Identified the likelihood of success for treatment of five cancers of interest by comparing novel drug combinations treated cell line gene signatures (predicted by bioinformatic analysis), with disease gene signatures (calculated by fitting linear model on gene expression data).
DATA SUMMIT 24 Building Real-Time Pipelines With FLaNK
Project Presentation
1. PROPRIETARY & CONFIDENTIAL
Repurposing Drugs for
Combination Therapy in Cancer
BMI 217, Winter 2016
Jennifer Epler, Ivan Ip, Yooli Light,
Tiffany Wang, Yiying Wang
1
2. +Introduction
• Oncology drug development is lengthy and expensive
• Drug repurposing promises reduced time and cost
• Bioinformatics techniques can be used to predict efficacy of new
drug combinations
• Clinical trials are underway for new drug combinations that
combine existing non-cancer drugs with existing
chemotherapeutic agents
3. +Existing Non-Cancer Drugs
• There is evidence that Metformin (diabetes) and HDAC inhibitors
(Depakote, an anticonvulsant/mood stabilizer, and Zolinza, approved for
some lymphomas) may have potent anti-tumor effects, particularly when
combined with more established oncology drugs.
• Metformin is an AMPK agonist, and also corrects insulin resistance.
• HDAC inhibitors may enhance the DNA damaging effects of traditional
chemotherapy.
• Both drugs can alter cell growth signaling pathways which are often
dysregulated in cancer. These pathways have already been the subject
of numerous new drug development efforts.
Valproic acid (Depakote) and Vorinostat
(Zolinza) are HDAC inhibitors. Metformin is an AMPK agonist.
4. + Drug Combinations
All “Partner B” drugs were curated from open studies at clinicaltrials.gov; all
are currently being tested in cancer patients in combination with a “Partner
A” drug.
Connectivity Map contains data for all drugs in the indicated cell lines.
For simplicity, create signatures for all pairings of Partner A drugs + Partner
B drugs.
For a more novel approach, integrate gene expression data with KEGG
pathway data and see how the up- and down- regulated genes would be
affected
5. +Collecting Data Step 1: Process normal and
disease GEO datasets to
identify significant up and down
regulated genes by building a
linear model
Step 2: Process drug
combinations to identify
significant up and down
regulated genes using
Connectivity Map
Step 3: Prepare a final gene list
of potential targets that each
drug combination pair can treat
Step 4: Perform functional
analysis of genes using MSigDB
We chose to focus on three
cells lines: breast, prostate and
myeloid/heme based on
Connectivity Map database
Giant set of gene expression
data from CCLE was
manipulated and analyzed
We found that SW900_LUNG
and HUPT3_PANCREAS cell
lines that clustered close to PC3
We applied the drug signatures
from PC3 to these two cell lines
Diseases of interest: breast
cancer, prostate cancer, acute
myelogenous leukemia (AML),
pancreatic cancer and lung
cancer
6. + Gene expression
profile of drug A
Gene expression
profile of drug B
Estimate expression
change of drug A
Estimate expression
change of drug A
Estimate the drug
combination effect
- Integrated expression data
with KEGG pathway data
- Examined sharing of 7 cell
growth pathways known to be
oncogenic
Yang et al. Drug-Induced Genomic Residual Effect Model
for Successful Prediction of Multidrug Effect. CPT
Pharmacometrics. (2015)
Cell Growth Pathway
Approach
7. +
32 upstream and downstream
KEGG pathways included in GP
Yang et al. Drug-Induced Genomic
Residual Effect Model for Successful
Prediction of Multidrug Effect. CPT
Pharmacometrics. (2015)
Additional
Growth Pathway
Approach
8. + Results
There was not enough
data for the prostate cell
line for us to evaluate
new drug combinations
The MsigDB also
showed little correlation
for the pancreatic and
lung cell lines
We shifted the bulk of
our analysis to
breast/AML cancers
Total Transcripts per Disease Signature
#Transcripts
9. + Breast Cancer: hundreds of genes up- &
down-regulated.
Metformin drug combinations showed significant genes overlapping in the
CHARAFE_BREAST_CANCER_LUMINAL_VS_BASAL_UP and _DN, and the
SMID_BREAST_CANCER_BASAL_DN gene sets.
Depakote drug combinations showed significant genes overlapping in the
MASSARWEH_TAMOXIFEN_RESISTANCE_UP gene set.
Zolinza drug combinations showed significant genes overlapping in the
ELVIDGE_HYPOXIA_UP gene set.
#Transcripts
10. + Breast Cancer
The two HDAC inhibitors showed similar results in MSigDB.
Depakote drug combinations had significant amount of genes
overlapping in:
NAGASHIMA_NGR1_SIGNALING_UP gene set. Upregulation of
Neuregulin-1 in breast cancer is significant, because tumors evade
apoptosis by down regulating that gene.
MASSARWEH_TAMOXIFEN_RESISTANCE_UP gene set, indicating
beneficial second-line therapy for breast cancer patients
Zolinza drug combinations had significant amount of genes
overlapping in:
ELVIDGE_HYPOXIA_UP gene set. As tumor adapt for growth in hypoxic
conditions, upregulation of these genes may counter this survival
advantage.
Promising drug combinations include:
Depakote + Letrozole, Azacitidine, Doxorubicin, and Irinotecan
Zolina + Azacitidine
Metformin + Prednisone, Simvastatin, and Doxorubicin
11. + AML: hundreds of genes up- & down-
regulated.
Several Metformin drug combinations showed significant genes overlapping
in: PILON_KLF1_TARGETS_DN gene set.
Depakote drug combinations had significant amount of genes overlapping in:
HALLMARK_TNF-ALPHA_SIGNALING_VIA_NFKB gene set.
Zolinza drug combinations had significant amount of genes overlapping in:
HELLER_HDAC_SILENCED_BY_METHYLHYLATION_UP,
HELLER_HDAC_TARGETS_UP, and KRIGE_RESPONSE_TO_TOSEDOSTAT gene
sets.
#Transcripts
12. + AML
The two HDAC inhibitors did not show similar results in MSigDB, this
is not surprising since Zolinza has already been approved for a
different heme indication
As an HDAC inhibitor, Depakote only covers 8 HDAC’s in Classes I
and IIa whereas Zolinza can effectively target all 11 HDAC.
Metformin drug combinations affecting the PILON_KLF1_TARGETS_DN gene
set will alter KLF activities, likely restoring normal cell cycle progression .
Depakote drug combinations affecting the HALLMARK_TNF-
ALPHA_SIGNALING_VIA_NFKB gene set may improve overall survival. High
TNFα levels are associated with poor prognosis in AML..
Zolinza drug combinations affecting the
KRIGE_RESPONSE_TO_TOSEDOSTAT gene sets , indicate beneficial
second-line therapy for leukemia patients.
Promising drug combinations include:
Depakote + Prednisone or Gefitinib
Zolinza + Prednisone, Hydralazine, or Letrozole
Metformin + Aspirin, Simvastatin, or Letrozole
13. + Tripled Drug Combinations
In addition to our double drug combinations, we proposed another approach
which was to triple our drug combinations
Results from MSigDb showed similar results to the double drug combination
but there was an increase in the number of overlapping genes for both breast
and AML cancers
Most promising triple therapies:
Breast
Metformin + either HDACi + Prednisone
Metformin + either HDACi + Doxorubicin
AML
Metformin + either HDACi + Hydralazine
Metformin + either HDACi + Letrozole
Metformin
Valproic acid
(Depakote)
Aspirin
Metformin
Vorinostat
(Zolinza)
Aspirin
Hydralazine Hydralazine
Prednisone Prednisone
Dexamethasone Dexamethasone
Simvastatin Simvastatin
Letrozole Letrozole
Azacitidine Azacitidine
Doxorubicin Doxorubicin
Irinotecan Irinotecan
Paclitaxel Paclitaxel
Gefitinib Gefitinib
Sirolimus Sirolimus
14. + KEGG Pathway Approach
Integrating the gene expression data with the KEGG pathway
data produced four data sets that showed similar trends to
simple drug addition
Primary focus was placed on cell growth pathways, which is
important in cancer studies
0
100
200
300
400
500
600
700
acetylsalicylicacid
hydralazine
prednisone
dexamethasone
simvastan
letrozole
azacidine
doxorubicin
irinotecan
paclitaxel
gefinib
sirolimus
acetylsalicylicacid
hydralazine
prednisone
dexamethasone
simvastan
letrozole
azacidine
doxorubicin
irinotecan
paclitaxel
gefinib
sirolimus
acetylsalicylicacid
hydralazine
prednisone
dexamethasone
simvastan
letrozole
azacidine
doxorubicin
irinotecan
paclitaxel
gefinib
sirolimus
me ormin valproic acid (depakote) Vorinostat
#ofgenes
Drug Combina ons
Breast/MCF7 - Cell Growth Path
Drug - / Disease +
Drug + / Disease -
15. + KEGG Pathway Approach
This approach allowed us to examine up- and down-stream
signaling factors that would be affected by our initial up- and
down- regulated genes
Genes that were regulated in same direction were considered to
have an inductive effect on the cell growth pathway
From MSigDB, we saw similar gene sets up and down
regulated, we also saw significant amount of genes overlapping
in:
KEGG_MAPK_SIGNALING_PATWAY and KEGG_CELL_CYCLE
Promising drug combinations
Breast: Metformin + Azacitadine, Depakote + Azacitidine, Vorinostat
+ Azacitadine
AML: Metformin + Gefitinib, Depakote + Gefitinib, Vorinostat +
Hydralazine
16. + Side Effects
While we have proposed many promising drug combinations, we must also consider the side
effects of these drug combinations
Examination of the combined side effects of some promising drug combinations using drug-
drug interaction side effects database TWOSIDES
TWOSIDES database: only synergistic interactions of drug pairs are reported. Exclude drug
pair with one drug of the pair that is likely responsible for the adverse event
If no synergistic interactions are associated, recapitulations of known effects of each drug is
considered
Tatonetti et.al.,Data-Driven Prediction of Drug Effects
and Interactions. Sci Transl Med (2012).
OFFSIDES
Database
1332 drugs
10,097 side effects
adverse event
report in AERS
from FDA
TWOSIDES
Database
59,220 drug pairs
868,221 significant
associations
1301 side effects
List of Promising
drug combinations
Drug pairs
with only
synergistic
interactions
Canada’s
MedEffect
FDA drug labels
Only significant
associations
(p<0.05)
Prediction and
evaluation of
side effects of
the target drug
combinations
No significant synergistic interactions
17. +
Side Effects
Side effects counts associated with promising drug pairs
List of serious side effects associated with the drug pairs
18. + Conclusions
Using a bioinformatics approach, we have proposed several promising
drug combinations that might have a therapeutic effect for breast or
AML cancer.
Our approach builds upon earlier studies, which cast a wider net on
drug repurposing, but looked only at monotherapies.
Our approach simulated combination therapies, which are likely to have
longer-lived efficacy in clinical practice. We were also careful to
preserve tissue specificity, increasing the likelihood of predicting
successful combinations.
With more comprehensive, publically available data sets, bioinformatic approaches will have
greater power to predict new drug combinations.
Confirming our results, clinicaltrails.gov currently shows these drug combinations in trials:
Metformin + Aspirin for prostate cancer
Metformin + Irinotecan for brain tumors
Metformin + Letrozole for endometrial cancer
Depakote + Prednisone for B-cell Lymphoma
Depakote + Azacitidine for AML
Vorinostat + Dexamethasone for mantle cell lymphoma
Vorinostat + Gefitinib for Non-Small-Cell Lung Carcinoma