1. Preclinical strategies to help better identify responder populations in the clinic
NorCal SOT Fall Symposium: New Frontiers in Oncology Drug Development
September 27, 2012
Jocelyn Holash Novartis Institutes for Biomedical Research

2. Building a robust pre-clinical translational infrastructure
Limitations of existing preclinical model systems
The number of models used in preclinical evaluation typically under-represents the number of distinct tumor types
Models often inadequately characterized at the molecular level (making alignment to human disease impossible)
In vitro selection pressures alleviate dependence on key oncogenic pathways
In vitro models do not replicate stromal-tumor cell interactions
Use of immuno-compromised animals eliminates the ability to study modulators of antitumor immunity, and suffer from cross-species imperfections in ligand-receptor interactions etc
Low throughput of in vivo models limits the number of tumor types that can be tested
| NorCal SOT Fall Symposium | Holash | September 2012

3. Building a robust pre-clinical translational infrastructure
Improving indication selection and patient stratification – what can we do pre-clinically ?
Human Cancer
Model Systems
Therapeutic-Profiling
Build databases from public sources
Fill gaps with external and internal efforts: - Sequencing project
Cancer Cell Line Encyclopedia
Murine tumor allograft models
Primary tumor models
CLiP (cell line profiling)
Drug Combinations Zalicus screen
Drug resistance
shRNA pooled screens
| NorCal SOT Fall Symposium | Holash | September 2012

4. Human cancer databases provide the foundation
3 key human cancer databases
integrate internal and external (GEO, TCGA, COSMIC) data
meticulous sample curation
consistent QC and data normalization
accessible for analysis and visualization
Human Cancer
OncExpress
Onc*Base
OncCNA
N= ~50,000 samples
Expression data for tumors, models and normal tissue
Analytical tools: Prevalence analysis, Outlier analysis, Correlation analysis
N= ~580,000
Allows retrieval of mutation data in human cancer
Analytical tools: Find gene mutation frequency, recurrent gene mutations, lineage patterns, structure
N = ~6,000
Copy number data base derived from SNP array analysis
Analytical tools: Find recurrent chromosome alterations, delineate regions of conserved copy number change, correlate with expression
| NorCal SOT Fall Symposium | Holash | September 2012 4

5. Cancer Cell Line Encyclopedia: 1000 cell lines annotated by molecular profiles and response Goal: to identify predictive biomarkers for patient selection
Expression profiling Affy U133 Plus2 array
Compound Sensitivity
2000 cpd x 500 lines x 8 pt IC50
shRNA Sensitivity
145 cell lines x ~2000 genes
Copy number/LOH analysis, genotyping:
Affy SNP6.0 array
Mutation analysis:
Oncomap Ver 3.0, 353 mutations (Sequenom iPlex)
Hybrid (exon) capture ~1600 genes (Illumina)
| NorCal SOT Fall Symposium | Holash | September 2012

6. Cell lines model the genetic diversity of tumors
| NorCal SOT Fall Symposium | Holash | September 2012

7. CCLE portal http://www.broadinstitute.org/ccle
| NorCal SOT Fall Symposium | Holash | September 2012

8. Progress towards targeting the RAS/RAF/MAPK pathway
LGX818 Potency
Growth Factors
pERK
Ki67
Kinase Assay
IC50 (nM)
BRAFV600E
0.4
BRAF
0.5
CRAF
0.3
Cell-Based – A375
pERK 3
Proliferation 4
RTKs
Vehicle
Grb2/SOS
10 mg/kg
RAS
A375 (BRAFV600E) xenograft
RAF
MEK
ERK
| NorCal SOT Fall Symposium | Holash | September 2012

9. The Cancer Cell Line Encyclopedia: CLiP
Therapeutic Profiling
LGX818
Cpds (2000)
Sensitive Insensitive
Cell Lines (>500)
CliP: Large-scale compound profiling using CCLE cell lines
| NorCal SOT Fall Symposium | Holash | September 2012

10. LGX818 is a potent and highly selective RAF kinase inhibitor
LGX818 Cellular GI50(uM) across 501 cell lines 1-
0.1-
0.01-
| NorCal SOT Fall Symposium | Holash | September 2012

11. LGX818 is selective for cells expressing BRAFV600E/D/K
Sensitivity is greatest within melanoma and CRC lineages
Melanoma and CRC
V600E/D/K lines are the
most sensitive
IC50 (uM)
BRAF mutation
status
V600E
V600D/K
Non -V600
Wildtype
| NorCal SOT Fall Symposium | Holash | September 2012

12. A genotype centric view of the “cube” (CLE) Pharmacologic sensitivity for 1300 compounds in >500 cell lines
BRAF V600E mutated cell line
Cpds (1300)
Sensitive Insensitive
Cell Lines (>500)
| NorCal SOT Fall Symposium | Holash | September 2012

13. BRAF mutant cancer cell lines are more sensitive to RAF or MEK inhibitors
BRAF mut Selectivity
Compound Rank
MEK inh
BRAF mut Selectivity
RAF inhib
Class Mechanism of Action Rank
| NorCal SOT Fall Symposium | Holash | September 2012

14. A Computational Framework for Identifying Predictive Biomarkers
Assign Response class
Copy number
Explicit feature selection (Fisher or Wilcoxon + local FDR)
Expression
Categorical machine learning and cross-validation
(predict class)
Mutation
Lineage
IC50
Output of predictive “features”
Effect vs control
AUC
Amax
Concentration (log)
| NorCal SOT Fall Symposium | Holash | September 2012

15. Genetic features identified as predictors of sensitivity
Cell Lines
Compound
Target(s)
Feature
Rank
LGX818
BRAF
BRAF mut 1*
RAF265
lapatinib
EGFR, ERBB2
ERBB2 amp 1
BYL719
PI3Ka
PIK3CA mut PD
MEK
10*
AZD6244 PF
MET
Met amp 3
Features
LGX818 “feature matrix” 50K features #1 predictive feature: BRAF mutation
*additional MAPK pathway features in top 50 (DUSP, SPRY, ETV, NRAS for MEK inhibitors)
| NorCal SOT Fall Symposium | Holash | September 2012

16. Finding synthetic lethality through shRNA screening Pooled shRNA screens across many cell lines
PooledshRNA POOLS
T24
Count
Cell Line
Count by NextGen Sequencing
Day 0
Day 7
Day 14
BRAF
MEK1
ERK2
Growth: Inhibition Induction
| NorCal SOT Fall Symposium | Holash | September 2012

17. Towards the systematic study of combination effects
Large-scale systematic discovery of combination activity on-going in collaboration with Zalicus.
Mid-scale hypothesis directed screens enabled through a global effort e.g. PI3K-RAS pathway combination screens across melanoma and CRC cell lines
Combinations
Mutations
| NorCal SOT Fall Symposium | Holash | September 2012

18. Building predictive preclinical models
Primary human tumor xenograft models
Primary Tumor
Implant
Rationale/Goal
Human tumors propagated in culture undergo artificial selection
e.g. p16 deletion increases
loss of Hedgehog signaling
loss of Wnt signaling
Primary propagation in nude mice may lead to more predictive models or provide models that otherwise are very limited
e.g. Pancreatic xenograft models vs. pancreatic cancer cell lines
| NorCal SOT Fall Symposium | Holash | September 2012

19. Novartis Primary Human tumor model bank
Tumor type
Received
Established
Colon
125 63
Lung
196 55
Breast
362 47
Pancreas
159 52
Ovary
231 39
Sarcoma
199
Kidney
328 25
Melanoma 61 27
Uterus 10
Esophagus 21
Brain 49 8
Lymphoma 73 4
Stomach 43
Liver 56 2
Intestine 3
Others
179 15
Total
2115
410
Primary Tumor
Implant
Molecular Annotations
Progress
(total = 410)
SNP 6.0
286
Affy U133
288
RNA-Seq 55
Whole Exome 12
2K Exome
185
| NorCal SOT Fall Symposium | Holash | September 2012

20. In vitro to in vivo translation: LGX818 is efficacious only in human tumor xenografts expressing BRAFV600E
HMEX1655 (BRAFwt, KITN822K)
HMEX1906 (BRAFV600E)
Malme-3M (BRAFV600E)
SW620 (BRAFwt, KRASG13D) *
| NorCal SOT Fall Symposium | Holash | September 2012

21. Taking Advantage of Primary Tumors: Modeling Resistance to RAF inhibitors in BRAF(V600E) Human Melanoma Xenografts
HMEX1906 Human melanoma xenograft
Tumor volume (mm3)
resistant
Time
Concentration
Serial biopsy of resistant tumors for PD analysis
Treatment with vemurafinib models human clinical response: initial tumor regression followed by the emergence of resistant tumors
| NorCal SOT Fall Symposium | Holash | September 2012

22. A mini-randomized trial of therapeutics in pancreatic cancer
Tested 7 NVS compounds in 9 pancreas primary tumors.
One therapeutic candidate was the most active.
HPAX2402
HPAX1317
HPAX1959
HPAX2406
HPAX2198
HPAX2633
HPAX1948
HPAX2043
HPAX2428
Gene signature predicted from cell line sensitivity screening accurately predicted 3/3 responses on primary tumor models
stasis
Tumor Growth
| NorCal SOT Fall Symposium | Holash | September 2012

23. Primary Mouse Tumor Allograft Models
Rationale
Primary Tissue
Monitor
Tumor growth
Passage (P1)
Fix Tissue
Flash Freeze
Histology
IHC
Fix / freeze
SNP arrays
Exp Prof.
Exome seq.
Efficacy studies
Freeze
(master stock)
Passage (P2)
Passage (P3)
Passage (P4)
(working stock)
Implant
Use of immunocompetent mice supports species-matched host- tumor interactions
Tumor immunology
Tumor stroma interactions
Developmental signaling pathways
GEMM-derived tumors may provide broader coverage of disease progression
Include tissues not available from patients
Forward genetics approach
| NorCal SOT Fall Symposium | Holash | September 2012

24. Towards curative therapy for cancer
Model Systems
Compound Profiles
Cancer
Molecular Profiles
Human Cancer
Define to completion the genetic basis of cancer
Create a large collection of genetically annotated human cancer cell lines and preclinical animal models representative of human cancers for therapeutic profiling
Define resistance mechanisms early (prior to clinical entry) and develop either second-generation inhibitors or combination strategies
Define highly active combinations that can lead to curative therapies
| NorCal SOT Fall Symposium | Holash | September 2012

25. NIBR Acknowledgements Shanghai Basel Emeryville, Cambridge
GNF San Diego
External Collaborators and partners
Patients and their families
| NorCal SOT Fall Symposium | Holash | September 2012