University of Southern California Dissertations and Theses

Investigating the complexity of the tumor microenvironment's role in drug response

(USC Thesis Other)

Investigating the complexity of the tumor microenvironment's role in drug response

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

Investigating the complexity of the tumor microenvironment’s role in drug response

by

Colleen M. Garvey

A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(Cancer Biology & Genomics)

May 2020

2
TABLE OF CONTENTS
CHAPTER 1 ..................................................................................................................... 4
1.1 INTRODUCTION .................................................................................................... 4
1.2 HYPOXIA ............................................................................................................... 5
1.3 CANCER ASSOCIATED FIBROBLASTS .............................................................. 5
CHAPTER 2 ..................................................................................................................... 7
2.1 INTRODUCTION .................................................................................................... 7
2.2 RESULTS ............................................................................................................... 8
Birth and death rates vary across cell types and microenvironments 8
Subclassifying heterocellular populations 10
Comparison of HCS to traditional cell biology assays 11
2.3 DISCUSSION ....................................................................................................... 13
2.4 METHODS ........................................................................................................... 16
Cell culture and reagents 16
Cellular assay conditions 16
Image acquisition & analysis 17
MTS assay 18
Flow cytometry 18
2.5 ADDITIONAL THOUGHTS ................................................................................... 19
CHAPTER 3 ................................................................................................................... 20
3.1 INTRODUCTION .................................................................................................. 20
3.2 RESULTS ............................................................................................................. 20
Model Parameterization 21
Effect of Erlotinib, Evofosfamide, and Oxygen on Growth Kinetics 22
Growth Kinetics During Combination Therapy 24
Toxicity Constraints 24
Comparison of combination strategies with standard monotherapy schedules 25
Optimized combination strategies 28
3.3 DISCUSSION ....................................................................................................... 31
3.4 ADDITIONAL THOUGHTS ................................................................................... 34
CHAPTER 4 ................................................................................................................... 35
4.1 INTRODUCTION .................................................................................................. 35
4.2 RESULTS ............................................................................................................. 36
CAFs decrease cancer cell sensitivity to cetuximab 36
CAFs secrete more EGF in response to cetuximab treatment 38
Exogenous EGF causes cetuximab resistance in 2D and 3D models 39
Secreted EGF from cetuximab-treated CAFs is sufficient to render cancer cells resistant to
cetuximab 41
3
4.3 DISCUSSION ....................................................................................................... 43
4.4. METHODS .......................................................................................................... 44
Cell culture and reagents 44
Primary cell culture: Human tumor organoids and cancer-associated fibroblasts 44
Imaging growth rate assays 44
Collection and processing of conditioned media 45
Secretome Analysis 45
Western Blotting 45
Organoid Viability Assay 45
4.5 FUTURE DIRECTIONS ........................................................................................ 46
Experiments in Progress 46
Suggested Future Experiments 47
4.6 ADDITIONAL THOUGHTS ................................................................................... 48
REFERENCES .............................................................................................................. 49

4
CHAPTER 1
An introduction into cancer as an evolutionary process and the role of the
tumor microenvironment in therapeutic response

1.1 INTRODUCTION
Cancer is a highly complex, rapidly evolving disease that relies on more than the
underlying genetics of a cell to fuel progression. It is now well established that the tumor
microenvironment (TME) plays an active role in tumor progression and treatment
response. It is heterogeneous in nature, consisting of naturally occurring gradients of
oxygen, nutrients, cytokines, and drug, other cell types (i.e. fibroblast, immune, and
endothelial cells), and extracellular matrix factors, which create physical environmental
niches in the tumor that drive cellular adaption.
The importance of a tumor’s surroundings on its response to therapy has been
appreciated for some time. Over twenty years ago, cancer cells were implanted either
subcutaneously or into different visceral organs prone to metastasis. While the tumors
were genetically identical upon implantation, they varied in their response to therapy. The
subcutaneous tumors were sensitive to the chemotherapeutic agent doxorubicin,
whereas tumors implanted in the lung or liver did not respond
1
. This result, among others,
highlights the importance of “context” in cancer - where cancer cell surroundings influence
how they progress and respond to therapy.
This dissertation focuses on inhibition of EGFR through molecular targeting kinase
inhibitors (TKIs) and molecular antibodies (mAb) in non-small cell lung cancer (NSCLC)
and colorectal cancer (CRC), respectively. While these EGFR targeting agents have been
approved for clinical use, resistance is almost inevitable. Investigating environmental
pressures that support the development of or inherently cause this resistance is the gap
this work begins to fill. Below I will discuss the impact of various microenvironmental
factors on therapeutic response. While there are many factors to consider, I will focus on
hypoxia and cancer-associated fibroblasts.

5
1.2 HYPOXIA
In solid tumors, the vasculature is notoriously disorganized, with regions of increased
interstitial fluid pressure and extracellular matrix resulting in irregular blood flow
2
. Tumor
hypoxia, or low-oxygen conditions, is one consequence of this irregular vasculature.
Hypoxia can lead to cellular adaptions or select for clones that have decreased drug
sensitivity through various mechanisms. One such mechanism is the result of reduced
cell proliferation in oxygen deprived regions of a tumor, which limits the cells’
responsiveness to chemotherapy agents whose mechanisms of action target rapidly
dividing cells
3
. In addition, the effectiveness of radiation therapy to kill cells is highly
dependent on oxygen availability and thus regions of low oxygen are considered radiation
resistant
4
. Tumor cells also actively upregulate expression of specific markers in low
oxygen conditions that lead to drug resistance including P-glycoprotein (also known as
multidrug resistance protein, MDR1), which is an ATP-dependent efflux pump involved in
actively exporting drugs out of the cell
5
. Higher expression of this protein results in lower
efficacy of many anticancer therapies. In addition, HIF-1a, which is only stable in low
oxygen conditions, has been shown to increase pyruvate dehydrogenase kinase-3
expression (an enzyme involved in the regulation of glucose metabolism) and results in
metabolic re-wiring that confers chemoresistance in cervical
6
and colon
7
cancer models.
Hypoxia has been known to contribute to genomic diversification in tumors,
resulting in production of more aggressive phenotypes including enhanced motility and
invasiveness
8,9
. In particular, genomic instability (i.e. point mutations, gene amplifications,
etc.) can be induced by low oxygen levels through a variety of mechanisms, including
deregulation of DNA damage checkpoint signaling
10
and selection for clonal populations
with loss of DNA mismatch repair
11
. Therefore, the likelihood of cancer cells to acquire
advantageous mutations in oxygen scarce regions is also higher
10,12
.
1.3 CANCER ASSOCIATED FIBROBLASTS
Cancer cells are in physical and biochemical contact with many different cells native to
the host environment (termed stromal cells), including fibroblasts, immune, vascular, fat,
glial, and smooth muscle cells. A balance exists between these different cell types, with
6
each cellular species competing for nutrients and cytokines and uniquely responding to
their availability. The interplay between tumor and stromal cells has been shown to drive
many processes in cancer progression, including response to therapeutics
13,14
. Often
times it is thought that therapeutic resistance is due to tumor cell intrinsic mechanisms.
However, more recent evidence points to the bidirectional feedback between tumor and
stromal cells leading to the tumor microenvironment as a source of acquired resistance.
Cancer-associated fibroblasts form the dominant component in most tumor
stroma. Fibroblasts found in normal tissue are responsible for the deposition of non-
cellular scaffolding; upon appropriate stimuli, they become ‘activated’ to orchestrate the
wound healing process. In the context of cancer, so called cancer-assocated fibroblasts
are found to be in this ‘activated state’ while in the absence of stimuli. It has become
apparent that these cells have an integral role on cancer progression and response to
therapy. Indeed, fibroblast composition in patient tumors is prognostic of overall survival
in many cancer types, including colorectal cancer, with fibroblast percentages over 50%
associated with decreased survival rate
15
.
Various factors secreted by fibroblasts are known to alter therapeutic response in
cancer
16
. Stroma-mediated resistance, especially to targeted therapies, has been shown
to be a common occurrence in in vitro studies. Straussman et al.
13
observed that HGF
secreted by fibroblasts was responsible for innate resistance to RAF inhibitors in
melanoma cells. In ovarian cancer, fibroblast-mediated resistance to platinum-based
chemotherapies has been shown to occur through reducing the intracellular content of
platinum in cancer cells. However, CD8+ T-cells abolish this resistance by altering
fibroblast metabolism through the IFNg pathway
17
.

7
CHAPTER 2
Development of a high-content imaging method to quantitatively study cell
populations response to microenvironmental perturbations

Associated publications:
Garvey, C. M. et al. A high-content image-based method for quantitatively studying context-dependent cell
population dynamics. Sci. Rep. 6, 29752; doi: 10.1038/srep29752 (2016).

Garvey, C.M., Gerhart, T.A., Mumenthaler, S.M. Discrimination and Characterization of Heterocellular
Populations Using Quantitative Imaging Techniques. J. Vis. Exp. (124), e55844, doi:10.3791/55844 (2017).

2.1 INTRODUCTION
For years, in vitro cellular assays have been performed using conditions that do not
recapitulate physiological conditions. While great advances have been made with such
experimental strategies, it is now acknowledged that the results of some of these
experiments may be misleading given the lack of relevant context. This is especially true
in the context of drug response in tumors – whose genetic and environmental makeup
are notoriously complex and evolving. Many traditional viability assays are unable to
decouple the effects from multiple cell types or resolve the heterogeneity found within a
single cell type/subpopulation.
The power of this approach comes from the ability to easily and quantitatively
examine cellular phenotypes across different timescales and in high throughput providing
novel insights into cellular dynamics that were overlooked when directly compared to
traditional assay readouts from MTS and flow cytometry. As a result, in using this platform
one can investigate a large number of conditions at multiple time points to determine
“context dependent rates”, rather than sampling at a single moment in time or a single
concentration value, where the outcomes might be vastly different depending on when
you look. Tumors are continually adapting; what they look like today may not be what they
look like tomorrow. The quantitative nature of this platform provides a more accurate
representation of biological processes and can feed into additional applications (e.g.,
mathematical model predictions) that provide further insights into population dynamics.
8
Many sources of cell-to-cell variation exist in tumors, including different types of
cells. For the purpose of my work, I focused on a subset of cells termed cancer-associated
fibroblasts (as introduced in Chapter 1), although I acknowledge that there are many other
important cell types – including those involved in the immune component of tumors – that
are not being considered.
Here, we apply our quantitative imaging pipeline to the context of
microenvironment-mediated drug resistance in non-small cell lung cancer (NSCLC). We
investigate the effect of various microenvironmental conditions (glucose, O2, stromal
cells) on the response of NSCLC cell lines to erlotinib, a tyrosine kinase inhibitor against
EGFR that is used in the treatment of NSCLC.

2.2 RESULTS
Birth and death rates vary across cell types and microenvironments
Drug treatments may elicit different cellular responses depending on concentration and
cell type. This is important to consider in many applications, such as drug development,
where it may be beneficial to have cells undergoing apoptosis as opposed to transitioning
to a quiescent state, or vice versa
18
. Our HCS method is able to distinguish between
cytotoxic and cytostatic responses to drug treatments by differentiating between a
decrease in cell birth versus an increase in cell death, as depicted in Figure 2.1a.
Treatment with erlotinib on one of our NSCLC cell lines, H3255, results in a
cytotoxic response to drug, while our other cell line, HCC4011, displayed a cytostatic
response. This is important information – as one can anticipate that if cells are not actively
dying they will still be present after drug removal and allow for tumor recurrence.
Due to the high-throughput capacity of this platform, we are able to capture cellular
responses to various microenvironmental factors individually and concurrently
19
. Birth
rates decreased in both cell lines when exposed to low (0,1%) oxygen conditions,
although overall drug response did not seem to be affected (Figure 2.1a). The high-
throughput nature of this assay is exemplified in Figure 2.1b. H3255 cells were exposed
to forty different microenvironmental contexts, with perturbations including oxygen, drug,
9
glucose, and fibroblasts. We reveal that the presence of fibroblasts increases H3255 cell
growth and renders them less sensitive to low glucose concentrations in comparison to
standard laboratory culture conditions (21% O2, 2 g/L glucose, no fibroblasts).

Figure 2.1: Quantitating the birth-death processes of multiple cell populations under selective
pressures
(a) Representative images showing Hoechst (white) and propidium iodide (orange) staining of H3255
and HCC4011 cells treated with erlotinib (0.1 μM). Live and dead cell counts were obtained by
segmenting nuclei and identifying dead cells based on propidium iodide intensity levels. Birth and
death rates were then calculated for each cell type in response to erlotinib (0–10 μM) and oxygen
perturbation (21%- normoxia and 0.1% O2– severe hypoxia). (b) Net growth rates (units = 1/hour),
represented by heatmap color, of H3255 cell line exposed to 40 different microenvironmental contexts.
Drug (x-axis), oxygen (O2), glucose (gluc), and fibroblasts (F) were the micreonvironmetal conditions
perturbed.

10
Subclassifying heterocellular populations
In most physiologically relevant systems, there is often more than one cell type involved.
Cell populations are in a constant flux that is governed by the evolutionary pressures
being imposed on the system, including the interaction between cell types. Thus, it is
valuable to include multiple cell types in experimental design, and, more importantly, be
able to distinguish between the populations. For example, recent work has shown that
clonal subpopulations of lung cancer cells resistant to erlotinib exist at a low frequency
prior to treatment
20
. Therefore, it is of interest to jointly study mixed populations of cells
in order to investigate the targeted effects of drug or other microenvironmental conditions
that may exist on particular subpopulations. As shown in Figure 2.2, we are able to
visualize and quantitatively measure the actual cell counts (live and dead) of HCC4011
sensitive and resistant subpopulations across time and under drug treatment. This
analysis is carried out using a machine-learning algorithm (as described in the methods
section) that allows one to train and identify live and dead, sensitive (RFP-positive) and
resistant (GFP-positive) cells within a population in an unbiased fashion. Alternatively,
intensity thresholds can be used to separate out populations; however, the heterogeneity
in fluorescence expression across individual cells can make the multi-parameter
machine-learning approach the ideal choice as it is less prone to errors (data not shown).
The importance of stromal cells, such as cancer-associated fibroblasts, in cancer
progression has resulted in the desire to have co-culture assays where tumor and
fibroblast populations can easily be distinguished. However, creating stable fluorescent
cell lines, a common approach to differentiate cell types, is time consuming and may not
be desirable for all cells, such as those obtained from patient samples, due to potential
off target effects. To overcome this, I developed an additional methodology to distinguish
between populations by utilizing differing differences in morphological features.
In a direct comparison to the fluorescence classification, the two agreed
approximately 93% of the time and this concordance was consistent over time and even
when drug was added (Figure 2.2b). As a ground truth assessment, we manually
annotated over 300 cells and categorized them as either tumor or fibroblast. We observed
11
drug effects on cancer cell viability, without toxic effects on non-cancerous cells, which is
an important consideration in drug development.

Figure 2.2 (a) HCC4011 (red) and HCC4011R (green) cells were admixed at a starting ratio of 1:1 and
treated with 2 μM erlotinib (E) or control as indicated. Representative images of cell populations on day 0
and 3 in the presence or absence of drug are shown. Cell counts were generated based off of Hoechst
nuclei segmentation and dead cells were identified by DRAQ7 staining. HCC4011 and HCC4011R cells
were differentiated based upon their fluorescent intensity of RFP and GFP, respectively. (b) Top –
Representative images of H3255 (red) and CCD-19Lu (blue) classification as compared with labeled
fibroblasts (pseudo-stained blue). Bottom Left – Venn diagram illustrates concordance between fluorescent
(assumed truth) and morphologic classification of H3255 and CCD-19Lu, which was determined to be
92.6%. Bottom Right - Classification based off of morphological characteristics was evaluated against
fluorescence intensity over time and in the presence of erlotinib (1 μM).

Comparison of HCS to traditional cell biology assays
I next performed a direct comparison of our HCS approach to common MTS and flow
cytometry assays (Figure 2.3). H3255 and H3255R-RFP cells were admixed at a starting
ratio of 1:1 (determined via HCS to be 54% H3255 and 46% H3255R-RFP) and the three
assay types were set up simultaneously (Figure 2.3a). As shown in Figure 2.3b, MTS is
unable to differentiate between the two cell populations (i.e. sensitive versus resistant)
nor discriminate between cells that are alive and growth arrested versus dead, resulting
12
in an erlotinib viability curve that more closely resembles the resistant population. The
discordant percent viability of the total population between MTS, flow cytometry, and HCS
may be due to: assay sensitivity (i.e. MTS – colorimetric absorbance readout; flow
cytometry – defined event number; HCS – absolute cell count), loss of dead cells during
sample preparation (flow cytometry), or distinctive metabolic activities between cell types
(MTS). Analysis of tumor population composition was comparable between HCS and flow
cytometry (Figure 2.3c, top); however, flow cytometry detected less than 0.01% dead
cells, likely due to their loss while preparing the samples. Although flow cytometry reveals
an increasing overall percentage of the H3255R population as erlotinib concentration
increases, it is through HCS that we determine that the resistant population counts are in
fact not increasing, but the overall population size is actually decreasing (Figure 2.3c,
bottom). In a clinical setting, it is important to not only detect changes in the total
population size, but also the cellular composition. A decrease in tumor size could be
mistaken for successful treatment, yet if the remaining population is completely resistant
to the current therapy, the prognosis may be poor depending on alternative treatments
available.
In addition to generating live and dead cell counts of the H3255 and H3255R
populations as a function of erlotinib concentration (Figure 2.3c, bottom), with the same
HCS experimental setup we calculate morphology features on a single-cell level (Figure
2.3d). There is a distinct difference in nuclear area between the sensitive and resistant
subpopulations in addition to changes in size over time as a result of drug treatment, with
H3255 cells showing a more dramatic shrinkage in nuclear size as drug concentration is
increased. The additional features of HCS provide a more accurate, detailed
representation of each subpopulation’s response to increasing drug. The advantages of
our HCS approach, including time saved and outputs generated, are summarized in
Figure 2.3a. The utility of MTS and flow cytometry assays is limited, especially in
heterogeneous systems such as cancer.
13

Figure 2.3. Comparison of HCS platform to standard cell biology assays, MTS and flow
cytometry (a) Illustrates schematic of experimental design. Cells were admixed at a starting ratio of
1:1 H3255 to H3255R-RFP cells and treated with erlotinib. Following a 72-hour incubation, cell
populations were analyzed via MTS, flow cytometry, or our HCS approach. The table summarizes
important features of each experimental design including processing time and data
outputs. (b) Viability curve displaying the data generated from each assay. The flow cytometry curve
remains constant across drug concentrations (at approximately 100% viability) due to loss of dead
cells during sample processing and the acquisition of relative counts (not absolute counts) of live
cells. (c) Top – Population percentages of H3255 live, H3255 dead, H3255R live, and H3255R dead
as determined by flow cytometry and HCS analyses. Bottom – Absolute cell counts generated by HCS
over time and in response to erlotinib treatment. (d) Box plot showing the distribution of H3255 and
H3255R nuclear area in response to increasing concentrations of erlotinib.
2.3 DISCUSSION
Our quantitative, image based HCS platform improves upon current cell viability assays
by providing more comprehensive insights into phenotypic dynamics of multiple cell types
14
within the contexts of a heterogeneous tumor microenvironment. Due to the high-
throughput nature of this approach, we are able to explore a variety of physiologically
relevant selective pressures (including different types and quantities), resulting in tens to
hundreds of conditions probed simultaneously. Here I highlighted the unique ability to
measure the temporal dynamics of phenotypic properties, specifically growth rate, that
are known to be important in cancer progression
21,22
. Significant work was also done on
evaluating morphological changes as a function of drug and environmental perturbation.
While the majority of this data is not shown, its application highlights the multiplexing
capabilities of this platform.
With our platform, we can track cellular fitness dynamics being driven by
evolutionary selective pressures that more closely mimic a tumor, which has a direct
impact on the tumor composition. Knowledge of such dynamical changes in cellular
heterogeneity and treatment response (e.g. changes in drug resistant fraction of cells)
may have a significant clinical impact at the time of treatment by influencing drug choice
and dosing strategy.
Our system has several advantages for scientists in the cancer biology field. On
the biological side, it allows one to (1) investigate phenotypic contributions at the single
cell or subpopulation level, capturing the heterogeneity of tumors; (2) perform more
complex experiments that encapsulate the relevant cellular and environmental features
of a tumor; and (3) conduct a more thorough multidimensional analysis of the data. On
the technical side, our platform, in comparison to traditional viability assays, has the ability
to visualize cells to identify whether confounding issues (i.e. seeding error, contamination,
morphology changes) may have occurred. Further, the enzymatic reactions that assays
such as MTS rely on may not be accurate in some microenvironmental contexts, such as
varying glucose conditions
23
, and are unable to differentiate between multiple cell
populations, which limits their ability to probe heterogeneous systems. Specifically, we
show drastic differences in cell viability outcome when directly evaluating results of an
erlotinib-treated co-culture of drug sensitive and resistant cells acquired by HCS
compared with MTS and flow cytometry assays. These differences can be attributed to
the shortcomings of traditional assays and include: the inability to distinguish between
15
two cell populations, an incomplete phenotypic characterization (cell birth versus death),
and technical challenges associated with assay preparation (Fig. 2.3).
It is also essential to acquire images of good quality to achieve the most from
further downstream analyses. If cells grow in clumps rather than in a uniform layer,
nuclear segmentation and subsequent morphological analyses may not be possible. If
cell populations are not morphologically distinct or are densely compact, automated
classification may be bias. Other important aspects to consider during experimental
design and analysis are reviewed in Boutros et al., 2015
24
. We also recognize that HCS
platforms are not available in all research labs or institutions, and therefore stress that the
described protocols can be accomplished on a smaller scale using any imaging platform
coupled with analysis protocols generated from open source software (i.e. CellProfiler
25
,
ImageJ) or other high content analysis software. Here we illustrate comparable
morphological outputs (nuclear and cell area) obtained from CellProfiler and Harmony.
Along with the benefits of using patient derived material come the technical hurdles
centered on the complexity of its composition. Therefore, the ability to distinguish between
cellular populations based on fluorescence or morphological features as described here
(Fig. 2.3d & 2.3e) may help mitigate some of the technical challenges with patient
samples.
The quantitative and standardized nature of HCS datasets allows for amenable
integration into physical science applications such as mathematical modeling of
population dynamics or emerging data formats (e.g. generation of digital cell lines –
MultiCellDS, www.multicellds.org). Integration of the physical and biological sciences has
proven to be an effective interdisciplinary approach for understanding cancer progression
and identifying novel treatment strategies
19,26-28
, and we believe this quantitative imaging
platform coupled with computational modeling will help further advance this field, as
exemplified in Chapter 3.

16
2.4 METHODS
Cell culture and reagents
CCD-19Lu (lung fibroblasts) were acquired from ATCC and maintained in EMEM media.
NSCLC isogenic erlotinib sensitive and resistant (R) cell lines (HCC4011, HCC4011R-
Met amplified; H3255, H3255R-T790M; PC9, and PC9R-T790M; HCC827, and
HCC827R-T790M) were acquired from Dr. William Pao (while at Vanderbilt University)
and cultured in RPMI 1640 media. All culture media was supplemented with 10% fetal
bovine serum (Gemini) and 1% penicillin/streptomycin solution and cells were kept under
standard laboratory conditions (5% CO2, 37°C) unless otherwise noted. Erlotinib resistant
cell lines were derived through dose escalation methods, as described in Ohashi et al.
(2012)
29
, and were maintained in 1 μM erlotinib (LC Laboratories # E-4007). Fluorescent
cell lines were generated by transduction with MISSION pLKO.1-puro-CMV-Turbo GFP
or RFP lentivirus particles (Sigma SHC003V, SHC012V) and maintained in 1 μg/mL
puromycin (Life Technologies #A11138-03). For primary cell isolation, colon tumors were
washed in PBS and minced using two scalpels. Tumors were then digested with 1.5
mg/mL collagenase, 20 µg/mL hyaluronidase, and 10 µM Ly27632. Cells were incubated
for 30 minutes, strained, and washed with PBS. Primary cell cultures were grown in
DMEM/F12 supplemented with 10% FBS and 1% penicillin/streptomycin. Experimental
protocols utilizing patient samples were approved by the University of Southern California
Health Science Campus Institutional Review Board (IRB protocol number HS-06-00678).
Cellular assay conditions
For HCS experiments, cells were seeded at either 4,000 (HCC4011, H3255) cells per well
at least eighteen hours prior to drug treatment on 96-well plates (Corning #3904). For a
standard assay setup, one plate per time point was seeded (all at the same time) and the
addition of dyes resulted in endpoint analysis. For hypoxia assays, cells were directly
placed in a Biospherix C-Chamber set at 0.1% O2 following seeding and subsequent work
was done in a hypoxia glove box to maintain constant oxygen control. For glucose
modulation, glucose-free media (0 g/L) was supplemented with D-(+)-Glucose solution
(Sigma, G8644) to achieve the desired concentration. Prior to imaging, cells were stained
with 5 μg/mL Hoechst 33342 (Invitrogen #H21492) and either 5 μg/mL Propidium Iodide
17
(PI) (Invitrogen #P1304MP), 5 μg/mL TO-PRO-3 Iodide (Life Technologies T3605), or 5
µM DRAQ7 (Biolegend, #424001) to identify cells as live or dead, respectively, depending
on the fluorescent channels being used for imaging. For morphology assays, cells were
stained with 10 μM CellTracker dye (Life Technologies, Orange CMRA #C34551) for thirty
minutes and washed with PBS to reduce background.
Image acquisition & analysis
Monolayer Experiments: Images were acquired on an Operetta High Content Screening
(HCS) System (Perkin Elmer) equipped with climate control (37°C, 5% CO2) using a 10X
objective lens. Each condition was assayed in at least triplicate wells and a minimum of
24 fields per well per time point (0, 48, 72 hours post drug addition) were imaged. Image
analysis was performed using the Harmony 3.5.2 software (Perkin Elmer) (or CellProfiler
2.1.1 when stated
25
). All data points considered for analysis were taken before any
confluence effects were apparent.
Calculation of Birth and Death Rates: Birth and death rates were calculated using time
series data consisting of live and dead cell counts obtained over a period of 0-72 hours.
To determine live and dead cell counts, nuclei were segmented using the Hoechst
channel. Propidium Iodide, TO-PRO-3 Iodide, or DRAQ7 intensity was used to classify
cells as dead. For Propidium Iodide, cells were identified as dead based upon an intensity
threshold. For the dead cell stains in the far red channel (TO-PRO-3 Iodide and DRAQ-
7), the background intensity -30 pixels (+/- 15, depending on cell type) outside the nucleus
was measured, and cells with a nuclear to background ratio greater than 1.2 were
classified as dead. To obtain net growth rates (birth minus death rates), the experimental
counts of live cells at various time points under each microenvironmental perturbation
were fit to an exponential growth model. A linear regression of the log-transformed data
was performed to obtain fitted rates at each drug concentration. Using the assumption of
a constant death rate per live-cell unit of time, death rates were calculated using the
accumulated amount of cell deaths over the time period. The cell birth rate was then
equated to the sum of the net growth rate and the death rate for each cell population
under the microenvironmental conditions perturbed.
18
Calculation of Cell Morphology Features: Nuclei were segmented using the Hoechst
channel and cytoplasmic boundaries were segmented based off of CellTracker stain. Cell
morphology features were calculated on a single-cell level. Cells with poor segmentation
were excluded.
Cellular Classifications Based on Fluorescence or Morphology Features: In order to
classify fluorescent cells into subpopulations (e.g. GFP-labeled erlotinib resistant versus
RFP-labeled erlotinib sensitive cells), intensity thresholds were applied to filter out cells
that did not express the fluorescent label of interest. To classify cells based on differences
in morphology, cells were segmented as described above. Data sets were then subjected
to linear classification analysis (PhenoLOGIC). Briefly, training sets consisting of 100 cells
per population were generated and thirty parameter values for various morphological
properties were calculated for every cell to identify the most relevant properties for cell
classification. For manual classification to determine ground truth, >300 cells were blindly
classified into subpopulations based on appearance and these results were compared to
the automated classification.
MTS assay
H3255 sensitive and resistant cells were admixed at a ratio of 1:1 and seeded at 4,000
cells per well (in Costar 96-well plates, #3596) in phenol-red free RPMI (Gibco, #11835-
030). Twenty-four hours after seeding, cells were treated with erlotinib at the designated
concentrations and incubated for 72 hours. CellTiter 96 AQUEOUS One Solution Cell
Proliferation MTS Assay (Promega G3582) was then performed following manufacturer’s
instructions and plates were read on SpectraMax M5 Microplate Reader (Molecular
Devices). All data points were performed in triplicate.
Flow cytometry
H3255 sensitive and resistant cells were admixed at a ratio of 1:1 and seeded at 2.2 x
10
6
cells per well in 10 cm plates. Twenty-four hours after seeding, cells were treated with
desired concentrations of erlotinib and incubated for 72 hours. Cells were trypsinized,
washed with PBS, and stained with Hoechst and DRAQ7. Samples were run in triplicate
on a BD FACSAria.

19
2.5 ADDITIONAL THOUGHTS
This project represents my first first author paper. I was fortunate that the framework for
the methodology was already in place from Dr. Mumenthaler’s previous work. I was able
to join the project at the stage of designing and executing experiments that would be most
relevant in showcasing the assay’s utility. There was still some troubleshooting and
optimizing involved which was very satisfying to solve and really gave me a positive
experience of the scientific process. It was also a great learning experience to work
through what goes into a full publication, as well as the overall process of submitting and
revising a manuscript.

20
CHAPTER 3
Integrating mathematical modeling and hypoxia-activated prodrugs to
prevent drug resistance in non-small cell lung cancer

Associated publication:
Lindsay D, Garvey CM, Mumenthaler SM, Foo J (2016) Leveraging Hypoxia-Activated Prodrugs to Prevent
Drug Resistance in Solid Tumors. PLoS Comput Biol 12(8): e1005077.
https://doi.org/10.1371/journal.pcbi.1005077.

3.1 INTRODUCTION
How and when drugs are delivered is an important variable in overall therapeutic
response. Recently, mathematical modeling has been recognized for its ability to model
population dynamics and treatment responses
30
and utilized to identify better drug
administration regimens
26
. Another potential key factor driving the emergence of drug
resistance is the spatial heterogeneity in the distribution of drug and oxygen throughout
a tumor caused by disorganized tumor vasculatures. Recently, researchers have
developed a class of novel drugs termed hypoxia activated prodrugs (HAPs) - that
penetrate to hypoxic regions where they are activated to kill tumor cells. One such
compound is evofosfamide, which consists of a radical anion linked to a potent DNA-
alkylating agent
31
. Under hypoxic conditions, the radical anion undergoes irreversible
fragmentation and releases the activated drug into the tumor
32,33
. We hypothesize that
the efficacy of treatments involving these prodrugs depends heavily on identifying the
correct treatment schedule and that mathematical modeling can be used to help design
potential therapeutic strategies combining HAPs with standard therapies to achieve long-
term tumor control or eradication. We develop this framework in the specific context of
EGFR-driven non-small cell lung cancer, which is commonly treated with the EGFR
targeting tyrosine kinase inhibitor erlotinib.
3.2 RESULTS
In order to evaluate and optimize the impact of erlotinib-evofosfamide combination
therapy on NSCLC tumor populations, an evolutionary mathematical modeling approach
21
where each tumor cell response is dependent upon local environmental concentrations
of oxygen and drug was utilized. The pseudo-spatial model is comprised of a weighted
series of environmental compartments whose oxygen and drug profiles mirror tumor
physiologic data. The model is parameterized using (i) experimentally calculated growth
rates under a spectrum of environmental perturbations of oxygen and erlotinib
concentration, (ii) published experimental results on cell viability in response to
evofosfamide therapy, (iii) tumor oxygenation measurements, and (iv) pharmacokinetic
data mapping evofosfamide and erlotinib dose to plasma concentration. I will not describe
in detail the specific equations that were used to define the model, as this was work that
was spear-headed by our collaborators. Instead, I will focus on the biological rational for
the parametrization of the models and the results that were generated.
Model Parameterization
In order to model tumor evolution within an environment with heterogeneous oxygen and
drug concentrations, we consider a stochastic population dynamic process in which the
cell population is distributed amongst a series of compartments with varying oxygen and
drug profiles. Oxygen concentration within the tumor decays exponentially as a function
of distance from the nearest blood vessel. This decay rate is parameterized in the model
based on estimates of the half-length away from the blood vessel
19
. This is used to define
the oxygen concentration in each microenvironmental compartment; hence every
compartment corresponds to a volume some distance from the nearest blood vessel. To
estimate the relative contributions of each of these compartments to the tumor
microenvironment, we utilize experimental data capturing relative frequencies of a
spectrum of oxygen partial pressures throughout solid tumors
34
. We consider a total of 32
environmental compartments to mirror this data. We then construct a mixture model of
compartments in which the weighting of each compartment is determined based on the
relative frequency of its corresponding oxygen partial pressure in the experimental profile.
22
A schematic of this process is depicted in Figure 3.1.

Fig 3.1. Tumor microenvironment modeling process.
This schematic shows the process used to model the tumor microenvironment as a set of discrete
compartments. A series of compartments is defined based on various distances from the nearest blood
vessel, and the oxygen concentration in each compartment is calculated accordingly. The relative weights
of the compartments are determined based on experimental observations of oxygen partial pressure
distribution in solid tumors.
Tumors consist of a heterogenous population of cells. In this framework, we chose
to study populations that are sensitive and resistant to erlotinib. Each cell population will
have different corresponding birth and death rates when faced with environmental
pressures, such as drug and oxygen. These rates also vary as drug concentrations
change over time. To reflect this, we first define distinct functions describing the individual
effects of erlotinib and evofosfamide on these birth and death rates. The growth kinetics
of the cancer cell population during treatment by each of these drugs are estimated using
a combination of pharmacokinetic and experimental cell viability data. Under the
assumption that the relative selective differences observed between different cell types
in vitro are preserved in vivo, the model predictions are useful for comparing dosing
schedules. However, the specific time scale and cell population sizes of the model
predictions are relevant to the in vitro setting only.
Effect of Erlotinib, Evofosfamide, and Oxygen on Growth Kinetics
Our recent experimental results have demonstrated that the response of non-small cell
lung cancer tumor cells to erlotinib is dependent on oxygen concentration. Using our high-
content imaging based platform, we calculated birth and death rates of isogenic sensitive
and resistant cell lines as functions of oxygen and erlotinib concentrations using an
23
exponential growth model (as described in Chapter 2). At baseline conditions, the
resistant cell line has a slightly higher growth rate compared to the sensitive line (Figure
3.2).

Fig 3.2. Net growth rate and plasma concentration functions for erlotinib and evofosfamide.
Examples of net growth rates of sensitive and resistant cells are shown as functions of erlotinib
concentration in (A) and evofosfamide concentration in (B). These rates are shown in blue for a low oxygen
concentration (0.33%), corresponding to that which is found in the compartment furthest from the blood
vessel, as well as in red for high oxygen concentration (10.5%), corresponding to that which is found in the
compartment closest to the blood vessel. Solid lines represent sensitive cell growth rates, and dotted lines
represent resistant cell growth rates.
We observe that overall erlotinib response for both cell types is more pronounced
in high oxygen compartments (which comprise the tumor bulk) than in low oxygen
compartments (Figure 3.2A). Resistant cells display essentially complete resistance
under low oxygen levels but only partial resistance under high oxygen levels, whereas
the sensitive cells exhibit only a minor increase in drug tolerance under low-oxygen levels
Next we consider the effect of evofosfamide on the erlotinib-sensitive and resistant
cell populations in each microenvironmental compartment. The mechanism of action
differs greatly between erlotinib and evofosfamide, so we made the assumption that the
sensitive and resistant populations will have the same response to evofosfamide
treatment. The net growth rate dependence on evofosfamide concentration and oxygen
concentration was calculated from cell viability experiments performed in Meng et al
33

(Figure 3.2B).
A
B
24

Growth Kinetics During Combination Therapy
We define each of our combination dosing regimen cycles as having two parts. The first
t1 hours are dedicated to treatment with erlotinib and the last t2 hours are used for
evofosfamide treatment. Because evofosfamide is eliminated so quickly from the blood
stream (half-life of 0.81 hours
35
), we assume that there is no residual evofosfamide during
the erlotinib treatment phase, given a sufficient amount of time between doses. Therefore
the birth and death rates during the first t1 hours in every cycle are governed by erlotinib
response kinetics. Since erlotinib has a much longer half-life than evofosfamide, during
the hours of evofosfamide treatment (t2) it is possible that some erlotinib will remain in the
blood stream by the beginning of this period. Therefore, the cellular birth and death rates
must reflect responses to both drugs. However, since erlotinib is primarily cytostatic while
evofosfamide is primarily cytotoxic, we assume that during this period of time, cellular
birth rates reflect the response to erlotinib while cellular death rates reflect the response
to evofosfamide.
Toxicity Constraints
An important consideration for this work was estimating a treatment space where the drug
combination would not elicit toxic effects. To achieve this, we utilized clinical trial data on
drug tolerability to parametrize toxicity constraints and estimate the space of all tolerated
single-agent and combination therapy dosing schedules.
For each single-agent treatment we defined a toxicity constraint curve representing
the relationship between frequency of drug administration and maximum tolerated dose.
In addition, we analyzed the overlapping toxicities between the two drugs as well as each
drug elimination rate to determine the necessary conditions for safely administering both
drugs in succession. This combination therapy constraint, together with the toxicity
constraint curves corresponding to each of the monotherapies, defines the space of all
tolerated dosing schedules for erlotinib (Figure 3.3A) and evofosfamide (Figure 3.3B).

25

Figure 3.3. Toxicity constraint curves for erlotinib and evofosfamide.
These curves depict the maximum tolerated doses of erlotinib (A) and evofosfamide (B) as functions of
frequency of dose administration. The black points are the coordinates corresponding to tolerated dosing
schedules, and the red points are the ordered pairs associated to dosing schedules that were not tolerated
in clinical trials. All points contained in the areas on and below these two curves make up the space of
tolerated monotherapy dosing schedules, and all points contained in the areas above these two curves
make up the space of dosing schedules which lead to dose-limiting toxicities. The curves themselves
represent the space of all monotherapy maximum tolerated dosing schedules.
Comparison of combination strategies with standard monotherapy schedules
To investigate the potential of combination therapies (and whether they would even
portentially provide overall benefit), we compared the simulated treatment outcomes from
several hypothetical combination therapies with the monotherapy schedules currently in
clinical use. The standard dosing schedule for erlotinib is 150 mg/day. Two evofosfamide
dosing schedules have been tested in a clinical trial and designated as maximum tolerated
dosing schedules: 670 mg/m
2
given every 3 weeks and 575 mg/m
2
given weekly. We
consider combination schedules which are clinically feasible and satisfy the toxicity
constraints described in the previous section. We examined a total of ten dosing
schedules and predicted the mean tumor size and probability of resistance over the
course of treatment using the model.
Overall, we find that all the combination therapies considered produce treatment
outcomes superior to the standard monotherapy schedules. The results of these
calculations up to recurrence time (the time at which the cancer cell population reaches
its initial size once again) are plotted in Figure 3.4A and Figure 3.4B, respectively. The
red curves show the evolutionary dynamics of a tumor during erlotinib monotherapy, the
blue curves show the dynamics during evofosfamide monotherapy, and the green curves
show the evolutionary dynamics of the cancer cell population during combination therapy.
26
The combination schedules result in lower average tumor sizes over the course of
treatment than those resulting from either of the monotherapies (Figure 3.4A). Even more
significantly, the probability of developing resistance decreases dramatically with the use
of combination therapy (Figure 3.4B). Under monotherapy with either drug, the probability
of resistance eventually reaches one (in agreement with clinical results); this is due to the
fact that sensitive cell division is not sufficiently inhibited by therapy to prevent the
emergence of resistance before eradication of the tumor. However the model predicts
that for a significant fraction of patients tumor eradication is possible under combination
therapy.
The means of the sensitive and resistant cell populations are shown separately in
Figure 3.4C for each type of dosing schedule: erlotinib alone, evofosfamide alone, and
combination therapy. Erlotinib monotherapy yields a steady but slow decline of the
sensitive cell population, due to the fact that the tumor oxygen distribution consists
primarily of hypoxic regions and erlotinib does not penetrate well to these areas. On the
other hand, treatment with evofosfamide alone targets hypoxic regions which comprise
the majority of the cell population, leading to an initial steep decline of the sensitive cell
population. However, the toxicity constraint during the subsequent break in treatment
causes the mean of the sensitive cells to quickly surpass the initial population size and
drives the production of erlotinib-resistant mutants. During combination therapy, however,
the cancer cell population demonstrates an initial steep decline due to evofosfamide,
followed by a long-term controlled phase due to the combination of evofosfamide and
erlotinib. This tight control over the sensitive cell population during combination therapy is
possibly due to the fact that cancer cells close to blood vessels are receiving lethal
concentrations of erlotinib while cancer cells in hypoxic regions are targeted by
evofosfamide.

27

Figure 3.4 Tumor evolutionary dynamics over time, given a variety of single-agent and combination
therapies.
Mean tumor size (A) and probability of resistance (B) are calculated up to recurrence time for a tumor with
an initial population of 1.6 ⋅ 10
6
sensitive cells undergoing treatment with each of the ten dosing schedules
defined in Table 3. Each labeled curve corresponds to the dosing schedule with the matching letter in Table
3. For the sake of comparison, results due to dosing schedules using erlotinib alone are shown in red,
results due to dosing schedules using evofosfamide alone are shown in blue, and results due to
combination therapies are shown in green. Mean tumor size for one of each of these three types of dosing
schedules is broken down into the means of sensitive and resistant cells in (C). (D) shows the expected
tumor size for combination strategies, conditioned upon the event of developing resistance.
Figure 3.4D demonstrates that for the patients who develop resistance, the
average length of time until tumor recurrence is longer for all of the combination therapy
dosing schedules than it is for either of the standard monotherapies. For example, it takes
patients who develop resistance 40.54% longer to rebound on Schedule I compared with
standard erlotinib therapy. However, the length of time until recurrence as well as the
overall probability of resistance varies between specific combination schedules; this
serves as motivation for identifying the optimal timing and dosage sequence for
A B
C D
28
combination schedules in the following section. Finally, we note that the combination
strategies were also compared to optimized monotherapies (subject to the toxicity
constraints developed in the previous section), and we found that no tolerated
monotherapy schedule could outperform combinations in delaying or preventing
resistance (data not shown).
Optimized combination strategies
We next utilize the mathematical model to optimize the space of tolerated combination
treatment strategies (constrained by toxicity constraints derived in the previous section)
to minimize the probability of developing resistance or maximally delay recurrence.
We consider three distinct classes of combination therapies:
Class 1 - combining standard erlotinib monotherapy of 150 mg/day with a variety of
evofosfamide dosing schedules given 24 hours before erlotinib
Class 2 – decreasing the concentration of erlotinib to 7 mg twice daily (to reduce toxicity)
and decreasing the amount of time after erlotinib and before evofosfamide dosing
Class 3 – decreasing the amount of time after varying evofosfamide doses to 6 hours
before given 150 mg erlotinib
For each class, we start with a base erlotinib dosing schedule complying with the
monotherapy toxicity constraint curve in Figure 3.3A. Modifications to this schedule are
then made to incorporate varying doses of evofosfamide, from 0 (corresponding to
erlotinib monotherapy) to a maximal value (corresponding to evofosfamide monotherapy),
in a three-week period. The dose of evofosfamide is determined by the toxicity constraint
curve in Figure 3.3B. Whenever necessary, the minimum number of erlotinib doses are
removed to comply with the combination toxicity constraint described in the previous
section.
Combination schedules outperform monotherapy endpoints. We calculated the
means of the sensitive and resistant cells as well as the probability of resistance, after
nine weeks of treatment for every dosing schedule in each optimization class. We observe
that all of the combination therapies we have investigated yield a considerable benefit
over any of the optimal monotherapies in each dosing class (Figure 3.5). Furthermore, at
the end of treatment with evofosfamide, the tumor primarily consists of sensitive cells,
29
which agrees with our previous observation that evofosfamide is unable to control the
sensitive cell population without erlotinib. Note that some of the means are unrealistically
high for an in vivo setting (> 10
12
), due to the fact that
the model is parametrized using in vitro growth rates; however we are interested in the
relative difference between cell population sizes resulting from different therapies rather
than absolute population sizes.
The means of the sensitive and resistant cell populations for combination therapies
are plotted in Figure 3.5A and Figure 3.5B, respectively. The sums of these means, or
the mean tumor sizes, are plotted in Figure 3.5C. The probability of resistance for each
dosing schedule is plotted in Figure 3.5D. All four panels include results for Class 1 in
blue, results for Class 2 in red, and results for Class 3 in yellow. Note that in these figures,
every integer on the x-axis corresponds to a combination dosing schedule given by the
number of evofosfamide doses in a 3-week period, as defined in the previous paragraph.
All three classes lead to similar results, suggesting that our findings regarding the
characteristics of optimal combination dosing strategies are quite robust. However, the
tumor population sizes under the Class 3 schedules are generally less than half the
population sizes under Class 1, and also less than those under Class 2 (Figure 3.5). This
suggests that designing schedules that minimize the amount of time after a dose of
evofosfamide and before a dose of erlotinib may lead to better control of the tumor
population. This finding is in agreement with our previous observations that the tumor
population response to evofosfamide is strong but short-lived; hence quickly intervening
in the subsequent population growth phase is important.

30

Figure 3.5. Probability of resistance and means of sensitive and resistant cells at the end of
treatment with combination therapy.
For every dosing schedule in each optimization class, means of the sensitive cells (A), resistant cells (B),
and total tumor size (C), as well as probability of resistance (D), are calculated according to the model at
the end of nine weeks of treatment for a tumor initially consisting of 1.6 ⋅ 10
6
sensitive cells. The results
shown here only include dosing schedules from Class 1 (blue), Class 2 (red), and Class 3 (yellow) which
use a combination of both erlotinib and evofosfamide. Every integer on the x-axis represents a combination
dosing schedule defined by the number of evofosfamide doses administered in three weeks.
As we move along the spectrum of combination densities (horizontal axis) from
monotherapy with erlotinib to monotherapy with evofosfamide, there is a clear region in
the interior (approximately n = 9–17 evofosfamide doses) where the tumor size and
sensitive and resistant population size are minimized, as well as the overall probability of
developing resistance. Sequential alternating sequences are optimal. We observe that all
optimal dosing schedules correspond to values of n between 12 and 17. This implies that
combination therapies incorporating more frequent, smaller doses of evofosfamide result
in better treatment outcomes. Even more interestingly, all values of n correspond to the
same type of dosing schedule. Besides n = 12 in Class 2, every other optimal n
A B
C D
31
corresponds to a dosing schedule which alternates between a single dose of erlotinib and
a single dose of evofosfamide. We call these alternating dosing schedules. The dosing
schedule corresponding to n = 12 in Class 2 consists of two low doses of erlotinib for
every dose of evofosfamide, which is still quite similar to the alternating dosing schedules.
Thus, even though the optimization ranged over a full spectrum of treatment schedules
incorporating variable dose densities for each drug, the optimal therapies were those that
utilized close to an equal number of doses of evofosfamide and erlotinib in a sequential
alternating fashion.
3.3 DISCUSSION
In this work we considered an approach to investigate the use of hypoxia-activated
prodrugs (HAPs) to enhance the effectiveness of targeted therapies and prevent the
(usually inevitable) emergence of drug resistance. To this end we developed a model
reflecting the heterogeneity of oxygen and drug concentrations throughout a tumor to
describe the evolutionary dynamics of resistance emerging under combination HAP-
targeted therapy strategies. The model was parametrized using experimental and clinical
pharmacokinetic data to investigate potential combinations of the HAP evofosfamide with
the targeted tyrosine kinase inhibitor erlotinib against EGFR-activated non small cell lung
cancer.
We investigated combinations in which doses were not given simultaneously (to
avoid toxicities) and found that the complementary action of evofosfamide and erlotinib
results in a combined ability to control the tumor’s evolution and growth. In particular:
(i) Combination therapies outperform standard clinical monotherapies. This is most
significantly realized in reduction of the probability of developing resistance. The time to
progression, for those who develop resistance, is 40.54% longer using an optimal
combination therapy rather than standard monotherapy with erlotinib.
(ii) Sequentially alternating single doses of each drug leads to a maximal reduction in the
probability of developing resistance and an overall smaller tumor burden. Deviating
significantly from an equal number of evofosfamide and erlotinib doses leads to an
increase in both average tumor burden and the probability of developing resistance.
32
(iii) Strategies minimizing the length of time after an evofosfamide dose and before
erlotinib confer further benefits in reduction of tumor burden. The tumor population
response to evofosfamide is strong but short-lived; hence quickly intervening in the
subsequent population growth phase is important.
These alternating dosing schedules (and other similar dosing schedules) are likely
the most effective because the constant switching between erlotinib and evofosfamide
allows the strengths of these drugs to complement one another. Too much time spent
taking erlotinib without evofosfamide allows the sensitive cell population to remain quite
substantial for a long period of time (due to the lack of targeting the hypoxic regions),
which, in turn, leads to a high probability of a resistance mutation arising. On the other
hand, too much time spent on evofosfamide without erlotinib allows the sensitive cell
population to expand drastically since evofosfamide is unable to control its long-term
growth. Alternating between these two drugs allows each one to provide the necessary
control over the cancer cell population the other one is lacking. In addition, it is important
to consider the subpopulation of cancer cells each drug acts on. Erlotinib acts primarily
on portions of the tumor microenvironment close to blood vessels, whereas evofosfamide
acts primarily on hypoxic regions that are further from the blood stream. Because of this,
alternating frequently between the two drugs allows the entire population of cancer cells
in the tumor microenvironment to be constantly controlled by the drugs.
At the time the paper was published, all clinical trials utilizing HAPs for combination
therapy were administrating the drugs simultaneously, rather than sequential. We
observed a reduction in the probability of developing resistance that was dependent on
the exact timing and sequence of combination therapy. Therefore, we hypothesize that
clinical trial outcomes could be different – and potentially better - if drug scheduling was
altered based upon our mathematical modeling predictions.
In addition to its promising clinical implications, this work provides insight into the
biological factors which can cause a treatment strategy to either succeed or fail.
Specifically, analysis and comparison of the tumor evolutionary dynamics during single-
agent and combination therapy suggests that erlotinib and evofosfamide may be effective
together because they target separate subpopulations within the tumor microenvironment
33
and on much different scales of time with differing degrees of strength. This theory can
be generalized to predict which types of drugs have the potential to be strong partners in
combination therapy; specifically, this methodology can be applied to determine the
biological and pharmacokinetic parameters that may lead to treatment success or failure
with monotherapy or combination therapy. These findings highlight the importance of
designing combination therapies with drugs whose strengths complement each other in
order to maximize the therapeutic benefits. Another important implication of this work, and
something to consider when designing combination dosing regimens using two or more
drugs, is the role that variability in timing between the dosing of different drugs plays in
treatment outcomes.
This work gave rise to multiple promising improvements that could be made in the
treatment of non-small cell lung cancer. However, the dosing strategies proposed here
need to be tested in vivo to verify these model predictions. In addition, this work provided
a novel framework for defining drug toxicity constraints, which is sufficiently general to be
extended to any drug or combination of drugs. Possible extensions of this work include
considering the possibility of pre-existing resistance as well as modeling the bystander
effect, which refers to the idea that evofosfamide, once activated in a hypoxic region of
the tumor, diffuses outward and affects cancer cells in normoxic regions as well
33,36
. In
addition, it would be useful to explore the effect of HAPs other than evofosfamide on the
probability of developing resistance in order to determine whether the results presented
here are specific to evofosfamide or rather are a general phenomenon of HAPs used in
combination with tyrosine kinase inhibitors. Another possible extension of this work is to
model therapies which alter the tumor microenvironment by changing the distribution of
oxygen within the tumor, for example, and investigate how this affects treatment
outcomes. Since evofosfamide is hypoxia-activated and birth and death rates due to
erlotinib are microenvironment-dependent, there is good reason to suspect that
alterations to the tumor microenvironment would have a large impact on treatment
outcomes with both single-agent and combination therapy.

34
3.4 ADDITIONAL THOUGHTS
This work was done in close collaboration with Dr. Jasmine Foo (University of Minnesota)
and her graduate student, Danika Lindsay. It was a great opportunity to work with
researchers from another discipline to formulate and test hypotheses. A roadblock we
faced on our end was the difficulty of using evofosfamide in a laboratory setting. We had
major issues with drug solubility that did not allow me to perform experiments where I was
confident with the results. This is why there was no in vitro evofosfamide data generated
to parametrize the model. Not having this data, as well as having to rely on sometimes
unfounded assumptions, does leave me with some skepticism of the overall biological
conclusions. But, evofosfamide failed two Phase III clinical trials and it is interesting to
think whether optimizing how this drug was given would have pushed some of the
observed trends into statistically significant outcomes.

35
CHAPTER 4
Differential effects of anti-EGFR therapy on cells in the colorectal cancer
microenvironment and its consequence of drug resistance

Manuscript in progress
4.1 INTRODUCTION
Colorectal cancer has been well studied over the years, leading to a relative
understanding of the genetics involved in disease progression and the identification of
proteins that may benefit from therapeutic targeting. For example, supplementing
chemotherapy regimens with EGFR targeting agents, such as cetuximab, is now standard
of care for some patient cohorts. However, such treatment offers only modest benefit.
Many genetic alterations in the epithelial cells of tumors have been identified as sufficient
to confer resistance to cetuximab, such as mutations in KRAS, BRAF, and alterations in
the PIK3CA/PTEN pathway
37,38
. Yet, the mechanism of an estimated 10-30% of patients
with initial clinical resistance still remains unknown
38,39
.
The epithelial cells of tumors are surrounded by various types of non-transformed
cells, collectively termed stroma. A major subset of the tumor stroma consists of cancer-
associated fibroblasts (CAFs), which have been implicated across cancer types in
promoting various stages of tumorigenesis, including therapeutic resistance
13
. However,
to our knowledge, no one has investigated the potential role of CAFs in contributing to
cetuximab resistance in colorectal cancer.
CAFs arise from multiple origins, including resident fibroblasts, epithelial cells, and
distant bone marrow mesenchymal stem cells
40
. They are molecularly characterized by
expression of CAF-specific markers (such as aSMA, fibronectin, etc.), but also express
proteins fundamental to cellular processes, including EGFR. CAFs surround cancer cells
and therefore are also exposed to drug at comparable levels. The effect of targeting
proteins such as EGFR in CAFs, which are not addicted to this pathway, has not
previously been investigated.
In this final, more prominent stage of my PhD work, I set out to investigate the
effects of CAFs themselves on cancer cell response to cetuximab. Ultimately, it led to the
36
observation that targeted therapies, such as cetuximab, may have unintended
consequences on surrounding cells through the identification of a novel environment-
mediated mechanism of cetuximab resistance.
4.2 RESULTS
CAFs decrease cancer cell sensitivity to cetuximab
Using the experimental setup discussed in Chapter 2, I studied the composition and
viability of cancer cells and CAFs over time in response to cetuximab treatment. When
co-cultured at a starting ratio of approximately 1:1, the presence of CAFs prevented DiFi
cell death, even at higher concentrations of cetuximab (Figure 4.1A, 4.1B). Further, when
we increased the starting fraction of CAFs-to-cancer cells, we observed a stronger
protective effect against cetuximab as evidenced by an increased growth rate (Figure
4.1C).
Next, I sought to identify whether the CAF protective effect was dependent on
some sort of physical cellular interaction, or whether it was based on something found in
the CAF secretome. Keeping in mind that CAF secretome may change in response to
drug treatment, I collected media from untreated and treated CAFs. (Cetuximab was
removed from the treated CAF media using a/g resin). The conditioned media from
cetuximab treated CAFs (CMtx) provided more protection than untreated CAF
conditioned media (CM) (Figure 4.1D).

37

Figure 4.1. CAFs protect cancer cells from cetuximab treatment. Co-culture of cancer cells and CAFs
in an approximate 1:1 ratio results in cancer cell protection against cetuximab treatment, as shown in
representative images (A) and calculated growth rates in a dose response curve (B). This protective
38
effect is also dependent on the starting fraction of CAFs, with a increasing proportion of CAFs resulting
in increased DiFi growth rate in the presence of cetuximab. (D) The difference (D) between untreated
and cetuximab treated (1 µg/mL) DiFi growth rate is smaller in cetuximab-treated CAF conditioned
media (CMtx) compared to CAF conditioned media (CM), signifying CMtx is more protective than CM.

CAFs secrete more EGF in response to cetuximab treatment
In order to determine which CAF secreted factors may be contributing to cetuximab
resistance, a cytokine array was performed to compare CM versus CMtx. As expected,
there was inter-patient heterogeneity among the factors secreted; however, surprisingly,
the only common cytokine that was differentially expressed in the treated vs. untreated
samples was EGF (Figure 4.2A). This cetuximab-induced increase in CAF secretion of
EGF was confirmed via ELISA, with at least a two-fold increase seen for each patient-
derived CAF (Figure 4.2B). We next wanted to test whether this was a CAF specific effect,
or something seen across all cell types. Cancer cell lines, patient-derived organoids, and
normal patient-derived colon fibroblasts were all assayed and found to secrete very low
baseline levels of EGF and did not increase EGF secretion upon treatment with cetuximab
(Figure 4.2B). Furthermore, EGFR inhibition by erlotinib, a small molecule inhibitor that
binds to the intracellular tyrosine kinase domain, or treatment with oxaliplatin, a
chemotherapy used to treat colorectal cancer, did not cause increased secretion of EGF
(Figure 4.2C). This suggests that the increased EGF secretion by CAFs is specific to
cetuximab binding to the extracellular region of EGFR and not a general response to
inhibition of the EGFR pathway or a general stress response.
39

Figure 4.2. (A) Secretion of one cytokine – EGF - was significantly upregulated in all three patient-
derived CAF lines after treatment with cetuximab. Increase in EGF secretion in response to cetuximab
in CAFs, but not cancer or normal fibroblasts, was validated via ELISA (B). Other drugs, including
chemotherapy and molecular targeting EGFR agent, did not elicit an increase in EGF secretion in any
of the CAF lines (C).

Exogenous EGF causes cetuximab resistance in 2D and 3D models
Standard 2D culture media contains levels of EGF that are barely detectable by ELISA.
In these conditions, our cell lines are sensitive to cetuximab. However, when EGF is
introduced in conjunction with cetuximab, the DiFi cells growth rates increases at in
proportion with EGF concentration at each cetuximab dose (Figure 4.3A).
We hypothesized that cancer cell survival in the presence of EGF and cetuximab
was due to sustained signaling through the MAPK pathway. Analysis of protein levels
downstream from EGFR support this hypothesis. When stimulated with EGF, we
observed increased levels of pEGFR and pHER2 (EGFRs favored heterodimer partner in
40
the presence of EGF
41
). Activation of the downstream ERK1/2 protein was also observed.
In the presence of cetuximab, this pathway was shut down as evidenced by no activation
of EGFR and HER2 and undetectable pERK1/2 levels. When cells were pre-incubated
with cetuximab and treated with EGF, we observed MAPK activation that was dependent
on the concentration of EGF with increased levels of pEGFR, pHER2, and pERK1/2
observed two-hours post stimulation in the higher EGF concentrations (Figure 4.3B).

3D patient-derived organoid models more accurately resemble patient tumors given the
genetic and microenvironmental heterogeneity that exists in these systems. However, this
primary culture media contains various supplements, including EGF, to support long term
growth. Previous studies have warned about the potential bias these exogenous factors
may impart, especially in the context of drug response
42
. When we lowered EGF
concentration in the media (from the previously defined 50 ng/mL), we restored cetuximab
sensitivity in the organoids (Figure 4.3C).

41

Figure 4.3. (A) DiFi sensitivity to cetuximab is decreased in proportion to increasing concentrations of
exogenous EGF. (B) Additionally, downstream EGFR signaling in cetuximab treated DiFi cells was
also increased in proportion to added EGF. (C) 3D patient-derived ORG12620 sensitivity to cetuximab
is also decreased in proportion to increasing concentrations of exogenous EGF.

Secreted EGF from cetuximab-treated CAFs is sufficient to render cancer cells
resistant to cetuximab
CAFs secrete many factors. We wanted to verify that EGF was the factor in the CMtx that
was rescuing cancer cells from cetuximab treatment. To do this, we incubated CMtx with
EGF-neutralizing antibody (CMtx-EGF) and observed cancer cell response to cetuximab.
Decrease in EGF levels between CMtx and CMtx-EGF was verified via ELISA. I found
that when DiFi cells were exposed to CMtx-EGF, they were resensitized to cetuximab
treatment at a level resembling baseline response (Figure 4.4A, 4.4B). This supports the
42
overall hypothesis that it is EGF secreted by cetuximab treated CAFs that causes cancer
cell non-response to cetuximab.

Figure 4.4. Cetuximab-treated CAF13000 conditioned media neutralized with EGF antibody (CMtx-
EGF) returned DiFi sensitivity to cetuximab closer to baseline in (A). (B) The difference (D) between
untreated and cetuximab treated (1 µg/mL) DiFi growth rate is smaller in cetuximab-treated CAF
conditioned media (CMtx) compared to CMtx, suggesting EGF is the factor in CMtx that is increasing
cancer cell resistance.

43
4.3 DISCUSSION

Molecular targeting agents have revolutionized the treatment of cancer. However, a large
portion of the proteins that are being targeted are expressed in cell types other than
cancer cells. There is limited research being done to investigate potential phenotypic
responses to targeted agents other than viability that may occur in cells throughout the
body, including the stromal cells of the tumor microenvironment. We discovered CAF
secretion of EGF is increased in response to cetuximab and the presence of exogenous
EGF results in resistance to cetuximab treatment. The observation that EGF can
outcompete EGFR antibody in cell models has been previously reported
42-45
, however, a
source of EGF secretion in the tumor microenvironment had not previously been
identified.
The ratio of CAFs to tumor varies across patient tumors. If this ratio is low, it is
likely the concentration of EGF secreted by the CAFs is not sufficient to rescue the cancer
cells from cetuximab treatment. However, as the tumor shrinks from cetuximab treatment,
the ratio will increase (since CAFs viability is not affected) and it is possible EGF levels
will be adequate for protection from cetuximab and therefore also be a cause of relapse
to treatment.
There have been multiple clinical studies looking at biomarkers for cetuximab
response that may supplement genetic alterations currently used for treatment
stratification The CMS4 subtype of tumors– which are characterized by a high stromal
density
39
– were found to be prognostic for response to anti-EGFR treatment. Further,
when looking at plasma levels of EGFR ligand, increase in EGF levels from 2 weeks post-
treatment compared to initial treatment levels was significantly higher in nonresponders
46
.
Another independent study also identified significant increase in EGF serum levels after
cetuximab treatment which corresponded to disease progression
47
. These clinical
observations support our finding that CAFs treated with anti-EGFR therapy increase
secretion of EGF which leads to cancer cell resistance.
Most therapeutic agents used for treating cancer are given systemically and
therefore have the potential to effect cells throughout the body. While this concept is
44
considered extensively in the context of adverse side effects, the potential of one’s body
contributing to better or worse overall response to drug has just recently begun garnering
attention. For example, microbiome composition is indicative of overall response to PD-1
based immunotherapy
48
.
In conclusion, our data suggests that EGF secretion by cetuximab treated CAFs is
a previously unknown mechanism of resistance to anti-EGFR treatment in colorectal
cancer. Our findings emphasize the importance of considering how therapies may
influence the tumor surroundings and ultimately alter tumor response.
4.4. METHODS
Cell culture and reagents
DiFi and LIM1215 cell lines were obtained from Dr. Alberto Bardelli (University of Torino)
and cultured in DMEM and RPMI, respectively, supplemented with 10% FBS and 1%
penicillin/streptomycin (P/S) under standard laboratory conditions (5% CO2, 37°C).
Primary cell culture: Human tumor organoids and cancer-associated fibroblasts
Tumor tissues were received from colorectal-cancer patients under Institutional Review
Board (IRB) approval at the Norris Comprehensive Cancer Center of USC. Patient-
derived metastatic colorectal tumor organoids (CTOs) were developed following
previously described methods
49
.
Cancer associated fibroblasts (CAFs) were generated from the same tumor tissue
by culturing cells on plastic tissue culture plates and letting the fibroblasts grow out over
1-2 passages. Cells were then verified as CAFs via qPCR and immunofluorescent
staining for common CAF markers. For all experiments, CAFs were used between
passages 2 and 8.
Imaging growth rate assays
Cells were seeded in four 384-well plates 24-hours prior to treating with cetuximab. On
day 0, cells were treated with drug at the desired concentration. Prior to imaging, cells
were stained with 5 μg/mL Hoechst 33342 (nuclear dye) and 5 μg/mL Propidium Iodide
(PI) (vital dye) to identify cells as live or dead, respectively. Individual plates were imaged
on day 0, 2, 3 and 5. Cells were then segmented based upon the nuclear dye using
PerkinElmer Harmony software. In order to differentiate cell types in co-culture assays,
45
morphological features were calculated and used to train a machine learning algorithm to
classify cells as either ‘CAF’ or ‘tumor,’ as described in Garvey et al.
50,51
. Propidium iodide
intensity levels were calculated and cells were classified as ‘dead’ if their intensity was
above the established threshold. Growth rates for each cell type were calculated as
previously described
19,50
by fitting the live cell counts over time to an exponential growth
model.
Collection and processing of conditioned media
When CAF cultures reached about 80% confluency, media was changed to DMEM
supplemented with 1% P/S and 10% FBS. Cells were treated with 1 µg/mL cetuximab or
IgG isotype control and incubated for 72 hours (unless otherwise specified). Media was
collected, spun to remove debris, and stored at -80°C. Media was thawed and incubated
overnight with Protein A/G Agarose to remove remaining drug or IgG isotype control. After
separation of the media from the agarose pellets, 0.5 µg/mL EGF neutralizing antibody
or IgG isotope control was added and media was incubated at 37°C for one hour.
Secretome Analysis
When cells were at ~70% confluence, culture media was replaced with FBS and P/S free
DMEM for three days. This media was then collected, spun down to remove debris,
aliquoted, and stored at -80°C. Frozen conditioned media was thawed and subjected to
cytokine arrays or ELISAs, both following manufacturer’s instructions.
Western Blotting
Cells were serum-starved overnight and treated with EGF or cetuximab for the time
specified. Cells that were treated with EGF and cetuximab were incubated with cetuximab
for 1 hour prior to addition of EGF for the specified time. Cells were harvested on ice and
needle treated using RIPA buffer supplemented with protease and phosphatase
inhibitors. The protein lysates (30 µg) were then resolved on 4-12% Bis-Tris gradient pre-
cast gels (Invitrogen), transferred onto a PVDF membranes via semi-dry transfer.
Immunoblotting was then performed with corresponding antibodies.
Organoid Viability Assay
Organoids were digested to single cells using Organoid Harvesting Solution (Cultrex) and
40 µm cell strainer. 2,500 cells were seeded per well in a white 96-well plate. After three
days to allow organoids to form, wells were treated with drug in quadruplicate in the
46
specified media conditions. Following 5 days, media was removed, and organoids were
incubated in harvesting solution for 30 minutes at 4C to degrade BME. CellTiter-Glo 3D
Reagent was then added, cells were incubated for 30 minutes at RT, and the plate was
read on a luminometer.
4.5 FUTURE DIRECTIONS
At the time of writing this draft, there are still ongoing experiments I hope to complete
myself as well as others I hope will be completed by others to continue research into this
subject matter.
Experiments in Progress
• Prognostic ability of CAF composition to predict cetuximab response
Colorectal cancer tumors with high stromal content have been correlated with
worse overall survival
15
. Additionally, the CMS4 subclass of colorectal cancers,
which is classified as having a high mesenchymal component (which includes
stromal cells such as CAFs), has been shown to be correlated with worse
outcomes to cetuximab
39
. While these studies provide evidence that CAFs may be
involved in cetuximab non-response, they are still not a direct correlation.
We have worked to acquire samples taken from colorectal cancer patients prior to
treatment. We have a tissue microarray with 50 patient cores and slides from 47
patients stained with H&E. Currently, we are awaiting MTA and IRB approval.
Once this is approved, the samples will be sent to Dr. Brent Larson at Cedars Sinai
for stratification based upon stromal cell composition (low, medium, high). We will
use the outcome data we have for all of these patients to see if we can prove a
direct correlation between high CAF percentage and worse cetuximab outcome.
• Cetuximab-treated CAF conditioned media (CMtx) protects patient-derived
organoids from cetuximab treatment
Extensive work has been done in our 2D cell culture lines to demonstrate the
protective effect of CMtx. This work is still in the process of being translated to our
patient derived organoid models. We have shown that their response to cetuximab
is dependent on EGF concentration. However, performing comparable assays in
our 3D model has proved troublesome for the following reasons:
47
A. The ratio of cells in organoid assays to CAFs media is much greater than 2D
assays (and corresponds to ‘low’ CAF tumors)
B. Factors secreted by CAFs appear to degrade BME organoids are grown in
C. Organoids require some EGF to grow and die in EGF deficient conditions

Suggested Future Experiments
• Elucidating the mechanism of increased EGF secretion by cetuximab treated
CAFs.
One of the most common questions I and others receive when presenting this work
is why do CAFs respond to anti-EGFR therapy with increasing EGF secretion. It
seems to be some sort of a feedback loop, however, what is most confounding to
me is why this increase in secretion is not seen when EGFR is targeted with a TKI.
EGFR signaling is elegantly complex, so there must be something about the
inhibition caused by targeting extracellular versus intracellular regions. It will also
be important to investigate whether this phenomenon is caused solely by the cells
increasing rate of secretion, or if transcription and/or translational regulation are
the steps that are affected.
• Pharmacologically preventing CAF-mediated cetuximab resistance
Many current clinical trials in the cancer field are studying combination therapy to
combat the overwhelming rates of resistance that occurs. EGF preferentially
signals through EGFR-HER2 dimerization. I hypothesize that combining a HER2
inhibitor (e.g. trastuzumab, lapatinib, pertuzumab) with cetuximab may have a
modest effect on increasing cetuximab response in colorectal cancer patients with
high stromal tumors. Preliminary in vitro experiments have confirmed this, with 2C4
and cetuximab and trastuzumab and cetuximab combinations increasing CMtx
treated cancer cells sensitivity to cetuximab. There are current clinical trials testing
this combination, however, the outcomes have not been as advantageous as
promised. However, specifically testing this regiment on patients with high stromal
content could have more promising effects. Additionally, there could be other
combination therapies that prove more effective. Directly targeting EGF, in addition
to it being a challenging drug target, likely would not be advantageous given the
48
necessary role of EGF throughout the body. Further work into the signaling
cascades caused by CAF-secreted EGF and cetuximab in cancer cells could
provide further insight into potential therapeutic strategies to overcome resistance.
4.6 ADDITIONAL THOUGHTS
This project was developed from an overall hypothesis that CAFs could cause cetuximab
resistance and a goal to identify a mechanism. This wasn’t necessarily a novel idea –
CAFs have been shown to cause resistance in a variety of cancer and drug contexts. It
was surprising and exciting to ‘stumble’ across a more complicated scenario – where it
wasn’t just the interaction of the cell types that drove resistance, but the dependence of
both cells ‘feeling’ and responding to the drug pressure. Honestly, it’s quite impressive
how the cells work in harmony, but also quite scary. This system contains only a few
factors of the tumor microenvironment. It is quite overwhelming to think of how the
interaction between the whole microenvironment drives certain aspects of tumorigenesis
and how much we are likely missing and misinterpreting in our cell culture models.

49

REFERENCES
1 Fidler, I. J. et al. Modulation of tumor cell response to chemotherapy by the organ
environment. Cancer Metastasis Rev 13, 209-222 (1994).
2 Tan, Q., Saggar, J. K., Yu, M., Wang, M. & Tannock, I. F. Mechanisms of Drug
Resistance Related to the Microenvironment of Solid Tumors and Possible
Strategies to Inhibit Them. Cancer J 21, 254-262,
doi:10.1097/PPO.0000000000000131 (2015).
3 Sullivan, R., Pare, G. C., Frederiksen, L. J., Semenza, G. L. & Graham, C. H.
Hypoxia-induced resistance to anticancer drugs is associated with decreased
senescence and requires hypoxia-inducible factor-1 activity. Mol Cancer Ther 7,
1961-1973, doi:10.1158/1535-7163.MCT-08-0198 (2008).
4 Brown, J. M. & Wilson, W. R. Exploiting tumour hypoxia in cancer treatment. Nat
Rev Cancer 4, 437-447, doi:10.1038/nrc1367 (2004).
5 Wartenberg, M. et al. Regulation of the multidrug resistance transporter P-
glycoprotein in multicellular tumor spheroids by hypoxia-inducible factor (HIF-1)
and reactive oxygen species. FASEB J 17, 503-505, doi:10.1096/fj.02-0358fje
(2003).
6 Lu, C. W., Lin, S. C., Chen, K. F., Lai, Y. Y. & Tsai, S. J. Induction of pyruvate
dehydrogenase kinase-3 by hypoxia-inducible factor-1 promotes metabolic switch
and drug resistance. J Biol Chem 283, 28106-28114,
doi:10.1074/jbc.M803508200 (2008).
7 Lu, C. W. et al. Overexpression of pyruvate dehydrogenase kinase 3 increases
drug resistance and early recurrence in colon cancer. Am J Pathol 179, 1405-1414,
doi:10.1016/j.ajpath.2011.05.050 (2011).
8 Zeng, M., Kikuchi, H., Pino, M. S. & Chung, D. C. Hypoxia activates the K-ras
proto-oncogene to stimulate angiogenesis and inhibit apoptosis in colon cancer
cells. PLoS One 5, e10966, doi:10.1371/journal.pone.0010966 (2010).
50
9 Hongo, K. et al. Hypoxia enhances colon cancer migration and invasion through
promotion of epithelial-mesenchymal transition. J Surg Res 182, 75-84,
doi:10.1016/j.jss.2012.08.034 (2013).
10 Luoto, K. R., Kumareswaran, R. & Bristow, R. G. Tumor hypoxia as a driving force
in genetic instability. Genome Integr 4, 5, doi:10.1186/2041-9414-4-5 (2013).
11 Kondo, A. et al. Hypoxia-induced enrichment and mutagenesis of cells that have
lost DNA mismatch repair. Cancer Res 61, 7603-7607 (2001).
12 Reynolds, T. Y., Rockwell, S. & Glazer, P. M. Genetic instability induced by the
tumor microenvironment. Cancer Res 56, 5754-5757 (1996).
13 Straussman, R. et al. Tumour micro-environment elicits innate resistance to RAF
inhibitors through HGF secretion. Nature 487, 500-504, doi:10.1038/nature11183
(2012).
14 Meads, M. B., Gatenby, R. A. & Dalton, W. S. Environment-mediated drug
resistance: a major contributor to minimal residual disease. Nat Rev Cancer 9,
665-674, doi:10.1038/nrc2714 (2009).
15 Huijbers, A. et al. The proportion of tumor-stroma as a strong prognosticator for
stage II and III colon cancer patients: validation in the VICTOR trial. Ann Oncol 24,
179-185, doi:10.1093/annonc/mds246 (2013).
16 Rasanen, K. & Vaheri, A. Activation of fibroblasts in cancer stroma. Exp Cell Res
316, 2713-2722, doi:10.1016/j.yexcr.2010.04.032 (2010).
17 Wang, W. et al. Effector T Cells Abrogate Stroma-Mediated Chemoresistance in
Ovarian Cancer. Cell 165, 1092-1105, doi:10.1016/j.cell.2016.04.009 (2016).
18 Millar, A. W. & Lynch, K. P. Rethinking clinical trials for cytostatic drugs. Nat Rev
Cancer 3, 540-545, doi:10.1038/nrc1124 (2003).
19 Mumenthaler, S. M. et al. Evolutionary modeling of combination treatment
strategies to overcome resistance to tyrosine kinase inhibitors in non-small cell
lung cancer. Mol Pharm 8, 2069-2079, doi:10.1021/mp200270v (2011).
51
20 Bhang, H. E. et al. Studying clonal dynamics in response to cancer therapy using
high-complexity barcoding. Nat Med 21, 440-448, doi:10.1038/nm.3841 (2015).
21 Anderson, A. R., Weaver, A. M., Cummings, P. T. & Quaranta, V. Tumor
morphology and phenotypic evolution driven by selective pressure from the
microenvironment. Cell 127, 905-915, doi:10.1016/j.cell.2006.09.042 (2006).
22 Greaves, M. & Maley, C. C. Clonal evolution in cancer. Nature 481, 306-313,
doi:10.1038/nature10762 (2012).
23 Huang, K. T., Chen, Y. H. & Walker, A. M. Inaccuracies in MTS assays: major
distorting effects of medium, serum albumin, and fatty acids. Biotechniques 37,
406, 408, 410-402, doi:10.2144/04373ST05 (2004).
24 Boutros, M., Heigwer, F. & Laufer, C. Microscopy-Based High-Content Screening.
Cell 163, 1314-1325, doi:10.1016/j.cell.2015.11.007 (2015).
25 Carpenter, A. E. et al. CellProfiler: image analysis software for identifying and
quantifying cell phenotypes. Genome Biol 7, R100, doi:10.1186/gb-2006-7-10-
r100 (2006).
26 Leder, K. et al. Mathematical modeling of PDGF-driven glioblastoma reveals
optimized radiation dosing schedules. Cell 156, 603-616,
doi:10.1016/j.cell.2013.12.029 (2014).
27 Chmielecki, J. et al. Optimization of dosing for EGFR-mutant non-small cell lung
cancer with evolutionary cancer modeling. Sci Transl Med 3, 90ra59,
doi:10.1126/scitranslmed.3002356 (2011).
28 Altrock, P. M., Liu, L. L. & Michor, F. The mathematics of cancer: integrating
quantitative models. Nat Rev Cancer 15, 730-745, doi:10.1038/nrc4029 (2015).
29 Ohashi, K. et al. Lung cancers with acquired resistance to EGFR inhibitors
occasionally harbor BRAF gene mutations but lack mutations in KRAS, NRAS, or
MEK1. Proc Natl Acad Sci U S A 109, E2127-2133, doi:10.1073/pnas.1203530109
(2012).
52
30 Michor, F. & Beal, K. Improving Cancer Treatment via Mathematical Modeling:
Surmounting the Challenges Is Worth the Effort. Cell 163, 1059-1063,
doi:10.1016/j.cell.2015.11.002 (2015).
31 Liu, Q. et al. TH-302, a hypoxia-activated prodrug with broad in vivo preclinical
combination therapy efficacy: optimization of dosing regimens and schedules.
Cancer Chemother Pharmacol 69, 1487-1498, doi:10.1007/s00280-012-1852-8
(2012).
32 Phillips, R. M. Targeting the hypoxic fraction of tumours using hypoxia-activated
prodrugs. Cancer Chemother Pharmacol 77, 441-457, doi:10.1007/s00280-015-
2920-7 (2016).
33 Meng, F. et al. Molecular and cellular pharmacology of the hypoxia-activated
prodrug TH-302. Mol Cancer Ther 11, 740-751, doi:10.1158/1535-7163.MCT-11-
0634 (2012).
34 Weickhardt, A. J. et al. Local ablative therapy of oligoprogressive disease prolongs
disease control by tyrosine kinase inhibitors in oncogene-addicted non-small-cell
lung cancer. J Thorac Oncol 7, 1807-1814, doi:10.1097/JTO.0b013e3182745948
(2012).
35 Araten, D. J. et al. A quantitative measurement of the human somatic mutation
rate. Cancer Res 65, 8111-8117, doi:10.1158/0008-5472.CAN-04-1198 (2005).
36 Sun, J. D. et al. Selective tumor hypoxia targeting by hypoxia-activated prodrug
TH-302 inhibits tumor growth in preclinical models of cancer. Clin Cancer Res 18,
758-770, doi:10.1158/1078-0432.CCR-11-1980 (2012).
37 Zhao, B. et al. Mechanisms of resistance to anti-EGFR therapy in colorectal
cancer. Oncotarget 8, 3980-4000, doi:10.18632/oncotarget.14012 (2017).
38 Misale, S., Di Nicolantonio, F., Sartore-Bianchi, A., Siena, S. & Bardelli, A.
Resistance to anti-EGFR therapy in colorectal cancer: from heterogeneity to
convergent evolution. Cancer Discov 4, 1269-1280, doi:10.1158/2159-8290.CD-
14-0462 (2014).
53
39 Trinh, A. et al. Practical and Robust Identification of Molecular Subtypes in
Colorectal Cancer by Immunohistochemistry. Clin Cancer Res 23, 387-398,
doi:10.1158/1078-0432.CCR-16-0680 (2017).
40 Madar, S., Goldstein, I. & Rotter, V. 'Cancer associated fibroblasts'--more than
meets the eye. Trends Mol Med 19, 447-453, doi:10.1016/j.molmed.2013.05.004
(2013).
41 Sweeney, C. et al. Growth factor-specific signaling pathway stimulation and gene
expression mediated by ErbB receptors. J Biol Chem 276, 22685-22698,
doi:10.1074/jbc.M100602200 (2001).
42 Luraghi, P. et al. A Molecularly Annotated Model of Patient-Derived Colon Cancer
Stem-Like Cells to Assess Genetic and Nongenetic Mechanisms of Resistance to
Anti-EGFR Therapy. Clin Cancer Res 24, 807-820, doi:10.1158/1078-0432.CCR-
17-2151 (2018).
43 Derer, S. et al. Impact of epidermal growth factor receptor (EGFR) cell surface
expression levels on effector mechanisms of EGFR antibodies. J Immunol 189,
5230-5239, doi:10.4049/jimmunol.1202037 (2012).
44 Matsuo, T. et al. Analysis of the anti-tumor effect of cetuximab using protein
kinetics and mouse xenograft models. BMC Res Notes 4, 140, doi:10.1186/1756-
0500-4-140 (2011).
45 Yoshida, T. et al. Matuzumab and cetuximab activate the epidermal growth factor
receptor but fail to trigger downstream signaling by Akt or Erk. Int J Cancer 122,
1530-1538, doi:10.1002/ijc.23253 (2008).
46 Loupakis, F. et al. EGFR ligands as pharmacodynamic biomarkers in metastatic
colorectal cancer patients treated with cetuximab and irinotecan. Target Oncol 9,
205-214, doi:10.1007/s11523-013-0284-7 (2014).
47 Han, S. W. et al. Phase II study and biomarker analysis of cetuximab combined
with modified FOLFOX6 in advanced gastric cancer. Br J Cancer 100, 298-304,
doi:10.1038/sj.bjc.6604861 (2009).
54
48 Routy, B. et al. Gut microbiome influences efficacy of PD-1-based immunotherapy
against epithelial tumors. Science 359, 91-97, doi:10.1126/science.aan3706
(2018).
49 Sato, T. et al. Long-term expansion of epithelial organoids from human colon,
adenoma, adenocarcinoma, and Barrett's epithelium. Gastroenterology 141, 1762-
1772, doi:10.1053/j.gastro.2011.07.050 (2011).
50 Garvey, C. M. et al. A high-content image-based method for quantitatively studying
context-dependent cell population dynamics. Sci Rep 6, 29752,
doi:10.1038/srep29752 (2016).
51 Garvey, C. M., Gerhart, T. A. & Mumenthaler, S. M. Discrimination and
Characterization of Heterocellular Populations Using Quantitative Imaging
Techniques. J Vis Exp, doi:10.3791/55844 (2017).

Abstract (if available)

Abstract Targeted agents have improved the efficacy of chemotherapy for cancer patients, yet there remains a lack of understanding of how these therapies affect the unsuspecting bystanders of the stromal microenvironment. Cetuximab, a monoclonal antibody therapy targeting the epidermal growth factor receptor (EGFR), is given in combination with chemotherapy as standard of care for a subset of metastatic colorectal cancer patients. Overall response to this treatment is underwhelming and, while genetic mutations that confer resistance have been identified, it is still not known why this drug is ineffective for some patients. We discovered that cancer-associated fibroblasts (CAFs), a major cellular subset of the tumor stroma, are able to provide a source of cancer cell resistance. Specifically, we observed that upon treatment with cetuximab, cancer-associated fibroblasts increased their secretion of EGF, which was sufficient to render neighboring tumor cells resistant to cetuximab treatment. Furthermore, we show the cetuximab-induced EGF secretion to be specific to CAFs and not tumor cells nor normal fibroblasts. Altogether, this work emphasizes the importance of considering the potential unintended consequences of therapeutically targeting cancer-driving proteins on non-tumorigenic cell types.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Development of a colorectal cancer-on-chip to investigate the tumor microenvironment's role in cancer progression

PDF

Exploration of the roles of cancer stem cells and survivin in the pathogenesis and progression of prostate cancer

PDF

Contribution of cancer associated fibroblasts to cancer progression

PDF

Bone marrow derived mesenchymal stem cells promote survival and drug resistance in tumor cells

PDF

Utilizing neurodevelopment to discover molecular pathways promoting medulloblastoma metastasis

PDF

Study of bone morphogenetic protein-2 and stromal cell derived factor-1 in prostate cancer

PDF

Bone marrow-derived mesenchymal stromal cells in the tumor microenvironment of neuroblastoma

PDF

An image-based identification of aggressive circulating tumor cell subtypes

PDF

Neuronal master regulator SRRM4 in breast cancer cells facilitates CNS-acclimation and colonization leading to brain metastasis

PDF

Bimodal effects of bone morphogenetic proteins in prostate cancer

PDF

Molecular classification of breast cancer specimens from tissue morphology

PDF

Pre-transplant conditioning and the role of the microenvironment during hematopoietic reconstitution

PDF

Identification of CBP/FOXM1 as a molecular target in triple negative breast cancer

PDF

Applying multi-omics in cancer liquid biopsy for improved patient monitoring and biomarker discovery

PDF

The role of endoplasmic reticulum chaperone protein GRP78 in breast cancer

PDF

Molecular mechanisms of chemoresistance in breast cancer

PDF

Targeting BCL-2 family proteins and plasminogen activator inhibitor-1 in turmor cell apoptosis

PDF

Developing a robust single cell whole genome bisulfite sequencing protocol to analyse circulating tumor cells

PDF

Evaluating novel chemotherapeutic strategies in colorectal and gastric cancer: the role of histone deacetylase inhibitors and human epidermal receptor family inhibitors

PDF

RNA methylation in cancer plasticity and drug resistance

Asset Metadata

Creator Garvey, Colleen Margaret (author)

Core Title Investigating the complexity of the tumor microenvironment's role in drug response

School Keck School of Medicine

Degree Doctor of Philosophy

Degree Program Cancer Biology and Genomics

Publication Date 01/27/2020

Defense Date 10/07/2019

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag CAFs,cetuximab,colorectal cancer,EGF,OAI-PMH Harvest,tumor microenvironment

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor DeClerck, Yves (committee chair), Agus, David (committee member), Mumenthaler, Shannon (committee member), Yu, Min (committee member)

Creator Email cgarvey@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-263185

Unique identifier UC11674746

Identifier etd-GarveyColl-8132.pdf (filename),usctheses-c89-263185 (legacy record id)

Legacy Identifier etd-GarveyColl-8132.pdf

Dmrecord 263185

Document Type Dissertation

Rights Garvey, Colleen Margaret

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA