999 and 2019 from several facilities in Europe: 400 dogs at Ludwig Maximilian University, Munich, Germany, and 31 dogs at the University Small Animal Hospital, University of Helsinki, Helsinki, Finland (“main cohort”); 65 dogs at Utrecht University, Utrecht, the Netherlands, and 9 dogs at the University Small Animal Clinic, University of Ljubljana, Ljubljana, Slovenia (“Utrecht cohort”); and 35 dogs at the Swedish University of Agricultural Sciences (SLU), Uppsala, Sweden (“Uppsala cohort”) (Additional file 2: Table S1). Clinical examinations consisted of a physical examination and a standard echocardiographic examination, including an assessment of left ventricular size by either Simpson’s method of disks (SMOD) or M-mode and, for all dogs of the main cohort, a 24-h Holter monitoring. The cutoff values, presented in Table 1, were based on the recommendations of Wess et al. [21, 41,42,43].

Table 1 Measurements and cutoff values for classifying the examined Dobermanns into groups with normal or abnormal echocardiography and 24-h ECG (Holter) results. The cutoffs were based on the recommendations of Wess et al. [12]

From the measured values, fractional shortening (FS) was calculated from left ventricular internal end-diastolic and end-systolic dimensions ((LVIDd − LVIDs)/LVIDd × 100) and left atrial to aortic root diameter ratio (LA/Ao) from the diameters of the left atrial appendage and the aortic root. Congestive heart failure (CHF) was confirmed by thoracic radiographs showing pulmonary edema as indicated by an interstitial or alveolar lung pattern in the caudo-dorsal part of the lung, plus enlargement of the left atrium and cardiomegaly.

We classified the dogs with detailed clinical information (N = 431, from Finland and Germany—“main cohort,” Table 2) into several subcohorts based on the findings as follows: (i) “echo only” (N = 45): dogs with an abnormal echocardiography result, a normal Holter result, and no CHF; (ii) “echo + arrhythmia” (N = 113): dogs with an abnormal result for both echocardiographic examination and Holter monitoring and no CHF; (iii) “arrhythmia only” (N = 70): dogs with a normal echocardiography result, an abnormal Holter result, and no CHF; (iv) “CHF” (N = 55): dogs at the overt stage of DCM; and (v) “healthy” (N = 148): dogs with a normal result for both echocardiographic examination and Holter monitoring that were at least 6 years old at the time of the examination.

Table 2 Characteristics of the three canine cohorts and their role in the genetic association study

To enrich for dogs with a very high likelihood of DCM phenotype, we used cutoff values of SMOD EDVI ≥ 100 ml/m2 and ESVI ≥ 60 ml/m2, which are slightly higher than the recommended values [21]. For the control group (“healthy”), we used the recommended SMOD EDVI < 95 ml/m2 and ESVI < 55 ml/m2 values. Dogs in the “gray zone” or when only one of the parameters was abnormal were excluded from this study. Only dogs fulfilling the Holter criteria clearly were included, and dogs in gray zones were excluded.

Dogs that had supraventricular premature contractions were included, but they were only put into the arrhythmia group if there were also ventricular premature contractions detected according to the criteria stated before. According to a recent study, dogs with only supraventricular premature contractions without VPCs were classified according to the ECHO criteria and not counted as arrhythmias, because supraventricular premature contractions were not early markers of DCM [44]. Dogs with atrial fibrillation were included in the arrhythmia groups and the respective ECHO classification because all dogs with atrial fibrillation also had VPCs on Holter examinations.

In addition to the 431 dogs of the main cohort included in GWAS, we evaluated a cohort of 74 dogs examined at Utrecht University, Utrecht, the Netherlands, and Small Animal Clinic, University of Ljubljana, Ljubljana, Slovenia (“Utrecht cohort”). These dogs were not included in the main cohort due to the lack of comprehensive phenotypic information. Clinical parameters collected include LVIDd, LVIDs, FS, and LA/Ao. Dogs were confirmed as unaffected when older than 9 years old and without clinical findings or signs indicative of DCM.

To enable genetic analyses, we collected EDTA blood samples from the study cohorts with the owners’ informed consent and stored them at −20 °C or −80 °C until genomic DNA was extracted. Extraction was performed with either a semi-automated Chemagen extraction robot (PerkinElmer Chemagen Technologie, Waltham, MA, USA), a Maxwell RSC 48 instrument with the Maxwell RSC Whole Blood DNA Kit (Promega, Madison, WI, USA), a Nucleon BACC2 kit (GE Healthcare, Chicago, IL, USA), or the QiaSymphony platform using the Mini-prep kit (Qiagen, Hilden, Germany). DNA concentration was measured with NanoDrop ND-UV/Vis Spectrophotometer or Denovix DS-11 Spectrophotometer. The samples were collected under the permission of the animal ethical committee of the County Administrative Board of Southern Finland (ESAVI/6054/04.10.03/2012, ESAVI/343/04.10.07/2016, and ESAVI/25696/2020); the ethical committee of the Center of Clinical Medicine, Munich, LMU University (60-18-11-2015); and the ethical committee of the Swedish Board of Agriculture (No. C2/12 (2012-02-24) and No. C12/15 (2015-02-27)). All sampling procedures in clinics conformed to the guidelines from Directive 2010/63/EU of the European Parliament on the protection of animals.

Association analyses

To identify new DCM loci, we performed genome-wide array SNP genotyping of genomic DNA prepared from blood samples taken from 431 clinically examined Dobermanns in the “main cohort” in three batches with the Axiom Canine Genotyping Array Sets A and B (1,269,218‬ markers) (Thermo Fisher Scientific, Waltham, MA, USA). Quality control (QC) was performed with PLINK v1.9 [45]. Each batch was pruned for a genotyping rate of > 95% per marker and individual before merging. Pre-analytical QC was conducted separately for each analysis and included genotyping rate of > 95% per marker and individual as well as a minor allele frequency of > 5% and Hardy-Weinberg test score of > 1 × 10−8. Three samples that failed the X-chromosomal sex check were discarded. Only markers classified into PolyHighResolution, NoMinorHom, or MonoHighResolution categories in every batch were retained. Finally, only one marker at each duplicate position was retained, and the rest were excluded from the analysis.

We analyzed the data with univariate linear mixed model association with a likelihood ratio test implemented in GEMMA (version 0.98.1) [46] with sex as a covariate to account for the sex bias in DCM [18] and a genetic relatedness matrix to correct for population stratification. Multiple testing correction was implemented with both Bonferroni correction and by correcting with the effective number of independent tests (Meff) estimated with simpleM [47,48,49]. We used the LD-based simpleM method as the primary correction approach, as Bonferroni correction is overly conservative for data with substantial LD. Finally, post-analytical QC included evaluation of population stratification from multidimensional scaling (MDS) plots and quantile-quantile (Q-Q) plots and assessment of SNP correlation structure calculated with PLINK [45].

Odds ratios and interaction analysis

To explore the genome-wide significant loci discovered in the association analyses, we examined the association and interaction of the risk SNPs chr5:60,531,090 and chr5:53,109,178 with left ventricular systolic dysfunction and dilatation using logistic regression, with 213 cases from the “echo only,” “echo + arrhythmia,” and “CHF” subcohorts as events and 148 controls from the “healthy” subcohort as non-events. Of these dogs, 52% were male and 48% female. Due to technical genotyping errors, the genotype for chr5:60,531,090 was missing for two dogs and chr5:53,109,178 for one dog, and thus, the final number of dogs included in the analysis was 358. We explained the echocardiographic phenotype with genotypes at chr5:60,531,090 and chr5:53,109,178 and the dog’s sex.

We carefully assessed model fit. First, we evaluated and plotted influential data points with packages “broom” [50], “dplyr” [51], and “ggplot2” [52] and did not find any outliers. Second, we evaluated multicollinearity with the package “car” [53]. The generalized variance inflation factor estimate was less than 1.05 for all variables, indicating no multicollinearity. Third, we calculated the area under the receiver operating characteristic curve (AUC) to evaluate how well the model is able to classify cases and controls. AUC was 0.789, indicating good classification.

After fitting the model, we estimated the overall effect of the explanatory variables with analysis of variance (ANOVA) with the package “car” [53]. Furthermore, we calculated the estimated marginal means for all explanatory variables using the package “emmeans” [54] to obtain means for variable levels and evaluate their pairwise differences.

The significance cutoff value was P < 0.05. Logistic regression as well as assessment of model fit and evaluation of the results were conducted in R (version 3.6.2) [55].

Whole-genome sequence analysis

We performed whole-genome sequencing of genomic DNA prepared from blood samples from twelve affected and one unaffected Dobermann using the Illumina HiSeq X ultra-high-throughput sequencing platform (Illumina Inc., San Diego, CA, USA) with 30 × target coverage (paired-end reads, 2 × 150 bp) (Novogene Bioinformatics Institute, Beijing, China). We originally sequenced ten dogs, including five from the “echo only” subcohort and five from the “echo + arrhythmia” subcohort in the Finnish and German populations, for our WGS analysis. Samples representing opposite homozygous haplotypes (Fig. 1e) were selected to discover possible candidate variants in the associated major locus in different clinical subgroups. Later during the study, three dogs from the Uppsala cohort, including one dog from the “echo only” subcohort, one dog from the “CHF” subcohort, and one dog from the “healthy” subcohort, were sequenced for an auxiliary analysis of a candidate gene; they were not included in the full WGS analysis. In addition, we used variant data from 393 whole-genome sequences from wolves and 87 non-Dobermann breeds that were either publicly available from the Dog Biomedical Variant Database Consortium (DBVDC) [56] or sequenced for our other studies (Additional file 2: Table S2).

Fig. 1

GWAS with a univariate linear mixed model. The analysis included 235 cases from the “echo only,” “echo + arrhythmia,” and “CHF” subcohorts and Utrecht cohort and 143 controls from the “healthy” subcohort. In a–c, Bonferroni-corrected and Meff significance thresholds are indicated with solid and dashed red lines, respectively. In d and e, the colors indicate the homozygous genotype for the case major allele (light gray), heterozygous genotype (middle gray), and homozygous genotype for the case minor allele (dark gray). In h and i, error bars indicate 95% confidence limits and asterisks ** and *** p-values < 0.01 and < 0.001, respectively. a A genome-wide significant two-locus signal occurs on chromosome 5: the major locus resides at 60 Mb and the minor locus at 53 Mb. The most significant SNP (praw = 1.40 × 10–9, pMeff = 8.80 × 10–5) is located at chr5:60,531,090. b A locus plot of chr5:45.0–69.0 Mb and SNP correlation structure of the index SNP chr5:53,109,178. c A locus plot of chr5:45.0–69.0 Mb and SNP correlation structure of the index SNP chr5:60,531,090. d Genotype plot of chr5:59–54 Mb. e Genotype plot of chr5:58–63 Mb. f Q-Q plot of the p-values (likelihood ratio test). g A multi-dimensional scaling (MDS) plot of the cases (red) and controls (black). h Probability of case status (P (case)) by genotype at chr5:53,109,178 (N = 372). i Probability of case status (P (case)) by genotype at chr5:60,531,090 (N = 372). j Frequency of case status by joint genotypes at ch5:53,109,178 and chr5:60,531,090 (N = 372)

For our in-house data, mapping, variant calling, and annotation for single nucleotide variants (SNV), small insertions/deletions (indel), structural variants (SV), and mobile element insertions (MEI) were performed as previously described using canFam3.1 as the reference genome [57, 58]. To account for inaccuracies in indel calling, genotypes of indel variants were considered heterozygous and homozygous if an individual had one or two copies of any non-reference allele, respectively. Additional annotations were obtained from the catalogs of Gene Ontology (GO) [59, 60], Mouse Genome Informatics (MGI) [61], and Human Protein Atlas [62]. The impact of exonic variants was predicted with SIFT [63].

To identify the variants that tag the risk alleles at chr5:60,531,090 and chr5:53,308,774, we retrieved all SNVs, indels, SVs, and MEIs in the regions in the ten initially sequenced affected dogs. In the chr5:60 Mb locus, we required that cases with one or two risk alleles shared the variants in heterozygous or homozygous state. In the chr5:53 Mb locus, we analyzed the two groups of cases separately according to genotype at chr5:53,109,178 (G/G or A/A), as based on allele frequency it could not be determined whether the G allele was introduced in the Dobermann breed formation resulting in an increased risk of echocardiographic changes or if the A allele was introduced resulting in a protective effect (Additional file 2: Table S3). In the filtering step, we omitted breeds with a known high prevalence of cardiac disease and utilized 366 control genomes with SNVs/indels and 211 with SVs/MEIs. As DCM occurs across dog breeds, we allowed a maximum of 10% of controls to carry the alternate allele in the heterozygous or homozygous state.

Sanger sequencing

To screen a candidate variant at chr5:60,111,983, we genotyped it in 35 dogs from the Uppsala cohort with standard PCR and Sanger sequencing. Using Primer3 [64], the following primers were designed: 5′-TCTCCCTCTTCCTCTCACCA-3′ (forward) and 5′-TGTCTCCCATGATCTCGGC-3′ (reverse). The PCR products were sequenced with an Applied Biosystems ABI3730XL DNA Analyzer capillary sequencer (Thermo Fisher Scientific) at the Institute for Molecular Medicine Finland (FIMM) core facility. The sequencing results were analyzed with UGENE (version 1.32.0) [65]. Finally, the association of the variant to left ventricular systolic dysfunction and dilatation was assessed with the one-tailed chi-squared test assuming the A allele to be associated.

RNA samples

We performed RNA sequencing on cardiac tissue to evaluate the differences in RNA expression between DCM-affected and unaffected Dobermanns. Following signed informed owner consent, we collected the tissues from four healthy and five affected Dobermanns (“Uppsala RNA cohort,” a subset of the Uppsala cohort) euthanized for reasons unrelated to this study at the University Animal Teaching Hospital in association with SLU, Uppsala, Sweden. Euthanization was initiated by intravenous injection in the cephalic vein of propofol (2,6 diisopropylfenol) at a dose between 2 and 6 mg/kg, followed by an intravenous administration of sodium pentobarbital at an approximate amount of 100 mg/kg.

The samples were collected from the same anatomical location of the heart for each dog, snap-frozen in liquid nitrogen, and stored at −80 °C until the date of process. Briefly, less than 25 mg of septal tissue was cut into smaller pieces on ice. Homogenization was performed in CK14-tubes (Bertin Technologies, Montigny-le-Bretonneux, France) for 2 × 20 s, with a 10-s delay in between, at 6000 rpm with the Precellys Evolution tissue homogenizer (Bertin Technologies) under cool conditions. The tissue was homogenized in lysis buffer according to the manufacturer’s recommendations for the AllPrep DNA/RNA Mini Kit (Qiagen), followed by DNA and RNA extraction according to the protocol. Samples were stored at −80 °C until further processed.

As a replication dataset, we also included myocardial tissue previously described by Cheng et al. [66] collected at several facilities in Ontario, Canada. Written consent was obtained from the clients for all procedures. Briefly, to euthanize the dogs, an intravenous injection of pentobarbital sodium at about 1 mL per pound of body weight, administered through a peripheral IV catheter. Left ventricular free wall tissue was collected from three affected Dobermanns and three unaffected large mixed-breed dogs. DCM diagnosis was made based on echocardiography (fractional shortening: unaffected 24.0 ± 7.2%, affected 7.7 ± 4.7%), gross pathology and histopathology, and CHF on the basis of history, physical examination, and thoracic radiographs. Samples were snap-frozen in liquid nitrogen and stored at −80 °C until further processed. Total RNA was prepared using TRIzol (Invitrogen, Waltham, MA, USA) as previously described [67].

RNA libraries and sequencing

We processed the RNA samples collected in Sweden as follows: TapeStation 2200 (Agilent Technologies, Santa Clara, CA, USA) was used for quantification and integrity evaluation of the RNA samples prior to library preparation. Samples with a RIN value of 7.0 or higher were chosen for processing. Sequencing libraries were prepared from 0.5 μg of total RNA using the TruSeq® Stranded mRNA LT-sample prep kit, including poly(A) selection (Illumina Inc.). Quantification of libraries was performed with KAPA Biosystems Illumina Library Quantification Kit for ABI Prism® (Roche, Basel, Switzerland) on StepOnePlus™ Real-Time PCR Systems (Applied Biosystems, Foster City, CA, USA). Libraries were normalized to 10 nM and pooled prior to sequencing. RNA sequencing was performed using the Illumina NextSeq550 system (Illumina Inc.). Paired-end 75-bp reads were obtained, generating 1.87–2.31 Gb of data per sample.

Second, we processed the RNA samples collected in Canada as follows: 1–3 ug of total RNA was used for the isolation of poly-A RNA (Dynabeads™ mRNA purification kit, Ambion, Life Technologies, AS, Norway). The poly-A RNA was reverse transcribed to double-stranded cDNA (SuperScript™ Double-Stranded cDNA Synthesis Kit, Life Technologies, Carlsbad, CA, USA). Random hexamers (New England BioLabs, Ipswich, MA, USA) were used for priming the first strand synthesis reaction and SPRI beads (Agencourt AMPure XP, Beckman Coulter, Brea, CA, USA) for purification of cDNA. RNA-seq libraries were prepared using 60 ng of cDNA for fragmentation and tagging with Nextera™ Technology (Illumina, Inc.). After the tagmentation reaction, the fragmented cDNA was purified with SPRI beads. To add Illumina-specific bridge PCR compatible sites and enrich the library, limited-cycle PCR (5 cycles) was done according to instructions of the Nextera system with minor modifications. To generate barcoded libraries, 50 X Nextera Adaptor 2 was replaced with barcoded Illumina-compatible adapters from the Nextera Bar Codes kit (Illumina, Inc.) in the PCR setup. SPRI beads were used to purify the PCR products, and library QC was performed with Agilent Bioanalyzer (Agilent Technologies). Finally, each transcriptome was loaded to occupy 1/3 of the lane capacity in a flow cell. C-Bot (TruSeq PE Cluster Kit v3, Illumina Inc.) was used for cluster generation and Illumina HiSeq2000 platform (TruSeq SBS Kit v3 reagent kit) for paired-end sequencing with 93-bp read length.

RNA sequencing data analysis

The generated RNA sequencing data in the FASTQ format were first checked with FastQC (version 0.11.8) [68], followed by trimming of adapter sequences and low-quality reads with Trimmomatic (version 0.32) [69]. SortMeRNA (version 2.1b) [70] was used to reduce noise from contaminating rRNAs before mapping the reads to the dog reference genome CanFam3.1 with STAR (version 2.7.2b) [71]. Of the filtered reads, 95.0–96.5% (Uppsala RNA cohort) or 86.7–88.8% (Ontario cohort) were mapped to the reference genome. We counted uniquely mapped reads with the software program featureCounts [72] and analyzed them for differential expression patterns with a generalized linear model approach and quasi-likelihood F-test in edgeR (version 3.10) [73] and Wald test in DESeq2 [74] with R (version 3.6.1) [55]. Finally, we inspected selected transcripts visually with Integrative Genomic Viewer (IGV, version 2.5.0) [75] and compared them with cardiac transcripts in Broad Improved Canine Annotation v1 [76] and dog mRNAs and ESTs from GenBank [77, 78].

Transcript analyses

To verify the gene expression data obtained by RNA sequencing, we performed direct Sanger sequencing and Droplet Digital PCR (ddPCR) (Bio-Rad Laboratories, Hercules, CA, USA). Briefly, total RNA from the previous extraction of four unaffected and five affected Dobermanns from the Uppsala cohort were re-quantified using Qubit RNA BR Assay Kit (Invitrogen). A volume of 8 µl corresponding to 500 ng total RNA was used as input for cDNA synthesis with the RT2 First Strand Kit (Qiagen), according to the manufacturer’s instructions. Quantification of synthesized cDNA was performed with Qubit ssDNA Assay Kit (Invitrogen).

We used the BigDye® Direct Cycle Sequencing Kit (Applied Bioscience, Waltham, MA, USA) for direct sequencing of RNF207, PRKAA2, and PLPP3 in three dogs to verify alternate transcripts observed in the RNA-seq data. The initial PCR amplification was performed on 4 ng cDNA using gene-specific primers (Additional file 2: Table S4) designed with Primer3 [64] tagged with M13 sequences according to the manufacturer’s instructions on the 5′-end. The following adjustment was made to the PCR protocol: 10 min enzyme activation at 95 °C, 35 cycles of denaturation for 3 s at 96 ºC, annealing at the temperature specified in Additional file 2: Table S4 for 15 s, and extension at 68 °C for 30 s. A final extension for 2 min at 72 °C was run before holding at 4 °C until further processed. Cycle sequencing was performed according to the manufacturer’s recommendations, and purification of products was done with the BigDye XTerminator® Purification Kit. Finally, we ran the samples on a 3500 Genetic Analyzer (Applied Bioscience) and evaluated the generated sequences with CodonCode Aligner v8.0.2 (http://www.codoncode.com).

We next evaluated the differential gene expression for selected candidate genes RNF207 and PRKAA2 and reference genes B2M and SRP14 with EvaGreen-based ddPCR. Particular emphasis was placed on optimizing the ddPCR to differentiate between fragments of different lengths for exon 7 of PRKAA2. Briefly, QX200™ ddPCR™ EvaGreen Supermix (Bio-Rad Laboratories) was mixed with forward and reverse primers (Additional file 2: Table S4), and 5 ng cDNA/reaction. QX200™ Droplet Generation Oil (Bio-Rad Laboratories) was used to generate droplets for each sample in an Automated Droplet Generator (Bio-Rad Laboratories). The droplets were transferred to a new plate, sealed with foil, and run in a ProFlex™ (Applied Bioscience). The protocol for the PCR was initiated with an enzyme activation at 95 °C for 5 min. Forty cycles were carried out with a first step of 30-s denaturation at 95 °C, followed by annealing and extension for 1 min at primer-specific temperatures (Additional file 2: Table S4). The signal was stabilized by 5 min at 4 °C and 5 min at 95 °C, followed by infinite hold at 4 °C. Each step was ramped 2 °C/s, and the lid temperature was set to 105 °C. After the run, the plate was placed in a QX200 Droplet Reader (Bio-Rad Laboratories) for sample quantification. Data evaluation was performed in the QX Manager Software (Bio-Rad Laboratories). We normalized the expression levels to the geometrical mean of B2M and SRP14 and performed statistical evaluation in JMP Pro (version 16, SAS Institute, Cary, NC, USA). Accounting for additional factors was not possible due to insufficient sample size.

CAGE-seq data

We inspected our WGS variants in the GWAS loci in the context of the CAGE-seq data from the DoGA consortium (https://www.doggenomeannotation.org/, manuscript in preparation). This data set contains CAGE-seq data from 118 samples, spanning 36 different tissue types from 8 dogs. To review variants residing in putative enhancer sites, myocardial-expressed bidirectional CAGE-seq reads, known to indicate active enhancers [79], that resided within 1 kb of gene start sites were overlapped with the WGS variants and catalogued in the variant tables.

Cardiac tissue stainings

Immunofluorescent staining was performed on canine cardiac tissue to inspect differences in protein expression and localization. Frozen tissue sections were cut at 4 µm, treated with acetone, and blocked with 3% BSA in PBS. ACTN2 (A7811, Sigma-Aldrich, St. Louis, MA, USA) combined with RNF207 (orb185936, Biorbyt) primary antibodies were incubated for 1 h at RT (1:250 in 1% BSA in PBS). Sections were washed with 0.5% Tween20 in PBS and incubated for 1 h at RT with Hoechst 33342 and the secondary antibodies (Invitrogen) Alexa Fluor 488 and Alexa Fluor 568. Fluorescence staining was mounted with Mowiol. Images were acquired with the Leica SP8 confocal microscope at × 63 magnification.

Human variant analysis

To ascertain the role of RNF207 and PRKAA2 as potential DCM risk genes in humans, we studied three different cohorts of patients referred for diagnostic genetic testing related to a diagnosis of cardiomyopathy between 2014 and 2020 at the University Medical Center Utrecht (UMCU), the Netherlands, and at Amsterdam Medical Center (AMC), the Netherlands. Genetic testing was performed using whole-exome sequence analysis in three patient groups comprising altogether 721 people: first, a cohort of 63 patients with DCM and ventricular arrhythmias without pathogenic or likely pathogenic variants in established cardiomyopathy genes; second, a cohort of 13 patients diagnosed with arrhythmogenic cardiomyopathy without pathogenic or likely pathogenic variants in established genes; and third, 645 patients referred for diagnostic cardiovascular genetic testing. In these patients, all coding fragments and exon-intron boundaries of RNF207 (NM_207396.3) and PRKAA2 (NM_006252.4) were analyzed, including evaluation of pathogenicity using variant counts and MAF as presented in gnomAD version 3.1.2 [80] and prediction of impact with PolyPhen-2 [81] and SIFT [63]. Clinical information of associated patients was retrieved from medical records. The study was conducted in accordance with the principles laid out in the Declaration of Helsinki and in line with the guidelines provided by the ethics committee of the University Medical Centre Utrecht, the Netherlands.

In addition to our patient cohorts, we utilized two large-scale datasets: the UK Biobank [82] and the public data of FinnGen (freeze 8) [83]. The UK Biobank dataset was comprehensively analyzed due to the availability of individual phenotypic data, while the FinnGen data was inspected on a less detailed level. For UK Biobank data, participants were included as DCM cases if they were ever diagnosed with left ventricular systolic dysfunction (LVSD) and had not been diagnosed with coronary artery disease (CAD), valvular, or congenital heart disease more than 100 days before the report of LVSD. The definitions of LVSD, CAD, and valvular and congenital heart diseases are listed in Additional file 2: Table S5. Furthermore, we used a previously developed and validated deep-learning methodology (AI-CMRQC) to extract left ventricular ejection fraction (LVEF) and left ventricular end-diastolic volume (LVEDV) measurements from the cardiac magnetic resonance data [84]. Any negative values and outliers (defined as larger or smaller than the median ± 3 × interquartile range) were removed as quality control. Then, participants were also included as DCM cases if they had an LVEF below 45% and an indexed LVEDV larger than 2 standard deviations from normal [85, 86]. Variant analysis in RNF207 and PRKAA2 was performed as described above, and an additional analysis of variant enrichment was performed for RNF207 by comparing the number of variants in affected and non-affected participants with Fisher’s exact test. Finally, we investigated the FinnGen data for RNF207 and PRKAA2 by querying each gene for phenotype-associated variants in the online interface, using a significance threshold of p < 5 × 10−8, and filtering the results for cardiovascular phenotypes.