Samples

In order to recruit EOAD-affected and normal hearing dogs, we conducted an online survey in 2020 to private dog owners of Rhodesian Ridgebacks who were registered in the Embark customer database and agreed to participate in scientific research. The survey asked if 1) a dog is deaf (bilaterally or unilaterally) and, if yes, 2) suspected age of onset of deafness. In addition, EOAD-affected dogs were also recruited via ProjectDOG (https://www.projectdog.org/), where dog owners reported their dogs showing early development of hearing impairment with clinical signs of EOAD. Hearing loss was initially determined by owners to ensure that a dog likely had hearing capability at birth but subsequently exhibited hearing loss before reaching adulthood (less than 2 years old). In addition, the loss of hearing was clinically confirmed in all case dogs by Brainstem Auditory Evoked Response (BAER) test [21, 22]. BAER testing was coordinated by ProjectDog through individual clinics and in-home visits. Selected control dogs were also BAER tested to confirm normal hearing. Control dogs with normal hearing were recruited by the survey. All control dogs were at least five years old at the time of the study to ensure that they did not exhibit hearing loss. This dataset was used for GWAS to identify genomic regions and variants associated with EOAD (hereafter referred to as a discovery dataset).

Three additional EOAD-affected and 12 control Rhodesian Ridgebacks were recruited in 2021 after the initial discovery of the associated genomic region. These three cases were confirmed to be EOAD by observing early loss of hearing by the owners, followed by BAER testing, which was conducted at individual clinics. Two were full siblings (dog_1187 and dog_1188), and their sire (dog_1186) was one of 12 control dogs. The remaining EOAD-affected dog was unrelated to any of the dogs used in this study. These dogs were used as a validation dataset to confirm the association between a putative causal mutation and EOAD. All samples used for the initial discovery and follow-up validation are listed in S1 Table.

Genotyping and genome-wide association

Buccal swabbing was conducted by dog owners to collect cheek cell samples for DNA extraction. DNA was extracted by Illumina, Inc. These DNA samples were genotyped using Embark’s custom high-density SNP arrays (221,412 markers on 38 autosomes and 6,649 markers on the X chromosome). Genotyping rate calculation, pi_hat estimate, and marker filtering were performed by PLINK v1.9 [23].

A genomic region associated with EOAD in the Rhodesian Ridgeback was identified by using a multivariate linear mixed model implemented in GEMMA v.0.98 [24]. To minimize the confounding effects of shared ancestry between dogs, a kinship matrix was generated by GEMMA from the filtered genotype dataset and used as a random effect in the model. A Wald test was applied to detect significant association between markers and phenotype with a threshold of P < 5.0 x 10⁻⁸. To evaluate marginally significant associations identified in the main GEMMA analysis, we repeated the analysis with the same number of case and control samples (10 and 107, respectively) out of the entire dataset including related individuals (23 case and 162 control dogs) by randomization with replacement for 100 times. We only considered regions showing significant association consistently across the 100 replications as a candidate EOAD-associated region in downstream analyses. To identify haplotypes associated with EOAD, we phased genotypes of markers on CFA18 of all genotyped Rhodesian Ridgebacks with Beagle v4.1 with default parameter settings [25].

Sequencing analysis

Genomic DNA was extracted from one EOAD-affected dog and sequenced on an Illumina Genome Analyzer IIx to produce 100 bp paired-end reads. Raw sequence reads were mapped to the CanFam3.1 dog reference assembly by following the GATK best practice [26]. Briefly, adapter-marked sequence reads by MarkIlluminaAdapters were mapped to the reference genomes by using the BWA-MEM algorithm in BWA 0.7.17 [27], followed by duplicate marking by MarkDuplicates in Picard tools (http://sourceforge.net/projects/picard/). Base quality score was calibrated with BaseRecalibrator and ApplyBQSR in GATK version 4.1.9 [28]. Approximately 570 million reads were produced and 99.2% of the reads mapped to the dog reference genome. Sequence variants were called using HaplotypeCaller in GATK. Potential functional impacts of these variants were evaluated by using the Variant Effect Predictor (VEP) version 99.2 (http://www.ensembl.org/vep).

To identify large structural variations (SVs) in the EOAD-associated region on CFA18, we used Manta [29], which uses paired and split-read evidence for SV detection in mapped sequencing reads. In addition to the EOAD-affected Rhodesian Ridgeback, we downloaded whole-genome sequence data for 4 Rhodesian Ridgebacks and 37 dogs of the three additional breeds from the NCBI Sequence Read Archive (https://www.ncbi.nlm.nih.gov/sra) (S2 Table). We selected Pembroke Welsh Corgis, German Shepherd Dogs, and Labrador Retrievers as additional reference breeds because 1) a core EOAD-associated haplotype identified in Rhodesian Ridgebacks was identified in these breeds, and 2) these breeds are not known to exhibit EOAD. Sequence reads of these samples were mapped to the CanFam3.1 reference genome by using the method described above.

To characterize the genetic variations in a candidate gene EPS8L2, we designed 50 PCR primers to amplify regions in and around this gene (S3 Table). Amplicons were purified by Monarch PCR and DNA purification kit and were sequenced by the Sanger method at the Cornell Genomics Facility.

Ethics statement

Participating dogs were part of the Embark Veterinary, Inc., customer base. Owners provided informed consent to use their dogs’ data in scientific research. Non-invasive methods for genotype and phenotype collection were used, and dogs were never handled directly by researchers. Owners were given the opportunity to opt-out of the study at any time during data collection. All published data have been de-identified of all personal information as detailed in Embark’s privacy policy (embarkvet.com/privacy-policy/).

Early-Onset of Adult Deafness (EOAD) in Rhodesian Ridgeback

We identified 23 EOAD-affected Rhodesian Ridgebacks (11 males and 12 females) that were confirmed deaf by BAER testing; these dogs lost their hearing approximately from 6 months to 24 months after birth. All of the dogs were bilaterally deaf. We also identified 162 adult dogs (five years or older) with normal hearing as reported by their owners in the online surveys. Among these, six dogs were BAER tested, confirming their normal hearing. Some of the 23 EOAD-affected dogs were closely related with each other with the average relatedness being higher than the control group (pi-hat 0.226 vs 0.176). Our EOAD case dogs included one 2-generation family with seven members where two offspring both had EOAD, while their parents and other relatives exhibited normal hearing (S1 Fig). These samples were used in the downstream analyses to identify EOAD-associated regions by GWAS and screen candidate variants by haplotype analysis.

A novel association on chromosome 18 with EOAD

Closely related dogs, defined as pairs of dogs with pi_hat > 0.45, were removed from the dataset used for the initial discovery of the associated genomic region by GWAS; 13 EOAD case dogs and 53 control dogs were removed. After removing markers with more than 3% missingness across the remaining dogs, we used 192,249 markers for GWAS, for which the genotyping rate was 99.8% in 117 dogs (10 case dogs and 107 control dogs) (S1 Table).

Our GWAS of 10 unrelated EOAD-affected dogs and 107 control dogs identified one highly significant region on CFA18 and two other marginally significant regions on CFA24 and CFA26 (Figs 1 and S2). The most significant marker was at the position 25,448,444 on CFA18, an intronic region of SEPTIN14 gene (P = 5.8 x 10⁻³⁵). All affected dogs were GG homozygotes at this marker, whereas none of the control dogs carried the GG genotype (94 and 13 control dogs were AA homozygotes and AG heterozygotes, respectively) (S1 Table). The other two significant markers were at CFA24:42,714,739 (intergenic region, P = 8.8 x 10⁻⁸) and CFA26:23,554,170 (intronic region of MTMR3, P = 2.0 x 10⁻⁹). These two markers were not consistently detected as significantly associated regions in replicated runs, where we repeated the GWAS analysis 100 times by randomly selecting 10 case and 107 control dogs from the entire dataset of 23 case and 162 control dogs (S4 Table). CFA24:42,714,739 was not significantly associated with EOAD in any of the replicated runs, and CFA26:23,554,170 was significantly associated with EOAD in 14 of the 100 replicated runs. In contrast, the CFA18 region was consistently significantly associated with EOAD across these replicated analyses, and thus, we only consider the region on CFA18 as a candidate region of EOAD in the following analysis.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. A Manhattan plot of association with Early Onset Adult Deafness (EOAD) in Rhodesian Ridgebacks.

Red and blue horizontal lines are significant (P < 5 x 10⁻⁸) and suggestive (P < 1 x 10⁻⁵) associations, respectively.

https://doi.org/10.1371/journal.pone.0264365.g001

Out of all EOAD affected dogs including closely related individuals (n = 23), 19 dogs were homozygous for a common haplotype in the associated region on CFA18, defined as a 1-Mb region from the position 24,891,130 to 25,897,356 where all markers showed Bonferroni corrected P values < 1.0 x 10⁻⁸ in the GWAS (Fig 2). The remaining four EOAD-affected dogs had one or two copies of closely related haplotypes, which were different from the common haplotype by one to five SNPs out of 140 markers. By including these closely related haplotypes as EOAD-associated haplotypes, all of the EOAD affected dogs were homozygous for an EOAD-associated haplotype. In the 162 control dogs, the most common EOAD-associated haplotype was found in 20 dogs (12%) in a heterozygous form, but none of the control dogs carried two copies of an EOAD-associated haplotype. Both parents of the two affected siblings (S1 Fig) had one copy of the EOAD-associated haplotypes, indicating an autosomal recessive mode of inheritance where both the sire and dam were a carrier of the risk variant. One of the half-siblings of the affected dogs (Dog_1174) was a carrier of an EOAD-associated haplotype, whereas the other half-sibling (Dog_1106) and the dam of these two dogs (Dog_1175) did not have any EOAD-associated haplotype.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Haplotypes near the marker on CFA18 that are significantly associated with Early Onset Adult Deafness (EOAD) in Rhodesian Ridgebacks.

Random subsets of control dogs were selected for visualization purpose: Control-1 dogs (blue) do not carry any of the EOAD-associated haplotypes, whereas Control-2 dogs (orange) carry one copy of the EOAD-associated haplotype (i.e., a carrier). Affected dogs are indicated in red. Rows correspond to haplotypes (two rows/individual), and columns correspond to markers. The positions of the first and last markers are 24,891,130 and 25,897,356, respectively. The most significant marker in the GWAS is indicated by a red triangle (position 25,448,444). The marker closest to the 12-bp deletion is indicated by a blue triangle (position 25,752,381). Magenta triangles indicate markers where Control-1 dogs are homozygous for a reference allele and affected dogs are homozygous for an alternate allele. These 13 informative markers, along with the most significant marker (red triangle) are used for defining a core haplotype predictive of EOAD in Rhodesian Ridgebacks. Positions of the informative markers are 25545949, 25557090, 25660957, 25696066, 25704136, 25708602, 25857995, 25861018, 25862585, 25863350, 25863894, 25883389, and 25897356. The genome coordinates in canFam3.

https://doi.org/10.1371/journal.pone.0264365.g002

In order to define a core haplotype predictive of EOAD in Rhodesian Ridgebacks and identify other dog breeds that carried the same core haplotype, we chose 14 markers across a 449-kb window that were homozygous for the alternate allele in all EOAD-affected Rhodesian Ridgebacks (Fig 2). We looked for this haplotype in a sample of 110,205 purebred dogs submitted for commercial genetic testing and found 1,399 of them (1.3%) were homozygous for the haplotype. These dogs were of various breeds, with a significant number being Pembroke Welsh Corgis (1,304 dogs or 93.2%). As Pembroke Welsh Corgis were not known to exhibit EOAD, we chose this breed as a reference to screen and validate candidate causal variants linked with this core haplotype, assuming that any causal mutation found on this risk haplotype in EOAD-affected Rhodesian Ridgebacks would not be present in the Pembroke Welsh Corgis carrying the haplotype (see the Validation of the association section).

Whole genome sequencing of one EOAD affected dog identified 2,089 non-reference homozygous variants within the associated region. Two SNPs (CFA18:25494223T>A and CFA18:25660957T>C) were predicted to be deleterious by introducing missense mutations in PKP3 and LRRC56 (SIFT score = 0.02). However, these variants were probably not associated with EOAD because whole genome variants of 722 dogs and wild canid species in a NCBI database [30] showed that 63 and 65 dogs were homozygous for the non-reference alleles at these two SNP positions (allele frequencies = 0.206 and 0.162, respectively). While auditory phenotypes of these samples in the public database were unknown, they likely had normal hearing. Two deletions relative to the reference sequence were identified within the associated region, and the EOAD affected dog was homozygous for both (50-bp deletion at CFA18:25874759–25874809 and 221-bp deletion at CFA18:25885742–25885963). However, these two deletions were also found in four other Rhodesian ridgebacks reported in Plassais et al. [30] (S2 Table). Since the 50-bp deletion was not in linkage disequilibrium with the associated variant identified in the GWAS (CFA18:25,448,444), we concluded this deletion was unrelated to EOAD. The other 221-bp deletion was in linkage disequilibrium with the most associated GWAS marker CFA18:25,448,444. This deletion was likely accompanied by SINEC_Cf (CFA18:25885768–25885963) located at an intron of ALKBH3. Since this deletion was also identified in Pembroke Welsh Corgis, German Shepherd Dogs, and Labrador Retrievers (S2 Table), we did not further consider this deletion as a candidate for EOAD since this condition has not been reported in these breeds.

Among 33 protein coding genes in the EOAD-associated region based on the Ensembl gene annotation (S5 Table), EPS8L2 (Ensembl gene ID: ENSCAFG00000025183) was of particular interest because this gene was identified as a putative causal gene for hearing loss in consanguineous families in humans [31, 32]. The Ensembl gene annotation predicted 22 exons in this gene in one of the latest dog reference genomes (UMICH_Zoey_3.1/canFam5), whereas the canFam3.1 reference genome had fewer annotated exons. The difference in the gene annotation between the two canine genome assemblies was likely due to the high GC-content, which introduced misassemblies in canFam3.1 (65.9% and 66.6% GC-content in canFam3.1 and canFam5, respectively). To identify candidate causal variants associated with EOAD, we designed primers to amplify the entire EPS8L2 region based on the UMICH_Zoey_3.1/canFam5 reference genome (S3 Table and S3 Fig). Out of these primer pairs, 15 exonic and 11 intronic regions were partially amplified (a total length of 6.7-kb, mean fragment size of 514 bp). Sequencing these fragments identified three SNPs and two indels in 11 EOAD affected dogs. A SNP in the first intron at CFA18:25859452G>A was not perfectly associated with EOAD, whereas the other two SNPs in intron 13 (CFA18:25869159C>T) and intron16 (CFA18:25869858G>A) were perfectly or nearly perfectly associated with the EOAD status (S6 Table) (coordinates in UMICH_Zoey_3.1/canFam5). Among the two deletions that we found, the 12-bp deletion in exon 12 (CFA18:25868739–25868751) was perfectly associated with EOAD status, where all EOAD-affected dogs were deletion homozygotes (n = 22 with available DNA, one DNA sample was not available at the time of analysis), 17 dogs with one copy of the EOAD-associated haplotypes (Fig 2) were heterozygous for the deletion, including the obligate heterozygote dogs (parents of the EOAD-affected dogs; Dog_1029 and Dog_1038; S1 Fig), and 33 dogs without the EOAD-associated haplotypes did not have the deletion (S6 Table). The other deletion was found in intron 17 (CFA18:25870908–25870912) and was not associated with the condition by having both a wildtype homozygote and a deletion homozygote in the EOAD case group (S6 Table).

Functional prediction of the 12-bp deletion in EPS8L2

Ensembl Variant Effect Predictor (VEP) did not predict any strong pathogenicity in the four intronic variants in EPS8L2. In contrast, the 12-bp deletion likely causes an inframe shift by removing four amino acids from the EPS8L2 protein (valine, histidine, phenylalanine, and leucine: VHFL). A translated amino acid sequence of exon 12 consists of 25 amino acid residues and was highly conserved across vertebrates (S4 Fig). The four amino acid residues that were deleted in the EOAD-associated haplotypes were identical in all analyzed vertebrate species in the RefGene dataset except the Malayan flying lemur (Galeopterus variegatus, galVar1) and the zebrafish (Danio rerio, danRer11). The variant was detected as deleterious by PROVEAN (Protein Variation Effect Analyzer) [33] with the score = -29.992 (score smaller than -2.5 considered as deleterious), but MutPred-Indel [34] did not identify a sufficiently strong statistical signal to call this deletion to be pathogenic by using a Human Gene Mutation Database as a training dataset (score = 0.43471, where the score > 0.5 are considered to be pathogenic).

Folding Unit predictor (FUpred) [35] predicted that the region containing the deleted VHFL amino acids has a capability of forming an alpha helix, which was a part of a predicted protein domain D5 (S5 Fig). A predicted tertiary structure of the human and mouse EPS8L2 proteins by AlphaFold (UniProtKB: Q9H6S3 and Q99K30) showed that the VHFL amino acid residues are indeed a part of an alpha helix structure (S6 Fig).

Validation of the association

To test the association between the 12-bp deletion at CFA18:25868739–25868751 and EOAD, we genotyped three additional EOAD-affected dogs and 12 additional control dogs by Sanger sequencing. We confirmed that all of the three EOAD-affected dogs were homozygous for the 12-bp deletion at CFA18:25868739–25868751, whereas none of the control dogs carried two copies of the deletion. Consistent with the predicted recessive mode of inheritance, the sire (dog_1186) of the two full-sibling EOAD-affected dogs (dog_1187 and dog_1188) was a carrier by having one copy of CFA18 with the deletion (S1 Table).

The core haplotype predictive of EOAD in Rhodesian Ridgebacks (Fig 2) was also common in Pembroke Welsh Corgis, but this breed is not known to exhibit EOAD. To validate the association between the 12-bp deletion and EOAD, we chose eight adult Pembroke Welsh Corgis that were homozygous for the core haplotype with no owner-reported hearing problems as controls and genotyped them for the 12-bp deletion. This analysis confirmed the absence of the deletion in these Pembroke Welsh Corgis, further supporting the association between the deletion and EOAD in Rhodesian Ridgebacks.