ATO54260 EORTC Biomarkers 11-12_Layout 1 11/12/10 5:37 PM Page 1

Tivozanib Biomarker Identifies Tumor-infiltrating Myeloid Cells Contributing to Tivozanib Resistance in Both Preclinical Models and Human Jie Lin, Xiaojian Sun, Bin Feng, Feng Jiang, Richard Nicoletti, Guangmu Li, M. Isabel Chiu, Brooke Esteves, Pankaj Bhargava, Murray O. Robinson AVEO Pharmaceuticals, Inc., Cambridge, MA, USA.

*Presenting author.

Abstract Inter-tumor variation of breast tumor archives by microarray and aCGH Tivozanib mechanism of action in preclinical tumors Correlation matrix identifies a set of coherently expressed resistance genes Background: To identity biomarkers associated with tivozanib ● ● Variation in genomic ● response, a population-based, genetically engineered breast A total of 107 Her2-driven Heterogeneity in gene expression profiles (aCGH) Complete tumor growth inhibition Typical histologic change for blockade A key measure of the robustness of a signature is whether it retains consistent behavior across platforms breast tumors were collected profile observed of tumor model comprising 107 tumors was developed, ● Correlation heat maps were created to assess correlation among genes in any dataset characterized, and used to test the efficacy of tivozanib, a and propagated from the – All 107 tumors profiled at 1800 VEGFR-1, -2, and -3 kinase inhibitor that has shown clinical primary chimeric animals Vehicle activity in renal cell carcinoma (RCC) [ASCO 2009, Abstract baseline (in duplicate) 1600 Tivozanib 5 mpk qd Expression correlations among a list of genes

) 1400 Tivozanib 20 mpk qd represented as a two-dimensional matrix #5032]. 3 Correlation heat map approach m Necrotic m 1200

( 1. Input a gene list (signature, pathway gene set, X Y ⌺ ⌺

e center ⌺XY – Results: Twenty-five tumors from the archive were treated with 1000 biologically related genes, etc) 4 N m Y Y Y

r =

u

l 2 2 e e Different BH tumors exhibit distinct e Data points (⌺X) (⌺Y) o n tivozanib, revealing both responding and non-responding tumors n 2 2 800 n X Y v 2. Use linear regression to calculate correlation of – – e e (⌺ )(⌺ ) e

g g focal gains and losses in many human r N N g Regression

3 f f o (40% responders, 60% resistant). Bioinformatics analysis of RNA 600 Viable the expression values of each gene in the list f o o o

relevant genes: CDK6, Myc, PTEN m SS (xy) e e u rim within a given dataset (eg, Pearson, Spearman) e T u microarray expression profiles of pretreatment tumors identified a u

u r = l 400 l l a a a 2 v v SS (x) SS (y) v

set of 200 genes that were significantly associated with 3. Plot the correlation values as a matrix y = ␣ + ␤x + E n 200 n i i i n o o o i i (correlation heat map) to quantify the degree i s s resistance. A novel coherence-based bioinformatics approach s s s ⌺(x – x)(y – y) 0 s 1 ˆ i i e e

) ஶ e ␤ –

r of correlation of each gene to each other r r 2 r

n 13 15 17 19 21 23 25 p p incorporating multiple human tumor datasets led to a 42-gene p ⌺(x i – x) o x x w (coherence). Each dataset has its own E E

m Days o

resistance signature representing components of hematopoietic u h 0

t correlation matrix s

H gene expression. Immunohistochemistry (IHC) quantitation of 8 0.2 0.4 0.6 0.8 1 B 4. High-quality gene lists exhibit coherence across 1

( Vehicle Tivozanib 5 mpk myeloid markers in the tumors identified the presence of multiple datasets Expression value of gene X infiltrating myeloid cells, whose percentage composition in the tumor correlated with both the 42-gene signature and resistance Chromosome maps showing 200-gene signature does not retain high A subset of the signature genes are 42-gene biomarker retains correlated to tivozanib. Examination of both the signature and myeloid extent and degree of amplicons correlation in other datasets correlated in human breast cancer expression across many human tumor datasets (red) and deletions (green) Tivozanib efficacy studies across the tumor population infiltration in human tumor microarray datasets indicated that this Tivozanib breast Breast cancer Gene logic Tivozanib breast Breast cancer Gene logic model population NKI dataset kidney cancer model population NKI dataset kidney cancer resistant phenotype is present in a significant subset of all 7 enabled the development of a signature Purple: positive correlation Blue: negative correlation human tumor types examined, including human kidney cancer. Black: no correlation The preclinically derived IHC marker was then used to Diversity of breast archive in VEGFa expression and tumor microvasculature Vehicle Tivozanib 5 mpk qd Tivozanib 20 mpk qd retrospectively examine the relationship between myeloid BH line 1 BH line 2 BH line 3 infiltration and maximum tumor shrinkage achieved in available BH469 VEGFa expression (mRNA) Vessel number 2000 1800 2000 100 1800 1600 1800 A few genes exhibit

2 1600 significant correlated )

1600 ) )

m 80 1400 3 3 patient samples from a phase 2 study of tivozanib in RCC [ASCO 0.8 3 Gene logic Gene logic Gene logic Gene logic m m m m 1400 1400 expression in NKI / m 60 m 1200 m colon cancer lung cancer NCI cell lines ( ( s ( colon cancer lung cancer NCI cell lines

l Tivozanib- 1200 e 1200 e e 2009, Abstract #5032]. IHC analysis of infiltrating myeloid cells 0.6 e (human breast cancer

s 40 1000 m m m s u u u

l 1000 l

l 1000 e P ≤0.001) o o sensitive o V 20 800 v v v

800 800 r r in 21 patient samples demonstrated a significant correlation 0.4 r o o 0 o 600 m m 600 tumor class m 600

BH469 BH216 BH374 u u u T T T 400 between the percent myeloid cell composition in the tumors and 0.2 400 400 200 200 200

maximum tumor shrinkage by RECIST criteria. 0 Vessel cross section area 0 0 0 0510 15 20 25 30 13 15 17 19 21 23 25 0 5 10 15 20 25 30 35 40 6000 Days Days Days –0.2 l 5000 e s

Conclusions: These observations suggest tivozanib-sensitive s 4000 e

v 3000 Correlated genes among the 200 genes were

–0.4 /

2 BH line 4 BH line 5 BH line 6 2000 and -insensitive angiogenesis mechanisms in both murine and m

µ 4000 2000 2400 Standard statistical approaches to gene signatures used as a seeding set to identify a 33-gene Standard statistical approaches to gene signatures 300 268 374 469 270 271 394 215 228 216 214 1000 1800 human solid tumors, provide a candidate response biomarker for 0 3500 set of coherently expressed genes BH469 BH216 BH374 2000 typically perform poorly across tumor types typically perform poorly across tumor types

) 1600 ) ) 3 3 3 3000 m m m 1400 m m

tivozanib, and demonstrate the utility of this population-based m 1600 ( ( (

Tivozanib- 2500 e e

e 1200 m m

Microvessel density m u u u l l

l 2000 1000 1200

preclinical model in predicting response in humans. a resistant o o o v v e 12 v

r

r 800 r r 1500 a o o o 10 800 l tumor class m m

m 600 e u u u s

8 1000 T T T s 400 e 6 400 42-gene predictive gene set predicts tivozanib response in mouse tumors and is coherent in human tumors v

500

t 200

n 4

Chimeric inducible breast and lung tumor models e 0 0 0 Microvessel density = % microvessel cross section c 2 r 8910 11 12 13 14 15 16 17 18 0246810 12 14 16 18 036912 15 e

2 P 0 BH374 BH216 areas per tumor area (1 mm ) BH469 BH216 BH374 Days Days Days This biomarker platform enables analysis across any tumor dataset Quantitative “index” score for each of Introduce Inject stem cells Mice seeded 200 most significantly different genes between sensitive and resistant tumors 107 model tumors complex into 3-day-old with inducible About half Lung exhibits a Biology not genetic Inducible mouse embryos oncongenic

e of kidney high proportion present in Mouse alterations oncogenic tissue r Tivozanib preclinical and clinical summary o exhibits of predicted cell lines mouse Predicted c

embryonic s

stem cell t resistance resistance stem cell Tivozanib-sensitive resistant e s

e

gene set n e g

Supervised clustering e v i t c ● Extremely potent (~200 pM) Reported potencies of VEGFR inhibitors* using 200 genes Correlation matrix i d e

identified by T test demonstrating high r In vivo propagation of primary tumors enables against all 3 VEGFRs (1,2,3) 1000 Tivozanib-resistant correlation among Heatmap of each of the P drug testing Genotype for Her2-driven 5 weeks on gene set 42 genes across the Predicted breast tumor (BH) model: doxycycline 42 biomarker probes tumor population Colon Kidney Lung HCC Pancreas Gastric Primary tumor • MMTV rTTA ● Highly selective sensitive Breast s t e • tetO Her2 e s First propagation n

X 100 primary tumors e

• INK4A(-/-) e g

● 4.5-day t in human studies 100 n PCA mapping (85.2%) Gene correlation matrix Second 1/2 l e

• Luciferase reporter a g

} u propagation VEGFR-1 e Tivozanib response d i v i v t Archived tumor collection ● Robust efficacy in 272-patient Expression of 42 i VEGFR-2 c

Genotype for Kras-driven d )

Tivozanib no i n d i M

lung tumor (LK) model: response signature e f

VEGFR-3 r n Tivozanib yes phase 2 RCC trial o ( p • CCSP-rtTA

10 n

0 genes across the e 5 o h i

• tetO-KRASG 12V t s

C s I tumor population Principle Component Analysis n i • INK4A(-/-) – ORR: 27 %- 40% (27% in all e r h t p • Luciferase reporter using 200 genes separates i x w RCC, independent radiology sensitive tumors from E review; 40% in clear cell 1 resistant tumors ABT-869 vatalanib RCC/nephrectomy, investigator (AMG-706) (GW-786034) (PTK787) (ZD6474) Diversity in the histopathological features of breast tumors (AZD-2171) Myeloid cell density highly correlates with tivozanib Retrospective analysis from tivozanib phase 2 trial: median review) 0.1 (AG13736) tivozanib resistance in the tumor population tumor volume stratified by CD68 biomarker low vs high ● Histological phenotypes are heritable through at least – PFS: 14.8 months in patients (AV-951) *Either biochemical or cell based IC 50 s; cell-based data shown when available. 3 serial propagations with clear cell RCC who had Summary undergone nephrectomy ● Measuring the degree of myeloid cell infiltration into the ● Of the 272 patients in the phase 2 trial, only 21 samples had Selectivity of VEGFR inhibitors Histological subtypes of Her2+ breast adenocarcinomas tumor revealed a correlation with tivozanib resistance data for tumor volume assessment, were not randomized to o (n = 176) t ● We have generated a population-based, engineered mouse tumor model of

e v ) i placebo, and had available formalin fixed paraffin t

after propagation s

t Her2-driven breast cancer

a 9 ● Percent myeloid infiltration exhibits good correlation to l e k ● Safety profile consistent with e r c

embedded (FFPE) material of sufficient quality for analysis a y

r 8 multi-gene biomarker score (Pearson R = 0.62, P = 0.0004) c ● b Significant inter-tumor variation in histology, gene expression, and/or DNA copy n on-mechanism inhibition (

e t e 7 s o number as well as angiogenesis was observed across the tumor population a p

Myeloid cells in tumor 20X Myeloid percentage in sensitive vs resistant tumors Overall median = –23.8% n i 3 -

k 6

● Most common adverse events are

t d

e Low n ● Twenty-five tumor lines were tested for sensitivity to the VEGFR

g 60 Wilcoxon a r 5 BH469 hypertension and hoarseness , a

t rank sum test: - 2 f - inhibitor tivozanib to understand the mechanism of resistance and to explore 50 Myeloid high median = –13.7% f

, 4 P value = 0.029 o

CI 1 - 40 d

H H R predictive biomarkers

e Myeloid F r 3

N N u

G 30

s high E a V e 2 ● A tivozanib biomarker, comprising infiltrating myeloid cells, was discovered. 20 CH : 3 m y

t t i x O N v 1 10

i Retrospective analysis of 21 tumor samples from a phase 2 clinical trial in RCC e O t Myeloid n c

O High e 0 l e 0 showed significant correlation between the biomarker and response low e h BH229 t Solid sheet–like tumor Tumors with some Tumors with stromal s 9 5 4 8 7 1 5 0 6 1 4 4 0 6 0 9 5 8 9 8 3 1 4 5 3 4

[c-] [c-kit] [PDGFR] [PDGFR][c-kit] [EGFR] [c-kit] [raf1][c-kit] [Cst1R] 6 0 0 3 2 9 6 9 2 3 2 3 9 3 0 3 3 8 2 6 3 8 3 3 0 0 O 4 5 5 3 2 3 2 2 2 2 3 2 2 2 5 2 2 4 2 2 4 2 5 5 4 4 Myeloid low median = –33.5% d

l H H H H H H H H H H H H H H H H H H H H H H H H H H •HCI•H 2O B B B B B B B B B B B B B B B B B B B B B B B B B B morphology stromal involvement involvement showing CH o

3 F (well differentiated (well differentiated nested pattern (poorly tivozanib pazopanib motesanib cediranib sunitinib CH (AV-951) (GW-786034) (AMG-706) (AZD-2171) tumor cells) tumor cells) differentiated cells) 3 Sensitive Resistant O N axitinib vatalanib vandetanib sorafenib ABT-869 Acknowledgements –80 –60 –40 –20 0 20 40 (AG13736) (PTK787) (ZD6474) 45% of tumors examined 40% of tumors examined 15% of tumors examined This study was supported by AVEO Pharmaceuticals, Inc., Cambridge, MA. Best response % change in tumor volume Editorial assistance was provided by Kimberly Brooks, PhD, of MedErgy.

POSTER PRESENTED AT THE EORTC-NCI-AACR INTERNATIONAL SYMPOSIUM ON MOLECULAR TARGETS AND CANCER THERAPEUTICS, 16-19 NOVEMBER 2010, BERLIN, GERMANY. ATO54260 EORTC Biomarkers 11-12_Layout 1 11/12/10 5:37 PM Page 2

Tivozanib biomarker identifies tumor-infiltrating myeloid cells contributing to tivozanib resistance in both preclinical models and human renal cell carcinoma

Jie Lin, Xiaojian Sun, Bin Feng, Feng Jiang, Richard Nicoletti, Guangmu Li, M. Isabel Chiu, Brooke Esteves, Pankaj Bhargava, Murray O. Robinson

AVEO Pharmaceuticals, Inc., Cambridge, MA.

POSTER PRESENTED AT THE EORTC-NCI-AACR INTERNATIONAL SYMPOSIUM ON MOLECULAR TARGETS AND CANCER THERAPEUTICS, 16-19 NOVEMBER 2010, BERLIN, GERMANY.