x :1 I 2:... “LI! .. .... £15.. , .3»?!th . . «SIZJ... . .. inn? . . . aaunagg. 33 . t3¢~xfnmn¢u42 117, oc‘: . i. .. .fi, 2 .. mum“; : Brain. anew... a .,. .y . .2 $3th 1&2 eicfer . , 3-35.33 .th . .2 3... i h .. ii. . z! Rh a. x. A , : 2.233... vigil): . . 511...... £331.... :11... 2. C 7.1. .x i: 1 fail) 1.11). $95,535.” A . .33.. _ .2 $1... i. E.‘ (s... .u‘... 3:19.... i: .v‘rfiv ll «7‘ S gift-3073} TEE SIS that LIBR 1‘ 'RY MiChiQ‘a.. State University This is to certify that the thesis entitled HER-2/HER-3 HETERODIMERS: A QUANTITATIVE MOLECULAR EPIDEMIOLOGY MARKER AND METHOD OF ANALYSIS presented by ERIC JON KORT has been accepted towards fulfillment of the requirements for 14.3. degree in EPIDEMIOLOGY Kveldajor professor Date 12-11-2001 0-7639 MS U is an Affirmative Action/Equal Opportunity Institution PLACE IN RETURN BOX to remove this checkout from your record. TO AVOID FINES return on or before date due. MAY BE RECALLED with earlier due date if requested. DATE DUE DATE DUE ' DATE DUE 6/01 c:/ClRC/DateDue.p65cp.15 HER-2 / 3 HETERODIMERS: A QUANTITATIVE MOLECULAR EPIDEMIOLOGY MARKER AND METHOD OF ANALYSIS By Eric Jon Kort A THESIS Submitted to Michigan State University in partial fulfillment of the requirements for the degree of MASTER’S OF SCIENCE Department of Epidemiology 2001 ABSTRACT HER-Z/ 3 HETERODIMERS: A QUANTITATIVE MOLECULAR EPIDEMIOLOGY MARKER AND METHOD OF ANALYSIS By Eric Jon Kort The emerging field of quantitative molecular epidemiology promises to expand our understanding of the molecular bases of disease, and treatments that can specif- ically target these pathways. Closely resembling our current approach to infectious disease, molecular epidemiology has already begun to allow classification of dis- ease on the basis of the molecular pathways leading to pathology, and thereby allowing treatment to target these pathways. This document was derived from an application for funding to refine meth- ods for quantifying the data contained within human tissues. Specifically, these methods will be applied to test the hypothesis that the Her-2 and Her-3 cell sur- face receptors must interact through non-covalent association (dimerization) to elicit pathophysiological activity in Her-2 positive human breast cancer. Through the use of fluorescent markers, the location of these proteins may be mapped as pixels within microscopy images. A molecular epidemiologic statistical parame- ter will be developed that will quantify the correlation of these imaged markers (colocalization—used as a proxy for the biological phenomenon of dimerization) The statistical power of this parameter to discriminate between differential prog- nostic groups will be measured and compared to existing methods. ACKNOWLEDGEMENTS Special thanks to: 0 James H. Resau, PhD, for his guidance on all issues related to pathology, mi- croscopy, histology, and politics. 0 Han Tsarfaty, PhD, for his biochemical and computational expertise. And to my thesis committee: 0 Nigel Paneth, M.D., M.P.H. 0 Michael Collins, M.D., MS. 0 Wen-Jiang Fu, PhD. And to: 0 Andrea Kort, M.D., for all of her support and for always being there to ask - ”So, when are you going to finish your thesis?” SOLI DEO GLORIA TABLE OF CONTENTS 1 Introduction 1 1.1 Human Tlssue Research: a growing imperative. ............ 2 1.2 Quantitative Molecular Epidemiology Defined ............. 6 1.3 Specific aims ................................. 8 2 Her-2 and Breast Cancer: Epidemiology and Cell Biology 10 2.1 Descriptive Epidemiology of Breast Cancer and Her—2 expression . . 11 2.2 Analytic Epidemiology of Breast Cancer and Her-2 expression . . . . 13 2.2.1 Breast cancer risk factors ...................... 13 2.2.2 Her-2 is elevated in a variety of cancers. ............ 15 2.2.3 Her-2 may distinguish between prognostic groups. ...... 23 2.2.4 Her-2 may distinguish between treatment groups. ...... 23 2.3 Her-2 exhibits complex ligand-receptor relationships. ......... 24 2.4 Her-2 has no confirmed human, cancer-specific mutations ...... 25 2.5 Her-2 is the target of the first receptor-specific cancer treatment. . . . 26 2.6 Issues for study ............................... 27 3 Design 29 3.1 Images as Data Sets ............................. 29 3.1.1 Confocal microscopy ........................ 33 3.1.2 Colocalization as an image phenomenon ............ 34 3.2 Application to Her-2/ Her-3 colocalization ................ 39 4 Methods 41 4.1 Phase one: algorithm development and data protocol development . 41 4.1.1 Tissues ............... . ................ 41 4.2 Imaging ................................... 42 4.3 Image analysis ................................ 42 4.3.1 Data Modeling ........................... 42 4.3.2 Algorithm Implementation .................... 44 4.4 Phase Two: Retrospective survival analysis of a cohort of breast can- cer cases ................................... 46 4.4.1 Followup data ............................ 47 4.4.2 Statistical analysis ......................... 47 5 Denouement 51 Glossary 52 1.1 1.2 2.1 2.2 3.1 3.2 3.3 3.4 3.5 3.6 4.1 LIST OF FIGURES Treatment effect prior to stratification .................. 4 Treatment effect after stratification .................... 5 Breast cancer incidence, 1973-1998 .................... 12 Breast cancer mortality, 1973—1998 ..................... 13 Microscopy Image Example ........................ 30 Map of population density as analogy to microscopy images ..... 32 Conceptualized effect of confocal microscopy . ............. 33 Idealized imaging of proteins ....................... 36 Summary of quantification strategy .................... 37 Illustration of the moment at two pixels ................. 38 Histogram of a representative image .................... 44 vi LIST OF TABLES 2.1 Her-2 and Survival in various cancers ................... 18 vii Chapter 1 Introduction This work, derived from a grant application written to obtain funding for this project, describes an approach to analysis of health-related events on the molec- ular level, which is the aim of the field of Quantitative Molecular Epidemiology. The immediate context for this discussion is a proposed study designed to analyze specific molecular markers of breast cancer prognosis (Her-2 and Her-3 cell surface receptors). The study will test the hypothesis that measuring Her-2 alone is an im- precise prognostic measure due to the fact that Her-2 and Her-3 must interact to elicit pathophysiological activity in Her-2 positive human breast cancer. Instead, quantitative measurement of Her-2/Her-3 interaction is proposed as a more pre- cise predictor of breast cancer prognosis. Such non-covalent interaction of two pro- teins to form a single functional unit (by virtue of their complementary structure and function) is referred to as dimerization. Specifically, when the two subunits are different proteins, the phenomenon is referred to as heterodimerization. The goal of the present work, then, is to quantify Her-2 and Her-3 heterodimerization and test whether this quantitative parameter represents an improved prognostic marker in breast cancer. Unfortunately, these proteins are too small to image in- dividually using light microscopy. Instead, fluorescent markers will be used to image the density of these proteins aggregated over very small units of space (pix- els), and the correlation of Her-2 and Her-3 density over space will be quantified using strategies detailed below. This correlation over space, or colocalization, will be used as a computational proxy for measuring heterodimerization directly. In the balance of this chapter, Quantitative Molecular Epidemiology is further defined and placed into context, and the specific aims of the proposed study are enumerated. 1.1 Human Tissue Research: a growing imperative. In recent years there has been an increasing emphasis on the use of human models and specimens in cancer research (1). Numerous gene and gene product markers of pathologic processes have been identified, including tumor suppressor genes such as p53 (2) and oncogenes and their products including telomerase (3), VEGF (4), and Met (5), to name a few. These markers provide substantial opportunities for new stratification and analysis of both past and future data. This family of identified markers will certainly continue to grow over the com- ing years, and as it does, the potential for medical progress presented by human specimen analysis will grow as well. The increasing library of markers will allow not only investigation of the role of single components, but elucidation of complex collaborations between genes and gene products that produce normal or patho- logical responses via (possibly competing) multi-factorial pathways. Archived samples—when connected to treatment, outcome and survival data—will allow these markers to be analyzed in sophisticated diagnostic, prognostic, and treat- ment effect models. Diagnosis and prognosis on the basis of visual examination of tissue on the microscopic level is well established (6). However, objective, quantifiable criteria require further development (7, 8). Subjective grading by pathologists (even those trained at the same institution) exhibits a high level of inter-observer variability whether using common pathological classification systems such as WHO, Bloom- Richardson, or Nottingham, (8—10) or more complex systems such as Bayesian Be- lief Networks (11). This impedes the transferability of prognostic models, and may lead to either type I or type II errors in models using these subjective interpreta- tions as predictive variables. The potential inherent in digital image analysis as a more quantitative, objective, and reproducible histologic tool has been described (12—14). The lack of objective, quantitative methodology has been cited as a signif- icant challenge in the emerging era of molecular epidemiology (15). As the molecular markers of disease etiology and progression become increas- ingly defined, our thinking about the epidemiology and treatment of disease must evolve. Consider the following hypothetical example: Figure 1.1 shows some hy- pothetical results from a clinical trial, which could be considered to indicate treat- ment failure (i.e., affirmation of the null hypothesis) since the odds ratio is essen- tially 1 (95% CI: 0.88-1.35), indicating no treatment effect. However, a different picture emerges in figure 1.2. The patients have been retro- spectively stratified on a ”new” hypothetical marker. It emerges that the treatment may be highly efficacious for those patients who were positive for the marker. If this post-hoc finding was confirmed through one or more prospective studies, fu- ture patients could be tested for this marker, and treated accordingly. This is more than a hypothetical possibility. Genetic profiling has already allowed the iden- tification of differential prognostic groups among B-cell lymphoma patients (16). Additionally, examination of cell lines from the National Cancer Institute’s in vitro drug screen has identified genetic profiles with unique drug susceptibilities based v- 'f ._‘..", .' ,‘.« '~-. 'IA ' x ' '-"\.‘ ._ "v n 0‘ 'r ‘1‘ .5..~"."'.. ’1' \ 4" . 1 v"": -5 , ‘ . J" ' .Io ‘ ‘ a "A " ' - \ , ‘- ". h.I 1". '"J—I' v ' l ' ‘v’a'fl-v ‘-’-f?-l-“'-III-'n in find- l—‘o-k'fi'wp'u-t |“FF.I:'WTJ" lu‘dfl'l'w a‘rfiu '5‘“ Marc-e :11 ("FF 47} 5 year survival Survived Died e :3 Treatment 320 180 500 >4 "0 3 Placebo 310 190 500 (I) 630 370 Odds ratio: 1.09 (95% CI: 0.88-1.35) Figure 1.1: Prior to stratification, it appears that the treatment had no effect (the odds ratio is essentially 1, (95% CI: 0.88—1.35). on activation of the ras oncogene (17). And as will be discussed fully below, Her-2 expression has been used to identify differential prognostic and treatment groups. This sort of approach is more akin to our current methods of treating infectious diseases than traditional approaches to cancer treatment. While the approach to infectious disease treatment has long emphasized the identification of a specific etiologic agent and treatment with a compound of demonstrated efficacy against that agent, cancer treatment has typically taken the form of generalized assault against the entire patient with the hope of some differential damage to the puta- tive cells. Molecularly targeted treatments hold great promise in elucidating opti- mal treatment modalities for specific groups of cancer patients while minimizing side effects. Even a retrospective look at treatment and diagnosis with new mark- ers could be very useful and reveal new information from past studies. Recent evidence suggests that even decades old specimens contain useful biological data waiting to be tapped (18, 19). This work describes an effort to further develop the analysis tool set that can be - ‘fi . .. -7 -. . r . . , 'r . . .1 . ' , .. . 7. _ . . .. - . r .. . - _ w 7" . D"? a, T. ' ‘.'. l "’l" ‘ '." '. Po 1 .‘L l. u .I. . 'L*r ' ‘v ‘ ‘1 . ’.’ , Z ' l . . ‘. r Q' . t I I ‘r — ‘ ' u' ‘u. . ‘ A a ‘ . ' '0 ‘ ’ . - ‘ [I ' h i l b D T ’7'- , .- - , w. . . . .. . ~ . -w....-~ .. . .7 . . 7,. .. . .. . . , .. . ~ . . . , . . v . . - . . . . ‘ -' .vfilw“ '2‘“ it 7‘. “..,~_ ' «war . -. .- ,7 . .13 . F. )\ .-- v ‘ .1, . .5 th¢5posfive{HIHIMl _ 5 year survival Survived Died E E Treatment 60 10 70 >1 1) 3 Placebo 70 40 110 U) 130 50 Odds ratio: 3.43 (95% CI:1.79-6.56) HYP negative: 5 year survival Survived Died E E Treatment 260 170 430 >, 'U 3 Placebo 240 150 390 a) 500 320 Odds ratio: 0.96 (95% CI:O.76-1.21) Figure 1.2: Stratification on the new genetic marker reveals a pronounced effect among the subgroup of marker positive patients (95% CI: 1.79-6.56), as opposed to the marker negative patients (95% CI: o.7e1.21). applied to human tissues to elucidate these molecular pathways and quantify these phenomena for use as epidemiologic markers. A quantitative, automated analysis tool to measure the interaction of two or more proteins (e.g. dimerization—the for- mation of single functional unit by two identical proteins—or heterodimerization— the dimerization of two different proteins) within human tissue samples is pro- posed. Furthermore, we propose to apply this methodology to quantifying the formation of Her-2/Her-3 heterodimers (via measurement of colocalization—the correlation or covariance of protein density over space as represented by fluores- cent microscopy images) on the cell surface as a marker of breast cancer prognosis. 1 .2 Quantitative Molecular Epidemiology Defined Molecular epidemiology seeks to apply the tools of cellular biology to measure— ment of disease related states and events for the purposes of epidemiologic anal- ysis. Quantitative molecular epidemiology takes this process one step further by quantifying these cellular phenomena as opposed to relying on subjective inter- pretation (an example salient to the present work is pathology, wherein molecular markers imaged microscopically may be interpreted subjectively by the patholo- gist or quantitatively using computer algorithms). This emerging field of quantita- tive molecular epidemiology promises to expand our understanding of the molec- ular bases of disease, and treatments that can specifically target these pathways. Molecular epidemiology represents a paradigm shift in cancer research. As op- posed to the former emphasis on anatomical site as a classification scheme (e.g. ”Ductal Carcinoma In Situ”), today we are increasingly able to classify tumors based on their genetic and molecular fingerprint (e.g. ”Her-2 overexpressing, es- trogen receptor negative”). This allows us to direct treatments not at the imprecise target of anatomical site, but at specific causal factors elucidated through knowl- edge of the molecular pathway leading to carcinogenesis, invasion, growth, and metastasis. Therefore, molecular epidemiology affords two critical opportunities: 1. The ability to classify disease based on molecular fingerprints, allowing for classification and stratification that has a high degree of granularity and phys- iologic relevance. This leads to an improved typology when studying cause, effect, or treatment. 2. The ability to target treatments at specific cellular events that lead directly to pathologic events Molecular epidemiology has already elucidated predisposing genetic polymor- phisms associated with increased risk (e.g., the BRCA family (20)) as well as pro- teins involved in carcinogenesis such as the growth factor HGF and its receptor Met. These molecular markers have significantly improved the resolving power of prognostication over histological interpretation alone. Camp, et a1. (21) have demonstrated that high Met expression in node negative breast cancer patients is associated with increased risk of dying within five years (RR=5, p=0.03) inde- pendent of tumor size and exhibits a synerigistic relationship with nuclear grade (RR=33.4, p<0.001 for advanced nuclear grade combined with high Met expres- sion). Similarly, in a land mark study, Yamashita, et al. (22) reported that in mul- tivariate survival analysis of 258 breast cancer patients, high HGF expression was a more potent predictor of poor survival (RR=3.3, p=0.001) than any histological indicator or even nodal status (RR=2.5). Such an approach will surely have much to offer the ongoing study of breast cancer. First, exposures such as exercise, chemicals, and hormonal events are up- stream components of causal chains leading to downstream molecular pathways in the cells giving rise to tumors. By elucidating the molecular details of these pathways, exposures can be better understood and treatment responses better de- signed. Furthermore, existing risk factors explain only a fraction of total breast cancer risk. Much of the remaining risk may only be explained through analy- sis of molecular events that either lack measurable upstream causes or manifest heterogenous cellular responses to known exposures. The story behind the HER-2 receptor illustrates wonderfully this full spectrum of molecular epidemiologic effort, from molecular cause to molecular treatment. It further highlights new tools that must be developed in order to practice this kind of science. Some of these tools have been developed, and some will be refined through the work described in this proposal. 1.3 Specific aims Through this study, methods for quantifying the data contained within human tis- sues will be refined, and details of the role of Her-2 and Her-3 in human breast cancer will be elucidated. A molecular epidemiologic parameter will be devel- oped that will quantify the colocalization of Her-2 and Her—3 protein densities in breast cancer cells as a proxy for quantifying heterodimerization of these proteins (which are too small to image individually using light microscopy). This measure will be used to test the hypothesis that heterodimerization of these receptors is a more precise measure of pathophysiologic activity and, therefore, prognosis than measurement of Her-2 expression alone. The power of this parameter as a prog- nostic marker will be compared to existing methods. Specifically, this project has the following component aims: 1. Test the relative power of two approaches to quantifying Her-2 and Her- 3 colocalization in cancer tissue—joint moment of standard images analysis and correlation of globally standardized images analysis—as compared to each other and traditional methods. 2. Test whether variation in a quantitative, continuous variable describing the level of Her receptor colocalization in breast cancer tissue is a significant pre- dictor of time-to-event in univariate and multivariate models of breast cancer outcome. 3. Develop an algorithm and software that are as simple in design as possible to promote the extension of these findings to clinical practice. In the following chapters, a synopsis of the literature pertaining to the growing role of quantitative molecular epidemiology and the role of Her-2 in breast cancer will be presented, followed by a description of the design of this study and the proposed methods to be used to execute the study. Chapter 2 Her-2 and Breast Cancer: Epidemiology and Cell Biology Her-2, also known as neu and c-erbB-Z, is a proto-oncogene that encodes a 185kDa trans-membrane cell surface receptor (p185, or Her-2) with tyrosine kinase activ- ity. (Note that throughout this document, gene names are placed in italics, whereas their protein products—typically bearing the same name—are rendered in plain text. I.e., Her-2 is a gene, whereas Her-2 is a protein). It is a member of the EGFR type I family together with epidermal growth factor receptor (EGFR), Her-3 and Her-4. The story of Her—2 is highly illuminating in terms of clarifying the potential and the pitfalls inherent in the dawning era of molecular epidemiology. The science surrounding Her-2 reveals the potential these molecular markers offer to quantita- tively differentiate between differing prognostic and treatment groups. In 1974, a group of closely related oncogenes was identified in rat neuroblas- toma cells(23), and was therefore termed neu. Over the subsequent decade, it be- came clear that this gene family encoded p185 (a common tumor antigen) (24-26), was homologous to but not the same as the erbB gene that encoded the epithelial growth factor receptor (27, 28), and encoded a growth factor receptor tyrosine ki- nase (18). These genes were therefore labelled erbB-2 and erbB-3. Neu/erbB-Z was capable of transforming certain cell lines into a malignant state, and was found to 10 be activated in many rat neuroblastomas (29). Subsequently, similar erbB-Z related genes were probed from human genomic libraries and isolated independently by two groups (27, 30). The human analog was termed human Human Epidermal- growth-factor—receptor-like Receptor, or Her-2. The matter reached its denouement with the revelation from sequencing analysis that erbB-Z, neu, and Her—2 were not only related but also, in fact, one and the same (27, 30). The gene had already been found to be elevated (i.e., amplified) in human mammary carcinoma (28). Growth factors and their receptors are integrally linked to the cell cycle, and regulate apoptosis, locomotion, differentiation, and division. It is important to note that Her-2, like numerous other growth factor receptors and their ligands, are essential components of normal development and physiology. These proteins may become important in the development and progression of cancer, however, when they are mutated or over-expressed (either through gene amplification or break down of normal regulatory pathways). Therefore, as a growth factor receptor can- didate, both the Her-2 gene and its protein product quickly became the subject of intensive study to discover what role, if any, it may play in carcinogenesis. What has been found makes clear the fact that molecular epidemiology will continue to improve the outlook for patients even while presenting unique challenges to the field. 2.1 Descriptive Epidemiology of Breast Cancer and Her-2 expression Metastatic breast cancer kills more than 40,000 women each year (31). The inci- dence of invasive breast cancer has been rising over the past several decades (32) as shown in figure 2.1. Mortality has decreased nearly 20 percent over the past decade from 15.5 deaths per 100,000 cases in 1988 to 12.6 in 1998 (figure 2.2). Both 11 these facts can no doubt be explained in part by improved screening and the re- sulting biases in detection and lead time (33—37). Breast Cancer Incidence by year of diagnosis as V .N N O O 01 .N o (a) 9’ o 1 Cases por100,000 population {3 S O O 12.0 i‘ T ' I If TI ' I I T _T T r j I I I fi' I I I I I T —I 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 Yearofdlagnoclc Figure 2.1: Incidence of invasive breast cancer per 100,000 females (age adjusted to 1970 standard population) by year of diagnosis. Source: SEER cancer registry data (32). In addition to invasive breast cancer, a comparable number of women are af- fected by non-invasive ductal carcinoma in situ of the breast. The incidence of this diagnosis increased 14 fold between 1973 and 1998 (from 2 cases per 100,000 women in 1973 to 28 per 100,000 in 1998) (32, 38). One area of extensive investigation is the role of the Human epidermal-growth- factor-receptor-like receptor (HER) family receptors, particularly Her-2, in breast cancers. A 1998 review of 22,616 patients indicated that about a quarter of breast cancer patients express Her-2 in tumor tissue at detectable, though not necessarily elevated, levels. Also of interest is the proportion of patients who exhibit higher than normal Her-2 levels. Elevated protein levels may occur through two path- ways. One is gene amplification (an abnormally high number of copies of the 12 Breast Cancer Mortality by year of death 16.0 _‘“‘“—’"‘— ______,_ 15.5 i 15.0 14.5 . 14.0 « per 100,000 population .5 01 Death: .s .3 10 OD 0| 0 12.0 I I fi' T I I I I ' I l I I l I I I T I I : I I 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 Yearotdoath Figure 2.2: Mortality from invasive breast cancer per 100,000 female population (age adjusted to 1970 standard population) by year of death. Source: SEER cancer registry data (32). gene in the genome as detected by microarray, fluorescent in-situ hybridization, or southern blot), and the other is over-expression (the increased production of transcription products from either a normal or abnormal amount of mRNA, as detected by western blot or immunohistochernistry ). Her-2 gene amplification oc— curs in 20-30% of breast cancer cases (39, 40). Immunohistochemistry reveals that over-expression of the Her-2 gene product occurs in about 20% of breast cancer cases (41). 2.2 Analytic Epidemiology of Breast Cancer and Her-2 ex- pression 2.2.1 Breast cancer risk factors According the American Cancer Society, the following are currently accepted risk factors for breast cancer (taken from www.cancer.org): 13 . Age. Less than 1% of breast cancer are diagnosed prior to age 30. . Genetics. About 10% of breast cancer risk can presently be accounted for by known gene mutations. (Note that in the current work we are concerned not with a gene mutation, but with excess protein production from a gene, such that Her-2 over-expression and / or dimerization would not be included in this figure). . Family history. Having one first degree relative with breast cancer increases a woman’s risk of breast cancer 1 fold, while having two first degree relative with breast cancer increases a woman’s risk 5 fold. . Personal history of breast cancer. In addition to recurrence, a woman with breast cancer is at a 3-4 fold increased risk of developing a second breast can- cer that is not a metastasis or recurrence of the first tumor. . Race. White women are about 1.2 times as likely to be diagnosed with breast cancer as opposed to African American women (42). . Previous breast biopsy. Prior diagnosis of non-malignant hypertrophic breast disease is associated with a 1.5-5 fold increase in risk. . Previous breast irradiation. Prior irradiation for another cancer is associated with increased risk. . Menstrual periods. Early menarche or late menopause is associated with a small increase in risk. . Oral contraceptive use. While unclear, recent reports suggest that oral contra- ceptive use may be associated with a slight increase in risk (RR 1.2, p<0.05) (43). 14 10. Not having children. Having no children, or having a first child after age 30, is associated with a slight increase in risk. 11. Hormone replacement therapy (HRT). HRT has been shown to slightly in- crease the risk of breast cancer, however this risk is perhaps offset by other health benefits. 12. Alcohol. Consumption of 2-5 drinks per day is associated with a 1.5 fold in- crease in rick of breast cancer. 13. Obesity/ High Fat Diet. Obesity is associated with increased risk, and some ecological studies suggest that high fat diet may increase the risk of breast cancer as well. In addition, physical inactivity, environmental pollution, and not breast feeding have been suggested as possible risk factors by some studies. Of course, all of these risk factors account for only a fraction of the risk of breast cancer underscoring the need for ongoing research into the causes of breast cancer. Reviews of known breast cancer risk factors demonstrate that only 30-40% of breast cancer risk has been explained (44). 2.2.2 Her-2 is elevated in a variety of cancers. Information from published studies reporting on the relationship between Her-2 and prognosis for various cancers is presented in table 2.1. The various studies used a variety of measures to measure the effect of Her-2 expression on progno— sis, including Cox regression models, Kaplan-Meier models, and simple t-tests be- tween 5-year survival rates. For this reason, interpretation of the results of these studies must proceed with care. 15 Some general trends can be deduced from the literature summarized in table 2.1. It is unclear what the role of Her-2 may be in Ovarian cancers. While several of fairly large studies found over-expression to correlate with worse prognosis, including the single quantitative study, numerous other studies fail to bear this out. Her-2 seems to play a clearer role in salivary gland tumors. Interestingly, the largest study (45) found only a ”suggestive” association. However, the small num- ber of Her-2 expressing tumors (2 of 201) calls into question the methodology or definitions used in this study, as does the fact that all the other studies (with the ex- ception of one very small study) found a correlation between over-expression and decreased survival. Similarly, the majority of studies examining the role of Her-2 in bladder cancer report a marked decrease in survival among patients with Her- 2 expressing tumors (although the failure of many of these studies to distinguish between expression and over-expression is cause for some skepticism). However, it should be noted that one study found a positive correlation between Her-2 ex- pression and survival. The results with regard to Esophageal cancer are more ambiguous. Nonetheless, these studies, taken together, seem to suggest that Her-2 is a prognostic factor in Barrett’s adenocarcinoma, particularly those of an intestinal cell type. Equally ambiguous are the results pertaining to renal cancer and pancreatic can- cer. However, the report of correlation of her-2 expression with decreased survival in a number of large studies should motivate further investigation of the role of Her-2 in these cancers. Expression of Her-2 appears to be an important predictor of outcome in gas- trointestinal cancers, particularly those of advanced intestinal cell type. 16 Her-2 does not appear to be a major predictor of Lung cancer survival, with the possible exception of small cell lung carcinomas. It will be important to con- tinue to monitor the role of known molecular markers in lung cancer given that the distribution of lung cancer pathophysiology may evolve over time as a result of changing smoking behaviors. An overarching message that can be taken from these studies is that both method- ology and definitions vary and involve considerable subjectivity. This fact under- scores the importance of an approach that is not only molecular, but quantitative as well if we are to fully exploit the information inherent in the patterns of molecular pathways with respect to disease processes. One hazard of multivariate analysis in the context of molecular epidemiology bears emphasizing. When incorporating molecular markers with environmental, demographic, or ”macroscopic” clinical indicators, one is virtually guaranteed of identifying factors that share a pathway—one factor upstream, another down- stream. Careful consideration must be given in selecting interaction terms in a meaningful way to ensure that covariance of upstream factors do not lead to elim- ination of downstream factors when those downstream factors are the predictors of interest in the study. 17 1352:0523 522d @2232 35:3 Eucau 55.5 322%. oz gasses... :mm 2 a. a9 concu§_d6m 550 555 12.55 oz 5.: 2 28282 Ema 2 a. as 52.3 c2235 595$ BEE—om oZ coaookmxoogo UT: 3 3 RB soouev 3233 gas .56ch :2.ng _a>_>.3m c.5665 8.83 an .m> n _o>m_ “.2965 $52 83% e mm 39 Catt—5 .0: “an 6930 5:: «5228.8 coamfimxagov .8280 5530 3333,»... 02 co_mm8mx8o>0 Or: a NV :9 aaoseeo 5.530 5.23 oz eoeaoeeaca .22. 5e : mm as coup—mu 52.230 33.3.26 02 connoE—mfio x5 mun— wn 03 $3 .5228 5:56 3.95:? 2 co 5 M680 .9655 02 wSESm wcobm Um: mm m2 :3; .Adodum EPA—ace 228.; .53: co «egocEw_m 958??“ Go as? :06 3298353 c353 5:530 3:055“. beam REP—3m oZ wEESm xaoZEELoUOE Ow: {mm 9. Rev $0.980 :86an 45:32.. 15.5: o. venous—co 5858 53.85 3.92% poo—235‘ 13:56 :85 92:05 am .m> m” .868..on £92 0:. MN mm 3$ >m<>O Emocmfio 2:850 eczema—:2 coimoaxocoO cocEcmp Egomxm 9:232 + meoI 2 “mm fittsm 9 conflmtou decency— sfio oacoaboa doe s82 28.855663 case: 252m {mam 68533»: new .5 Eoufiocsofi .EmE ugébamflocsg Um: ”econoSEnn< .3636 c8253 wart? coca—awe 2: 5.3 2.53833 Eon go =a .88? S no} 22829830.. 35 802 .356 9.: 5. voted“: “Cots 2.: 21:95 no :92» a «Seaweed :0 Como Sandra—«=6 do ooze—tom be a €355?» $3 wEocw 2580a a 98:3 65 axe: 3&6on n. ~29, onoocwoa do one; gunman 52w 2.: go: .5 82:5 .228 + «L03 2: 5 52» 0.3 8:38 6033 338%»: 63: 2.53 8.385 min—=6 89:5 .2528 “5:333 93 5. 53m .acanmov can 85¢. -38 35.3 St 3 Bacon no 3580.... «.32 Bo? 0:3 8833 Lo 3855 05 .3 330:2 Exam 3 ~Zv o5» aunt 9a. 699 590 .3 £356 noon; 5 133:» v5 «Luz Ad 2ch 18 mm00<4m 28583 8:: .0953 833% 02 8838880 0E o a $8 88% 8953 85 8 5.8588 085838832 880 .885 5:: 08.5.. :02880 >2 0:: 0 mm K8 852w 828m 88:. 8 «:8:an 0:886 $0.on 6.5m .88 5 885980088 880588082 mum .m> 733.3» 88» m anon :08880 >3 01— 2 mm 33 .898 ~:::w=§.:-:0: 2 8.89:8 88.30 8.3»: 88.0"“: -5: 8 83888.3— 8; 3.8 :23208 «-85 NNN .9» 83.53 882 .228 85 8 89:3 >828» 852 8333 :38:— nficofi 0.: :0_m§mx88>0 UT: 3 Na. 88 8063 083 >853 ESE-E 8>z8mwzm :03880 %:< UT: ~ Sm 83 2803800 :80on mm” .9» 13.sz .Nos .9» 122:» :ozmuE_mE<\wE:_3a 22w 8958 00 «5898 2063382,: .323 .538 2.85 a 92.86 3.86 «a 258...: 8:85 a 5 mm 2.3 DZM<>E:3m .m> .be 838m 88% a A :onab:uu:00 0:850 :< UT: on mm ANS 9:05.88 :8 88298: «waam 88>0< 8225 02 :2880 >:< 05 m; on :8 5:05.88 8885 0:: 8:88: Edcwzfi: 30— 8 895.: 8.85.8 8:30 822:6 02 :ozauazmfia >2 ImE 9 «R 58 :88: 83 max 05 $83 8838: 85 :08335 88:88.... 3 80583 5:80 832% oZ 02 838883808883 Emfiaom\um_ w\ 8 8: any 38:00.8 8: 8:: £638: 328.»: -12: :0 88233-8 828m :0 :038588 0882-883— 3 2% 2:523 .856 3.8%... 5:50 1323 Oz 80-3 cofiemxemééonsaai 0:38.. a; :23 $9 3568 #8 88:3 5:88 88588 .0355 733:3 m> $3 8333 88» m :038888-4. 85 Eufisom mm mo" va 388-0 25:83 8 8.5388 858mg 8885 .1333 889: -8 YE? :8 «.8: 8820 80:8 8530 8323 «5:9: 3 395 E :03880 >5 DE. 3. wow 33 £89.85 «E880 cos—BEBE :2889880 8:88: 8:8:xm 8.3882 + W81 2 8M 828m 0. :08—880 8:588 £88 88> E fiziam :5 «.8: AN «Bah 19 28% SEE 2 «922889 “on ma? c0683 fichm £me .5 «-33 ”302 .mquzguocuVa 333.3 735:5 .n> 123:» .32» m cg wEESm 88952 UE n 8 88 mDU94. U5 3. omm 33 49:33 32895 *0 9506th as? $553.» 75$ 2558 $62 $85.88 =8 58.93; iatsm A883: awonxm cofieaxemé US 8 8“ ac .ufifiw .563 5:5 83% 33 50235—9 -Ec N15: ”302 .wEoEBmu :8 _§o_=§h $323 02 ”BEBE—mg xUm «N No :5 £853 3233 =8 3.2225 333.3 oz 55% $233: uE 58 SN 8: def—0&2 .0: 22: no.2 Amodev govoE .25 322%.. ”802 £80583 :8 30:85:. Ezism E. 323:» commfluun cozufimv ~93 UE mN mm Amt .owflm 65 avg» 5:5 Sooodumv figs—9:8 03¢ Eggnog—ms; «Luz .votom 862-6293. .2 «o: 83.. 13?:6 ”202 “.858 $355 Elfizm .3 3232.. .885 conauasmfilx MU; 9 mm :b .vutommu 8: 8‘8 _a>_>.5m 6qu x89 Amodvmv «may «:8 m0; .939? 3 U385 E85... .3233 wings fiztsm .3 123:» Btozm c2833 >91 0:. R mw 85 68.3: :06mb05 .8 «E38 3 1333.8 6 .n> .1333 9:53» .0: 6qu .2858 :3 35:35.: F3955 1323.. :85 was?» ad 198$ .5 88282.. DE N 8 $8 macawflo 3.03:0 533535 comman—xfloa sowing 233m 93932 + WEI Z wax 125m 9 cons—utou .3: £28 .9838 3555 E 1.:sz v5. «.8: AN 03¢ 20 2:00.88 :00 9.06033. .a>.?_ 0m 02 cofimflmxm U2. mm 3 9.0. mac..a0....mfic\w:.53m 0805230000.. 03:30. .m>.>.=.m 02 02.802 .05 .00 \ U2. .R. om . 3.0. 98.3.. ate .8050 was. :8 2.002 .02sz .m> .355» .3» m Gem... 00.80.00.830 U2. mm EN .8. 3.48 5.3 :0: -0525. .58...in $050 was. :8 =~Emé02 .6233 02 :2398530 U2. 8. RH 80. 802503. was. 05 .0 «505880000< 10.95 20233 8» 00.829905 U2. 8 Nu. .0... 302503. .356"... .50ch was. =0U zuEm \ .0233 gin 0:8 935:. 003029.805 U. .. wow .wa .88.?3 .930 «5N .u> 1223 .850 was :6 :20 12:3,. 5.86 2.6 m msssm .323. 0:. z B. .3. 023‘. 3:8. 38%.. 5.2:. 26% 3?. .8 .0: 8.3. ”802 .afiocfiuaU =0U .003. .5556 .0052: 3.0.2.503. 638 52000 030.3 .339558. <2 ww. 3w. «6050;000:000 .23. .0223 02 0030300 >2 U2. N. mm. Am... 8:09:63. anEBmU ..0U .58». \ .9555 02 .86.. soamamxm U2. om on .3... >mZQ.v. .9133. 0. 0828.50 3.500%. .005 .0: :5 dem 0. 030.230 v.03 00.3m .5 NE... “202 60305.. .auwczaogohmaw 00:03.. 9&2008 2.. .0 «60588053.. .0223 02 005899.805 U2. 0. .v 8.... 60:05.. utummwowa £0.03 05. 32008 05 .0 0505800053.. .u>.>.§m 02 0030200 >3. U2. 3 N.‘ a... 66305.. 0.0.0.0.»...38 :00: 0202.3 0.0 12>...» 0. 8.3.8.00 05 8.530: ‘35; .302 .05. 0.3% 0. 030.8000 no? 00.80.... .20: 301%....an co... .xm man... ”302 690058300000 9:83.. .m>.>.3m u 530 .040... 02.0.8 00.89.90.830 U2. m. ow .5. 385.05 0500.30 :0:8c....6< no.82nx805 003.530 33805.. 9:532 + «.002 Z .3. .a>.>.5m 0. 0020.2.0U .33....28 .3356 030.53 S. .9:an 93 $32 ”fim 030... 21 I] 28.323: a“ .n> 13.53. magnum Emto .826 .0 89:3 533:3 Witsm :38 92:08 n macaw 9 28982 U3_ 3 8 Ame: «50583953 33:6 Uzmflucmm _m>_>.3m oZ conuflmv >c< um: um um: A3: SHEEP—305.3 unmflucsm Eatsm oZ map—um 9 39532 \ be UIH 3 \ no on A8: agave can .a> «3.3.5» Sung unsung—am fittsm :35 mxuv va 339mg EEwm Ammmmocsfiafi um 93 ANOC . m .9655 2560a NLmI ~52 202 680583 “.530 iitzm :36 5:08 8 wEESn «53:82 Ur: : no So: .6226 :o_oU 332.6 02 c2322. be UE mm om" 8o: £60588 Rifficzmflfi wwfim Bung? 2 3:5: mm? coca—9:00 9:95.89 953.0 _m>_>.3m A80: 93 wag—3m 283832 UT: NM mm Aog .355 gave -EuE ho uwfimfimoiuv wfififlm magi >5 mom .a> 1233a wESEm 5:3 «5.22:6 _a>w>..:m £80583 3.330 :38 mxav 5mm BHEEQE \ uficmcloiu UE vam cm Awov .3»...— mod «5 a 289:»; 6: £3 @325 $5 :5 ._n>_>._=m 83:55 $539 95.5 a 3305. v.38 958m 28563.4 awn—0588 £3me 73:56 TV 02 MEESm mamBEuZ U5 w mm 98 4 1333 953238me =86 9.8; z 835x230 uE 8 mm aé amocwfio 3:830 cozmuczmg cggmxflgo coficamu Eamonxm 23me“). + Nme 2 Ex 3.555 2 configou 3:528 .9883 30C...» E 1333 .23 $3: Hm mick 22 2.2.3 Her-2 may distinguish between prognostic groups. A year 2000 review of 47 studies presenting data on Her-2 abnormalities (ampli- fication or over-expression) measured in over 15,000 breast cancer patients indi- cates that Her-2 overexpression or amplification is correlated with worse prognosis (106). Several smaller studies in the late 19803 had failed to show any Her-2 effect in multivariate models, or even in univariate models in about half of the studies. However, since then, the overwhelming evidence of numerous large studies sug- gests that level of Her-2 expression is an independent predictor of poor outcome, though the magnitude of the effect is difficult to quantify due the variability in the study designs (and for this reason the review does not attempt to quantify the magnitude of effect). This same review points out that immunohistochemical methods were more likely to result in negative or univariate only results compared to gene-based meth- ods (9 studies vs. 3). However, gene-based studies can not identify overexpression, only amplification, and they can not quantify the interaction of protein products. Nonetheless, this finding underscores the need for objective, quantitative methods to be applied to immunohistochemical techniques. 2.2.4 Her-2 may distinguish between treatment groups. Several studies have indicated that Her-2 status is a powerful predictor of tamox- ifin resistance (107-109). This is likely due to the fact that the Her-2 pathway initi- ates kinase-dependent cell proliferation pathways independent of estrogen recep- tors blocked by tamoxifen. 23 2.3 Her-2 exhibits complex ligand-receptor relationships. Despite the homology of Her-2 to EGFR, EGFR ligands such as TNF-alpha and EGF do not activate erbB-Z. However, the following Her-2 agonist ligands have been suggested: 1. Gp30, a 30kDa TNF-a like protein, homologous to heregulin (110) 2. 50kDa factor (111) 3. Neu differentiation factor, also known as heregulin and p44 (110) The role of Gp30 and 50kDa factor are unclear. Heregulin acts as an agonist ligand in breast and colon tissues, but not in ovarian tissue, despite demonstrated comparability of the amino acid sequences of the receptors (112). This indicates that the presence of Her-2/heregulin interaction alone is not sufficient for stimu- lating the tyrosine kinase activity of the receptor (112). Further study of the Her-2 receptor has elucidated additional intriguing characteristics of its ligand-receptor relationships. First, it was found that divalent antibodies directed against Her-2 can induce its kinase activity, but monovalent fragments do not (113). Second, molecular studies of Her-2 indicate that activating mutations observed in vitro in- volve trans-membrane domains likely involved in receptor-receptor interactions (114). Third, disruption of Her-2 dimers by herceptin (see below) inhibits tyrosine kinase activity and growth of transformed cells (115). These findings suggested that Her-2 is activated via dimerization. Subsequent work has demonstrated that Her-2 preferentially forms heterodimers with Her— 3 (116, 117), that these heterodimers exhibit enzymatic activity not present when only one or the other receptor is expressed alone (118), and that these heterodimers can lead to neoplasia when activated by heregulin (119). It has been shown that 24 Her-3, and not Her-2, binds heregulin (120, 121), explaining the lack of heregulin induced kinase activity in ovarian cells lacking Her-3. Interestingly, Her-3 has lim- ited kinase activity (122) but numerous ligand docking sites, while Her-2 has sub- stantial kinase activity but no ligand of known pathophysiologic importance. The formation of a heterodimer results in a catalytic unit with high ligand affinity and high kinase activity. Unlike Her-2, Her-3 and Her-4 amplification has not been ob- served, indicating that the availability of Her-2 kinase activity is rate limiting. It is presumably for this reason that the limited evidence available suggests that Her-3 expression level is not associated with cancer risk (reviewed in (123)). Nonethe- less, Her-3 targeted treatment has been suggested, with an aim towards interrupt- ing Her-2/ Her-3 heterodimers (118). Interestingly, heterodimerization with EGFR inactivates Her-2 in vitro (122). Dimerization dependent kinase activity is not unique to Her-2 and Her-3. EGFR, another member of the Her family, also exhibits dimerization dependent phospho- rylation. However, homodimerization of EGFR is not sufficient for phosphory- lation, as induced homodimerization in the absence of agonist ligands does not result in kinase activity (122). 2.4 Her-2 has no confirmed human, cancer-specific muta- tions Her—2 gene amplification and protein overexpresion have been identified in a sub- set of breast cancers as described above. Until very recently, no cancer related Her-2 mutations (i.e., a change in the genotype, or DNA sequence, leading to a change in the protein sequence of Her-2) had been discovered in human cancer tis- sues (124, 125), despite the fact that oncogenic Her-2 point mutations and deletions have been identified in animal and cell line models (126—130). This is in contrast to 25 other human oncogenes such as ras for which human oncogenic polymorphisms have been identified (17). A population based study has suggested that a valine to isoleucine point mutation at codon 655 of the Her-2 receptor may be associated with increased breast cancer risk, but the relative risk was of borderline signifi— cance (131). Nonetheless, this continues to be an important area of investigation. In the mean time, the lack of clearly oncogenic polymorphisms means that genetic testing is presently of limited use with respect to Her-2, because there are no clearly prognostic mutations to look for as there are with other genes such as the BRCA group. Instead, novel quantitative techniques paired with statistical methods that fully account for the interaction of continuous variables must be applied to detect those patients who may be at increased risk of disease progression or treatment failure, based not on genotype, but on gene copy, expression level, and interaction (i.e., based on phenotype). The present study seeks to expand current abilities to measure protein expression patterns in order to better quantify phenotype in this fashion. 2.5 Her-2 is the target of the first receptor-specific cancer treatment. Antibodies directed against Her-2 result in homodimerization and sequestration of the receptor, and subsequent inhibition of cancer cell growth (132-135) via cell cycle arrest at the G1 phase (136). This observation led to the development of her- ceptin, an antibody based treatment with demonstrated efficacy in the treatment of breast cancer both alone (137) and in combination with other therapies (138). In these studies, 12—20% of patients with severe metastatic breast cancer unresponsive to traditional therapies responded to herceptin treatment (”response” was deter- mined by a blinded review board, but was not clearly defined), and the combina- 26 tion therapy was shown to improve response rate (17% vs. 42%, p = 0.001), prolong the response duration (from 4.5 to 10.4 months, p=0.01) prolong time to disease progression (from 3.0 to 6.9 months, p=0.0001). These findings demonstrate the full potential of the molecular approach to diagnosis and treatment: patients re- ceive treatments not based on anatomy, but based on the specifics of the metabolic pathways causing the disease. 2.6 Issues for study As can be seen, a great deal is known about Her-2. Its role in the molecular path— way of breast cancer has been identified, and it has been targeted in treatment intervention. However, there is still more information to be gleaned about this interesting receptor. As described above, Her-2 exhibits complex receptor-ligand interactions. Her-2/ Her-3 heterodimers exhibit biological function not exhibited by either receptor alone, and, therefore, these heterodimers should be the unit of analysis if the most accurate measure of receptor activity is desired. Therefore, methods to quantitate these interactions are required. In a static environment such as a fixed tissue section, interaction of proteins at the moment of fixation can be quantified by staining the proteins if interest with different fluorophore—conjugated antibodies and quantifying the resulting fluores- cent intensities within tiny spatial regions (pixels) and the interaction of these intensities. Careful consideration must be given, of course, to how the surgical handling and subsequent fixation and staining of the tissue may have impacted the spatial relationships of the proteins of interest. In this project, we propose to modify and further develop these methods for application to Her-2 and Her-3 interaction (dimerization), and use the resulting data in multi-variate models of 27 prognosis. Because of the nature of its possible role in cancer, Her-2 and Her-3 are particu- larly amenable to quantitative molecular epidemiologic techniques. In particular, the fact that Her-2/ 3 heterodimers are believed to play an interactive role in hereg- ulin activated kinase dependent cell processes makes these proteins an extremely interesting candidate for colocalization analysis. If successful, such analytic methodology could clearly be applied to other inter- esting protein interactions as well. 28 Chapter 3 Design As discussed above, Her-2 has no known pathophysiologically relevant ligand, and Her-3 has little enzymatic ability. We hypothesize that interaction of these two proteins is necessary for signal transduction. Therefore, measuring the interaction of these proteins, rather than the level of Her-2 alone, will more accurately portray the level of activation of this potentially neoplastic pathway. In this way, we may improve the resolving power of Her-2 expression as a prognostic factor in breast cancer. In the proposed study, we will use flourescently tagged antibodies against Her— 2 and Her-3 to allow microscopic imaging of the location of these proteins within cancer and control tissues. Quantitative algorithms will be applied to determine the level to which these two protein are interacting. These concepts are further discussed below. 3.1 Images as Data Sets Digitized images are stored as an array of pixel intensities, with the format of these arrays and associated file information arranged in a file format dependant manner. Each pixel represents a small square area of the image and the pixel is displayed as a unifome colored area (i.e. the pixel displays a single color, which in the case of 29 a grey scale image will be some shade of the color white). The digital image is then a mosaic of pixels. However, unless highly magnified, the digital image appears as a coherent image as opposed to a jagged array of square pixels thanks to the very small size of the pixel and the facilities of the human visual system. An example of this can be seen in figure 3.1. A Figure 3.1: A representative confocal microscopy image of intestinal epithelium. A. Green fluorescent antibodies against the Met receptor are excited and digitally captured. B. Red fluorescent antibodies against HGF are excited and digitally captured. C. The previous two images are inserted into the Red and Green channels, respectively, of an RGB image to allow for simultaneous viewing of the location (and possible colocation) of Met and HGF. D-E. The area indicated by the arrow in A for each image A-C is magnified to allow visualization of individual pixels within each field. Such a representation of data is similar to that employed in ecological studies. For example, smoking or population density may be measured and aggregated for some given geographic subdivision (say for example, census block groups). The resulting data may be visualized by color coding each geographic unit according to the level of the variable of interest. An illustration is given in figure 3.2, where 30 population density is represented by the gray level of each block group. This data could then be correlated to an outcome of interest—say, cancer rate for each census block group. In this example, the map is simply a way of visualizing the numer- ical data. In the same way, fluorescent markers may be conjugated to antibodies against specific proteins, and the resulting fluorescent intensity measured and ag- gregated on a pixel by pixel basis. The result is a digital map of labelled tissues, with each pixel color coded according to protein density. The numerical data un- derlying this visual representation may be analyzed statistically-just as with the geographic map, the graphical representation is simply a tool for visualizing the data. However, minimal attention has traditionally been given to the numerical data underlying the visual representation of the image. Instead, subjective scoring of the visual representation is typically performed by pathologists. It is as if we did not calculate a correlation coefficient between population density and cancer rate, but rather simply looked at the color coded maps and subjectively determined the degree to which cancer and population density were covariant over space. One final important concept is that of the image ”channel”. As described above, multiple markers may be applied to a tissue, and individually imaged (by using the appropriate excitation wavelength and filters to elicit and capture the fluo- rescence of only one marker at a time). Therefore, at a given pixel, data on the intensity of each marker can be stored. This can be conceptualized as a stack of images, each depicting the same slice of material but representing the density of different proteins. Each slice is referred to as a ”channel” of the composite image, and stores data unique to an individual marker. To return to our ecological exam- ple, one could have one map representing population density, and a second map which, while depicting the identical geographical area, is color coded to represent 31 5 Figure 3.2: A familiar example of a color coded image representing data (population density) ag- gregated for each subdivision. This is precisely analogous to microscopy images which allow us to visualize data (protein density) at each subdivision (pixel) of the image. Both types of images allow visualization of underlying numeric data. cancer incidence. The two maps could be printed on transparencies and super- imposed to visualize the correspondence of the two variables, or the underlying numerical data could be computationally manipulated. In the same way, multiple image channels could be superimposed to allow visualization of the covariance of multiple proteins (by means of, for example, a color image where the detected fluorescence for one labelled protein is loaded into the red channel, and another into the green channel—areas where both protein densities are high would then be bright yellow due to the interaction of red and green). However, more important than the visualization is the fact that these multiple channels can be used to com- putationally measure the influence of one protein upon another by analyzing the interaction of the multiple data points corresponding to multiple proteins for any given pixel or group of pixels. 32 3. 1 . 1 Confocal microscopy Confocal microscopy allows for exclusive resolution of a specific slice within a specimen. Conventional microscopy provides an image that is a representation of all material in the light path, resulting in blurring if the specimen is relatively thick or dense. However, confocal microscopy isolates a layer of interest and fluores- cence from markers above or below the plain of focus are filtered out (139). This allows for measurement of structure colocalization, for one may be assured that if the structures of interest are colocalized in the x-y (horizontal) plane, they are colo- calized in the 2 (vertical) plane as well since only those structures in the plane of focus are being visualized. As an illustration, consider the four pennies depicted in figure 3.3. When viewed from above (3.3-A), the upper two pennies and the lower two pennies appear to be closer together than the middle two pennies. However, when viewed from the side (3.3-B), it become apparent that the middle two pen- nies are in fact closer together. By filtering out those pennies above or below the plane of focus (3.3-C), only truly colocalized pennies remain visible. Figure 3.3: A conceptualized illustration of confocal microscopy. A. Viewed from above, the top two pennies and the bottom two pennies appear to be closer together (more colocalized) than the middle two pennies. B. However, when viewed from the side, it becomes apparent that this is not the case due to their differing heights (or position on the z-axis). C. Confocal microscopy optically eliminates items outside the plane of focus, such that objects near each other in the horizontal (or x-y) plane, can be assumed to be near each other in the vertical (or z) plane as well. By using multiple fluorescent markers, the interaction of multiple compounds 33 may be visualized using confocal methods(139). One fluorescent marker may be excited and imaged, and then the second marker excited and imaged. This pro- duces two slices or channels of the same field, with each channel documenting the density of a different protein. At any given pixel, then, the density of one protein can be compared to the density of the other protein, and these interactions may be quantified in various ways. This technique has been applied to elucidate pro- tein and gene interactions in a variety of cellular processes. The use of multiple laser frequencies to image up to three markers has been employed for a decade (140—142). Most of these efforts have described such colocalization qualitatively. However, some techniques for quantifying confocal microscopy images have been described (143, 144). Typically, these methods have been applied to examine the interaction of cellular structures (membranes and chromosomes, etc), as opposed to molecular interactions. 3.1.2 Colocalization as an image phenomenon Colocalization as an image phenomenon can be conceptualized as correlated vari- ations in intensity of two or more different markers over a region of interest, in- dicating that the concentration or density of the marked compounds are spatially related over that region. Of particular interest, then, is quantifying the degree to which the intensity of the respective markers are varying over space in a related fashion. In the proposed study, we will stain her-2 and Her-3 in sections of breast tumor tissue with different fluorescent antibodies, allowing the density of each protein to be imaged. The colocalization of Her-2 and Her-3 proteins will be mea- sured by applying traditional measures of covariance to the fluctuation of Her-2 and Her-3 protein density (as measured by fluorescent marker intensity at each pixel) over space. An idealized representation of this is given in figure 3.4. In 3.4- 34 A we see an idealized nucleated cell. In 3.4-B, fluorescent labelling of one protein is depicted, as would be visualized in a microscopic image of the cell—a grid of pixel intensities which may be represented either visually (as shown), or simply as an array or matrix of numerical data. In 3.4-C, the same methodology is used to measure the density of a second protein throughout the cell. As can be seen, both proteins are somewhat localized to the membrane region of the cell (more so for the protein in 3.4-C), and no protein is seen in the nuclear region. (This may be verified computationally by measuring the intensity of the proteins’ fluorescence in the region indicated by the nuclear fluorescent dye represented in 3.4-D. In this case, where the nuclear marker fluorescence is greater than 0, there is little or no fluorescence in this region in images 3.4-B and 3.4-C). There seems to be some colo- calization of the protein (both proteins exhibit bright fluorescence at some of the same pixels), but certainly not perfect colocalization. To quantify this colocalization, we will calculate the rate of change in protein density (also known as the ”moment”—further described below) at each pixel for each protein, and then calculate the degree to which the Her-2 and Her-3 moments are correlated for the entire image. By calculating a correlation coefficient between the Her-2 moments and the Her-3 moments, we can quantify the degree to which the presence of one protein is correlated with the presence of the second protein (as would be the case if they are forming dimers). This overall strategy is sum- marized in figure 3.5. The resulting correlation coefficient provides a quantitative score for each image which may then be included in survival analysis to test if the parameter-and therefore the colocalization and, by proxy, the heterodimerization of these two proteins—correlates to survival. To accomplish this, the joint moment of the pixels from two or more markers 35 Figure 3.4: Idealized representation of digital imaging of flourescently labelled proteins. A. An idealized nucleated cell. B. The same cell, with a protein flourescently labelled, and the resulting fluorescence digitally captured (pixels enlarged for clarity). C. Same as B., but now with a different protein labelled. D. Here, a fluorescent nuclear dye has been applied to allow imaging of the nuclear region. Such an exercise is useful to enable the computer to count nuclei, and thereby count cells, or two compare nuclear vs. cytoplasmic expression of protein (the dye tells the computer where the nucleus is, so it can separate nuclear expression from cytoplasmic expression in the other channels) can be compared. The moment at a given pixel simply refers to the rate of change of pixel intensity near that pixel. In other words, is the specified pixel much dif- ferent from its neighbors? This can be quantified by calculating the deviation of a given pixel’s intensity from the mean of its neighbors (the ”local" mean). This concept is illustrated in figure 3.6 and expressed in mathematical terms in the methods section below. Where the intensities are ”flat”—as in large regions of constant intensity—the moment can be defined as zero because the denominator (the variance) is zero. Where the intensities are changing over small intervals of space (edges of structures, etc), the moment is non-zero because the variance is 36 Calculated Moment at Image Data Each Pixel Channel1 (e.g. Her-2) ,..-Hf"uu~ . Channel 2 (e.g. Her-3) Correlation Coefficient Between Channels : " 1.-.. “ ”168566.11" as?“ i rPier l 1.0 l 0.9%; 222 '2"::: if: :22 l _. [Pixels __ 18W 16 J. Figure 3.5: Summary of quantification strategy. First, flourtscently labeled proteins are imaged (with her-2 fluorescence captured in one channel, and Her-3 fluorescence captured in a second channel). The moments are calculated (as described in the next chapter) for each pixel. Then, a correlation coefficient is calculated describing the correlation of the her-2 and Her-3 moments to eachother. non-zero. The joint moment then refers the relatedness of the moments of two channels at each pixel. The moments for each marker at each pixel can be tested for correlation using standard (or novel) statistical techniques (143, 145, 146). If the intensities of the markers are rising and falling, as measured by their moments at each pixel, in an unrelated fashion, the compounds they mark can be assumed to be non-colocalized. The degree to which this is the case can be assigned a numerical score (such as a Spearman’s correlation coefficient), and this score can be included in multivariate analysis. This method is called the ”Joint Moment of Standardized Images” or JMSI analysis and has been described previously (147). 37 Pixel A Figure 3.6: channels)lllustration of the moment at two pixels. Here, the moments of two pixels of one of the channels from figure 3.4 is shown. A 3x3 window is used to calculate the local mean in the neighborhood of each pixel. These neighborhoods are blown up for clarity, and the numerical intensity of each pixel is specified in each pixel. By calculating the standard deviation of the 9 pixels in each neighborhood, the deviation of the intensity of the central pixel (the pixel of interest) from the local mean may be expressed in standard deviations (the z-score). This value represents the moment, or the degree of flux near the pixel of interest. These values for the two pixels shown are given in the tables. As can be seen, "flat" areas (such as surrounding the pixel on the right) have smaller moments than pixels in regions of fluctuating intensity (as on the left). By correlating these values for each pixel in two or more channels, the joint moment for the entire image may be calculated. However, by comparing the value of a pixel to the mean of its neighbors ef- fectively blurs the image (the larger that neighborhood used to calculate the local mean, the more blurred the image becomes). In fact, a similar algorithm is used to digitally retouch images by softening them using a gaussian filter. This is fine where the coincidence of relatively large structures are of interest (for example, the localization of mRNA to a nuclear membrane), and is a particularly effective method for detecting edges of cellular structures, which are relatively large. Where 38 interactions of individual proteins are of interest, as in this proposal, such method- ology may not be appropriate since the proteins are somewhat smaller than a sin- gle pixel—therefore, incorporating data from neighboring pixels via local averaging may not be indicated. Instead, simple pixel by pixel correlation within globally standardized images may be preferable, wherein each pixel is expressed in terms of its deviation from the mean of the entire image field—not a local neighborhood—thereby standardizing the pixel intensities without reference to local windows. The normalized intensity of each pixel can then be correlated between the two images using the Spearman rank correlation coefficient. These two methods will be compared in the proposed analysis. 3.2 Application to Her-2/Her-3 colocalization In the proposed study, we will use digital image analysis to determine the degree of colocalization of Her-2 and Her-3 within tissues. As described above, we will store the fluorescent intensity resulting from fluorophore conjugated anti-Her-2 antibody in one color channel, and that resulting from anti-Her-3 antibody in an- other color channel. At each pixel within the field, then, we can determine the relative level of Her-2 protein and the relative level of Her-3 protein based on the intensity of fluorescence stored in their respective color channels at each pixel. If the intensities or moments (depending on which of the two methods described is employed) in each pixel within a given image are correlated, this suggests that the presence of one of these proteins is some how driving the presence of the other (or perhaps they are synergistically causing eachother to be located in the same place), as would be the case if molecular interactions were occurring leading to formation 39 of heterodimers between the receptors. If, on the other hand, there were no attrac- tive forces between these receptors, their location (and, therefore, there relative intensities at all pixels within the image) would not be expected to be correlated. As described above, there is more than one way to quantify the correlation of pixel intensities. This proposal describes two methods, and will compare their efficacy in conducting this type of molecular interaction analysis. In future work, these findings could be compared to other methodologies. Most notable among alternatives is Fluorescent Resonance Energy Transfer which allows much finer resolution of fluorescent particles. This technology has been rejected for the current study because of its complexity, but it will provide an important point of reference for future study. Furthermore, other epidemiological fields have approaches to offer. For example, Kulldorff, et al. (148) describe an approach to detection of geographic clustering of epidemiologic phenomena. One of the strengths of their approach is its independence of ad hoc assumptions about the nature of location of the clustering, an objective shared by the approach described in the present work. Extension of their approach to microscopic events may prove fruitful in the future. 40 Chapter 4 Methods 4.1 Phase one: algorithm development and data protocol development In phase one, we will further develop and test our computer algorithm for quan- tifying colocalization of the proteins of interest. Sections of test cases will be ob- tained and stained to optimize the parameters and algorithm, as described below, to ensure that the algorithm captures molecular events of interests. The methods for these analyses are further described below. I The second major task for phase one is to develop, in cooperation with the Grand Rapids Clinical Oncology Program, a mechanism whereby supplemental outcome data can be added to our dataset in a manner that protects the anonymity and confidentiality of patients. This mechanism will need to be fully reviewed and approved by the IRBs of both our institutions. 4.1.1 Tissues Breast carcinoma tissue specimens will be obtained from the VARI tissue reposi- tory using our web-based query system. These samples will be prepared accord- ing to existing protocols in our lab and as described previously (143). Briefly, the paraffin block will be sectioned in our laboratory at 5pm. These sections will be 41 flourescently stained for Her-2 and Her-3. We proposed using Fitc and Rhodamine markers (each having a unique excitation profile allowing them to be individually isolated optically). The exact array of antibodies and markers used may require adaptation and optimization through initial testing of the system. 4.2 Imaging After the tissues are prepared as described above, each section will be imaged by our Zeiss/ LSM-410 upright confocal microscope in 4 areas at 40x: two tumor tissue fields and two fields of surrounding normal tissue. Each field will be imaged two times. The first image of each field will be imaged using a 665nm low pass filter and 633nm excitation to elicit fluorescence of the Cy5 marker. The second image of each field will be imaged using a 590nm low pass filter and 568nm excitation to elicit fluorescence of the rhodamine marker. (Again, this protocol may require optimization). The images will be stored as 512 x 512 pixel tagged inline file format (TTFF) files. Each marker will be imaged as a separate 8 bit (256 shades of gray) tiff files (multilayer tiffs could be used to store both images in a single file, but we have found separate files to be simpler to deal with in the image processing phase). 4.3 Image analysis 4.3. 1 Data Modeling Two key questions need to be addressed in conducting the analyses described be- low. First, to what extent does the arbitrariness arising from the orientation of the tissue within the tissue block (and, therefore, within the section being imaged with the microscope) as well as the selection of fields within the section to be imaged 42 introduce additional variance into the analysis? Second, how should variance of pixel intensities be modelled? Are these intensities normally distributed? If not, how are they distributed and how can they be modelled? To answer the first question, a test series of cases will be analyzed. Stacks of images (say, at 0.1 micron intervals) representing the entire thickness of each tissue section will be isolated by means of confocal microscopy and imaged. These im- age stacks can then be used to reconstruct a three dimensional model of the tissue section which can be computationally rotated in space. Then, new sections taken at different orientations through this virtual tissue block can be computationally created. The analysis can be conducted on these various sections to see to what ex- tent the orientation of the tissue affects the analysis. If it is found that there is high variability between orientations, this will need to be factored into our analysis, and may point out the need to conduct the analysis in three dimensions rather than two. Since proteins clearly interact in three dimensions, such an approach may be preferably any way. However, the computational burden of such an approach is an order of magnitude greater than the approach here described. Therefore, if the two dimensional approach proves to provide an effective predictor of prognosis, it would be preferable by virtue of its relative simplicity. To answer the second question, the distributions of the pixel intensities will be analyzed, both in terms of entire images and small windows used in the joint mo- ment of standardized images method described below. As shown in figure 4.1, these images tend not to be normally distributed on the whole. It may be possible to restore a normal distribution through various techniques (background correc- tion, or analysis of the square root of the intensity of each pixel, or ranking the intensities of the pixels and analyzing the rank rather than the raw intensity). The 43 20 XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX l 30 xxxxxxxxx 35xxxxxx 40xxxx g; §§§§§xxxxxxxxxxxxxxx o momomomomomomomomomo oo v-NNCOO') in o F.. gmomnnmmmmFPSFFrrrifir Figure 4.1: Histogram of a representative image. Note the chi-square distribution of intensities. Each x represents 1% of all pixels. discussion that follows assumes that normally distributed data can be extracted from the images for the purposes of measuring the variance of the intensities of the pixels over space. If this is not possible, non-parametric measures of central tendency and variance will need to be employed instead of the mean and Z-score as described below. 4.3.2 Algorithm Implementation Software already developed by the author for performing the computational anal- ysis here described will be used to quantify the colocalization between the two markers. Two methods will be employed and compared: 1. Joint moment of standardized images (J M81) (147) e The moments at each pixel for each of the two markers (her-2 and Her-3) will be quantified in terms of the variability of the pixel intensities within a lo- cal ”window” centered on the pixel of interest. The local window will be a small region (for example, 5 x 5 pixels) around the region of interest. If (as discussed above), the intensities of the pixels in these windows can be rep- 44 resented in a normally distributed fashion, the moment will be quantified as the Z-score of each pixel (it’s deviance from the mean intensity of the local window expressed in terms of the variance of the local window). From this point forward, it will be assumed that this is the case (if not, adjustments will need to be made as mentioned above). 0 A correlation score will then be assigned to the image by calculating the Spear- man Rank-Correlation Coefficient between the moments of the her-2 and Her- 3 markers flourescent markers. The Spearman Rank-Correlation Coefficient will be used because the intensities of such images are not normally dis- tributed, and normality is not required by this non-parametric statistical test. e All pixels with an intensity less than an empirically determined threshold will be discarded as background (non-tissue areas). 2. Correlation of globally standardized images (CGSI) e The images will be standardized by assigning a Z-score to each pixel as cal- culated as the deviation of the pixel from the mean of the entire image (not a local window) mean, measured in standard deviations. e A correlation score will then be assigned to the image by calculating the Spear- man Rank-Correlation Coefficient between the z-scores of each of the two markers (her-2 and Her-3). This will be the measure of colocalization for each image. In either case, each pixel will be assigned a score based on its variation from either the local or global mean. Spearman Rank-Correlation Coefficients will be used because even if the pixel intensities are represented in a normally distributed 45 fashion, the manipulation of this data (which involves squaring deviations) will al- most certainly result in a chi-square distribution of Z-scores. The Spearman Rank- Correlation Coefficient will be calculated as follows: 1. Pixel Z-scores will be sorted in ascending order. 2. Each pixel will be assigned a rank. 3. Where multiple pixels have identical scores, they will each be assigned the mean rank for the entire group. 4. The correlation coefficient is then calculated as follows: (4.1) Where 3 Lu; = 20971; '— firm) X (pryi — firy) i=1 And pm, is the rank of the score of the 2"” pixel in image x (and respectively for y), and 13,, is the mean rank for all pixel scores in image x (and respectively for y). 4.4 Phase Two: Retrospective survival analysis of a cohort of breast cancer cases In phase two, the technologies and data transfer protocols will be applied to full analysis of a cohort of breast cancer tissues from the repository at VARI. 46 4.4. 1 Followup data The exact protocol for data handling in this study will be developed in cooperation with the Grand Rapids Clinical Oncology Program, as described in phase one. The protocol could follow the following strategy, but the final strategy will be agreed upon in the course of Phase 1, and approved by the IRBs at both institutions. The Van Andel Research Institute maintains a repository of paraffin fixed hu- man tissue specimens from surrounding pathology departments. Each specimen retains the original accession number imprinted on its cassette, and is linked to a data file that contains age, gender, and diagnosis field from the pathology report. Sections from the blocks to be used in this study will be cut, and a new num- ber placed on the resulting slides. The accession number for specimens identified through our online query tool together with the new number assigned to the slides will be transmitted securely to the Grand Rapids Clinical Oncology Program, who could add treatment and survival data (what treatment, if any, was attempted, date of last contact, status at last contact) and strip the file of the accession number, leaving only the new identifier. This file would then be transmitted back to the Van Andel Research Institute. The file linking the slide identifier to the accession num- ber would be destroyed, removing any means of linking the slides and associated data to the original accession number or medical record. 4.4.2 Statistical analysis 1. Subject selection and censoring Survival analysis will be conducted upon patients corresponding to the breast cancer blocks in our repository. This analysis will be done retrospectively. Survival time will be calculated from the date of diagnosis. For each patient, the time to 47 event (see dependant variables below) will either be known through the medical record or vital record, or the patient will be censored as of last contact with the data system (i.e. the last medical record date that confirms the disposition of the patient as of that date). 2. Dependent variables Two dependent variables will be individually and separately analyzed: e Survival e Time to metastasis 3. Models The JMSI and CGSI coefficients will be separately analyzed as in- dependent variables in Cox proportional hazard models. This model has been selected because of its robustness—Le, it performs well under a variety of distri- butions citepkleinsurv—and the distribution of our correlation coefficients is not likely to be normal. Initial covariates included in the model will be: 1. Age 2. Nodal status 3. Tumor size 4. Tumor grade 5. Treatment 6. Mean Her-2 intensity 7. Mean Her-3 intensity Thus, our initial hazard model will be as follows: 48 h(t, X) : h0(t)efilrs+22:2 ficxc+2?:2 LMX. xrs) (4.2) Where 61 is the coefficient for the correlation coefficient 7*, calculated as de- scribed above, BC is the coefficient for each covariate, X c, above, and fl, is the coef- ficient for each interaction term between rs and each covariate listed above. Beginning with the least significant interaction term, non-contributory parame- ters will be removed from the model in a step-wise fashion. At each step, the in- tegrity of the model will be assessed using the likelihood ratio test (the alternative— the Wald test—has been rejected at the behest of statistical texts such as (149)). For each parameter removed, the following calculation will be performed: LR = (—2long.) — (—2long) (4.3) This number has a chi-square distribution, and should be non-significant if the subtraction of the parameter (in vector X") did not impact the predictive power of the hazard function. Alternatively, entire sets of parameters (such as all interaction terms) may be removed so long as the resulting difference in LR is non—significant. The point estimate of interest is the adjusted hazard ratio due to the colocalization of Her-2 and Her-3 receptors. The hazard ratio is calculated as follows: [-772 2 8B" (4.4) Where the confidence interval is given by exponentiating 6,, :l: 1.96 x SE (SE being the standard error of the correlation coefficient as calculated from the data set). If this interval is substantially narrower with the inclusion of a parameter that is non-significant by the LR test above, it may be retained. 49 Note that the model must at all times be well formed. That is, no term may be removed, even if non-significant by the LR test, if an interaction term involving that term has been retained. For example, the Nodal Status term must be retained if Nodal Status xrs has been retained for any reason. This analysis will be performed using the Proc Logisitic procedure of the SAS statistical analysis program. If the level of interaction between Her-2 and Her-3 is truly prognostic, we would expect the hazard ratio as given above to be significantly greater than 1. Further- more, the the CGSI method is indeed superior to the JMSI method for this sort of application, the hazard ratio from the correlation coefficient as calculated by the CGSI method should be significantly larger than that from the JMSI method. We have hypothesized that this analysis will provide additional information beyond simple measurement of Her-2 level. By including the Her-2 and Her-3 terms in our model, we will be able to identify any independent predictive power of the colocalization term above and beyond that of measuring the intensities of either or both proteins without attempting to account for their interaction in space. 50 Chapter 5 Denouement It was the intent of this work to describe the approach of quantitative molecu- lar epidemiology, motivate its implementation, and demonstrate its application through a proposed study of the role of the interaction of two receptors, Her-2 and Her-3, as a quantitative prognostic parameter in breast cancer. This work arose from a grant application to the National Institutes of Health, which as of this date remains pending. Regardless of the outcome of this application, the author and his colleagues will continue to pursue the vital agenda of quantitative molecular epi- demiology, and they invite the inquiry and/ or collaboration of interested parties. 51 Glossary Because of the cross-disciplinary nature of this work, some key terminology is here defined: Amplification - The increase in gene copy of a specified gene in the genomic DNA. This may be detected by Flourescent In-Situ Hybridization (FISH), a tech- nique that hybridizes flourescent probes targeted to specific gene sequences to de- tect how many copies of the gene are in the chromosomes. 'Dimer - The molecular interaction (via non-covalent forces such as hydrogen bonding) of two molecules to form a complex. These molecules may cell surface receptors or other molecules. Dimerization is the process by which this occurs. Colocalization - An imaging phenomena in which imaged markers are found in the same point in space as documented by the multi-channel image of those markers. Flourescent tags attached to dimerized proteins would be expected to be colocalized in an image of those tags. Granularity - The ability of a measure to distinguish fine gradations in state. For example a True/ False question has less granularity than a multiple choice question with 5 possible answers, which in turn has less granularity than an essay question with respect to measuring exactly what a student knows about a given topic. Heterodimer - A dimer composed of two different proteins. Homodimer - A dimer composed of a pair of identical proteins. Overexpression - The production of protein above normal physiologic levels re— 52 sulting from gene amplification, or an increase in transcription products (mRNA), or an increase in translation products (these latter two manifesting a breakdown in the normal regulation of these processes). Proto-oncogene - A gene encoding a protein necessary for normal physiologic function but which may become carcinogenic as a result of mutation or disruption of normal regulatory processes. 53 Bibliography [1] Grizzle W, Aamodt R, Clausen K, LiVolsi V, Pretlow T, Qualman S. Providing human tissues for research: how to establish a program. Arch Pathol lab Med 1998;122:1065-76. [2] Hanahan D, Weinberg R. The hallmarks of cancer. Cell 2000;100:57-70. [3] Wallis Y, Macdonald F. Demystified oncogenes. Mol Pathol 1999;52:55-63. [4] Banks R, Forbes M, Kinsey S, Stanley A, Ingham E, Walters C, Selby P. Re- lease of the angiogenic cytokine vascular endothelial growth factor (VEGF) from platelets: significance for VEGF measurements and cancer biology [see comments] Br J Cancer 1998;77:956-64. [5] Tsarfaty I, Alvord W, Resau J, Altstock R, Lidereau R, Bieche I, Bertrand F, Horev J, Klabansky R, Keydar I, Vande-Woude G. Alteration of Met pro- tooncogene product expression and prognosis in breast carcinomas. Anal Quant Cytol Histol 1999;21:397-408. ‘ [6] Page D. Prognosis and breast cancer. Recognition of lethal and favorable prognostic types. Am J Surg Pathol 1991;15:334-49. [7] Jones D, Fletcher C. How shall we apply the new biology to diagnostics in surgical pathology? J Pathol 1999;187:147—54. [8] Delides G, Garas G, Georgouli G, Jiortziotis D, Lecca J, Liva T, Elemenoglou J. Intralaboratory variations in the grading of breast carcinoma. Arch Pathol Lab Med 1982;106:126-8. [9] Boiesen P, Bendahl P, Anagnostaki L, Domanski H, Holm E, Idvall I, Johans- son S, Ljungberg O, Ringberg A, Ostberg G, Femo M. Histologic grading in breast cancer-reproducibility between seven pathologic departments. South Sweden Breast Cancer Group. Acta Oncol 2000;39:41-5. [10] Sloane J, Amendoeira I, Apostolikas N, Bellocq J, Bianchi S, Boecker W, Bus- solati G, Coleman D, Connolly C, Eusebi V, De-Miguel C, Dervan P, Dri- jkoningen R, Elston C, Faverly D, Gad A, Jacquemier J, Lacerda M, Martinez- Penuela J, Munt C, Peterse J, Rank F, Sylvan M, Tsakraklides V, Zafrani B. Consistency achieved by 23 European pathologists from 12 countries in di- agnosing breast disease and reporting prognostic features of carcinomas. Eu- ropean Commission Working Group on Breast Screening Pathology. Virchows Arch 1999;434:3-10. 54 [11] Kronqvist P, Montironi R, Kquio T, Collan Y. Subjective breast cancer grad- ing. Analyses of reproducibility after application of Bayesian belief net- works. Anal Quant Cytol Histol 1997;19:423-9. [12] Toczek H, Fritz P, Schwarzmann P, Wu X, Mahner G. Breast carcinoma, corre- lations between visual diagnostic criteria for histologic grading and features of image analysis Analyt Quant Cytol Histol 1996;18:481-493. [13] Layfield L, Saria E, Mooney E, Liu K, Dodge R. Tissue heterogeneity of immunohistochemically detected estrogen receptor. Implications for image analysis quantification. Am I Clin Pathol 1998;110:758-64. [14] Neri T, Carnevali L, Orlandini G, Gatti R, Scandroglio R, Savi M, Allegri L. Fluourescent in situ hybridisation on tissue sections: a quantitative approach with confocal laser scanning microscopy. Eur J Histochem 2000;44:193-8. [15] Bogardus S, Concato J, Feinstein A. Clinical epidemiological quality in molecular genetic research: the need for methodological standards. [AMA 1999;281:1919-26. [16] Alizadeh A, Eisen M, Davis R, Ma C, Lossos I, Rosenwald A, Boldrick J, Sabet H, Tran T, Yu X, Powell J, Yang L, Marti G, Moore T, Hudson J, Lu L, Lewis D, Tibshirani R, Sherlock G, Chan W, Greiner T, Weisenburger D, Armitage J, Wamke R, Staudt L, S . Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling [see comments] Nature 2000;403:503- 11. [17] Koo H, Monks A, Mikheev A, Rubinstein L, Gray-Goodrich M, McWilliams M, Alvord W, Oie H, Gazdar A, Paull K, Zarbl H, Vande—Woude G. En- hanced sensitivity to 1-beta-D-arabinofuranosylcytosine and topoisomerase II inhibitors in tumor cell lines harboring activated ras oncogenes. Cancer Res 1996;56:5211-6. [18] Iwamoto K, Mizuno T, Ito T, Akiyama M, Takeichi N, Mabuchi K, Seyama T. Feasibility of using decades-old archival tissues in molecular oncol- ogy/ epidemiology. Am J Pathol 1996;149:399-406. [19] Kleeberger C, Lyles R, Margolick J, Rinaldo C, Phair J, Giorgi J. Viability and recovery of peripheral blood mononuclear cells cryopreserved for up to 12 years in a multicenter study. Clin Diagn Lab Immunol 1999;6:14-9. [20] Dillon D, Howe C, Bosari S, Costa J. The molecular biology of breast cancer: accelerating clinical applications. Crit Rev Oncog 1998;9z125-40. [21] Camp R, Rimm E, Rimm D. Met expression is associated with poor out- come in patients with axillary lymph node negative breast carcinoma. Cancer 1999;86:2259-65. [22] Yamashita J, Ogawa M, Yamashita S, Nomura K, Kuramoto M, Saishoji T, Shin S. Immunoreactive hepatocyte growth factor is a strong and indepen- dent predictor of recurrence and survival in human breast cancer. Cancer Res 1994;54:1630-3. 55 [23] Schubert D, Heinemann S, Carlisle W, Tarikas H, Kimes B, Patrick J, Stein- bach J, Culp W, Brandt B. Clonal cell lines from the rat central nervous sys- tem. Nature 1974;249:224-7. [24] Drebin J, Stern D, Link V, Weinberg R, Greene M. Monoclonal antibodies identify a cell-surface antigen associated with an activated cellular oncogene. Nature 1984;312:545-8. [25] Padhy L, Shih C, Cowing D, Finkelstein R, Weinberg R. Identification of a phosphoprotein specifically induced by the transforming DNA of rat neu- roblastomas. Cell 1982;28:865-71. [26] Schechter A, Stern D, Vaidyanathan L, Decker S, Drebin J, Greene M, Wein- berg R. The neu oncogene: an erb-B-related gene encoding a 185,000-Mr tu- mour antigen. Nature 1984;312:513—6. [27] Semba K, Kamata N, Toyoshirna K, Yamamoto T. A v-erbB-related pro- tooncogene, c-erbB-2, is distinct from the c-erbB-l / epidermal growth factor- receptor gene and is amplified in a human salivary gland adenocarcinoma. Proc Natl Acad Sci U S A 1985;82:6497—501. [28] King C, Kraus M, Aaronson S. Amplification of a novel v-erbB-related gene in a human mammary carcinoma. Science 1985;229:974-6. [29] Bargmann C, Hung M, Weinberg R. The neu oncogene encodes an epidermal growth factor receptor-related protein. Nature 1986;319:226-30. [30] Coussens L, Yang-Feng T, Liao Y, Chen E, Gray A, McGrath J, Seeburg P, Libermann T, Schlessinger J, Francke U, P . Tyrosine kinase receptor with extensive homology to EGF receptor shares chromosomal location with neu oncogene. Science 1985;230:1132-9. [31] Landis S, Murray T, Bolden S, Wingo P. Cancer statistics, 1998. CA Cancer J Clin 1998;48:6-29. [32] SEER . SEER Cancer Statistics Review, 1973-1998 seer.cancer.gov/Publications/CSRI973-1998 2001. [33] Keller D, Peterson E, Silberman G. Survival rates for four forms of cancer in the United States and Ontario. Am ] Public Health 1997;87:1164-7. [34] Shwartz M. Estimates of lead time and length bias in a breast cancer screen- ing program. Cancer 1980;46:844-51. [35] Pelikan S, Moskowitz M. Effects of lead time, length bias, and false-negative assurance on screening for breast cancer. Cancer 1993;71:1998-2005. [36] Morrison A. The effects of early treatment, lead time and length bias on the mortality experienced by cases detected by screening. Int J Epidemiol 1982;11:261—7. 56 r [37] Swain S. Noninvasive breast cancer: Part 1. Ductal carcinoma in situ— incidence, presentation, guidelines to treatment. Oncology (Huntingt) 1989;3z25-31. [38] Emster V, Barclay J, Kerlikowske K, Grady D, Henderson C. Incidence of and treatment for ductal carcinoma in situ of the breast. [AMA 1996;275:913-8. [39] Slamon D, Clark G, Wong S, Levin W, Ullrich A, McGuire W. Human breast cancer: correlation of relapse and survival with amplification of the HER- 2 / neu oncogene. Science 1987;235:177-82. [40] Revillion F, Bonneterre J, Peyrat J. ERBB2 oncogene in human breast cancer and its clinical significance. Eur J Cancer 1998;34:791-808. [41] Koeppen H, Wright B, Burt A, Quirke P, McNicol A, Dybdal N, Sliwkowski M, Hillan K. Overexpression of HERZ/neu in solid tumours: an immunohis- tochemical survey. Histopathology 2001;38:96-104. [42] Kelsey J, Bernstein L. Epidemiology and prevention of breast cancer. Annu Rev Public Health 1996;17:47-67. [43] Breast Cancer C. G. Breast cancer and hormonal contraceptives: collabo- rative reanalysis of individual data on 53,297 women with breast cancer and 100,239 women without breast cancer from 54 epidemiological studies. Iancet 1996;347:1713-1727. [44] Kelsey J, Gammon M. Epidemiology of breast cancer. Epidemiol Rev 1990;12:228-240. [45] Shrestha P, Huang J, Tsuji T, Shinozaki F, Maeda K, Sasaki K, Ueno K, Yamada K, Mori M. Rare expression of the c—erbB-2 oncoprotein in salivary gland tumors: an immunohistochemical study. ] Oral Pathol Med 1992;21:477-80. [46] Berchuck A, Kamel A, Whitaker R, Kerns B, Olt G, Kinney R, Soper J, Dodge R, Clarke-Pearson D, Marks P, et . Overexpression of HER-2/neu is asso- ciated with poor survival in advanced epithelial ovarian cancer. Cancer Res 1990;50:4087-91. [47] Rubin S, Finstad C, Federici M, Scheiner L, Lloyd K, Hoskins W. Prevalence and significance of HER-Z/neu expression in early epithelial ovarian cancer. Cancer 1994;73:1456-9. [48] Rubin S, Finstad C, Wong G, Ahnadrones L, Plante M, Lloyd K. Prognostic significance of HER-Z/neu expression in advanced epithelial ovarian cancer: a multivariate analysis. Am ] Obstet Gynecol 1993;168:162-9. [49] Hruza C, Dobianer K, Beck A, Czerwenka K, Hanak H, Klein M, Leodolter S, Medl M, Mullauer—Ertl S, Preiser J, et . HER-2 and INT-2 amplification estimated by quantitative PCR in paraffin- embedded ovarian cancer tissue samples. Eur J Cancer 1993;29A:1593-7. 57 [50] Csokay B, Papp J, Besznyak I, Bosze P, Sarosi Z, Toth J, Zalay Z, Olah E. Onco- gene patterns in breast and ovarian carcinomas. Eur ] Surg Oncol 1993;19:593- 9. [51] Eltabbakh G, Belinson J, Kennedy A, Biscotti C, Casey G, Tubbs R. p53 and HER-2/neu overexpression in ovarian borderline tumors. Gynecol On- col 1997;65:218-24. [52] Meden H, Marx D, Schauer A, Wuttke W, Kuhn W. Prognostic significance of p105 (c-erbB-2 HERZ/neu) serum levels in patients with ovarian cancer. Anticancer Res 1997;17:757-60. [53] Kowalski L, Kanbour A, Price F, Finkelstein S, Christopherson W, Seski J, Naus G, Bumham J, KanbourShakir A, Edwards R. A case-matched molec- ular comparison of extraovarian versus primary ovarian adenocarcinoma. Cancer 1997;79:1587-94. [54] Young S, Liu W, Brock J, Smith S. ERBB2 and chromosome 17 centromere studies of ovarian cancer by fluorescence in situ hybridization. Genes Chro- mosomes Cancer 1996;16:130—7. [55] Wang Z, Liu W, Smith S, Parrish R, Young S. c-myc and chromosome 8 centromere studies of ovarian cancer by interphase FISH. Exp Mol Pathol 1999;66:140-8. [56] Meden H, Marx D, Roegglen T, Schauer A, Kuhn W. Overexpression of the oncogene c-erbB-2 (HERZ/neu) and response to chemotherapy in patients with ovarian cancer. Int J Gynecol Pathol 1998;17:61-5. [57] Yang H, Zhang G, Xu K. [c-erbB2 gene amplification in human primary ep- ithelial ovarian cancer and its clinical significance] Zhonghua Zhong Liu Za Zhi 1998;20:367-70. [58] Fajac A, Benard J, Lhomme C, Rey A, Duvillard P, Rochard F, Bernaudin J, Riou G. c-erbB2 gene amplification and protein expression in ovarian epithe- lial tumors: evaluation of their respective prognostic significance by multi- variate analysis. Int J Cancer 1995;64:146—51. [59] Katsaros D, Theillet C, Zola P, Louason G, Sanfilippo B, Isaia E, Arisio R, Gia- rdina G, Sismondi P. Concurrent abnormal expression of erbB-2, myc and ras genes is associated with poor outcome of ovarian cancer patients. Anticancer Res 1995;15:1501-10. [60] Ross J, Yang F, Kallakury B, Sheehan C, Ambros R, Muraca P. HER-2/neu oncogene amplification by fluorescence in situ hybridization in epithelial tu- mors of the ovary. Am J Clin Pathol 1999;111:311-6. [61] Gershenson D, Baker V, Price J, Hung M, El-Naggar A, Tortolero-Luna G, Silva E. Molecular profile of advanced-stage transitional cell carcinoma of the ovary. Am J Obstet Gynecol 1997;177:120-5. 58 It]. ‘73-'51: .r‘KAh’M A! - ,_ . [62] Costa M, Walls J. Epidermal growth factor receptor and c-erbB-2 oncoprotein expression in female genital tract carcinosarcomas (malignant mixed mulle- rian tumors). Clinicopathologic study of 82 cases. Cancer 1996;77:533-42. [63] Heinze T, Jonas S, Karsten A, Neuhaus P. Determination of the oncogenes p53 and C-erb B2 in the tumour cytosols of advanced hepatocellular carci- noma (HCC) and correlation to survival time. Anticancer Res 1999;19:2501-3. [64] Press M, Pike M, Hung G, Zhou J, Ma Y, George J, Dietz-Band J, James W, Slamon D, Batsakis J, et . Amplification and overexpression of HER-2/neu in carcinomas of the salivary gland: correlation with poor prognosis. Cancer Res 1994;54:5675-82. [65] Giannoni C, Naggar A, Ordonez N, Tu Z, Austin J, Luna M, Batsakis J. c- erbB-2/neu oncogene and Ki-67 analysis in the assessment of palatal salivary gland neoplasms. Otolaryngol Head Neck Surg 1995;112:391-8. [66] Sugano S, Mukai K, Tsuda H, Hirohashi S, Furuya S, Shimosato Y, Ebihara S, Takeyama I. Irnmunohistochemical study of c-erbB-2 oncoprotein overex- pression in human major salivary gland carcinoma: an indicator of aggres- siveness. Iaryngoscope 1992;102:923-7. [67] Cho K, Kim J, Lee S, Oh K. Mucoepidermoid carcinoma of the salivary gland—a clinico-pathologic and immunohistochemical study for c-erbB-2 on- coprotein. J Korean Med Sci 1997;12:499-504. [68] Martinez-Barba E, Cortes-Guardiola J, Minguela-Puras A, Torroba-Caron A, Mendez-Trujillo S, Bermejo-Lopez J. Salivary duct carcinoma: clini- copathological and immunohistochemical studies. I Craniornaxillofac Surg 1997;25:328-34. [69] Lipponen P, Eskelinen M, Syrjanen S, Tervahauta A, Syrjanen K. Use of im- munohistochemically demonstrated c-erb B-2 oncoprotein expression as a prognostic factor in transitional cell carcinoma of the urinary bladder. Eur Urol 1991;20:238-42. [70] Ravery V, Grignon D, Angulo J, Pontes E, Montie J, Crissman J, Chopin D. Evaluation of epidermal growth factor receptor, transforming growth factor alpha, epidermal growth factor and c-erbB2 in the progression of invasive bladder cancer. Urol Res 1997;25:9-17. [71] Miyamoto H, Kubota Y, Noguchi S, Takase K, Matsuzaki J, Moriyama M, Takebayashi S, Kitamura H, Hosaka M. C-ERBB-2 gene amplification as a prognostic marker in human bladder cancer. Urology 2000;55:679-83. [72] Lee S, Chow N, Chi Y, Tzai T, Yang W, Lin S. Expression of c-erbB-2 protein in normal and neoplastic urothelium: lack of adverse prognostic effect in human urinary bladder cancer. Anticancer Res 1994;14:1317-24. [73] Lipponen P. Expression of c-erbB-2 oncoprotein in transitional cell bladder cancer. Eur J Cancer 1993;29A:749-53. 59 [74] Orlando C, Sestini R, Vona G, Pinzani P, Bianchi S, Giacca M, Pazzagli M, Selli C. Detection of c-erbB-2 amplification in transitional cell bladder carcinoma using competitive PCR technique. I Urol 1996;156:2089-93. [75] Korkolopoulou P, Christodoulou P, Kapralos P, Exarchakos M, Bisbiroula A, Hadjiyannakis M, Georgountzos C, Thomas-Tsagli E. The role of p53, MDM2 and c-erb B-2 oncoproteins, epidermal growth factor receptor and prolifer- ation markers in the prognosis of urinary bladder cancer. Pathol Res Pract 1997;193:767-75. [76] Tetu B, Fradet Y, Allard P, Veilleux C, Roberge N, Bernard P. Prevalence and clinical significance of HER/Zneu, p53 and Rb expression in primary super- ficial bladder cancer. I Urol 1996;155:1784-8. [77] Sato K, Moriyama M, Mori S, Saito M, Watanuki T, Terada K, Okuhara E, Akiyama T, Toyoshima K, Yamamoto T, et . An immunohistologic evaluation of C-erbB-2 gene product in patients with urinary bladder carcinoma. Cancer 1992;70:2493—8. [78] Lonn U, Lonn S, Friberg S, Nilsson B, Silfversward C, Stenkvist B. Prognos- tic value of amplification of c-erb-B2 in bladder carcinoma. Clin Cancer Res 1995;]:1189-94. [79] Mellon], Lunec J, Wright C, Horne C, Kelly P, Neal D. C-erbB-2 in bladder cancer: molecular biology, correlation with epidermal growth factor recep- tors and prognostic value. I Urol 1996;155:321-6. [80] Flejou J, Paraf F, Muzeau F, Fekete F, Henin D, Jothy S, Potet F. Expression of c-erbB-2 oncogene product in Barrett’s adenocarcinoma: pathological and prognostic correlations. I Clin Pathol 1994;47:23-6. [81] Nakamura T, Nekarda H, Hoelscher A, Bollschweiler E, Harbeck N, Becker K, Siewert J, HarbecN[correctedtoHarbeck N. Prognostic value of DNA ploidy and c-erbB-2 oncoprotein overexpression in adenocarcinoma of Bar- rett’s esophagus. Cancer 1994;73:1785-94. [82] Duhaylongsod F, Gottfried M, Iglehart J, Vaughn A, Wolfe W. The signifi- cance of c-erb B-2 and p53 immunoreactivity in patients with adenocarci- noma of the esophagus. Ann Surg 1995;221:677-83. [83] Polkowski W, vanSandick J, Offerhaus G, tenKate F, Mulder J, Obertop H, vanLanschot J. Prognostic value of Lauren classification and c-erbB-2 oncogene overexpression in adenocarcinoma of the esophagus and gastroe- sophageal junction. Ann Surg Oncol 1999;6z290-7. [84] Hofmockel G, Riess S, Bassukas I, Dammrich J. Epidermal growth factor family and renal cell carcinoma: expression and prognostic impact. Eur Urol 1997;31:478-84. [85] Lipponen P, Eskelinen M, Hietala K, Syrjanen K, Gambetta R. Expression of proliferating cell nuclear antigen (PClO), p53 protein and c-erbB-2 in renal adenocarcinoma. Int I Cancer 1994;57:275-80. 60 1E [86] Rasmuson T, Grankvist K, Ljungberg B. Soluble ectodomain of c-erbB-2 on- coprotein in relation to tumour stage and grade in human renal cell carci- noma. Br I Cancer 1997;75:1674-7. [87] Micke P, Hengstler J, Ros R, Bittinger F, Metz T, Gebhard S, Beeh K, Oesch F, Buhl R. c-erbB-2 expression in small-cell lung cancer is associated with poor prognosis. Int I Cancer 2001;92:474—9. [88] D’Amico T, Massey M, Herndon J, Moore M, Harpole D. A biologic risk model for stage I lung cancer: immunohistochemical analysis of 408 patients with the use of ten molecular markers. I Thorac Cardiovasc Surg 1999;117:736- 43. [89] Hsieh C, Chow K, Fahn H, Tsai C, Li W, Huang M, Wang L. Prognostic sig- nificance of HER-2/neu overexpression in stage I adenocarcinoma of lung. Ann Thorac Surg 1998;66:1159-63. [90] Kim Y, Park K, Kern J, Park C, Lirn S, Jang A, Yang]. The interactive effect of Ras, HER2, P53 and Bcl-2 expression in predicting the survival of non-small cell lung cancer patients. Lung Cancer 1998;22:181-90. [91] Harpole D, Herndon J, Wolfe W, Iglehart J, Marks J. A prognostic model of recurrence and death in stage I non-small cell lung cancer utilizing presenta- tion, histopathology, and oncoprotein expression. Cancer Res 1995;55:51-6. [92] Harpole D, Marks J, Richards W, Herndon J, Sugarbaker D. Localized ade- nocarcinoma of the lung: oncogene expression of erbB-2 and p53 in 150 pa- tients. Clin Cancer Res 1995;]:659-64. [93] Volm M, Efferth T, Mattem J. Oncoprotein (c-myc, c-erbBl, c—erbB2, c-fos) and suppressor gene product (p53) expression in squamous cell carcinomas of the lung. Clinical and biological correlations. Anticancer Res 1992;12:11-20. [94] Kern J, Schwartz D, Nordberg J, Weiner D, Greene M, Torney L, Robinson R. p185neu expression in human lung adenocarcinomas predicts shortened survival. Cancer Res 1990;50:5184-7. [95] Cantero R, Torres A, Maestro M, Fernandez C, Hernando F, delBarco V, Sanz T, Balibrea J. Use of possible synergistic expression of p53 and p185 as a prog- nostic tool for stage I non-small-cell lung cancer. World J Surg 1999;23:1294—9. [96] Giatromanolaki A, Gorgoulis V, Chetty R, Koukourakis M, Whitehouse R, Kittas C, Veslemes M, Gatter K, Iordanoglou I. C-erbB—2 oncoprotein expres- sion in operable non-small cell lung cancer. Anticancer Res 1996;16:987-93. [97] Hilton D, West K. c-erbB-2 oncogene product expression and prognosis in gastric carcinoma. I Clin Pathol 1992;45:454-6. [98] Lee E, Cibull M, Strode] W, Haley J. Expression of HER-2/neu oncoprotein and epidermal growth factor receptor and prognosis in gastric carcinoma. Arch Pathol lab Med 1994;118:235-9. 61 [99] Jaehne J, Urmacher C, Thaler H, Friedlander-Klar H, Cordon-Cardo C, Meyer H. Expression of Her2/neu oncogene product p185 in correlation to clinic0pathological and prognostic factors of gastric carcinoma. I Cancer Res Clin Oncol 1992;118:474-9. [100] Hu J, Wang Z, Jiang Y. [Relations between p53 and p185 expression and prog- nosis of patients with colon cancers] Zhonghua Zhong Liu Za Zhi 1996;18:247- 9. [101] Jain S, Filipe M, Gullick W, Linehan J, Morris R. c-erbB-2 proto-oncogene expression and its relationship to survival in gastric carcinoma: an immuno- histochemical study on archival material. Int I Cancer 1991;48:668-71. [102] Okada N, Ohshio G, Yamaki K, Imamura T, Imamura M. Elevated serum c- erbB-2 protein levels in patients with pancreatic cancer: correlation to metas- tasis and shorter survival. Oncology 1995;52:392—6. [103] Dugan M, Dergham S, Kucway R, Singh K, Biernat L, Du W, Vaitkevicius V, Crissman J, Sarkar F. HER-2/neu expression in pancreatic adenocarcinoma: relation to tumor differentiation and survival. Pancreas 1997;14:229-36. [104] Kawesha A, Ghaneh P, Andren-Sandberg A, Ograed D, Skar R, Dawiskiba S, Evans J, Campbell F, Lemoine N, Neoptolemos J. K-ras oncogene sub- type mutations are associated with survival but not expression of p53, p16(INK4A), p21(WAF—1), cyclin D1, erbB-2 and erbB-3 in resected pancre- atic ductal adenocarcinoma. Int I Cancer 2000;89:469-74. [105] Lei S, Appert H, Nakata B, Domenico D, Kim K, Howard J. Overexpression of HER2/neu oncogene in pancreatic cancer correlates with shortened survival. Int I Pancreatol 1995;17:15-21. [106] Ross], Fletcher]. The HER-2/neu oncogene in breast cancer: prognostic fac- tor, predictive factor, and target for therapy. Stem Cells 1998;16:413-28. [107] Tetu B, Brisson J. Prognostic significance of HER-Z/neu oncoprotein expres- sion in node-positive breast cancer. The influence of the pattern of immunos- taining and adjuvant therapy. Cancer 1994;73:2359-65. [108] Borg A, Baldetorp B, Femo M, Killander D, Olsson H, Ryden S, Sigurdsson H. ERBB2 amplification is associated with tamoxifen resistance in steroid- receptor positive breast cancer. Cancer Lett 1994;81:137-44. [109] De-Placido S, Carlomagno C, De-Laurentiis M, Bianco A. c-erbB2 expression predicts tamoxifen efficacy in breast cancer patients. Breast Cancer Res Treat 1998;52:55-64. [110] Lupu R, Dickson R, Lippman M. The role of erbB-2 and its ligands in growth control of malignant breast epithelium. Princess Takamatsu Symp 1991;22:49- 60. 62 [1 11] De-Corte V, De-Potter C, Vandenberghe D, Van-Laerebeke N, Azam M, Roels H, Mareel M, Vandekerckhove J. A 50 kDa protein present in conditioned medium of COLO-16 cells stimulates cell spreading and motility, and ac- tivates tyrosine phosphorylation of Neu/HER-2, in human SK-BR-3 mam- mary cancer cells. I Cell Sci 1994;107:405-16. [112] Peles E, Ben-Levy R, Tzahar E, Liu N, Wen D, Yarden Y. Cell-type specific interaction of Neu differentiation factor (NDF/heregulin) with Neu/HER-2 suggests complex ligand-receptor relationships. EMBO I 1993;12:961-71. [113] Yarden Y. Agonistic antibodies stimulate the kinase encoded by the neu pro- tooncogene in living cells but the oncogenic mutant is constitutively active. Proc Natl Acad Sci U S A 1990;87:2569-73. [114] Chantry A. The kinase domain and membrane localization determine intra- cellular interactions between epidermal growth factor receptors. I Biol Chem 1995;270:3068-73. [115] Burke C, Stem D. Activation of Neu (ErbB-2) mediated by disulfide bond- induced dimerization reveals a receptor tyrosine kinase dimer interface. Mol Cell Biol 1998;18:5371-9. [116] Wallasch C, Weiss F, Niederfellner G, Jallal B, Issing W, Ullrich A. Heregulin- dependent regulation of HER2/neu oncogenic signaling by heterodimeriza- tion with HER3. EMBO I 1995;14:4267-75. [117] Tzahar E, Waterman H, Chen X, Levkowitz G, Karunagaran D, Lavi S, Ratzkin B, Yarden Y. A hierarchical network of interreceptor interactions de- termines signal transduction by Neu differentiation factor/neuregulin and epidermal growth factor. Mol Cell Biol 1996;16:5276—87. [1 18] Ram T, Schelling M, Hosick H. Blocking HER-2/HER—3 function with a dom- inant negative form of HER-3 in cells stimulated by heregulin and in breast cancer cells with HER-2 gene amplification. Cell Growth Difi‘er 2000;11:173-83. [119] Alimandi M, Romano A, Curia M, Muraro R, Fedi P, Aaronson S, Di-Fiore P, Kraus M. Cooperative signaling of ErbB3 and ErbB2 in neoplastic transfor- mation and human mammary carcinomas. Oncogene 1995;10:1813-21. [120] Marikovsky M, Lavi S, Pinkas-Kramarski R, Karunagaran D, Liu N, Wen D, Yarden Y. ErbB-3 mediates differential mitogenic effects of NDF/heregulin isoforms on mouse keratinocytes. Oncogene 1995;10:1403—11. [121] Tzahar E, Levkowitz G, Karunagaran D, Yi L, Peles E, Lavi S, Chang D, Liu N, Yayon A, Wen D, W . ErbB-3 and ErbB-4 function as the respective low and high affinity receptors of all Neu differentiation factor / heregulin isoforms. I Biol Chem 1994;269:25226-33. [122] Arteaga C, Ramsey T, Shawver L, Guyer C. Unliganded epidermal growth factor receptor dimerization induced by direct interaction of quinazolines with the ATP binding site. I Biol Chem 1997;272:23247—54. 63 [123] Gullick W, Srinivasan R. The type 1 growth factor receptor family: new lig- ands and receptors and their role in breast cancer. Breast Cancer Res Treat 1998;52:43-53. [124] Chozick B, Benzil D, Stopa E, Pezzullo J, Knuckey N, Epstein M, Finkel- stein S, Finch P. Immunohistochemical evaluation of erbB-2 and p53 pro- tein expression in benign and atypical human meningiomas. I Neurooncol 1996;27:117-26. [125] Iemoine N, Staddon S, Dickson C, Barnes D, Gullick W. Absence of acti- vating transmembrane mutations in the c-erbB-2 proto-oncogene in human breast cancer. Oncogene 1990;5z237-9. [126] Siegel P, Muller W. Mutations affecting conserved cysteine residues within the extracellular domain of Neu promote receptor dimerization and activa- tion. Proc Natl Acad Sci U S A 1996;93:8878-83. [127] Nikitin A, Jin J, Papewalis I, Prokopenko S, Pozharisski K, Winterhager E, Flesken-Nikitin A, Rajewsky M. Wild type neu transgene counteracts mu- tant homologue in malignant transformation of rat Schwann cells. Oncogene 1996;12:1309-17. [128] Marte B, Graus-Porta D, Jeschke M, Fabbro D, Hynes N, Taverna D. NDF/heregulin activates MAP kinase and p70 / p85 S6 kinase during prolif- eration or differentiation of mammary epithelial cells. Oncogene 1995;10:167- 75. [129] Hynes N, Stern D. The biology of erbB-2/neu/HER-2 and its role in cancer. Biochim Biophys Acta 1994;1 198:165-84. [130] Chan R, Muller W, Siegel P. Oncogenic activating mutations in the neu/erbB- 2 oncogene are involved in the induction of mammary tumors. Ann N Y Acad Sci 1999;889:45-51. [131] Xie D, Shu X, Deng Z, Wen W, Creek K, Dai Q, Gao Y, Jin F, Zheng W. Population-based, case-control study of HER2 genetic polymorphism and breast cancer risk. I Natl Cancer Inst 2000;92:412-7. [132] Drebin J, Link V, Greene M. Monoclonal antibodies reactive with distinct do- mains of the neu oncogene-encoded p185 molecule exert synergistic anti- tumor effects in vivo. Oncogene 1988;2z273—7. [133] Hudziak R, Lewis G, Winget M, Fendly B, Shepard H, Ullrich A. p185HER2 monoclonal antibody has antiproliferative effects in vitro and sensitizes hu- man breast tumor cells to tumor necrosis factor. Mol Cell Biol 1989;9z1165-72. [134] Stancovski I, Hurwitz E, Leitner O, Ullrich A, Yarden Y, Sela M. Mechanis- tic aspects of the opposing effects of monoclonal antibodies to the ERBB2 receptor on tumor growth. Proc Natl Acad Sci U S A 1991;88:8691-5. [135] Harwerth I, Wels W, Schlegel J, Muller M, Hynes N. Monoclonal antibodies directed to the erbB-2 receptor inhibit in vivo tumour cell growth. Br I Cancer 1993;68:1140-5. 64 Elm [136] Le X, McWatters A, Wiener J, Wu J, Mills G, Bast R. Anti-HER2 antibody and heregulin suppress growth of HER2-overexpressing human breast can- cer cells through different mechanisms. Clin Cancer Res 2000;6z260-70. [137] Baselga J, Tripathy D, Mendelsohn J, Baughman S, Benz C, Dantis L, Sklarin N, Seidman A, Hudis C, Moore J, Rosen P, Twaddell T, Henderson 1, Nor- ton L. Phase II study of weekly intravenous recombinant humanized anti- p185HER2 monoclonal antibody in patients with HER2/neu-overexpressing metastatic breast cancer. I Clin Oncol 1996;14:737-44. [138] Pegram M, Slamon D. Combination therapy with trastuzumab (Herceptin) and cisplatin for chemoresistant metastatic breast cancer: evidence for receptor-enhanced chemosensitivity. Semin Oncol 1999;26:89—95. [139] Brelje T, Wessendorf M, Sorenson R. Multicolor laser scanning confocal im- munofluorescence microscopy: practical application and limitations. Meth- ods Cell Biol 1993;38:97-181. [140] Montag M, Kukulies J, Jorgens R, Gundlach H, Trendelenburg M, Spring H. Working with the confocal scanning UV-laser microscope: specific DNA 10- calization at high sensitivity and multiple-parameter fluorescence. I Microsc 1991;163:201-10. [141] Schubert W. Triple immunofluorescence confocal laser scanning microscopy: spatial correlation of novel cellular differentiation markers in human muscle biopsies. Eur I Cell Biol 1991;55:272-85. [142] Arndt-Jovin D, Robert-Nicoud M, Jovin T. Probing DNA structure and func- tion with a multi-wavelength fluorescence confocal laser microscope. I Mi- crosc 1990;157:61-72. [143] Altstock R, Stein G, Resau J, Tsarfaty 1. Algorithms for quantitation of protein expression variation in normal versus tumor tissue as a prognostic factor in cancer: Met oncogene expression, and breast cancer as a model. Cytometry 2000;41:155-65. [144] Tanke H, Florijn R, Wiegant J, Raap A, Vrolijk J. CCD microscopy and image analysis of cells and chromosomes stained by fluorescence in situ hybridiza- tion. Histochem I 1995;27:4-14. [145] Kask P, Palo K, Fay N, Brand L, Mets U, Ullmann D, Jungmann J, Pschorr I, Gall K. Two-dimensional fluorescence intensity distribution analysis: theory and applications. Biophys I 2000;78:1703-13. [146] Kask P, Palo K, Ullmann D, Gall K. Fluorescence-intensity distribution anal- ysis and its application in biomolecular detection technology. Proc Natl Acad Sci LI 5 A 1999;96:13756-61. [147] Demandolx D, Davoust J. Multicolour analysis and local image correlation in confocal microscopy [Microscopy 1996;185:21-36. 65 [148] Kulldorff M, Nagarwalla N. Spatial disease clusters: detection and inference. Stat Med 1995;14:799-810. [149] Kleinbaum D. G. Survival Analysis, a Self-Learning Text. Springer-Verlag, New York 1996. 66 IIIIIIIIIIIIIIIIIIIIIIIIIIIIIII lulll[III]IIIIIIIIIIIII