VHL and clear cell Renal Cell Carcinoma...Renal Cell Carcinoma (RCC) • Originates in the renal cortex • Most common solid lesion occurring in the kidney (80‐85% of all primary

11/7/2014

1

October 23, 2014

Title

Gene expression profiles in renal cell carcinoma

W. Kimryn Rathmell, MD, PhD

VHL and clear cell Renal Cell Carcinoma

• VHL syndrome hallmark cancer:– Clear cell renal cell carcinoma (ccRCC)

• VHL mutations also a hallmark of sporadic ccRCC

• VHL mutationHIF stabilization and gene expression changes associated with the hypoxia response.

Renal Cell Carcinoma (RCC)• Originates in the renal cortex• Most common solid lesion occurring in the kidney (80‐85% of all primary renal neoplasms)

3

Diseased Kidney

Expression profiling reveals cell of origin

Cancer Genome Atlas Consortium, Nature, 2013

11/7/2014

2

VHL/HIF Regulatory PathwayHIF2α

HIF2α

HIFβ

HIF1α

HIF1α

HIF1α

HIFβ

VEGF

PDGF CCD1GLUT1

HIF2α

PDK

LDHAFLT1 TGFα

OCT4

Glycolysis

BNIP3

ApoptosisAngiogenesis Invasion

Metastasis

MMP2

CXCR4

De-differentiation

Proliferation

Outline

• Gene expression profiles in clear cell renal cell carcinomas.

• Validating a prognostic signature in RCC.• Exploring gene expression profiles in VHL syndrome RCC tumors.

Sporadic VHL mutant tumors express HIF1 and HIF2, or HIF2 alone.

Gordan et al, Cancer Cell, 2008

Gene expression patterns show HIF‐specific variability

• HIF1 drives glycolytic genes, mTOR pathway

• HIF2 drives genes involved in cell cycle and DNA damage.

• Overlap in angiogenesis and motility targets.

Gordan et al, Cancer Cell, 2008

11/7/2014

3

Using gene expression to identify tumor subgroups. ccA and ccB

Brannon, et al, Genes and Cancer, 2010

K=3, K=4 still fall into two groups


ccA (classical angiogenic), ccB (bad)


Validation in a historical dataset


11/7/2014

4

Extent of Disease at Diagnosis

• Most cancers of the kidney and renal pelvis are diagnosed when the disease is still localized to the primary site

13National Cancer Institute. SEER Stat Fact Sheets. Available at: http://seer.cancer.gov/statfacts/html/kidrp.html. Accessed August 28, 2008.

Loco-regional Spread 19%

Localized Disease 56%

Metastatic Spread 20%

Unknown 5%

Determining Prognosis: Anatomic Extent of Disease• Most consistent factor used to determine RCC prognosis

14Reprinted with permission from Tsui KH, et al. J Urol. 2000;163:1090-1095.

5-year Cancer-specific Survival Based on TNM Stage

TNM Stage

5-year Cancer-specific Survival

Stage I 91 ± 2.5%

Stage II 74 ± 6.9%

Stage III 67 ± 6.1%Stage IV 32 ± 3.2%

12 24 36 48 60 72 84 96 108 120 132

1.00

.75

.50

.25

0

Log Rank P Value<.001 Stage IV(N=318)

Stage III(N=83)

Stage II (N=57)

Stage I (N=185)

Months of Postsurgery

Prob

abili

ty o

f Sur

viva

l

ccA/ccB predicts for cancer specific and overall survival outcomes


Validation dataset reveals a small distinct tumor set

Brannon, et al, Eur Urol, 2012

11/7/2014

5

Cluster 3 are highly divergent from ccAand ccB in metabolic genes.


VHL mutants in both ccA and ccB


Developing a clinic tool: ClearCode34

LAD and ConsensusCluster analysisSet aside arrays with non-concordant assignments

43 ccA arrays(42 tumors, 1 replicate)

72 arrays (microarray standard set)(69 tumors, 3 replicates)

29 ccB arrays(27 tumors, 2 replicates)

Prediction Analysis for Microarrays (PAM)

95 clear cell tumors

Predictive biomarkers:ClearCode34

Brooks, et al, Eur Urol, 2014

Prognostic value of ClearCode34 evaluated in TCGA

0 20 40 60 80 100

0.0

0.2

0.4

0.6

0.8

1.0

Time (months)

0 20 40 60 80 100

0.0

0.2

0.4

0.6

0.8

1.0

Time (months)0 20 40 60 80 100

ccA ccB(n=205) (n=175)

No. of events 50 75Median RFS, months 91 53HR, 2.3; 95% CI, 1.6 to 3.3; P=4.3e-060.

0 0

.2

0.4

0.6

0.

8 1

.0 ccAccB

0 20 40 60 80 100

0.0

0.2

0.

4 0

.6

0.8

1.0

Can

cer-S

peci

fic S

urvi

val

(pro

babi

lity)

ccAccB

Time (months) Time (months)

Rec

urre

nce-

free

Sur

viva

l (p

roba

bilit

y)ccA ccB

(n=205) (n=175)No. of events 14 29Median CSS, months -- --HR, 2.9; 95% CI, 1.6 to 5.6; P=0.0005


11/7/2014

6

Prognostic value of ClearCode34 validated in UNC cohort

0 50 100 150 200

0.0

0.2

0.4

0.6

0.8

1.0

Time (months)

0 50 100 150 200

0.0

0.2

0.4

0.6

0.8

1.0

Time (months)

Can

cer-S

peci

fic S

urvi

val

(pro

babi

lity)

0.0

0.2

0.

4 0

.6

0.8

1.0

Time (months)0 50 100 150 200

ccAccB

ccA ccB (n=69) (n=88)

No. of events 7 25Median CSS, months -- 151HR, 3.0; 95% CI, 1.3 to 7.0; P=.005R

ecur

renc

e-Fr

ee S

urvi

val

(pro

babi

lity)

0.0

0.2

0.

4 0

.6

0.8

1.0

ccAccB

ccA ccB (n=69) (n=88)

No. of events 26 59Median RFS, months 88 52HR, 2.1; 95% CI, 1.3 to 3.4; P=.001

0 50 100 150 200Time (months)


Prognostic value of ClearCode34 validated in TCGA

Abbreviation: HR, hazard ratioSubtype ccA was used as reference in univariate and multivariate analysis.$ Stage I was used as reference in univariate and multivariate analysis. Stage was encoded as an ordinal variable with three levels.|| Grade 1 and 2 were combined and used as reference in univariate and multivariate analysis. Grade was encoded as an ordinal variable with three levels.


Integrated prognostic models can evaluate risk outcomes

Group Risk Score Low 0-0.5Intermediate 0.5-1.5High >1.5

0 50 100 150 200

0.0

0.2

0.4

0.6

0.8

1.0

Time (months)

Rec

urre

nce-

free

Sur

viva

l (p

roba

bilit

y)0.

0 0

.2

0.4

0

.6

0.8

1.

0

0 50 100 150 200Time (months)

Log-rank P=3.04e-09

Low Risk (N=98)Intermediate Risk (N=140)High Risk (N=28)

0 50 100 150 200

0.0

0.2

0.4

0.6

0.8

1.0

Time (months)

Log-rank P=2.03e-08

Low Risk Intermediate RiskHigh Risk

0.0

0.2

0

.4

0.6

0

.8

1.0

Time (months)

Can

cer-S

peci

fic S

urvi

val

(pro

babi

lity)

0 50 100 150 200


RCC Algorithms for cancer-specific survival

UCLA Integrated Staging System (UISS)

11/7/2014

7

ClearCode34 Model outperforms established algorithms

UISS ClearCode34 Model

C-in

dex

0.50

0.55

0.60

0.65

UISS ClearCode34 Model

.50

.5

5

.6

0

.6

5C

-inde

x

CC34Model->UISS

Cha

nge

in C

hi-s

quar

ed S

tatis

tic0

24

68

100

2

4

6

8

10

C

hang

e in

chi

-squ

ared

sta

tistic

Model->UI UI->Model7.3

1.9

2.5

6.7*

*CC34Model->SSIGN

Cha

nge

in C

hi-s

quar

ed S

tatis

tic0

510

1520

2530

35

SSIGN ClearCode34 Model

C-in

dex

0.50

0.55

0.60

0.65

0.70

SSIGN ClearCode34 Model

.50

.5

5

.

60

.65

.70

C-in

dex

0

5

10

15

2

0

25

30

35

Cha

nge

in c

hi-s

quar

ed s

tatis

tic

26.9

3.2

15.9

14.2*

*Model->SS SS->Model

*


ClearCode34 Summary

• ClearCode34 can accurately classify ccRCCtumors

• Prognostic value of ccA/ccB classification validated in a TCGA and UNC cohort

• Integrated model for recurrence-free and cancer-specific survival constructed using ccRCC subtypes and traditional clinical variables stage and grade.

• This classifier adds value to predicting cancer-specific survival above and beyond established algorithms.

How do we know this is meaningful for VHL patients?

*VHL disease tumors and sporadic ccRCCs (the majority VHL mutated) are not distinguishable by gene expression.

TCGA includes 14 tumors from VHL patients:2‐class comparison reveals NO significantly different upregulatedgenes, 209 significantly downregulated.

VHL cases in the TCGA‐A mixture of ccA and ccB

TCGA Sample ID ClearCode34 Status GradeSize (cm) T Stage N Stage M StatusStage

KIRC TCGA‐A3‐xxxx ccA G2 3 T1a NX M0 Stage IKIRC TCGA‐A3‐xxxx ccA G3 4.9 T1b N0 M0 Stage IKIRC TCGA‐A3‐xxxx ccB G2 4 T1a N0 M0 Stage IKIRC TCGA‐AK‐xxxx G2 5.5 T1 N0 M0 Stage IKIRC TCGA‐AK‐xxxx ccA G2 12 T3b N0 M0 Stage IIIKIRC TCGA‐AK‐xxxx ccA G2 7.5 T2 N0 M0 Stage IIKIRC TCGA‐AK‐xxxx G3 6.5 T3b N1 M0 Stage IIIKIRC TCGA‐AK‐xxxx ccB G3 11 T2 NX M0 Stage IIKIRC TCGA‐AK‐xxxx ccA G2 7 T1b NX M0 Stage IKIRC TCGA‐AK‐xxxx G2 9 T2 N0 M1 Stage IVKIRC TCGA‐AK‐xxxx G2 6 T1b NX M0 Stage IKIRC TCGA‐AK‐xxxx ccB G3 5.5 T3a NX M0 Stage IIIKIRC TCGA‐AK‐xxxx ccA G2 4 T1a NX M0 Stage IKIRC TCGA‐AK‐xxxx ccA G3 9 T2 N0 M0 Stage II

11/7/2014

8

Summary

• The hypoxia gene expression signature dominates VHL mutated RCC, but many other pathways emerge.

• The expression profile matches ccRCC to the early proximal tubule.

• Tumor expression profiles can reveal relevant biology and aid in disease prognosis .

AcknowledgmentsRathmell LabKate Hacker, PhDAlex Arreola, PhDSamira BrooksZufan Debebe, PhDSneha Sundaram, PhDCatherine FaheyAdam Sendor

Rathmell Lab Past MembersRose Brannon, PhDOishee SenShufen Chen, MD, PhDLance Cowey, MDCaroline Martz Lee, MD, PhDTricia Wright, PhDNeal Rasmussen, PhD

Translational Pathology LaboratoryGenomics Core, Tissue Procurement Facility

RutgersGyan Bhanot, PhDAnupama Reddy, PhD

Joel Parker, PhD

UNC Biomedical Research Imaging CenterWeili Lin, PhDAmir Khandani, MDJulia Fielding, MD

The Cancer Genome AtlasParticularly: Chad Creighton, Marston Linehan, Richard Gibbs, Kenna Shaw

Clinical TCGAPartiularly: James Hsieh, Ari Hakimi, Toni Chouieri

Funding: AACR INNOVATOR Award, NIH (TCGA), NIH (K24), V Foundation for Cancer Research

Documents

VHL and clear cell Renal Cell Carcinoma...Renal Cell Carcinoma (RCC) • Originates in the renal cortex • Most common solid lesion occurring in the kidney (80‐85% of all primary