This study was designed to use LD to exploit a possible founder effect suspected to be responsible for IE in these breeds by testing as many candidate genes as possible using microsatellite markers. Microsatellites have been shown to be a practical and useful tool in candidate gene studies [29
]. Marker locations were selected within average blocks of canine LD so that one marker's result could, in theory, accurately represent the entire gene. The use of multiple breeds allowed for the possibility of assessing whether a discovered mutation was unique to one breed, suggesting the suspected founder effect specific to that breed, or was observed in several breeds, indicating a much older mutation. While no mutations were identified to draw definitive conclusions, it is not unreasonable to speculate that in dogs the underlying genetic basis of IE may vary by breed, as it is known that the rare human monogenic IEs are typically isolated by family [30
]. These experiments incorporated both linkage analysis and association analysis, each having different strengths. While linkage analysis is the more powerful of these two statistical analysis methods for detecting Mendelian disease mutations, it requires samples from appropriate family structures. Conversely, association analysis does not require family-based samples, and it is generally considered more powerful than linkage analysis in detecting polygenic disease associations. Sequencing of the subset of human IE-associated candidate genes was not pursued for this study, as brain tissues for obtaining cDNAs for most cost efficient sequencing were not available.
Overall marker informativeness was highly influenced by sample population. The Vizslas and Beagles had excellent average heterozygosities (0.569 and 0.574, respectively), while the ESS and GSMD were much lower (0.425 and 0.341, respectively). This may be due, in part, to the sample cohorts: the ESS and GSMD were set up as discordant full- or half-sibling pairs, whereas the Vizslas were constructed as large pedigrees, and the Beagles were case/control matched pairs that did not share common ancestors to at least the grandparent level. Reduced marker heterozygosity across breeds could also be influenced by a higher degree of inbreeding within the ESS and GSMD sample cohorts in this study, and the lower average heterozygosities of these two breeds may suggest a founder effect that could eventually help uncover associations. However, marker informativeness is not entirely dependent on sample population, as evidenced by CACNA1A, which was polymorphic enough to be confidently insignificant after analyzing one or two microsatellites in ESS, GSMD, and Beagles, but remained inconclusive through three markers in the Vizslas and ultimately required a fourth microsatellite to achieve a suitable heterozygosity in this breed. The accepted marker heterozygosity of 0.3 is perhaps low, especially when analyzing only a single microsatellite for each candidate gene, however, for many of the insignificant microsatellite results in all four breeds the heterozygosity was > 0.5.
Ultimately, 16 of the 20 human epilepsy-associated candidate genes (ARX, CACNA1A, CACNA1H, CACNB4, CHRNA4, CHRNB2, CLCN2, GABRA1, GABRG2, KCNQ2, KCNQ3, LGI1, NHLRC1, SCN1A, SCN1B, and SCN2A) and both of the mouse model genes (CACNA2D2 and CACNG2) were either excluded (with linkage) or demonstrated insignificant association to IE in all four of the breed cohorts. Another two of them (KCNA1 and ME2) were insignificant in three out of four breed cohorts, remaining inconclusive only in the GSMD. One gene, GABRD, was inconclusive in two breeds (ESS and GSMD), and CHRNA2 was inconclusive in ESS and was not tested in Vizslas. Insignificant association to IE was demonstrated in most of the additional candidate genes tested. For every breed, however, there were a handful of markers that were inconclusive, due to low heterozygosity.
The few markers with potentially interesting results (uncorrected p-values of < 0.05) that were not excluded by follow-up markers, including CACNB1 in the Vizsla, CHRNB2 in the GSMD, and KCNQ3 and LGI1 in the Beagle, were of minimal interest when correcting for multiple testing. With the Bonferroni correction, a p-value of 0.05 considered as statistically significant would be lowered to 0.0025 for twenty tests, and further lowered to 0.00125 for forty tests; p-values of this magnitude were not obtained for any marker, and all breeds were tested on more than forty genes. The potential of false positives due to population stratification must also be considered for these few genes with suggestive results in the association studies, despite attempts to control for the degree of relatedness. Haplotype association analysis was not possible with this data since there were not enough closely-spaced markers to generate haplotypes. Conclusive confirmation or exclusion of these loci can be performed in the future with newer technologies such as whole-genome SNP arrays, which can generate vast amounts of data in less time.
It is possible that the populations tested were underpowered to detect association if it existed. Utilizing a sibling-pair case-control design could decrease power because the control dog may also carry the risk allele. However, this situation is difficult to avoid in highly inbred dog populations, and the sibling-pair case-control design should aid in avoiding population stratification, which can create false positive results. Although sample size requirements for canine association studies have not been precisely defined, based on average linkage disequilibrium estimates in dog breeds, a starting point of approximately 25 cases and 25 controls appears to be adequate to find statistical significance for a completely penetrant recessive trait [36
]. The number increases to approximately 50 cases and 50 controls when a trait is dominant, and for more complex traits as many as 100 dogs in each group may be necessary. Therefore, it is possible that there were not adequate numbers of dogs in the association studies, particularly if IE is not monogenic. A meta-analysis might improve power by pooling data from the three association study breeds (ESS, GSMD, and Beagle). However, this proved to be impractical because 1) many microsatellites had varying informativeness in each breed, so that pooling data would only work for a single microsatellite that was informative for a gene in all three breeds and 2) microsatellites by their very nature are highly polymorphic and there was often little overlap in alleles between breeds. Lastly, if IE has a different genetic basis in each of these breeds, pooling data across breeds seems unlikely to yield interesting results.
In the linkage analysis of Vizsla pedigrees, it is possible to conclude that these candidate gene loci are truly excluded, due to convincingly negative LOD scores at zero centiMorgans. However, if the inheritance and age dependent penetrance assumptions for the Vizsla model are incorrect, then the present linkage study is flawed. These input assumptions were based on previous study of IE in this breed [23
] and on the best information available. Conversely, in the association studies of the other three breeds, it is not possible to specifically state that the loci are "excluded", as there is some degree of doubt that one microsatellite marker is adequate to provide convincing evidence for exclusion of a locus with this type of study. The addition of more markers would help to verify these insignificant results and truly exclude a locus. That work is beyond the scope of this study and is better addressed with whole-genome SNP analysis. The latest commercial canine SNP array, with over 170,000 SNPs, became available well after these studies were initiated. Further, the excluded and insignificant genes reported in the present study may only reflect the specific breeds examined, or even lines within these breeds, and results should not necessarily be extrapolated to other breeds.
It is possible that canine IE, like the majority of human IE, is a genetically complex disease in most breeds and that multiple loci contribute to susceptibility in any given breed. The present study would very likely have detected a major contributing locus, and it is unlikely that a truly causative locus is being excluded as a false negative. In humans, the vast majority of IE remains genetically unexplained and is considered to be polygenic [37
]. A recent study by Oberbauer et al. [40
] utilized microsatellites in a genome-wide linkage scan for epilepsy loci in the Belgian shepherd dog and concluded that the disease was highly polygenic, reporting a tentative six QTLs. These results further support the conclusion that canine IE is a more complex disease than originally hypothesized. Whole-genome association analyses with SNP arrays are likely the platform of choice for further studies of IE, as they can query tens of thousands of markers simultaneously across the genome and are better able to identify multiple susceptibility loci. Additionally, copy number variants (CNVs) have been increasingly shown to be involved in neurologic disorders such as autism [41
] and schizophrenia [43
], as well as epilepsy [45
]. CNV studies of canine epilepsy may reveal this as a similar mechanism for disease in both species.
Ultimately, canine IE found to be significantly associated with DNA markers or with a mutation in a specific gene would allow genetic tests to be developed to assist dog breeders with decreasing the incidence of this disease. This would be most effective for monogenic, highly penetrant forms of IE, but if the disease proves to be genetically complex, with multiple genes contributing and less than 100% penetrance, then such a test could still possibly be used to provide a relative risk for IE development within a breed. In any event, discovery of IE associated gene loci in dogs may not only improve the understanding of canine health, but could advance the study of neurobiology and human health as well.