In this study, the majority of densely affected families had average genotypic loads for the five common SNPs associated with AMD that would be expected on the basis of a simulation of genotypic load. This assumes that the five variants account for the known familial variance in AMD.
The ability of simulation to closely mirror the actual average genotype distribution also supports the validity of this novel analytical approach and its underlying assumptions regarding allele frequency, disease prevalence, and estimates of the effect sizes of individual risk SNPs. As gene discovery proceeds for other common diseases to the point at which a significant amount of familial variance can be explained by known genetic variation, this approach may be useful to try to understand the genetic architecture of undiscovered familial variance. This could provide a basis on which to decide whether to pursue additional common additive variants or rarer, more penetrant variants. It may also serve to identify families who are particularly likely to harbor additional, undiscovered variants. On the basis of this approach alone, however, we cannot distinguish a priori for a specific family whether the remainder of the unexplained genetic variance for AMD is because of highly penetrant loci or because of multiple common ones.
Although the genotypic load for most of our AMD families did not deviate significantly from the expected load, we did identify some families, particularly in the four out of four and four out of five affected configurations, who had a lower than expected genotypic load. There are several explanations that could explain this finding. There may be rare, more penetrant genetic variants in these families contributing to their disease. To further explore this possibility, we also performed a simulation that included an additional rare, penetrant risk allele. We show that the distribution of slopes of the mean genotypic load from the actual families vs the mean genotypic load from this additional simulation is shifted to lower values (). This is because, under these conditions, a family with a particular disease density now has a possible explanation for their increased burden of disease that does not involve the five common SNPs. If our genotypic load score recognized and accounted for the existence of this rare allele, the median slope in the distribution of slopes would once again be 1. Therefore, our current genotypic load, which only takes into account the five known common SNPs, is incomplete in its ability to explain some of the disease burden in densely affected families. One possible explanation for the difference between the observed vs simulated genotypic loads is an undiscovered, rare, high-penetrance variant.
Alternatively, these families could also share a common environmental risk factor so that the density of disease is not explained on a genetic basis. This is less likely, given that environmental risk factors were well documented as part of the selection of these families, and no strong, common, environmental risk factor that can explain these differences has been identified in these families. A third possibility is that the AMD in these families is a distinct subphenotype of AMD, with an underlying genetic architecture that is different from the AMD in the general population. However, this is unlikely, as the phenotypes observed represent the typical phenotypes seen in other AMD populations, and all families had a proband with advanced AMD. To further explore the possibility that environmental risk factors or subphenotypes could account for our findings, we compared environmental risk factors (smoking and body mass index)16, 17
and the two main subtypes of AMD, advanced dry (geographic atrophy) and wet (choroidal neovascularization) AMD, in families whose genotypic load was as expected and in families with a lower than expected genotypic load (). There is no statistically significant difference between these two groups for any of these variables. A final possibility for explaining our findings is that the assumption of a liability model or the assumption of additivity of the known loci and the polygenic portion on the liability scale may be incorrect.
Comparison of macular degeneration subtypes and environmental risk factors among families with expected genotypic load and families with genotypic loads that deviate from the expected
It should be noted that our method of selecting densely affected families is subject to ascertainment bias, because some siblings were selected as affected sibling pairs for a linkage analysis. Therefore, to design a study that was independent of the ascertainment scheme strategy, we adopted an ascertainment-free analysis approach by considering the mean genetic load according to family configuration.
There are some limitations to our study. Genotype and/or phenotype information was unavailable on some siblings primarily because of illness or death of the sibling. There is a possibility of misclassification bias. Some unaffected siblings were ascertained at a younger age than their affected proband sibling. We did, however, follow all siblings prospectively to obtain the most recent AMD grade used in this study, and all subjects had a grade assigned after an age of 60 years. It is still possible that a few of these siblings could develop AMD or progress with time. We have limited numbers of densely affected families for some of the family configurations. This limits our power to detect true differences between actual vs simulated families. Finally, we recognize that some of the reported P-values are marginal and may be due to chance.
Despite these limitations, we believe that the novel analytic approach used in this study is a valid way to determine the expected genotypic load for densely affected families for diseases in which a significant proportion of the genetic variance is already known. In our AMD families, this method allowed us to identify a subset of densely affected families who have a lower than expected genotypic load. Given that these families may harbor rare and more penetrant variants for AMD, linkage analyses targeting these families would be one way to search for potential additional implicated genes. Resequencing of the known associated genes could reveal additional implicated variants.