Embed Size (px)
Citation preview
Early Detection and Treatment of Breast Cancer
by Random Peptide Array in neuN Transgenic Mouse Model
by
Hu Duan
A Dissertation Presented in Partial Fulfillment
of the Requirements for the Degree
Doctor of Philosophy
Approved June 2015 by the:
Graduate Supervisory Committee:
Stephen Albert Johnston, Chair
Leland Hartwell
Valentin Dinu
Yung Chang
ARIZONA STATE UNIVERSITY
August 2015
i
ABSTRACT
Breast cancer is the most common cancer and currently the second leading
cause of death among women in the United States. Patients’ five-year relative
survival rate decreases from 99% to 25% when breast cancer is diagnosed late.
Immune checkpoint blockage has shown to be a promising therapy to improve
patients’ outcome in many other cancers. However, due to the lack of early
diagnosis, the treatment is normally given in the later stages. An early diagnosis
system for breast cancer could potentially revolutionize current treatment
strategies, improve patients’ outcomes and even eradicate the disease. The current
breast cancer diagnostic methods cannot meet this demand. A simple, effective,
noninvasive and inexpensive early diagnostic technology is needed.
Immunosignature technology leverages the power of the immune system to find
cancer early. Antibodies targeting tumor antigens in the blood are probed on a
high-throughput random peptide array and generate a specific binding pattern
called the immunosignature.
In this dissertation, I propose a scenario for using immunosignature
technology to detect breast cancer early and to implement an early treatment
strategy by using the PD-L1 immune checkpoint inhibitor. I develop a
methodology to describe the early diagnosis and treatment of breast cancer in a
FVB/N neuN breast cancer mouse model. By comparing FVB/N neuN transgenic
mice and age-matched wild type controls, I have found and validated specific
immunosignatures at multiple time points before tumors are palpable.
Immunosignatures change along with tumor development. Using a late-stage
ii
immunosignature to predict early samples, or vice versa, cannot achieve high
prediction performance. By using the immunosignature of early breast cancer, I
show that at the time of diagnosis, early treatment with the checkpoint blockade,
anti-PD-L1, inhibits tumor growth in FVB/N neuN transgenic mouse model. The
mRNA analysis of the PD-L1 level in mice mammary glands suggests that it is
more effective to have treatment early.
Novel discoveries are changing understanding of breast cancer and
improving strategies in clinical treatment. Researchers and healthcare
professionals are actively working in the early diagnosis and early treatment fields.
This dissertation provides a step along the road for better diagnosis and treatment
of breast cancer.
iii
DEDICATION
I dedicate this dissertation to my wife Xinyao for her unconditional love and
support, and my parents who accompany and help me during these days
And
To my son Roger who is the best gift for me before my graduation
iv
ACKNOWLEDGEMENTS
Six years of my PhD studies have led to this dissertation. I have
experienced a lot during these years, and many people have influenced me
positively. I want to offer my most sincere thanks to all the people who helped
make me who I have become during these six years.
Special thanks to my advisor, Dr. Stephen Albert Johnston. Dr. Johnston
has inspired me and patiently taught me the most during my PhD studies. When I
faced a complex problem, he always inspired and guided me to break it into small
steps and push my thoughts further. One of his favorite expressions is “Keep it
simple, stupid”—the KISS principle—I cannot tell express how much I benefited
from this advice. I also learned a very important way to make decisions when the
choices are controversial. Dr. Johnston fully supported me in my dissertation
project through both his efforts to prepare me intellectually and his willingness to
offer me specific scientific advice. Dr. Leland Hartwell has been more than a
committee member for me. Besides his insightful comments on my dissertation,
what I learned most came from his spirit. His constant eagerness to learn
knowledge in a new field, his integrity and his modesty set an excellent example
for me to follow. I have also consulted with him on my career development and
shared many thoughts about my life and dreams. Dr. Hartwell always encouraged
and advised me, which made me more confident in pursuing my dream. Every
time I talked with Dr. Valentin Dinu, he provided me with new knowledge on
bioinformatics analysis. Although Dr. Dinu was mainly at the Mayo Clinic, and it
was a long way for him to come to Biodesign, he was always very willing to meet
v
me with great patience and kindness. I learned immunology from Dr. Yung Chang
when I took her classes in 2010. I would like to thank her for sharing her
knowledge of immunology and for supporting my dissertation project.
I also want to thank Dr. Kathy Sykes, Dr. Sid Hecht and Dr. Tim Karr,
who were on my comprehensive exam committee. They helped and supported me
in becoming a PhD candidate. During my PhD studies, it was a pleasure to work
with the people in the Center for Innovations in Medicine in the Biodesign
Institute at ASU. I am grateful to be a member of the Biological Design Graduate
Program, which provides funding to support me. I thank Dr. Joann Williams,
Maria Hanlin, Laura Hawes and Lauren Dempsey, who guided me in each step
toward graduation and provided managerial support for my PhD program.
With regard to my dissertation studies, I would like to specially thank the
following individuals. Dr. Luhui Shen was the major collaborator in my
dissertation project. Dr. Bart Legutki was my first guide when I did my rotation in
CIM back in 2009. John Lainson taught me a lot about microarray production and
automatic robot systems, and we had many amusing talks that made life at CIM
more interesting. I shared an office with Dr. Josh Richer for two years, and we
spent quite a lot time together discussing each other’s experiment. The many
arrays that were run in the past six years would not have been possible without the
support of Dr. Zbigniew Cichacz and the Peptide Array Core team.
I would like to give special thanks to my friends Andy Ge, Jie Wang,
Xiaofang Bian and Xiaobo Yu. They provided me with tremendous emotional
support and have encouraged me all along the way.
vi
TABLE OF CONTENTS
Page
LIST OF TABLES ............................................................................................ xii
LIST OF FIGURES .......................................................................................... xiii
CHAPTER
1 INTRODUCTION ....................................................................................... 1
1.1 Breast Cancer ....................................................................................... 1
1.1.1 Overview of Cancer ...................................................................... 1
1.1.2 Overview of Breast Cancer ........................................................... 2
1.1.3 Overview of Transgenic Mouse Model ......................................... 4
1.2 Current Breast Cancer Diagnosis and Biomarkers ................................ 6
1.2.1 Statistics Preparation for Diagnosis and Biomarker ....................... 6
1.2.2 Current Clinical Diagnostic Methods of Breast Cancer ................. 7
1.2.2.1 Breast Self-Examination (BSE) ............................................. 7
1.2.2.2 Clinical Breast Examination (CBE)....................................... 8
1.2.2.3 Mammography ..................................................................... 8
1.2.2.4 Adjunctive Testing: Ultrasound Imaging and MRI .............. 10
1.2.2.5 Molecular Diagnosis Platform ............................................. 11
1.2.3 Benefit of Early Detection and Treatment ................................... 12
1.2.3.1 Early Diagnosis of Breast Cancer Improve Survival Rate13
vii
CHAPTER Page
1.2.3.2 Proposed Early Diagnosis and Treatment Scenario ......... 14
1.3 Early Detection of Breast Cancer by Antibodies ................................. 15
1.3.1 Overview of Blood Based Breast Cancer Biomarkers ................. 15
1.3.2 Challenges in Finding Early Breast Cancer Biomarker................ 17
1.3.3 Cancer Antibodies as Biomarkers ............................................... 19
1.3.3.1 Patients can Develop Autoantibodies Against Tumor Early . 19
1.3.3.2 Advantage of Using Antibodies as Biomarker ..................... 21
1.3.3.3 Breast Cancer Associated Antibodies .................................. 24
1.3.4 Emerging Technologies for Profiling Antibody Responses ......... 25
1.3.4.1 SEREX ............................................................................... 25
1.3.4.2 Peptide and Peptoid Microarrays ......................................... 27
1.3.4.3 SERPA ............................................................................... 28
1.3.4.4 Protein microarray .............................................................. 29
1.4 Immunosignature ............................................................................... 31
1.4.1 Definition of Immunosignature ................................................... 31
1.4.1.1 Building a Complex Chemical Surface................................ 32
1.4.1.2 Probing Sera ....................................................................... 33
1.4.1.3 Analysis of Binding Pattern ................................................ 33
1.4.2 Advantage of Immunosignature .................................................. 34
viii
CHAPTER Page
1.4.3 Published Studies with Immunosignature.................................... 36
1.5 Breast Cancer Treatment .................................................................... 36
1.5.1 Clinical Treatment of Breast Cancer ........................................... 36
1.5.1.1 Surgery and Radiotherapy ................................................... 37
1.5.1.2 Cancer Drugs ...................................................................... 39
1.5.1.2.1 Chemotherapy ............................................................. 39
1.5.1.2.2 Hormone Therapy ....................................................... 40
1.5.1.2.3 Targeted Therapy ........................................................ 40
1.5.2 Emerging Strategy of Immune Checkpoint Blockade .................. 41
2 Early Detection Of Breast Cancer .............................................................. 45
2.1 Introduction........................................................................................ 45
2.2 Results ............................................................................................... 48
2.2.1 HER2 Transgenic Mouse Model ................................................. 48
2.2.2 Scheme of Experiment Design .................................................... 50
2.2.3 Influence of Genetic Background on Immunosignature ............... 51
2.2.4 Pre-tumor Immunosignature Prior to Tumor Palpation ................ 52
2.2.5 Earliest Time Point of the Detectable Immunosignature .............. 53
2.2.6 Time Course of Immunosignatures During Tumor Progression .. 54
2.2.7 Cross-stage Prediction ................................................................ 59
ix
CHAPTER Page
2.3 Methods ............................................................................................. 60
2.3.1 Mice and Sample Collection ....................................................... 60
2.3.2 Preparation of Microarray ........................................................... 61
2.3.3 Probing Serum Antibodies on Peptide Microarrays ..................... 62
2.3.4 Microarray Data Processing ........................................................ 62
2.3.5 Statistical Analysis ..................................................................... 63
2.4 Discussion .......................................................................................... 63
2.5 Supplements ....................................................................................... 68
2.5.1 Selection of multiple batches of array ......................................... 68
2.5.2 Removal of Batch Effect............................................................. 69
3 A MODEL FOR THE EARLY DETECTION AND TREATMENT OF
CANCER .......................................................................................................... 74
3.1 Abstract.............................................................................................. 74
3.2 Introduction........................................................................................ 75
3.3 Results: .............................................................................................. 77
3.3.1 Early Detection of Breast Cancer and PD-L1 Treatment ............. 77
3.3.2 Inhibition of Tumor Growth Rate by Early Treatment ................ 79
3.3.3 RNA Expression of PD-L1 in Mammary Glands ........................ 80
3.3.4 Changes of Immunosignatures After PD-L1 Treatment ............... 82
x
CHAPTER Page
3.4 Methods ............................................................................................. 84
3.4.1 Mice and Sample Collection ....................................................... 84
3.4.2 Binding of Sera to the Peptide Microarrays ................................. 84
3.4.3 PD-L1 Treatment Regime ........................................................... 85
3.4.4 RNA Expression in Mammary Gland ......................................... 86
3.4.5 Statistical Analysis ..................................................................... 87
3.5 Discussion .......................................................................................... 87
REFERENCES .................................................................................................. 93
APPENDIX
A DETECTION OF OSTEOSARCOMA RECURRENCE IN CLINICAL
CANINE SAMPLES BY IMMUNOSIGNATURE .......................................... 118
AP1.1 Introduction ................................................................................. 119
AP1.1.1 Background of Osteosarcoma .............................................. 119
AP1.1.2 Current Diagnosis and Treatment of OSA: ........................... 120
AP1.1.3 Recurrence of OSA .............................................................. 120
AP1.2 Result and Discussion ................................................................. 121
AP1.2.1 Canine Osteosarcoma Clinical Sample Summary ................. 121
AP1.2.2 Correlation Analysis Of Overall Antibody During OSA....... 122
AP1.2.3 Comparison of Immunosignature at Multiple Time Points ... 124
xi
CHAPTER Page
AP1.2.4 Conclusion and Future Direction ......................................... 127
AP1.3 Methods ...................................................................................... 128
AP1.3.1 Sample Summary ................................................................ 128
AP1.3.2 Binding of Sera to the Peptide Microarrays .......................... 129
AP1.3.3 Statistical Analysis .............................................................. 129
B CHARACTERIZATION OF ANTIBODY PROFILE DURING CANCER
DEVELOPMENT............................................................................................ 131
AP 2.1 Quantitative Characterization Of Antibody Profiles During Tumor
Development ................................................................................................... 132
AP2.1.1 Introduction ......................................................................... 132
AP2.1.2: Result and Discussion ......................................................... 133
AP2.2 A Time Series Study of Antibody Profile Along Tumor
Development ................................................................................................... 138
AP2.2.1 Introduction ......................................................................... 138
AP2.2.2 Result and Discussion .......................................................... 139
AP2.3 Robustness of immunosignature .................................................. 146
AP2.3.1 Introduction ......................................................................... 146
AP2.3.2 Result and discussion:.......................................................... 147
C PERMISSIONS ....................................................................................... 150
BIO SKETCH ................................................................................................. 152
xii
LIST OF TABLES
Table Page
2.1: Cross-Stage Prediction Performance in the Training Set ............................. 60
2.2: Correlation Analysis for Different Printing Batches of Slides ...................... 68
2.3: Statistic Characterization of Different Batches of Slides ............................. 69
3.1: The Timeframe for Occurrence of Signatures, Palpable Tumor and
Treatments ......................................................................................................... 78
AP1.1: Canine Osteosarcoma Clinical Sample Summary ................................. 121
AP1.2: Correlation Analysis of Overall Antibody Profile ................................. 123
AP1.3: Distribution of Fold Change in Selected Peptides ................................. 123
AP1.4: Proportion of Overlapped Peptides ....................................................... 126
AP2.1: Statistic characterization of All Time Points for Each Mouse ............... 136
AP2.2: Sample Information for Tg8-1-1 and WT2-3 ........................................ 140
AP2.3: Sample Information for Tg8-3-9 and WT1-3 ........................................ 141
AP2.4: Sample Information for Robustness Test .............................................. 147
xiii
LIST OF FIGURES
Figure Page
1.1 Biology of Breast Tissue ................................................................................ 2
1.2: Illustration of Mammogram and MRI ........................................................... 7
1.3: Five-Year Relative Survival Rates (%) by Stage in US ............................... 13
1.4: Distribution of Human Plasma Protein. ....................................................... 22
1.5: Principle of Immunosignature ..................................................................... 32
1.6: Introduction of CTLA-4 and PD-1 Treatment ............................................. 43
1.7: Biology and Mechanism of Anti-PD-L1 Antibody Treatment ..................... 44
2.1: Tumor Free and Growth Curve of FVB/N NeuN Mouse Model .................. 49
2.2: Scheme of Experiment Design .................................................................... 50
2.3: Influence of Genetic Background on Immunosignature ............................... 52
2.4: Immunosignature Can Show a Separation Between Transgenic Mice and Age
Matched Controls Two Weeks Before First Palpable Tumor .............................. 53
2.5: Heatmap of Immunosignatures of Different Time Points in Tumor
Development…. ................................................................................................ 54
2.6: Time Course of Immunosignatures During Tumor Progression in FVB/N
NeuN Mice ........................................................................................................ 56
2.7: Venn Diagram of Different Immunosignatures Along Tumor Development 58
2.8: Numbers of Selected Peptides Overlap Between Two Adjacent Stages During
Tumor Development .......................................................................................... 58
xiv
Figure Page
2.9: Numbers of Selected Peptides Overlap Between Earliest Tumor Signature
with Other Stages Signatures ............................................................................. 59
2.10: Principle Component Analysis of Different Batches .................................. 70
2.11: Correlation of Technical Replicates Before and After SVA Adjusted and
PCA Analysis of Batch Effects .......................................................................... 71
2.12: Correlation of Technical Replicates Before and After Combat Adjusted and
PCA Analysis of Batch Effects .......................................................................... 72
2.13: Comparison of Different Batch Effect Removal Methods.......................... 73
3.1: Cartoon of Experimental Design Outline..................................................... 78
3.2: Tumor Growth Characteristics in FVB/N NeuN α-PD-L1 Treatment
Experiment ........................................................................................................ 80
3.3: RT-PCR Analysis of MMTV Promotor and PDL1 Expression in Established
Tumors and Adjacent Normal Mammary Glands ............................................... 81
3.4: Heatmap Showing PD-L1 Generated Immunosignature .............................. 83
AP1.1: Proportion of Overlapped Peptides Shared Among Different Time
Points .............................................................................................................. 125
AP1.2: Intensity Plot of Significant Peptides Between Primary and Recurrent
OSA ................................................................................................................ 126
AP2.1: Summary of Mice Samples .................................................................. 134
AP2.2: Tumor Growth Curve for Tg8_1-1 and Tg8_3-9 Mice ......................... 134
AP2.3: AbStat Time Course for Tg8-1-1 and WT2-3 Mice .............................. 135
xv
Figure Page
AP2.4: AbStat Time Course for Tg8-3-9 and WT1-3 Mice .............................. 136
AP2.5: Entropy Time Course for Transgenic and Wild Type Mice .................. 137
AP2.6: Concept of Autocorrelation Function and Durbin-Watson Test ............ 142
AP2.7: Self-Organizing Map to Cluster 249 Autocorrelated Peptides in
Tg8-3-6. .......................................................................................................... 143
AP2.8: Self-Organizing Map to Cluster 849 Autocorrelated Peptides in
Tg8-3-6. .......................................................................................................... 144
AP2.9: Intensity of a Cluster of 24 Autocorrelated Peptides Correlated with
Tumor Development ........................................................................................ 145
AP2.10: Concept of Robustness in Immunosignature ....................................... 147
AP2.11: Correlation of Selected Peptides in Transgenic (Tg) and Wild-Type (WT)
Control Mice. .................................................................................................. 148
AP2.12: Correlation of All Peptides on the 10K Array ..................................... 149
1
CHAPTER 1
INTRODUCTION
1.1 Breast Cancer
1.1.1 Overview of Cancer
Cancer is a group of diseases characterized by the uncontrolled growth
and spread of abnormal cells [1]. Hanahan and Weinberg propose the following
hallmarks of cancer: self-sufficiency in growth signals; evasion of growth
suppressors; limitless replicative ability; resistance to apoptosis; sustained
angiogenesis; activation of invasion and metastasis; evasion of immune
destruction; and deregulation of energy metabolism [2, 3]. Genome instability and
mutation and tumor-promoting inflammation are enabling characteristics of
cancer cells [3]. Cancer is the leading cause of morbidity and mortality in the
world. In 2012, there were approximately 14 million new cancer cases, and cancer
accounted for 8.23 million deaths [4]. Worldwide in 2012, the five most common
cancers in women were breast, colorectal, lung, cervical, and stomach. For men,
they were lung, prostate, colorectal, stomach and liver cancer. In the year of 2015,
cancer accounted for 589,430 deaths in US, and 78% of those diagnosed cancer
are in people 55 years or older. In the next two decades, the number of new cancer
cases worldwide is expected to increase by more than 70% [4]. Many factors can
increase cancer risk, such as genetic background, viral infections, obesity, low
fruit and vegetable intake, use of tobacco and alcohol and others [1, 4].
2
1.1.2 Overview of Breast Cancer
Most of following information is from the SEER Cancer Statistics Review
and the American Cancer Society [5, 6].
Breast cancer is a type of malignant tumor initiated in breast. A normal
breast is made of lobules to secret breast milk, ducts that carry milk from lobules
to the nipple, connective tissue, blood vessels and lymph nodes (Figure 1.1). The
lymph system of the breast is the primary way that tumor cells spread outside the
breast. It includes lymph nodes and vessels that carry lymph. When breast cancer
cells spread into the lymph system, cancer cells will reside in the local lymph
nodes or later spread to other organs through metastasis.
Figure 1. 1 Biology of Breast Tissue
Figure adapted from American Cancer Society with permission [6].
3
There are several major types of breast cancer. When abnormal cells
originate in the ducts, the cancers are ductal. According to whether or not
abnormal cells invade the walls of ducts and spread to nearby lymph nodes and
other breast tissues, ductal carcinoma is either ductal carcinoma in situ (DCIS) or
invasive ductal carcinoma (IDC). IDC is the most common type of breast cancer.
Different from DCIS and IDC, invasive lobular carcinoma (ILC) start in the cells
lining the lobules and become invasive. Inflammatory breast cancer (IBC) is a
rare type of invasive breast cancer. Patients with IBC typically have breast skin
that is red, thick and pitted in texture. IBC patients may also feel their breast
become bigger, hard or itchy. IBC does not present with abnormal lump and is
likely to be missed by a mammogram.
Cancer stages describe the extent of a cancer in a patient. In general,
cancer stages can be divided into localized when cancer cells are only found in
breast; regional when cancer cells spread into regional lymph nodes; or metastatic
when the cancer cells spread to distant organs.
Breast cancer is the most common cancer and the second leading cause of
death among women in the United States. In 2012, nearly three million women
were living with breast cancer in the US. It is estimated that, in 2015, there will be
231,840 new cases of invasive breast cancer (14% of all new cancer cases) and
40,290 breast cancer deaths (6.8% of all cancer deaths) in the United States. 12.3%
of women (one in eight) will be diagnosed with breast cancer during their lifetime.
The five-year relative survival rate is used to calculate the proportion of patients
expected to be alive five years after diagnosis compared with a general population
4
of the same age, race and sex that haven’t been diagnosed with cancer. Based on
data from the Surveillance, Epidemiology and End Results Program (SEER)
2005- 2011, breast cancer has a five-year relative survival rate of 89.4%. If we
consider different stages of breast cancer, localized breast cancer account for 61%
of new breast cancer cases, while regional and distant account for 32% and 6% of
new cases, respectively.
According to American Cancer Society data, in the year of 2015, 61% of
breast cancer cases are detected at a localized stage, with a five-year relative
survival rate of 99% [1]. However, when tumors spread to nearby lymph nodes or
other tissues, the survival rate decreases to 85% [1]. If the breast cancer is in a
more advanced stage, where tumors are found in lymph nodes around the
collarbone or in distant organs, the survival rate falls to 25% [1]. Cancer can
reoccur well after the 5 year time period. Women who have had breast cancer live
in the fear of recurrence all the time.
1.1.3 Overview of Transgenic Mouse Model
The mouse has played an important role in studying the basic biology of
breast cancer. With more understanding of the molecular basis of breast cancer
development, different oncogenes are engineered into mice to manifest
overexpression of a specific oncogene or depletion of a specific tumor suppressor
gene in mouse germ line cells. Over 100 different genetically engineered—also
called transgenic—mouse models have been constructed and studied, including,
HER2, p53, c-myc, TGFa, Cyclin D1 and many others [7-10]. The use of
transgenic mouse models allows researchers to study breast cancer samples that
5
are difficult to access in clinical trials such as the Women's Health Initiative (WHI)
study and the Prostate, Lung, Colorectal and Ovarian (PLCO) cancer-screening
trial. Multiple time points of samples that cover different stages of tumor
development can be collected in a well-controlled manner. This creates a huge
advantage in the study of early-stage breast cancer, as human early-stage cancer
samples are scarce.
In order to correctly represent breast cancer development in humans, the
mouse models must have enough similarity with human breast cancer
development—in both genetic and protein-based comparisons—so that the
discoveries in mice may be applied to human breast cancer diagnostics.
Herschkowitz and colleagues have shown that transgenic breast cancer mouse
models have significant genetic similarities to human breast cancer. They used six
models with different transgenes to represent different human subtypes of breast
cancer. Many of the defining characteristics of human breast cancer can be seen in
the mouse models [9]. In the area of immune response, Lu and colleagues used a
transgenic mouse model with overexpression of the neu oncogene that developed
a serum autoantibody repertoire similar to that in cancer patients [11].
In this study, I use the MMTV-neuN mouse model, which is the most
extensively studied breast cancer model in prevention studies [8]. The neuN gene
is the mouse homolog of the human Erbb2 gene, which is overexpressed in about
20% - 30% of human breast cancers [12]. The neuN gene is overexpressed in
mouse mammary glands under the transcriptional control of the mouse mammary
tumor virus promoter long terminal repeat (MMTV-LTR) [13]. This mouse model
6
has been used to test different cancer vaccines to prevent tumor development [14-
18]; to find biomarkers for breast cancer [9, 11, 19]; to test different drug
treatments [20-23]; and to study basic mechanisms of breast cancer [9, 13, 24-27].
The MMTV-neuN mouse model shares a great similarity with human
breast cancer. However, it also has its limits. The cellular morphology is different
from that of human breast cancer. This mouse model can represent human ductal
carcinoma in situ (DCIS) but not lobular intraepithelial neoplasia. At late stages
of tumor development, bone and brain metastases are missing from this model
[12]. Overall, transgenic mouse models serve as a relevant, rapid and inexpensive
system to investigate breast cancer in humans [8, 10].
1.2 Current Breast Cancer Diagnosis and Biomarkers
1.2.1 Statistics Preparation for Diagnosis and Biomarker
In order to measure the performance of a diagnostic test or a biomarker,
sensitivity, specificity and accuracy are often used. In a diagnostic test of a
disease, a positive call means that the test score of a sample is positive for the
disease. A negative call means that the sample’s test score is negative for the
disease. In a test with an overall population of (A+B+C+D), A is the number of
positive samples that actually have the disease. B means non-diseased samples
that score positive. C means diseased samples that score negative. D is the
number of negative samples that do not have the disease.
Sensitivity is the measure of the proportion of positive samples in overall
samples actually with the disease -A/(A+C). Specificity describes the proportion
of negative samples among the samples without the disease –D(B+D). The false
7
positive rate, which equals 1-specificity, is more frequently used. Accuracy is the
proportion of positive and negative samples that are correctly predicted to the
actually disease condition in the overall population- (A+D)/(A+B+C+D).
1.2.2 Current Clinical Diagnostic Methods of Breast Cancer
The American Cancer Society provides screening guidelines for “early
detection of cancer in average-risk asymptomatic people” for 2015[1]. Breast self-
examination (BSE), clinical breast examination (CBE) and mammography are
recommended for women over age 20 (Figure 1.2). Ultrasound (US) and magnetic
resonance imaging (MRI) are also additional diagnostic tests widely used in
clinical settings [28].
Figure 1. 2: Illustration of Mammogram and MRI
Figure adapted from American Cancer Society with permission [6].
1.2.2.1 Breast Self-examination (BSE)
Adult women at the age of 20 should begin to do BSE at least once a
month and report any new breast symptoms to healthcare professionals.
Symptoms include changes in how the breast feels, its appearance, and any
8
discharge from nipple. Specific symptoms include lumps, nipple tenderness,
thickening of the breast, skin texture changes and others. BSE can be done
routinely and allows patients to monitor their breasts regularly. Although BSE
serves as a good tool to self-monitor for any abnormal breast symptoms, when
such symptom occurs, a clinical breast examination should be performed [1, 29].
1.2.2.2 Clinical Breast Examination (CBE)
A clinical breast examination is a clinical exam performed by a healthcare
professional that is trained to diagnosis abnormal breast symptoms. The goal of
CBE is to check any lumps or physical changes in order to achieve early diagnosis
of breast cancer. During the exam, a healthcare professional will check the
patient’s breast to discover any unusual texture, lumps, and suspicious areas. If a
lump is hard and cannot move easily, further tests are needed.
For women in their 20s and 30s, CBE is recommended to be a part of
patient’s periodic health examination at least once every 3 years. For
asymptomatic women over 40, CBE is recommended annually. When
mammography and other sophisticated tests are not available, CBE is an easy way
to perform breast cancer screening [1, 29].
1.2.2.3 Mammography
A mammogram is an x-ray image of the breast used to screen for breast
cancer. During a mammogram, a patient’s breasts are compressed between two
firm surfaces to spread out the breast tissue. The breasts then are exposed to small
dose of iodizing radiation to produce an image of the breast tissue. A trained
healthcare professional will examine the image to look for any sign of cancer.
9
Mammography is recommended annually for women over 40. Women with a high
risk of breast cancer should consult with professionals on the advisability of
mammography before age 40. The advantage of mammography is that it can often
detect a breast lump before the patient can feel the lump. If the mammogram
reveals an abnormal area in the breast, additional tests, such as ultrasound or MRI,
should be offered, and a biopsy should be performed when test results show that
the mass is solid[1, 29]. A mammogram is much more expensive than CBE, and,
thus, it not feasible as a routine test for all women.
The estimated sensitivity of mammography ranges from 29-97% with a
mean of 77%, and the false-positive rate ranges from 1-29% with a mean of 10%
[30]. Several clinical trials showed that mammography screening produced a
projected 15-20% reduction in breast-cancer-related mortality [29]. Although
mammography has played an important role in breast cancer diagnosis and has
helped save lives, it has many critics. Mammography has limited sensitivity and a
high false-positive rate [31, 32]. It has been shown to detect only 70% of breast
cancers [33]. 54% of all patients that were diagnosed to have breast cancer by
screening mammograms, up to 54% are results of overdiagnosis [34]. In addition,
mammography tends to have low sensitivity in women with dense breast tissue
[35, 36]. Women with aggressive breast cancer, such as triple negative breast
cancer [37], and those younger than 50 years old [38] also benefit less [39].
Around 80% of the false positives in mammography are caused by benign
growths [40, 41]. Benign growths can be classified as proliferative, hyperplasia
and hyperplasia/atypical growths [42, 43]. All three classes are associated with
10
different breast cancer risks compared to the risk for normal women.
Hyperplasia/atypical growths account for a 4.5-fold increase in the risk for breast
cancer [44]. However, mammography does not perform well to differentiate the
three classes. Absolute mortality benefit for women screened annually for 10
years is around 1% on average [45].
Another concern about mammography is that the value of mammograms
is associated with the experience of physician who interprets them [30].
Physicians who interpret 2500-4000 mammograms annually with a high screening
focus have a 50% lower false-positive rate than physicians who interpret 480-750
mammograms annually with a low screening focus[30]. This association between
the physician’s experience and the performance of mammography may bias breast
cancer diagnosis.
1.2.2.4 Adjunctive Testing: Ultrasound Imaging and Magnetic Resonance
Imaging (MRI)
Ultrasound imaging uses high-frequency sound waves to view the breast
tissue and can capture images in real time. During an ultrasound exam, a
transducer is placed on the skin with a thin layer of gel in between to transmit
ultrasound waves from the transducer into the patient’s body. MRI uses magnets
and radio waves to produce images of the breast. MRI is the most sensitive
imaging method for breast cancer detection [46].
Ultrasound and MRI are adjunctive diagnostic tests used when abnormal
areas have been found in the breast. Although current evidence on adjunctive
testing is limited, it has proved beneficial in identifying breast cancer in patients
11
who have dense breasts, as indicated by mammograms [29]. Most of the cancers
detected by adjunctive tests are invasive ductal cancer (IDC) rather than ductal
carcinoma in situ (DCIS) [47]. However, due to the high cost of MRI, it typically
is used as adjunctive testing when something suspicious has been found, and not
for screening of the general population. Also, it is important to note that
adjunctive tests can increase the false-positive rate [29].
1.2.2.5 Molecular Diagnosis Platform
Polymerase chain reaction (PCR) and DNA microarray are often used to
detect specific mRNA or DNA sequences of breast cancer. The advantage of PCR
is that it can perform low-noise amplification even when the signal is low-level.
DNA microarray technology can simultaneously detect multiple sequences. The
FDA has approved Oncotype DX, HOXB13-IL17BR assays and the MammaPrint
assay to measure gene expression of breast cancer [48].
Oncotype DX detects a 21-gene profile by reverse-transcriptase PCR to
predict the risk of recurrence in patients taking Tamoxifen. Sixteen of the 21
genes are cancer-related genes with reference genes. A mathematical algorithm is
derived from empirical retrospective study to calculate a score to estimate the risk
of distant recurrence [49].
MammaPrint is a gene expression assay by Agendia that detects a 70-gene
profile by reverse-transcriptase PCR. It used 117 patients with axillary lymph
node-negative primary breast cancer to find genes highly correlated with a short
interval from primary tumor to distant metastases. MammaPrint profiling is used
to identify a sub-population of patients by their prognosis performance [50].
12
HOXB13-IL17BR assays use complementary DNA microarray
technology to detect the expression level of HOXB13/IL17BR genes. It is used to
predict the outcome of Tamoxifen in untreated ER-positive/node-negative patients
[51].
Current breast cancer diagnostic methods suffer from low sensitivity and
high false-positive rates. Breast tumors with high breast density and aggressive
growth are not detected during current diagnostic procedures [52, 53]. Moreover,
CBE, mammography and adjunctive tests are not effective in the early detection
of breast cancer.
1.2.3 Benefit of Early Detection and Treatment
Early detection of cancer identifies tumors before they are diagnosed can
bring more opportunities for medical intervention. With early detection, it seems
reasonable to expect that systematic therapy involving smaller amounts of less-
toxic medications can be initiated earlier for a shorter period of time and be more
effective. This could potentially revolutionize current treatment strategies and
improve patients’ outcomes—and, perhaps, even eradicate the disease. Take
ovarian cancer for example. Patients with stage I ovarian cancer, a 42-mm tumor
diameter, on average, have a five-year survival rate of 92%. The five-year
survival rate decreases to around 30% when patients are in stage III to stage IV
[54]. Thus, researchers and medical professionals have spent considerable effort
in seeking a noninvasive, simple and inexpensive means for the early detection of
cancer.
13
1.2.3.1 Early Diagnosis of Breast Cancer Can Improve Survival Rate
In cancer treatment, patients with localized cancer have better survival
rates than those with metastatic disease (Figure 1.3) [55]. Failure to detect local
cancer at an early stage is the major problem in cancer diagnosis and treatment.
In breast cancer, in the localized stage, tumors have not spread to lymph
nodes, nearby structures outside of the breast. According to American Cancer
Society projected data for 2015, 61% of breast cancer cases are detected at a
localized stage with a five-year relative survival rate of 99% [1]. However, when
tumors spread to nearby lymph nodes or other tissues, the survival rate decreases
to 85% [1]. If the breast cancer is in a late stage, in which tumors are found in
lymph nodes around the collarbone or more distant, the survival rate falls to 25%
[1]. Therefore, an early detection system could eliminate the mortality caused by
late-stage metastasis and largely improve the five-year survival rate for patients.
We currently lack a screening strategy to make a noninvasive, timely diagnosis.
Figure 1. 3: Five-year Relative Survival Rates (%) by Stage at Diagnosis in
US from 2004-2010
Figure adapted from American Cancer Society with permission [1].
14
1.2.3.2 Proposed Early Diagnosis and Treatment Scenario
In order to be successful, an early diagnosis system should have the
following merits. First, it should have high sensitivity and a low false-positive rate
for patients in the early stage of breast cancer. One of the main challenges with
mammography is its high false-positive rate. The proposed early detection system
is a screening system for the general population or the sub-population of patients
that have found an abnormality during self-exam. Second, the cost of diagnosis by
this system should be low so patients can be examined routinely. Although MRI
has the highest sensitivity to detect small tumors, its high cost prevents patients
from having MRIs frequently. The cost of an early detection system should be
less than the cost of clinical breast examination. Another advantage of routine
examinations is that they can increase prediction accuracy [56]. For example, a
test with 95% accuracy per test will be have 99.75% accuracy after two tests and
99.99% after three tests. Routine testing allows establishing a baseline which
should also decrease false positives. In order to allow for routine examinations,
the system needs to use non-invasive techniques to screen asymptomatic
populations. Blood, saliva or other fluids could be good candidates to be used in
the test.
I propose a scenario for early detection of breast cancer through routine
examinations. Patients will perform breast self-examination to find any lump and
use an early detection system to monitor any aberration. If the system was simple
and inexpensive, patients could get the diagnosis result well before lump was
detected. Further testing by mammography, MRI, ultrasound and others will be
15
performed when the test shows an increased risk of breast cancer. Early surgery
and therapy will be used to kill the cancer using a low-dose, non-toxic treatment.
Some people may argue that the early diagnosis of breast cancer will
increase the financial burden on patients and society. Their logic assumes that if a
test has 99% specificity, and over 100 million US women are over age 20[57],
annual detection will lead to one million false-positive cases. If the cost of a
follow-up mammogram for every false positive is around $100, this will lead to
100 million dollars in unnecessary costs per year. However, this description
totally ignores the increased sensitivity of the early diagnosis system. Also, new
therapies may appear with the early detection system. The regimen of drugs
targeting early-stage cancer could be different from the current ones targeting
later-stage cancer. Because the target for early diagnosis is the general population,
the market for drugs to treat early breast cancers will increase dramatically and,
thus, reduce the costs of drug development and manufacture. This has already
been seen in many drugs; a prime example is aspirin, first marketed by Bayer in
1932 and, today, a very low-cost widely-used drug.
1.3 Early Detection of Breast Cancer by Antibodies
1.3.1 Overview of Blood Based Breast Cancer Biomarkers
Current molecular approaches for detecting breast cancer can fall into two
categories: 1. direct detection of the tumor or molecular tumor cells shed; 2.
detection of the patient’s immune response to the tumor.
16
Biomarkers that detect the tumor or the cells it sheds include proteins,
miRNAs, circulating tumor cells (CTC), aberrant glycosylation and others.
Biomarkers that detect the immune response of patients include antibodies that
will be discussed in the following section.
CA 15-3 and CA27.29 are biomarkers to detect the circulating MUC-1
antigen in the blood. They are different epitopes on the same MUC-1 antigen. A
high CA 15.3 level is correlated with a large tumor size and presence of lymph
node metastases [58]. These biomarkers have been shown to have prognostic
value in early-stage breast cancer [58]. According to the American Society of
Clinical Oncology guidelines, CA 15-3 and CA27.29 are not recommend for
screening, diagnosis and staging of primary and recurrent breast cancer because
their role in early-stage breast cancer and treatment decisions is not clear [59].
HER2 is a member of the epidermal growth factor receptor (EGFR) family
70. 15% -30% of diagnosed breast cancers have an overexpression of HER2 [60].
It is recommended that HER2 expression be evaluated in primary invasive breast
cancer from the time of diagnosis to recurrence [60],[61]. It can help to guide the
use of trastuzumab in the adjuvant setting [62].
Circulating tumor cells (CTC) are tumor cells present in the blood that
represent a specific cancer type. CTC may predict the presence of micrometastasis
or aggressive primary breast cancer [63]. However, the low concentration of
CTCs (1 in 106~107 leukocytes) [63, 64] makes it difficult to be captured by
immunomagentic beads or to use RT-PCR to amplify signals [65-67]. The CTC
assay is not recommended in the guidelines for diagnosing breast cancer.
17
Changes in circulating microRNAs (miRNAs) have been found to be
significantly related to early and minimally invasive breast cancer. A panel of 31
candidate miRNAs has shown significant differences between 20 women with
early-stage breast cancer and 20 control patients [68]. Another study, using a
panel of seven miRNA candidates in 148 patients with breast cancer and 44
controls, showed significantly altered miRNA expression levels [69]. Tumor-
specific miRNAs are tissue-specific, dysregulated in cancer, highly abundant and
stable in the blood [70, 71]. miRNAs could serve as a good candidate for non-
invasive biomarkers.
Aberrant glycans and glycoforms of proteins could also be a source of
breast cancer biomarkers to monitor disease progression. Abd Hamid and
colleagues discovered that a trisialylated triantennary glycan containing alpha1, 3-
linked fucose with a twofold increase in breast cancer patients compared with
controls [72].
Although early diagnosis of breast cancer has many merits, only limited
biomarkers have been discovered and widely used in the clinical setting. Methods
to identify novel breast cancer biomarkers are still urgently needed. Each
approach has its strengths and weaknesses. No single technology can provide the
ideal detection performance with high sensitivity and specificity.
1.3.2 Challenges in Finding Early Breast Cancer Biomarker
In order to be useful in early detection of breast cancer, a biomarker
should have high sensitivity and a low false-positive rate, and it should show
evidence of value in determining treatment after the early detection. The
18
biomarker should also be stable and robust and available in a non-invasive test.
However, with thousands of candidate biomarkers identified, only a few are
approved each year and widely used in the clinic [29, 59]. Clearly, searches for
biomarkers for early detection have not yet been generally successful [73-75].
Researchers in early detection of breast cancer face several challenges, including
over-fitting of the algorithm by small sample size, unstandardized sample
preparation and specimen annotation [76, 77].
In early detection studies, one problem is that most biomarkers are
discovered from diagnosed cancers and then are used in an attempt to diagnose
early cancer. From the biology of cancer, tumors change dynamically at different
stages. Samples at the time of diagnosis are most likely to be irrelevant for early-
stage samples. Antigens in early high-risk lesions may have a different expression
panel from later stages.
The reason that many researchers still use samples with diagnosed tumors
is that samples from the same patients before and after tumor diagnosis are hard to
get. It is difficult to study the correlation of a biomarker with tumor progression
and early-stage breast cancer without the proper early-stage samples.
Many clinical studies also have suggested that using biomarkers
discovered and widely used in diagnosing cancer perform poorly in pre-diagnostic
patients [78-81]. Zhu and colleagues studied 118 patients one to two years before
ovarian cancer diagnosis [79]. Among 28 evaluated biomarkers, CA125 achieved
the highest score, with a sensitivity of 69.2% and a specificity of 96.6% in
diagnosed samples but failed in pre-diagnostic samples. Other models had poorer
19
performance than CA125 alone. Zhu et al. concluded that “biomarker panels
discovered in diagnostic samples may not validate in pre-diagnostic samples.” In
the Carotene and Retinol Efficacy Trial, Anderson and colleagues showed they
could achieve only limited discriminatory power in pre-diagnostic ovarian
specimens and that the power increased as the time of diagnosis approached [78].
In breast cancer, Lu and colleagues used a sample 150 days before diagnosis, and
HER2 achieved a ROC score of only 0.63 and p53 with 0.63 [81]. In the Prospect-
EPIC study, Opstal-van Winden and colleagues used samples from 68 women
with a median of 21.3 months before breast cancer diagnosis and 68 controls.
After evaluation of ten breast cancer biomarkers, none of those resulted in correct
classification [80]. In these studies, researchers concluded that markers
discovered in diagnostic samples may not be validated in pre-diagnostic samples.
1.3.3 Cancer Antibodies as Biomarkers
1.3.3.1 Cancer Patients can Develop Autoantibodies Against Tumor Early
The immune system in our bodies protects us from pathogens and
abnormal conditions. It continuously monitors any for foreign invader. B cells
from the humoral immune system and T cells from the cellular immune system
play important roles in finding non-self-antigens. Antibodies produced by B cells
bind to specific antigens and neutralize the foreign invader. After binding, B cells
are activated and form a memory response to encounter future events from the
same antigen.
T cells are also triggered and help B cells to carry out cytotoxic killing of
infected cells. In order to correctly recognize foreign antigens instead of self-
20
molecules, B cells and T cells are negatively selected to remove any B cells and T
cells that can bind to self-proteins.
Cancer cells start as normal cells and enter into atypical growth later. The
abnormal changes lead to changes in the abundance of protein expression,
different post-translational modification patterns, and other changes. Burnet and
Thomas developed an immune surveillance hypothesis in 1971 after Ehrlich
proposed that the immune system can recognize and destroy nascent transformed
cells in the human body [82]. The formal immune surveillance describes that
thymus-dependent immune cells can produce an effective immune response to
tumor antigens and protect the host from nascent transformed cells [83-85]. Later,
this theory was refined by an immune-editing concept that shows that there is a
dynamic interaction between the immune system and the tumor. In the refined
theory, immune surveillance inhibits tumor development from tumor initiation,
and the tumor evolves through multiple mechanisms to avoid immune system
surveillance [86, 87].
A classic example that reflects the role of antibodies early in cancer can be
found in the paraneoplastic syndrome (PNS) [88]. PNS is a syndrome that results
from a cancer causing an autoimmune effect on the nervous system. When a
tumor and nervous tissue share common an antigen, then remote pathologic
effects will influence the nervous system. Sometimes the symptoms of PNS are
detected months to years before the formation of a malignant tumor because
tumor cells express neuronal antigens and trigger an anti-tumor immune response
[89]. This is further supported by Marcia Wilkinson and colleagues, who
21
discovered a neuronal antibody in the serum of patients with small-cell lung
cancer [88, 90]. After removal of the cancers, PNS is usually cured. These
findings reveal that tumors elicit an anti-tumor antibody response at early stages
[88, 91].
More research has shown that autoantibodies can be found several months
to years before the appearance of symptoms of lung cancer [92], esophageal
squamous cell carcinoma [92], prostate cancer [93], breast cancer [94], and others
[91, 95, 96].
B cells have been detected in invasive breast cancer [97], a phenomenon
called immune infiltration. This shows that the immune system closely monitors
cancer cells. In 2005, p53 antibodies were detected in 12 of 49 individuals who
developed cancer later, compared with four out of 54 individuals who did not [98].
Autoantibodies against HER2 have also been detected in early-stage breast cancer
patients [99]. Evidence of autoantibodies to tumor-associated antigens (TAAs)
precedes manifestations of diagnosed breast cancer, making autoantibodies a
source of biomarkers to detect breast cancer early.
1.3.3.2 Advantage of Using Antibodies as Biomarker
Although the discovery of biomarkers for early diagnosis of breast cancer
holds great promise, the low concentrations of biomarkers in blood impede the
development of effective tests. Current plasma protein biomarkers in the human
proteome organization (HUPO) project are in the range of ug/ml to mg/ml (Figure
1.4). When using proteins and miRNA secreted from tumors or circulating cells as
biomarkers to detect early-stage cancer, they are 10-fold4 too low to be detected
22
by state-of-the-art technology [100-102]. At the time of diagnosis by current
biomarker technology, presumably a tumor would be larger than 1mm3 in size,
with more than three million cells, and would have been developing in body for
several years. It is not feasible to detect such a low signal without any
amplification technology to extract and amplify the tumor-specific signal. Tumor-
specific antibodies are the perfect biomarker candidate to amplify tumor-specific
signals [103].
Figure 1. 4: Distribution of Human Plasma Protein.
Figure adapted from Anderson et al. with permission [101].
23
After activation, a B cell can produce 300,000-1,200,000 antibodies per
hour [104, 105] and replicates every 70 hours [106]. It has a lifespan of up to 4.5
months [107, 108]. B cells towards a tumor antigen can amplify the production of
tumor-specific antibodies up to ~10-fold11 in one week [109, 110]. Memory B
cells that generated those antibodies can exist years after the immunogenic event
occurs.
Antibodies are also stable and resistant to common types of proteolysis.
Other proteins and small molecules will be either rapidly degraded or cleared
from the blood. Antibodies have a half-life of over seven days in the blood. Even
after separation from the blood, antibodies are stable for several years [111, 112].
This means that we can store achieved blood samples in sera or a standard filter
spot [113]. This allows researchers to use stored, historical samples to find
antibody biomarkers [114].
Antibodies are also easy to detect. Each subtype of human
immunoglobulin shares the same Fc structure on heavy chains. Commercial
antibodies specifically binding to each subtype of human immunoglobulin are
widely available with high specificity and affinity. This means that researchers
can use the same secondary antibody to detect different tumor-specific antibodies
(primary antibody), thus largely reducing the complexity of developing a
biomarker.
24
1.3.3.3 Breast Cancer Associated Antibodies
Much effort has gone into finding antibody biomarkers to diagnosis breast
cancer. ELISA, protein array and other technologies have been used to study
single biomarkers and a panel of multiple biomarkers.
Early in 1997, Disis and colleagues found the presence of the HER2
antibody in 12 of 107 breast cancer patients and none among the 200 normal
controls. They showed that nine of 44 patients with HER2- positive tumors had
antibodies against HER2 [115]. Autoantibodies against HER2 have been detected
in early-stage breast cancer patients [99]. Later, in 2005, p53 antibodies were
detected in 12 of 49 individuals who developed cancer later, compared with four
out of 54 individuals without cancer [98]. This study, for the first time, showed
the relationship between p53 autoantibodies and the subsequent development of
malignancy. A panel of multiple antibody biomarkers was investigated to increase
prediction performance. Chapman and colleagues reported a panel of six antigens
(p53, c-myc, HER2, NY-ESO-1, BRCA2 and MUC1) to distinguish patients’
samples of primary breast cancer from normal and ductal carcinoma in situ.
Antibody response was observed in at least one of six antigens in 64% of primary
breast cancer samples and 45% of DCIS samples with 85% specificity [116].
Anderson and colleagues used a protein microarray to discover 28 tumor antigens
that can distinguish invasive breast cancer (stages 1-3) from benign breast disease
[95]. Desmetz and colleagues focused on autoantibody biomarkers in in-situ
carcinoma (CIS) in younger women under age 50) with breast cancer. They used a
panel of antibodies against PPIA, PRDX2, FKBP52, HSP60 and MUC1 to study
25
60 primary breast cancer patients, 82 CIS patients and 93 healthy controls. The
panel of biomarkers significantly distinguished primary breast cancer with AUC
of 0.73 and CIS with AUC of 0.80 from healthy controls [117]. Five biomarkers
(GAL3, PAK2, PHB2, RACK1 and RUVBL1) were investigated in DCIS and
node-negative early-stage breast cancer samples. The five markers significantly
distinguished healthy controls from early-stage cancer with AUC of 0.81 and
DCIS with AUC of 0.85 [118].
1.3.4 Emerging Technologies for Profiling Antibody Responses
Humoral response can provide us with important information about a
person’s disease progression and overall health status. In order to get information
about humoral response, we need high-throughput methods to profile an antibody
repertoire in a patient at a given time. In order to do that, we need a large library
of binding ligands to represent possible antigens. The ligand could be the real
antigen, a peptide sequence representing the epitope, or a mimotope sharing the
same binding with a disease- specific antibody. Several technologies have been
proposed in the past and are summarized below [119, 120].
1.3.4.1 SEREX
The serological screening of recombinant cDNA library using phage
display (SEREX) is as follows: lambda phages are used to display proteins from a
recombinant cDNA library of interests and probed with sera of interest to select
antigens with high binding interactions with antibodies in the sera. Michael
Pfreundschuh and colleagues [121] first introduced SEREX to isolate tumor
antigens that trigger a high-titer humoral response in cancer patients. More than
26
2700 immunogenic antigens have been identified by SEREX [120, 122, 123]. The
methodology of SEREX includes three main steps: building a cDNA library;
probing with sera; and iteratively selecting sub-clones of phages.
Building the cDNA library: The cDNA library is generated by reverse
transcription of RNA from tissues or cell lines of interest. After the reverse
transcription, all cDNA is inserted into constructed lambda phage vectors. After
plating phages and transfer onto a nitrocellulose membrane, proteins will be
expressed in an E. coli expression system.
Probing with sera: Phage clones will be probed with diluted serum
samples from cancer patients and proper controls. High-binding clones will be
detected by an antibody-coupled enzymatic reaction and selected to be sequenced.
Normally, candidates will be validated by independent methods such as ELISA or
microarray.
SEREX has been used to identify tumor-associated antigens (TAA) NY-
BR-1 to NY-BR-7 [124] and p33ING1[125] in breast cancer; NY-CO-37 and NY-
CO-38 in colon cancer [126]; NY-ESO-1 and SSX2 in esophageal cancer and
melanoma [127, 128]; and NY-ESO-1, LAGE-1, and XAGE-1[129] in prostate
cancer.
Although SEREX has been used to discover thousands of tumor-
associated antigens, it has internal limitations. First, a large serum volume is
required to screen for specific phage clones due to the extensive screening.
However, many historical sample databases have only limited aliquots of each
patient’s sample. Second, SEREX is biased to high-expression mRNA—a limit
27
that is internally inherent from its beginning. As we recall from its definition,
SEREX aims to select antigens with high-binding interactions with antibodies in
the sera. Although the cDNA library could have as many as 106 copies of phages,
the constructed cDNA library will overrepresent the reverse transcriptions of
those mRNA with a high expression level. When cancer-associated antigens are
expressed at a lower level, those that are important indicators in early-stage will
be substantially diluted and not even be represented during the screening. Third,
because SEREX uses prokaryotic expression library, the structure of the antigens
may not be correct—e.g., unnatural folding and post-translational modification.
This will lead SEREX to miss some TAA and to an increased false-positive rate.
Finally, the decrease in amount of an antibody will not be picked up by this
system, which could also be as biomarker for diagnosis.
1.3.4.2 Peptide and Peptoid Microarrays
Peptide microarrays are comparable to SEREX in that they both use a
large library of short peptide sequences. However, peptide microarrays overcome
the bias of high-expression mRNA and the large volume of serum required in
SEREX. The core of peptide microarrays is a high-throughput way to display a
large library of peptides with different sequences. Peptides can be chemically
synthesized in large quantities in advance, and pure products are then linked to
microarrays. This gives the peptide library a long shelf life. Sigma-
Genosys provides a synthesis platform to allow the rapid parallel synthesis of
custom libraries. Peptides can also be directly generated in situ[130]. More than
330,000 peptides can be generated on silicon wafers per assay, using
28
semiconductor manufacturing technology. Diluted serum samples will be probed
on peptide microarrays, and the signal will be collected by labeled secondary
antibodies with dye.
Peptide microarrays have been used to study cell interaction[131], binding
sequences of antibodies [132], enzymes [133, 134], proteins [135], and DNA and
small molecules [136-138]. One important focus of peptide microarrays is on
epitope mapping. Tiled peptide sequences with partial overlaps reveal the epitope
of a monoclonal antibody [139]. The immunosignature section of this paper will
include a more-detailed analysis of peptide microarrays.
Other non-natural molecules can also be used for antibody profiling. One
example is peptoid, a small molecule of N-substituted oligoglycines [140]. A
peptoid microarray is constructed to display thousands of peptoids reproducibly
on a chemically functionalized glass surface [141]. Diluted serum samples are
probed on peptoid microarrays, and the signal will be collected by a labeled
secondary antibody with dye. Muralidhar Reddy and his colleagues have shown
the application of peptoid microarrays in their discovery of two candidate IgG
biomarkers for Alzheimer’s disease [140].
1.3.4.3 SERPA
Serological proteomic analysis (SERPA) uses a different approach to
overcome problems in peptides or protein synthesis. SERPA takes advantage of
two-dimensional polyacrylamide gel electrophoresis (2D-PAGE) and mass
spectrometry analyses. Instead of using chemically synthesized peptides or
expressed proteins to mimic antigens in the patients’ samples, SERPA directly
29
uses cell lines or tissue lysates. It first separates proteins in the samples by their
isoelectric point and then further separates them by molecular mass in 2D-PAGE.
The separated proteins are transferred onto membranes and then are probed with
patient or control serum samples. The corresponding blot will be extracted from
the 2D-PAGE and identified by mass spectrometry. SERPA overcomes some of
the major limitations of other platforms, such as how to ensure a correct protein
structure. But unlike other methods using microarray, it lacks the ability to
perform high-throughput screening[142] and also requires a large number of
serum samples [143].
1.3.4.4 Protein microarray
The definition of a protein microarray is the technology that displays an
array of proteins on a slide surface in a reproducible and addressable manner. The
binding of each protein can be analyzed in a high-throughput way. Roger Ekins
first introduced the concept of the protein array in 1989 [144], and MacBeath et al.
developed the first mature protein array in 2000 [145]. Heng Zhu and his
colleagues also printed protein arrays with 5,800 yeast proteins on a slide in 2002
[146].
Protein microarray can be divided into three categories: analytical protein
array, reverse-phase protein array and functional protein array. Analytical protein
arrays use antibodies printed on the array surface to capture proteins labeled with
fluorescence in cell lysate [147]. Reverse-phase protein arrays, on the contrary,
print cell lysate on a glass surface and probe proteins in cell lysate with
fluorescence-labeled antibodies of interest. This platform requires high-quality
30
antibodies to achieve reliable results. Functional protein arrays aim to investigate
the biochemistry properties and interactions of proteins with their targets, such as
protein, DNA, RNA, drug, and enzyme. In a functional protein array, proteins are
synthesized by a protein expression system and then spotted on the array surface
or directly synthesized on the array surface. Cell-based expression systems
include bacteria, insect cells or yeast, as well as the in-vitro expression system
[145, 146, 148-150]. A widely used human protein microarray is Protoarray®,
manufactured by Thermo Scientific [151]. In order to represent the correct protein
structure, Protoarray uses the Sf9 insect cell line, which is the close to the
mammalian expression system.
Protein microarrays have been used extensively to profile humoral
response to diseases and to screen for disease-specific autoantibody biomarkers
[152-157].
Although thousands of tumor-associated antigens have been discovered
using the protein microarray, it also has limitations. First, because proteins need
to have the correct structure and be purified to ensure high-quality proteins that
can be spotted on arrays, much effort is required to express the proteins correctly
and to isolate and purify proteins from the expression system mix [158]. Post-
translational modifications may not be represented. Second, after collecting pure
proteins, storage is a big issue. Proteins have only a limited shelf life and may
aggregate or even become denatured after a few weeks. This will influence any
functional study or protein-antibody bindings that require conformational epitopes
[159]. A cell-free in situ expression system, Nucleic Acid Programmable Protein
31
Array (NAPPA), has been proposed to address these two problems [148]. Proteins
will first be encoded by plasmids with a fusion tag and spotted on an array surface.
After being expressed through a cell-free expression system, proteins will be
captured by an affinity reagent in situ. However, some steps will still be required:
building a plasmid system; inserting each protein into plasmids; and printing each
plasmid to each spot on the microarray. Before probing with sera, proteins on a
NAPPA array need to be expressed by an in vitro expression system. In addition,
these systems are expensive and do not lead themselves to high-throughput
production so are primarily useful as discovery platforms.
1.4 Immunosignature
1.4.1 Definition of Immunosignature
The immunosignature is the binding pattern of a complex mixture of
antibodies. Immunosignature technology uses a peptide array with 10,000 spotted
peptides selected without bias from a non-biological peptide sequence space
(Figure 1.5). It can reflect a disease state or, more generally, the health status of
an individual. Immunosignature is a multidimensional reflection of overall
humoral immune response. It can serve as a universal diagnostic platform to
detect and identify any disease with a significant antibody response.
32
Figure 1. 5: Principle of Immunosignature
Left is the scan image of CIM 10K array. Right is the major steps in
immunosignature. Diluted Serum samples are probed on arrays. Biotinylated anti-
mouse IgG (H+L) antibodies are incubated on arrays next. Bound secondary
antibodies are visualized by Alexa Fluor-647 labeled streptavidin.
1.4.1.1 Building a Complex Chemical Surface
At the core of immunosignature technology is an array with 10,000
random-sequence peptides spotted in an addressable, machine-readable fashion
with 20-mer peptides.[160]
A random number generator generated the peptides sequences. We
designed the algorithm of the random number generator to evenly cover as many
combinations of amino acid sequences, potentially epitopes, as possible. These
peptide sequences were not based on protein sequences in nature. For the choice
of amino acids, we included all natural amino acids except cysteine to synthesize
10,000 peptides. Cysteine is used only on the C-terminus of each peptide to form
a chemical link to the activated surface in an oriented manner. The sequence of
peptides can be a true epitope of that antibody or a serve as a mimotope of the
actual epitopes due to the cross-reactive properties of antibodies [161].
33
Each peptide was spotted to an array surface in an addressable manner, so
the sequence of each spot is known. We spotted 10,000 peptides on an array to
cover a substantial portion of the molecular recognition space of all circulating
antibodies. The aim of this design is to provide chemical complexity on an array
to allow an unbiased display of antibody binding.
1.4.1.2 Probing Sera
To generate an immunosignature, a drop of blood or serum (2ul) is diluted
at 1:500 in buffer. At this dilution, essentially only the antibodies from the serum
bind the surface. The diluted serum is then applied to the 10,000 peptides array.
After applying the diluted serum on an array, we incubate the array for one
hour and then wash it to remove unbound antibodies. Antibodies that are bound to
peptides are detected by a secondary antibody and fluorescence dye. The array is
then washed, dried and scanned to determine the array’s antibody binding profile.
Different secondary antibodies can be used for specific isotypes of interest, which
could provide more layers of information about an individual’s immune status.
1.4.1.3 Analysis of Binding Pattern
We read this pattern with our peptide microarray, which captures enough
information about a patient that the health status is legible. Once the arrays are
probed with patient sera, the immunosignature of each disease state is determined
by selecting peptides that show common reactivity among patients with the
disease of interest and different reactivity in controls. Selected peptides are used
to train classification algorithms to evaluate diagnostic efficacy and classification
error.
34
1.4.2 Advantage of Immunosignature
As mentioned in the previous section, we can find the immunosignature
that is unique to each disease and is stable over time [160]. The main reason for
this advantage is that antibodies are stable in both circulating blood and stored
sera samples. B cells that generated those antibodies can exist years after the
immunogenic event has occurred, with a lifespan of up to four-and-a-half months
[107, 108]. Moreover, after sera separation, antibodies are stable for several years
in solutions [111, 112]. This means that we can use archived blood samples stored
in sera or a standard filter spot [113].
The random peptide array is universal. We designed the algorithm of the
random number generator to evenly cover as many combinations of amino acid
sequences, potentially epitopes, as possible. These peptide sequences were not
based on protein sequences in nature. The aim of this design is to provide
chemical complexity on an array to allow an unbiased display of antibody binding.
In this way, we can use same array to make diagnostic assays of other diseases or
any condition that can be reflected in a circulating antibody repertoire.
Immunosignature technology also does not require pre-knowledge of a
complex disease. Unlike PCR, ELISA and protein arrays that require isolation of
molecular targets to form an assay, we use random peptide sequences instead of
known protein or epitope information. We can investigate diseases even if we do
not known their antigens. The sequence of a peptide can be a true epitope of that
antibody or serve as a mimotope of the actual epitopes due to the cross-reactive
properties of antibodies [161]. We have shown that antibodies generated against
35
multiple types of targets, such as proteins and sugar, can bind peptides on the
array [162]. This gives immunosignature the advantage of dealing with an
outbreak of an unknown pathogen, and we can use same array for other diseases
in different species. Although the peptide sequences are randomly generated, we
can “decipher” diseases’ specific immunosignature based on sequencing analysis
to guide us find real epitopes at times [163, 164].
Immunosignature technology can be adapted to have multiple antibody
isotypes measured simultaneously on an array. This will provide orthogonal
measurements of a disease. Studies have shown that multiple isotypes can
increase diagnostic performance [19, 109, 110, 113, 160]. We have also routinely
detected both IgG and IgM on the same array in other studies.
Because the immunosignature measures a host’s humoral response to a
disease, we can evaluate how the host’s immune system responds to that disease.
Some patients may have a high titer of protective antibodies and may not need
any treatment later. We can use immunosignature technology to identify those
patients and devote limited medical resources to the highly vulnerable patients
who need treatment.
In addition, the arrays are inexpensive and can be used to assay large
numbers of samples quickly.
36
1.4.3 Published Studies with Immunosignature
Immunosignature has been demonstrated in multiple diseases (infectious
and chronic) or immunizations and in animal models (multiple mouse models),
dog and human samples. We have demonstrated that the resulting
immunosignature is unique for each disease that we tested and is stable over time
[110, 160, 165-173].
Take valley fever (Coccidioidomycosis) as an example of an infectious
disease. We have demonstrated that by using a training set from 55 infected
individuals and 55 uninfected individuals, we can achieve 100% prediction
accuracy in an independent testing set. The immunosignature can detect the
disease in samples from patients at a point in time when the standard ELISA test
shows no titer, but the patient later develops valley fever[174].
Also in chronic diseases such as cancers, immunosignature can detect the
cancer- specific signature for each type of cancer. We have used a training cohort
with five cancers (20 samples for each cancer) and 20 non-cancer samples to
generate reference immunosignatures to distinguish each disease. The
immunosignatures gave 95% classification accuracy in a blinded test with 120
blinded samples with the same disease structure [175].
In Alzheimer’s disease, Lucas Restrepo has developed immunosignature
to detect the disease in both mouse-model and human samples[169].
1.5 Breast Cancer Treatment
1.5.1 Clinical Treatment of Breast Cancer
37
Current clinical breast cancer treatment includes three main methods:
surgery, radiation, and drug treatment, including chemotherapy, hormone therapy,
and targeted therapy. More recently, great achievements have been made in
cancer immunotherapy, which may become an important part of future cancer
treatment. Adjuvant therapy is a way of providing patients with more treatment
after surgical removal of cancers. Neoadjuvant therapy is different in that a
systemic treatment or radiation is giving before surgery to decrease tumor size.
Healthcare professionals and medical researchers have investigated many
different strategies for cancer treatment, and a combination of different treatments
instead of a single one is the trend in the future.
1.5.1.1 Surgery and Radiotherapy
Removing the tumor through surgery remains the major way that breast
cancer is treated. Surgery for cancer has been performed since the early 19th
century and was later combined with anesthesia and antisepsis in the following
half century. Surgery remained the only option before the appearance of radiation
by the middle of the 20th century.
The purpose of surgery is to remove as much of the cancer from a patient
as possible. By using imaging technology, more-accurate surgery can be
performed to almost completely remove the cancer cells and preserve the healthy
tissues. Based on tumor size, extent of spread and the patient’s preference,
different surgeries can be performed.
Breast-conserving surgery (BCS): BCS, also called partial mastectomy,
removes only a part of the breast, depending on the size and place of the tumor
38
and others. The removed breast tissue will be examined to see whether the tumor
has spread to the edges. If it has, more surgery is needed to completely remove all
tumor tissues. BCS is often followed by radiation to kill any remaining tumor
cells.
Mastectomy is a surgery that removes the entire breast and, sometimes,
nearby tissue. If only the breast is removed, it is called a simple mastectomy.
When a simple mastectomy is combined with an axillary lymph node dissection,
it becomes a modified radical mastectomy. Axillary lymph nodes will be
examined for any tumor cells. A sentinel lymph node biopsy examines the first
lymph nodes to which cancer may spread. Studies have shown that for early non-
invasive breast cancers, long-term survival is similar between BSC with radiation
and mastectomy.
The discovery of X-ray in 1895 by Wilhelm Roentgen made cancer
radiation therapy possible. Radiation therapy uses high-energy rays such as x-rays
or particles to kill cancer cells and shrink tumor size in the human body. External
beam radiation uses a machine outside of the body to emit radiation.
Brachytherapy is the placement of radioactive pellets into the breast tissue to kill
targeted tumor tissues. Its long-term results may not be as good as those with
external beam radiation. With the development of better radiation machines and
computer imaging technology, radiotherapy developed significantly during the
20th century. This contributed to a 30% increase in the cancer cure rate in in the
1950s, when radiotherapy was combined with surgery [176].
39
1.5.1.2 Cancer Drugs
Besides surgery and radiotherapy, breast cancer treatment can also use
drugs to kill cancer, including chemotherapy, hormone therapy, and targeted
therapy. These drugs can be used to prevent cancer or treat diagnosed cancer.
1.5.1.2.1 Chemotherapy
Chemotherapy is the use of cancer-killing drugs. Drugs enter the human
body and spread to different organs by blood circulation. Chemotherapy can be
used before or after surgery for early-stage breast cancer. When chemotherapy is
performed before surgery, it can shrink tumor size so that only BCS is needed
instead of mastectomy. It can also decrease the risk of recurrence after surgery.
Chemotherapy is given in cycles, with each cycle of a treatment followed by a
rest period. For early-stage breast cancer, a normal course of treatment may last
for three to six months. However, if necessary, the chemotherapy can be
continued as long as it is effective.
After the use of chemotherapy in curing childhood leukemia and advanced
Hodgkin’s lymphoma in the 1960s [177, 178], more drugs were discovered to
treat major types of cancer. Chemotherapy has greatly improved patients’
outcome. However, chemotherapy has side effects, including menstrual changes,
nerve damage, heart damage, nausea, hair loss and others. Because chemotherapy
does not specifically target tumor cells, it will also affect normal cells. Diagnostic
tests have been developed to determine which patients will benefit most from
chemotherapy. MammaPrint is a gene expression assay by Agendia to detect a 70-
genes profile by reverse-transcriptase PCR. It uses 117 patients with axillary
40
lymph node-negative primary breast cancer to find genes highly correlated with a
short interval from primary tumor to distant metastases. MammaPrint profiling
can help doctors to identify the sub-population of patients that will benefit most
from chemotherapy [50].
1.5.1.2.2 Hormone Therapy
Hormone receptors—such as those for estrogen and progesterone—are on
the surface of cancer cells. They promote the growth of the cancer. Hormone
therapy works by blocking the hormone receptors in the receptor-positive tumor
cells. It has been used to reduce the risk of recurrence after surgery and to treat
advanced breast cancer. Hormone therapy may include tamoxifen, toremifene,
and fulvestrant, which are estrogen blockers. Anastrozole, exemestane, and
letrozole are aromatase inhibitors. Hormonal therapy is beneficial for women
with early-stage breast cancer that is positive for hormone receptors. A diagnostic
assay has been developed to identify the subpopulation of patients that benefit
most from hormone therapy. Oncotype DX detects a 21-gene profile by reverse-
transcriptase PCR to predict the risk of recurrence in patients taking tamoxifen.
Sixteen of the 21 genes are cancer-related genes with reference genes as reference.
A mathematical algorithm is derived from an empirical retrospective study to
calculate a score to estimate the risk of distant recurrence [49].
1.5.1.2.3 Targeted Therapy
Specific genes targets have been found to promote breast cancer
development. New drugs that specifically target those candidates are called
targeted therapy. Trastuzumab, lapatinib and pertuzumab are targeted drugs that
41
are available to treated women with breast cancer overexpressing the growth-
promoting protein HER2. Most of these drugs are humanized monoclonal
antibody or small-molecule kinase inhibitors that bind to different epitopes on the
extracellular domain of HER2 receptors and inhibit its dimerization. Because
tumor growth requires blood vessel formation to supply nutrition, anti-
angiogenesis drugs that prevent blood vessel growth in breast cancer are also
tested in clinical trials. Chemotherapy, hormone therapy and targeted therapy
have improved many patients’ outcomes. However, the problem is that after the
primary tumor has been killed, the remaining sub-population of cancer cells will
develop resistance to the initial treatment and can result in recurrence.
1.5.2 Emerging Strategy of Immune Checkpoint Blockade
The immune system in our bodies protects us from pathogens and
abnormal conditions. In the immune-editing theory, immune surveillance inhibits
tumor development from tumor initiation, and the tumor evolves through multiple
mechanisms to avoid immune system surveillance [86, 87]. Much effort has been
devoted to leverage immune system to fight against disease.
William Coley performed the first clinical trial of cancer immunotherapy
in 1891—Coley’s toxins. It was used to treat erysipelas with live or attenuated
bacteria [179]. However, the efficacy of Coley’s toxins was controversial. Other
immunotherapy drugs have been developed to target cancer-associated proteins
and to induce tumor cell apoptosis, phagocytosis and antibody-dependent cellular
cytotoxicity. These will trigger the adaptive immune system to target cancer [180].
Immunotherapy using immune checkpoint inhibitors has been promising in recent
42
years. Immune checkpoint inhibitor therapy is based on the biology of immune
suppression of the immune system by tumor cells. One mechanism of immune-
editing and surveillance is that tumor cells can suppress immune surveillance.
Tumors can express immune suppression factors that inhibit immune infiltration
and recruit myeloid-derived suppressor cells (MDSCs) and Regulatory T cells
(Tregs) to the local tumor environment to inhibit the normal function of immune
cells[181-186]. This will result in apoptosis of activated anti-tumor immune cells
and induce tolerance.
Among those, Tregs play an important role in the battle between tumor
cells and the immune system [187, 188]. Tregs constitutively express CTLA-4
[189, 190], which bind to CD80 and CD86 on the surface of antigen-presenting
cells (APCs). The binding of CTLA-4/CD80 or CD86 blocks the co-stimulatory
signal transduction and results in immunosuppression (Figure 1.6) [191-193].
Inhibitory receptor programmed cell death 1 (PD-1) contributes to immune
tolerance of self-antigens and is expressed in many tumor-infiltrating
lymphocytes, such as natural killer cells, dendritic cells, activated monocytes, B
cells and T cells [194].
43
Figure 1. 6: Introduction of CTLA-4 and PD-1 Treatment
Figure adapted with permission from Merelli et al [195].
The binding of PD-1/PD-L1 is an important mechanism to convey an
inhibitory signal to T cells and let tumor cells evade the immune system (Figure
1.7) [196, 197]. Many clinical trials using antibodies targeting PD-1 or PD-L1
have shown promising results in improving the survival rate in human prostate
cancer, melanoma, and breast cancer [198, 199]. Expression of programmed death
ligand 1(PD-L1) has been found in different types of cancer, such as non-small
cell lung carcinoma [200], esophageal cancer [201], pancreatic cancer [202] and
breast cancer [203]. Ghebeh and colleagues reported that PD-L1 is expressed in
34% of primary breast cancer samples but not in healthy controls. The expression
level of PD-L1 is significantly correlated with grade 3 tumors [203, 204].
44
Figure 1. 7: Biology and Mechanism of Anti-PD-L1 Antibody Treatment
Figure adapted with permission from Patient Resource LLC
With a combination of different therapy methods and the discovery of
novel therapies, the mortality rate of many types of cancer has decreased during
the past decades. However, due to the complexity of metastasis and immune
suppression in late stages, a diagnostic tool that can detect tumors early will
provide more therapy options and eventually improve patients’ outcomes.
45
CHAPTER 2
Early Detection Of Breast Cancer
2.1 Introduction
In the United States in 2014, 235,030 new breast cancer cases were
diagnosed, and there were 40,430 breast cancer deaths, making breast cancer the
second leading cause of death among females [205]. Currently, the standard
diagnostic tool for breast cancer is mammography. However, with its low
sensitivity and specificity and its limitation to a specific population,
mammography detects only 70% of breast cancers [33, 36, 37, 53]. The five-year
relative survival rate decreases from 100% to 22% when breast cancers are
detected at stage 4 instead of stage 1 [206]. Therefore, there is an urgent need for
early diagnosis of breast cancer, as it could improve disease outcome and be the
key to decreasing the death rate from breast cancer. This study proposes a serum
test called immunosignature to identify specific antibody signatures to diagnose
breast cancer early.
Serum proteins, circulating miRNAs, circulating tumor cells (CTCs) and
glycan biomarkers have been investigated to discover breast cancer biomarkers to
diagnose the disease [72, 207-210]. Current plasma protein biomarkers in the
human proteome organization (HUPO) project are in the range of ug/ml to mg/ml
and the concentration of target molecules shed from a small number of tumor
cells into the blood is 104 fold too low to be detected by current technology [100-
102]. This situation is especially true in early-stage breast cancer, when only a
small number of tumor cells initiate and secrete molecules. The challenge for the
46
foregoing approaches is the detection of low-abundance valuable biomarkers of
early stage breast cancer.
Circulating serum antibodies could provide a comprehensive reflection of
patients ‘health status. Immunosurveillance of B cells and T cells begins in the
early stage. Tumor-specific B cells can be activated and amplify the antibody
level, making the detection of a minimal exposure of antigen at the early stage
feasible. Pre-diagnostic antibodies have shown value in the early detection of
human breast cancer [19, 211, 212]. Chapman and colleagues found an antibody
response to six antigens (c-myc, p53, HER2, NY-ESO-1, BRCA2 and MUC1) in
early-stage primary breast cancer patients, although with low sensitivity [213].
Different panels of biomarkers all proved that a tumor-specific antibody response
exists in early-stage breast cancer, such as the carcinoma in situ stage [117, 118].
Multiple antibody-based approaches have been explored to leverage the specific
amplification effect of the immune system in the early-stage breast cancer.
Serological screening of the recombinant cDNA library (SEREX) uses phages to
display proteins from a recombinant cDNA library of interests and probed with
tumor patients’ sera of interest to select antigens with high binding interactions
with tumor related antibodies in the sera [122, 125, 126]. Serological proteomic
analysis (SERPA) separates proteins from cell lines or tissue lysates into spots by
two-dimensional polyacrylamide gel electrophoresis (2D-PAGE) [143]. Selected
spots will be analyzed by mass-spectrometry to find protein biomarkers for breast
cancer. Protein arrays printed with pre-synthesized recombinant proteins or
directly synthesizing proteins by an in-vitro expression system can also be used to
47
screen autoantibodies biomarkers in a high-throughput way. However, SEREX
and SERPA require large amounts of serum for profiling, and the initial
construction of the cDNA library in SEREX can easily cause biased result.
Proteins may not be folded and modified correctly in all three methods. More
importantly, these approaches can detect only antibody responses that target a
limited number of known proteins in each system. This cannot capture neo-
antigens by mutations or post-translational modification.
Here, we propose an unbiased antibody profiling technology using high-
density peptide microarrays for the early detection of breast cancer. The
technology, called immunosignature (ImS), uses a peptide array with 10,000
spotted peptides selected without bias from a non-biological peptide sequence
space. The assay consists of applying diluted sera to the array and then detecting
the binding pattern of the primary antibody with a labeled secondary antibody. In
previous work, we successfully used immunosignature technology to find
antibody signatures in early-stage pancreatic cancer and Alzheimer’s disease in
human patients [168, 175, 214]
This murine model develops invasive mammary carcinomas and can
recapitulate the morphologic, pathologic, and molecular features of human
luminal breast cancer at many stages [9]. We hypothesize that the FVB/N neuN
mouse model can recapitulate human humoral responses to breast cancer and
provide insights into the dynamic antibody profile from the early to late stages of
breast tumor development. It can test whether, in principle, immunosignature can
detect early stage of cancer or not.
48
We have found immunosignatures of early stage tumors as early as 12
weeks before the tumor is first palpable. The existence of different
immunosignatures at different stages reflects the dynamic changes in the antibody
profiles during tumor development. By using the immunosignature at each stage,
we achieved high sensitivity and specificity by cross-validation in multiple
classification methods. Applying the late-stage immunosignature to early-stage
samples resulted in low-prediction performance. This implies that the late-stage
tumor immunosignature cannot be used to develop the early-stage signature. This
study could determine the potential application of immunosignature technology in
preclinical diagnosis of early-stage breast cancer.
2.2 Results
2.2.1 HER2 Transgenic Mouse Model
This study uses FVB/N neuN transgenic mice, a well-characterized murine
model for HER2 breast cancer, to compare antibody profiles at different stages of
breast cancer. Mice are engineered with neuN with an MMTV promoter that
restricts the neuN gene from being expressed in mammary glands. We collected
blood samples from 23 transgenic mice and ten wild-type mice aged 12 weeks to
terminal age. The mice were examined every week for any tumor formation, and
any tumors found were measured once a week. Among the 23 transgenic mice,
the first tumor was detected, on average, at 33 weeks of age (Fig. 2.1a) with an
average tumor size of 96.3 mm3. We defined the point in time when the first
palpable tumor was found in a mouse to be window 0 week. A window is used to
calculate intervals between a certain age and the age at which the mouse had the
49
first tumor detected. The phenotypes of tumors at different stages are well
characterized into hyperplasia, dysplasia and carcinoma in situ, multiple foci of
carcinoma in situ or small lobular carcinoma, and carcinoma. Studies from
Boggio and Abe have shown a transition from normal mammary glands to
hyperplasia around 8.6 to 15 weeks [23, 215]. Figure 1b shows the tumor growth
curve of this mouse model. The average tumor size and its standard error are
plotted against days after the first palpable tumor was detected.
Figure 2. 1: Tumor Free and Tumor Growth Curve of FVB/N NeuN
Transgenic Mouse Model
23 FVB/N NeuN transgenic mice and 10 FVB/N wild type controls were
monitored every week for any palpable tumor in mammary glands. The median
tumor free time is at 33.1 weeks with an average tumor size of 96.3 mm3. 94.3%
of FVB/N NeuN transgenic mice will develop palpable tumors in 47.6 weeks
monitoring.
Tumor free curve of FVB/N NeuN mouse model
0 10 20 30 40 500
50
100
Transgenic
Wild Type
Median Survival
at 33.1 weeks
Age of weeks
Perc
en
t tu
mo
r fr
ee
T u m o r G r o w th C u r v e
D a y s a fte r 1 s t p a lp a b le tu m o r
Tu
mo
r s
ize
(m
m3
)
0 1 4 2 8 4 2 5 6 7 0 8 4 9 8
0
5 0 0
1 0 0 0
1 5 0 0
2 0 0 0
2 5 0 0
3 0 0 0
A B
50
2.2.2 Scheme of Experiment Design
Figure 2. 2: Scheme of Experiment Design
The scheme of the early diagnosis of breast cancer was layout in Figure
2.2. Serum samples from FVB/N transgenic mice and wild type controls were
collected every two weeks during tumor development. Mice were palpated every
week to find any tumor occurrence. Serum samples were categorized into four
stages: ‘Stage-1’ for samples at 12 weeks before 1st tumor, ‘Stage-2’ for samples
at 8 weeks before 1st tumor, ‘Stage-3’ for samples at 4 weeks before 1st tumor,
‘Stage-4’ for samples at at 1st tumor. For each time point, transgenic mice were
compared with age-matched wild type controls to find significant peptides that
passed statistic test.
Tumor measured every week, blood collected every 2wks
Healthy Early stage 1st palpable tumor
Stage 1 12 weeks before
Stage 2 8 weeks before
Stage 3 4 weeks before
Stage 4 1st tumor
Early stage Age matched
VS
Transgenic Wild Type
51
2.2.3 Influence of Genetic Background on Tumor-specific Immunosignature
In order determine the extent to which transgenes might affect the ImS, we
compared a sample set of seven transgenic mice at 12 weeks and seven age-
matched wild-type controls. Principle component analysis of these samples did
not reveal clustering by groups (Fig. 2.3a). Fig. 2b shown a Heatmap analysis of
four transgenic mice and six wild-type controls at 12 weeks. Although transgenic
and wild type mice were not mixed after hieratical clustering, the intensity of 833
selected peptide did not show distinguishable pattern between two groups (Fig.
2.3b). No peptides could differentiate the two groups by hypothesis test in this
ten-sample set with a two-tail t-test with p value cut off at 0.05 (Fig. 2.3c). An
independent sample set of seven transgenic mice and nine wild- type controls
resulted in 37.5% accuracy using the top 833 peptides with the highest p value in
the previous set of seven transgenic and seven control mice (Fig. 2.3d). We
concluded that the ImS was not influenced by the transgenes before the tumor
activated.
52
Figure 2. 3: Influence of Genetic Background on Immunosignature
A) Principle component analysis of 7 transgenic and 7 wild type mice at 12 week
age by all peptides.
B) Heatmap of 4 Tg and 6 WT at 12 weeks age by 833 peptides with top p value
by t-test without MCC. No distinguishable patterns between Tg and WT were
observed.
C) Hypothesis test of 4 Tg and 6 WT at 12 week age. No peptides pass T-test. 1%
peptides have P<0.01. 0.5% have <1.3 fold change.
D) Classification result by 7 Tg and 9 WT at 12 weeks age by 833 peptides with
top p value by t-test without MCC gives 37.5% accuracy.
Each colored square represents relative median intensity. Blue: low expression;
Red: high expression; Yellow: average expression. Clustering using two-way
unsupervised hierarchical clustering with distance as measurement of similarity in
GeneSpring software. Intensity is median normalization, the classifier uses Naïve
Bayes by Weka.
2.2.4 Existence of Pre-tumor Immunosignature Prior to Tumor Palpation
To test whether immunosignature technology can detect changes at the
pre-palpable tumor stage, we used eight transgenic mice (two weeks before the
first tumor) and eight wild-type mice samples. We were able to detect a
significant difference in antibody binding of 51 peptides (p≤0.05 with Benjamin
and Hochberg FDR) between the transgenic and wild-type samples (Fig. 2.4a).
The 51 peptides signature was further validated using an independent set of
samples, including eight transgenic and eight wild-type mice. By using K-means
to classify, the 51 peptides gave 100% sensitivity and 66.7% specificity (Fig.
2.4b). Two experiments with different peptide libraries and samples all showed
Tg WT
A CB
D
P value cutoff 0.05 0.02 0.01T-test FDR 0 0 0T-test Bonferroni 0 0 0
Tg(Predict) WT(Predict)
Tg (True) 0 7
WT (True) 3 6
53
that significant peptides could be found between pre-tumor transgenic samples
and wild-type controls. The average age of these pre-tumor samples were greater
than 17 weeks, at which time the mammary glands had already progressed into
carcinoma in situ. We concluded we can use immunosignatures to segregate
transgenic and wild-type mice before the first palpable tumor was detected.
Figure 2. 4: Immunosignature Can Show a Separation between Transgenic
Mice and Age Matched Controls Two Weeks before First Palpable Tumor
A) Independent sample sets with 8 transgenic mice (2 weeks before 1st tumor)
and 8 wild type mice samples using CIM 10K Version 3 peptide library. 51
peptides (p≤0.05 with Benjamin and Hochberg FDR) were significant between
transgenic and wild type samples. Principle component analysis shows separation.
B) Cross validation result of training sample and prediction performance in an
independent samples gives a sensitivity of 87.5% (7 right in 8 cases) and
specificity of 87.5% (7 right in 8 cases).
2.2.5 Earliest Time Point of the Detectable Immunosignature
After showing the existence of immunosignature at two weeks before first
palpable tumor, we were interested in how early immunosignature could detect
tumor. In order to test this, we used mice at 12 weeks, and 16 weeks to estimate
the earliest time point in CIM-Version 2 arrays. Four transgenic and seven wild
type mice at age of 12 weeks were ran in the same batch of slides (Fig. 2.5). No
Cross validation Result
Sensitivity 87.5%
Specificity 87.5%
Independentvalidation
Result
Sensitivity 100.0%
Specificity 66.7%
AB
54
peptide passed T-test with p value cutoff at 0.05. At age of 16 weeks, fifty three
peptides were significant between five transgenic and seven wild type mice. The
lack of difference between transgenic mice and wild type control at age of 12
weeks and existence of significant difference at age of 16 weeks implied the
earliest detectable time point for breast tumor is between 12 to 16 weeks by
immunosignature in CIM-Version 2 arrays.
Figure 2. 5: Heatmap of Immunosignatures of Different Time Points in
Tumor Development
Four NeuT transgenic FVB/N (Tg) mice and seven wild type FVB/N (WT) mice
are assayed at age of 12 weeks and 16 weeks. Each colored square represents
relative median intensity. Blue: low expression; Red: high expression; Yellow:
average expression.
2.2.6 Time Course of Immunosignatures During Tumor Progression in
FVB/N neuN Mice
Previous studies have shown that tumors present different tumor antigens
at different stages [86, 87]. Also, biomarkers that were selected using late-stage
patients gave poor prognostic performance in an early-stage setting. These facts
imply that the immune system develops different antibody profiles along with the
TgWT
11.4 Age of weeks 16.6
55
development of a tumor. The immunosignature is a representation of the antibody
profile at a given point in time. Therefore, we were interested in comparing
immunosignatures from different stages to quantitatively investigate differences
in antibody profiles. In this experiment, we selected 23 transgenic mice with ten
wild-type controls from previous populations and collected serum at multiple
points in time. Both the transgenic and control samples were grouped into four
stages by the numbers of weeks before the first palpable tumor. We defined ‘IS-1’
as 12 weeks before the first tumor; ‘IS-2’ as eight weeks before the first tumor;
‘IS-3’ as four weeks before the first tumor; and ‘IS-4’ as the time of the first
tumor. Wild-type mice (aged 15.3 to 45.4 weeks) were used to make between-
stage comparisons comparable and to eliminate age influence.
By using 23 transgenic mice and ten wild-type controls, we selected the
top 200 significant peptides using an FDR-corrected T-test with a larger than 1.3-
fold change between transgenic and wild-type mice for each point in time (Fig
2.6). To visualize the differences between transgenic mice and the wild-type
controls, we plotted heatmaps containing these top 200 peptides between the
transgenic and wild-type groups.
We observed a change in immunosignatures from a greater to a less
signature difference between the wild-type and transgenic mice from early to late
stage tumors (Fig 2.6). This demonstrates that immunosignature can capture
tumor development. The IS-1 heatmap showed two clear clusters of transgenic
and wild-type mice. The majority of the peptides were highly reactive in the wild-
type controls and low reactive in the transgenic mice. This pattern reversed in IS-
56
2, when the majority of peptide intensities were higher in transgenic mice. In IS-3
and IS-4, the wild-type mice began to be clustered into the transgenic group.
Distinguishing patterns of peptides intensities in transgenic and wild- type mice
became less obvious in later stages of the tumor, while small clusters in these two
groups appeared instead.
Figure 2. 6: Time Course of Immunosignatures during Tumor Progression in
FVB/N NeuN Mice
23 NeuT transgenic FVB/N (Tg) mice are assayed at four time points- 12, 8, 4
weeks before the first tumor and at the first tumor. 10 wild type FVB/N (WT)
mice are assayed. We select the top 200 most significant peptides using FDR-
corrected T-test for each time point. Heatmap and PCA using each stage
signatures are used to show the difference between transgenic mice and wild-type
controls. Significant peptides are judged by p<=0.05 with fold change >=1.3.
Each colored square represents relative median intensity. Blue: low expression;
Red: high expression
The Venn diagrams show the overlap of immunosignatures at the four
stages (Fig. 2.7). We found that different tumor stages have different and partially
overlapping signatures. The adjacent stages share more peptide overlap (Fig. 2.8).
Tg WT
175 17525 165 16535 110 11090
Stage 1 Stage 3 Stage 4Stage 2
57
The immunosignatures of IS-1 and IS-2 shared 25 peptides. The
immunosignatures of IS-2 and IS-3 shared 35 peptides, while the
immunosignatures of IS-3 and IS-4 shared 90. Sixteen peptides were shared by all
four stages. These numbers are larger than the overlaps expected by peptides
chosen at random. The number of overlapping peptides between the earliest stages
(IS-1) and all the following stages remained consistent, but the number increased
when comparing the latest stage (IS-4) with all the previous stages (Fig. 2.9).
Taken together, this also shows that immunosignature can detect differences
between transgenic mice and healthy control mice caused by tumor development
at multiple stages.
58
Figure 2. 7: Venn Diagram of Different Immunosignatures Along Tumor
Development
Different stages in tumor development have different immunosignatures, a
potential reflection of antigens profile changes by immunoediting in tumor
development. 200 significant peptides with highest fold change between
transgenic and control group are selected at four stages, corresponding to 12
weeks before, 8 weeks before, 4 weeks before and at 1st tumor. Peptides shared
among 4 stages take <10% in any stage.
Figure 2. 8: Numbers of Selected Peptides Overlap between Two Adjacent
Stages during Tumor Development
N u m b e r o f S ig n if ic a n t P e p tid e s O v e r la p
B e tw e e n A ja c e n t S ta g e s
S ta g e C o m p a r is o n
Ov
erla
pp
ed
Pe
pti
de
Nu
mb
er
S ta g e 1 -S ta g e 2 S ta g e 2 -S ta g e 3 S ta g e 3 -S ta g e 4
0
2 0
4 0
6 0
8 0
1 0 0
59
Figure 2. 9: Numbers of Selected Peptides Overlap between Earliest Tumor
Signature with Other Stages Signatures
2.2.7 Cross-stage Prediction
Recent biomarker failures implied that using late-stage biomarkers for
early prediction leads to low performance. Here, we tried to quantify cross-stage
performance by using an immunosignature from one age to predict another. We
use SVD in a PAMR package to predict twelve combinations of any two of the
N u m b e r o f S ig n if ic a n t P e p tid e s O v e r la p B e tw e e n
E a r lie s t S ta g e w ith th e R e s t
S ta g e C o m p a ris o n
Ov
erla
pp
ed
Pe
pti
de
Nu
mb
er
S ta g e 1 -S ta g e 2 S ta g e 1 -S ta g e 3 S ta g e 1 -S ta g e 4
0
2 0
4 0
6 0
8 0
1 0 0
N u m b e r o f S ig n if ic a n t P e p tid e s O v e r la p B e tw e e n
1 s t tu m o r w ith th e R e s t
S ta g e C o m p a ris o n
Ov
erla
pp
ed
Pe
pti
de
Nu
mb
er
S ta g e 4 -S ta g e 3 S ta g e 4 -S ta g e 2 S ta g e 4 -S ta g e 1
0
2 0
4 0
6 0
8 0
1 0 0
60
four ages (Table 2.1). Using 200 peptides from IS-1 to predict IS-2, IS-3, and IS-4
samples gave 0% accuracy. Using IS-2 to predict IS-1, IS-3, and IS-4 samples
gave 0%, 4.2% and 8.3% accuracy, respectively. Using IS-3 to predict IS-1, IS-2,
and IS-4 samples gave 28.6%, 58.3% and 87.5% accuracy, respectively. Using IS-
4 to predict IS-1, IS-2, and IS-3 samples gave 9.5%, 41.7% and 75.0% accuracy,
respectively. This implies that late-stage tumor immunosignature cannot be used
to develop an early-stage signature.
Table 2. 1: Cross-Stage Prediction Performance in the Training Set
Using peptides selected from 10 week before, 4 weeks before or 6 weeks after
first tumor to predict samples from another time points. Accuracy is calculated by
the percent of correct prediction of both transgenic and wild type mice in both
groups.
2.3 Methods
2.3.1 Mice and Sample Collection
Female FVB/NJ mice (n=10, age 6 weeks) were purchased from Jackson
Laboratories (Bar Harbor, ME). Female FVB/N neuN transgenic mice were
kindly gifted from Dr. Chella David at Mayo, Rochester. Transgenic FVB/N
NeuN mice and wild-type FVB/NJ mice were maintained and weaned in the same
barrier isolation housing with standard rodent feed and water. Mice were palpated
every week to detect mammary tumor growth. Blood samples were collected by
submandibular venipuncture using a 5.0 mm lancet (MEDIpoint, Inc. NY). Serum
Accuracy -12 wks -8 wks -4 wks 1st tumor
-12wks to predict 0.0% 0.0% 0.0%
-8wks to predict 0.0% 4.2% 8.3%
-4wks to predict 28.6% 58.3% 87.5%
1st tumor to predict 9.5% 41.7% 75.0%
61
was separated by serum separating Microtainer (BD Biosciences, San Jose, CA)
and stored at -20 C with aliquots. Mice were euthanized by 2,2,2-tribromoethanol
at a dose of 125mg/kg through intraperitoneal injection when tumor size exceeded
10% of the body weight. All murine experiments were conducted following
animal protocol reviewed and approved by Arizona State University Institutional
Animal Care and Use Committee.
2.3.2 Preparation of Microarray
CIM10K peptide arrays were printed as earlier described [110, 160]. Two
libraries of peptides were used separately: CIM10K Version 2 and Version 3.
Both libraries contained 10,000 random 20-residue peptide of 17 random
sequence amino acids. Peptides were designed with random sequences with
nineteen amino acids (cysteine excluded), except for three residues at N-terminal
as a linker to attach peptides covalently on Nexterion A+ aminosilane slides
(Schott). The CIM10K Version 2 library contained L amino acid with Gly-Ser-
Cys-NH2 linker. Version 3 contained both L and D amino acids with Cys-Ser-
Gly-NH2 linker. Peptides were synthesized by Sigma Genosys (St. Louis, MO).
Slides were printed using the two-up format by piezoelectric non-contact printer
at Applied Microarrys (Tempe, AZ). Quality-control tests were performed using
subsamples of each batch with pooled human naïve serum to select high array-to-
array correlation and dynamic range. After being produced, arrays were stored in
containers, protected from light.
62
2.3.3 Probing Serum Antibodies on Peptide Microarrays
Slides were first washed by 33% isopropanol, 7.5% acetonitrile, and 0.5%
trifluoroacetic acid for five minutes, then dried by centrifuge at 800 rpm for five
minutes. After pre-wash, slide processing was automated by a TECAN HS4800-
Pro automated incubator, following a protocol optimized for antibody binding.
Briefly, arrays were blocking by 0.014% mercaptohexanol, 3% BSA, and 0.05%
Tween 20 in 1×PBS solution for 1 h at 23° C. Serum samples were then diluted at
1:500 in incubation buffer (3% BSA 0.05% Tween 20 in 1×PBS) and probed on
arrays for one hour at 37° C. Biotinylated anti-mouse IgG(H+L) antibodies
(Bethyl Laboratories, Montgomery, TX) were diluted in the incubation buffer at
5nM for one hour at 37° C. Bound secondary antibodies were visualized by Alexa
Fluor-647 labeled streptavidin (Invitrogen, CA) at 5nM for one hour at 37° C. The
injection volume of each step was 110 μl per array, and TBST solution was used
to wash the array between steps. The final wash used TBST and distilled water to
remove residual salt on slides. Correlation for technical replicates less than 0.85
were re-probed [175].
2.3.4 Microarray Data Processing
After the final wash, slides were scanned by an Agilent C scanner at
633nm emission (Agilent Technologies, CA) with 100% laser power and 100%
PMT. Images were extracted by GenePix Pro 6.0 (Molecular Devices, Sunnyvale,
CA) and saved as gpr files. Poor-quality peptide spots were excluded by flagging
them as “absent” by visual inspection. Samples with Pearson correlation
coefficient >0.85 among technical replicates were used. Raw data from gpr files
63
were collected and imported into GeneSpring GX 7.3 or R for later processing
(Agilent Technologies, Palo Alto, CA). Data were normalized to the 50th
percentile to minimize array- to-array variation. Signal intensities below 0.01
were set to 0.01. Batch effects were adjusted by SVA algorithm with irw as
adjustment method with 10 iterations. The biological group was used as a
covariate in SVA.
2.3.5 Statistical Analysis
Statistical analyses were done using Genespring GX 7.3 (Agilent
Technologies, Palo Alto, CA), R version 3.1.1, and JMP version 11. P-values
were corrected for multiple testing by the Benjamin-Hochberg false discovery rate
method. The Welch t test was used in R to select significant peptides with greater
than 1.3-fold changes and p<0.05 between two groups. We used a support vector
machine with settings including polynomial dot order at 1 and scaling order at 0
for cross-validation and prediction. Packages used in R were sva, caret, e1071
[216, 217].
2.4 Discussion
In this study, we questioned the feasibility of breast cancer early detection
by immunosignature technology. In comparing transgenic mice and their non-
transgenic littermates, we found the tumor-specific immunosignature of
significantly different peptides as early as 12 weeks before a tumor became
palpable. We also tested four more time points at early stages between transgenic
and wild-type groups and found the immunosignature of significant peptides for
each stage during tumor development. Fewer than 10% of the peptides were
64
shared among four time points. Using early-stage immunosignature to predict
later-stage sample results in low classification performance.
Early detection of breast cancer has the advantage of giving patients early
treatment intervention and improving their outcomes. In our previous work, we
successfully used immunosignature technology to find early antibody signatures
in Alzheimer’s disease in a mouse model [168]. Here, we wanted to test whether
we could detect the immunosignature of the early stages of breast cancer.
However, serum samples from the same cancer patients before and after clinical
diagnosis are hard to get. It is difficult to study biomarkers of early-stage breast
cancer without proper early-stage samples. The FVB/N neuN transgenic mouse
model is a well-established model to study breast cancer. In order to cover
samples at the early stages of tumors, we collected serum samples for each
transgenic mouse at time points from zero to twelve weeks before the first
palpable tumor—aged 12.36 to 45.4 weeks. In this mouse model, hyperplasia
occurred at around 8.6 to 15 weeks [23, 215]. So, selected samples could have
good coverage of early stage breast cancer. Samples at multiple time points were
tested to see whether we could find the immunosignature that distinguishes
transgenic mice from wild-type controls.
In human clinical trials, Lu and colleagues showed that autoantibodies
against HER2 and p53 can be detected by protein microarrays 150 days before a
breast cancer diagnosis. Opstal-van Winden and colleagues also showed that
breast cancer autoantibodies can be found in patients’ samples a median of 21.3
months before diagnosis. These findings proved the feasibility of early diagnosis
65
of breast cancer. However, a few protein biomarkers that were discovered in late-
stage cancer were evaluated each time, and their samples did not cover a wide
range of early-stage breast cancers in humans. Using the same mouse model,
Jianning Mao and colleagues used serological screening of cDNA expression
libraries to identify tumor-associated autoantibodies in serum one month prior to
the development of a palpable tumor. Six proteins were selected [19]. They found
that autoantibody responses of six proteins appeared in mice as early as 20-30
weeks of age. In our study, the earliest time point at which we found a difference
by immunosignature was between ages 12 and 16 weeks, which is still during the
time of hyperplasia development in this mouse model. This found
immunosignature is more sensitive to the detection of a minimal antibody
response in early-stage breast cancer. Our study also used more time points to
reflect antibody responses in early-stage breast cancer more comprehensively.
When we compared immunosignatures at different stages, the early
signature was quite different from the late one. Antibodies that specifically target
antigens that emerge in the late stages do not exist at an earlier stage. Fewer than
10% of the peptides overlapped the four stages. 34 significant peptides
overlapped Stages Ⅰ and Ⅳ, while 90 peptides overlapped Stages Ⅲ and Ⅳ. This
showed that the binding pattern of the extra 56 peptides had been generated since
Stage Ⅲ. In our previous study of Alzheimer’s disease, we also found that the
early-stage was different from the late-stage immunosignature. These findings
support the immunoediting theory of new antigens generated by tumors during an
active immunoediting process.
66
Several studies have shown that biomarkers discovered in established
tumor patients did not predict well in the early detection of breast cancer. Lu and
colleagues used eight antigens to detect breast cancer samples 150 days before
diagnosis. HER2 and p53 among the eight antigens gave a ROC score of 0.6 and
0.63, respectively. Opstal-van Winden and colleagues also showed that markers
discovered in diagnostic samples might not be valid in pre-diagnostic samples.
Here, we use late-stage immunosignature to predict early-stage samples, and the
performance of cross-stage prediction decreased significantly compared with
cross-validation within each stage. Our findings suggest that using biomarkers
discovered in diagnostic samples may be not valid in pre-diagnostic samples.
They further suggest the importance of using proper pre-diagnostic samples to
discover early-detection biomarkers.
In terms of intensity level, the majority of peptides in early breast cancer
sera are less reactive than non-cancer controls. We hypothesize that the down
regulation that of antibodies may be caused by immunosuppression. Edward Y.
Woo and colleagues reported CD4+ CD25+ T cells secrete immunosuppressive
cytokine transforming growth factor- β in early-stage non-small cell lung cancer
[218]. In breast cancer, Isabel Poschke and colleagues observed
immunosuppression in early-stage breast cancer patients in terms of the
exhaustion of tumor-associated T cells [219]. A report also showed that Atypical
T cell-suppressive neutrophils occurred during early tumor progression [220]. In
terms of antibody response, data from Jianning Mao and colleagues showed a low
reactivity of autoantibodies to four tumor-associated proteins in early-stage breast
67
cancer [19]. Our findings further showed that immunosuppression already occurs
at early stages of tumor development in humoral response.
We also observed that the immunosignature of transgenic mice is
homogeneous among different mice at Stage Ⅰ but become sparsely distributed
at later stages. Because the tumorigenesis of this mouse model is initiated by a
single oncogene (NeuN) expression, the immune system in different mice may
response in a similar way. However, when more mutations occur during tumor
development, the immune system in each mouse will respond differently to new
tumor antigens. At the scope of 12 weeks, the dramatic changes in the
immunosignature imply that the immune system has gone through significant
changes along with tumor development.
We recognize that the limited sample size and the use of a random
sequence array may not be the best conditions for finding gene and protein targets.
Because serum from each mouse needed to be collected as the tumors developed,
the tumor’s histology information could not be known. However, this study
established the methodology and demonstrated the feasibility of early detection of
breast cancer by immunosignature technology in a mouse model.
Early detection of breast cancer could lead to early treatment intervention
and improved patient outcomes. Future studies can use retrospective sample
collections in clinical trials to develop a specific immunosignature for early-stage
breast cancer in humans. If these immunosignatures are found, we could use
multiple immunosignatures at different stages in early breast cancer to predict
breast cancer early and reduce the false-positive rate in cancer prediction.
68
2.5 Supplements
2.5.1 Selection of multiple batches of array
Due to the large number of slides that were used in this study, slides from
multiple batches of printing were selected. In order to reduce the variance caused
by different printing of batches, we tested the correlation of the same quality
control sample in eight different batches. Statistic characterization were also
calculated to evaluate the overall performance for different batches, including
intensity of empty spots on a slides, ratio of median intensity of all peptides to
empty spots, coefficient of variance of all peptides, correlation between two
technical replicates using same quality control samples (Table 2.2, table 2.3).
From the analysis, batches from 474, 493, 494, 524,530,538 and 540 were
selected.
Table 2. 2: Correlation Analysis for Different Printing Batches of Slides
Using Quality Control Sample
First column and last row are the batch number. Square of Pearson Correlation is
calculated for each comparison between eight different batches.
474 0.464 0.441 0.420 0.428 0.385 0.292 0.188
486 0.570 0.360 0.325 0.288 0.248 0.130
493 0.615 0.584 0.532 0.400 0.225
494 0.855 0.837 0.638 0.324
524 0.928 0.613 0.318
530 0.688 0.354
538 0.477
540 486 493 494 524 530 538 540
69
Table 2. 3: Statistic Characterization of Different Batches of Slides
Correlation analysis for different printing batches of slides using quality control
sample. Empty: intensity of empty spots on a slides; f647 M/E: ratio of median
intensity of all peptides to empty spots; f647 CV: coefficient of variance of all
peptides; Correlation: correlation between two technical replicates using same
quality control samples.
2.5.2 Removal of Batch Effect
One important issue in microarray experiment is the variance caused by
batch effect. In this experiment, seven different batches of slides were used.
Principle Component Analysis (PCA) of 336 samples using all peptides with
median normalization was plotted to illustrate batch effect (Figure 2.10). From the
figure, we can see batch 474 and some samples in batch 493 were separated from
other batches. Batch removal methods were needed to remove these difference
from non-biological difference.
Batch Empty f647 M/E f647 CV Correlation Printing date Slide Order
474 1321.78 12.87 68.00% 0.95 5/18/2012 135 Schott A+ slides A1947
486 583.67 8.1 97.90% 0.9 5/29/2012 136 Schott A+ slides A1956
493 534.55 5.71 70.40% 0.87 6/7/2012 126 Schott A+ slides A1962
494 920.22 6.97 78.10% 0.92 6/13/2012 126 Schott A+ slides A1970
524 1646.39 7.72 66.50% 0.98 7/23/2012 136 Schott A+ slides A2002
530 1168.19 7.21 60.80% 0.96 7/27/2012 136 Schott A+ slides A2003
538 442.08 6.11 83.40% 0.92 8/1/2012 136 Schott A+ slides A2015
540 495.57 1.17 21.30% 0.56 8/3/2012 136 Schott A+ slides A2016
70
Figure 2. 10: Principle Component Analysis of Different Batches
Our lab had employed ComBat to remove batch effects [221, 222] in
previous studies. Surrogate variable analysis (SVA) is also a widely used batch
effect removal method [217]. Batch removal methods need to be selected based
for the specificity of different studies [223, 224]. Here, we evaluated SVA and
ComBat methods in a data set from array result of 168 different mice sera samples
each with two replicates (Fig 2.11, Fig 2.12, Fig 2.13). Packages from SVA and
ComBat were run in R. Correlations between median normalized data and SVA or
ComBat adjusted data for each sample were plotted. PCA was also plotted to
show the variance between different batches after adjustment of batch effects. In
Principle component one
Pri
nci
ple
co
mp
on
ent
two
71
SVA, we can choose irw and twostep as different methods. I tested the irw
method with 10 iterations and twostep method in SVA (Fig 2.11).
Figure 2. 11: Correlation of Technical Replicates Before and After SVA
Adjusted and PCA Analysis of Batch Effects
Principle component one
Pri
nci
ple
co
mp
on
ent
two
Principle component one
Pri
nci
ple
co
mp
on
ent
two
72
Figure 2. 12: Correlation of Technical Replicates Before and After Combat
Adjusted and PCA Analysis of Batch Effects
After both irw and twostep method in SVA, batch effects were largely reduced by
the visualization of PCA. Correlation between technical replicates for the same
sample increased. By comparing statistics of technical correlation after different
SVA and ComBat, we found irw with 10 iterations in SVA package gave the best
performance in both reducing batch effects and increasing correlation of technical
replicates for the same samples (Figure 2.).
Technical Correlation before ComBat Technical Correlation After ComBat
73
Figure 2. 13: Comparison of Different Batch Effect Removal Methods
Correlation of technical replicates for 168 samples after different batch effect
removal methods are summarized.
74
CHAPTER 3
A MODEL FOR THE EARLY DETECTION AND TREATMENT OF
CANCER
3.1 Abstract
Cancer kills ~8M people/year and the WHO predicts an epidemic of
cancer in the developing world over the next 20 years due to increased longevity.
Development of a simple method to detect and treat cancer early could be a
solution to this challenge. Here we test this approach in principle in a mouse
mammary tumor model. We had previously demonstrated that tumors could be
detected at an early stage in the mouse FVB-NeuN (MMTVneu/202Mul/J)
mammary cancer model by immunosignature diagnostics. Immunosignatures are
profiles of antibodies in the blood displayed on peptide arrays. We determined if
early treatment at the time of diagnosis with the checkpoint blockade, anti-PD-L1,
would inhibit the tumor growth. A course of 3 injections at the time of diagnosis
significantly retarded tumor growth relative to untreated controls. The early
treatment elicited long protection which could restrict the tumor growth even four
months after the treatment. The treated mice share a specific immunosignature 28
days post treatment and 28 days post the first palpable tumor, compared to the
untreated mice. This indicates that the early treatment elicited an anti-tumor
immune response that could inhibit tumor growth. This preliminary experiment
suggests that early diagnosis combined with systemic treatment might be a
method to control or eliminate cancer.
75
3.2 Introduction
Since most of the deaths from cancer occur in the developing world, any
solution to the cancer problem must be applicable to the whole world population.
A possible solution is to develop a simple, inexpensive technique to detect cancer
early and couple it to a systemic treatment that would eradicate the tumor. Early
detection and early treatment has the advantages that the tumor cell number is
small, the chance for evolving evasive mutations is less and the probability of
metastasis is greatly decreased. It also is likely a lower dose and shorter treatment
would be efficient, so the cost and side effect will be reduced.
We developed the immunosignature diagnostic technology for the early
detection of changes in health status, including early cancers [168, 175]. The
technology is based on the peptide array with 10,000 spotted peptides. These
peptides are disease agnostic as they are chosen from non-biological peptide
sequence space [225]. The assay consists of applying diluted sera to the array and
then detecting the binding pattern of primary antibody with a labeled secondary
antibody. The technology relies on the activation of humoral immune response to
diseased cells, such as nascent transformed cells. The humoral immune response
changes antibody profile and can be detected by the immunosignature. It has
been used to detect multiple diseases. We have shown that this technique can
detect Alzheimer’s disease at an early stage in a mouse model [168] and recently
showed that mammary tumors could be detected by this method at a very early
stage in FVB-NeuN mice, a mouse mammary tumor model.
76
Immune checkpoint blockades are showing promise as systemic cancer
therapeutics. A portion, but not all, patients treated have demonstrated definitive
responses [195]. The underlying concept is that an antitumor immune response is
naturally mounted against any tumor but it is suppressed in multiple ways by the
tumors cells directly (PD1, PD-L1) or through T-regulatory cells (CTLA4) [226].
Checkpoint blockades release this suppression allowing the native antitumor
response to limit or even kill the tumor cells. For example, anti-PD-L1 is an
antibody therapeutic that has been reported to have efficacy in treatment of late
stage melanoma and lung cancer. To date, three such drugs have been approved
by FDA. Many more human trials are in the progress. Patients in all clinical trials
are at late stage disease [198, 199].
We proposed that combining early detection and a checkpoint inhibitor
treatment (eg. anti-PD-L1) would be an effective way to treat tumors, before they
were imageable or symptomatic. The question is whether the immune response to
a tumor at an early stage, even if unsuppressed, would be sufficient to inhibit the
tumor. We tested this basic concept using a transgenic mouse model of breast
cancer. In the FVB-NeuN model the MMTV promoter activates the wild type rat
ERBB2 gene in mammary cells, and every mammary cell potentially turns to be
cancer cell. The mice develop 1-3 (2.7±0.33) tumors at 19-45 (32.6±1.2) weeks
of age in our hand. Studies from Boggio and Abe using same mouse model have
shown a transition from normal mammary glands to hyperplasia around 8.6 to 15
weeks [23, 215]. We had previously demonstrated that an immunosignature of the
initiation of the tumor could be detected at least 12 weeks before a palpable tumor.
77
We defined this signature as the early stage of the cancer, namely stage I
signature. Here we ask whether checkpoint treatment by PD-L1 blockade can be
effective at this early stage.
3.3 Results:
3.3.1 Early Detection of Breast Cancer and PD-L1 Treatment
The basic outline of the experimental design is cartooned in Figure 1.
Mice were monitored every four weeks for their immunosignature. The treatment
was immediately initiated when a mouse was scored positive for the stage I
cancer signature in two contiguous monitors. The summary heatmap of the
confirmed stage I signature comparing to the historical data is given in Figure 3.1.
The treatment consisted of 3 intraperitoneal injections of 200ug anti-PD-L1
(10F.9G2) every three days. Serum was collected in four week intervals after
treatment. The timeframe for occurrence of signatures, palpable tumor and
treatments is given in Table 3.1. A group of 3 mice were not treated after the
signature was identified. The anti-PD-L1 antibody injection step was done by
Luhui Shen.
78
Figure 3. 1: Cartoon of Experimental Design Outline
Mice were monitored every four weeks for their immunosignature. The treatment
was immediately initiated when a mouse was scored positive for the stage I
cancer signature in two continue monitors. Treatment consisted of 3 injections of
anti-PDl1 (10F.9G2) over 9 days. Serum was collected in four week intervals
after treatment.
Table 3. 1: The Timeframe for Occurrence of Signatures, Palpable Tumor
and Treatments
Sample from anti-PD-L1 treatment group and non-treated control are labeled in
the rows. Columns are ages of mice at seven time points. “+” means samples at
that time point scores positive for the stage I cancer signature. Green shows the
Early detection Drug treatment
Stage 1 Signature Positive
Stage 1 Signature Positive
0 days +28 days +37 days
3 inject α-PD-L1 Ab
Sample 16 wks 20 wks 24 wks 28 wks 32 wks 36 wks 40 wks
Treated #1 + - + + Tumor
Treated#2 + - + +
Treated#3 + +
Treated#4 + + Tumor
Treated#5 + + Tumor
Treated#6 + + Tumor
Treated#7 + - + + Tumor
Treated#8 + + Tumor
Treated#9 + +
Treated#10 + +
Non-treated#1 Tumor
Non-treated#2 Tumor
Non-treated#3 Tumor
79
time point of giving the treatment. Red “Tumor” shows the time point of having
first palpable tumor.
3.3.2 Inhibition of Tumor Growth Rate by Early Treatment
The tumor volumes were measured every week and plotted in Figure 3.2.
As evident the PDL1 treatment had several effects. Four of ten treated mice did
not developed a palpable tumor at 40 weeks old. Tumor growth rate was
significantly retarded in the treated mice that eventually developed tumors. This
tumor growth inhibition was a long term protection. It was still effective to hinder
the growth of tumors in the mice that started to develop a tumor 4 months after of
the treatment. Tumors from the untreated controls and the treated mice were
examined at the same time point after detection.
80
Figure 3. 2: Tumor Growth Characteristics in FVB/N neuN α-PD-L1
Treatment Experiment
The tumor volumes were measured every week and plotted. Tumor growth rate
was significantly retarded in the α-PD-L1 treated mice that eventually developed
tumors. Non-treated mice share same characteristics with historical non-treated
mice.
3.3.3 RNA Expression of PD-L1 in Mammary Glands
PD- An implication of the effect of anti-PD-L1 treatment is that the gene
is expressed in the early stage of tumor development. To test this we performed
an RNA analysis on mammary glands with normal morphology and established
tumors with different size. (Figure 3.3). The PD-L1 expression level was
significantly higher in normal mammary glands than established tumors,
especially in mammary gland with increasing expression of MMTV, which
indicates the initiation of tumor development by increasing the expression of rat
T u m o r G r o w th C u r v e
D a y s a fte r 1 s t p a lp a b le tu m o r
Tu
mo
r s
ize
(m
m3
)
0 1 4 2 8 4 2 5 6 7 0 8 4 9 8
0
1 0 0 0
2 0 0 0
3 0 0 0
H is to r ic a l
N o n -tre a t
P D -L 1 tre a te d
81
ERBB2. The PD-L1 expression was decreased after the tumor was established.
All of the tumors had similar expression level of PD-L1 regardless the size and
expression level of rat ERBB2 (Figure 4). We concluded the L1 gene and
presumably protein were expressed on mouse mammary cells and the expression
was lower in established tumor.
The RT-PCR step was done by Luhui Shen.
Figure 3. 3: RT-PCR Analysis of MMTV Promotor and PDL1 Expression in
Established Tumors and Adjacent Normal Mammary Glands in No-Treated
FVB-Neun Mice
RNA analysis on different stages of established tumors and mammary
glands. MMTV is the promotor of Rat Her2 gene in FVB-NeuN mouse. Both
PDL1 and rat Her2 were at similar expression levels in different stages of the
established tumors. PDL1 is express high in normal mg and decrease in tumor.
82
3.3.4 Changes of Immunosignatures After PD-L1 Treatment
The long term protection of the short term early treatment suggested that
the treatment elicited an efficient anti-tumor immune response. The immune
checkpoint blockades can induce unique immunologic changes in cancer patients,
which may be related to the anti-tumor immune response [227]. It is possible
these changes are reflected in the antibody profiles. We found a specific
immunosignature of 321 peptides that can distinguish the treated mice 28 days
after the treatments with both the age matched untreated mice and their own
signature before the treatment. The 321 peptides gave a separated clustering
between the treated mice 28 days after the treatments PD-L1 and all samples
without PD-L1 treatment at three time points (Figure 3.4a). The same
immunosignature was also detected in the treated mice 28 days after treatment
and 35 days after their first palpable tumors, compared with non-treated mice at
age of 16 weeks (Figure 3.4b). In 46.4% selected peptides, their antibody
activities were significantly increased after the tumor established. Taken together,
this immunosignature suggests that the anti-tumor immune response was elicited
in the treated mice for long term protection.
83
Figure 3. 4: Heatmap Showing PD-L1 Generated Immunosignature
Sera were collected from 8 FVB neuN mice in the PD-L1 treatment group at 28
days after the treatment (AT_T28), and at 35 days after 1st tumor (AT_F35). 8
age-match transgenic controls without any treatment were also collected for sera
at corresponding time points ( NT_T28, NT_F35). Sera were collected from all
mice at age of 16 weeks (NT_Wk16) before any PD-L1 treatment. The
immunosignature of 321 peptides selected distinguished samples of AT_T28 and
NT_T28.
A) Heatmap of selected peptides depicting simultaneous analysis of samples at 28
days after the treatment (AT_T28) compared with samples without PD-L1
treatment at 3 time points (NT_Wk16, NT_T28, NT_F35).
B) Heatmap of selected peptides showing the immunosignatures at one time point
without treatment (NT_Wk16) and two following time points with treatment
(AT_T28, AT_F35)
A B
84
3.4 Methods
3.4.1 Mice and Sample Collection
Female FVB/NJ mice (n=10, age 6 weeks) were purchased from Jackson
Laboratories (Barharbor, ME). Female FVB/N neuN transgenic mice were kindly
gifted from Dr. Chella David at Mayo, Rochester. Transgenic FVB/N NeuN mice
and wild-type FVB/NJ mice were maintained and weaned in the same barrier
isolation housing with standard rodent feed and water. Mice were palpated every
week to detect mammary tumor growth. Blood samples were collected by
submandibular venipuncture using a 5.0 mm lancet (MEDIpoint, Inc. NY). Serum
was separated by serum separating Microtainer (BD Biosciences, San Jose, CA)
and stored at -20 C with aliquots. Mice were euthanized by 2,2,2-tribromoethanol
at a dose of 125mg/kg through intraperitoneal injection when tumor size exceeded
10% of the body weight. All murine experiments were conducted following
animal protocol reviewed and approved by Arizona State University Institutional
Animal Care and Use Committee.
3.4.2 Binding of Sera to the Peptide Microarrays
CIM10K Version 3 peptide arrays were printed as earlier described [110,
160]. CIM10K Version 3 library contained 10,000 random 20-residue peptide of
17 random sequence amino acids. Peptides were designed with random sequences
with nineteen amino acids (cysteine excluded), except for three residues at N-
terminal as a linker to attach peptides covalently on Nexterion A+ aminosilane
slides (Schott). Version 3 contained both L and D amino acids with Cys-Ser-Gly-
NH2 linker. Peptides were synthesized by Sigma Genosys (St. Louis, MO). Slides
85
were printed using the two-up format by piezoelectric non-contact printer at
Applied Microarrys (Tempe, AZ). Quality-control tests were performed using
subsamples of each batch with pooled human naïve serum to select high array-to-
array correlation and dynamic range. After being produced, arrays were stored in
containers, protected from light.
Slides were first washed by 33% isopropanol, 7.5% acetonitrile, and 0.5%
trifluoroacetic acid for five minutes, then dried by centrifuge at 800 rpm for five
minutes. After pre-wash, slide processing was automated by a TECAN HS4800-
Pro automated incubator, following a protocol optimized for antibody binding.
Briefly, arrays were blocking by 0.014% mercaptohexanol, 3% BSA, and 0.05%
Tween 20 in 1×PBS solution for 1 h at 23° C. Serum samples were then diluted at
1:500 in incubation buffer (3% BSA 0.05% Tween 20 in 1×PBS) and probed on
arrays for one hour at 37° C. Biotinylated anti-mouse IgG(H+L) antibodies
(Bethyl Laboratories, Montgomery, TX) were diluted in the incubation buffer at
5nM for one hour at 37° C. Bound secondary antibodies were visualized by Alexa
Fluor-647 labeled streptavidin (Invitrogen, CA) at 5nM for one hour at 37° C. The
injection volume of each step was 110 μl per array, and TBST solution was used
to wash the array between steps. The final wash used TBST and distilled water to
remove residual salt on slides. Correlation for technical replicates less than 0.85
were re-probed [175].
3.4.3 PD-L1 Treatment Regime
Stage I cancer signature of FVB/N-NeuN model was established in
Chapter 2. The age when the mice showed the stage I signature was 16 weeks. 10
86
mice were start immunosignature analysis every four weeks since 16 weeks old.
The mouse was confirmed with stage I cancer immunosignature in two continual
analysis was treated 200ug anti-mouse PD-L1 antibody in PBS every three days
through i.p. for total three treatments. Monoclonal anti mouse PD-L1 antibody
(10F.9G2) were purchased from BioXcell (Cat#: BE0101).
3.4.4 RNA Expression in Mammary Gland
All tissues and tumor were collected and storage in RNA-later (Ambion,
Cat#AM7020). Total RNA were purified with TRIzol (Life Tech. Cat#15596-018)
and PureLink RNA purification system (Life Tech. Cat#12183-018). Genomic
DNA were removed with PapidOut DNA Removal kit (Thermo Sci. Cat#K2981).
cDNAs were synthesized from 2ug of total RNA from each sample with cDNA
synthesis kit (Thermo Sci. Cat#K1641). PCR reactions were set up with PCR
SuperMix (Life Tech. Cat#10572-014) in 20ul reaction volume. The expression of
β-actin, PD-L1 and MMTV promotor (the promotor for Rat Her2 gene) were
analyzed.
PCR primers:
β-actin: 5’:CCTGTATGCCTCTGGTCG;3’: TGTCACGCACGATTTCC.
PDL1:5’:TTCACAGCCTGCTGTCACTT; 3’:GGGCATTGACTTTCAGCGTG.
MMTV: 5’: ATCGGTGATGTCGGCGATAT; 3’:GTAACACAGGCAGATGT
AGGA.
PCR condition: β-actin: 95℃ 2’, (95℃ 30”, 60℃ 30”, 72℃ 30”) X 25
cycles, 72℃ 5’. PDL1: 95℃ 2’, (95℃ 30”, 65℃ 30”, 72℃ 30”) X 27 cycles, 72℃
5’. CMV: 95℃ 2’ , (95℃ 30”, 60℃ 30”, 72℃ 30”) X 30 cycles, 72℃ 5’.
87
3.4.5 Statistical Analysis
After the final wash, slides were scanned by an Agilent C scanner at
633nm emission (Agilent Technologies, CA) with 100% laser power and 100%
PMT. Images were extracted by GenePix Pro 6.0 (Molecular Devices, Sunnyvale,
CA) and saved as gpr files. Poor-quality peptide spots were excluded by flagging
them as “absent” by visual inspection. Samples with Pearson correlation
coefficient >0.85 among technical replicates were used. Raw data from gpr files
were collected and imported into R for later processing (Agilent Technologies,
Palo Alto, CA). Data were normalized to the 50th percentile to minimize array-
to-array variation. Batch effects were adjusted by SVA algorithm with irw as
adjustment method with 10 iterations. The biological group was used as a
covariate in SVA. Statistical analyses were done using R version 3.1.1, and JMP
version 11. P-values were corrected for multiple testing by the Benjamin-
Hochberg false discovery rate method. The Welch t test was used in R to select
significant peptides with greater than 1.3-fold changes and p<0.05 between two
groups. We used a support vector machine with settings including polynomial dot
order at 1 and scaling order at 0 for cross-validation and prediction. Packages
used in R were sva, caret, e1071 [216, 217].
3.5 Discussion
We had shown that transgenic mice display an immunosignature of having
a tumor at least 12 weeks before the tumor is palpable. Short treatment at this
early stage with i.p. injection of anti-PD-L1 retarded the tumor growth compared
to the untreated mice. This early treatment has long term protection, which can
88
slow the tumor growth four months after the treatment with similar efficiency in
the mice that develop tumor late. High PD-L1 gene expression was evident in the
transforming initiated mammary glands, consistent with the effects of anti-PD-L1
treatment at early stages. The early treatment elicited a specific antibody profile
which can be detected one month after the treatment and the first palpable tumor
established in the treated mice.
It was not obvious that early treatment of a tumor would be effective. It
would depend first on the tumor actually producing PD-L1 at this early stage and
second, that there is a potential anti-tumor immune response at the same time that
was being inhibited by the PD1/PD-L1 immune checkpoint pathway. In other
studies, higher mRNA expression of PD-L1 in the diagnosed tumor were
observed in the patient’s tissue and also in multiple breast cancer cell lines [204,
228-232]. Triple-negative breast cancer (TNBC), basal type breast cancer cells
expressed higher levels of PD-L1 constitutively [228, 232]. Although subtypes of
breast cancer can influence clinical treatment strategy, timing is also an important
factor when giving treatments. However, few study investigate PD-L1 mRNA
expression level at prediagnosed stage and its relationship between anti-PD-L1
treatment effects. Here we demonstrated that PD-L1 was highly expressed in the
early stage of the tumor when transformation was just initiated.
The immune checkpoint blockade treatment in cancer patients could
induce unique systemic immunologic changes [227], which are possible to elicit a
long term anti-tumor immune response against tumor developing in different
location. In the late stage cancer, increasing number of tumor-infiltrating
89
lymphocytes (TIL) in tumors were reported to be significantly correlated with
anti-PD-L1 treatment [198, 199, 229]. Therefore, we hypothesize that for
protection, one underline mechanism could be anti-PD-L1 treatment can also
boost the immune system even at early stage by increasing tumor infiltration.
The long term protection of the early treatment in this experiment
indicates the anti-tumor immune responses were elicited and boosted by the
treatment. This is supported by a specific and consistent immunosignature of the
treated mice in both one month after the treatment and the first palpable tumor
was established, comparing to the non-treated mice with matched age and tumor
status. Also in multiple advanced human cancers, BMS-936559, an anti-PD-L1
antibody drug, showed long term protection against cancer with a response rate of
6-17% [198]. By using breast cancer [233], melanoma, lung and colon mouse
models [234] injected with cancer cells, it was shown that anti-PD-L1 treatment
can provide a protection against tumors after the first injection and rechallenges of
cancer cells.
Taken together, this suggests that there are cancer antigens at an early
stage, which can induce the specific anti-tumor immune response and anti-PDL1
treatment is more beneficial at the early stage of tumor development.
In humans, generally a single primary tumor develops that can metastasize.
This contrasts to the mouse mammary tumor model in this study, where multiple
tumors can arise as all the cells in the mammary gland can overexpress the
oncogenic rat Her2. Also in contrast, therapeutic treatment in humans with
checkpoint inhibitors to date, has involved treatment up to 111 weeks [235]. We
90
choose to only treat 3 times after early diagnosis to determine if even a brief
intervention was beneficial. As shown in Figures 3 the early treatment had a
significant inhibition on tumor growth late. We suspect that in the case of natural
cases of tumors, the early treatment would be more efficacious. It still may
require a longer treatment regime than we tested to get better tumor inhibition.
We note that the treatment was not started until the diagnosis was confirmed 4
weeks later. While a second monitoring is likely for human early detection and
treatment scenarios, in the mouse model this would allow for significant tumor
development. It should improve the protection if we treat the mice earlier.
One thing worth noticed is that the average time of first palpable tumor in
anti-PD-L1 treatment group was at 30.8 weeks, 35.8 weeks for three mice in non-
treatment group, and 33.8 weeks for all non-treatment mice we recorded. There
was no significant difference between those three groups. This difference of
average time of first palatable tumor by anti-PD-L1 treatment may be due to high
variance in small sample size or may be caused by increasing TIL in tumor tissue
resulting in increasing tumor size. Topalian et al. shown that in cancer patients
after anti-PD-1 treatment the observed tumor size first increased after 2 months
treatment then decreased at 4 months [198, 199].
A combination of immune checkpoint blockades therapeutics has been
tested in different mouse tumor models and shown better responses than the
mono-blockade treatment [236, 237]. The clinical trial of combined nivolumab
and ipilimumab treatment on human also showed better response than the mono-
treatment with a manageable safety profile [238]. The combined treatment
91
induced distinct immunologic changes in the patients [227]. Combination therapy
may also improve the outcome for the cancer early treatment. Future studies will
explore the effect of earlier, longer and combined treatment regimes.
We propose that the work described here could be applied as a general
plan for effectively eradicating cancer. People would regularly send in a blood
sample to be screened for the early signature of cancer. If a positive was detected
the person would be rescreened soon after to confirm the diagnosis and assess the
vector of the signature. A second positive diagnosis may be sufficient to initiate
treatment. This decision could depend on factors discussed below. The response
to the treatment could also be monitored by immunosignature. We believe the
principle of all the elements of this scenario have been demonstrated in this work.
One major concern relative to this scenario is the false positive rate. If a
tumor was misidentified, or more likely, was a tumor that would self-resolve or
remain indolent, the concern is that false positive treatments would lead to
unnecessary costs and increasing risk of side effects. We believe that if the
immunosignature protocol was both inexpensive and simple, it could substantially
relieve the cost of false positives through retesting and assessment of the vector of
change in tumor signatures. Studies to date indicate the immunosignature
diagnostic can be both simple and inexpensive [225].
The other consideration is the cost and side-effects of the treatments. If
the systemic treatment with immune checkpoint inhibitors, or other systemic
treatments, had little side effects, then it would not be a risk to treat positives,
even though the eradicated tumor was benign or otherwise not life-threatening.
92
One, as yet to be tested possibility, is that treatment at an early stage would
decrease the time of treatment and amount of therapeutic required, therefore
reducing the risk and cost of treatment. Checkpoint inhibitor trials to date have
been in cases of advanced tumors and have had significant occurrences of adverse
effects [226].
Cost of treatment is a significant concern for the model depicted in Figure
6. Currently, the cost of a course of treatment with anti-PD-1 is reported to be
over $12,500 a month [239]. If such a treatment were to be administered to kill
any tumor arising the cost issue could be of major concern. There are ~1.7M new
cancer cases/year in the US. At current checkpoint therapeutic treatment costs, it
would require ~$200B/yr to treat all cancers. This is still less than the estimated
current cost of treating cancer. Worldwide there are ~14M new cases/yr. ~70%
of the new cases are in the developing world. Treatment of these cases at current
costs would incur ~$1T/year which is approximately the current cost per year for
cancer effects estimated in this sector (ACS/Livestrong report). Even though
current costs of treatment might be justified, there is a clear incentive to develop
less costly equivalent therapeutics. We also point out that any drug, not only
checkpoint inhibitors, that could be administered systemically and be effective
could be used in this scenario.
93
REFERENCES
1. Society, A.C., Cancer facts and figures 2015. 2015(American Cancer
Society).
2. Hanahan, D. and R.A. Weinberg, The hallmarks of cancer. Cell, 2000.
100(1): p. 57-70.
3. Hanahan, D. and R.A. Weinberg, Hallmarks of cancer: the next
generation. Cell, 2011. 144(5): p. 646-74.
4. Stewart, B.W., Wild, C. P., World Cancer Report 2014. 2014(IARC
Nonserial Publication).
5. Howlader N, N.A., Krapcho M, Garshell J, Miller D, Altekruse SF,
Kosary CL, Yu M, Ruhl J, Tatalovich Z,Mariotto A, Lewis DR, Chen HS,
Feuer EJ, Cronin KA (eds), SEER Cancer Statistics Review, 1975-2012,
National Cancer Institute. 2015.
6. Society, A.C., Breast Cancer. 2014(American Cancer Society).
7. Borowsky, A., Special considerations in mouse models of breast cancer.
Breast Dis, 2007. 28: p. 29-38.
8. Green, J.E. and T. Hudson, The promise of genetically engineered mice for
cancer prevention studies. Nat Rev Cancer, 2005. 5(3): p. 184-98.
9. Herschkowitz, J.I., et al., Identification of conserved gene expression
features between murine mammary carcinoma models and human breast
tumors. Genome Biol, 2007. 8(5): p. R76.
10. Allred, D.C. and D. Medina, The relevance of mouse models to
understanding the development and progression of human breast cancer. J
Mammary Gland Biol Neoplasia, 2008. 13(3): p. 279-88.
11. Lu, H., et al., The tumor antigen repertoire identified in tumor-bearing
neu transgenic mice predicts human tumor antigens. Cancer Res, 2006.
66(19): p. 9754-61.
94
12. Ursini-Siegel, J., et al., Insights from transgenic mouse models of ERBB2-
induced breast cancer. Nat Rev Cancer, 2007. 7(5): p. 389-97.
13. Guy, C.T., et al., Expression of the neu protooncogene in the mammary
epithelium of transgenic mice induces metastatic disease. Proc Natl Acad
Sci U S A, 1992. 89(22): p. 10578-82.
14. Dakappagari, N.K., et al., Prevention of mammary tumors with a chimeric
HER-2 B-cell epitope peptide vaccine. Cancer Res, 2000. 60(14): p. 3782-
9.
15. Pupa, S.M., et al., Prevention of spontaneous neu-expressing mammary
tumor development in mice transgenic for rat proto-neu by DNA
vaccination. Gene Ther, 2001. 8(1): p. 75-9.
16. Park, J.M., et al., Therapy of advanced established murine breast cancer
with a recombinant adenoviral ErbB-2/neu vaccine. Cancer Res, 2008.
68(6): p. 1979-87.
17. Amici, A., F.M. Venanzi, and A. Concetti, Genetic immunization against
neu/erbB2 transgenic breast cancer. Cancer Immunol Immunother, 1998.
47(4): p. 183-90.
18. Venanzi, F.M., et al., Co-expression of Flt-3 ligand gene ablates tumor
immunity elicited by HER-2/neu DNA vaccine in transgenic mice. Vaccine,
2010. 28(22): p. 3841-7.
19. Mao, J., et al., Mining the pre-diagnostic antibody repertoire of TgMMTV-
neu mice to identify autoantibodies useful for the early detection of human
breast cancer. J Transl Med, 2014. 12(1): p. 121.
20. Abe, F., et al., Therapeutic activity of sunitinib for Her2/neu induced
mammary cancer in FVB mice. Int Immunopharmacol, 2010. 10(1): p.
140-5.
21. Pegram, M. and D. Ngo, Application and potential limitations of animal
models utilized in the development of trastuzumab (Herceptin): a case
study. Adv Drug Deliv Rev, 2006. 58(5-6): p. 723-34.
95
22. Leach, D.R., M.F. Krummel, and J.P. Allison, Enhancement of antitumor
immunity by CTLA-4 blockade. Science, 1996. 271(5256): p. 1734-6.
23. Boggio, K., et al., Interleukin 12-mediated prevention of spontaneous
mammary adenocarcinomas in two lines of Her-2/neu transgenic mice. J
Exp Med, 1998. 188(3): p. 589-96.
24. Bouchard, L., et al., Stochastic appearance of mammary tumors in
transgenic mice carrying the MMTV/c-neu oncogene. Cell, 1989. 57(6): p.
931-6.
25. Stancovski, I., et al., Mechanistic aspects of the opposing effects of
monoclonal antibodies to the ERBB2 receptor on tumor growth. Proc Natl
Acad Sci U S A, 1991. 88(19): p. 8691-5.
26. Lin, E.Y., et al., Progression to malignancy in the polyoma middle T
oncoprotein mouse breast cancer model provides a reliable model for
human diseases. American Journal of Pathology, 2003. 163(5): p. 2113-
2126.
27. Kelly-Spratt, K.S., et al., A mouse model repository for cancer biomarker
discovery. J Proteome Res, 2008. 7(8): p. 3613-8.
28. Yahalom, G., et al., An Antibody-based Blood Test Utilizing a Panel of
Biomarkers as a New Method for Improved Breast Cancer Diagnosis.
Biomark Cancer, 2013. 5: p. 71-80.
29. Force, U.S.P.S.T., Screening for breast cancer: U.S. Preventive Services
Task Force recommendation statement. Ann Intern Med, 2009. 151(10): p.
716-26, W-236.
30. Smith-Bindman, R., et al., Physician predictors of mammographic
accuracy. J Natl Cancer Inst, 2005. 97(5): p. 358-67.
31. Hendrick, R.E. and M.A. Helvie, United States Preventive Services Task
Force screening mammography recommendations: science ignored. AJR
Am J Roentgenol, 2011. 196(2): p. W112-6.
96
32. Elmore, J.G., et al., Ten-year risk of false positive screening mammograms
and clinical breast examinations. N Engl J Med, 1998. 338(16): p. 1089-
96.
33. Esserman, L., Y. Shieh, and I. Thompson, Rethinking Screening for Breast
Cancer and Prostate Cancer. Jama-Journal of the American Medical
Association, 2009. 302(15): p. 1685-1692.
34. Zahl, P.H., B.H. Strand, and J. Maehlen, Incidence of breast cancer in
Norway and Sweden during introduction of nationwide screening:
prospective cohort study. BMJ, 2004. 328(7445): p. 921-4.
35. Dogan, B.E., et al., Multimodality Imaging of Triple Receptor-Negative
Tumors With Mammography, Ultrasound, and MRI. American Journal of
Roentgenology, 2010. 194(4): p. 1160-1166.
36. Dogan, B.E. and L.W. Turnbull, Imaging of triple-negative breast cancer.
Annals of Oncology, 2012. 23: p. 23-29.
37. Dent, R., et al., Triple-negative breast cancer: Clinical features and
patterns of recurrence. Clinical Cancer Research, 2007. 13(15): p. 4429-
4434.
38. Foulkes, W.D., I.E. Smith, and J.S. Reis, Triple-Negative Breast Cancer.
New England Journal of Medicine, 2010. 363(20): p. 1938-1948.
39. USPSTF, Screening for breast cancer: U.S. Preventive Services Task
Force recommendation statement. Ann Intern Med, 2009. 151(10): p. 716-
26, w-236.
40. Hartmann, L.C., et al., Benign breast disease and the risk of breast cancer.
N Engl J Med, 2005. 353(3): p. 229-37.
41. Benson, J.R., et al., Early breast cancer. Lancet, 2009. 373(9673): p.
1463-79.
42. Guray, M. and A.A. Sahin, Benign breast diseases: classification,
diagnosis, and management. Oncologist, 2006. 11(5): p. 435-49.
97
43. Pearlman, M.D. and J.L. Griffin, Benign breast disease. Obstet Gynecol,
2010. 116(3): p. 747-58.
44. Dupont, W.D., et al., Breast cancer risk associated with proliferative
breast disease and atypical hyperplasia. Cancer, 1993. 71(4): p. 1258-65.
45. Moss, S.M., et al., Effect of mammographic screening from age 40 years
on breast cancer mortality at 10 years' follow-up: a randomised
controlled trial. Lancet, 2006. 368(9552): p. 2053-60.
46. Linda, A., et al., Outcome of initially only magnetic resonance
mammography-detected findings with and without correlate at second-
look sonography: distribution according to patient history of breast
cancer and lesion size. Breast, 2008. 17(1): p. 51-7.
47. Berg, W.A., et al., Detection of breast cancer with addition of annual
screening ultrasound or a single screening MRI to mammography in
women with elevated breast cancer risk. JAMA, 2012. 307(13): p. 1394-
404.
48. Harris, L., et al., American Society of Clinical Oncology 2007 update of
recommendations for the use of tumor markers in breast cancer. J Clin
Oncol, 2007. 25(33): p. 5287-312.
49. Cronin, M., et al., Measurement of gene expression in archival paraffin-
embedded tissues: development and performance of a 92-gene reverse
transcriptase-polymerase chain reaction assay. Am J Pathol, 2004. 164(1):
p. 35-42.
50. van 't Veer, L.J., et al., Gene expression profiling predicts clinical
outcome of breast cancer. Nature, 2002. 415(6871): p. 530-6.
51. Ma, X.J., et al., A two-gene expression ratio predicts clinical outcome in
breast cancer patients treated with tamoxifen. Cancer Cell, 2004. 5(6): p.
607-16.
52. Esserman, L.J., et al., A role for biomarkers in the screening and diagnosis
of breast cancer in younger women. Expert Review of Molecular
Diagnostics, 2007. 7(5): p. 533-544.
98
53. Tamimi, R.M., et al., Endogenous hormone levels, mammographic density,
and subsequent risk of breast cancer in postmenopausal women. Journal
of the National Cancer Institute, 2007. 99(15): p. 1178-1187.
54. Jemal, A., et al., Cancer statistics, 2009. CA Cancer J Clin, 2009. 59(4): p.
225-49.
55. Takikita, M., et al., Biomarkers of apoptosis and survival in esophageal
squamous cell carcinoma. BMC Cancer, 2009. 9: p. 310.
56. Bar-Joseph, Z., A. Gitter, and I. Simon, Studying and modelling dynamic
biological processes using time-series gene expression data. Nat Rev
Genet, 2012. 13(8): p. 552-64.
57. U.S. Census Bureau, P.D., Population Estimates. U.S. Census Bureau,
2015.
58. Martin, A., et al., Prognostic value of pre-operative serum CA 15.3 levels
in breast cancer. Anticancer Res, 2006. 26(5B): p. 3965-71.
59. Khatcheressian, J.L., et al., American Society of Clinical Oncology 2006
update of the breast cancer follow-up and management guidelines in the
adjuvant setting. J Clin Oncol, 2006. 24(31): p. 5091-7.
60. Slamon, D.J., et al., Human breast cancer: correlation of relapse and
survival with amplification of the HER-2/neu oncogene. Science, 1987.
235(4785): p. 177-82.
61. Lipton, A., et al., Elevated serum Her-2/neu level predicts decreased
response to hormone therapy in metastatic breast cancer. J Clin Oncol,
2002. 20(6): p. 1467-72.
62. Lipton, A., et al., Serum HER-2/neu and response to the aromatase
inhibitor letrozole versus tamoxifen. J Clin Oncol, 2003. 21(10): p. 1967-
72.
99
63. Mostert, B., et al., Circulating tumor cells (CTCs): Detection methods and
their clinical relevance in breast cancer. Cancer Treatment Reviews, 2009.
35(5): p. 463-474.
64. Hong, B. and Y.L. Zu, Detecting Circulating Tumor Cells: Current
Challenges and New Trends. Theranostics, 2013. 3(6): p. 377-394.
65. Cristofanilli, M., et al., Circulating tumor cells, disease progression, and
survival in metastatic breast cancer. N Engl J Med, 2004. 351(8): p. 781-
91.
66. Weigelt, B., et al., Marker genes for circulating tumour cells predict
survival in metastasized breast cancer patients. Br J Cancer, 2003. 88(7):
p. 1091-4.
67. Demel, U., et al., Detection of tumour cells in the peripheral blood of
patients with breast cancer. Development of a new sensitive and specific
immunomolecular assay. J Exp Clin Cancer Res, 2004. 23(3): p. 465-8.
68. Zhao, H., et al., A pilot study of circulating miRNAs as potential
biomarkers of early stage breast cancer. PLoS One, 2010. 5(10): p.
e13735.
69. Heneghan, H.M., et al., Circulating microRNAs as novel minimally
invasive biomarkers for breast cancer. Ann Surg, 2010. 251(3): p. 499-
505.
70. Grady, W.M. and M. Tewari, The next thing in prognostic molecular
markers: microRNA signatures of cancer. Gut, 2010. 59(6): p. 706-8.
71. Mitchell, P.S., et al., Circulating microRNAs as stable blood-based
markers for cancer detection. Proc Natl Acad Sci U S A, 2008. 105(30): p.
10513-8.
72. Abd Hamid, U.M., et al., A strategy to reveal potential glycan markers
from serum glycoproteins associated with breast cancer progression.
Glycobiology, 2008. 18(12): p. 1105-18.
100
73. Pepe, M.S., et al., Phases of biomarker development for early detection of
cancer. J Natl Cancer Inst, 2001. 93(14): p. 1054-61.
74. Chung, L. and R.C. Baxter, Breast cancer biomarkers: proteomic
discovery and translation to clinically relevant assays. Expert Rev
Proteomics, 2012. 9(6): p. 599-614.
75. Kern, S.E., Why your new cancer biomarker may never work: recurrent
patterns and remarkable diversity in biomarker failures. Cancer Res, 2012.
72(23): p. 6097-101.
76. Issaq, H.J., T.J. Waybright, and T.D. Veenstra, Cancer biomarker
discovery: Opportunities and pitfalls in analytical methods.
Electrophoresis, 2011. 32(9): p. 967-75.
77. Marchiò, C., M. Dowsett, and J.S. Reis-Filho, Revisiting the technical
validation of tumour biomarker assays: how to open a Pandora's box.
BMC Med, 2011. 9: p. 41.
78. Anderson, G.L., et al., Assessing lead time of selected ovarian cancer
biomarkers: a nested case-control study. J Natl Cancer Inst, 2010. 102(1):
p. 26-38.
79. Zhu, C.S., et al., A framework for evaluating biomarkers for early
detection: validation of biomarker panels for ovarian cancer. Cancer Prev
Res (Phila), 2011. 4(3): p. 375-83.
80. Opstal-van Winden, A.W., et al., A bead-based multiplexed immunoassay
to evaluate breast cancer biomarkers for early detection in pre-diagnostic
serum. Int J Mol Sci, 2012. 13(10): p. 13587-604.
81. Lu, H., et al., Evaluation of known oncoantibodies, HER2, p53, and cyclin
B1, in prediagnostic breast cancer sera. Cancer Prev Res (Phila), 2012.
5(8): p. 1036-43.
82. P, E., Ueber den jetzigen stand der karzinomforschung (About the current
state of the art of cancer research). Ned Tijdschr Geneeskd, 1909(5): p.
18.
101
83. Burnet, F.M., Immunological surveillance in neoplasia. Transplant Rev,
1971. 7: p. 3-25.
84. Thomas, L., On immunosurveillance in human cancer. Yale J Biol Med,
1982. 55(3-4): p. 329-33.
85. Stutman, O., Tumor development after 3-methylcholanthrene in
immunologically deficient athymic-nude mice. Science, 1974. 183(4124):
p. 534-6.
86. Dunn, G.P., L.J. Old, and R.D. Schreiber, The three Es of cancer
immunoediting. Annu Rev Immunol, 2004. 22: p. 329-60.
87. Kim, R., M. Emi, and K. Tanabe, Cancer immunoediting from immune
surveillance to immune escape. Immunology, 2007. 121(1): p. 1-14.
88. Lernmark, A., Immune surveillance: paraneoplastic or environmental
triggers of autoimmunity. Crit Rev Immunol, 1997. 17(5-6): p. 437-47.
89. Darnell, R.B. and L.M. DeAngelis, Regression of small-cell lung
carcinoma in patients with paraneoplastic neuronal antibodies. Lancet,
1993. 341(8836): p. 21-2.
90. Dalmau, J. and F. Graus, Paraneoplastic syndromes: from remote to clear
and present knowledge. Brain, 2012. 135(5): p. 1650-1653.
91. Anderson, K.S. and J. LaBaer, The sentinel within: exploiting the immune
system for cancer biomarkers. J Proteome Res, 2005. 4(4): p. 1123-33.
92. Xu, Y.W., et al., Autoantibodies as Potential Biomarkers for the Early
Detection of Esophageal Squamous Cell Carcinoma. American Journal of
Gastroenterology, 2014. 109(1): p. 36-45.
93. Wang, X., et al., Autoantibody signatures in prostate cancer. N Engl J
Med, 2005. 353(12): p. 1224-35.
102
94. Tabernero, M.D., L.L. Lv, and K.S. Anderson, Autoantibody profiles as
biomarkers of breast cancer. Cancer Biomarkers, 2010. 6(5-6): p. 247-256.
95. Anderson, K.S., et al., Protein microarray signature of autoantibody
biomarkers for the early detection of breast cancer. J Proteome Res, 2011.
10(1): p. 85-96.
96. Stockert, E., et al., A survey of the humoral immune response of cancer
patients to a panel of human tumor antigens. J Exp Med, 1998. 187(8): p.
1349-54.
97. Mahmoud, S.M., et al., The prognostic significance of B lymphocytes in
invasive carcinoma of the breast. Breast Cancer Res Treat, 2012. 132(2): p.
545-53.
98. Li, Y., et al., p53 autoantibodies predict subsequent development of
cancer. Int J Cancer, 2005. 114(1): p. 157-60.
99. Goodell, V., et al., Level of HER-2/neu protein expression in breast cancer
may affect the development of endogenous HER-2/neu-specific immunity.
Mol Cancer Ther, 2008. 7(3): p. 449-54.
100. Hori, S.S. and S.S. Gambhir, Mathematical model identifies blood
biomarker-based early cancer detection strategies and limitations. Sci
Transl Med, 2011. 3(109): p. 109ra116.
101. Anderson, L. and C.L. Hunter, Quantitative mass spectrometric multiple
reaction monitoring assays for major plasma proteins. Molecular &
Cellular Proteomics, 2006. 5(4): p. 573-588.
102. Lutz, A.M., et al., Cancer screening: a mathematical model relating
secreted blood biomarker levels to tumor sizes. PLoS Med, 2008. 5(8): p.
e170.
103. Brennan, D.J., et al., Antibody-based proteomics: fast-tracking molecular
diagnostics in oncology. Nat Rev Cancer, 2010. 10(9): p. 605-17.
103
104. Cenci, S. and R. Sitia, Managing and exploiting stress in the antibody
factory. FEBS Lett, 2007. 581(19): p. 3652-7.
105. Sulzer, B., et al., Memory in idiotypic networks due to competition
between proliferation and differentiation. Bull Math Biol, 1993. 55(6): p.
1133-82.
106. Cooperman, J., et al., Cell division rates of primary human precursor B
cells in culture reflect in vivo rates. Stem Cells, 2004. 22(6): p. 1111-20.
107. Forster, I. and K. Rajewsky, The bulk of the peripheral B-cell pool in mice
is stable and not rapidly renewed from the bone marrow. Proc Natl Acad
Sci U S A, 1990. 87(12): p. 4781-4.
108. Hao, Z. and K. Rajewsky, Homeostasis of peripheral B cells in the
absence of B cell influx from the bone marrow. J Exp Med, 2001. 194(8):
p. 1151-64.
109. Halperin, R., Characterization and analysis of a novel platform for
profiling the antibody response. 2011, Arizona State University: Tempe,
Arizona. p. 290.
110. Stafford, P., et al., Physical characterization of the "immunosignaturing
effect". Mol Cell Proteomics, 2012. 11(4): p. M111 011593.
111. Usami, A., et al., The effect of pH, hydrogen peroxide and temperature on
the stability of human monoclonal antibody. J Pharm Biomed Anal, 1996.
14(8-10): p. 1133-40.
112. Geijersstam, V., et al., Stability over time of serum antibody levels to
human papillomavirus type 16. The Journal of Infectious Diseases, 1998.
177(6): p. 1710-1714.
113. Chase, B.A., S.A. Johnston, and J.B. Legutki, Evaluation of biological
sample preparation for immunosignature-based diagnostics. Clin Vaccine
Immunol, 2012. 19(3): p. 352-8.
104
114. af Geijersstam, V., et al., Stability over time of serum antibody levels to
human papillomavirus type 16. J Infect Dis, 1998. 177(6): p. 1710-4.
115. Disis, M.L., et al., High-titer HER-2/neu protein-specific antibody can be
detected in patients with early-stage breast cancer. Journal of Clinical
Oncology, 1997. 15(11): p. 3363-3367.
116. Chapman, C., et al., Autoantibodies in breast cancer: their use as an aid to
early diagnosis. Annals of Oncology, 2007. 18(5): p. 868-873.
117. Desmetz, C., et al., Identification of a new panel of serum autoantibodies
associated with the presence of in situ carcinoma of the breast in younger
women. Clin Cancer Res, 2009. 15(14): p. 4733-41.
118. Lacombe, J., et al., Identification and validation of new autoantibodies for
the diagnosis of DCIS and node negative early-stage breast cancers. Int J
Cancer, 2013. 132(5): p. 1105-13.
119. Wang, L.F. and M. Yu, Epitope identification and discovery using phage
display libraries: applications in vaccine development and diagnostics.
Curr Drug Targets, 2004. 5(1): p. 1-15.
120. Luna Coronell, J.A., et al., The current status of cancer biomarker
research using tumour-associated antigens for minimal invasive and early
cancer diagnostics. J Proteomics, 2012. 76 Spec No.: p. 102-15.
121. Chen, Y.T., A.O. Gure, and M.J. Scanlan, Serological analysis of
expression cDNA libraries (SEREX): an immunoscreening technique for
identifying immunogenic tumor antigens. Methods Mol Med, 2005. 103: p.
207-16.
122. Scanlan, M.J., et al., Cancer/testis antigens: an expanding family of
targets for cancer immunotherapy. Immunological Reviews, 2002. 188: p.
22-32.
123. Zhou, S., et al., Mapping the high throughput SEREX technology
screening for novel tumor antigens. Comb Chem High Throughput Screen,
2012. 15(3): p. 202-15.
105
124. Jager, D., et al., Identification of a tissue-specific putative transcription
factor in breast tissue by serological screening of a breast cancer library.
Cancer Research, 2001. 61(5): p. 2055-2061.
125. Jager, D., et al., Identification of a tissue-specific putative transcription
factor in breast tissue by serological screening of a breast cancer library.
Cancer Res, 2001. 61(5): p. 2055-61.
126. Scanlan, M.J., et al., Characterization of human colon cancer antigens
recognized by autologous antibodies. International Journal of Cancer,
1998. 76(5): p. 652-658.
127. Sahin, U., et al., HUMAN NEOPLASMS ELICIT MULTIPLE SPECIFIC
IMMUNE-RESPONSES IN THE AUTOLOGOUS HOST. Proceedings of
the National Academy of Sciences of the United States of America, 1995.
92(25): p. 11810-11813.
128. Chen, Y.T., et al., A testicular antigen aberrantly expressed in human
cancers detected by autologous antibody screening. Proceedings of the
National Academy of Sciences of the United States of America, 1997.
94(5): p. 1914-1918.
129. Fossa, A., et al., Serological cloning of cancer/testis antigens expressed in
prostate cancer using cDNA phage surface display. Cancer Immunol
Immunother, 2004. 53(5): p. 431-8.
130. Legutki, J.B., et al., Scalable high-density peptide arrays for
comprehensive health monitoring. Nat Commun, 2014. 5: p. 4785.
131. Falsey, J.R., et al., Peptide and small molecule microarray for high
throughput cell adhesion and functional assays. Bioconjug Chem, 2001.
12(3): p. 346-53.
132. Shreffler, W.G., et al., IgE and IgG4 epitope mapping by microarray
immunoassay reveals the diversity of immune response to the peanut
allergen, Ara h 2. J Allergy Clin Immunol, 2005. 116(4): p. 893-9.
133. Fu, J., J. Reinhold, and N.W. Woodbury, Peptide-modified surfaces for
enzyme immobilization. PLoS One, 2011. 6(4): p. e18692.
106
134. Fu, J., et al., Exploring peptide space for enzyme modulators. J Am Chem
Soc, 2010. 132(18): p. 6419-24.
135. Diehnelt, C.W., et al., Discovery of high-affinity protein binding ligands--
backwards. PLoS One, 2010. 5(5): p. e10728.
136. Morales Betanzos, C., et al., Bacterial glycoprofiling by using random
sequence peptide microarrays. Chembiochem, 2009. 10(5): p. 877-88.
137. Foong, Y.M., et al., Current advances in peptide and small molecule
microarray technologies. Curr Opin Chem Biol, 2012. 16(1-2): p. 234-42.
138. Boltz, K.W., et al., Peptide microarrays for carbohydrate recognition.
Analyst, 2009. 134(4): p. 650-2.
139. Cretich, M., et al., Epitope mapping of human chromogranin A by peptide
microarrays. Methods Mol Biol, 2009. 570: p. 221-32.
140. Reddy, M.M., et al., Identification of candidate IgG biomarkers for
Alzheimer's disease via combinatorial library screening. Cell, 2011.
144(1): p. 132-42.
141. Reddy, M.M. and T. Kodadek, Protein "fingerprinting" in complex
mixtures with peptoid microarrays. Proc Natl Acad Sci U S A, 2005.
102(36): p. 12672-7.
142. Anderson, K.S. and J. LaBaer, The sentinel within: Exploiting the immune
system for cancer biomarkers. Journal of Proteome Research, 2005. 4(4):
p. 1123-1133.
143. Seliger, B. and R. Kellner, Design of proteome-based studies in
combination with serology for the identification of biomarkers and novel
targets. Proteomics, 2002. 2(12): p. 1641-51.
144. Ekins, R.P., Multi-analyte immunoassay. Journal of pharmaceutical and
biomedical analysis, 1989. 7(2): p. 155-68.
107
145. MacBeath, G. and S.L. Schreiber, Printing proteins as microarrays for
high-throughput function determination. Science, 2000. 289(5485): p.
1760-1763.
146. Zhu, H., et al., Global analysis of protein activities using proteome chips.
Science, 2001. 293(5537): p. 2101-2105.
147. BB., H., Multiplexed protein analysis using antibody microarrays and
label-based detection. Methods Mol Med, 2005. 114: p. 183-194.
148. Ramachandran, N., et al., Self-assembling protein microarrays. Science,
2004. 305(5680): p. 86-90.
149. Ramachandran, N., et al., Next-generation high-density self-assembling
functional protein arrays. Nature Methods, 2008. 5(6): p. 535-538.
150. Hurst, R., et al., Protein-protein interaction studies on protein arrays:
Effect of detection strategies on signal-to-background ratios. Analytical
Biochemistry, 2009. 392(1): p. 45-53.
151. Yang, L.N., et al., Protein microarrays for systems biology. Acta
Biochimica Et Biophysica Sinica, 2011. 43(3): p. 161-171.
152. Kijanka, G. and D. Murphy, Protein arrays as tools for serum
autoantibody marker discovery in cancer. Journal of Proteomics, 2009.
72(6): p. 936-944.
153. Ramachandran, N., S. Srivastava, and J. LaBaer, Applications of protein
microarrays for biomarker discovery. Proteomics Clinical Applications,
2008. 2(10-11): p. 1444-1459.
154. Anderson, K.S., et al., Protein Microarray Signature of Autoantibody
Biomarkers for the Early Detection of Breast Cancer. Journal of Proteome
Research, 2011. 10(1): p. 85-96.
155. Nam, M.J., et al., Molecular profiling of the immune response in colon
cancer using protein microarrays: Occurrence of autoantibodies to
ubiquitin C-terminal hydrolase L3. Proteomics, 2003. 3(11): p. 2108-2115.
108
156. Qiu, J., et al., Development of natural protein microarrays for diagnosing
cancer based on an antibody response to tumor antigens. Journal of
Proteome Research, 2004. 3(2): p. 261-267.
157. Hueber, W., et al., Autoantibody profiling for the study and treatment of
autoimmune disease. Arthritis Research, 2002. 4(5): p. 290-295.
158. Qiu, J. and J. LaBaer, NUCLEIC ACID PROGRAMMABLE PROTEIN
ARRAY: A JUST-IN-TIME MULTIPLEXED PROTEIN EXPRESSION
AND PURIFICATION PLATFORM, in Methods in Enzymology, Vol 500:
Methods in Systems Biology, D. Jameson, M. Verma, and H.V. Westerhoff,
Editors. 2011, Elsevier Academic Press Inc: San Diego. p. 151-163.
159. Romanov, V., et al., A critical comparison of protein microarray
fabrication technologies. Analyst, 2014. 139(6): p. 1303-1326.
160. Legutki, J.B., et al., A general method for characterization of humoral
immunity induced by a vaccine or infection. Vaccine, 2010. 28(28): p.
4529-37.
161. Kroening, K., S.A. Johnston, and J.B. Legutki, Autoreactive antibodies
raised by self derived de novo peptides can identify unrelated antigens on
protein microarrays. Are autoantibodies really autoantibodies? Exp Mol
Pathol, 2012. 92(3): p. 304-11.
162. Boltz, K., et al., High throughput technology for the identification and
characterization of glycan binding peptides. Glycobiology, 2006. 16(11):
p. 1155-1156.
163. Richer, J., S.A. Johnston, and P. Stafford, Epitope identification from
fixed-complexity random-sequence peptide microarrays. Mol Cell
Proteomics, 2015. 14(1): p. 136-47.
164. Halperin, R.F., et al., GuiTope: an application for mapping random-
sequence peptides to protein sequences. BMC Bioinformatics, 2012. 13: p.
1.
165. Kukreja, M., Analysis of immunosignaturing case studies. 2012, Arizona
State University: Tempe, Arizona. p. 217.
109
166. Kukreja, M., S.A. Johnston, and P. Stafford, Immunosignaturing
microarrays distinguish antibody profiles of related pancreatic diseases.
Proteomics and Bioinformatics, 2012.
167. Brown, J.R., et al., Statistical methods for analyzing immunosignatures.
BMC Bioinformatics, 2011. 12: p. 349.
168. Restrepo, L., P. Stafford, and S.A. Johnston, Feasibility of an early
Alzheimer's disease immunosignature diagnostic test. J Neuroimmunol,
2013. 254(1-2): p. 154-60.
169. Restrepo, L., et al., Application of immunosignatures to the assessment of
Alzheimer's disease. Ann Neurol, 2011. 70(2): p. 286-95.
170. Hughes, A.K., et al., Immunosignaturing can detect products from
molecular markers in brain cancer. PLoS One, 2012. 7(7): p. e40201.
171. Lee, H., Identification of neo-antigens for a cancer vaccine by
transcriptome analysis. 2012, Arizona State University: Tempe, Arizona.
p. 168.
172. Shen, L., Investigation of tumor frame shift antigens for prophylactic
cancer vaccine, cancer detection and tumorigenicity. 2012, Arizona State
University: Tempe, Arizona. p. 256.
173. Johnston, S.A. and P.E. Stafford, Compound arrays for sample profiling.
2010, The Arizona Board of Regents, Arizona State University: WO.
174. Navalkar, K.A., et al., Application of immunosignatures for diagnosis of
valley fever. Clin Vaccine Immunol, 2014. 21(8): p. 1169-77.
175. Stafford, P., et al., Immunosignature system for diagnosis of cancer. Proc
Natl Acad Sci U S A, 2014. 111(30): p. E3072-80.
176. DeVita, V.T., Jr. and S.A. Rosenberg, Two hundred years of cancer
research. N Engl J Med. 366(23): p. 2207-14.
110
177. Frei, E., 3rd, et al., The effectiveness of combinations of antileukemic
agents in inducing and maintaining remission in children with acute
leukemia. Blood, 1965. 26(5): p. 642-56.
178. Moxley, J.H., 3rd, et al., Intensive combination chemotherapy and X-
irradiation in Hodgkin's disease. Cancer Res, 1967. 27(7): p. 1258-63.
179. Coley, W.B., The treatment of malignant tumors by repeated inoculations
of erysipelas. With a report of ten original cases. 1893. Clin Orthop Relat
Res, 1991(262): p. 3-11.
180. Mellman, I., G. Coukos, and G. Dranoff, Cancer immunotherapy comes of
age. Nature. 480(7378): p. 480-9.
181. Perillo, N.L., et al., Apoptosis of T cells mediated by galectin-1. Nature,
1995. 378(6558): p. 736-9.
182. Garin, M.I., et al., Galectin-1: a key effector of regulation mediated by
CD4+CD25+ T cells. Blood, 2007. 109(5): p. 2058-65.
183. Pandiyan, P., et al., CD4+CD25+Foxp3+ regulatory T cells induce
cytokine deprivation-mediated apoptosis of effector CD4+ T cells. Nat
Immunol, 2007. 8(12): p. 1353-62.
184. Scheffold, A., J. Huhn, and T. Hofer, Regulation of CD4+CD25+
regulatory T cell activity: it takes (IL-)two to tango. Eur J Immunol, 2005.
35(5): p. 1336-41.
185. Rubartelli, A. and R. Sitia, Chemo-metabolic regulation of immune
responses by Tregs. Nat Chem Biol, 2009. 5(10): p. 709-10.
186. Yan, Z., S.K. Garg, and R. Banerjee, Regulatory T cells interfere with
glutathione metabolism in dendritic cells and T cells. J Biol Chem.
285(53): p. 41525-32.
187. Wakkach, A., et al., Characterization of dendritic cells that induce
tolerance and T regulatory 1 cell differentiation in vivo. Immunity, 2003.
18(5): p. 605-17.
111
188. Ghiringhelli, F., et al., Tumor cells convert immature myeloid dendritic
cells into TGF-beta-secreting cells inducing CD4+CD25+ regulatory T
cell proliferation. J Exp Med, 2005. 202(7): p. 919-29.
189. Liyanage, U.K., et al., Prevalence of regulatory T cells is increased in
peripheral blood and tumor microenvironment of patients with pancreas
or breast adenocarcinoma. J Immunol, 2002. 169(5): p. 2756-61.
190. Wing, K., et al., CTLA-4 control over Foxp3+ regulatory T cell function.
Science, 2008. 322(5899): p. 271-5.
191. Sharma, M.D., et al., Plasmacytoid dendritic cells from mouse tumor-
draining lymph nodes directly activate mature Tregs via indoleamine 2,3-
dioxygenase. J Clin Invest, 2007. 117(9): p. 2570-82.
192. Liang, B., et al., Regulatory T cells inhibit dendritic cells by lymphocyte
activation gene-3 engagement of MHC class II. J Immunol, 2008. 180(9):
p. 5916-26.
193. Bopp, T., et al., Cyclic adenosine monophosphate is a key component of
regulatory T cell-mediated suppression. J Exp Med, 2007. 204(6): p.
1303-10.
194. Muenst, S., et al., Expression of programmed death ligand 1 (PD-L1) is
associated with poor prognosis in human breast cancer. Breast Cancer
Research and Treatment, 2014. 146: p. 15-24.
195. Merelli, B., et al., Targeting the PD1/PD-L1 axis in melanoma: Biological
rationale, clinical challenges and opportunities. Critical Reviews in
Oncology / Hematology, 2014. 89: p. 140-165.
196. Dong, H., et al., Tumor-associated B7-H1 promotes T-cell apoptosis: a
potential mechanism of immune evasion. Nat Med, 2002. 8(8): p. 793-800.
197. Iwai, Y., et al., Involvement of PD-L1 on tumor cells in the escape from
host immune system and tumor immunotherapy by PD-L1 blockade. Proc
Natl Acad Sci U S A, 2002. 99(19): p. 12293-7.
112
198. Brahmer, J.R., et al., Safety and activity of anti-PD-L1 antibody in patients
with advanced cancer. N Engl J Med, 2012. 366(26): p. 2455-65.
199. Topalian, S.L., et al., Safety, activity, and immune correlates of anti-PD-1
antibody in cancer. N Engl J Med, 2012. 366(26): p. 2443-54.
200. Konishi, J., et al., B7-H1 expression on non-small cell lung cancer cells
and its relationship with tumor-infiltrating lymphocytes and their PD-1
expression. Clin Cancer Res, 2004. 10(15): p. 5094-100.
201. Ohigashi, Y., et al., Clinical significance of programmed death-1 ligand-1
and programmed death-1 ligand-2 expression in human esophageal
cancer. Clin Cancer Res, 2005. 11(8): p. 2947-53.
202. Nomi, T., et al., Clinical significance and therapeutic potential of the
programmed death-1 ligand/programmed death-1 pathway in human
pancreatic cancer. Clin Cancer Res, 2007. 13(7): p. 2151-7.
203. Ghebeh, H., et al., The B7-H1 (PD-L1) T lymphocyte-inhibitory molecule
is expressed in breast cancer patients with infiltrating ductal carcinoma:
correlation with important high-risk prognostic factors. Neoplasia, 2006.
8(3): p. 190-8.
204. Ghebeh, H., et al., FOXP3+ Tregs and B7-H1+/PD-1+ T lymphocytes co-
infiltrate the tumor tissues of high-risk breast cancer patients: Implication
for immunotherapy. BMC Cancer, 2008. 8: p. 57.
205. Siegel, R., et al., Cancer Statistics, 2014. Ca-a Cancer Journal for
Clinicians, 2014. 64(1): p. 9-29.
206. Chen, L., et al., Trends in 5-year survival rates among breast cancer
patients by hormone receptor status and stage. Breast Cancer Res Treat,
2014. 147(3): p. 609-16.
207. Pietrowska, M., et al., Mass spectrometry-based serum proteome pattern
analysis in molecular diagnostics of early stage breast cancer. Journal of
Translational Medicine, 2009. 7: p. 13.
113
208. Berger, F. and M.F. Reiser, Micro-RNAs as potential new molecular
biomarkers in oncology: have they reached relevance for the clinical
imaging sciences? Theranostics, 2013. 3(12): p. 943-52.
209. Mostert, B., et al., Diagnostic applications of cell-free and circulating
tumor cell-associated miRNAs in cancer patients. Expert Rev Mol Diagn,
2011. 11(3): p. 259-75.
210. Mostert, B., et al., Circulating tumor cells (CTCs): detection methods and
their clinical relevance in breast cancer. Cancer Treat Rev, 2009. 35(5): p.
463-74.
211. Barderas, R., et al., Sporadic colon cancer murine models demonstrate the
value of autoantibody detection for preclinical cancer diagnosis. Sci Rep,
2013. 3: p. 2938.
212. Piura, E. and B. Piura, Autoantibodies to tailor-made panels of tumor-
associated antigens in breast carcinoma. J Oncol, 2011. 2011: p. 982425.
213. Chapman, C., et al., Autoantibodies in breast cancer: their use as an aid to
early diagnosis. Ann Oncol, 2007. 18(5): p. 868-73.
214. Kukreja, M., Immunosignaturing Microarrays Distinguish Antibody
Profiles of Related Pancreatic Diseases. Journal of Proteomics &
Bioinformatics, 2013. 01(S6).
215. Abe, F., et al., Myeloid-derived suppressor cells in mammary tumor
progression in FVB Neu transgenic mice. Cancer Immunol Immunother,
2010. 59(1): p. 47-62.
216. Kuhn, M., Building predictive models in R using the caret package.
Journal of Statistical Software, 2008. 28(5): p. 1-26.
217. Leek, J.T., et al., The sva package for removing batch effects and other
unwanted variation in high-throughput experiments. Bioinformatics, 2012.
28(6): p. 882-883.
114
218. Woo, E.Y., et al., Regulatory CD4(+)CD25(+) T cells in tumors from
patients with early-stage non-small cell lung cancer and late-stage
ovarian cancer. Cancer Res, 2001. 61(12): p. 4766-72.
219. Poschke, I., et al., Tumor-induced changes in the phenotype of blood-
derived and tumor-associated T cells of early stage breast cancer patients.
Int J Cancer, 2012. 131(7): p. 1611-20.
220. Casbon, A.J., et al., Invasive breast cancer reprograms early myeloid
differentiation in the bone marrow to generate immunosuppressive
neutrophils. Proc Natl Acad Sci U S A, 2015. 112(6): p. E566-75.
221. Johnson, W.E., C. Li, and A. Rabinovic, Adjusting batch effects in
microarray expression data using empirical Bayes methods. Biostatistics,
2007. 8(1): p. 118-27.
222. Johnston, S.A., D.H. Thamm, and J.B. Legutki, The immunosignature of
canine lymphoma: characterization and diagnostic application. BMC
Cancer, 2014. 14: p. 657.
223. Luo, J., et al., A comparison of batch effect removal methods for
enhancement of prediction performance using MAQC-II microarray gene
expression data. Pharmacogenomics Journal, 2010: p. S48-S61.
224. Chen, C., et al., Removing batch effects in analysis of expression
microarray data: an evaluation of six batch adjustment methods. PLoS
One, 2011. 6(2): p. e17238.
225. Sykes, K.F., J.B. Legutki, and P. Stafford, Immunosignaturing: a critical
review. Trends Biotechnol. 31(1): p. 45-51.
226. Drake, C.G., E.J. Lipson, and J.R. Brahmer, Breathing new life into
immunotherapy: review of melanoma, lung and kidney cancer. Nat Rev
Clin Oncol, 2014. 11(1): p. 24-37.
227. Das, R., et al., Combination therapy with anti-CTLA-4 and anti-PD-1
leads to distinct immunologic changes in vivo. J Immunol, 2015. 194(3): p.
950-9.
115
228. Mittendorf, E.A., et al., PD-L1 expression in triple-negative breast cancer.
Cancer Immunol Res, 2014. 2(4): p. 361-70.
229. Schalper, K.A., et al., In situ tumor PD-L1 mRNA expression is associated
with increased TILs and better outcome in breast carcinomas. Clin Cancer
Res, 2014. 20(10): p. 2773-82.
230. Ghebeh, H., et al., The B7-H1 (PD-L1) T Lymphocyte-Inhibitory Molecule
Is Expressed in Breast Cancer Patients with Infiltrating Ductal
Carcinoma: Correlation with Important High-Risk Prognostic Factors.
Neoplasia, 2006. 8: p. 190-198.
231. Ghebeh, H., et al., Expression of B7-H1 in breast cancer patients is
strongly associated with high proliferative Ki-67-expressing tumor cells.
Int J Cancer, 2007. 121(4): p. 751-8.
232. Soliman, H., F. Khalil, and S. Antonia, PD-L1 Expression Is Increased in
a Subset of Basal Type Breast Cancer Cells. PLoS ONE, 2014. 9: p.
e88557.
233. Kmieciak, M., et al., CD4+ T cells inhibit the neu-specific CD8+ T-cell
exhaustion during the priming phase of immune responses against breast
cancer. Breast Cancer Res Treat, 2011. 126(2): p. 385-94.
234. Nagato, T., et al., Combinatorial Immunotherapy of Polyinosinic–
Polycytidylic Acid and Blockade of Programmed Death-Ligand 1 Induce
Effective CD8 T-cell Responses against Established Tumors. Clinical
Cancer Research, 2014. 20: p. 1223-1234.
235. Brahmer, J.R., et al., Safety and Activity of Anti–PD-L1 Antibody in
Patients with Advanced Cancer. The New England journal of medicine,
2012. 366: p. 2455-2465.
236. Duraiswamy, J., et al., Dual blockade of PD-1 and CTLA-4 combined with
tumor vaccine effectively restores T-cell rejection function in tumors.
Cancer Res, 2013. 73(12): p. 3591-603.
116
237. Chen, S., et al., Combination of 4-1BB agonist and PD-1 antagonist
promotes antitumor effector/memory CD8 T cells in a poorly
immunogenic tumor model. Cancer Immunol Res, 2015. 3(2): p. 149-60.
238. Wolchok, J.D., et al., Nivolumab plus ipilimumab in advanced melanoma.
N Engl J Med, 2013. 369(2): p. 122-33.
239. Carroll, J., Bristol-Myers' pioneering PD-1 drug Opdivo OK'd by FDA for
melanoma. 2014.
240. Mueller, F., B. Fuchs, and B. Kaser-Hotz, Comparative biology of human
and canine osteosarcoma. Anticancer Res, 2007. 27(1A): p. 155-64.
241. Gorlick, R., Current concepts on the molecular biology of osteosarcoma.
Cancer Treat Res, 2009. 152: p. 467-78.
242. De Vlaminck, I., et al., Temporal response of the human virome to
immunosuppression and antiviral therapy. Cell, 2013. 155(5): p. 1178-87.
243. Selvarajah, G.T. and J. Kirpensteijn, Prognostic and predictive
biomarkers of canine osteosarcoma. Vet J, 2010. 185(1): p. 28-35.
244. Boston, S.E., et al., Evaluation of survival time in dogs with stage III
osteosarcoma that undergo treatment: 90 cases (1985-2004). J Am Vet
Med Assoc, 2006. 228(12): p. 1905-8.
245. Norrdin, R.W., et al., Characterization of osteosarcoma cells from two
sibling large-breed dogs. Am J Vet Res, 1989. 50(11): p. 1971-5.
246. Ru, G., B. Terracini, and L.T. Glickman, Host related risk factors for
canine osteosarcoma. Vet J, 1998. 156(1): p. 31-9.
247. McNeill, C.J., et al., Characterization of the biological behaviour of
appendicular osteosarcoma in Rottweilers and a comparison with other
breeds: a review of 258 dogs. Vet Comp Oncol, 2007. 5(2): p. 90-8.
117
248. Holmes, D., P. Moody, and D. Dine, Research Methods for the
Biosciences. 2011: Oxford University Press.
249. Agarwal, B.L., Programmed Statistics (Question-Answers). 2007: New
Age International.
250. Black, K., Business Statistics: For Contemporary Decision Making, 8th
Edition. 2013: Wiley Global Education. 880.
251. Shannon, C.E., A mathematical theory of communication. ACM
SIGMOBILE Mobile Computing and Communications Review, 2001.
5(1): p. 3-55.
252. Chen, R., et al., Personal omics profiling reveals dynamic molecular and
medical phenotypes. Cell, 2012. 148(6): p. 1293-307.
118
APPENDIX A
DETECTION OF OSTEOSARCOMA RECURRENCE IN CLINICAL CANINE
SAMPLES BY IMMUNOSIGNATURE
119
AP1.1 Introduction
AP1.1.1 Background of Osteosarcoma
Osteosarcoma (OSA) is a malignant tumor of the bone [240]. It often
starts in the shoulder, wrist, knee, cranium, spinal column, or ribs. Appendicular
osteosarcoma, which is OSA of the limbs, accounts for 75-85% of all bone cancer
in dogs. Bones with OSA are destroyed from the inside out. Intensive pain,
obvious swelling-like lumps and lameness are evidence of tumor growth in OSA.
Tumorous bones are easy to break and difficult to heal. OSA is also highly
aggressive. Over 90% of clinical cases have micro-metastasized to the lungs and
other organs by the time of diagnosis[241]. OSA is the most common canine bone
tumor[242] and accounts for 85% of all primary canine bone tumors[243]. Bone
tumors account for approximately 5% of all canine cancer. OSA has a high
prevalence in middle-aged to elderly dogs (8 to 10 years) [244] with around 8,000
cases diagnosed each year [240]. Large and giant breed dogs, including
Rottweilers, Great Danes, Greyhounds, Saint Bernards and Doberman Pinschers
[245-247], tend to develop OSA at younger ages [242]. Male dogs have a higher
rate of OSA than female dogs [243].
120
AP1.1.2 Current Diagnosis and Treatment of OSA:
Currently, radiographic imaging, age, gender and breed are the main
diagnostic tools. Tissue biopsy is the gold standard to determine OSA. After the
diagnosis of OSA, dogs can be treated with radiation therapy and analgesic
medication (carprofen, aspririn, butorphanol and others). Surgery can also be
performed. Limb-sparing techniques are adapted from human medicine and
involve replacing the tumorous part of the bone with a bone graft.
AP1.1.3 Recurrence of OSA
Because of the high metastasis rate, many the dogs with OSA will already
have micro- metastasis or recurrence after treatment of the primary tumor.
Metastasis is the key factor that influence patient’s survival rate both in dog and
human. In human cancer treatment, patients with localized cancer have better
survival rates than those with metastatic disease [55]. According to American
Cancer Society projected data for 2015, 61% of breast cancer cases are detected at
a localized stage with a five-year relative survival rate of 99% [1]. However,
when tumors spread to nearby lymph nodes or other tissues, the survival rate
decreases to 85% [1]. If the breast cancer is in a late stage, in which tumors are
found in lymph nodes around the collarbone or more distant, the survival rate falls
to 25% [1]. The early detection of primary osteosarcoma and its recurrence will
improve prognostic capability and treatment modality, relieve pain in dogs, and
potentially extend their lives.
121
AP1.2 Result and Discussion
AP1.2.1 Canine Osteosarcoma Clinical Sample Summary
Dog samples were generously provided by Dr. Douglas H. Thamm from
Colorado State University. Five dogs were serially collected for four time points.
The four time points were labeled as Initial, A5, PTOR and OR, respectively.
Sample IDs and dates for each time point are described below (Table AP1.1). The
disease course interval is the difference of days between two time points. Initial
was the time point of diagnosis before surgery. A5 was the time point of the fifth
chemotherapy treatment, approximately ten weeks after diagnosis. PTOR was the
time point of the last visit prior to relapse. OR was the time point of the visit at
which the relapse was detected. Age-matched normal serum samples were are
selected from clinically normal client-owned dogs.
Table AP1. 1: Canine Osteosarcoma Clinical Sample Summary
Sample
Sample ID 165155 171854 172133 168077 111098
Initial 1998/6/16 1999/5/27 1999/6/8 1998/10/29 1997/7/8
A5 1998/9/18 1999/8/19 1999/8/23 1999/1/6 1998/6/9
PTOR 1998/12/11 1999/9/20 2000/2/15 1999/3/19 1998/11/24
OR 1999/3/12 1999/9/27 2000/5/18 1999/6/10 1999/2/16
Disease course interval time (days)
Initial to A5 94 84 76 69 336
A5 to PTOR 84 32 176 72 168
PTOR to OR 91 7 93 83 84
122
Blood samples from five dogs were serially collected for four time points.
The four time points are labeled as Initial, A5, PTOR and OR, respectively.
Sample IDs and dates for each time point have been described. The disease course
interval is the difference of days between two time points.
AP1.2.2 Correlation Analysis Of Overall Antibody Profile During OSA
Progression
For this time series study, it was interesting to know how the overall
antibody profile changes at different time points by immunosignature technology.
At each time point, the intensity of each peptide for the five samples was
averaged. Pearson correlation, using all peptides between two time points, was
calculated (Table AP1.2).
The highest correlation was 0.948 between OR (the time point of the visit
at which the relapse was detected) and PTOR (the time point of the last visit prior
to rela thatpse). This means the antibody response at PTOR was similar to that at
OR and implies that the samples at PTOR might already have been at the
recurrence stage and have a signature of recurrence. The lowest correlation was
0.739, between PTOR and Normal. This implies that compared with the primary
tumor, the recurrence was more distinguishable from normal.
123
Table AP1. 2: Correlation Analysis of Overall Antibody Profile
The experiment was done by using the CIM 10K version-2 peptide arrays
from Batch 324. Each sample was run with two to three replicates. Informative
peptides were selected by comparing different stages with p value and fold change
cutoff. P values were calculated by GeneSpring software using a two-tail T-test
with FDR as a multiple comparison correction. The fold change of a peptide was
calculated by using the average intensity of samples from one condition divided
by the average intensity of samples from the other condition.
Table AP1. 3: Distribution of Fold Change in Selected Peptides
Fold is calculated by using the average intensity of each peptide at one time point
divided by the other. Samples of recurrent disease include samples at both OR and
PTOR.
#: Numbers of peptides have different fold change between two conditions. An
increase means that peptides have a certain fold change from one condition to the
other, and vice versa for a decrease.
Table AP1.3 lists the distribution of the numbers of selected peptides by
fold change in three comparisons. I compared dog samples without canine cancers
from normal to initialFold # increase decrease
1.5 115 99 16
2 20 18 2
2.5 4 3 1
3 2 1 1
4 2 1 1
5 0 0 0
from normal to recurrenceFold # increase decrease
1.5 278 211 67
2 90 85 5
2.5 51 49 2
3 31 30 1
4 8 7 1
5 2 1 1
5.5 1 0 1
from normal to ORFold # increase decrease
1.5 197 170 27
2 71 67 4
2.5 35 34 1
3 18 17 1
4 1 0 1
6 1 0 1
6.5 0 0 1
All sample Initial A5 OR PTOR
Initial
A5 0.943
OR 0.95 0.942
PTOR 0.889 0.946 0.948
Normal 0.912 0.83 0.854 0.739
124
(normal) with samples at initial diagnosis to find the immunosignature for
primary OSA. As mentioned above in the introduction, OSA is highly aggressive,
and micro-metastases may already be present when primary OSA is detected. The
time difference between PTOR and OR was 71.6 days on average (ranging from
7-93 days) for the five dogs. At the time of PTOR, the dogs could already have
had micro-recurrent tumors. By combining samples at PTOR and OR together,
with label “recurrent,” this could cover all stages of recurrent OSA
comprehensively. By comparing normal with recurrent samples, I could
investigate the immunosignature of recurrent OSA. OR was the time point when a
relapse was determined clinically. By comparing OR with normal samples, I
could investigate the immunosignature of recurrent OSA for that specific time
point.
The result shows that most selected peptides had 1.5- to two-fold change.
A few had a greater than four-fold change. Most selected peptides were increased,
with fewer than 20% decreased, on average. The number of decreased peptides
dropped faster than the increased ones with the increase of fold change. These
data show that we could find immunosignatures to distinguish primary OSA and
recurrent OSA from healthy controls. They also show that most of the selected
peptides had higher intensity levels in dogs with OSA compared with health
controls.
AP1.2.3 Comparison of Immunosignature at Multiple Time Points
After finding informative peptides for each time point, I wanted to ask
how the intensities for those peptides changed along tumor development. This
125
could reveal the role of tumor-related antibodies during the disease progression of
OSA.
I first selected two peptide sets: 1) the 94 overlapped peptides between
115 peptides of primary OSA and 278 for recurrent OSA at both OR and PTOR;
and 2) the 86 overlapped peptides between 115 peptides of primary OSA and 197
for recurrent OSA at OR. 85 peptides of the 86-peptides set were shared in the 94-
peptide set.
Figure AP1. 1: Proportion of Overlapped Peptides Shared Among Different Time
Points
Intensity level of the overlapped peptides are plotted against each time point.
21 184
Primary RecurrenceOR+PTOR
94 29 111
Primary RecurrenceOR
86
overlapped 94 overlapped 86
126
Table AP1. 4: Proportion of Overlapped Peptides Shared Among Different Time
Points
From the Figure AP1.1, the intensity of the majority of the overlapped
peptides intensity increased from Initial to A5 and to PTOR, but decreased in OR.
From the Table AP1.4, the number of significant peptides overlapped between
primary OSA and recurrent OSA was 74.8%- 81.7% in the 115-peptide set and
decreased to 43.7% - 47.7%.
These data mean that different antibodies were dominant at different time
points. This might be due to the immunoediting process between OSA and the
immune system in dogs. Antigens that presented in the primary OSA disappeared
due to high immunogenicity. After the tumor developed, more neo-antigen may
have appeared and triggered different antibody responses at the recurrent stage.
Figure AP1. 2: Intensity Plot of Significant Peptides between Primary and
Recurrent OSA
Percentage of the overlap peptides in each set
Overlap Initial(115) Recurrence(278) OR(197)
94 81.7% 33.8% 47.7%
86 74.8% 30.9% 43.7%
127
In order to further test whether there was a difference between primary
tumor and recurrent tumor, I asked how many peptides were statistically
significant between primary and recurrent by comparing these two time points. 35
peptides were selected. As Figure AP1.2 shows, most of them decrease from
primary to recurrent. The PCA result on the 35 peptides shows cluster of primary
samples and recurrent samples by each time point. These show that primary OSA
differed from recurrent OSA in terms of their immune response to tumor antigens.
AP1.2.4 Conclusion and Future Direction
I recognize the limited sample size for this experiment, and I don’t have
further validation of the discovered immunosignature. Some of the selected
peptides may have been due to randomness. The signature discovered may not
well represent different breeds of dogs with osteosarcoma. However, in this study,
I have used multiple comparison methods to reduce the possibility of randomness
and carefully test each sample. This study could still be beneficial to studying
osteosarcoma.
With the development of OSA, the overall antibody profiles changes
substantially. Different immunosignatures were found for primary OSA and
recurrent OSA. The significant peptides are increased from normal controls to
tumor samples, showing increasing anti-tumor immune response. The difference
between primary OSA and recurrent OSA exists. Most significant peptides
decrease from primary OSA to its recurrence. Based on the data, the selected
peptides can be grouped into the following categories: Intensities were high in
128
normal and low in initial. This may be due to the immunosuppression or Treg by
the primary tumor. Intensities increased after treatment: suppression was relieved
and IS increased in the recurrence. Intensities decreased at late recurrence: the
recurrent tumor seemed to suppress these peptides but not significantly. It
behaved differently from the primary tumor. This may imply that the immune
suppression from primary and recurrent tumors was different.
In order to get a more solid immunosignature of primary and recurrent
osteosarcoma, more dog samples with initial, A5, PTOR, OR time points are
needed. Current PTOR is the time point before relapse is detected, with an
average of 71.6 days before recurrence. In order to address the early detection of
OSA recurrence, it will be better to get a time point that precedes recurrence
detection by a longer period. It will be better to have samples for which the
interval between “initial to A5” and “PTOR to OR” is long enough to have more
distinguished stages.
AP1.3 Methods
AP1.3.1 Sample Summary
Dr. Douglas H. Thamm from Colorado State University generously
provided five canine osteosarcoma clinical samples. The breeds of the five canine
patients are Borzoi, Labrador retriever, golden retriever, and mix breeds. Each
dog had been confirmed with osteosarcoma histologically or cytologically. Sera
from each dog were serially collected for four time points. The four time points
were labeled as Initial, A5, PTOR and OR respectively. Initial is the time point of
diagnosis before surgery. A5 is the time point of firth chemotherapy treatment,
129
approximately ten weeks after diagnosis. PTOR is the time point of the last visit
prior to relapse. OR is the time point of the visit at which relapse was detected.
Sera were collected and stored at -80°C through all collections.
Chemotherapy-naïve primary tumor samples were selected from the
Colorado State University Animal Cancer Center's tissue archive based on the
criteria that the patient had undergone surgical amputation followed by
chemotherapy with doxorubicin and/or a platinum-based drug. Doxorubicin
(Adriamycin) was given every two weeks for five treatments.
AP1.3.2 Binding of Sera to the Peptide Microarrays and Microarray Data
Processing
CIM standard clinical protocol with four steps was used. Detailed methods
can be found in Chapter 2 and 3.
AP1.3.3 Statistical Analysis
Statistical analysis of microarray data was done with Gene- Spring 7.3.1
(Agilent, Inc., Palo Alto, CA) by importing image-processed data from GenePix
Pro 6.0 (Axon Instruments, Union City, CA). Calculations based on the GenePix-
prepared gpr text files were done on the median signal intensity per spot. Poor
quality spots were excluded from analysis by flagging as “absent” upon visual
inspection. Prior to analysis, each array was normalized to the 50th percentile to
eliminate array-to-array variation, and signal intensities of less than 0.01 were set
to 0.01. Values from replicate arrays were averaged and used in the analysis. P-
values were corrected for multiple testing by the Benjamin-Hochberg false
discovery rate method in Genespring GX 7.3 software (Agilent Technologies,
130
Palo Alto, CA). The Welch t test was used in GeneSpring to select significant
peptides with p<0.05 between two groups.
131
APPENDIX B
CHARACTERIZATION OF ANTIBODY PROFILE DURING CANCER
DEVELOPMENT
132
AP 2.1 Quantitative Characterization Of Antibody Profiles During Tumor
Development
AP2.1.1 Introduction
Immunosignatures allow us to study the overall antibody profile of a
sample. We can quantitatively evaluate the distribution of overall antibodies at a
certain time point in a disease to potentially reveal its internal patterns. A
quantitatively method to measure statistical dispersion of antibodies can indirectly
show antigen heterogeneity. This can help researchers to understand the
interaction between the immune system and cancer.
Different statistical parameters are available to characterize the
distribution of information with a single quantitative value. With the increasing
popularity of mobile and wearable devices, customers can collect data about their
health easily. A simple visualization of major biological information in our bodies
will make it possible for everyone monitor their health. In Dr. Kurt Whittemore’s
dissertation research, he used AbStat as a tool to characterize the antibody profile
of a patient through immunosignature technology. The current work is a
collaboration with Dr. Whittemore and uses his AbStat algorithm to calculate
entropy, coefficient of variation, kurtosis, skew, and dynamic range [248-251].
Entropy is a measurement of the homogeneity of a frequency distribution.
If all information in the distribution is ordered (with homogeneous value), its
entropy decreases. The minimum value is zero when all values in that distribution
are identical. The value increases when the distribution is more heterogeneous.
The maximum entropy is achieved when values of every element are different
133
from each other. Kurtosis measures the peakedness of a frequency distribution,
with higher values indicating a sharper peak. Skewness measures symmetry: a
larger absolute amount of skewness indicates a more asymmetric distribution,
which means that values in that distribution lean to one side of the mean.
Dynamic range is the ratio of the value of 95% intensity to the value of 5%
intensity. A high dynamic range indicates a wider range of distribution. The
coefficient of variation is the ratio of the standard deviation to the mean. A high
coefficient of variation indicates a high degree of dispersion.
We performed a time series study to characterize the overall antibody
profile during tumor development, using two transgenic mice and two wild type
controls (Figure AP2.1, Figure AP2.2). By doing this, we hope to find some
interesting general pattern between the immune system and cancer.
AP2.1.2: Result and Discussion
Sera were collected every two weeks for two transgenic mice and two age-
matched wild type mice (Table AP2.1, Table AP2.2). Sera from these samples
were run on the CIM Version-2 arrays using a standard CIM clinical protocol
with one replicate at each time point. We used one replicate because the
abundance of continuous time points allowed us to get a good representation of
the antibody profile during tumor development with one replicate. Slides were
randomly chosen from all available slides in the same batch of slides by “sample”
function in R. Arrays, and samples were also randomly assigned to different
positions on the Tecan.
134
Figure AP2. 1: Summary of Mice Samples
Sixteen time points of Tg8-1-1, twelve time points of Tg8-3-9 and WT1-3, and
eight time points for wild type mice WT2-3 were used.
Figure AP2. 2: Tumor growth Curve for Tg8_1-1 and Tg8_3-9 Transgenic Mice
Entropy, kurtosis, skewness, dynamic range, and CV were calculated for
two transgenic mice and two wild type controls at multiple time points during
tumor development (Figure AP2.3, AP2.4). The results show that the value of
kurtosis, skewness, dynamic range, and CV were similar and correlated between
0
5
10.0 20.0 30.0 40.0 50.0
Age(weeks)
Time series collection
Tg8_3-9(12)
Tg8_1-1(16)
WT1-3(12)
WT2-3(8)
0
1000
2000
3000
4000
5000
0.0 10.0 20.0 30.0 40.0 50.0
Tumor size(mm3)
Age(weeks)
Tumor Growth
Tg8_3-9 Tg8_1-1
135
transgenic mice and control mice. The chaotic pattern of these parameters showed
dynamic changes occurring in the antibody repertoire during mouse development
(Table AP2.1). These changes might be the result of the aging process instead of
tumor development, as there was no significant difference between the transgenic
mice and the wild type controls for these parameters.
Figure AP2. 3: AbStat Time Course for Transgenic Tg8-1-1 and Wild Type WT2-
3 Mice
0
50
100
150
200
250
10 20 30 40 50
Kurtosis
Tg8-1-1 WT2-3
0
2
4
6
8
10
12
14
16
10 20 30 40 50
Skewness
Tg8-1-1 WT2-3
0
2
4
6
8
10
12
10 20 30 40 50
Dynamic Range
Tg8-1-1 WT2-3
5.2
5.4
5.6
5.8
6
6.2
10 30 50
Entropy
Tg8-1-1 WT2-3
0
1
2
3
4
5
6
10 20 30 40 50
CV
Tg8-1-1 WT2-3
136
Figure AP2. 4: AbStat Time Course for Transgenic Tg8-3-9 and Wild Type WT1-
3 Mice
Table AP2. 1: Coefficient of Variance of Entropy, Kurtosis, Skewness and
Dynamic Range of All Time Points for Each Mouse
Only entropy showed some level of difference between the transgenic
samples that developed tumor later and the wild type controls (Figure AP2.5). In
WT 2-3, the entropy for the wild type controls was linear increasing with less
variance, compared with Tg8-1-1. The entropy for Tg8-1-1 fluctuated along
tumor development with significant variance. However, this pattern was not
CV Entropy Kurtosis SkewnessDynamic
Range
Tg8-1 4.8% 27.6% 13.3% 26.2%
WT2-3 5.8% 32.3% 15.0% 17.5%
Tg8-3-9 3.0% 36.2% 16.0% 22.8%
WT1-3 2.8% 35.8% 16.2% 11.2%
0
50
100
150
200
250
300
350
400
15 25 35 45
Kurtosis
Tg8-3-9 WT1-3
0
2
4
6
8
10
12
14
16
18
15 25 35 45
Skewness
Tg8-3-9 WT1-3
0
1
2
3
4
5
6
7
15 25 35 45
Dynamic Range
Tg8-3-9 WT1-3
4.8
4.9
5
5.1
5.2
5.3
5.4
5.5
5.6
5.7
15 25 35 45
Entropy
Tg8-3-9 WT1-3
0
1
2
3
4
5
6
7
15 25 35 45
CV
Tg8-3-9 WT1-3
137
evident in the wild type mouse. Dr. Kurt Whittemore’s study showed a positive
correlation between entropy and age. The constant increase in wild-type mice may
be the result of the aging process, when more antibodies were produced during
their lifetimes. However, in the transgenic mice, entropy of the antibody
repertoire was chaotic instead of showing a constant increase. This might have
been the result of an immunoediting process between tumor and immune system.
Different antibodies were produced to target tumors at different development
stages. The immune system might have produced a large amount of antibodies
when antigens were highly immunogenic in tumor cells and then decreased
antibody production because of mutations in those antigens that made them not
immunogenic. This distinct difference of entropy in transgenic mice and wild-
type mice may be an indicator for tumor development.
Figure AP2. 5: Entropy Time Course for Transgenic and Wild Type Mice
In conclusion, this study provided an interesting approach to evaluating
5.2
5.4
5.6
5.8
6
6.2
10 20 30 40 50
Tg8-1-1
entropy
5.2
5.4
5.6
5.8
6
6.2
10 20 30 40 50
WT2-3
entropy
4.8
5
5.2
5.4
15 20 25 30 35 40 45
Tg8-3-9
entropy
5
5.2
5.4
5.6
15 20 25 30 35 40 45
WT1-3
entropy
138
the antibody repertoire during tumor development. A time series study of cancer
was first conducted in the immunosignature platform to discover hidden patterns
over time. The results implied that entropy could be a useful parameter to evaluate
tumor development and to measure statistical dispersion of antibodies and
indirectly show antigen heterogeneity.
AP2.2 A Time Series Study of Antibody Profile Along Tumor Development
AP2.2.1 Introduction
Although immunoediting theory suggests that tumor cells present different
antigens along tumor development, few studies have investigated the dynamic
process of the immune system during tumor development. Immunosignature can
reflect different antibody profiles at different stages. It provides us an opportunity
to evaluate the antigen heterogeneity along tumor development by analyzing
immunosignature in a time series study.
Compared with static comparisons, a time series study of biomarkers has
the advantage of improving the performance of diagnostic methods. The pitfall of
current biomarkers in bigger populations is that there is substantial heterogeneity
even in populations with similar genetic backgrounds. Although common profiles
can be found and an absolute cutoff can be drawn, the results may not be relevant
to each individual. Therefore, dynamic measurements that allow the assessment of
within-patient expression changes will be more useful [56]. The recent study on
continuous monitoring of multiple omics by Rui Chen and colleagues showed the
promise of self-monitoring in personalized medicine and diagnosis [252].
139
The method for analyzing multiple time points of a feature falls under time
series analysis. Time series studies have a number of advantages: they can use the
time series structure in the data; can increase the accuracy of prediction; and can
capture information about features with transient expression changes. These are
not possible in static studies that compare only two populations at a single time
point.
In this study, an individual dynamic monitoring of the antibodyome was
conducted in a murine breast cancer model.
AP2.2.2 Result and Discussion
In order to study the time serial pattern of the antibodyome, we used
serum samples from the same individual at multiple time points. In fields in which
time series analyses is widely used, such as economics, astronomy, physics and
industrial engineering, time series data normally include more than 30 time points.
However, in biology, such an extended sequence of samples is rarely accessible.
For this study, I had a time series serum collection for FVB/N neuN transgenic
mice and FVB/N wild-type mice with more than ten time points for each mouse.
Serum for each mouse was collected every two weeks, from mice aged 5-10
weeks to 40-45 weeks, which covered the complete tumor development stages.
In this study, I chose two transgenic mice (Tg8-1-1, Tg8-3-9) and two
wild-type mice (WT1-3, WT2-3) (Table AP2.2, Table AP2.3). Sixteen time points
of Tg8-1-1, twelve time points of Tg8-3-9 and WT1-3, and eight time points for
wild type mice WT2-3 were used. The samples used in this study were the same
samples run in the AP2.1 section. The interval was the weeks between two time
140
points. It should be noted that in order to have a better result in time series
analysis, a constant interval is important. In Tg8-1-1, the average interval is
2.02±0.37 weeks. In Tg8-3-9, the average interval is 1.98±0.4 weeks.
Table AP2. 2: Sample Information for Tg8-1-1 and WT2-3
Tg8-1-1 WT2-3
Tumor size (mm3) Age (wks) Window (wks) Interval (wks) Age (wks)
15.1 -18 2.3 15.3
17.4 -15.7 1.9
19.3 -13.8 2.1 19.1
21.4 -11.7 2.7
24.1 -9 1.2 24.1
25.3 -7.8 2.1
27.4 -5.7 2
29.4 -3.7 2.2 29.1
31.6 -1.5 1.5
0.5(1) 33.1 0 2.5 33.1
89.43(2) 35.6 2.5 1.8
261.71(2) 37.4 4.3 1.7 37.4
423.64(2) 39.1 6 2
711.49(3) 41.1 8 2.2 41.3
1626.01(4) 43.3 10.2 2.1
3326.36(4) 45.4 12.3 45.4
141
Table AP2. 3: Sample Information for Tg8-3-9 and WT1-3
Autocorrelation function (ACF) is a standard method to analyze time
series data. A feature with time series patterns is autocorrelated, and the Durbin-
Watson test can calculate and test its significance (Figure AP2.6). The Durbin-
Watson Test in the R package was used to calculate the significance of ACF with
a cutoff of p value at 0.05. Peptides passing the test should have autocorrelation
structure instead of randomness.
Tg8_3-9 WT1-3
Tumor size (mm3) Age (wks) Window (wks) Interval (wks) Age (wks)
0 19.3 -12.3 2.1 17.3
0 21.4 -10.2 2.7 19.1
0 24.1 -7.5 1.2 21.1
0 25.3 -6.3 2.1 24.1
0 27.4 -4.2 2 27.3
0 29.4 -2.2 2.2 29.1
35.93(1) 31.6 0 1.5 31.3
165.38(1) 33.1 1.5 2.5 33.1
493.32(1) 35.6 4 1.8 35.1
1105.01（2） 37.4 5.8 1.7 37.4
2215.85（2） 39.1 7.5 2 39.1
2932.92（2） 41.1 9.5 41.3
142
Figure AP2. 6: Concept of Autocorrelation function (ACF) and Durbin-Watson
Test
249 peptides in Tg8-1-1 and 849 peptides in Tg8-3-9 passed Durbin-the
Watson test with p value less than 0.05. These peptides showed an autocorrelated
structure. Self-Organizing Map was used to cluster the significant peptides into
Tg8-1-1 and Tg8-3-9 and to reveal potential patterns (Figure AP2.7). Several
clusters showed a correlated trend between signal intensity and tumor
development, although patterns were different for each individual. Tg8-1-1 had
increased peptides around a tumor-initiated point, while Tg-3-9 only had
decreased ones. This pattern could be a representation of the immunoediting
process in the mouse immune system.
Autocorrelation function (ACF)
143
Figure AP2. 7: Self-Organizing Map to cluster 249 autocorrelated peptides in
Tg8-3-6.
Setting for SOM in GeneSpring are Rows 3, Columns 3, Iterations 10000, Radius
4.0.
Tg8-1_ACF_249: Rows 3, Columns 3, Iterations 10000, Radius 4.0.
144
Figure AP2. 8: Self-Organizing Map to cluster 849 autocorrelated peptides in
Tg8-3-6. Setting for SOM in GeneSpring are Rows 3, Columns 3, Iterations
10000, Radius 4.0.
One cluster of the 249 peptides by K-means included a cluster of 24
peptides (Figure AP2.9). The intensity of these peptides increased 5.7 weeks
before the first tumor occurred and decreased later. The intensity changes of these
peptides may be caused by the primary tumor. In order to show that the changes
are tumor-related, the same 24 peptides were also plotted in the age-matched
wild-type mice. The autocorrelated pattern appeared only in the transgenic mice
with the development of tumors but not in the wild-type controls, which indicated
its tumor specificity. The changing pattern implies that tumor antigens may have
appeared early, before the primary tumor triggered a strong immune response, and
that those antigens disappeared when the tumor matured.
145
Figure AP2. 9: Intensity of a cluster of 24 autocorrelated peptides correlated with
tumor development
The orange arrow is the time point for first primary tumor.
In this study, an individual dynamic monitoring of antibodyome was
conducted in a murine breast cancer model. Autocorrelation function was
calculated for each peptide in two transgenic mice and two wild-type mice. The
Durbin-Watson test was used to find significant peptides with autocorrelation
structure. Selected autocorrelated peptides showed dynamic and tumor-specific
changes with tumor development. The pattern of changes can be tracked back five
weeks before the tumor was palpable. Although this study used only four mice
and the results have not been repeated, the results might attract the interest of
other researchers in this field. The method used in this study can also be applied
to genomic, transcriptomic, proteomic, and metabolomic data.
Tg WT
146
AP2.3 Robustness of immunosignature
AP2.3.1 Introduction
A good, robust biomarker should have low variance in a population, even
when that population is diverse. On the one hand, since many biomarkers have
high variance in both healthy and diseased populations, they are likely to fail in
clinical trials, as they cannot differentiate the overlap between the healthy and
diseased populations. This kind of biomarkers I called limited biomarkers. They
will lead to high false-positive rate. A robust biomarker, on the other hand, should
have low variance in both healthy and diseased populations (Figure AP2.10) so
that a sample can be classified as either disease-positive or disease-negative with
high accuracy. We use correlation to measure the heterogeneity in a population. If
the correlation among samples in a population is high, that means that there is low
variance in that population.
147
Figure AP2. 10: Concept of Robustness in immunosignature
AP2.3.2 Result and discussion:
Table AP2. 4: Sample Information for Robustness Test
In this study, serum samples from four transgenic (Tg) and eight age-
matched wild-type (WT) control mice at a time point before the first breast tumor
was detected were run in CIM V2 arrays using the standard CIM clinical protocol
Healthy Disease
Leve
l of
cert
ain
bio
mar
ker
Healthy Disease
Low correlation High correlation
RobustBiomarker
LimitedBiomarker
Treatment Control
Tg-1_31 WT Cage1-1_19 WT Cage2-1_17
Tg-10_16.6 WT Cage1-2_17 WT Cage2-2_19
Tg-11_19.4 WT Cage1-3_19 WT Cage2-3_17
Tg-8_16.8 WT Cage1-4_17 WT Cage2-4_19
148
(Table AP2.4). Sixty-seven peptides were significant by t-test (p<0.05), with a
1.3-fold change as the cutoff. The selected peptides were specific
immunosignatures for breast cancer at two weeks before the first tumor. Here, we
wanted to test whether these peptides were robust or not.
The correlations of 67 peptides in either the transgenic group or the
control group were high and had a low variance (Figure AP2.11). These were
called in-group correlations. The calculated correlation between one transgenic
sample and one wild-type sample and all possible combinations were called
between-group correlations. The correlation of 67 peptides was lower than the in-
group correlation and had a much wider range. This showed that the selected 67
peptides had lower variance in in-group correlations and had higher variance in
between-group correlations (p<0.01 between among and WT).
Figure AP2. 11: Correlation of selected peptides in transgenic (Tg) and wild-type
(WT) control mice.
Correlations of 67 peptides in any two samples of Tg group are label with “Tg”,
same as “WT”. Correlations between one transgenic sample and one wild-type
sample and all possible combinations are labelled with “among”.
among vs Tg P = 0.14among vs WT P < 0.01
Tg vs WT P > 0.05
149
In order to show that the difference between in-group and between-group
correlations were not due to internal array bias, I calculated the correlations for
both using all peptides on the 10K array (Figure AP2.12). This should reflect the
overall antibody profiles of samples. The result was that in-group and between-
group correlations were at around same level, and their variance was not as great
as in the correlation of the 67 selected peptides.
Figure AP2. 12: Correlation of all peptides on the 10K array
In sum, I tested the robustness of selected peptides and showed that
robustness could be a criterion for evaluating the performance of an
immunosignature.
150
APPENDIX C PERMISSIONS
151
Permissions have been obtained to reprint copyrighted material.
American Cancer Society has granted permission automatically to use
Figure 1. 1 Biology of Breast Tissue and Figure 1. 2: Illustration of Mammogram
and MRI and Figure 1. 3: Five-year Relative Survival Rates (%) by Stage in US.
Molecular & Cellular Proteomics has granted permission automatically to
use Figure 1. 4: Distribution of Human Plasma Protein.
Figure 1. 6: Introduction of CTLA-4 and PD-1 Treatment has obtained
permission for Elsevier with license number 3654470549052.
Use of Figure 1. 7: Biology and Mechanism of Anti-PD-L1 Antibody
Treatment has been approved by Leann Sandifar, Vice President – Operations,
from Patient Resource LCC on June 11th.
152
BIO SKETCH
I grew up surrounded by outstanding healthcare professionals and
biomedical researchers. As a child, I was curious about the fancy equipment that
doctors used to perform surgery and amazed by how easily the drugs a doctor
prescribed could cure a patient’s disease. Seeing patients relieved of their pain
and the happiness of parents when their child was cured made me feel that doing
something that could improve people’s health would be a meaningful way to
spend my life. I chose biotechnology as my undergraduate major at Wuhan
University in China. The concept that, in the future, a person’s health condition
could be monitored conveniently and that diseases could be detected before
symptoms occur became clear to me. I realized that early diagnosis and
prevention were more crucial than treatment at the late stage of disease. Since
that time, I have been interested in all kinds of new technologies to diagnose a
disease. I came to ASU because it was one of only a few universities that have
many professors focusing on developing diagnostic tools for human diseases. My
work with Dr. Johnston is to find new strategies to diagnose breast cancer early
and to improve cancer outcomes through novel treatment. It has been an amazing
journey, and I feel sincerely grateful to Dr. Johnston and ASU for giving me this
opportunity. Upon graduation, I plan to pursue a career in building novel
technology to improve people’s health.
![Automated breast cancer detection and classification using ...fs.unm.edu/neut/AutomatedBreastCancerDetection.pdf · of ultrasound images for breast cancer detection [6–8]. Statistics](https://img.dokumen.tips/doc/110x75/5f0610337e708231d4161b2c/automated-breast-cancer-detection-and-classification-using-fsunmeduneutautomatedb.jpg)
