复旦大学医学遗传学casecontrolTDT分析

1、Fudan Genetic Workshop 2005Genetic Epidemiology in Populations Genetic Association Studies: Studies with Unrelated IndividualsFudan Genetic Workshop, Jan 04 2005 Assessing risk for disease due to specific alleles/genotypes Contrast direct / indirect genetic association studies Discuss genetic models

2、 Tests and estimates of association between alleles/genotypes and disease Designs to be usedLearning Objectives for This Lecture Applications for genetic association analyses Direct versus indirect association Indirect association: marker-based disease association Exploits LD between markers and uno

3、bserved disease variants Examples of direct and indirect association studies Study designs for direct and marker-based (LD) associationGenetic Association Studies in EpidemiologyApplications: Fine mapping Candidate gene effects Whole genome scan Gene environment effects Drug response modificationGen

4、etic Association Studies in EpidemiologyCommon Designs: Cross-sectional Case-control Cohort Clinical Trial (drug response hypothesis is really a case-control) Case-family (trios, sibs, extended families) Case-onlyLearning Objectives for This LectureApplications for association analyses Direct versus

5、 indirect association Indirect association: marker-based association Exploits LD Examples of direct and indirect association studies Study designs for direct and marker-based (LD) association1.Direct methodTesting whether a particular allele is a disease - predisposing (causative) alleleexposure sta

6、tus directly measuredGenetic epidemiology studies two different concepts .GACTAAGGCCC CCGTTCAAGGAA. C/TAPOE gene on c19Eg: A particular APOE allele (e4) changes protein isoform Genotype that particular site for association study 2.LD Mapping (indirect method)exposure status not directly measuredRely

7、 on markers correlated with true exposure statusThis correlation is due to linkage disequilibriumGenetic epidemiology studies two different concepts .GACTAAGGCCC CCGTTCAAGGA CCTG. C/TA/GAPOE gene on c19Eg: Genotype a nearby genetic marker among study participantsRely on correlation (LD) between thes

8、e alleles to detect association!Direct Tests Test for association between observed genotypes and disease outcomeCaseControlTotalCCa bm1CTcdm2TTefm3TotalndnhNCCCCCG CCCCCG CCCCCG CCTCCG CCTCCG CCTCCG Directly measure potentially predisposing allelesIndirect Tests Test for association between marker g

9、enotypes and disease outcomeMarkerCaseControlTotalAAa bm1AGcdm2GGefm3TotalndnhNCC?CCGCAGCC?CCGCAG Use marker genotypes as surrogate for functional genotype by exploiting any correlation between C and ACC?CCGCAGCC?CCGCGGCC?CCGCGGCC?CCGCGGDirect Method: Candidate Allele Testing Must know potentially f

10、unctional polymorphisms SNPs may offer set of candidate loci for this method Eg, look at non-synonymous cSNPs (those in coding regions that are likely to be functional) Note: Assuming disease is caused by (relatively) common alleles Common disease-common variant hypothesis This hypothesis is controv

11、ersial If genome-wide project: multiple comparisons - 30,000 genes. Even if only one functional locus/gene tested, very high number of false +s due to chanceIndirect Method: LD Gene MappingGeneral idea: Exploit the phenomenon of linkage disequilibrium (LD) between alleles of closely linked markers t

12、o identify genetic regions associated with disease status. i.e., Exploit LD-induced correlation between observed marker alleles and potentially unobserved disease alleles .GACTAAGGCCC CCGTTCAAGGA CCTG. C/TA/GRely on correlation (LD) between these alleles to detect association!Indirect Association St

13、udies in EpidemiologyApplications: Localization (what is the best estimate of the disease gene location?) Fine mapping Whole genome scan Candidate gene effects Gene environment effects Drug response modificationLocalizationIndirect Association Studies in EpidemiologyApplications: Localization (what

14、is the best estimate of the disease gene location?) Fine mapping Whole genome scan Candidate gene effects Gene environment effects Drug response modificationSource: Uhl et al, 2001. AJHG 69:1290-1300.Indirect Association Studies in EpidemiologyApplications: Localization (what is the best estimate of

15、 the disease gene location?) Fine mapping Whole genome scan Candidate gene effects Gene environment effects Drug response modificationCandidate gene LD studies Instrumental in identifying genetic risk factors for disease HLA Type 1 diabetes Rheumatoid arthritis Multiple sclerosis APOE Alzheimers dis

16、ease Atherosclerosis Angiotensin Converting Enzyme (ACE) Myocardial infarction AtherosclerosisCandidate gene LD studies Polymorphism in candidate genes can be used as a marker for variation in the gene. The higher the LD between the causative variant and the marker alleles, the better the surrogate

17、information provided by the marker.MarkerCaseControlTotalAAa bm1AGcdm2GGefm3TotalndnhNCC?CCGCAGCC?CCGCAGCC?CCGCAGCC?CCGCGGCC?CCGCGGCC?CCGCGGThe power of G to detect disease risk at ? is only as strong as the LD between the loci!Candidate gene LD studies Use genetic markers to test for variation in t

18、hat gene that may predispose to disease risk Marker alleles are surrogates for the (unobserved) disease allele due to LDMarkerCasesControlsAabMAacdMaMcaseMctrl1(surrogate for unobserved disease allele)AdxGenotype-level tests for disease association Test for association between genotypes and disease

19、outcomeGCaseControlTotal11a bm112cdm222efm3TotalndnhN61222)(iiiiEEOEx: Alzheimers disease and APOE SNPsSingle-marker 2 tests for disease association:M1GCaseControlTotalCC101 (.50)116 (.75)217CT75 (.37)32(.21)107TT26(.13)6 (.04)32Total20215435661222)(iiiiEEO80.2484.13)84.136(16.18)16.1826(29.46)29.46

20、32(71.60)71.6075(87.93)87.93116(13.123)13.123101(22222222Modeling issues If we do measure the putative variation of interest, what kinds of statistical analyses should be performed? Most general genetic model for risk: assume to a priori relationship between heterozygote and homozygote risk Consider

21、 each genotype as a separate exposure category:GenotypeRisk parameterRelative RiskAARAARRAAA*RA*RRA*R*1*21)(1)(lnAAAIIDPDP In a logistic regression model:00* *10A*01AAIAaIAAGenotype00* *10A*01AAIAaIAAGenotypeModeling issues This requires 2 risk parameters What genetic models could be considered to i

22、mprove power (reduce # of parameters to estimate)? May have more power if particular models can correctly be assumed: DominantRRAA = RRA* 1 = RR* Additive allele effectsRAA = 2*RA* R* = 1 Multiplicative allele effectsRRAA = (RRA*)2 RR* = 1 RecessiveRRAA 1 = RRA* = RR*GenotypeRisk parameterAARRAAA*RR

23、A*RR*GenotypeRisk parameterDomAARRAARRA.A*RRA*RRA.*RR*1GenotypeRisk parameterDomRecAARRAARRA.RRAAA*RRA*RRA.1*RR*11GenotypeRisk parameterDomRecAddAARRAARRA.RRAA2RRA*A*RRA*RRA.1RRA*RR*111GenotypeRisk parameterDomRecAddMultAARRAARRA.RRAA2RRA*(RRA*)2A*RRA*RRA.1RRA*2RRA*RR*1111Logistic regressionH0: i =

24、0 Genetic model interpretations: Assume “11” genotype coding represents genotype with lowest absolute risk (baseline)1 = 2 = 0no association with that polymorphism1 = 0, 2 0(completely) recessive1 = 2 0(completely) dominant0 1 Possible interaction!SmokeTGFCPControlORNoG-721271NoG+12191.11YesG-25152.

25、94YesG+1555.29Logistic regression with interactionTest for interaction: H0: gi = 0 Modeling issues For Cleft example, dominance was modeled, so logistic model would look like:)*()*()(1)(ln2221213222121EGEGEGGDPDPgg)*()(1)(ln11211EGEGDPDPg222121)(1)(lnGGDPDPInterpretations of Indirect Association Stu

26、diesA positive association can mean:The targeted allele is causalThe targeted allele is in LD with a causal alleleThere is confounding due to population stratificationThere is confounding or bias for some other reasonType I errorA negative finding can mean:The gene or region under study is not assoc

27、iated with disease riskThe targeted allele is not in LD with the causal allele Appropriate stratification or other accommodation of heterogeneity was not identified Type II error (not enough power)Learning Objectives for This LectureLinkage versus associationUses of association analysesDirect versus

28、 indirect associationIndirect association: marker-based associationExploits LD Examples of direct and indirect association studies Study designs for direct and marker-based (LD) associationChoice of Design and Analysis Sampling design Unrelated individuals Family-based sampling Unit of analysis Sing

29、le-locus Haplotypes Statistical Procedures Chi-square tests Likelihood-based tests Asymptotic p values Empirical significance from resampling Appropriate significance thresholds Choice of Design and Analysis Design options for association studies Sampling unrelated individuals Family-based designs T

30、hese differ on how controls are definedStudy Designs for LD Mapping StrategiesUnrelated Samples (Eg, Case control) Unrelated controls are sampled from the same population as the cases Matched or unmatched on other factors Perform chi-squared test for associationFamily-based Ex: TDT, looks for excess

31、 transmission of particular alleles from parents to affected children Controls are untransmitted allelesMarker alleleCase ControlAAabAacdaaefFor each individual, have 2x2 table of 0s, 1s, or 2sUse all such tables to get a matched chi-square test for excess occurrence in cells b and c McNemars testA-

32、Not transmitteda Not transmittedA - Transmitted02a - Transmitted00A,aA,aA,AContrasts between epidemiologic and family-based association designs Unrelated individuals Do not need parental genetic information Provides estimates of allele frequencies (if appropriately sampled) May be more powerful and

33、simpler to collect for some phenotypes Opens the potential for confounding due to population stratification Family-based designs Need parental genotypes (or at least other family members) Do not need unrelated controls Avoids population stratification confounding (analysis is matched by family)Desig

34、ns for LD studies of unrelated individuals All unrelated sampling LD methods look for association between markers and the outcome of interest across individuals The utility of this approach will be a function of the actual LD between the markers and the true disease allele(s) Cross-sectional Case-co

35、ntrol Cohort Analyses are similar, interpretations may be different because sampling and temporal relationships are different Clinical trial? Such data are used to assess pharmacogenetic response outcomes. However, this is often a case-control design (the genotypes are not randomized!)Study Designs

36、used for LD mappingCase - Control design (unrelated individuals)Advantages: Commonly used tool in epidemiology - methodology is well-known Convenient to collect - opportunity to draw very large samples More efficient recruitment than family-based sampling methods Population-based designs can allow t

37、he simultaneous characterization of disease allele frequency, penetrance, and attributable risk Unrelated controls can provide increased power in many situationsStudy Designs used for LD mappingCase - Control design (unrelated individuals)Limitations: Difficult to establish phase when focusing on ha

38、plotypes May be susceptible to confounding due to stratification Difficult to measure parent-of-origin effects or other parent-specific effects Difficult to estimate recombination fractions (localization) using LD in an unrelated-subject settingCase-Control Susceptibility to population stratificatio

39、n TDT and other family-based association methods commonly used to guard against confounding due to population stratificationBut, Requires recruitment of additional family members Added cost Family members may not be available Inefficient if many members required or members are uninformative Probably

40、 not as prevalent a problem as once thought Can be dealt with through assessment and adjustment of population stratification within a case-control data setStudy Designs used for LD mappingCase - Control designLimitations: May be susceptible to confounding due to stratification Solution: Measure pote

41、ntial confounding via genomic control Significant substructure is rare in preliminary empirical findings! Doesnt hurt to assess anyway! May still want to adjust for other confounding reasons1) Adjustment in Analysis Assess and correct for possible stratification Genomic control (Devlin and Roeder, 1

42、999, Genomic Control for association studies. Biometrics, 1999. 55: p. 997-1004) Cluster analysis of genetic substructure based on markers (STRUCTURE, Pritchard et al. Inference of population structure using multilocus genotype data. Genetics, 2000. 155(2): p. 945-59) FSTSolutions to Population Stra

43、tification Problem: How to distinguish LD from allelic associations due to sub-structure?Ex: Use Fst to detect gross genetic distances between cases and controls as indication of substructure Study 1 Study 2 Cases Controls Cases Controls Normotensives Cases 1 1 0.00059 0.0007 0.0016 -0.0005 Controls

44、 1 0.307 1 0.001 0.002 0.002 Cases 2 0.218 0.049 1 0.0004 0.001 Controls - 2 0.039 0.039 0.267 1 0.148 Normotensives 0.742 0.029 0.059 0.148 1 Genetic Distance (FST) P-valueGenetic Distances Between Sample Populations in Renal failure Susceptibility StudySolutions to Population Stratification Proble

45、m: How to distinguish LD from allelic associations due to sub-structure?2) Matching in design Ethnicity-matched controls- Self-reported- Correlation between broad ethnicity and genetic background may not be very good Family controls Matched by parents, so genetic substructure controlled Next lecture

46、!梅毒螺旋体的超微结构Candidate gene LD studies Use genetic markers to test for variation in that gene that may predispose to disease risk Marker alleles are surrogates for the (unobserved) disease allele due to LDMarkerCasesControlsAabMAacdMaMcaseMctrl1(surrogate for unobserved disease allele)AdxModeling issu

47、es If we do measure the putative variation of interest, what kinds of statistical analyses should be performed? Most general genetic model for risk: assume to a priori relationship between heterozygote and homozygote risk Consider each genotype as a separate exposure category:GenotypeRisk parameterR

48、elative RiskAARAARRAAA*RA*RRA*R*1*21)(1)(lnAAAIIDPDP In a logistic regression model:00* *10A*01AAIAaIAAGenotype00* *10A*01AAIAaIAAGenotypeModeling issues If we do measure the putative variation of interest, what kinds of statistical analyses should be performed? Most general genetic model for risk: assume to a priori relationship between heterozygote and homozygote risk Consider each genotype as a separate exposure category:GenotypeRisk parameterRelative RiskAARAARRAAA*RA*RRA*R*1*21)(1)(lnAAAIIDPDP In a logistic regression model:00* *10A*01AAIAaIAAGenotype00* *10A