Transcript Modeling Endocrine Resistance in Breast Cancer

Modeling Endocrine Resistance in Breast Cancer
Robert Clarke, Ph.D., D.Sc.
Professor of Oncology
Director, Center for Cancer Systems Biology
Dean for Research
Georgetown University Medical Center
Systems Biology in Cancer Research
Robert Clarke, Ph.D., D.Sc.
A systems biology approach is required to integrate knowledge from
cancer biology with computational and mathematical modeling
Study of an organism viewed as an integrated and interacting network
of genes, proteins, and biochemical reactions that give rise to life…*
● Systems biology goals
– interactions among the components of a biological system
– how these interactions control system function and behavior
– integrate and analyze complex data from multiple sources using
interdisciplinary tools
– build in silico models of system (network) function
Systems Biology Research Cycle
Endocrinologist 94: 13, 2010
Biological cycle
Integration with modeling
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
*Lee Hood - Institute for Systems Biology
Resistance to Endocrine Therapies
Age (Menopausal Status)
Risk Reduction1
Recurrence: <50 years (ER+)
Recurrence: 60-69 years (ER+)
Recurrence (ER-)
45 ± 8%
54 ± 5%
6 ± 11% (not significant)
Death: any cause <50 years (ER+)
Death: any cause 60-69 years (ER+)
Death: any cause (ER-)
33 ± 6%
32 ± 10%
-3 ± 11% (not significant)
1Proportional
Robert Clarke, Ph.D., D.Sc.
} Benefit from TAM
reduction in the 10-year risk of recurrences or death from the Early Breast Cancer Trialists Group meta analyses
To understand how some ER+ breast cancers become (or already are at
diagnosis) resistant to endocrine therapies, we invoke an integrated,
multimodal, network hypothesis
– network is modular and exhibits both redundancy and degeneracy
– signaling is highly integrated and coordinates many cellular functions
In the face of the stresses induced by endocrine therapies, the network
modules of interest are those that regulate cell fate, i.e., the cell’s choice:
– to live or die (e.g. control/execution of apoptosis, autophagy, necrosis)
– if to live, whether or not to proliferate (i.e., cell cycle control/execution)
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
Are all Tamoxifen Failures the Same?
Computational Modeling: task = classification
● Compare failures “on-treatment” (early; ≤3yrs) with those that
recurred (distant recurrence) later “off-treatment” (later; ≥5 yrs)
● Construct molecular classifiers using gene expression microarray
data from breast tumors collected at diagnosis
– integrated resampling workflow to ease the “gene selection bias” problem
– Support Vector Machine with recursive feature elimination
ER+
ER-
human cost (mortality)
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
health care cost (treatment)
Classifying Early vs. Later TAM Recurrences
c: Repeat 100 times
b: 10-fd CV
Validation
Performance
Validation
Set
Split in 1/3 and 2/3 parts
All Samples
Split in 10 equal parts
a: Repeat 100 times
Test
Set
Optimal gene set
Learning
Set
LOOCV to optimal
number of genes
Combination of 10 validation gene lists
Validation gene list
Training
Set
Average Validation
Performance
Classifier
Gene list
= 100% occurrence

Resampling approaches used to ease the “gene selection bias” problem

Must outperform random gene sets of the same size (10,000 random sets)1

Must meet n=7 pre-established performance benchmarks2

Clinical characteristics
– training procedure (block a)
– validation step (block b)
– n=131 cases; >95% ER+; almost all Invasive Ductal Carcinomas
– Tamoxifen only after surgery and radiotherapy
– ≥15 years of clinical follow-up
et al., PLoS Comp Biol, 2011 report that >60% (in some cases up to 90%) of breast cancer molecular predictors are no better than random gene sets
Mackay et al., JNCI, 2011 report that the molecular subgroup classifications for the LumA, LumB, and normal-like breast cancer subgroups are not robust
1Venet
2
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
Early (≤3 yr) and Later (≥5 yr) TAM Recurrences
BC030280
Accuracy
Specificity
Sensitivity
AUC
PPV
NPV
Hazard Ratio
P-value
0.90
0.95
0.81
0.87
0.91
0.89
3.45
<0.0001
1
RASD2
RAB6B
STXBP5L
TAAR3
LOC728683
RFX3
0.9
USP36
BCL2L14
ADAMTS1
PLCH1
LOC144874
C1orf96
ATXN7L1
ME3
PNPLA3
SLC7A5
C1orf187
SOD2
LRP8
BATF2
TMEM4
CR1L
LOC284801
IKZF1
DCLK3
LOC150763
LOC440292 /// LOC647995
C12orf65
PRO0471
OR10A3
GGN
NAP1L4
SLC6A6
% Survival
Sensitivity
0.8
0.7
0.6
0.5
0.4
0.3
OFCC1
C8orf12
RHD
LOC651964
KCNJ12
LOC283079
MGC52498
ZNF704
RNF133
STK35
C1orf86
TNNI1
FERD3L
SEC14L2
MS4A7
MS4A7
ITGA8
PTGER3
PTGER3
PTGER3
CX3CR1
EFHC2
FLJ14959
DEGS2
MUM1L1
NCOA7
MAOB
STK32B
CMYA5
KIAA1467
RBM24
SLC7A8
0.2
0.1
0
ERBB4
THSD4
THSD4
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1 - Specificity
Loi et al.
Time
Accuracy
Specificity
Sensitivity
AUC
PPV
NPV
Hazard Ratio
P-value
0.77
0.83
0.74
0.81
0.88
0.67
3.11
0.0004
Performance exceeds all
(n=7) pre-established
benchmarks in both datasets
1
0.9
% Survival
0.7
sensitivity
Sensitivity
0.8
0.6
0.5
0.4
0.3
0.2
(and outperforms all of 10,000
randomly selected gene sets)
0.1
0
Lombardi
0
0.1
0.2
0.3
0.4
0.5
0.6
1-specificity
0.7
1 - Specificity
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
0.8
0.9
1
Time
Minetta Liu (Georgetown; Mayo)
Mike Dixon; Bill Miller (Edinburgh)
Jason Xuan (Virginia Tech)
Joseph Wang (Virginia Tech)
Approach to Network Modeling
● The module(s) of interest exist within an immense search space (the human
interactome) and we don’t know all of the genes/proteins in each module
● Networks are high dimensional and so the data have unique properties, e.g., curse
of dimensionality; confound of multimodality; scale free; small world; etc.
Clarke et al., Nature Rev Cancer, 2008; Wang et al., Br J Cancer 2008
● We take a systems biology approach to integrate knowledge from cancer biology
with computational and mathematical modeling to make both qualitative and
quantitative predictions on how a system functions
Computational modeling
Physical modeling
● We apply both computational and mathematical modeling tools
– computational models can find local topologies or modules within high dimensional data
using multiple different methods (top down)
– mathematical models can represent local topologies or modules by a series of
differential equations, stochastic reaction networks, etc. (bottom up)
Chen et al. Nucl Acid Res, in press, 2013
Gusev et al., Cancer Informatics, 12: 31-51, 2013
Zhang et al., PLoS ONE, 5 (4): e10268, 2010
Zhang et al., Bioinformatics 25: 526-532, 2009
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
Wang et al., J Mach Learn Res, in press, 2013
Gu et al. Bioinformatics, 28: 1990-1997, 2012
Yu et al., J Mach Learn Res, 11;2141-2167, 2010
Clarke et al., Nature Rev Cancer 8: 37-49, 2008
Yu et al, Bioinformatics, in revision, 2013
Tyson et al., Nature Rev Cancer, 11: 523-532, 2011
Chen et al., Bioinformatics, 26: 1426-1422, 2010
Wang et al., Bioinformatics, 23: 2024-2027, 2007
Network Modeling: Where to Start?
● We have selected our key modules of interest
– live or die (e.g., apoptosis, autophagy, necrosis)
– proliferate or growth arrest (i.e., cell cycling)
● We know that ERα is relevant and will coordinate several cell functions
– key regulator in normal mammary gland development and function1
– most tumors acquiring endocrine resistance retain ERα expression2
– responses to 2nd and 3rd line endocrine therapies are relatively common2
– small molecule inhibitors and RNAi against ERα inhibit resistant cells3
● We don’t know precisely how ERα signaling is regulated or wired
● We need an ERα-driven network model to guide our studies
ERα
et al., Nat Med , 2003
et al. Pharmacol Rev, 2002
3Kuske et al., Endocr Relat Cancer, 2006
Wang et al., Cancer Cell, 2006
1Johnson
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
2Clarke
Roadmap for Modeling ER-Related Signaling
Hypothesis: With the stresses induced by endocrine therapies, the network modules
of interest are those that regulate cell fate, i.e., the breast cancer cell’s choice
– to live or die (e.g. control/execute apoptosis, autophagy, necrosis, UPR)
Primary Outputs
– if to live, whether or not to proliferate (i.e., cell cycle control/execution)
Primary Inputs/Regulators
Estrogen Receptors
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
Growth Factor Receptors (e.g., EGFR; Her2)
John Tyson et al., Nature Rev Cancer, 2011
ERα as a “Master” Regulator of Cell Fate
 ER is the most upstream regulator of cell fate decisions
 ER can be mutated, phosphorylated, degraded, recycled
– mutations appear to be relatively rare in clinical samples
– Fulvestrant acts by targeting the receptor for ubiquitin-mediated
degradation

ER can activated by ligand or by growth factors
– several growth factors and their receptors signaling to MAPKs that can
activate ER through phosphorylation
 Regulation of ER activation may be a central determinant of
endocrine responsiveness
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
ER and EGFR/HER2 Crosstalk
Mathematical Modeling: task = nature of ER regulation of cell fate
Primary data from multiple clones of MCF-7 cells
transfected with either HER2 or EGFR and assayed
for E2-dependent or E2-independent growth
Liu et al., Breast Cancer Res Treat, 1995
Miller et al., Cell Growth Diff, 1994
Crosstalk between ER and GFR
Parameters
Description
Value
γEPI
Rate of EPI reaching its steady state
3×10−4 min-1
γGFR
Rate of GFR reaching its steady state
ωEPI
Basal inhibition of EPI
−1.92
ωGFR
Basal inhibition of GFR
−4
ωE2ER
Basal inhibition of E2ER
−2.1
ωERP
Basal inhibition of ERP
−1.5
ωEPI,GFR
EPI activation by GFR
6
ωGFR,GFR
GFR activation by EPI
5
ωGFR,E2ER
GFR inhibition by E2ER
−2
ωGFR,ERP
GFR activation by ERP
1.85
ωGFR,GFRover
GFR activation by GFRover
0.15
ωE2ER,E2
E2ER activation by E2
3
ωERP,GFR
GFR activation by ERP
3
E2 level in MCF7 cells
Parameter determining total ER level in
MCF7 cells
1 (normal); 0 (E2-depleted cells)
Excess GFR in GFR-transfected MCF7 cells
0 (normal); >0 (GFR-transfected cells)
E2
ERT
GFRover
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
5×10−2 min-1
1 (normal); >1 (ER-overexpressed cells)
Chun Chen et al., FEBS Lett, 2013
GFR = growth factor receptor (HER2 or EGFR)
GFRover = transfected with GFR
EPI = epigenetic components
ERP = estrogen-independent
E2ER= estrogen-dependent
ERT = total ER levels
ER is a Bistable Switch for EGFR/HER2 Crosstalk
Mathematical Modeling: task = nature of ER regulation of cell fate
Bistability: resting in two different minimum states separated by a maximum
John Tyson et al., Nature Rev Cancer, 2011
●
Breast cancer cells can switch reversibly and robustly between E2
and GFR dependence
–
–
GFR can inhibit ER expression and/or activate (phosphorylate) any remaining ER
cells can eliminate or silence GFR plasmid (epigenetic) and upregulate ER
●
Model can explain some of the molecular heterogeneity in cell
populations
●
Blocking either pathway increases the likelihood that the other
pathway will be activated
●
E2-dependence  GFR-dependence (ER-independence) occurs
more easily/rapidly than the reverse
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
Chun Chen et al., FEBS Lett, 2013
Robert Clarke, Ph.D., D.Sc.
Phenotype Transitions Support Intermittent Therapy
Mathematical Modeling: task = ER-driven phenotype transitions
Shifting E2 dose response
Ligand
Dependent
Ligand
Supersensitive
Minimum Action Paths characterize state transitions
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
Ligand
Independent
Intermittent therapy opens a 2nd response window
Chun Chen et al., in preparation
Robert Clarke, Ph.D., D.Sc.
Factors Affecting Endocrine Responsiveness
●
What molecular events are associated with endocrine
resistance?
●
When are these changes acquired (early, late)?
●
Which changes are functionally/mechanistically important?
●
How do cells coordinate their functions to make and
execute a cell fate decision?
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
ERα Signaling: Early vs. Late Recurrences
Computational Modeling: task = network topology
Number of nodes
● Identify closest protein partners to ERα using
a novel Random Walk (RW) based algorithm with
Metropolis Sampling (MS; Markov Chain-Monte
Carlo) technique to walk 8 PPI (protein-protein
interaction) databases
– 2-steps per iteration (walk)
– 300,000 iterations
– 1,452 neighbors selected; n=50 are frequently visited
● Model the n=50 using the microarray data and
MS/RW method
red = overexpressed in ‘Early’
SRC
ERβ
ERβ
ERα
BCL2
EGFR
ERα
AKT
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
Gene Ontology
p-value
23/50
“Apoptosis”
2.9E-13
14/50
“Cell proliferation”
6.8E-5
AR
EGFR
SRC NFκB
BCL2
MAPK
MAPK
green = overexpressed in ‘Late’
yellow = inconsistent
Genes
AR
Circles = nodes
Circles = nodes
Lines = edges
Lines = edges
Minetta Liu et al., in review
Bai Zhang et al., in preparation
Minetta Liu (Georgetown; Mayo)
Mike Dixon; Bill Miller (Edinburgh)
Jason Xuan (Virginia Tech)
Joseph Wang (Virginia Tech)
Some Changes are Acquired Early
Computational Modeling: Differential Dependency Network (DDN) analysis
 Represent the local structures of a network by a set of local conditional
probability distributions – decompose the entire expression profile
into a series of local networks (nodes; parents)
– local dependency is learned
– local conditional probabilities are estimated from linear regression model
– allow more than one conditional probability distribution per node
– Lasso technique is used to limit overfitting
 Identify motifs and “hot spots” within motifs
– time series data from T47D cells ± E2; ± Fulvestrant (Lin et al., Genome Biol, 2004)
– key nodes identified include AKT, XBP1, NFκB, several BCL2 family members,
several MAPKs
extracellularly exposed
plasma membrane
BCL2 (large family) regulate apoptosis/survival
cytosol
XBP1 is a key component of the
Unfolded Protein Response (UPR)
nucleus
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
Yue Wang et al., Bioinformatics, 2009
Some Early Changes are Retained
Selected from molecular comparison of sensitive (LCC1) vs. stable resistant variant (LCC9)
Gene Name
Gene Symbol1
Difference
p-value
Genes Up-regulated in LCC9 vs. LCC1
Cathepsin D
CTSD
5-fold
<0.001
autophagy
X-box Binding Protein-1 (TF)
XBP1
4-fold
<0.001
UPR
B-cell CLL/lymphoma 2
BCL2
4-fold
<0.001
apoptosis/UPR
Epidermal growth factor receptor
EGFR
2-fold
0.002
Heat Shock Protein 27
HSBP1
2-fold
0.001
UPR
NFκB (p65) (TF)
RELA
2-fold
<0.05
UPR/apoptosis
apoptosis
Genes Down-regulated in LCC9 vs. LCC1
Death Associated Protein 6
DAXX
6-fold
0.049
Early Growth Response-1 (TF)
EGR1
3-fold
<0.05
Interferon Regulatory Factor-1 (TF)
IRF1
2-fold
<0.05
apoptosis
Tumor Necrosis Factor-α
TNF
2-fold
<0.05
apoptosis
TNFRSF1A
2-fold
<0.05
apoptosis
TNF-Receptor 1
Data are mean values of the relative level of expression for each gene to the nearest integer; 1HUGO Gene Symbols
UPR = Unfolded Protein Response; TF = transcription factor
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
Zhiping Gu et al., Cancer Res, 2002
Todd Skaar et al, J Steroid Biochem Mol Biol, 1998
XBP1(s) May Control Some Retained Changes
Symbol
Gene Name
Change
p-value
# CREs
APBB2
amyloid beta (A4) precursor protein-binding
-1.3
0.001
1
BCL2
B-cell CLL/lymphoma-2
3.1
0.029
3
CRK
v-crk sarcoma virus CT10 oncogene homolog
-2.0
0.003
2
ESR1
estrogen receptor alpha (ERα)
2.8
0.040
0*
IL24
interleukin 24
-9.7
<0.001
1
MYC
v-myc myelocytomatosis viral oncogene homolog
1.6
0.04
1
PHLDA2
pleckstrin homology-like domain, family A, member 2
-3.3
0.004
2
S100A6
S100 calcium binding protein A6 (calcyclin)
2.3
0.001
1
XRCC6
X-ray repair complementing defective repair 6
1.6
0.016
1
*several ATF6 sites that may be regulated by ATF6:XBP1 heterodimers
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
Bianca Gomez et al., FASEB J, 2007
Some Retained Changes are Functionally Important
XBP1(s) confers Antiestrogen Resistance
XBP1 cDNA increases BCL2
Relative Bcl-2:actin ratio
4
T47D/XBP1
T47D/c
MCF7/c
MCF7/XBP1
XBP1 siRNA reduces BCL2
p< 0.001 for ANOVA,
*p=0.029
^p=0.019
*
3
^
2
1
0
EtOH
TAM
FAS
Rebecca Riggins et al., Mol Cancer Ther, 2005
Bianca Gomez et al., FASEB J, 2007
Inhibition of both BCL2 and BCLW is better
BECN1 (siRNA) and 3-MA each reverse antiestrogen resistance
when combined with BCL2 (YC137) inhibition
proliferation
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
Anatasha Crawford et al., PLoS ONE, 2010
Yanxia Ning et al., Mol Cancer Ther, 2010
apoptosis
BCL2 and Total-BH3 Predicts Level of Apoptosis
Mathematical Modeling: task = explore role of BCL2 family in apoptosis

d[BAX ]
 (k f 1  k f 2[ BH 3])[ BAX]  kb [BAXm]F
dt
kb [BAXm : BCL2]
d[BAXm : BCL 2]
 k as [ BAXm]F [ BCL2]F  k ds [ BAXm : BCL2]
dt
k b [BAXm : BCL2]

d[BH 3]F
 ks  ks Stress  kd [ BH 3]F  kasBH 3[BH 3]F [BCL2]F
dt
kdsBH 3[BH 3 : BCL2]

d[BH 3 : BCL2]
 k asBH 3[BH 3]F [BCL2]F  k dsBH 3[BH 3 : BCL2]
dt
k d [ BH 3 : BCL2]
17 nonlinear ordinary differential equations
and 44 parameters for the various molecular species
Model predicts %apoptosis and provides an approximate measure
of responsiveness based on the concentrations of BCL2 and
the total of all BH3 members of the BCL2 family
PCD = programmed cell death/apoptosis
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
Bill Bauman, Tongli Zhang in preparation
Coordinated Functions: BCL2 Family and Cell Fate
Apoptosis
(cell death)
Autophagy
(cell survival)
altered cell metabolism?
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
Autophagy (self-eating)
Normal process through which aged or damaged subcellular organelles
are degraded and their components recycled into intermediate
cellular metabolism
BECN1 (siRNA) and 3-MA (inhibit
autophagy) reverse antiestrogen resistance
when combined with BCL2 (YC137) inhibition
Anatasha Crawford et al., PLoS ONE, 2010
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
XBP1(s) Induces Pro-Survival Autophagy
Monodansylcadaverine-labeled Vesicles
Vehicle
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
ICI 182, 780
Ayesha Shajahan et al., submitted
LC3-GFP expression
MCF7/EV Vehicle
MCF7/EV 1uM Fas
MCF7/XBP1 Vehicle
MCF7/XBP1 1uM FAS
Coordinated Functions: Metabolism
How does a cell coordinate its resources to allow execution of a cell fate decision?

Metabolome: collection of metabolites (~2500 identified in humans)
e.g., within a cell
– reflects the physiological state of a cell

Intermediates and products of metabolism (<1 kDa in size)

Metabolites separated by mass and charge using UPLC-MS (Ultra
Performance Liquid Chromatography-Mass Spectrometry)

Data processed using Random Forest algorithm to identify most
robust discriminant metabolites
– e.g., amino acids, antioxidants, nucleotides, sugars, etc.
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
High Confidence Interaction Network
Mapping metabolome onto transcriptome (LCC1 vs. LCC9)
METABOLITE
GENE/PROTEIN
Insulin/IGF signaling
MET. – PROT./MET.
PROT. – PROT.
Cell survival signaling
Energy metabolism
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
Ayesha Shajahan et al., submitted
Antiestrogens Reduce Intracellular ATP
ATP levels relative to LCC1 Vehicle
1.2
Vehicle
E2
TAM
FAS
PAC
ATP
1.0
0.8
0.6
0.4
0.2
0.0
LCC1
LCC9
ATP levels drop with treatment in sensitive cells
Resistant cells have lower basal ATP levels that are
refractory to endocrine treatment
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
Ayesha Shajahan et al., submitted
Vehicle=ethanol and no E2
E2=17β-estradiol
TAM=Tamoxifen
FAS=Fulvestrant/Faslodex
PAC=Paclitaxel
MYC, Glutamine, and UPR Enable LCC9 Survival
Complete medium
Glutamine (no glucose) medium
UPR Activation
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
Ayesha Shajahan (GU) et al., submitted
Cellular Sensing of Nutrient/Energy Deprivation
GRP78 and AMPK may be energy sensors and autophagy switches
XBP1
BCL2  BCL2:BECN1
XBP1  BCL2:BECN1 may confer
degenerancy on autophagy induction
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
Katherine Cook et al., Cancer Res, 2012
A Mechanistic Topology of Endocrine Resistance
Cellular metabolism may be an essential determinant of cell fate
or Glutamine (poor vascularization; loss of growth factor stimulation, etc.)
Autophagy
UPR
BECN1
BCL2,
et al.
Apoptosis
GRP78 = HSPA5 = BiP
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
Clarke et al., Cancer Res, 2012
Katherine Cook et al., Cancer Res, 2012
System Coordination: Network Modeling
Metabolic
Adaptations
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
John Tyson et al., Nature Rev Cancer, 2011
Summary
● Systems biology approaches provide one way to explore phenotypes and
to integrate cellular and molecular features to understand mechanism(s)
● Cells appear to experience EnR stress and can use GRP78 to activate the
UPR, which then integrates signaling to determine cell fate
— inhibits apoptosis (e.g., antiapoptotic BCL2 family members)
— induces autophagy (e.g., BECN1, antiapoptotic BCL2 family members, AMPK, mTOR)
— initiates/coordinates changes in metabolism required to execute the cell fate decision
● Antiestrogens modify cellular energy metabolism leading to changes in
glutamate/glutamine/glucose uptake and intracellular AMP levels
— autophagy also provides intermediate metabolites to fuel the cell fate decision
● ER acts as a bistable switching mechanism to affect phenotype, making
intermittent therapy a more effective strategy
● Some early adaptations to treatment are retained in resistant cells
● Resistance may not require many new nodes but does change the
nature/usage of existing edges among nodes
(it’s mostly the same network of nodes, its just wired differently)
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
Acknowledgments
The patients who contributed to the clinical studies
Harini Aiyer
Younsook Cho
Ahreej Eltayeb
Leena Hilakivi-Clarke
Mike Johnson
Habtom Ressom
Jessica Schwartz
Anni Wärri
Amrita Cheema
Katherine Cook
Caroline Facey
Rong Hu
Lu Jin
Rebecca B. Riggins
Ayesha Shajahan
Alan Zwart
J. Michael Dixon
William R. Miller
Lorna Renshaw
Andrew Simms
Alexey Larionov
University of Edinburgh, Breast Unit
University of Edinburgh, Breast Unit
University of Edinburgh, Breast Unit
University of Edinburgh, Breast Unit
University of Edinburgh, Breast Unit
Bill Baumann
Engineering & Computer Science
Engineering & Computer Science
Engineering & Computer Science
Biological Sciences & Virginia Bioinformatics Institute
Biological Sciences & Virginia Bioinformatics Institute
Biological Sciences & Virginia Bioinformatics Institute
Engineering & Computer Science
Engineering & Computer Science
Engineering & Computer Science
Chun Chen
Li Chen
Iman Tavasolly
John Tyson
Anael Verdugo
Yue Wang
Jianhua Xuan
Bai Zhang
Erica Golemis
Rochelle Nasto
Ilya Serebriiskii
Lombardi
COMPREHENSIVE
CANCER CENTER at GEORGETOWN UNIVERSITY
Sandra Jablonski
Yongwei Zhang
Lou Weiner
Subha Madhavan
Yuriy Gusev
Robinder Gauba
Minetta Liu (now at Mayo)
U54-CA149147 ICBP Center for Cancer Systems Biology
29XS194 NCI In Silico Research Center of Excellence
R01-CA131465; R01-CA149653
BC073977
BC122874
KG090245