Human Cell Systems Biology for Drug Discovery and Chemical Safety. Presentation at the 7th Brazilian Symposium on Medicinal Chemistry, November 12, 2014, Campos do Jordao-SP, Brazil. Ellen Berg.
1. Human Cell Systems for Drug Discovery
and Chemical Safety
Ellen L. Berg, Scientific Director
The 7th Brazilian Symposium on Medicinal Chemistry
Campos do Jordão-SP, Brazil, November 9-12, 2014
2. Agenda
• Challenges in pharmaceutical research
• Primary human cell systems – BioMAP
platform
• Case studies
- Understanding ADRs - thrombosis-related side
effects
- Drug combinations
2
3. • Problem:
- Pharmaceutical productivity is at an all time low
- We are swimming in oceans of data
• A need for new approaches
- Better physiological relevance
- More predictive of clinical effects
Challenges in Drug Discovery
We need to do something different: A Turning Point
3
4. Complexity of Biology
Scale (meters)
molecules pathways cells tissues humans
10-9 M 10-8 M 10-7 M 10-6 M 10-5 M 10-4 M 10-3 M 10-2 M 10-1 M 1 M
Human exposureMolecular targets
4
• Human biology is complex
- Modular, redundant, highly networked, & full of feedback loops
5. Complexity of Biology
Scale (meters)
molecules pathways cells tissues humans
10-9 M 10-8 M 10-7 M 10-6 M 10-5 M 10-4 M 10-3 M 10-2 M 10-1 M 1 M
Human exposureMolecular targets
5
• Human biology is complex
- Modular, redundant, highly networked, & full of feedback loops
• Prediction (and understanding!) is difficult
- Emergent properties
Primary human cell systems
6. Solution: Primary Human Cell Systems
• BioMAP® Profiling:
- In Vitro testing in primary human cell based tissue and
disease models
• Data driven chemical biology approach
- Data-driven research methodology
- Leverages the analysis of a large chemical biology dataset
• Applications in drug discovery
- Compound characterization across a broad range of biology
- Drug mechanisms of action – anchored on clinical outcomes
- Guidance for translational studies, indications & biomarkers
Confidential6
9. Data Driven Research
Issues
Many hypotheses are generated
Each hypothesis requires validation
Validation requires both computational
and “domain” expertise
Solution
Incorporate “domain” expertise upfront
10. BioMAP® Technology Platform
BioMAP®
Assay Systems
Reference
Profile Database
Predictive
Informatics Tools
Human primary cells
Disease-models
30+ systems
Biomarker responses to drugs
are stored in the database
>3000 drugs
Custom informatics tools are
used to predict clinical outcomes
High Throughput Human Biology
10
11. BioMAP® Systems – Key Features
11
Primary human cell types
Physiologically relevant “context”
Complex activation settings
Co-cultures
Translational biomarker endpoints
12. Feature Mice Man
Lifespan 2 Years 70 Years
Size 60 g 60 kg
Environment
Animal facility,
cage-mates
Outside world, people,
animals, etc.
Why Human?
Key differences:
DNA repair mechanisms
Control of blood flow, hemostasis
Immune system status
12
13. Closer to the disease process
Downstream of multiple pathways and integrate information
“Decision-making”
Used by clinicians to guide therapy - Provide clinical “line of site”
Predictive
Why Translational Biomarkers?
mRNA,
epigenome
Phospho-sites,
intracellular proteins,
metabolome
Cell surface,
secreted molecules
13
16. • Challenges
- Cells and assays are expensive
- Primary cells (all cell-based assays!) are variable
- Very large number of assay components / choices
• Cell types, media, additives, time points, endpoints
Experimental Design
16
17. • Solutions
- Automation
• Microwell plate-based
- Standardized methods
• Quality management system (SOPs)
• Strict assay acceptance criteria
- Incorporate methods to reduce variability
• Cells from pooled donors, prequalified
• Normalize data within plate (Log10 ratio of
compound/vehicle)
• 6+ vehicle replicates, two positive controls per plate
Experimental Design
17
18. • Compromises:
- Single well per endpoint, but:
• Multiple concentrations (4+) per compound
• Multiple assay systems per compound
• Multiple endpoints per assay system
- Single timepoint
• Suboptimal for some endpoints, but optimal for most
endpoints (24 hr – 6 days)
- Pauciparameter (7-22 endpoints per assay system), but:
• Highly informative disease biomarker endpoints
Experimental Design
18
19. BioMAP Profile of Positive Control
• Colchicine is an inhibitor of microtubules
- It is active in every system and used as a positive control on every plate
• Colchicine profile has a distinctive pattern of activities or “shape”
BioMAP Systems
Readout Parameters (Biomarkers)
Cytotoxicity Readouts
Colchicine 1.1 μM
Logexpressionratio
(Drug/DMSOcontrol)
Vehicle Control
(no drug)
95%
significance
envelope
19
20. Reproducibility of Profiles
• 16 Experiments over many months
• Pairwise correlation of profiles (Pearson’s) were > 0.8
BioMAP Systems
Readout Parameters (Biomarkers)
Houck, K.A., J. Biomolecular Screening, 2009, 14:1054-66.20
Logexpressionratio
(Drug/DMSOcontrol)
Vehicle Control
(no drug)
95%
significance
envelope
21. • Assess cytotoxicity in primary human cells
- Cytotoxicity mechanisms are cell type and activation dependent
- Note: cytotoxicity is a confounder
• Flag compounds (concentrations) that are overtly cytotoxic
• Analyze overall activity profiles
- Profile characteristics
- Unsupervised and supervised approaches to compare profiles
• Focus on individual endpoints
- Correlate to external data
- Build an understanding of clinical mechanisms
What Can We Do With BioMAP Profile Data?
21
22. Types of BioMAP Profiles
InactiveActive – Sharp dose-response
Active – Dose resistantActive – Selectively
22
23. Rapamycin (mTOR) Genistein (multi-target)
Dose Resistance
A Profile “Characteristic”
• “Dose resistant” compounds have similar activity profiles over a
wide range of concentrations
- No sharp activity jumps; Rapamycin > Genistein
• Characteristic of approved drugs & target-selective compounds
- Rapamycin is highly selective for mTOR; Genistein has multiple targets
- The dose resistance index of Rapamycin is > 60,000x
23
24. BioMAP Profiling: Example Profile
Reference p38 MAPK Inhibitor
Logexpressionratio
(Drug/DMSOcontrol)
Control (no drug)
99%
significance
envelope
BioMAP Systems
Readout Parameters (Biomarkers)
Dose
Response
Cytotoxicity Readouts
24
This profile shows dose-resistance – similar over a range of
concentrations
25. BioMAP Profiling: Example Profile
Reference p38 MAPK Inhibitor
Logexpressionratio
(Drug/DMSOcontrol)
25
Activities relevant to the role of p38 in monocyte / Th1-type inflammation
p38 kinase is important for Th1-dependent inflammatory responses
Takanami-Ohnishi Y, et al., Essential role of p38 mitogen-activated protein kinase
in contact hypersensitivity. J Biol Chem. 2002, 277:37896-903.
IL-8
HLA-DR
Monocyte
activation
IL-6IL-1aCD38
HLA-DR
TNF-a
26. BioMAP Profiling: Example Profile
Reference p38 MAPK Inhibitor
Logexpressionratio
(Drug/DMSOcontrol)
26
Activities relevant to anti-thrombotic effects of p38 inhibitors
Tissue factor is the primary cellular initiator of coagulation
p38α deficiency impairs thrombus formation
Sakurai K, et al. Role of p38 mitogen-activated protein kinase in thrombus
formation. J Recept Signal Transduct Res. 2004;24(4):283-96.
Tissue
Factor
27. BioMAP Profiling: Example Profile
Reference p38 MAPK Inhibitor
Logexpressionratio
(Drug/DMSOcontrol)
27
Activities relevant to side effects – clinical finding: skin rash
Upregulation of VCAM and ITAC are characteristic of skin hyperreactivity
Melikoglu M, et al., Characterization of the divergent wound-healing responses
occurring in the pathergy reaction and normal healthy volunteers. J Immunol.
2006, 177:6415-21.
ITAC
VCAM
MMP1
VCAM
28. 28
BioMAP® Data Analysis
Predictive
Informatics Tools
Custom informatics tools are
used to predict clinical outcomes
• Unsupervised Analyses
- Similarity Search of our reference
database
- Clustering
• Supervised Analyses
- Computational models (classifiers)
for mechanism of action
29. 29
BioMAP® Reference Database
BioMAP®
Reference Database
Biomarker responses to drugs
are stored in the database
>3000 drugs
• More than 3000 agents
- Drugs – Clinical stage, approved, and failed
- Experimental Chemicals - Research tool
compounds, environmental chemicals,
nanomaterials
- Biologics – Antibodies, cytokines, factors,
peptides, soluble receptors
• Availability of reference data
- EPA ToxCast and selected reference data
are published and have been made
available (Houck, 2009; Berg, 2010; Berg,
2013; Kleinstreuer, 2014)
30. Similarity Analysis of Profiles
Highly correlated Similar
Pearson’s correlation of r > 0.7
Low correlation Not similar
Pearson’s correlation of r < 0.7
30
36. Building Support Vector Machine Classifiers
• 88 Compounds
• 28 Target/Pathway
mechanisms
• 1-8 concentrations
• 327 Profiles
• 84 endpoints (8 BioMAP
Systems)
• Support Vector Machine
• 2-class models
• Mechanism class versus “Null”
set
• Result = Decision Value (DV)
• PPV – positive predictive value
(fraction of profiles that are correctly
classified)
• PPV = TP / (TP + FP))
• Sensitivity (fraction of profiles that
are assigned to the class)
• Sensitivity = TP / (TP + FN))
Mitochondrial
Inhibitor
Microtubule
Stabilizer Hsp90 Inhibitor
Classifier Performance: Examples
PDE IV
Inhibitor
Generate Data
Set
Build
Classifiers
Test Performance
of Classifiers
Berg, Yang & Polokoff, 2013, J. Biomol Screen. 18:1260.36
37. • AhR agonist (Aryl Hydrocarbon)
• Calcineurin
• EGFR (Epidermal Growth Factor R)
• SERCA (SR Ca++ ATPase)
• EP agonist
• Estrogen R agonist
• Glucocorticoid R agonist
• H1R Antagonist (Histamine)
• HDAC
• HMG-CoA-Reductase
• Hsp90 Inhibitor
• IKK2
• IL-17 R agonist
• JAK
Confidential37
List of Classifiers (SVM Mechanism Models)
• MEK
• Microtubule Disruptor
• Microtubule Stabilizer
• Mitochondrial Inhibitor
• mTOR
• p38 MAPK
• PDE IV (Phosphodiesterase
• PI3K
• PKC (c+n)
• Proteasome
• RAR-RXR agonist
• Src family
• TNF (Tumor Necrosis Factor)
• VDR agonist (Vitamin D R)
Berg, Yang & Polokoff, 2013, J. Biomol Screen. 18:1260.
38. • Compound characterization
- Broad biological fingerprint
- Cell types, pathways, possible clinical indications
• Mechanism of action
- Triage hits from phenotypic drug discovery programs
- Unexpected off-targets (toxicity)
• Support therapeutic hypotheses
- Compare to competitor molecules, clinical standards of care
- Identify translational biomarkers
Applications
38
40. EPA ToxCastTM Program – BioSeek
Goal is identification of in vitro assays that can help
forecast in vivo toxicity of environmental and other
agents (including pharmaceuticals)
40
Task Order Compound Number Compound Type
TO1 320 Environmental compounds
TO2 500 Environmental compounds
TO3 200 Environmental and Failed Pharma Compounds
TO4 39 Nanomaterials
TO5 31 Nanomaterials
TO6 100 Failed Pharma Compounds, etc.
TO7 39 Nanomaterials
Total 1229
40
Houck, K.A., J. Biomolecular Screening, 2009, 14:1054-66;
Kleinstreuer, 2014, Nature Biotechnology, 32:583-91.
41. Chemical Groups & Classes in ToxCast
Most active
Least active
41
Overall: 73% Active
(33 – 83%)
Kleinstreuer, 2014, Nature Biotechnology, 32:583-91.
42. Results of Supervised Analysis
Performance of SVM Mechanism Classifiers
Mechanism Class1 Number of
Compounds
Correctly
Assigned
% Comment
p38 MAPK Inhibitor 2 2 100%
Estrogen R Agonist 10 6 60%
Not classified: meso-Hexestrol, 4-
nonylphenol and diethystilbestrol
HMG-CoA Reductase Inhibitor 3 3 100%
Histamine R1 Antagonist 1 1 100%
Microtubule Inhibitor 2 1 50% Herbicides
GR Agonist 3 3 100%
Mitochondrial Inhibitor 2 2 100% Fungicides
PDE IV Inhibitor 8 6 75%
RAR/RXR Agonist 2 2 100%
Total 33 26 79%
• 1Mechanisms for which classifiers were available and mechanisms were known
- Dataset: Kleinstreuer, Nature Biotechnology, 2014, 32:583
- Classifiers: Berg, Yang and Polokoff, JBS, 2013, 18:1260
42
43. Unsupervised Analysis (Self Organizing Maps)
AhR Phenotypic Signature
• Phenotypic signature of
compounds in SOM cluster #57
- Box and whisker plot for cluster
57 representing a signature for
AhR activation
• Confirmation of AhR activity
- 85% of members of clusters 57,
67 (adjacent in the 10X10 SOM)
were active in an AhR reporter
gene assay (examples shown
here).
Tissue Factor
Kleinstreuer, 2014, Nature Biotechnology, 32:583-91.43
44. Unsupervised Analysis (Self Organizing Maps)
Estrogen R Actives: Phenotypic Signatures
• Two clusters of chemicals defined by their BioMAP signatures
- Blue = Estradiol, Estrogen Receptor Agonists
- Red = Estrogen Receptor Antagonists, “Selective Estrogen R Modulators”
• Increased levels of Tissue Factor by SERMs and ER antagonists
Kleinstreuer, 2014, Nature Biotechnology, 32:583-91.
Estrogen
Receptor
Antagonists
Estrogen
Receptor
Agonists
Tissue Factor
44
46. • Pathologic setting – aberrant coagulation thrombosis
- The formation of a blood clot (coagulation) within a vein
- Deep vein thrombosis (DVT), stroke
- Pulmonary embolism thrombi break off and get lodged in the
lung
Thrombosis
SMC
Endothelial cells
Vessel Lumenplatelets in fibrin clot
46
48. • Associated with:
- Exposure to Smoking & Pollution
• Polycyclic aromatic hydrocarbons (“Aryl Hydrocarbons”)
- Contraceptives, hormonal replacement therapy
- Various other drugs
• mTOR inhibitors (everolimus)
• 2nd generation anti-psychotics
Thrombosis-Related Side Effects
48
49. • Aryl Hydrocarbon receptor agonists
- PAHs, Benz(a)anthracene
- Smoking (Cigarette smoke extract)
• mTOR inhibitors
- Everolimus (Baas, 2013, Thromb Res 132:307)
• Anti-Estrogens / SERMS, oral contraceptives
- Tamoxifen, Clomiphene, Cyproterone
• Second generation anti-psychotics
- Clozapine
• Others
- Crizotinib
Mechanisms / Drugs Associated with
Thrombosis-Related Side Effects
All show increased Tissue Factor levels in 3C and LPS Systems
49
50. • Search our reference database for all compounds / test
agents that increase TF in the 3C system
- What are the mechanisms represented?
- Do they share any common biology?
• Issues
- Large chemical-biology datasets will have errors
• Inactive concentrations, toxic concentrations, variability
- How do we increase our confidence?
• Require compound effects at more than one concentration
• Effect size >20% (4 SD)
• Multiple compounds with same target mechanism
Is There A Connection?
50
52. Mechanisms that Increase TF
Test Agents Mechanism
Confidence in
Mechanism
2-Mercaptobenzothiazole AhR agonist High
3-Hydroxyfluorene AhR agonist High
Benzo(b)fluoranthene AhR agonist High
C.I Solvent yellow 14 AhR agonist High
FICZ AhR agonist High
Abiraterone CYP17A Inhibitor High
Ketoconazole CYP17A Inhibitor High
Clomiphene citrate Estrogen R Antagonist High
Histamine H1R agonist High
Histamine Phosphate H1R agonist High
Cobalt(II) Chloride Hexahydrate HIF-1α Inducer High
Tin(II) Chloride HIF-1α Inducer High
Chloroquine Phosphate Lysosome Inhibitor High
Primaquine Diphosphate Lysosome Inhibitor High
Temsirolimus mTOR Inhibitor High
Torin-1 mTOR Inhibitor High
Torin-2 mTOR Inhibitor High
Bryolog PKC activator High
Bryostatin PKC activator High
Bryostatin 1 PKC activator High
Phorbol 12-myristate 13-acetate PKC activator High
Phorbol 12,13-didecanoate PKC activator High
Picolog PKC activator High
3,5,3-Triiodothyronine Thyroid H R agonist Good
Concanamycin A Vacuolar ATPase Inhibitor Good
Mifamurtide NOD2 agonist Good
Oncostatin M OSM R agonist Good
Ethanol Organic Solvent Good
PAz-PC Oxidized phospholipid Good
Z-FA-FMK Cysteine protease Inhibitor Good
8-Hydroxyquinoline Chelating agent Unknown
A 205804 ICAM, E-selectin inhibitor Unknown
AZD-4547 FGFR Inhibitor Unknown
Crizotinib ALK, c-met Inhibitor Unknown
Desloratadine H1R antagonist Unknown
Dodecylbenzene Industrial chemical Unknown
Fenaminosulf Fungicide Unknown
GDC-0879 B-Raf Inhibitor Unknown
GW9662 PPARγ agonist Unknown
Imatinib PDGFR, c-Kit, Bcr-Abl Inhibitor Unknown
KN93 CaMKII Inhibitor Unknown
Linoleic Acid Ethyl Ester Fatty Acid Unknown
Mancozeb Fungicide Unknown
MK-2206 AKT Inhibitor Unknown
Mometasone furoate GR agonist Unknown
N-Ethylmaleimide Alkylating agent Unknown
PP3 SRC Kinase Inhibitor Unknown
Primidone GABA R agonist Unknown
Sulindac Sulfide NSAID Unknown
Terconazole Anti-fungal Unknown
Tris(1,3-dichloro-2-propyl) phosphate Flame retardant Unknown
TX006146 Unknown Unknown
TX006237 Unknown Unknown
TX011661 Unknown Unknown
U-73343 Unknown Unknown
UO126 MEK Inhibitor Unknown
ZK-108 PI-3K Inhibitor (βγ-selective) Unknown
Mechanisms that Increase TF
AhR Agonist
CYP17A Inhibitor
Estrogen R Antagonist
H1R Agonist
HIF-1α Inducer
Lysosomal Inhibitor
mTOR Inhibitor
PKC Activator
Thyroid H R Agonist
Vacuolar ATPase Inhibitor
NOD2 Agonist
OSM R Agonist
52
53. Mechanisms that Increase TF
Mechanisms that Increase TF
AhR Agonist
CYP17A Inhibitor
Estrogen R Antagonist
H1R Agonist
HIF-1α Inducer
Lysosomal Inhibitor
mTOR Inhibitor
PKC Activator
Thyroid H R Agonist
Vacuolar ATPase Inhibitor
NOD2 Agonist
OSM R Agonist
Implicate Autophagy
53
54. Autophagy
• Cellular response to nutrient deprivation
• Also contributes to recycling of dysfunctional
organelles, handling of protein aggregates, bacteria and
viruses54
60. • Summary
- Compounds that increase TF are associated with thrombosis
related side effects
- Compounds that increase TF also increase autophagic vacuoles
(increase formation or decrease breakdown)
- Mechanistic Hypothesis: thrombosis-related side effects are
associated with alterations in the process of autophagy that
increase TF cell surface levels
• Take home message:
- This case study illustrates how chemical biology datasets,
combined with external knowledge, can give rise to higher level
mechanistic understanding of toxicity mechanisms
Tissue Factor, Autophagy & Thrombosis
60
61. Adverse Outcome Pathway Framework
MIE
Key
Event
Adverse
Outcome
Key
Event
Key
Event
Molecular
Initiating Event Clinical Effect
• Framework for integrating mode of action hypotheses to
outcomes for chemical risk assessment (OECD)
- http://www.oecd.org/chemicalsafety/testing/adverse-outcome-pathways-
molecular-screening-and-toxicogenomics.htm
• Focused on the clinical outcome
- Anchored at both ends
61
62. AOP for DVT
MIE
Key
Event
Adverse
Outcome
Inhibition of
mTOR
Upregulation
of Tissue
Factor
Deep Vein
Thrombosis
Initiation of
Coagulation
Key
Event
Key
Event
Molecular
Initiating Event Clinical Effect
Increase in
Autophagic
Vacuolization
62
63. AOP for DVT
MIE
Key
Event
Adverse
Outcome
Inhibition of
mTOR
Upregulation
of Tissue
Factor
Deep Vein
Thrombosis
Initiation of
Coagulation
Key
Event
Key
Event
Molecular
Initiating Event Clinical Effect
MIE
Activation of
AhR
Increase in
Autophagic
Vacuolization
Key
Event
Inhibition of
NPC1
Key
Event
HDF3CGF
In vitro
disease model
3C
3C 4H LPS
Endothelial
Cells
Endothelial
Cells
PBMC +
Endothelial
Cells
P
En
Th1 Th2 TLR4
BioMAP System
Primary Human Cell
Types
Stimuli
! ! !
63
65. • Challenges for studying drug combinations:
- System must include both targets
- Physiologically relevant setting (ideally all human)
- Suitably robust to capture combination effects
• Case Example
- BioMAP Oncology systems that model tumor-host
microenvironments
- Trametinib (MEK kinase inhibitor) + Dabrafenib (Braf inhibitor)
• Combination approved for treatment of melanoma
Drug Combinations
65
69. Dabrafenib (B-raf) Trametinib (MEK) Dabrafenib +Trametinib
• Combination effects of Dabrafenib (B-raf) and Trametinib (MEK)
- Tumor cell marker (CEACAM5) is reduced only in the combination (green
arrow)
- Consistent with the combination being more efficacious against tumors in vivo
69
Combination Study Example:
B-Raf + MEK Inhibitor
70. Dabrafenib (B-raf) Trametinib (MEK) Dabrafenib +Trametinib
• Combination effects of Dabrafenib (B-raf) and Trametinib (MEK)
- Tumor cell marker (CEACAM5) is reduced only in the combination
- Consistent with the combination being more efficacious against tumors in vivo
- Reduced levels of Inflammatory endpoints; collagen III (grey arrows)
- Consistent with reduced Trametinib-related skin side effects (Flaherty, 2012,
NEJM 367:1694).
70
Combination Study Example:
B-Raf + MEK Inhibitor
71. • Chemical profiling in human cell systems generates
activity profiles that can be used to:
- Group chemicals into bioactivity classes
- Generate MoA hypotheses
- Identify activities that may correlate with in vivo outcomes
• High throughput in vitro data is most informative when
combined with external information
- Known targets
- In vivo bioactivities
Summary
Confidential71
72. • Applications for predicting in vivo effects must
also consider:
- Exposure - level and route
- Distribution
- Metabolism
- Human variability
Challenges and Considerations
Confidential72
73. • BioSeek
- Mark A. Polokoff
- Dat Nguyen
- Xitong Li
- Antal Berenyi
- Alison O’Mahony
- Jian Yang (Oracle)
• UCSF
- Kevan Shokat
Acknowledgements
• EPA
- Keith Houck
- Nicole Kleinstreuer
- Richard Judson
• Support
- NIH/NIAID (SBIR)
- EPA (EP-D-12-047, EP-
W-07-039)
73
74. Contact:
Ellen L. Berg, PhD
Scientific Director
BioSeek, a division of DiscoveRx
eberg@bioseekinc.com
Editor's Notes
The complexity of biological systems makes it difficult to predict outcomes from both target-based as well as phenotypic drug discovery efforts.
The BioMAP platform of cell-based assay platform designed to include more of the biological complexity of human disease, but yet in a practical format with sufficient throughput to be used in early discovery.
So what do we mean by biological complexity?
The complexity of biological systems makes it difficult to predict outcomes from both target-based as well as phenotypic drug discovery efforts.
The BioMAP platform of cell-based assay platform designed to include more of the biological complexity of human disease, but yet in a practical format with sufficient throughput to be used in early discovery.
So what do we mean by biological complexity?