Expression-based variant impact phenotyping of somatic mutations in cancer

Angela Brooks, Ph.D. Alice Berger, Ph.D. Xiaoyun Wu, Ph.D.

Matthew Meyerson, M.D., Ph.D.

Jesse Boehm, Ph.D.

May 11, 2015

Cancer sequencing studies are underpowered to find all functionally relevant mutated genes

Lawrence et al. Nature 2014

No.

of s

ampl

es n

eede

d fo

r 90%

pow

er in

90

% o

f gen

es

20,000

10,000

5,000

3,000 2,000

1,000

500

300 200

100

50

0.1 0.2 0.3 0.5 Somatic mutation frequency (per Mb)

1 2 3 5 10 20

1% 2% 3% 5%

10%

Mutation rate

Rhabdoid Medulloblastoma

AML

Carcinoid

Neuroblastoma

CLL

Prostate

Breast Multiple myeloma

Ovarian

Kidney clear cell

GBM Endometrial

Colorectal

DLBCL

Esophageal adeno.

Head and neck Bladder

Lung adenocarcinoma Lung squamous

Melanoma

Sequencing studies are even more underpowered to find impactful variants from passengers

AML Bladder

Breast CRC

DLBCL Endometrial Esophageal

GBM

Head/Neck Kidney (CC) Lung adeno.

Lung squamous Melanoma

MM Neuroblastoma

Ovarian Prostate

www.tumorportal.org

0 100 200 300 400 500 600 700 800 900

EGFR (amino acid coordinates)

Approaches to high-throughput experimental variant impact phenotyping

Somatic variants from cancer genome studies

Expression profiling

Multiplexed tumorigenesis

Variant-specific ORFs and CRISPRs

Morphologic profiling

Genetic interactions

Lab of Jesse Boehm

Variant-specific reagents: open-reading frame (ORF) or CRISPR

Pathway-specific assays Gene-agnostic assays

Expression-based variant impact phenotyping

Somatic variants from cancer genome studies

Expression profiling

Multiplexed tumorigenesis

Variant-specific ORFs and CRISPRs

Morphologic profiling

Genetic interactions

Lab of Jesse Boehm

Pathway-specific assays Gene-agnostic assays

Variant-specific reagents: open-reading frame (ORF) or CRISPR

Testing functional impact of somatic mutations in lung adenocarcinomas

Lawrence et al. Nature 2014

No.

of s

ampl

es n

eede

d fo

r 90%

pow

er in

90

% o

f gen

es

20,000

10,000

5,000

3,000 2,000

1,000

500

300 200

100

50

0.1 0.2 0.3 0.5 Somatic mutation frequency (per Mb)

1 2 3 5 10 20

Rhabdoid Medulloblastoma

AML

Carcinoid

Neuroblastoma

CLL

Prostate

Breast Multiple myeloma

Ovarian

Kidney clear cell

GBM Endometrial

Colorectal

DLBCL

Esophageal adeno.

Head and neck Bladder

Lung adenocarcinoma Lung squamous

Melanoma

1% 2% 3% 5%

10%

Mutation rate

Approach: Compare gene expression changes upon introduction of WT and mutant variants

ORF library • 47 genes, 303 alleles • Missense and indel variants

found in >400 lung adenocarcinomas*

• KEAP1: 38 alleles + WT • KRAS: 12 alleles + WT • EGFR: 11 alleles + WT

96h

A549 lung adenocarcinoma

cell line

(luminex-based profiling of 1000

transcripts)

384 well plates & replicates

L1000 gene expression profiling WT

MUT

*Imielinski et al. 2012, TCGA Nature 2014

Broad Genetics Perturbation Platform Broad Connectivity Map

Comparing changes in gene expression can predict variant impact

rep 1

……

Likely inert: ARAF V145L

1 ……. . 8

rep 8

-1

0

1

Correlation (connectivity score)

rep MUT

MU

T

rep

WT

WT

MU

T

WT

WT WT vs. MUT MUT

Comparing changes in gene expression can predict variant impact

rep 1

……

1 ……. . 8

WT WT vs. MUT MUT

Functional impact:

ARAF S214F

rep 8

-1

0

1

Correlation (connectivity score)

Likely inert: ARAF V145L

MUT

MU

T

WT

WT

MU

T

MUT

MU

T

WT

MU

T

WT

WT

rep

rep WT

rep 1

rep 8

……

Likely inert: ARAF V145L

1 ……. . 8

Variant impact prediction is determined by a Kruskal-Wallis test (FDR 5%)

Impact: ARAF S214F

WT WT vs. MUT MUT 1

-1

0

row

med

ian

scor

e

1

-1

0

row

med

ian

scor

e

Corrected p-value = 0.10

Corrected p-value = 0.0001

WT

WT vs.

MUT MUT

rep 1

rep 8

……

Likely inert: ARAF V145L

1 ……. . 8

Impact direction score can predict loss-of-function or gain-of-function change

-1

0

1

connectivity score

Impact-GOF: CTNNB1 S33N

Impact-LOF: STK11 D194Y

Impact: ARAF S214F

WT MUT

Sparkler plot of variant impact phenotyping for 151 lung cancer somatic mutations

ARAF V145L

Ryo Sakai Bang Wong

Impa

ct d

irect

ion

scor

e

-log10(corrected p-value) Gain-of-function (n=23) Loss-of-function (n=49)

Likely inert (n=45) Impact (n=34)

-3

0

1

2

-2

-1

3

0 2 3 1 4

CTNNB1 S33N

STK11 D194Y

ARAF S214F

Gene-by-gene variant impact phenotyping

Known oncogenes

Known tumor suppressors

Genes of unknown function

Gain-of-function Loss-of-function

Likely inert Impact

CTNNB1 EGFR KRAS

KEAP1 STK11 FBXW7

DCAF8 TPK1

Gain- and change-of-function predictions of rare mutations in known oncogenes

Known oncogenes

Known tumor suppressors

Genes of unknown function

Gain-of-function Loss-of-function

Likely inert Impact

CTNNB1 EGFR KRAS

KEAP1 STK11 FBXW7

DCAF8 TPK1

CTNNB1

V600G G367V I140T F777S

S45F

G34V S33N

KRAS

G12A G12C G12D G12F G12R G12S G13V

G12Y

D33E

Loss-of-function mutations in known tumor suppressor genes

Known oncogenes

Known tumor suppressors

Genes of unknown function

Gain-of-function Loss-of-function

Likely inert Impact

CTNNB1 EGFR KRAS

DCAF8 TPK1

STK11

G56V

G251F G251V

D194Y G242W

FBXW7

D211N

P620R

V464E

Predicted impact of mutations in genes with unknown function

Known oncogenes

Known tumor suppressors

Genes of unknown function

Gain-of-function Loss-of-function

Likely inert Impact

CTNNB1 EGFR KRAS

DCAF8 TPK1

DCAF8

E63K

G588C

R559P P156S

TPK1

T213S P152T K111M

T205S

E81Q

G48C

L185I

How do we know these predictions are correct?

1) Literature Benchmarks - 100% accuracy with 21 previously studied mutations 2) Subsampling from high replicate experiment to assess false positive rate

- FDR 5% cutoff gives 4.84% false positive calls 3) Correspondence with genetic patterns (e.g. hotspot)

Functional mutation in FBXW7 is adjacent to a hotspot mutation

D211N

V464E

Impa

ct d

irect

ion

scor

e

-log10(corrected p)

0

-3

3

0 2 4 1 3

FBXW7

D211N

P620R

V464E

P620

How do we know these predictions are correct?

1) Literature Benchmarks

2) Simulation to assess false positive rate 3) Correspondence with genetic patterns (e.g. hotspot) 4) Signatures are similar to alleles of known gene function

ORF: 1

272

1 272 …………………………………………….

……

……

……

……

……

……

……

……

….

Correlation

1 0.8

-0.8 -1

EGFR/RAS pathway driver cluster

KEAP1-NFE2L2 anti-correlation

STK11 WT cluster

Identification of major transcriptional classes of lung adenocarcinoma genes

Rare variants cluster with known EGFR/RAS pathway drivers

Rare variants cluster with known EGFR/RAS pathway drivers

KRAS

D33E

Impa

ct d

irect

ion

scor

e

0

3

-3 -log10(corrected p) 0 4

Impa

ct d

irect

ion

scor

e

0

3

-3 -log10(corrected p) 0 4

ARAF

S214C S214F

Impa

ct d

irect

ion

scor

e

0

3

-3 -log10(corrected p) 0 4

Gain-of-function Loss-of-function

Likely inert Impact EGFR

K754E

S645C

Unique signature of a rare non-canonical EGFR mutation is currently being investigated

EGFR/RAS cluster

EGFR variants

Non-canonical signature

EGFR

H1129Y

Gain-of-function Loss-of-function

Likely inert Impact

How do we know these predictions are correct?

1) Literature Benchmarks

2) Simulation to assess false positive rate 3) Correspondence with genetic patterns (e.g. hotspot) 4) Signatures are similar to alleles of known gene function 5) Orthogonal assays - Genetic rescue screens, multiplex tumor xenografts, cell morphologic profiling

Variant impact phenotyping using a gene-agnostic assay

• Variant impact phenotyping is critical for understanding and treating cancer and other diseases.

• Gene-agnostic bioassays can more rapidly characterize

variant impact

• This approach can be used for any gene, regardless of disease type and regardless of knowledge of gene function.

Jesse Boehm Matthew Meyerson Todd Golub Yashaswi Shrestha Candace Chouinard John Doench Itay Tirosh Pablo Tamayo Bang Wong Ryo Sakai

Aravind Subramanian Larson Hogstrom Ted Natoli David Lahr Jackie Rosains John Davis Dan Lam Corey Flynn Josh Gould Rajiv Narayan Ian Smith Federica Piccioni Mukta Bagul Sasha Pantel Cong Zhu Thomas Green Dave Root Xiaoping Yang Steven Corsello

Meyerson Lab Tanaz Sharifnia Heidi Greulich Josh Campbell Jenny Chen Peter Choi Fujiko Duke Hugh Gannon Marcin Imielinski Bethany Kaplan Craig Strathdee Alison Taylor Luc DeWaal Xiaoyang Zhang Kasper Lage Johnathan Mercer Reza Kashani

Gaddy Getz Bill Hahn Nina Ilic Lihua Zou Atanas Kamburov Eejung Kim Mike Lawrence Anne Carpenter Shantanu Singh Mark Bray Paula Keskula Sara Seepo

Alice Berger Xiaoyun Wu

Jesse Boehm Matthew Meyerson Todd Golub Yashaswi Shrestha Candace Chouinard John Doench Itay Tirosh Pablo Tamayo Bang Wong Ryo Sakai

Aravind Subramanian Larson Hogstrom Ted Natoli David Lahr Jackie Rosains John Davis Dan Lam Corey Flynn Josh Gould Rajiv Narayan Ian Smith Federica Piccioni Mukta Bagul Sasha Pantel Cong Zhu Thomas Green Dave Root Xiaoping Yang Steven Corsello

Meyerson Lab Tanaz Sharifnia Heidi Greulich Josh Campbell Jenny Chen Peter Choi Fujiko Duke Hugh Gannon Marcin Imielinski Bethany Kaplan Craig Strathdee Alison Taylor Luc DeWaal Xiaoyang Zhang Kasper Lage Johnathan Mercer Reza Kashani

Gaddy Getz Bill Hahn Nina Ilic Lihua Zou Atanas Kamburov Eejung Kim Mike Lawrence Anne Carpenter Shantanu Singh Mark Bray Paula Keskula Sara Seepo

Alice Berger Xiaoyun Wu

Brooks Lab positions available in cancer transcriptome analysis and functional genomics at UC Santa Cruz! anbrooks@ucsc.edu