A Splice Acceptor Variant in DLL3 Is Associated with Spondylocostal Dysostosis in a Litter of Mixed-Breed Dogs

Abstract

Background/Objectives: Spondylocostal dysostosis (SCDO) is a rare disorder characterized by congenital malformations of the spine and ribs. SCDO affects 1 in 40,000 human births, with rare cases also reported in dogs. Mutations in DLL3, encoding a critical Notch signaling pathway ligand, account for a majority of human SCDO cases. The remaining cases have variants in HES7, LFNG, MESP2, RIPPLY2, TBX6, and DLL1, which code for proteins in the Notch pathway. A mixed-breed litter of three dogs presented with varying degrees of spinal malformations and underwent comprehensive phenotyping including radiographic and neurologic examination. Two littermates demonstrated classic SCDO features including shortened torsos, vertebral malformations, and rib abnormalities, while a third showed only caudal vertebral truncation. Methods: Short-read whole-genome sequencing was performed on all three animals, followed by variant filtering and analysis using the two severely affected dogs as cases and 173 control dogs of various breeds. Variants were prioritized based on segregation patterns, population frequency, and predicted functional impact using established bioinformatics tools. Results: Variant analysis identified a novel splice acceptor variant in DLL3 (c.650-2A>C). This mutation, located at the splice acceptor site preceding exon 5, is predicted to disrupt critical EGF-like domains and O-fucosylation sites essential for DLL3 protein function. Conclusions: This study identifies a DLL3 splice variant causing SCDO in dogs, demonstrating phenotypic conservation with humans. These findings refine our understanding of genotype–phenotype correlations and demonstrate the value of comparative genomics for rare developmental disorders.

1. Introduction

Spondylocostal dysostosis (SCDO) is characterized by malformation of the spine and ribs. It is a rare congenital disease that reportedly affects 1 in 40,000 human births [1] and has been reported in dogs both historically and in contemporary cases [2,3,4,5]. Characteristics of SCDO in people include a short trunk, mild scoliosis, and a short neck. SCDO is diagnosed based on radiographic abnormalities [6]. Vertebral malformations include a reduced number of vertebrae, hemivertebrae, and butterfly vertebrae [6,7]. Rib abnormalities are often characterized by proximal or distal fusion, reduction in number, and malalignment [6,7]. Disease management depends on the severity of the phenotype but can include respiratory support in cases of compromised respiratory function due to reduced thoracic capacity and surgery for patients with severe scoliosis [1,6,8]. Males with SCDO are at a higher risk for inguinal hernias, and neurologic complications are uncommon but can occur [9]. Although respiratory dysfunction can rarely lead to infant mortality, most SCDO patients live into adulthood with normal life expectancy [6].

SCDO exhibits genetic heterogeneity with mutations in multiple genes producing similar phenotypes, and SCDO is classified into subtypes based on these mutations in people. Mutations include MESP2 (SCDO type 2), LFNG (SCDO type 3), HES7 (SCDO type 4), TBX6 (SCDO type 5), RIPPLY2 (SCDO type 6), and most recently, DLL1 (SCDO type 7), although pathogenic variants in DLL3 (SCDO type 1) remain the most common [9,10,11,12,13,14,15,16,17]. All these genes function within or interact with the Notch signaling pathway, which controls somite segmentation during embryonic development [7]. SCDO is usually the result of autosomal recessive mutations, although in one family, a TBX6 mutation was reported to have autosomal dominant inheritance [16].

SCDO has been studied using the pudgy mouse model, first described in 1961 and characterized by shortened vertebral columns with marked vertebral and rib irregularities and shortened, twisted tails [18]. In 1998, these malformations were attributed to a protein-terminating mutation in DLL3, which disrupts the Notch signaling pathway [19]. SCDO has also been reported in dogs. Historical reports from the mid-1900s describe dogs with shortened spines and normal legs and skull, including ‘Baboon Dogs’ in South Africa, which was suspected to be caused by an autosomal recessive mutation, and short-spine dogs studied in Japan that reportedly lived normal lifespans despite severe vertebral malformations [2,3,4]. Several 17th century paintings by David Klocker Ehrenstrahl also depict dogs with similar malformations [20]. More recently, a frameshift mutation in HES7 was identified as the cause of SCDO in an outbred family of miniature schnauzers [5]. The disorder is known as “Comma defect” because of the comma-shaped morphology seen in affected puppies [5]. The phenotype in these cases was severe, characterized by sacrococcygeal agenesis, hemivertebrae, and a poorly developed rib cage with missing and fused ribs [5]. All affected puppies in the studied litter were stillborn or died shortly after birth, likely due to reduced respiratory function [5].

Here, we report a mixed-breed litter presenting with spondylocostal dysostosis characterized by malformations of varying severity. Given the unrelated breed background to the previously described miniature schnauzer cases and the survival of affected puppies, we sought to identify the genetic cause in these dogs using short-read whole-genome sequence data.

2. Materials and Methods

2.1. Sample Collection

Blood was collected from the three mixed-breed dogs in this study as part of clinical workup, with consent obtained (IACUC, protocol 21942). DNA was extracted using EDTA blood with the QIAGEN Puregene Blood Kit (Qiagen, Hilden, Germany) using whole blood protocol. Following extraction, DNA concentration and purity was evaluated by spectrophotometry (NanoDrop ND-1000, Thermo Scientific, Waltham, MA, USA). DNA samples were maintained at −20 °C prior to whole-genome sequencing. DNA samples collected previously (IACUC, protocol 22035) from 354 dogs were used for genotyping and whole-genome sequence analysis (Supplementary Data S1).

2.2. Clinical Evaluation

Three mixed-breed puppies were presumed to be littermates since they were found together at 14 weeks of age. Extensive phenotyping was performed at the William R. Pritchard Veterinary Medical Teaching Hospital. Physical and neurologic examinations were performed by a board-certified veterinary neurologist (K.V.). Radiographs were interpreted by a board-certified veterinary radiologist (C.B.).

2.3. Dog Breed Composition Identification

Because of unknown history from this litter, pedigree information is not available. The breed composition of this litter cannot be determined by appearance alone, as these dogs are of varying size and coat color. DNA from puppies 1 and 3 were submitted to Embark Breed + Health testing (Embark Veterinary Inc., Ithaca, NY, USA) to determine breed composition and assess inbreeding.

2.4. Whole-Genome Sequencing and Analysis

Short-read whole-genome sequencing was performed on the three mixed-breed littermates. These cases were compared to 173 unaffected dogs representing multiple breeds. Out of 173 controls, 147 were previously published, with BioProject, BioSample, and SRR Accession numbers listed in Supplementary Data S1. Twenty-six control samples were deposited under the NCBI BioProject associated with this publication, PRJNA1371007. Library preparation and 150bp Illumina paired-end sequencing were performed by the UC Davis DNA Technologies Core. Raw sequencing data processing and alignment to the canine reference genome CanFam4 (UU_Cfam_GSD_1.0) were performed as described by Chevallier et al. [21]. Variant calling was performed on the combined 176 samples. SnpEff (v5.1) was used to add effect predictions to the VCF, and then variants were filtered for coding variants only. The coding variant VCF was converted into PLINK format (.bed, .bim, .fam) using PLINK v1.9.0-b.7.7 [22]. These files were used for variant analysis and filtering. Identity-by-descent (IBD) analysis was completed using --genome in PLINK to assess pairwise relatedness (Pi-hat, proportion of genome shared IBD).

2.5. Variant Analysis and Filtering

Allele frequency analysis was performed using PLINK v1.9.0-b.7.7 using the --assoc command on whole-genome sequence data from 175 dogs, with the two severely affected littermates (puppies 2 and 3) as cases and 173 dogs as controls (Supplementary Data S1). Puppy 1 was excluded from initial analysis due to an ambiguous phenotype that could represent a heterozygous carrier state, reduced penetrance, or an unrelated tail defect. Only variants that were homozygous mutant in cases and absent in controls were retained as candidate variants. A multi-step filtering strategy was then applied. Variants present in the Dog 10K database [23], which contains 34.6 million variants from 1987 canids representing 321 dog breeds, village dogs, and wolves [24], were excluded. Puppy 1’s genotype was then incorporated under a recessive inheritance model, excluding variants that were homozygous in this phenotypically normal littermate. Finally, variants were prioritized based on SnpEff functional impact annotations, and only high and moderate impact variants were retained. Variant effects were predicted using multiple computational tools. Combined Annotation-Dependent Depletion (CADD) scores were calculated to assess variant deleteriousness [25]. Missense variants were evaluated using AlphaMissense [26]. Conservation across vertebrate species was evaluated using PhyloP [27,28]. Potential splicing effects were assessed using Splice Site Prediction by Neural Network (SSPnn) [29].

2.6. Genotyping Assay

Sanger sequencing was used to validate the DLL3 variant in affected dogs and screen for carrier frequency in 105 German Shepherd Dogs and 76 Australian Cattle Dogs from the Bannasch Lab DNA Repository. These breeds were selected based on predominant breed ancestry of the subjects of this study. PCR amplification was performed in 20 μL reactions containing 2 μL of 10X buffer (Qiagen, Hilden, Germany), 0.8 μL 25 nM MgCl₂, 0.5 μL 10 mM dNTPs, 0.5 μL of 20 μM primers (DLL3_F: 5′-TCCATCTGCCACTTCCCATC-3′, DLL3_R: 5′CACATGGGTTCCCGTCACAG-3′), 0.2 μL HotStar Taq Plus polymerase (Qiagen, Hilden, Germany), and 1 μL template DNA. Thermocycling conditions followed the manufacturer’s protocol for HotStar Taq Plus. PCR products were treated with ExoSAP-IT (Thermo Fisher Scientific, Waltham, MA, USA) and sequenced by Azenta Life Sciences (Genewiz, South Plainfield, NJ, USA).

3. Results

3.1. Phenotyping

Three 14-week-old, presumptive sibling mixed-breed puppies were presented to the Wiliam R. Pritchard Veterinary Medical Teaching Hospital for evaluation of their health. There was one grossly phenotypically normal male (puppy 1), one affected male (puppy 2), and one affected female (puppy 3).

Puppy 1 had a short tail but was otherwise physically normal (Figure 1a), with unremarkable physical and neurological examinations. Puppy 2 exhibited a shortened appearance of the cervical and thoracolumbar vertebral column and a truncated tail (Figure 1b) but had normally proportioned head and limbs. Puppy 3 was smaller in stature compared to her two littermates, with a similarly shortened cervical and thoracolumbar vertebral column and a truncated tail (Figure 1c). No voluntary tail movement was observed and there was reduced range of motion of the tail upon manipulation. Both affected puppies (2 and 3) were thin with a body condition score (BCS) of 3/9, and demonstrated reduced cervical range of motion, but physical and neurological examination findings were otherwise within normal limits.

Vertebral column radiographs were obtained on all three puppies under sedation.

Puppy 1 exhibited truncation of the tail with absence of the caudal endplate of the sixth caudal vertebrae (Cd6) and all succeeding caudal vertebrae along with mild flattening of the caudal occiput. No other abnormalities were noted on radiographs (Figure 2).

In puppies 2 and 3, there were a reduced number of vertebral bodies, multiple congenital vertebral malformations, including hemivertebrae, butterfly vertebrae, and block vertebrae, as well as fusion and bifid deformity of spinous processes and fusion of the rib heads throughout the cervical, thoracic, and lumbar vertebral column and tail, where applicable (Figure 3 and Figure 4). Puppies 2 and 3 also exhibited mild flattening of the caudal occiput, scoliosis, kyphosis, narrowing of intervertebral disc spaces, and truncation of the caudal vertebral column, though the changes were less pronounced in puppy 2. Puppy 3 additionally had asymmetric rib formation and variable vertebral malalignment. Both puppies had evidence of skeletal dysplasia characterized by disproportionate limb shortening, most pronounced in the proximal appendicular skeleton (Figure 5).

3.2. Dog Breed Composition Identification and Inbreeding

Embark Breed + Health testing on puppies 1 and 3 revealed they share 64% of their DNA and are predominantly German Shepherd Dog (41.6% and 36.7%, respectively) and Australian Cattle Dog (36.4% and 38.7%, respectively) mixes, with the remaining amount a mix of other breeds. Puppy 1 had a coefficient of inbreeding (COI) of 20% and puppy 3 had a COI of 27%. According to the Embark website, the COI for an average mixed-breed dog is 3%. PLINK identity-by-descent analysis confirmed the three puppies were full siblings (pairwise Pi-hat = 0.716–0.732), with elevated IBD sharing consistent with recent common ancestry.

3.3. Short-Read Whole-Genome Sequencing and Variant Analysis

The three puppies were whole-genome-sequenced, and variants were called along with 173 unaffected controls. Protein coding variants shared in puppies 2 and 3 but differing from the reference genome assembly were filtered based on their absence in control dogs. One hundred and eighty-six variants were absent in control dogs, including 175 variants within a 9 Mb haplotype on chromosome 1. Sequential filtering reduced these candidates to a prioritized set (Figure 6). First, 174 variants were excluded after comparison with the Dog 10K database revealed their presence in 1987 phenotypically normal dogs [23]. Three variants that were homozygous in puppy 1 were removed, given the phenotypic differences between the puppies. The remaining 9 variants were all located on chromosome 1. SnpEff annotations classified these as high impact (1), moderate (3), low (3), and modifier (2). The 4 high and moderate impact variants were prioritized for further analysis (

Table 1. Four variants remained after variant filtering and analysis.

Location and Mutation (CanFam4)	Gene	Predicted Effect	Alpha- Missense	PhyloP	CADD
Chr1:114,529,289 c.650-2A>C p.?	DLL3	High: splice acceptor variant	N/A	10.9	23.3
Chr1:115,448,238 c.1525G>A p.Ala509Thr	GGN	Moderate: missense	Not conserved	Not conserved	Not conserved
Chr1:115,708,525 c.947C>G p.Pro316Arg	SIPA1L3	Moderate: missense	0.18	7.6	25.6
Chr1:119,875,029 c.1164C>A p.Asn388Lys	GPATCH1	Moderate: missense	0.37	−1.2	14.2

).

The variants in GGN, SIPA1L3, and GPATCH1 appear less significant, either because predicted effect scores such as AlphaMissense indicate a benign or ambiguous change, or because the base is not conserved in humans or mice, suggesting that the mutation is less likely to be deleterious or damaging [26,30]. Variant analysis identified a novel splice acceptor variant in DLL3 (NC_049222.1(XM_038528906.1): c.650-2A>C, p.?). The variant has a PhyloP score of 10.9, indicating a highly conserved nucleotide across 470 mammalian species [27,28]. This c.650-2A>C substitution disrupts the canonical splice acceptor AG dinucleotide, which is essential for spliceosome recognition. The variant received a CADD score of 23.3 [25]. The splice acceptor mutation occurs immediately before exon 5 (Figure 7A). Splice site prediction analysis using SSPnn confirmed disruption of the canonical splice acceptor, with the wildtype sequence showing a high confidence acceptor site prediction score of 0.98 that did not appear in the mutant sequence [29]. A weak cryptic splice acceptor site was identified 164 nucleotides downstream within exon 5 (SSPnn score 0.47). While the precise splicing outcome cannot be determined without functional validation, several aberrant splicing patterns are predicted based on in silico analysis (Figure 7B).

3.4. Allele Frequency of DLL3 Variant

WGS genotypes of puppies 1, 2, and 3 were confirmed with targeted genotyping assay. Puppies 2 and 3 are homozygous for the DLL3 variant, while Puppy 1 is heterozygous. Genetic analysis identified this litter as predominantly Australian Cattle Dog and German Shepherd Dog. The DLL3 variant was not detected in 105 German Shepherd Dogs or 76 Australian Cattle Dogs, further supporting this as a rare variant associated with a recent inbreeding event.

4. Discussion

DLL3, or Delta Like Canonical Notch Ligand 3, encodes for the DLL3 protein, an important ligand in the Notch signaling pathway. The Notch signaling pathway mediates cell-to-cell communication during development, controlling cell fate decisions and tissue patterning through direct contact between neighboring cells [31]. As a cis-acting inhibitory ligand, DLL3 functions to suppress Notch signaling within the same cell that expresses it, providing essential negative regulation during somitogenesis [31,32]. Loss of DLL3 function would disrupt the oscillatory signaling required for proper somite formation, resulting in vertebral segmentation defects characteristic of spondylocostal dysostosis (SCDO) [31]. Components of this pathway are highly conserved among vertebrates [31]. SCDO is characterized by congenital malformations of the axial skeleton including a compressed trunk, a short neck, a protruding abdomen, and scoliosis [6,8]. As of 2021, there were 31 pathologic variants in DLL3 reported to cause SCDO type 1 in humans, including insertions, deletions, frameshift, and nonsense mutations that lead to premature truncation or protein function impairment [7,9]. SCDO type 1 is the most common form of SCDO and is usually caused by consanguinity [6,8]. While we do not have any history on these mixed-breed littermates prior to 14 weeks of age, the inbreeding coefficients identified through genetic analysis and the amount of DNA shared between siblings suggest these dogs are recently inbred.

The phenotype observed in these mixed-breed dogs closely mirrors both human SCDO type 1 patients and the pudgy mouse model. In humans, SCDO is diagnosed radiographically based on characteristic patterns of vertebral and rib malformations [6]. The radiographic findings in puppies 2 and 3, combined with the identification of a pathogenic DLL3 variant, confirm the diagnosis of spondylocostal dysostosis in these dogs. Pudgy mice exhibit shortened tails and trunks with extensive axial skeleton malformations resulting from a 4 bp deletion causing a frameshift and early truncation of DLL3 [18,19]. The mutation is autosomal recessive and heterozygotes appear phenotypically normal [18]. Radiographic similarities between pudgy mice, human SCDO patients, and the cases described in this study include the “pebble beach sign” characterized by misshapen and smooth vertebrae seen in young cases, missing ribs, proximal and distal rib fusions, hemivertebrae, and in mice and dogs, shortened and twisted tails. Analysis of pudgy mice revealed phenotypic variability arising from the same mutation, suggesting the influence of additional factors such as gestational hypoxia, maternal health during pregnancy, or other environmental factors [8,18,33]. This phenotypic variability may explain the size differences observed between puppies 2 and 3, both of whom are homozygous for the DLL3 variant. Alternatively, since they are not purebred, this could also be due to differential segregation of size variants between littermates.

The variant filtering analysis was performed under a recessive inheritance model based on the predominant inheritance pattern of SCDO. Given the severity of this phenotype and the suspected recent inbreeding origin based on high inbreeding coefficients (20–27%), despite their mixed-breed compositions, variants were filtered against the Dog 10K database, which contains phenotypically normal dogs from diverse populations. SCDO is a developmental disease that would have been apparent at any age of sampling, making it unlikely that affected individuals would be included in this database. The suspected recent inbreeding origin of this mutation further suggested it would be absent from geographically diverse dog populations. Both severely affected puppies were homozygous for the DLL3 variant, while puppy 1 was heterozygous, consistent with recessive inheritance.

DLL3 protein consists of several key structural domains, including a Delta/Serrate/Lag2 domain, which is essential for Notch receptor binding, six epidermal growth factor (EGF)-like repeats, which are necessary for DLL3 function, a single transmembrane region to anchor the protein in the cell membrane and a short cytoplasmic tail for intracellular signaling [7,9,11]. The c.650-2A>C variant disrupts the canonical splice acceptor site preceding exon 5, which encodes EGF-like domains 1–3. While the precise splicing outcome has not been experimentally validated, canonical splice acceptor site mutations commonly lead to exon skipping of the exon immediately following the splice acceptor mutation or use of cryptic splice sites, either of which is predicted to disrupt the first three EGF-like repeats. EGF-like repeats 2 and 5 are O-fucosylated in wild type DLL3 [34]. One in vivo study showed that O-fucosylation-deficient DLL3 is not functional in vivo, suggesting that O-fucosylation is essential to DLL3 function, and missense mutations to the EGF-like repeats involved in DLL3 O-fucosylation are enough to cause a loss of function [34]. Pathologic variants in DLL3 affecting EGF-like domains are well-documented causes of the SCDO phenotype in humans [9]. No strong alternative splice acceptor sites were identified within exon 5. If the weak cryptic site identified using SSPnn were utilized, this would result in deletion of 166 nucleotides from exon 5, causing a frameshift mutation and premature termination. Alternatively, if exon 5 were skipped entirely, this would also result in a frameshift and early truncation of DLL3 protein, since exon 5 is 215 nucleotides long, which represents an incomplete number of codons and therefore cannot maintain the reading frame. While in silico splice site predictors have inherent limitations in determining precise splicing outcomes, disruption of this highly conserved canonical splice acceptor site is expected to result in loss-of-function DLL3 protein, as evidenced by the SCDO phenotype in affected puppies.

Splice sites are highly conserved during evolution, and canonical splice site mutations at the −2 position are well-established as functionally significant [35]. While DLL3 expression is limited to brain and gonadal tissue after development [36,37,38], precluding direct functional validation in these cases, the convergence of conservation analysis, predicted functional impact, segregation patterns, and phenotypic concordance with human SCDO type 1 and pudgy mouse models supports classification of the splice acceptor variant in DLL3 as a loss-of-function allele and the causative mutation for SCDO in this litter. Additionally, using standardized animal variant classification guidelines, the variant meets criteria for a pathogenic classification based on its predicted null effect in a gene with established loss-of-function disease mechanism, as well as its absence from 2341 control dogs.

While DLL3 represents the most likely causative variant, the presence of a missense mutation in GPATCH1 (p.Asn388Lys) warrants consideration as a potential modifier. GPATCH1 is a component of catalytically active spliceosomes that promotes splicing fidelity through its interaction with the helicase DHX35 [39]. Given that the primary DLL3 mutation is a splice acceptor variant, impaired GPATCH1 function could theoretically exacerbate splicing defects by reducing overall splicing efficiency or cellular tolerance to splicing errors. However, the GPATCH1 variant more likely represents a neutral polymorphism that co-segregates with the causative DLL3 mutation due to linkage, given the ambiguous AlphaMissense score (0.37), the unclear relationship between known GPATCH1 mutations and human phenotypes, and its location within the 9 Mb haplotype shared by both cases. Functional studies would be required to determine whether these variants interact significantly and contribute to the SCDO phenotype. The remaining variants identified in this chromosome 1 haplotype (GGN p. Ala509Thr and SIPA1L3 p.Pro316Arg) appear to be neutral polymorphisms based on conservation analysis and pathogenicity prediction scores.

The cause of puppy 1’s truncated caudal vertebrae remains undetermined. It is possible the tail was docked after birth, but it is also possible the short tail is due to haploinsufficiency resulting from the heterozygous DLL3 mutation described, or another unidentified tail mutation. There are several known mutations for bobtails in dogs, such as a missense mutation in TBXT [40], which can be found in Australian Cattle Dogs, among other breeds, and a deletion in DVL2 [41] leading to a frameshift, but neither of these mutations is present in these littermates. Haploinsufficiency of Notch pathway genes may affect tail morphology in some carriers. In heterozygous HES7-knockout mice, 43% exhibited kinked tails [42], demonstrating incomplete penetrance of tail defects. Since both HES7 and DLL3 function in the Notch signaling pathway [15,32], haploinsufficiency affecting caudal vertebral development is plausible. However, in the miniature schnauzer HES7-associated SCDO study, three heterozygous carriers showed no external signs of vertebral malformation, though one had a docked tail that may have masked tail abnormalities [5]. Additionally, heterozygous pudgy mice have normal, long tails, suggesting that if the DLL3 splice acceptor variant causes tail truncation in canine carriers, this may represent an allele-specific effect. Further investigation of additional DLL3 carriers would be needed to determine whether this variant exhibits haploinsufficiency in dogs.

This report describes DLL3-associated spondylocostal dysostosis in dogs, expanding on the previous identification of HES7-associated SCDO in the miniature schnauzer breed [5]. The clinical presentation in these mixed-breed littermates closely mirrors human SCDO type 1, demonstrating phenotypic conservation of DLL3-mediated somitogenesis defects across mammalian species. This finding expands the known genetic basis of canine SCDO and reinforces the conservation of Notch signaling pathway disruption as the underlying mechanism for this developmental disorder.