Association Studies in Asthma Genetics

Scott T. Weiss, M.S.

Channing Laboratory, Brigham and Women's Hospital, Harvard Medical School, Boston, Massachusetts

	ARTICLE

TOP ARTICLE REFERENCES

In this issue of the AJRCCM (pp. 2036-2044), Hakonarson and colleagues present a case-control association study of asthma and its associated phenotypes with 24 candidate genes for asthma (1). This study design has become the most popular design in asthma genetics because it is easy to implement. It is, however, difficult to implement well. My laboratory has spent time considering what goes into performing a good genetic association study (2, 3). Table E1 in the online data supplement to this editorial provides a list of 10 specific issues that should be addressed in any case-control association study of a candidate gene. It is worthwhile evaluating this particular article with regard to these criteria.

Could the lack of significant association in the article by Hakonarson and coworkers (1) be due to population stratification? Probably not. Population stratification can be thought of as confounding by ethnicity. Differing allele frequencies in different ethnic groups could lead to spurious associations or lack thereof by creating different (or the same) allele frequencies in case and control subjects. The ethnic homogeneity of the Icelandic population would make this explanation unlikely (4). Confirmation of ethnic homogeneity could be pursued by typing random markers across the genome (5).

Is there something unique about the environment in Iceland that could explain these results? Unique effects of these genes or interactions with specific environmental exposures, that is, population differences, are always a possibility that is difficult to rule out completely. It would appear that Iceland is similar to Northern Europe with regard to its environmental exposures and disease prevalence (6), suggesting that this is unlikely.

One of the strengths of the article by Hakonarson and coworkers (1) is the defined universe of patients used as cases. The 185,000 patients initially identified belonged to 409 family clusters and each patient was related to at least one other within four meioses (i.e., four meiotic events separating a proband from a first cousin). The 94 patients selected from the 409 available family clusters were not selected from the same families to avoid familial correlation in the case-control analysis. Phenotypic information was available for a doctor's diagnosis of asthma, skin test reactivity to 12 aeroallergens, total IgE levels, pulmonary function tests, and a methacholine challenge test. Whereas the case subjects were carefully assessed both phenotypically and from a disease perspective, the control subjects were much less well assessed. This is frequently a problem in case-control association studies. In this particular study, the 94 control subjects were selected from more than 300 healthy subjects who donated blood to serve as control subjects for other association studies. Although these individuals completed a questionnaire about asthma, they did not undergo assessment of intermediate phenotypes such as allergy skin testing and methacholine challenge testing. Misclassification could have occurred in this study because the control subjects did not undergo the same degree of rigorous phenotypic assessment as did the case subjects. In addition, control selection may be subject to selection bias because of the possibility that health concerns may have influenced participation. Both of these factors could potentially diminish or abolish a true association. A more rigorous study design would have been to select case and control subjects from within an ongoing cohort study. This would provide the same degree of phenotypic assessment of both case and control subjects and would provide a greater sense that selection bias and information bias were not operating in the study.

A second concern for case-control studies is sample size. The 94 case subjects and 94 control subjects described in the article by Hakonarson and coworkers (1) provide only moderate power to identify allelic association with asthma. With an allele frequency of 25%, the power for such a study would be only 80% for a risk of 2-fold or greater. The power would fall off dramatically for allele frequencies less than 20% with this sample size. Power in genetic association studies is a function of three things: the strength of the association or degree of difference between the case and control subjects, the allele frequency, and the sample size. In this case, power is only modest at best and, hence, this study is biased in the direction of null results or no effect.

In a nongenetic case-control association study, the interpretation of a putative association is dependent on the association being causal (i.e., there is a true relationship between the risk factor and the disease), being due to chance (which is assessed by the p value), or being due to bias (usually confounding or the type of selection bias discussed above in relationship to the control subjects). In a genetic association study, additional possibilities emerge. Although it is true that the allele itself may be functional and directly affect the expression of the phenotype, a more likely event is that the allele is in linkage disequilibrium with another allele at another nearby locus that is the true causal allele.

Linkage disequilibrium occurs because of the tendency of alleles of closely linked loci to segregate together with an increased frequency across a population. Linkage disequilibrium or, more properly, allelic association is a function of evolutionary history and population genetics and may be due to mutations arising in a population, natural selection, genetic drift, or population admixture. Haplotypes are a set of related alleles within a particular genetic region on a chromosome that share a similar evolutionary history. Hakonarson and coworkers (1) genotyped 42 single-nucleotide polymorphisms (SNPs) in the 24 genes or, roughly, 2 SNPs per gene. Although some of these SNPs were in linkage disequilibrium, many were not. Given that the pattern of linkage disequilibrium varies across the genome, and that relatively few SNPs per gene were genotyped, it is possible that most of the SNPs were not in linkage disequilibrium with other SNPs in the genes of interest. This might create a false picture of no association because the linkage disequilibrium pattern within the genes and regions of interest was inadequately characterized.

Another unique feature of the study by Hakonarson and coworkers is the use of linkage subsequent to the candidate gene association study. This is the opposite of what is usually done in human genetics of complex traits, where the traditional approach to the identification of new genes is to perform linkage analysis first, identify a candidate region of interest, and then explore candidate genes within that linked region. Association analyses are extremely useful for localizing susceptibility loci within a region of linkage. Despite the carefully selected family data with information about 269 relatives from the 94 Icelandic families, no evidence of linkage to chromosomal region 5q, 12q, 11q, or 3p was noted in this population. This is at odds with much of the published linkage work in asthma, where several studies have demonstrated linkage to asthma phenotypes in these regions (7, 8). Again, inadequate statistical power may explain these negative results.

It is important to recognize that the negative results observed by Hakonarson and coworkers are atypical of reported association studies, among which false-positive results are frequent and by far and away the more common problem in the literature. Obviously the results presented here may be truly negative. Replication with a family-based design would help confirm this. Careful attention to the study design issues outlined in Table E1 in the online data supplement to this editorial will help to minimize both false-positive and false-negative results in this rapidly evolving field.

	Footnotes

	References

TOP ARTICLE REFERENCES

1. Hakonarson H, Bjornsdottir US, Ostermann E, Arnason T, Adalsteinsdottir AE, Halapi E, Shkolny D, Kristjansson K, Gudnadottir SA, Frigge ML, Gislason D, Gislason T, Kong A, Gulcher J, Stefansson K. Allelic frequencies and patterns of single-nucleotide polymorphisms in candidate genes for asthma and atopy in Iceland. Am J Respir Crit Care Med 2001; 164: 2036-2044 [Abstract/Free Full Text].

2. Silverman EK, Palmer LJ. Case-control association studies for the genetics of complex respiratory diseases. Am J Respir Cell Mol Biol 2000; 22: 645-648 [Free Full Text].

3. Weiss ST, Silverman EK, Palmer LJ. Case-control association studies in pharmacogenetics (editorial). Pharmacogenetics J. (In press)

4. Helgason A, Siguroardottir S, Gulcher JR, Ward R, Stefansson K. mtDNA and the origin of the Icelanders: deciphering signals of recent population. Am J Hum Genet 2000; 66: 999-1016 [Medline].

5. Pritchard JK, Rosenberg NA. Use of unlinked genetic markers to detect population stratification in association studies. Am J Hum Genet 1999; 65: 220-228 [Medline].

6. European Community Respiratory Health Survey Group. Genes for asthma? An analysis of the European Community Respiratory Health Survey. Am J Respir Crit Care Med 1997;156:1773-1780.

7. Cookson WO, Moffatt MF. Genetics of asthma and allergic disease. Hum Mol Genet 2000; 9: 2359-2364 [Abstract/Free Full Text].

8. Hakonarson H, Wjst M. Current concepts on the genetics of asthma. Curr Opin Pediatr 2001; 13: 267-277 [Medline].