A call for accurate phenotype definition in the study of complex disorders

To the editor:

Gretarsdottir et al. recently reported on the association between risk of ischemic stroke and the gene encoding phosphodiesterase 4D (PDE4D)¹. There was no significant association with ischemic stroke as a whole¹, and so the authors focused on various etiological subgroups of stroke. They found that different haplotypes were associated with different risk of combined carotid and cardiogenic strokes. Their work warrants serious criticism. There is an important methodological problem relating to combining individuals with cardiogenic and carotid stroke into one subgroup for genetic analysis. Combined carotid and cardiogenic strokes represented 40–45% of all ischemic strokes in the authors' sample^1,2, which means that they excluded 55–60% of individuals with stroke from the genetic association study. Excluding that many individuals (and losing that much genetic information) is appropriate only if it refines the stroke phenotype in the remaining sample. Unfortunately, this was not the case in the study of Gretarsdottir et al.

The authors claim that cardiogenic and carotid strokes are most clearly related to atherosclerosis. This is true for carotid strokes (14–16% of all ischemic strokes in their population^1,2) but not for cardiogenic strokes (24–29% of total). Data from the literature show that only 25% of cardiogenic strokes directly result from complications of coronary atherosclerosis³, whereas ∼65% are due to valvular heart disease or atrial fibrillation (with less than 20% of all atrial fibrillations being attributable to coronary heart disease⁴). The combined subgroup resulting from the admixture of carotid and cardiogenic strokes is, in fact, highly heterogeneous and, in our opinion, unsuitable for genetic analysis. Readers should receive with caution the authors' assumptions regarding the putative role of PDE4D in atherogenesis and subsequent risk of stroke. In agreement with this caveat, we note that PDE4D genotypes or haplotypes do not influence the risk of peripheral artery occlusive disease or coronary artery disease (see Supplementary Table 7 from ref. 1), although atherosclerosis is the cause of both disorders in more than 90% of cases.

The genetic analysis of the whole group of individuals with ischemic stroke showed a possible though nonsignificant association of anonymous markers at the PDE4D locus with stroke. In this context, the additional finding of downregulation of several PDE4D mRNA isoforms in individuals with ischemic stroke further suggests that PDE4D is a very good candidate for genetic modulation of ischemic stroke risk. Gretarsdottir et al. only considered the possible involvement of PDE4D in atherogenesis, which is supported by few experimental data¹. By contrast, phosphodiesterase 4 has repeatedly been reported to be involved in susceptibility to ischemic brain damage in animal models^5,6,7,8. Such a mechanism directly related to cerebral ischemia could explain why the PDE4D locus was found to be in linkage with stroke regardless of cause in the Icelandic population².

It seems that the PDE4D functional variant that influences the risk of ischemic stroke has yet to be identified, as statistically significant associations were found only in a pathophysiologically inadequate subgroup of individuals in the present study. Genotyping studies are becoming more comprehensive and including growing numbers of polymorphisms in each study. Accurate phenotype definition seems more than ever to be a necessary prerequisite to establish reliable genotype-phenotype relationships in the study of complex disorders.

See Reply to “A call for accurate phenotype definition in the study of complex disorders” by Gulcher et al.