Tools for analyzing cancer variation

Ekta Khurana, PhD Assistant Professor

Meyer Cancer Center Englander Institute for Precision Medicine Institute for Computational Biomedicine

Department of Physiology and Biophysics Weill Cornell Medicine, New York, NY

1

ekk2003@med.cornell.edu

@ekta_khurana

2

FirstcancerWGS,

Ley,Mardis,etal.,Nature,2008

~500cancerWGSAlexandrov,etal.,Nature,

2013

~3000WGSfromICGC/TCGA

2014

3

International Cancer Genome Consortium & The Cancer Genome Atlas

~3000WGS(tumor&normal),~1600RNA-Seq,~1500methylaQon

4

Most variants are

in noncoding

regions

Khurana et al, Nature Rev Genet, 2016

MB:medulloblastomaDLBC:BcelllymphomaSTAD:gastricBRCA:breastPAAD:pancreaHcPRAD:prostateLIHC:liverPA:pilocyHcAstrocytomaLUAD:Lungadenocarcinoma

5

CGGAGG

CGGAAG mRNA

Gain-of-motif TF

TF 0.0

1.0

2.0WT

Mutated

Loss-of-motif

Loss-of-motif

TATCTAT X

TF

TATTTAT TF

WT

Mutated

T0.0

1.0

2.0

CGT

ATG

TGCA5G

CTA

CTG

ATT10AG

TGA

G

ACT

CGAT

ACT15GTA

C

Altered binding effects 2.0

Promoter Gene

X

•  MYBmoHfcreated&drivesTAL1

overexpressioninT-ALL(Mansouretal,Science,2014)

Modes of action of noncoding variants: transcription factor binding disruption

TERT promoter mutated in many different cancer types

Killela et al, PNAS, 2013 Horn et al, Science, 2013 Huang et al, Science, 2013

Co-variates of mutation rates: Increased mutation density at TF binding sites in melanoma and lung

cancer

6

Perera et al, Nature, 2016 Sabarinathan et al, Nature, 2016 Khurana, Nature News & Views, 2016

Outline

•  Variants with high functional impact: FunSeq

•  Driver elements w/ more recurrent & high functional impact mutations than expected randomly: CompositeDriver

7

Khurana et al, Nature Rev Genet, 2016

IdenQfyingnoncodingvariants

associatedwithcancer

Khurana et al, Nature Rev Genet, 2016

IdenQfyingnoncodingvariants

associatedwithcancer

FunSeq

10

EvoluQonaryconservaQon- Typicallydefinedbycomparisonacrossspecies

ConservaQonamonghumans-DepleHonofcommonvariants/Enrichmentofrarevariants

1

2

4

3

Estimating negative selection

Commonvariant Rarevariants

FracHonofrarevariants=(Numofrarevariants/Totalnumofvariants)

Enrichment of rare SNPs as a metric for negative selection

11

0.55

0.60

0.65

0.70

0.4

0.5

0.6

0.7

0.8

0.9

A

Frac

tion

of ra

re S

NPs

(non

syn)

All C

odin

g

LOF-

tol.

Rece

ssive

GW

AS

Dom

inan

t

Esse

ntia

l

Canc

er

B

Frac

tion

of ra

re S

NPs

DHS Coding

Bone

Urot

heliu

mEm

br s

tem

cel

lBr

east

Bloo

dPa

ncr d

uct

Panc

reas

Skin

Eye

Live

rM

uscle

Bloo

d ve

ssel

Hear

tLu

ngBl

astu

laM

yom

etriu

mEp

ithel

ium

Pros

tate

Brai

nEm

br lu

ngG

ingi

val

Kidn

eySp

inal

cor

dFo

resk

inCo

nnec

tive

Adip

ose

Live

rCe

rvix

Trac

hea

Sple

enPr

osta

teEs

opha

gus

Lung

Test

isPl

acen

taBl

adde

rCo

lon

Hear

tTh

yroi

dTh

ymus

Kidn

eyBr

ain

Ova

ry

(rare=derivedallelefreq<0.5%)

•  Depletion of common polymorphisms in regions under selection

Negative selection restricts the allele frequency of deleterious mutations.

•  Results for coding genes consistent with known phenotypic impacts

•  Other metrics for selection •  EvoluHonaryconservaHon

(e.g.GERP)•  SNPdensity

(confoundedbymutaHonrate)

LOF-tol(Loss-of-funcQontolerant):leastnegaQveselecQonCancer:mostselecQon Khurana et al., Science, 2013

Organism-level negative selection in noncoding elements

12Khurana et al., Science, 2013

Negative selection and tissue-specificity of coding and noncoding regions

13

q Ubiquitously expressed genes and bound regions show stronger selection

q Differences in constraints amongst tissues q Constraints in coding genes and regulatory genes are

correlated across tissues

Which noncoding categories are under very strong “coding-like” selection ?

q  Topcategoriesamongranked102categories

q  BindingpeaksofsomegeneralTFs(egFAM48A)

q  CoremoHfsofsomeTFfamilies(egJUN,GATA)

q  DHSsitesinspinalcordandconnecHveHssue

~0.4%genomiccoverage(~top25)

~0.02%genomiccoverage(top5)

~400-fold

~40-fold

Enrichmentofknowdisease-causingmutaHonsfromHumanGeneMutaHon

database

14

Human regulatory network from ENCODE ChIP-Seq

Peak Calling (ChIP-Seq) Assigning TF binding sites to targets Filtering high confidence edges

~28K proximal edges

PotenHalDistalEdge

StrongProximalEdge

TF TF

Nodes119TFsand~9000

targetgenesEdges

28,000interacHons

UsingcorrelaHonwithexpressiondata 15Gerstein¶…..Khurana¶…., Nature, 2012 (¶ co-first authors) Yip et al, Genome Res, 2012

Gene essentiality and human regulatory network

16

Non-TF target

In-degree = 1

Out-degree = 5

TF

Gerstein¶…..Khurana¶…., Nature, 2012 (¶ co-first authors)

LoF-tolerant Essential

0.0

0.5

1.0

1.5

2.0

2.5

Reg

ulat

ory

degr

ee

Wilcoxon pvalue=1.29e-2

Totalde

gree(IN+OUT)

(logscale)

Essential genes tend to be central

Khurana et al., PLoS Comp. Bio., 2013

16

EssenHal

LoF-tolerant

Sizeofnodesscaledbytotaldegree

ZGumusiCAVEmovie

17

Identification of noncoding mutations with high impact: FunSeq

•  Feature weight - Weighted with mutation patterns in natural polymorphisms

(features frequently observed weighed less) - entropy based method

!! = 1+ !!!"#!!! + 1− !! !"#! 1− !! !!!

!!"#$%! = ! !!!!!!!"!!"#$%&$'!!"#$%&"'!

18

HOTregion

SensiHveregion

Polymorphisms

Genome

p=probabilityofthefeatureoverlappingnaturalpolymorphisms

Featureweight:

Foravariant:

wdp

FunSeq2:weightedscoringscheme

Fu et al., Genome Biology, 2014 hkps://github.com/khuranalab/FunSeq_PCAWGhkp://funseq2.gersteinlab.org

IdenQfyingnoncodingvariants

associatedwithcancer

CompositeDriverFunSeq

CompositeDriver for detecting driver coding & noncoding elements

(A)AlteraHonsarefuncHonallyannotatedbyFunSeq2pipelineFSi=originalFunSeq2score

(B)CalculateposiHonalrecurrenceofeachmutaHoninthecohortSample1Sample2Sample3

(C)WithineachfuncHonalregion,compositefuncHonalscore(CFSr)issumofrecurrencemulHpliedbyFunSeq2scoreineachposiHonwithalteraHon.

r = region (cds, promoter, enhancer and lincRNA)n = number of variants in rWi = number of samples with variant i

(D)P-valueforeachregionisproducedfrompermutaHontestandBenjaminiandHochbergmethodtocorrectmulHplehypothesistesHng.

FS1 FS3 FS4 FS5 FS7FS6FS2 FS8 FS9

20Eric Minwei Liu

Results from 40 lung adenocarcinoma samples Codingseq. Promoters

lincRNA

Expected(-logP)Expected(-logP)

Expected(-logP)

DatafromTCGA

0 1 2 3 4 5 60

1

2

3

4

5

6Q−Q plot for SNV's in promoter regions

Expected (−logP)

Obse

rved (−

logP

)

0 1 2 3 40

1

2

3

4Q−Q plot for SNV's in coding regions

Expected (−logP)

Obse

rved (−

logP

)

SPOP

0 1 2 3 4 5 6 70

1

2

3

4

5

6

7Q−Q plot for SNV's in enhancer regions

Expected (−logP)

Obse

rved (−

logP

)

Q-Q plot of SNVs in coding regions Q-Q plot of SNVs in promoters Q-Q plot of SNVs in enhancers

22

Results from 188 prostate cancer samples

DatafromICGC,BacaetalCell2013,BergeretalNature2011

Functional validation of candidates in prostate cancer

WDR74 promoter

q  Sanger sequencing in 19 additional samples confirms the recurrence

q  WDR74 shows increased expression in tumor samples

benign

PCa

23

RET promoterIncreased activity

In collaboration w/ Mark Rubin

EIF4EBP3 promoterReduced activity

24

Acknowledgements

Yale Yao Fu (now at Bina), Xinmeng Mu (now at

Broad), Jieming Chen, Lucas Lochovsky, Arif Harmanci, Alexej Abyzov,

Suganthi Balasubramanian, Cristina Sisu, Declan Clarke, Mike Wilson, Yong Kong, Mark

Gerstein

Sanger Vincenza Colonna, Yuan Chen, Yali Xue, Chris

Tyler-Smith

Cornell Steven Lipkin, Jishnu Das, Robert Fragoza,

Xiaomu Wei, Haiyuan Yu

Andrea Sboner, Dimple Chakravarty, Naoki Kitabayashi, Vaja Liluashvili,

Zeynep H. Gümüş, Kellie Cotter, Mark A. Rubin

U of Michigan Hyun Min Kang U of Geneva

Tuuli Lappalainen (NYGC), Emmanouil T. Dermitzakis

Baylor Daniel Challis, Uday Evani, Donna

Muzny, Fuli Yu, Richard Gibbs EBI

Kathryn Beal, Laura Clarke, Fiona Cunningham, Paul Flicek, Javier Herrero, Graham R. S. Ritchie

Boston College Erik Garrison, Gabor Marth

Mass Gen Hospital Kasper Lage, Daniel G. MacArthur,

Tune H. Pers Rutgers

Jeffrey A. Rosenfeld

FuncQonalInterpretaQon

Group~50parHcipants

~40InsHtutes~550parHcipants

Khurana lab Eric Minwei Liu Priyanka Dhingra Alexander Fundichely Tawny Cuykendall Andrea Sboner Mark Rubin Dimple Chakravarty Kellie Cotter Steve Lipkin Chason Lee

Sandra and Edward Meyer Cancer Center Englander Institute for Precision Medicine

Institute for Computational Biomedicine

PostdocposiQonsavailablekhuranalab.med.cornell.edu/jobs

ekk2003@med.cornell.edu