How to Troubleshoot Apps for the Modern Connected Worker
Bio variance j_scheiber_bioit_repurposingworkshop2013_draft
1. Drug Repurposing –
Fishing for Pearls with a very wide net
Josef Scheiber, PhD
BioIT Workshop
April 9, 2013
2. Significant unmet medical need
100%
90%
80%
NSAIDS 80 % response rate
Drug response rate
70%
60%
50%
40%
30% Alzheimer 25 % response rate
20%
10% Several thousand diseases without
0%
known treatment
diseases
3. Disease understanding is getting better and better
Example: Leukemia and Lymphoma
5 Year 1950 Disease of
the Blood
Survival
~0% 1960 Leukemia Lymphoma
Increasing understanding of underlying biology
1970 Chronic
Leukemia
Acute
Leukemia
Preleukemia
Indolent
Lymphoma
Aggressive
Lymphoma
opens up new hypotheses
2010
~ 70%
4. Overview – Drug Repurposing
• Background
• The Sirota/Butte approach: Rebuilding the paper
– Workflow & Results
• Extensions: Orthogonal evidence from biological
networks & rare disease information
5. Drug repurposing
• Has becomes a matter of intense interest during the
past few years
≡ • The concept originally evolved in the early 1990s
• Drug repositioning • Is a strategic approach to drug development to extract
added value from prior research and development
• Drug reprofiling investments
• Drug retasking • Reinvestigation of drug candidates that have not
succeeded in advanced clinical trials, for reasons other
than safety, for potential new therapeutic applications
6. What we ultimately want to have
Molecular inititating event(s) What signalling pathways are affected?
How does this translate into gene expression changes?
How does this impact the phenotype? What drug targets are good to interfere with
The phenotype? Does a known drug interact with one of these targets?
8. Drug target network as reminder
Key message:
Every drug binds a significant number of targets
Yildirim MA, Goh KI, Cusick ME, Barabási AL, Vidal M. Drug-target network.
Nat Biotechnol. 2007 Oct;25(10):1119-26.
9. Dealing with a very complex environment –
i.e. many opportunities
DNA Target
RNA Off-targets
Protein Metabolites
Interactions Additional indications
Clinical parameters Unspecific effects
Treatment History Similar drugs
Tissue anatomy
Surgical History
Epigenetic Profiles from many
patients at different
timeponits
Adapted from: J. Scheiber; How can we enable drug discovery informatics for personalized healthcare?
Expert Opinion on Drug Discovery, 1-6; 2/2011
10. There are quite a few successful examples
Drug Original indication New indication
Sildenafil Angina Male erectile dysfunction
Eflomithine Anti-infective Reduction of unwanted facial hair in women
Finasteride Benign prostatic hyperplasia Hair loss
Raloxifene Breast and prostate cancer Osteoporosis
Paclitaxel Cancer Restenosis
Zidovudine Cancer HIV/AIDS
Topiramate Epilepsy Obesity
Minoxidil Hypertension Hair loss
Phentolamine Hypertension Impaired night vision
Tadalafil Inflammation and cardiovascular disease Male erectile dysfunction
Mecamylamine Moderately severe to severe essential hypertension ADHD
and uncomplicated cases of malignant hypertension
Celecoxib Osteoarthritis and adult Familial adenomatous polyposis, colon and breast cancer
Mifepristone Pregnancy termination Psychotic major depression
Thalidomide Sedation, nausea and insomnia Cutaneous manifestations of moderate to severe erythema
nodosum leprosy and multiple myelome
Dapoxetine Analgesia Depression Premature ejaculation
Chlorpromazine Anti-emetic / antihistamine Non-sedating tranquillizer
Tofisopam Anxiety-related conditions Irritable bowel syndrome
Fluoxetine Depression Premenstrual dysphoria
Sibutramine Depression Obesity
Bupropion Depression Smoking cessation
Duloxetine Depression Stress urinary incontinence
Milnacipran Depression Fibromyalgia syndrome
Ropinirole Hypertension Parkinson’s disease and idiopathic restless leg syndrome
Lidocaine Local anaesthesia Oral corticosteroid-dependent asthma
Atomoxetine Parkinson’s disease ADHD
Galantamine Polio, paralysis Alzheimer’s disease
11. The two major “schools” of
repurposing
Disease Profiles Compound Binding Profiles, compound
Differences in Activity Profiles, can be mediated indications linked to immediate targets
through pathways
Diseases with similar gene expression profiles Compound binding profiles are similar
are treated in a similar way Drug-target network
Match drug profiles (chembank, connectivity Explain drug profiles
map)
Extend to similar Diseases (Barabasi
Diseasasome)
OMIM, GWAS ,,
,,
Focus for today
12. Core Idea: Every biological state can be
described by a given gene expression signature
2259 Genes
Gsc
siE-Cadherin
TGFβ
Twist
Snail
Gene Expression Value
-3.0 3.0
Harrison, C. Translational genetics: Signatures for drug repositioning. Nat Rev Genet, 2011
Lukk, M et al. A global map of human gene expression. Nat Biotech, 2010
Dudley, J.T. et al. Disease signatures are robust across tissues and experiments. Mol Sys Biol, 2009
Culhane, A.C. et al. GeneSigDB--a curated database of gene expression signatures
Nucleic Acids Res, 2009
Nevins, J.R. & Potti, A. Mining gene expression profiles: expression signatures as cancer phenotypes. Nat Rev Genet, 2007
Lamb, J.et al. The Connectivity Map: using gene-expression signatures to connect small molecules, genes, and disease
Science, 2006
Source: adapted from Taube et al, PNAS 2010
13. Starting data: Connectivity map
Every drug response can be described by a given
gene expression signature
Genome-wide differential expression
profiles as ranked lists of genes
UP-regulated genes
Null effect
Down-regulated genes
Small molecules: Cell lines: Concentration and treatment
• 1309 perturbagens tested • MCF7 (human epitelial breast cancer) • 10mM (when the optimal concentration is
(FDA approved and nondrug bioactive • PC3 (human epitelial prostate cancer) unknown) x 6h
compounds) • HL60 (human leukemia)
Negative control
• SKMEL5 (human melanoma)
• ssMCF7 (MCF7 grown in a different vehicle) • Cell in the same plate and treated with
vehicle alone (medium, DMSO…)
Lamb, J. The connectivity map: a new tool for biomedical research. Nat Rev Cancer 7, 54-60 (2007)
Lamb, J. et al. The connectivity map: using gene-expression signatures to connect
small molecules, genes, and disease. Science 313, 1929-1935 (2006)
14. Drugs ”reverting” a phenotype signature
”revert the phenotype”
A B
Reference database of drug gene expression
Genes
Disease individuals Treated samples
Drugs
Healthy controls Untreated samples
Disease gene expression Drug gene
signature expression profile
Discovery and preclinical validation of drug indications using
compendia of public gene expression data
Source: Sirota et al. Sci Trans Med 2011
15. Genome-based drug re-purposing approaches
1 2
Drugs able to “revert” a phenotype signature Drugs eliciting similar transcriptional
could “revert the phenotype”, i.e. influence responses could share therapeutic effects
the disease and its progression
Disease Drug response
signature signature
Therapeutic effect
No mechanistic insights
16. Disease neighbors & drug neighbors
• Neighboring diseases may share the same drugs
• Computational drug repurposing
Chiang AP, Butte AJ
Clinical Pharma and Therapeutics. 2009
Dudley J, Tibshirani R, Butte AJ
Molecular Systems Biology, 2009
Source: Sirota et al. Sci Trans Med 2011
17. Predicting new indications –
Cimetidine as an example
“To perform an initial experimental evaluation
of our approach, we chose to evaluate
one of the therapeutic predictions
for lung adenocarcinoma (LA), because
lung cancer contributes the greatest burden
of cancer mortality and incidence
in Europe and the United States.
Although our methodology predicted
multiple new therapeutic relationships
for LA, we chose to test cimetidine because
it is an off-patent and inexpensive
drug available over the counter in the
United States and has a favorable side effect
profile . Our prediction score of
−0.088 for cimetidine was moderate, but
still more significant than the score of
−0.075 for gefitinib, a well-known therapy
for LA.”
Source: Sirota et al. Sci Trans Med 2011
18. One more example:
Topiramate (antiepileptic) for Crohn’s disease
Drug-disease score A
-0,30 -0,20 -0,10 0,00
5186324
Topiramate
Prednisolone
12,13-EODE
Tolbutamide
Yohimbine
5213008
Tomelukast
Phenanthridinone
5162773 B. Gross pathology score
5151277
Clotrimazole 5
5140203 4
Genistein
Fasudil 3 ****
5230742 2
5182598
1
Computational Repositioning of the Anticonvulsant Topiramate for 0
Inflammatory Bowel Disease TNBS + veh TNBS + pred TNBS + top Vehicle
* Prednisolone = Established compound for Crohn’s disease
** Trinitobenzene Sulfonic Acid (TNBS)
Source: Dudley et al, Sci Trans Med 2011
19. General impression - Caveats
• Need for gene expression profile measurements on the candidate drugs, data are publicly
available for > 1000, numerous relevant compounds not yet tested
• not clear how drug performance in a breast cancer cell line is relevant to all types of
diseases effects of drugs on gene expression across disease tissues
• Disease-related microarray data can and should be combined with other types of
knowledge on drugs
• Therapeutic efficacy is more complex than a simple matching of expression profiles.
Compounds have to reach the appropriate tissue to have an effect. Tissueagnostic
Methodology (the disease and drug gene expression was not measured on the same
tissues) might be suitable to find both direct and distant effects of drugs.
• Although findings for cimetidine will need further preclinical testing and demonstration in
larger clinical trials, the concept of computational analysis of public gene expression
databases as a potentially useful approach to drug discovery that may uncover additional
uses for approved drugs is given
20. Rebuilding the Sirota/Butte analysis
• chose 3 datasets in GEO concerning lung tumors in order to compare results to those obtained in
the paper : GSE2514 as analysed in the paper, GSE10072 and GSE7670 (both using Affymetrix
U133A chips) for further validation
• used gcRMA algorithm to normalize each of them and as recommended by Sirota et al., applied a
rank normalization on expression data.
• Concerning the GSE2514: converted the U95av2 probeset-ID into U133A probeset-ID (as the drug-
signatures, availables in Connectivity Map are based on U133A chips, not specifically mentioned in
the paper). In cases where several U95av2 probes map to the same U133A probe, averaged the
signal and calculated the normalized rank.
• Compared control samples to disease samples using SAM algorithm (“wilc.stat” option used to
perform Wilcoxon test on ranked data) with a FDR threshold of 0.05 for q-values
• obtained large lists of up- and down regulated genes (thousands of genes) in each datasets. As the
tool available on the Connectivity Map website only accept lists up to 300 genes, selected each the
150 most significative up- and down-regulated genes in each dataset to create disease signatures
in the analysis
Special thanks: J.P Meyniel, ISoft
.
21. The tool used: AMADEA from ISoft
Courtesy: J.P Meyniel, St. Graziani; ISoft
.
22. Rebuilding the Sirota/Butte analysis
• Download the signatures from Connectivity Map website, obtain results for each dataset
The results are relevant and consistent between the datasets:
• PHA-00665752 is found at the top of the 3 lists when you order enrichment from -1 to 1.
This molecule was shown to suppress KRAS-induced lung preneoplasia by inhibition of the Met receptor tyrosine kinase
(http://www.aacrmeetingabstracts.org/cgi/content/abstract/2006/1/416 ) ==> “Short-term treatment with a small
molecule inhibitor of c-met (PHA-00665752) decreased the size of atypical alveolar hyperplasia (AAH) and adenomas and,
within these lesions, reduced the phosphorylation of AKT, a pro-survival mediator of c-met, and induced apoptosis of
vascular endothelial cells and alveolar epithelial cells”
• Top hits all with known modes of action in Oncology
HDAC inhibitor MS-275 is found in the top 5 molecules in GSE7670 and GSE10072 datasets and in
position 174 in GSE2514 (enrichment -0.499).
This molecule is used in several clinical trials: Hodgkin’s lymphomas, breast cancer (in combination with aromatase
inhibitors) and metastatic lung cancer (in combination with erlotinib). This drug promotes differenciation and apoptosis
(http://cancerres.aacrjournals.org/content/63/13/3637 )
• Another HDAC inhibitor, HC-toxin, is found in the top 5 molecules in GSE2514.
• Celastrol, a natural proteasome inhibitor is found in the top 5 molecules in each dataset .
This drug is a potent antioxidant and anti-inflammatory molecule was shown to suppress androgen-independent prostate
cancer progression (http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0014153 )
23. Surprise
However, contrary to the paper of Sirota
et al., didn’t find Cimetidine as significant
hit:
GSE2514: enrichment 0.249
GSE10072: enrichment 0.249
GSE7670: enrichment -0.237
24. Possible reasons for the difference
with Cimetidine
• Time difference when analyses where run
• Selection by Sirota not driven by computational
results only (explicitly mentioned in the paper)
25. Going a step further
1 2
?
Drugs eliciting similar transcriptional
Drugs able to “revert” a phenotype signature
responses could share therapeutic effects
could “revert the phenotype”
Disease Drug response
signature signature
How can we get to mechanistic insights?
Therapeutic effect
26. Going a step further
• Trying to find a molecular initiating event that leads
to gene expression changes
• A computational dive into the biology underlying the
hypotheses
• Workflow:
Take most relevant genes from profiles map
into pathways retrieve most-affected
pathways retrieve small molecules active in
those pathways / annotate targets in the
pathways with additional valuable data sources
to link diseases
27. On-Pathway Repurposing
Primary or secondary target of Drug X
Gene expression level affected
Genetic link to rare disease
Treating a disease/symptom
29. Data sources
Rare disease fishing
for initial indication
Retrieve and sort
Take affected Match into (Further mapping),
pathways by
genes from pathways from e.g. rare diseases
number of
previous analysis Reactome or GWAS
affected targets
What biology most affected?
Several immune-system relevant pathways
Gene expression-affecting pathways
Developmental biology
Insuline metabolism
PI3K/AKT signalling
30. Annotate targets with diseases
Individually Rare – Collectively common (10%)
Diseases for which testing is available
2443 diseases (~6000 genes) are highly predictive & medically actionable
2.000
1.800
1.600
1.400
1.200
3 rare diseases linked to Lung cancer example
1.000 (example: a rare genetically determined asthma)
800
600
400
200
0
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
Source: Genetests.org
31. More annotation:
Genome-Wide Association Studies (GWAS)
As of 2011,
1200 human
GWASs have
been
published on
over 400 traits
Manolio TA. N Engl J
Med 2010;363:166-176.
32. One more step: Small molecules
from chemogenomic DBs,
• Example: ChEMBL
• What indications are they approved for
• Can these compounds used in the indication
being scrutinized
33. Outlook
• The right drug for the right patient at the right time & right dose is
only possible if you have the right knowledge within the right context
right in place
• Number of known off-targets increases repurposing opportunities
gamble with safety
• massive quantity of public available gene expression data has not yet
been fully exploited (e.g. disease profile fishing with drug expression
profile, we are attempting this now)
• Identified rare diseases can be used as point of entry for clinical
trials