In Response: ME2 association analysis in adolescent‐onset genetic generalized epilepsy

Abstract

To the Editors: Dr. Sander asserts that our replication results are inconsistent, despite our reporting three separate lines of evidence: ME2 gene expression, association, and replication; all supporting our finding that ME2 (malic enzyme 2) influences genetic generalized epilepsy (GGE). In 1000 Genomes, the correlation between selected ME2 single nucleotide polymorphisms (SNPs) is high (Figure 1A). Dr. Sander assumes that the correlation should be high in our GGE cases too. On the basis of this assumption, he concludes that the evidence for association (ie, P‐values) at these selected ME2 SNPs should be the same. However, these SNPs are not all highly correlated in our GGE cases (Figure 1B). Specifically, the SNP with the smallest P‐value, rs608781, is not highly correlated (r2 = 0.48) with any of the other ME2 SNPs. Moreover, cases and controls may well have different patterns of correlation at associated SNPs.1 Therefore, our replication results are not inconsistent as Sander claims. Furthermore, our main finding, that ME2 is involved in GGE, is supported by multiple lines of evidence including differences in predicted ME2 gene expression in human brain, replication, and increased evidence from both family based and population‐based association. In his letter, Dr. Sander also implies that his failure to replicate provides evidence against ME2 s involvement in GGE. However, a fundamental pillar of statistical reasoning is that “the absence of evidence is not evidence of absence.” This means that Sander s P‐value of .17, for example, is NOT evidence that the null hypothesis is true. Misinterpreting P‐values in this way has prompted the American Statistical Association to release strict guidelines concerning the proper interpretation of P‐values.2 Given that our replication results are consistent and supported by other evidence, the more relevant question is: How do our cases differ from Sander s cases? It is likely that our GGE cases are more homogeneous than Sander s cases because our cases were rigorously phenotyped, and many had a positive family history of disease. This means that our cases are more likely to have the same basis for GGE, and if that basis is related to genetic variation at ME2, then we should (in principle) see a difference between cases and controls. By contrast, Sander s sample of GGE cases (n = 3893) likely contains increased heterogeneity due to diagnostic variation across centers and genetic differences in population substructure. This means his cases likely have GGE for different genetic reasons (ie, heterogeneity is present). As we have written elsewhere,3 heterogeneity is the most likely reason that a genetic study misses a true signal. For example, despite the high heritability of GGE (65%‐80%), the reason that GGE genes have not been convincingly found by large‐scale