Population structure of UK Biobank and ancient Eurasians

reveals adaptation at genes influencing blood pressure.

Kevin Galinsky

Harvard T.H. Chan School of Public Health

American Society of Human Genetics

October 21, 2016

Outline

1. Introduction

2. Population Structure

3. Natural Selection

2

Recent work was able to study population structure in the UK • People of British Isles (PoBI)

study contained 2,039 samples whose grandparents were born within 80km of each other

• Clustered into 17 groups using fineSTRUCTURE

• Sample size precluded natural selection analysis

Leslie et al. 2015 Nature 3

UK Biobank provides large dataset to detect population structure and natural selection

• Study follows 500k volunteers in UK

• Phase 1 data release contains 150k samples

• ~120k have UK ancestry

4

Sarkar, Webster & Galacher 2014 Healthy Cities: Public Health through Urban Planning

Previous work used PC-based approach to detect natural selection in 55k European Americans

Galinsky et al. 2016a AJHG 5

Population structure from FastPCA Natural selection along PC1

Outline

1. Introduction

2. Population Structure

3. Natural Selection

6

UK Biobank population structure analysis

7

SNPs Samples

847k 153k

Original 2101210k 113k

LD QC Filtering

Remove: Non-UK Ancestry Missing data SNPs Low MAF SNPs Missing data samples

LD pruning

HWE filter LD < 0.2 Prune Related

FastPCA

http://www.hsph.harvard.edu/alkes-price/software/

PCs

511k 119k

QC*

k-means clustering

Clusters

External Ref Pops

Project

Galinsky et al. 2016b AJHG

First 5 PCs contained interesting population structure

8

Galinsky et al. 2016b AJHG

210k SNPs × 113k samples FastPCA – 2 hours run time, 6.5GB memory

k-means clustering with 6 clusters produced stable clusters

9

Projecting PoBI samples identified the clusters Several north-south PCs, separate Welsh PCs

10

Leslie et al. 2015 Nature

Tracing UK ancestry to ancient populations

• 3 ancient populations thought to be origin of modern Europeans • Steppe

• Early farmers

• Hunter-gatherers

• Approach: project ancient populations onto UK Biobank PCs

Lazaridis et al. 2014 Nature Haak et al. 2015 Nature Mathieson et al. 2015 Nature

Steppe

Early Farmers

Hunter-gatherers

11

Ancient populations projected along north-south PCs; Steppe projected farther north

12

Outline

1. Introduction

2. Population Structure

3. Natural Selection

13

Unusual population differentiation is a powerful way to detect recent selection Sequencing of 50 Human Exomes Reveals Adaptation to High Altitude Yi et al. 2010 Science One single-nucleotide polymorphism (SNP) at EPAS1 shows a 78% frequency difference between Tibetan and Han samples, representing the fastest allele frequency change observed at any human gene to date.

14

Unusual population differentiation is a powerful way to detect recent selection The Impact of Natural Selection on an ABCC11 SNP Determining Earwax Type Ohashi, Naka & Tsuchiya 2011 Mol Bio & Evol

In addition, we show that absolute latitude is significantly associated with the allele frequency of rs17822931-A in Asian, Native American, and European populations, implying that the selective advantage of rs17822931-A is related to an adaptation to a cold climate.

15

Unusual population differentiation is a powerful way to detect recent selection

Greenlandic Inuit show genetic signatures of diet and climate adaptation Fumagalli et al. 2015 Science

To detect signals of positive selection, we used the population branch statistic (PBS), which identifies alleles that have experienced strong changes in frequency in one population relative to two reference populations.

16

UK Biobank natural selection analysis

SNPs Samples http://www.hsph.harvard.edu/alkes-price/software/

17

LD pruning 2101210k 113k

LD

FastPCA

PCs

511k 119k

QC* 511k 113k

QC QC SNPs + Unrelated

Selected SNPs

External Selection Statistic

Combined Selection Statistic

PC-based Natural Selection

Evaluating SNPs for unusual differentiation along principal components • Previous work compared

allele frequencies in discrete populations

• Statistic can be extended to SNP loadings – correlation of PC and genotypes

• 𝑀𝑈𝑖𝑘2 =

𝑀 𝑋𝑖𝑉𝑘2

Σ2=

𝑋𝑖𝑉𝑘2

Λ𝑘~𝜒1

2

18

Bhatia et al. 2011 AJHG

An

cest

ral

Pop

ula

tio

n

𝑝

Pop 1 𝑝1

Pop 2 𝑝2

𝐹𝑆𝑇 𝑝1 − 𝑝2

Sample 𝑋: genotypes

𝑋 = 𝑈Σ𝑉 𝑋𝑇𝑋

𝑀= 𝑉𝑇Λ𝑉

Galinsky et al. 2016a AJHG

Detect selection at FUT2 within UK PC1 selection statistic for each SNP in the genome

19

FUT2 protects against norovirus

• Fucosyltransferase 2 is part of the Lewis blood group

• Top hit rs601338 is coding variant, rs601338*A codes for nonsecretor allele (more common in northern UK)

• Associations: • Protects against Norwalk Norovirus (Thorven et al. 2005 J Virol)

• Vitamin B12 level (Hazra et al. 2008 Nature)

• Crohn’s disease (McGovern et al. 2010 Hum Mol Genet)

• Celiac & IBD (Parmar et al. 2012 Tissue Antigens)

20

Combine our selection statistic with ancient Eurasian selection statistic for more power • Compares four modern and

three ancient populations

• Model each modern population as mixture of ancient ones

• Detects allele frequency deviations from expected

• Produces 𝜒42 statistic – adding to

PC stat produces 𝜒52 statistic

21

Mathieson et al. 2015 Nature

Find two new signals at F12 and CYP1A2/CSK Ancient selection scan combined with PC2 selection statistic for each SNP in the genome

22

Coagulation factor XII (F12) was detected under this combined scan • rs2545801 suggestive in ancient population scan (𝑝 = 5.3 × 10−8)

• Associated with activated partial thromboplastin time (Tang et al. 2012 AJHG)

• Associations of second significant SNP rs2731672 • Expression of F12 in liver (Innocenti et al. 2011 PLoS Genet)

• Plasma levels of factor XII (Guerrero et al. 2011 Haematologica)

23

Mathieson et al. 2015 Nature

CSK (rs1378942) is known target of selection UK Biobank selection more recent

24 Ding and Kullo 2011 BMC Medical Genetics

CYP1A2/CSK is associated with blood pressure

• CSK rs1378942 associated with: • Blood pressure (Newton-Cheh et al. 2009 Nat Genet)

• Systemic sclerosis (Martin et al. 2012 Hum Mol Genet)

• CYP1A2 rs2472304 associated with: • Esophageal cancer (Xie et al. 2005 BMC Bioinformatics)

• Caffeine consumption (Cornelis et al. 2011 PLoS Genet)

• UK Biobank associations: • Diastolic blood pressure (DBP) (𝑝 = 3.6 × 10−19)

• Hypertension (𝑝 = 4.8 × 10−9)

25

ATXN2/SH2B3 associated with blood pressure

• Additional signal of selection found within ATXN2/SH2B3 LD block

• ATXN2 associated with blood pressure • DBP: 𝑝 = 8.00 × 10−33

• Hypertension: 𝑝 = 1.30 × 10−9

26

SH2B3

ATXN2

PTPN11

Mathieson et al. 2015 Nature The International Consortium for Blood Pressure Genome-Wide Association Studies 2011 Nature

PC3 & PC4 are associated with diastolic blood pressure Phenotype PC1 PC2 PC3 PC4

BMI 2.36 × 10−27 2.59 × 10−11

Height < 1 × 10−50 < 1 × 10−50

DBP 2.40 × 10−8 6.16 × 10−13

SBP 6.15 × 10−7

Asthma 2.50 × 10−6 2.89 × 10−5

27

Galinsky et al. 2016b AJHG

Conclusions

• UK Biobank dataset enabled study of subtle population structure

• PC-based selection statistic detected selection at FUT2

• Demonstrated new approach combining multiple selection statistics that follow 𝜒2 distribution

28

Acknowledgements

This research has been conducted using the UK Biobank Resource

Alkes Price Po-Ru Loh

Swapan Mallick David Reich Iain Mathieson

Nick Patterson

For more details, see Galinsky et al. 2016b AJHG!