On the Significance of Linkage Studies of Complex Traits

To the Editor:

We read with interest the recent article by Wang et al. (2004) reporting linkage of premature myocardial infarction (MI) to a locus on 1p34-36. The authors have recruited a large sample of families with premature coronary artery disease (CAD), as detected by catheterization, revascularization, or MI. Such large-scale approaches will be crucial to the identification of genetic variation underlying complex traits, including atherosclerotic CAD and MI, the leading killer of men and women in the Western world. We commend the authors for undertaking such an important study.

Although their article reports the nominal LOD score of 11.68 for linkage of premature MI to 1p34-36 and a corrected pointwise P value of .00011, we note that the genomewide significance of the linkage statistic is not clear. We have some methodological concerns regarding the initial emphasis on results presented in their table 2 and figure 1:

• 1. The finding of 11 independent −log₁₀P values >3.13 in the multipoint linkage approach (regions identified in single-point and multipoint W2 analyses overlap significantly) raises questions regarding the assertion that such a threshold corresponds to a LOD score >2.2, as proposed by Lander and Kruglyak (1995). A LOD score >2.2 should occur once in a maximally informative genomewide scan, under the null hypothesis that no genetic linkage is present. It seems unlikely that 11 true genetic loci influencing a complex phenotype would be detected in a single study. It is more likely that the asymptotic P value statistic generated by the authors’ modified Haseman-Elston regression model is inflated.
• 2. The marked attenuation of the multipoint P value of <10⁻¹² to 10⁻⁴ on pointwise permutation testing at the 1p34-36 locus suggests that the nominal asymptotic P values are inflated. It is possible that the method of linkage analysis may have inflated the P value estimates. In particular, the treatment of the dichotomous MI phenotype as a continuous variable may not be appropriate. The assumption of equal variances required for a quantitative trait may not be valid for different numbers of affected and unaffected individuals in each family.
• 3. The authors’ attempt to correct the pointwise empirical P value estimates for the number of markers tested is quite important for establishing the significance of their findings. However, the attempt may be inadequate to account for the testing of multiple markers. The authors refer to the simulation analyses by Wiltshire et al. (2002), which explored the influence of experimental deviations from the Lander-Kruglyak assumptions of completely informative linkage analyses. We are uncertain whether the empirical genomewide P value estimates derived from the Wang et al. (2004) data correspond to the same nominal LOD-score thresholds identified in the Wiltshire et al. (2002) study using simulated data. Permutation testing of the Wang et al. (2004) data by use of the Wiltshire approach might provide greater confidence regarding the genomewide significance of the study findings, but this approach admittedly represents a significant computational burden.
• 4. Genomewide linkage analyses at 10 cM may not extract maximally the identity-by-descent information for the sample under study. A fine-mapping study at higher density across a region of interest may show a change in the maximum-LOD-score estimate. An increase in the LOD-score estimate with better information extraction might be reassuring, but a fall in the LOD score may signal a false-positive finding. We would encourage the authors to perform and publish the results of a higher-density map.
• 5. The study cohort was recruited on the basis of a composite definition of premature CAD. Was the broader CAD phenotype (including MI) the primary phenotype prespecified in the linkage analysis? The reported MI linkage analysis represents a subgroup of the subjects enrolled; the “less-restrictive” CAD phenotype was also tested and revealed no suggestive or significant linkage evidence. Could the authors clarify their original primary analysis and whether additional subgroups were analyzed? A true empirical P value would also account for the multiple phenotypes tested.

On review of the study, the declaration of a finding of genomewide significance may not be as strongly supported as suggested by the authors. The results of this linkage analysis do not contain much overlap with those of similar analyses, and this certainly could result from differences in phenotype definition, environmental exposures, or study design (Pajukanta et al. 2000; Francke et al. 2001; Broeckel et al. 2002; Harrap et al. 2002; Chiodini and Lewis 2003). Replication of linkage analyses for complex cardiovascular traits has often proven challenging, and the difficulty in achieving replication for MI underscores the many difficulties in the conduct and interpretation of such linkage analyses (Altmuller et al. 2001).

Identifying genetic factors underlying linkage peaks in this and related studies of MI will require considerable expenditure of resources and should proceed on the basis of the strongest possible evidence. We encourage the systematic comparison of available and accruing linkage data across studies in various CAD phenotypes, including continued assessment of the most appropriate linkage methods.