A predictive model for canine dilated cardiomyopathy—a meta-analysis of Doberman Pinscher data

Introduction

Dilated cardiomyopathy (DCM) is a prevalent and often fatal disease requiring clinical management in humans and dogs (Egenvall, Bonnett & Häggström, 2006; Hershberger, Morales & Siegfried, 2010). DCM is the second most common cardiac disease in dogs and is characterised by ventricular chamber enlargement and systolic dysfunction which often leads to congestive heart failure (Egenvall, Bonnett & Häggström, 2006). The aetiology of DCM is complex. Genetic factors, myocardial ischemia, hypertension, toxins, infections and metabolic defects have been implicated (McNally, Golbus & Puckelwartz, 2013). To date, mutations in over 50 genes have been associated with DCM in humans; however, mutations in the most prevalent DCM related genes only account for approximately 50% of patients with DCM (Posafalvi et al., 2013). In human DCM genetic testing where a panel of approximately 50 loci are tested concurrently, often more than one locus can be implicated in the disease (McNally, Golbus & Puckelwartz, 2013), suggesting multiple genetic factors cooperate in DCM aetiology.

Canine DCM is phenotypically similar to human DCM (Shinbane et al., 1997). As outlined below, to date mutations in only two genes (PDK4 and STRN) and a single nucleotide polymorphism (SNP) on chromosome 5 have been associated with canine DCM (Mausberg et al., 2011; Meurs et al., 2012; Meurs et al., 2013), suggesting additional genetic causes remain unknown. While canine studies have sometimes been limited by small sample size (typically less than 10 individuals), those studies with larger sample numbers (greater than 50 individuals) have also frequently failed to find significant associations with DCM (e.g., Philipp et al., 2007; Philipp, Vollmar & Distl, 2008; Wiersma et al., 2008). One possible explanation for the challenges in identifying DCM associated loci in humans and dogs is that even within an extended family or breed, genetic variation at a single locus cannot explain the development of DCM. Indeed dog breeds can be considered as large families, with dogs within a breed more related to each other than dogs of other breeds (Parker et al., 2004). In the same way that some human families are affected by DCM, a subset of dog breeds are affected by DCM more frequently than others (Egenvall, Bonnett & Häggström, 2006). Dobermans Pinschers (hereafter Dobermans) are particularly affected by DCM, with both a high prevalence (58.2% in European Dobermans) and severity with DCM associated death often occurring within 8 weeks of diagnosis (Calvert et al., 1997; Wess et al., 2010). In dogs, diagnosis is usually at the onset of clinical symptoms of heart failure. But there is an extended pre-clinical phase, during which if treatment can be effective by prolonging the onset of heart failure (Summerfield et al., 2012). In this phase left ventricular dilation and dysfunction begins, and can be accompanied by ventricular premature complexes (Singletary et al., 2012), Median life expectancy of DCM affected European Dobermans is 7.8 years, compared with 11 years for unaffected European Dobermans (Proschowsky, Rugbjerg & Ersbøll, 2003; Egenvall, Bonnett & Häggström, 2006). A deletion in a splice site of the PDK4 gene (Meurs et al., 2012) and a SNP on chromosome 5 (Mausberg et al., 2011) in Dobermans are two of only three canine DCM mutations identified to date. While these two loci are associated with Doberman DCM, individually neither locus explains all cases of Doberman DCM (Mausberg et al., 2011; Meurs et al., 2012). Individuals heterozygous at the Chr5 SNP are more likely to develop DCM, but there are many DCM cases homozygous for the healthy allele (Mausberg et al., 2011). While PDK4 genotypes are less definite predictors of DCM, with both affected and unaffected individuals possessing the three possible genotypes, the 16bp PDK4 splice site deletion is found more frequently in North American Dobermans with DCM than those without DCM (Meurs et al., 2012). However an analysis of European Dobermans failed to identify an association between PDK4 and DCM (Owczarek-Lipska et al., 2013), suggesting additional unknown factors influence the effect of PDK4 in predisposing individuals to DCM. Thus novel genetic causes of canine DCM remain to be identified (Mausberg et al., 2011; Philipp et al., 2012).

In this study we developed genetic models to test the influence of unknown genetic factors to predict which DCM-associated genotype combinations are likely to develop DCM. Using this method, we provide evidence for a sex-linked genetic influence on known DCM loci in the pathogenesis of canine DCM. Our study is the first to propose a multigenic contribution to canine DCM.

Materials and Methods

Model development

A literature search of the Pubmed and Web of Science database using the following search terms: “Doberman DCM loci,” “Doberman Dilated Cardiomyopathy loci,” “Doberman DCM gene,” “Doberman Dilated Cardiomyopathy gene,” “Doberman DCM locus” and “Doberman Dilated Cardiomyopathy locus” (Fig. 1) identified thirty unique records. This search identified two loci associated with DCM in North American Dobermans (Meurs et al., 2012; Mausberg et al., 2011). By combining the genotypes from the identified Doberman DCM associated loci, and additional putative loci, predictive models were developed and tested against observed DCM incidence data. All genotype combinations for the DCM associated SNP identified on chromosome 5 (TIGRP2P73097:CFA5:g.53,941,386T>C, CanFam2.1) (Mausberg et al., 2011) and the PDK4 (GeneID:482310) splice site deletion (CFA14:g.20,829,667_20,829,682del, CanFam3.1) (Meurs et al., 2012) were determined. Further analysis determined which genotype combinations were likely to lead to DCM. Some genotypes are definitive; all individuals homozygous for the susceptibility allele at CFA5:g.53,941,386T>C develop DCM (Mausberg et al., 2011).

Results

Following the constraints stated in the methods and using biologically feasible reasoning each model was optimised to best fit the observed data. For each model the genotype-phenotype decision descriptions are shown in Table 4. Tables of each model are in Supplemental Information.

Discussion

This study used publicly available data to test the prediction that genetic models incorporating multiple factors can better explain and predict the incidence of canine DCM than those utilising a single factor. Until now, the possibility that multiple genes combine to influence DCM phenotype has been proposed, but has not yet been tested, despite an established role for multiple loci in related diseases (Ingles et al., 2005; Xu et al., 2010; Rampersaud et al., 2011; Posafalvi et al., 2013). This is the first study to investigate the combined effect of multiple factors on the predisposition to DCM. Although our models do not explain all cases of canine DCM, by combining three factors (PDK4, Chr5 TIGRP2P73097 SNP and an X-linked locus) we show that DCM incidence can be more accurately predicted (Tables 6–12). Furthermore, as noted above the PDK4 splice site deletion is not significantly associated with DCM in the European population. But in the model incorporating only the two known loci, the PDK4 variant improves the odds ratio for the Chr5 SNP. Collectively these findings indicate that models incorporating multiple factors are more effective than those incorporating a single factor. This result is important because it has implications for future studies of the genetics and management of DCM. A better understanding of the genetic basis of DCM will permit the monitoring and earlier clinical intervention of high risk individuals thus potentially improving the outcome for affected individuals.

To assess the accuracy of each model, we performed several statistical tests. For any model to be considered an accurate representation of observed data it should predict similar numbers of affected and unaffected individuals at each genotype as have been reported in the published data. It should also predict a similar DCM frequency to that found in the population. Secondly, the odds ratios of genotypes and alleles should support an association of the specific variants with DCM. The models incorporating the two known DCM loci and an additional X-linked locus with a susceptible allele frequency of 0.46 for the novel susceptible allele met all such conditions. It is important to note that this susceptible allele frequency should have been identified by the previous GWAS studies (Mausberg et al., 2011; Meurs et al., 2012). It is therefore possible that additional cases and controls are required to complete a comprehensive GWAS analysis of DCM in Dobermans to establish the function and frequency of this predicted DCM associated locus.

Most predictive models are based on either known or simulated genotypes at multiple loci (Janssens et al., 2006; Pencina, D’Agostino & Vasan, 2008). Such models do not account for known effects of genotypes or allow the inclusion of additional as yet unknown, loci. For example, in this study all individuals possessing the Chr5 CC genotype have DCM. Our methodology is unique and useful where there are multiple known and unknown factors which do not fully account for the phenotype. In particular, our approach accommodates specific gene combinations to lead to disease, rather than incremental risk factors as is the case in other predictive models (Janssens et al., 2006; Pencina, D’Agostino & Vasan, 2008). Limitations to our methodology include the number of factors that can be modelled is limited by the available data. Despite this, our methodology could be used in other situations. While many phenotypes are the consequence of multiple loci, there can be some loci which have comparatively more important contribution to the phenotype (e.g., Strange et al., 2011; Papa et al., 2013). Identifying these loci can be the first steps in predicting phenotypes (e.g., Hayes et al., 2010; Papa et al., 2013). Following the identification of loci associated with a trait, our methodology can be used to indicate what type of additional loci may be influencing the trait of interest, which may simplify the identification of additional loci.

Conclusions

There are many unknown factors involved in the aetiology of canine and human DCM. In Dobermans, we have identified multigenic effects and a possible X-linked locus as novel variables influencing DCM risk. While the PDK4 splice site deletion and the Chr5 SNP have both been tested for association with DCM in the European population of Dobermans, the combined genotype of individuals has not yet been considered (Mausberg et al., 2011; Owczarek-Lipska et al., 2013). Our model would benefit from further genotyping of Dobermans at both the PDK4 and Chr5 variants to further validate the model. Future work is also required to identify X-linked DCM loci if the model is verified for the known loci. It is also possible that the different combinations of alleles leading to DCM in the model could affect the time taken to progress from one disease stage to the next as reported by Wess et al. (2010). If validated, our model has implications for current canine breeding practices and welfare of individuals within the breed. Individuals with allele combinations more likely to develop DCM can be monitored more intensely than those with less genetic risk, and mating pairs resulting in deleterious genotypes can be avoided. This will have improve welfare by reducing the prevalence of DCM-associated alleles within the population and potentially improving the longevity of affected dogs by enabling monitoring and earlier clinical management. By utilising similar methodology, equivalent multigenic effects and possible additional loci could be identified in human DCM, giving similar benefits to those described for Dobermans.

Supplemental Information