A Genomics and Mathematical Modeling Approach for the study

of Helicobacter pylori associated Gastritis and Gastric Cancer

A dissertation submitted to the Graduate School of the University of Cincinnati in partial fulfillment of the requirements of degree of

Doctor Of Philosophy in Systems Biology & Physiology Program in the Department of Molecular and Cellular Physiology by

Shruti Marwaha Post Graduate Diploma in Bioinformatics, Institute of Bioinformatics and Applied Biotechnology, India, 2007 BSc. Zoology, University of Delhi, India, 2005

Committee Chair: Dr. Hamid Eghbalnia, Ph.D.

Committee Members: Dr. Mario Medvedovic, Ph.D; Dr. Marshall Montrose, Ph.D; Dr. Nelson Horseman Ph.D; and Dr. Yana Zavros Ph.D

1

Abstract

Gastric cancer is the fifth most common malignancy in the world and third the leading cause of cancer-related mortality worldwide, with five-year survival rate of only 20-

29%. In order to develop better drugs, diagnostics and preventive measures for gastric cancer, it is critical to understand the underlying molecular biology of the disease and factors that increase the risk for the disease. Helicobacter pylori-induced chronic gastritis is a major risk factor associated with gastric cancer development.

We analyzed publically available gene expression data from patients with gastric cancer and patients with H. pylori mediated gastritis, to identify genes and pathways that play an important role in the two diseases. We further integrated the identified disease signature with Broad Institute’s Connectivity Map to identify and prioritize drugs that can potentially reverse the molecular signature of gastric cancer cells and that of gastric tumors resistant to Cisplatin-Flurouracil (CF) chemotherapy. Our analysis identified vorinostat, trichostatin A and thiostrepton as potential therapeutic compounds for gastric cancer treatment. We identified genes and pathways that are differentially expressed (57 up- regulated and 86 down-regulated) in both gastric cancer and H. pylori mediated atrophic gastritis. The topmost pathways enriched for these genes include - cell-cell adhesion/communication, tight junctions, leukocyte transendothelial migration, gastric acid secretion, potassium ion transport and creatine pathways. Analysis of CF resistant and sensitive tumors suggests the role of metabolic and statin pathways towards resistance to the chemotherapy.

2

We also developed a mathematical model of a sub-network comprising of sonic hedgehog (SHH), pro-inflammatory cytokines and anti-inflammatory cytokines, which play a critical role in H. pylori mediated gastritis. We integrated qPCR results, mathematical modeling technique and microarray data from H. pylori infected mice to explore the temporal behavior of the cytokine-SHH sub-network. Our mathematical model suggests that

NFĸB, SHH and the cytokines engage in a feedback loop which can result in damped oscillations. The model helps to bring out emergent properties of the network and generate testable hypotheses. Future experiments capturing cytokines and SHH profile over time can reveal more insights about the relationship between the different genes, their regulation and improve our current understanding of the dynamics and sequence of the events in the system.

3

4

Acknowledgements

I would like to acknowledge all the people in my professional and personal network for the scientific advice, support and encouragement I received during my PhD. First and foremost, I would like to thank my research advisor, Dr. Hamid Eghbalnia, for his inputs and insights through the development of this work, for the freedom to try different ideas.

I wish to express my gratitude towards my thesis committee, Dr. Mario Medvedovic,

Dr. Marshall Montrose, Dr. Nelson Horseman and Dr. Yana Zavros, who have helped greatly towards the refinement of this work through their reviews, inputs and suggestions. I am deeply grateful to them for agreeing to serve on my dissertation committee and for their advice and motivation. I am thankful to Dr. Zavros for her time, advice on the H. pylori and gastric cancer biology and giving me the opportunity for the hands-on experience in experimental lab. Many thanks to Dr. Medvedovic, for his suggestions, time and advice on statistics and his feedback on my ideas. I will also like to thank Dr. Montrose for his valuable advice, willingness to help & for being always accessible.

It was a great experience studying in the Systems Biology and Physiology program at

University of Cincinnati (UC). I wish to thank the Department and the University for the resources provided to me for my graduate studies.

Completing Grad school would have not been possible without the support and motivation of family and friends. A special thanks to my friends at UC - Sayali, Ravi, Gopi and

Balaji, for being the soundboard to my ideas. I am thankful to Mike and Mindy for helping me with my endeavor with wet lab. I big thank you to Priyanka, Divya, Swati, Preeti, Kavita, Hari,

5

Sudhir, Ganesh, Jai and all other friends from Cincy who made this place special. I am indebted to my parents and in-laws for their constant encouragement and support. I would like to dedicate this work to Maa - my mother, my support system, to my sister - Ritu didi, for her unconditional love and to my husband, my best friend - Ravi, without whose support this journey would neither be complete nor enjoyable.

And last but not least, I would like to thank all the patients who kindly contributed their data to public repositories for biomedical research.

6

Table of Contents

Chapter 1: Introduction and Motivation ...... 16

1.1 Gastric Cancer ...... 16

1.2 Objectives ...... 20

Chapter 2: Literature Review ...... 21

2.1 Anatomy and Histology of Normal Stomach ...... 21

2.1.1 Anatomy ...... 21

2.1.2 Histology ...... 22

2.2 Gastric Cancer ...... 23

2.2.1 Classification of Gastric Cancer ...... 23

2.2.2 Stages of Gastric Cancer ...... 24

2.2.3 Risk factors for Gastric Cancer ...... 28

2.2.4 Gastric Cancer Diagnosis ...... 33

2.2.5 Currently Available Therapy for Gastric Cancer ...... 34

2.3 Drug Repurposing: Finding new uses of old drugs ...... 35

2.4 Mathematical Modeling ...... 36

2.4.1 Why mathematical models? ...... 36

2.4.2 Modeling biology using Ordinary Differential Equations ...... 37

2.4.3 Using Eigenvalues to determine system’s stability ...... 39

Chapter 3: Mathematical model for studying Helicobacter pylori mediated inflammation in host gastric tissue...... 41

3.1 Synopsis ...... 41

3.2 Background ...... 42 7

3.3 Aim ...... 44

3.4 Methods ...... 44

3.4.1 Animal Model ...... 44

3.4.2 Interaction Map ...... 45

3.4.3 Statistical Analysis ...... 47

3.4.4 Mathematical model ...... 48

3.4.5 Sensitivity Analysis ...... 50

3.4.6 GEO Data ...... 50

3.5 Results ...... 51

3.5.1 SHH positively regulates cytokine expression during H. pylori infection ...... 51

3.5.2 Mathematical Model Behavior ...... 55

3.5.3 Temporal Analysis of GEO Data ...... 61

3.5.4 Sensitivity and Stability Analysis ...... 61

3.6 Key Findings ...... 62

Chapter 4: Meta-analysis of gastric cancer and H. pylori mediated gastritis microarray data...... 72

4.1 Background ...... 72

4.2 Aim ...... 73

4.3 Methods ...... 73

4.3.1 Datasets: ...... 73

4.3.2 Differential Gene Expression Analysis ...... 74

4.3.3 Connectivity Map Analysis ...... 75

4.4 Results ...... 75

8

4.5 Discussion ...... 77

4.6 Limitations ...... 81

4.7 Conclusions ...... 81

Chapter 5: Drug repurposing for gastric cancer using gene expression data...... 89

5.1 Background ...... 89

5.2 Aim ...... 90

5.3 Methods ...... 91

5.3.1 Datasets ...... 91

5.3.2 Differential Gene Expression Analysis ...... 92

5.3.3 Connectivity Map Analysis ...... 92

5.4 Results ...... 93

5.4.1 Gastric Cancer Data Analysis ...... 93

5.4.2 Connectivity Map Analysis ...... 94

5.4.3 Acquired Resistance to Cisplatin and Flurouracil in Gastric Tumors ...... 99

5.5 Limitations ...... 100

5.6 Conclusions ...... 101

Chapter 6: Regulation of cytokines by Sonic Hedgehog in macrophages...... 102

6.1 Background ...... 102

6.2 Aim ...... 103

6.2.1 Hypothesis ...... 103

6.2.2 Significance ...... 103

6.2.3 Expected Results: ...... 103

6.3 Materials and Methods ...... 103

9

6.3.1 Cell Culture ...... 103

6.3.2 Quantitative Real-Time PCR (qPCR) ...... 104

6.3.3 Statistical Analysis ...... 105

6.4 Results ...... 105

6.5 Conclusion ...... 106

Chapter 7: Using empirical probability (frequency) of change for analyzing microarray data...... 108

7.1 Background and Motivation ...... 108

7.2 Hypothesis ...... 109

7.3 Methods ...... 109

7.3.1 Datasets ...... 109

7.3.2 Calculation of Frequency Score ...... 110

7.3.3 Calculation of Fold Score ...... 111

7.3.4 Integration of Frequency of change and Fold change with pathways

knowledge ...... 111

7.4 Preliminary Results ...... 112

7.5 Limitations ...... 112

Chapter 8: Discussion ...... 114

8.1 Mathematical modeling of cytokine-SHH sub-network in H. pylori infected gastric

tissue...... 115

8.1.1 Summary of Major Findings ...... 115

8.1.2 Significance of the findings ...... 115

8.1.3 Future Directions ...... 115

10

8.2 Meta-analysis of microarray data from gastric cancer and H. pylori infected

patients...... 116

8.2.1 Summary of Major Findings ...... 116

8.2.2 Significance of the findings ...... 116

8.2.3 Future Directions ...... 116

8.3 Drug Repurposing for gastric cancer using gene expression data...... 117

8.3.1 Summary of Major Findings ...... 117

8.3.2 Significance of the findings ...... 117

8.3.3 Future Directions ...... 118

Chapter 9: Bibliography ...... 119

11

List of Figures

Figure 2-1. Parts of the stomach [27]...... 22

Figure 2-2. Histology of stomach [29] ...... 23

Figure 2-3. Multifactorial pathway leading to gastric carcinoma [51] ...... 32

Figure 2-4. Determination of system’s behavior by eigenvalues derived from Jacobian matrix of system [90]...... 40

Figure 3-1. Interaction Map of signaling pathways activated in host stomach in response to H. pylori...... 46

Figure 3-2. Graphical representation of mathematical model of cytokine-SHH network, during H. pylori infection...... 50

Figure 3-3. Interaction between infection status and genotype...... 53

Figure 3-4. Effect of H. pylori on SHH and cytokines's expression in WT and PC-SHHKO mouse stomachs, day 7 and day 180 post-inoculation...... 54

Figure 3-5. In-silico SHH KO results show a decrease in cytokines as compared to WT...... 56

Figure 3-6. Temporal profiles of model species in uninfected and infected conditions...... 59

Figure 3-7. In-silico IL-10 knock-out and overexpression...... 60

Figure 3-8. Trajectories of cytokines in mock-infected and H. pylori infected mice from GEO microarray dataset. A) IL-10, (B) IL-1β and (C) IFNγ for day 2, 7, 14 and 28 from chief, parietal and pit cell of mock-infected and H. pylori infected mice. The temporal profiles

12

indicate that these cytokines potentially display a cyclic expression pattern in response to H. pylori infection...... 64

Figure 3-9. Temporal profiles of model species, in absence of negative feedback on NFĸB by

IL-10...... 64

Figure 4-1. Venn diagram showing intersection between Differentially Expressed Genes

(DEG) from gastric cancer and H. pylori mediated atrophic gastritis...... 77

Figure 5-1. Overview of the methodology...... 93

Figure 5-2. Venn diagram showing intersection between Differentially Expressed Genes

(DEG) from the two gastric cancer datasets...... 94

Figure 6-1. Effect of SHH on the expression of cytokines in H. pylori stimulated macrophages...... 106

Figure 7-1. Differentially expressed genes predicted by Limma and frequency-based approach...... 113

13

List of Tables

Table 2-1. Stage wise treatment options and survival rates in gastric cancer patients [34]. ... 27

Table 2-2. Determination of system’s behavior by eigenvalues derived from Jacobian matrix of system...... 40

Table 3-1. Sensitivity Analysis of model parameters for damped oscillations. Key parameters and their range for which the model shows damped oscillations...... 64

Table 3-2 Model Assumptions ...... 65

Table 3-3 Mathematical equations used in the model...... 66

Table 3-4 Species’ parameters used in the model...... 68

Table 3-5 Kinetic parameters used in the model...... 69

Table 4-1. Sample size of GEO datasets used in the study. GSE27411 represents the H. pylori mediated atrophic gastritis dataset while GSE27342 and GSE13861 are the gastric cancer datasets...... 74

Table 4-2. Number of Differentially Expressed Genes (DEG) in gastric cancer and H. pylori mediated atrophic gastritis datasets. n = sample size...... 76

Table 4-3. Enriched GO processes and pathways in DEG common to gastric cancer and H. pylori mediated gastritis datasets...... 79

Table 4-4. Top ranked compounds with expression profiles opposite to that of common DEG signature identified from gastric cancer and H. pylori mediated atrophic gastritis...... 82

14

Table 4-5. The role of compounds predicted by CMap in gastric cancer...... 82

Table 4-6. Genes overexpressed in gastric cancer and Helicobacter pylori mediated gastric atrophy datasets...... 84

Table 4-7. Genes under-expressed in gastric cancer and Helicobacter pylori mediated gastric atrophy datasets...... 86

Table 5-1. Sample size of GEO datasets used in the study. CF: Cisplatin and Flurouracil. .... 91

Table 5-2. Number of Differentially Expressed Genes (DEG) in gastric cancer datasets. n = sample size...... 94

Table 5-3. List of the compounds predicted by CMap for each gastric cancer dataset. The compounds highlighted in bold and with asteric have been reported in literature to have implications in gastric cancer at in-vitro or in-vivo level...... 96

Table 5-4. Characteristics of CMap compounds that have expression profiles opposite to that of gastric cancer tumors in this study...... 98

Table 5-5. CMap pathway predicted for each pathway...... 98

Table 5-6. Pathway enrichment for genes differentially expressed in cisplatin- resistant tumors...... 100

15

Chapter 1: Introduction and Motivation

1.1 Gastric Cancer

Gastric cancer is the fifth most common malignancy in the world and the third leading cause of cancer-related mortality worldwide [1], with a five year survival rate of only 20% [2] to 29% [3]. Risk factors for development of gastric cancer include chronic atrophic gastritis, Helicobacter pylori infection, smoking, heavy alcohol use, and high salt intake [4].

Helicobacter pylori is one of the most significant risk factor associated with gastric cancer

[5]. In 1994, H. pylori was classified as class I carcinogen by the International Agency for

Research on Cancer [6]. H. pylori-induced chronic gastritis is the first step of the cascade of gastric cancer development. Correa proposed the “human model of gastric carcinogenesis” as a cascade that includes the non-atrophic chronic gastritis, multifocal atrophic gastritis, intestinal metaplasia, low-grade dysplasia (low-grade non-invasive neoplasia), high-grade dysplasia (high-grade non-invasive neoplasia), and invasive adenocarcinoma [7,8]. Non- atrophic gastritis can be either cured by clearing H. pylori infection, or it may evolve in of the following ways: a) it can remain as non-atrophic or b) it can progress in severity, damaging gastric glands. The progression depends on interplay of three factors: virulence of infectious agent, host’s genetic susceptibility and external environment [9].

H. pylori infection of gastric epithelium leads to recruitment and activation of immune cells. Non-specific immune response involves the release of chemokines such as IL-8 and

GRO-alpha, and pro-inflammatory cytokines liberated by mononuclear phagocytes (TNF-α,

IL-1 and IL-6) [10,11]. Antigen-specific cellular immunity results in a predominant Th1

16

lymphocyte response with increase in IFN-γ secreting T-helper cells, while humoral responses lead to the production of anti-H. pylori antibodies and complement activation

[12]. The complex network of cytokines implicated in these inflammatory responses include counter-regulatory elements such as IL-10 which may serve to damp down inflammation

[11,13]. The balance between pro-inflammatory and immunosuppressive cytokines is likely to be a critical determinant of the severity of H. pylori-associated inflammation [12]. Chronic cellular damage leads to increase in reactive oxygen and nitrogen species [14,15], epithelial cyclooxygenase-2 and activation of the pro-inflammatory transcription factor NF-κB, which may all contribute to gastric cancer development [16]. Sonic Hedgehog (SHH) is another important gene that has been reported to play a role in pathogenesis of chronic H. pylori infection [100,101] and gastric cancer [102–105].

In spite of the advancement in identifying the host susceptibility genes, the precise molecular and cellular details of how H. pylori associated inflammatory response promotes gastric cancer development, are not well understood. Identification of a gene signature that is common to gastric cancer and H. pylori mediated gastritis can enhance our current knowledge about the molecular details which link the two diseases. Moreover, a systemic understanding of the gene network is a prerequisite for enabling rational manipulation of specific genes and pathways that would help to develop in-vivo and in-vitro disease models.

However, such experiments are associated with high cost, especially time-series experiments. Mathematical models have been useful tools for understanding and predicting emergent properties of such complex systems. Models provide useful insights towards generating new hypotheses, and can be helpful in the design of subsequent experimental

17

studies. In the context of studying dynamics involving interactions that drive the immune response to H. pylori, alternatives for interactions among components of the system can be explored computationally in order to yield insight into possible dynamic modes. By comparing experimental data and possible dynamical modes, it may be possible, in some cases, to significantly optimize experimental models and help ameliorate complexity and cost of bench-top experiments.

Key challenges in clinical management of gastric cancer are late diagnosis and limited success rate of currently available drugs. Early diagnosis of gastric cancer is difficult as most patients are asymptomatic in early stages. Weight loss and abdominal pain are often late signs of tumor progression [17]. Treatment options include surgery, chemotherapy and radiation therapy. Unfortunately, most cases are diagnosed in later stages [17] and advanced gastric cancer is generally refractory to chemotherapy [18]; with the five year survival rate of less than 40% for patients who undergo surgery [19]. Target based therapy, which blocks the growth of cancer cells by interfering with specific targeted molecules that are required for carcinogenesis and tumor growth, has emerged as a new tool in treatment of cancer [20]. Combination of target based drugs with standard cytotoxic chemotherapy has shown promising results in some cancers; for example, breast cancer (Trastuzumab,

Lapatinib), non-small cell lung cancer (Erlotinib, Gefitinib) and renal cell carcinoma

(Sunitinib) [20]. However development of target based drugs for gastric cancer treatment is challenging due to limited or absence of biomarker knowledge, and heterogeneous nature of the disease. Currently, there are only two FDA approved target based drugs for gastric cancer treatment – Trastuzumab (targeting HER2) [21] and Ramucirumab (targets VEGF)

18

[22]. Clearly, there is an urgent need for development of more drugs for gastric cancer.

However, drug development is a long and expensive process. The average cost and time for developing a new drug is approximately 1-2 billion dollars and 10 years, respectively [23].

Moreover, the rate of failure of a new lead compound in clinical phase is high, especially for oncology drugs [24].

One of the approaches that may help reduce the high cost and time associated with drug discovery is repurposing drugs approved for other diseases. A major advantage of repositioning drugs is their established safety profile (in cases where doses regiments do not change significantly). Identification of molecular signature derived from disease omics data, with or without drug treatment, can be help to prioritize existing drugs for treatment of the disease. Gene expression profiling can help identify genes that may serve as candidate biomarkers and the profiles be can used for other applications such as for diagnosis, prognosis, sub-classification of the disease, response to specific drugs and as potential drug targets. For example, OncotypeDx, MammaPrint and Mammostrat are commercially available genomic assays for early stage breast cancer that predict the likelihood of chemotherapy benefit as well as recurrence of the disease [25]. Similar multi-gene expression tests are required for better treatment and prognosis of gastric cancer.

Integration of pathway knowledge with genomics data can further help provide key targets for repositioned drugs.

19

1.2 Objectives

The long-term goal of this work is to understand the mechanism by which Helicobacter pylori mediated gastritis increases the risk for development of stomach cancer. There are two objective of this thesis – (I) to understand crosstalk between inflammatory cytokines and Sonic Hedgehog (SHH), which are activated during H. pylori infection. (II) (a) to identify the molecular signature common to gastric cancer and H. pylori mediated atrophic gastritis.

(II) (b) drug repurposing for gastric cancer. These objectives are addressed using computational approaches by the following aims. These computational approaches act as tools for hypotheses generation and can complement future experimental studies to help advance our current understanding of the disease. Each aim is described in detail in the chapters 3-5.

AIM I: To develop a mathematical model of SHH and inflammatory cytokines that are activated during H. pylori infection.

AIM II is divided into two sub-aims:

I. Meta-analysis of gene expression data to identify the molecular signature common

to gastric cancer and H. pylori mediated atrophic gastritis.

II. Identify Therapeutic Compounds that can be potentially used for Treatment of

Gastric Cancer.

20

Chapter 2: Literature Review

2.1 Anatomy and Histology of Normal Stomach

2.1.1 Anatomy

The human stomach can be divided into different sections (Figure 2-1):

I. Cardia: is a small space at the esophageal entrance to the stomach that sits just

under the diaphragm and the heart.

II. Fundus: The upper part of the stomach next to the cardia.

III. Body (corpus): The main part of the stomach, between the upper and lower parts.

IV. Antrum: The lower portion (near the intestine), where the food is mixed with gastric

juice.

V. Pylorus: The last part of the stomach, which acts as a valve to control emptying of

the stomach contents into the small intestine.

The first three parts of the stomach (cardia, fundus, and body) are sometimes called the proximal stomach. Some cells in these parts of the stomach make acid and pepsin (a digestive enzyme), the parts of the gastric juice that help digest food. They also make a protein called intrinsic factor, which the body needs to absorb vitamin B12. The lower 2 parts (antrum and pylorus) are called the distal stomach [26].

21

Figure 2-1. Parts of the stomach [27].

2.1.2 Histology

The stomach wall is made up of four layers (Figure 2-2):

a) The innermost layer is the mucosa. The mucosa has 3 parts: epithelial cells, which lie

on top of a layer of connective tissue (the lamina propria), which is on top of a thin

layer of muscle (the muscularis mucosa).

b) Under the mucosa is a supporting layer called the submucosa.

c) Below this is the muscularis propria, a thick layer of muscle that moves and mixes

the stomach contents. 22

d) The next two layers, the subserosa and the outermost serosa, act as wrapping layers

for the stomach [28].

Figure 2-2. Histology of stomach [29]

2.2 Gastric Cancer

2.2.1 Classification of Gastric Cancer

Histologically, gastric tumors can be classified as adenocarcinoma (cancer starts in the gland cells in the stomach lining), GISTs (Gastrointestinal stromal tumor), Lymphoma or

Neuroendocrine tumors. The most common form of stomach cancer is adenocarcinoma [30]. Lauren classified gastric adenocarcinoma into two histological types, intestinal and diffuse, according to morphological features of tumor [31]. Intestinal-type tumors are characterized by a corpus-dominated gastritis with gastric atrophy and intestinal metaplasia, whereas diffuse-type tumors show gastritis throughout the stomach but no atrophy [32]. Intestinal-type adenocarcinoma, progresses through a series of histological

23

steps Figure 2-3 that are initiated by the transition from normal mucosa to chronic superficial gastritis, which then leads to atrophic gastritis and intestinal metaplasia, and finally to dysplasia and adenocarcinoma [8].

2.2.2 Stages of Gastric Cancer

Gastric cancer is routinely classified according to the tumor–node–metastasis (TNM) parameters of the primary tumor, lymph nodes, and metastasis [33]. Table 2-1 summarizes stage wise treatment options and survival rates in gastric cancer patients.

The TNM system for staging contains 3 key pieces of information:

§ T describes the extent of the primary tumor (how far it has grown into the wall of the

stomach and into nearby organs).

§ N describes the spread to nearby (regional) lymph nodes.

§ M indicates whether the cancer has metastasized (spread) to distant parts of the body.

Numbers or letters appear after T, N, and M to provide more details about each of these factors:

1) The numbers 0 through 4 indicate increasing severity.

2) The letter X means “cannot be assessed” because the information is not available.

3) The letters “is” refer to carcinoma in situ, which means the tumor is only in the top layer

of mucosa cells and has not yet invaded deeper layers of tissue.

24

T categories of stomach cancer

Nearly all stomach cancers start in the innermost layer of the stomach wall (the mucosa).

The T category describes how far through the stomach’s 5 layers the cancer has invaded.

TX: The main (primary) tumor cannot be assessed.

T0: No signs of a main tumor can be found.

Tis: Cancer cells are only in the top layer of cells of the mucosa (innermost layer of the stomach) and have not grown into deeper layers of tissue such as the lamina propria or muscularis mucosa. This stage is also known as carcinoma in situ.

T1: The tumor has grown from the top layer of cells of the mucosa into the next layers below such as the lamina propria, the muscularis mucosa, or submucosa.

§ T1a: The tumor is growing into the lamina propria or muscularis mucosa.

§ T1b: The tumor has grown through the lamina propria and muscularis mucosa and into

the submucosa.

T2: The tumor is growing into the muscularis propria layer.

T3: The tumor is growing into the subserosa layer.

T4: The tumor has grown into the serosa and may be growing into a nearby organ (spleen, intestines, pancreas, kidney, etc.) or other structures such as major blood vessels.

§ T4a: The tumor has grown through the stomach wall into the serosa, but the cancer

hasn’t grown into any of the nearby organs or structures.

§ T4b: The tumor has grown through the stomach wall and into nearby organs or

structures.

25

N categories of stomach cancer

NX: Nearby (regional) lymph nodes cannot be assessed.

N0: No spread to nearby lymph nodes.

N1: The cancer has spread to 1 to 2 nearby lymph nodes.

N2: The cancer has spread to 3 to 6 nearby lymph nodes.

N3: The cancer has spread 7 or more nearby lymph nodes.

§ N3a: The cancer has spread to 7 to 15 nearby lymph nodes.

§ N3b: The cancer has spread to 16 or more nearby lymph nodes.

M categories of stomach cancer

M0: No distant metastasis (the cancer has not spread to distant organs or sites, such as the liver, lungs, or brain).

M1: Distant metastasis (the cancer has spread to organs or lymph nodes far away from the stomach).

Treatment of stomach cancer and survival rate depends to a large degree on where the cancer started in the stomach and how far it has spread.

26

5 year Stage TNM Status Treatment Options observed survival

0 Tis, N0, M0 Surgery -

IA T1, N0, M0 Surgery 71%

T1, N1, M0 Primary: Surgery; IB 57% T2, N0, M0 Secondary: Chemotherapy or chemoradiation

T1, N2, M0 Primary: Surgery; IIA T2, N1, M0 46% Secondary: Chemotherapy or chemoradiation T3, N0, M0

T1, N3, M0

IIB T2, N2, M0 Primary: Surgery; 33% T3, N1, M0 Secondary: Chemotherapy or chemoradiation

T4a, N0, M0

T2, N3, M0 IIIA Primary: Surgery; T3, N2, M0 20% Secondary: Chemotherapy or chemoradiation T4a, N1, M0

T3, N3, M0 Primary: Surgery; IIIB T4a, N2, M0 14% Secondary: Chemotherapy or chemoradiation T4b, N0 or N1, M0

IIIC T4a, N3, M0 Primary: Surgery; 9% T4b, N2 or N3, M0 Secondary: Chemotherapy or chemoradiation

Surgery, Chemo and/or radiation therapy, IV Any T, any N, M1 4% Targeted therapy

Table 2-1. Stage wise treatment options and survival rates in gastric cancer patients [34].

27

2.2.3 Risk factors for Gastric Cancer

Incidence rates for gastric cancer vary widely geographically, and, in general, more males than females are affected [35]. Although incidence rates of gastric cancer have been declining worldwide for several decades, Japan, China, Eastern Europe, and certain Latin

America countries still remain as high-risk areas [35]. Several factors like chronic atrophic gastritis, Helicobacter pylori infection, smoking, heavy alcohol use, and high salt intake have been linked to increased risks for gastric cancer [36]. Epidemiological studies have shown that individuals infected with H. pylori have an increased risk of gastric adenocarcinoma

[37–39]. The odds ratio for the relative risk for gastric carcinoma in H. pylori-infected individuals is about 13.3 (95% confidence interval (CI) 5.3-35.6) [40] to 16.7 (95% confidence interval (CI) 9.6–29.1) [41]. The odds ratio further varies depending on ethnic background, sex and age of the individual [41] and the strain of H. pylori. Among infected patients, those with severe atrophy accompanying intestinal metaplasia, corpus-predominant gastritis, or both are at particularly high risk [37].

2.2.3.1 Role of Helicobacter pylori in gastritis and gastric cancer

H. pylori-induced gastritis is associated with three phenotypes that correlate closely with clinical outcome.

§ antrum-predominant/corpus-sparing pattern associated with high acid secretion and

increased risk of duodenal ulcer disease.

§ mild mixed antrum/corpus gastritis with no major effect on acid secretion and,

generally, no serious clinical outcome.

28

§ corpus-predominant or severe pangastritis pattern that is associated with gastric

atrophy, hypochlorhydria, and an increased risk of gastric cancer. Inhibition of gastric

acid pharmacologically can lead to a shift from an antrum-predominant pattern to a

corpus-predominant one with onset of gastric atrophy. [42]

H. pylori causes damage by initiating chronic inflammation in the gastric mucosa. This inflammation is mediated by an array of pro- and anti-inflammatory cytokines [42]. H. pylori induced chronic gastric inflammation may progresses to atrophy, metaplasia, dysplasia and ultimately gastric cancer [43]. The bacterial colonization results in development of an immune reaction, characterized by proinflammatory TH1 response [44,45]. Although the immune response decreases H. pylori numbers [46], it is ineffective in completely eradicating the bacteria [44] and does not provide resistance against reacquisition after antimicrobial therapy [47]. Activation of pro-inflammatory and oncogenic transcription factor NFkB, increased expression of epithelial cyclooxygenase-2 can potentially contribute to gastric cancer [16]. Chronic inflammation of gastric mucosa can also increase cellular damage and turnover, thus promoting carcinogenesis [48,49]. Studies focusing on the mechanism by which H. pylori mediated inflammation promotes gastric cancer can help to improve our current understanding of the disease and lead to prevention and treatment strategies for gastric cancer. Overall, the risk of development of gastric cancer in the presence of H. pylori infection depends on a variety of bacterial, host, and environmental factors that mostly relate to the pattern and severity of gastritis [50].

a) H. pylori virulence factors

29

H. pylori infects 50% of the world population but most of the carriers do not develop gastric cancer [51]. Cancer risk is associated with H. pylori strain differences, inflammatory responses governed by host genetics, and specific interactions between host and microbial determinants. H. pylori populations are extremely diverse [52], owing to point mutations, substitutions, insertions and/or deletions in their genomes [53]. Several genomic loci encoding virulence factors, such as the cag pathogenicity island (cagPAI), the toxin VacA, and the Bab2 adhesin, have been strongly associated with an increased risk of developing gastric cancer and peptic ulcer disease [54,55].

b) Host susceptibility factors

It has become apparent that not only the pathogen but also host genetics play an important role in determining the clinical manifestation of H. pylori infections. Genetic polymorphisms in inflammatory genes tend to increase the risk of gastric cancer, as demonstrated for IL-1, a potent proinflammatory cytokine and the most prominent inhibitor of gastric acid secretion [56]. Reduced acid secretion is linked to corpus-predominant colonization by H. pylori , which results in pangastritis formation of atrophic gastritis and thus an increased risk of gastric cancer and gastric ulcer disease [57–59]. Specific IL-10 haplotypes lead to higher cytokine expression levels, thereby, shifting the balance towards an anti-inflammatory host cell [60,61]; this is associated with the colonization of more virulent H. pylori strains [62]. However, specific IL-10 haplotypes actually induce lower IL-10 expression levels, favouring pro-inflammatory responses and an associated increased risk of gastric cancer [63]. Analysis of high-throughput gene expression from H. pylori infected

30

stomach tissues can help to identify more of such genes that influence the final outcome of the disease. These genes may serve as candidate biomarkers and can be used for several applications such as for diagnosis, prognosis, sub-classification of the disease, response to specific drugs and as potential drug targets.

c) Molecular Mimicry

One of the mechanisms for the failure of host’s immune system in eliminating H. pylori includes the ability of the bacteria to evade host adaptive and innate responses by frequent antigenic variation, and host antigen mimicry [16,64]. Molecular mimicry refers to a condition where microbial antigen can include an epitope that is structurally similar to an auto-antigen epitope [65]. H. pylori infection can mediate the autoimmune destruction of the gastric corpus mucosa and lead to autoimmune gastritis (AIG) or pernicious anemia through molecular mimicry between host HK-ATPase and H. pylori antigens of T-cell epitopes [66].

31

Figure 2-3. Multifactorial pathway leading to gastric carcinoma [51]

32

2.2.4 Gastric Cancer Diagnosis

2.2.4.1 Symptoms

The physical symptoms of gastric cancer, which include poor appetite, weight loss, abdominal pain, heartburn, nausea, and anemia, are not specific to gastric cancer alone and are often associated with stomach virus or ulcer [67].

2.2.4.2 Diagnosis

The initial diagnosis of gastric carcinoma often is delayed because up to 80 percent of patients are asymptomatic during the early stages of stomach cancer [68]. In Japan, a higher incidence of adenocarcinoma and rigorous screening processes have led to a greater number of cases of gastric cancer being detected in an early stage (i.e., when limited to the mucosa and submucosa, with or without lymph node involvement) [17]. The diagnostic techniques include Esophagogastroduodenoscopy (EGD), Endoscopic ultrasonography, computed tomographic (CT) scanning [17].

2.2.4.3 Genetic Screening

Hereditary diffuse gastric cancer (HDGC) is an autosomal dominant genetic condition associated with an increased risk of gastric cancer. It is associated most frequently with a heterozygous germline mutation in CDH1, the gene that encodes E-cadherin. CDH1 has an essential role in cell–cell adhesion [69]. A mutation in the CDH1 gene increases the risk of developing gastric cancer and other cancers associated with HDGC. Not everyone who inherits a gene mutation for HDGC will develop cancer. In people who have a mutation in

33

the CDH1 gene, the lifetime risk for advanced diffuse gastric cancer is estimated to be greater than 80% for men and women by age 80. Screening for stomach cancer is suggested for people known to be at risk for HDGC [70,71]. High-throughput analysis of gene expression data from patient and healthy population can help to improve our current understanding of the molecular profile of gastric cancer and identify potential diagnostic biomarkers and candidate drug targets.

2.2.5 Currently Available Therapy for Gastric Cancer

There are three main modes of therapy for the treatment of gastric cancer – surgery, radiation and chemotherapy. Treatment with surgery alone is not very promising in improving the survival rate [17]. Gastric cancer has not been shown to respond successfully to radiation alone. Chemotherapy has demonstrated limited success with multi-drug regimens [72]. The common drugs approved for gastric cancer include fluorouracil, ramucirumab, docetaxel, trastuzumab, mitomycin C, doxorubicin [73].

Since, chemotherapeutic drugs target rapidly dividing cells, which includes both cancer cells and certain normal tissues, they are associated with severe adverse effects and toxicities. Although traditional cytotoxic chemotherapy remains the treatment of choice for many malignancies, with recent advancement in molecular cancer biology, targeted-based drugs have emerged as a component of treatment for many types of cancer, including that of breast, colorectal, lung, pancreatic cancers, lymphoma, leukemia, and multiple myeloma

[20]. Targeted therapy aims to act on a specific target or biologic pathway that, when inactivated, causes regression or destruction of the malignant process [74]. Currently, there are only two FDA approved target based drugs for gastric cancer treatment – Trastuzumab 34

(targeting HER2) [21] and Ramucirumab (targeting VEGF) [22]. Trastuzumab, which binds to

HER2 receptors, was originally approved by FDA for breast cancer treatment in 1998.

Overexpression and/or amplification of HER2 has been found in about 20–30% of breast cancers [75], in >20% of gastric cancers and in 33% of gastroesophageal junction (GEJ) adenocarcinoma [76]. In 2010, FDA granted approval for trastuzumab, in combination with cisplatin and a fluoropyrimidine (either capecitabine or 5-fluorouracil), for the treatment of patients with HER2-overexpressing metastatic gastric or gastroesophageal junction (GEJ) adenocarcinoma who have not received prior treatment for metastatic disease [77]. It is now recommended that all patients with gastric cancer should have their tumors tested for

HER2 status at the time of initial diagnosis. Studies focusing on molecular profiling of the gastric tumors can help to identify specific biomarkers like HER2 and help in selecting drugs that target those biomarkers.

2.3 Drug Repurposing: Finding new uses of old drugs

Development of a new drug usually takes around 10 years and can cost of several billion dollars [23]. One of the major reasons for this extraordinary cost is high rate of failure in the preclinical and clinical phase. Only one out of ten new drugs in clinical trials reach the

FDA approval stage [78]. One of the approaches to reduce the high cost and time associated with drug discovery is through repurposing of drugs already approved for other diseases. A key advantage of drug repurposing is the established clinical safety of the drug.

Drug repurposing strategy has been proven to be therapeutically useful in several diseases. A well-known example of a successful repurposed drug is Sildenafil (Viagra), which 35

was originally developed to treat erectile dysfunction. Later it was discovered that the vasodilatory effects of sildenafil also help reduce symptoms of pulmonary arterial hypertension (PAH) and FDA subsequently approved the drug for treatment of PAH [79].

Raloxifene was first used to prevent osteoporosis (bone loss) in postmenopausal women.

Raloxifene is now used for the reduction in the incidence of invasive breast cancer in postmenopausal women with osteoporosis or have a high risk for developing breast cancer

[80]. In 2010, FDA approved use of a breast cancer drug - trastuzumab with chemotherapy for treatment for metastatic stomach cancer in HER2 positive patients [81]. Identification of existing drugs that have potential in treatment of gastric cancer can help improve the survival rate for the disease. Chapter 5 explains in detail how we used genomics data and

Broad Institute’s Connectivity Map to identify FDA approved drugs and nondrug bioactive compounds that can be potentially used for treatment of gastric cancer.

2.4 Mathematical Modeling

2.4.1 Why mathematical models?

To understand complex biological systems, in addition to the identification and characterization of the individual components, it is also important to understand the interactions between molecules and pathways. However, this understanding is often difficult to achieve experimentally due to high cost and time associated with such studies.

Mathematical models are abstractions of complex biological processes and allow prediction of the system behavior (under given condition). Mathematical models enable and facilitate studies aimed at investigating the mechanism through which different components of the

36

system are interlinked and how disruptions of these links may contribute to the development of disease [82]. Investigation of a signaling network’s positive and negative feedback loops can give insight into the network’s behavior, and identify important system properties such as multi-stability, excitability, and oscillations [83]. Dynamic models identify emergent patterns of the system that may be non-observable and non-intuitive when individual components of the network are studied in isolation. For example, Lev Bar-Or et al. developed a mathematical model suggesting that, given a sufficient time delay for p53- induced Mdm2 transcription, the core p53 circuit can give rise to oscillations. To model this behavior, the authors created a set of Ordinary Differential Equations to describe the system and added a hypothetical intermediate that directs Mdm2 transcription with kinetics governed by a Hill function. Using immunoblots, they verified that the levels of Mdm2 and p53 oscillate in response to DNA damage–inducing irradiation. Although the model parameters were estimated rather than measured or fit to experimental data, Lev Bar-Or et al. were able to infer qualitative behaviors such as the broadening of the p53 response to weaker DNA damage signals (Lev Bar-Or et al., 2000 ; Hughey, Lee, & Covert, 2009).

Although in-silico models cannot replace laboratory experiments, simulations enable the exploration of many different aspects of the system’s behavior at low cost and in much less time.

2.4.2 Modeling biology using Ordinary Differential Equations

Different approaches have been employed in order to model the dynamics of biological systems, for example: Boolean, Ordinary Differential Equations, Flux Balance

Analysis, stochastic, cellular-automata [86]. The choice of modeling approach depends on

37

the goal of the study. In Boolean modeling, a gene’s expression or protein level is represented by binary code - 1 for active (on) or 0 for inactive (off) and hence that its products are present or absent. Although Boolean networks are efficient for analyzing large regulatory networks, in Boolean formalism, a gene is considered to be either on or off, and intermediate expression levels are ignored. Also, transitions between the activation states of the genes are assumed to occur synchronously. However, in many biological processes, transitions may not take place simultaneously [86].

Kinetic ordinary differential equation (ODE), assumes that molecular concentrations are continuous (it ignores the discrete nature of molecules), that reactions occur in a homogeneous, well-stirred volume and that these reactions occur in a deterministic manner. In an ODE model, a reaction network is expressed as a set of differential equations with one equation per chemical and with terms that represent the reactions [87]. However, a key challenge with ODE modeling is the lack of knowledge about kinetic parameters. The rate of change of product P with respect to time can be represented by following equation:

k1 k 2 S ⎯⎯→P ⎯⎯→Q

dP = k1[S]− k2[P] dt

Stochastic models incorporate randomness and may be closer to reality than continuous and deterministic approaches, however they are numerically more difficult to solve and computationally expensive. Flux Balance Analysis is often used for studying metabolic networks and it is most appropriate for identifying optimal fluxes for biomass production. 38

Cellular automata can be used to study change in spatial patterns or movement of cells over time [87].

2.4.3 Using Eigenvalues to determine system’s stability

Stability of an ODE (ordinary differential equations) system can be determined by eigenvalues derived from Jacobian matrix of that system. The Jacobian matrix summarizes how every variable changes with each of the other variables at a given point.

Mathematically, it is always a square matrix (i.e., number of rows of matrix = number of columns of matrix). The eigenvalues (λ) are complex numbers (of form a + bi) with real and imaginary part and inform about stability of the system. The imaginary part of an eigenvalue determines if the system will oscillate while the real part determines whether the amplitude of oscillations increases or decreases with time. Table 2-2 and Figure 2-4 summarize the different forms of eigenvalues and its impact on stability of the system. When λ has negative real part and zero imaginary part, the system decays exponentially; when all λ are real, and at least one is positive, the system increases exponentially. If λ has negative real part and nonzero imaginary part, the system exhibits damped oscillations; if λ has zero real part and nonzero imaginary part, the system shows sustained oscillations; for λ with at least one positive real part and some imaginary parts, the system will oscillate with increasing amplitude [88,89].

39

Eigenvalue (λ) Type Mathematical form for λ System’s Stability System’s Behavior

All λ are real and -a + 0i Stable Exponential decay negative

All λ are real and positive +a + 0i Unstable Exponential increase

All λ have negative real parts, some imaginary -a + bi Stable Damped oscillation parts At least one λ has positive real parts, some +a + bi Unstable Increase with oscillation imaginary parts All λ have zero real parts and nonzero imaginary 0 + bi Unstable Undamped oscillation parts

Table 2-2. Determination of system’s behavior by eigenvalues derived from Jacobian matrix of system.

Figure 2-4. Determination of system’s behavior by eigenvalues derived from Jacobian matrix of system [90]. 40

Chapter 3: Mathematical model for studying Helicobacter pylori mediated inflammation in host gastric tissue.

“A mathematical model is neither a hypothesis nor a theory. Unlike scientific hypotheses, a model is not verifiable directly by an experiment. For all models are both true and false.... The validation of a model is not that it is "true" but that it generates good testable hypothesis relevant to important problems.” - R. Levins, Am. Scientist 54:421-31, 1966

3.1 Synopsis

Helicobacter pylori infection of gastric tissue results in an immune response dominated by Th1 cytokines and has also been linked with deregulation of Sonic Hedgehog

(SHH) signaling pathway in gastric tissue. However, interactions between the cytokines and

SHH during H. pylori infection are not well understood. Here, we use mathematical modeling aided by restraints of experimental data to understand the temporal behavior of

H. pylori activated cytokine circuit. Statistical analysis of qPCR data from uninfected and H. pylori infected wild-type and parietal cell-specific SHH knockout (PC-SHHKO) mice for day 7 and day 180 suggest role of SHH in cytokine regulation. The experimentally observed changes are further investigated using a mathematical model that examines dynamic crosstalks among pro-inflammatory (IL1β, IL-12, IFNγ, MIP-2) cytokines, anti-inflammatory

(IL-10) cytokines and SHH during H. pylori infection. Analysis of the resulting model demonstrates that circuitry, as currently known, is inadequate for explaining of the experimental observations, suggesting the need for additional specific regulatory interactions. A key advantage of a computational model is the ability to propose putative 41

circuit models for in-silico experimentation. We use this approach to propose a parsimonious model that incorporates crosstalks between NFĸB, SHH, IL-1β and IL-10, resulting in a feedback loop capable of exhibiting cyclic behavior. Separately, we show that analysis of an independent time-series microarray data for IL-1β, IFNγ and IL-10 in mock and

H. pylori infected mice further supports the proposed hypothesis that these cytokines may follow a cyclic trend. Predictions from the in-silico model provide useful insights for generating new hypotheses and design of subsequent experimental studies.

The work described in this chapter is published as “Crosstalks between Cytokines and Sonic

Hedgehog in Helicobacter pylori Infection: A Mathematical Model.” [91].

3.2 Background

Helicobacter pylori is a significant risk factor for atrophic gastritis [92,93], gastric ulcer

[93,94] and gastric cancer [38,93,94]. The host cell detects the presence of bacteria and produces an immune response to eliminate the bacteria. The final outcome depends on a balance between pro-inflammatory and anti-inflammatory cytokines produced during H. pylori infection [60,95]. A strong pro-inflammatory response may allow eradication of the bacteria however at a cost of increased risk for gastritis [96] while anti-inflammatory cytokines may protect against gastritis, but H. pylori may continue to persist [97,98].

Another gene that has been recently identified to play an important role in pathogenesis of chronic H. pylori infection [99,100] and gastric cancer [101–104] is sonic hedgehog (SHH). In both human and mouse stomachs, it is expressed in the parietal cells [105]. Under normal conditions, SHH regulates differentiation of the gastric epithelium [105,106] and T-cell

42

[107,108]. During chronic H. pylori infection, SHH-dependent proliferation of parietal cells plays a key role in gastric mucosal repair [99,109]. It has been shown that SHH is involved in early immune response to H. pylori [110,111] and act as a monocyte/macrophage chemoattractant during H. pylori infection [111]. Schumacher et al have shown that parietal cell-specific SHH knock-out (PC-SHHKO) mice failed to develop gastritis, even after 6 months of H. pylori infection in contrast to infected control group (WT) which developed significant inflammatory response [111]. Another group has shown that overexpression of ShhWT induced gastritis while CMV-ShhF200H mice (carrying mutant SHH) did not develop gastritis.

They also reported that SHH overexpression exacerbated the histologic severity observed with Helicobacter felis infection and increased the amount of myeloid cells recruited to the inflamed stomach as compared to that in non-transgenic mice [110].

Although recent studies have highlighted the immunoregulatory role of SHH in stomach [110,111], a model studying temporal relationship between SHH and cytokines activated during H. pylori infection is still lacking, and it is unexplored what effect SHH may have in the context of regulating cytokine expression in the H. pylori infected stomach. Such temporal studies are not readily amenable to experimental approaches because of high cost and time associated with time series in-vivo experiments. In-vitro experiments of immune responses often face experimental limitations – for example, lack of a host immune cell response. Mathematical modeling is a powerful technique to complement such studies as it allows predicting dynamic behavior of the system under various perturbations and generating new hypotheses. Currently, there are only a handful of mathematical models that study the H. pylori - host immune response, and though they provide high-level or cellular-level details [112–115], they do not focus on quantifiable biomarkers involved. 43

3.3 Aim

To develop a mathematical model of Sonic Hedgehog and inflammatory cytokines, which are activated during Helicobacter pylori infection.

3.4 Methods

3.4.1 Animal Model

qPCR (Quantitative real-time Polymerase Chain Reaction) data from uninfected and

H. pylori infected mice, was received from Zavros lab. Briefly, the mouse model with parietal cell-specific deletion of SHH (PC-SHHKO) was generated as previously described (C57Bl/6,

129/Sv background) [106] and HKCre mice expressing Cre transgene under the control of

H+,K+-adenosine triphosphatase (ATPase) β subunit promoter (C57Bl/6, FVB/N background) were used as control [111]. HKCre (WT) and PC-SHHKO mice, aged 8 weeks were either infected with H. pylori or left uninfected. The uninfected group received 200 µl of Brucella broth over 3 consecutive days whereas the infected group was inoculated with 108 H. pylori

SS1 (Sydney strain 1) bacteria per 200 µl of Brucella broth over 3 consecutive days. Mice

(n=3-4 per group) were sacrificed on day 7 and 180 post-infection and levels of gastric SHH and cytokines were assayed by qPCR. H. pylori colonization as measured by bacterial cultures and analysis of CFU/g Tissue (colony-forming units per gram tissue) for WT and PC-

SHHKO infected mouse stomachs were shown to be equivalent [111]. Total RNA was isolated from stomachs of uninfected and infected WT and PC-SHHKO mice.

44

3.4.2 Interaction Map

An interaction map - a topological network of signaling pathways, activated by H. pylori, was manually curated from literature [116,117] (Figure 3-1). H. pylori virulence factors like

CagA [43,118,119], VacA [118] and PGN [118] activate a cascade of signaling pathways in epithelial cells of host stomach. Virulence factor CagA (cytotoxin-associated gene A) activates ERK [43,109,120] and AKT [121] pathways which further stimulate nuclear translocation of NFĸB [121–123]. Effector molecule PGN (peptidogylcan) is sensed by intracellular receptor NOD1 (nucleotide-binding oligomerization domain 1) [109,118] which activates NFĸB [109,118,119,124], ERK, p38 and AP-1 [124]. Protein VacA (Vacuolating cytotoxin A) also stimulates ERK and p38 pathways [125] which activate transcription factor

AP-1 [43,126–128]. Both NFĸB and AP-1 positively regulate chemokine IL-8/MIP-2 transcription [43,119,129]. (MIP-2 is a functional homolog of human IL-8 in mouse [130]). H. pylori colonization of gastric epithelium eventually leads to recruitment of monocytes which secrete pro-inflammatory cytokines like IL-12, IL-1β, TNFα, IL6 [131–133] and IL8 [134,135].

IL-12 and other H. pylori antigens synergistically stimulates release of IFNγ from natural killer cells [133,136–138]. SHH, a crucial player in early immune response to H. pylori [111] has been reported as a putative target of NFĸB in gastric [139] and pancreatic cancer

[140,141]. Interestingly, Waghray et al have shown that IL-1β suppresses SHH expression in parietal cells by inhibiting acid secretion [142]. Along with the activation of pro- inflammatory cytokines, IL-10, an anti-inflammatory cytokine is also produced during H. pylori infection [131,138,143]. IL-10 has been reported to inhibit NFĸB activity, and subsequently IL-8 transcription in gastric epithelial cells during H. pylori infection [144] and

45

also in macrophage [145] and T-cells [146]. In summary, different virulence factors of H. pylori activate various signaling pathways in the stomach tissue leading to increased activity or expression of NFĸB, chemokines and cytokines [147,148]. The severity of inflammation and exact details of signaling pathways described above can vary depending on the strain of

H. pylori and strain of mice used in the experiment [149,150].

Figure 3-1. Interaction Map of signaling pathways activated in host stomach in response to H. pylori.

H. pylori virulence factors (CagA, VacA and PGN, shown in orange) activate cascade of signaling pathways in host gastric epithelium that leads to nuclear translocation of NFĸB. NFĸB further activates IL8/MIP-2 and SHH. Immune response to the bacteria involves recruitment of monocytes to gastric epithelium where they secrete cytokines like IL-12, IL-1β, TNFα, IL6, IL10 and IL8. Blue arrows show activation while red lines represent inhibition. The network was built using Cytoscape using information based on current literature. However, the current knowledge does not inform about any role of SHH in regulation of cytokines as suggested by our analysis.

46

3.4.3 Statistical Analysis

For each sample, dCT value for target gene was calculated by subtracting CT value of calibrator gene (GAPDH) from CT value of target gene. To test if expression of target gene in gastric mucosa was affected by H. pylori and if this effect was same in all genotypes, a two- way ANOVA (Analysis of variance) test was performed, comparing dCT values from all four condition: uninfected wild type, infected wild type, uninfected PC-SHHKO and infected PC-

SHHKO. Bonferroni test was performed as post ANOVA test to assess the effect of H. pylori on target genes in each genotype. dCT values from uninfected mice were compared with dCT values from H. pylori infected mice in both WT and PC-SHHKO conditions. For graphical presentation of qPCR results, for each gene in WT and PC-SHHKO conditions, ddCT was calculated by subtracting dCT value of H. pylori infected group from dCT of uninfected group. The data were plotted as 2-ddCT (mean fold change). A P-value <0.05 was considered statistically significant. Interaction test was also performed to study the interaction between infection status and genotype. Since CT value is inversely proportional to gene’s expression level, negative dCT values were used for interaction test. The interaction plot displays levels of treatment (absence and presence of H. pylori) on the x-axis and mean of negative dCT values for each treatment on the y-axis. A separate line connects the means corresponding to each level of the trace factor – genotype (WT and PC-SHH-KO). The qPCR data was analyzed using R statistical software. Link to the R script used for analyzing the data and generating graphs: http://rpubs.com/marwahsi/20168.

47

3.4.4 Mathematical model

A comprehensive interaction map was developed by manual curation of published literature, capturing key signaling pathways activated in gastric epithelium and in macrophages by H. pylori virulence factors. For our mathematical model, a selected subset of biomolecules from the interaction map was included. The molecules were selected for inclusion if: a) they were reported in the literature to be regulated with H. pylori SS1 strain

[147,148] (as SS1 strain was used in our experiments) and b) their experimental trends were available in our data. Mathematical modeling of very large networks can be impractical, primarily due to unavailability of parameter values, especially in case of higher organisms

[116,151–153]. In order to overcome these challenges, the complex interaction map was pruned to a simpler reduced network Figure 3-2 focusing on cytokines and SHH, for which experimental trends were available [154]. The reduced model can capture key characteristics of the larger network and conserve the regulatory mechanisms present in the system [151–153]. Our reduced model compresses the details of a complex pathway into a

“black box”, which can be scaled up, by adding detailed components as needed. The reduced model was further enriched with a new “influence” link between SHH and cytokines as predicted by our experimental data. It must be noted that this link may not be a direct interaction between SHH and cytokines and rather may involve an indirect mechanism. This unknown mechanism is currently shown by an intermediate species “X” in the model which links SHH to all the cytokines. Equations based on Michaelis-Menten and mass action kinetics were used in the reduced model in order to construct a dynamic system that describes the evolution of the biomolecules over time. The sampling frequency (control

48

equaling time zero, day 7 equaling early, and day 180 equaling late) was unlikely to be sufficient for obtaining detailed dynamics of concentration and oscillation frequencies.

Therefore, unit-free measures were used to express “time” and “amount” in order to focus on the qualitative behavior of the cytokine dynamics. The kinetic parameters initially selected for the model, represent a range of biologically feasible values [155–157].

Subsequent computational optimization (iterative trial and error) was used to select the parameter set that best satisfies the trends observed in experimental data and are within biologically relevant limits. This approach is similar to the methodology used for parameter estimation for computational models when no experimental values are available

[84,156,158]. In addition to the reactions activated through H. pylori signaling, each model species is activated by a small “constitutive” flux which accounts for some basal levels of the species formed through a pathway either not represented in the model or currently unknown. Each model species is connected to a source and sink. The source represents inactive form of the protein. The sink accounts for downstream signaling and half-life of the protein. Complete list of model assumptions are provided in Table 3-2. Ordinary Differential

Equations (ODE) were simulated to characterize the system using modeling tool

CellDesigner [159]. Knock-out (KO) of different genes was modeled by setting the concentration of its source to zero. See Table 3-3, Table 3-4 and Table 3-5 for model equations and parameter values.

49

Figure 3-2. Graphical representation of mathematical model of cytokine-SHH network, during H. pylori infection.

This reduced network derived from interaction map, represents the key cytokines activated as host’s immune response to H. pylori. Blue arrows represent activation whereas while red arrows depict inhibition. Model species with suffix “i” represent the inactive form. The link between SHH and cytokines, as predicted by our experimental data is modeled through unknown model species “X” (grey colored).

3.4.5 Sensitivity Analysis

To ensure that the cyclic nature of our mathematical model is not depended on specific parameter and concentration values, but rather observed for a wide range of biologically feasible values, we performed sensitivity analysis. Jacobian matrix and eigenvalues for the model were calculated using simulation tool – Copasi [160]. The parameters and concentrations that had influence on the imaginary part of eigenvalues were selected and varied over a wide range of values. These parameter values were varied

(+- 50% or more) to assess the boundary value for the parameter beyond which the imaginary part of eigenvalues will be zero and the system moves towards stable point.

3.4.6 GEO Data

A time-series microarray dataset – GSE37938, available through GEO (Gene Expression

Omnibus) database was used as an auxiliary data set in order to analyze the temporal

50

behavior of the genes present in our model. The microarray data set contains six-week-old female BALB/c mice. The mice were uninfected or infected with SS1 strain of H. pylori for

2,7,14 & 28 days. A total of 71 samples (35 infected + 36 uninfected) were collected from 3 cell types: chief cell, parietal cell and mucous producing pit cell [161]. This dataset was selected because it used SS1 strain of H. pylori, which was also used as in our experimental data and this dataset contained data for parietal cells. We extracted and analyzed time- series data for IL-10, IFNγ, IL-1β from GSE37938 dataset. Data was not available for SHH and

MIP-2 while IL-12 data was not used for analysis due to large number of missing values. For a given probe, median of its expression values was calculated for each time point. Probes with two or more missing values for more than one time point, were not considered for further analysis. The median values of a probe from both uninfected and H. pylori infected mice were plotted against time to study their trajectory in the two conditions. To address the problem of probe selection in case of multiple probes representing the same gene, median value was taken for probes with same Genbank ID (GB_list ID) while probes with different Genbank ID were analyzed separately. Plots are shown only for Genbank IDs that are represented by at least three probes. Link to the R script used for extracting and analyzing the data: http://rpubs.com/marwahsi/8932

3.5 Results

3.5.1 SHH positively regulates cytokine expression during H. pylori infection

Interaction plots shown in Figure 3-3 examines if H. pylori affects cytokine expression and if this effect is different in WT and PC-SHHKO mice. Non-parallel lines that do not cross or

51

crossing lines, imply an interaction effect between the two factors (genotype and infection status) [162]. An interaction effect between genotype and infection was observed for expression of all cytokines. Parallel lines would have implied that the effect of H. pylori on cytokine expression is the same in WT and PC-SHHKO mice. On day 7, the interaction effect was statistically significant for IL-12 and IL-10 but not for IL-1β and MIP-2. For day 180, statistically significant interaction effect between genotype and infection was observed for all cytokines.

Two-way ANOVA test was performed, comparing dCT values from all four conditions

(WT+ Brucella Broth, WT + H. pylori, PC-SHHKO + Brucella Broth, PC-SHHKO + H. pylori), followed by Bonferroni test to compare specific groups (Figure 3-4). All cytokines showed a statistically significant increase with H. pylori in WT mice, on both day 7 and day 180 (except

MIP-2 on day 7, which increased with a P-value of 0.05). However, this increase in cytokines’ expression with H. pylori was not observed in PC-SHHKO mice. These observations suggest the potential role of SHH in positive regulation of expression of IL-12, IL-1β, IL-10, IFNγ and

MIP-2 during H. pylori infection. An unexpected trend was observed for IL-10 on day 180 in

PC-SHHKO mice - IL-10 expression was significantly lower in H. pylori infected mice as compared with uninfected mice on day 180, although it did not change significantly with H. pylori on day 7.

It is interesting to note that our analysis of experimental data indicates a positive regulation of the cytokines’ expression involving SHH. However, current pathway databases

(for example Strings, Pathway Commons), and molecular circuits (wiring diagram of interactions among genes and proteins) derived through manual curation of current literature, fail to identify the positive regulation. 52

Figure 3-3. Interaction between infection status and genotype.

RNA was extracted from stomachs of uninfected (-HP) and H. pylori-infected (+HP) wild type (WT) and parietal cell specific SHH knock out (PC-SHH-KO) mice 7 and 180 days post-inoculation. Expression of genes was measured by qPCR and interaction test was performed. Parallel lines would imply that H. pylori has same effect on gene’s expression in WT and PC-SHHKO mice whereas intersecting or non-parallel lines would indicate an interaction between genotype and infection. The graphs show interaction plot between infection status and genotype for (A) IL1β on day 7, (B) IL1β on day 180, (C) IL-12 on day 7, (D) IL-12 on day 180, (E) MIP-2 on day 7 (F) MIP-2 on day 180 (G) IL-10 on day 7 and (H) IL-10 on day 180. P-value for interaction between infection and genotype were calculated by two-way ANOVA test. Y-axis: mean of negative dCT value of cytokine, X-axis: infection status, trace-factor: genotype. 53

Figure 3-4. Effect of H. pylori on SHH and cytokines's expression in WT and PC-SHHKO mouse stomachs, day 7 and day 180 post-inoculation.

RNA was extracted from stomachs of uninfected and H. pylori-infected wild type (WT) and parietal cell- specific SHH knock out (PC-SHH-KO) mice 7 and 180 days post-inoculation. Expression of genes was measured by qPCR and two-way ANOVA test was performed, followed by Bonferroni test to compare uninfected (-HP) with H. pylori infected group (+HP) in each genotype. The graphs show average fold change in expression of IL-1β (A, B), IL-12 (C, D), MIP-2 (E, F), IL-10 (G, H) and SHH (I, J) upon H. pylori infection relative to uninfected condition. Bars represent the mean ±SEM, n=3-4 per group. 54

3.5.2 Mathematical Model Behavior

An important goal of model building is to ensure that the model can recapitulate the experimental trends. SHH knock-out condition was tested in the model that is curated using current pathway databases and literature. This model lacks the putative positive regulation link (kcat9=0) suggested by our qPCR analysis. The model exhibited no change in levels of IL-

12, IL-10, IL-1β, IFNγ or MIP-2 upon simulating SHH KO condition. In our model with the putative link, all cytokines: IL-12, IL-1β, IFNγ, IL-10 and MIP-2 decreased with SHH knock-out

(Figure 3-5), aligning the model closer to the experimental results. The decrease in MIP-2 was not considerable as MIP-2 is also formed through a parallel pathway via NFĸB.

55

Figure 3-5. In-silico SHH KO results show a decrease in cytokines as compared to WT.

SHH KO condition was simulated by setting SHHi to zero. Graph A-F shows profiles of (A) SHH (B) IL-1β (C) IL12 (D) IFNγ (E) MIP2 (F) IL10. Wild type condition (SHHi=1) is shown in green and in-silico SHH KO condition (SHHi=0) is represented in red.

56

Next, we assessed the effect of H. pylori on the temporal profile for all model species in-silico. The simulation results align qualitatively with the experimental trends of cytokines.

All cytokines increased with switching H. pylori “on” in the model. Temporal profile for key model species is compared in the presence and absence of H. pylori in Figure 3-6. Dynamic modeling of the reduced network shows that with H. pylori stimulation, there is increase in

NFĸB which transcribes MIP-2 and SHH. SHH activates IL-1β, IL-12 and IL-10 through unknown mechanism (shown through “X” in the model). IL-10 in turn inhibits NFĸB, forming a negative feedback loop. This inhibition decreases NFĸB, which lowers SHH and eventually decreases IL-10 and other cytokines. The decrease in IL-10 relieves the inhibition on NFĸB, allowing it to rise again. These interactions between NFĸB, SHH and IL-10 form a negative feedback loop and give rise to damped oscillations in the model (Figure 3-6 B).

To understand the role of the anti-inflammatory cytokine IL-10 in our model circuit, we simulated overexpression and knock-out IL-10 conditions in the model. Overexpression of IL-10 led to decrease in NFĸB and in all the cytokines examined while knock-out of IL-10 in the model relieved the inhibition on NFĸB and resulted in increase in NFĸB and in pro- inflammatory cytokines downstream (Figure 3-7). Our in-silico outcomes qualitatively matched published results of similar experiments. Robinson et al studied the effect of IL-10 in gastric epithelial cells (AGS cell line) [144] where they showed that addition of recombinant human IL-10 to co-cultures of H. pylori strain 60190 with AGS cells caused decrease in nuclear NFĸB and IL-8. Similar decreasing trends in NFĸB and MIP-2 (murine homologue of human IL-8) were observed with increase of IL-10 in our model (Figure 3-7 A,

B). Knock-out of IL-10 in our model relieved the inhibition on NFĸB and resulted in increase

57

in NFĸB and in pro-inflammatory cytokines downstream. Figure 3-7 C shows increase in pro- inflammatory cytokine IFNγ with IL-10 KO as compared to that in WT, thus suggesting the potential protective role of IL-10 against H. pylori mediated gastritis. Bodger et al have also suggested that IL-10 secretion during H. pylori infection may serve a protective role, reducing local tissue damage caused by inflammation [163]. Ismail et al have shown that

Helicobacter felis infection in IL-10 knock-out in mice resulted in a higher inflammation score and severe gastritis as compared to that in infected wild type mice [97]. They also cultured splenocytes from control uninfected and H. felis-infected WT and IL-10 KO mice, with sonicated H. felis Ag and the culture supernatants were evaluated for the concentration of IFN-γ. Splenocytes from H. felis-infected WT mice produced low levels of

IFN-γ while splenocytes from H. felis-infected IL-10−/− mice produced large amounts of IFN-γ

[97].

58

Figure 3-6. Temporal profiles of model species in uninfected and infected conditions.

Simulation results comparing temporal profiles of model species in (A) absence and (B) presence of H. pylori. Graph C-F show temporal profiles of (C) SHH (D) MIP-2 (E) IL-1β and (F) IL-10 in absence and presence of H. pylori.

59

A B

C D

E F

Figure 3-7. In-silico IL-10 knock-out and overexpression.

Effect of IL-10 knock-out and overexpression on (A) NFkB (B) MIP-2 and (C) IFNγ (D) IL-1β (E) IL-12 and (F) SHH. Wild type condition (IL10i=1) is shown in black, in-silico IL-10 knok-out (IL10i=0) in red and in- silico IL-10 overexpression (IL10i=10) in green.

60

3.5.3 Temporal Analysis of GEO Data

We also analyzed a microarray time-series dataset, available through the GEO database.

This data allowed us to perform a separate investigation of the temporal profile of IL-1β, IL-

10 and IFNγ for both mock-infected and H. pylori infected mice using the alternative technology of microarrays. Figure 3-8 shows the trajectory of the three cytokines in chief cells, parietal cells, and pit cells, for day 2, 7, 14 and 28. This time-based analysis of the data further suggests that these cytokines show a cyclic behavior as opposed to a linear trend.

3.5.4 Sensitivity and Stability Analysis

Steady state or oscillatory behavior of a system that is defined by a set of ordinary differential equations can be determined by eigenvalues derived from Jacobian matrix of that system Table 2-2. We used the above mathematical principles to investigate the behavior of our model. The damped oscillatory nature of our model, for example, depends on the negative inhibition of NFĸB by IL-10 and it is deduced from the complex eigenvalues

(with negative real part) of the system’s response. Removal of this inhibitory link in the model results in steady state behavior (Figure 3-9) - which can be mathematically identified with the vanishing imaginary part of the eigenvalues. A sensitivity analysis is also performed to find the range of parameters across which the model displays damped oscillations. These results indicate that the cyclic behavior of the model remains viable within a large, biologically feasible, parameter region and is not limited to specific parameters and concentration values. The results presented in Table 3-1 show that the model exhibits oscillatory trend even when original value of key parameters are increased or decreased by

50%.

61

3.6 Key Findings

• SHH positively regulates expression of cytokines like IL-12, IL-1β, MIP-2, IL-10, however

the mechanism remains unclear.

• Our model suggests that NFĸB, SHH and the cytokines engage in a feedback loop which

can result in damped oscillations.

• We use the mathematical model as a tool to gain insights into cytokine and SHH

relationship during H. pylori infection and as a hypotheses generating tool to predict host

responses that may be associated with gastric disease or clinical treatments that may

provide a better outcome.

62

63

Figure 3-8. Trajectories of cytokines in mock-infected and H. pylori infected mice from GEO microarray dataset. A) IL-10, (B) IL-1β and (C) IFNγ for day 2, 7, 14 and 28 from chief, parietal and pit cell of mock- infected and H. pylori infected mice. The temporal profiles indicate that these cytokines potentially display a cyclic expression pattern in response to H. pylori infection.

Figure 3-9. Temporal profiles of model species, in absence of negative feedback on NFĸB by IL-10.

The damped oscillatory nature of the model depends on the negative inhibition of NFĸB by IL-10. Removal of this inhibitory link in the model results in steady state behavior.

Table 3-1. Sensitivity Analysis of model parameters for damped oscillations. Key parameters and their range for which the model shows damped oscillations.

Parameter Name (re20) k1 (re23) k1 (re18) k1 (re25) k1 Hpylori

Parameter Range 0.005 to 0.5 0.02 to 0.2 0.005 to 0.5 0.01 to 0.2 1 to 100

64

Table 3-2 Model Assumptions

• Model compresses the details of a complex pathway into a “black box”, which can be

scaled to larger, detailed components when needed. For example activation (nuclear

transport) of NFĸB involves activation of ERK, AKT pathways by H. pylori virulence factors

that includes many steps, has been compressed to a single step. Similarly, cellular details

of transcription, translation and post translation modifications are not shown. The focus

of this model is to provide a high level understanding of cytokine-SHH circuit during H.

pylori infection.

• Equations were based on Michaelis Menten and mass action kinetics.

• “Time” and “concentration” are expressed in arbitrary units as we do not have enough

experimental data to confidently measure proteins’ concentration or frequency of

oscillations.

• Kinetic parameters initially selected for the model, represented a range of biologically

feasible values [155–157]. Subsequent computational optimization (iterative trial and

error) was used to select the parameter set that best satisfies the trends observed in

experimental data and are within biologically relevant limits.

• The interaction map focuses on key signaling pathways activated in gastric epithelium

and in macrophages, by H. pylori virulence factors. Pathways in other immune cells are

currently not included but in future it will be of great value to include them in both

interaction map and model.

• A constant, steady state level of H. pylori is shown in current model. 65

Table 3-3 Mathematical equations used in the model.

Reaction Id Reactants Products Modifiers Equation

Hpylori, kcat1 * Hpylori * NFKBi / (km1 * (1 + IL10 re1 NFKBi NFKB IL10 / Ki1) + NFKBi)

re2 NFKBi NFKB v1 * NFKBi / (k1 + NFKBi)

re3 NFKB NFKBi v1 * NFKB / (k1 + NFKB)

re4 MIP2i MIP2 v1 * MIP2i / (k1 + MIP2i)

re5 MIP2i MIP2 NFKB kcat5 * NFKB * MIP2i / (km5 + MIP2i)

re6 MIP2i MIP2 X kcat6 * X * MIP2i / (km6 + MIP2i)

kcat7 * NFKB * SHHi / (km7 * (1 + IL1b / re7 SHHi SHH NFKB, IL1b Ki7) + SHHi)

re8 SHHi SHH v1 * SHHi / (k1 + SHHi)

re9 Xi X SHH kcat9 * SHH * Xi / (km9 + Xi)

re10 IL1bi IL1b X kcat10 * X * IL1bi / (km10 + IL1bi)

re11 IL1bi IL1b v1 * IL1bi / (k1 + IL1bi)

re12 IL12i IL12 X kcat12 * X * IL12i / (km8 + IL12i)

re13 IL12i IL12 v1 * IL12i / (k1 + IL12i)

re14 IL10i IL10 X kcat14 * X * IL10i / (km14 + IL10i)

re15 IL10i IL10 v1 * IL10i / (k1 + IL10i)

66

IL12, kcat16 * Hpylori * IL12 * IFNgi / (km16 + re16 IFNgi IFNg Hpylori IFNgi)

re17 IFNgi IFNg v1 * IFNgi / (k1 + IFNgi)

re18 NFKB Sink NFKB * k1

re19 MIP2 Sink MIP2 * k1

re20 SHH Sink SHH * k1

re21 IL1b Sink IL1b * k1

re22 IL12 Sink IL12 * k1

re23 IL10 Sink IL10 * k1

re24 IFNg Sink IFNg * k1

re25 X Sink X * k1

67

Table 3-4 Species’ parameters used in the model.

Species Name Initial Quantity Constant

Hpylori 10 TRUE

NFKB 0.1 FALSE

NFKBi 1 TRUE

MIP2 0.1 FALSE

MIP2i 1 TRUE

IL1b 0.1 FALSE

SHHi 1 TRUE

SHH 0.1 FALSE

IL10 0.1 FALSE

X 0 FALSE

IL1bi 1 TRUE

Xi 1 TRUE

IL10i 1 TRUE

IL12i 1 TRUE

IL12 0.1 FALSE

IFNgi 1 TRUE

IFNg 0.1 FALSE

Sink 0 TRUE

68

Table 3-5 Kinetic parameters used in the model.

Reaction Id Parameter Name Value

(re1) km1 0.521

(re1) kcat1 0.516

(re1) Ki1 0.702

(re2) v1 0.2

(re2) k1 0.5

(re3) v1 0.5

(re3) k1 1

(re4) v1 0.1

(re4) k1 2

(re5) kcat5 0.1

(re5) km5 2

(re6) kcat6 0.7

(re6) km6 4

(re7) kcat7 1.01

(re7) km7 16.25

(re7) Ki7 3.98

(re8) v1 0.01

69

(re8) k1 8

(re9) kcat9 0.211

(re9) km9 1.03

(re10) km10 10

(re10) kcat10 0.5

(re11) v1 0.1

(re11) k1 10

(re12) km8 8

(re12) kcat12 0.2

(re13) v1 0.05

(re13) k1 8

(re14) kcat14 0.253

(re14) km14 20.1

(re15) v1 0.1

(re15) k1 20

(re16) kcat16 0.03

(re16) km16 15

(re17) v1 0.1

(re17) k1 10

70

(re18) k1 0.1

(re19) k1 0.2

(re20) k1 0.05

(re21) k1 0.05

(re22) k1 0.04

(re23) k1 0.04

(re24) k1 0.015

(re25) k1 0.04

71

Chapter 4: Meta-analysis of gastric cancer and H. pylori mediated gastritis microarray data.

4.1 Background

Although the majority of H. pylori infected individuals remain symptom-free, infection by this bacteria is considered to be the the most primary cause of chronic gastritis and a high risk factor for peptic ulcer and gastric cancer [50,51]. Meta-analyses based on epidemiologic studies suggest that the presence of H. pylori increases the risk for development of gastric cancer by 2-fold to 5-fold [38,53], whereas more sensitive and strain-specific studies suggest a 20-fold increased risk [16,164]. H. pylori-infected gastric mucosa evolves through stages of chronic gastritis, glandular atrophy and intestinal metaplasia which is a main cause of intestinal type of gastric adenocarcinoma [7]. Atrophic gastritis, accompanied with loss of parietal cells and hypochlorhydria, is highly associated with development of gastric cancer [165]. However, the molecular details that link H. pylori mediated atrophic gastritis and gastric cancer are not well understood. Identification of genes and pathways which are dis-regulated during gastric cancer and H. pylori mediated atrophic gastritis will enhance our current knowledge about the molecular details which link the two diseases.

72

4.2 Aim

To identify the molecular signature that is common to gastric cancer and H. pylori

mediated gastric atrophy.

4.3 Methods

4.3.1 Datasets:

Microarray data comparing healthy vs. disease tissue was obtained from the Gene

Expression Omnibus (GEO) [166] database Table 4-1. Gse27411 [165] contains data from

patients who were admitted for endoscopy due to dyspepsia, malabsorption or anemia.

Patients not suffering from extragastric malignancy or inflammatory disease were selected

and classified into two groups:

i. without current H. pylori infection (HP- ,n=6). ii. with current or past H. pylori-infection and corpus-predominant atrophic gastritis (Atr,

n=6).

From these patients, biopsies from the antrum and corpus mucosa were obtained during

endoscopy. GSE27342 and GSE13861 with sample size of 160 and 84 respectively represent

the human gastric cancer expression datasets analyzed in this study.

73

Table 4-1. Sample size of GEO datasets used in the study. GSE27411 represents the H. pylori mediated atrophic gastritis dataset while GSE27342 and GSE13861 are the gastric cancer datasets.

Sample size (Stomach tissue) GEO ID Treatment Control

GSE27411 6 HP- 6 Atr

GSE27342 80 Tumor 80 Adjacent Normal

GSE13861 65 Tumor 19 Adjacent Normal

4.3.2 Differential Gene Expression Analysis

Differential expression analysis for each dataset was performed in R version 3.12

(programming language and environment for statistical computing) using Limma package

[167] from Bioconductor. Limma fits a linear model to the expression data for each gene. It uses emperical Bayes method to moderate the standard errors of the estimated log fold changes. Benjamini-Hochberg test was used for multiple correction and genes with False

Discovery Rate (FDR) < 0.05 were considered to be differentially expressed. For Gse27411, corpus and antrum samples were analyzed separately as they represent different tissues; different genes are preferentially expressed in antrum and corpus tissue [165]. Since

Gse27342 represents paired data, paired limma test was used for its analysis. Functional enrichment of Differentially Expressed Genes (DEG), common among the 3 datasets was performed by submitting entrez ids of DEG to the ToppFun tool [168].

74

4.3.3 Connectivity Map Analysis

Entrez gene identifiers of Differentially Expressed Genes (DEG) from the microarray datasets were mapped to Affymetrix HG-U133A probesets using hgu133a.db package [169] in R. The DEG common across the three datasets were used as input query for Connectivity

Map (CMap) version 2 [170] analysis. Compounds with negative enrichment scores and permutation P-value < 0.05 were selected as a basis for to identifying compounds which can maximally reverse the gene signature.

4.4 Results

The number of Differentially Expressed Genes (DEG) in each dataset is summarized in

Table 4-2. To identify genes that are differentially expressed during H. pylori mediated atrophic gastritis, Atrophy group (Atr) was compared to uninfected (HP-) group from

GSE27411. In the corpus tissue, 684 genes were observed to be up-regulated and 665 genes to be down-regulated during H. pylori mediated atrophic gastritis as compared to uninfected corpus. However, there were no DEG observed in the antrum samples (between

HP- and Atr (Atrophy patients)) because for most of the patients in this dataset, atrophy was predominantly in the corpus tissue and usually absent or mild in the antrum region. Venn diagrams of DEG common among the gastric cancer and H. pylori mediated atrophic gastric are shown in Figure 4-1. There were 57 up-regulated genes and 86 down-regulated genes that were differentially expressed in all the three datasets. Functional Enrichment of DEG

75

common among the three datasets was performed using ToppFun tool. Topmost biological processes and pathways enriched for the common DEG are shown in Table 4-3.

Table 4-2. Number of Differentially Expressed Genes (DEG) in gastric cancer and H. pylori mediated atrophic gastritis datasets. n = sample size.

GEO ID Number of Differentially Expressed Genes (DEG)

Up-regulated Down-regulated

Gse273411 – Corpus (n=6) 684 665

Gse273411 – Antrum (n=6) 0 0

Gse27342 (n=160) 652 341

Gse13861 (n=84) 3366 2329

76

Up-regulated Genes Down-regulated Genes

Figure 4-1. Venn diagram showing intersection between Differentially Expressed Genes (DEG) from gastric cancer and H. pylori mediated atrophic gastritis.

4.5 Discussion

Genes like claudins that were up-regulated in both gastric cancer and H. pylori mediated atrophic gastritis are involved in pathways related to cell-cell adhesion, cell-cell communication, tight junctions and leukocyte transendothelial migration. An independent microarray study (not used in our meta-analysis) has compared gastric adenocarcinoma with adjacent normal tissues from 20 patients and it has demonstrated that claudin-1

(CLDN1) is one of the most consistently up-regulated genes in the tumors. The same group has reported CLDN1 as marker for poor post-operative prognosis and has suggested that up-

77

regulated genes were mapped to cell-adhesion and collagen-related processes [171]. In another study, Tsai et al analyzed microarray data from a randomized, placebo-controlled trial of H. pylori therapy. In their study, 54 gastric biopsies were obtained from 27 subjects

(13 from the treatment and 14 from the placebo group) with chronic gastritis, atrophy, and/or intestinal metaplasia. Each subject contributed one biopsy before and another biopsy 1 year after the intervention. S100A10, a gene involved in cell cycle differentiation and protein hetero-tetramerization, was reported to increase over time in placebo group and decrease in treatment group [172]. An independent investigation by another group has shown that S100A10 is overexpressed in gastric cancer [173]. EPCAM (epithelial cell adhesion molecule), CDH17 (cadherin 17) and ANXA2 (annexinA2) are among other genes that were predicted by our analysis to be overexpressed in both gastric cancer and H. pylori mediated gastric atrophy. EPCAM [174], CDH17 [175,176] and ANXA2 [177] have also been reported to be up-regulated in gastric cancer.

Our analysis shows that genes involved in gastric acid secretion, potassium ion transport and creatine pathways were down-regulated in both gastric cancer and corpus predominant gastric atrophy associated with H. pylori infection. It has been previously reported that in subjects with predominant body (corpus) gastritis, acid secretion is impaired and the subjects have an increased risk of developing gastric cancer [178,179].

Qing et al have also shown that creatine kinase – brain (CKB) is under-expressed in gastric cancer cells. CKB is expressed in parietal cells and couples functionally to H-K-ATPase to effectively provide ATP for proton pumping [180]. Complete list of genes, which were

78

differentially expressed in both gastric cancer and H. pylori mediated gastric atrophy are summarized in Table 4-6 and Table 4-7.

Table 4-3. Enriched GO processes and pathways in DEG common to gastric cancer and H. pylori mediated gastritis datasets.

q-value Pathways Hit in Up-regulated Gene List (Bonferroni)

1.76E-05 Tight junction interactions CLDN4,CLDN3,CLDN7,CLDN2,CLDN1

Cell-cell junction organization 5.71E-04 CLDN4,CLDN3,CLDN7,CLDN2,CLDN1

Leukocyte transendothelial 1.71E-02 CLDN4,CLDN3,CLDN7,CLDN2,CLDN1 migration

Cell-Cell communication CLDN4,CLDN3,CLDN7,CLDN2,CLDN1 2.81E-02

Hepatitis C CLDN4,CLDN3,CLDN7,CLDN2,CLDN1 3.02E-02

q-value Pathways Hit in Down-regulated Gene List (Bonferroni)

KCNE2,SLC26A7,KCNJ15,KCNQ1,CHRM3,ATP4A,ATP4B,CC 2.39E-06 Gastric acid secretion KBR

creatine-phosphate CKB,CKM,CKMT2 2.53E-04 biosynthesis

Pirenzepine, Omeprazole, CHRM3,ATP4A,ATP4B,CCKBR 7.34E-04 Ranitidine Pathway

2.18E-03 Creatine pathway CKB,CKM,CKMT2

Urea cycle and metabolism of CKB,CKM,CKMT2 4.08E-02 amino groups

q-value Biological Process Hit in Up-regulated Gene List (Bonferroni)

calcium-independent cell-cell CLDN4,CLDN3,CLDN7,CLDN2,CLDN1 1.71E-05 adhesion

79

DSG3,CLDN4,CLDN3,CLDN7,CLDN2,CLDN1,CDH17,CEACA 1.84E-04 cell-cell adhesion M1

S100A10,PDSS1,ANXA2 1.50E-02 protein heterotetramerization

cell adhesion TGFBI,S100A10,IL32,DSG3,OLFM4,CLDN4,CLDN3,CLDN7,E 1.75E-02 PCAM,CLDN2,CLDN1,CDH17,CEACAM1

biological adhesion TGFBI,S100A10,IL32,DSG3,OLFM4,CLDN4,CLDN3,CLDN7,E 1.93E-02 PCAM,CLDN2,CLDN1,CDH17,CEACAM1

q-value Biological Process Hit in Down-regulated Gene List (Bonferroni)

1.38E-04 creatine metabolic process CKB,GHR,CKM,CKMT2

cellular potassium ion 1.71E-02 transport KCNMB2,KCNE2,HPN,KCNJ15,KCNQ1,ATP4A,ATP4B

potassium ion transmembrane KCNMB2,KCNE2,HPN,KCNJ15,KCNQ1,ATP4A,ATP4B 1.71E-02 transport

response to endogenous CA9,MICALL1,GHR,DUOX1,IGFBP2,GPER1,NRG4,HPN,ESRR 2.76E-02 stimulus G,ANGPTL3,KCNQ1,NCAM1,APLP1,CAB39L,GCNT2,AQP4,S

LC1A2,HOMER2,ATP4B

phosphocreatine metabolic CKM,CKMT2 3.45E-02 process

To identify potential compounds that can maximally reverse the common gene signature identified above, Broad Institute’s Connectivity Map (CMap) was used.

Connectivity Map is a collection of microarray expression data from cancer cell lines treated with bioactive small molecules. CMap portal accepts disease signature (lists of up-regulated and down-regulated genes) as an input and uses a pattern-matching algorithm to compare it against reference gene signature of each compound. CMap assigns a positive enrichment

80

score to a compound that has an expression signature similar to that in the disease, and a negative score to a perturbagen that has an expression signature opposite to that in the disease. Probes corresponding to the 57 upregulated and 86 downregulated genes from our study were used as input query for CMap. The top ranked compounds are shown in Table 4-

4. The role of these compounds in gastric cancer is summarized in Table 4-5.

4.6 Limitations

Gastric cancer and gastric atrophy data were obtained from different patients and thus the identified genes are subject to heterogeneity across population.

4.7 Conclusions

Using genomics data, we have tried to identify the molecular signature and pathways, which are common between gastric cancer and H. pylori mediated atrophic gastritis. Further analysis of this gene signature can help us identify the underlying molecular mechanism through which H. pylori mediated chronic gastritis increases the risk for development of gastric cancer.

81

Table 4-4. Top ranked compounds with expression profiles opposite to that of common DEG signature identified from gastric cancer and H. pylori mediated atrophic gastritis.

rank cmap name mean n enrichment p

1 vorinostat -0.691 12 -0.752 0

2 trichostatin A -0.344 182 -0.356 0

3 nalbuphine -0.517 5 -0.761 0.00146

5 pargyline -0.641 4 -0.816 0.00209

6 alpha-estradiol -0.412 16 -0.441 0.00233

7 levamisole -0.698 4 -0.8 0.0031

8 xylometazoline -0.598 4 -0.797 0.00338

11 cefsulodin -0.418 4 -0.775 0.00523

12 piperlongumine -0.829 2 -0.951 0.00531

13 DL-thiorphan -0.786 2 -0.949 0.00561

Table 4-5. The role of compounds predicted by CMap in gastric cancer.

Implications in Approved Implications in Compound MOA / Target other cancers & Implication Gastric cancer other diseases inhibits the Clinical trials: cutaneous T cell enzymatic Glioblastoma In-vitro : Yes lymphoma activity of multiforme (GBM), vorinostat Clinical Trials: (CTCL) - skin histone advanced non- Yes cancer deacetylases small-cell lung HDAC cancer Inhibitor of Antifungal Agent histone In-vitro: Yes trichostatin A - In-vivo: breast deacetylase In-vivo: Yes cancer (HDAC) activity

82

Kappa-type nalbuphine pain medication opioid receptor - - agonist monoamine pargyline antihypertensive - - oxidase B 17 beta-estradiol has protective role weakly binds to alpha-estradiol against development of Helicobacter - estrogen pylori-induced gastric cancer in INS- receptors GAS mice. restore levamisole Clinical Trials: Antihelminthic depressed Colorectal Cancer Gastric Cancer immune function increases In-vitro: GBM, in- reactive oxygen vivo: B-lymphoma, piperlongumine - species (ROS) - breast, lung and apoptotic cancer cell death

83

Table 4-6. Genes overexpressed in gastric cancer and Helicobacter pylori mediated gastric atrophy datasets.

Entrez Gene ID Gene Symbol Gene Name

7045 TGFBI transforming growth factor, beta-induced, 68kDa

6281 S100A10 S100 calcium binding protein A10

220042 DDIAS DNA damage-induced apoptosis suppressor

113802 HENMT1 HEN1 methyltransferase homolog 1 (Arabidopsis)

90381 TICRR TOPBP1-interacting checkpoint and replication regulator

126353 MISP mitotic spindle positioning

346389 MACC1 metastasis associated in colon cancer 1

2706 GJB2 gap junction protein, beta 2, 26kDa

9235 IL32 interleukin 32

1044 CDX1 caudal type homeobox 1

5268 SERPINB5 serpin peptidase inhibitor, clade B (ovalbumin), member 5

10903 MTMR11 myotubularin related protein 11

26 AOC1 amine oxidase, copper containing 1

55711 FAR2 fatty acyl CoA reductase 2

83879 CDCA7 cell division cycle associated 7

1830 DSG3 desmoglein 3

23590 PDSS1 prenyl (decaprenyl) diphosphate synthase, subunit 1

2984 GUCY2C guanylate cyclase 2C

430 ASCL2 achaete-scute family bHLH transcription factor 2

302 ANXA2 annexin A2

196410 METTL7B methyltransferase like 7B

51645 PPIL1 peptidylprolyl isomerase (cyclophilin)-like 1

3904 LAIR2 leukocyte-associated immunoglobulin-like receptor 2

80704 SLC19A3 solute carrier family 19 (thiamine transporter), member 3

10562 OLFM4 olfactomedin 4

1604 CD55 CD55 molecule, decay accelerating factor for complement (Cromer blood group)

55365 TMEM176A transmembrane protein 176A

84

5445 PON2 paraoxonase 2

79814 AGMAT agmatine ureohydrolase (agmatinase)

80201 HKDC1 hexokinase domain containing 1

carcinoembryonic antigen-related cell adhesion molecule 6 (non-specific cross 4680 CEACAM6 reacting antigen)

79949 PLEKHS1 pleckstrin homology domain containing, family S member 1

717 C2 complement component 2

1871 E2F3 E2F transcription factor 3

286676 ILDR1 immunoglobulin-like domain containing receptor 1

1364 CLDN4 claudin 4

1365 CLDN3 claudin 3

1366 CLDN7 claudin 7

91862 MARVELD3 MARVEL domain containing 3

1111 CHEK1 checkpoint kinase 1

344 APOC2 apolipoprotein C-II

4312 MMP1 matrix metallopeptidase 1 (interstitial collagenase)

9052 GPRC5A G protein-coupled receptor, class C, group 5, member A

4192 MDK midkine (neurite growth-promoting factor 2)

9700 ESPL1 extra spindle pole bodies homolog 1 (S. cerevisiae)

4072 EPCAM epithelial cell adhesion molecule

283375 SLC39A5 solute carrier family 39 (zinc transporter), member 5

3692 EIF6 eukaryotic translation initiation factor 6

55789 DEPDC1B DEP domain containing 1B

9843 HEPH hephaestin

9075 CLDN2 claudin 2

9076 CLDN1 claudin 1

4982 TNFRSF11B tumor necrosis factor receptor superfamily, member 11b

1015 CDH17 cadherin 17, LI cadherin (liver-intestine)

634 CEACAM1 carcinoembryonic antigen-related cell adhesion molecule 1 (biliary glycoprotein)

85

Table 4-7. Genes under-expressed in gastric cancer and Helicobacter pylori mediated gastric atrophy datasets.

Entrez Gene ID Gene Symbol Gene Name

1152 CKB creatine kinase, brain

768 CA9 carbonic anhydrase IX

85377 MICALL1 MICAL-like 1

2690 GHR growth hormone receptor

10242 KCNMB2 potassium channel subfamily M regulatory beta subunit 2

1158 CKM creatine kinase, muscle

2694 GIF gastric intrinsic factor (vitamin B synthesis)

9992 KCNE2 potassium channel, voltage gated subfamily E regulatory beta subunit 2

1160 CKMT2 creatine kinase, mitochondrial 2 (sarcomeric)

3977 LIFR leukemia inhibitory factor receptor alpha

285704 RGMB repulsive guidance molecule family member b

5005 ORM2 orosomucoid 2

124817 CNTD1 cyclin N-terminal domain containing 1

53905 DUOX1 dual oxidase 1

130576 LYPD6B LY6/PLAUR domain containing 6B

137872 ADHFE1 alcohol dehydrogenase, iron containing, 1

5909 RAP1GAP RAP1 GTPase activating protein

1558 CYP2C8 cytochrome P450, family 2, subfamily C, polypeptide 8

284434 NWD1 NACHT and WD repeat domain containing 1

4118 MAL mal, T-cell differentiation protein

27159 CHIA chitinase, acidic

222235 FBXL13 F-box and leucine-rich repeat protein 13

146456 TMED6 transmembrane emp24 protein transport domain containing 6

3354 HTR1E 5-hydroxytryptamine (serotonin) receptor 1E, G protein-coupled

401052 LOC401052 uncharacterized LOC401052

80157 CWH43 cell wall biogenesis 43 C-terminal homolog (S. cerevisiae)

3485 IGFBP2 insulin-like growth factor binding protein 2, 36kDa

2852 GPER1 G protein-coupled estrogen receptor 1

86

115111 SLC26A7 solute carrier family 26 (anion exchanger), member 7

5799 PTPRN2 protein tyrosine phosphatase, receptor type, N polypeptide 2

221476 PI16 peptidase inhibitor 16

145957 NRG4 neuregulin 4

79917 MAGIX MAGI family member, X-linked

3249 HPN hepsin

10930 APOBEC2 apolipoprotein B mRNA editing enzyme, catalytic polypeptide-like 2

348093 RBPMS2 RNA binding protein with multiple splicing 2

2104 ESRRG estrogen-related receptor gamma

204474 PDILT protein disulfide isomerase-like, testis expressed

3899 AFF3 AF4/FMR2 family, member 3

3772 KCNJ15 potassium channel, inwardly rectifying subfamily J, member 15

252995 FNDC5 fibronectin type III domain containing 5

92737 DNER delta/notch-like EGF repeat containing

2752 GLUL glutamate-ammonia ligase

27329 ANGPTL3 angiopoietin-like 3

2243 FGA fibrinogen alpha chain

7108 TM7SF2 transmembrane 7 superfamily member 2

25924 MYRIP myosin VIIA and Rab interacting protein

6340 SCNN1G sodium channel, non voltage gated 1 gamma subunit

25928 SOSTDC1 sclerostin domain containing 1

3784 KCNQ1 potassium channel, voltage gated KQT-like subfamily Q, member 1

148808 MFSD4 major facilitator superfamily domain containing 4

64714 PDIA2 protein disulfide isomerase family A, member 2

4684 NCAM1 neural cell adhesion molecule 1

51660 MPC1 mitochondrial pyruvate carrier 1

333 APLP1 amyloid beta (A4) precursor-like protein 1

1358 CPA2 carboxypeptidase A2 (pancreatic)

27087 B3GAT1 beta-1,3-glucuronyltransferase 1

81617 CAB39L calcium binding protein 39-like

64850 ETNPPL ethanolamine-phosphate phospho-lyase

151126 ZNF385B zinc finger protein 385B

114132 SIGLEC11 sialic acid binding Ig-like lectin 11

87

222166 MTURN maturin, neural progenitor differentiation regulator homolog (Xenopus)

8789 FBP2 fructose-1,6-bisphosphatase 2

135892 TRIM50 tripartite motif containing 50

54102 CLIC6 chloride intracellular channel 6

84952 CGNL1 cingulin-like 1

375775 PNPLA7 patatin-like phospholipase domain containing 7

27098 CLUL1 clusterin-like 1 (retinal)

2651 GCNT2 glucosaminyl (N-acetyl) transferase 2, I-branching enzyme (I blood group)

168544 ZNF467 zinc finger protein 467

1000 CDH2 cadherin 2, type 1, N-cadherin (neuronal)

361 AQP4 aquaporin 4

154091 SLC2A12 solute carrier family 2 (facilitated glucose transporter), member 12

6506 SLC1A2 solute carrier family 1 (glial high affinity glutamate transporter), member 2

1131 CHRM3 cholinergic receptor, muscarinic 3

79981 FRMD1 FERM domain containing 1

495 ATP4A ATPase, H+/K+ exchanging, alpha polypeptide

9455 HOMER2 homer scaffolding protein 2

10863 ADAM28 ADAM metallopeptidase domain 28

496 ATP4B ATPase, H+/K+ exchanging, beta polypeptide

5104 SERPINA5 serpin peptidase inhibitor, clade A (alpha-1 antiproteinase, antitrypsin), member 5

887 CCKBR cholecystokinin B receptor

79739 TTLL7 tubulin tyrosine ligase-like family member 7

2940 GSTA3 glutathione S-transferase alpha 3

7036 TFR2 transferrin receptor 2

146556 C16orf89 chromosome 16 open reading frame 89

88

Chapter 5: Drug repurposing for gastric cancer using gene expression data.

- Teaching new tricks to old drugs

5.1 Background

Gastric cancer is the 3rd leading cause of cancer-related deaths worldwide [1], with 5- year survival rate of 20% [2] to 29% [3]. Oncology drug development efforts are often faced with high failure rate, limited efficacy, high cost, and long development time. Only one of every 5000–10,000 prospective anticancer agents receives FDA approval and only 5% of oncology drugs entering Phase I clinical trials are ultimately approved [181]. Drug repurposing is one of the alternate approaches to reduce the high cost and failure rates associated with development of a new drug de novo. Drug repurposing is the process of re- using an approved drug for a different indication or rescuing a lead compound that failed in clinical trials and develop it to treat a different condition. The key advantage of this approach relates to the established safety profile of the drugs; the pharmacokinetics, pharmacodynamics, and toxicity of the compounds are usually known because of the prior preclinical or Phase I studies [182].

With decreasing cost of microarray and next generation sequencing, and increasing availability of open source genomic data, signature-based method offers a promising approach for computational drug repurposing. In this study, we analyze genomic data to identify gene signature for gastric cancer and compare it to the gene signature associated 89

with known drugs. We then prioritize potential therapeutic compounds for the disease based on this signature comparison. We used microarray data from GEO database and

Broad Institute’s Connectivity Map (CMap) portal to link known drugs with the genes involved in the disease. The Connectivity Map method is based on the Kolmogorov–Smirnov statistic [183] using a nonparametric, rank-based, pattern-matching strategy called the Gene

Set Enrichment Analysis (GSEA) [184]. In CMap approach, the disease query signature is comprised of a nonrank ordered list of genes correlated with the disease state, along with a sign indicating whether the gene is up- or down-regulated. The compound reference gene expression profiles are also represented in a nonparametric manner, but instead each gene is rank-ordered by its degree of differential expression relative to untreated controls. Each gene in the disease query signature is then compared against the rank-ordered list of each compound reference signature, and the strength of its positive or negative connectivity yields a score ranging from +1 to −1 for each compound signature [170,185]

5.2 Aim

To identify FDA approved drugs or nondrug bioactive agents that can be potentially used for treating gastric cancer.

90

5.3 Methods

5.3.1 Datasets

Microarray data comparing control vs. treatment tissue was obtained from the Gene

Expression Omnibus (GEO) [166] database Table 5-1. A key challenge while comparing gene expression between tissues obtained from cancer patients and healthy individuals is the difference contributed by genetic heterogeneity among individuals. Thus, we selected only those datasets from GEO database, where both tumor and (adjacent) non-tumor tissues were obtained from the same patient, thus reducing the error due to genetic heterogeneity among individuals. For GSE27342, gastric cancer tissues and their adjacent noncancerous tissues were collected from 80 non-treated patients in China [186]. In GSE13861 dataset, 65 gastric adenocarcinoma tissues with 19 normal surrounding tissue samples from gastric cancer patients in South Korea were used [187]. For drug resistance signature, we used data from GSE14210. Here gastric adenocarcinomas were obtained from 22 patients in South

Korea who initally responded to Cisplatin-Flurouracil (CF) but later developed resistance to the chemotherapy [188].

Sample size (Stomach tissue) GEO ID Treatment Control

GSE27342 80 Tumor 80 Adjacent Normal

GSE13861 65 Tumor 19 Adjacent Normal

GSE14210 22 CF – Resistant Tumor 22 CF - Sensitive Tumor

Table 5-1. Sample size of GEO datasets used in the study. CF: Cisplatin and Flurouracil.

91

5.3.2 Differential Gene Expression Analysis

Differential expression analysis for each dataset was performed in R (version 3.1.2) using Limma package [167] from Bioconductor. Limma fits a linear model to the expression data for each gene. It uses emperical Bayes method to moderate the standard errors of the estimated log fold changes. Benjamini-Hochberg test was used for multiple correction and genes with False Discovery Rate (FDR) < 0.05 were considered to be differentially expressed.

Since Gse27342 represents paired data, paired Limma test for was used for its analysis.

Functional enrichment of Differentially Expressed Genes (DEG), common among the 2 datasets was performed by submitting entrez ids of DEG to the ToppFun tool [168].

5.3.3 Connectivity Map Analysis

Entrez gene identifiers of DEG from the microarray datasets were mapped to

Affymetrix HG-U133A probesets using hgu133a.db package [169] in R. For analysis of each microarray study, lists of DEG of size “k” were used as input query to CMap (version 2), where “k” was varied from 400, to 1000, using an increment of 200. Compounds with negative enrichment scores at permutation P-value < 0.05 from each iteration were selected to identify compounds which can maximally reverse the gene signature. Intersection of compounds from the above iteration results represents potential therapeutic compounds.

92

Figure 5-1. Overview of the methodology.

5.4 Results

5.4.1 Gastric Cancer Data Analysis

The number of Differentially Expressed Genes (DEG) in each dataset is summarized in Table 5-2. Venn diagram of DEG common between the two gastric cancer data sets is shown in Figure 5-2. There were 468 up-regulated genes and 204 down-regulated genes that were differentially expressed in both the two datasets.

93

Number of Differentially Expressed Genes (DEG)

Up-regulated Down-regulated

Gse27342 (n=160) 652 341

Gse13861 (n=84) 3366 2329

GSE14210 (n=44) 12481 3

Table 5-2. Number of Differentially Expressed Genes (DEG) in gastric cancer datasets. n = sample size.

Up-regulated Genes Down-regulated Genes

Figure 5-2. Venn diagram showing intersection between Differentially Expressed Genes (DEG) from the two gastric cancer datasets.

5.4.2 Connectivity Map Analysis

To identify potential compounds that can maximally reverse the gastric cancer gene signature identified above, Connectivity Map was used. Probes corresponding to the upregulated and downregulated genes were used as input query for CMap. However, 94

instead of using standard input list of top 500 up-regulated and 500 down-regulated genes, we submitted four lists of DEG of different sizes. List of DEG of size “k” were used as input query to CMap, where “k” was varied from 400 to 1000, with increment of 200 in each iteration. The CMap results for gastric cancer dataset GSE27342 and GSE13861 are summarized in Table 5-3. The role of these compounds in gastric and other cancers are summarized in Table 5-4. Four out of the eleven compounds identified by our appraoch, have been shown to have implications in gastric cancer in-vivo or in-vitro studies.

Vorinostat, trichostatin A and thiostrepton are the compounds identified with both gastric cancer datasets. Vornistat is currently being tested in clinical trials for treatment of gastric cancer – i) in combination with Capecitabine and Cisplatin for 1st line of treatment of metastatic or recurrent gastric cancer [ClinicalTrials.gov ID: NCT01045538]. ii) with irinotecan, fluorouracil, and leucovorin in treating patients with advanced upper gastrointestinal cancer [ClinicalTrials.gov ID: NCT00537121].

The DEG in gastric cancer identified above were enriched for GO (Gene Ontology) biological processes and pathways using ToppFun. Topmost pathways enriched for DEG from GSE27342 included cell cycle, extracellular matrix organisation, RB in cancer and collagen formaton. Genes from each pathway were submitted to CMap separately to identify compounds that can reverse the expression of genes representing the enriched pathway Table 5-5.

95

GSE27342 GSE13861

vorinostat* vorinostat*

trichostatin A* trichostatin A*

thiostrepton thiostrepton

phenoxybenzamine scriptaid

resveratrol* gossypol*

monobenzone

antimycin A

bufexamac

0175029-0000

Table 5-3. List of the compounds predicted by CMap for each gastric cancer dataset. The compounds highlighted in bold and with asteric have been reported in literature to have implications in gastric cancer at in- vitro or in-vivo level.

Implications Implications in Compound Approved Implication MOA / Target in Gastric other cancers & cancer

other diseases

inhibits the Clinical trials: enzymatic cutaneous T cell In-vitro : Yes Glioblastoma activity of lymphoma Clinical multiforme, Vorinostat* histone (CTCL) - skin cancer Trials: Yes deacetylases advanced non- HDAC small-cell lung cancer

96

inhibitor of Antifungal histone In-vitro: Yes Agent

trichostatin A* deacetylase In-vivo: Yes In-vivo: breast (HDAC) activity cancer

FOXM1 In-vitro: breast (transcription cancer,

thiostrepton antibiotic factor forkhead leukemia and box M1) liver cancer

adrenergic, pheochromocytoma alpha-receptor phenoxybenzamine blocking agent

suppresses NF- Herpes labialis kappaB (NF- infections kappaB) In-vitro: Yes Clinical Trials: resveratrol* activation in In-vivo: Yes HSV infected Colon cancer,

cells rectal cancer

to treat the loss of in-vivo: monobenzone Tyrosinase skin color (vitiligo) melanoma

mitochondrial In-vitro: Antibiotic complex III antimycin A Lung Cancer inhibitor

Anti-inflammatory, COX, class II bufexamac topical agent HDAC inhibitor

97

CDK inhibitor 0175029-0000

In-vivo: breast cancer

In-vitro: HDAC Inhibitor scriptaid colorectal

cencer, leukemia

Inhibits several dehydrogenase including ribonucleotide In-vitro: In-vitro: Yes gossypol* reductase the prostate cancer

rate limiting enzyme activity in DNA synthesis.

Table 5-4. Characteristics of CMap compounds that have expression profiles opposite to that of gastric cancer tumors in this study.

Extracellular Matrix Cell Cycle RB in Cancer Collagen formation organization

phenoxybenzamine verteporfin phenoxybenzamine amrinone

tanespimycin CP-690334-01 tanespimycin CP-690334-01

Thioguanosine Prestwick-675 Thioguanosine PNU-0293363

Apigenin moracizine Apigenin merbromin

pyrvinium cicloheximide pyrvinium iopamidol

Table 5-5. CMap pathway predicted for each pathway.

98

5.4.3 Acquired Resistance to Cisplatin and Flurouracil in Gastric Tumors

One of the key challenges in cancer treatment is development of resistance to chemotherapy. To address this challenge, we analyzed GSE14210 data to identify the genes and pathways differentially regulated in tumors which had acquired resistance to Cisplatin-

Flurouaracil (CF). Four lists of DEG of size 400, 600, 800 and 1000 were run through CMap and intersection of compounds from the four iterations identified 29 compounds.

Trichostatin A, identified through the CMap approach has previously been shown to induce sensitivity to cisplatin treatment in cisplatin-resistant human lung adenocarcinoma cell line

(A549 cells) [189].

Most of the DEG were over-expressed (12,481) and only 3 genes were under- expressed. Functional enrichment of DEG identifed two major pathways – statin pathway and metabolism pathway (including metabolism of lipids and lipoproteins, The Citric Acid

(TCA) cycle and respiratory electron transport and metabolism of amino acids and derivatives) Table 5-6. Two independent studies have shown that fatty acid synthesis pathway contributes to resistance to cisplatin [190] and to flurouracil [191] in breast cancer cells. GLUD1, GLUD2 - genes that code for enzymes involved in glutaminolysis (conversion of the amino acid glutamine to α-ketoglutarate) were also differentially expressed in the CF- resistant tumor as compared to CF-sensitive tumors. Cancer cells have been reported to be dependent on glutamine to maintain TCA cycle [192]. Glutaminolysis co-induced by glutamine and leucine can activate mTORC1 signaling, which triggers cell growth and inhibits autophagy [193]. Further study of the DEG involved in statin and metabolic pathways can

99

elucidate the possible mechanism by which the gastric tumors developed resistance to

Cisplatin – Flurouracil.

Pathway FDR B&H Genes from Input Genes in Annotation

Statin Pathway 1.656E-2 10 32

Metabolism 1.656E-2 130 1575

Table 5-6. Pathway enrichment for genes differentially expressed in cisplatin- resistant tumors.

5.5 Limitations

Although Connectivity Map is a unique and powerful strategy to use genomic disease signature to for drug repurposing, it has a few limitations. The therapeutic compounds identified are limited to the compounds present in CMap library. Secondly, CMap uses a simple algorithm to represent degree of connectiveness between the query disease signature and the compound signature. This algorithm has been useful in a number of cases, but it may not be the most productive representation of connectivity. However, it is possible for the users to implement an advanced algorithm and scoring criteria by using original gene expression data for the CMap compounds. In addition, it is unclear to how a drug’s potential in breast cancer, prostrate cancer, melanoma and leukemia cell lines, which were used to generate CMap signatures, will be relevant in gastric cancer. However, there is a growing evidence of success rate of identifying potential therapeutic compounds from CMap for treatment of other cancers and non-cancerous diseases and their validation using in-vitro models.

100

5.6 Conclusions

• Analysis of gene expression profiles from multiple disease datasets can help to identify

important pathways and potential novel therapeutic targets for the disease.

• Connectivity Map helps to identify and prioritize existing drugs or compounds that can

be of therapeutic value for the disease.

101

Chapter 6: Regulation of cytokines by Sonic Hedgehog in macrophages.

6.1 Background

In Chapter 3, qPCR data from uninfected and H. pylori infected - wild type and PC-SHH-

KO mice, was analyzed. The statistical analysis suggests that Sonic Hedgehog (SHH) positively regulates expression of cytokines like IL-12, IL-1β, MIP-2, IL-10, however the mechanism remains unclear. We developed a simple mathematical model of the cytokines and SHH activated during H. pylori infection. Our model captures the positive regulation of cytokines by SHH through an intermediate model species ‘X’. We speculate the following mechanism through which SHH may play a key role in regulating the cytokines during H. pylori infection. Schumacher et al have shown that SHH acts as a macrophage chemoattractant in response to H. pylori infection [111] and this mechanism involves smoothened receptor, which is downstream of SHH signaling. It has been reported by independent studies that monocytes and macrophages treated with H. pylori secrete IL-

1β, IL-12, IL-10, TNFγ, IL-6 [131–133]. Thus the positive influence on the expression of cytokines by SHH, predicted by our qPCR data may be mediated through recruitment of macrophage by SHH to the gastric epithelium. The IL-12 secreted in turn activates release of

IFNγ from natural killer cells [136,137].

102

6.2 Aim

6.2.1 Hypothesis

SHH can enhance the expression of cytokines like IL-12, IL-1β, MIP2 (CXCL-2) and TNF in H. pylori stimulated macrophages and this mechanism involves smoothened receptor

(SMO).

6.2.2 Significance

Results from this experiment can help to identify unknown molecule ‘X’ in our mathematical model and reduce the existing knowledge gap between SHH and cytokines, which are activated during H. pylori infection.

6.2.3 Expected Results:

In WT group, macrophages stimulated with SHH would express higher levels of cytokines as compared to that in macrophages stimulated with PBS. This would suggest that positive regulation of cytokines by SHH involves macrophages. Whereas, (Smoothened gene knock-out) SMO-KO macrophages stimulated with SHH would express lower levels of cytokines as compared to WT macrophages stimulated with SHH. This would suggest that

SHH-SMO pathway is involved in cytokines production from macrophages.

6.3 Materials and Methods

6.3.1 Cell Culture

H. pylori SS1 (kindly donated by Dr. K. A. Eaton, University of Michigan, Ann Arbor,

MI) bacteria were grown in Brucella broth supplemented with 5% fetal calf serum in a 103

humidified microaerophilic chamber (BBL Gas System, with CampyPak Plus packs; BD

Microbiology, Sparks, MD) in a shaking incubator at 37oC for 16 hours. Macrophages were generated from bone marrow (femur) of WT (LysMCre) and SMO-KO

(LysMCre/SmoothenedFF) mice [111]. Briefly, bone marrow from femur and tibia of mice was isolated, filtered through a 40 micron sieve, and grown on 12 well-plate containing 1ml of DMEM media (DMEM + 1% penicillin + 1% streptomycin + sodium pyruvate) and 50 ul of

MCSF (Macrophage Colony Stimulating Factor) media per well. After 3 days, non-adherent cells were removed and the remaining differentiated macrophages were used for experimentation. Cells were incubated with a 10:1 multiplicity of infection (MOI) of H. pylori

(Sydney strain 1) for 2 hours. WT macrophages were divided into two groups, one was stimulated with 1ug/ml SHH (n=4) and other with PBS (n=4) for 10 hours. Similarly, SMO-KO macrophages were stimulated either with SHH (n=4) or by PBS (n=4) for 10 hours.

Previously, macrophages were also treated with only SHH, without pre-stimulation with H. pylori and tested for expression of cytokines. These macrophages did not express the cytokines in absence of H. pylori stimulation. Currently, there is no literature studying the effect of SHH treatment on macrophages that could be used as a reference. We estimated

10 hours of SHH exposure time based on a study that showed production of cytokines (IL-

12) from dendritic cells within 10 hours of treatment with H. pylori [194].

6.3.2 Quantitative Real-Time PCR (qPCR)

Total RNA was isolated from the macrophages and High Capacity cDNA Reverse

Transcription Kit was used for cDNA synthesis from 100ng of RNA following the manufacturer’s protocol (Applied Biosystems). Predesigned real-time PCR assays were

104

purchased for the following genes (TaqMan Gene Expression Systems; Applied Biosystems):

IL-12, MIP2, IL-1Β, and HPRT. HPRT was used as an internal control. PCR amplifications were performed in a total volume of 20 uL containing 20X TaqMan Expression Assay primers, 2X

TaqMan Universal Master Mix (TaqMan Gene Expression Systems; Applied Biosystems), and complementary DNA template. Each PCR amplification was performed in duplicate wells in a

StepOne Real-Time PCR System (Applied Biosystems) using the following conditions: 50°C for 2 minutes, 95°C for 10 minutes, 95°C for 15 seconds (denature), and 60°C for 1 minute

(anneal/extend) for 40 cycles.

6.3.3 Statistical Analysis

For each sample, dCT value for target gene was calculated by subtracting CT value of calibrator gene (HPRT) from CT value of target gene. To test if expression of the cytokines in

H. pylori stimulated macrophages was affected by SHH, a t-test was performed. dCT values from PBS treated WT macrophages were compared with dCT values from SHH treated WT macrophages. P-value <0.05 was considered statistically significant. For graphical presentation of qPCR results, for each gene, ddCT was calculated by subtracting dCT value of

SHH treated group from dCT of PBS treated group. The data were plotted as 2-ddCT (fold change). The qPCR data was analyzed using R statistical software.

6.4 Results

H. pylori stimulated macrophages were treated with SHH or PBS and expression of IL-

12, IL-b, MIP2 (CXCL2) and TNF were measured by qPCR. No statistically significant

105

difference was observed in the expression of cytokines between SHH and PBS treated groups.

Figure 6-1. Effect of SHH on the expression of cytokines in H. pylori stimulated macrophages.

6.5 Conclusion

Our current results show that there is not enough evidence to reject the null hypothesis that - SHH does not enhance the expression of cytokines like IL-12, IL-b, CXCL2

106

and TNF in H. pylori stimulated macrophages. However, optimization of the experimental design may be required to further explore the role of SHH in regulation of these cytokines.

For example, further optimization of time period for which macrophages are treated with

SHH and the SHH concentration used, may allow to better understanding of the effect of

SHH on cytokines’ expression and their release from macrophages.

107

Chapter 7: Using empirical probability (frequency) of change for analyzing microarray data.

This chapter describes the framework of integrating (i) probability (frequency) of change, (ii) fold change and (iii) pathway knowledge for identifying disease biomarkers form microarray data. We discuss the advantages and limitations of the proposed method.

7.1 Background and Motivation

Gastric cancer is the 3rd leading cause of cancer-related deaths worldwide. The five- year survival rate for the disease is high (71-57%) if diagnosed in early stage as compared to that in late stages (20-4%) [3]. The two major factors contributing to the low survival rate of gastric cancer are late diagnosis due to lack of diagnostic biomarkers and limited target based drugs available for treatment of the disease. Identification of genes and pathways that are differentially regulated in gastric cancer is critical for discovery of new biomarkers and drug targets for the disease. One of the major challenges while comparing gene expression between tissues obtained from cancer patients and healthy people is the difference contributed by genetic heterogeneity among individuals. Thus, we selected only those datasets from GEO database, where both tumor and (adjacent) non-tumor tissues were obtained from the same patient, thus reducing the error due to genetic heterogeneity among individuals.

Currently, selection of differentially expressed genes relies heavily on the magnitude of fold change of the gene in disease to that in control. The standard process involves comparison of control and treatment group by a parametric or non-parametric statistical 108

test, followed by selection of genes with adjusted p-value < 0.05. The genes are further sorted and selected based on their fold change. However, some genes, like transcription factors that can play a crucial role in disease, may exhibit low fold change in expression

[195]. So, we suggest integrating two other features while selecting differentially expressed genes – (i) consistency (frequency) of change across the sample population. (ii) topological network of biological pathways.

7.2 Hypothesis

Integration of ‘frequency of change’ in gene expression and pathway knowledge along with fold change can enhance the method of screening the list of DEG (differentially expressed genes) identified based on FDR statistics than using only fold change criteria.

7.3 Methods

7.3.1 Datasets

Three microarray data sets with gastric tumor and adjacent non-tumor tissues, are obtained from GEO [166] database - GSE27342 (n=160), GSE3438 (n=100) and GSE13861

(n=84). For each data set, patients are randomly divided into training and test set using three fold cross validation approach. Two-third of the patients are assigned to training set and one-third to test set and this process is repeated 3 times, with each of the 3 subsamples of patients used exactly once as the test set. Differentially expressed genes (DEG) are predicted using training set and then the error rate for the prediction is calculated using test set. The estimated errors from the three iterations are averaged to calculate overall F1

109

score and AUROC (Area Under Receiver Operating Characteristic). Plotting Recall over False

Positive Rate creates ROC curve and AUROC value above 0.8 is considered good while value less than 0.6 is considered as poor performance of the algorithm. The advantage of k-fold method over repeated random sub-sampling is that all observations are used for both training and validation, and each observation is used for validation exactly once.

2 ∗ Precision ∗ Recall �1 = Precesion + Recall

���� �������� ��������� = ���� �������� + ����� ��������

���� �������� ������ = ���� �������� + ����� ��������

����� �������� ����� �������� ���� = ����� �������� + ���� ��������

7.3.2 Calculation of Frequency Score

We first calculate log expression ratio (tumor expression/normal expression) for each gene in each patient. Let aij represent log expression ratio (tumor expression/normal expression) for gene ‘i’ in patient ‘j’. For each gene, calculate the number of times aij is greater than or less than zero and the associated p-value using Binomial test.

For each gene, calculate aij in all patients:

• Frequency.Increase = number of times aij > 0

• Frequency.Decrese = number of times aij < 0

• p-value (Binomial test) 110

If the p-value for the gene is less than 0.05, Frequency Score for that gene is assigned a value that is greater between Frequency.Increase and Frequency.Decrease. The Frequency

Score is multiplied by -1 if Frequency.Decrease is greater than Frequency.Increase. Genes with p-value greater than 0.05 are given a Frequency Score of zero.

7.3.3 Calculation of Fold Score

For each data set, we compare gene expression between gastric tumor and adjacent non-tumor tissue using Limma package in R. Since our data sets contain paired samples, paired Limma test is used for the analysis. Benjamini-Hochberg test is used for multiple correction and genes with False Discovery Rate (FDR) < 0.05 are considered to be differentially expressed. For genes with FDR < 0.05, the log fold change is normalized such that it’s value varies between 0 to 1 and can be easily integrated with Frequency Score later.

Maximum (logFCmax) and minimum (logFCmin) log fold change shown by all genes is calculated. Normalized log fold change for each gene (Norm.logFCi) is calculated by dividing the log fold change for that gene by the absolute value of the difference between logFCmax and logFCmin. For genes with FDR > 0.05, Norm.logFCi is assigned a value of zero. A

Cumulative Score is calculated for each gene by adding Frequency Score and Normalized

Fold Score of that gene.

7.3.4 Integration of Frequency of change and Fold change with pathways knowledge

In future, the Fold Score and Frequency Score can be combined with biological networks to identify differentially expressed pathways. Topological network of protein- protein interactions can be obtained from pathway databases like pathway commons, reactome and NCI-PID. The network nodes representing genes will be assigned the 111

Cumulative Score calculated above. The central idea of this approach is - the probability that a gene is showing differential expression just by chance, is less if it is connected to other differentially expressed genes and thus it gets higher ranking (score). Along with frequency and fold change, inclusion of this approach would also take into account - the direction of change (overexpression or under-expression) and pathway topology.

7.4 Preliminary Results

Figure 7-1 shows the differentially expressed genes predicted for 160 samples of

GSE27342. The top panel in blue shows genes that were predicted as DEG by frequency- based method but not by Limma. The second panel shows that there were two genes that were predicted to be DEG by Limma but were missed by frequency-based method. The bottom panel in green shows the genes predicted to be DEG by both the methods. Although frequency-based approach captures more genes (which may include more True Positives) than fold change method, it comes with a cost of increased noise (may include more False

Positives). So, we combine the two approaches to identify genes that change consistently across the sample population and with high fold change.

7.5 Limitations

Currently, the frequency-based method does not incorporate correction for multiple testing (for the p-values that are calculated by Binomial test). In future, FDR method should to be incorporated to calculate adjusted p-values for all the genes. Although, the frequency of change can be a useful feature along with fold-change to further screen the list of DEG

112

identified based on FDR statistics, this approach is limited to datasets that have large sample size and contain paired samples.

(A). Genes predicted by

frequency-based method

but not by Limma.

(B). Genes predicted by

Limma but not by

frequency-based method.

(C). Genes predicted by

both the methods.

Figure 7-1. Differentially expressed genes predicted by Limma and frequency-based approach.

A) Genes predicted as DEG exclusively by frequency-based method B) Genes predicted as DEG exclusively by Limma C) Genes predicted by both the methods.

113

Chapter 8: Discussion

Gastric cancer is the fifth most common malignancy in the world and third leading cause of cancer-related mortality worldwide [1], with five-year survival rate of only 20-29%

[3]. In order to develop better drugs, diagnostics and preventive measures for gastric cancer, it is critical to understand the underlying molecular biology of the disease and factors that increase the risk for the disease. Helicobacter pylori-induced atrophic gastritis is a major risk factor associated with gastric cancer development [7]. H. pylori infection of gastric tissue results in an immune response dominated by Th1 cytokines [96] and has also been linked with deregulation of Sonic Hedgehog (SHH) signaling pathway in gastric tissue

[110,111]. However, a model studying temporal relationship between SHH and cytokines activated during H. pylori infection is lacking. Also, the molecular details that link gastric cancer and H. pylori mediated atrophic gastritis are not well understood.

The objective of this dissertation research were: 1) to study the cytokine-SHH sub-network activated during H. pylori infection. 2) to identify the molecular signature common to gastric cancer and H. pylori mediated atrophic gastritis. 3) drug repurposing for gastric cancer. We have used mathematical modeling and publically available microarray data to explore the temporal behavior of the cytokine-SHH subnetwork, which plays a critical role in H. pylori mediated gastritis. We also identified genes and pathways that are differentially expressed in gastric cancer and H. pylori mediated atrophic gastritis. We use signature-based drug repurposing approach to identify potential drugs for gastric cancer treatment.

114

8.1 Mathematical modeling of cytokine-SHH sub-network in H. pylori

infected gastric tissue.

8.1.1 Summary of Major Findings

Analysis of qPCR data from uninfected and H. pylori infected - WT and PC-SHHKO mice, shows that SHH positively regulates expression of cytokines like IL-12, IL-1β, MIP-2, IL-

10, however the mechanism remains unclear. Our mathematical model suggests that NFĸB,

SHH and the cytokines engage in a feedback loop which can result in damped oscillations.

8.1.2 Significance of the findings

Our analysis of experimental data indicates a positive regulation of the cytokines’ expression involving SHH. However, current pathway databases (for example Strings,

Pathway Commons), and molecular circuits (wiring diagram of interactions among genes and proteins) derived through manual curation of current literature, fail to identify the positive regulation. The mathematical model helps to bring out emergent properties of the network, which can guide future experimental studies and enhance our current understanding of the system. We use the model as a tool to gain insights into cytokine and SHH relationship during H. pylori infection and as a hypotheses generating tool.

8.1.3 Future Directions

Temporal analysis of cytokines from an independent microarray dataset also indicates that IL-1β, IFNγ and IL-10 from H. pylori infected mice, show a cyclic behavior rather than a linear trend. Based on our preliminary results from analysis of Geo dataset, we propose that experiments capturing expression of cytokines and other genes at different time points (on scale of days) can inform about time dependent variation in expression of genes and also help to study the correlations among pairs of genes and understand their relationships.

115

8.2 Meta-analysis of microarray data from gastric cancer and H. pylori

infected patients.

8.2.1 Summary of Major Findings

Our analysis shows that cell-cell adhesion, cell-cell communication, tight junctions and leukocyte transendothelial migration involving claudin genes (CLDN1, CLDN2, CLDN3,

CLDN4, CLDN7) are the key pathways that were up-regulated in both gastric cancer and H. pylori mediated atrophic gastritis. Whereas, genes involved in gastric acid secretion, potassium ion transport and creatine pathways were down-regulated in both the diseases.

8.2.2 Significance of the findings

Although, atrophic gastritis accompanied with loss of parietal cells and hypochlorhydria is highly associated with development of gastric cancer [165], the molecular details that link gastric cancer and H. pylori mediated atrophic gastritis are not well understood. Identification of genes and pathways which are dis-regulated in both gastric cancer and H. pylori mediated atrophic gastritis will enhance our current knowledge about the molecular details which link the two diseases.

8.2.3 Future Directions

Further analysis of the genes and pathways identified above can elucidate the underlying molecular mechanism through which H. pylori mediated atrophic gastritis increases the risk for development of gastric cancer.

116

8.3 Drug Repurposing for gastric cancer using gene expression data.

8.3.1 Summary of Major Findings

Genes upregulated in the two gastric cancer datasets were mostly enriched for cell cycle pathways while the down-regulated genes were enriched for pathways involved in gastric acid secretion. Connectivity Map (CMap) analysis identified eleven compounds that can potentially reverse the gastric cancer signatures dervied from the two disease data sets.

Vorinostat, trichostatin A and thiostrepton are the compounds that were identified with both the datasets.

We also identified the genes and pathways differentially regulated in gastric tumors which had acquired resistance to Cisplatin-Flurouaracil (CF) chemotherapy. Functional enrichment of differentially expressed genes identifed two major pathways – statin pathway and metabolism pathway (including metabolism of lipids and lipoproteins, The citric acid

(TCA) cycle and respiratory electron transport and metabolism of amino acids and derivatives). CMap analysis identified 29 compounds that can potentially reverse the cisplatin-flurouracil resistant signature.

8.3.2 Significance of the findings

One of the approaches to reduce the high cost and time associated with drug discovery is through repurposing of drugs approved for other diseases. The key advantage of this approach is established safety profile of the drugs. We use molecular signature of gastric cancer derived from microarray data and Broad Institute’s Connectivity Map (CMap)

117

to identify and prioritize compounds that can be further tested in lab for treatment of gastric cancer.

8.3.3 Future Directions

In future, three fold cross validation approach can be used for identifying differentially expressed genes; for each data set, patients would be randomly divided into training and test set. Two-third of the patients should be assigned to training set and one- third to test set and this process is repeated 3 times, with each of the 3 subsamples of patients used exactly once as the test set. Differentially expressed genes (DEG) will be predicted using training set and then the error rate for the prediction will be calculated using test set. The estimated errors from the three iterations will be averaged to calculate overall F1 score and AUROC (Area Under Receiver Operating Characteristic). Along with

CMap, other resources for therapeutic compounds’ genomic data can also be used – LINCS

(Library of Integrated Network-based Cellular Signatures) [196], CCLE (Cancer Cell Line

Encyclopedia: 24 anticancer drugs across 479 cell lines) [197], caArray GSKdata (19 compounds in a panel of 311 cancer cell lines) [198]. Integration of genomics data with available signaling and metabolic pathways and protein interaction networks to reconstruct gastric cancer specific pathways will help to identify molecular targets for repositioned drugs. Differentially expressed genes identified from genomics data will be enriched for

KEGG pathways and known drug targets. Connectivity Map can be used to identify drugs that target the enriched pathways. A topological network that connects drugs and their target pathways will help to elucidate possible mechanism of action of the identified drug.

118

Chapter 9: Bibliography

1. Ferlay J, Soerjomataram I, Ervik M, Dikshit R, Eser S, Mathers C, Rebelo M, Parkin DM, Forman D, Bray F. Stomach Cancer Estimated Incidence, Mortality and Prevalence Worldwide in 2012. In: GLOBOCAN 2012 v1.0, Cancer Incidence and Mortality Worldwide: IARC CancerBase No. 11. [Internet]. 2013 [cited 26 Apr 2015]. Available: http://globocan.iarc.fr/Pages/fact_sheets_cancer.aspx

2. Chun N, Ford JM. Genetic testing by cancer site: stomach. Cancer J. 18: 355–63. doi:10.1097/PPO.0b013e31826246dc

3. Survival rates for stomach cancer, by stage. In: American Cancer Society [Internet]. 2014 [cited 22 Apr 2015]. Available: http://www.cancer.org/cancer/stomachcancer/detailedguide/stomach-cancer-survival-rates

4. Yoon H, Kim N. Diagnosis and management of high risk group for gastric cancer. Gut Liver. 2015;9: 5–17. doi:10.5009/gnl14118

5. Peek RM, Blaser MJ. Helicobacter pylori and gastrointestinal tract adenocarcinomas. Nat Rev Cancer. 2002;2: 28–37. doi:10.1038/nrc703

6. Schistosomes, liver flukes and Helicobacter pylori. IARC Working Group on the Evaluation of Carcinogenic Risks to Humans. Lyon, 7-14 June 1994. IARC Monogr Eval Carcinog Risks Hum. 1994;61: 1–241.

7. Correa P. Chronic gastritis: a clinico-pathological classification. Am J Gastroenterol. 1988;83: 504–9.

8. Correa P. Human gastric carcinogenesis: a multistep and multifactorial process--First American Cancer Society Award Lecture on Cancer Epidemiology and Prevention. Cancer Res. 1992;52: 6735–40.

9. Carrasco G, Corvalan AH. Helicobacter pylori-Induced Chronic Gastritis and Assessing Risks for Gastric Cancer. Gastroenterol Res Pract. 2013;2013: 393015. doi:10.1155/2013/393015

10. Shimoyama T, Crabtree JE. Bacterial factors and immune pathogenesis in Helicobacter pylori infection. Gut. 1998;43 Suppl 1: S2–5.

11. Bodger K, Crabtree JE. Helicobacter pylori and gastric inflammation. Br Med Bull. 1998;54: 139–150.

12. Bodger K, Crabtree JE. Helicobacter pylori and gastric inflammation. Br Med Bull. 1998;54: 139–50.

119

13. Bodger K, Wyatt JI, Heatley R V. Gastric mucosal secretion of interleukin-10: relations to histopathology, Helicobacter pylori status, and tumour necrosis factor-alpha secretion. Gut. 1997;40: 739–44.

14. Mannick EE, Bravo LE, Zarama G, Realpe JL, Zhang XJ, Ruiz B, et al. Inducible nitric oxide synthase, nitrotyrosine, and apoptosis in Helicobacter pylori gastritis: effect of antibiotics and antioxidants. Cancer Res. 1996;56: 3238–43.

15. Smoot DT, Elliott TB, Verspaget HW, Jones D, Allen CR, Vernon KG, et al. Influence of Helicobacter pylori on reactive oxygen-induced gastric epithelial cell injury. Carcinogenesis. 2000;21: 2091–5.

16. Moss SF, Blaser MJ. Mechanisms of disease: Inflammation and the origins of cancer. Nat Clin Pract Oncol. 2005;2: 90–7; quiz 1 p following 113. doi:10.1038/ncponc0081

17. Layke JC, Lopez PP. Gastric cancer: diagnosis and treatment options. Am Fam Physician. 2004;69: 1133–40.

18. Kim HS, Kim HJ, Kim SY, Kim TY, Lee KW, Baek SK, et al. Second-line chemotherapy versus supportive cancer treatment in advanced gastric cancer: a meta-analysis. Ann Oncol. 2013;24: 2850–4. doi:10.1093/annonc/mdt351

19. Wanebo HJ, Kennedy BJ, Chmiel J, Ph D. Cancer of the Stomach A Patient Care Study by the American. 1993;218: 583–592.

20. Gerber DE. Targeted therapies: a new generation of cancer treatments. Am Fam Physician. 2008;77: 311–9.

21. US Food and Drug Administration. Genentech US BL 103792 supplement: Trastuzumab— Genentech I. [Internet]. 2010 [cited 26 Apr 2015]. Available: http://www.accessdata.fda.gov/drugsatfda_docs/label/2010/103792s5250lbl.pdf

22. FDA approves Cyramza for stomach cancer. In: U.S. Food and Drug Administration [Internet]. 2014 [cited 26 Apr 2015]. Available: http://www.fda.gov/NewsEvents/Newsroom/PressAnnouncements/ucm394107.htm

23. Key Facts. In: 2013 Profile Biopharmaceutical Industry [Internet]. 2013 [cited 26 Apr 2015]. Available: http://www.phrma.org/sites/default/files/pdf/PhRMA Profile 2013.pdf

24. Arrowsmith J, Miller P. Trial watch: phase II and phase III attrition rates 2011-2012. Nat Rev Drug Discov. 2013;12: 569. doi:10.1038/nrd4090

25. Kittaneh M, Montero AJ, Glück S. Molecular profiling for breast cancer: a comprehensive review. Biomark Cancer. 2013;5: 61–70. doi:10.4137/BIC.S9455

26. What is stomach cancer. In: American Cancer Society [Internet]. 2014 [cited 26 Apr 2015]. Available: http://www.cancer.org/cancer/stomachcancer/detailedguide/stomach-cancer- what-is-stomach-cancer 120

27. Yang I, Nell S, Suerbaum S. Survival in hostile territory: the microbiota of the stomach. FEMS Microbiol Rev. 2013;37: 736–61. doi:10.1111/1574-6976.12027

28. The Digestive System. Anatomy & Physiology. Boundless; 2013.

29. Anatomy & Physiology: Energy, Maintenance and Environmental Exchange [Internet]. 2015 [cited 26 Apr 2015]. Available: http://cnx.org/contents/4ee7e2a2-fa0a-462c-adf4- [email protected]:13/Anatomy_&_Physiology:_Energy,_

30. Klaus J. Lewin HDA. Tumors of the Esophagus & Stomach (Atlas of Tumor Pathology, 3rd Series, Vol. 18). 1996.

31. LAUREN P. THE TWO HISTOLOGICAL MAIN TYPES OF GASTRIC CARCINOMA: DIFFUSE AND SO- CALLED INTESTINAL-TYPE CARCINOMA. AN ATTEMPT AT A HISTO-CLINICAL CLASSIFICATION. Acta Pathol Microbiol Scand. 1965;64: 31–49.

32. Cuello C, López J, Correa P, Murray J, Zarama G, Gordillo G. Histopathology of gastric dysplasias: correlations with gastric juice chemistry. Am J Surg Pathol. 1979;3: 491–500.

33. How is stomach cancer staged? In: American Cancer Society. 2014.

34. Diagnosis & Staging [Internet]. [cited 26 Apr 2015]. Available: http://www.debbiesdream.org/portal/diagnosis

35. Bertuccio P, Chatenoud L, Levi F, Praud D, Ferlay J, Negri E, et al. Recent patterns in gastric cancer: a global overview. Int J Cancer. 2009;125: 666–73. doi:10.1002/ijc.24290

36. Fox JG, Wang TC. Review series Inflammation , atrophy , and gastric cancer. 2007;117. doi:10.1172/JCI30111.60

37. Uemura N, Okamoto S, Yamamoto S, Matsumura N, Yamaguchi S, Yamakido M, et al. Helicobacter pylori infection and the development of gastric cancer. N Engl J Med. 2001;345: 784–9. doi:10.1056/NEJMoa001999

38. Huang JQ, Sridhar S, Chen Y, Hunt RH. Meta-analysis of the relationship between Helicobacter pylori seropositivity and gastric cancer. Gastroenterology. 1998;114: 1169–79.

39. Parsonnet J, Friedman GD, Vandersteen DP, Chang Y, Vogelman JH, Orentreich N, et al. Helicobacter pylori infection and the risk of gastric carcinoma. N Engl J Med. 1991;325: 1127– 31. doi:10.1056/NEJM199110173251603

40. Kikuchi S, Wada O, Nakajima T, Nishi T, Kobayashi O, Konishi T, et al. Serum anti-Helicobacter pylori antibody and gastric carcinoma among young adults. Research Group on Prevention of Gastric Carcinoma among Young Adults. Cancer. 1995;75: 2789–93.

41. Miehlke S, Hackelsberger A, Meining A, von Arnim U, Müller P, Ochsenkühn T, et al. Histological diagnosis of Helicobacter pylori gastritis is predictive of a high risk of gastric carcinoma. Int J Cancer. 1997;73: 837–9. 121

42. Macarthur M, Hold GL, El-Omar EM. Inflammation and Cancer II. Role of chronic inflammation and cytokine gene polymorphisms in the pathogenesis of gastrointestinal malignancy. Am J Physiol Gastrointest Liver Physiol. 2004;286: G515–20. doi:10.1152/ajpgi.00475.2003

43. Fox JG, Wang TC. Inflammation, atrophy, and gastric cancer. J Clin Invest. 2007;117: 60–9. doi:10.1172/JCI30111

44. Sommer F, Faller G, Konturek P, Kirchner T, Hahn EG, Zeus J, et al. Antrum- and corpus mucosa-infiltrating CD4(+) lymphocytes in Helicobacter pylori gastritis display a Th1 phenotype. Infect Immun. 1998;66: 5543–6.

45. D’Elios MM, Amedei A, Del Prete G. Helicobacter pylori antigen-specific T-cell responses at gastric level in chronic gastritis, peptic ulcer, gastric cancer and low-grade mucosa-associated lymphoid tissue (MALT) lymphoma. Microbes Infect. 2003;5: 723–730. doi:10.1016/S1286- 4579(03)00114-X

46. Mohammadi M, Nedrud J, Redline RAY, Lycke N, Czinn SJ. Murine CD4 T-Cell Response to. 1997; 1848–1857.

47. Ramirez-ramos A, Gilman RH, Leon-barua R, Recavarren-arce S, Salazar G, Checkley W, et al. Rapid Recurrence of Helicobacter pylori Infection in Peruvian Patients after Successful Eradication. : 1027–1031.

48. Mohammadi M, Nedrud J, Redline R, Lycke N, Czinn SJ. Murine CD4 T-cell response to Helicobacter infection: TH1 cells enhance gastritis and TH2 cells reduce bacterial load. Gastroenterology. 1997;113: 1848–57.

49. Fox JG, Beck P, Dangler C a, Whary MT, Wang TC, Shi HN, et al. Concurrent enteric helminth infection modulates inflammation and gastric immune responses and reduces helicobacter- induced gastric atrophy. Nat Med. 2000;6: 536–42. doi:10.1038/75015

50. Kusters JG, van Vliet AHM, Kuipers EJ. Pathogenesis of Helicobacter pylori infection. Clin Microbiol Rev. 2006;19: 449–90. doi:10.1128/CMR.00054-05

51. Wroblewski LE, Peek RM, Wilson KT. Helicobacter pylori and gastric cancer: factors that modulate disease risk. Clin Microbiol Rev. 2010;23: 713–39. doi:10.1128/CMR.00011-10

52. Go MF, Kapur V, Graham DY, Musser JM. Population genetic analysis of Helicobacter pylori by multilocus enzyme electrophoresis: extensive allelic diversity and recombinational population structure. J Bacteriol. 1996;178: 3934–8.

53. Islami F, Kamangar F. Helicobacter pylori and esophageal cancer risk: a meta-analysis. Cancer Prev Res (Phila). 2008;1: 329–38. doi:10.1158/1940-6207.CAPR-08-0109

54. Van Doorn LJ, Figueiredo C, Sanna R, Plaisier A, Schneeberger P, de Boer W, et al. Clinical relevance of the cagA, vacA, and iceA status of Helicobacter pylori. Gastroenterology. 1998;115: 58–66.

122

55. Gerhard M, Lehn N, Neumayer N, Borén T, Rad R, Schepp W, et al. Clinical relevance of the Helicobacter pylori gene for blood-group antigen-binding adhesin. Proc Natl Acad Sci U S A. 1999;96: 12778–83.

56. Calam J. Helicobacter pylori modulation of gastric acid. Yale J Biol Med. 1999;72: 195–202.

57. Furuta T, Shirai N, Xiao F, El-Omar EM, Rabkin CS, Sugimura H, et al. Polymorphism of interleukin-1beta affects the eradication rates of Helicobacter pylori by triple therapy. Clin Gastroenterol Hepatol. 2004;2: 22–30.

58. Machado JC, Figueiredo C, Canedo P, Pharoah P, Carvalho R, Nabais S, et al. A proinflammatory genetic profile increases the risk for chronic atrophic gastritis and gastric carcinoma. Gastroenterology. 2003;125: 364–71.

59. Rad R, Prinz C, Neu B, Neuhofer M, Zeitner M, Voland P, et al. Synergistic effect of Helicobacter pylori virulence factors and interleukin-1 polymorphisms for the development of severe histological changes in the gastric mucosa. J Infect Dis. 2003;188: 272–81. doi:10.1086/376458

60. Zambon C-F, Basso D, Navaglia F, Belluco C, Falda A, Fogar P, et al. Pro- and anti-inflammatory cytokines gene polymorphisms and Helicobacter pylori infection: interactions influence outcome. Cytokine. 2005;29: 141–52. doi:10.1016/j.cyto.2004.10.013

61. Hellmig S, Hampe J, Fölsch UR, Schreiber S. Role of IL-10 promoter haplotypes in Helicobacter pylori associated gastric inflammation. Gut. 2005;54: 888.

62. Rad R, Dossumbekova A, Neu B, Lang R, Bauer S, Saur D, et al. Cytokine gene polymorphisms influence mucosal cytokine expression, gastric inflammation, and host specific colonisation during Helicobacter pylori infection. Gut. 2004;53: 1082–9. doi:10.1136/gut.2003.029736

63. Bauer B, Meyer TF. The Human Gastric Pathogen Helicobacter pylori and Its Association with Gastric Cancer and Ulcer Disease. Ulcers. 2011;2011: 23. doi:10.1155/2011/340157

64. Blaser MJ, Atherton JC. Helicobacter pylori persistence: biology and disease. J Clin Invest. 2004;113: 321–33. doi:10.1172/JCI20925

65. Amedei A, Bergman MP, Appelmelk BJ, Azzurri A, Benagiano M, Tamburini C, et al. Molecular mimicry between Helicobacter pylori antigens and H+, K+ --adenosine triphosphatase in human gastric autoimmunity. J Exp Med. 2003;198: 1147–56. doi:10.1084/jem.20030530

66. D’Elios MM, Appelmelk BJ, Amedei A, Bergman MP, Del Prete G. Gastric autoimmunity: the role of Helicobacter pylori and molecular mimicry. Trends Mol Med. 2004;10: 316–23. doi:10.1016/j.molmed.2004.06.001

67. Stomach Cancer: Symptoms and Signs. In: Cancer.Net [Internet]. 2014 [cited 26 Apr 2015]. Available: http://www.cancer.net/cancer-types/stomach-cancer/symptoms-and-signs

68. Feldman M, Friedman LS SM. Tumors of the stomach. Sleisenger & Fordtran’s Gastrointestinal and liver disease. 7th ed. 2002. 123

69. McLean MH, El-Omar EM. Genetics of gastric cancer. Nat Rev Gastroenterol Hepatol. 2014;11: 664–674. doi:10.1038/nrgastro.2014.143

70. Hereditary Diffuse Gastric Cancer. In: Cancer.Net [Internet]. 2014. Available: http://www.cancer.net/cancer-types/hereditary-diffuse-gastric-cancer

71. CDH1 (E-cadherin) Gene Mutation Analysis.

72. GASTRIC CANCER: INTRODUCTION. In: hopkinsmedicine.org [Internet]. 2014 [cited 26 Apr 2015]. Available: http://www.hopkinsmedicine.org/gastroenterology_hepatology/_pdfs/esophagus_stomach/g astric_cancer.pdf

73. Drugs Approved for Stomach (Gastric) Cancer. In: National Cancer Institute [Internet]. 2015 [cited 26 Apr 2015]. Available: http://www.cancer.gov/cancertopics/druginfo/stomachcancer

74. PAYNE S, MILES D. Mechanisms of anticancer drugs. Scott-Brown’s Otorhinolaryngology: Head and Neck Surgery. 2008. pp. 34–46.

75. Vu T, Claret FX. Trastuzumab: updated mechanisms of action and resistance in breast cancer. Front Oncol. 2012;2: 62. doi:10.3389/fonc.2012.00062

76. Albarello L, Pecciarini L, Doglioni C. HER2 testing in gastric cancer. Adv Anat Pathol. 2011;18: 53–9. doi:10.1097/PAP.0b013e3182026d72

77. Bang Y-J, Van Cutsem E, Feyereislova A, Chung HC, Shen L, Sawaki A, et al. Trastuzumab in combination with chemotherapy versus chemotherapy alone for treatment of HER2-positive advanced gastric or gastro-oesophageal junction cancer (ToGA): a phase 3, open-label, randomised controlled trial. Lancet. 2010;376: 687–97. doi:10.1016/S0140-6736(10)61121-X

78. FDA Issues Advice to Make Earliest Stages Of Clinical Drug Development More Efficient. In: U.S. Food and Drug Administration [Internet]. 2006 [cited 26 Apr 2015]. Available: http://www.fda.gov/NewsEvents/Newsroom/PressAnnouncements/2006/ucm108576.htm

79. Ghofrani HA, Osterloh IH, Grimminger F. Sildenafil: from angina to erectile dysfunction to pulmonary hypertension and beyond. Nat Rev Drug Discov. 2006;5: 689–702. doi:10.1038/nrd2030

80. Wishart DS, Knox C, Guo AC, Cheng D, Shrivastava S, Tzur D, et al. DrugBank: a knowledgebase for drugs, drug actions and drug targets. Nucleic Acids Res. 2008;36: D901–6. doi:10.1093/nar/gkm958

81. Luis M, Tavares A, Carvalho LS, Lara-Santos L, Araújo A, de Mello RA. Personalizing therapies for gastric cancer: molecular mechanisms and novel targeted therapies. World J Gastroenterol. 2013;19: 6383–97. doi:10.3748/wjg.v19.i38.6383

82. Fischer HP. Mathematical modeling of complex biological systems: from parts lists to understanding systems behavior. Alcohol Res Health. 2008;31: 49–59. 124

83. Hughey JJ, Lee TK, Covert MW. Computational modeling of mammalian signaling networks. Wiley Interdiscip Rev Syst Biol Med. 2: 194–209. doi:10.1002/wsbm.52

84. Lev Bar-Or R, Maya R, Segel L a, Alon U, Levine a J, Oren M. Generation of oscillations by the p53-Mdm2 feedback loop: a theoretical and experimental study. Proc Natl Acad Sci U S A. 2000;97: 11250–5. doi:10.1073/pnas.210171597

85. Hughey JJ, Lee TK, Covert MW. Computational modeling of mammalian signaling networks. 2009; doi:10.1002/wsbm.052

86. De Jong H. Modeling and simulation of genetic regulatory systems: a literature review. J Comput Biol. 2002;9: 67–103. doi:10.1089/10665270252833208

87. Andrews SS, Arkin AP. Simulating cell biology. Curr Biol. 2006;16: R523–7. doi:10.1016/j.cub.2006.06.048

88. Ferrell JE, Tsai TY-C, Yang Q. Modeling the cell cycle: why do certain circuits oscillate? Cell. Elsevier Inc.; 2011;144: 874–85. doi:10.1016/j.cell.2011.03.006

89. A.K. Konopka. Systems Biology: Principles, Methods, and Concepts. 2006.

90. Woolf P. Dynamic Systems Analysis II: Evaluation Stability, Eigenvalues. Chemical Process Dynamics and Controls. Open.Michigan – educational resources; 2008.

91. Marwaha S, Schumacher MA, Zavros Y, Eghbalnia HR. Crosstalks between Cytokines and Sonic Hedgehog in Helicobacter pylori Infection: A Mathematical Model. PLoS One. 2014;9: e111338. doi:10.1371/journal.pone.0111338

92. Kuipers EJ, Uyterlinde AM, Peña AS, Roosendaal R, Pals G, Nelis GF, et al. Long-term sequelae of Helicobacter pylori gastritis. Lancet. 1995;345: 1525–8.

93. Sipponen P, Hyvärinen H. Role of Helicobacter pylori in the pathogenesis of gastritis, peptic ulcer and gastric cancer. Scand J Gastroenterol Suppl. 1993;196: 3–6.

94. Blaser MJ, Chyou PH, Nomura A. Age at establishment of Helicobacter pylori infection and gastric carcinoma, gastric ulcer, and duodenal ulcer risk. Cancer Res. 1995;55: 562–5.

95. Ernst P. Review article: the role of inflammation in the pathogenesis of gastric cancer. Aliment Pharmacol Ther. 1999;13 Suppl 1: 13–8.

96. Sawai N, Kita M, Kodama T, Tanahashi T, Yamaoka Y, Tagawa Y, et al. Role of gamma interferon in Helicobacter pylori-induced gastric inflammatory responses in a mouse model. Infect Immun. 1999;67: 279–85.

97. Ismail HF, Fick P, Zhang J, Lynch RG, Berg DJ. Depletion of neutrophils in IL-10(-/-) mice delays clearance of gastric Helicobacter infection and decreases the Th1 immune response to Helicobacter. J Immunol. 2003;170: 3782–9.

125

98. Algood HMS, Cover TL. Helicobacter pylori persistence: an overview of interactions between H. pylori and host immune defenses. Clin Microbiol Rev. 2006;19: 597–613. doi:10.1128/CMR.00006-06

99. Katoh Y, Katoh M. Hedgehog signaling pathway and gastric cancer. Cancer Biol Ther. 2005;4: 1050–4.

100. Katoh Y, Katoh M. Hedgehog signaling pathway and gastrointestinal stem cell signaling network (review). Int J Mol Med. 2006;18: 1019–23.

101. Lee S-Y, Han HS, Lee KY, Hwang TS, Kim JH, Sung I-K, et al. Sonic hedgehog expression in gastric cancer and gastric adenoma. Oncol Rep. 2007;17: 1051–5.

102. Myung K, Sang J, Sik H, Kyun D, Sun H, Hahm K. Late reactivation of sonic hedgehog by Helicobacter pylori results in population of gastric epithelial cells that are resistant to apoptosis : Implication for gastric carcinogenesis. Cancer Lett. Elsevier Ireland Ltd; 2010;287: 44–53. doi:10.1016/j.canlet.2009.05.032

103. Wan J, Zhou J, Zhao H, Wang M, Wei Z, Gao H, et al. Sonic hedgehog pathway contributes to gastric cancer cell growth and proliferation. Biores Open Access. 2014;3: 53–9. doi:10.1089/biores.2014.0001

104. Martin J, Donnelly JM, Houghton J, Zavros Y. The role of sonic hedgehog reemergence during gastric cancer. Dig Dis Sci. 2010;55: 1516–24. doi:10.1007/s10620-010-1252-z

105. Van den Brink GR, Hardwick JC, Tytgat GN, Brink MA, Ten Kate FJ, Van Deventer SJ, et al. Sonic hedgehog regulates gastric gland morphogenesis in man and mouse. Gastroenterology. 2001;121: 317–28.

106. Xiao C, Ogle S a, Schumacher M a, Orr-Asman M a, Miller ML, Lertkowit N, et al. Loss of parietal cell expression of Sonic hedgehog induces hypergastrinemia and hyperproliferation of surface mucous cells. Gastroenterology. Elsevier Inc.; 2010;138: 550–61, 561.e1–8. doi:10.1053/j.gastro.2009.11.002

107. Lowrey JA, Stewart GA, Lindey S, Hoyne GF, Dallman MJ, Howie SEM, et al. Sonic hedgehog promotes cell cycle progression in activated peripheral CD4(+) T lymphocytes. J Immunol. 2002;169: 1869–75.

108. Crompton T, Outram S V, Hager-Theodorides AL. Sonic hedgehog signalling in T-cell development and activation. Nat Rev Immunol. 2007;7: 726–35. doi:10.1038/nri2151

109. Katoh M. Dysregulation of stem cell signaling network due to germline mutation, SNP, Helicobacter pylori infection, epigenetic change and genetic alteration in gastric cancer. Cancer Biol Ther. 2007;6: 832–9.

110. El-Zaatari M, Kao JY, Tessier A, Bai L, Hayes MM, Fontaine C, et al. Gli1 deletion prevents Helicobacter-induced gastric metaplasia and expansion of myeloid cell subsets. PLoS One. 2013;8: e58935. doi:10.1371/journal.pone.0058935 126

111. Schumacher MA, Donnelly JM, Engevik AC, Xiao C, Yang L, Kenny S, et al. Gastric Sonic Hedgehog acts as a macrophage chemoattractant during the immune response to Helicobacter pylori. Gastroenterology. 2012;142: 1150–1159.e6. doi:10.1053/j.gastro.2012.01.029

112. Blaser MJ, Kirschner D. Dynamics of Helicobacter pylori colonization in relation to the host response. Proc Natl Acad Sci U S A. 1999;96: 8359–64.

113. Kirschner DE, Blaser MJ. The dynamics of Helicobacter pylori infection of the human stomach. J Theor Biol. 1995;176: 281–90. doi:10.1006/jtbi.1995.0198

114. Bisset K, Alam MM, Bassaganya-Riera J, Carbo A, Eubank S, Hontecillas R, et al. High- Performance Interaction-Based Simulation of Gut Immunopathologies with ENteric Immunity Simulator (ENISI). 2012 IEEE 26th Int Parallel Distrib Process Symp. Ieee; 2012; 48–59. doi:10.1109/IPDPS.2012.15

115. Carbo A, Bassaganya-Riera J, Pedragosa M, Viladomiu M, Marathe M, Eubank S, et al. Predictive computational modeling of the mucosal immune responses during Helicobacter pylori infection. PLoS One. 2013;8: e73365. doi:10.1371/journal.pone.0073365

116. Franke R, Müller M, Wundrack N, Gilles E-D, Klamt S, Kähne T, et al. Host-pathogen systems biology: logical modelling of hepatocyte growth factor and Helicobacter pylori induced c-Met signal transduction. BMC Syst Biol. 2008;2: 4. doi:10.1186/1752-0509-2-4

117. Ghosh S, Matsuoka Y, Asai Y, Hsin K-Y, Kitano H. Software for systems biology: from tools to integrated platforms. Nat Rev Genet. 2011;12: 821–32. doi:10.1038/nrg3096

118. Polk DB, Peek RM. Helicobacter pylori: gastric cancer and beyond. Nat Rev Cancer. Nature Publishing Group; 2010;10: 403–14. doi:10.1038/nrc2857

119. Viala J, Chaput C, Boneca IG, Cardona A, Girardin SE, Moran AP, et al. Nod1 responds to peptidoglycan delivered by the Helicobacter pylori cag pathogenicity island. Nat Immunol. 2004;5: 1166–74. doi:10.1038/ni1131

120. Higashi H, Tsutsumi R, Muto S, Sugiyama T, Azuma T, Asaka M, et al. SHP-2 tyrosine phosphatase as an intracellular target of Helicobacter pylori CagA protein. Science. 2002;295: 683–6. doi:10.1126/science.1067147

121. Suzuki M, Mimuro H, Kiga K, Fukumatsu M, Ishijima N, Morikawa H, et al. Helicobacter pylori CagA phosphorylation-independent function in epithelial proliferation and inflammation. Cell Host Microbe. 2009;5: 23–34. doi:10.1016/j.chom.2008.11.010

122. Nozawa Y, Nishihara K, Peek RM, Nakano M, Uji T, Ajioka H, et al. Identification of a signaling cascade for interleukin-8 production by Helicobacter pylori in human gastric epithelial cells. Biochem Pharmacol. 2002;64: 21–30.

123. Backert S, Naumann M. What a disorder: proinflammatory signaling pathways induced by Helicobacter pylori. Trends Microbiol. 2010;18: 479–86. doi:10.1016/j.tim.2010.08.003 127

124. Allison CC, Kufer TA, Kremmer E, Kaparakis M, Ferrero RL. Helicobacter pylori induces MAPK phosphorylation and AP-1 activation via a NOD1-dependent mechanism. J Immunol. 2009;183: 8099–109. doi:10.4049/jimmunol.0900664

125. Nakayama M, Kimura M, Wada A, Yahiro K, Ogushi K, Niidome T, et al. Helicobacter pylori VacA activates the p38/activating transcription factor 2-mediated signal pathway in AZ-521 cells. J Biol Chem. 2004;279: 7024–8. doi:10.1074/jbc.M308898200

126. Naito Y, Yoshikawa T. Molecular and cellular mechanisms involved in Helicobacter pylori- induced inflammation and oxidative stress. Free Radic Biol Med. 2002;33: 323–36.

127. Yamada H, Aihara T, Okabe S. Mechanism for Helicobacter pylori stimulation of interleukin-8 production in a gastric epithelial cell line (MKN 28): roles of mitogen-activated protein kinase and interleukin-1beta. Biochem Pharmacol. 2001;61: 1595–604.

128. Seo JH, Lim JW, Kim H, Kim KH. Helicobacter pylori in a Korean isolate activates mitogen- activated protein kinases, AP-1, and NF-kappaB and induces chemokine expression in gastric epithelial AGS cells. Lab Invest. 2004;84: 49–62. doi:10.1038/sj.labinvest.3700010

129. Aihara M, Tsuchimoto D, Takizawa H, Azuma A, Wakebe H, Ohmoto Y, et al. Mechanisms involved in Helicobacter pylori-induced interleukin-8 production by a gastric cancer cell line, MKN45. Infect Immun. 1997;65: 3218–24.

130. Remick DG, Green LB, Newcomb DE, Garg SJ, Bolgos GL, Call DR. CXC chemokine redundancy ensures local neutrophil recruitment during acute inflammation. Am J Pathol. 2001;159: 1149– 57. doi:10.1016/S0002-9440(10)61791-9

131. Fehlings M, Drobbe L, Moos V, Renner Viveros P, Hagen J, Beigier-Bompadre M, et al. Comparative analysis of the interaction of Helicobacter pylori with human dendritic cells, macrophages, and monocytes. Infect Immun. 2012;80: 2724–34. doi:10.1128/IAI.00381-12

132. Mai UE, Perez-Perez GI, Wahl LM, Wahl SM, Blaser MJ, Smith PD. Soluble surface proteins from Helicobacter pylori activate monocytes/macrophages by lipopolysaccharide-independent mechanism. J Clin Invest. 1991;87: 894–900. doi:10.1172/JCI115095

133. Bimczok D, Smythies LE, Waites KB, Grams JM, Stahl RD, Mannon PJ, et al. Helicobacter pylori infection inhibits phagocyte clearance of apoptotic gastric epithelial cells. J Immunol. 2013;190: 6626–34. doi:10.4049/jimmunol.1203330

134. Bliss CM, Golenbock DT, Keates S, Linevsky JK, Kelly CP. Helicobacter pylori lipopolysaccharide binds to CD14 and stimulates release of interleukin-8, epithelial neutrophil-activating peptide 78, and monocyte chemotactic protein 1 by human monocytes. Infect Immun. 1998;66: 5357– 63.

135. De Jonge R, Kusters JG, Timmer MS, Gimmel V, Appelmelk BJ, Bereswill S, et al. The role of Helicobacter pylori virulence factors in interleukin production by monocytic cells. FEMS Microbiol Lett. 2001;196: 235–8.

128

136. Lindgren A, Pavlovic V, Flach C-F, Sjöling A, Lundin S. Interferon-gamma secretion is induced in IL-12 stimulated human NK cells by recognition of Helicobacter pylori or TLR2 ligands. Innate Immun. 2011;17: 191–203. doi:10.1177/1753425909357970

137. Yun CH, Lundgren A, Azem J, Sjöling A, Holmgren J, Svennerholm A-M, et al. Natural killer cells and Helicobacter pylori infection: bacterial antigens and interleukin-12 act synergistically to induce gamma interferon production. Infect Immun. 2005;73: 1482–90. doi:10.1128/IAI.73.3.1482-1490.2005

138. Pellicanò A, Sebkova L, Monteleone G, Guarnieri G, Imeneo M, Pallone F, et al. Interleukin-12 drives the Th1 signaling pathway in Helicobacter pylori-infected human gastric mucosa. Infect Immun. 2007;75: 1738–44. doi:10.1128/IAI.01446-06

139. Kim J-H, Choi YJ, Lee SH, Shin HS, Lee IO, Kim YJ, et al. Effect of Helicobacter pylori infection on the sonic hedgehog signaling pathway in gastric cancer cells. Oncol Rep. 2010;23: 1523–8.

140. Singh AP, Arora S, Bhardwaj A, Srivastava SK, Kadakia MP, Wang B, et al. CXCL12/CXCR4 protein signaling axis induces sonic hedgehog expression in pancreatic cancer cells via extracellular regulated kinase- and Akt kinase-mediated activation of nuclear factor κB: implications for bidirectional tumor-stromal interactions. J Biol Chem. 2012;287: 39115–24. doi:10.1074/jbc.M112.409581

141. Nakashima H, Nakamura M, Yamaguchi H, Yamanaka N, Akiyoshi T, Koga K, et al. Nuclear factor-kappaB contributes to hedgehog signaling pathway activation through sonic hedgehog induction in pancreatic cancer. Cancer Res. 2006;66: 7041–9. doi:10.1158/0008-5472.CAN-05- 4588

142. Waghray M, Zavros Y, Saqui-Salces M, El-Zaatari M, Alamelumangapuram CB, Todisco A, et al. Interleukin-1beta promotes gastric atrophy through suppression of Sonic Hedgehog. Gastroenterology. Elsevier Inc.; 2010;138: 562–72, 572.e1–2. doi:10.1053/j.gastro.2009.10.043

143. Haeberle H a, Kubin M, Bamford KB, Garofalo R, Graham DY, El-Zaatari F, et al. Differential stimulation of interleukin-12 (IL-12) and IL-10 by live and killed Helicobacter pylori in vitro and association of IL-12 production with gamma interferon-producing T cells in the human gastric mucosa. Infect Immun. 1997;65: 4229–35.

144. Robinson K, Kenefeck R, Pidgeon EL, Shakib S, Patel S, Polson RJ, et al. Helicobacter pylori- induced peptic ulcer disease is associated with inadequate regulatory T cell responses. Gut. 2008;57: 1375–85. doi:10.1136/gut.2007.137539

145. Wang P, Wu P, Siegel MI, Egan RW, Billah MM. Interleukin (IL)-10 inhibits nuclear factor kappa B (NF kappa B) activation in human monocytes. IL-10 and IL-4 suppress cytokine synthesis by different mechanisms. J Biol Chem. 1995;270: 9558–63.

146. Romano MF, Lamberti A, Petrella A, Bisogni R, Tassone PF, Formisano S, et al. IL-10 inhibits nuclear factor-kappa B/Rel nuclear activity in CD3-stimulated human peripheral T lymphocytes. J Immunol. 1996;156: 2119–23.

129

147. Kundu P, De R, Pal I, Mukhopadhyay AK, Saha DR, Swarnakar S. Curcumin alleviates matrix metalloproteinase-3 and -9 activities during eradication of Helicobacter pylori infection in cultured cells and mice. PLoS One. 2011;6: e16306. doi:10.1371/journal.pone.0016306

148. Algood HMS, Gallo-Romero J, Wilson KT, Peek RM, Cover TL. Host response to Helicobacter pylori infection before initiation of the adaptive immune response. FEMS Immunol Med Microbiol. 2007;51: 577–86. doi:10.1111/j.1574-695X.2007.00338.x

149. Lee A, O’Rourke J, De Ungria MC, Robertson B, Daskalopoulos G, Dixon MF. A standardized mouse model of Helicobacter pylori infection: introducing the Sydney strain. Gastroenterology. 1997;112: 1386–97.

150. Thompson LJ, Danon SJ, Wilson JE, O’Rourke JL, Salama NR, Falkow S, et al. Chronic Helicobacter pylori infection with Sydney strain 1 and a newly identified mouse-adapted strain (Sydney strain 2000) in C57BL/6 and BALB/c mice. Infect Immun. 2004;72: 4668–79. doi:10.1128/IAI.72.8.4668-4679.2004

151. Koschorreck M, Conzelmann H, Ebert S, Ederer M, Gilles ED. Reduced modeling of signal transduction - a modular approach. BMC Bioinformatics. 2007;8: 336. doi:10.1186/1471-2105- 8-336

152. Anderson J, Chang Y-C, Papachristodoulou A. Model decomposition and reduction tools for large-scale networks in systems biology. Automatica. Elsevier Ltd; 2011;47: 1165–1174. doi:10.1016/j.automatica.2011.03.010

153. Maurya MR, Bornheimer SJ, Venkatasubramanian V, Subramaniam S. Reduced-order modelling of biochemical networks: application to the GTPase-cycle signalling module. Syst Biol (Stevenage). 2005;152: 229–42.

154. Brazhnik P, Tyson JJ. Cell cycle control in bacteria and yeast: a case of convergent evolution? Cell Cycle. 2006;5: 522–9.

155. Novak B, Pataki Z, Ciliberto A, Tyson JJ. Mathematical model of the cell division cycle of fission yeast. Chaos. 2001;11: 277–286. doi:10.1063/1.1345725

156. Ciliberto A, Petrus MJ, Tyson JJ, Sible JC. A kinetic model of the cyclin EyCdk2 developmental timer in Xenopus laevis embryos. Biophys Chem. 2003;104: 573–589. doi:10.1016/S0301-4622

157. Chen KC, Calzone L, Csikasz-nagy A, Cross FR, Novak B, Tyson JJ. Integrative Analysis of Cell Cycle Control in Budding Yeast □. Mol Biol Cell. 2004;15: 3841–3862. doi:10.1091/mbc.E03

158. Zhang T, Brazhnik P, Tyson JJ. Computational analysis of dynamical responses to the intrinsic pathway of programmed cell death. Biophys J. Biophysical Society; 2009;97: 415–34. doi:10.1016/j.bpj.2009.04.053

159. Funahashi A, Morohashi M, Kitano H, Tanimura N. CellDesigner : a process diagram editor for gene-regulatory and. 2003;1: 159–162.

130

160. Hoops S, Sahle S, Gauges R, Lee C, Pahle J, Simus N, et al. COPASI--a COmplex PAthway SImulator. Bioinformatics. 2006;22: 3067–74. doi:10.1093/bioinformatics/btl485

161. Mueller A, Merrell DS, Grimm J, Falkow S. Profiling of microdissected gastric epithelial cells reveals a cell type—specific response to Helicobacter pylori infection. Gastroenterology. 2004;127: 1446–1462. doi:10.1053/j.gastro.2004.08.054

162. Reinard JC. Communication Research Statistics. 2006. p. 218.

163. Bodger K, Wyatt JI, Heatley R V. Gastric mucosal secretion of interleukin- 10 : relations to histopathology , Helicobacter pyloni status , and tumour necrosis factor-a secretion. 1997; 739–744.

164. Ekström AM, Held M, Hansson LE, Engstrand L, Nyrén O. Helicobacter pylori in gastric cancer established by CagA immunoblot as a marker of past infection. Gastroenterology. 2001;121: 784–91.

165. Nookaew I, Thorell K, Worah K, Wang S, Hibberd ML, Sjövall H, et al. Transcriptome signatures in Helicobacter pylori-infected mucosa identifies acidic mammalian chitinase loss as a corpus atrophy marker. BMC Med Genomics. 2013;6: 41. doi:10.1186/1755-8794-6-41

166. Edgar R, Domrachev M, Lash AE. Gene Expression Omnibus: NCBI gene expression and hybridization array data repository. Nucleic Acids Res. 2002;30: 207–10.

167. Smyth GK. Limma : Linear Models for Microarray Data. : 397–420.

168. Chen J, Bardes EE, Aronow BJ, Jegga AG. ToppGene Suite for gene list enrichment analysis and candidate gene prioritization. Nucleic Acids Res. 2009;37: W305–11. doi:10.1093/nar/gkp427

169. Carlson M. hgu133a.db: Affymetrix Human Genome U133 Set annotation data (chip hgu133a).

170. Lamb J, Crawford ED, Peck D, Modell JW, Blat IC, Wrobel MJ, et al. The Connectivity Map: using gene-expression signatures to connect small molecules, genes, and disease. Science. 2006;313: 1929–35. doi:10.1126/science.1132939

171. Eftang LL, Esbensen Y, Tannæs TM, Blom GP, Bukholm IRK, Bukholm G. Up-regulation of CLDN1 in gastric cancer is correlated with reduced survival. BMC Cancer. 2013;13: 586. doi:10.1186/1471-2407-13-586

172. Tsai CJ, Herrera-Goepfert R, Tibshirani RJ, Yang S, Mohar A, Guarner J, et al. Changes of gene expression in gastric preneoplasia following Helicobacter pylori eradication therapy. Cancer Epidemiol Biomarkers Prev. 2006;15: 272–80. doi:10.1158/1055-9965.EPI-05-0632

173. El-Rifai W, Moskaluk CA, Abdrabbo MK, Harper J, Yoshida C, Riggins GJ, et al. Gastric cancers overexpress S100A calcium-binding proteins. Cancer Res. 2002;62: 6823–6.

131

174. Wenqi D, Li W, Shanshan C, Bei C, Yafei Z, Feihu B, et al. EpCAM is overexpressed in gastric cancer and its downregulation suppresses proliferation of gastric cancer. J Cancer Res Clin Oncol. 2009;135: 1277–85. doi:10.1007/s00432-009-0569-5

175. Lin Z, Zhang C, Zhang M, Xu D, Fang Y, Zhou Z, et al. Targeting cadherin-17 inactivates Ras/Raf/MEK/ERK signaling and inhibits cell proliferation in gastric cancer. PLoS One. 2014;9: e85296. doi:10.1371/journal.pone.0085296

176. Wang J, Kang W-M, Yu J-C, Liu Y-Q, Meng Q-B, Cao Z-J. Cadherin-17 induces tumorigenesis and lymphatic metastasis in gastric cancer through activation of NFκB signaling pathway. Cancer Biol Ther. 2013;14: 262–70. doi:10.4161/cbt.23299

177. Sun X, Hao D, Zheng Z, Fu H, Xu H, Wang M, et al. [Screening and analysis of associated genes in the carcinogenesis and progression of gastric cancer]. Zhonghua Yi Xue Yi Chuan Xue Za Zhi. 2005;22: 31–4.

178. Axon ATR. Relationship between Helicobacter pylori gastritis, gastric cancer and gastric acid secretion. Adv Med Sci. 2007;52: 55–60.

179. Sipponen P, Kekki M, Seppälä K, Siurala M. The relationships between chronic gastritis and gastric acid secretion. Aliment Pharmacol Ther. 1996;10 Suppl 1: 103–18.

180. He Q-Y, Cheung YH, Leung SY, Yuen ST, Chu K-M, Chiu J-F. Diverse proteomic alterations in gastric adenocarcinoma. Proteomics. 2004;4: 3276–87. doi:10.1002/pmic.200300916

181. Kola I, Landis J. Can the pharmaceutical industry reduce attrition rates? Nat Rev Drug Discov. 2004;3: 711–5. doi:10.1038/nrd1470

182. Gupta SC, Sung B, Prasad S, Webb LJ, Aggarwal BB. Cancer drug discovery by repurposing: teaching new tricks to old dogs. Trends Pharmacol Sci. 2013;34: 508–17. doi:10.1016/j.tips.2013.06.005

183. Hollander M, Wolfe DA, Chicken E. Nonparametric Statistical Methods. 2nd ed. New York: Wiley.; 1999.

184. Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A. 2005;102: 15545–50. doi:10.1073/pnas.0506580102

185. Mazzarella Ri, Webb C. APPLICATION OF TECHNOLOGY PLATFORMS TO UNCOVER NEW INDICATIONS AND REPURPOSE EXISTING DRUGS. Drug Repositioning: Bringing New Life to Shelved Assets and Existing Drugs. 2012. pp. 107 – 112. doi:10.1002/9781118274408

186. Cui J, Li F, Wang G, Fang X, Puett JD, Xu Y. Gene-expression signatures can distinguish gastric cancer grades and stages. PLoS One. 2011;6: e17819. doi:10.1371/journal.pone.0017819

132

187. Cho JY, Lim JY, Cheong JH, Park Y-Y, Yoon S-L, Kim SM, et al. Gene expression signature-based prognostic risk score in gastric cancer. Clin Cancer Res. 2011;17: 1850–7. doi:10.1158/1078- 0432.CCR-10-2180

188. Kim HK, Choi IJ, Kim CG, Kim HS, Oshima A, Michalowski A, et al. A gene expression signature of acquired chemoresistance to cisplatin and fluorouracil combination chemotherapy in gastric cancer patients. PLoS One. 2011;6: e16694. doi:10.1371/journal.pone.0016694

189. Wu J, Hu C, Gu Q, Li Y, Song M. Trichostatin A sensitizes cisplatin-resistant A549 cells to apoptosis by up-regulating death-associated protein kinase. Acta Pharmacol Sin. 2010;31: 93– 101. doi:10.1038/aps.2009.183

190. Warmoes M, Jaspers JE, Xu G, Sampadi BK, Pham T V, Knol JC, et al. Proteomics of genetically engineered mouse mammary tumors identifies fatty acid metabolism members as potential predictive markers for cisplatin resistance. Mol Cell Proteomics. 2013;12: 1319–34. doi:10.1074/mcp.M112.024182

191. Vazquez-Martin A, Ropero S, Brunet J, Colomer R, Menendez JA. Inhibition of Fatty Acid Synthase (FASN) synergistically enhances the efficacy of 5-fluorouracil in breast carcinoma cells. Oncol Rep. 2007;18: 973–80.

192. Zhao Y, Butler EB, Tan M. Targeting cellular metabolism to improve cancer therapeutics. Cell Death Dis. 2013;4: e532. doi:10.1038/cddis.2013.60

193. Durán R V, Oppliger W, Robitaille AM, Heiserich L, Skendaj R, Gottlieb E, et al. Glutaminolysis activates Rag-mTORC1 signaling. Mol Cell. 2012;47: 349–58. doi:10.1016/j.molcel.2012.05.043

194. Kranzer K, Eckhardt A, Aigner M, Knoll G, Deml L, Speth C, et al. Induction of maturation and cytokine release of human dendritic cells by Helicobacter pylori. Infect Immun. 2004;72: 4416– 23. doi:10.1128/IAI.72.8.4416-4423.2004

195. Tarca AL, Romero R, Draghici S. Analysis of microarray experiments of gene expression profiling. Am J Obstet Gynecol. 2006;195: 373–88. doi:10.1016/j.ajog.2006.07.001

196. Duan Q, Flynn C, Niepel M, Hafner M, Muhlich JL, Fernandez NF, et al. LINCS Canvas Browser: interactive web app to query, browse and interrogate LINCS L1000 gene expression signatures. Nucleic Acids Res. 2014;42: W449–60. doi:10.1093/nar/gku476

197. Barretina J, Caponigro G, Stransky N, Venkatesan K, Margolin AA, Kim S, et al. The Cancer Cell Line Encyclopedia enables predictive modelling of anticancer drug sensitivity. Nature. 2012;483: 603–7. doi:10.1038/nature11003

198. Greshock J, Bachman KE, Degenhardt YY, Jing J, Wen YH, Eastman S, et al. Molecular target class is predictive of in vitro response profile. Cancer Res. 2010;70: 3677–86. doi:10.1158/0008-5472.CAN-09-3788

133