MOLECULAR AND CLINICAL NETWORK ANALYSIS OF COLORECTAL
CANCER
By
Ertugrul Dalkic

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
DOCTOR OF PHILOSOPHY
Cell and Molecular Biology
2012

ABSTRACT
MOLECULAR AND CLINICAL NETWORK ANALYSIS OF COLORECTAL CANCER
By
Ertugrul Dalkic
Cancer is a large class of diseases and colorectal cancer is one of the leading types of
cancer. Systems level analysis of complex diseases like cancer requires the analysis of
relationships between different types of clinical data as well as molecular data. Common or
specific network features of colorectal cancer together with the other cancer types could be
identified by using different network approaches, such as the analysis of clinical data
associations, molecular signaling pathways of cancers, and specific interaction networks of
cancers. Firstly, a clinical network analysis has been performed on relationships between
different types of cancer and the drugs. We generated two cancer networks, one of cancer types
that share Food and Drug Administration (FDA) approved drugs, and another of cancer types
that share clinical trials of FDA approved drugs. Breast cancer is the only cancer type with
significant weighted degree values in both cancer networks. Lung cancer is significantly
connected in the FDA approval based cancer network, whereas ovarian cancer and lymphoma
are significantly connected in the clinical trial based cancer network. We defined global and
local lethality values representing death rates relative to other cancers vs. within a cancer.
Correlation and linear regression analyses suggests that global lethality impacts the drug
approval and trial numbers, whereas, local lethality impacts the amount of drug sharing in trials
and approvals. However, this effect may not apply to pancreatic, liver, and esophagus cancers as
the sharing of drugs for these cancers is very low. We also showed a weak overlap between the
mutation and drug target based cancer networks. Secondly, we analyzed the cancer pathways in
the KEGG (Kyoto Encyclopedia of Genes and Genomes) database, which provides a collective

of signaling pathway members involved in cancer progression. However, the KEGG cancer
pathways, unlike signaling pathways, were analyzed extensively with gene expression and
mutation data. We transformed the colorectal cancer pathway into subgroups based on their
position and analyzed the relative expression levels of adenoma and carcinoma samples as well
as the distribution of mutation targets. The gene expression values of the early stage pathway
members are significantly higher than the rest of the pathway members in colorectal adenoma
tissues. The colorectal cancer pathway shows some degree of coherence in only the carcinoma
samples. The correlated gene pairs responsible for the coherence of the colorectal cancer
pathway in the carcinoma samples are supported, in part, by the literature and may suggest novel
regulatory associations. Thirdly, we compared colorectal cancer samples not only to a control
sample set but against a wide variety of samples and conditions, in contrast to current integrative
network approaches that identify specific genes by comparing pair-wise control (i.e. normal) to
treated (i.e. disease) samples. We were able to identify a distinctly expressed set of genes which
were significantly associated with colorectal cancer in the literature unlike the pair-wise
approach. We integrated these specific genes with the PPI data to construct a colorectal cancerspecific network. We identified a potential regulatory relationship between glucocorticoid
receptor (GR) and ring finger protein 43 (RNF43) which may play a role in colorectal cancer. In
HCT116 colorectal cancer cell line, knocking-down GR levels with siRNA resulted in increased
RNF43 levels and inducing the colorectal cancer cells with dexamethasone, which is an
activating ligand for GR, resulted in decreased RNF43 levels. On the other hand, knocking-down
RNF43 levels with siRNA resulted in decreased GR levels. Our study suggests GR might
regulate RNF43 negatively, whereas there might not be such a negative regulation from RNF43
to GR.

ACKNOWLEDGMENTS

I would like to thank my family; my father Mesut Dalkic, my mother Emine Dalkic, and
my brother Mete Dalkic for their support for not only my studies but for my life.
I would like to thank Prof. Christina Chan for supporting me during my PhD. I would like
to thank her for being very open to any scientific discussion, organizing research meetings and
courses for systems biology.
I would like to thank my committee members; Prof. David Arnosti, Prof. Phillip
Duxbury, Assoc. Prof. Shin-Han Shiu, and Assoc. Prof. Michael Feig.
I would like to thank to the current and previous members of Cellular and Molecular
Biology Lab at the Department of Chemical Engineering and Materials Science; Li, Hyun-Ju,
Linxia, Amanda, Xueruei, Betul, Jason, Phil, Danielle, and all others.
I would like to thank to the current and previous Computational Biology/Systems
Biology people of our group; Zheng, Xuewei, Ming, Aritro, Hussein, Daniel, Mohammad
Kasim.
I would like to thank to the members of Cell and Molecular Biology Program at MSU.
I would like to thank Prof. Christina Chan for nominating me for Sigma Xi Graduate
Student Award and Prof. David Arnosti and others in Sigma Xi for the Graduate Student Award.

iv

TABLE OF CONTENTS

LIST OF TABLES

vii

LIST OF FIGURES

ix

INTRODUCTION

1

BIBLIOGRAPHY

13

CHAPTER 1: CANCER-DRUG ASSOCIATIONS
Introduction
Results and Discussion
FDA cancer drug approvals and clinical cancer drug trials
Global and local lethality values for cancer types
Effect of lethality on FDA approvals and clinical trials
Weighted cancer networks
Effect of lethality on the cancer networks
Specific and originally approved drugs
Clinical and molecular target based cancer networks
Conclusion
Materials and Methods
Drug-cancer pairs
Cancer death and survival statistics
Network construction
Statistical analysis
Appendices
Appendix A
Appendix B
Bibliography

15
15
18
18
19
19
21
25
26
27
30
32
32
33
34
35
38
38
55
76

CHAPTER 2: INTEGRATIVE ANALYSIS OF CANCER PATHWAYS
Introduction
Materials and Methods
Pathway data
X/Y scale
Expression data
Drug and mutation targets
Statistical analysis
Results and Discussion
Colorectal cancer pathway subgroups
Coherence of the colorectal cancer pathway
Mutation target analysis
Conclusion

78
78
82
82
83
84
85
86
88
88
91
94
96

v

Appendix
Bibliography

98
117

CHAPTER 3: SPECIFIC INTERACTION NETWORK ANALYSIS FOR
COLORECTAL CANCER
Introduction
Results and Discussion
Colorectal cancer specific genes
Colorectal cancer specific network
GR-RNF43 regulation in colorectal cancer
Conclusion
Materials and Methods
Transcriptome and Interactome data
Determining the colorectal cancer specific gene list
Cell culture
Quantitative real-time polymerase chain reaction
Western blot
Dexamethasone and siRNA treatment
Appendix
Bibliography

120
120
123
123
124
125
126
127
127
127
128
129
129
130
132
142

CONCLUSION

145

BIBLIOGRAPHY

157

vi

LIST OF TABLES

Table 1

Clinical trial and FDA approvals of drugs for different cancer types

38

Table 2

FDA approvals, clinical trial, weighted degree values and death
statistics of cancers in this study

44

Correlation values of weighted degree, approval number values and
of FDA specific drug percentage with global and local lethality values

46

Table 4

Cancer pairs with a weight difference of at least 0.5 or lower than 0

47

Table 5

Weighted degree values of drug target and mutation target based networks

48

Table 6

Mutation target- and drug target-based weight values of cancer pairs which
have a positive difference between the drug and mutation target-based values

49

Table 7

Cancers with at least one common mutation and drug target

50

Table 8

Drug and cancer-type association with a year tag, based on FDA labels

55

Table 9

Targets of FDA approved cancer drugs

59

Table 3

Table 10 Mutation targets of different cancer types

62

Table 11 Global lethality ratio values for different cancers from 2001 to 2007

65

Table 12 Local lethality ratio values for different cancers from 2001 to 2007

66

Table 13 Correlation values of weighted degree, approval number values with global
lethality values for 2001-2007

67

Table 14 Correlation values of weighted degree, approval number values with local
lethality values for 2001-2007

68

Table 15 Weighted degree values for FDA cancer network from 2000 to 2007

69

Table 16 Weighted degree p-values for FDA cancer network from 2000 to 2007

70

Table 17 Clinical trial numbers along with distinct drug number and specific drug
number for clinical trials

71

Table 18 ANOVA p values

98

Table 19 Pairwise t-test p values

98
vii

Table 20 Apoptosis and colorectal cancer pathway coherence at correlation
coefficient of 0.5

98

Table 21 X/Y group coherence at correlation coefficient of 0.5

99

Table 22 Correlated pairs in the colorectal cancer pathway for normal, adenoma
and carcinoma samples

100

Table 23 Mutation frequency values of X/Y groups

102

viii

LIST OF FIGURES

Figure 1 Cancer drug approval and clinical trial percentages. FDA cancer
drug approval and clinical drug trial percentages for 23 cancers

51

Figure 2 Weighted degree values breast and lung cancers in the previous years

52

Figure 3 FDA cancer network weighted degree vs. local lethality ratio

53

Figure 4 FDA specific drug percentages

54

Figure 5 Time dependent characteristics of the FDA approvals and FDA cancer
network

72

Figure 6 Cluster dendogram of cancer types based on global and local lethality
values

74

Figure 7 The ratio of carcinoma expression ratio to the adenoma expression
ratio given for the maximum and minimum two genes and the average

103

Figure 8 X/Y-dependent analysis of colorectal cancer pathway gene expression
levels

104

Figure 9 Pathway correlation and cumulative fraction distributions

110

Figure 10 Vend diagrams of correlated pairs of genes in colorectal normal,
adenoma and carcinoma samples

114

Figure 11 Literature comparison of differentially expressed genes in colorectal
cancer samples

132

Figure 12 Construction of the colorectal cancer specific network

134

Figure 13 Modulation of mRNA levels of GR and RNF43

136

Figure 14 Modulation of protein levels of GR and RNF43

138

ix

INTRODUCTION
Cancer is a large class of different diseases which result from the deregulation of cell
growth [1]. There are more than 200 cancer types, such as lung cancer, breast cancer, colorectal
cancer, etc. [2]. Cancer is a major cause of death in the United States. The United States has the
highest number of cancer deaths in the world [2]. Approximately 12 million Americans alive in
2008 had a history of cancer, some of whom are currently cancer free. Around one and a half
million Americans are expected to be diagnosed with cancer and approximately half a million
Americans are expected to die of cancer in 2012 [1]. Cancer is the second leading cause of death
in the United States. These statistics support the significance of cancer for clinical research and
shows the need for developing efficient therapy for cancer. Colorectal cancer is one of the
leading types of cancers based on the number of new cases and number of expected deaths [1].
However, it does not have as many clinical drug trials and FDA drug approvals as lung cancer,
leukemia, etc. This suggests that further clinical studies for colorectal cancer are needed, which
should also be accompanied by basic molecular research.
Clinical treatment of cancer includes chemotherapy, radiotherapy, surgery,
transplantation, etc. [3]. Chemotherapy is the use of drugs to slow down or prevent the growth of
the cancer cells. The goal of chemotherapy might be to cure the cancer by destroying the tumor
cells, control the growth of cancer by preventing the spread of the tumor cells, and slow the
growth of the cancer, or ease the symptoms (i.e., pain) associated with the growth of the cancer.
Classical cancer drugs include i) mitotic poisons or tubulin inhibitors such as Paclitaxel, which
disrupt the formation of mitotic spindle and thereby progression of the cell cycle, ii) antimetabolites like Cytarabine, which are nucleic acid base analogues that are incorporated into the
DNA to stop DNA synthesis, iii) topoisomerase inhibitors like Epipodophyliotoxins, which are
1

inhibitors of topoisomerase I or topoisomerase II, that prevent DNA replication, and iv) cell
cycle phase-independent agents such as Cisplatin, Doxorubicin, etc., which alkylate the guanine
bases to crosslink two strands of DNA, or intercalate between the neighbor bases, or locate in the
external grooves of the DNA to disrupt the double helix topology [2]. There may be side-effects
of chemotherapy such as nausea, diarrhea, hair loss, fatigue, secondary cancers, etc., because the
non-cancer cells might also be damaged [4]. Managing the dosage and the delivery of the cancer
drugs is also important aspect of the chemotherapy. There is only a small range between the
toxicity and efficacy of most cytotoxic cancer drugs, which necessitates caution in determining
the dosage of the chemotherapy [4].
Radiation therapy is the use of high energy radiation, such as X-rays, gamma rays, etc., to
damage the DNA and thereby destroy the tumor cells [3]. There also might be side effects to
radiotherapy as the normal cells are damaged with this therapy. Furthermore, different types of
cancer therapies might be combined; for instance, surgery can be followed by radiotherapy and
chemotherapy [3]. A challenge of cancer therapy is to avoid recurrence of the tumor growth [2].
For this reason, tumor markers are monitored to check the progression of the disease.
As mentioned previously, cancer drugs generally target the DNA structure, function, and
synthesis to halt or stop cell division. Such drugs are not likely to selectively kill a specific tumor
type or even differentiate between normal and tumor cells. Especially in recent years, there has
been progress in targeted cancer therapies, which are drugs that target molecular markers of a
specific cancer type [3]. For example, the SRC family protein tyrosine kinases are downstream
factors of several cell membrane receptors and relay the signal to proliferation-promoting genes.
Dasatinib is a direct inhibitor of the SRC family tyrosine kinases and is approved for the
treatment of leukemia. Targeted therapies are more promising, as they are more specific to the
2

tumor cells, and thus are less likely to harm other (i.e. normal) cells. The success of the cancer
treatment is dependent on the cause of the cancer, external causes such as smoking or internal
causes such as mutation, and the type of the cancer. For example, cancers caused by smoking can
be prevented [3]. Also, if a certain mutation is found in a patient, if there is a targeted
chemotherapy for that molecule, it would be more likely to be successful. For example, the
cancer drug Vemurafenib inhibits the activated mutated form of the BRAF serine/threonine
kinase protein in melanoma patients [3]. In recent years, combination chemotherapy for cancer
has gained significant attention [2]. When a single protein is targeted by a drug, the cell can use a
redundant pathway to continue proliferation. In combination chemotherapy, different drugs can
target different molecules, avoiding the redundant pathway activity; as well as reducing toxicity
due to the lower doses of the drugs. Another targeted therapy for cancer is vaccine therapy, in
which the immune system of the patient is exploited to recognize and destroy the tumor cells [3].
Cancer vaccines contain inactivated cancer cells, tumor antigen-expressing viruses, or
overexpressed tumor antigens along with supporting factors for the immune system such as
interleukin-2. Cancer vaccines enhance the immune activity of the patient towards the cancer by
helping it to recognize the tumor cells.
Since most of the current cancer drugs target general mechanisms such as DNA
synthesis, which do not provide specificity and also result in drug resistance, clinical trials for
new targeted cancer drugs are of high priority [2]. For this purpose, both basic and translational
research are important. While basic cancer research aims to find novel molecular mechanisms
involved in the cancer, translational research aims to improve cancer therapy by capitalizing on
the basic research findings [2]. Basic research investigates the modified genes in cell lines
through functional studies, such as analyzing the upregulated genes with apoptosis assays. On
3

the other hand, translational research investigates the potential of clinical usage of this target
gene, i.e., screening drugs for inhibitors of the upregulated protein product. This is followed by
in vivo studies in transgenic mouse models and then clinical trials [2].
Clinical trials for novel drugs provide evidence for the efficacy, risks and optimal use of
the drugs in patients [5]. Phase 1 clinical trials are performed on small groups of individuals to
evaluate the pharmacodynamics, pharmacokinetics, and dose toxicity of the drug [2]. Phase 2
clinical trials are performed to assess the safety and efficacy of the drug. Phase 3 clinical trials
are performed on a large group with similar purposes as well as to provide comparisons with
previously approved drugs. The US Food and Drug Administration (FDA) regulates the approval
and labeling of the drugs by considering the safety, efficacy, and security of the cancer drugs for
patient use [6]. Since basic molecular studies are fundamental to the cancer drug development, it
is necessary to monitor the relationship between the basic cancer research results with the
clinical results. If a common molecular mechanism is found for two different cancer types, this
raises the question of whether they also share clinical drug trials or FDA drug approvals. This
provides an idea of whether the molecular research findings are reflected in the cancer therapy
that is developed.
In recent years the biomedical and clinical sciences have embraced the era of systems
level research which is quantitative, integrative, and predictive. Systems science is an
interdisciplinary research that focuses on the principles of abstract organization rather than
specific properties and produces models to describe the global characteristics [7]. Analyzing part
of a system in isolation from the rest cannot provide a complete understanding of its properties
and thus cannot readily yield predictions of its behavior. This is significant for biomedical
research since the interactions between the genes, proteins, cells, tissues, and organs are crucial

4

for the proper function of the organism. In other words, biomedical systems cannot be
completely understood by analyzing the properties of the molecules in isolation, but rather is
better understood by analyzing the relationships between the molecules [7]. For example, the
identification of feedback mechanisms is a systems level observation of a biomedical system.
Systems medicine investigates the regulatory interactions in the clinical systems such as
feedback and feed-forward loops between the patients, clinicians, literature knowledge and data
from both basic and translational research [7]. The relationships between different disease types,
drugs, and clinical trials are also important just like the relationships between different
molecules. Analyzing the links between the diseases provides a system level observation of the
relationships between the different diseases [8]. These studies are based on the assumption that
two diseases that share a common molecular mechanism, like a mutation target, should have a
common origin. This is supported by the finding that similar disease phenotypes, like different
cancer types, cluster together when they are linked by mutation targets [9]. These network
analyses of disease networks provided new insight into comorbidities, which are secondary
diseases that occur concurrently with the primary disease [8]. A patient with a particular disease
was shown to be more likely to develop a secondary disease that is a neighbor in the disease
network [8]. Such analysis of disease networks provides a systems level approach to
understanding disease relationships as compared with the traditional classification of human
diseases that neglects this higher level of connectivity of diseases. For example, there might be a
mutation in a gene which can cause a primary disease and the interacting neighbor of that
specifically mutated gene can lead to secondary disease phenotypes, thereby linking these
diseases to each other [8].

5

Previously, a network level analysis was performed for all the approved drugs and
disease therapies [10]. The global topological analysis of the drug and disease associations
provided some emergent properties of the FDA approved drug therapies, i.e., most diseases are
connected to each other at most through 3 connections; revealing the high compactness of the
network as well as the presence of drug hubs that connect many disease therapies to each other
[10]. These results can be used in future clinical trials, i.e., the hub drugs may be given priority.
My hypothesis is that systems level analysis can improve our understanding and treatment of
cancer, which should have many subtypes that are interconnected with each other in a complex
way. Therefore I propose an isolated in-depth systems analysis of cancer. The network of cancer
types is a smaller-scale network than the disease and drug network [8, 9, 10] but it provides a
more focused analysis of the individual members and their interactions in the network. The
presence of a common drug therapy between two different cancer types hints to a relationship
between these drugs, therefore I analyzed the collection of these relationships at the network
level in Chapter 1. The list of cancer drugs in this study included both the drugs that target
general cell division mechanisms and those that pursue specific targets, as mentioned before.
Systems medicine investigates the relationships between different types of information at
the clinical or basic molecular level, such as the clinical decisions for different diseases vs. death
statistics. Only by performing such systematic analyses is it possible to have a more
comprehensive clinical therapy that is regulated by clinical and research results. Therefore, I
analyzed the potential influences of death statistics on the FDA approved cancer drugs in
Chapter 1 and showed the presence of only a few cancers with a significant connection in the
FDA approval based cancer network such as breast and lung cancers. I also showed a significant
correlation of the death statistics with the drug sharing of different cancers. Furthermore, for a

6

complete understanding of diseases such as cancer, it is vital to perform such analysis together
with molecular biology studies. For example, different cancer types were shown to be closely
related to each other as they share some common mutation targets like TP53 [9]. This raises the
need for comparing the molecular target information for the various cancers with the drug target
information in order to assess for any overlaps or differences. I addressed this issue in Chapter 1
and showed through network analysis that there is a low overlap between the cancer networks
based on drug target based associations versus mutation target based associations, suggesting a
low impact of molecular biological research findings on clinical decisions for cancer.
Even within the same cancer type, there are several sub-classes of the cancer. Thus it is
also important to describe the differences and relationships between these sub-classes. One of the
differentiating factors is the severity level of the cancer. For example, colorectal cancer can be
classified into colorectal adenoma and colorectal carcinoma, where the latter is differentiated by
the invasive properties of the tumor. The KEGG (Kyoto Encyclopedia of Genes and Genomes)
database provides cancer pathways which gives an opportunity to analyze the signaling events in
different sub-classes of a cancer [11]. The KEGG cancer pathways are different from other
pathways such as signaling pathways, metabolic pathways, etc., as they are different signaling
pathway members that are integrated into a single pathway. These cancer pathways contain
information about the molecular mechanisms on the different stages of a specific cancer like
colorectal cancer, pancreatic cancer, glioma, etc., and the different signaling pathways involved
in the different stages of the specific cancer.
The molecular events which play a major role in the different stages of colorectal cancer
are described in databases such as KEGG; which is based on current literature information. [12,
13, 14, 15]. For example, alteration of the Wnt pathway leads to activation of β-catenin, which in

7

turn activates its target genes and thereby leads to the progression of colorectal cancer [12, 13].
Because these events are described as earlier molecular mechanisms of colorectal cancer, they
take place in the upper portions of the KEGG colorectal cancer pathway. On the other hand,
TGF-beta pathway is another important pathway implicated in colorectal cancer. The members
of the TGF-beta pathway are located in the lower portions of the pathway image since they play
a role in the more advanced stages of colorectal cancer [14, 15]. The actual value of the KEGG
cancer pathways is the combination of these different signaling pathways such that a cancer is
described as an interconnection of different signaling modules. Module level analysis is an
important systems level methodology that aims to find significantly altered modules in a
phenotype of interest, shifting towards the analysis of crosstalk between modules recently [16].
Therefore, the analysis of the cancer pathways could provide a systems level identification of the
relationships between the different pathway modules in colorectal cancer.
Previously, pathways and protein complexes were studied for the collective behavior of
their members in terms of their gene expression levels. Although integrating signaling pathways
with genome level expression data was studied widely, it had yet to be realized with cancer
pathways. For example, the members of the protein complexes or pathways, such as ribosome,
proteasome, and certain metabolic and signaling pathways, have been found to have a strong
correlation to each of the members within the pathway with respect to their gene expression
levels or cis-element profiles, which is an indicator of similar transcriptional regulation [17, 18].
In another study with a similar goal, the coherence of a pathway was found to have a significant
fraction of its members correlated with each other based on their gene expression levels, as
compared with a random set of genes of the same size within the pathway of interest [19]. In
order to evaluate whether a cancer pathway, which itself is composed of several other pathways,

8

could be coherent, I performed a similar analysis with the KEGG colorectal cancer pathway in
Chapter 2 and found that the colorectal cancer pathway as a whole is coherent.
Since these pathway analyses depend on the current literature information, it is necessary
to improve the current information on the molecular events involved in cancer. Network level
analysis of protein-protein interactions (PPI) is an important tool for generating specific
networks and adding novel regulatory information to our understanding of cancer [20, 21, 22,
23]. Network biology and network medicine are closely related to each other [8, 24]. As network
biology investigates the interdependencies between the genes, proteins, metabolites, etc.,
network medicine relates these dependencies with human diseases. Therefore, the relationships
between a collection of genes is as important as the characteristics of a single gene for
understanding the function of biological systems and diseases. PPIs are one of the common ways
genes and their products regulate or interact with each other, in addition to other mechanisms
such as transcriptional interactions. PPI datasets are large-scale collections of interactions
between proteins and their topological analysis constituted some of the earliest network biology
studies [24]. For example, yeast PPI network was found to have a specific degree distribution,
namely scale-free, where the degree of a protein is defined as the number of connections to other
proteins in the network. In a scale-free topology very few proteins are hubs, i.e. highly
connected, whereas most proteins are lowly connected [24]. Nevertheless, these previous studies
are limited to the global statistical observations of the characteristics of the network structure.
These global analyses did not provide enough information on the role of the interactions or the
network modules in the biological process, phenotype or disease.
PPI datasets are a collection of protein interactions obtained from various sources and
conditions and are not condition-specific [25]. However, certain PPIs, excluding permanent

9

protein complexes, can be condition-specific [26]. Since such condition-specific interaction
information is not available on a large-scale to perform network level research, therefore an
alternative approach is required to constructed specific interaction networks from the large-scale
datasets. Only then can these specific networks be associated with a phenotype such as cancer.
Condition-specific datasets such as microarray datasets, which contain global mRNA expression
levels of a particular condition, was integrated with the PPIs, genetic interactions, etc. in order to
construct condition-specific interaction networks [20, 21]. In these studies, the PPI datasets
provide the pairs of proteins which were shown to interact physically at some condition and the
differentially expressed genes or differentially correlated gene pairs are calculated from
condition-specific microarray datasets. These condition-specific pairs of genes are checked to
determine if they interact. If so, they are considered to take place in the specific network. Such
networks constructed for cancer metastasis was shown to be represent the hallmarks of cancer
like apoptosis, cell growth, metabolism, etc. [20]. These specific networks can be used to
generate hypotheses about the involvement of genes and their interactions in the different
conditions, such as the involvement of a regulatory interaction between two genes in a particular
cancer [21].
A shortcoming of these previous integrative studies that use PPI is that they were based
on differential gene expression using a pairwise comparison approach, such as tumor versus
normal samples; metastatic versus non-metastatic samples [20, 21]. However, it was recently
shown that the pairwise comparison approach is insufficient for detecting the most relevant
genes of a condition of interest [27]. This limitation was addressed with a multiple comparison
approach, in which a sample set from a condition of interest was compared not only to a single
reference set such as normal samples but rather to a collection of samples from many different

10

conditions; including different normal samples, different cancer types, different human diseases,
etc. The diverse condition space more readily uncovers distinctive expression patterns for a
particular condition. For example p53, a well-known mutated target in cancer was not
differentially expressed based on a pairwise analysis of the gene expression data but through this
multiple comparison approach p53 was identified to be an important gene in irradiation [27].
Based on this, I used the multiple comparison approach to identify the top ranked, significantly
and differentially expressed genes in colorectal cancer in Chapter 3 and constructed a network of
these genes by integrating the gene expression with the human PPI data. I showed that the
colorectal cancer specific network included more genes that are significantly related to
‘Colorectal cancer’ in the literature, than the genes selected based on the pairwise differential
expression approach. I also showed that there might be a regulatory mechanism between ring
finger protein 43 (RNF43) and glucocorticoid receptor (GR, NR3C1) in this network which is
confirmed experimentally.

11

BIBLIOGRAPHY

12

BIBLIOGRAPHY

1.

American Cancer Society. Cancer Facts & Figures. Atlanta: American Cancer Society.
2012

2.

Rachel, A. Cancer Chemotherapy: Basic Science to the Clinic. Wiley-Blackwell,
Hoboken, NJ, USA, 2009

3.

http://www.cancer.gov/cancertopics/treatment/types-of-treatment

4.

Priestman, T. Cancer Chemotherapy in Clinical Practice. Springer, New York, USA,
2012

5.

http://www.clinicaltrials.gov

6.

http://www.fda.gov

7.

Deisboeck TS, Kresh JY. Complex Systems Science in Biomedicine. Springer, New
York, USA, 2006

8.

Barabasi AL, Gulbahce N, Loscalzo J. Network medicine: a network-based approach to
human disease. Nat Rev Genet. 2011, 12(1):56-68

9.

Goh KI, Cusick ME, Valle D, Childs B, Vidal M, Barabasi AL. The human disease
network. Proc Natl Acad Sci U S A. 2007, 104(21):8685-90

10.

Nacher JC, Schwartz JM. A global view of drug-therapy interactions. BMC Pharmacol.
2008, 8:5

11.

Kanehisa M, Goto S. KEGG: Kyoto Encyclopedia of Genes and Genomes. Nucleic Acids
Res. 2000, 28, 27-30

12.

Behrens J. The role of the Wnt signalling pathway in colorectal tumorigenesis. Biochem
Soc Trans. 2005, 33, 672-5

13.

Burgess AW, Faux MC, Layton MJ, Ramsay RG. Wnt signaling and colon
tumorigenesis--a view from the periphery. Exp Cell Res. 2011, 317(19):2748-58

14.

Roman C, Saha D, Beauchamp R. TGF-beta and colorectal carcinogenesis. Microsc Res
Tech. 2001, 52, 450-7

15.

Lampropoulos P, Zizi-Sermpetzoglou A, Rizos S, Kostakis A, Nikiteas N, Papavassiliou
AG. TGF-beta signalling in colon carcinogenesis. Cancer Lett. 2012, 314(1):1-7
13

16.

Wang X, Dalkic E, Wu M, Chan C. Gene module level analysis: identification to
networks and dynamics. Curr Opin Biotechnol. 2008, 19(5):482-91

17.

Jansen R, Greenbaum D, Gerstein M, Relating whole-genome expression data with
protein-protein interactions. Genome Res. 2002, 12, 37-46.

18.

Hannenhalli S, Levy S. Transcriptional regulation of protein complexes and biological
pathways. Mamm Genome. 2003, 14, 611-9

19.

Yang HH, Hu Y, Buetow KH, Lee MP. A computational approach to measuring
coherence of gene expression in pathways. Genomics 2004, 84, 211-7

20.

Chuang HY, Lee E, Liu YT, Lee D, Ideker T. Network-based classification of breast
cancer metastasis. Mol Syst Biol. 2007, 3:140

21.

Ahn J, Yoon Y, Park C, Shin E, Park S. Integrative gene network construction for
predicting a set of complementary prostate cancer genes. Bioinformatics. 2011,
27(13):1846-53

22.

Ideker T, Ozier O, Schwikowski B, Siegel AF. Discovering regulatory and signalling
circuits in molecular interaction networks. Bioinformatics. 2002, 18 Suppl 1:S233-40

23.

de Lichtenberg U, Jensen LJ, Brunak S, Bork P. Dynamic complex formation during the
yeast cell cycle. Dynamic complex formation during the yeast cell cycle. Science. 2005,
307(5710):724-7

24.

Barabasi AL, Oltvai ZN. Network biology: understanding the cell's functional
organization. Nat Rev Genet. 2004, 5(2):101-13

25.

Brown KR, Jurisica I. Online predicted human interaction database. Bioinformatics.
2005, 21(9):2076-82

26.

Levy ED, Pereira-Leal JB. Evolution and dynamics of protein interactions and networks.
Curr Opin Struct Biol. 2008, 18(3):349-57

27.

Wu M, Liu L, Chan C. Identification of novel targets for breast cancer by exploring gene
switches on a genome scale. BMC Genomics. 2011, 12:547

14

CHAPTER 1
CANCER-DRUG ASSOCIATIONS
INTRODUCTION
Cancer is a complex disease, with many subtypes, affecting various tissues in diverse
ways, thus giving rise to an abundance of chemotherapies. Taken together, cancers are the
second leading cause of death in the United States [1]. The common features of cancer include
uncontrolled cell growth, reduction in apoptosis, and loss of cell cycle regulation, while other
features are more tissue specific and thus differentiate them and their chemotherapies.
In a global network level analysis of different diseases, where the vertices represented
diseases and the edges represented connections between diseases that share common genetic
background, most diseases were less connected, while a limited number of diseases, mostly
cancers, were highly connected hubs [2]. Similarly, a network analysis of drugs, where the
vertices represented drugs and the edges represented connections between drugs that share
common protein targets, showed that drugs of similar types clustered together, and most proteins
were targeted by a few drugs, whereas only a few proteins were targeted by many drugs [3, 4].
Cancers have fewer drugs that are used to treat them as compared with the other diseases, and the
targets for the cancer drugs are at a shorter distance from the genes that are mutated in the
cancers [3]. Quantitative analysis of the drug targets showed that proteins with at least 3 proteinprotein interactions are more likely to be targeted by drugs [5]. A recent network study
characterized the global map of many diseases, including cancers, and their associations with
drugs, where the vertices represented diseases and the edges represented connections between
diseases that share common drugs [6]. This study was also concerned with the global description
of the network, and found that only a few diseases are highly connected by drugs, while most

15

diseases are less connected; and most diseases, even those unrelated to each other, are connected
by a few links [6]. These studies constitute the global topological analysis aspect of the emerging
areas of network medicine [7] and network pharmacology [8]. However, these studies do not
focus on the specific relationships between diseases and drugs, to address questions, such as,
how might these relationships arise, or what factors may affect these relationships.
The field of medical sciences includes both basic molecular and clinical research, the
latter involves clinical trials. Clinical trials apply biomedical protocols to humans that aim to
intervene or observe a disease, e.g., testing drugs on cancers (http://clinicaltrials.gov). Clinical
trials provide preliminary evidence of the efficacy, risks and optimum usage of the drugs. Phase
1 and 2 clinical trials are performed on small groups of individuals to evaluate their safety and
efficiency. Phase 3 clinical trials are performed on a large group of individuals, to evaluate their
efficiency, side effects and how they compare with approved drugs. Phase 4 clinical trials are
performed after the drug has been approved for use, to obtain additional information. The United
States Food and Drug Administration (FDA) regulates the approval and labeling of the drugs
with regard to their safety, efficacy, and security to humans (http://www.fda.gov). In addition to
the clinical drug trial and FDA approval data, death statistics, such as the estimated cases and
estimated deaths over the years are available for the different cancer types [9]. Cancer is a large
class of disease with various types, each with its own specific approvals, trials, death statistics,
and molecular information, i.e., mutation targets. These diverse data provide opportunities to
perform an integrative, systems level analysis of the cancers to reveal potential relationships
between the various types of cancer and the drugs used to treat them and possible trends or
factors that influence these relationships.

16

Global network analyses have been previously applied to describe the overall topology of
disease and drug relationships, i. e, very few diseases and drugs are highly connected, while most
members of these networks are less connected [2, 3, 4, 6].Smaller network systems, such as in
this study, are amendable to a more focused analysis of individual members of the network,
whereas larger networks are not, and hence are more amendable to statistical topological
analyses, such as degree distribution analysis [10]. We propose that a drug approved or used in
clinical trials for treating several cancers may hint to a relationship between those cancers.
Similarly, a mutation involved in or a drug target used in treating different cancers may suggest a
relationship between these cancers. System level analysis of these relationships could reveal
potential factors involved in the development of these complex relationships that are not readily
apparent from the data itself. In contrast to the previous medical network analyses, the analysis
of smaller networks of cancer-drug and cancer-target associations permits a more detailed
evaluation of the specific relationships between individual cancers. Through correlation and
linear regression analyses of the number of approvals and trials, and weighted degree values,
with the cancer lethality values, we assessed whether the death statistics impact the formation of
associations between the cancers and drugs. Our analyses suggest that global lethality has an
effect on the number of FDA approved and clinical trial cancer drugs. Comparative analysis of
the cancer networks based on the FDA approved drugs and clinical trial drugs showed that some
cancers are significantly and highly connected in the clinical trial cancer network but not in the
FDA cancer network, and vice versa. Correlation and linear regression analyses suggest that
local and global lethality differentially impact the sharing of FDA approved cancer drugs and the
sharing of clinical trial drugs. Further, a comparison of the mutation target-based with the FDA

17

drug target-based cancer networks suggests that the molecular information about a cancer does
not strongly influence the cancer drug approvals.

RESULTS AND DISCUSSION
FDA cancer drug approvals and clinical cancer drug trials
We collected the drugs approved through 2009 by the FDA for 23 cancer types and the
clinical trials completed by 2009 for these same drugs (see Appendix B). We compared these 81
drugs for the 23 cancer types, and checked which drugs had i) completed Phase 1 and 2 trials but
were not listed under Phase 3 clinical trials and thus were not FDA approved, ii) completed
Phase 3 clinical trial but were not FDA approved, iii) were FDA approved and in Phase 3 clinical
trial (Table 1), and iv) were FDA approved and were not in clinical trials. There are several
drugs for which Phase 3 clinical trial was completed but were not FDA approved (item ii). For
example, cisplatin was approved for only testicular and bladder cancers, and has undergone and
completed Phase 3 clinical trials for many types of cancer but has yet to be listed as approved by
the FDA for those cancers (Table 1). The clinical trial data is incomplete (see Materials and
Methods section for details). For example, there are some drugs which were FDA approved but
not listed under any past clinical trials, completed or otherwise, which suggests that the analysis
of the clinical trials will not be comprehensive. Leukemia, breast cancer, lung cancer, and
lymphoma have the highest number of drug approvals and the highest number of clinical trials
(Table 2). The percentage of clinical trials or FDA approvals for the different cancers were
calculated as the number of clinical drug trials or FDA drug approvals for a specific cancer type,
divided by the total number of clinical drug trials or FDA drug approvals for the 23 cancers
analyzed in this study. The clinical trial and FDA approval percentages are similar for many of

18

the cancers in this study (Figure 1). There are a few notable exceptions, namely breast cancer and
myeloma, which have much higher percentages of FDA approvals than of clinical trials.
Global and local lethality values for cancer types
Death and survival ratios have been predominantly used to describe the values of global
and local significance of cancer deaths [9]. It is confusing to use these values since one uses
death and the other uses survival numbers to describe global and local death statistics of a
specific cancer. Therefore, we defined two different death-based statistics, a global and a local
lethality rate by using the estimated death and new case numbers of each cancer (Table 2, see
Appendix B). The percentage of global lethality is calculated as the ratio of estimated number of
deaths for a cancer to the estimated number of deaths for all cancers. The percentage of local
lethality is calculated as the ratio of estimated number of deaths to the estimated number of cases
for a particular cancer. The global lethality provides a perspective of a particular cancer with
respect to the other cancers, whereas, the local lethality is specific to each cancer type. A cancer
with a high local lethality suggests that it has a high number of deaths within its own incidences,
while its global lethality may or may not be high. For example, pancreatic cancer is a locally
lethal but not globally lethal cancer; it has a local lethality value of 0.91 but a global lethality
value of 0.06 (Table 2). This is because most of the pancreatic cancer patients have low survival
rates, but comparatively there are fewer cases of pancreatic cancer.
Effect of lethality on FDA approvals and clinical trials
We hypothesize that there are factors, such as the lethality values of a cancer, that may
influence the number of clinical trials and, in turn, FDA approvals. To quantitatively evaluate
whether lethality values are related to the number of FDA drug approvals and clinical drug trials,
Spearman correlation coefficients were calculated between the global/local lethality measures

19

and the trial/approval numbers. The correlation analyses suggest that global lethality is
correlated, whereas local lethality is not correlated, to both the clinical trial and FDA approval
numbers (Table 3). To further evaluate the impact of lethality values on the FDA drug approvals
and clinical drug trials, we performed a linear regression analysis. Linear fit of the clinical trial
numbers with global lethality suggests a slight but albeit significant relationship (r2 = 0.25, p =
0.03). This suggests the higher clinical drug trial numbers could be explained, in part, by the
higher global lethality rates. Next, we considered whether the relationships found by correlation
and linear regression analyses are affected by lung cancer, the most globally lethal cancer, and
pancreatic, esophagus, and liver cancers, the most locally lethal cancers (see Table 2 and the
Materials and Methods section). We re-calculated the correlations by removing the globally or
locally lethal cancers. No significant change in the correlations resulted upon removing lung
cancer. However, a linear fit of the FDA approval numbers with global lethality suggests a slight
relationship which is significant, when lung cancer is excluded (r2 = 0.20, p = 0.05). The
significance of the correlation and the linear fit between local lethality with FDA approval and
clinical trial numbers increased upon removing the most locally lethal cancers, pancreatic, liver
and esophagus cancers (Table 3). Local lethality has a significant correlation with clinical trial
drug numbers for the cancers other than the most locally lethal ones. This suggests the number of
FDA approvals and clinical trials are much lower for pancreatic, liver and esophagus cancers as
compared to other cancers despite their very high local lethality. Linear fits suggest that the
values of the lethal cancers affect the linear relationship, therefore we estimated the values based
on the equation with the other cancers. Based on the linear regression of the local lethality ratio
with the FDA cancer network weighted degree values (y = 0.96x + 0.40), pancreatic, liver and

20

esophagus cancers should have been expected to have a weighted degree of around 1.2; however,
they have a weighted degree of around 0.07-0.5.
Although, the linear fit p-values of local lethality with FDA approval numbers and
clinical trial numbers decreased, when pancreatic, liver and esophagus cancers are excluded, they
are not very significant. We also analyzed whether the FDA approval numbers from previous
years correlated with the lethality values. The correlation of global lethality with the FDA
approval numbers has mostly been present in previous years (see Appendix B). The correlation
and linear regression analyses suggest that global lethality has an impact on the drug trial and
approval numbers, for the cancers in this study.
Weighted cancer networks
The global relationships between drugs and diseases have been analyzed topologically in
large-scale networks of drugs and diseases [2, 3, 4, 6]. Complex relationships between the types
of cancer and drugs constitute a smaller network structure. Unlike the larger networks, a smaller
network system, as in this study, are amendable to a more focused analysis of individual
members of the network rather than statistical topology-based parameters [10]. We applied this
more focused analysis, where individual members and interactions in the networks were studied
rather than their global structure, to elucidate the drug therapy based relationships between
various cancers and the factors that may influence these relationships.
The collection of cancer-drug pairs make up a bipartite network, which we transformed
into a unipartite weighted network consisting of only cancers. To construct a weighted network
of cancers, an edge between any two cancers was assigned, if there is at least one drug which
was approved by FDA to treat both types of cancers. The weight of an edge was defined by the
Jaccard index, which is the fraction of drugs which were approved for both cancers over all the

21

drugs which were approved for each of the two cancers, separately (see Materials and Methods).
Weighted degree values were not significantly correlated with the number of FDA approvals
(Pearson correlation coefficient of 0.34, p = 0.11), suggesting that the number of drugs approved
for a cancer does not implicate the number of drugs shared with other cancers. We, further,
assessed the significance of the weighted degree values by a permutation test, while keeping the
number of drugs per cancer type constant, and found the degree of drug sharing is not significant
for most of the cancers (Table 2), except for lung and breast cancer These two cancers have
significant weighted degree values in the FDA cancer network. Lung cancer shares FDA drugs
with many other cancers. Leukemia, the cancer type with the highest number of FDA approvals,
does not have a significant weighted degree value in the FDA cancer network (Table 2). This is
because leukemia does not share many of its FDA approved drugs with other cancers. Indeed, as
discussed later, leukemia has many specific drugs (see section “Drugs specific to particular
cancer types”). We also analyzed the FDA cancer network over time, by including the cancer
drug approvals for the different years.
Using the date at which a drug was approved by the FDA, we analyzed the cancer
networks at earlier time points. Based on the first cancer drug, mechlorethamine, approved in
1949, leukemia, lymphoma and lung cancer can be connected in a FDA cancer network. These 3
cancers for a long time were the only cancers for which a drug was approved by the FDA. This
initial network began to grow as drugs for other cancers were approved by the FDA, joining the
network from 1986 onwards. We defined the average weight as the average of the weights of all
the edges of a network (0 weights are not included). The average weight of the FDA cancer
networks was calculated to capture the changes in how the drugs are shared between the different
cancers over the years (see Appendix B). The average weight of the FDA cancer network was at

22

the lowest in 1997, suggesting that the sharing of drugs between cancer types was not very high
in 1997 as compared to the other years. Sharing of drugs between cancer types increased
between 1998 and 2004. We also determined the component numbers of the FDA cancer
networks. The network consisted of a single component except between 1991-1998 when the
network contained 2 components (see Appendix B). There have been only two cases where a
drug led to a jumping growth as new drugs were approved for cancers. Pamidronate was
approved for breast cancer and myeloma in 1991, making a connection between the two cancers,
separate from the other cancers. Paclitaxel was approved for skin cancer in 1997, making a
connection to ovarian cancer, causing another jump in the growth of the network. Some FDA
approvals could be terminated later, however, this analysis is only concerned with the drugs
which were not terminated by 2009 (the time of completion of the clinical trials is not available
for the majority of the trials, therefore a time dependent study of the clinical trial-based cancer
networks was not performed). Based on the average weight values of the networks, there is no
major change over the years. Weighted degree values for most of the cancers also are not
significant in the previous years’ networks. However the breast cancer weighted degree value has
been significant since 2000 and the lung cancer degree value has become significant recently.
Weighted degree values of lung and breast cancer have been increasing and significantly higher
than the other cancers since 2006 (Figure 2). In recent years, FDA approved drugs for these
cancers (the 1st and 3rd most globally lethal) have a high overlap with other cancers.
A weighted clinical trial-based cancer network was also constructed (herein denoted as
clinical trial cancer network), where two cancers were connected if there is at least one FDA
approved drug (approved for at least a cancer) in the clinical trial data for both cancers. The
clinical trial cancer network is almost a complete network, because of the large number of drugs

23

that were used in clinical trials for the different cancers, thereby connecting many of the cancers,
albeit not all, to each other. The significance of the weighted degree values was evaluated by a
permutation test, with the number of drug trials kept constant. Breast cancer, ovarian cancer, and
lymphoma have significant weighted degree values in the clinical trial cancer network (Table 2).
Also, the weighted degree values of lung cancer and head and neck cancer are close to being
significant. This indicates that these cancers shares clinical trial drugs, significantly, with other
cancers. In addition, we calculated the difference in the edge weights between the FDA and
clinical trial cancer networks for each cancer pair, and identified that most pairs are strongly
connected in the clinical trial but not in the FDA cancer network. For example, stomach and
esophagus cancers are strongly connected in the clinical trial cancer network (Table 4). There are
many drugs used in clinical trials for both types of cancers, i.e., capecitabine, cisplatin,
doxorubicin, erlotinib, fluorouracil, irinotecan, ixabepilone, leucovorin, oxaliplatin, paclitaxel,
and vinorelbine, and thus strongly connecting these two cancers. However, they are not
connected in the FDA cancer network, i.e. no drug is approved by the FDA for both stomach and
esophagus cancers; porfimer was approved for esophagus cancer while docetaxel, fluorouracil,
imatinib, and sunitinib were approved for stomach cancer. There are a few pairs of cancers
which are more highly connected in the FDA cancer network than in the clinical trial cancer
network (Table 4). For example, sarcoma and endometrial cancer pair has a weight of 0.5; they
share methotrexate which is the only drug approved for endometrial cancer and one of the two
drugs approved for sarcoma. On the other hand, there are many drugs in the clinical trial data for
each of these two cancers which are not shared between them, such as altretamine, capecitabine,
etoposide, etc. Weighted networks of cancers based on FDA approvals and clinical trials show
different characteristics. Breast cancer is the only cancer with a significant degree value in both

24

the FDA and the clinical trial cancer networks. While lung cancer is more significantly
connected only in the FDA cancer network, ovarian cancer and lymphoma are more significantly
connected in the clinical trial cancer network (Table 2). This suggests that ovarian cancer and
lymphoma have a high overlap of drugs in clinical trials but not in FDA approvals.
Effect of lethality on the cancer networks
Given that the lethality of a cancer impacts the number of drug trials and approvals, it
raises the question of whether it could also influence the FDA and clinical trial cancer networks
and if there could be differences in their influence on these two networks. We analyzed the
correlation and the linear fit between the weighted degree values of the FDA/clinical trial cancer
networks and the global/local lethality values. The weighted degree values for the clinical trial
cancer network are correlated with local lethality (Table 3). Linear regression between the
weighted degree values and the lethality values shows a partial but significant relationship
between local lethality and clinical trial network weighted degree (r2 = 0.26, p = 0.02). This
suggests that sharing of drugs in clinical trials is impacted positively by local lethality values.
The weighted degree values of the FDA cancer network are not significantly correlated
with the global and local lethality values (Table 3). Next, we analyzed the effect of the most
globally lethal (lung cancer) and the most locally lethal cancers (pancreatic, esophagus and liver
cancers) on these correlations and linear fits. Weighted degree values of the FDA cancer network
are significantly correlated with local lethality after removing pancreatic, liver, and esophagus
cancers (Table 3, Figure 3A-3B). Linear fit analysis suggests that the weighted degree of a
cancer in the FDA cancer network tend to be high if its local lethality value is high. However, the
most locally lethal cancers (pancreatic, esophagus and liver cancers) are excluded from this
effect since they have lower than expected weighted degree values, as compared to the other

25

cancers. We also analyzed if the FDA cancer networks from previous years correlated with the
lethality values. Global lethality and local lethality do not have a significant correlation in the
older FDA cancer networks. However, more recently (2007) the cancer network has become
correlated with local lethality, with the exclusion of pancreatic, liver, and esophagus cancers (see
Appendix B).
Analysis of the weighted degree values of the cancer networks provides information on
the level of drug sharing between cancers. We showed that local lethality has an effect on the
clinical cancer drug trial sharing as well as FDA approved drug sharing, the latter appears to be a
recent trend. However, the most locally lethal cancers, pancreatic, liver, and esophagus cancers,
are biased towards having lower levels of sharing of FDA approved drug. For the most local
lethality cancers, although sharing of drugs in clinical trials correlates positively with local
lethality values, the sharing of the approved drugs does not correlate with local lethality values.
Specific and originally approved drugs to particular cancer types
Network analysis captured the overlap in cancer drug use, however, only 26 of the total
81 cancer drugs were approved for more than one cancer type. Therefore we analyzed the
distribution of the remaining 55 drugs which were approved specifically for only one type of
cancer. A drug which was approved by the FDA solely for a single cancer is denoted as a
“specific” FDA drug. We calculated the specific drug percentage for a cancer as the ratio of the
number of specific drugs to the total number of drugs approved by the FDA. Prostate cancer,
leukemia, breast cancer, and lymphoma have the highest specific drug percentage approved by
the FDA (Figure 4). The most locally lethal cancers, pancreatic, liver, and esophagus cancers,
have no specific drugs (Table 2). Globally lethal cancer, i.e., lung cancer, has a low percentage
of FDA specific drugs (Table 2, Figure 4). The number of specific drugs in clinical trials is very

26

low, therefore it was not analyzed further (see Appendix B). We also analyzed the possible effect
of lethality on the percentage of FDA specific drug approvals and showed that there is no
significant effect (Table 3).
There is also a notable difference among the non-specific (shared) drugs, such that some
of the drugs were first approved for a cancer type and then approved for other cancer types,
while other drugs might be approved for more than one type of cancer at the same time. We
defined whether a drug was “originally approved” by the FDA for a specific cancer type and then
approved for other cancers after at least a year. Colorectal cancer has the highest number of
“originally approved” FDA drugs (Table 2). There is only one originally approved FDA drug,
erlotinib for lung cancer (Table 2, see Appendix B). Many more drugs were approved for other
cancers that were subsequently approved for lung cancer (11 drugs) than were “originally
approved” for lung cancer (only one).
Comparison of clinical and molecular target based cancer networks
In addition to the death statistics, we asked whether molecular information impacted the
cancer-drug associations. To compare the molecular target-based relationships to the clinical
target-based relationships for the different cancer types, we constructed weighted molecular and
clinical cancer networks based on mutation targets and FDA approved drug targets respectively.
The edges between two cancers in the mutation target based network was assigned if there is at
least one mutation target associated with both cancers and the edges between two cancers in the
drug target based network was assigned if there is at least one drug target associated with both
cancers. The weights of the edges were defined by the Jaccard index. To compare the mutation
target-based and the drug target-based cancer networks, we included only the cancers that have
both mutation and drug target data. We calculated the weighted degree values for the different

27

cancers and evaluated the significance of the weighted degrees with permutation test, keeping the
distribution of target numbers for each cancer constant (Table 5). The weighted degree values of
the mutation and drug-target based cancer networks are not strongly correlated (Pearson
correlation coefficient of 0.37, p = 0.11). Lung and breast cancers have significant and high
weighted degree values in the drug target-based network but not in the mutation target-based
network (Table 5). On the other hand, colorectal, ovarian and brain cancers have significant
weighted degree values in the mutation target-based network but not in the drug-target based
network (Table 5). Leukemia is the only cancer which has significant weighted degree values in
both networks.
The overlap between the two networks is very low and for the overlapping edges, we
calculated the difference in mutation and drug target weight values (Table 6). Colorectal-ovarian,
ovarian-endometrial, and endometrial-colorectal cancer pairs have higher mutation target-based
weights than drug target-based weight values (Table 6). These cancers are connected to each
other in the mutation target-based network through the following mutations: PMS1, PMS2,
MLH1, MSH2, and MSH6, which are proteins responsible for DNA mismatch repair. On the
other hand, all three cancers share no drug targets. Since they share many mutation targets, this
suggests that they could have similar molecular mechanisms, and thus raises the question if they
should share drug targets. On the other hand, kidney and liver cancers, which do not share any
mutation targets, have a high overlap of drug targets. They share drug targets such as FLT4,
PDGFRB, BRAF, etc. There could be mutation targets common to these cancers which may not
have been identified or is absent in the current dataset. Alternatively, they could share molecular
mechanisms without sharing mutation targets, i.e., similar pathways may be affected in both
cancers despite different mutated genes.

28

We also evaluated the cancers that are associated with proteins that are both mutation and
drug targets (Table 7). Only leukemia, lung, and breast cancers have mutation targets that are
also drug targets. For example, ERBB2, a member of EGFR family, has long been known as a
mutation target for breast cancer [11]. Lapatinib, letrozole, and trastuzumab are drugs that target
ERBB2 in our data and all have been used in clinical drug trials for only breast cancer and
approved by the FDA for only breast cancer. Furthermore, ERBB1, a member of the EGFR
family, is known as a mutation target for lung cancer [12]. There are several drugs which target
ERBB1, such as cetuximab, erlotinib, gefitinib, lapatinib, panitumumab, and trastuzumab,
among which, only erlotinib and gefitinib are approved only for lung cancer (see Apendix B).
The remaining drugs have not completed Phase 3 clinical trials. Cetuximab, and trastuzumab
have completed Phase 1 and 2 trials, whereas clinical trials using lapatinib and panitumumab for
lung cancer have not yet completed Phase 1 and 2 trials (Table 1). Overall, very few mutations
have been approved as targets for cancer therapy.
Comparison of mutation and drug-target based cancer networks indicate that the overlap
is very low. Various cancers have strong associations in one but not in the other network. For
instance, lung and breast cancers have significant drug-target based associations but not
mutation-target based associations. Similarly there are pairs of cancers, such as the pair of
colorectal and endometrial cancers, with relatively high weights in the mutation-target based
network but not in the drug-target based network. This analysis suggests that the influence of
molecular information on the cancer-associations is not strong, and there are very few proteins
which are both mutation-targets and drug-targets.

29

CONCLUSION
In this study, we present a systems level view of the cancer drugs. Comparing clinical
trial and FDA approval based cancer networks, we showed that only breast cancer is
significantly connected in both networks. Lung cancer is significantly connected in the FDA
cancer network, whereas ovarian cancer and lymphoma are significantly connected in the clinical
trial cancer network. This suggests that lung cancer has a high degree of sharing of FDA
approved drugs with the other cancers. Indeed, it has the highest number of FDA approved drugs
which are shared with other cancers. In contrast, ovarian cancer and lymphoma have a high
degree of drug sharing in clinical trials but not in FDA approvals.
We also assessed whether death statistics and molecular information are related to the
cancer-drug associations. We showed that the cancer-drug associations are differentially
impacted by the type of lethality. Global lethality appears to have an effect on the number of
FDA approved drugs and clinical drug trials, but not on the FDA approval and clinical trialbased drug sharing, as determined by the cancer network weighted degree values. On the other
hand, local lethality has an effect on the FDA approval and clinical trial-based drug sharing, but
not on the number of FDA approved drugs and clinical drug trials. The effect of local lethality on
the sharing of FDA approved drugs is not present or captured by the most locally lethal cancers,
pancreatic, liver and esophagus cancers. These cancers are biased towards having very low
overlap of FDA approved drugs with other cancers. For example, there is only one drug
approved for liver cancer, sorafenib, which is shared with lung cancer; however there are 13
more FDA approved drugs for lung cancer, which are not approved for liver cancer, leading to
the lower weight for liver cancer. Although sharing of drugs in clinical trials correlates positively
with local lethality values, however, it does not translate to increase sharing of the approved

30

drugs for the most locally lethal cancers. There could be a number of reasons for this; the drugs
in clinical trial are not being approved for the most locally lethal cancers or they have not been
approved yet. For example, liver cancer and lung cancer share 13 drugs out of total 32 drugs
used in clinical trials for these cancers. 5 of these 15 common/overlapping clinical trial drugs are
approved for lung cancer by FDA but they are still in clinical trials for liver cancer. Therefore
they have a higher connection weight in the clinical trial cancer network than the FDA cancer
network. These findings support network-based analysis and their ability to reveal relevant
information distinct from the raw data. It is not surprising that clinical decisions may be
impacted by death statistics. However, it is interesting that different types of death statistics
(global lethality vs. local lethality) show different results. It should be kept in mind that this
study does not capture all aspects of the clinical drug data. For example, this analysis does not
account for the differential efficiencies of the various drugs used in treating a particular cancer,
which could have an impact on why some cancers have few while others may have many more
drugs that target it. The current analysis of the clinical trials is limited to those which have
already been approved by the FDA for at least one cancer type and therefore do not include all
cancer drugs currently in clinical trial.
Currently, most cancer drugs are designed to target the general mechanisms of cell
division, which may not directly address the specific molecular mechanisms that drive the
development of the type of cancer it aims to treat. We compared mutation and drug targets for
various cancer types. We identified a number of differences and noted that some cancer types
share mutation targets but not drug targets while others share drug targets but not mutation
targets, thereby hinting at the possibility that new drug targets or mutation targets could be
identified for these cancers. Nevertheless, there are many other factors to consider when

31

evaluating the data. Although two cancer types may not have the same mutation targets, they
may have the same genes that are differentially expressed, which could suggest the involvement
of similar molecular mechanisms. Given that cancer treatment includes surgery, radiotherapy in
addition to chemotherapy (http://www.cancer.gov/cancertopics/treatment/types-of-treatment)
thus, this study provides a systems level analysis of the trends of one aspect of clinical cancer
research, namely from the perspective of the drugs that are FDA approved or undergoing clinical
trials.
In closing, we demonstrated a systems level view of the drugs that have been approved
and how they have been shared between cancer types. Thus we envision that this study could be
informative to medical researchers from both the basic and clinical sciences alike. The trends
revealed in this study could be monitored in the following years for any changes and these
analyses could be used to guide more in-depth analysis of potential targets that could be involved
in future clinical cancer drug trials and approvals. For example, one could followed whether the
FDA approved drug sharing continues to be significant for breast and lung cancers which
appears to be recent trends, beginning in the 2000s, and whether the overlap between the
molecular target based and the drug target based cancer networks increases.

MATERIALS AND METHODS
Drug-cancer pairs
We obtained lists of cancer drugs from the National Cancer Institute Drug Information
Summaries (http://www.cancer.gov/cancertopics/druginfo/alphalist), and the FDA Center for
Drug Evaluation and Research (http://www.fda.gov/Drugs/default.htm). We used the indication
information by 2009 from the drug labels from the Drugs@FDA database

32

(http://www.accessdata.fda.gov/scripts/cder/drugsatfda/index.cfm?fuseaction=Search.Search_Dr
ug_Name) to generate a list of drug-cancer associations that included 23 types of cancer (see
Appendix B). We renamed some cancers, for example, Kaposi’s sarcoma is listed under skin
cancer, glioma is listed under brain cancer, and different types of leukemia and lymphoma are
listed more generally as leukemia and lymphoma, respectively. The time information tag of the
FDA approved label files is also used. Drugs discontinued in the market were excluded. We
obtained clinical trial information for all drug trials completed by 2009 from the Clinical Trials
database (http://clinicaltrials.gov) and collected the clinical trials for the drugs and the cancer
types that are in the list of FDA data. We differentiated between Phase ½ and Phase 3 trials since
the Phase ½ are initial trials on small groups of patients, whereas Phase 3 trials are performed on
large groups of patients. We excluded Phase 4 trials since they are post-approval. We did not
include neoplasms in our analysis. Names of drugs and cancers have been organized according to
the FDA data. In addition, we only collected the trials which were listed as drug trials. These
limitations could lead to loss of information, such that we have FDA approval information for
some drugs without completed clinical trial information (Table 1). We observed that these
limitations could affect the clinical trial information prior to 2000s, namely, there could be cases
in which there is an approval of a drug earlier than the trial dates. Therefore, we did not perform
a time analysis of the clinical trials.
Cancer death and survival statistics
The cancer statistics for 2001-2008 of the estimated number of new cases and the
estimated number of deaths for the different types of cancers were obtained from the American
Cancer Society [9]. We defined two kinds of lethality values. Global lethality is defined as the
ratio of deaths of a particular cancer over all cancers. Local lethality is defined as the ratio of

33

deaths of a particular cancer over the cases of that particular cancer. For breast, ovarian, cervical
cancers only the female population values were considered. Likewise, for prostate and testicular
cancers only the male population values were used. For the other cancer types, both the male and
female population values were included. Lung cancer has the highest global lethality value,
whereas pancreatic cancer has the highest local lethality value. To determine which other cancers
are similar to lung and pancreatic cancers with respect to their global and local lethality values,
we performed hierarchical clustering, based on Euclidean distance of lethality values with single
linkage. Lung cancer clustered by itself, and pancreatic, liver and esophagus cancers clustered
together (see Appendix B). Therefore only lung cancer is considered globally lethal cancer,
whereas pancreatic, liver and esophagus cancers are considered locally lethal cancers.
Network construction
We constructed weighted clinical networks of cancer types, FDA cancer network and
clinical trial cancer network, from the drug-cancer pairs. In the clinical cancer networks an edge
was defined between two cancer types when there is at least one drug which was approved or
used in clinical trials for both types of cancer. The weight of the edge was defined by the Jaccard
index, which is the fraction of common drugs for both cancer types over all the drugs for each of
the cancer types. For example, there is only one drug which was approved for both pancreatic
and stomach cancers, fluorouracil, whereas there are 2 more drugs, erlotinib and gemcitabine,
which were approved for pancreatic cancer but not for stomach cancer, and there are 3 more
drugs, docetaxel, imatinib, and sunitinib, which were approved for stomach cancer but not for
pancreatic cancer (see Appendix B). Therefore the weight of the edge between these two cancers
is 1/(1+2+3)=0.17 (Table S3). The resulting FDA drug approval-based cancer network (herein
denoted as FDA cancer network) contains 23 types of cancer (vertices or nodes) with 70

34

interactions (edges). We defined the weighted degree value for a cancer as the sum of the
weights of the edges for that cancer. For example, pancreatic cancer shares drugs with stomach,
lung, colorectal and breast cancers, therefore its weighted degree is the sum of the weights of the
edges with these cancers, which is 0.17+0.13+0.1+0.1=0.5 (Table 1). This parameter provides an
account of the allocation of drugs for a particular cancer and its neighbors in the network. If
more drugs, which are approved for other cancers, are approved for pancreatic cancer (regardless
of whether the drug is shared with stomach, lung, colorectal and breast cancers or other cancers)
in the future, its weighted degree value will increase. Its weighted degree value will decrease if
more drugs are approved for stomach, lung, colorectal and breast cancers but not for pancreatic
cancer.
Similarly, we also constructed molecular target and clinical target-based cancer networks,
using mutation target data from the Cancer Gene Census database
(http://www.sanger.ac.uk/genetics/CGP/Census/) [13] and FDA approved drug target data from
the DrugBank database (http://www.drugbank.ca) (Dataset S4), respectively. Mutation target
data from the Cancer Gene Census database used was updated in January 2009 and includes
mutation targets which have been implicated in the cancer. This database was chosen because it
is based on literature curation and thus captures information on the molecular mechanisms that
clinical researchers should have information on. Cytoscape version 2.4 was used to visualize the
networks [14].
Statistical analysis
The significance of the weighted degree values in the cancer networks was analyzed by
permutation tests. The distribution of the number of drugs or drug and mutation targets was kept
constant while the cancer-drug or cancer-target associations were randomized, respectively. The

35

p-value for the weighted degree of a cancer type is calculated as the fraction of the randomly
generated networks with a weighted degree value for a particular cancer which is equal to or
greater than the actual weighted degree value of that particular cancer (Table 2 and 5).
Conventional cutoff of 0.05 was used as a significance threshold. No multiple test correction has
been applied to the p-values. Therefore, given the number of statistical tests performed, some of
the associations reported, particularly borderline significant, could be spurious. In the FDA
cancer network, Wilcoxon test is used to determine if the weighted degree values of breast and
lung cancer are higher than the rest of the cancers in the network.
Shapiro-Wilk test suggests that some of the datasets used in this study are not normally
distributed. If p-values obtained from this test are lower than 0.05, the null hypothesis that data
are normally distributed is rejected. p-values for the 23 cancers and the cancers minus head and
neck cancer, mesothelioma and sarcoma are, respectively, 4.775e-05 and 0.0003658 for FDA
approval numbers, 0.002419 and 0.02321 for specific FDA drug percentage values, 0.001606
and 0.003714 for clinical trial numbers, 0.1582 and 0.3365 for FDA cancer network weighted
degree values, 0.07316 and 0.0341 for clinical trial cancer network weighted degree values. pvalue is 1.075e-06 for the global lethality values, and 0.07739 for the local lethality values.
Therefore, we used Spearman correlation coefficient values for the analysis of the relationships
between lethality values and the clinical trial and approval numbers, and the network weight
values (Table 3). The significance of the correlations was determined by a permutation based
algorithm [15].

36

APPENDICES

37

APPENDIX A
Table 1 Clinical trial and FDA approvals of drugs for different cancer types
Cancer type

Phase 3 trial only

Phase 1, 2 only

Phase 3 and FDA

FDA only

bladder cancer doxorubicin,
gemcitabine, paclitaxel

carboplatin, ifosfamide, bortezomib,
trastuzumab, ixabepilone

cisplatin

-

brain cancer

procarbazine, cisplatin,
ifosfamide, carboplatin,
thalidomide, etoposide

cladribine, irinotecan, bortezomib,
gefitinib, lenalidomide, busulfan,
erlotinib, oxaliplatin, imatinib,
temsirolimus, ixabepilone, topotecan,
methotrexate, lapatinib, capecitabine,
sorafenib

carmustine,
temozolomide,
cyclophosphamide

-

breast cancer

carboplatin, carmustine,
cisplatin, doxorubicin,
zoledronate, vinorelbine

decitabine, busulfan, etoposide,
fludarabine, leucovorin,
temozolomide, gefitinib, oxaliplatin,
pemetrexed, dasatinib, irinotecan,
bortezomib, erlotinib, imatinib,
vorinostat, alemtuzumab

anastrozole,
capecitabine,
docetaxel,
fulvestrant,
gemcitabine,
lapatinib, paclitaxel,
trastuzumab,
cyclophosphamide

epirubicin,
pamidronate,
toremifene,
fluorouracil

38

Table 1 (cont’d)
cervical
cancer

cisplatin

gemcitabine, capecitabine, paclitaxel,
fluorouracil, oxaliplatin, arsenic
trioxide, erlotinib, docetaxel, gefitinib

topotecan

-

colorectal
cancer

cisplatin, doxorubicin

gefitinib, erlotinib, gemcitabine,
trastuzumab, carmustine, ixabepilone,
imatinib

panitumumab

endometrial
cancer

cisplatin, doxorubicin,
paclitaxel

capecitabine, topotecan, pemetrexed,
oxaliplatin, raloxifene, thalidomide

bevacizumab,
capecitabine,
cetuximab,
irinotecan,
oxaliplatin,
fluorouracil,
leucovorin
-

esophagus
cancer

cisplatin, fluorouracil

oxaliplatin, capecitabine, carboplatin,
paclitaxel, irinotecan, topotecan,
decitabine, vinorelbine, doxorubicin,
docetaxel, erlotinib, arsenic trioxide,
leucovorin, ixabepilone

-

porfimer

eye cancer

-

busulfan, carboplatin, topotecan

-

cyclophosphamide

39

methotrexate

Table 1 (cont’d)
head and neck
cancer

cisplatin, fluorouracil,
paclitaxel

irinotecan, oxaliplatin, bevacizumab,
erlotinib, azacitidine, capecitabine,
carboplatin, temozolomide,
ixabepilone, doxorubicin, topotecan,
carmustine, cyclophosphamide,
etoposide, porfimer, thalidomide,
gefitinib, bortezomib, sorafenib,
gemcitabine

cetuximab, docetaxel

-

kidney cancer

carboplatin,cyclophosph
amide,doxorubicin,
etoposide

paclitaxel, irinotecan, oxaliplatin,
temozolomide, busulfan, fludarabine,
bevacizumab, cetuximab, erlotinib,
fluorouracil, pentostatin, topotecan,
arsenic trioxide, gefitinib,
methotrexate, capecitabine,
gemcitabine, lenalidomide, imatinib,
denileukin diftitox, thalidomide

-

sunitinib

leukemia

cytarabine, etoposide,
leucovorin, doxorubicin,
ifosfamide

topotecan, bexarotene, carboplatin,
bortezomib, temozolomide, rituximab,
bevacizumab, sorafenib, thalidomide,
denileukin diftitox, docetaxel,
ixabepilone, temsirolimus

alemtuzumab,
busulfan,
daunorubicin,
fludarabine,
idarubicin, imatinib,
mitoxantrone,
methotrexate,
cyclophosphamide

mechlorethamine,
nilotinib,
teniposide,
bendamustine

40

Table 1 (cont’d)
liver cancer

cisplatin, fluorouracil,
doxorubicin

irinotecan, oxaliplatin, temozolomide,
erlotinib, gemcitabine, pemetrexed,
capecitabine, carboplatin, topotecan,
thalidomide, docetaxel, epirubicin

-

-

lung cancer

cisplatin,carboplatin

docetaxel, erlotinib,
etoposide, gefitinib,
gemcitabine,
paclitaxel,
pemetrexed

mechlorethamine,
nofetumomab,
porfimer,
methotrexate

lymphoma

etoposide, doxorubicin,
ifosfamide, leucovorin,
cisplatin, idarubicin,
mitoxantrone,
daunorubicin

sunitinib, busulfan,
cyclophosphamide, fludarabine,
cetuximab, decitabine, imatinib,
irinotecan, doxorubicin, bortezomib,
ifosfamide, sorafenib, fluorouracil,
azacitidine, trastuzumab,
temozolomide, thalidomide,
temsirolimus
topotecan, paclitaxel, irinotecan,
oxaliplatin, busulfan, imatinib,
temozolomide, fludarabine, cladribine,
decitabine, alemtuzumab, arsenic
trioxide, altretamine, gemcitabine,
carboplatin, azacitidine, pentostatin,
thalidomide, ixabepilone,
temsirolimus, tretinoin, bevacizumab,
sorafenib

carmustine,
cytarabine, rituximab,
methotrexate,
cyclophosphamide

bexarotene,
methoxsalen,
procarbazine,
vorinostat,
bendamustine

mesothelioma

cisplatin

decitabine, doxorubicin, gemcitabine,
epirubicin, gefitinib, bevacizumab

pemetrexed

-

41

Table 1 (cont’d)
myeloma

-

arsenic trioxide, fludarabine,
etoposide, cisplatin, clofarabine

bortezomib,
thalidomide,
cyclophosphamide

carmustine,
doxorubicin,
lenalidomide,
zoledronate

ovarian cancer epirubicin, mitoxantrone vinorelbine, temozolomide, docetaxel,
ixabepilone, cisplatin, capecitabine,
etoposide, ifosfamide, gemcitabine,
bortezomib, lapatinib, erlotinib,
imatinib, gefitinib, anastrozole,
letrozole, pemetrexed, oxaliplatin,
alemtuzumab, leucovorin,
methotrexate, irinotecan, sorafenib,
toremifene, bevacizumab, cetuximab,
vorinostat

carboplatin,
paclitaxel,
cyclophosphamide

altretamine

pancreatic
cancer

oxaliplatin

lapatinib, irinotecan, bevacizumab,
cetuximab, cisplatin, pemetrexed,
imatinib, trastuzumab, capecitabine,
leucovorin, paclitaxel, docetaxel,
ixabepilone, bortezomib, arsenic
trioxide, temsirolimus,
cyclophosphamide

erlotinib,
gemcitabine,
fluorouracil

-

prostate
cancer

mitoxantrone,
zoledronate, toremifene

doxorubicin, paclitaxel, carboplatin,
epirubicin, temsirolimus, ixabepilone,
pemetrexed, oxaliplatin, sunitinib,
azacitidine, imatinib, trastuzumab,
arsenic trioxide, bevacizumab,
thalidomide, lapatinib

leuprolide, degarelix

-

42

Table 1 (cont’d)
sarcoma

cisplatin, doxorubicin,
etoposide, ifosfamide,
daunorubicin,
cyclophosphamide,
topotecan

irinotecan, oxaliplatin, temozolomide,
busulfan, erlotinib, carboplatin,
altretamine, leucovorin, paclitaxel,
thalidomide, gemcitabine,
trastuzumab, ixabepilone,
bevacizumab, cytarabine

methotrexate

-

skin cancer

cisplatin

lenalidomide, decitabine, irinotecan,
oxaliplatin, busulfan,
cyclophosphamide, etoposide,
fludarabine, docetaxel, leucovorin,
sorafenib, temozolomide, thalidomide,
denileukin diftitox, tretinoin,
carmustine, temsirolimus, bortezomib,
ixabepilone

-

daunorubicin,
doxorubicin,
imiquimod

stomach
cancer

-

irinotecan, cisplatin, gemcitabine,
vinorelbine, doxorubicin, paclitaxel,
leucovorin, oxaliplatin, ixabepilone,
erlotinib, capecitabine

-

imatinib, sunitinib

testicular
cancer

carboplatin,
cyclophosphamide,
paclitaxel

busulfan, fludarabine, temozolomide,
topotecan, ixabepilone, alemtuzumab,
arsenic trioxide, imatinib

etoposide,
ifosfamide, cisplatin

-

43

lung cancer
colorectal
cancer
breast cancer
pancreatic
cancer
prostate
cancer
leukemia
lymphoma
liver cancer
endometrial
cancer
ovarian cancer
esophagus
cancer
bladder cancer
brain cancer
kidney cancer
skin cancer
myeloma

Local lethality
ratio

Global lethality
ratio

Clinical trial
w.d.p

Clinical trial
w.d.

FDA w.d.p.

FDA w.d.

Clinical drug
trial number

FDA original
drug approval
number

FDA specific
drug
percentage (%)

FDA specific
drug approval
number

FDA drug
approval
number

Cancer type

Table 2 FDA approval numbers, clinical trial numbers, weighted degree values (w.d.), weighted degree p-values (w.d.p.) and death
statistics (global and local lethality ratio) of cancers in this study

14
8

3
4

21.4
50.0

1
3

121
61

1.32
0.46

0.029
0.840

7.58
6.15

0.066
0.899

0.286
0.088

0.753
0.336

19
3

10
0

52.6
0.0

2
1

97
35

1.17
0.5

0.003
0.898

6.81
6.96

0.037
0.576

0.072
0.061

0.222
0.910

4

3

75.0

0

48

0.41

0.937

4.32

0.998

0.051

0.154

23
16
1
1

16
9
0
0

69.6
56.3
0.0
0.0

0
0
0
0

170
121
19
10

0.65
0.83
0.33
1.06

0.157
0.195
0.899
0.653

5.16
6.41
6.57
4.43

0.656
0.013
0.714
0.996

0.038
0.036
0.033
0.028

0.490
0.276
0.862
0.186

6
1

2
0

33.3
0.0

2
1

71
26

1.16
0.07

0.488
0.970

7.42
6.8

0.018
0.605

0.027
0.025

0.717
0.867

2
3
3
4
8

1
1
1
1
3

50.0
33.3
33.3
25.0
37.5

0
1
1
2
1

10
62
30
31
13

0.25
0.89
0.5
0.48
0.86

0.967
0.748
0.899
0.885
0.563

4.34
6.57
7.22
5.84
2.81

0.997
0.606
0.221
0.924
1.000

0.025
0.023
0.023
0.020
0.019

0.205
0.599
0.239
0.165
0.537

44

Table 2 (cont’d)
stomach
cancer
cervical
cancer
testicular
cancer
eye cancer
head and neck
cancer
mesothelioma
sarcoma

4

0

0.0

1

19

1.13

0.609

6.32

0.759

0.019

0.506

1

0

0.0

0

20

0.24

0.939

5.48

0.932

0.007

0.350

3

1

33.3

2

11

0.31

0.966

5.31

0.984

0.001

0.047

1
3

0
0

0.0
0.0

0
0

2
45

0.78
1.35

0.691
0.558

1.58
8.03

1.000
0.000
0.100
0.056 -

1
2

0
0

0.0
0.0

0
1

40
7

0.07
1.21

0.984
0.636

3.23
7.39

1.000 0.259 -

45

-

Table 3 Correlation values of weighted degree, approval number values and of FDA specific
drug percentage with global and local lethality values

Global lethality

All cancer
types

Local lethality

All cancer
All cancer
types except
types except
locally lethal
globally
All cancer
cancers
lethal
types
(pancreatic,
cancers (lung
esophagus
cancer)
and liver
cancers)
Spearman
statistic
Spearman p-value

0.50

0.44

0.05

0.42

0.03

0.06

0.85

0.09

0.67

0.63

0.34

0.53

0.00

0.00

0.15

0.03

FDA cancer
network
weighted degree

0.25

0.12

0.14

0.53

0.29

0.62

0.57

0.03

Clinical trial
cancer network
weighted degree

0.42

0.33

0.61

0.55

0.06

0.17

0.00

0.03

FDA specific
drug
percentage

0.35

0.44

-0.32

-0.05

Spearman
statistic

0.13

0.06

0.17

0.85

Spearman p-value

FDA approval
number

Clinical trial
number

46

Spearman
statistic
Spearman p-value
Spearman
statistic
Spearman p-value
Spearman
statistic
Spearman p-value

Table 4 Cancer pairs with a weight difference of at least 0.5 or lower than 0
Cancer type 1

Cancer type 2

stomach cancer
head and neck
cancer
kidney cancer
ovarian cancer

esophagus cancer
kidney cancer

leukemia
ovarian cancer
cervical cancer
head and neck
cancer
head and neck
cancer
stomach cancer
brain cancer
ovarian cancer
head and neck
cancer
ovarian cancer
eye cancer
brain cancer
sarcoma

Clinical
trial
cancer
network
weight
0.71
0.56

FDA
Difference
cancer
network
weight
0.00
0.00

0.71
0.56

lung cancer
head and neck
cancer
lymphoma
breast cancer
esophagus cancer
brain cancer

0.54
0.54

0.00
0.00

0.54
0.54

0.68
0.61
0.50
0.50

0.15
0.09
0.00
0.00

0.53
0.53
0.50
0.50

liver cancer

0.50

0.00

0.50

cervical cancer
myeloma
myeloma
endometrial
cancer
eye cancer
myeloma
eye cancer
endometrial
cancer

0.50
0.17
0.10
0.25

0.00
0.22
0.17
0.33

0.50
-0.05
-0.06
-0.08

0.06
0.00
0.12
0.22

0.17
0.13
0.33
0.50

-0.11
-0.13
-0.21
-0.28

47

Table 5. Weighted degree values of drug target and mutation target based networks

leukemia
lung cancer
breast cancer
colorectal cancer
ovarian cancer
brain cancer
sarcoma
pancreatic cancer
endometrial
cancer
eye cancer
stomach cancer
lymphoma
testicular cancer
skin cancer
bladder cancer
head and neck
cancer
kidney cancer
liver cancer
myeloma
prostate cancer

Drug
Drug
Mutation Mutation
target
target
target
target
based
based
based
based
network network network network
weighted weighted weighted weighted
degree
degree
degree degree pvalue
p-value
value
value
2.28
0.000
0.25
0.003
3.35
0.000
1.12
0.210
3.29
0.001
1.08
0.160
2.75
0.290
1.57
0.000
2.53
0.634
1.43
0.011
1.66
0.674
0.9
0.016
1.76
0.961
0.42
0.509
0.62
1.000
0.82
0.706
0.66
0.996
0.78
0.732
1.79
2.12
2.17
1.83
2.22
2.32
2.75

0.309
0.774
0.287
0.817
0.808
0.330
0.275

0.17
0.53
0.19
0.33
0.35
0.14
0.06

0.880
0.909
0.952
0.980
0.981
0.998
1.000

1.38
1
1.24
0.88

0.996
0.999
0.999
1.000

0.13
0.26
0.16
0.19

1.000
1.000
1.000
1.000

48

Table 6 Mutation target- and drug target-based weight values of cancer pairs which have a
positive difference between the drug and mutation target-based values
Cancer type 1

Cancer type 2

colorectal cancer

endometrial
cancer
ovarian cancer
ovarian cancer

colorectal cancer
endometrial
cancer
brain cancer
ovarian cancer
colorectal cancer
brain cancer
brain cancer
breast cancer
endometrial
cancer
liver cancer
pancreatic cancer
brain cancer
colorectal cancer
head and neck
cancer
liver cancer
brain cancer
endometrial
cancer
eye cancer
lung cancer
brain cancer
brain cancer
brain cancer
brain cancer
breast cancer
breast cancer
colorectal cancer
pancreatic cancer
bladder cancer
brain cancer
breast cancer
eye cancer
prostate cancer

Difference
of mutation
target from
drug target
based
weight
0.26
0.24
0.15

colorectal cancer
pancreatic cancer
liver cancer
sarcoma
endometrial
cancer
stomach cancer
stomach cancer

0.14
0.14
0.1
0.09
0.08

pancreatic cancer
testicular cancer
lung cancer
pancreatic cancer
kidney cancer

0.08
0.08
0.07
0.06
0.06

ovarian cancer
prostate cancer
prostate cancer

0.06
0.05
0.05

lung cancer
stomach cancer
breast cancer
liver cancer
pancreatic cancer
stomach cancer
eye cancer
kidney cancer
sarcoma
skin cancer
sarcoma
kidney cancer
sarcoma
sarcoma
sarcoma

0.05
0.05
0.03
0.03
0.03
0.03
0.03
0.03
0.03
0.03
0.02
0.02
0.02
0.02
0.01

0.08
0.08

49

Table 7 Cancers with at least one common mutation and drug target
Cancer
type
lung
cancer
breast
cancer
leukemia

Common mutation
and drug target name
and Entrez Gene ID
ERBB1(1956)
ERBB2(2064)
FCGR2B(2213),
ABL1(25),
PDGFRB(5159),
KIT(3815), ABL2(27),
LCK(3932),
BCL2(596)

50

Figure 1 Cancer drug approval and clinical trial percentages. FDA cancer drug approval and clinical drug trial percentages for 23 cancers

51

Figure 2 Weighted degree values breast and lung cancers in the previous years. Weighted
degree values of breast cancer, lung cancer, and the remaining cancers in the FDA cancer
networks from 2000 to 2008. Average and the standard deviation of the weighted degree
values are shown. Wilcoxon test was performed for greater values of lung and breast cancer
than the other cancers. The networks with p-values lower than 0.05 are indicated by asterisk
(*).

52

Figure 3 FDA cancer network weighted degree vs. local lethality ratio. FDA cancer network
weighted degree values are plotted against local lethality ratio for (A) 23 cancers (r2 = 0.01, p
= 0.78), (B) the cancers except pancreatic, liver and esophagus cancers (r2 = 0.35, p = 0.01).
Lung cancer is shown as an open triangle and pancreatic, liver, esophagus cancers are shown
as open circles.
A
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0

0

0.2

0.4
0.6
Local lethality

0.8

1

B
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0
0

0.2

0.4
Local lethality

53

0.6

0.8

Figure 4 FDA specific drug percentages. FDA and clinical trial specific drug percentages for
the cancers except cervical, endometrial, esophagus, liver, pancreatic, eye, sarcoma,
mesothelioma, and stomach cancers, which do not have specific drugs

54

APPENDIX B
Table 8 Drug and cancer-type association with a year tag, based on FDA labels
Drug name
Alemtuzumab
Altretamine
Anastrozole
Arsenic trioxide
Azacitidine
Bendamustine
Bendamustine
Bevacizumab
Bevacizumab
Bevacizumab
Bexarotene
Bortezomib
Bortezomib
Busulfan
Capecitabine
Capecitabine
Carboplatin
Carmustine
Carmustine
Carmustine
Cetuximab
Cetuximab
Cisplatin
Cisplatin
Cladribine
Clofarabine
Cyclophosphamide
Cyclophosphamide
Cyclophosphamide
Cyclophosphamide
Cyclophosphamide
Cyclophosphamide
Cyclophosphamide
Cytarabine
Dasatinib
Daunorubicin

Other names of the drug
Campath
Hexalen
Arimidex
Trisenox
Vidaza
Treanda, Bendamustine
hydrochloride
Treanda, Bendamustine
hydrochloride
Avastin
Avastin
Avastin
Targretin
Velcade
Velcade
Myleran, Busulfex
Xeloda
Xeloda
Paraplatin
Gliadel, Bicnu
Gliadel, Bicnu
Gliadel, Bicnu
Erbitux
Erbitux

Cancer type
leukemia
ovarian cancer
breast cancer
leukemia
leukemia
leukemia
lymphoma

colorectal cancer
lung cancer
breast cancer
lymphoma
myeloma
lymphoma
leukemia
breast cancer
colorectal cancer
ovarian cancer
brain cancer
lymphoma
myeloma
colorectal cancer
head and neck
cancer
Platinol, Platinol-AQ, Cisplatin testicular cancer
Platinol, Platinol-AQ, Cisplatin bladder cancer
Leustatin
leukemia
Clolar
leukemia
Lyophilized cytoxan
brain cancer
Lyophilized cytoxan
breast cancer
Lyophilized cytoxan
eye cancer
Lyophilized cytoxan
leukemia
Lyophilized cytoxan
lymphoma
Lyophilized cytoxan
myeloma
Lyophilized cytoxan
ovarian cancer
DepoCyt, Cytosar-U
lymphoma
Sprycel
leukemia
Daunorubicin citrate,
skin cancer
DaunoXome, Daunorubicin
hydrochloride, Cerubidine

55

Label date
2001
1990
1995
2000
2004
2008
2008
2004
2006
2008
1999
2003
2006
1954
1998
2001
2003
1996
1997
1997
2004
2006
1978
1993
1993
2004
2000
2000
2000
2000
2000
2000
2000
1999
2006
1996

Table 8 (cont’d)
Daunorubicin

Decitabine
Degarelix
Denileukin diftitox
Docetaxel
Docetaxel
Docetaxel
Docetaxel
Docetaxel
Doxorubicin
Doxorubicin
Doxorubicin
Epirubicin
Erlotinib
Erlotinib
Estramustine
Etoposide
Etoposide
Exemestane
Fludarabine
Fluorouracil
Fluorouracil
Fluorouracil
Fluorouracil
Fulvestrant
Gefitinib
Gemcitabine
Gemcitabine
Gemcitabine
Gemtuzumab

Daunorubicin citrate,
DaunoXome, Daunorubicin
hydrochloride, Cerubidine
Dacogen
Ontak
Taxotere
Taxotere
Taxotere
Taxotere
Taxotere
Doxil, Doxorubicin
hydrochloride
Doxil, Doxorubicin
hydrochloride
Doxil, Doxorubicin
hydrochloride
Ellence, Epirubicin
hydrochloride
Tarceva, Erlotinib
hydrochloride
Tarceva, Erlotinib
hydrochloride
Emcyt, Estramustine phosphate
sodium
Etopophos, Etoposide
phosphate, Vepesid
Etopophos, Etoposide
phosphate, Vepesid
Aromasin
Fludara, Fludarabine phosphate
Fluoroplex
Fluoroplex
Fluoroplex
Fluoroplex
Faslodex
Iressa
Gemzar, Gemcitabine
hydrochloride
Gemzar, Gemcitabine
hydrochloride
Gemzar, Gemcitabine
hydrochloride
Mylotarg, Gemtuzumab
ozogamicin
56

leukemia

1998

leukemia
prostate cancer
lymphoma
breast cancer
lung cancer
prostate cancer
head and neck
cancer
stomach cancer
skin cancer

2006
2008
1999
1996
1999
2004
2006

ovarian cancer

1999

myeloma

2007

breast cancer

1999

lung cancer

2004

pancreatic cancer

2005

prostate cancer

1981

testicular cancer

1983

lung cancer

1986

breast cancer
leukemia
colorectal cancer
breast cancer
pancreatic cancer
stomach cancer
breast cancer
lung cancer
pancreatic cancer

1999
1991
1962
1998
1998
1998
2002
2003
1996

lung cancer

1998

breast cancer

2004

leukemia

2000

2006
1995

Table 8 (cont’d)
Ibritumomab
tiuxetan
Idarubicin

Ifosfamide
Imatinib
Imatinib
Imatinib
Imiquimod
Irinotecan
Ixabepilone
Lapatinib
Lenalidomide
Letrozole
Leucovorin
Leuprolide

Mechlorethamine
Mechlorethamine
Mechlorethamine
Methotrexate
Methotrexate
Methotrexate
Methotrexate
Methotrexate
Methotrexate
Methotrexate
Methoxsalen
Mitoxantrone
Nelarabine
Nelarabine

Zevalin

lymphoma

2002

Idamycin PFS, Idarubicin
hydrochloride, Idarubicin
hydrochloride PFS,
Ifex, Ifex/Mesnex kit,
Ifex/Mesna kit, Mesna
Gleevec, Imatinib mesylate
Gleevec, Imatinib mesylate
Gleevec, Imatinib mesylate
Aldara
Camptosar, Irinotecan
hydrochloride
Ixempra, Ixempra kit
Tykerb, Lapatinib ditosylate
Revlimid
Femara
Leucovorin Calcium
Eligard, Leuprolide acetate,
Lupron, Lupron depot, Lupron
depot-3, Lupron depot-4,
Lupron depot-ped, Viadur
Mustargen, Mechlorethamine
hydrochloride
Mustargen, Mechlorethamine
hydrochloride
Mustargen, Mechlorethamine
hydrochloride
Methotrexate LPF,
Methotrexate sodium, Trexall
Methotrexate LPF,
Methotrexate sodium, Trexall
Methotrexate LPF,
Methotrexate sodium, Trexall
Methotrexate LPF,
Methotrexate sodium, Trexall
Methotrexate LPF,
Methotrexate sodium, Trexall
Methotrexate LPF,
Methotrexate sodium, Trexall
Methotrexate LPF,
Methotrexate sodium, Trexall
Uvadex
Novantrone, Mitoxantrone
hydrochloride
Arranon
Arranon

leukemia

1997

testicular cancer

1988

leukemia
stomach cancer
sarcoma
skin cancer
colorectal cancer

2003
2003
2006
2004
1996

breast cancer
breast cancer
myeloma
breast cancer
colorectal cancer
prostate cancer

2007
2007
2006
1997
1952
1998

leukemia

1949

lung cancer

1949

lymphoma

1949

sarcoma

1988

breast cancer

2003

endometrial cancer

2003

head and neck
cancer
leukemia

2003

lung cancer

2003

lymphoma

2003

lymphoma
leukemia

1999
1987

leukemia
lymphoma

2005
2005

57

2003

Table 8 (cont’d)
Nilotinib
Nofetumomab
Oxaliplatin
Paclitaxel
Paclitaxel
Paclitaxel
Paclitaxel
Pamidronate
Pamidronate
Panitumumab
Pemetrexed
Pemetrexed
Pentostatin
Porfimer
Porfimer
Procarbazine
Raloxifene
Rituximab
Sorafenib
Sorafenib
Sunitinib
Sunitinib
Temozolomide
Temsirolimus
Teniposide
Thalidomide
Topotecan
Topotecan
Topotecan
Toremifene
Tositumomab
Trastuzumab
Tretinoin
Valrubicin
Vinorelbine
Vorinostat
Zoledronate

Tasigna, Nilotinib
hydrochloride monohydrate
Verluma
Eloxatin
Abraxane, Taxol
Abraxane, Taxol
Abraxane, Taxol
Abraxane, Taxol
Aredia, Pamidronate disodium
Aredia, Pamidronate disodium
Vectibix
Alimta, Pemetrexed disodium
Alimta, Pemetrexed disodium
Nipent
Photofrin, Porfimer sodium
Photofrin, Porfimer sodium
Matulane, Procarbazine
hydrochloride
Evista, Raloxifene
hydrochloride
Rituxan
Nexavar, Sorafenib tosylate
Nexavar, Sorafenib tosylate
Nexavar, Sorafenib tosylate
Nexavar, Sorafenib tosylate
Sutent, Sunitinib malate
Torisel
Temodar
Thalomid
Hycamtin, Topotecan
hydrochloride
Hycamtin, Topotecan
hydrochloride
Hycamtin, Topotecan
hydrochloride
Fareston, Toremifene citrate
Bexxar, Tositumomab and
iodine I 131 tositumomab
Herceptin
Vesanoid
Valstar
Navelbine, Vinorelbine tartrate
Zolinza
Zometa, Zoledronic acid

58

leukemia

2007

lung cancer
colorectal cancer
ovarian cancer
skin cancer
breast cancer
lung cancer
breast cancer
myeloma
colorectal cancer
lung cancer
mesothelioma
leukemia
esophagus cancer
lung cancer
lymphoma

1996
2005
1992
1997
1998
1998
1991
1991
2006
2004
2004
1991
1995
1998
1969

breast cancer

2007

lymphoma
kidney cancer
liver cancer
stomach cancer
kidney cancer
brain cancer
kidney cancer
leukemia
myeloma
ovarian cancer

1997
2005
2007
2006
2007
1999
2007
1992
2006
1996

lung cancer

1998

cervical cancer

2006

breast cancer
lymphoma

1997
2003

breast cancer
leukemia
bladder cancer
lung cancer
lymphoma
myeloma

1998
2004
1998
1994
2006
2002

Table 9 Targets of FDA approved cancer drugs
Drug
Alemtuzumab

Target
CD52(1043), FCGR1A(2209), FCGR3B(2215), C1R(715),
C1QA(712), C1QB(713), C1QC(714), FCGR3A(2214),
FCGR2A(2212), FCGR2B(2213), FCGR2C(9103)
Altretamine
DNA
Anastrozole
CYP19A1(1588)
Arsenic trioxide
ATP2C1(27032), AKT1(207), IL6(3569), MAPK1(5594),
ERGIC2(51290), CCND1(595), ABCB1(5243), JUN(3725),
MAPK3(5595), IKBKB(3551)
Azacitidine
DNA, DNMT1(1786)
Bevacizumab
VEGF(7422), FCGR1A(2209), FCGR3B(2215), C1R(715),
C1QA(712), C1QB(713), C1QC(714), FCGR3A(2214),
FCGR2A(2212), FCGR2B(2213), FCGR2C(9103)
Bexarotene
RXRB(6257)
Bortezomib
PSMD1(5707), PSMD2(5708), PSMB1(5689), PSMB5(5693),
PSMB2(5690)
Busulfan
DNA
Capecitabine
TYMS(7298), DPYD(1806)
Carboplatin
DNA, ALB(213)
Carmustine
DNA, GSR(2936)
Cetuximab
FCGR1A(2209), EGFR(1956), FCGR3B(2215), C1S(716),
C1R(715), C1QA(712), C1QB(713), C1QC(714),
FCGR3A(2214), FCGR2A(2212), FCGR2B(2213),
FCGR2C(9103)
Cisplatin
DNA
Cladribine
NP(4860), DNA, ADA(100)
Clofarabine
RRM1(6240), POLA1(5422)
Cyclophosphamide DNA
Cytarabine
POLB(5423), ALB(213)
Dasatinib
ABL1(25), PDGFRB(5159), KIT(3815), ABL2(27), SRC(6714),
FYN(2534), YES1(7525), EPHA2(1969), LCK(3932),
STAT5B(6777)
Daunorubicin
ABCC1(4363), DNA, ABCB1(5243)
Decitabine
DNA, DNMT1(1786)
Degarelix
Denileukin diftitox IL2RB(3560), IL2RA(3559), IL2RG(3561)
Docetaxel
TUBB1(81027), BCL2(596)
Doxorubicin
TOP2A(7153), DNA
Epirubicin
CHD1(1105), TOP2A(7153)
Erlotinib
EGFR(1956)
Estramustine
ESR1(2099), ESR2(2100), ABCB1(5243), MAP2(4133),
MAP1A(4130)
Etoposide
TOP2A(7153), MAP2K7(5609)
Exemestane
CYP19A1(1588)
Fludarabine
BCL2(596), RRM1(6240), POLA1(5422), STAT1(6772),
ADA(100)
59

Table 9 (cont’d)
Fluorouracil
Fulvestrant
Gefitinib
Gemcitabine
Gemtuzumab

Ibritumomab
tiuxetan

Idarubicin
Ifosfamide
Imatinib

Imiquimod
Irinotecan
Ixabepilone
Lapatinib
Lenalidomide
Letrozole
Leucovorin
Leuprolide
Mechlorethamine
Methotrexate
Methoxsalen
Mitoxantrone
Nelarabine
Nilotinib
Nofetumomab
Oxaliplatin
Paclitaxel
Pamidronate
Panitumumab
Pemetrexed
Pentostatin
Porfimer
Procarbazine
Raloxifene
Rituximab

TYMS(7298), DPYD(1806)
ESR1(2099)
EGFR(1956)
TYMS(7298), RRM1(6240), CMPK1(51727)
CD33(945), FCGR1A(2209), FCGR3B(2215), C1S(716),
C1R(715), C1QA(712), C1QB(713), C1QC(714),
FCGR3A(2214), FCGR2A(2212), FCGR2B(2213),
FCGR2C(9103)
MS4A1(931), FCGR1A(2209), FCGR3B(2215), C1S(716),
C1R(715), C1QA(712), C1QB(713), C1QC(714),
FCGR3A(2214), FCGR2A(2212), FCGR2B(2213),
FCGR2C(9103)
TOP2A(7153)
DNMT1(1786)
ABL1(25), PDGFRB(5159), KIT(3815), PDGFRA(5156),
CSF1R(1436), ABCB1(5243), NTRK1(4914), ABCG2(9429),
RET(5979), DDR1(780)
TLR7(51284)
TOP1MT(116447), TOP1(7150)
ERBB2(2064), EGFR(1956)
PTGS2(5743)
ESR1(2099), ERBB2(2064), CYP19A1(1588)
TYMS(7298)
GNRHR(2798)
DNA
DHFR(1719), ALB(213)
DNA, CYP2A6(1548)
TOP2A(7153), ABCG2(9429)
DNA
DNA
TUBB1(81027), BCL2(596)
FDPS(2224)
EGFR(1956)
TYMS(7298), DHFR(1719), GART(2618)
ADA(100)
LDLR(3949), FCGR1A(2209)
DNA
ESR1(2099), ESR2(2100)
MS4A1(931), FCGR1A(2209), FCGR3B(2215), C1S(716),
C1R(715), C1QA(712), C1QB(713), C1QC(714),
FCGR3A(2214), FCGR2A(2212), FCGR2B(2213),
FCGR2C(9103)

60

Table 9 (cont’d)
Sorafenib
Sunitinib

Temozolomide
Temsirolimus
Teniposide
Thalidomide
Topotecan
Toremifene
Tositumomab

Trastuzumab

Tretinoin

Valrubicin
Vinorelbine
Vorinostat
Zoledronate

FLT4(2324), RAF1(5894), FLT3(2322), PDGFRB(5159),
KDR(3791), KIT(3815), BRAF(673)
FLT4(2324), FLT1(2321), FLT3(2322), PDGFRB(5159),
KDR(3791), KIT(3815), PDGFRA(5156), CSF1R(1436),
RET(5979),
DNA
FRAP1(2475)
TOP2A(7153)
PTGS2(5743), TNF(7124), NFKB1(4790)
TOP1MT(116447), TOP1(7150), ABCG2(9429)
ESR1(2099)
MS4A1(931), FCGR1A(2209), FCGR3B(2215), C1R(715),
C1QA(712), C1QB(713), C1QC(714), FCGR3A(2214),
FCGR2A(2212), FCGR2B(2213), FCGR2C(9103)
ERBB2(2064), EGFR(1956), FCGR1A(2209), FCGR3B(2215),
C1S(716), C1R(715), C1QA(712), C1QB(713), C1QC(714),
FCGR3A(2214), FCGR2A(2212), FCGR2B(2213),
FCGR2C(9103)
NR0B1(190), RXRB(6257), RARG(5916), ALDH1A2(8854),
ALDH1A1(216), RARRES1(5918), GPRC5A(9052),
RXRG(6258)
TOP2A(7153)
TUBB(203068)
HDAC8(55869), HDAC1(3065), HDAC2(3066), HDAC3(8841),
HDAC6(10013)
FDPS(2224)

61

Table 10 Mutation targets of different cancer types
Cancer
type
Bladder
cancer
Brain
cancer

Mutation targets
FGFR3(2261), HRAS(3265)

ALK(238), APC(324), ATM(472), BRAF(673), COPEB(1316),
EGFR(1956), GOPC(57120), HRAS(3265), IDH1(3417),
KIAA1549(57670), MDM2(4193), MLH1(4292), MN1(4330),
MYCN(4613), NBS1(4683), NBS1(4683), NF1(4763), NF2(4771),
PALB2(79728), PHOX2B(8929), PIK3CA(5290), PIK3R1(5295),
PMS2(5395), PTCH(5727), PTEN(5728), ROS1(6098), SDHB(6390),
SDHC(6391), SDHD(6392), SUFU(51684), TP53(7157), WRN(7486)
Breast
AKT1(207), BRCA1(672), BRCA2(675), BRIP1(83990), CCND1(595),
cancer
CDH1(999), CHEK2(11200), EP300(2033), ERBB2(2064), ETV6(2120),
MAP2K4(6416), NTRK3(4916), PALB2(79728), PIK3CA(5290),
RB1(5925), TP53(7157)
colorectal
AKT1(207), APC(324), BRAF(673), CTNNB1(1499), EP300(2033),
cancer
FBXW7(55294), KRAS(3845), MADH4(4089), MAP2K4(6416),
MDM2(4193), MLH1(4292), MSH2(4436), MSH6(2956),
MUTYH(4595), PIK3CA(5290), PIK3R1(5295), PMS1(5378),
PMS2(5395), TP53(7157)
endometrial FBXW7(55294), FGFR2(2263), JAZF1(221895), MLH1(4292),
cancer
MSH2(4436), MSH6(2956), PMS1(5378), PMS2(5395), PTEN(5728),
SUZ12(23512)
eye cancer RB1(5925)
head and
MET(4233)
neck cancer
kidney
BHD(201163), FH(2271), GPC3(2719), MET(4233), NONO(4841),
cancer
PALB2(79728), PRCC(5546), PRO1073(29005), SFPQ(6421),
TFE3(7030), TFEB(7942), TSC1(7248), TSC2(7249), VHL(7428),
WT1(7490), WTX(139285)
leukemia
ABL1(25), ABL2(27), AF15Q14(57082), AF1Q(10962), AF3p21(51517),
AF5q31(27125), ARHGEF12(23365), ARNT(405), ATM(472),
BCL11A(53335), BCL11B(64919), BCL2(596), BCL3(602), BCL5(603),
BCL6(604), BCL9(607), BCR(613), BLM(641), BRCA2(675),
BRIP1(83990), BTG1(694), CBFA2T1(862), CBFA2T3(863),
CBFB(865), CBL (867), CCND1(595), CCND1(595), CCND2(894),
CDK6(1021), CDX2(1045), CEBPA(1050), CHIC2(26511),
CREBBP(1387), D10S170(8030), DDX10(1662), DEK(7913),
ELF4(2000), ELL(8178), ELN(2006), EP300(2033), EPS15(2060), ERG
(2078), ETV6(2120), EVI1(2122), EWSR1(2130), FACL6(23305),
FANCA(2175), FANCC(2176), FANCD2(2177), FANCE(2178),
FANCF(2188), FANCG(2189), FBXW7(55294), FCGR2B(2213),
FLT3(2322), FNBP1(23048), FOXO3A(2309), FOXP1(27086),
FSTL3(10272), FUS(2521), GAS7(8522), GATA1(2623), GATA2(2624),
GMPS(8833), GPHN(10243), GRAF(23092), HEAB(10978), HLF(3131),
HLXB9(3110), HOXA11(3207), HOXA13(3209), HOXA9(3205),
HOXC11(3227), HOXC13(3229), HOXD11(3237), HOXD13(3239)
62

Table 10 (cont’d)
leukemia

IGH@(3492), IKZF1(10320), JAK2(3717), JAK3(3718), KIT(3815),
KRAS(3845), LAF4(3899), LASP1(3927), LCK(3932), LCX(80312),
LMO1(4004), LMO2(4005), LPP(4026), LYL1(4066), MDS1(4197),
MKL1(57591), MLF1(4291), MLL(4297), MLLT1(4298),
MLLT10(8028), MLLT2(4299), MLLT3(4300), MLLT4(4301),
MLLT6(4302), MLLT7(4303), MN1(4330), MSF(10801),
MSI2(124540), MTCP1(4515), MYC(4609), MYH11(4629),
MYST4(23522), NCOA2(10499), NOTCH1(4851), NPM1(4869),
NRAS(4893), NSD1(64324), NUMA1(4926), NUP214(8021),
NUP98(4928), OLIG2(10215), PALB2(79728), PAX5(5079),
PBX1(5087), PCM1(5108), PDGFRB(5159), PER1(5187),
PICALM(8301), PML(5371), PMX1(5396), PNUTL1(5413),
PRDM16(63976), PSIP2(11168), PTPN11(5781), RANBP17(64901),
RAP1GDS1(5910), RARA(5914), RBM15(64783), RPL22(6146),
RPN1(6184), RUNX1(861), RUNXBP2(7994), SBDS(51119),
SEPT6(23157), SET(6418), SH3GL1(6455), SIL(6491),
SSH3BP1(10006), STL(7955), TAF15(8148), TAL1(6886), TAL2(6887),
TCF3(6929), TCL1A(8115), TCL6(27004), TFPT(29844), TIF1(8805),
TLX1(3195), TLX3(30012), TOP1(7150), TRA@(6955), TRB@(6957),
TRD@(6964), TRIP11(9321), TTL(150465), WHSC1L1(54904),
ZNF145(7704), ZNF384(171017), ZNF521(25925), ZNFN1A1(10320)
liver cancer APC(324), CTNNB1(1499), IL6ST(3572), TCF1(6927)
lung cancer AKT1(207), ALK(238), BRAF(673), EGFR(1956), EML4(27436),
ERBB2(2064), FGFR2(2263), KRAS(3845), MYCL1(4610), RB1(5925),
STK11(6794), TP53(7157)
lymphoma ALK(238), ALO17(57714), ARHH(399), ATIC(471), ATM(472),
BCL10(8915), BCL2(596), BCL6(604), BCL7A(605), BIRC3(330),
BLM(641), CARD11(84433), CARS(833), CCND2(894), CEP1(11064),
CLTC(1213), CLTCL1(8218), DDX6(1656), EIF4A2(1974),
ETV6(2120), FGFR1(2260), FGFR1OP(11116), FGFR3(2261),
FVT1(2531), HIST1H4I(8294), HSPCA(3320), HSPCB(3326),
IGH@(3492), IGK@(50802), IGL@(3535), IL2(3558), IL21R(50615),
IRTA1(83417), ITK(3702), LCP1(3936), MALT1(10892),
MHC2TA(4261), MSN(4478), MUC1(4582), MYC(4609), MYH9(4627),
NACA(4666), NBS1(4683), NFKB2(4791), NPM1(4869),
PAFAH1B2(5049), PAX5(5079), PCSK7(9159), PIM1(5292),
POU2AF1(5450), REL(5966), SFRS3(6428), SOCS1(8651 ),
SYK(6850), TFG(10342), TFRC(7037), TNFRSF17(608),
TNFRSF6(355), TPM3(7170), TPM4(7171), WAS(7454),
ZNF198(7750), ZNFN1A1(10320)
myeloma
CCND3(896), FGFR3(2261), HCMOGT1(92521), HIP1(3092),
IGH@(3492), IRF4(3662), MAF(4094), MAFB(9935), NRAS(4893),
PDGFRB(5159), PER1(5187), PTPN11(5781), RAB5EP(9135),
WHSC1(7468)
ovarian
AKT1(207), AKT2(208), BRAF(673), BRCA1(672), BRCA2(675),
cancer
CTNNB1(1499), ERBB2(2064), MLH1(4292), MSH2(4436),
MSH6(2956), PIK3R1(5295), PMS1(5378), PMS2(5395), STK11(6794)
63

Table 10 (cont’d)
pancreatic
cancer
prostate
cancer
sarcoma

skin cancer

stomach
cancer
testicular
cancer
thyroid
cancer

AKT2(208), APC(324), BRCA2(675), CDKN2A(1029), EP300(2033),
KRAS(3845), MADH4(4089), MAP2K4(6416), MEN1(4221),
STK11(6794)
C15orf21(283651), COPEB(1316), ERG (2078), ETV1(2115),
ETV4(2118), ETV5(2119), HNRNPA2B1(3181), PTEN(5728),
SLC45A3(85414), TMPRSS2(7113)
ASPSCR1(79058), BUB1B(701), CHN1(1123), CIC(23152),
COL1A1(1277), CREB1(1385), CREB3L2(64764), DDIT3(1649),
DUX4(22947), ERG (2078), ETV1(2115), ETV4(2118), ETV6(2120),
EWSR1(2130), EXT1(2131), EXT2(2132), FEV(54738), FLI1(2313),
FOXO1A(2308), FUS(2521), FUS(2521), HRAS(3265), MDM2(4193),
NBS1(4683), NCOA1(8648), NR4A3(8013), NTRK3(4916),
PAX3(5077), PAX7(5081), PDGFB(5155), POU5F1(5460), RB1(5925),
RECQL4(9401), SS18(6760), SS18L1(26039), SSX1 (6756),
SSX2(6757), SSX4(6759), TAF15(8148), TCF12(6938), TFE3(7030),
TP53(7157), WRN(7486), ZNF278(23598)
ATF1(466), BLM(641), BRAF(673), CDK4(1019), CDKN2A(1029),
DDB2(1643), ERCC2(2068), ERCC3(2071), ERCC4(2072),
ERCC5(2073), GNAQ(2776), KIT(3815), MITF(4286), NRAS(4893),
PTCH(5727), RECQL4(9401), SMO(6608), TNFRSF6(355), XPA(7507),
XPC(7508)
CDH1(999), ERBB2(2064), FGFR2(2263), PIK3CA(5290)
KIT(3815), STK11(6794), TNFRSF6(355)
AKAP9(10142 ), BRAF(673), D10S170(8030), ELKS(23085),
GOLGA5(9950), HMGA1(3159), HRPT2(3279), KRAS(3845),
KTN1(3895), MEN1(4221), NCOA4(8031), NRAS(4893),
NTRK1(4914), PAX8(7849), PCM1(5108), PPARG(5468),
PRKAR1A(5573), RET(5979), TFG(10342), TPM3(7170), TPR(7175),
TRIM33(51592), TSHR(7253), ZNF331(55422)

64

Table 11 Global lethality ratio values for different cancers from 2001 to 2007
Cancer
type

lung cancer
colorectal
cancer
breast
cancer
pancreatic
cancer
prostate
cancer
leukemia
lymphoma
liver cancer
endometrial
cancer
ovarian
cancer
esophagus
cancer
bladder
cancer
brain
cancer
kidney
cancer
skin cancer
stomach
cancer
myeloma
cervical
cancer
testicular
cancer
eye cancer

2001
Global
lethality
ratio
0.284
0.103

2002
Global
lethality
ratio
0.280
0.102

2003
Global
lethality
ratio
0.283
0.103

2004
Global
lethality
ratio
0.285
0.101

2005
Global
lethality
ratio
0.287
0.099

2006
Global
lethality
ratio
0.288
0.098

2007
Global
lethality
ratio
0.287
0.093

0.150

0.148

0.147

0.147

0.147

0.150

0.150

0.052

0.054

0.054

0.056

0.056

0.057

0.060

0.110

0.106

0.101

0.103

0.103

0.094

0.093

0.039
0.050
0.026
0.025

0.039
0.047
0.026
0.025

0.039
0.044
0.026
0.025

0.041
0.037
0.025
0.026

0.040
0.036
0.027
0.027

0.039
0.036
0.029
0.027

0.039
0.035
0.030
0.027

0.052

0.052

0.053

0.059

0.059

0.056

0.057

0.023

0.023

0.023

0.024

0.024

0.024

0.025

0.022

0.023

0.023

0.023

0.023

0.023

0.025

0.024

0.024

0.024

0.023

0.022

0.023

0.023

0.022

0.021

0.021

0.022

0.022

0.023

0.023

0.018
0.023

0.017
0.022

0.018
0.022

0.018
0.021

0.019
0.020

0.019
0.020

0.019
0.020

0.020
0.017

0.020
0.015

0.020
0.015

0.020
0.014

0.020
0.014

0.020
0.014

0.019
0.014

0.001

0.001

0.001

0.001

0.001

0.001

0.001

0.000

0.000

0.000

0.000

0.000

0.000

0.000

65

Table 12 Local lethality ratio values for different cancers from 2001 to 2007
Cancer type

lung cancer
colorectal
cancer
breast cancer
pancreatic
cancer
prostate
cancer
leukemia
lymphoma
liver cancer
endometrial
cancer
ovarian
cancer
esophagus
cancer
bladder
cancer
brain cancer
kidney cancer
skin cancer
stomach
cancer
myeloma
cervical
cancer
testicular
cancer
eye cancer

2001
Local
lethality
ratio
0.929
0.360

2002
Local
lethality
ratio
0.914
0.328

2003
Local
lethality
ratio
0.914
0.680

2004
Local
lethality
ratio
0.923
0.267

2005
Local
lethality
ratio
0.947
0.268

2006
Local
lethality
ratio
0.931
0.259

2007
Local
lethality
ratio
0.752
0.232

0.209
0.990

0.195
0.980

0.188
0.977

0.186
0.981

0.191
0.988

0.192
0.958

0.227
0.898

0.159

0.160

0.131

0.130

0.131

0.117

0.124

0.682
0.434
0.870
0.172

0.704
0.424
0.849
0.168

0.716
0.405
0.832
0.170

0.697
0.333
0.754
0.176

0.648
0.323
0.879
0.179

0.635
0.305
0.875
0.178

0.492
0.276
0.876
0.189

0.594

0.597

0.563

0.629

0.729

0.759

0.681

0.947

0.962

0.935

0.933

0.935

0.946

0.896

0.228

0.223

0.218

0.211

0.208

0.213

0.205

0.762
0.393
0.174
0.590

0.771
0.365
0.165
0.574

0.716
0.373
0.167
0.540

0.690
0.349
0.173
0.519

0.690
0.350
0.160
0.528

0.681
0.330
0.156
0.513

0.621
0.252
0.167
0.527

0.778
0.341

0.740
0.315

0.747
0.336

0.725
0.371

0.707
0.358

0.683
0.381

0.542
0.329

0.056

0.053

0.053

0.040

0.049

0.045

0.048

0.095

0.091

0.091

0.086

0.108

0.097

0.094

66

Table 13 Correlation values of weighted degree, approval number values with global
lethality values for 2001-2007
FDA approval
number

FDA cancer network
weighted degree

All
cancers

All
cancers
except
lung
cancer

All
cancers

All
cancers
except
lung
cancer

0.440047
0.100702
0.434419
0.105646
0.384073
0.141915
0.454699
0.076807
0.489158
0.046291
0.4698
0.049159
0.415906
0.076549

-0.01325
0.961142
0.010302
0.969796
0.295524
0.249477
0.256285
0.320752
0.273761
0.271658
0.414946
0.077301
0.301958
0.195699

-0.01431
0.959628
-0.00179
0.994956
0.207506
0.440621
0.133922
0.620973
0.159509
0.540859
0.313503
0.205218
0.185413
0.447281

2001 0.507883
0.044594
2002 0.493703
0.05195
2003 0.453218
0.067694
2004 0.529063
0.028981
2005 0.549142
0.018258
2006 0.525673
0.020802
2007 0.479393
0.032453

67

Spearman statistic
Spearman p-value
Spearman statistic
Spearman p-value
Spearman statistic
Spearman p-value
Spearman statistic
Spearman p-value
Spearman statistic
Spearman p-value
Spearman statistic
Spearman p-value
Spearman statistic
Spearman p-value

Table 14 Correlation values of weighted degree, approval number values with local lethality
values for 2001-2007
FDA approval
number

All
cancers

2001 0.062646
0.817717
2002 0.078756
0.771877
2003 0.203359
0.433722
2004 0.164304
0.528597
2005 0.197299
0.432609
2006 0.174395
0.475188
2007 0.072584
0.761048

FDA cancer network
weighted degree

All
All
cancers
cancers
except
pancreatic,
liver and
esophagus
cancers
0.40361
0.152398
0.425737
0.129076
0.553168
0.03244
0.522921
0.045488
0.39853
0.126272
0.389871
0.121863
0.449279
0.070419

0.267647
0.312408
0.270588
0.306974
0.144697
0.57952
0.137255
0.595315
0.168215
0.500542
0.124561
0.607889
0.171429
0.466237

All
cancers
except
pancreatic,
liver and
esophagus
cancers
0.481319
0.082451
0.485714
0.079402
0.37891
0.163683
0.407143
0.131493
0.358824
0.170818
0.330882
0.192679
0.573529
0.017639

68

Spearman statistic
Spearman p-value
Spearman statistic
Spearman p-value
Spearman statistic
Spearman p-value
Spearman statistic
Spearman p-value
Spearman statistic
Spearman p-value
Spearman statistic
Spearman p-value
Spearman statistic
Spearman p-value

Table 15 Weighted degree values for FDA cancer network from 2000 to 2007
breast
cancer
lung cancer
leukemia
lymphoma
ovarian
cancer
head and
neck cancer
myeloma
stomach
cancer
sarcoma
endometrial
cancer
eye cancer
brain cancer
colorectal
cancer
skin cancer
pancreatic
cancer
kidney
cancer
liver cancer
prostate
cancer
cervical
cancer
testicular
cancer
bladder
cancer
esophagus
cancer
mesotheliom
a

2000
0.944

2001
1.013

2002
0.936

2003
1.192

2004
1.269

2005
1.250

2006
1.327

2007
1.139

0.774
0.559
0.802
1.258

0.771
0.525
0.797
1.255

0.763
0.514
0.716
1.223

1.049
0.832
0.983
1.031

1.092
0.706
0.957
0.967

1.158
0.711
0.934
0.962

1.315
0.681
0.858
1.097

1.296
0.649
0.839
1.164

NA

NA

NA

2.312

2.282

2.273

1.411

1.389

1.401
0.917

1.397
0.833

1.138
0.827

0.937
0.663

0.923
0.591

0.913
0.488

0.771
1.009

0.873
1.156

0.000
NA

0.000
NA

0.000
NA

2.312
2.312

2.282
2.282

2.273
2.273

1.237
1.086

1.223
1.072

1.144
1.319
0.655

1.138
1.314
0.593

1.037
1.191
0.583

0.921
1.067
0.525

0.906
1.054
0.391

0.896
1.041
0.336

0.824
0.936
0.434

0.793
0.897
0.422

0.567
0.927

0.562
0.877

0.558
0.871

0.484
0.683

0.416
0.681

0.414
0.629

0.406
0.525

0.487
0.505

NA

NA

NA

NA

NA

0.000

0.000

0.500

NA
0.000

NA
0.000

NA
0.000

NA
0.000

NA
0.125

NA
0.125

NA
0.488

0.333
0.479

NA

NA

NA

NA

NA

NA

0.238

0.238

0.341

0.341

0.341

0.327

0.317

0.317

0.313

0.313

0.250

0.250

0.250

0.250

0.250

0.250

0.250

0.250

0.111

0.111

0.111

0.091

0.077

0.077

0.071

0.071

NA

NA

NA

NA

0.077

0.077

0.071

0.071

69

Table 16 Weighted degree p-values for FDA cancer network from 2000 to 2007
breast cancer
lung cancer
leukemia
lymphoma
ovarian
cancer
head and
neck cancer
myeloma
stomach
cancer
sarcoma
endometrial
cancer
eye cancer
brain cancer
colorectal
cancer
skin cancer
pancreatic
cancer
kidney cancer
liver cancer
prostate
cancer
cervical
cancer
testicular
cancer
bladder
cancer
esophagus
cancer
mesothelioma

2000
0.063
0.327
0.383
0.299
0.315

2001
0.035
0.289
0.391
0.298
0.291

2002
0.027
0.343
0.401
0.292
0.286

2003
0.028
0.164
0.169
0.188
0.515

2004
0.007
0.088
0.181
0.176
0.535

2005
0.012
0.073
0.151
0.161
0.528

2006
0.007
0.022
0.166
0.203
0.509

2007
0.009
0.024
0.155
0.212
0.466

NA

NA

NA

0.369

0.388

0.401

0.544

0.522

0.379
0.618

0.415
0.596

0.426
0.617

0.627
0.817

0.615
0.832

0.641
0.843

0.664
0.694

0.563
0.582

1.000
NA

1.000
NA

1.000
NA

0.367
0.380

0.386
0.387

0.374
0.379

0.612
0.632

0.623
0.617

0.520
0.434
0.706

0.506
0.431
0.732

0.541
0.450
0.711

0.723
0.672
0.838

0.723
0.650
0.897

0.690
0.653
0.878

0.714
0.747
0.856

0.724
0.741
0.868

0.741
0.596

0.787
0.627

0.775
0.619

0.858
0.797

0.890
0.786

0.885
0.814

0.925
0.885

0.894
0.881

NA
NA
1.000

NA
NA
1.000

NA
NA
1.000

NA
NA
1.000

NA
NA
0.966

1.000
NA
0.983

1.000
NA
0.880

0.891
0.907
0.898

NA

NA

NA

NA

NA

NA

0.931

0.931

0.862

0.862

0.858

0.919

0.929

0.931

0.942

0.945

0.915

0.913

0.896

0.939

0.946

0.942

0.949

0.962

0.940

0.944

0.928

0.956

0.987

0.980

0.979

0.985

NA

NA

NA

NA

0.966

0.973

0.979

0.984

70

Table 17 Clinical trial numbers along with distinct drug number and specific drug number
for clinical trials
Cancer type

leukemia
lymphoma
lung cancer
breast cancer
ovarian cancer
brain cancer
colorectal cancer
prostate cancer
head and neck cancer
mesothelioma
pancreatic cancer
skin cancer
kidney cancer
esophagus cancer
cervical cancer
liver cancer
stomach cancer
myeloma
testicular cancer
bladder cancer
endometrial cancer
sarcoma
eye cancer

Clinical
drug
trial
number
170
121
121
97
71
62
61
48
45
40
35
31
30
26
20
19
19
13
11
10
10
7
2

Clinical
trials
distinct
drug
number
37
42
30
37
34
25
16
23
26
8
21
21
27
16
11
16
13
9
14
10
9
24
3

Clinical
trials
specific
drug
number
2
3
0
2
0
1
0
3
1
0
0
0
0
0
0
0
0
1
0
1
0
0
0

71

Figure 5 Time dependent characteristics of the FDA approvals and FDA cancer network (A)
Number of cancers in the network from 1980-2008, (B) Number of FDA approvals from
1980-2008, (C) Average weight of the network from 1980-2008, (D) Number of components
of the network from 1980-2008.
A

Number of cancers

25
20
15
10
5

08
20

06
20

04
20

02
20

00
20

98
19

96
19

94
19

92
19

90
19

88
19

86
19

84
19

82
19

19

80

0

B

120
100
80
60
40
20

72

08
20

06
20

04
20

02
20

00
20

98
19

96
19

94
19

92
19

90
19

88
19

86
19

84
19

82
19

80

0

19

Number of FDA approvals

140

73

20

20

20

20

20

19

19

19

19

19

19

19

19

19

19

08

06

04

02

00

98

96

94

92

90

88

86

84

82

80

Number of components

20

20

20

20

20

19

19

19

19

19

19

19

19

19

19

08

06

04

02

00

98

96

94

92

90

88

86

84

82

80

Average weight

Figure 5 (cont’d)

C
0.14

0.12

0.10

0.08

0.06

0.04

0.02

0.00

D
3

2

1

0

Figure 6 Cluster dendogram of cancer types based on global and local lethality values

74

BIBLIOGRAPHY

75

BIBLIOGRAPHY

1.

Miniño AM, Heron MP, Murphy SL, Kochanek KD; Centers for Disease Control and
Prevention National Center for Health Statistics National Vital Statistics System. Deaths:
final data for 2004. Natl Vital Stat Rep. 2007, 55:1-119

2.

Goh KI, Cusick ME, Valle D, Childs B, Vidal M, et al. The human disease network. Proc
Natl Acad Sci U S A. 2007, 104: 8685-8690

3.

Yildirim MA, Goh KI., Cusick ME, Barabasi AL, Vidal M. Drug-target network. Nat
Biotechnol. 2007, 25: 1119-1126

4.

Ma'ayan A, Jenkins SL, Goldfarb J, Iyengar R. Network analysis of FDA approved drugs
and their targets. Mt Sinai J Med. 2007, 74: 27-32

5.

Sakharkar MK, Li P, Zhong Z, Sakharkar KR. Quantitative analysis on the characteristics
of targets with FDA approved drugs. Int J Biol Sci. 2008, 4:15-22

6.

Nacher JC, Schwartz JM. A global view of drug-therapy interactions. BMC Pharmacol.
2008, 8:5

7.

Barabasi AL. Network Medicine — From Obesity to the "Diseasome". N Engl J Med.
2007, 357: 404-407

8.

Hopkins, AL. Network pharmacology. Nat Biotechnol. 2007, 25: 1110-1111

9.

American Cancer Society. Cancer Facts & Figures. Atlanta: American Cancer Society
2001-2008

10.

Newman MEJ. The structure and function of complex networks. SIAM Review. 2003,
45: 167-256

11.

Yokota J, Yamamoto T, Toyoshima K, Terada M, Sugimura T, et al. Amplification of cerbB-2 oncogene in human adenocarcinomas in vivo. Lancet. 1986, 1:765-767

12.

Testa JR, Siegfried JM. Chromosome abnormalities in human non-small cell lung cancer.
Cancer Res. 1992, 52:2702-2706

13.

Futreal PA, Coin L, Marshall M, Down T, Hubbard T, et al. A census of human cancer
genes. Nat Rev Cancer. 2004, 4:177-183

76

14.

Shannon P, Markiel A, Ozier O, Baliga NS, Wang JT, et al. Cytoscape: a software
environment for integrated models of biomolecular interaction networks. Genome Res.
2003, 13: 2498-2504

15.

Best DJ, Roberts DE. Algorithm AS 89: The Upper Tail Probabilities of Spearman's rho.
Applied Statistics 1975, 24: 377–379

77

CHAPTER 2
INTEGRATIVE ANALYSIS OF CANCER PATHWAYS
INTRODUCTION
Cancer is a complex disease, with many subtypes, affecting various tissues, and
according to the severity of the abnormality, giving rise to different stages and classifications,
such as carcinoma, sarcoma, primary, etc. Physiological and genetic studies identified different
stages in the progression of the various types of cancers. For instance, colorectal tumorigenesis
begins with normal epithelium which proceeds through stages of hyperproliferative epithelium,
early adenoma, intermediate adenoma, late adenoma, carcinoma, and metastasis [1]. Genetic
alterations, such as mutations or deletions of genes, accumulate as the cancer progresses to the
next, more severe and proliferative stage. The accumulation of changes (e.g., mutations,
deletions, etc.) is a key factor in tumor progression. For example, a more severe stage will have a
higher probability than a less severe stage of having more mutations [1]. On the other hand, the
specific number and the identity of the genetic alterations, such as point mutations,
amplifications, or deletions in a stage, are not determining factors, but rather are general features
that characterize the severity of a stage, and varies from one cancer to another [1].
Omics technologies, such as cDNA and oligonucleotide arrays, and comparative genome
hybridization, have dominated tumor characterization [2]. Genome level analysis have been
applied successfully to differentiate between the different stages of cancer, and revealed
differentially expressed genes and genomic alterations that play important roles in the
development of cancer [2]. More recently, gene set and pathway centric analysis of cancer
progression have gained popularity. These analyses confirmed results currently known about the
role of cell cycle in cancer development, as well as produced novel findings, i.e. the involvement

78

of the ERBB4 gene in primary prostate cancer [3]. Approaches that integrate gene expression
and pathway analysis of colorectal cancer have proved useful in finding potential prognostic and
diagnostic markers, as well as therapeutic targets [2].
Biological pathways consist of molecular interactions or other biochemical events among
a group of proteins, genes, and other chemicals. They are used to characterize metabolism, signal
transduction, specific cellular processes, diseases, or drug mechanisms. The Kyoto Encyclopedia
of Genes and Genomes (KEGG) database contains pathway information on the progression of
different types of cancer, as combinations of signaling pathways
(http://www.genome.ad.jp/kegg/pathway.html#disease) [4, 5, 6]. Signaling pathways capture the
molecular interactions and reactions that take signals from the outside to the nucleus of the cell,
where transcriptional regulation occurs. Signaling pathways, e.g. the MAPK, Wnt, TGF-beta
signaling, are well studied in the context of cell proliferation [7]. Cancer is the only human
disease for which pathway information of the disease progression is available, i.e., at different
stages of the cancer, from normal tissue to the advanced tumor phase in KEGG [4, 5, 6]. KEGG
cancer pathways are different in that they contain members of different signaling pathway
members within a single pathway. This is because they contain information on various stages of
the cancer and the various stages involve different signaling pathways. The pathway information
contained in KEGG, i.e., the genes in each stage of the disease progression that are genetically
altered in the colorectal cancer pathway, is derived from the literature [8, 9]. In the KEGG
colorectal cancer pathway, the Wnt pathway members are at the upper positions of the image,
corresponding to normal or initial stages of the cancer progression, which is supported by the
literature [8, 9, 10]. Constant activation of Wnt signaling targets occurs in adenomatosis
polyposis coli (APC) or beta-catenin mutations, as well as upon phosphorylation by glycogen

79

synthase kinase-3-beta. These events lead to early dysplastic lesions, which is an early event in
colorectal cancer [10]. Similar support exists in the literature to suggest that genes, i.e., DCC and
KRAS, should be below the Wnt pathway members since their alterations occur later in the
disease progression. Furthermore, DCC is located below KRAS in the KEGG pathway because
DCC alterations are believed to occur later than KRAS alterations in the development of
colorectal cancer [11]. Finally, TGF-beta pathway members are located in the lower portions of
the pathway image because the literature suggests that their alterations occur during more
advanced stages of the cancer [12]. The significance of the KEGG cancer pathways is the
integration of the cancer stages with signaling pathways. Although the integration of signaling
pathways with genome level expression data has been widely performed, it has yet to be realized
with cancer pathways.
Using pathway information to understand genome level expression data has been
extensively applied [13, 14, 15]. The approach integrates a priori knowledge of a gene's
functional role with expression data to detect for concerted expression changes in a set of genes
responsible for producing a phenotype [13]. Pathway-centric analysis of tumor microarray data
has been successfully applied to identify signaling pathway members [16]. IL-1 and ER-induced
pathways were found to be significantly coexpressed in breast cancer data. In addition to
analyzing the entire dataset, analysis of individual samples or a subset of the data identified
significant pathway activities that were relevant to the biological context of the tissue or organ.
For example, pathway analysis uncovered the expected association of estrogen-induced
pathways within a group of clinical breast cancer data [16], and signaling and metabolic
pathways involved in the development of type 2 diabetes [17]. On the other hand, gene
expression level analysis within pathways is an area that mostly has been ignored. Some studies

80

have analyzed the relationship among the members of protein complexes or pathways in terms of
their gene expression levels [18, 19, 20, 21]. Protein complexes, such as ribosome and
proteasome, show significant correlation in their gene expression levels [18]. In addition, the ciselement profiles are highly similar for members within a signaling pathway, such as the KEGG
apoptosis pathway, and functionally related interacting proteins (i.e., protein complexes). This
suggests that a strong relationship in the gene expression levels between members of these
pathways should exist [19]. Coherence is a measure of the level of correlation among a group of
genes. A coherent group of genes may share similar regulation of their gene expression levels.
Indeed, genes in the same pathway with similar functions have been shown to be coherent as
compared to a random group of genes from the genome [20].
Previous studies that integrate gene expression data with pathway information have not
incorporated the dimensionality of the pathways. Most studies have focused on the members of
the pathways. Thus far, there has been one study of pathways that incorporated the position in
the pathways in the analysis of the genome level data. They developed a statistical impact
analysis that used the pathway position to calculate the significance of the pathway [21]. With
this approach they identified the Focal Adhesion Pathway as a significant pathway for lung
cancer, which was not found using classical approaches, such as gene set enrichment and gene
ontology analyses, thereby enhancing the information content extracted from analyzing genome
level data. Impact analysis considers expression alteration in receptors, such as integrin, receptor
tyrosine kinase (RTK), and the receptor ligand vascular endothelial growth factor (VEGF), as
important parameters in the analysis, since they affect the downstream molecules in the pathway.
The Focal Adhesion Pathway, on the other hand, was not found to be significant using classical
approaches because the other genes in the pathway were not significantly altered. Classical

81

approaches analyze pathways as a whole without special emphasis on receptor molecules or
other position(s) in a pathway.
In this study, we capitalized upon the progressive nature of the cancer disease captured in
the biological knowledge represented in the KEGG cancer pathways. We analyzed the gene
expression levels, coherence, and mutation target data of the pathway members to determine if
there is a significant relationship or correlation within any group of pathway members from the
rest of the pathways. Analyzing the KEGG colorectal cancer pathway with microarray data
identified that different parts of the pathways were up-regulated or coherent at the mRNA level,
at the different stages of cancer progression, i.e., adenoma vs. carcinoma. Since the KEGG
cancer pathways integrate different signaling pathways of the various cancer stages, unlike
classical signaling pathways, we analyzed the coherence of the colorectal cancer pathway, and
found the carcinoma expression data was more coherent than the normal or adenoma data. In
addition, mutation targets were found to be localized primarily in the nucleus of the cell and
concentrated at the later stages of the cancer.

MATERIALS AND METHODS
Pathway data
We used KEGG as our source of pathway information [4, 5, 6]. We focused on the
cancer, apoptosis, oxidative phosphorylation and proteasome pathways. We collected a list of
human genes for these pathways. For each protein/gene in the pathways we used all the different
homologues provided by KEGG database.

82

X/Y scale
The X scale represented the direction from receptor to nucleus. The Y scale (analyzed
only for the colorectal cancer pathway) represented the direction from normal tissue to advanced
tumor or metastasis. We ignored the sub-pathway distinction (such as Chromosome Unstable
Pathway and Microsatellite Unstable Pathway in colorectal cancer pathway). We generated 3 Xdependent and 4 Y-dependent groups in the KEGG colorectal cancer pathway, and 2 Xdependent groups in the KEGG apoptosis pathway. For X, receptor ligands and receptors are
designated by a value of 1. If there is a nuclear distinction in the pathway; molecules in the
cytosol are designated by a value of 2, and molecules in the nucleus are designated by a value of
3, otherwise molecules in the cytoplasm are designated by a value of 2. For Y (analyzed only for
the KEGG colorectal cancer pathway), the first stage of the pathway, which signified the first
initial molecular events (in the colorectal cancer pathway, the first stage is represented by the
transition from normal epithelium to early adenoma) is designated by a value of 1 and the next
stages in the cancer progression are designated by values of 2, 3, 4 (the last stage). The last stage
is defined by the latest events in the progression of the disease in the KEGG pathway. For
example, in the colorectal cancer pathway, the last stage is represented by the transition from late
adenoma to carcinoma. The biomolecules involved in the transition to the different stages
provided by KEGG are supported by the literature, as discussed in the Introduction above. The
KEGG information was used as is, unless the stage is an insignificant one. For example,
dysplastic aberrant crypt foci stage was ignored in our analysis of the colorectal cancer pathway,
because this stage covered a very small part of the pathway and it was unclear which molecules
were associated with this stage. In the colorectal cancer pathway, normal epithelium to early
adenoma is designated by a Y value of 1 and includes proteins such as, Frizzled, glycogen

83

synthase kinase-3-beta (GSK-3β), adenomatosis polyposis coli (APC), T cell factor/lymphoid
enhancer factor (TCFLEF), Survivin, etc. Early adenoma to intermediate adenoma is designated
by a Y value of 2 and includes proteins such as, RTK, K-Ras, protein kinase B (PKB),
extracellular signal regulated protein kinase (ERK), C-Fos, etc. Intermediate adenoma to late
adenoma is designated by a Y value of 3 and includes proteins such as, deleted in colon cancer
(DCC), caspase 3 (CASP3), human mutL homolog 1 (hMLH1), etc. Late adenoma to carcinoma
is designated by a Y value of 4 and includes proteins such as cytochrome c (Cytc), p53, etc.
Since our analysis focused on the potential of the genes in contributing to the progression of
cancer, we used the larger value of Y for a gene if there were more than one value associated
with the gene. For consistency, the same approach was taken for the X value, i.e., the larger of
the two values was used. For example, if a protein is present in more than one location, the larger
value is assigned, for instance, CyclinD1 is present in two Y locations, 1 and 2, but was assigned
a Y value of 2, and transforming growth factor beta receptor 2 (TGFBR2) is located in two X
locations, 1 and 3, but was assigned a X value of 3.
Expression data
Normalized microarray data (GSE4183 and GSE8671) were downloaded from the NCBI
GEO database (http://www.ncbi.nlm.nih.gov/geo/) GSE4183 includes normal colon tissue, colon
adenoma and colon carcinoma gene expression datasets for several biopsy samples obtained
from individuals. The GSE8671 dataset includes the whole genome mRNA expression level for
32 colorectal adenomas paired with the normal mucosa from the same individuals. In these largescale datasets, we focused on the expression values of the pathway members, as oppose to
analyzing all the genes. Both datasets contained expression values for all the gene members in
the colorectal cancer pathway. If there were more than one value for a gene in a sample (same

84

column), the mean values were used. We combined the normal and the adenoma samples from
both datasets, to obtain a set of expression values for normal, adenoma and carcinoma samples.
We calculated expression ratio for adenoma (or carcinoma) by dividing the average value of
adenoma (or carcinoma) samples to the average value of normal samples. For adenoma, the ratio
was calculated by dividing the average of the combined adenoma values to the average of the
combined normal values. Since carcinoma data is present only in GSE4183, the average of the
carcinoma values in GSE4183 were divided by the average of the normal values in GSE4183.
For the calculation of carcinoma/adenoma ratio, the carcinoma ratio was divided by the adenoma
ratio for each gene.
Drug and mutation targets
We collected drug information from the National Cancer Institute web page
(http://www.cancer.gov/cancertopics/druginfo/alphalist). We limited the drug list to ones which
targeted a cellular signaling protein. Drugs that targeted molecules with a general role in the cell
were excluded; for example, Bortezomib, a drug approved for leukemia, was excluded because it
targeted the proteasome. We collected mutation target information from two sources. The first
source was the Cancer Gene Census of the Cancer Genome project [22]
(http://www.sanger.ac.uk/genetics/CGP/Census/). We used only the cancer types provided by
this dataset. For some mutation targets, such as p53, “others” was mentioned in addition to the
cancer type listed, therefore we were not able to include the cancer types referred to as “others”
in the analysis. The second source was a list of genetically altered genes provided by KEGG for
each pathway. We combined these two sources of data to obtain a list of mutation targets for
each cancer pathway. The list from the two sources did not overlap, for example, BRAF was
identified as a mutation target only in the Cancer Gene Census dataset, while TGFBR2 was

85

identified as a mutation target only in the KEGG dataset. For mutation targets we calculated the
frequency as the number of mutation targets in a group divided by the total number of genes in
the group.
Statistical Analysis
Analysis of groups of genes or members of a pathway was performed for their coherence
in the normal, adenoma, and carcinoma tissues. We calculated the Pearson’s correlation
coefficient for the expression values of every pair of genes in both datasets. We selected the
same number of random genes from the entire microarray dataset, whether the genes belonged in
the pathway or not, and performed the same calculation for this random group. For the range
between 0-1, at increments of 0.01, we calculated the fraction of pairs with a correlation
coefficient for both the real groups (pathways, subgroups inside pathways) and the random
groups. For each gene, we performed the randomizations 1000 times for the pathways and 100
times for the subgroups within the pathways, and calculated the fraction of random pairs with a
correlation coefficient threshold of 0.5. In addition to the colorectal cancer pathway, we analyzed
the apoptosis pathway because it was previously shown to be coherent in colorectal cancer [20].
We also analyzed the oxidative phosphorylation and proteasome pathways which were shown to
be coherent both in normal and cancer tissues [20]. The oxidative phosphorylation pathway has
more genes than the colorectal cancer and apoptosis pathways, while the proteasome has fewer
genes. Therefore any size effect of the pathway should be accounted for. For each X/Y
dependent subgroup, randomizations of the same size were performed and whether the
subgroups differed significantly from random was determined at a correlation coefficient
threshold of 0.5. For the analysis of expression ratios, we performed t-test on groups of 2 (i.e.,
comparing gene expression profiles of pathway members in group Y=1 to members in the rest of

86

the pathway, namely Y=2, 3, 4). For more than two groups (i.e., comparing gene expression
profiles of pathway members in groups X=1, X=2, and X=3), we used ANOVA.

87

RESULTS AND DISCUSSION
Colorectal cancer pathway subgroups
We analyzed the tumor expression levels of the members of the KEGG colorectal cancer
pathway. We investigated whether the pathway members are differentially expressed with
respect to their X (cellular location), and Y (stage of the tumor) values in the KEGG pathway,
see Materials and Methods for details. The colorectal cancer pathway in KEGG provides some of
the molecular events that underlie the progression of cancer, from normal to carcinoma. We
assessed whether a relationship existed between the progression of colorectal cancer and the
gene expression levels of the members of the colorectal cancer pathway. The normal and the
adenoma samples from both datasets (GSE4183, GSE8671) were combined and the carcinoma
dataset came from GSE4183.
Analyzing the genes that were highly differentiated among the pathway members in the
carcinoma and adenoma datasets, we found AXIN2 and FZD3 genes were the most downregulated in the carcinoma and the most upregulated in the adenoma samples. These genes had
the lowest carcinoma to adenoma expression ratio, whereas PDGFRB and FZD2 were the most
upregulated genes in the carcinoma samples and thus had the highest carcinoma to adenoma
expression ratio (Figure 7). The average expression level of all the genes in the pathway did not
change significantly for the adenoma and carcinoma samples (or stages), which centered around
a ratio of 1. Analyzing the members at particular locations of the pathway suggested that
different genes are differentially expressed and thus possibly differentially regulated, depending
on the stage of the cancer tissues from which the expression data were obtained (Figure 7). This
demonstrated one of the findings that could be obtained with this integrative approach and would
have been lost by analyzing all the pathway members. Analysis of all the pathway members as a

88

whole, suggested no difference between the stages (i.e., ratio = 1) and did not identify a potential
for any genes to be differentially regulated. FZD2 and FZD3 are different homologues of
Frizzled, which is the receptor for Wnt signaling molecules and is known to be important in early
development of colorectal cancer [8]. AXIN2 (Axin) is also a member of Wnt signaling and has
been shown to be a mutation target in the development of colorectal adenoma [8]. Our analysis
suggests that Wnt signaling members, such as different homologues of Frizzled receptor (FZD2,
FZD3), may play a role in early (adenoma) and late (carcinoma) events of colorectal cancer
progression. Wnt signaling is known to be activated during the earlier stages of colorectal cancer
progression and is suggested to be involved in also the later stages of the progression [10].
Platelet-derived growth factor receptor beta (PDGFRB) is known to play a role in advanced and
metastatic stages of colorectal cancer development [23]. It is noteworthy that most genes that
showed differential expression in adenoma and carcinoma samples were receptors, i.e. Frizzled
and PDGFRB. In support of this integrative pathway analysis, a previous study [21] also found
that pathway information, i.e. whether it is a receptor, was important in identifying
physiologically relevant functional groups. Currently, the drugs, Erbitux and Vectibix, used to
treat colorectal cancer, target a receptor, EGFR (RTK). The location of the gene in the colorectal
cancer pathway corresponds to the early cancer stage and the receptor region. The analysis
appears to suggest that receptors could be important regulatory regions in colorectal cancer
development, and as such, other receptors, i.e. FZD2, FZD3 and PDGFRB, could be possible
candidates for drug development. However, more data is needed to confirm this relationship.
Next, we analyzed the possibility of dimensional (X and Y) distinction of the overall
pathway, in other words, whether a group of pathway members defined by a stage (Y) or a
cellular location (X) showed a significant difference in terms of expression levels, and the

89

presence of mutation and drug targets, as compared to the other members of the pathway. We set
the X values to vary from 1 to 3, corresponding to the following locations: receptor/ligand
(denoted as 1), cytosol (denoted as 2), and nucleus (denoted as 3). We examined the significance
of grouping the expression profiles of the pathway members according to their X values. There
was a significant grouping of both the adenoma and carcinoma expression profiles of the
pathway members across the X groups (Table 18). In the adenoma and the carcinoma datasets,
cytosolic members of the colorectal cancer pathway (X=2) had significantly lower gene
expression values (Figure 8A, 8B, Table 19), on the other hand, nuclear members of the
colorectal cancer pathway (X=3) had significantly higher gene expression values relative to other
X groups (Figure 8C, 8D, Table 19).
Similar analysis was performed for the Y values, where the Y values ranged from 1 to 4,
to correspond to the different stages, from normal tissue to early adenoma (denoted as 1), early to
intermediate adenoma (denoted as 2), intermediate to advanced adenoma (denoted as 3), and
advanced adenoma to carcinoma (denoted as 4). There was a significant difference across the Y
groups for the adenoma but not the carcinoma datasets (Table 18), which may be attributed to a
significant difference in the expression values of the pathway members with Y values of 1 as
compared to the other pathway members (Y values of 2-4) in adenoma (Figure 8E, Table 19). In
the adenoma tissue samples, the gene expression values of the pathway members, which play a
role in the normal epithelium to early adenoma stage (Y=1) were expressed significantly higher
than the other pathway members. The key genes that contribute to these results were BIRC5
(Survivin), FZD3, and AXIN2 (Axin), which are highly expressed and are located at Y=1, and
PIK3CG, and PDGFRA, which are lowly expressed in the colorectal adenoma samples and
located at Y=2. This result is in line with our previous observation that a group of genes in a

90

particular pathway could be differentially expressed from the other members in the pathway
depending on the stage of the cancer. In addition to the adenoma tissue samples, we analyzed the
expression values of the colorectal pathway members in the carcinoma tissue samples for
possible distinctive patterns but found no significant grouping of the carcinoma expression with
respect to Y (Figure 8F, Table 18, 19). If we consider that the colorectal cancer pathway includes
the stages from normal epithelium to carcinoma, with several adenoma stages but only a single
stage of carcinoma development, and also only one set of gene expression data from carcinoma
tissues, there is likely insufficient information to distinguish the molecular events in the
carcinoma pathway.
Coherence of the colorectal cancer pathway
The KEGG cancer pathways represent a collective behavior of a group of proteins that
underlies the disease, and as such, are built from proteins that belong to multiple signaling
pathways. A coherence indicator has been defined as the ratio of the number of correlated gene
pairs to the total number of gene pairs in a pathway, which is deemed significant based upon a
statistical measure [20]. Using this indicator, the gene expression levels of the signaling and
metabolic pathway members were shown to be coherent, suggesting that coherence may be an
important measure of functionally related genes [20]. Cancer pathways, due to their very nature
of involving multiple signaling groups, may not be expected to show the same level of
coherence. Therefore, we examined whether the gene expression levels of the KEGG colorectal
cancer pathway members were coherent, as compared to apoptosis, oxidative phosphorylation
and proteasome pathway members, the latter were previously shown to be coherent [20]. The
degree of coherence is determined by plotting the correlation in gene expression of the pathway
members. We analyzed normal colorectal, adenoma and carcinoma samples. In normal colorectal

91

tissue expression data, the oxidative phosphorylation and proteasome pathways show a distinct
positive correlation among their pathway members, the apoptosis pathway show a slightly
negative correlation, whereas the correlation distribution of the colorectal cancer members are
closer to the random distributions and hence uncorrelated (Figure 9A). In addition to the
correlation distribution of the expression levels, we analyzed the cumulative distributions of the
absolute values of the correlation coefficients. Similar results for the cumulative and noncumulative correlation distributions are observed for normal colorectal tissue, i.e. the pathway is
uncorrelated (Figure 9A, 9B). Oxidative phosphorylation and proteasome pathways are coherent
in all 3 groups (normal, adenoma, and carcinoma tissues), while the colorectal cancer pathway is
coherent only in the carcinoma samples (Figure 9B-9D). We compared whether the colorectal
cancer and apoptosis pathways differed significantly from random at a correlation coefficient of
0.5 (Table 20). The colorectal cancer pathway appears to be coherent for colorectal carcinoma
but not for the normal and adenoma datasets. This suggests that a cancer pathway may be
coordinately regulated to achieve a biological function or phenotype, which, in this case, is the
progression of the colorectal tissue to the tumor stage. The apoptosis pathway appears to be
coherent in normal colorectal sample but not in the colorectal adenoma or carcinoma data (Table
20). This is in contrast to a previous report which used different gene expression data and found
the data was coherent in the colorectal tumor but not in the normal samples [20]. In addition to
the entire pathway, we studied the possibility of coherence of dimensional groupings in
colorectal cancer and apoptosis pathways. Our analysis suggests that members within the
pathways could be coherent, for example X=2 in the colorectal cancer pathway could be
coherent in normal and adenoma samples even though the entire pathway may not be coherent
(Table 20, 21). However, more data is required to confirm these analyses.

92

In order to determine which proteins contribute to the coherence of the colorectal cancer
pathway in the carcinoma stage, we obtained pairs of genes with an absolute correlation
coefficient of at least 0.5. There were 683 correlated gene pairs in the carcinoma samples, most
of which were specific to carcinoma (Figure 10A). On the other hand, the normal and adenoma
samples had fewer correlated gene pairs and shared more than half of them with each other. The
pairs of correlated genes specific to normal and adenoma tissues included mostly genes in the
early stage members of the pathway, such as AKT homologues (Gene ID of 207 and 208), DVL
homologues (Gene ID of 1856 and 1857), and FZD homologues (Gene ID of 8321, 8322, 8323,
8324). On the other hand, pairs of correlated genes specific to carcinoma included mostly pairs
of genes from different parts of the pathway, such as TGFBR2 homologue (Gene ID of 91) from
late stage of the pathway, APC (Gene ID of 324) from early stage of the pathway, KRAS (Gene
ID of 3485) and MET (Gene ID of 4233) from mid stage of the pathway. Next, we analyzed an
absolute correlation coefficient of at least 0.8, suggesting these gene pairs are highly correlated,
which are provided in Table 22 (0 indicates absolute correlation below 0.8 and 1 indicates
absolute correlation of at least 0.8). We identified 10 gene pairs which were highly correlated in
both the normal and adenoma but not in carcinoma samples (Figure 10B, Table 22). These pairs,
unlike the carcinoma specific gene pairs, suggest coordination within the members of the Wnt
signaling pathway, such as FZD, DVL and AXIN. In addition, these members are highly
correlated with RAC, which is downstream of RAS oncogene (in the adenoma stage of the
colorectal cancer pathway). The coordination between the members of the Wnt signaling
pathway and RAC is supported by protein level interactions [10]. On the other hand, there were
38 gene pairs in the carcinoma samples, none of which were correlated in the normal or adenoma
samples (Figure 10B, Table 22). These strongly correlated gene pairs may suggest direct or

93

indirect protein-protein or transcriptional interactions specific to the carcinoma stage of the
colorectal cancer pathway. For example, the list included the gene pair TGFBR1 and TGFBR2,
whose protein products are known to interact directly and play a role in advanced stages of
colorectal cancer [12]. The presence of a strong correlation between these two genes suggests
that the interaction of these two receptors may be relevant only in the carcinoma stage, for the
samples we analyzed in this study. Another significantly correlated pair was MYC and MSH2
(Table 22). MYC is a transcriptional regulator of MSH2 [24]. Therefore, MYC driven regulation
of MSH2 may be important in carcinoma but not in the normal or adenoma samples. In addition,
BIRC5 was identified to be highly correlated with various other genes, such as ACVR1B and
EGFR, suggesting that BIRC5 may also affect the regulation of ACVR1B and EGFR. While
some of the strongly correlated gene pairs of the colorectal cancer pathway are supported by the
literature, our analysis suggests there are others that may be correlated but currently are not
supported by the literature. These results could suggest potentially novel direct or indirect
interactions, or common regulations by upstream molecules. Since the analysis identified the
colorectal cancer pathway to be significantly coherent only in the carcinoma samples, further
experimental investigation of these correlations is needed to confirm these novel regulatory
mechanisms for colorectal carcinoma.
Mutation target analysis
In addition to gene expression, the distribution of mutation targets of the colorectal
cancer pathway members was compared with respect to their X and Y values. The pathway
members with high X values had higher frequencies of mutation targets. The nuclear members
(X=3) had the highest number of mutation targets, followed by the cytosolic members (X=2) and
the receptor/ligand members (X=1) (Table 23). Most of the mutation targets were concentrated at

94

the later two stages of the pathway (Y=3, Y=4); covering the progression from intermediate
adenoma to carcinoma, with the highest number of mutation targets found in the last stage, from
late adenoma to carcinoma. In other words, colorectal cancer mutation targets were represented
more in the late tumor stages of the KEGG pathway. This result is supported by the current view
that mutations accumulate and increase as the tumor grows and cancer progresses [8]. Lastly, we
analyzed whether the mutation targets were differentially expressed in the adenoma or carcinoma
tissues as compared to the other pathway members, and found there was no statistically
significant difference between the mutation targets and the other pathway members. This may be
because that the expression and mutation data are sparse. Note that our list of mutations targets is
a collection of several genes which were reported to be mutated in at least a single sample.
Therefore only some of these mutation targets may actually be mutated in the samples from
which the expression data were collected. Furthermore, we do not know whether these mutations
are the cause or the effect of the gene expression level changes. Currently, not enough mutation
and gene expression data are available to perform an extensive mutation analysis with respect to
the X and Y groups.
Mutations alter important residues or domains of proteins that could lead to alteration in
the binding properties of proteins to other proteins or regulatory DNA sequences, thus the levels
of other genes and proteins may change. Previous reports have suggested an association between
mutation events and changes in the gene expression levels [25, 26]. There is no evidence to
suggest a correlation between the mutation of a gene and its own mRNA expression levels,
however, a mutation in the TP53 gene has been shown to be correlated with its protein
expression level in cancer [27]. This current study also did not find a strong relationship between
a mutation target and its gene expression level for the colorectal cancer pathway members.

95

CONCLUSION
In this study, we analyzed expression and mutation data of the KEGG colorectal cancer
pathway members. Previous studies focused predominantly on analyzing gene expression data
for an entire signaling pathway and assessing whether the entire pathway is differentially
expressed in a microarray dataset. Here we demonstrate an integrative analysis that investigates
the distribution of expression values of the pathway members. Our analysis incorporated the
location of the pathway members and found the expression values and the number of mutation
targets varied depending on the cellular location (X) and stage of the cancer (Y) (Figure 9,
Tables 19, 20). Previous studies found signaling and metabolic pathways to be coherent, and we
show that the colorectal cancer pathway is coherent as well, depending on the stage at which the
expression data was obtained. The members of the KEGG colorectal cancer pathway showed
some degree of correlation in the expression profile of the carcinoma data (GSE4183).

This analysis has the potential to help uncover the roles of the different genes in the pathways in
the progression of colorectal cancer. We were able to analyze only the colorectal cancer pathway
in this study. However, we anticipate that as more information on cancer pathways and
expression data for the various cancer stages becomes available, this approach to pathway
analysis could be more widely applicable and may help contribute to our understanding of the
similarities and differences in the progression of cancer in the different tissues. Similarly, this
integrative analysis could be applied to analyze the progression of other diseases or biological
processes.

96

APPENDIX

97

APPENDIX
Table 18 ANOVA p values
Group
X=1, 2, 3
X=1, 2, 3
Y=1, 2, 3, 4
Y=1, 2, 3, 4

Sample set
Adenoma
Carcinoma
Adenoma
Carcinoma

P value
0.0005
0.0058
0.0265
0.6136

Table 19 Pairwise t-test p values
Group 1 Group 2
X=1
X=2
X=3
Y=1
Y=2
Y=3
Y=4

X=2, 3
X= 1, 3
X=1, 2
Y=2, 3, 4
Y=1, 3, 4
Y=1, 2, 4
Y=1, 2, 3

Adenoma Carcinoma
p value
p value
0.9470
0.2051
0.0038
0.0023
0.0002
0.0149
0.0049
0.8826
0.1654
0.3116
0.0882
0.3050
0.7675
0.5209

Table 20 Apoptosis and colorectal cancer pathway coherence at correlation coefficient of 0.5
Sample set

Normal
Adenoma
Carcinoma

Apoptosis
pathway

Colorectal
cancer
pathway
0.021
0.255
0.182
0.244
0.119
0.002

98

Table 21 X/Y group coherence at correlation coefficient of 0.5
Pathway

Groups Normal Adenoma Carcinoma

Colorectal X=1
cancer
X=2
pathway
X=3

0.79

0.42

0.35

0.02

0.02

0.95

0.91

0.72

0.21

Y=1

0.84

0.94

0.19

Y=2

0.21

0.07

0.92

Y=3

0.34

0.07

0.18

Y=4

1.00

1.00

1.00

X=1

0.91

0.61

0.17

X=2

0.00

0.07

0.84

Apoptosis
pathway

99

Table 22 Correlated pairs (with absolute correlation coefficient level of 0.8 as the threshold) in
the colorectal cancer pathway for normal, adenoma and carcinoma samples
Gene
ID 1
91
208
324
332
332
332
332
332
332
332
332
842
2353
2956
2956
3845
3845
3845
4233
4609
4609
4609
5880
6654
6654
6655
7040
7043
7976
8313
8322
8326
83439
83439
83439
130399
130399
130399
1857
1857

Gene
Gene
Name 1
ID 2
ACVR1B
1956
AKT2
4087
APC
5602
BIRC5
91
BIRC5
324
BIRC5
1956
BIRC5
2956
BIRC5
4436
BIRC5
5602
BIRC5
6932
BIRC5
8313
CASP9
6934
FOS
7046
MSH6
1630
MSH6
8313
KRAS
1630
KRAS
2956
KRAS
6934
MET
5604
MYC
332
MYC
4436
MYC
6932
RAC2
5293
SOS1
4436
SOS1
8322
SOS2
324
TGFB1
5159
TGFB3
23533
FZD3
5881
AXIN2
6932
FZD4
4436
FZD9
4087
TCF7L1
332
TCF7L1
4436
TCF7L1
5602
ACVR1C
91
ACVR1C
1956
ACVR1C
5295
DVL3
5159
DVL3
10000

Gene
Normal Adenoma Carcinoma
Name 2
EGFR
0
0
1
SMAD2
0
0
1
MAPK10
0
0
1
ACVR1B
0
0
1
APC
0
0
1
EGFR
0
0
1
MSH6
0
0
1
MSH2
0
0
1
MAPK10
0
0
1
TCF7
0
0
1
AXIN2
0
0
1
TCF7L2
0
0
1
TGFBR1
0
0
1
DCC
0
0
1
AXIN2
0
0
1
DCC
0
0
1
MSH6
0
0
1
TCF7L2
0
0
1
MAP2K1
0
0
1
BIRC5
0
0
1
MSH2
0
0
1
TCF7
0
0
1
PIK3CD
0
0
1
MSH2
0
0
1
FZD4
0
0
1
APC
0
0
1
PDGFRB
0
0
1
PIK3R5
0
0
1
RAC3
0
0
1
TCF7
0
0
1
MSH2
0
0
1
SMAD2
0
0
1
BIRC5
0
0
1
MSH2
0
0
1
MAPK10
0
0
1
ACVR1B
0
0
1
EGFR
0
0
1
PIK3R1
0
0
1
PDGFRB
0
1
0
AKT3
0
1
0
100

Table 22 (cont’d)
5296
5879
5879
5879
5879
5881
7040
8313
8323
207
207
208
208
208
208
1857
1857
7157
7157
8313
8313
8313
8313
8313
8313
1857
1857
7157
8313
8313
8321
8321
8323
8323
8324

PIK3R2
RAC1
RAC1
RAC1
RAC1
RAC3
TGFB1
AXIN2
FZD6
AKT1
AKT1
AKT2
AKT2
AKT2
AKT2
DVL3
DVL3
TP53
TP53
AXIN2
AXIN2
AXIN2
AXIN2
AXIN2
AXIN2
DVL3
DVL3
TP53
AXIN2
AXIN2
FZD1
FZD1
FZD6
FZD6
FZD7

23533
1857
5881
8321
8324
23533
1630
5159
10000
1857
5881
207
1857
5291
5881
5291
8321
8323
8324
207
208
7157
8321
8323
8324
5881
8323
5881
1857
5881
5881
8324
5881
8324
5881

PIK3R5
DVL3
RAC3
FZD1
FZD7
PIK3R5
DCC
PDGFRB
AKT3
DVL3
RAC3
AKT1
DVL4
PIK3CB
RAC3
PIK3CB
FZD1
FZD6
FZD7
AKT1
AKT2
TP53
FZD1
FZD6
FZD7
RAC3
FZD6
RAC3
DVL5
RAC3
RAC3
FZD7
RAC3
FZD7
RAC3

0
0
0
0
0
0
0
0
0
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1

101

1
1
1
1
1
1
1
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
1
1
1
1
1
1
1
1

0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0

Table 23 Mutation frequency values of X/Y groups
Group

X=1
X=2
X=3
Y=1
Y=2
Y=3
Y=4

Average
mutation
frequency
0.0455
0.1702
0.3333
0.0833
0.0789
0.3333
0.5714

102

Figure 7 The ratio of carcinoma expression ratio to the adenoma expression ratio given for the
maximum and minimum two genes and the average

Carcinoma/Adenom expression ratio

3.5
3
2.5
2
1.5
1
0.5
0
PDGFRB

FZD2

Average

103

AXIN2

FZD3

Figure 8 X/Y-dependent analysis of colorectal cancer pathway gene expression levels (A)
Comparison of expression ratio values of X = 2 to other X groups in adenoma. (B) Comparison
of expression ratio values of X = 2 to other X groups in carcinoma. (C) Comparison of
expression ratio values of X = 3 to other X groups in adenoma. (D) Comparison of expression
ratio values of X = 3 to other X groups in carcinoma. (E) Comparison of expression ratio values
of Y = 1 to other Y groups in adenoma. (F) Comparison of expression ratio values of Y = 1 to
other Y groups in carcinoma.
A

104

Figure 8 (cont’d)
B

105

Figure 8 (cont’d)
C

106

Figure 8 (cont’d)
D

107

Figure 8 (cont’d)
E

108

Figure 8 (cont’d)
F

109

Figure 9 Pathway correlation and cumulative fraction distributions (A) Pearson’s correlation
coefficient distributions among members of apoptosis, colorectal cancer, oxidative
phosphorylation, and proteasome pathways shown together with 10 random selections of size 84
for normal colorectal tissue. (B–D) Cumulative absolute value fractions of correlation coefficient
distributions among members of apoptosis, colorectal cancer, oxidative phosphorylation, and
proteasome pathways shown together with 10 random selections of size 84 for (B) normal
colorectal tissue, (C) colorectal adenoma, and (D) colorectal carcinoma.
A

110

Figure 9 (cont’d)
B

111

Figure 9 (cont’d)
C

112

Figure 9 (cont’d)
D

113

Figure 10 Vend diagrams of correlated pairs of genes in colorectal normal, adenoma and
carcinoma samples (A) Correlated pairs with at least 0.5 absolute correlation coefficient. (B)
Correlated pairs with at least 0.8 absolute correlation coefficient.
A

114

Figure 10 (cont’d)
B

115

BIBLIOGRAPHY

116

BIBLIOGRAPHY

1.

Fearon, E.R., and Vogelstein, B., A genetic model for colorectal tumorigenesis. Cell
1990, 61, 759-767

2.

Cardoso, J., Boer, J., Morreau, H., Fodde, R., Expression and genomic profiling of
colorectal cancer. Biochim Biophys Acta. 2007, 1775, 103-37

3.

Edelman, E.J., Guinney, J., Chi, J.T., Febbo, P.G., Mukherjee, S. Modeling Cancer
Progression via Pathway Dependencies. PLoS Comput Biol. 2008, 4, e28

4.

Kanehisa, M., Goto, S., KEGG: Kyoto Encyclopedia of Genes and Genomes. Nucleic
Acids Res. 2000, 28, 27-30

5.

Kanehisa, M., Goto, S., Hattori, M., Aoki-Kinoshita, K.F. et al., From genomics to
chemical genomics: new developments in KEGG. Nucleic Acids Res. 2006, 34, D354357

6.

Kanehisa, M., Araki, M., Goto, S., Hattori, M. et al., KEGG for linking genomes to life
and the environment. Nucleic Acids Res. 2008, 36, D480-D484

7.

Dreesen, O., Brivanlou, A.H., Signaling Pathways in Cancer and Embryonic Stem Cells,
Stem Cell Rev. 2007, 3, 7-17

8.

Grady, W.M., Genomic instability and colon cancer. Cancer Metastasis Rev. 2004, 23,
11-27

9.

Söreide, K., Janssen, E.A., Söiland, H., Körner, H., Baak, J.P., Microsatellite instability
in colorectal cancer. Br J Surg. 2006, 93, 395-406

10.

Behrens, J., The role of the Wnt signalling pathway in colorectal tumorigenesis. Biochem
Soc Trans. 2005, 33, 672-5

11.

Mehlen, P., Fearon, E.R., Role of the dependence receptor DCC in colorectal cancer
pathogenesis. J Clin Oncol. 2004, 22, 3420-8

12.

Roman, C., Saha, D., Beauchamp, R., TGF-beta and colorectal carcinogenesis. Microsc
Res Tech. 2001, 52, 450-7

13.

Subramanian, A., Tamayo, P., Mootha, V.K., Mukherjee, S. et al., Gene set enrichment
analysis: a knowledge-based approach for interpreting genome-wide expression profiles.
Proc Natl Acad Sci USA 2005, 102, 15545-50

117

14.

Li, Z., Srivastava, S., Findlan, R., Chan, C., Using dynamic gene module map analysis to
identify targets that modulate free fatty acid induced cytotoxicity. Biotechnol Prog. 2008,
24, 29-37

15.

Li, Z., Srivastava, S., Yang, X., Mittal, S. et al., A hierarchical approach employing
metabolic and gene expression profiles to identify the pathways that confer cytotoxicity
in HepG2 cells. BMC Syst Biol. 2007, 11, 1-21

16.

Breslin, T., Krogh, M., Peterson, C., Troein, C. Signal transduction pathway profiling of
individual tumor samples. BMC Bioinformatics 2005, 6, 163

17.

Mootha, V.K., Lindgren, C.M., Eriksson, K.F., Subramanian, A. et al., PGC1alpharesponsive genes involved in oxidative phosphorylation are coordinately
downregulated in human diabetes. Nat Genet. 2003, 34, 267-273

18.

Jansen, R., Greenbaum, D., Gerstein, M., Relating whole-genome expression data with
protein-protein interactions. Genome Res. 2002, 12, 37-46

19.

Hannenhalli, S., Levy, S., Transcriptional regulation of protein complexes and biological
pathways. Mamm Genome. 2003, 14, 611-9

20.

Yang, H.H., Hu, Y., Buetow, K.H., Lee, M.P., A computational approach to measuring
coherence of gene expression in pathways. Genomics 2004, 84, 211-7

21.

Draghici, S., Khatri, P., Tarca, A.L., Amin, K. et al., A systems biology approach for
pathway level analysis. Genome Res. 2007, 17, 1537-45

22.

Futreal, P.A., Coin, L., Marshall, M., Down, T. et al., A census of human cancer genes.
Nat Rev Cancer 2004, 4, 177-83

23.

Wehler, T.C., Frerichs, K., Graf, C., Drescher, D. et al., PDGFRalpha/beta expression
correlates with the metastatic behavior of human colorectal cancer: a possible rationale
for a molecular targeting strategy. Oncol Rep. 2008, 19, 697-704

24.

Menssen, A., Hermeking, H., Characterization of the c-MYC-regulated transcriptome by
SAGE: identification and analysis of c-MYC target genes. Proc Natl Acad Sci USA
2002, 99, 6274-9

25.

Whyte, D.B., Holbeck, S.L., Correlation of PIK3Ca mutations with gene expression and
drug sensitivity in NCI-60 cell lines. Biochem Biophys Res Commun. 2006, 340, 469-75

26.

Austinat, M., Dunsch, R., Wittekind, C., Tannapfel, A. et al., Correlation between betacatenin mutations and expression of Wnt-signaling target genes in hepatocellular
carcinoma. Mol Cancer. 2008, 7, 21

118

27.

Martinez-Delgado, B., Robledo, M., Arranz, E., Infantes, F. et al., Correlation between
mutations in p53 gene and protein expression in human lymphomas. Am J Hematol.
1997, 55, 1-8

119

CHAPTER 3
SPECIFIC INTERACTION NETWORK ANALYSIS FOR COLORECTAL CANCER
INTRODUCTION
Previous studies used gene expression levels to derive condition specific networks from
large scale protein-protein interaction (PPI) datasets [1-4]. The limitations of having static PPI
datasets were somewhat mitigated with the integration of sample-specific gene expression level
changes so that condition specific/dynamic, and possibly novel regulatory interactions can be
obtained. Differential gene expression analysis was used for groups of genes in a close
neighborhood in the human PPI network to derive modules which were shown to increase the
classification accuracy of cancer [1, 3]. Biologically relevant and significant condition specific
molecular mechanisms have been revealed in such studies. Integrative analyses of PPI data have
mostly been based on modular analysis in which the human PPI network is assumed to have a
modular structure and the condition specific modules were retrieved by estimating the activity of
the modules from the gene expression alterations of the interacting pairs. With this approach,
modules representing deregulated events, such as cell cycle, apoptosis, etc. were revealed [1, 3,
4].
Integrative analyses of PPI data use microarray datasets to provide the condition specific
network information. PPI datasets provide the pairs of proteins that are likely to interact
physically. Differentially expressed genes obtained from microarray datasets in an experimental
or disease condition can provide additional support for the interaction between the genes at the
protein level. These pairs of genes are specific to the conditions in the microarray dataset unlike
the PPI dataset pairs. Therefore the specific gene pairs could play a role under one condition
120

specifically. Indeed, with the integration of condition specific information, the PPI networks
become more suitable for predicting novel regulatory interactions and providing novel
hypotheses for the experimental condition under investigation. Previous integrative studies using
PPI data were limited to comparisons of gene expression levels between two conditions such as
comparing tumor to normal samples, long survival to poor survival samples, or metastatic to
non-metastatic samples [1, 3, 4]. However, the analysis of a multi-conditional large-scale gene
expression dataset also provides useful information, such as identifying genes with switch-like
behavior that were not easily uncovered with the analysis of a pair-wise dataset [5]. Analyzing
the expression levels of a gene in a diverse condition space provides a better understanding of the
specificity of the expression level for a particular condition. For example p53, a well-known
gene mutated commonly in many cancer types, does not appear to have a very significant
differential gene expression in a small dataset that compares only two conditions, but was
identified in the large dataset comprising of multiple conditions [5].
Colorectal cancer is one of the leading cancer types based on the number of new cases
and number of expected deaths [6]. Colorectal cancer shares many mutation targets with other
cancers such as endometrial cancer and ovarian cancer, but very few drug targets with other
cancer types [7]. The insufficient number of drug therapies suggests that further studies are
necessary for colorectal cancer to find novel targets. Although certain molecular mechanisms
such as the involvement of TGF-beta pathway in colorectal cancer is known [8]; there is a need
for more and specific molecular mechanisms to be elucidated for colorectal cancer.
RNF43 is an E3 ubiquitin ligase that resides in the ER and nuclear membranes, however
no ubiquitination target for RNF43 has been identified and thus far only auto-ubiquitination
activity has been shown [9, 10]. RNF43 was found to be upregulated in colorectal cancer and is
121

also a tumor associated antigen marker for colorectal cancer [11]. There are peptide vaccine
clinical trials of RNF43 for colorectal cancer [12, 13]. GR is a nuclear receptor for
glucocorticoid hormones, primarily involved in maintaining homeostasis in response to stress.
However, it has diverse roles in various cell types and under different conditions. For example
GR can induce the killing of lymphocytes, and promote proliferation of endometrial and liver
cells, etc. [14]. Upon binding of glucocorticoids, GR is phosphorylated (activated) and
transported into the nucleus where it can 1) directly bind to the promoters of its targets genes, or
2) modulate the activity of other transcription factors [14]. GR expression levels and function
vary among different colorectal cancer patient samples and different cell lines. GR was found to
be epigenetically downregulated by promoter hypermethylation or absent at the protein level in
some patient tissue derived colorectal cancer samples as well as in certain cell lines [15-19].
Some studies have suggested an apoptotic role for glucocorticoids, and have associated it, in
part, to GR activity. In those cases, GR enhanced the activity of drugs like genistein. GR
expression has been shown to correlate with pRB and p16 tumor suppressor protein expression
levels [18, 20].
In our study, we used a multi-condition dataset to define the top ranked significantly
differentially expressed genes in colorectal cancer, which were found significantly associated
with colorectal cancer in the literature. We constructed a network of these genes using the human
protein-protein interaction data. Our analysis was not restricted to a modular structure, thus any
gene could be linked to any other gene; thereby increasing the number of potential gene pairs.
The ring finger protein 43 (RNF43), which is upregulated in colorectal cancer, is linked to the
glucocorticoid receptor (GR, NR3C1), which is downregulated. In the HCT116 colorectal cancer

122

cell line, modulation of GR and RNF43 levels by siRNA and glucocorticoid treatment supported
the link between these two proteins.

RESULTS AND DISCUSSION
Colorectal cancer specific genes
Unlike previous studies that defined specific genes based on the gene expression level
changes by comparing tumor to normal samples, long survival to poor survival samples, or
metastatic to non-metastatic samples, etc. [1, 3, 4], we defined the specific genes by differential
expression of genes in colorectal cancer samples compared to a large set of samples from diverse
conditions, including normal samples from various tissues, different cancer samples, cell lines, as
well as other disease samples, etc. In other words, in prior approaches the comparison is made
between condition A and an “other” condition (i.e. condition B). In contrast, we compare a
condition against many conditions (i.e. condition A against conditions B,C,D, where B,C,D all
fall under the “other condition”) to identify more specific targets.
We calculated the separation of the expression values by D value, which is a normalized
absolute difference of the mean values between two populations [5]. We compared the lists of
differentially expressed gene based on a comparison of the colon cancer samples with the normal
samples vs. all samples minus the colorectal cancer samples, for their enrichment of colorectal
cancer related publications in the literature. We searched NCBI Pubmed database for the Official
Full Name together with ‘Colorectal cancer’ or ‘Colon cancer’ and counted the number of genes
with at least one publication. We also performed the same calculation for a random set of genes
with the same number of gene expression profiles. Comparison to all the samples always yielded

123

more genes and more significant genes that are related to colorectal cancer than the comparison
to only normal samples (Figure 11). The greater fluctuation in the fraction of relevant genes for
the multiple comparison approach is because of the smaller size of the list.
As D has to be at least 2 for two distinct populations, we used it as a threshold to obtain a
list of colorectal cancer specific genes which has distinct expression profiles in colorectal cancer
samples compared to a large spectrum of samples and which were significantly associated with
colorectal cancer in the literature. In this way, we collected a list of specific genes for colorectal
cancer which have significantly distinct expression profile from the other samples of various
conditions.
Colorectal cancer specific network
We aimed to find the most significant regulatory interactions or associations between
different genes in colorectal cancer such that these genes are distinctly expressed in colorectal
cancer samples as compared to not only non-cancer colorectal samples but also all other cancer
and disease samples of various types and conditions. Therefore our objective was to obtain novel
molecular mechanisms specific to colorectal cancer, which is not readily revealed with previous
two-condition or pair-wise comparisons. We used colorectal cancer specific genes to construct a
colorectal cancer specific network so that we could generate hypotheses on potentially novel
molecular mechanisms of colorectal cancer. We used the direct and indirect PPI neighborhood to
construct a colorectal cancer specific network (Figure 12). We defined the resulting network as
the colorectal cancer specific network (Figure 12). The colorectal cancer specific network
includes genes that are well associated with colorectal cancer in the literature. In this network,
only NR3C1 has a D value greater than 2, when the pairwise approach is used. This shows that

124

this network would not be obtained by a pairwise differential expression approach. The network
also provides potential regulatory mechanisms for colorectal cancer. For example, a previously
identified interaction between the HOXB7 protein and MAPK pathway [21], was identified in
the network.
Moreover, we aimed to confirm the distinct expression levels of the members of the
colorectal cancer specific network in an independent expression dataset. We used a patient
derived paired normal and adjacent tissue sample set to calculate the D values. In the colorectal
cancer specific network, only NR3C1 (GR) and RNF43 have D values greater than 2 in this
independent dataset.
GR-RNF43 regulation in colorectal cancer
NR3C1 (glucocorticoid receptor, GR) is the only gene in the colorectal cancer specific
network that also had a high D value when the colorectal cancer samples were compared
pairwise to normal samples. While NR3C1 was downregulated, one of its neighbors in the
network, RNF43 was upregulated.
To validate a possible regulatory mechanism between these two proteins, we knocked
down GR and RNF43 levels by specific siRNAs in HCT116 colorectal cancer cell lines. While
knock-down of GR increased RNF43 mRNA levels, knock-down of RNF43 levels did not
change the GR mRNA levels (Figure 13). This result suggests a possible transcriptional
regulation of RNF43 by GR, therefore we induced GR with dexamethasone and measured the
RNF43 mRNA levels. The induction of GR by dexamethasone led to decreased RNF43 mRNA
levels that is abolished when GR is simultaneously silenced by siRNA treatment (Figure 13),
suggesting that the effect of dexamethasone is specific to GR. This result suggests that GR is

125

transcriptional repressor of the RNF43 gene. Furthermore, GR has a binding site at 138 kb
upstream of RNF43 transcription start site in lung carcinoma cells (FDR=5.9%) [22]. The
position of this binding site is consistent with the results of other genes that are negatively
regulated by GR, where GR binding is at distant sites (with a median of 146kb in contrast to the
median of 11kb for positively regulated genes) from the transcription start site on the promoter
of these genes. When GR was silenced, RNF43 protein levels were also upregulated (Figure 14).
RNF43 protein levels were also downregulated in response to dexamethasone treatment in
HCT116 colorectal cancer cells (Figure 14). These results confirm the negative regulation of
RNF43 by GR. On the other hand, GR protein levels, unlike the mRNA levels, were
downregulated in response to RNF43 silencing (Figure 14). Therefore there could be a positive
regulation of GR by RNF43, which might be mediated by the ubiquitin ligase activity of RNF43
towards a negative regulator of GR.
CONCLUSION
In contrast to current approaches that identify targets by comparing pair-wise control (i.e.
normal) to treated (i.e. disease) or across samples, we compared colorectal cancer samples not
only to a control sample set but against a wide variety of samples and conditions. We were able
to identify more specific genes for colorectal cancer which are significantly associated with
colorectal cancer in the literature. We constructed a colorectal cancer-specific network based on
the expression levels of the neighboring genes obtained from the human protein-protein
interaction network. We identified a potential negative relationship between glucocorticoid
receptor (GR) and ring finger protein 43 (RNF43) which may play a role in colorectal cancer. In
HCT116 colorectal cancer cell line, knocking-down GRα levels with siRNA resulted in
increased RNF43 levels and inducing the colorectal cancer cells with dexamethasone, which is
126

an activating ligand for GR, resulted in decreased RNF43 levels. On the other hand, knockingdown RNF43 levels with siRNA resulted in decreased GRα levels. Our study suggests GR
negatively regulates RNF43 whereas there is no such a negative regulation from RNF43 to GR,
indeed there could be a positive regulation.

MATERIALS AND METHODS
Transcriptome and Interactome data
We collected the expression data from an integrated multi-condition human microarray
dataset (E-TABM-185) [23] from the ArrayExpress (http://www.ebi.ac.uk/arrayexpress/) and
integrated GSE24514 [24] from NCBI GEO database (http://www.ncbi.nlm.nih.gov/geo/) as
there were not many colorectal tissue samples. After combining the raw data, we re-normalized
the dataset by GCRMA [25]. We also used another expression data of a set of colorectal cancer
patient-derived paired normal and tumor samples (GSE18105; GSM452629- GSM452662) [26]
from NCBI GEO database. We used the human protein-protein interaction data from I2D
database (Version 1.9) (http://ophid.utoronto.ca/ophidv2.201/), which includes experimentally
confirmed as well as computationally predicted interactions [27, 28].
Determining the colorectal cancer specific gene list
We used the separation value (D value) to quantify the distinction of the expression
values between sample sets similar to [5]. We calculated the D value between the colorectal
cancer samples and the non-cancer colorectal samples as well as all the samples except the
colorectal cancer samples. We tested the significance of the D values by permutation test. We
also calculated Mann-Whitney test p-values. We used each affymetrix ID individually when
127

calculating the D values and Mann-Whitney test p-values, which corresponded to different gene
expression profiles. We used the Entrez Gene IDs matching to the affymetrix IDs to access the
Official Full Name of the gene. In order to obtain the colorectal cancer associated genes, we used
this full name to search Pubmed (http://www.ncbi.nlm.nih.gov/pubmed/) for this gene together
with ‘Colorectal cancer’ or ‘Colon cancer’ using Bipython [29]. We collected the fraction of the
number of genes (each distinct Gene ID) that was found to have at least one publication when
searched together with ‘Colorectal cancer’ or ‘Colon cancer’ in Pubmed and divided by the
number of genes (the number of distinct Gene ID s). We also obtained a random set of genes
with the same number of distinct gene expression profiles (affymetrix IDs) and performed the
same calculation for the random gene lists to obtain a random distribution for the colorectal
cancer associated genes.
We used indirect interaction neighbors to construct the network, since it has been shown
that they have significant functional similarity [30]. We integrated the colorectal cancer specific
genes, which have a D value of at least two when colorectal cancer samples were compared to a
multiple condition set of samples, with the first and second degree neighborhood in the parent
network which is the human protein-protein interaction network from I2D database.
Cell culture
HCT116 colorectal carcinoma cell line (ATCC, Manassas, VA, USA) was cultured in
McCoy’s media (ATCC, Manassas, VA, USA) with 10% fetal bovine serum and 1%
penicillin/streptomycin and maintained at 37oC and 5% CO2.

128

Quantitative real-time polymerase chain reaction
mRNA from HCT116 cells were isolated with the RNA isolation kit (Qiagen, Valencia,
GA, USA). cDNA synthesis from mRNA was done with the cDNA synthesis kit (Bio-Rad,
Hercules, CA, USA). For specific mRNA quantification by PCR, based on SYBR Green
detection (Bio-Rad, Hercules, CA, USA) the following primers (Operon, Huntsville, AL, USA)
were used; GR-alpha forward (5’-TCAGTTCCTAAGGACGGTCTG-3’), GR-alpha reverse (5’CCACTTCATGCATAGAATCCAA-3’), RNF43 forward (5’CAAATTCACAGCCAGTGTGG-3’), RNF43 reverse (5’- GCTCCTCGAGTTCCTCCTCT-3’).
The PCR cycle threshold values were based on the MyIQ software.
Western Blot
HCT116 cells were washed twice with cold PBS and lysed in RIPA lysis buffer with
protease inhibitor. The cell lysate was centrifuged at 8000 rcf for 10 min, and the supernatant
was collected. Total protein levels were quantified by BCA assay kit from Pierce Inc. (Rockford,
IL, USA). A total of 20-40 g of total protein was resolved by SDS-PAGE and transferred to
nitrocellulose membranes, and probed with primary antibodies with RNF43 antibody (ab84125
from Abcam), GRα (ab3580 from Abcam), and anti-beta-actin (Sigma-Aldrich). Biotinylated
protein ladders (Cell Signaling, Beverly, MA, USA) were used together with the samples antibiotin antibody was used to detect the protein ladders on the western blots. Primary antibody
incubation was done at 4 °C for overnight, secondary anti-rabbit and anti-mouse antibody (Pierce
Biotechnology Inc) incubation was done for 1 hour at room temperature. Antibody detection was
performed using the enhanced chemiluminescence kit from Pierce Biotechnology and imaged on
the Molecular Imager ChemiDoc XRS System from Bio-Rad.

129

Dexamethasone and siRNA treatment
Dexamethasone (Sigma-Aldrich, St.Louis, MO, USA) was dissolved in ethanol and used
as 250nM in HCT116 cell culture. Validated siRNA s for GR-alpha and RNF43 (Applied
Biosystems, Carlsbas, CA, USA) and a scrambled negative control siRNA was transfected by
reverse-transfection using Lipofactemine RNAiMAX (Invitrogen, Grand Island, NY, USA) for
48-72 hours.

130

APPENDIX

131

APPENDIX
Figure 11 Literature comparison of differentially expressed genes in colorectal cancer samples
with respect to only non-cancer colorectal samples (pairwise) vs. all other samples (multiple) (A)
Fraction of genes that are relevant to colorectal cancer in Pubmed database when different D
value cut-offs around 2 are chosen. (B) Significance of the genes that are relevant to colorectal
cancer in Pubmed database when different D value cut-offs around 2 are chosen, based on
permutation test. For interpretation of the references to color in this and all other figures, the
reader is referred to the electronic version of this dissertation.
A

132

Figure 11 (cont’d)
B

133

Figure 12 Construction of the colorectal cancer specific network (A) Network construction
method based on both direct and indirect protein-protein interactions, which are integrated with
differentially expressed genes with D>2. (B) Colorectal cancer specific network

134

Figure 12 (cont’d)
B

135

Figure 13 Modulation of mRNA levels of RNF43 and GRα (A) All values were first normalized
to the beta-actin levels. Specific siRNA treatment of RNF43 and GR was done in HCT116
colorectal cancer cell lines and normalized to the scrambled control siRNA treatment (B)
HCT116 cells treated with 250nM dexamethasone, dissolved in EtOH, were compared with
control cells treated with EtOH (less than 0.1%) treated cells. All values were first normalized to
the beta-actin levels. Dexamethasone and GR siRNA treated HCT116 cells were normalized
with EtOH and negative control siRNA treated cells. Real-time PCR results of primers specific
for RNF43 and GRα are normalized with the results from β-actin primers. Fold change values
are based on 9 values representing the comparison of all of the 3 treatment replicates to 3 control
replicates. * indicates p ≤0.06 based on the comparison of expression values of the treatment
replicates to control replicates.
A

136

Figure 13 (cont’d)
B

*

137

Figure 14 Modulation of protein levels of RNF43 and GRα (A) 250nM dexamethasone,
dissolved in EtOH, treated HCT116 cells were compared with EtOH (less than 0.1%) treated
cells. Dexamethasone and GR siRNA treated HCT116 cells were compared with EtOH and
negative control siRNA treated cells. Western blot with RNF43 antibody (ab84125 from Abcam)
(B) Specific siRNA treatment of RNF43 and GR was done in HCT116 colorectal cancer cell
lines and compared to the negative control siRNA treatment. (C) Quantitated value for the
RNF43 and GR western blot images in siRNA experiments by ImageJ (D) Quantitated value for
the RNF43 and GR western blot images in Dexamethasone experiments by ImageJ. siRNA
treatments are normalized with negative control siRNA treatment and dexamethasone treatments
are normalized with EtOH treatments.
A

B

138

Figure 14 (cont’d)
C

139

Figure 14 (cont’d)
D

140

BIBLIOGRAPHY

141

BIBLIOGRAPHY

1.

Ideker T, Ozier O, Schwikowski B, Siegel AF. Discovering regulatory and signalling
circuits in molecular interaction networks. Bioinformatics. 2002, 18 Suppl 1:S233-40

2.

de Lichtenberg U, Jensen LJ, Brunak S, Bork P. Dynamic complex formation during the
yeast cell cycle. Dynamic complex formation during the yeast cell cycle. Science. 2005,
307(5710):724-7

3.

Chuang HY, Lee E, Liu YT, Lee D, Ideker T. Network-based classification of breast
cancer metastasis. Mol Syst Biol. 2007, 3:140

4.

Ahn J, Yoon Y, Park C, Shin E, Park S. Integrative gene network construction for
predicting a set of complementary prostate cancer genes. Bioinformatics. 2011,
27(13):1846-53

5.

Wu M, Liu L, Chan C. Identification of novel targets for breast cancer by exploring gene
switches on a genome scale. BMC Genomics. 2011, 12:547

6.

American Cancer Society. (2012) Cancer Facts & Figures. Atlanta: American Cancer
Society.

7.

Dalkic E, Wang X, Wright N, Chan C. Cancer-drug associations: a complex system.
PLoS One. 2010, 5(4):e10031

8.

Lampropoulos P, Zizi-Sermpetzoglou A, Rizos S, Kostakis A, Nikiteas N, Papavassiliou
AG. TGF-beta signalling in colon carcinogenesis. Cancer Lett. 2012, 314(1):1-7

9.

Yagyu R, Furukawa Y, Lin YM, Shimokawa T, Yamamura T, Nakamura Y. A novel
oncoprotein RNF43 functions in an autocrine manner in colorectal cancer. Int J Oncol.
2004, 25(5):1343-8

10.

Sugiura T, Yamaguchi A, Miyamoto K. A cancer-associated RING finger protein,
RNF43, is a ubiquitin ligase that interacts with a nuclear protein, HAP95. Exp Cell Res.
2008, 314(7):1519-28

11.

Uchida N, Tsunoda T, Wada S, Furukawa Y, Nakamura Y, Tahara H. Ring finger protein
43 as a new target for cancer immunotherapy. Clin Cancer Res. 2004, 10(24):8577-86

12.

Yoshimatsu K, Yokomizo H, Osawa G, Fujimoto T, Otani T, Tsunoda T, Nakamura Y,
Ogawa K. Phase I study of combination therapy with peptide vaccine and anti-cancer
drug for colorectal cancer]. Gan To Kagaku Ryoho. 2008, 35(12):2268-70

142

13.

Yasuda S, Tsuchiya I, Okada K, Tanaka A, Suzuki T, Sadahiro S, Takeda K, Yamamoto
S, Nakui M. Significant clinical response of advanced colon cancer to Peptide vaccine
therapy: a case report. Tokai J Exp Clin Med. 2012, 37(2):57-61

14.

Oakley RH, Cidlowski JA. Cellular processing of the glucocorticoid receptor gene and
protein: new mechanisms for generating tissue-specific actions of glucocorticoids. J Biol
Chem. 2011, 286(5):3177-84

15.

Lind GE, Kleivi K, Meling GI, Teixeira MR, Thiis-Evensen E, Rognum TO, Lothe RA.
ADAMTS1, CRABP1, and NR3C1 identified as epigenetically deregulated genes in
colorectal tumorigenesis. Cell Oncol. 2006, 28(5-6):259-72

16.

Ahlquist T, Lind GE, Costa VL, Meling GI, Vatn M, Hoff GS, Rognum TO, Skotheim
RI, Thiis-Evensen E, Lothe RA. Gene methylation profiles of normal mucosa, and benign
and malignant colorectal tumors identify early onset markers. Mol Cancer. 2008, 7:94

17.

Lien HC, Lu YS, Shun CT, Yao YT, Chang WC, Cheng AL. Differential expression of
glucocorticoid receptor in carcinomas of the human digestive system. Histopathology.
2008, 52(3):314-24

18.

Theocharis S, Kouraklis G, Margeli A, Agapitos E, Ninos S, Karatzas G, Koutselinis A.
Glucocorticoid receptor (GR) immunohistochemical expression is correlated with cell
cycle-related molecules in human colon cancer. Dig Dis Sci. 2003, 48(9):1745-50

19.

Alford TC, Do HM, Geelhoed GW, Tsangaris NT, Lippman ME. Steroid hormone
receptors in human colon cancers. Cancer. 1979, 43(3):980-4

20.

Park JH, Oh EJ, Choi YH, Kang CD, Kang HS, Kim DK, Kang KI, Yoo MA. Synergistic
effects of dexamethasone and genistein on the expression of Cdk inhibitor
p21WAF1/CIP1 in human hepatocellular and colorectal carcinoma cells. Int J Oncol.
2001, 18(5):997-1002

21.

Liao WT, Jiang D, Yuan J, Cui YM, Shi XW, Chen CM, Bian XW, Deng YJ, Ding YQ.
HOXB7 as a prognostic factor and mediator of colorectal cancer progression. Clin
Cancer Res. 2011, 17(11):3569-78

22.

Reddy TE, Pauli F, Sprouse RO, Neff NF, Newberry KM, Garabedian MJ, Myers RM.
Genomic determination of the glucocorticoid response reveals unexpected mechanisms of
gene regulation. Genome Res. 2009, 19(12):2163-71

23.

Parkinson H, Kapushesky M, Kolesnikov N, Rustici G, Shojatalab M, Abeygunawardena
N, Berube H, Dylag M, Emam I, Farne A, Holloway E, Lukk M, Malone J, Mani R,
Pilicheva E, Rayner TF, Rezwan F, Sharma A, Williams E, Bradley XZ, Adamusiak T,
Brandizi M, Burdett T, Coulson R, Krestyaninova M, Kurnosov P, Maguire E, Neogi SG,
Rocca-Serra P, Sansone SA, Sklyar N, Zhao M, Sarkans U, Brazma A. ArrayExpress

143

update--from an archive of functional genomics experiments to the atlas of gene
expression. Nucleic Acids Res. 2009, 37(Database issue):D868-72
24.

Alhopuro P, Sammalkorpi H, Niittymäki I, Biström M, Raitila A, Saharinen J,
Nousiainen K, Lehtonen HJ, Heliövaara E, Puhakka J, Tuupanen S, Sousa S, Seruca R,
Ferreira AM, Hofstra RM, Mecklin JP, Järvinen H, Ristimäki A, Orntoft TF, Hautaniemi
S, Arango D, Karhu A, Aaltonen LA. Candidate driver genes in microsatellite-unstable
colorectal cancer. Int J Cancer. 2012, 130(7):1558-66

25.

Wu Z, Irizarry RA, Gentleman R, Martinez-Murillo F, Spencer F. A Model-Based
Background Adjustment for Oligonucleotide Expression Arrays. Journal of the American
Statistical Association 2004, 99(468): 909-917

26.

Matsuyama T, Ishikawa T, Mogushi K, Yoshida T, Iida S, Uetake H, Mizushima H,
Tanaka H, Sugihara K. MUC12 mRNA expression is an independent marker of prognosis
in stage II and stage III colorectal cancer. Int J Cancer. 2010, 127(10):2292-9

27.

Brown KR, Jurisica I. Online predicted human interaction database. Bioinformatics.
2005, 21(9):2076-82

28.

Brown KR, Jurisica I. Unequal evolutionary conservation of human protein interactions
in interologous networks. Genome Biol. 2007, 8(5):R95

29.

Cock PJ, Antao T, Chang JT, Chapman BA, Cox CJ, Dalke A, Friedberg I, Hamelryck T,
Kauff F, Wilczynski B, de Hoon MJ. Biopython: freely available Python tools for
computational molecular biology and bioinformatics. Bioinformatics. 2009, 25(11):14223

30.

Chua HN, Sung WK, Wong L. ing indirect neighbours and topological weight to predict
protein function from protein-protein interactions. Bioinformatics. 2006, 22(13):1623-30

144

CONCLUSION
Systems level analysis of biomedical relationships, systems medicine or systems biology,
not only interactions between genes, proteins, metabolites but also diseases, drugs, drug targets,
and mutation targets, is necessary for understanding the global characteristics of diseases such as
cancer [1, 2, 3]. For example, the structure of the PPI networks is a specific structure which
contains only a few hub proteins that connect a lot of proteins to each other; therefore, a random
dysfunction of a member in the network is very unlikely to hit one of these very few hub
proteins, thus making the network robust against random problems. This finding can be used in a
perturbation analysis; for instance, proteins with different number of interactions could be
targeted separately to examine the differences, i.e., hub proteins could be more important for the
survival of the organism, or cancer mutation targets are more connected from the rest of the
proteins in a human PPI network [4, 5]. Therefore, systems medicine and systems biology
approaches are useful for obtaining important characteristics of the organization of the
biomedical system which could be exploited for advancing our understanding of cancer and its
therapy. As cancer is a major cause of death in USA and colorectal cancer is one of the leading
types of cancer, we aimed to identify common or distinct network features of colorectal cancer
together with the other cancer types such as the analysis of clinical data associations, molecular
signaling pathways of cancers, and specific interaction networks.
Previously, global network analysis was done on disease-gene, drug-target, disease-drug
associations, however, driving factors of the networks or topological properties of the networks
was not done extensively. For example, some disease phenotypes were shown to be highly
connected in a disease network study, but, whether the high connectivity is related to the high
number of cases or deaths for those diseases was not investigated [6].

145

Firstly, we collected the cancer-drug associations to generate cancer networks and
analyzed the correlation of the network properties with cancer death statistics as a likely factor
affecting cancer drugs. The cancers with the highest number of FDA approvals and clinical trials
are leukemia, lung cancer, lymphoma and breast cancer; however, only breast and lung cancers
have high and significant weighted degree values in the FDA cancer network, which is based on
the cancer-drug associations. Leukemia and lymphoma have high number of drug approvals and
trials but they don’t share most of these drugs with other cancers unlike breast and lung cancers.
This implies that the drug therapy for leukemia and lymphoma are more specialized for their
own cases. Furthermore, cancer drug approvals and clinical drug trials are correlated to each
other; a cancer type with a high number of drug approvals is more likely to have a high number
of clinical drug trials. However, FDA cancer network, based on the sharing of drugs in FDA
approvals, is very different from the clinical trial cancer network, based on the allocation of
drugs in clinical trials. Moreover, lung cancer, which is significantly connected in the FDA
cancer network, is not significantly connected in the clinical trial cancer network, whereas,
lymphoma and ovarian cancer, which are not significantly connected in the FDA cancer network,
are significantly connected in the clinical trial cancer network. As a result, there is a significant
difference between the FDA approval based drug sharing and the clinical trial based drug
sharing. This raises questions regarding the cancer drugs that need to be addressed, such as, if
there is a bias in favor of lung cancer in FDA drug approvals since it shares many drugs in FDA
approvals but not in clinical trials, also what the reason could be for less drug sharing for ovarian
and lymphoma in FDA approvals in contrast to clinical trials.
In order to understand some of these differences, we compared the edge weights for the
FDA and clinical trial cancer networks for each cancer pair and found out that most pairs are

146

strongly connected in the clinical trial but not in the FDA cancer network. As a specific example,
stomach and esophagus cancers share many clinical drug trials such as capecitabine, cisplatin,
doxorubicin, etc. but not any FDA drug approvals. There could be various reasons for this; i.e.,
the clinical trials for these cancers may not be successful. Our data includes any kind of clinical
drug trial which is completed. It should be investigated further if the FDA approved drugs of
these two cancer types could be used for each other. This analysis provides guidance for future
FDA approvals and clinical decision making. When a decision is going to be made for a certain
drug to be approved for a certain cancer type, the neighborhood of that cancer type in the clinical
trial cancer network could be searched for the FDA approved drugs of the neighbor cancer types.
For example, the approved drugs of stomach cancer; docetaxel, fluorouracil, imatinib, and
sunitinib, could be given priority for future drug approvals for esophagus cancer.
Unquestionably, there will be other factors such as the suitability of the deliverance of these
drugs for esophagus cancer, but this information could be used as an additional factor in clinical
decision making.
We also constructed FDA cancer networks for different years and observed that while
breast cancer is significant since earlier years, lung cancer is significantly connected only in later
years. In addition, these two cancers are significantly more connected than the rest of the
network only recently. This shows that clinical decisions for cancer drugs in different years can
change the structure of this network. There are more cancer drug approvals for lung cancer in
later years. This observation shows a bias in favor of lung cancer that needs to be monitored in
the following years.
We tested the correlation of cancer death statistics, global and local lethality values, with
the network degree values and compared them with the FDA approval or clinical trial numbers.

147

This showed us that while there is a significant correlation between the FDA cancer network
degree values and the local lethality values, there is not a significant correlation between the
FDA approval numbers and the local lethality values. On the other hand, while the FDA
approval numbers are significantly correlated with the global lethality values, the FDA cancer
network degree values are not correlated. This implies that global lethality of cancer could affect
the number of FDA approved drugs, but not the FDA approved drug sharing. On the other hand,
local lethality might influence the FDA approval drug sharing, but not the number of FDA
approved drugs. There could be similar effect for the clinical trial drugs also.
Interestingly, we also observed exceptions for certain groups of cancer types; the reasons
for which should be investigated further. For example, the potential effect of local lethality on
the sharing of FDA approved drugs is not present for the most locally lethal cancers, pancreatic,
liver and esophagus cancers. These cancers have very low overlap of FDA approved drugs with
other cancers. While liver and lung cancers have only one common FDA approved drug, they
have 13 common drugs out of total 32 drugs used in clinical trials for both. 5 of these clinical
trial drugs are approved for lung cancer by FDA, but they are still in clinical trials for liver
cancer. In the future, these drugs might be expected to be given priority for liver cancer.
We also showed the unlikeness of the drug target based relationships of cancer types with
the mutation target based relationships. While lung and breast cancers are significantly
connected with respect to drug targets, they are not significantly connected based on mutation
targets. This might indicate that their relatedness to the other cancer is not very high based on
molecular mechanisms but they are highly related to other cancers in drug approvals. On the
other hand, colorectal, ovarian and brain cancers could be highly related to the other cancers in
drug approvals but not in molecular mechanisms based on their drug target based and mutation

148

target based degree values. These results might guide the future drug trials and approvals of
colorectal, ovarian, and brain cancers to look for overlaps between these 3 cancer types and other
cancer types. This is based on the assumption that the mutation target information is complete
and represents the molecular events underlying cancer. It should be noted that there could be
other mutation events, as well as other molecular alterations such as the transcriptional regulation
of genes, protein modifications, etc. On the other side, we only analyzed the clinical drugs for
cancer types, which is also not complete as there are many other drugs in clinical trials which are
not in our list. In addition, there can be other drugs used in patients than the list of FDA approval
drugs. In this study, we only analyzed the drug therapy aspect of cancer therapy excluding
radiotherapy, surgery, etc. There may be some cancer types, for which chemotherapy might not
be as important as other cancer types which can impact some of our conclusions.
In our next study, we showed that cancer pathways which consist of various signaling
pathways can be analyzed in combination with gene expression levels to investigate the
coherence of the pathway. Coherence of a pathway is defined as having close expression levels
among the members of a pathway [7]. While previous research has focused on mostly metabolic,
cellular, signaling pathways for coherence, the analysis of cancer pathways was ignored. We
showed that the KEGG colorectal cancer pathway is coherent in only the carcinoma stage but not
the normal or adenoma stages. This implies the collective regulation of the colorectal cancer
pathway genes in carcinoma. These genes belong to several different signaling pathways such as
Wnt signaling and TGF-beta pathways, which are well-known to be involved in colorectal cancer
[8, 9]. Therefore our study implies enhanced communication between these pathways in later
stages. Networks of pathways or modules can be analyzed to have global descriptions underlying
biological processes and diseases like cancer [10, 11, 12]. For example, analysis of a pathway

149

crosstalk network for gene expression changes in colorectal cancer metastasis revealed a module
of cell cycle related pathways and a modules cell migration related pathways [11]. Therefore, the
potential interaction between Wnt signaling, MAPK, TGF-beta pathways should be investigated
further. They can either have common regulators or they could co-regulate each other. We found
more support for the latter case. A positive crosstalk between Wnt signaling and MAPK
pathways was shown to be involved in colorectal cancer recently [13]. There is also evidence for
crosstalk between Wnt signaling and TGF-beta pathways, mediated by Smad or Dishevelled
proteins [14, 15]. In the future, the crosstalk between these pathways specifically only in
colorectal carcinoma but in colorectal adenoma should be confirmed. For example, the
association between Smad and Dishevelled was shown to be induced by Wnt signaling and this
association prevented ubiquitination and deregulation of Smad [15]. First of all, it is necessary to
show this association in colorectal carcinoma cells and test for its absence in colorectal adenoma
samples. Then the reasons for the absence in colorectal adenoma samples can be analyzed
further. Elucidation of such mechanisms is critical for understanding the progression of the
invasive carcinoma phenotype from the adenoma phenotype.
We observed that Wnt signaling pathway members are upregulated compared to the other
members of the KEGG colorectal cancer pathway. This may be one explanation for the loss of
coherence in adenoma. Wnt pathway was known to be involved in the early progression of
colorectal cancer [8]. Thus, our study suggests that the regulation of pathways in cancer can be
captured at the gene expression level.
We showed that there are a lot more pairs of genes in the colorectal cancer pathway that
have correlated gene expression levels in carcinoma samples compared to normal or adenoma
samples. As a result of this, we observed a significant coherence of the pathway in only

150

carcinoma stage. In normal and adenoma samples, most pairs of correlated genes are from Wnt
signaling pathway; such as pairs of Dishevelled and Frizzled genes. In contrast, carcinoma
samples have correlated pairs of genes from different pathways corresponding to the different
sections of the colorectal cancer pathway, such as Tgfbr2 from TGF-beta pathway with Apc
from Wnt signaling pathway, Kras with Met, etc. These pairs of genes might have direct or
indirect protein-protein interactions or they might regulate each other at the transcriptional level
directly or indirectly. They can also be regulated by some common factors. For example, Myc
and Msh2 genes are correlated only in the carcinoma samples. This might be because Myc is a
positive transcriptional regulator of Msh2 or they have an indirect protein-protein interaction
between themselves mediated by Max [16, 17]. Therefore, it is necessary to confirm the
transcriptional regulation of Msh2 by Myc or the Max-mediated interaction between them in
colorectal carcinoma samples. Then, these can also be checked in colorectal adenoma samples so
that any differential regulation between these two proteins can be tested. This kind of further
studies might support our observation for more integrative behavior of the colorectal cancer
pathway in carcinoma stage. Then, the correlated pairs, specific for carcinoma stage, might be
important candidates for perturbation analysis to check for their effect on the cancer progression.
Lastly, we analyzed specific networks for colorectal cancer in order to identify novel
mechanisms involved in colorectal cancer. We used a novel approach to define the differentially
expressed genes in colorectal cancer. We compared the expression levels in colorectal cancer
samples to a large set of samples from various sources, conditions, etc. (multiple comparison
approach) rather than the classical way to compare the cancer samples to normal samples
(pairwise approach). The top ranked genes with a highly separated expression patterns obtained
by using the multiple comparison approach are more significantly associated with colorectal

151

cancer in the literature. This observation is based on counting the genes which gives at least one
paper in the NCBI Pubmed database when their official names are searched together with
‘Colorectal cancer’ or ‘Colon cancer’. This analysis might miss some genes which have articles
associated with colorectal cancer. Also the number of papers for a gene is omitted. A gene with
many articles associated with colorectal cancer might be more related to colorectal cancer than a
gene with only one article. Overall, our analysis suggests that multiple comparison approach to
obtain the differentially expressed genes gives more relevant lists of genes to colorectal cancer.
This should also be tested for other cancers, diseases, etc.
Interestingly, there is only one gene common to the specific genes obtained by pairwise
and multiple comparison approaches. This shows that the multiple comparison approach gives us
also a very different set of genes. Our study does not imply that the pairwise comparison
approach should be completely abandoned and the multiple comparison approach should be used
alone. Instead, the two approaches should be combined as they capture genes with different
expression patterns.
The colorectal cancer specific network is constructed by mapping the specific genes
obtained by the multiple comparison approach to the direct and indirect neighborhoods in the
human protein-protein interaction network. It can be used for generating hypotheses for novel
molecular mechanisms for colorectal cancer. One such example, that is between HOXB7 and
MAPK13, is confirmed in the literature. Protein level expression of HOXB7 was upregulated in
advanced, metastatic, highly proliferative stages of colorectal cancer patient samples and the
induction of HOXB7 expression in colorectal cancer cells induce MAPK pathway [18]. The
other interactions in this network can be tested in the future experimentally to check for their
involvement in colorectal cancer. We focused on glucocorticoid receptor (GR, NR3C1) and ring

152

finger 43 (RNF43) proteins in this network as they shows a high and significant separation of
expression levels for colorectal cancer in an independent paired dataset. GR is downregulated
while RNF43 is upregulated, which is also confirmed in most colorectal cancer cell lines
including HCT116.
We knocked down GR levels in HCT116 cell line by specific siRNA and showed that
both mRNA and protein levels of RNF43 increase in response to GR downregulation. We also
treated HCT116 cell line with Dexamethasone, which is a glucocorticoid ligand for GR, in order
to activate GR function. We observed a decline in RNF43 levels in response to Dexamethasone,
but it is less significant than the change of RNF43 levels in response to GR knock-down.
However, when we induced the HCT116 cells with both Dexamethasone and GR siRNA the
decrease in RNF43 is abolished. Therefore the negative effect of Dexamethasone on RNF43
might be significant. However, the most significant observation for the decline of RNF43 is
when the GR is silenced without Dexamethasone treatment. GR has some isoforms which do not
need glucocorticoid ligands to get activated and are always active in the nucleus [19]. These
isoforms might be responsible for the effect we observed without Dexamethasone treatment. As
a result, GR might be a negative regulatory factor for RNF43 transcription. This is partly
supported by a study in lung cancer, showing that GR regulates genes from distant sites [20]. GR
has binding sites further away from the transcription start site (TSS), for the genes that it
regulates negatively. Genes activated by GR are around 11 kb of TSS, whereas genes repressed
by GR are around 146 kb of TSS. They showed that GR has a binding site at 138 kb upstream of
RNF43 transcription start site in lung carcinoma cells [20]. This indicates that GR could be a
negative regulator of RNF43 in lung cancer. It should be tested in the future, whether GR could

153

bind to the distant promoter site of RNF43 and negatively regulate the transcription of RNF43 in
colorectal cancer.
Recently, β-catenin was found to be a positive regulator of RNF43 transcription, by direct
binding to its promoter together with TCF4 in HCT116 colorectal cancer cell line [21]. RNF43 is
induced by Wnt signaling pathway in colorectal cancer [22]. Wnt pathway is well-known to
induce the progression of colorectal cancer [8]. RNF43 is also upregulated in colorectal cancer
and knock-down of RNF43 suppresses the growth of colorectal cancer [23]. Therefore, it is
possible that Wnt signaling pathway, as an early event in colorectal cancer, induces RNF43,
which may in turn downregulate proteins like p53, or other unknown targets [24]. On the other
hand, GR downregulates Wnt signaling pathway by direct inactivation of β-catenin [25]. By
downregulating Wnt signaling pathway GR represses the targets of the pathway such as Cyclin
D1, therefore it is likely that GR could also downregulate RNF43, another target of Wnt
pathway, through inactivation of β-catenin. It is necessary to show the inhibition of β-catenin by
GR in colorectal cancer. Then the downregulation of RNF43 by GR could be explained by the
inactivation of β-catenin in colorectal cancer.
When we knocked-down RNF43 by siRNA, GR mRNA levels didn’t change whereas
there was a decline in protein levels, yet not statistically significant. This might show a positive
feedback from RNF43 to GR, yet it needs to be supported by further experiments. For example,
there might be a negative regulator of GR, which in turn might be a ubiquitination target of
RNF43. However, the analysis of RNF43 function in HCT116 cell line is questionable, as a
recent study showed that RNF43 is mutated in this cell line [26]. Interestingly RNF43 was shown
to inhibit Wnt signaling pathway and this was shown to be mediated by the ubiquitination of
Frizzled receptor. This observation is abolished in HCT116 colorectal cancer cell line as RNF43

154

is mutated, thus it cannot ubiquitinate Frizzled receptor. Therefore, it is necessary to analyze the
possible effect of RNF43 on GR in a different colorectal cancer cell line.
Overall, we were able to show a potential mechanism between GR and RNF43 in
colorectal cancer as the downregulation of GR could be involved in the upregulation of RNF43.
This observation has important clinical implications, as there are successful vaccine therapy
studies using RNF43 derived antigens in colorectal cancer patients [27]. The potential use of
glucocorticoids in colorectal cancer patients together with the RNF43 antigens needs to be
investigated.

155

BIBLIOGRAPHY

156

BIBLIOGRAPHY

1.

Deisboeck TS, Kresh JY. Complex Systems Science in Biomedicine. Springer, New
York, USA, 2006

2.

Barabasi AL, Gulbahce N, Loscalzo J. Network medicine: a network-based approach to
human disease. Nat Rev Genet. 2011, 12(1):56-68

3.

Barabasi AL, Oltvai ZN. Network biology: understanding the cell's functional
organization. Nat Rev Genet. 2004, 5(2):101-13

4.

Jeong H, Mason SP, Barabasi AL, Oltvai ZN. Lethality and centrality in protein
networks. Nature. 2001, 411(6833):41-2

5.

Jonsson PF, Bates PA. Global topological features of cancer proteins in the human
interactome. Bioinformatics. 2006, 22(18):2291-7

6.

Goh KI, Cusick ME, Valle D, Childs B, Vidal M, Barabasi AL. The human disease
network. Proc Natl Acad Sci U S A. 2007, 104(21):8685-90

7.

Yang HH, Hu Y, Buetow KH, Lee MP. A computational approach to measuring
coherence of gene expression in pathways. Genomics 2004, 84, 211-7

8.

Burgess AW, Faux MC, Layton MJ, Ramsay RG. Wnt signaling and colon
tumorigenesis--a view from the periphery. Exp Cell Res. 2011, 317(19):2748-58

9.

Lampropoulos P, Zizi-Sermpetzoglou A, Rizos S, Kostakis A, Nikiteas N, Papavassiliou
AG. TGF-beta signalling in colon carcinogenesis. Cancer Lett. 2012, 314(1):1-7

10.

Wang X, Dalkic E, Wu M, Chan C. Gene module level analysis: identification to
networks and dynamics. Curr Opin Biotechnol. 2008, 19(5):482-91

11.

Li Y, Agarwal P, Rajagopalan D. A global pathway crosstalk network. Bioinformatics.
2008, 24(12):1442-7

12.

Liu KQ, Liu ZP, Hao JK, Chen L, Zhao XM. Identifying dysregulated pathways in
cancers from pathway interaction networks. BMC Bioinformatics. 2012, 13(1):126

13.

Jeong WJ, Yoon J, Park JC, Lee SH, Lee SH, Kaduwal S, Kim H, Yoon JB, Choi KY.
Ras stabilization through aberrant activation of Wnt/β-catenin signaling promotes
intestinal tumorigenesis. Sci Signal. 2012, 5(219):ra30

14.

Letamendia A, Labbé E, Attisano L. Transcriptional regulation by Smads: crosstalk
157

between the TGF-beta and Wnt pathways. J Bone Joint Surg Am. 2001, 83-A Suppl 1(Pt
1):S31-9
15.

Mamidi A, Inui M, Manfrin A, Soligo S, Enzo E, Aragona M, Cordenonsi M, Wessely O,
Dupont S, Piccolo S. Signaling crosstalk between TGFβ and Dishevelled/Par1b. Cell
Death Differ. 2012, doi: 10.1038/cdd.2012.50

16.

Menssen A, Hermeking H, Characterization of the c-MYC-regulated transcriptome by
SAGE: identification and analysis of c-MYC target genes. Proc Natl Acad Sci USA
2002, 99, 6274-9

17.

Mac Partlin M, Homer E, Robinson H, McCormick CJ, Crouch DH, Durant ST,
Matheson EC, Hall AG, Gillespie DA, Brown R. Interactions of the DNA mismatch
repair proteins MLH1 and MSH2 with c-MYC and MAX. Oncogene. 2003, 22(6):819-25

18.

Liao WT, Jiang D, Yuan J, Cui YM, Shi XW, Chen CM, Bian XW, Deng YJ, Ding YQ.
HOXB7 as a prognostic factor and mediator of colorectal cancer progression. Clin
Cancer Res. 2011, 17(11):3569-78

19.

Oakley RH, Cidlowski JA. Cellular processing of the glucocorticoid receptor gene and
protein: new mechanisms for generating tissue-specific actions of glucocorticoids. J Biol
Chem. 2011, 286(5):3177-84

20.

Reddy TE, Pauli F, Sprouse RO, Neff NF, Newberry KM, Garabedian MJ, Myers RM.
Genomic determination of the glucocorticoid response reveals unexpected mechanisms of
gene regulation. Genome Res. 2009, 19(12):2163-71

21.

Bottomly D, Kyler SL, McWeeney SK, Yochum GS. Identification of {beta}-catenin
binding regions in colon cancer cells using ChIP-Seq. Nucleic Acids Res. 2010,
38(17):5735-45

22.

Hao HX, Xie Y, Zhang Y, Charlat O, Oster E, Avello M, Lei H, Mickanin C, Liu D,
Ruffner H, Mao X, Ma Q, Zamponi R, Bouwmeester T, Finan PM, Kirschner MW,
Porter JA, Serluca FC, Cong F. ZNRF3 promotes Wnt receptor turnover in an R-spondinsensitive manner. Nature. 2012, 485(7397):195-200

23.

Yagyu R, Furukawa Y, Lin YM, Shimokawa T, Yamamura T, Nakamura Y. A novel
oncoprotein RNF43 functions in an autocrine manner in colorectal cancer. Int J Oncol.
2004, 25(5):1343-8

24.

Shinada K, Tsukiyama T, Sho T, Okumura F, Asaka M, Hatakeyama S. RNF43 interacts
with NEDL1 and regulates p53-mediated transcription. Biochem Biophys Res Commun.
2011, 404(1):143-7

25.

Takayama S, Rogatsky I, Schwarcz LE, Darimont BD. The glucocorticoid receptor
represses cyclin D1 by targeting the Tcf-beta-catenin complex. J Biol Chem. 2006,
158

281(26):17856-63
26.

Koo BK, Spit M, Jordens I, Low TY, Stange DE, van de Wetering M, van Es JH,
Mohammed S, Heck AJ, Maurice MM, Clevers H. Tumour suppressor RNF43 is a stemcell E3 ligase that induces endocytosis of Wnt receptors. Nature. 2012, 488(7413):665-9

27.

Yasuda S, Tsuchiya I, Okada K, Tanaka A, Suzuki T, Sadahiro S, Takeda K, Yamamoto
S, Nakui M. Significant clinical response of advanced colon cancer to Peptide vaccine
therapy: a case report. Tokai J Exp Clin Med. 2012, 37(2):57-61

159