Alkes Price Harvard School of Public Health January 24 & January 26, 2017

Transcription

1 EPI 511, Advanced Population and Medical Genetics Week 1: Intro + HapMap / 1000 Genomes Linkage Disequilibrium Alkes Price Harvard School of Public Health January 24 & January 26, 2017

2 EPI 511: Course structure Week 1: HapMap, 1000G / Linkage disequilibrium Week 2: Population structure and admixture Week 3: Population stratification Week 4: Fine-mapping / Natural selection Week 5: Heritability / Genetic risk prediction Week 6: Mixed models / Rare variant analysis Week 7: Functional interpretation

3 EPI 511: How to address the instructor Alkes Dr. Price Professor Price Honorable Professor Price Honorable Distinguished Dr. Professor Price

4 EPI 511: Office Hours Instructor: Alkes Office Hours: Thu 3:30-4:30pm, Building 2, Room 211 Address: (Please put EPI511 in the subject of your ) Teaching Assistant: Armin Office Hours: Fri + Mon 2-3pm, Building 2, Room 209 Address: arminschoech@g.harvard.edu

5 EPI 511: Course components Advance reading 1 required paper + 1 optional paper per course session

6 EPI 511: Course components Advance reading 1 required paper + 1 optional paper per course session Lecture + Discussion

7 EPI 511: Course components Advance reading 1 required paper + 1 optional paper per course session Lecture + Discussion discussants: each student to sign up as discussant for 1 class

8 EPI 511: Course components Advance reading 1 required paper + 1 optional paper per course session Lecture + Discussion discussants: each student to sign up as discussant for 1 class

9 EPI 511: Course components Advance reading 1 required paper + 1 optional paper per course session Lecture + Discussion discussants: each student to sign up as discussant for 1 class Video of each class will be posted on the course www site <1hr after class.

10 EPI 511: Course components Advance reading 1 required paper + 1 optional paper per course session Lecture + Discussion discussants: each student to sign up as discussant for 1 class

11 EPI 511: Course components Advance reading 1 required paper + 1 optional paper per course session Lecture + Discussion discussants: each student to sign up as discussant for 1 class Experiences 5 take-home projects due Tue Jan 31,, Tue Feb 28

12 EPI 511: Course components Advance reading 1 required paper + 1 optional paper per course session Lecture + Discussion discussants: each student to sign up as discussant for 1 class Experiences 5 take-home projects due Tue Jan 31,, Tue Feb 28

13 EPI 511: Course components Advance reading 1 required paper + 1 optional paper per course session Lecture + Discussion discussants: each student to sign up as discussant for 1 class Experiences 5 take-home projects due Tue Jan 31,, Tue Feb 28 short Research Paper due Fri Mar 10

14 EPI 511: Course components Advance reading 1 required paper + 1 optional paper per course session Lecture + Discussion discussants: each student to sign up as discussant for 1 class Experiences 5 take-home projects due Tue Jan 31,, Tue Feb 28 short Research Paper due Fri Mar 10 self-assessment Opportunity 20min exam (date will not be announced in advance)

15 EPI 511: Outcome measures Advance reading (0% of course grade) 1 required paper + 1 optional paper per course session Lecture + Discussion (0% of course grade) discussants: each student to sign up as discussant for 1 class

16 EPI 511: Outcome measures Advance reading (0% of course grade) 1 required paper + 1 optional paper per course session Lecture + Discussion (0% of course grade) discussants: each student to sign up as discussant for 1 class Experiences (60% of course grade) 6 take-home projects (data and programming intensive)

17 Approaches to Scientific Understanding Love is Understanding. -- Madonna Data is Understanding. -- Alkes

18 EPI 511: Outcome measures Advance reading (0% of course grade) 1 required paper + 1 optional paper per course session Lecture + Discussion (0% of course grade) discussants: each student to sign up as discussant for 1 class Experiences (60% of course grade) 6 take-home projects (data and programming intensive)

19 Approaches to Scientific Understanding Understanding Data requires Fixing Bugs.

20 Genetics + data + programming = bright future Gewin 2007 Nature Hayden 2012 Nature

21 EPI 511: Outcome measures Advance reading (0% of course grade) 1 required paper + 1 optional paper per course session Lecture + Discussion (0% of course grade) discussants: each student to sign up as discussant for 1 class Experiences (60% of course grade) 5 take-home projects (data and programming intensive) short Research Paper (40% of course grade) 1,000-1,500 words (suggested topics provided on Feb 16)

22 EPI 511: Outcome measures Advance reading (0% of course grade) 1 required paper + 1 optional paper per course session Lecture + Discussion (0% of course grade) discussants: each student to sign up as discussant for 1 class Experiences (60% of course grade) 5 take-home projects (data and programming intensive) short Research Paper (40% of course grade) 1,000-1,500 words (suggested topics provided on Feb 16) self-assessment Opportunity (0% of course grade) 20min exam (date will not be announced in advance)

23 EPI 511: Policy on group work Experiences (60% of course grade) 6 take-home projects (data and programming intensive) OK to discuss experiences with your colleagues Each piece of code you write should be your own short Research Paper (40% of course grade) 1,000-1,500 words (suggested topics provided on Feb 16) OK to discuss the project with your colleagues Each piece of code you write should be your own Each piece of text you write should be your own

24 EPI 511, Advanced Population and Medical Genetics Week 1: Introduction + HapMap Project Linkage Disequilibrium

25 Outline 1. Introduction to Population Genetics 2. HapMap and HapMap2 projects 3. F ST 4. HapMap3 and 1000 Genomes projects

26 Outline 1. Introduction to Population Genetics 2. HapMap and HapMap2 projects 3. F ST 4. HapMap3 and 1000 Genomes projects

27 What is Population Genetics? Population genetics is the study of genetic variation both within and between human populations.

28 Are different human populations actually genetically different?

29 Slightly. Are different human populations actually genetically different? 5-7% of worldwide human genetic variation is due to genetic differences between human populations. The remaining 93-95% of human genetic variation is due to genetic variation within human populations (Rosenberg et al Science).

30 Why study differences between human populations? Learn about human migration patterns and ancient history.

31 Why study differences between human populations? Learn about human migration patterns and ancient history. Improve our power to identify and localize disease genes.

32 Rosenberg et al Nat Rev Genet

33 Bustamante et al Nature; also see Popejoy & Fullerton 2016 Nature

34 Why study differences between human populations? Learn about human migration patterns and ancient history. Improve our power to identify and localize disease genes. Williams et al Nature

35 Why study differences between human populations? Learn about human migration patterns and ancient history. Improve our power to identify and localize disease genes. - Use differences in linkage disequilibrium for fine-mapping. - Avoid false positives due to population stratification. - Signals of natural selection at genes related to disease.

36 Does race exist?

37 For a fun time: go to a population genetics party and ask, Does race exist? Worldwide patterns of human genetic variation are best described using continuous clines instead of discrete clusters. (Serre & Paabo 2004 Genome Res) Racial classifications are inadequate descriptors of the distribution of human genetic variation. (Tishkoff & Kidd 2004 Nat Genet)

38 Isn t it politically incorrect to study differences between human populations?

39 Isn t it politically incorrect to study differences between human populations? No. It is not politically incorrect.

40 Isn t it politically incorrect to study differences between human populations? No. It is not politically incorrect. Studies of human population genetics have generated the strongest proof that there is no scientific basis for racism. (Cavalli-Sforza 2005 Nat Rev Genet) also see Cavalli-Sforza et al The History and Geography of Human Genes

41 Outline 1. Introduction to Population Genetics 2. HapMap and HapMap2 projects 3. F ST 4. HapMap3 and 1000 Genomes projects

42 The International HapMap Project (International HapMap Consortium 2005 Nature) CEU (European) YRI (Nigerian) CHB (Chinese) JPT (Japanese)

43 The International HapMap Project: 270 samples from 4 populations CEU northern European USA 90 CHB Chinese China 45 JPT Japanese Japan 44 YRI Yoruba Nigeria 90

44 The International HapMap Project (International HapMap Consortium 2005 Nature) Phase I HapMap: >1,000,000 SNPs CEU (European) YRI (Nigerian) CHB (Chinese) JPT (Japanese)

45 The International HapMap Project (International HapMap Consortium 2007 Nature) Phase II HapMap: >3,000,000 SNPs CEU (European) YRI (Nigerian) CHB (Chinese) JPT (Japanese)

46 What is a SNP? A Single Nucleotide Polymorphism (SNP) is a letter of the genome that differs in different individuals (e.g. G/T).

47 What is a SNP? A Single Nucleotide Polymorphism (SNP) is a letter of the genome that differs in different individuals (e.g. G/T). Each SNP corresponds to one single mutation event in history, e.g. G mutated to T in one single ancestor. G = ancestral allele, T = derived allele. Coalescent tree Rosenberg & Nordborg 2002 Nat Rev Genet

48 What is a SNP: physical position Each SNP has a physical position on a chromosome. physical chrom. position (bp) rs rs

49 What is a SNP: physical vs. genetic position Each SNP has a physical and genetic position on a chromosome. physical genetic position chrom. position (Morgans) rs rs recombination event per Morgan per generation. Genome-wide recombination rate is about 1cM / Mb. [cm = centimorgan = 1/100 Morgan, Mb = Megabase = 10 6 bp] Thus, 1 Morgan is roughly 100Mb = 10 8 bp on average.

50 HapMap project: Summary of main results 3.1 million SNPs successfully genotyped using Perlegen genotyping technology (Hinds et al Science). These 3.1 million SNPs: about 30% of all common SNPs (defined as SNPs with minor allele frequency >5%).

51 HapMap: 270 samples from 4 populations CEU northern European USA 90 CHB Chinese China 45 JPT Japanese Japan 44 YRI Yoruba Nigeria 90 Affymetrix and Illumina chips

52 HapMap project: Summary of main results 3.1 million SNPs successfully genotyped using Perlegen genotyping technology (Hinds et al Science). These 3.1 million SNPs: about 30% of all common SNPs (defined as SNPs with minor allele frequency >5%). Properties of SNPs are influenced by discovery sampling HapMap relied on nearly any piece of information available. Clark et al Genome Res; also see Keinan et al Nat Genet

53 Summary of main results, continued Understanding genetic differences between populations. Patterns of linkage disequilibrium both within and across populations. Most common SNPs in the human genome are in strong linkage disequilibrium with at least one HapMap SNP [avg r in 10 sequenced ENCODE regions].

54 Genetic differences between HapMap populations (International HapMap Consortium 2005 and 2007 Nature) 50% frequency C allele of rs % frequency 77% frequency

55 Genetic differences between HapMap populations (International HapMap Consortium 2005 and 2007 Nature) F ST = 0.11 F ST = 0.16 Note: F ST accounts for sampling error due to finite sample size. F ST = 0.19

56 Populations can be distinguished using a large number of genetic markers using 100 markers Principal Components Analysis

57 Populations can be distinguished using a large number of genetic markers using 3 million markers Principal Components Analysis

58 Outline 1. Introduction to Population Genetics 2. HapMap and HapMap2 projects 3. F ST 4. HapMap3 and 1000 Genomes projects

59 Genetic differences between HapMap populations (International HapMap Consortium 2005 and 2007 Nature) F ST = 0.11 F ST = 0.16 F ST = 0.19

60 Weir & Hill 2002 Annu Rev Genet, Bhatia et al Genome Res Defining vs. Estimating F ST F ST is an underlying parameter that depends on the two populations, but does not depend on a particular finite sample. ^ F ST is an estimate of the underlying F ST that depends on a particular finite sample that is analyzed.

61 Weir & Hill 2002 Annu Rev Genet, Bhatia et al Genome Res Defining F ST Definition: The F ST between two populations is the value such that the allele frequency difference between the two populations has mean 0 and variance 2F ST p(1 p), where p is the allele frequency in the ancestral population. p F ST p(1 p) F ST p(1 p) p 1 p 2

62 Weir & Hill 2002 Annu Rev Genet, Bhatia et al Genome Res Defining F ST Definition: The F ST between two populations is the value such that the allele frequency difference between the two populations has mean 0 and variance 2F ST p(1 p), where p is the allele frequency in the ancestral population. p F ST p(1 p) F ST p(1 p) p 1 p 2 p 1 ~ N(p, F ST p(1 p))

63 Weir & Hill 2002 Annu Rev Genet, Bhatia et al Genome Res Defining F ST Definition: The F ST between two populations is the value such that the allele frequency difference between the two populations has mean 0 and variance 2F ST p(1 p), where p is the allele frequency in the ancestral population. p F ST p(1 p) F ST p(1 p) p 1 p 2 p 1 ~ Beta(p(1 F ST )/F ST, (1 p)(1 F ST )/F ST )

64 Weir & Hill 2002 Annu Rev Genet, Bhatia et al Genome Res Defining F ST Definition: The F ST between two populations is the value such that the allele frequency difference between the two populations has mean 0 and variance 2F ST p(1 p), where p is the allele frequency in the ancestral population. OR The F ST between two populations is equal to the proportion of genotypic variance in a set of N individuals from each population that is attributable to population differences.

65 Defining F ST Theorem 1: The F ST between two populations is the value such that the allele frequency difference between the two populations has mean 0 and variance 2F ST p(1 p), where p is the allele frequency in the ancestral population. => The F ST between two populations is equal to the proportion of genotypic variance in a set of N individuals from each population that is attributable to population differences.

66 Defining F ST Proof: Let p avg = (p 1 + p 2 )/2. Total genotypic variance is 2p avg (1 p avg ) 2p(1 p) [Note that individuals are diploid: genotype = 0 or 1 or 2. Binomial sampling with n=2.]

67 Defining F ST Proof: Let p avg = (p 1 + p 2 )/2. Total genotypic variance is 2p avg (1 p avg ) 2p(1 p) [Note that individuals are diploid: genotype = 0 or 1 or 2. Binomial sampling with n=2.] Genotypic variance attributable to population differences: Suppose we have N data points with value 2p 1, N with value 2p 2 After subtracting the average value (p 1 + p 2 ), we have N data points with value (p 1 p 2 ), N with value (p 2 p 1 ). Since p 1 and p 2 each have variance F ST p(1 p), it follows that (p 1 p 2 ) and (p 2 p 1 ) each have variance 2F ST p(1 p)

68 Defining F ST Proof: Let p avg = (p 1 + p 2 )/2. Total genotypic variance is 2p avg (1 p avg ) 2p(1 p) [Note that individuals are diploid: genotype = 0 or 1 or 2. Binomial sampling with n=2.] Genotypic variance attributable to population differences: Suppose we have N data points with value 2p 1, N with value 2p 2 After subtracting the average value (p 1 + p 2 ), we have N data points with value (p 1 p 2 ), N with value (p 2 p 1 ). Since p 1 and p 2 each have variance F ST p(1 p), it follows that (p 1 p 2 ) and (p 2 p 1 ) each have variance 2F ST p(1 p) 2F ST p(1 p) / 2p(1 p) = F ST. Q.E.D.

69 Defining F ST Theorem 1 : The F ST between two populations is the value such that the allele frequency difference between the two populations has mean 0 and variance 2F ST p(1 p), where p is the allele frequency in the ancestral population. => The proportion of genotypic variance in a set of αn individuals from population 1 and (1 α)n individuals from population 2 that is attributable to population differences is equal to 4α(1 α) F ST.

70 Genetic differences between HapMap populations (International HapMap Consortium 2005 and 2007 Nature) F ST = 0.11 F ST = 0.16 F ST = 0.19

71 Genetic differences between HapMap populations (International HapMap Consortium 2005 and 2007 Nature) F ST = 0.11 [2F ST p(1 p)] 1/2 = 0.23 for p = 0.5 F ST = 0.16 [2F ST p(1 p)] 1/2 = 0.28 for p = 0.5 F ST = 0.19 [2F ST p(1 p)] 1/2 = 0.31 for p = 0.5

72 Genetic distances (F ST ) between European American subpopulations Ashkenazi F ST = F ST = Southeast Eur. Northwest Eur. F ST = Price, Butler et al PLoS Genet

73 Genetic distances (F ST ) between European American subpopulations Ashkenazi [2F ST p(1 p)] 1/2 = for p = 0.5 F ST = F ST = [2F ST p(1 p)] 1/2 = for p = 0.5 Southeast Eur. Northwest Eur. F ST = [2F ST p(1 p)] 1/2 = for p = 0.5 Price, Butler et al PLoS Genet

74 Genetic distances (F ST ) between East Asian subpopulations Chinese Japanese F ST = [2F ST p(1 p)] 1/2 = for p = 0.5 International HapMap Consortium 2007 Nature

75 Genetic distances (F ST ) between West African subpopulations Yoruba (Nigeria) F ST = Luhya (Kenya) [2F ST p(1 p)] 1/2 = for p = 0.5 International HapMap3 Consortium 2010 Nature

76 How do we estimate F ST? p 1 and p 2 are allele frequencies in 2 populations Var(p 1 p 2 ) = 2F ST p(1 p). Thus, estimate F ST = Var((p 1 p 2 ) / [2p(1 p)] 1/2 ). = E((p 1 p 2 ) 2 / [2p(1 p)]).

77 How do we estimate F ST? p 1 and p 2 are allele frequencies in 2 populations Var(p 1 p 2 ) = 2F ST p(1 p). Thus, estimate F ST = Var((p 1 p 2 ) / [2p(1 p)] 1/2 ). = E((p 1 p 2 ) 2 / [2p(1 p)]). A PROBLEM: we don t get to observe p (ancestral frequency) SOLUTION: approximate p p avg = (p 1 + p 2 )/2.

78 How do we estimate F ST? p 1 and p 2 are allele frequencies in 2 populations Var(p 1 p 2 ) = 2F ST p(1 p). Thus, estimate F ST = Var((p 1 p 2 ) / [2p(1 p)] 1/2 ). = E((p 1 p 2 ) 2 / [2p(1 p)]). A BIGGER PROBLEM: we don t get to observe p 1 and p 2. We only get to observe sample allele frequencies p^ 1 and p^ 2 in sample sizes N 1 (from pop. 1) and N 2 (from pop. 2).

79 How do we estimate F ST? p 1 and p 2 are allele frequencies in 2 populations Var(p 1 p 2 ) = 2F ST p(1 p). Thus, estimate F ST = Var((p 1 p 2 ) / [2p(1 p)] 1/2 ). = E((p 1 p 2 ) 2 / [2p(1 p)]). SOLUTION: Since Var(p^ 1 p^ 2 ) [2F ST + 1/(2N 1 ) + 1/(2N 2 )] p(1 p), estimate F ST = E([(p^ 1 p^ 2 ) 2 (1/(2N 1 ) + 1/(2N 2 ))p(1 p)] / [2p(1 p)]) (where we approximate p (p^ 1 + p^ 2 )/2) some details omitted; see Bhatia et al Genome Res

80 How do we estimate F ST? p 1 and p 2 are allele frequencies in 2 populations Var(p 1 p 2 ) = 2F ST p(1 p). Thus, estimate F ST = Var((p 1 p 2 ) / [2p(1 p)] 1/2 ). = E((p 1 p 2 ) 2 / [2p(1 p)]). SOLUTION: Since Var(p^ 1 p^ 2 ) [2F ST + 1/(2N 1 ) + 1/(2N 2 )] p(1 p), estimate F ST = E([(p^ 1 p^ 2 ) 2 (1/(2N 1 ) + 1/(2N 2 ))p(1 p)] / [2p(1 p)]). ^ ^ OR F ST = Σ i [(p i1 p i2 ) 2 (1/(2N 1 ) + 1/(2N 2 ))p i (1 p i )] Σ i [2p i (1 p i )] (where i indexes SNPs) some details omitted; see Bhatia et al Genome Res

81 Drift vs. Divergence Divergence (per 1000bp of DNA) Drift (F ST ) YRI CEU CHB YRI YRI CEU CEU CHB CHB NA18488 NA06989 NA18597 Keinan et al Nat Genet

82 Drift (F ST ) Drift vs. Divergence Divergence (generations) Based on mut. rate x 10-8 (Kong et al Nature, Sun et al Nat Genet) ~30K gen. ~20K gen. ~20K gen YRI CEU CHB YRI YRI CEU CEU CHB CHB NA18488 NA06989 NA18597 Keinan et al Nat Genet

83 Outline 1. Introduction to Population Genetics 2. HapMap and HapMap2 projects 3. F ST 4. HapMap3 and 1000 Genomes projects

84 HapMap: 270 samples from 4 populations CEU northern European USA 90 CHB Chinese China 45 JPT Japanese Japan 44 YRI Yoruba Nigeria 90 Affymetrix and Illumina chips Perkel 2008 Nat Methods

85 The HapMap Project: Work is done, relax on beach?

86 Beyond HapMap: what the world still needs Larger sample sizes for analyses of linkage disequilibrium More complete representation of world population diversity e.g. South Asian and Native American genetic variation Analyses of copy number variation (CNV) Low-frequency variants (minor allele frequency <5%)

87 The International HapMap3 Project: 1,260 samples from 11 diverse populations International HapMap3 Consortium 2010 Nature

88 HapMap3: 1,260 samples from 11 populations CEU northern European USA 180 CHB Chinese China 90 JPT Japanese Japan 90 YRI Yoruba Nigeria 180 TSI Tuscan Italy 90 CHD Chinese USA 100 LWK Luhya Kenya 90 MKK Maasai Kenya 180 ASW African-American USA 90 MXL Mexican-American USA 90 GIH Gujarati-American USA 90

89 The HapMap3 project Larger sample sizes for analyses of linkage disequilibrium More complete representation of world population diversity e.g. South Asian and Native American genetic variation Analyses of copy number variation (CNV) Low-frequency variants (minor allele frequency <5%) International HapMap3 Consortium 2010 Nature

90 Data generation: SNPs and CNVs Affymetrix 6.0 array 900K SNPs 940K copy-number probes Illumina Infinium 1M array 1M SNPs, of which 80K targeted at CNV regions 1.5M SNPs passed QC in all populations (99.3% concordance for 250K SNPs on both arrays) Note: only 1.5M SNPs, versus 3.1 million SNPs in HapMap2 International HapMap3 Consortium 2010 Nature

91 Not all HapMap3 populations are similar to a population from HapMap HapMap3 population Closest pop. from HapMap F ST TSI (Tuscan) CEU CHD (Chinese) CHB LWK (Luhya) YRI MKK (Maasai) YRI 0.03 ASW (African-American) YRI 0.01 MXL (Mexican-American) CEU 0.04 GIH (Gujarati-American) CEU 0.04

92 Approaches to Scientific Understanding Love is Understanding. -- Madonna Data is Understanding. -- Alkes

93 CEU.ind: HapMap3 data: individual files NA06989 F CEU NA11891 M CEU NA11843 M CEU NA12341 F CEU NA12739 M CEU [sample ID] [sex] [popname]

94 CEU.snp: HapMap3 data: SNP files rs C T rs C T rs C T rs A G rs G A [SNP ID] [chr] [0.0] [position] [ref] [var]

95 CEU.geno: HapMap3 data: genotype files [Each line is 1 SNP, each column is 1 indiv.] [Number of copies of reference allele: 0 or 1 or 2. 9 denotes missing data.] Note: the HapMap3 data files for this course are restricted to ~700K SNPs that are common (MAF>5%) in every population.

96 Beyond HapMap: what the world still needs Larger sample sizes for analyses of linkage disequilibrium More complete representation of world population diversity e.g. South Asian and Native American genetic variation Analyses of copy number polymorphisms (CNV) Low-frequency variants (minor allele frequency <5%)

97 Common Disease/Common Variant hypothesis For common diseases, there will be one or a few predominating disease alleles with relatively high frequencies at each of the major underlying disease loci Lander 1996 Science; Reich & Lander 2001 Trends Genet reviewed in Gibson 2012 Nat Rev Genet, Visscher et al Am J Hum Genet

98 Are rare and low-frequency variants important? (to be continued, Thu of Week 6) Visscher et al Am J Hum Genet

99 Are rare and low-frequency variants important? (to be continued, Thu of Week 6) Gibson 2012 Nat Rev Genet

100 Are rare and low-frequency variants important? (to be continued, Thu of Week 6) Kaiser 2012 Science

101 HapMap3 1Mb pilot sequencing study and 1000 Genomes pilot projects HapMap3 pilot sequencing: kb regions spanning 1Mb (high coverage: Sanger sequencing) 692 individuals from 10 HapMap3 populations 1000 Genomes Trio pilot project: Genome-wide (high coverage: 42x) 6 individuals (one CEU trio and one YRI trio) 1000 Genomes Low-coverage pilot project: Genome-wide (low coverage: 2x-6x) 179 individuals from CEU, YRI, CHB, JPT populations 1000 Genomes Exon pilot project: 8,140 exons spanning 1.4Mb from 906 genes (high coverage: >50x) 697 individuals from 7 HapMap3 populations International HapMap3 Consortium 2010 Nature 1000 Genomes Project Consortium 2010 Nature

102 Sample size and SNP discovery (per Mb) International HapMap3 Consortium 2010 Nature

103 The 1000 Genomes (1000G) Project Sequence the entire genomes of 1,092 individuals: 379 of European ancestry (Europe and USA) 286 of East Asian ancestry (Asia) 246 of African ancestry (Africa and USA) 181 of Latino ancestry (Latin America and USA) Use next-generation sequencing technologies (~4x coverage): e.g. Illumina, 454, SOLiD (read lengths bp) (Metzker 2010 Nat Rev Genet, Davey et al Nat Rev Genet, also see Nielsen et al Nat Rev Genet) 1000 Genomes Project Consortium 2012 Nature

104 1000G project: Summary of main results 38 million SNPs discovered and successfully genotyped. Most of these are rare and low-frequency variants. The 38 million SNPs include 99.7% of all SNPs with minor allele frequency 5% 98% of all SNPs with minor allele frequency 1% *** 50% of all SNPs with minor allele frequency 0.1% based on an independent UK European sample. ***: stated goal to identify >95% of SNPs with frequency 1% was successfully achieved Genomes Project Consortium 2012 Nature

105 Common variants are shared across populations, but rare variants are often population-private 1000 Genomes Project Consortium 2012 Nature

106 1000G project: the final phase Sequence the entire genomes of 2,504 individuals: 503 of European ancestry (Europe and USA) 504 of East Asian ancestry (Asia) 661 of African ancestry (Africa and USA) 347 of Latino ancestry (Latin America and USA) 489 of South Asian ancestry (South Asia and USA) Use next-generation sequencing technologies (~7x coverage): Illumina only (read lengths bp only) 85 million SNPs, of which 64 million have MAF<0.5% Related resource: UK10K project: 7x WGS of 3,781 UK samples (UK10K Consortium 2015 Nature; also see Gudbjartsson et al Nature) 1000 Genomes Project Consortium 2015 Nature

107 1000G project: the final phase Sequence the entire genomes of 2,504 individuals: 503 of European ancestry (Europe and USA) 504 of East Asian ancestry (Asia) 661 of African ancestry (Africa and USA) 347 of Latino ancestry (Latin America and USA) 489 of South Asian ancestry (South Asia and USA) Use next-generation sequencing technologies (~7x coverage): Illumina only (read lengths bp only) 85 million SNPs, of which 64 million have MAF<0.5% 1000 Genomes Project Consortium 2015 Nature; also see UK10K Consortium 2015 Nature, Gudbjartsson et al Nat Genet, McCarthy et al Nat Genet

108 What about rare variants? The 1000G project has identified most low-frequency variants (minor allele frequency 1%-5%). These variants can be placed on genotyping arrays or imputed (see Thu of Week 1)

109 What about rare variants? The 1000G project has identified most low-frequency variants (minor allele frequency 1%-5%). These variants can be placed on genotyping arrays or imputed (see Thu of Week 1) Rare variants: most have not been identified by 1000 Genomes! Must sequence disease samples directly. Past focus has been mostly on exome sequencing, but now shifting to whole-genome sequencing. (to be continued, Thu of Week 6) Kiezun et al Nat Genet, Tennessen et al Science, Pasaniuc et al Nat Genet, Purcell et al Nature, Do et al Nature, Cai et al Nature. Reviewed in Goldstein et al Nat Rev Genet, Lee et al Am J Hum Genet, Zuk et al PNAS

110 Conclusions Human populations are slightly genetically different. These differences may be important for disease mapping. (see Thu slides: Linkage Disequilibrium.) F ST quantifies differences between human populations. HapMap, HapMap2, HapMap3 and 1000 Genomes projects provide a valuable resource for common & low-frequency variants (but most rare variants have not yet been identified).

111 EPI 511, Advanced Population and Medical Genetics Week 1: Intro + HapMap / 1000 Genomes Linkage Disequilibrium

112 EPI 511: Course components Advance reading 1 required paper + 1 optional paper per course session Lecture + Discussion discussants: each student to sign up as discussant for 1 class

113 Outline 1. Introduction to Linkage Disequilibrium 2. LD and Tag SNPs 3. LD and imputation 4. LD and fine-mapping

114 Outline 1. Introduction to Linkage Disequilibrium 2. LD and Tag SNPs 3. LD and imputation 4. LD and fine-mapping

115 Linkage Disequilibrium Definition: Linkage Disequilibrium (LD) refers to correlations between genotypes of nearby markers.

116 Linkage Disequilibrium Definition: Linkage Disequilibrium (LD) refers to correlations between genotypes of nearby markers. Linkage Disequilibrium Association Studies Linkage Disequilibrium Linkage Mapping (reviewed in Ott et al Nat Rev Genet)

117 Linkage Disequilibrium: Example Individuals billion letters A A G A T T A A C G T T G G C C A A... A A G G T T A A C C T T G G C T A A... A A A A T T A A G G T T G G T C A A... A A G G T T A A C C T T G G T T A A... A A G A T T A A C G T T G G C T A A... A A G G T T A A C C T T G G C T A A... A A G A T T A A C C T T G G C C A A... A A G G T T A A C C T T G G T T A A... SNP 1 SNP 2 YES, in LD

118 Linkage Disequilibrium: Example Individuals billion letters A A G A T T A A C G T T G G C C A A... A A G G T T A A C C T T G G C T A A... A A A A T T A A G G T T G G T C A A... A A G G T T A A C C T T G G T T A A... A A G A T T A A C G T T G G C T A A... A A G G T T A A C C T T G G C T A A... A A G A T T A A C C T T G G C C A A... A A G G T T A A C C T T G G T T A A... SNP 1 SNP 2 SNP 3 YES, in LD NOT in LD

119 Linkage Disequilibrium: Example Individuals billion letters SNP 1 SNP 2 SNP 3 r 2 =1, in LD r 2 =0, NOT in LD r 2 is squared correlation

120 Linkage Disequilibrium: Example Individuals billion letters SNP 1 SNP 2 SNP 3 r 2 =1, in LD r 2 =0.7, partial LD r 2 is squared correlation

121 Linkage Disequilibrium: Example Individuals billion letters SNP 1 SNP 2 SNP 3 r 2 =1, in LD r 2 =0.7, partial LD r 2 is squared correlation

122 Genotypes vs. Haplotypes: phasing Genotypes Haplotypes Individuals PHASING Individuals Stephens et al Am J Hum Genet, Browning et al Nat Rev Genet, Williams et al Am J Hum Genet, Delaneau et al Nat Methods, Loh et al. 2016a Nat Genet, Loh et al. 2016b Nat Genet

123 Genotypes vs. Haplotypes: phasing Genotypes Haplotypes Individuals PHASING Individuals Stephens et al Am J Hum Genet, Browning et al Nat Rev Genet, Williams et al Am J Hum Genet, Delaneau et al Nat Methods, Loh et al. 2016a Nat Genet, Loh et al. 2016b Nat Genet

124 Genotypes vs. Haplotypes: phasing Genotypes Haplotypes Individuals PHASING Individuals Fact: r 2 between SNP1 and SNP2 (phased haplotype data) equals r 2 between SNP1 and SNP2 (unphased genotype data), assuming Hardy-Weinberg equilibrium holds

125 Linkage Disequilibrium: Haplotype Blocks Individuals SNP SNP SNP These 3 SNPs form a haplotype block with two main haplotypes 3 billion letters

126 LD with phased haplotypes: r 2 vs. D Slatkin 2008 Nat Rev Genet Consider two SNPs with frequencies p A and p B of alleles A, B. Let g A refer to # copies (0, 1) of allele A for the first SNP. Let g B refer to # copies (0, 1) of allele B for the second SNP. ) (1 ) (1 ) ( ) ( ) ( )] ( ) ( ) ( [ B B A A B A AB B A B A B A p p p p p p p g Var g Var g E g E g g E r

127 LD with phased haplotypes: r 2 vs. D Slatkin 2008 Nat Rev Genet Consider two SNPs with frequencies p A and p B of alleles A, B. Suppose p A < p B < 0.5. ) (1 ) (1 2 2 B B A A B A AB p p p p p p p r B A A B A AB p p p p p p D

128 LD with phased haplotypes: r 2 vs. D Slatkin 2008 Nat Rev Genet Consider two SNPs with frequencies p A and p B of alleles A, B. Suppose p A < p B < 0.5. r 2 and D are maximized when p AB = p A. 1 B A A B A AB p p p p p p D B A B B A A B B A A B A AB p p p p p p p p p p p p p r ) (1 ) (1 2 2

129 LD with phased haplotypes: r 2 vs. D Slatkin 2008 Nat Rev Genet Consider two SNPs with frequencies p A and p B of alleles A, B. Suppose p A < p B < 0.5. r 2 and D are maximized when p AB = p A. e.g. p A = 0.25, p B = 0.4, p AB = 0.25 => r 2 = 0.5, D = 1 1 B A A B A AB p p p p p p D B A B B A A B B A A B A AB p p p p p p p p p p p p p r ) (1 ) (1 2 2

130 LD with unphased diploid genotypes Consider two SNPs with frequencies p A and p B of alleles A, B. Let g A refer to # copies (0, 1, 2) of allele A for the first SNP. Let g B refer to # copies (0, 1, 2) of allele B for the second SNP. r 2 D [ E( g p p AB A A Var ( g g B p p A ) A p p A B E( g B A ) Var ( g 1 ) E( g B ) B )] 2... cannot be directly computed, since p AB relies on phased data! Slatkin 2008 Nat Rev Genet

131 Approaches to Scientific Understanding Love is Understanding. -- Madonna Data is Understanding. -- Alkes

132 200 kb Linkage Disequilibrium: Haplotype Blocks Haplotype blocks in 216kb region (MHC, chr 6) x-axis = y-axis = SNP position in region 100 kb 0 kb D and L are measures of LD (related to r 2 ) Red indicates high LD Black indicates low LD Also see Haploview program, Barrett et al Bioinformatics Slatkin 2008 Nat Rev Genet

133 Linkage Disequilibrium: Haplotype Blocks Europeans and Asians Africans Gabriel et al Science also see Reich 2001 Nature, Daly 2001 Nat Genet

134 Linkage Disequilibrium: Haplotype Blocks African chromosomes: 50% of the genome lies in haplotype blocks >22kb. Europeans and Asians: 50% of the genome lies in haplotype blocks >44kb. Longer haplotype blocks in Europeans/Asians due to out-of-africa population bottleneck: descended from small number of ancestors who left Africa kya. Gabriel et al Science also see Reich 2001 Nature, Daly 2001 Nat Genet

135 A brief history of modern humans Cavalli-Sforza & Feldman 2003 Nat Genet; also see Ramachandran et al PNAS, Mellars 2006 Science, Armitage et al Science, Henn et al PNAS

136 A brief history of modern humans, contradicted All non-african populations have ~2% of their genomes descended from Neanderthals. Melanesian populations have ~5% of their genomes descended from Denisovans, a relative of Neanderthals. Green et al Science, Reich et al Nature, Meyer et al Science, Sankararaman et al Nature, Vernot & Akey 2014 Science reviewed in Racimo et al Nat Rev Genet

137 Population bottlenecks increase LD population bottleneck population bottleneck Cavalli-Sforza & Feldman 2003 Nat Genet; also see Ramachandran et al PNAS, Mellars 2006 Science, Armitage et al Science, Henn et al PNAS

138 Population bottlenecks increase LD Individuals billion letters SNP 2 SNP 3 r 2 =0, NOT in LD r 2 is squared correlation

139 Population bottlenecks increase LD due to subsampling haplotypes (genetic drift) Individuals billion letters SNP 2 SNP 3 r 2 =0, NOT in LD r 2 is squared correlation

140 Population bottlenecks increase LD due to subsampling haplotypes (genetic drift) Individuals billion letters SNP 2 SNP 3 r 2 =0.5, partial LD

141 Population bottlenecks increase LD due to subsampling haplotypes (genetic drift) Individuals billion letters SNP 2 SNP 3 r 2 =0.5, partial LD r 2 is squared correlation

142 Population bottlenecks increase LD Average number of haplotypes per genomic region Conrad et al Nat Genet

143 Outline 1. Introduction to Linkage Disequilibrium 2. LD and Tag SNPs 3. LD and imputation 4. LD and fine-mapping

144 Linkage Disequilibrium and tag SNPs Direct association: genotype SNP1 in Cases and Controls. Cases Individuals Controls 3 billion letters SNP 1: causal SNP

145 Linkage Disequilibrium and tag SNPs Indirect association: genotype SNP2 in Cases and Controls. If SNP1 affects disease risk, then SNP2 will also be associated! Individuals Cases Controls 3 billion letters SNP 1 SNP 2 r 2 =1, in LD

146 Linkage Disequilibrium and tag SNPs Indirect association: genotype SNP3 in Cases and Controls. If SNP1 affects disease risk, then SNP3 will also be associated! Individuals Cases Controls 3 billion letters SNP 1 SNP 2 SNP 3 r 2 =0.7, partial LD

147 Linkage Disequilibrium and tag SNPs Theorem 2 (Pritchard and Przeworski 2001 Am J Hum Genet): If SNP1 is causal and LD(SNP1,SNP2) = r 2, then Power of an association study of SNP1 with N samples = Power of an association study of SNP2 with N/r 2 samples.

148 Linkage Disequilibrium and tag SNPs Theorem 2 (Pritchard and Przeworski 2001 Am J Hum Genet): If SNP1 is causal and LD(SNP1,SNP2) = r 2, then Power of an association study of SNP1 with N samples = Power of an association study of SNP2 with N/r 2 samples. Proof: Let g1 and g2 be genotypes of SNP1 and SNP2 respectively and π be phenotype, all normalized to mean 0 and variance 1. Armitage Trend Test (χ 2 = Nρ(g, π) 2 ; Armitage 1955 Biometrics).

149 Linkage Disequilibrium and tag SNPs Theorem 2 (Pritchard and Przeworski 2001 Am J Hum Genet): If SNP1 is causal and LD(SNP1,SNP2) = r 2, then Power of an association study of SNP1 with N samples = Power of an association study of SNP2 with N/r 2 samples. Proof: Let g1 and g2 be genotypes of SNP1 and SNP2 respectively and π be phenotype, all normalized to mean 0 and variance 1. Armitage Trend Test (χ 2 = Nρ(g, π) 2 ; Armitage 1955 Biometrics): SNP1 with N samples: Nρ(g1, π) 2 = NE(g1 π) 2 SNP2 with N/r 2 samples: (N/r 2 )ρ(g2, π) 2 = (N/r 2 )E(g2 π) 2 = (N/r 2 )E([rg1 + (g2-rg1)] π) 2 = (N/r 2 )E(rg1 π) 2 = NE(g1 π) 2. Q.E.D.

150 Linkage Disequilibrium: Haplotype Blocks Risk haplotype Case Case Case Case Case Control Control Control Control Control Question: Which SNP to genotype? Answer: Choose 1 SNP per haplotype block, and take advantage of indirect association!

151 Linkage Disequilibrium: Haplotype Blocks Risk haplotype Case Case Case Case Case Control Control Control Control Control Needed: a resource describing the haplotypes at each location in the genome.

152 The International HapMap Project: 270 samples from 4 populations CEU European USA trios YRI Yoruba Nigeria trios CHB Chinese China 45 unrelated JPT Japanese Japan 45 unrelated

153 Genetic differences between populations are small 50% frequency C allele of rs % frequency 51% frequency 11kb away on chr 1 A allele of rs % frequency

154 LD differences between populations are large! 50% frequency C allele of rs % frequency r 2 = kb away on chr 1 r 2 = % frequency A allele of rs % frequency

155 HapMap project: a resource for SNP tagging Individuals billion letters SNP 1 SNP 2 SNP 3 SNP1 tags this entire haplotype block at an r 2 of 0.7

156 HapMap project: a resource for SNP tagging How to select SNPs to genotype in an association study: Choose genomic region(s) of interest. Look up HapMap SNPs in the genomic region(s). Choose a subset of HapMap SNPs which tag haplotype blocks in the genomic region(s). (e.g. Tagger algorithm, de Bakker et al Nat Genet) Note: because LD patterns vary by population, it is important to choose tag SNPs using a HapMap population similar to the population in the association study.

157 HapMap project: a resource for SNP tagging How many tag SNPs are required? For the entire genome, the answer is: Thus, to choose tag SNPs at an r 2 of 0.8, we need roughly 1 SNP per 3kb in YRI, or 1 SNP per 5kb in CEU or CHB+JPT International HapMap Consortium 2007 Nature; also see Barrett et al Nat Genet, Smith et al Genomics, International HapMap Consortium 2005 Nature

158 Things aren t always what they seem

159 Things aren t always what they seem Estimating LD using a small number of HapMap samples may lead to overfitting. HapMap SNPs are not a random subset of SNPs.

160 Things aren t always what they seem Estimating LD using a small number of HapMap samples may lead to overfitting. HapMap SNPs are not a random subset of SNPs. Bhangale et al Nat Genet

161 Things aren t always what they seem According to International HapMap Consortium 2007 Nature: 82% of common SNPs are tagged at r2 0.8 by Affymetrix 6.0 According to Bhangale et al Nat Genet: 66% of common SNPs are tagged at r2 0.8 by Affymetrix 6.0 Bhangale et al Nat Genet

162 Multi-SNP tagging Haplotype [freq. 25% for each haplotype] (causal) SNP1 A A C C SNP2 A C C A SNP3 A C A C r 2 =0, NOT in LD

163 Multi-SNP tagging Haplotype [freq. 25% for each haplotype] (causal) SNP1 A A C C SNP2+3 A+A C+C C+A A+C r 2 =1, YES in LD

164 Multi-SNP tagging Pe er et al Nat Genet also see Zaitlen et al Am J Hum Genet

165 Outline 1. Introduction to Linkage Disequilibrium 2. LD and Tag SNPs 3. LD and imputation 4. LD and fine-mapping

166 What is imputation? Marchini et al Nat Genet, Howie et al PLoS Genet, Li et al Genet Epidemiol, Howie et al Nat Genet, Fuchsberger et al Bioinformatics

167 Marchini et al Nat Genet, Howie et al PLoS Genet, Li et al Genet Epidemiol, Howie et al Nat Genet, Fuchsberger et al Bioinformatics What is imputation??

168 Imputation: Why try? Increase power to detect disease association at untyped causal SNP (imputed causal SNP may have stronger association than tag SNP)

169 Marchini et al Nat Genet, Howie et al PLoS Genet, Li et al Genet Epidemiol, Howie et al Nat Genet, Fuchsberger et al Bioinformatics Imputation: Why try? r 2 = 0.8 Causal SNP

170 Marchini et al Nat Genet, Howie et al PLoS Genet, Li et al Genet Epidemiol, Howie et al Nat Genet, Fuchsberger et al Bioinformatics Imputation: Why try? Causal SNP

171 Imputation: Why try? Increase power to detect disease association at untyped causal SNP (imputed causal SNP may have stronger association than tag SNP)

172 Imputation: Why try? Increase power to detect disease association at untyped causal SNP (imputed causal SNP may have stronger association than tag SNP) Enable meta-analysis of studies on Affymetrix + Illumina chips

173 Imputation: Why try? Increase power to detect disease association at untyped causal SNP (imputed causal SNP may have stronger association than tag SNP) Enable meta-analysis of studies on Affymetrix + Illumina chips Improve genotype data quality

174 Imputation: Algorithms Hidden Markov Model (HMM) based approaches: IMPUTE (Marchini et al Nat Genet, Howie et al PLoS Genet, Howie et al Nat Genet) MACH (Li et al Genet Epidemiol) fastphase/bimbam (Scheet/Stephens 2006 AJHG, Servin/Stephens 2007 PLoS Genet, Guan/Stephens 2008 PLoS Genet) GEDI (Kennedy et al ISBRA) Localized Haplotype Clustering: BEAGLE (Browning/Browning 2007 AJHG, Browning/Browning 2009 AJHG) Likelihood-based approaches: UNPHASED (Dudbridge 2008 Hum Hered) SNPMStat (Lin et al AJHG) reviewed in Marchini et al Nat Rev Genet; also see Li et al ARGHG

175 Imputation: What do the algorithms output? Integer-valued genotypes at untyped SNPs e.g. genotype = 2 OR Continuous genotype dosages at untyped SNPs e.g. genotype dosage = 1.79 OR Continuous genotype probabilities at untyped SNPs e.g. genotype probabilities P(0) = 0.01, P(1) = 0.19, P(2) = 0.80

176 Imputation: People do it. reviewed in Marchini et al Nat Rev Genet; also see Li et al ARGHG

177 reviewed in Marchini et al Nat Rev Genet; also see Li et al ARGHG HMM-based imputation approaches hap1 hap2 hap3 hap4 hap5 Imp.??? Note: current paradigm is to first phase the data, then run imputation on phased data (Howie et al Nat Genet, Fuchsberger et al Bioinformatics)

178 reviewed in Marchini et al Nat Rev Genet; also see Li et al ARGHG HMM-based imputation approaches hap1 hap2 hap3 hap4 hap5 Imp. Note: current paradigm is to first phase the data, then run imputation on phased data (Howie et al Nat Genet, Fuchsberger et al Bioinformatics)

179 Measuring imputation accuracy Concordance rate: % of genotypes (or alleles) imputed correctly Natural analogue of genotyping error rate in QC analyses Concordance rate is often in the range of 95-99%. Squared correlation (r 2 ) between true and imputed genotype Natural analogue of r 2 between causal SNP and tag SNP r 2 << concordance rate, particularly for rare SNPs.

180 Measuring imputation accuracy Concordance rate: % of genotypes (or alleles) imputed correctly Natural analogue of genotyping error rate in QC analyses Concordance rate is often in the range of 95-99%. Squared correlation (r 2 ) between true and imputed genotype Natural analogue of r 2 between causal SNP and tag SNP r 2 << concordance rate, particularly for rare SNPs.

181 Measuring imputation accuracy Concordance rate: % of genotypes (or alleles) imputed correctly Natural analogue of genotyping error rate in QC analyses Concordance rate is often in the range of 95-99%. Squared correlation (r 2 ) between true and imputed genotype Natural analogue of r 2 between causal SNP and tag SNP r 2 << concordance rate, particularly for rare SNPs.

182 Measuring imputation accuracy Concordance rate: % of genotypes (or alleles) imputed correctly Natural analogue of genotyping error rate in QC analyses Concordance rate is often in the range of 95-99%. Squared correlation (r 2 ) between true and imputed genotype Natural analogue of r 2 between causal SNP and tag SNP r 2 << concordance rate, particularly for rare SNPs. Normalized difference between true and imputed allele frequency Measures whether imputation is biased towards ref or var allele

183 Imputation using HapMap data common SNPs imputed using HapMap2 CEU (N=120): r 2 = 0.95 (European-ancestry WTCCC samples, Affymetrix & Illumina chips) International HapMap3 Consortium 2010 Nature

184 Imputation using HapMap data common SNPs imputed using HapMap2 CEU (N=120): r 2 = 0.95 common SNPs imputed using HapMap3 CEU+TSI (N=410): r 2 = 0.96 (European-ancestry WTCCC samples, Affymetrix & Illumina chips) International HapMap3 Consortium 2010 Nature

185 Imputation using HapMap data x-axis: MAF<5% SNPs, imputed using HapMap2 CEU (N=120) y-axis: MAF<5% SNPs, imputed using HapMap3 CEU+TSI (N=410) International HapMap3 Consortium 2010 Nature

186 Imputation using HapMap data x-axis: MAF<5% SNPs, imputed using HapMap2 CEU (N=120) y-axis: MAF<5% SNPs, imputed using HapMap3 CEU+TSI (N=410) International HapMap3 Consortium 2010 Nature

187 Imputation using HapMap data x-axis: MAF<5% SNPs, imputed using HapMap2 CEU (N=120) y-axis: MAF<5% SNPs, imputed using HapMap3 CEU+TSI (N=410) International HapMap3 Consortium 2010 Nature

188

189 Low-coverage sequencing + imputation increases power vs. genotyping arrays Effective sample size of a GWAS with a $300,000 budget: Cost per sample Actual #samples Average imputation r 2 Effective #samples Illumina 1M array $ x sequencing $83* 3,60.81** 2, x sequencing $43* 7,00.64** 4,500 *Based on sample preparation cost of $30/sample, which is conservatively double the $15/sample reported by Rohland & Reich 2012 Genome Res, and on $133 per 1x sequencing (Illumina Network cost). **Imputation r 2 attained at Illumina 1M SNPs by downsampling reads from real off-target exome sequencing data. Relative performance of low-coverage sequencing will be even higher at non-illumina 1M SNPs. Pasaniuc et al Nat Genet; also see Cai et al Nature, Davies et al Nat Genet

190 Outline 1. Introduction to Linkage Disequilibrium 2. LD and Tag SNPs 3. LD and imputation 4. LD and fine-mapping (to be continued, Tue of Week 4)

191 Definition of fine-mapping Which of these SNPs on chr 6 is the biologically causal SNP? (Ditto for chr 5, 8, 12, 19) Manhattan plot from Ikram et al PLoS Genet

192 WTCCC fine-mapping study Maller et al Nat Genet

193 LD and fine-mapping in Europeans GWAS in Europeans SNP1: P-value = 10-8

194 TCF7L2 locus in T2D: 1 top signal Maller et al Nat Genet

195 LD and fine-mapping in Europeans Fine-mapping in Europeans SNP1: P-value = 10-8 CAUSAL?? SNP2: P-value = 10-8 CAUSAL??

196 FTO locus in T2D: many top signals Maller et al Nat Genet

197 LD and cross-population fine-mapping Fine-mapping in Europeans Fine-mapping in Africans SNP1: P-value = 10-8 SNP1: P-value = 10-5 SNP2: P-value = 10-8 SNP2: P-value = 0.62 SNP3: P-value = 0.41 SNP3: P-value = 10-5 r 2 LD in Europeans SNP1 SNP2 SNP3 SNP SNP SNP r 2 LD in Africans SNP1 SNP2 SNP3 SNP SNP SNP

198 LD and cross-population fine-mapping Fine-mapping in Europeans Fine-mapping in Africans SNP1: P-value = 10-8 SNP1: P-value = 10-5 CAUSAL SNP2: P-value = 10-8 SNP2: P-value = 0.62 SNP3: P-value = 0.41 SNP3: P-value = 10-5 r 2 LD in Europeans SNP1 SNP2 SNP3 SNP SNP SNP r 2 LD in Africans SNP1 SNP2 SNP3 SNP SNP SNP

199 LD and multi-ethnic fine-mapping Zaitlen*, Pasaniuc* et al Am J Hum Genet also see Morris 2011 Genet Epidemiol, Udler et al Hum Mol Genet, Wu et al PLoS Genet, Peters et al PLoS Genet, Liu et al Am J Hum Genet