Genetics of complex human diseases: genome screening,

association studies and fine mapping

J. XU, D. G. WIESCH and D. A. MEYERS

Center for the Genetics of Asthma and Complex Diseases, University of Maryland School of Medicine, Baltimore,Maryland, USA

Summary

Positional cloning has been applied successfully to many Mendelian disorders. Because of the publichealth significance, there is strong interest in mapping susceptibility genes for common disorders,such as asthma and allergy, that have a genetic component. Genome-wide screening has been veryuseful in detecting regions of the genome likely to contain susceptibility genes. There are multiplechromosomal regions implicated in asthma and now the difficult process of finding the genes andrelevant mutations is underway. Two approaches that are being utilized are those of associationstudies in candidate genes, and haplotype sharing or identical by descent (IBD) mapping. Althoughthese are useful approaches, it is important to realize the strengths and limitations of each. The levelof significance needed for an initial study or a replication study should be considered in light of theprior evidence for studying a specific gene polymorphism. Haplotype-sharing approaches, althoughdifficult to use in outbred heterogeneous populations, may provide important insight into finemapping and gene localization.

Introduction

Genetic studies of common complex disorders withmajor public health significance are of great interest tomany investigators [1]. Family ascertainment methodsare available and have been used successfully to collectsamples of families with asthma for linkage studies [2±4].Several areas of the genome have been shown to containsusceptibility loci based on evidence from linkagestudies. This manuscript will focus on the next step ofthe process: fine mapping and delineation of relevantsequence variants in susceptibility genes for the devel-opment of asthma and atopy.

Genome screening

Genome-wide screening using highly polymorphic mar-kers typed thoughout the genome is very useful indelineating chromosomal locations of susceptibility loci.Several genome-wide scans of families characterized for

asthma and atopy have been performed [3±5]. Severalregions of the genome have been seen in multiple studiesand in different ethnic populations including chromo-some 5q, 6p, 11p, 12q, 13 and 14. The majorsusceptibility regions have been identified althoughthere are a few additional regions that need replication.However, genome-wide screening of families withasthma has been successful in several populations. Thisis true for several other common disorders with aheritable component. The next step that has not yet beenaddressed is the actual step of positional cloning, findingthe susceptibility genes and the relevant mutations orsequence variants.

Fine mapping

Fine mapping is more difficult for common complexdisorders due to the presence of multiple susceptibilityloci, unclear models of inheritance, incomplete pene-trance and phenocopies [1]. In Mendelian diseases, thegenomic region can be narrowed using information fromgene recombinants and linkage disequilibrium. How-ever, this is more difficult in complex disorders since keyrecombinants may be due to phenocopies (affected

Clinical and Experimental Allergy, 1998, Volume 28, Supplement 5, pages 1±5

AhedBhedChedDhedRefmarkerFigmarkerTablemarkerRef endRef start

# 1998 Blackwell Science Ltd 1

Correspondence: D. A Meyers, Professor of Pediatrics, Center for the

Genetics of Asthma and Complex Diseases, University of Maryland

School of Medicine, Suite 119, 108 N Greene St, Baltimore, MD

21201, USA.

Paper 024 Disc

family members who do not have the gene of interest)and linkage disequilibrium may be difficult to detect dueto genetic heterogeneity. Approaches to fine mapping incomplex diseases tended to be concentrated in the areasof association studies of candidate genes and haplotype-sharing approaches (Fig. 1).

Association studies of candidate genes

A sequence variant is associated with a disease if itoccurs at a significantly higher frequency among affectedcompared to control individuals. It is well known thatassociation can occur in three different situations: (1)the sequence variant increases susceptibility to thedisease; (2) the sequence variant does not increasesusceptibility to the disease, but is in linkage disequili-brium with the actual mutation; and (3) is due to popu-lation stratification. In the past several years, progresshas been made to develop approaches and study designsthat can detect association while guarding against theartifact of population admixture, such as the transmis-sion disequilibrium test (TDT) in young, isolatedpopulations [6,7]. There have been many applicationsof these approaches to the complex diseases such asasthma, diabetes, and psychiatric disorders. However,little attention has been paid to several fundamental andpractical questions: What is the important assumptionof association and under what circumstances can as-sociation be detected? What are the criteria for declaringa `significant association'? How do we use associationapproaches in fine mapping of complex diseases?

The basic requirements for association The two keyrequirements of observing an association of a sequencevariant (a mutation or a polymorphism of a gene orDNA marker) with increased susceptibility to the diseaseare that the sequence variant has large effect and high

frequency in affected individuals (less locus and allelicheterogeneity). When both requirements are met, theassociation can be observed among unrelated subjects,even in an outbred population, using either case-controlstudies or TDT trio studies. No founder effect is requiredto detect the association; however, a founder effect ishelpful, as well as other approaches to increase allelichomogeneity. The mutation may be sporadic or inherited.

One way to measure the effect of the sequence variantis the genotypic risk ratio (GRR), which is defined as theincreased chance that an individual with a particulargenotype has the disease. GRR is different from thecommon measurement of risk ratio (l) since GRR refers toa specific sequence variant while l refers to the combina-tion of all involved genes. An association is likely to beobserved when the GRR is large and when the sequencevariant is frequent. For example, Risch and Merikagas[8] estimated that a sample size of 185 affected siblingpairs are needed with a GRR of 4 and a sequencevariant frequency of 5% (using TDT for sib-pairs) toobtain 80% power to detect association. However, witha GRR of 2 or less, the sample sizes are generally beyondreach (well over 2000 sib-pairs). It is difficult to obtain agood estimate of GRR before the sequence variant(mutation or polymorphism) is studied although estimatesmay be obtained from genetic models of the disorder.

Two examples will be used to illustrate the two keyrequirements for such an association study. A strongallelic association between APOE e4 allele and suscept-ibility to Alzheimer disease has been reported manytimes [9±11]. A significantly higher frequency of thee4 allele is observed in late-onset familial cases (about42±50%), autopsy-confirmed cases (about 40%), sporadiccases (about 34±38%), compared to the unaffectedcontrol group (about 14±16%). The association hasbeen observed in many populations including Cauca-sians, Japanese, African Americans, and Hispanics. Theestimated GRRs for individuals who are heterozygousor homozygous for the e4 allele are approximately 5 and18, respectively. The frequency of e4 is high enough todetect the association. The other example is the breastcancer genes. Although individuals with specific muta-tions of BRCA1 certainly have a higher risk of developingbreast cancer, the low frequency of a specific mutation(there are many different mutations of BRCA1 [12])makes it impossible to observe an association in thegeneral population of women with breast cancer.

More requirements are necessary to observe associa-tion of a marker allele that is only due to linkagedisequilibrium with the actual mutation. Besides the twopreviously described requirements, a strong foundereffect is necessary. Linkage disequilibrium occurs whena proportion of affected individuals are descendants of aFig. 1. Approaches to fine mapping.

# 1998 Blackwell Science Ltd, Clinical and Experimental Allergy, 28, Supplement 5, 1±5

2 J. Xu, D. G. Wiesch and D. A. Meyers

Paper 024 Disc

common affected ancestor. Insufficient recombinantevents between the susceptibility gene and nearbymarkers lead to over-representation of the chromosomalregion containing the gene among the affected indivi-duals. The length of the region depends on the numberof generations. When the number of generations is large,the region that is in linkage disequilibrium becomessmall enough to pinpoint the location of the gene; this isthe reasoning behind linkage disequilibrium mapping(LD mapping) [7]. Obviously, a founder effect is thecritical requirement for the association. Even with a highGRR, it is difficult to find an association with nearbymarker alleles if there is no founder effect. Differentfounders can introduce different sequence variants atnearby markers. The association studies of Alzheimerdisease and the loci near APOE illustrate this point. Onestudy reported a sequence variant of APOCII (which ispart of gene cluster of APOE, APOCI, and an APOCIpseudogene that spans about 40 kb) in linkage dis-equilibrium with Alzheimer disease in Volga Germankindreds [13]. These German families immigrated to theVolga River region of Russia in the 1760s. However, thisassociation has not been observed in other populations[14,15]. The founder effect in these kindreds made itpossible to observe the association.

There have been several successful association studiesfor Mendelian diseases in isolated populations, whichhave led to the identification of the disease gene.However, so far there is little or no comparable evidencefor complex diseases. Either the low values of GRR fora specific mutation or the low frequency of the specificsequence variant in the study population due to locusand allelic heterogeneity makes it very difficult toobserve an association in complex diseases.

The statistical significance of an association and replica-tion The statistical significance of a postulated asso-ciation receives less attention than it should in thecommunity of gene mapping of complex diseases. It isrelatively common to see a reported association betweena complex disease and a marker sequence variant or apolymorphism of a candidate gene, with a P-value of0.001 in a small sample. The false-positive rate of thesereported associations is not negligible. In the linkageanalyses, it is well accepted to use stringent criteria, forexample a LOD score of 3+ or a P-value of 0.00002[16,17], to declare a positive linkage. These criteria arechosen to guard against false-positives due to the lowprior odds for linkage for any chosen pair of loci andmultiple tests. In association studies, these two problemsare very important and should not be ignored. The priorodds for association between a disease and a specificsequence variant of a randomly selected locus are much

smaller than that for linkage for several reasons.Heterogeneity in association studies comes from twosources, both locus and allelic heterogeneity, whileallelic heterogeneity does not affect linkage studies. Inaddition to allelic heterogeneity, a founder effect isrequired in association studies using LD mapping whileit is not necessary for linkage studies. Furthermore, themarker resolution required for association studies ismuch denser, compared to that of linkage studies. In thelinkage studies, we normally depend on recombinationevents in several generations within the families, thus a10 cM marker resolution is usually sufficient. In LDmapping, we depend on recombination events in manygenerations, thus a marker resolution of less than 1 cM isnecessary. These characteristics of association studiesdetermine the large number of independent associationtests, because of locus and allelic heterogeneity.

Because the prior likelihood is so low for association,Crowe [18] pointed out that the significance levelrequired for having only a 5% false-positive rate is aP-value less than 0.00001. Risch and Merikangas [8]proposed statistical significant for association studiesfrom another point of view. Assume there are 100 000genes in the human genome, each with five diallelicpolymorphisms. The total number of independent testsneeded for association studies are 106105=106. Themultiple tests will lead to an inflation of type I error. Inorder to have fewer than 5% genome-wide false-positives, the nominal P-value should be 56108. Thisis equivalent to the false-positives using a LOD score of3.0 in the genome-wide linkage studies [17].

One can argue that for most association studies oneonly performs a limited number of independent tests,usually for those biologically meaningful candidate genesand markers in the region supported by linkage studies.This commonly used approach should increase the priorprobability for association and decrease the number oftests. This is a difficult practical issue and answers aredifferent from statisticians and geneticists. However, weshould realize the potential problem of choosingcandidate genes. It is not possible to make a strong casefor any particular gene being the major cause of asthmaand allergy. Multiple genes appear to be involved in theregulation of asthma and allergy associated traits. Manygenes continue to be identified and our knowledge on the`genetic architecture' of complex diseases advancesquickly. For each candidate gene that is proposed, thereare likely to be multiple other genes that, a priori, areequally likely candidates but have not been identified.

It is worthwhile to point out, following the sameargument of prior probability, that we should notpenalize ourselves too much in association studies forthose candidate genes that are proven to be involved in

# 1998 Blackwell Science Ltd, Clinical and Experimental Allergy, 28, Supplement 5, 1±5

Genetics of complex diseases 3

Paper 024 Disc

the regulation of disease, and for which a sequencevariant with defined function has been described. Weprobably should use less stringent criteria to declareassociation in these cases.

It is practically very difficult to propose a unifiedstandard to report a significant association. Thus, thereplication of previously reported association becomesvery important. Most researchers accept consistentreplication as the best evidence for a true association.However, there are two major problems in replicationstudies for association. First, the association, especiallythe one that is due to linkage disequilibrium, is usuallyobserved in very strict circumstances. They are limitedto a specific allele of a specific marker, in an isolatedpopulation of specific ethnic groups, and for a specificphenotype. The failed replication in other ethnic groupsshould not pose any negative impact on the originalreport, if they did not meet the same conditions as thefirst report. Second, it is common to see the phenotypedefinition `drift' in the replication of association studies.As various studies attempt to replicate an originalfinding, each slightly revises the definition of thedisorder based on the maximum association found inthe new study. An example would be the first associa-tion found between the asthma phenotype to a specificpolymorphism `a' for a candidate gene. The secondreported did not find the same association, instead, aP-value of 0.04 was found between the atopic phenotypeand the polymorphism `a' of the candidate gene. Thesetype of results tend to be reported as a confirmatoryassociation study. One may argue that this is acceptablebecause asthma and atopy are correlated. However, in astrict sense, each of these refinements in the definition ofdisease is in fact a new post hoc hypothesis and notstrictly a replication of the original findings. Thestatistical significance requirements for the replicationand the post hoc hypotheses are different.

Haplotype sharing: IBD walking

Isolated populations such as some areas in Finland andThe Netherlands are ideal for association studies. It wasassumed that these populations were founded over 100generations ago, thus the extent of linkage disequili-brium between the disease and the surrounding marker(about 1 cM) is small enough to be meaningful and largeenough to be practically observed. The number ofaffected ancestors may be small in these populations,which increases the chance that a large proportion ofaffected individuals in the population are descendentsfrom a common affected ancestor. On the other hand, inthe newly founded isolated populations such as thePennsylvania Amish, linkage disequilibrium is easier to

observe. However, since only 11 or 12 generations haveoccurred since the original founding of the population, theshared chromosomal regions are likely to be too large tobe used to clone the gene.

Association should be more likely to be observed in asubgroup of samples that showed linkage to the regionof interest. It is still possible that families who showedlinkage in the same region may carry different suscept-ibility genes, but the chance of locus heterogeneity willbe smaller. The affected individuals who have the sameallele as the affected ancestor at a marker near the geneshould be at least consistent with the linkage for themarker in families. It is thus very useful to have familydata before undertaking association studies.

Recently, IBD mapping, which is a haplotype-sharingapproach, has been proposed as a useful approach tomap position of genes in a founder population [19]. IBDmapping uses haplotype sharing at several markersrather than difference in allele frequencies at individualmarkers to identify regions of interest. IBD mappingdepends on the overlap of haplotypes in affectedindividuals. Consistent overlap occurs if, and only if,the disease allele originates from a common founder.However, one has to make sure the haplotype sharing isIBD, not identical by state. This can be guarded by usinga very dense resolution of markers in the region. Thereare several advantages of this approach. First, it utilizesthe linkage data but extends its usage. IBD mapping canimmediately follow a linkage study where haplotypeinformation can be obtained. After identifying a sharehaplotype, more markers inside the region can be typedto confirm they are IBD. Second, locus heterogeneityand allelic heterogeneity is not as detrimental for IBDmapping. As long as there are overlaps in the severalindependent haplotypes affected, they will be identified.The affected individuals with disorders caused by othergenes or from other founders or environmental risks willnot seriously prevent the success of IBD mapping,although they will decrease the power.

A partial reanalysis of a breast cancer mutation studyin The Netherlands by Petrij-Bosch et al. [20] presents agood example to illustrate association and IBD mappingin complex diseases. They identified a mutation of510-bp deletion in exon 22 of BRAC1 gene in two of fourfamilies with evidence for linkage. They then screened137 breast cancer families from The Netherlands. Theyfound six new cases and the deletion cosegregated withthe disease in these families. They also genotyped fiveintragenic markers (THRA1, D17S855, D17S1322,d17S1323, and D17S1327) in the BRCA1 gene, whichspans a 3 cM region. A common haplotype (169-151-122-151-133) was found in all eight cases (two from linkedfamilies and six new cases). Suppose the BRCA1 gene

# 1998 Blackwell Science Ltd, Clinical and Experimental Allergy, 28, Supplement 5, 1±5

4 J. Xu, D. G. Wiesch and D. A. Meyers

Paper 024 Disc

has not yet been cloned but we do know thechromosomal region that contains a breast cancersusceptibility gene based on linkage studies. First wetry the association approach. The allele frequencies inThe Netherlands population for these alleles are 0.08 forallele 169 of THRA1, 0.21 for allele 151 of D17S855,0.51 for allele 122 of D17S1322, 0.68 for allele 151 ofD17S1323, and 0.60 for allele 133 of D17S1327. If wecompared the allele frequencies between the independentcases from 137 breast cancer families and the generalpopulation, we would not have found significant results,even without adjusting for multiple tests. This is becausethe proportion of the cases who have the 510-bp deletion istoo small in the disease sample (2 + 6)/(4+ 137)=5.6%.As is likely to occur by chance, the original mutationoccurred in a common haplotype. If we use IBDmapping, we would identify these eight cases by theoverlap of the common haplotype. More markers withinthe 3 cM could then be genotyped to confirm that theseeight individuals are IBD for this haplotype and aredescended from a common ancestor. IBD mapping thuscan narrow the region and be very useful for localizingthe position of the susceptibility gene.

Summary

Association studies, if successful, can dramatically narrowthe chromosomal region containing the susceptibilitygene. Understanding the difficulties of association studiesfor complex diseases should not discourage studies; onthe contrary, it is meant to provide clues to increase thechance of success. Association studies still remain one of thebest approaches in the fine mapping of complex diseases.

For most complex diseases, there is a wide spectrumof phenotypes, ranging from mild to severe cases, early-age onset to late-age onset, and co-occurrence withother subphenotypes. It is possible that different genescause the varieties of phenotypes. Focusing on a narrowdefinition of phenotype may increase the homogeneity.

The best chance to observe association is to performthe study in a sample of homogeneous affectedindividuals. Effort should be made to increase theproportion of affected individuals who are descendentsfrom a common ancestor. Collecting samples in anisolated or genetically restricted population, focusing onthe families that showed linkage to the region, refiningphenotypic definition, and using haplotype-sharingmethods are useful approaches to reach the goal.

References

1 Lander ES, Schork NJ. Genetic dissection of complex traits[Review]. Science 1994; 265:2037±48.

2 Postma DS, Bleecker ER, Amelung PJ et al. Genetic

susceptibility to asthma±bronchial hyperresponsivenesscoinherited with a major gene for atopy. N Engl J Med1995; 333:894±900.

3 Daniels SE, Bhattacharrya S, James A et al. A genome-wide search for quantitative trait loci underlying asthma.Nature 1996; 383:247±50.

4 Collaborative Study of the Genetics of Asthma (CSGA). A

genome-wide search-wide search for asthma susceptibility lociin ethnically diverse populations. Nat Genet 1997; 15:389±97.

5 Ober C, Cox N, Parry R et al. and the CSGA. Genome-

wide search for asthma genes in an inbred population. AmJ Respir Crit Care Med 1997; 155:A257.

6 Spielman RS, McGinnis RE, Ewens WJ. Transmission test

for linkage disequilibrium: the Insulin gene region andinsulin-dependent diabetes mellitus (IDDM). Am J HumGenet 1993; 52:506±16.

7 Jorde LB. Linkage disequilibrium as a gene-mapping tool.

Am J Hum Genet 1995; 56:11±4.8 Risch N, Merikangas K. The future of genetic studies of

complex human diseases. Science 1996; 273:1516±7.

9 Strittmatter WJ, Saunders AM, Schmechel D et al. High-avidity binding to beta-amyloid and increased frequency oftype 4 allele in late-onset familial Alzheimer disease. Proc

Natl Acad Sci USA 1993; 90: 1977±81.10 Saunders AM, Schmader K, Breitner JC et al. Apolipo-

protein E epsilon 4 allele distributions in late-onset

Alzheimer's disease and in other amyloid-forming diseases.Lancet 1993; 342:710±1.

11 Payami H, Kaye J, Heston LL, Bird TD, Schellenberg GD.Apolipoprotein E genotype and Alzheimer's disease.

Lancet 1993; 342:738.12 Collins FS. BRCA1±lots of mutations, lots of dilemmas. N

Engl J Med 1996; 334:186±8.

13 Schellenberg GD, Boehnke M, Wijsman EM, Moore DK,Martin GM, Bird TD. Genetic association and linkageanalysis of the apolipoprotein CII locus and familial

Alzheimer's disease. Ann Neurol 1992; 31:223±7.14 Liddell M,Williams J, Bayer A, Kaiser F, OwenM. Confirma-

tion of association between the e4 allele of apolipoprotein E

and Alzheimer's disease. J Med Genet 1994; 31:197±200.15 Tsuda T, Lopez R, Rogaeva EA et al. Are the associations

between Alzheimer's disease and polymorphisms in theapolipoprotein E and the apolipoprotein CII genes due to

linkage disequilibrium? Ann Neurol 1994; 36:97±100.16 Morton NE. Sequential tests for the detection of linkage.

Am J Hum Genet 1955; 7:277±318.

17 Lander ES, Kruglyak L. Genetic dissection of complextraits: guidelines for interpreting and reporting linkageresults. Nat Genet 1995; 11:241±7.

18 Crowe RR. Candidate genes in psychiatry: an epidemio-logical perspective. Am J Med Genet 1993; 48:74±7.

19 Te Meerman GJ, Van der Meulen MA, Sandkuijl LA.Perspectives of identity by descent (IBD) mapping in founder

populations. Clin Exp Allergy 1995; 25 (Suppl. 2):97±102.20 Petrij-Bosch A, Peelen T, van Vliet M, et al. BRCA1

genomic deletions are major founder mutations in Dutch

breast cancer patients. Nat Genet 1997; 17:341±5.

# 1998 Blackwell Science Ltd, Clinical and Experimental Allergy, 28, Supplement 5, 1±5

Genetics of complex diseases 5

Paper 024 Disc