Heritability enrichment differentially expressed genes expressed genes ... • Adrenal Gland • Artery ‐Aorta ... – What should Y (window size around gene) be? 17

Heritability enrichment of differentially expressed genes

Hilary FinucanePGC Statistical Analysis Call

January 26, 2016

1

Functional genomics + GWAS gives insight into disease‐relevant tissues

See also Maurano et al. 2012 Science, Pickrell 2014 AJHG, Kichaev et al. 2014 PLoS Genet.

Trynka et al. 2013 Nature Genetics Farh et al. 2014 Nature

2

LD score regression uses the whole genome, improving power for polygenic traits.

Finucane*, Bulik‐Sullivan*, et al. 2015 Nature Genetics3

Gene expression + GWAS gives insight into disease‐relevant tissues

5

4

0

1

2

3

Obs

erve

d –l

og10

(p)

10 2 3 4

Expected –log10 (p)

5

4

0

1

2

3

Obs

erve

d –l

og10

(p)

10 2 3 4


10 2 3 4


5

4

0

1

2

3

120

80

0

40

12

8

0

4

Obs

erve

d –l

og10

(p)

Bipolardisorder

Crohn’sdisease

Type 1diabetes

Type 2diabetes

Rheumatoidarthritis

Hypertension

Coronaryarterydisease

Whole bloodMuscleLungNerve, tibialThyroid

AdiposeArtery, tibialHeartSkin

10 2 3 4


5

5

4

0

1

2

3O

bser

ved

–log

10 (p

)

Hu et al. 2011 AJHG

GTEx Consortium, 2015 Science4

Our goal: combine ENCODE/Roadmap data with gene expression data to identify

disease‐relevant tissues

• By combining regulatory marks with gene expression data, we should be able to:– Replicate results in orthogonal data– Increase power

• Using LD score regression will increase power for very polygenic traits.

• First step: how much signal is there in GTEx, and how can we get it out?

5

Outline

• Review of stratified LD score regression.

• Differentially expressed genes in GTEx.

• Validation: Brain v Blood in SCZ and RA.

• Results in 7 Brain‐related traits and RA.

6

Outline





7

Stratified LD score regression• Our model is

Where • Y is an individual’s phenotype, • Xj is an individual’s genotype at the j‐th SNP(normalized to mean 0 and variance 1),

• βj is the effect of SNP j, and • ε is noise and random environmental effects.

Finucane*, Bulik‐Sullivan* et al. 2015 Nat Genet8

Stratified LD score regression• We model SNP effects as random and mean 0.• For disjoint categories,

where

• Intuition: LD to a category that is enriched for heritability will increase the chi‐square statistic of a SNP more than LD to other categories.


Stratified LD score regression with overlapping categories

• Var(βj) = Expected per‐SNP heritability of SNP j.• With disjoint categories, if SNP j is in category C,

• With overlapping categories,

• i.e., τc is the contribution of category C to per‐SNP heritability of SNPs in category C, after controlling for all other categories in the model.


• With overlapping categories,

Stratified LD score regression:• Estimate LD scores from a reference panel with matching LD.

• Perform (weighted) regression of chi‐square on LD scores to estimate the

• Optional: convert back to heritability.

Stratified LD score regression with overlapping categories


Stratified LD score regression to identify disease‐relevant tissues

• Have a baseline model with many annotations (exon, DHS, H3K27ac, “weak enhancer,” etc.)

• Add tissue‐specific annotation; e.g. H3K27ac in Liver.

• Test for positive for this annotation.


Finucane*, Bulik‐Sullivan* et al. 2015 Nat Genet

Stratified LD score regression identifies disease‐relevant tissues

Results with Roadmap/ENCODE data:

To apply stratified LD score regression to GTEx, have to define new tissue‐specific annotations.

13

Outline





14

GTEx tissues• Brain ‐ Amygdala• Brain ‐ Anterior cingulate cortex

(BA24)• Brain ‐ Caudate (basal ganglia)• Brain ‐ Cerebellar Hemisphere• Brain ‐ Cerebellum• Brain ‐ Cortex• Brain ‐ Frontal Cortex (BA9)• Brain ‐ Hippocampus• Brain ‐ Hypothalamus• Brain ‐ Nucleus accumbens (basal

ganglia)• Brain ‐ Putamen (basal ganglia)• Brain ‐ Spinal cord (cervical c‐1)• Brain ‐ Substantia nigra

• Adipose ‐ Subcutaneous• Adipose ‐ Visceral (Omentum)• Adrenal Gland• Artery ‐ Aorta• Artery ‐ Coronary• Artery ‐ Tibial• Breast ‐Mammary Tissue• Cells ‐ EBV‐transformed

lymphocytes• Cells ‐ Leukemia cell line (CML)• Cells ‐ Transformed fibroblasts• Colon ‐ Transverse• Esophagus ‐Mucosa• Esophagus ‐Muscularis• Fallopian Tube• Heart ‐ Atrial Appendage• Heart ‐ Left Ventricle

• Kidney ‐ Cortex• Liver• Lung• Muscle ‐ Skeletal• Nerve ‐ Tibial• Ovary• Pancreas• Pituitary• Prostate• Skin ‐ Not Sun Exposed (Suprapubic)• Skin ‐ Sun Exposed (Lower leg)• Stomach• Testis• Thyroid• Uterus• Vagina• Whole Blood

• 46 tissues total– Average of 66 samples per

tissue– min = 1 sample– max = 607 samples

• Of these, 13 are brain tissues– Average of 42 samples per

tissue– min = 28 samples– max = 60 samples

15

Defining the annotation

• For each tissue:– Rank genes by differential expression.– Choose the top X genes in this ranking.– Include these genes, with a window of Y kb.

16



• Choices:– How to rank?– What should X (# genes) be?– What should Y (window size around gene) be?

17




18

We rank genes using HKNorm and a t‐test for differential expression.

• We normalize RPKM using the HKNorm algorithm [Byrnes et al. in prep].

• For each gene, we compute a t statistic for whether HKNorm expression levels are different for samples in the tissue vs. not in the tissue.

• Rank by the t statistic.

19




20

Choosing #genes and window size.• How many genes to choose?

– We try top 2%, 5%, 10% of all genes.– After QC, this gives us 369, 8921, and 1842 genes, respectively.

• What window size?– We try 20kb and 100kb from transcribed region.– For all genes, this gives us 45% and 58% of the genome, respectively.

• In total, we try 3 x 2 = 6 parameter settings.• Note: For LD score, want an annotation that covers at least roughly 1% of the genome.

21

Outline





22

Brain annotation• Brain = union of 13 brain tissues

• GO enrichment (10% set):

Blood annotation• Blood = Whole blood and LCLs

• GO enrichment (10% set):

23

Brain v Blood, RA v SCZ: should use 10% of genes with a 100kb window

24

Outline





25

Brain vs. Blood in many traits also identifies Brain for Bipolar Disorder

26

Schizophrenia

27

Bipolar Disorder

28

BMI

29

Age at menarche

30

Rheumatoid Arthritis

31

Brain vs Blood in many traits also identifies Brain for Bipolar Disorder

32

Previous anlaysis with histone marks has higher power

33

Gene sets derived from histone marks have similar power to gene sets

derived from GTEx.

34

Future directions

• Improve power by combining signal from multiple tissue‐relevant annotations (histone marks, gene sets, etc.).– Still method in progress. – One idea: for a single tissue, add histone marks, gene expression, etc. to the model together. Test for enrichment using the sum of the coefficients. (Or some other linear combination.)

• Difficulty: need data on the same cell types, or to group cell types together.

35

Acknowledgements

• Alkes Price• Ben Neale• Andrea Byrnes• Verneri Anttila• Samuela Pollack• Yakir Reshef• The GTEx Consortium

36

Documents

Heritability enrichment differentially expressed genes expressed genes ... • Adrenal Gland • Artery ‐Aorta ... – What should Y (window size around gene) be? 17