allergen / vancouver - 01/03//2009 meta-analysis of gabriel gwas asthma & ige f. demenais, m....

22
AllerGen / Vancouver - 01/03//2009 Meta-Analysis of GABRIEL GWAS Asthma & IgE F. Demenais, M. Farrall, D. Strachan GABRIEL Statistical Group

Upload: kaila-dome

Post on 14-Dec-2015

223 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: AllerGen / Vancouver - 01/03//2009 Meta-Analysis of GABRIEL GWAS Asthma & IgE F. Demenais, M. Farrall, D. Strachan GABRIEL Statistical Group

AllerGen / Vancouver - 01/03//2009

Meta-Analysis of GABRIEL GWASAsthma & IgE

F. Demenais, M. Farrall, D. StrachanGABRIEL Statistical Group

Page 2: AllerGen / Vancouver - 01/03//2009 Meta-Analysis of GABRIEL GWAS Asthma & IgE F. Demenais, M. Farrall, D. Strachan GABRIEL Statistical Group

AllerGen / Vancouver - 01/03//2009

GABRIEL Phase I GWAS

GWAS (Illumina 300K) of UK & German data

→ 17q21 locus (ORMDL3) associated with asthma Moffat et al, Nature, 2007

Replication of this association by several studies

Genetic heterogeneity at 17q21 locus (French EGEA data) → Effect of 17q21 variants restricted to early-onset asthma and enhanced by early-life exposure to ETS Bouzigon et al, New Engl J Med, 2008

Page 3: AllerGen / Vancouver - 01/03//2009 Meta-Analysis of GABRIEL GWAS Asthma & IgE F. Demenais, M. Farrall, D. Strachan GABRIEL Statistical Group

AllerGen / Vancouver - 01/03//2009

Aim of Phase II Gabriel GWAS

To identify associations of genetic variants with:

- susceptibility to asthma

(childhood onset, adult onset, industrial)

- total IgE levels

across populations of European ancestry using Illumina Human 610-Quad beadchip

by conducting a meta-analysis of all studies

Page 4: AllerGen / Vancouver - 01/03//2009 Meta-Analysis of GABRIEL GWAS Asthma & IgE F. Demenais, M. Farrall, D. Strachan GABRIEL Statistical Group

AllerGen – Vancouver – 01/03/2009

DATA AVAILABLE for Phase II GWAS

- Most datasets are cases/controls

- A few datasets include families: MRC (UK), EGEA (French), Canadian, Russian, GSK…

Childhood onset Asthma 20 012 subjects

Adult onset Asthma

Industrial Asthma 1356 subjects

Page 5: AllerGen / Vancouver - 01/03//2009 Meta-Analysis of GABRIEL GWAS Asthma & IgE F. Demenais, M. Farrall, D. Strachan GABRIEL Statistical Group

ALSPAC J . Henderson/W. McArdle 1292 1249BAMSE G. Pershagen 567 in progressB58C D. Strachan 911 904ECRHS D. J arvis/M. Wjst 2433 2370EGEA F. Demenais/F. Kaufmann 2034 2034FINRISK L. Vonhertzen 152 151GABRIEL AS E. von Mutius/M. Kabesch 1727 1710INDUS I. Wouters/D. Heederick 1356 1317MAS Y. Lee/I. Marenholz 184 183PIAMA D. Postma/G. Koppelman 423 406POLONIKOV A. Polonikov 682 634SAPALDIA N. Probst/M. Imboden 1640 1640SEVERE M. Moffatt/W. Cookson 366 362TOMSKA E. Bragina/M. Freidin 804 736UFA E. Khusnutdinova 727 715

BUSSELTON A. J ames 1598 1530CANADA - CAPPS & SAGE T. McDonald/D. Daley 1310 1109CANADA - SLSJ C. Laprise 1282 1268GSK-GAIN S. Pillai 1880 1844

TOTAL: 20729

POPULATION/COHORTContact Name/Principal

InvestigatorSamples selected

for Phase IIPassed QC

GWAS Phase II DATA

Page 6: AllerGen / Vancouver - 01/03//2009 Meta-Analysis of GABRIEL GWAS Asthma & IgE F. Demenais, M. Farrall, D. Strachan GABRIEL Statistical Group

Genotyping at CNG (Y. Gut, M. Lathrop, Evry, France)

Using Illumina Human 610-Quad beadchip

Initial QC processing at CNG (S.Heath, CNG) • % genotype calls - by individuals (< 95%: individuals excluded) • Relationship analysis to confirm known & identify cryptic relationships • Sex checks based on X-chromosome SNPs • Principal components analysis to identify cryptic non-European ancestry

Phase II GWAS: Overall Strategy

Analysis study by study (M Farrall, Oxford)

From Phenotypic data (each group) & Genotypic Data (CNG)

Meta-analysis of all studies: Phase II + Phase I (imputation)Asthma (F Demenais, Paris) IgE (D Strachan, London)

childhood onset, adult onset, all controls & cases separatelyindustrial asthma

Page 7: AllerGen / Vancouver - 01/03//2009 Meta-Analysis of GABRIEL GWAS Asthma & IgE F. Demenais, M. Farrall, D. Strachan GABRIEL Statistical Group

AllerGen – Vancouver – 01/03/2009

Phenotypes

Asthma :Cases : doctor-diagnosed asthma or self-reported

+ age onset of asthma

Controls: unaffecteds (not selected as « hypernormal »

and may include other forms of wheezing)

→ Childhood Onset / Adult Onset Asthma using a cutoff of 16yrs

Controls drawn at random for childhood onset/ adult onset cases

IgE (log10) IgE wadjusted on sex and age-at-measurement by study and by case-control status

Page 8: AllerGen / Vancouver - 01/03//2009 Meta-Analysis of GABRIEL GWAS Asthma & IgE F. Demenais, M. Farrall, D. Strachan GABRIEL Statistical Group

AllerGen – Vancouver – 01/03/2009

Method used for Study by Study Analysis

Single SNP analysis based on logistic regression models (linear

regression for IgE) allowing for familial clustering using STATA

Different models considered:

- additive model (1df)

- additive and non-additive effects ( 2 x 1 df)

- genotype association model (2 df)

Population Stratification:

Eigenvectors from PCA included in regression model

PCA uses HapMap data + CNG data (European controls)

Page 9: AllerGen / Vancouver - 01/03//2009 Meta-Analysis of GABRIEL GWAS Asthma & IgE F. Demenais, M. Farrall, D. Strachan GABRIEL Statistical Group

AllerGen – Vancouver – 01/03/2009

Population stratificationPCA on European controls from French National Genotyping Center

Heath et al, Eur J Hum Genet, 2008

Page 10: AllerGen / Vancouver - 01/03//2009 Meta-Analysis of GABRIEL GWAS Asthma & IgE F. Demenais, M. Farrall, D. Strachan GABRIEL Statistical Group

AllerGen – Vancouver – 01/03/2009

Meta-Analysis for Asthma & IgE

From the study-by-study analysis, tables generated including for each SNP:

QC metrics (MAF, SNP Call Rate, HW..) Number of cases / controls by genotype Regression coefficients & Standard errors Various test statistics

QC Filtering based on MAF (1% or 5%), SNP Call Rate (≥ 97%) HW (p > 10-4)

Meta-analysis using different methods

Page 11: AllerGen / Vancouver - 01/03//2009 Meta-Analysis of GABRIEL GWAS Asthma & IgE F. Demenais, M. Farrall, D. Strachan GABRIEL Statistical Group

AllerGen – Vancouver – 01/03/2009

Methods used for Primary Meta-Analysis

• Fixed-effect (inverse variance weighted) models assumes that observed effects are estimates of a single effect average effect computed by weighting each study’s log OR

according to the inverse of their sampling variance

→ Test of homogeneity for SNP effect across studies using Cochran Q test

• Random-effect models (DerSimonian & Laird, 1986)

allows for effects to vary across studies

variance = between study variation + intra-study variation

preferred if # of study-specific estimates ≥ 5

Page 12: AllerGen / Vancouver - 01/03//2009 Meta-Analysis of GABRIEL GWAS Asthma & IgE F. Demenais, M. Farrall, D. Strachan GABRIEL Statistical Group

AllerGen / Vancouver - 01/03//2009

Fixed vs Random effect Models

Example: Type 2 Diabetes (Ionnadis et al, PLoS one, 2007)

Meta-analysis of FUSION, DGI, WTCC

Gene SNP Q (p) I2 (95% CI) Random p Fixed p

rs9300039 0.019 75% (0-90) 0.015 4.3x10-7

FTO rs8050136 0.013 77% (0-91) 0.015 1.3x10-12

CDKAL1 rs10946398 0.16 46% (0-84) 3.2x10-6 4.1x10-11

PPARG rs1801282 0.15 47% (0-84) 0.0003 1.7x10-6

CDKN2B rs564398 0.48 0% (0-73) 1.2x10-7 1.2x10-7

HHEX rs5015480 0.45 0% (0-73) 5.7x10-10 5.7x10-10

Page 13: AllerGen / Vancouver - 01/03//2009 Meta-Analysis of GABRIEL GWAS Asthma & IgE F. Demenais, M. Farrall, D. Strachan GABRIEL Statistical Group

AllerGen / Vancouver - 01/03//2009

Other Methods of Meta-analysis: Meta-Regression Bag & Nikolopoulos, Stat Appl Mol Biol, 2007

Study

i = 1, 2..k

Cases

yij = 1

Controls

yij =0

Genotype

j =1,2,..r

1 38 72 1

1 52 78 2

1 6 22 3

2 44 38 1

2 27 45 2

2 11 17 3

Logit (pij) = i + 2zi2 + 3zi3 if genotype effect cst between studies

Logit (pij) = i + 2zi2 + 3zi3 + i2 izi2 + i3 izi3 if gentoypexstudy int

→ Test for heterogeneity between studies using Multivariate Wald test

Possible to include random effect + various covariates

Page 14: AllerGen / Vancouver - 01/03//2009 Meta-Analysis of GABRIEL GWAS Asthma & IgE F. Demenais, M. Farrall, D. Strachan GABRIEL Statistical Group

AllerGen / Vancouver - 01/03//2009

Other Approaches of Meta-Analysis

● Combining p-values or Z scores

● Local Score method (Guedj et al, 2006; Aschard et al, 2007) can detect aggregation of association signals

flexible approach which can use any test statistic

Page 15: AllerGen / Vancouver - 01/03//2009 Meta-Analysis of GABRIEL GWAS Asthma & IgE F. Demenais, M. Farrall, D. Strachan GABRIEL Statistical Group

COHORT/POPULATION STATUS

ECRHS Data released to Martin FarrallEGEA Data released to Martin FarrallFINRISK Data released to Martin FarrallINDUS_1 Data released to Martin FarrallMAS Data released to Martin FarrallPIAMA Data released to Martin FarrallSAPALDIA Data released to Martin FarrallTOMSKA Data released to Martin FarrallUFA Data released to Martin Farrall

ALSPAC Awaiting data releaseB58C Awaiting data releaseGABRIEL AS Awaiting data releasePOLONIKOV Awaiting data releaseSEVERE Awaiting data release

GSK_GAIN Genotyping in progressINDUS_2 Genotyping in progressCANADA -SLSJ Genotyping scheduled Feb/MarCANADA -CAPPS & SAGE Genotyping scheduled Feb/MarBAMSE DNA QC in progressBUSSELTON DNA QC in progress

Page 16: AllerGen / Vancouver - 01/03//2009 Meta-Analysis of GABRIEL GWAS Asthma & IgE F. Demenais, M. Farrall, D. Strachan GABRIEL Statistical Group

Outcome of Meta-Analysis

Identify Top SNPs (genome-wide significant)

Phase III Gabriel

Genotype top SNPs in 40 000 individuals

Page 17: AllerGen / Vancouver - 01/03//2009 Meta-Analysis of GABRIEL GWAS Asthma & IgE F. Demenais, M. Farrall, D. Strachan GABRIEL Statistical Group

AllerGen / Vancouver - 01/03//2009

Gene-Gene Interactions

Page 18: AllerGen / Vancouver - 01/03//2009 Meta-Analysis of GABRIEL GWAS Asthma & IgE F. Demenais, M. Farrall, D. Strachan GABRIEL Statistical Group

AllerGen / Vancouver - 01/03//2009

Various Methods to investigate GxG

- Regression-based methods (one stage, 2 stages…)

- Bayesian based approaches

- Data Reduction based-methods / Machine Learning

‘Combinatorial Partitioning Method (CPM), MDR)

- Pattern recognition models (neural networks)

- Combination of test statistics (meta-statistics)

Gabriel provides opportunity to compare these methods

by pooling data or in the context of meta-analysis

Page 19: AllerGen / Vancouver - 01/03//2009 Meta-Analysis of GABRIEL GWAS Asthma & IgE F. Demenais, M. Farrall, D. Strachan GABRIEL Statistical Group

AllerGen / Vancouver - 01/03//2009

Gene-Environment Interactions

Page 20: AllerGen / Vancouver - 01/03//2009 Meta-Analysis of GABRIEL GWAS Asthma & IgE F. Demenais, M. Farrall, D. Strachan GABRIEL Statistical Group

AllerGen / Vancouver - 01/03//2009

2 Step-Analysis to identify genes involved in GxEMurcray et al, Am J Epidemiol, 2008

Step 1: Screening test: case only analysis (combined case/control sample )

For each of N SNPs: LR Test for association between G and E

→ Select m SNPs with P < 1

Step 2 : Case- Control analysis

LR Test for GxE applied to m SNPs selected at step 1

→ Significance based on P < /m

Comparison with classical one-step approach applied to case-controls

→ Significance based on P < /N

Page 21: AllerGen / Vancouver - 01/03//2009 Meta-Analysis of GABRIEL GWAS Asthma & IgE F. Demenais, M. Farrall, D. Strachan GABRIEL Statistical Group

AllerGen / Vancouver - 01/03//2009

Power for one-step and two-step analyses to detect GxE

for varying levels of interaction effect size

10,000 markers and 500 cases/500 controls

Page 22: AllerGen / Vancouver - 01/03//2009 Meta-Analysis of GABRIEL GWAS Asthma & IgE F. Demenais, M. Farrall, D. Strachan GABRIEL Statistical Group

AllerGen / Vancouver - 01/03//2009

GABRIEL Working Groups

GW search for G X smoking in asthma M Boezen, D Postma, The Netherlands

Childood Asthma (M Kabesch) & Adult Asthma (D. Jarvis) to summarize data available in each study (phenotypes, environment)

Main areas of interest for collaborations: Phenotypes Environmental exposures : GxE Pathways: GxG Other types of variation: CNVs

Methodological issues

New opportunities that are going to emerge from the

AllerGen meeting