UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

A review of oral/pharyngeal cancer and a review of 328 cases of lingual squamous cell carcinoma Oakley, Carol 1997

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


831-ubc_1998-0038.pdf [ 18.89MB ]
JSON: 831-1.0088416.json
JSON-LD: 831-1.0088416-ld.json
RDF/XML (Pretty): 831-1.0088416-rdf.xml
RDF/JSON: 831-1.0088416-rdf.json
Turtle: 831-1.0088416-turtle.txt
N-Triples: 831-1.0088416-rdf-ntriples.txt
Original Record: 831-1.0088416-source.json
Full Text

Full Text

A Review of Oral/Pharyngeal Cancer and A Review of 328 Cases of Lingual Squamous Cell Carcinoma By Carol Oakley D.D.S., University of Alberta, 1978 Dip. Periodontics, University of British Columbia, 1990 Ph.D., University of British Columbia, 1995 A THESIS SUBMITTED IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF 8N ! V SCIENCE in THE FACULTY OF GRADUATE STUDIES THE FACULTY OF DENTISTRY (Department of Oral Biological and Medical Sciences) We accept this thesis as conforming to the required standard THE UNIVERSITY OF BRITISH COLUMBIA July 1997 © Carol Oakley, 1997 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. The University of British Columbia Vancouver, Canada Date YlfV G{?7 DE-6 (2/88) 11 Abstract Between 1979 and 1994, 328 cases of squamous cell carcinoma (SCC) of the tongue were admitted to the BCCA, Vancouver Branch. These cases were reviewed by Drs. Hay, Epstein and van der Meij who also established the data fields that were used in statistical analyses performed by Dr. Le. The analyzed data set comprises Chapter 3, Results, and the data set is critiqued in Chapter 4, Discussion. The major component of the thesis is Chapter 1, Introduction which offers a broad overview of oral/pharyngeal cancer with respect to carcinogenesis, premalignancy, and treatment. In addition, Chapter 1 includes a brief review of epidemiological concepts and provides a perspective for the balance of the thesis in terms of understanding the strengths and shortcomings of previous investigations into the diagnosis, epidemiology, treatments and outcomes of oral/pharyngeal SCC. Conclusions across studies over time for trend data of oral/pharyngeal cancer cannot easily be drawn because of inconsistencies in criteria defining cancer cases and risk factors, and in data analyses and reporting. Nevertheless, analyses of the 328 lingual SCCs revealed that this case series was consistent with other reports of lingual SCC with respect to patient demographics, risk factors, tumour characteristics and survival. Among 328 patients, the mean age was 61 years with a male:female ratio of 1.5:1.0. The majority of patients had a history of alcohol and tobacco use, and the most common symptom was a sore tongue. Most cancers were early stage, well-differentiated keratinizing tumours of the oral tongue. Treatment was primarily radiation alone, followed by a combination of surgery and radiation; complications included necrosis of the soft tissues (10%) and mandible (6%). Patient follow-up ranged from 0 to 154 months. The overall all-cause survival was 41% and the stage of disease was significantly (p<0.001) related to survival. SCCs of the tongue have a good prognosis if they are detected early; consequently, screening and case finding strategies are essential and should concentrate efforts on individuals at high risk due to alcohol and tobacco use. s iii T A B L E OF CONTENTS Abstract ii Table of Contents iii List of Figures ix List of Tables x Acknowledgements xii CHAPTER 1 - INTRODUCTION I. Overview 1 II. Principles of Epidemiology 4 A. Definitions andTaxonomy 4 B. Descriptive Studies 7 1. Correlational Studies 8 2. Case Reports and Series 9 3. Cross-Sectional Surveys 10 C. Analytic Observational Studies 10 1. Case Control Studies 10 2. Cohort Studies 11 D. Measures of Disease Frequency 11 1. Ratios, Proportions and Rates 14 2. Prevalence and Incidence . 15 a. Prevalence 15 b. Incidence 15 3. Crude, Category-Specific and Adjusted Rates 16 a. Crude Rates 17 b. Category-Specific Rates 17 c. Adjusted Standardized Rates 18 E. Measures of Association 19 1. Relative Risk 21 2. Odds Ratios 24 3. Attributable Risk 25 F. Chance, Confounding and Bias 26 1. Chance 26 a. p values 27 b. Confidence Interval Estimates 28 2. Confounding 29 3. Bias 30 a. Selection Bias 31 b. Observation Bias 34 iv G. Measurement Reliability 36 1. Pearson Correlation Coefficients 37 2. Kappa 38 H. Measurement Validity 40 1. Sensitivity and Specificity of Measurement Techniques 41 2. Diagnostic Tests 43 J. Early Diagnosis and Survival of Disease 49 K. Summary 53 EI. Development and Anatomy of the Tongue 59 A. Embryological Development 59 B. Clinical Anatomy of the Tongue , 60 C. Structural Anatomy of the Tongue 62 1. Mucosa 62 a. Lining Mucosa 62 b. Gustatory Mucosa 63 2. Musculature of the Tongue 63 a. Extrinsic Muscles 64 i. Genioglossus 64 ii. Styloglossus 64 iii. Hyoglossus 65 iv. Palatoglossus 65 b. Intrinsic Muscles 65 i. Vertical 66 ii. Superior Longitudinal 66 iii. Inferior Longitudinal 66 iv. Transverse 66 3. Innervation of the Tongue 66 a. Sensory Innervation 66 i. Tactile 66 ii. Gustatory 67 b. Motor Innervation 67 c. Autonomic Innervation 67 4. Vasculature of the Tongue 67 a. Arterial 67 b. Venous 68 5. Lymphatic Tissues 68 6. Lymphatic Vessels and Drainage 69 7. Papillae and Taste Buds 70 a. Papillae 70 i. Filiform 70 ii. Fungiform 70 iii. Foliate 71 iv. Circumvallate 71 b. Taste Buds 71 8. Lingual Salivary Glands 72 V D. Functions of the Tongue 72 1. Speech 72 2. Mastication 73 3. Taste 73 4. Swallowing 74 5. Protective Reflexes 74 IV. Epithelium 77 A. General Characteristics 77 B. Protective Role of Oral Epithelium 78 1. Desquamation 80 2. Permeability Barriers 81 3. Cellular Defense Mechanisms 84 C. Differentiation of Oral Epithelium 85 1. Cornified Masticatory Epithelium 86 a. Stratum Basale 87 b. Stratum Spinosum 87 c. Stratum Granulosum 88 d. Stratum Corneum 88 i. Orthokeratin 88 ii. Parakeratin 88 2. Noncornified Lining Epithelium 89 3. Keratins as Markers of Differentiation 89 4. Influence of Retinoids on Epithelial Differentiation 91 D. Epithelial Proliferation 95 1. Cell Cycle 96 2. Control of the Cell Cycle 97 a. Intrinsic Mechanisms 97 i. Restriction Point 97 ii. p53 Gene and Protein 98 iii. myc Genes 98 iv. Apoptosis 99 v. Telomeres 99 b. Extrinsic Mechanisms 100 i. Regulation by Cell-Matrix Interactions 100 ii. Regulation by Cell-Cell Interactions 101 iii. Regulation by Growth Factors 102 iv. Regulation by Subepithelial Connective Tissue 104 V. Carcinogenesis 111 A. Models of Transformation 112 1. Monoclonal Transformation 112 2. Polyclonal Transformation and Field Cancerization 112 3. Contiguous Transformation 113 B. Tumour Initiation, Promotion and Progression 113 C. Molecular Basis and Biomarkers for Squamous Cell Carcinoma of the Head and Neck 116 1. Tumour Suppressor Gene p53 and Protein p53 119 vi a. p53 Protein 119 b. p53 Gene 120 2. c-erb B Gene, Epidermal Growth Factor Receptor and its Growth Factors 122 3. ras Oncogene Family 123 4. myc Oncogene Family 124 5. Bel-1 and Bcl-2 Oncogenes 124 6. TGF|3 Growth Factor and Attachment Receptors 125 7. Retinoic Acid Receptors and Keratin Markers 125 8. DNA Content 127 D. Tumour Kinetics 128 1. Implications for Radio and Chemo Therapy 130 E. Invasion and Metastasis 133 F. Risk Factors Associated with Oral/Pharyngeal Squamous Cell Carcinoma 136 1. Tobacco 137 a. Smoking Tobacco 138 b. Smokeless Tobacco 141 2. Alcohol 142 a. Alcoholic Beverages 142 b. Alcohol-Containing Mouthwashes 144 3. Tobacco Smoking and Alcohol 144 4. Diet, Oral Health, Dental Restorations and Socioeconomic Class 147 5. Candida 149 6. Viruses 150 a. Epstein-Barr Virus 150 b. Herpes Simplex Virus 151 c. Human Papillomavirus 151 7. Genetic Factors 154 VI. Oral Premalignancy 165 A. Histopathology 165 B. Clinical Features 170 1. Premalignant Lesions 170 a. Leukoplakia 171 i. Homogeneous Leukoplakias 172 ii. Nonhomogeneous Leukoplakias 172 b. Erythroplakia 173 c. Epidemiology of Premalignant Oral Lesions 173 d. Management of Premalignant Oral Lesions 177 2. Premalignant Conditions 180 a. Lichen Planus 180 b. Submucous Fibrosis 184 c. Tertiary Syphilis 186 d. Plummer-Vinson Sydrome 187 e. Dyskeratosis Congenita 187 VII. Oral Squamous Cell Carcinoma A. Histopathology 190 191 vii B. Clinical Features 195 1. Adjuctive Diagnostic Investigations 200 C. Classification and Staging of Oral/Pharyngeal Cancers 202 D. Epidemiology of Oral/Pharyngeal Squamous Cell Carcinoma 204 1. Overview 204 2. Cancer Statistics for Canada 210 3. Cancer Statistics for British Columbia 212 E. Treatment Modalities for Oral/Pharyngeal Cancer 213 1. Surgical Resection 214 2. Radiotherapy 217 a. Mechanisms of Action 217 i. Radiation Physics 217 ii. Radiation Biology 220 b. Clinical Considerations 226 c. Complications of Radiation Therapy 227 i. Acute Complications 227 ii. Chronic Complications 228 d. Clinical Applications of Radiotherapy for Oral Cancer 230 e. Radiotherapy for Lingual Cancer 232 i. Oral Tongue 232 ii. Base of Tongue 232 3. Chemotherapy 233 a. Mechanisms of Action 233 b. Clinical Applications 235 4. Treatment of Tongue Cancer at the Memorial Sloan-Kettering Cancer Center 236 F. Follow-up and Prevention 240 G. Summary 245 CHAPTER 2 - METHODS 249 CHAPTER 3 - RESULTS 252 I. Descriptive/Association Statistics 252 A. Cases by Year and Sample Demographics 252 1. Cases by Year 252 2. Gender, Age and Stage of Disease 252 B. Risk Factors 253 1. Alcohol 253 2. Tobacco 254 C. Symptoms 254 D. Tumour Characteristics 255 1. Tumour Location 255 2. Cell Type 256 viii 3. Cell Differentiation 257 4. Tumour Status, Node Status and Stage of Disease 257 E. Treatment 258 1. Treatment Modalities 258 2. Complications of Treatment 259 II. Survival Analyses 277 A. Overall All-Cause Survival 277 B. Disease-Specific Survival 278 C. Local Recurrence 278 D. Regional Recurrence 278 E. Distant Metastasis 279 III. Summary of Results 279 C H A P T E R 4 - D I S C U S S I O N 286 I. Descriptive Associations 286 A. Demographics, Stages of Disease, Symptoms and Risk Factors 286 B. Tumour Characteristics 287 C. Treatment 288 1. Treatment Modalities 288 2. Complications of Treatment 289 II. Survival 292 C H A P T E R 5 - F U T U R E D I R E C T I O N S 301 N O M E N C L A T U R E 307 B I B L I O G R A P H Y 310 ix Figure Number CHAPTER 1 Section II Figure 1.1 Figure 1.2 Figure 1.3 Section III Figure 1.4 Section IV Figure 1.5 Figure 1.6 CHAPTER 3 Section I Figure 3.1 Figure 3.2 Figure 3.3 Figure 3.4 Figure 3.5 Figure 3.6 Figure 3.7 List of Figures Title Page Taxonomy of Epidemiologic Studies 54 Designs of Case-Control Study and Cohort Study 55 The Natural History of a Disease 58 The Dorsal Surface of the Tongue 76 The Cell Cycle 107 Control of the Cell Cycle 108 Number of Cases Admitted to Vancouver Branch, BCCA, by Year, 1979 - 1994 262 Gender and Age Distribution of Patients 263 Distribution of Stage of Disease by Age Groups 264 Distribution of Stage of Disease by Gender 265 Patient Symptoms and Duration of Symptoms 270 Distribution of Patient Symptoms by Tumour Location 271 Distribution of Patients by T Status and Clinically Positive Neck Nodes 274 List of Tables Table Number Title Page CHAPTER 1 Section II Table 1.1 Table 1.2 Contingency Table to Aid in the Calculation of Measures of Association Comparison of a New Diagnostic Test to the Gold Standard 56 57 Section IV Table 1.3 Table 1.4 Table 1.5 Table 1.6 Distribution and Classes of Keratin Filaments in Oral Mucosa 105 Summary of Some Effects of Retinoic Acid on Oral Epithelium 106 Some Factors that Enhance Proliferation of Epithelial Cells 109 Some Factors that Suppress Proliferation of Epithelial Cells 110 Section V Table 1.7 Table 1.8 Table 1.9A Table 1.9B Table 1.10 Chromosomal Alterations in Squamous Cell Carcinoma of the Head and Neck 156 Biomarkers and Possible Roles in Carcinogenesis 157 Keratin Markers in Oral Epithelial Dysplasia 159 Keratin Markers in Oral Squamous Cell Carcinoma 160 Risk Factors, Odds Ratios and Relative Risks Associated with Oral and Pharyngeal Cancer 161 Section VI Table 1.11 Examples of Transformation Rates of Oral Leukoplakia to Malignancy 189 Section VII Table 1.12 Table 1.13 Table 1.14 TNM Classification of the Lips, Oral Cavity and Pharynx Summary of Cancer Statistics for Canada Summary of Cancer Statistics for British Columbia 246 247 248 CHAPTER 2 Table 2.1 Criteria for Clinical Status at Last Follow-up used in Survival Analyses xi 251 CHAPTER 3 Section I Table 3.1 Table 3.2 Table 3.3 Table 3.4 Table 3.5 Table 3.6 Table 3.7 Cases of Lingual SCC by Year and Patient Demographics 260 Risk Factors Associated with Patients 266 Alcohol and Tobacco Consumption and Percentage Distribution by Age, Stage of Disease and Tumour Location 267 History of Patient Symptoms 268 Tumour Characteristics 272 Summary of Treatments for SCC of the Tongue by Stage of Disease 275 Surgical Treatment, Radiation Therapy, Surgical Pathology and Complications 276 Section II Table 3.8 Table 3.9 Table 3.10 Prognostic Factors for Survival and p Values 280 Prognostic Factors and Survival Proportions at 5 years 281 Summary of Patient and Tumour Characteristics, Treatment and Survival of Patients with SCC of the Tongue Admitted to Vancouver Branch, BCCA Between 1979-1994 283 CHAPTER 4 Table 4.1 Survival of SCC of the Oral Tongue and BOT by Study, Site, Stage of Disease and Treatment Focus 296 Table 4.2 Complications of Radiotherapy by Study, Site and Treatment Focus 299 Acknowledgements xii I would like to acknowledge the Master of Dental Science Thesis Committee, Dr. N. Le, Dr. J. Epstein, and Dr. B. Blasberg as well as Dr. L. Zhang who withdrew from the Committee but kindly agreed to act as Faculty Examiner of the thesis. In particular, I am indebted to Dr. Nhu Le for his analysis not only once, but twice of the data set that comprised the Epstein BCCA case series! I would like to thank Dr. Joel Epstein for his review of the thesis and for his helpful comments. I would also like to thank and acknowledge the contributions of Dr. J. Epstein and Dr. E. van der Meij for generating the data set that was provided to Dr. Le for analysis, and subsequently to myself for review and critique. I would like to thank and acknowledge Mr. Jack Zolty who very kindly contributed his computer expertise to create the seven graphs in Chapter 3. I would also like to acknowledge the helpful librarians at the BCCA who provided numerous journal articles for review in the thesis. I am grateful to Dr. Don Brunette, Associate Dean, Faculty of Dentistry, for providing generous access to a Macintosh computer with Word Perfect Software and a laser printer, as well as for providing the photocopied versions of this thesis for the use of the Committee members. I am indebted to Dr. Michelle Williams, Dr. Helen Scott and Ms. Bonnie Craig for their friendship and moral support. Finally, I would like to express my gratitude to my family and especially to my husband, John. There have been too many sacrifices made and priorities misplaced; it has been a long, hard lesson but it was well-learned. 1 CHAPTER 1 INTRODUCTION I. Overview In the United States, oral/pharyngeal cancers represent 3% (Wingo et al., 1995) to 4% (Boring et al, 1991) of all body cancers in males and 2% of all body cancers in females (Boring et al., 1991; Wingo et al., 1995). For all cancer-associated deaths in the United States, oral/pharyngeal cancer is attributable for 2% of deaths in males and 1% of deaths in females (Boring et al., 1991; Wingo et al., 1995). In Canada, cancer statistics demonstrate similar trends: in 1991, there were a total of 3017 new cases of oral/pharyngeal cancers, representing 2.8% of all body cancers, and the number of Canadian deaths attributable to oral/pharyngeal cancer in 1993 was 995 cases which represented 1.8% of all deaths attributable to cancer (National Cancer Institute of Canada 1996). For 1996, the estimated age-standardized incidence rate for oral/pharyngeal cancer was 15 per 105 Canadian male population and 5 per 105 Canadian female population (National Cancer Institute of Canada 1996). In general, incidence rates of oral/pharyngeal cancer are 2.5 times as high in males as in females and tumours typically occur in the fifth and sixth decades of life (Shaha and Strong, 1995). The 5-year survival rates for oral/pharyngeal cancer "average 52%" and "rates have shown little significant change over the last 15 years" (Garfinkel 1995a). However, survival rates vary considerably for different sites with lips having the most favourable survival rate at 90%, the mouth 53%, the tongue 42%, and the pharynx 32% (Garfinkel 1995a). Squamous cell carcinoma (SSC) is a malignant neoplasm of the stratified squamous epithelium that lines the oral cavity and pharynx. SCC accounts for 80% to 90% of all oral/pharyngeal malignancies (Krolls and Hoffman, 1976; van der Waal and Pindborg, 1986), and nearly half of all SCCs of the upper aerodigestive tract occur in the oral cavity (Shaha and Strong, 1995). Excluding lip lesions, between 25% and 40% of all oral SCCs occur on the tongue (Regezi and Scuibba, 1989; Spitz 1994). In the United States, 6400 new cases of tongue cancer and 1820 deaths from tongue cancer were estimated for 1997 (Parker et al., 1997). In British Columbia, 2 Canada, 81 new cases of tongue cancer and 32 deaths from tongue cancer were estimated for 1996 (BC Cancer Agency et al., 1995-1996 Annual Report). In the 16-year period between 1979 and 1994, 328 patients with SCC of the tongue were admitted to the Vancouver branch of the British Columbia Cancer Agency (BCCA). The purpose of this thesis is to provide a retrospective, descriptive analysis of that case series (Chapters 2, 3, 4). In addition, Chapter 1 of the thesis provides a general overview of oral/pharyngeal cancer and tongue cancer in particular. Chapter 1 was written with the intent of compiling an instructional aid for graduate students or dental residents who have an interest in oral oncology from either a clinical- or laboratory-based research focus. In addition to providing a broad understanding of topics such as cell biology, carcinogenesis, precancer, and cancer treatment, an introduction to Epidemiology was included to provide a perspective for the balance of the thesis in terms of understanding the strengths and shortcomings of previous investigations into oral/pharyngeal cancer, and for appreciating the potential problems encountered in integrating laboratory-based research with the clinical setting. Chapter 1 begins with an introduction to fundamental epidemiological concepts (Section Lf) that should provide a perspective for subsequent sections which discuss the diagnosis, epidemiology, treatments and outcomes of oral SCC. A critique of the oral/pharyngeal cancer literature is included in Section II to illustrate certain epidemiological concepts as well as the strengths and shortcomings of previous investigations. As epidemiology is the study of disease in a population, its distribution and determinants, it is also helpful to understand the biology underlying the disease being studied. Cancer is basically a disease of cells characterized by the loss of control mechanisms that govern cell proliferation and differentiation; but in order to appreciate these aberrant processes, the normal functions and structure of the affected tissues must first be understood. Hence, the development, anatomy and function of the tongue are reviewed in Section HI; the structure and function of oral epithelium and its normal proliferation and differentiation are 3 reviewed in Section IV. Carcinogenesis, biomarkers of malignancy and risk factors for oral/pharyngeal cancer are reviewed in Section V; oral premalignancy is discussed in Section VI, and oral/pharyngeal SCC and lingual SCC in particular are reviewed in Section VII. In Section II, the reader will become aware of different terminologies and different methodologies that have been employed by the numerous investigators of head and neck cancers. For example, the terms "oral", "pharyngeal" or "oropharyngeal" cancer often appeared in the literature without further clarification or specification of which anatomic sites were included by the investigators. The descriptors "oral cancer" and "pharyngeal cancer" can designate a variety of different anatomic sites which may or may not include the lips, nasopharynx, oropharynx or hypopharynx, and may or may not include tumours of the salivary glands. As well, the descriptor "tongue cancer" may or may not include both the anterior two-thirds and posterior third of the tongue, and again, details were not consistently provided by the investigators (Section II. D). Throughout this thesis, the term "oral/pharyngeal" was used as a general term encompassing the upper aerodigestive tract, but details about the specific sites included in an investigation were included if they were described by the authors, and if that information was relevant to the discussion. In similar manner, the authors' phrases/descriptors to report results and conclusions were used; consequently, if some statements appear ambiguous or lack detail, it is because additional information to enable better understanding of the data was not provided by the authors of the cited study. In addition, if results were stated to be statistically significant, p values or confidence intervals were included if they were available in the cited study. Throughout this thesis, figures and tables are numbered consecutively in order of appearance in the text within each chapter, hence figure or table 1.1, 1.2, etc. for Chapter 1; 3.1, 3.2 for Chapter 3, etc. Within each chapter, the figures and tables are included, in order of appearance, at the end of the section in which they are cited. 4 II. Principles of Epidemiology This Section will provide definitions and descriptions of the epidemiological strategies that are encountered in the subsequent sections of Chapter 1. The intent of this section is not to provide a comprehensive review of epidemiology but, rather to provide a perspective or point of reference for interpreting the literature on oral/pharyngeal cancer. A. Definitions and Taxonomy Epidemiology is predicated on two assumptions; first, that human health and disease are not randomly distributed and secondly, that human disease has causal and preventive factors that can be identified (Hennekens and Buring, 1987). Implicit in these assumptions is the premise that the diseased state can be clearly distinguished from the healthy state so that the treatment that will provide the best prognosis can be selected; consequently, the classification of disease is also a predictive process and it is related to different concepts of disease (Wulff, 1976; Hennekens and Buring, 1987). Wulff (1976) distinguished two major concepts of disease, the nominalistic or patient-oriented concept, and the essentialistic concept which emphasizes disease as an independent entity. The nominalistic concept (Wulff 1976) is based on the view that "disease" does not exist as an independent entity and that disease classification is really a classification of sick people or patients. Thus, a particular "disease" is defined by the group of characteristics which occur more often in the patients of concern than in other people; patients will have a pattern of symptoms that resemble each other, and their prognosis and treatment have some common features. The nominalistic approach does not require a definition of "normality" and recognizes that definitions of disease may vary between different societies (Wulff 1976). The essentialistic view (Wulff 1976) is closely related to a modern concept of disease termed biochemical fundamentalism (reviewed by Dabelsteen and Mackenzie, 1987) which is based upon 5 the idea that disease can be described in terms of biochemistry and molecular biology. Diseases are assumed to follow regular patterns and once the underlying biochemical events are understood, the course of the disease can theoretically be predicted. Hence, the classification of disease becomes a matter of biotechnology yet defining the "normal" state is avoided by relying upon statistical terms to define disease. That is, disease is defined by the distribution of certain features in a particular population of people, and the extent to which that distribution differs from a similar assessment of a group whom the investigators consider "healthy" or not diseased (Wulff 1976; Dabelsteen and Mackenzie, 1987). This statistical approach forms the basis for utilizing biomarkers as diagnostic or screening tests, and as prognostic indicators (Sections II.H and II.J). Patients, clinicians and researchers generally agree that disease infers a derangement in anatomy, biochemistry, physiology or psychology, yet they rarely agree on the exact criteria defining the disease or disorder that is the target of the diagnostic process. Patients, as a result of having the target disorder, exhibit symptoms which are manifestations of the disorder that they themselves perceive, and signs which are manifestations perceived by the clinician (Sackett et al., 1991). Sackett et al. (1991) suggest that the cluster of symptoms and signs comprise the "illness", but the illness is also associated with social, psychological and economic factors of the patient's environment and they are collectively known as the "predicament". Thus, the act of clinical diagnosis focuses on the illness in order to identify the target disorder or disease, but it must do so in the context of the predicament (Sackett et al., 1991). The need to correctly assess the disease or target disorder appears self-evident, yet it remains a problem in many epidemiological studies including those investigating oral/pharyngeal cancer (eg. see Section U.D). Significantly, the failure to achieve clear definitions and accurate diagnoses of the target disease will cause systematic errors in the classification of the disease which in turn, distorts the natural history of the disease, its etiology (determinants), outcomes and treatments of the disease. As well, the suspected determinants or exposures of interest must be clearly defined 6 and assessed. Risk markers are determinants that cannot be modified and they are intrinsic characteristics of an individual such as age, gender, genetics, race, etc. Risk factors are determinants that can be modified such as lifestyle behaviours of smoking, diet, exercise, etc. Epidemiology is the study of the distribution, frequency and determinants of a disease in human populations. Distribution addresses the "who, when and where" aspects of the disease, and frequency involves quantifying the existence or occurrence of the disease (Hennekens and Buring, 1987). The distribution and frequency of disease are essential to formulating and testing hypotheses about the determinants or the causal/preventive factors of the disease to which individuals in a population are exposed. An epidemiologic hypothesis begins with a suspicion concerning the possible influence of a particular factor on the occurrence of a particular disease (Hennekens and Buring, 1987). This suspicion may arise from disease patterns in a population, from clinical practice, from laboratory research or even from theoretical speculation, but in each instance, the hypothesis must make biological and scientific sense. While basic research provides biological understanding of why an exposure causes or prevents disease, only epidemiology can quantify the magnitude of the exposure-disease relationship (risk) in humans and subsequently offer the possibility of altering risk through intervention (Hennekens and Buring, 1987). The taxonomy of epidemiological studies differs in specifics but generally separates studies according to whether they focus on describing the distribution of the disease or elucidating determinants of the disease by testing specific hypotheses (Figure 1.1). Descriptive epidemiology is a series of observations about the distribution (who, where, when) of the disease and how the frequency of disease varies over time. The observer may a have a suspicion about the relationship between variables (i.e. the exposure to causative agents and the disease distribution) but the observer is passive and does not interfere or manipulate the variables. The data obtained from, descriptive/observational studies are used to formulate hypotheses (a posteriori hypotheses). Analytic epidemiology focuses on determinants of disease by testing hypotheses (a priori 7 hypotheses) that were generated by descriptive/observational studies. In analytic studies there is an explicit comparison of exposure and outcome status and the use of an appropriate comparison group permits investigation into whether exposure to particular agent(s) causes or prevents the outcome. Depending upon the degree of intervention and control of the variables by the investigator, analytic studies can be observational or experimental (Figure 1.1), (Hennekens and Buring, 1987; Sheps 1995; Brunette 1996). The degree of exposure to a particular causative/preventive agent may change over time in a cohort study (Section II.C.2) even though this change was not induced or controlled by the investigator; if the effects of disease frequency due to changes in exposure can be analyzed, the study is analytic rather than descriptive but it is not experimental (Sheps 1995). This Section focuses on descriptive and analytic/observational studies because the epidemiologic studies relevant to this thesis have generally been limited to case series, case-control studies and a few cohort studies rather than intervention studies (clinical trials). B. Descriptive Studies Descriptive epidemiology describes the general characteristics of the distribution of a disease in relation to who, where and when. "Who" includes demographic factors such as age, gender, race, marital status, occupation as well as life-style related variables such as diet or medication use. "Where" refers to the geographic distribution of the disease such as rural, urban, and variations between and among countries. "When" may refer to a specific time intervals such as seasons or years, or may compare the disease frequency between different time periods such as the present and 100 years ago. Three main types of descriptive studies are the correlational study, case reports/series and cross-sectional surveys (eg. Sheps 1995). 8 1. Correlational Studies Correlational or ecological studies are widely used in cancer research because these studies are inexpensive, the data are usually available and the hypotheses they generate could prove useful for future research (Brunette 1996). Correlational studies use data obtained from entire populations to compare outcome (disease) frequencies between different groups within the population during the same time period or, in the same population at different points in time. The whole population is used to correlate an exposure with an outcome but it is not known how many of the exposed individuals had the outcome, or how many of individuals with the outcome had the exposure. That is, it is not known on an individual basis, the relationship between the exposure and the outcome. The data from correlational studies may be used to generate hypothesis that can subsequently be tested in individuals. However, the data from correlational studies cannot be extrapolated to individuals from the population (ecological fallacy) and it is not possible to link an exposure to disease in the same individual (Hennekens and Buring, 1987; Sheps 1995; Brunette 1996). Data in any epidemiological study is obtained in a sample from a particular source population. Although it is generally assumed that the sample is representative of the source population, it is not known if the findings obtained are generalizable and applicable to other populations, or in other words, if the study has external validity (Hennekens and Buring, 1987; Brunette 1996). For example, correlational studies in India suggested that chilli consumption was linked to submucous fibrosis (Section VI.B.2.b) because this condition is common among Indians and other populations who used chillies to spice their food. Yet in Mexico, South America and other Asian countries where consumption of chillies is also widespread, submucous fibrosis is unknown. Epidemiological studies generally follow a hierarchy or progression in which correlational studies often provide the first suggestion of a link between a disease and its determinants. Case reports and series (Section II.B.2) may also raise suspicion and together with case-control and cohort studies, generate a hypothesis that can be tested in intervention studies. For example, correlational 9 studies in South Africa revealed that submucous fibrosis was common among women of Indian origin but rare among blacks, and corresponded with the observation that areca nut chewing was often practiced by Indian women but rarely by South African blacks. Over a 5-year period, case reports and series of submucous fibrosis in a particular Indian community corresponded with the increased popularity of areca nut chewing in that area, and indicated the presence of the habit among patients with submucous fibrosis. These observations led to the formulation of a hypothesis linking areca nut chewing to submucous fibrosis. Case control studies provided estimates of the risks for areca nut chewing and provided evidence for a dose-response relationship. Eventually, the hypothesis linking submucous fibrosis and areca nut chewing was tested in and supported by intervention studies (reviewed by Murti et al., 1995). 2. Case Reports and Series Chapter 3 of this thesis reviews a case series of 328 SCCs of the tongue. Case reports and series are the most basic type of descriptive study, based on either a single patient or a number of patients, respectively. Cases may be selected for either a disease or exposure of interest and the report should specify which criterion was used. The proposed correlation between the exposure and outcome must be biologically plausible but the correlations obtained from individuals cannot be used to infer the same relationship in a population (atomistic fallacy), (Hennekens and Buring, 1987; Sheps 1995). Surveillance programs such as communicable disease and cancer registries typically use accumulating case reports to suggest the emergence of new diseases or trends such as epidemics. For example, a cluster of cases of Pneumocystis carinii pneumonia in previously healthy homosexual men was the first indication of a previously-unknown disease, subsequently called AIDS (Hennekens and Buring, 1987). 10 3. Cross-Sectional Surveys In cross-sectional surveys, the status of individuals with respect to both the exposure and disease are assessed at the same point in time. These surveys are useful for obtaining prevalence data (Section II.D.2.) but the criteria for cases must be very precisely defined in order to distinguish between prevalent (old and new cases) and incident cases (new cases only). Cross-sectional surveys cannot determine cause or effect because the temporal relationship between exposure and outcome cannot be clearly determined. That is, cross-sectional surveys cannot distinguish whether the exposure preceded the outcome or whether the outcome affected the level of exposure. Nevertheless, the study should be structured so that the exposure does not appear to be contiguous with the outcome (disease) and that the exposure can logically be seen to precede the outcome (Hennekens and Buring, 1987; Sheps 1995). C. Analytic Observational Studies Analytic studies include appropriate comparison groups for generating or testing hypotheses. In analytic observational studies, the investigator only observes the natural course of events, noting who is exposed or not exposed and who developed or did not develop the outcome, but the investigator does not manipulate the exposure. Analytic observational studies include the case control and cohort studies (Figure 1.2). 1. Case Control Studies In case-control studies, investigators look backwards in time to assess the effect of an exposure on a disease that has already occurred (Figure 1.2). A series of patients with the disease of interest and a control or comparison group without the disease are selected for investigation, and their respective exposures are determined retrospectively. Cases and controls must be similar in all respects so that controls could have been cases if they had developed the disease of interest. Case control studies are relatively inexpensive and less time-consuming than cohort studies. Case-11 control studies are well-suited for the study of rare diseases or diseases with a long latency, and to investigate a number of risk factors or potential exposures for a single outcome (such as SCC of the tongue). In case control studies both the disease and exposure have already occurred; cause and effect is therefore more difficult to establish and there is increased potential for sampling and measurement/observation bias (Sackett 1979), (Section H.F.3.). In addition, case-control studies provide odds ratios which estimate the relative risks of the exposure for the disease (Section HE.), (Hennekens and Buring, 1987; Sheps 1995; Brunette 1996). 2. Cohort Studies In cohort studies, investigators assemble a group of individuals without the disease of interest and classify them on the basis of the presence or absence of exposure to a factor of interest (Figure 1.2). Controls must be comparable to cases in all respects except for exposure to the determinant being investigated. In prospective cohort studies, the exposure has occurred but the outcome has not; therefore the subjects are followed for a specified period of time to determine the development of disease in each exposure group. In retrospective cohort studies, both the exposure and outcome of interest have already occurred at the time the study is started. Cohort studies are well-suited for evaluating rare exposures or multiple outcomes of a single exposure but they are more expensive and time-consuming than case-control studies which cohort studies usually follow. Retrospective cohort studies are susceptible to biases similar to case-control studies. Prospective cohort studies have less potential for selection bias but face increased potential for confounding factors (II.F.2) and loss of subjects to follow-up (Hennekens and Buring, 1987; Sheps 1995; Brunette 1996). D. Measures of Disease Frequency Measures of disease frequency are used to quantify the occurrence of disease relative to the size of the source population and to the time period during which the data were collected. However, unless a common time frame and common unit of population are employed, it is not possible to make direct comparisons of disease frequencies between different populations. In addition, the 12 definition of the disease and the reporting/recording criteria for the disease must be consistent over time and between different populations. For example, throughout the world and even within Canada, there are differences in the criteria for defining cancer cases, and registration and follow-up of cases are not consistent over time or between populations (see below). In the United States, there is no nationwide cancer registry so there is no way of knowing exactly how many new cases of cancer are diagnosed each year, and although the SEER program (Surveillance, Epidemiology and End Results) was instituted in 1973 to collect cancer data, only about 10% of the US population is covered (Wingo et al., 1995). Common problems with collecting cancer data include failure of the reported cases to be histologically verified, failure to specify site and histopathologic type, and failure to report cases altogether (Ostman et al., 1995). Regional differences in early detection activities or availability of overall diagnostic procedures can also affect cancer rates, and may explain why different Canadian provinces report different rates for cancers such as prostate and breast; moreover, increased incidence of prostate or breast cancers may simply reflect improved diagnostic procedures over time (National Cancer Institute of Canada, 1996). Incomplete or delayed reporting of cases to registries or changes in reporting also affect disease frequency; for example, in Manitoba changes in coding have considerably reduced previous artifactual overestimations of female cancers since originally-reported invasive tumours of the cervix and breast were recoded as in situ tumours (National Cancer Institute of Canada, 1996). Differences in the classification of disease prevents the comparison of data from different sources. For example, comparisons of oral/pharyngeal cancer data are thwarted by inconsistent groupings of different cancer sites. That is, a variety of anatomic sites are grouped together although the lesions at these different sites may not reflect the same disease process (eg. Sauter et al., 1992; Greenblatt et al., 1994). Some reports classify sites according to specified revisions of the ICD (International Classification of Diseases; eg. ICD-9) wherein lip is designated as 140, oral tongue and base of tongue are designated as 141, major salivary glands as 142, gum as 143, FOM as 144, other and unspecified parts of the mouth as 145, oropharynx as 146, nasopharynx as 147, 13 hypopharynx as 148, other and ill-defined sites within the lip, oral cavity and pharynx as 149, larynx as 161, etc. The National Cancer Institute of Canada (1996) collectively groups ICD-9 sites 140-149 together under the heading of "oral" cancer. The BC Cancer Agency et al., (1995-1996) consider the lip, tongue, mouth and pharynx as separate cancer sites in their data analysis, but it is not clear whether base of tongue lesions are included under tongue or pharynx (see Section HI). In addition to official registries, different investigators of oral/pharyngeal cancer have also included a variety of site categories in their analyses; for example, Blot et al (1988) and Winn et al., (1991) included ICD-9 sites 141-149 as "oral and pharyngeal" but excluded 142 (salivary glands) and 147 (nasopharynx); Cox et al. (1995) and Ostman et al. (1995) used the term "oral cancer" to include the lip, tongue, salivary glands, mouth and all parts of the pharynx; Boffetta et al. (1992) used the term "oral cancer" to include the oropharynx and oral cavity but it was not clear whether lip and salivary cancers were included; Brugere et al. (1986) included lips, oral cavity, larynx and pharynx but it was unclear whether nasopharynx was also included; Oreggia et al., (1991) limited their analyses to the "tongue" but did not specify whether the base of tongue (BOT) was also included. An obvious advantage to grouping together all cancers of the oral cavity and/or pharynx is the inclusion of a greater number of cases which enhances the power of statistical analysis. However, this approach precludes the analysis of data by specific sites which may represent site-specific etiologic effects and site-specific mutational differences (eg. Sauter et al., 1992; Greenblatt et al., 1994) . That is, carcinogens like tobacco and alcohol may have different effects in different sites, but risks for individual sites can not be determined from pooled data. Moreover, while most reports included men and women, some investigations (eg. Brugere et al., 1986; Macfarlane et al., 1995) were limited to men and excluded women. Overall, general conclusions across studies and between countries over time for trend data of oral/pharyngeal cancer cannot easily be drawn. 14 1. Ratios, Proportions and Rates The number of cases of a disease or outcome relative to the population in which they occurred can be described by three types of mathematical relationships: ratios, proportions and rates. Ratios take the general form "a/b" in which "a" and "b" are not necessarily related to one another and no specific relationship between them is implied, such as the number of females to males, or the number of smokers to nonsmokers. Ratios have no units and are a general term for more specific measures such as proportions, percentages and rates. Proportions are a type of ratio with the general form "a/a+b" where observations in the numerator ("a") are included in the denominator ("a+b"). Ratios of a part to the whole are often expressed as percentages; for example, 66% of all oral/pharyngeal cancers occur in smokers and alcohol drinkers (Blot et al., 1988). A rate is a ratio in which the numerator and denominator are distinctly related to one another and the denominator is a function of time which is reflected in time units (Hennekens and Buring, 1987; Sheps 1995; Brunette 1996). A rate may be a crude rate or a category-specific rate (see II.D.3 below). The term 'rate' is often used indiscriminantly to refer to rates, proportions or ratios and it is often unclear what measures constitute the numerator and denominator. That is, it is often unclear whether the numbers represent the number of events, or the number of individuals in whom the events occurred. For example, in describing the transformation frequency of premalignant lesions (Section VLB) to malignancy, some investigators utilized the number of lesions (eg. Mashberg et al., 1973; Cawson and Binnie, 1980) as their measure; some investigators designated the number patients (eg. Einhorn and Wersall, 1967; Banoczy 1977; Murti et al., 1986) as their measure even though more than one lesion occurred in individual patients (eg. Mincer et al., 1972; Silverman et al., 1984, 1985), and some investigators utilized both lesions and patients without clear distinction (eg. Lummerman et al., 1995), (Table 1.11, Section VI). Moreover, the proportion of premalignant lesions that subsequently become malignant is typically referred to as the "transformation rate" (Section VLB) yet the relevant time intervals are often obscured elsewhere in the article, are not included directly in the rate, and vary from 1 year to over 40 years without 15 conversion to an annual rate (Table 1.11, Section VI). As a result, the literature is confusing and generally non-comparable. 2. Prevalence and Incidence Prevalence and incidence are the most commonly used measures of disease frequency in epidemiology. a. Prevalence Prevalence is the proportion of individuals in a population which have the disease at either a specific point in time (point prevalence) or during a specified time period (period prevalence). Time units are generally not included in prevalence and therefore prevalence is not a "true rate" (Sheps 1995). Prevalence also provides a crude estimate of the probability or risk that a given individual in the source population will have the disease at a certain time, either calendar time or during their lifetime (Hennekens and Buring, 1987; Sheps 1995).. Number of Existing Cases Prevalence = Total Population The number of existing cases includes both "old" and "newly-diagnosed" cases of the disease (Hennekens and Buring, 1987; Sheps 1995). b. Incidence Incidence is a measure of the natural frequency of disease in a population and it provides an estimate of the probability or absolute risk than an individual will develop the disease during a specified period of time (Sheps 1995). Incidence is a crude rate (see Section HE. below) and there are two measures of incidence, cumulative incidence (CI) and incidence rate or density (ID). Incidence = (CI) Number of New Cases During a Given Period of Time Total Population at Risk 16 Cumulative incidence refers to the number of new cases of disease that develop in a population of individuals at risk during a specified time interval. "At risk" refers to individuals who have not yet developed the disease but who could potentially develop the disease. Consequently, individuals who already had or currently have the disease, or individuals who cannot develop the disease should be eliminated from the denominator (Hennekens and Buring, 1987; Sheps 1995). For example, in calculating the incidence of caries, individuals with existing restorations but without present caries should be excluded from the numerator because they represent "old" cases of caries rather than "new" cases; if they were included, the resultant measure would overestimate the true incidence of caries. In like manner, individuals without natural teeth should be excluded from the denominator because they are not at risk for the disease. When persons not at risk are included in the denominator, the resultant measure underestimates the true incidence of disease. Cumulative incidence assumes all individuals at risk have an equal chance for being diagnosed as a case, and that the entire population at risk at the start of time period was followed for the entire time period. These assumptions are often unrealistic as individuals are lost to follow-up, resulting in variable lengths of observation for the participants. Therefore, incidence rate or incidence density which measures the instantaneous rate of development of disease in a population, is used (Hennekens and Buring, 1987; Sheps 1995). Number of New Cases of Disease During a Given Time Period Incidence Rate = Total Person-Time of Observation 3. Crude, Category-Specific and Adjusted Rates Rates can be presented for an entire population (crude rate) or for segments or strata of the population based on particular characteristics such as age, gender, etc (category-specific rates). 17 a. Crude Rates Crude rates refer to an entire population and they are a summary measure of the total number of cases of the outcome in the population divided by the total number of individuals in that population in a specified time period. For example, in 1990 the number of new cases of oral/pharyngeal cancer in the United States was 10.4 per 105 population, and the mortality rate from oral/pharyngeal cancer was 3.0 per 105 population (Garfinkel 1995a). Crude rates are easy to calculate and comprehend, and they represent the actual experience of a particular population. Consequently, crude rates are useful for making decisions regarding health resource utilization and public health planning. However, a problem arises in comparing crude rates among different populations because different populations may differ with respect to certain underlying characteristics such as age, gender or race which may confound (Section n.F.2) the outcome. For example, cancer mortality rates rise dramatically with increasing age and a population in which the elderly represent a large segment of the population will have a higher crude cancer mortality rate than other populations in which all age groups are more evenly represented. In order to account for differing distributions of a characteristic between populations being compared, category-specific rates in each population can be compared (Hennekens and Buring, 1987). b. Category-specific Rates Category-specific rates are calculated for specific categories of the population based on particular characteristics such as age, gender, or race, etc. By grouping or stratifying the data for the different characteristics, the rates are unconfounded for that characteristic and provide the most detailed information about the pattern of disease in that subgroup. For example, the crude rate of new tongue cancers in BC in 1994 of 3.5 per 105 population can be separated into category-specific rates such as by gender: rate of new tongue cancers for BC males (2.6 per 105 population) and females (0.9 per 105 population). The crude rate can be also be categorized by age which 18 demonstrates that in 1994, the highest rates of new tongue cancers for males and females in BC occurred between ages 60 and 79 years (BC Cancer Agency Annual Report 1995-1996). The category-specific rates can be compared more readily than crude rates to similar data from another time period or another geographic location, but a large number of comparisons are required. Therefore, it would be useful to have a single summary rate for each population that could account for any differences in the structure of the populations (Hennekens and Buring, 1987) and this approach is used in adjusted standardized rates. c. Adjusted Standardized Rates Adjusted standardized rates are crude or category specific rates that are adjusted for differences between populations. That is, they are "summary rates that take into account differences with respect to underlying characteristics that differ in distribution among two populations" (Sheps 1995). Once rates have been adjusted for a particular characteristic (eg age), any remaining observed differences between the populations cannot be attributed to confounding by that characteristic. Adjusted standardized rates are typically used to compare morbidity and mortality rates between different geographic regions or between different time periods in relation to a standard or reference population that is typically derived from Census data. However, it is not appropriate to compare age-standardized rates when different reference populations have been used to standardize the rates. For example, the BC Cancer Agency (1995-1996 Annual Report) used the 1971 Canadian population as the standard for calculating age standardized incidence of cancers in 1994. Unfortunately, these rates cannot be compared to the National Cancer Institute of Canada age-standardized rates from 1987-1994 or to the 1995/96 rates because different reference populations were used in each instance. The National Cancer Institute of Canada used the World Standard Population to age-standardize cancer incidence and mortality rates from 1987 to 1994, but used the 1991 Canadian population to age-standardize rates for 1995 and 1996. The reason for the change 19 in reference populations was that the World Standard Population is much younger than the 1991 Canadian population; consequently, the National Cancer Institute rates in 1995 and 1996 are about 30-50% higher than those calculated using the World Standard Population between 1987 and 1994. Hence, the increased rates in 1995/96 do not reflect a sudden increase in the number of cancer cases and deaths; instead, they more closely reflect the actual incidence of cancer per 105 Canadian population which has a much higher proportion of people in the older age groups, in which cancer is much more common (National Cancer Institute of Canada, 1996). Similar discrepancies in reference populations exist in age-standardized cancer mortality rates calculated for the United States. Prior to 1992, US Cancer mortality statistics were standardized to the 1970 US population; starting in 1992, mortality rates were age-adjusted to the WHO standard world population (Wingo et al., 1995). E. Measures of Association An epidemiologic hypothesis attempts to link a specific exposure to a disease. The relationship between an exposure and disease is known as an "association" and it refers to the statistical dependence between two variables. Association is also the degree to which the rate of disease in individuals with a specific exposure is either higher or lower than the rate of disease among individuals without that exposure (Hennekens and Buring, 1987). However, the presence of an association does not imply that the observed relationship is one of cause and effect. Judgements of causal association must first determine whether the observed association between an exposure and disease is valid. That is, whether or not the data reflect the true relationship between the exposure and the disease, and this becomes a matter of determining the likelihood that alternative explanations such as chance, bias or confounding (Section II.F) could account for the findings. If these factors are determined to be unlikely explanations for the data, then it can be concluded that a valid statistical association exists between the exposure and the disease (Hennekens and Buring, 1987). 20 However, the presence of a statistical association (Section JJ.F) is still not sufficient to establish the observed relationship as one of cause and effect. Causal association must be evaluated with respect to strict criteria (eg. Hennekens and Buring, 1987; Brunette 1996) that include consistency of the findings with other investigations, the strength of the association, biologic gradient (dose-response relationships), temporal sequence of causes and events, biological plausibility, and substantiation by experimental research (Hennekens and Buring, 1987; Sheps 1995). For example, Newcomb and Carbone (1992) reviewed the data for support of a causal association between cigarette smoking and cancer. First, consistency has been demonstrated through numerous studies of different designs in diverse populations over many time periods. Secondly, the strength of the association (odds ratios, relative risks) has been consistently observed to be positive, and thirdly, a dose-response is evident for all cancers with a suspected link to cigarette smoking. That is, gradients of risk are observed with increasing numbers of cigarettes smoked, earlier age at initiation, greater total number of years smoked, degree of inhalation and type of cigarettes smoked. Temporality has been demonstrated by numerous prospective studies which established that smoking exposure occurred prior to the cancer, and by the higher proportion of premalignant changes observed in smokers as compared with nonsmokers. Moreover, risks of most smoking-related cancers diminish after cessation and with increased duration of abstinence. Finally, biological plausibility is supported by the natural history of cancer, and by the biological effects of cigarette smoke demonstrated by molecular biology techniques and experimental animal models. Thus, there is persuasive evidence that cigarette smoking has a causal role in the development of cancer at many sites (reviewed by Newcomb and Carbone, 1992). The magnitude of the observed association is useful to judge the likelihood that the exposure itself affects the risk of developing the disease and therefore, increases the likelihood of a causal relationship (Hennekens and Buring, 1987). Measures of association compare the frequency of disease between two populations, using an overall summary measure that estimates the association between an exposure of interest and an outcome of interest. Thus, the risk or probability of 21 developing a disease over a specified period of time, can be compared in relation to the absence or presence of an exposure and the most frequently used measures of association are the relative risk, odds ratios and attributable risk (Table 1.1). 1. Relative Risk In cohort studies, potential exposures or risk factors are identified at the start of the study, prior to the development of disease. Therefore, incidence data regarding the natural frequency of disease are generated and relative risk (RR) can be calculated by using either cumulative incidence or incidence density. Relative risk is the ratio of the incidence of disease (absolute risk) in the exposed group to the corresponding incidence of disease or absolute risk in the nonexposed group, and it indicates the likelihood of developing the disease in the exposed group relative to the group that was not exposed (Hennekens and Buring, 1987; Greenstein and Lamster, 1995), (Table 1.1). Absolute Risk (Incidence) when the Risk Factor is Present Relative Risk = Absolute Risk (Incidence) when the Risk Factor is Absent The value of relative risks calculated as the ratio of two absolute risks depends upon the time period over which the risks were calculated because the frequency of disease (incidence) may change depending upon the length of observation; that is, the relative risk after 10 years may differ considerably from that after 1 year. Therefore, it is important to specify the time period on which the calculation of the risk ratio was based (Hennekens and Buring, 1987), (eg. see Table 1.10, Section V). A relative risk of 1.0 indicates that the incidence of disease in the exposed and nonexposed group are identical and therefore, there is no association between the exposure and the disease. A value less than 1.0 reflects an inverse relationship between the exposure and disease, a decreased risk or protective effect of the exposure. A value greater than 1.0 indicates a positive association, or an 22 increased risk of disease among those exposed to the factor of interest (Hennekens and Buring, 1987). Some epidemiologists contend that the mere fact that a factor has a risk ratio greater than 1.0 is not in itself a sufficient basis for implicating that factor in the causation of the disease in question. Taubes (1995) and Barnett and Mathisen, (1997) maintain that "unless there is a high degree of biological plausibility for a given factor causing a specific disease, a given risk factor should have a risk ratio of at least 3.0 or 4.0 before it is implicated in the causation of the disease". Although relative risk provides an estimate of the strength of the association between an exposure and disease, it is of limited value in predicting risk of impending disease for any individual. For example, a relative risk of 3.9 calculated for individuals who smoke tobacco products (Table 1.10, Brugere et al., 1986) means that these individuals have 3.9 times the risk or are 290 percent (i.e. 3.9 minus the null value of 1.0, times 100; based on Hennekens and Buring, 1987, page 78) more likely to develop oral SSC than nonsmokers. However, the important unanswered question for an individual is, "3.9 times the risk" of what? The information that is needed, is knowledge of the absolute risk or incidence of SCC in nonsmokers. In addition, extrapolating from relative risks calculated in a population to absolute risks for an individual must be done cautiously because it may have different significance, depending upon the circumstances. As Greenstein and Lamster, (1995) illustrate, a relative risk of 10 can indicate risks of 1 per 106 in those without the factor : 1 per 105 in those with the risk factor. A relative risk of 10 can also indicate a risk of 1 per 102 in those without the factor : 1 per 10 in those with the factor; thus, the same relative risks have different absolute risks for the individual (Greenstein and Lamster, 1995). Only a few studies (eg. Brugere et al., 1986; Murti et al, 1986; Gupta et al., 1989; La Vecchia et al., 1991; Oreggia et al., 1991; Jovanovic et al., 1993a) have reported relative risks rather than odds ratios (see II.E.2. Odds Ratios below) as the measure of association between risk factors and oral/pharyngeal cancer. In one instance (Jovanovic et al., 1993a), the authors clearly stated that odds ratios were used to obtain estimates of the relative risks, and descriptions of methodologies in 23 the balance of the reports suggest that odds ratios were also used to estimate the relative risks which were reported. For example, La Vecchia et al. (1991) used a case control study to assess the relationship between diet and oral/pharyngeal cancer and clearly, incidence data were not available to calculate RR directly. Unless certain assumptions are made and are clearly stated, odds ratios can not be assumed to be the equivalent measure of association to relative risk (see Section II.E.2 below; Hennekens and Buring, 1987). Gupta et al. (1989) selected a group of 12,212 individuals from a rural region in India who used tobacco. Over 8 years, they performed periodic examinations for the presence of clinically-evident oral precancerous lesions (Section VI) or malignancy, and performed selected biopsies when indicated by clinical suspicion. A control group of nonsmokers was not included and neither the prevalence nor incidence of premalignant lesions or malignancy in nonsmokers was provided, yet relative risks for various oral conditions were reported. The average periods of follow-up varied among the different types of precursor lesions rendering comparisons between relative risks of the different lesions invalid. Moreover, neither p values nor confidence intervals (Section II.F.l) were provided and assessment of significance was limited to a single statement in regards to lichen planus (see also Table 1.10, Section V). The apparent differences in risk calculations for oral/pharyngeal cancer (Table 1.10, Section V) between different studies may reflect inherent, true differences in the populations that were studied. However, the selection of cases and controls is prone to a multitude of biases (Section II.F.3) which can distort the results. In addition, the definition of exposures and the classification of groups by degree of exposure varies considerably between studies. For example, in assessing the association between alcohol consumption and oral/pharyngeal cancer some investigators (eg. Brugere et al., 1986; La Vecchia et al., 1991; Jovanovic et al., 1993a) used the quantity of pure alcohol consumption as the unit of exposure. La Vecchia et al. (1991) converted each beverage type to a standard measure of ethanol content; thus, 150 ml wine = 330 ml of beer = 30 ml spirits = 12 ml ethanol; consumption was ranked as low (<4 drinks of 12 ml ethanol/day), moderate (4-6 24 drinks/day) or high (>6 drinks/day) and corresponded to RR of 1.7 (NS-not significant) for moderate consumption and 5.8 for high consumption (La Vecchia et al., 1991). Brugere et al. (1986) considered that 1 glass of any alcoholic beverage contained the same quantity of pure alcohol (i.e. 15 grams (g) of ethanol) so that 1 liter of wine = 80 grams ethanol = 6 glasses; they estimated a RR of 1.0 (NS) for consumption of 0-39 g/day, RR of 2.7 for 40-99 g/day, RR of 13 for 100-159 g/day, and RR of 70 for >160 g/day. Jovanovic et al. (1993a) assumed that the amount of alcohol per beverage of hard liquor, wine or beer was equivalent to 10 grams of alcohol. Consumption of more than 4 drinks/day was considered heavy and corresponded to OR of 3.3 for SCCoftheFOM. In contrast, Mashberg et al. (1981) used "whiskey equivalents" where one ounce 86-proof whiskey equalled 12 ounces of beer, or four ounces of dry wine with an alcohol content of 11-12%. Consumption of 6-9 "whiskey equivalents'Vday corresponded to a RR of 15.2 and heavy consumption of over 10 "whiskey equivalents'Vday corresponded to a RR of 10.6 for oral/pharyngeal SCC. Oreggia et al. (1991) used total volumes of consumption as equivalent measures between the different types of alcoholic beverages so that >200 ml/day of wine had a RR of 5.8 and >200 ml hard liquor/day had a RR of 3.3. Blot et al. (1988) used the number of drinks per week as their unit of comparison and for >30 drinks/week, RR for wine was 2.5 (NS), RR for beer was 4.7 and RR for hard liquor was 5.5. Overall, it is difficult to determine to what degree the exposures in the different studies are equivalent and to what extent the risks may be compared. 2. Odds Ratios In case control studies, participants are selected on the basis of disease which has already occurred, and the size of the diseased and nondiseased groups do not necessarily reflect the natural frequency of disease. Consequently, it is not possible to calculate the rate of development of disease, and incidence data are not available to calculate relative risk. However, the relative risk can be estimated by calculating the ratio of the odds of exposure among the cases to that among the 25 controls but odds ratios can not be used to predict an individual's chance of developing the disease. Odds that a Case (Disease Present) is Exposed Odds Ratios = Odds that a Control (Disease Absent) is Exposed In like manner to relative risk, an odds ratio of 1.0 indicates that there is no association between the exposure and the disease; values greater than 1.0 indicate an association between the exposure and the disease. While odds ratios provide an estimate of relative risk, there are also differences between odds and the probability (risk) that an event will occur. In risk calculations (Table 1.1), the numerator indicates the number of times an event occurred, and the denominator displays the number of times the event could have occurred. In odds, the numerator also includes the number of events that occurred but the denominator indicates the number of times it did not occur. For example, the probability of drawing an ace from a deck of cards is 4/52 or 1/13 but the odds of drawing an ace are 4/(52-4) = 4/48 or 1/12 (Greenstein and Lamster, 1995; Brunette 1996). Although odds are slightly different than probability calculations, odds provide a valid estimate of risk under conditions that prevail in most case-control studies: the disease must have a prevalence of less than 10%, the cases of disease must be newly diagnosed (incident) and old (prevalent) cases must not be included, the selection of cases and controls must not be based on exposure and finally, the ratio of cases:controls must be 1:1 (Hennekens and Buring, 1987; Sheps 1995). 3. Attributable Risk In contrast to relative risk which is used to determine if a causal effect exists between the exposure and disease, attributable risk (Table 1.1) assumes that a cause-effect relationship exists. Attributable risk (AR) is also known as the risk difference because it measures the absolute differences in disease frequency between exposed and nonexposed populations and represents the excess risk of disease among the exposed that would remain if risks attributable to all other 26 competing exposures were removed. Attributable risk is well-suited for public health purposes since AR values greater than '0' represent the number of cases that could be prevented if the exposure was removed. If AR equals '0', there is no difference in disease frequency between exposed and nonexposed groups and therefore, no association between the exposure and disease. AR values less than '0' indicate that the exposure is beneficial (Hennekens and Buring, 1987; Sheps 1995). F. Chance, Confounding and Bias For any individual epidemiological study, the observed association between exposure and disease may be valid and reflect the true nature of relationship. It is also possible that the findings have an alternative explanation and result from chance (random errors), bias (systematic errors) or confounding. Chance can occur anytime a sample from a source population is selected and it reflects the "luck of the draw". Bias occurs when a systematic error is made in selecting the sample (selection bias) or in the way information was obtained and reported (observation or measurement bias). Confounding occurs when the effect of some other variable(s) that existed between groups in the sample was not recognized or controlled. 1. Chance Epidemiological studies assume that evaluation of a sample can be used to draw inferences about the source population but because of chance or random variation from sample to sample, it is unlikely that any two samples from the same total population will be identical. The degree to which chance affects the findings in any particular study is largely determined by the size of the sample. In general, the smaller the sample on which the findings are based, the more variability and less reliability or reproducibility there will be of the findings; conversely, the larger the sample size, the less variability and the more reliable the inference (Hennekens and Buring, 1987; Brunette 1996). 27 a. p Value Statistical tests are powerful tools used to quantify the degree to which random errors or chance may account for the results observed in any individual study. Tests of statistical significance report a measure, the "p value", which is the probability or likelihood that the observed results are due to chance alone. P values and levels of significance are typically used with the statistical test of a hypothesis to arrive at a conclusion such as whether the observed differences between two groups are due to chance or whether the differences are real. By convention, medical studies set the p value or level of significance at 5% (0.05) or 1 in 20 probability, and if p values calculated by the test are less than or equal to 0.05, the results are deemed statistically significant meaning that there is no more than a 5% probability that a result as extreme as that observed was solely due to chance. Mills (1993) explains that "a p value of 0.05 means that there is a 5% chance of concluding that the two groups differ when they actually do not (type error I)" and a "95% probability of correctly concluding that there is no difference when no difference is present". A statistically significant result does not prove that chance could not have accounted for the findings, only that chance is an unlikely explanation, and it can offer no information about the actual magnitude of the differences between the groups. Moreover, a significant p value cannot assess the adequacy of the study design or rule out the possibility that results may be due to bias or confounding. Conversely, p values greater than 0.05 do not mean that chance was responsible for the findings or that the association cannot be causal; it only means that chance cannot be excluded as a likely explanation (Norman and Streiner, 1986; Hennekens and Buring, 1987; Potter 1994; Brunette 1996). Unfortunately, many studies use statistical significance or lack thereof as the sole criterion for decision-making, failing to differentiate statistical significance from clinical significance or from biological plausibility (eg. Sackett et al., 1991; Barnett and Mathisen, 1997). It should also be realized that p values are a composite measure that reflect both the sample size and the magnitude of the difference between two groups. If the sample size is too small, even a large effect may not achieve statistical significance and conversely, even a small effect may be statistically significant if 28 the sample size is sufficiently large (Norman and Streiner, 1986; Hennekens and Buring, 1987; Potter 1994; Brunette 1996). To overcome this problem, a related but more useful measure, the confidence interval estimate, is used. b. Confidence Interval Estimates Confidence interval (CI) estimates provide a range of values within which the true magnitude of the effect lies. This range has a designated likelihood or probability (usually 95%) to include the real but unknown mean value. Technically, a 95% CI means that if the same study were repeated 100 times with subjects from the same source population, 95 of the 100 confidence intervals would contain the true value of whatever was being estimated in the study (Mills 1993). Confidence intervals provide all the information of p values in terms of deciding whether an association is statistically significant. The effect of sample size is reflected in the width of the confidence intervals; the narrower the intervals, the lower the variability in estimating the effect which in turn, reflects a larger sample size. In contrast, wider intervals and greater variability reflect smaller samples. If the null value (eg. 1.0 for relative risk and odds ratios) is included in a 95% confidence interval, then the corresponding p value is greater than 0.05 and the association is not statistically significant. In such instances, if the interval is narrow, there is likely no real effect of the exposure whereas a wide interval suggests that the sample size was inadequate and did not have sufficient statistical power to conclude that chance was not a likely explanation for the findings. Thus, the p value and confidence intervals, together, provide the most information about the role of chance (Hennekens and Buring, 1987; Mills 1993; Potter 1994), but they "still ignore the systematic errors, the biases and confounders, that can overwhelm the statistical variation" (Taubes 1995). In assessing risk factors for oral/pharyngeal cancers, many authors have included p values and confidence interval estimates (eg. Brugere et al., 1986; Boffetta et al., 1992; Franceschi et al., 1992; Oreggia et al., 1991; Barasch et al., 1994; Bundgaard et al., 1994), some included confidence intervals alone (eg. Blot et al., 1988; Winn et al., 1991) and some included neither for 29 the vast majority of calculations presented (eg. Gupta et al., 1989). To illustrate the usefulness of confidence intervals, the following examples of odds ratios for smoking and oral/pharyngeal cancer in a series of 150 oral SCC are given (Barasch et al., 1994), (see also Table 1.10, Section V). Smoking and SCC of the floor of mouth (FOM) had an OR of 38 (95% CI=4.6-316); whereas smoking and SCC of the tongue had an OR of 1.75 (95% CI=0.63-4.9). These authors correctly concluded that their data showed that smoking was more strongly associated with SCC of the FOM than SSC of the tongue. The CI for the OR of FOM did not include the null value and therefore, the association was statistically significant although the wide CI intervals reflect great variability which is characteristic of a small sample size. The CI of the OR for the tongue included the null value (1.0) and the CI interval was very narrow which together, indicate the lack of statistical significance and the likely absence of any real effect of the exposure. However, the authors (Barasch et al., 1994) advised caution in the interpretation of their data as the study was not a case control (no noncancer control group) and had no information about alcohol consumption which is a confounder for tobacco use (see below). 2. Confounding Confounding occurs when there is a mixing of the effects between the exposure, the disease, and a third factor that is associated with the exposure and independently affects the development of the disease (Hennekens and Buring, 1987). Thus, the observed relationship between the exposure and the disease can be attributed wholly or in part to the third factor, the confounder which can cause either an increase or decrease of the true association between the exposure and disease. Confounding factors are hidden variables in the population being studied and they can generate an association that may be real but is not what the investigator thinks it is. The association between the confounder and the disease does not have to be causal; for example, age is always considered to a potential confounder although it can act as a surrogate for other etiologic factors. The confounder must be associated with the disease but independent of its 30 association with the exposure, and it must be associated with disease among nonexposed individuals. For example, alcohol and smoking are both confounders for oral/pharyngeal cancer. Both are independent risk factors for oral/pharyngeal cancer and their effects are difficult to separate because in the general population, people who drink alcohol also tend to smoke, and people who smoke also tend to drink alcohol. In like manner, most oral/pharyngeal cancer patients have smoked and consumed alcohol (Section V.F.). Consumers of tobacco and/or alcohol typically under-report their habits and they may also use mouthwashes to disguise their habits. Consequently, the results of a study evaluating the association between oral/pharyngeal cancer and the use of alcohol-containing mouthwashes (Winn et al., 1991) was confounded due to the under-ascertained exposure to alcohol and tobacco, and resulted in overestimation of the association (Shapiro et al., 1996). A confounder cannot be identified by a statistical test of the association between an exposure and disease. Instead, confounders may be identified after re-analysis or stratification of the data for the suspected confounder reveals a change in the measures of association. Ideally, potential confounders should be considered in the design phase of study but they can be corrected for during analysis (Hennekens and Buring, 1987; Sheps 1995; Taubes 1995) as is typically done in oral/pharyngeal cancer studies (eg. Blot et al. 1988; Winn et al., 1991). 3. Bias Bias is any systematic error that is introduced into the design or conduct of a study. Bias results in incorrect estimates of association between the exposure and disease, and bias is more difficult to evaluate than chance or confounding which can be quantified. Bias can occur during the process of identifying the study population (selection bias) or during the measurement of information about the exposure or the outcome (observation bias). Sackett (1979) has catalogued no less than 56 types of bias and the following is a brief review of common biases that affect epidemiological studies in general and have affected studies of oral/pharyngeal cancer. 31 a. Selection Biases Selection bias is due to differential diagnosis, surveillance or referral in a study which results in an observed relationship between the exposure and disease that is different among those who participated in the study and those who would have been eligible but were not selected to participate (Hennekens and Buring, 1987; Sacket et al., 1991). Selection bias is a great concern in case control and retrospective cohort studies because the disease and exposure have already occurred. Selection bias is less likely to occur in prospective cohort studies because the exposure is ascertained before the outcome. All types of epidemiological studies can incur selection bias if the criteria used to define a disease or an exposure differs between studies, rendering the studies non-comparable. This problem may occur with contemporaneous studies or in studies conducted over different points in time. Differential diagnosis, surveillance or referral may occur because the same condition has received different diagnostic labels (diagnostic vogue bias), prevalent cases may not be clearly distinguished from incident cases (prevalent/incident case bias), the common starting point for a disease or an exposure is not identified (starting time bias), or the referral of cases and controls differs (referral filter bias and centripetal bias) (Sackett 1979; Hennekens and Buring, 1987; Sacket etal., 1991; Taubes 1995). Patients with oral premalignancy or malignancy are typically seen in specialty clinics associated with a dental faculty or a hospital (eg. Silverman and Rozen, 1968). Epidemiological studies which rely on this patient pool suffer from a distorted perception of the disease and its determinants because of centripetal and referral filter biases. Centripetal bias occurs when the reputations of certain clinicians or institutions cause individuals with specific disorders or exposures to gravitate towards them. Referral filter bias occurs when individuals with a specified disease are referred from primary to secondary to tertiary care so that the concentration of rare diseases, rare causes and multiple diagnoses may increase (Sackett 1979). If controls in case-control and cohort studies are also drawn from a clinic or hospital setting, the study is predisposed to Berkson's admission rate bias because the hospitalization or admission rates of exposed and nonexposed cases and controls 32 will differ, and their relative odds of exposure to the putative cause will be distorted (Sackett 1979). Some case-control studies of oral cancer have used only hospital-based controls (eg. Oreggia et al., 1991; Boffetta et al., 1992; Franceschi et al., 1992) whereas others have used community-based controls (eg. Blot et al., 1988; Winn et al., 1991; Kulasegaram et al., 1995). However, even the "random" selection of controls from the community or population at large is prone to selection bias. For example, the selection of telephone numbers by random digits will omit those individuals without a telephone, who for economic and other factors, may differ significantly from individuals who do have a telephone. Moreover, individuals who are at home to answer the telephone may differ from those individuals who are not at home. Several examples of differing criteria for diagnosis of disease or assessment of exposure exist in the oral/pharyngeal cancer literature. For example, in studies of submucous fibrosis (Section VI.B.2.b), not all investigations have included the presence of palpable fibrous bands among the diagnostic criteria, nor have the criteria for including areca nut chewing as an exposure been very clear (reviewed by Murti et al., 1995). In 1978, the World Health Organization (WHO) defined leukoplakia (Section VI.B.a) as a "white patch or plaque that cannot be characterized clinically or pathologically as any other disease". In essence, any white lesion that could not be identified became "leukoplakia" and such a diagnosis was directly dependent upon the clinical experience of the investigator. Moreover, the term leukoplakia has no histological connotation yet in assessing the prevalence of leukoplakia, some investigators relied solely on clinical presentation (eg. reviews by Pindborg 1980, 1994; Banoczy et al., 1993) whereas others included histopathology for selected cases (eg. Banoczy 1977; Silverman et al., 1968; Gupta et al., 1989; Brown et al., 1993) or routinely for all cases (eg. Silverman et al., 1984; Murti et al., 1986), presumably in attempts to rule out other disease entities in order to leave a "pure" sample of "leukoplakias". Moreover, in instances where microscopic examinations provided a diagnosis of fungal infection which is a possible etiologic agent (Section 33 V.F.5), the clinical diagnosis of leukoplakia or erythroleukoplakia was typically retained, in some instances even after successful antifungal therapy (Section V.F.5). However, in many cases of leukoplakia, fungal infection was not assessed nor considered as an etiology in order to rule out a diagnosis of "idiopathic leukoplakia". Therefore, it is not surprising that the use of biased samples, inconsistent criteria, inconsistent diagnostic methods and the questionable reliability among investigators (see below and Section HG) have resulted in a wide range of prevalence data for leukoplakia and its transformation to malignancy (Section VI.B.l). Reports of the rate of malignant transformation (Section VI.B.l.c) among patients with oral leukoplakia (including fungal infection) vary from less than 1% over 2 years (Pindborg 1980) to 79% over 8 years (Gupta et al., 1989) and as high as 100% (Cawson 1975). These disparate reports may reflect differences in ethnic origin or risk factors among the different groups of patients investigated. However, the differences could also be due to the inclusion in some studies, of a broader range of benign white lesions of some other type among those classified as leukoplakias, thus giving a lower incidence of malignant transformation than for other groups (Dabelsteen and Mackenzie, 1987); conversely, some studies may have included lesions with a high malignant potential (eg. Gupta et al., 1989), thus raising the incidence of transformation. It is difficult to reconcile disparate transformation rates over time from the same investigators in the same clinic environment. Silverman and Rozen (1968) reported an incidence of malignant transformation of 6% over 1-11 years of observation, and 16 years later reported an incidence of 18% over a mean observation period of 8.1 years (Silverman et al., 1984). Do these data reflect an altered disease pattern, or the increased clinical experience of the investigators that resulted in the exclusion of benign lesions from the latter group but that would have been included in the first (Dabelsteen and Mackenzie 1980)? Similar conflicts may be involved in the disparate reports of malignant transformation of lichen planus (Section VI.B.2.a). Some studies (eg. Silverman et al., 1985; Murti et al., 1986; Holmstrup et al., 1988) included patients with documented oral lichen 34 planus; another (Sigurgeirsson and Lindelof, 1991) included patients with cutaneous lichen planus but without any information about oral involvement, yet the authors estimated the risk of oral transformation based on the incidence reported by other investigators (Holmstrup 1992). Meanwhile, Eisenberg and Krutchkoff (1992) argued that true lichen planus, based on precise histological criteria, was less prevalent than commonly accepted and had no inherent malignant predisposition. The confusion may result from a wide variety of clinical lesions that are diagnosed by examiners with varying clinical experience and are included under a single label (eg. leukoplakia or lichen planus), (Ephros and Samit, 1997). Fortunately, the problems associated with previous definitions and classifications of oral lesions have been recognized and addressed by recent attempts to provide a more uniform clinical staging procedure (Axell et al., 1996). However, other discrepancies persist. For example, in assessing the recurrence of oral dysplasia or malignancy three questions arise: (1) how can the success of treatment of either oral dysplasia or malignancy be compared in different studies (2) is it important to distinguish recurrence of disease (oral dysplasia or malignancy) from second primary events and if so (3) how can they be distinguished? Mincer et al. (1972) evaluated lesions according to the concept of field cancerization (Section V.A.2) and considered the patient's entire oral epithelium as a single site; all lesions that arose subsequent to the initial lesion were considered recurrences or transformations of the first event. Einhorn and Wesall (1967) described tumours as clinically or histologically discrete neoplasms in specific oral sites, and lesions that arose in sites that differed from the first tumour were considered new tumours. Banoczy (1977) failed to define their criteria altogether yet discussed the recurrence and transformation of oral lesions without respect to specific sites. In contrast, Day et al. (1994a, b) defined second cancers using specific criteria that included location, time of detection and histological evaluation in relation to the first lesion. b. Observation Bias Observation biases occur when methods of measuring exposures and outcomes, in analyzing or 35 interpreting the data are consistently dissimilar between the groups under study. Recall bias is a major concern in case control and retrospective cohort studies where individuals are asked to accurately recall exposure levels to given factors. Individuals with the disease may recall and report past events differently than individuals without the disease, and there may be a tendency to under-report certain behaviours which are considered socially questionable such as excessive alcohol consumption and smoking. Knowledge by the investigator of a subjects's disease (exposure suspicion bias) or exposure (diagnostic suspicion bias) status may influence the intensity and outcome of a search for the exposure or disease, respectively. Subjects may alter their behaviours when they know they are being observed (attention bias), or they may discontinue the study (or die) so that losses to follow-up between exposed and nonexposed groups differ (withdrawal bias), (Sackett 1979; Sacket et al., 1991; Taubes 1995). Bias can also be introduced during analysis of the data by selecting levels of significance after the statistical tests have been completed (post-hoc significance bias), by deleting outlying data (tidying-up bias), by repeatedly evaluating the accumulating data before completion of the study (repeated peeks bias), degrading or collapsing measurement scales to obscure differences between groups under comparison (scale degradation bias) or to alter interpretation of the data (magnitude bias), confusing statistical significance for biological and clinical significance (significance bias) or equating correlation with causation (correlation bias), (Sackett 1979; Sacket et al., 1991). A common bias occurs when all possible associations between variables are examined by many independent statistical tests (data dredging bias). In the search for chance associations, the 95% probability of a correct conclusion decreases drastically. Mills (1993) calculated that "by performing just 2 independent tests, the probability that the "significant" differences found by the investigators will reflect true differences is reduced to 90% (0.95 x 0.95). If 20 tests are performed, the probability is only 36% (0.9520), and 1 of every 20 independent comparisons will 36 yield a "significant" result" (Mills 1993). Results from such "fishing trips" (eg. Gorsky et al., 1994; Rubright et al., 1996) cannot be used for testing hypothesis but nevertheless, they are useful for generating hypotheses as long as the studies are identified as such and the authors do not disguise the results as conclusions of a priori hypotheses testing (Sackett 1979; Sacket et al., 1991; Mills 1993). In fact, chapters 2 and 3 of this thesis utilize multiple independent tests in a search for associations that may generate hypotheses. G . Measurement Reliability Measurement reliability refers to the ability to obtain the same measure consistently over sequential measures, and there are several sources of variability that can affect the reliability of a measurement. The first source relates to the normal biological variation inherent in the phenomenon that is being measured, such as hormone levels which may vary with the diurnal or menstrual cycle, or blood pressure which varies throughout the day and under different circumstances. Another source originates from the reliability of the measuring instrument itself. In laboratory-based studies, measurements typically involve the use of instruments such as rulers or weigh scales that are calibrated against recognized, established absolute standards, and observations are performed under well-specified conditions. Usually the results of these measurements are expressed as a standard deviation of the individual values or as confidence intervals around the calculated mean (Brunette 1996). In epidemiologic studies, investigators may need to make judgements using criteria that are not very specific, or about subject characteristics that are difficult to evaluate. In epidemiologic studies there are no absolute standards and the best that can be done is to determine if the investigators are consistent in their judgements. Comparisons can be made in which the same investigator examines the same subjects two or more times (intra-examiner reliability), or in which different investigators examine the same subjects (inter-examiner reliability). Interobserver variability is minimized when endpoints are well-defined and quantifiable as in measuring height, and is greater when criteria are 37 vague and subjective, as in diagnosis of leukoplakia or dysplasia. Agreement between observers also tends to improve when result categories are few and straightforward (i.e. dichotomous categories or outcomes), (Hennekens and Buring, 1987). Inter and intra examiner reliability can quantitated and one approach is to calculate Pearson correlation coefficients (r). 1. Pearson Correlation Coefficients Correlation coefficients are a measure of the association or agreement between two sets of data in which perfect agreement or the strongest positive correlation has a value of 1.0, no agreement or relationship is indicated by "0", and perfect disagreement or the strongest negative correlation has a value of -1.0. Correlation coefficients typically overestimate the true reliability and they do not explain how much of the correlation can be explained by variability of the data (Sackett et al., 1991; Brunette 1996). For example, a patient would reasonably expect that the histological diagnosis of oral dysplasia (Section VIA) or malignancy (Section VILA) was reliable (reproducible) and that it represented the truth (was valid). However, in comparing the agreement of six pathologists between their original sign-out diagnosis of oral epithelial dysplasia, and their diagnosis of the same slides made several months, Abbey et al. (1995) reported that the exact agreement ranged from 0.30 to 0.63 with an average of only 0.50. Another shortcoming of the Pearson correlation coefficient is that is cannot detect situations in which one set of data is systematically different than the others. Using the example of the 6 pathologists above, assume that pathologist A consistently assigned one lesser degree of dysplasia or severity relative to his original sign-out diagnosis, and pathologist B always agreed with his earlier sign-out diagnosis; the Pearson correlation coefficients measuring their respective 38 agreements between the sign-out and subsequent diagnoses would be the same. This shortcoming can be avoided by using the "intraclass correlation coefficient" which penalizes systematic errors and therefore, would assign a lower score to pathologist A. In addition, intraclass correlation coefficients may be interpreted just like kappa scores which are more commonly used (Sackett et al., 1991). 2 . Kappa A better approach in evaluating reliability is the kappa (K) statistic because it adjusts for the degree of agreement expected purely by chance. Observed Agreement - Chance Agreement Actual Agreement beyond Chance Kappa = 1 - Chance Agreement = Potential Agreement beyond Chance For a perfect association, K = 1.0 and for no association, K = 0. Qualitative terms in relation to kappa values vary (eg. Sackett et al., 1991; Karabulut et al., 1995; Brunette 1996), but Brunette (1996) suggests that K values below 0.4 indicates "poor" agreement, 0.4 - 0.75 is "fair" and 0.75 -1.0 is "excellent". Epidemiological investigations should include measures of reliability that assess the consistency of evaluations made by the examiners. Ideally, the study should not proceed before the investigators have been trained and calibrated with demonstrated high kappa scores (eg.K >0.6) which are especially crucial when definitive tests or gold standards for assessment are not available (Schechter 1995). For example, by definition a diagnosis of leukoplakia is based upon clinical rather than histological evaluation and although clinical criteria may be defined (eg. Axell et al., 1996), diagnosis is essentially a subjective judgement dependent upon the clinical experience of the investigator (Dabelsteen and Mackenzie, 1987). It is surprising, therefore, that to date no clinical investigations of oral/pharyngeal premalignancy or malignancy have investigated the reliability of 39 the investigators' clinical diagnoses by reporting kappa scores. Perhaps in attempts to avoid the unreliability of clinical diagnoses of leukoplakia, some investigators (eg. Silverman et al., 1984) have relied on histopathology for diagnosis of the presence or absence of dysplasia to classify leukoplakia. The classification of disease is traditionally based on pathological anatomy (Wulff 1976) and therefore, the histopathologist's diagnosis is typically regarded as the "gold standard" which is the acknowledged standard for definitive diagnosis (Sackett et al. 1991). However, when pathologists reach a diagnosis, they may be influenced by factors other than the histomorphology of the tissue on the slide. The pathologist's knowledge of the patient's clinical presentation may be considered and incorrectly weighted in reaching a diagnosis, so that the clinical data are "double counted" (Schwartz et al., 1981). For example, a pathologist's knowledge that a biopsy specimen was taken from an area of asymptomatic erythroleukoplakia on the FOM of a heavy smoker and alcohol drinker would raise the suspicion of malignancy (eg. Mashberg 1980; Ephros and Samit, 1997) even before the slide was placed on the microscope stage. In such instances, the dysplasia or carcinoma may be unconsciously graded as more severe than if the clinical information was not available to the pathologist (Schwartz et al., 1981). The histomorphologic criteria that characterize epithelial dysplasia (Section VIA) or malignancy (Section VILA) are well known to pathologists. However, the subjective interpretation of the histological features remains a problem (eg. Pindborg et al., 1985) and the "poor" to "fair" kappa scores of observer variability in the histologic assessment of oral dysplasia may be disconcerting to patients whose biopsy specimens are being examined. In the previous example of the 6 pathologists whose agreement between their original sign-out diagnosis of dysplasia and subsequent re-examinations of the same slides were compared, intraexaminer kappa scores ranged from 0.05 to 0.49 and compared to correlations of 0.30 to 0.63, respectively (Abbey et al, 1995). In the same study, interexaminer kappa scores for the presence or absence of dysplasia ranged 40 from 0.29 to 0.48 (Abbey et al., 1995). Karabulut et al. (1995) investigated the interobserver variability in grading sections of oral leukoplakia from no dysplasia to carcinoma in situ, and kappa values ranged from 0.27 to 0.45. In reaching a diagnosis, a pathologist recognizes tissue changes and generally assigns more weight to some histologic features than others (Sections VI.A and VILA). Attempts have been made to develop grading systems which focus on particular morphologic features of dysplasia or malignancy, and interobserver kappas for different histologic features ranged from 0.30 to 0.42 with the highest kappa measured for the epithelial pattern of invasion (Bryne et al., 1991a; 1991b). When stricter definitions of grading criteria are used, such as restricting assessment and diagnosis to only the most invasive margins of a tumour, an interobserver kappa of 0.63 was reported by Bryne et al. (1992). Overall, clinicians must appreciate that a pathologist's report rarely reflects only the morphologic findings (Schwartz et al., 1981), and that there are some cases that represent the pathologist's "best guess" so that, in the end, clinical judgement remains the guiding factor in management of the patient (Kaugars 1997). Research continues into the use of biomarkers (Sections II.H. and V.C) which may provide more objective criteria than traditional histomorphology for the diagnosis of dysplasia or malignancy but to date, a valid and reliable diagnostic test has not been identified. H. Measurement Validity Measurement validity refers to the truthfulness of the measurement, or in other words, "does it really measure what it claims to measure?" (Brunette 1996). Determining the validity of a measurement requires a comparison to a reference measure or "gold standard" that has been accepted as true. Sensitivity and specificity are two measures of the validity of a measurement; sensitivity is the true-positive rate and specificity is the true-negative rate. 41 1. Sensitivity and Specificity of Measurement Techniques Research into the prevention of a disease such as cancer requires well-defined endpoints so that the efficacy and success of a trial can be evaluated. If clinical disease (ie. the incidence of cancer) was used as an endpoint in cancer prevention studies, a large number of subjects and long durations of follow-up would be required (Pillai et al., 1992). However, if biological markers could be used as intermediate endpoints, they could act as surrogates for the disease, reveal responses in a shorter time and require fewer trial subjects to achieve statistical power. That is, biomarkers may represent changes in the continuum of events between the initiation of carcinogenesis and the final expression of clinically-evident disease (Pillai et al., 1992; Greenwald et al., 1995). Biomarkers may be used as a generic term for the target of assays that monitor critical aspects of tumour development, providing a "window into the biology of the epithelium" (Mulshine et al., 1993), (see Sections IV. and V). When biomarkers are used as potential measures of disease, the techniques for determining the presence of the biomarker must be valid and reliable. That is, the techniques must be sensitive and determine the true level or activity of the biomarker in the tissue or cells, and they must be specific and correctly identify the absence of the biomarker or biomarker activity. For example, the sensitivity of immuno-histochemistry (IHC) as a technique to detect an antigen of interest depends upon the method used to preserve the tissue because different methods may destroy and/or mask the epitopes or antigenic determinants of that antigen, or sterically hinder access of the antibodies. The type and quality of antibodies used will determine how sensitive and specific the antibody-antigen reaction (signal) is, in relation to the background staining (noise), (eg Bacallao et al., 1990; Denk 1987). For example, IHC has been used to identify p53 protein and staining is often considered a surrogate marker for gene p53 mutation (Section V) even in the absence of confirmatory DNA studies. Moreover, the absence of reactivity with p53 antibodies can not exclude genetic alterations since the antibodies may fail to detect truncated p53 protein resulting from frameshift or nonsense mutations (Gopalakrishnan et al., 1997; Scully and Field, 1997). In a 42 review of 84 studies that utilized both IHC and DNA sequencing, Greenblatt et al. (1994) reported the sensitivity of IHC for detecting p53 protein as only 75% (range 36-100%) and the positive predictive value (Table 1.2) as only 63% (range 8-100%), with considerable variation among tumour types. Consequently, the status of p53 protein as determined by IHC could not be equated with either the wild-type or mutant genotype (Greenblatt et al., 1994). Polymerase-chain techniques (PCR) are also subject to false-positive and false-negative results. PCR techniques are subject to error rates from 1/104 to more than 1/500 base pairs, depending upon the reaction conditions, and contamination of the PCR reaction may lead to false-positive or false-negative results (Greenblatt et al., 1994). In DNA studies of the p53 gene, it was established that the gene consists of 11 exons (coding sequences) and exons 2-11 code for protein p53. Most of the mutations in p53 gene occur in exons 5-8 and are single missense base substitutions (substitution of a single nucleotide pair resulting in the substitution of a single amino acid in the original gene product), although allelic loss, insertions, and deletions also occur (Greenblatt et al., 1994; Gopalakrishnan et al., 1997). Unfortunately, the predominance of mutations in exons 5-8 has caused a bias in searching for mutations so that most investigators confined their analysis to exons 5-8, effectively ignoring the remaining exons and consequently underestimating the prevalence of p53 mutations by over 20% in some tumours (Greenblatt et al., 1994). The recovery of human papillomavirus (HPV) DNA (Section V.F.6) from oral premalignant and malignant lesions also varies with the sensitivity and specificity of the technique. Low-sensitivity techniques such as immunoperoxidase IHC or in situ hybridization can detect over 10 copies of viral DNA per cell, and moderately-sensitive techniques such as the southern blot can detect 1-10 copies of viral DNA per cell (Miller and White, 1996). When low- or moderate-sensitivity techniques are used, the prevalence of HPV in normal oral mucosa is about 7%, and the prevalence in oral SCC is 17% and 25%, respectively. However, with the use of the highly-sensitive PC reactions which can detect less than 1 copy of viral DNA per cell, the prevalence of HPV in normal 43 mucosa increases to 25% and the prevalence in oral SCC increases to 37% (Miller and White, 1996). Significantly, even the use of a highly-sensitive method such as PCR is not sufficient to yield valid or reliable results if techniques are not standardized. For example, HPV DNA is detected significantly more often in frozen SCC (52%) samples than in paraffin-embedded (22%) tissue. Furthermore, if late region primers are used, PCR will detect only the late region genes which encode for the capsid proteins; significantly, only the early region HPV genes are associated with malignant transformation. Thus, HPV prevalence data should be based on the identification of early region DNA with documentation of viral DNA integration into host cell DNA (Miller and White, 1996; Eversole 1997). 2. Diagnostic Tests The potential use of biomarkers as measures of oral premalignancy or malignancy has been plagued by the inconsistency of techniques used in different investigations (eg. Greemblatt et al., 1994; see also Section V.C). In addition, many biomarker investigations have not set out to test a stated hypothesis; rather, investigators have scanned a tissue for the presence or activity of a range of suspected biomarkers which may not even be established as significant in the neoplastic process and often, the investigations are driven by the availability of a convenient technique. A major problem in interpreting the data from such studies arises from the paucity of information about the roles of the markers in normal human oral mucosa and of their variations between lining, masticatory and specialized mucosa in different oral sites. Moreover, the role of many biomarkers under inflamed, traumatized or healing conditions is not known, and unless the biomarker can be demonstrated to be absent in situations involving increased cell proliferation and cell movement such as in benign hyperplasia and wound healing, it can not be assumed that the biomarker alterations are specific to neoplastic transformation (reviewed by Johnson et al., 1980). Consequently, the use of biomarkers is hampered by the lack of a marker that is present in all malignant cells but is absent from normal mucosa, although the combined use of two or more markers may be more accurate indicators of precancer or cancer than single markers (Mulshine et al, 1993; Ogden 1997). In general, biomarkers are tumour-associated rather than tumour-specific, and the only way a biomarker can be evaluated is by means of follow-up studies, yet it is not possible to follow a lesion histologically without altering the lesion (Dabelsteen 1980). Several biomarkers (Section V.C) show potential as diagnostic aids but none are currently used in the routine diagnosis or screening of oral premalignancy or malignancy (eg. Lippman et al., 1990; Pillai et al., 1992; Mulshine et al., 1993). Moreover, a marker that performs well as an indicator of disease progression is not necessarily useful for the early detection of the same disease. That is, a Bayesian approach (eg. Brunette 1996) is relevant to the use of an early-detection tool in monitoring high-risk individuals for detection of second cancers, but the same tool may not be appropriate for screening the population for first cancers of the same type (see below; Schechter and Sheps, 1985; Mulshine et al., 1993; Epstein et al., 1997). The changes in activity or level of any physiologic, biochemical or molecular marker are typically reflected by continuous measures yet the presence or absence of an abnormality or disease is typically a dichotomous diagnosis (i.e. normal versus abnormal, or health versus disease), although gradations of abnormalities are also used (eg. mild, moderate, severe). Assuming that a biomarker has been associated with an abnormality or disease, different cut-off points in the levels of the biomarker will determine the fraction of true-positive, false positive, true-negative and false-negative results which in turn, produce different estimates of the sensitivity and specificity of the biomarker as a diagnostic test (Table 1.2). When used in relationship to a diagnostic test, sensitivity refers to the probability of a positive test in a patient with the disease. Specificity is the probability of a negative test in a patient without the disease, and significantly, sensitivity and specificity are calculated using the subset of patients from the trial population who have and do not have the disease, respectively. Sensitivity and specificity both range from 0 to 1 and they are converted to a percentage by multiplying by 100 (Sacket et al., 1991; Greenstein and Lamster, 1995; Brunette 1996). 45 The cut-off point selected to identify disease will dramatically affect the sensitivity and specificity. If a low threshold for disease activity is selected, the sensitivity is usually increased and the specificity is usually decreased. One of the best methods to evaluate the effect of different cut-offs is to use receiver operating characteristic (ROC) analysis. ROC analysis plots the true-positive fraction (sensitivity) as a function of the false-positive fraction (1.0 - specificity), and points along the curve represent different thresholds for the test. Thereby, selection of points towards the left of the curve yield higher specificity and points to the right yield higher sensitivity. ROC analysis also permits the comparison of different tests without any selection of upper or lower reference limits, or any particular sensitivity or specificity. Moreover, ROC curves are independent of the disease prevalence and therefore reflect the true performance of the diagnostic tests (Sackett et al., 1991; Greenstein and Lamster, 1995; Brunette 1996). In clinical practice, the selection of cut-off points is determined by several factors including mortality and morbidity of the disease, the consequences of over or under treatment, and the cost and time required to perform the diagnostic test. If it is important that all individuals with the disease or its progression are detected, then a low threshold cut-off is selected to provide high sensitivity and high positive predictive values (PTL+, Table 1.2), although they will be associated with an increased number of false-positive results (low specificity and low negative predictive values, Table 1.2). Such an approach would be useful when screening for a serious or life-threatening disease because disease status can be attained by use of confirmation tests (series approach, see below). Confirmation testing is required to ensure that risky or expensive therapy is not mistakenly undertaken and it requires high thresholds to limit the number of false-positives. Therefore, confirmation tests have high specificity and high negative predictive values but lower sensitivity (Greenstein and Lamster, 1995). Biomarkers (Section V.C) have not been developed for use as routine screening or diagnostic tests of oral/pharyngeal premalignancy or malignancy (eg. Lippman et al., 1990; Pillai et al., 1992; 46 Mulshine et al., 1993). However, Ogden et al. (1994) described the use of DNA ploidy (Section V.C.8 ) and keratin analysis (Section Ul. Table 1.3; Section V.C.7, Table 1.9) of oral exfoliative cytology in the detection of oral cancer. Smears were obtained from 33 biopsy-proven oral cancers and from contralateral normal sites of patients in a Dundee Dental Hospital. DNA profiles had 70% sensitivity, 90% specificity and 90% positive predictive value. Keratin 19 had 90% sensitivity, 50% specificity and 95% positive predictive value; Keratin 8 had 60% sensitivity, 100% specificity and positive predictive value. The combination (in parallel, see below) of an abnormal DNA profile and K19 expression had a positive predictive value of 95% and using these two parameters together, resulted in correct identification of a malignant tumour in 29 of 33 cases (Ogden et al., 1994). Yet, despite the high values of these test characteristics, the use of these methodologies as a screening test of the general population for oral SCC cannot be recommended (see below). Computer simulated neural networks have also been trained to categorise normal, premalignant and malignant oral smears that were previously classified from a pathologist's report (Brickley et al., 1996). The neural network was able to distinguish between smears obtained from normal mucosa or non-dysplastic lesions, and those collected from dysplastic or malignant lesions with a sensitivity of 76% and specificity of 82%. However, when asked to differentiate non-malignant from potentially malignant lesions the neural network misclassified half of the cases as false positives suggesting (as expected) that a greater difference exists between normal mucosa and mucosa affected by a lesion, than between a benign mucosal lesion and a premalignant one. Nevertheless, the consequences of a false positive diagnosis are merely that the site would be biopsied. The consequences of a false negative diagnosis could be more serious and therefore the authors considered it reassuring that the networks were more sensitive than specific (Brickley et al., 1996). Theoretically, sensitivity and specificity are considered to be stable properties of a test because they 47 are apparently not affected by the prevalence of the target disease; however, there is some evidence that sensitivity and specificity do change from one clinical population to another (Hlatky et al., 1984; Hlatky et al., 1987), especially if the stage of the disease varies in different groups of patients (Greenstein and Lamster, 1995). As noted above, sensitivity and specificity are calculated in defined populations in which the disease status of the individuals is known and where only extremes of disease (very sick) and health (very healthy) are represented. However, these circumstances do not represent the true clinical situation in which the diagnostic test is used to determine the disease status in a population comprised of healthy, diseased and equivocal cases. The predictive values (Table 1.2) of a test provide information about how often a test will provide a correct diagnosis in a mixed population, but the predictive values will vary widely as the prevalence of the disease changes (Hennekens and Buring 1987; Sackett et al., 1991; Greenstein and Lamster, 1995). The rationale for using any diagnostic test is based on the probability that the disease is present prior to the test (the pre-test probability). When considering the use of a test for the general population such as in screening for a disease (see Section n.J below), the prevalence of the disease is used as the pre-test probability. When a test is considered for a specific patient, the clinician estimates the probability about how likely it is that the patient has the disease of interest, or the clinician can assign the disease prevalence as the pre-test probability. The choice of diagnostic test for any particular disease is determined by the power or ability of the test to revise the pre-test probabilities, either upwards to rule-in the disease, or downwards to rule-out the disease. The cut-off probabilities for ruling-in or ruling-out a disease will depend upon the disease and the subsequent courses of action or follow-up that relate to either ruling-in or ruling-out the disease. That is, the consequences of false-positive and false-negative results must be weighed in each case. In addition, if a test is not powerful enough to alter the pre-test probabilities so that a positive or negative test result would alter the pre-test planned course of action, then the test should not be performed (Schechter and Sheps, 1985; Sackett et al., 1991). 48 For example, the use of toluidine blue (Section VIJ.B) has been advocated for the detection of oral SCC. The sensitivity of toluidine blue ranges from 93.5% to 97.8% and its specificity ranges from 73.3% to 92.9% (Rosenberg and Cretin, 1989). However, toluidine blue will have different predictive values if it used as a screening test (Section JJJ below) in the general population, or in a tertiary referral centre for oral cancer. The prevalence of SCC in the general population is only 3% (Parker et al., 1996), and therefore, the post test likelihood of a positive toluidine blue test is only 6% (see Schechter and Sheps, 1985 or Epstein et al., 1997 for calculation details). In contrast, the prevalence of SCC either as primary or recurrent disease, is greater (26%, Silverman, 1990; Parker et al., 1996; 33%, Epstein et al., 1997) in a tertiary care centre for oral SCC and consequently, the post test likelihood of a positive test is also greater (51%, Epstein et al., 1997). In a similar manner, if DNA profiles and keratin K19 analyses of oral smears (Ogden et al., 1994) were used as tests for the general population (assuming prevalence of oral SCC is 3%), the post test likelihoods of a positive test would be 18% and 5.3%, respectively. In contrast, if these same measures are used in an oral SCC tertiary care setting where prevalence is assumed to be 33% (Epstein et al., 1997), the post test likelihoods of a positive test would be 76% and 47%, respectively (see Schechter and Sheps, 1985 or Epstein et al., 1997 for calculation details). In the high-prevalence setting, the post-test likelihoods of the tests are considerably higher than the pre-test probabilities, meaning that there is a considerably increased probability that the disease (SCC) is actually present. In contrast, the post-test likelihoods of the same tests in the general population (low prevalence setting) are similar to the pretest probabilities, meaning that there is only a slight increase in the probability that the disease is actually present. Nevertheless, the significance of .each positive or negative test must be evaluated on an individual basis by the clinician who must then decide what the subsequent course of action will be (see also Section J below). Different tests for the same disease can also be used in combination with another, either in series such as confirmation testing, or in parallel, such as in the combined use of DNA profiles and keratins (Odgen et al., 1994). If Tests A and B are used in parallel, then all sites positive for Test 1 49 A or Test B are considered positive, and a negative result requires that both Test A and Test B are negative. If tests are used in series, either test A or B can be used first, but a positive result on the first test requires retesting with the other test (Greenstein and Lamster, 1995). In parallel testing, the combined sensitivities are greater than sensitivities of the individual tests but overall specificity is reduced and there is an increased percentage of false-positive results. Thus, parallel testing is more sensitive for detecting disease than series testing, but parallel testing is less efficient at confirming the presence of disease. In contrast, series testing is less sensitive in detecting disease, but series testing has greater specificity and is more efficient at specifically confirming the presence of disease (Sackett et al., 1991; Greenstein and Lamster, 1995). It is likely that diagnostic tests using biomarkers for premalignancy and malignancy will be developed in the future but their contention that they discriminate between health and disease must be carefully evaluated before histopathological diagnosis is abandoned as the gold standard. Nevertheless, biomarkers may serve as adjuncts to clinical diagnosis and they are potentially useful prognostic indicators in that they may provide information about the patient's clinical course. For example, the histological pattern of tumour invasion (Odell et al., 1994) and mutations in tumour suppressor genes and/or proto-oncogenes have been linked to metastasis and recurrence of oral SCC (see Section V.C). Biomarkers may also provide opportunities for the early detection of disease, before the disease becomes symptomatic or even clinically evident. J. Early Diagnosis and Survival of Disease For some diseases, it may be possible to improve the outcome of illness among affected individuals by reducing the severity of the clinical course of the disease or the rate of recurrence. One approach to achieving these objectives is to detect the disease at earlier stages, before the disease becomes symptomatic, and this may be accomplished by screening or case-finding strategies in which a test is applied to persons who are asymptomatic for the disease of interest. The test results are used to classify individuals with respect to their probability of having a 50 particular disease, and those with a positive test are further evaluated by subsequent tests to determine whether they do have the disease (Hennekens and Buring 1987). Screening may be voluntary as at health fairs, or it may be compulsory as when pilots submit to periodic electrocardiograms, immigrants to tuberculosis tests, and applicants for life and disability insurance to a combination of disparate tests (Hennekens and Buring 1987). Case finding refers to clinicians who can seek early diagnosis when patients come to them for annual health reviews or unrelated, intercurrent illnesses. For example, most individuals see a dentist or physician once in a while and the clinician can use the opportunity to scan the oral cavity and oropharynx for abnormalities, or measure blood pressures of every adult patient who attends the practice for any reason. In addition, targets for early diagnosis should include risk markers and factors for disease such as family history, alcohol and tobacco use, etc. (Hennekens and Buring 1987). The assumption underlying screening programs is that early detection prior to the development of symptoms, will lead to more favourable prognosis because treatment begun before the disease becomes clinically manifest, will be more effective that later treatment. The notion of early diagnosis is predicated on an ordered biological progression or 'natural history' of the disease which includes four fundamental time points: biologic onset, clinical onset, the point of usual clinical diagnosis and finally, the outcome which may be recovery, disability or death (Figure 1.3), and the amount of time between each time point will vary with each disease. In addition, there are two important critical points or thresholds: the detection threshold and the therapeutic threshold (Figure 1.3). The location of the detection and therapeutic thresholds in the natural history will vary with each disease and their locations are crucial to the value of early diagnosis. For a disease to be amenable to screening, the disease must have a detectable pre-clinical phase, (i.e. the detection threshold must occur prior to the clinical onset), the preclinical phase must be amenable to treatment, and the treatment must favourably alter the natural history of the disease so that survival, function, quality of life, or all three are improved (Sackett et al., 1991; Schechter 1995). 51 Patients whose cancers are diagnosed early, have better 5-year survivals than patients who are diagnosed in later, symptomatic stages. However, these observations do not prove the value of early diagnosis because even when cancer therapy is worthless, early diagnosis will always appear to improve survival. This paradox is the result of several biases and a result of how survival is measured. First, individuals who volunteer for screening programs and patients who present for periodic health examinations are generally healthier and they may exhibit exposures or outcomes which differ from those of non-volunteers or "late-comer" patients (volunteer bias, Sackett 1979). Survival analyses are used to determine the amount of time to an event or health outcome such as death, recurrence of disease, or the occurrence of a sign or symptom. After the clinical diagnosis of cancer, patients are followed over time (eg. 5 years) and each year a certain percentage of patients will die. For example, if 10% of the patients died every year, the 5-year survival would be 50%; if all the patients in this example were diagnosed at age 45, then 50% of them would be alive at age 50. If a screening test was able to detect the cancer a year before patients developed symptoms (i.e. at age 44), and if early treatment was no more effective than at the time of usual clinical diagnosis (age 45), then the survival curve is shifted towards the right because the starting point for the 5-year survival measurement has been shifted one year backwards. As before, only half of the patients will be alive at age 50 but instead of being given an extra year forward of life, they received an extra year backwards of disease. This "zero-time shift" is known as "lead time bias" (Sackett et al., 1991). Early diagnosis may also result in apparent survival because of "length-time bias" in which slow-growing tumours are preferentially detected. Slow-growing tumours are detectable over a longer time period than fast-growing tumours and consequently, slow-growing tumours are preferentially identified by early-detection programs. Even when therapy is worthless, patients detected through early diagnosis will have a longer survival than those detected at the time of usual clinical diagnosis. In general, patients with long preclinical periods tend to have long clinical durations of 52 disease, and patients with short preclinical durations tend to have short and rapidly fatal clinical periods (Sackett et al., 1991). Kowalski et al. (1994) recommended that screening or case-finding strategies be performed by professionals proficient in the diagnosis of oral premalignant and early-stage malignancies. These investigators evaluated the risks of presenting with advanced stage versus early stage (Section VH) oral/pharyngeal SCCs in 336 Brazilian patients, and their study was limited to lesions that could be accessible to self examination by the patients. The main reasons for diagnostic delays were attributed to the patient's ignorance about the disease although income level and educational levels were not associated with stage distribution! Female and older patients sought help more readily than males or younger patients, yet when patients sought medical or dental care for early symptoms, the lesions were frequently misdiagnosed as benign conditions. The majority (58%) of cases were symptomatic for over a month before help was sought and the most common first symptoms were a painful ulcer (63%) in the oropharynx, followed by odynophagia and/or dysphagia (21%). Upon seeking help, there was no delay of referral to a head and neck service for 18% of patients, but almost 12% of cases were delayed in diagnosis and treatment because medical doctors and dentists failed to recognize early lesions. Medical doctors delayed referral for a median of 12 months in 6% cases, and dentists for a median of 6.5 months in 3% of cases; the balance were attributed to delays caused by pharmacists and drug store clerks. Patient and professional delays were not related to the stage of disease but the consequences of advanced stage at diagnosis were considerable increases in treatment costs and longer hospital stays. Kowalski et al. (1994) concluded that early tumours were often asymptomatic and therefore could be detected only during a routine examination of high risk individuals, but the failure of doctors and dentists in recognising early lesions was a major concern. Cancer survival statistics may be reported as the observed survival rate for a specific patient group. Among any group of patients, some will be lost to follow-up so that there is no information 53 available concerning their survival. Some patients may die from causes other than the disease of interest, and some may develop diseases other than the disease of interest. As survival analysis calculates the amount of time to an event, that event must be clearly identified. For example, some investigators of oral/pharyngeal cancer have not clearly specified the methods and criteria used for calculating survival (eg. Spiro and Strong, 1974) so that it is unclear if survival refers to survival from death due to any cause (all-cause survival), survival until death from oral/pharyngeal cancer, its treatment or metastasis (cause-specific survival), or survival until local or regional disease recurrence (disease-free survival). In contrast, some investigators (eg. Callery et al., 1984; Franceschi et al., 1993; Zelefsky et al., 1992; Kraus et al., 1993) clearly identified survival criteria as survival to regional and neck failure as well as overall cause-specific survival. Throughout this thesis, survival data are reported using the authors' terms but in general, it was often unclear whether survival referred to overall survival from death due to the oral/pharyngeal cancer. Cancer survival statistics may also be reported as a five-year relative survival rate which is the ratio of the observed survival rate for the specific patient group to the expected survival rate for persons in the general population who are similar with respect to age, gender, race and calendar year of observation (Wingo et al., 1995). This method has been used to compare differences in survival among different races in the United States (Wingo et al., 1995). K . Summary This section has reviewed some basic concepts in epidemiology that should aid the readers of this thesis in their understanding of oral premalignancy (Section VI) and oral SCC (Section VII), and of the results presented in Chapter 3. 54 DESCRIPTIVE ANALYTIC study of disease distribution (who, where, when) » Hypothesis Generating * study of disease determinants correlational study case reports/series cross-sectional survey OBSERVATIONAL Hypothesis Generating I case control cohort-prospective cohort-retrospective EXPERIMENTAL (intervention study) Hypothesis Testing I clinical trial field trial community trial Figure 1.1. Taxonomy of Epidemiologic Studies. Epidemiological studies can be broadly divided into descriptive studies which focus on the distribution of disease, and analytic studies which focus on the determinants of disease. Descriptive studies are always observational studies and they are used to generate a posteriori hypotheses. Analytic studies and descriptive studies are inter-related in that analytic studies may be observational and used to generate hypotheses. The degree of intervention of the study variables determines whether an analytic study is observational or experimental. Experimental studies test a priori hypotheses and impose strict control over study variables. (Adapted from Hennekens and Buring, 1987; Sheps 1995) 55 Time Direction of Inquiry > Exposed _ ^ Disease Present Not Exposed POPULATIONS Exposed "^^> Disease Absent Not Exposed A . Case Control Study Design Time Direction of Inquiry POPULATION-People Without the Disease Exposed Not Exposed Disease No Disease < Disease No Disease B. Cohort Study Design Figure 1.2. Designs of Case-Control Study and Cohort Study. (A) . In a case control study, individuals are selected from the population on the basis of the presence (cases) or absence (controls) of the disease of interest. The exposure of cases and controls is determined retrospectively. (B) . In a cohort design, individuals without the disease of interest are selected on basis of the presence (cases) or absence (controls) of exposure to a factor of interest. In a prospective cohort design, the outcome of interest has not yet occurred and the subjects are followed for a specified period of time to determine the development of disease in each exposure group. In a retrospective cohort design, the exposure and the outcome of interest have already occurred at the time the investigator starts the study. (Sheps 1995) 56 Disease Disease Study Population Present Absent Totals Risk Factor Present a b a + b Risk Factor Absent g d_ c + d Totals a + c b + d a + b + c + d Table 1.1 Contingency Table to Aid in the Calculation of Measures of Association. Absolute Risks Risk Factor Present = a/a+b Risk Factor Absent = c/c+d Relative Risk= (a/a+b)/(c/c+d) Odds Ratio= ad/bc Attributable Risk = a/a+b - c/c+d 57 Gold Standard Disease Present Disease Absent Totals Disease New Present Diagnostic Test Disease Absent Totals a a + c a + b c + d b + d a+b+c+d Table 1.2. Comparison of a New Diagnostic Test to the Gold Standard. The comparison of a new test to criterion standards or the gold standard provides a variety of mathematical probabilities known as test characteristics which aid in the analysis and comparison of different tests. Accuracy a + d a + b + c + d the overall agreement between the test and the gold standard test Sensitivity a a + c the proportion of diseased people correctly identified by the test (true-positive rate) Specificity d b + d the proportion of non-diseased people correctly identified by the test (true-negative rate) PTL+ a a + b Post Test Likelihood of a Positive Test (Positive Predictive Value) For a patient with a positive test result, the probability that the disease is actually present PTL-c c + d Post Test Likelihood of a Negative Test For a patient with a negative test result, the probability that the disease is actually present. PTL- is not the negative predictive value. NPV d c + d Negative Predictive Value For a patient with a negative test result, the probability that the disease is absent Prevalence a + c a + b + c + d the overall probability that the disease is present prior to the test; the proportion of patients that have the disease (also known as the pre-test probability) 58 Biologic Onset (B) Clinical Onset (C) Pre-clinical Phase Detection Threshold Therapeutic Threshold Usual Diagnosis (U) Post-clinical Phase O U T C O M E Figure 1.3. The Natural History of a Disease. For a disease to be amenable to screening, the Detection Threshold and Treatment Threshold should occur in the Pre-clinical Phase and preferably, the two thresholds should be separated by a long time interval. Biologic Onset: Clinical Onset: Usual Diagnosis: Outcome: Detection Threshold Therapeutic Threshold the disease begins with the initial interaction between the patient, the causal factor(s) and the rest of the environment. For example, the interaction between a carcinogen such as nitrosamine in cigarette smoke and basal cells of the oral mucosa may result in point mutations in gene p53. the first signs are evident but the patient is not aware, is not concerned or denies the situation. For example, the patient may beware of a small, painless white lesion on the tongue in the absence of intervention or spontaneous disappearance, the disease progresses to the point where symptoms appear and the patient seeks clinical help recovery, disability or death due to the disease although there are no symptoms, the disease mechanisms produce functional or structural changes that could be detected with the appropriate test during screening or case finding. For example, cytologic smears may identify changes in DNA content or mutations in gene p53 the last point at which medical intervention has an important effect (usually curative) on altering the natural history of the disease Adapted from Sackett et al., 1991; Schechter 1995; see also Section V. 59 III. Development and Anatomy of the Tongue This Section reviews the embryological development, anatomy and function of the tongue in order to provide an appreciation for the consequences of lingual malignancy and its treatment. A. Embryological Development The development of the embryo is divided into 3 main periods: ovum, embryonic and fetal (Sperber 1973). The Period of the Ovum comprises the 7-8 days following conception during which time the ovum implants and the placenta is formed. The Embryonic Period extends from the 8th day to the 8th week and comprises the presomite (8th-20th day), somite (21st - 31st day) and late or post-somite (4th-8th week) periods. The Fetal Period extends from the 3rd month until birth (Sperber 1973). During the presomite period, the fetal membranes are established and the three primary germ layers, ectoderm, endoderm and mesoderm, are formed. During the late somite period, 5-6 mesodermal swellings known as the branchial arches develop bilaterally on the embryo's ventral aspect, caudal to the head fold of the future mandibulocervical region. Each of the five pairs of branchial arches contains a central cartilage rod that will form the skeleton of the arch, a muscular component, a vascular component, and a nervous element comprised of special visceral motor fibres of one or more cranial nerves that supply the branchial muscle arising from that arch (Sperber 1973). The mesoderm between the branchial arches does not proliferate, leaving the ectoderm (externally) and the endoderm (internally) in contact. The arches are thus separated by ectodermal/endodermal membranes which appear as four branchial grooves on the embryo's exterior and they correspond internally with the five endodermal pharyngeal outpocketings or pouches (Sperber 1973; Romanes 1986). The mucosa of tongue develops from the ventral endoderm lining the anterior internal aspects of the pharyngeal pouches and later, muscle invades these endodermal outpouchings (Atkinson and White 1992). 60 During the 4th week of development, the tongue rises into the developing mouth as a swelling that develops in two parts from the inner lining of the first four branchial arches. The first part or the anterior two-thirds of the tongue develops from three mesodermal swellings: bilateral swellings on the internal aspect of the 1st branchial arch (mandibular arch) which form the lingual swellings, and a median swelling in the floor of the mouth, the tuberculum impar, which occupies the groove between the mandibular and hyoid (second) arches (Sperber 1973; Scott and Symons, 1974). The lingual swellings and tuberculum impar enlarge and fuse to provide the mucosal covering of the anterior two-thirds of the tongue. Caudal to the tuberculum impar is a blind pit, the foramen caecum which marks the origin of an endodermal duct which migrates to the pharynx to form the thyroid gland (Sperber 1973). The second part or posterior one-third of the tongue is derived mainly from the third branchial arch which grows forwards, over the hyoid arch on the floor of the mouth (FOM), to join the back of the anterior part of the tongue. The second and third arches elevate into a midventral prominence known as the copula or hypobranchial eminence, from the back part of which develops the epiglottis (Scott and Symons 1974). The mucosa of the second to fourth arches and the copula provide the covering for the posterior third of the tongue (Sperber 1973; Scott and Symons, 1974). Between birth and adulthood, the tongue normally doubles in length, width and thickness. Failure to achieve normal growth results in microglossia (small tongue), macroglossia (excessively large tongue) or aglossia (failure to develop), and failure of fusion of its components results in a forked bifid or trifid tongue (Sperber 1973; van der Waal arid Pindborg 1986). B. Clinical Anatomy of the Tongue On the dorsal surface of the tongue, a 'V-shaped groove known as the sulcus terminalis divides the tongue into an anterior two-thirds and posterior one-third. The apex of the sulcus terminalis points posteriorly and is marked by a pit, the foramen caecum. A row of 7-12 circumvallate 61 papillae (Section III.C.7) lie immediately anterior to the sulcus terminalis and the median lingual sulcus divides the anterior tongue in half longitudinally (Romanes 1986; Atkinson and White 1992). The anterior two-thirds of the tongue is also known as the oral, mobile or palatine tongue and it comprises the body and tip of the tongue. The oral tongue is situated within the oral cavity which extends from the vermilion border of the lips to the junction of the hard and soft palate superiorly, the line of circumvallate papillae inferiorly, and laterally, to the palatoglossal arches or anterior pillars of the fauces which comprise the oropharyngeal isthmus. The oral tongue occupies the majority of the FOM where the tongue is separated from the teeth by the lingual sulcus formed by a reflection of the oral mucosa from the alveolar processes onto the ventral surface of the tongue. Posterior to the last molar, the lingual sulcus is filled in by the palatoglossus arch and anteriorly, the sulcus undermines the lateral margins of the tongue to extend beneath its free anterior third. In the anterior ventral midline, a thin crescent of mucosa called the lingual frenum connects the ventral surface of tongue to the FOM. On either side of the lingual frenum the deep lingual veins run beneath the mucous membrane medial to fringes of mucosa called the fimbriated folds. On the FOM, on either side of the frenum, are the sublingual papillae which represent the duct orifices of the submandibular salivary glands. Posterior and lateral to the sublingual papillae are two mucosal elevations, the sublingual folds, which contain the sublingual salivary gland and its numerous ductule openings as well as the duct from the submandibular gland (Romanes 1986; Atkinson and White 1992). The dorsum of the oral tongue is covered by numerous filiform papillae and scattered fungiform papillae (Section III.C.7). The lateral borders of the oral tongue, just anterior to the palatoglossal arches may contain 5 short, vertical folds of mucous membrane, the foliate papillae (Romanes 1986; Atkinson and White 1992). 62 The posterior third of the tongue is also known as the pharyngeal tongue or the root or base of the tongue (BOT). It is located in the oropharynx which extends posteriorly from the junction of the hard and soft palate superiorly, the sulcus terminalis inferiorly, the palatoglossal arches laterally, to the level of vallecula and includes the tonsils and posterior oropharyngeal wall. The dorsum of the pharyngeal tongue is smoother and thinner than the dorsum of the oral tongue and does not contain papillae. Instead, the pharyngeal dorsum contains small lymph follicles, the lingual tonsils (Section III.C.5), and multiple mucous-producing minor salivary glands (Section III.C.8). More posteriorly, the lingual mucosa is continuous with that of the vallecula and anterior surface of the epiglottis. The root of the tongue is attached to the hyoid bone and mandible by the hyoglossus and genioglossus muscles (Section III.C.2) respectively, and to the epiglottis by the glossoepiglottic fold (Romanes 1986; van der Waal and Pindborg, 1986; Atkinson and White 1992) C. Structural Anatomy of the Tongue 1. Mucosa The mucosa of the tongue consists of a stratified squamous epithelium (Section TV.A. and Section IV.C) and a layer of connective tissue, the lamina propria. Lamina propria is a dense cellular and collagenous tissue of variable thickness that contains collagen and elastic fibres, fibroblasts, blood and lymph vessels and small nerves. On a functional basis, lingual mucosa is divided into a lining mucosa and a gustatory mucosa. a. Lining Mucosa Lining mucosa (Sections IV.A. and IV.C.2) covers the ventral tongue and the root of the tongue in areas devoid of lingual tonsils and minor salivary glands. Lining mucosa is designed to permit compression and distention of the tissue, and the epithelium of lining mucosa varies in thickness which generally depends upon the trauma and mechanical loading received. Lining mucosa is generally nonkeratinized (noncornified) but may be slightly parakeratinized (Section IV.C). 63 Except for the ventral tongue, most areas of the oral cavity that are covered by a lining mucosa also have a submucosa which is a much looser connective tissue underlying the lamina propria and which binds the mucosa to the underlying muscles. The submucosa is not a distinct feature of the dorsal root of the tongue, but it also contains salivary gland acini and ducts, larger nerves and vessels that supply and drain the overlying mucosa (Ross and Reith 1985; Atkinson and White 1992). b. Gustatory Mucosa Gustatory mucosa lines the dorsum of the oral tongue. Gustatory mucosa displays the general features of masticatory mucosa (Sections IV.A and IV.C.l) and also contains numerous surface projections or papillae some of which contain taste buds (Section m.C.7). Masticatory mucosa is designed to resist mechanical stress and it is heavily parakeratinized (Section IV.C. 1). Many of the papillae are cornified and there are deep interdigitations between the epithelial rete pegs and underlying papillae of the lamina propria. A submucosa is absent as the mucosa is bound directly to the underlying lingual muscles (Section III.C.2), (Ross and Reith 1985; Atkinson and White 1992). 2 . Musculature of the Tongue During the Somite Period, the mesoderm alongside the notochord (primitive vertebral skeleton) divides into a series of 42-44 paired segmental blocks or somites including 4 occipital and 8 cervical somites which are located inferiorly to the ventral surfaces of the pharyngeal pouches, in the floor of the pharynx (Sperber 1973; Romanes 1986). During the 6th and 8th weeks, a strip of muscle from the occipital somites opposite the origin of the hypoglossal nerve, migrates cranially, carrying the hypoglossal nerve along with it to invade the tongue which is still a mucosal swelling on the FOM (Sperber 1973; Scott and Symons 1974; Romanes 1986; Atkinson and White 1992). With the exception of the palatoglossus muscle (Section III.C.2.a.iv) the muscles of the tongue are derived from the occipital somites and thus, their motor innervation is supplied by the hypoglossal 64 nerve. The palatoglossus muscle fibres and its motor innervation, the pharyngeal plexus, are derived from the third and fourth branchial arches (Sperber 1973). An incomplete fibrous median raphe or septum divides the tongue musculature longitudinally, into right and left halves. Posteriorly, the septum is attached to the hyoid bone and superiorly, the septum is separated from the dorsal mucous membrane by the superior longitudinal muscle (Romanes 1986). The musculature of each half of the tongue is divided into extrinsic and intrinsic groups. a. Extrinsic Muscles The extrinsic muscles originate from outside the tongue and they can move the tongue as well as alter its shape. Four pairs of extrinsic muscles, the genioglossus, styloglossus, hyoglossus and palatoglossus muscles attach the tongue to the mandible, styloid process, hyoid bone and soft palate, respectively (Romanes 1986; Atkinson and White 1992). (i) . Genioglossus The genioglossus muscle is the largest extrinsic muscle. It originates at the genial tubercles behind the symphysis of the mandible and extends in a fan shape, vertically upwards and backwards into the tongue to the tip (the superior/anterior fibres), the posterior third (the inferior/posterior fibres) and into the dorsum (the middle fibres); the right and left halves of this muscle are in contact in the median plane. Contraction of posterior fibres protrudes the tongue and if only one half of the muscle is active, the tongue deviates to the inactive side. Contraction of the posterior fibres also depresses the tongue in its centre and increases the volume of the tongue, such as in sucking. Contraction of superior and middle fibres depresses the tip of the tongue and retracts it (Romanes 1986; Atkinson and White 1992). (ii) . Styloglossus This muscle arises from the tip of the styloid process and adjacent part of the stylohyoid ligament. It runs downwards and forwards along the lateral walls of the pharynx to pass between the 65 superior and middle constrictors of the pharynx to insert into the whole length of the lateral tongue where it mingles with the hyoglossus muscle. Contraction of the styloglossus muscles pulls the posterior-lateral tongue margins backwards and upwards, such as in swallowing (Romanes 1986; Atkinson and White 1992). (iii). Hyoglossus The hyoglossus muscle is a flat, quadrilateral muscle arising from the body and superior surface of the greater horn of the hyoid bone. Fibres run upwards to enter the sides of the tongue, lateral to the genioglossus, where they mingle with fibres from the styloglossus. Contraction of the hyoglossus pulls the lateral borders of the tongue downwards and backwards and assists the genioglossus in enlarging the tongue during sucking motions when the hyoid bone is fixed by infrahyoid muscles (Romanes 1986; Atkinson and White 1992). Civ). Palatoglossus The palatoglossus muscles contribute to both the tongue and soft palate. The two halves of the muscle meet in the midline where they arise from the undersurface of the palatal aponeurosis and converge on the palatoglossal arch from where they insert from above, into the posterolateral aspect of the tongue to mingle with intrinsic transverse fibres. When the soft palate is fixed by other muscles, contraction of both palatoglossus muscles pulls the posterior third of the tongue upwards and backwards and pulls the palatoglossal arches together to narrow the oropharyngeal isthmus. When the soft palate is not fixed, contraction of the palatoglossus pulls the soft palate downwards towards the dorsum of the tongue to help separate the mouth from the pharynx (Romanes 1986; Atkinson and White 1992). b. Intrinsic Muscles The intrinsic muscles lie wholly within the tongue and can only modify the shape of the tongue. Intrinsic muscles are inserted in the deep fibrous connective tissue of the lamina propria of the mucosal covering or in the fibrous midline septa and run in longitudinal, transverse and vertical bundles. 66 Ci). Vertical The vertical group run from the dorsum inferiorly and laterally to flatten and widen the tongue and roll up the margins. fii). Superior Longitudinal The superior longitudinal group forms a layer on the dorsum; it curls the tip upwards and rolls it posteriorly. (iii) . Inferior Longitudinal The inferior longitudinal muscles lie lateral to the genioglossus, in the lower part of the tongue to turn the tip downwards and together with the superior muscle, retract and widen the tongue. (iv) . Transverse The transverse muscle fibres lie inferior to the superior longitudinal muscle and run between the vertical fibres of the genioglossus, hyoglossus and the vertical muscle from the septum to the margins. They narrow the tongue and increase its height (Romanes 1986; Atkinson and White 1992). 3. Innervation of the Tongue a. Sensory Innervation The sensory (tactile and gustatory) nerve supply of the mucous membrane of the tongue is explained by the different embryological origins of the tongue which retain their initially-established innervations. (i). Tactile The anterior two-thirds of the tongue is supplied by the nerve of the first branchial arch, the lingual branches of the mandibular division of the trigeminal nerve. The posterior one-third of root of the tongue is supplied primarily by the nerve of the third branchial arch, the glossopharyngeal nerve, with contributions from the nerve of the fourth arch, the vagus nerve via the superior laryngeal nerve to a small area adjacent to the epiglottis (Sperber 1973; Scott and Symons, 1974; Romanes 1986; Atkinson and White 1992). 67 (ii). Gustatory Taste sensation to the anterior two-thirds of the tongue is supplied by the nerve of the second branchial arch, the facial nerve, via the chorda tympani nerve. Taste sensation to the root of the tongue is supplied by the glossopharyngeal nerve (Sperber 1973; Scott and Symons, 1974; Romanes 1986). Although the circumvallate papillae lie in the anterior two-thirds of the tongue, they are derived embryologically from the same tissues as the posterior third of the tongue and therefore are supplied by the glossopharyngeal nerve (Atkinson and White, 1992). b. Motor Innervation The palatoglossus muscle is supplied from the vagus-accessory complex by fibres that reach the pharyngeal plexus through the pharyngeal branch of the vagus nerve (Romanes 1986). The balance of the extrinsic muscles and all of the intrinsic muscles are supplied by the hypoglossal nerve (Romanes 1986; Atkinson and White 1992). c. Autonomic Innervation Autonomic nerve fibres follow the course of the lingual blood vessels whose flow they regulate. Autonomic nerves also regulate the secretory activity of minor salivary glands in the tongue and both sympathetic and parasympathetic nerves innervate and stimulate vasodilation of the salivary glands (Ross and Reith 1985). 4 . Vasculature of the Tongue a. Arterial The main arteries of the tongue are the lingual arteries which spring from the external carotid artery opposite the tip of the greater horn of the hyoid and run anteriorly, underneath the hyoglossus muscle. The lingual artery gives off 3 branches, the suprahyoid branch, the dorsal lingual branches and sublingual branch, before continuing as the deep artery of the tongue. 68 Deep to the hypoglossal nerve, the lingual artery gives off the suprahyoid branch which runs along the superior border of the hyoid bone, lateral to the hypoglossal muscle. Deep to the hyoglossus muscle, the lingual artery gives off the dorsal lingual artery which branches upwards to supply the tongue musculature, mucosa of the pharyngeal tongue and palatine tonsils. The root of the tongue also receives arterial supply from the tonsillar branch of the facial artery and ascending pharyngeal artery (Romanes 1986; van der Waal and Pindborg, 1986; Atkinson and White 1992) At the anterior border of the hyoglossus muscle, the sublingual artery arises to run forwards and upwards to supply the sublingual gland and adjacent muscles. The sublingual artery anastomoses, through the mylohyoid muscle, with the submental artery from the external maxillary artery, which occasionally replaces the sublingual artery. The lingual artery continues as the deep artery of the tongue in the middle and anterior tongue, close to the mucosa of the ventral surface. The deep artery is a tortuous vessel permitting for elongation of the tongue and it anastomoses across the midline, with its partner (Romanes 1986; van der Waal and Pindborg, 1986; Atkinson and White 1992). b. Venous The arrangement of venous drainage is variable but all veins unite at the posterior border of the hyoglossus muscle to form the lingual veins which follow the artery deep to the hyoglossus to enter the internal jugular vein. The deep lingual vein is the main vein and it originates near the tip and runs backwards on the ventral surface, close to the mucosa, descending along the anterior margin of the hyoglossus. The dorsal lingual veins drain the dorsum and lateral borders before joining the lingual veins (Romanes 1986; van der Waal and Pindborg, 1986). 5. Lymphatic Tissues In the adult, the sites of the second and third branchial arches are marked by the anterior and posterior faucial pillars, respectively. The ventral aspects of the first and second pharyngeal 69 pouches are obliterated by growth of the third and fourth branchial arches as they contribute to the tongue, but the dorsal portions persist to develop into the auditory tubes and palatine tonsillar fossae, respectively. During the 3rd to 5th month of development, mesodermal lymphoid tissue invades the palatine, posterior pharyngeal and lingual tonsillar regions to form Waldeyer's ring (Sperber 1973). At birth, the oral mucosa of the posterior third of the tongue becomes pitted by deep crypts which form into rounded elevations, the lingual tonsil, and whose completion is marked by infiltration of lymphocytes (Sperber 1973). The lingual tonsils contain lymph nodules, often with germinal centres (Ross and Reith, 1985). 6. Lymphatic Vessels and Drainage The regional lymph nodes of the neck are often described by levels because the patterns of metastatic dissemination of epithelial cancers of the upper aerodigestive tract often occur in a sequential fashion to the regional lymph nodes (Shah and Lydiatt, 1995; Shaha and Strong, 1995). The grouping developed at the Memorial Sloan-Kettering Cancer Centre divides the cervical lymph nodes into five levels. The first echelon of nodes or Level I includes nodes of the submandibular triangle, the submental and submandibular nodes. Levels II to TV include nodes in the anterior triangle along the sternocleidomastoid muscle; Level U includes nodes in the upper jugular region such as the jugulodigastric nodes; Level III includes the mid jugular cervical nodes, Level IV includes the low jugular cervical and jugulo-omohyoid nodes. Level V includes nodes in the posterior triangle (Atkinson and White, 1992; Shah and Lydiatt, 1995; Shaha and Strong, 1995). A number of separate routes drain the lymph capillary plexus of the lingual mucous membrane. The majority of lymph vessels drain along the route of the blood vessels which supply the tongue; some parts drain bilaterally and some areas drain ipsilaterally. The factor limiting the extent of lymphatic drainage is the vertical midline fibrous septum which is impervious to lymph. The septum separates muscles from either side but it does not extend to the tip of the tongue nor to the mucosa covering the tongue. Therefore, ipsi or bilateral drainage depends upon whether lymph 70 originates from muscle or mucosa (Atkinson and White 1992). Muscle and mucosa of the tip of the tongue drain bilaterally via the submental nodes into the jugulo-omohyoid nodes on either side. The balance of the mucosa of the oral tongue drains bilaterally to the submandibular nodes into the jugulodigastric nodes, the balance of the musculature of the oral tongue drains unilaterally by the same route. Some lymphatics from the tip and dorsum may bypass the submandibular and submental nodes and drain directly into the deep cervical chain. Both the mucosa and musculature of the posterior third of the tongue drain bilaterally directly into the deep cervical chain or via retropharyngeal nodes (Scott and Symons 1974; Atkinson and White 1992). 7 . Papillae and Taste Buds a. Papillae The dorsum of the oral tongue contains 4 types of papillae: circumvallate, fungiform, filiform and foliate papillae. (i) . Filiform At 11 weeks of development, the mucosa of the dorsal oral tongue develops filiform and fungiform papillae. The filiform papillae are the smallest and most numerous papillae and they are evenly distributed over the dorsum. On the body, they are often arranged in rows parallel to the sulcus terminalis; in the tip they run transversely. Filiform papillae are conical projections, facing posteriorly, with a broad base of connective tissue covered by heavily cornified squamous epithelium. Filiform papillae do not contain taste buds and their primary function is to protect the dorsum from frictional and mechanical stresses evoked by speech and mastication (Ross and Reith, 1985; van der Waal and Pindborg, 1986; Atkinson and White, 1992) (ii) . Fungiform Fungiform papillae are globular mushroom-shaped projections interspersed among the filiforms and they are most numerous on lateral borders and the tip. Fungiform papillae are larger than 71 filiforms and they are bright red in colour. They consist of a highly vascular connective tissue core covered by a stratified non-cornified squamous epithelium and taste buds are located on their superior surface (Ross and Reith, 1985; van der Waal and Pindborg, 1986; Atkinson and White 1992). (iii) . Foliate When present and developed, foliate papillae are located on the lateral borders, anterior to the palatoglossal arches at the junction of the oral and pharyngeal tongue. They are red, leaf-like projections covered by cornified epithelium that is interspersed by taste buds (Ross and Reith, 1985; Atkinson and White 1992). (iv) . Circumvallate Circumvallate papillae develop at 2-5 months in utero. Usually 7-12 large (about 2-3mm in diameter) circumvallate papillae are situated in a row parallel and anterior to the sulcus terminalis. Circumvallate papillae are round structures surrounded by a deep trench or groove whose epithelium contains numerous taste buds. The papillae have a core of connective tissue covered by a lightly cornified squamous epithelium. Small serous glands, the glands of von Ebner, lie beneath the papillae and their secretions flood the base of the trench, dissolving solid matter and enabling it to be tasted (Ross and Reith, 1985; Atkinson and White 1992). b. Taste Buds Taste buds develop from epithelial cells starting in the 7th week of development and they appear to be functional, in their adult form, at 13-15 weeks in utero (Sperber 1973). Three types of taste buds can be distinguished histologically (Atkinson and White, 1992) but physiologically, 4 components of taste are identified: sweet, sour, bitter and salt. Rather than individual taste buds responding to only one taste component, it is more likely that individual taste buds, their cells and nerves can respond to different stimuli (Atkinson and White, 1992). Different regions of the tongue demonstrate relative differences in taste sensitivity: the tip is most sensitive to sweet, the lateral margins to sour, the base to bitter (Silloto 1975). Sensitivity to salty sensation is more 72 widespread but greatest at the tip (Silloto 1975). Taste sensation is conveyed by fibres of the facial (oral tongue) and glossopharyngeal (pharyngeal tongue) nerves which penetrate the basal lamina of the taste bud (Ross and Reith 1985). Prolonged exposure to a pure primary taste stimulus can result in adaptation to that sensation such that insensitivity to some tastes relative to others is produced, and previously neutral substances may then exhibit an apparent taste (Silloto 1975). Afferent taste fibres may also respond to warming or cooling of the tongue; sweet and bitter stimuli appear to be associated with warming temperatures whereas sour and salty stimuli are associated with cooling (Silloto 1975). 8. Lingual Salivary Glands Minor salivary glands arise from oral ectodermal and endodermal epithelium that remains as discrete acini and ducts scattered throughout the submucosa of mouth, including the tongue. The glands may be serous, mucous or mixed. Serous glands, the von Ebner glands, are located beneath circumvallate papillae where they open into the trough surrounding the papillae. Mucous glands are located in the dorsal surface of the root of the tongue, interspersed amongst the lingual tonsils. Some mucous glands are also located at the tip of the tongue and along the lateral margins. Mixed glands, the glands of Blandin-Nuhn, are located close to the ventral surface, at the tip of the tongue (Scott and Symons, 1974; Ross and Reith 1985; van der Waal and Pindborg, 1986). D. Functions of the Tongue The tongue is a mobile, tactile, muscular organ involved with the functions of speech, mastication and taste, swallowing and protective reflexes. These functions are briefly reviewed below but are discussed in depth elsewhere (eg. Matthews 1975; Lavelle 1975; Atkinson and White, 1992). 1. Speech The production of speech involves a complex coordination of phonation and articulation. Phonation comprises the regulation of exhaled air flow and the production of sound in the larynx 73 by the vibration of the vocal cords (Atkinson and White, 1992). The process of articulation modifies the sounds produced by phonation by varying the size of the oral cavity and position of the lips, tongue, palate, jaws and teeth. Although the degree of opening determines the size of the oral cavity, the position and shape of the tongue and lips determine the shape of the oral cavity. The tongue's role in articulation is particularly evident in the production of the consonants d, t, g and k but overall, even small disturbances in tongue function may cause speech abnormalities (Atkinson and White, 1992; van der Waal and Pindborg 1986). Muscle spindles of the lingual muscles provide proprioceptive feedback and positional sense but somatic sensory information is also required for speech as demonstrated by temporarily-altered speech following anaesthesia of the lingual nerve (Atkinson and White, 1992). 2. Mastication Mastication involves rendering food in the oral cavity into a state suitable for swallowing which occurs only when sensory information from the tongue indicates that the flavour, temperature and texture of the food is acceptable. Mastication involves muscles of mastication, suprahyoid and infrahyoid muscles, muscles of facial expression and extrinsic and intrinsic tongue muscles (Atkinson and White, 1992). The tongue is essential for placing food between the teeth and may also have a direct crushing effect on food by forcing it against the hard palate and pushing it onto the occluding surfaces of the teeth. The tongue also clears food displaced into the vestibules, mixes the food with saliva, forms food into a bolus and moves it posteriorly, into the oropharynx to be swallowed (Atkinson and White, 1992; van der Waal and Pindborg 1986). 3. Taste During the process of eating and drinking, normal taste sensation is due to the combined stimulation of taste and olfactory receptors, as evidenced by diminished taste sensation during the common cold which eliminates or reduces olfactory input. Taste provides essential information regarding the nature of the substances to be ingested, and tastes considered pleasant or unpleasant 74 are learnt by experience. Food is either accepted or rejected and reflexes initiated by the olfactory and taste systems either initiate salivation and secretion of gastric juices, or cause avoidance of food and nausea (Silloto 1975). Unilateral peripheral lesions of the cranial nerves associated with taste, the facial and glossopharyngeal nerves, do not affect overall taste sensitivity. A change in oral taste sensitivity requires a bilateral peripheral lesion of the chorda tympani which would simply result in an increase in the detection thresholds for salt and sweet relative to bitter and sour, but the ability to recognize the four taste qualities would remain (Silloto 1975). 4. Swallowing Swallowing is primarily a reflex response although the individual is conscious of certain aspects of swallowing. Swallowing is a continuous process which takes about 5 seconds for food to pass from the mouth to the upper oesophagus although oral, pharyngeal and oesophageal phases of swallowing have been described (Atkinson and White, 1992). Voluntary action initiates the swallowing response when the bolus is collected on the tongue, compressed between the tongue and hard palate, and propelled backwards into the pharynx. The tip of the tongue contacts the anterior hard palate, followed by the dorsum contacting the palate in an anterior-posterior direction which obliterates the oral cavity and forces the bolus backwards. Contact of food with the mucosa overlying the posterior oropharyngeal wall or the palatoglossal arches stimulates afferent sensory endings of the glossopharyngeal nerve in the pharyngeal plexus; this initiates a reflex chain of contractions in the pharynx and oesophagus (Atkinson and White, 1992; van der Waal and Pindborg 1986). Respiration is temporarily suspended and positioning of the epiglottis over the larynx prevents aspiration of the bolus. 5. Protective Reflexes The lower airway and gastrointestinal tract is protected from the entry of foreign bodies by the gag 75 reflex and from the ingestion of potentially toxic food by the vomiting reflex. The gag reflex occurs when a swallowing reflex is initiated by contact of the oropharynx by an object or material that cannot be swallowed; the mouth is opened and the posterior part of the tongue is elevated in attempts to expel the material. Vomiting occurs when contraction of the anterior abdominal wall forces stomach contents back through the cardiac sphincter and up into the oropharynx and oral cavity (Atkinson and White, 1992). Epiglottis Palatopharyn fold Palatine tonsil Lingual tonsil Palatoglossal Sulcus terminalis Circumvallate papillae Median furrow Fungiform papillae Figure 1.4. The Dorsal Surface of the Tongue. Adapted from Atkinson and White, 1992, page 329. 77 IV. Epithelium This Section reviews the normal organization, function, proliferation and differentiation of oral epithelium because they are relevant to understanding carcinogenesis and the behaviour of malignant epithelium. A. General Characteristics An epithelium can be described as a sheet of cells in which there is relatively little intercellular material. The cell sheets are characterized by intimate intercellular contacts (Section TV. B) which couple the cells to one another structurally, electrically, metabolically and mechanically so that the cells, rather than the extracellular matrix (ECM) bear the functional stresses. Epithelium is avascular and therefore dependent upon diffusion of nutrients from the underlying connective tissue from which it is separated by a basement membrane. Epithelial tissues are classified by the shape of the cells (flattened or squamous, cuboidal or columnar) and by the arrangement of the cells into single (simple) or multiple (compound) layers. In compound or multilayered epithelia such as epidermis and oral mucosa, the surface cells are mature, highly differentiated cells that are lost or shed (desquamated) from the surface and they are continuously replaced by the proliferation of less differentiated, more primitive cells that reside in subsurface sites (Scott and Dixon, 1972; Ross and Reith, 1985; Alberts et al., 1989; Atkinson and White, 1992). Epithelial cells called keratinocytes (Section IV.C) comprise 90% of the cells in oral epithelium; the remaining 10% consist of Langerhans cells (Section IV.B.3) and melanocytes (Lavelle 1976b; Squier 1980). Keratinocytes in the basal layer divide and change their appearance as they differentiate and migrate to the surface where they die (apoptosis, Section IV.D.2.a.iv.) and are shed. In different regions of the oral cavity the rate of cell proliferation, the thickness and number of cell layers, the type of differentiation and nature of the surface layer vary in relation to different functional demands. On a functional basis, oral mucosa is usually divided into 78 1. masticatory mucosa which is exposed to mechanical compression and friction, and covers the gingiva and hard palate. It is usually a cornified (keratinized) epithelium overlying a dense fibrous connective tissue that is firmly attached to the underlying structures. 2. lining mucosa which flexible and extensible and covers the lips, cheeks, floor of the mouth (FOM), alveolar mucosa, ventral tongue and soft palate. It is usually a noncornified epithelium overlying a loosely-fibrous connective tissue (submucosa) that is flexibly attached to bone or muscle. 3. specialized gustatory mucosa covering the dorsal tongue. Functionally, it is a masticatory mucosa but it contains specialized lingual papilla and taste buds (Squier 1980; Ross and Reith, 1985). Masticatory and lining epithelium also demonstrate different patterns of interdigitation with the underlying connective tissue. Different ratios of the area of interface to that of the epithelial surface reflect differences in mechanical attachment of the two tissues and in rates of metabolic exchange (Squier 1980). Noncornified lining mucosa is generally thicker than cornified masticatory mucosa but again regional modifications in relation to function exist; for example, noncornified cheek mucosa is thicker than noncornified mucosa lining the FOM. As well, masticatory mucosa exposed to chronic friction will demonstrate increased thickness of the keratin layer (Section IV.C), (Lavelle 1976b). These regional and functional modifications are reflected by differences in mitotic rate and by differences in permeability, both of which may play a role in a tissue's susceptibility to malignant transformation (Section V). B. Protective Role of Oral Epithelium The oral epithelium lines the oral cavity and protects the tissues it covers. Oral epithelium not only resists mechanical forces but it restricts the entry of microorganisms and acts as an impermeable 79 barrier to noxious substances. Many oral diseases are associated with the penetration of antigens and toxins and therefore mechanisms whereby this may be restricted or prevented are essential when considering the initiation of oral mucosal disease or malignancy. The protective functions of oral epithelium are mediated by 1. Desquamation. Desquamation is the normal turnover of tissue whereby the surface layer is continuously shed. This is an important mechanism for elimination of oral microorganisms. 2. Permeability Barriers. Permeability of the oral mucosa is determined by different kinds of macromolecular barriers that include a. the surface layer of adherent mucin derived from salivary glands b. keratin, an insoluble polymer located in the superficial layers of cornified epithelium (Section IV.C) c. the intercellular permeability barrier formed by specialized intercellular junctions, the tight (occluding) junctions, and anchoring junctions (desmosomes and adherens junctions), (Alberts et al., 1989). Tight junctions form independent, impermeable barriers of contact between cell membranes of adjacent cells, effectively sealing adjacent cells together, obliterating the intercellular space and blocking diffusion. Anchoring junctions rely on transmembrane proteins called cadherins, and anchoring junctions connect the cytoskeleton of one cell to adjacent cells via desmosomes and adherens junctions. Adherens junctions are intercellular connection sites for actin filaments of adjacent cells. Desmosomes are intercellular connection sites where bundles of intermediate filaments called tonofibrils (Section IV.C) from one cell are connected to tonofibrils in neighbouring cells within and between cell layers so that the filaments form a continuous network throughout the epithelium (Alberts et al., 1989). Desmosomes and tonofibrils are essential for maintaining the integrity of the epithelium and for disseminating locally-applied forces. Their 80 importance is illustrated by serious skin conditions such as pemphigus in which antibodies against desmosomal linker proteins disrupt the desmosomes (Alberts et al., 1989), and epidermolysis bullosa simplex in which mutations of tonofibrils (keratins) K5 and K14 cause fragility of the basal keratinocytes (reviewed by McLean and Lane 1995). Adjacent epithelial cells also communicate with one another via gap junctions which are 3-nm-wide gaps between the plasma membranes of adjacent cells. Gap junctions allow coupled cells to share small molecules such as inorganic ions, sugars, amino acids, nucleotides and vitamins which pass freely from one cell to another (Alberts et al., 1989; Atkinson and White, 1992). d. the barrier formed by the basement membrane which consists of two layers, the basal lamina produced by epithelial cells and the lamina reticularis produced by connective tissue cells. Basal lamina also controls the orientation, intracellular organization, attachment and migration of basal epithelial cells which are attached to the basal lamina by hemidesmosomes and focal contacts. Hemidesmosomes represent the termination site of intermediate filaments; focal contacts are connection sites mediated by integrins which are transmembrane receptors linking actin filaments to components of the ECM. (reviewed by Lavelle 1976b and Squier 1980) 3. Cellular defense mechanisms comprising Langerhans (dendritic) cells (eg. Daniels 1987) and polymorphonuclear leukocytes. 1. Desquamation The continual physical shedding of the surface epithelial layer is an important mechanism for removing adherent bacteria which are lost along with the desquamated cells. In order to maintain homeostasis, the cell loss at the surface must be balanced by the rate of mitosis and the rate of differentiation and migration of cells to the surface (transit time). In general, the turnover of oral mucosa is faster than skin but slower than intestinal mucosa. Within the oral mucosa, noncornified 81 regions turn over faster than cornified regions; for example, the turnover time for human oral lining mucosa is 5-16 days compared to 28-40 days for human gingiva (Squier 1987). An example of the faulty control of basal cell proliferation and surface desquamation is evident in psoriasis which is a common benign skin disorder characterized by erythematous, scaly plaques. In psoriasis the rate of basal cell proliferation is greatly increased and the cell turnover rate is up to eight times greater than normal. Hyperproliferation of the epidermis results in epithelial hyperplasia but cells are shed from the surface before they have had adequate time to fully cornify, generally within as little as a week after their formation in the basal layer (Laskaris 1988; Alberts et al., 1989; Regezi and Sciubba, 1989). Mitotic activity in gingival epithelium appears to increase with age and it has been suggested that this change reflects diminishing control of cell proliferation and might relate to the higher incidence of cancer in the elderly (Lavelle 1976b). Uncontrolled proliferation results in neoplasia which can be either benign or malignant (Sections V, VH) and the complex mechanisms controlling cell proliferation are reviewed in Section IV.D. 2. Permeability Barriers Saliva has a definite protective role in the oral cavity (reviewed by MacFarlane 1976) but its role as a solvent may also facilitate penetration of substances that are dissolved in saliva. Moreover, the dissolved substances are maintained in solution at the surface of the mucosa and some oral locations also serve as "puddling spots" (Moore and Catlin, 1967) which are continuously immersed in saliva. Significantly, 80% of oral malignancies occur in a horseshoe-shaped area that corresponds to "saliva reservoirs" but which comprises only 20% of the surface area of the entire oral mucosa (Moore and Catlin, 1967). For example, pipe smokers characteristically develop a symmetrical leukoplakia of the palate with inflammation and swelling of the minor salivary glands (nicotinic stomatitis); however, the palate is a rare site for carcinoma yet when a pipe smoker develops oral cancer, it is commonly in the retromolar region and FOM (Cawson 1975). Perhaps 82 the leukoplakia results from the physical effects of heat from the pipe; the malignant change may be the result of carcinogens dissolved in saliva, draining and accumulating in the FOM (Cawson 1975; Squier 1980). The most superficial barrier to diffusion is provided by salivary mucins which cover the surface of the oral epithelium. Mucins provide not only a permeability barrier, but they protect against desiccation of the epithelium and offer lubrication against surface abrasion. Mucins also play an important role in regulating microbial clearance and adherence in the oral cavity by aggregating organisms, and by trapping or concentrating protective molecules such as salivary IgA and lysozyme on the tissue surface (reviewed by Levine et al., 1987). Toxins and antigens produced by oral microorganisms, and carcinogens are potentially harmful if they are able to penetrate and cross the oral epithelium. Substances can penetrate the epithelium by traversing through the epithelial cells by endocytosis which includes pinocytosis (intake of fluids) and phagocytosis (intake of solid particles). In oral mucosa, cells of the stratum basale and spinosum (Section IV.C) are capable of endocytosis but this does not appear to be a likely transport mechanism across the entire epithelium (Squier and Johnson, 1975). In some epithelia, molecules and ions are transported by active transport but this has not been demonstrated in oral epithelium. Molecules can also diffuse across cell membranes through either a lipid phase or along aqueous channels but this would be an inefficient method in a complex stratified epithelium such as oral mucosa (Squier and Johnson, 1975). Although oral epithelial cells are closely apposed, tight junctions are not common in oral epithelium and therefore they are not the major intercellular barrier (Squier and Johnson, 1975). Hence, the intercellular space is sufficient to permit diffusion of some molecules and ions, and this appears to be the main mechanism of penetration through oral epithelium (Lavelle 1976b; Squier and Johnson, 1975; Squier 1987). The rate of diffusion and permeability of a substance are affected 83 by a number of factors including temperature, pH, molecular weight, interaction between the solvent and solute, concentration of the substance, duration of contact between the substance and the mucosa, thickness of the mucosa, and integrity of the permeability barriers (Lavelle 1976b; Squier 1980). In general, molecules penetrate more readily than ions, small molecules more readily than large molecules, and gases and volatile substances the most readily of all (Squier and Johnson, 1975). Substances that are lipid-soluble move more rapidly across mucosa than water-soluble ones, and substances with permeability in both penetrate most rapidly (Squier 1980). The skin is less permeable to water than any oral region. Within the oral cavity, the gingiva is the least permeable and most similar in permeability to epidermis, followed by the buccal mucosa and then the FOM which is most permeable (Squier and Hall, 1985). Keratin (Section IV.C) offers the major resistance to diffusion of water and both polar and nonpolar substances, and in terms of barrier function, parakeratinized epithelium is similar to orthokeratinized epithelium (Section IV.C). Polar molecules and electrolytes are also limited by the presence of lipid-containing membrane-coating granules which are evident throughout the superficial cell layers of cornified and noncornified regions. However, differences in the lipid content of the surface layers of cornified and noncornified epithelia may account for differences in their respective permeabilities. Consequently, cornified epithelium is impermeable to water and polar compounds; noncornified epithelium is less effective in resisting penetration of water and polar compounds but may be capable of resisting penetration of larger molecules such as proteins (Squier and Johnson, 1975; Squier 1987). Significantly, many oral lesions such as lichen planus and other premalignant conditions (Section VI) as well as SCC, are most frequently found in noncornified lining regions but not all noncornified areas are equally susceptible (Squier 1987). That is, the FOM is a 'high risk' site for SCC but buccal mucosa is not, yet both sites are lined by noncornified epithelium. The thickness of the buccal mucosa is typically thicker than FOM mucosa, suggesting that thickened epithelium is 84 less permeable and hence less susceptible (Lavelle 1976b, Squier 1987). However, increased thickness of epithelium does not appear to automatically confer improved barrier function and in fact, may reduce it. For example, hyperplastic and hyperkeratotic oral epithelia do not demonstrate improved barrier function, a finding that is consistent with measurements of increased permeability in hyperkeratotic palmar and plantar epidermis as compared to thin skin (Squier 1987). In fact, increased thickness of epithelium is often associated with less complete maturation of the surface layers and therefore barrier functions are reduced (Squier and Johnson, 1975). If noxious substances breach the superficial permeability barrier and the epithelium, further penetration of some substances may be limited by the basal lamina at the epithelial-connective tissue interface. The basal lamina is a differentially-permeable membrane that regulates metabolic exchange between the epithelium and connective tissue and allows passage of selected molecules and certain migratory cells such as polymorphonuclear leucocytes. Basal lamina also restricts penetration of many substances including endotoxin and immune complexes (Squier 1987). If the basal lamina is breached, the connective tissue of oral mucosa is not an effective barrier against penetration of polar substances although it may limit diffusion of macromolecules and non-polar substances (Squier and Johnson, 1975). 3. Cellular Defence Mechanisms The oral mucosa also contains immunocompetent antigen-presenting cells, the Langerhans cells (reviewed by Daniels 1987) which may have a role in induction of contact hypersensitivity or tolerance. The density of oral Langerhans cells displays regional variations in that the frequency of Langerhans cells varies inversely with the degree of cornification (Daniels 1984). In noricornified mucosa, Langerhans cells are located in deep, suprabasal epithelium approximately parallel to the basement membrane and their number is similar to the number of Langerhans cells in epidermis. In cornified mucosa of the hard palate and gingiva, Langerhans cells are located in midepithelium, parallel to the surface and their numbers are much lower, including irregular sites without 85 Langerhans cells. On the dorsal tongue, Langerhans cells are absent from the interpapillary epithelium but are abundant at the tips and one side of the filiform papillae (Daniels 1984). Erosion, ulceration, and mucosal atrophy, possibly in relation to age, result in the reduction or loss of the superficial epithelial permeability barrier; hence, permeability of the epithelium is increased, favouring the penetration of antigenic substances (Squier 1987). Antigenic material that penetrates into the epithelium may be trapped by antibodies as immune complexes within or beneath the epithelium. Inflammatory cells accumulate beneath the epithelium and elaborate proteolytic enzymes which damage the basal lamina, threatening the permeability barrier to other molecules. Neutrophils also infiltrate the epithelium, secrete proteolytic enzymes and thus reduce intercellular permeability barriers. Inflammation often produces hyperplastic changes characterized by acanthosis (thickening of the stratum spinosum) and hyperkeratosis due to accelerated cell division and accelerated passage of cells to the surface. Consequently, there is insufficient time for normal barrier layers to be formed and permeability is further increased (Squier and Johnson, 1975; Squier 1980). The effects of erosion and inflammation upon permeability are illustrated by denture-loaded palatal mucosa in which inflammation and erosion cause permeability to water to increase to twice that of normal noncornified buccal mucosa (Riber and Kaaber, 1978). C. Differentiation of Oral Epithelium Differentiation of basal epithelial cells is a complex process associated with the structural organization of the stratified squamous epithelium. It involves an ordered sequence of defined morphological changes accompanied by the sequential expression and modification of specific differentiation products. This Section reviews the biochemical and morphological changes that reflect the differentiation of basal cells as they migrate from the basal layer to the superficial layers of the mucosa. Discussion also includes differences in cornified and noncornified mucosa, differences in keratin expression and a brief review of vitamin A and connective tissue influences on oral epithelial differentiation. 86 In the literature, the use of the term "keratin" may be confusing because it is used to describe a class of intermediate filaments (Section IV.C.3) that are unique to epithelial cells (eg. Goldman and Steinert, 1990; Morgan and Su, 1994) as well as the tough, insoluble cross-linked keratin polymer formed by the accumulation of intermediate filaments in a filaggrin matrix and located in the superficial layers of ortho and parakeratinized epithelium (Ross and Reith, 1985). Keratinocytes are epithelial cells that produce keratin. Whereas all keratinocytes produce a keratin filament, not all keratinocytes (i.e. keratinocytes from noncornified epithelia) produce keratin-polymer in their surface layer (eg. Atkinson and White 1992). The term 'keratinized' is synonymous with the term 'cornified' and throughout this thesis, the term 'cornified' will be used to describe epithelium which contains the keratin polymer in its superficial layer. The term 'noncornified' is used to describe epithelium which does not contain the keratin polymer in its superficial layer. 1. Cornified Masticatory Epithelium Keratinization is the process of differentiation whereby viable keratinocytes transform or terminally differentiate into dead surface cells that are packed with keratin polymer (Atkinson and White 1992). Keratinization is an orderly, dynamic process of cellular proliferation and differentiation that is reflected by the structural and morphological organization of the cornified masticatory epithelium into four layers: stratum basale or germinativum, spinosum, granulosum and corneum. As cells migrate from the basal layer to the surface, they demonstrate an increasing accumulation of different intracellular proteins (eg. cytokeratin filaments, involucrin, filaggrin), an increasing number of desmosomes and an increased thickness of their cell membranes (eg. Ross and Reith, 1985; Atkinson and White, 1992). Differentiation is also reflected by the expression different keratin proteins which is discussed in Section IV.C.3. a. Stratum Basale In masticatory mucosa subjected to high mechanical loads, the junction between the basal epithelial layer and the connective tissue is highly interdigitated due to the interlocking arrangement of connective tissue papillae and epithelial rete ridges. Typically, the rete ridges are almost parallel-sided and they taper downwards to blunt ends (Kramer 1980). The basal layer is attached to the basal lamina and consists of a single layer of cuboidal or columnar cells. The basal cells are progenitor or stem cells whose continual division replenishes and compensates for the cells shed from the surface of the epithelium (Ross and Reith, 1985; Atkinson and White, 1992). Basal cells express major histocompatibility class I (MHC) antigens (HLA-A, B, C) which are retained throughout their differentiation (reviewed by Eversole 1993). Basal cells also express MHC class II antigens (HLA-DR, DP, DQ) (reviewed by Eversole 1993), and undifferentiated keratins K5/K14 (Table 1.3), (eg. Darmon 1991); however, neither the class II antigens nor the primary keratins are expressed in the suprabasilar layers. b. Stratum Spinosum As soon as the basal cells detach from the basement membrane, phenotypic changes occur. The cells produce involucrin, a protein that forms a cross-linked, thickened layer on the cytoplasmic surface of the cell membrane. A new set of keratin filaments (KI, K10; Table 1.3) are produced and the keratin filaments aggregate into bundles called tonofibrils which form an intracellular network that attaches to the plasma membrane at desmosomes (Ross and Reith, 1985; Darmon 1991). In this layer, the number of desmosomes increases and the cells are slightly separated from each other, assuming an irregular polyhedral shape with long "spiny" cytoplasmic processes by which the cells attach to each other via desmosomes. The spinous layer may be several layers thick and the cells become increasingly flattened as they move farther away from the basal layer (Atkinson and White, 1992). Occasionally, the basal and spinous layers are classified together as the Malpighian layer (eg. Leeson and Leeson, 1970) which is collectively responsible for 88 proliferation and the initiation of keratinization. c. Stratum Granulosum Desmosomes are structurally best-organized in the suprabasilar layers up to the granular layer where they begin to lose their organization. The granular layer consists of several layers of flattened cells whose long axis is parallel to the surface. The thickness of the plasma membrane increases and the cells synthesize filaggrin which accumulates in the cytoplasm as globular masses called keratohyaline granules. In the superficial granular cell layers, the number and size of the keratohyaline granules increases, and the cytoplasm also contains lipid-containing membrane-coating granules (Ross and Reith, 1985; Darmon 1991; Atkinson and White, 1992). d. Stratum Corneum ri). Orthokeratin In orthokeratinized epithelium, there is an abrupt change in appearance between the viable cells of the granular layer and the dead, cornified scale-like cells or squames of the corneal layer which contain the keratin polymer. Desmosomes have disappeared and the cells have a thickened, reinforced plasma membrane due to the tough, cross-linked layer containing involucrin. The cells are devoid of a nucleus and organelles but instead, they are packed with cytokeratin filaments which have aggregated in a matrix of filaggrin released by the keratohyaline granules. As well, the lipid content of the membrane-coating granules is excreted and the protective impermeable cornified stratum corneum is formed (Squier 1980; Ross and Reith, 1985; Darmon 1991; Atkinson and White, 1992). (ii). Parakeratin. In parakeratinized epithelium, cells retain their nucleus and some degenerate organelles. However, the cytoplasm is packed with keratin filaments embedded in filaggrin (Atkinson and White, 1992). 89 2. Noncornified Lining Epithelium Noncornified epithelium is relatively distensible, adjusting to the movements of the underlying muscles, and is generally thicker than cornified epithelium. Rete pegs of lining mucosa are fewer in number and are rounder and shorter than in masticatory mucosa. Lining mucosa consists of four layers: stratum basale, spinosum, intermedium and superficiale. Keratin filaments (K13, K4; Table 1.3) are expressed by lining epithelium but they are arranged into a loose network rather than into bundles of tonofibrils as in cornified mucosa. The frequency and size of desmosomes in all strata are less than in cornified mucosa. As well, the stratum intermedium lacks keratohyaline granules and the stratum superficiale is generally not cornified although it may be parakeratinized in some locations. The surface cells in lining mucosa are less flattened with highly folded cell walls, enabling the mucosa to elongate in response to tensile forces. Glycogen may be present in the superficial epithelium and some of the organelles including the nucleus, persist to the surface (Lavelle 1976b; Squier 1980; Ross and Reith, 1985; Atkinson and White, 1992). 3. Keratins as Markers of Differentiation Keratins are a class of intermediate filaments that are unique to epithelial cells. Intermediate filaments comprise part of the cytoskeleton and they are polymers of intracellular proteins organized to provide the principal structural support and stabilization for the epithelium. There are at least 30 distinct keratin types which consist of two classes: acidic (type I) and neutral/basic (type II). Equal numbers of type I and II keratin subunits are combined to form heterodimers called keratin or cytokeratin filaments (Table 1.3). Keratins are encoded by a large family of homologous genes and from the time a new keratinocyte in the basal layer is transformed into a superficial cell of either the stratum corneum or superficiale, it expresses a succession of different genes from its cytokeratin gene repertoire (Moll et al., 1982; Cooper et al., 1985; Darnell et al., 1986; Morgan and Su, 1994). Thus, epithelial cells express multiple keratins, usually in consistent pairs, and the expression of keratins is correlated to the degree of epithelial differentiation, making keratins the best marker of epithelial differentiation (Moll 1987). 90 Stratified epithelia demonstrate primary keratin markers K5/K14 in their basal compartment and secondary differentiation-specific keratins that are unique to the suprabasilar compartments of cornified and noncornified epithelium (Table 1.3). In cornified oral epithelia, suprabasilar layers are characterized by K1/K10 and K6/K16, the latter pair indicating a high cell turnover rate. In noncornified oral epithelia, the suprabasilar layers of are characterised by K4/K13 as well as K6/ K16. In addition, K19 is distributed throughout all layers of noncornified oral epithelium but predominately in the basal layer (Morgan and Su, 1994; Kautsky et al., 1995; Su et al., 1996). Furthermore, as part of a normal variation, the suprabasilar layers of small cell groups in an epithelium may display keratins that are uncharacteristic of that epithelium. That is, lining mucosa of the cheek may display small foci of K1/K10 (markers of cornification) and attached gingiva or hard palate may display foci of K4/K13 (noncornification markers), (Morgan and Su, 1994). The lateral and ventral surfaces of the tongue are lined by noncornified epithelium and thus demonstrate K19; K4/K13; K6/K16. The dorsal surface of the tongue demonstrates mixed patterns of cornification. The filiform and fungiform papillae demonstrate markers of cornified epithelium but the inter-papillary epithelium expresses noncornified markers. The taste buds (associated with some types of papillae) demonstrate K19 and simple epithelium markers K8/K18 (Morgan and Su, 1994). The profile of keratins expressed between and within different epithelia is a sensitive indicator of epithelial differentiation. However, keratins have a long half-life of at least 4 days (Denk et al., 1987) and therefore, their detection may mask actual changes in keratin gene activity. In fact, keratin mRNA may be present in tissues in which keratin protein is either not detected or would not be expected (Su et al., 1996). Ideally, both keratin protein and keratin mRNA should be assessed but there are limitations in sensitivity of both immunohistochemistry (eg. Bacallao et al., 1990; Denk 1987; Moll 1987) and in situ hybridization (eg. Terenghi and Fallon, 1990; Su et al., 1996) techniques, respectively (see also Section n.H.l). Close correlation in location between mRNA 91 and its protein indicates that control of gene expression is regulated at the transcriptional level. This transcriptional control is evident in oral epithelium for K14 (primary keratin of stratified epithelium) and noncornified oral epithelium for K19. In contrast, cornified epithelium expresses K19 rnRNA but not its protein, indicating that K19 expression in these cells is controlled post-transcriptionally (Su et al., 1996). 4. Influence of Retinoids on Epithelial Differentiation The differentiation of keratinocytes is due to many changes in gene expression, some of which are related to the control of cell division and others to the expression of the squamous differentiated phenotype, including keratin markers. The proliferation and differentiation of keratinocytes is mediated locally by vitamin A (retinol) and its derivatives (retinoids) and a greatly over-simplified summary of a complex topic is presented below (see also Table 1.4), (eg. Darmon 1991; Pfahl 1994; Love and Gudas, 1994; Gudas et al., 1994; Jetten et al, 1994). Retinol is oxidized to retinoic acid (RA) and its isomers which are the biologically-active forms of vitamin A in epithelia. In the adult, retinoids promote the normal differentiation of epithelial cells, and its effects on the synthesis of differentiation markers by keratinocytes occur at the transcriptional level and are mediated through two types of nuclear receptors (RAR for RA; RXR for other retinoids) and their isoforms (a, p, y), (Darmon 1991; Love and Gudas, 1994) which differ in their abilities to bind RA and retinoids (Xu et al., 1994). RARs belong to a large family of DNA-binding regulatory proteins that bind to promoter regions of specific genes and thereby modulate the expression of those genes. In order for a keratinocyte to respond to RA, the keratinocyte must contain nuclear RA receptors (RARs), and the genes that are expressed during differentiation must contain promoter regions that contain RA-responsive elements (RAREs) able to bind the RA receptors (Darmon 1991). Examples of genes that contain RAREs include genes for laminin, growth hormone, the receptor for epidermal growth factor, and keratin K14 (Darmon 1991). RARs may also facilitate cross-talk between different hormone signalling pathways, 92 suggesting a hormone-like control of cell differentiation by RA (van Poppel 1993; Pfahl 1994; Love and Gudas, 1994; Gudas et al., 1994; Jetten et al, 1994). Furthermore, two sets of cellular RA-binding proteins (CRABPs) sequester RA in the cytoplasm. CRABPs maintain intracellular RA concentrations for the appropriate differential regulation of gene transcription, possibly by preventing the interaction of RA with its nuclear receptors (Love and Gudas, 1994; Kautsky et al., 1995). Retinoids are multifunctional agents and similar to growth factors (Section IV.D.2.b.iii), their ability to elicit specific cellular responses depends upon the biological content or nature of the local ECM (Nathan and Sporn, 1991). Since RA controls the expression of a large variety of genes including genes encoding growth factors and their receptors, cellular enzymes and structural proteins including keratin, altered levels in any of these products can drastically alter the differentiation state of a particular cell. For example, retinoids control genes encoding gap junctional proteins; enhanced gap-junctional communication suppresses cell growth and suppresses malignant transformation, whereas inhibition of communication enhances these processes. Retinoids are also involved in restructuring of the ECM, including glycoproteins, metalloproteinases and their inhibitor, TIMP. Consequently, retinoids influence cell membrane permeability and cell-cell interactions including communication and adhesion not only amongst epithelial cells but also between the epithelium and connective tissue. In turn, the ability of retinoids to elicit specific cellular responses may rely on the nature of the underlying connective tissue (van Poppel 1993; Love and Gudas, 1994; Gudas et al., 1994). Terminal differentiation of oral epithelium may be influenced by at least two different RA-associated regulatory mechanisms. One is via a direct, RA-concentration-dependent mechanism that controls intrinsic properties of the epithelium itself. For example, under experimental conditions, low concentrations of RA decrease expression of K13 and K19, markers of noncornified oral epithelium, but increase the expression of profilaggrin and KI, markers of 93 cornification (Table 1.4). High concentrations of RA have the opposite effect; i.e. increase markers of noncornification and decrease markers of cornification (Kautsky et al., 1995) including the formation of involucrin involved in formation of the crosslinked cell membrane (Jetten et al., 1994). The clinical application of topical retinoids to the epidermis decreases the number of tonofilaments and desmosomal attachments. As a result, intercellular spaces are widened and cohesiveness of the stratum corneum is decreased which causes increased fragility of the upper epidermis. In addition, the function of the permeability barrier is impaired; transepidermal water loss is increased and the percutaneous absorption of topical agents is enhanced which can be either beneficial or potentially toxic (Peck and DiGiovanna, 1994). In essence, RA suppresses expression of the squamous (cornified) differentiated phenotype (Darmon 1991), and the maintenance of the noncornified state of squamous cells may depend on the continuous presence of retinol (Xu et al., 1994). Significantly, keratinocytes derived from either cornified or noncornified oral mucosa respond similarly to RA regulation of cornification, including KI, K13 and filaggrin expression. However, the expression of K19 in these two keratinocyte types differs with RA concentration and is due to the differential expression of RAR isoforms in the two cell types. RAR isoforms have differential control of cytokeratin and filaggrin expression and RARyin particular, may control expression of K13, KI and profilaggrin. RARa and RARymRNAs are equally expressed in both cornified and noncornified cell types and their levels are only minimally affected by RA. In contrast, RARp is expressed predominantly in noncornified cell types and is linked to expression of K19. Moreover, elevated concentrations of RA induce expression of both RARp and K19 mRNA levels (Crowe et al., 1991; Hu et al., 1991; Kautsky et al., 1995). A second level of differentiation control may be exerted by the subepithelial connective tissue but, the mesenchymal influences on epithelial differentiation vary depending upon the body site and the particular epithelium and connective tissue involved (eg. Mackenzie and Hill, 1984). That is, 94 expression of pre-existing patterns of keratin expression, basal proliferation and cell turnover may be maintained following recombination of the epithelium with a connective tissue from a site not normally associated with such patterns; in that instance the connective tissue permitted the epithelial phenotype, or alternatively, the epithelium was not capable of responding to the connective tissue signals (Hill and Mackenzie 1989). In contrast, some epithelia acquire keratin patterns and proliferative rates corresponding to those of the epithelium normally associated with the connective tissue, indicating directive or inducing influences from the connective tissue to an epithelium that was capable of responding (Mackenzie and Hill, 1984; Hill and Mackenzie 1989). Oral lining epithelium cornifies in response to inductive signals from masticatory subepithelial connective tissue and this behaviour forms the basis for grafting masticatory mucosa to sites of lining mucosa along cervical tooth margins (gingival grafting), (Karring et al., 1972, 1975). However, in contrast to subepithelial connective tissue, deep connective tissue does not have the full potential to induce cornification, and epithelium grafted onto deep connective tissue expresses a hybrid mixture of cornified and noncornified keratin markers (Ouhayoun et al., 1988). Depending upon the differentiation marker measured, there may be a continuous cross-talk between the oral fibroblasts and oral keratinocytes which leads to a terminal differentiation of the epithelial cells. For example, K19 expression may be intrinsically predetermined whereas KI, K13 and profilaggrin expression may be receptive to extrinsic influences (Kautsky et al., 1995). In vitro studies (Kautsky et al., 1995) have demonstrated that subepithelial fibroblasts can modulate the sensitivity of keratinocytes to RA and thereby influence differentiation of oral keratinocytes. Fibroblasts from cornified oral mucosa can inhibit the RA response of oral keratinocytes, resulting in increased expression of markers of cornification (profilaggrin, KI). In contrast, fibroblasts from noncornified oral mucosa potentiate the RA response of oral keratinocytes as indicated by their expression of K13. Oral fibroblasts appear to influence the apparent RA exposure of the keratinocytes independently of the actual RA concentration, and fibroblasts may elaborate diffusible factors which indirectly regulate the ability of CRABPs to 95 sequester RA in the epithelial cytoplasm (Boukamp et al.,1990; Kautsky et al. 1995). Thus, the regulation of keratinocyte growth and differentiation is controlled by intrinsic epithelial programs but it is also dependent upon underlying fibroblasts and the ECM. Components of the ECM bind growth factors thus affording the ECM a key regulatory mechanism in tissue organization. The turnover rate of ECM and basement membrane is carefully controlled by mechanisms which regulate the levels of metalloproteinases (eg. collagenase), levels of TJJVIP and the synthesis of new ECM components (Ansari and Hall 1992). RA also plays an important role in turnover of ECM as RA stimulates laminin production but inhibits the secretion of collagenase by keratinocytes (Table 1.4). RA influences fibroblasts and can either stimulate or inhibit synthesis of type I collagen and fibronectin by fibroblasts depending upon their source and the concentration of RA. In fibroblasts, RA regulates the expression of metalloproteinases and TIMP in an inverse manner by repressing the expression of metalloproteinases and favouring expression of TJJVIP (Gudas et al., 1994). These latter effects may be mediated by direct control of RA at the transcriptional level, or they may be mediated indirectly via the induction of a growth factor, transforming growth factor p (TGFp) which has an effect similar to RA on ECM turnover (Gudas et al., 1994). Mechanisms whereby retinoids affect communication between cells and regulate turnover of the ECM including neovascularization, are essential to maintaining tissue homeostasis as well as physiological repair of tissue injury. They are also essential features in carcinogenesis which is discussed in Section V. D. Epithelial Proliferation In normal tissue, the number of cells with a particular phenotype is strictly regulated through control of stem cell proliferation, cell differentiation and appropriate spatial organization balanced by the loss of cells to terminal differentiation and natural cell death (apoptosis, IV.D.2.a.iv). In 96 stratified squamous epithelium, proliferation limited to basal stem cells, is necessary to maintain homeostasis of the epithelium because the surface layers are terminally differentiated and incapable of dividing. The surface cells are continuously shed and this cell loss must be balanced by continual production of new cells from the basal layer. It is essential that each tissue maintains a size appropriate for the body's needs and under normal conditions, cells divide only when instructed to do so by other cells in their vicinity. Appropriate control of cell reproduction is carefully regulated by complex pathways that converge on the cell cycle (Figures 1.5, 1.6). 1. Cell Cycle The cell cycle (Figure 1.5) is a repetitive sequence of interphase, mitosis and return to interphase. It consists of 4 phases: Gap 1 (Gl), S (synthesis), Gap 2 (G2) and M (mitosis). Together Gl , S and G2 represent interphase and M represents mitosis. During Gap 1 phase, the cell increases in size and performs structural and synthetic functions in preparation for duplicating its DNA. The length of Gl phase is variable and its length determines the length of the cell cycle. In contrast, the length of time spent in phases S, G2 and M is constant and irrespective of the rate of cell production. The synthesis (S) phase requires about 6 hours during which time the DNA is duplicated in the nucleus. During Gap 2 phase, the components required for mitosis are assembled, requiring 2-4 hours. During the one-hour long M (mitosis) phase, the nuclear membrane breaks down, mitotic spindles form and the duplicated chromosomes are evenly separated into two newly formed daughter cells. Immediately after M phase, the daughters may rejoin the cell cycle in Gl and pass rapidly through to M phase again. Cells with a long life span typically enter a GO phase which is a nonproliferative phase in which they perform their normal functions for variable lengths of time (days or years) but are able to return to Gl as required (eg. Alberts et al, 1989; Atkinson and White, 1992). 97 2. Control of Cell Cycle The cell cycle is controlled by a system of checks and balances that safeguards against disordered growth. This system includes a network of intrinsic cellular mechanisms and extrinsic factors including "social" contacts between cells, cells and the ECM, and growth factors. A brief review is provided below and in Figure 1.6, Table 1.5 and Table 1.6 because alterations in these control mechanisms distinguish normal cells from their malignant counterparts. a. Intrinsic Mechanisms (i). Restriction Point Two classes of genes control the cell cycle: proto-oncogenes which promote cell division (Table 1.5) and tumour suppressor genes such as p53 and Rb (retinoblastoma) which suppress proliferation (Table 1.6). Proto-oncogenes are part of the normal vertebrate cell genome and they correspond to their counterpart oncogenes which are mutated tumour-promoting genes. Proto-oncogenes encode for growth factors such epidermal growth factor (EGF), growth factor receptors (R) such EGFR which is coded by gene c-erbB, cytoplasmic signalling proteins such as the ras family proteins, and nuclear DNA-binding transcription-activator proteins such as c-myc (reviewed by Hollywood and Lemoine, 1992, Weinberg 1996). Simplistically, the control of cell division can be depicted as a switch whereby, depending upon the signals received by the cell, the signal for division can be switched from "off to "on" (Weinberg 1996). A crucial switch or control point in the cell cycle occurs late in Gl at the Restriction Point which, once passed, commits cells to move into S, G2 and M phases. The Restriction Point is controlled by a protein, pRb, which is a product of tumour suppressor gene Rb. pRb is a powerful growth inhibitor and acts like a "master brake". When pRb is in its unphosphorylated form it sequesters the transcription factors necessary for proliferation and consequently, it acts like a "brake" actively inhibiting cell division. However, two cytoplasmic proteins, cyclins (cyclin D initially) and cyclin-dependent kinases (CDKs), can form a complex which phosphorylates pRb, 98 thereby releasing the "brake" and the sequestered transcription factors, allowing the cell to pass the restriction point. The levels of cyclins are strictly controlled. There are at least 8 distinct cyclin genes (eg. Bcl-1 for cyclin D) but cyclin and CDK levels are also controlled by other proteins and their genes such as pl5, pl6 and p21 which, in turn, are controlled by tumour suppressor gene, gene p53, and its protein p53 (Alberts et al., 1989; Atkinson and White 1992; Weinberg 1996). (ii) . p53 Gene and Protein p53 is an important tumour suppressor gene that is essential for normal cell growth. p53 has been described as a "molecular policeman", monitoring the integrity of the genome. Its protein product, wild-type p53 protein is normally present at very low levels in all normal cells and tissues. p53 protein has a role in regulating the transcription of genes that suppress cell proliferation and effect passage from late Gl to S phase. p53 also functions as a Gl checkpoint control because when cellular DNA is damaged, p53 protein levels significantly increase and arrest the cell cycle in Gl so that DNA can be repaired prior to cell division. However, if DNA repair fails, then p53 may trigger the programmed death of the cell called apoptosis (Raybaud-Diogene et al., 1996; Weinberg 1996). p53 protein has the ability to adopt two different conformations and it is therefore is a multifunctional protein. As noted above, the wild-type protein has a suppressor function for cell proliferation. In contrast, its mutant form has a promoting effect and therefore has been the focus of investigation in carcinogenesis, particularly of head and neck SCCs (Section V; eg. Warnakulasuriya and Johnson, 1994; Raybaud-Diogene et al., 1996; Weinberg 1996). (iii) . myc Genes The myc genes code for transcription factors that activate other growth-promoting genes. Therefore, myc genes have important roles in the maintenance of the proliferative state and the blockade of differentiation pathways. In particular, the c-myc gene controls exit from the cell cycle to the GO resting state, and the level of c-myc determines whether a cell will continue or cease proliferation. Transient repression of c-myc results in rapid disappearance of its protein which 99 signals the cell to enter GO; conversely, sustained levels of c-myc and its protein force continued cycles of proliferation (Hollywood and Lemoine, 1992). (iv). Apoptosis Apoptosis is the programmed death of a cell and is best demonstrated by the terminal differentiation process of keratinocytes.. Apoptosis also acts as a back-up system to destroy cells whose essential components such as DNA are irreparable or, cells whose control systems become deregulated by an oncogene or by the disabling of a suppressor gene. Although apoptosis is undesirable for the affected cell, it is beneficial for the tissue and body as a whole. Apoptosis is a metabolically-active process, often associated by the active synthesis of new proteins (Ansari and Hall, 1992). Under normal circumstances, apoptosis is triggered by p53 protein. A second gene, bcl-2, codes for an intracellular membrane protein which modulates apoptosis by inhibiting the effects of p53, and thereby protects the cell by preventing the triggering of apoptosis too easily (Duke et al., 1996; Atulaetal., 1996). (V). Telomeres The number of divisions that a stem cell and its daughters undergoes is tightly constrained and at some point, further cell division is no longer possible and the cell becomes senescent. The intracellular defense mechanism guarding against runaway proliferation may rely on telomeres which "count" and limit the total number of cell divisions. Telomeres have been compared to the plastic tip on the ends of shoelaces (Greider and Blackburn, 1996) in that they are a cap of special DNA sequences at the end of each chromosome. With each cell division, a small number of telomere sequences are lost, shortening the telomere, and when the telomere shrinks below a threshold length, the cells become senescent and attempts at further cell division result in cell death (Greider and Blackburn, 1996; Weinberg 1996). Abnormalities of cellular 'counting mechanisms' may underlie diseases such as psoriasis in which 100 there are increased numbers of divisions (Ansari and Hall 1992). Tumours also demonstrate uncontrolled proliferation and in some types, the tumour cells are able to repair or replace their telomeres by activating a gene that encodes for the enzyme telomerase. This enzyme is absent in most normal cells but is present in most tumour cells. Telomerase maintains the telomeres, enabling the tumour cells to divide endlessly and allowing the tumour to increase in size. Unfortunately, it also permits the pre- or already-cancerous cells to acquire additional mutations that may further increase their ability to proliferate, invade and metastasize (Greider and Blackburn, 1996). b. Extrinsic Mechanisms External or environmental regulation is crucial to the control of cell division and includes cell contact with its extracellular matrix (anchorage dependence), contact with neighbouring cells (contact inhibition) and the influence of growth factors (reviewed by Ansari and Hall, 1992 and Templeton and Weinberg, 1995). (i). Regulation by Cell-Matrix Interactions In order to divide, healthy cells require contact with a substrate and this phenomenon is called anchorage dependence (eg. Folkman and Moscona, 1978). Cell contact with the substratum maintains the organization and cohesiveness of a tissue, and normal cells that become detached from their surroundings cannot proliferate. Different cell types have different requirements for the amount or size of substratum necessary to promote cell growth (Maroudas 1973a, b), but anchorage is required in order for cyclin levels to rise to levels sufficient for the cell to pass through the restriction point in Gl of the Cell Cycle. Once the restriction point is passed, contact with the ECM is no longer required, and cells typically lose contact and their ability to divide (reviewed by Ruoslahti 1996). Anchorage dependence is mediated by integrins which couple contacts with the ECM to the 101 organization of the cell's cytoskeleton which, in turn, is linked to signal transduction ("outside-in" signals), (Hynes 1992; Haas and Plow, 1994). Integrins are transmembrane cell receptors, each type specific for a component of the ECM such as fibronectin, laminin or different types of collagen, but the affinity of an integrin for its ligand can also be influenced by cytoplasmic events ("inside-out" signals), (Hynes 1992; Horwitz and Thiery, 1994; Horwitz 1997). Interactions between the ECM and its integrin function like a "molecular address system" (Ruoslahti 1996) by directing and maintaining cells in their proper place in a tissue and in the body, and integrins may also play an important role in metastasis which is reviewed in Section V. Division of basal stem cells is usually asymmetric (eg. Alberts et al., 1989) because only one of the daughters remains a stem cell and the other daughter losses contact with the basal lamina, is forced to differentiate and ultimately die (apoptosis). Early in the differentiation of keratinocytes, there is a decreased expression of genes encoding certain integrins and ECM proteins (eg. fibronectin, laminin) and this down-regulation may be involved in the migration of cells from the basal to suprabasal layers. Significantly, this early stage of differentiation is not sensitive to retinoids (Jetten et al., 1994) which are involved in epithelial/connective tissue interactions across the basement membrane by influencing ECM synthesis, metalloproteinases and TJJVIP. (ii). Regulation by Cell-Cell Interactions The division of a basal stem cell can be asymmetric or it can result in two identical daughter stem cells by simple duplication of the stem cell. The number of stem cells and their growth pattern is strictly controlled and the fate of the daughters can be affected by the relationships of the stem cell to the other cells in the tissue. For example, denudation of the epithelium from the connective tissue stimulates undamaged adjacent epithelial cells from the stratum basale to migrate from the wound margins into the wound. While some basal cells are migrating, other basal cells a short distance back from the leading edge undergo simple duplication to restore the stem cell population. Once cells from different margins make contact across the wound, they stop their migration and 102 this behaviour is called contact inhibition of movement. Once a confluent sheet of cells has been re-established, asymmetric division restores the differentiated suprabasal layers. However, simple duplication also ceases, a phenomenon known as density-dependent inhibition of cell division or contact inhibition of cell division (eg. Odland and Ross, 1968; Lavelle 1975; Alberts et al. 1989; Atkinson an White, 1992). Cell proliferation is also regulated by the differentiated cell population to which the dividing cells belong. For example, the rate of production of keratinocytes is regulated by the thickness of the epithelium in that the presence of the outer differentiated layers appears to exert an inhibitory influence upon stem cells in the basal layer. If the normal epithelial thickness is reduced by stripping away just the surface differentiated layers, the rate of mitosis in the basal layers is increased; once the normal thickness is restored, the rate declines to normal, suggesting that there are signalling pathways from the superficial layers to the basal stem cells (Alberts et al., 1989; Atkinson and White, 1992). (iii). Regulation by Growth Factors Growth factors act as local chemical regulators, either stimulating or inhibiting cell division and differentiation, and most cell types probably depend upon a specific combination of growth factors rather than a single factor. Growth factors are present in blood and extracellular fluid and they are also bound to components of the ECM where they play a key regulatory role in tissue organization (Alberts et al., 1989). The binding of a growth factor to its specific receptor at the cell's surface triggers a cascade of biochemical changes called secondary-messenger systems. These systems relay information from the cell's surface to its nucleus where the response, appropriate to the surface signal, is generated (Alberts et al., 1989). Some of the internal signalling pathways switched on by growth factors may also be linked to activation of molecules that reside with integrins in focal adhesions, 103 suggesting that ECM molecules and growth factors may sometimes modulate one another's signals through convergent or intersecting pathways (Horwitz 1997). The main effect of growth factors occurs at the transition of cells from GO to Gl ; progression through the remainder of the cell cycle is largely independent of extracellular influences. Most growth factors possess both stimulatory and inhibitory growth properties, depending upon the cell type. Moreover, the same cell type may respond in different ways depending on the presence or absence of other growth factors (Nathan and Sporn, 1991). One group of growth factors with stimulatory growth effects for epithelial cells includes epidermal growth factor (EGF) which is produced by subepithelial fibroblasts (Alberts et al., 1989), and transforming growth factor a (TGFa) which is normally produced by epithelial cells (Kannan et al., 1996). EGF and TGFa interact with the same receptor, the epidermal growth factor receptor (EGFR), which is encoded by proto-oncogene c-erbB. Under normal conditions, TGFa is seen in the basal and suprabasal epithelial layers, and EGFR is found in high concentrations in proliferating basal keratinocytes. Binding of growth factors to EGFR results in signal transduction via second messenger systems and results in subsequent cell division (Alberts et al., 1989; Kannan et al., 1996). In contrast, transforming growth factor p (TGFp) inhibits the growth of most epithelial cells. TGFp inhibits the transcription of c-myc; as a result, c-myc protein levels are reduced and the cell is unable to proceed to Gl (Hollywood and Lemoine, 1992). As well, TGFp regulates the activity of gene pl5 whose protein product inhibits the activity of cyclin-dependent kinases (Weinberg 1996). TGFp also regulates the expression of receptors for EGF, promotes synthesis of collagens and TIMP but depresses synthesis of metalloproteinases (Ansari and Hall, 1992); these latter actions of TGFp are similar to those of retinoids but it is unclear whether the similarity is due to converging signalling pathways, or whether the action of retinoids are mediated via synthesis and 104 secretion of TGFp (Gudas et al., 1994). Civ). Regulation by Subepithelial Connective Tissue Connective tissue fibroblasts elaborate EGF which stimulates keratinocyte proliferation, and the interaction between the connective tissue and epithelium in relation to differentiation (keratin expression) and turnover of the ECM has already been discussed (Section IV.C.4). Further support for the role of connective tissue in regulating epithelial differentiation and proliferation stems from alterations observed in the subepithelial connective tissue in association with premalignancy and carcinoma discussed in Section V. 105 Class of Keratin Filaments A. Type of Keratin and Distribution Primary Keratins - Simple Primary Keratins - Stratified Basal cells Type I (acidic) K18 K14 Type II (basic/neutral) K8 K5 Differentiation-Specific Simple Epithelium Stratified Epithelium Cornified - suprabasal K19 K10 K7 KI Noncornified basal Noncornified suprabasal K19 K13 K4 Fast Turnover-Stratified Cornified - suprabasal Noncornified - suprabasal K16 K16 K6 K6 B. Type of Epithelium All Stratified Epithelia K5, K14 Keratin Markers primary markers in basal compartments Cornified Oral Epithelium basal layer K5, K14 suprabasal: KI, K10 differentiation markers K6, K16 fast turnover markers Noncornified Oral Epithelium basal layer K5, K14 and K19 suprabasal K4, K13 differentiation markers K6, K16 fast turnover markers Table 1.3. Distribution and Classes of Keratin Filaments in Oral Mucosa. (A) . Type of keratin (simple, differentiation, or fast turnover) and distribution in simple or stratified epithelia (cornified or noncornified, basal or suprabasal), and class of keratin (type I or Type II). (B) . Type of epithelium (cornified or cornified) and keratin markers in basal or suprabasal layers. Adapted from Morgan and Su, 1994. 106 A. in vitro Effects of Different Retinoid Acid Concentrations Low [RA] -decreased expression of markers of non cornification (K13, K19) -increased expression of markers of cornification (profillagrin, KI) High [RA] -increased expression of markers of noncornification -decreased expression of markers of cornification B. in vivo Effects of Retinoic Acid Suppresses -expression of cornified phenotype by decreasing -expression of involucrin . -number of tonofilaments -number of desmosomes -expression/secretion of collagenase by keratinocytes -expression/secretion of metalloproteinases (eg. collagenase) by fibroblasts -production of type I collagen and fibronectin by fibroblasts Enhances -gap junctional proteins and intercellular communication which suppress cell growth -expression of TGFp which also suppresses growth (Table 1.6) -expression of receptor for epidermal growth factor (EGFR) -expression/secretion of laminin by keratinocytes -expression/secretion of TIMP by fibroblasts -production of type I collagen and fibronectin by fibroblasts Table 1.4. Summary of Some Effects of Retinoic Acid on Oral Epithelium. A. Some in vitro effects of high and low concentrations of retinoic acid. B. Some in vivo effects of Retinoic Acid. 107 Figure 1.5. The Cell Cycle. The Cell Cycle consists of mitosis (cell division) and interphase which includes Gl (Gap 1), S (synthesis) and G2 (gap 2) phases. The Cell Cycle begins with Gl . G l -is of variable length and its length determines the length of the Cell Cycle -cell enlarges and synthesizes proteins in preparation for DNA duplication and cell division -includes the Restriction Point which is a control point that once passed, commits the cell to completion of the cell cycle S -DNA duplication in the nucleus G2 -cell prepares for division M -the nuclear membrane breaks down, mitotic spindles form and duplicated chromosomes are separated (nuclear division) -two new daughter cells are formed (cytoplasmic division) GO -shortly after completion of M phase, cells may re-enter Gl or exit the Cell Cycle to GO which is a nonproliferative phase of variable length 108 T CN o T 00 O CO to •J g 'o cu c o CO OH en. o •4—> TJ a CO O cO CO 3 CO £-3 p £ - - C L i-i. P i - . O PH ! CO ft PH 'a c^Tl ' • • ;j_ X Si CD ' CO PH-£H - , 13 3 > O co O J o u u o PH g ^ o <+H o co 13 •E o CO e r j a ft u — S 3 § U & S - a ^ « * £ HH ra is --H • - f H CO CD to o _ 2 ^ J=l 3-3 M O ,co M e - a _, ts CO c fn ft CO ft co +J O U <ft .2 CO CO -J3 TJ <H-H 'as i J P « co _, ft1 ft 5 <-> * § | CO co > CO 2P U .3 -° o u-a u 2 .— ft P 1) 8 ^ c j f l tJ c co 13 t3 w - co U C ft c = 5 ^ T 3 co.Sf^ § 3 ^  ^ * 6b' co co J J ' — 1 . S o — « M 1 - 1 OH 3 g 5 ®d § ^ 2 W)TJ © " •<§ 13 ^ -3 *>• - .3 . O > r H CO CO ft co "jg C3 Q . co 57 'ft bO ° C c ^ •2;a 109 Proto-oncogenes -Bcl-1 gene codes for cyclin D which is involved in the phosphorylation of protein Rb which is required for passage through the Restriction Point -Bcl-2 gene codes for intracellular membrane protein that modulates apoptosis and prevents triggering of apoptosis too easily -c-erb B gene codes for receptor for epidermal growth factor (EGFR) -myc genes code for transcription factors that activate other growth-promoting genes -high levels of c-myc protein sustain cell cycle; low levels signal cell to enter GO -ras family of genes codes for proteins which are involved in coupling growth factor receptors to effector proteins in the cell Cell-matrix Interactions -anchorage dependence ensures adequate cyclin levels for passage through Restriction Point Growth Factors -EGF produced by subepithelial fibroblasts }both interact with EGFR which results in -TGFa produced by epithelial cells cell division of basal cells Table 1.5. Some Factors that Enhance Proliferation of Epithelial Cells. 110 Tumour suppressor genes Rb: -Rb protein acts as "master brake" controlling the Restriciton Point by sequestering the transcription factors necessary for progression through the Cell Cycle -must be phosphorylated by cyclin/kinase complex to release transcription factors and permit entry to Cell Cycle p53: -protein p53 regulates transciption of genes that suppress cell proliferation and allow passage from Gl to S phase -arrests cell in Gl to permit repair of damaged DNA -triggers apoptosis genes pl5, pl6, p21: -the protein products of these genes inhibit cyclin-dependent kinases which are required for passage through Gl -these genes are controlled by p53 protein -gene pl5 activity is also increased by TGFp which has an suppressive effect on proliferation Telomeres -shrinkage below threshold level result in cell senescence Contact inhibition of cell division -simple cell duplication ceases once a confluent sheet of cells is re-established after wounding of the basal layer of epithelium Growth Factors -TGFp -inhibits transcription of c-myc so that cells are unable to proceed to Gl -regulates expression of EGFR -increases activity of gene pl5 whose protein inhibits cyclin-dependent kinase -promotes synthesis of collagen and TIMP -depresses syntheses of metalloproteinases Retinoic Acid Receptors -expression of different RA nuclear receptor isoforms can modulate effects of retinoic acid Table 1.6. Some Factors that Suppress Proliferation of Epithelial Cells. I l l V. Carcinogenesis Neoplasia occurs due to changes in gene expression and gene control which allows the altered cells to proliferate faster than cells in the surrounding tissue from which they arose, and thereby produce a mass of abnormal new tissue called a neoplasm. Depending upon their destructive effects, neoplasms are classified as either benign or malignant. Benign neoplasms do not metastasise. They are typically slow growing and non invasive although ameloblastomas (Regezi and Sciubba, 1989) are an exception. Benign tumours are sharply demarcated from adjacent normal tissue and cells comprising the benign neoplasm are usually indistinguishable from those of their tissue of origin. Benign neoplasms typically cause damage by the pressure they exert on the adjacent tissues and by occluding local blood supply (Atkinson and White, 1992). Malignant neoplasms are characterized by unregulated cell division and by abnormal cell differentiation. The appearance of malignant cells usually differs significantly from that of their normal counterparts in that they may be poorly differentiated (anaplastic) and so irregular in shape and size (pleomorphic) that it is difficult to recognize their tissue of origin. Transformed cells have escaped the controls that regulate orderly tissue growth and differentiation. Malignant cells lose their normal intercellular contacts such as desmosomes, and gap junctions are reduced or absent; their content of actin filaments is increased as is their secretion of proteolytic enzymes. Malignant cells no longer exhibit contact inhibition of movement or contact inhibition of cell division; they are anchorage independent and proliferate largely independent of control by external growth factors. Their high rate of proliferation increases demand for blood-borne nutrients and rapidly growing, poorly vascularized neoplasms may suffer necrosis of their central portions. Consequently, most tumours larger than several millimeters in diameter, produce angiogenic growth factors to promote neovascularization (reviewed by Templeton and Weinberg, 1995). Neovascularization is closely correlated to metastasis (Section V.E.) in which the transformed cells breach their basement membrane, penetrate the endothelium and enter the blood or lymph circulation where they must survive and evade immune surveillance. The malignant cells must also escape from the circulation 112 and finally implant and proliferate in foreign tissue to form secondary neoplasms (reviewed by Templeton and Weinberg, 1995 and Ruoslahti 1996). A. Models of Transformation Several mechanisms for malignant transformation are possible and significantly, more than one mechanism may be induced by the same carcinogenic agent (reviewed by Eversole 1993). 1. Monoclonal Transformation Haematopoietic tumours and possibly some oral tumours exhibit evidence of clonality in which all of the malignant cells are descendents of the single mutant cell. For oral/pharyngeal SCC, the model of monoclonal transformation predicts that a single keratinocyte is transformed to a malignant phenotype and subsequent simple duplication expands the malignant clone (Eversole 1993). 2. Polyclonal Transformation and Field Cancerization If several different basal keratinocytes are transformed either simultaneously or metachronously, their transformation results in the expansion of many different clones that may be anatomically widespread. Independent foci of malignant clones may progress to independent primary tumours or, several foci may eventually coalesce to produce a single large tumour (Eversole 1993). Slaughter et al. (1953) applied the term "field cancerization" to findings of two or more independent oral and upper respiratory SSCs. They reported the incidence of field cancerization as 11.2 % based on 43 of 88 patients who presented with two separate primary tumours of the same anatomical area in the oral cavity (Slaughter et al., 1953). Histologic premalignant changes (eg. Ogden et al., 1991) and biomarkers of malignancy (eg. Regezi et al., 1995; Yan et al., 1996; Scully and Field, 1997) have also been detected in the oral mucosa of patients with malignancies at a distant oral site, suggesting that oral cancer exerts a 113 regional effect upon normal oral mucosa. That is, a whole tissue region repeatedly exposed to carcinogens (eg. alcohol, tobacco) is at increased risk for developing multiple foci of malignant change (Slaughter et al., 1953). Field cancerization may also explain the local recurrence of oral cancer whereby pre-existing multicenters of premalignancy or malignancy persist at varying distances outside the area of treatment for the first premalignancy (eg. Mincer et al., 1972) or first primary SCC (Slaughter et al., 1953). 3. Contiguous Transformation Hypothetically, an infectious transforming agent such as a virus could be transmitted laterally or horizontally, from one basal cell to the next, so that contiguous cells become transformed. This mechanism could offer another explanation for the occurrence of anatomically diffuse longitudinal spans of malignancy (reviewed by Eversole 1993). B. Tumour Initiation, Promotion and Progression The transformation of cells from normalcy to full malignancy is a long, multistep process in which cells pass through a series of stages including initiation, promotion and progression. As a result, tumours are a heterogeneous, evolving population of cells; subpopulations of cells that have a selective growth advantage expand, acquire additional mutations, and ultimately dominate the tumour cell population (eg. Alberts et al., 1989; van Popple 1993; Vogelstein and Kinzler, 1993). Carcinogenesis is initiated with a heritable, irreversible genetic change in the cell's DNA sequence (Alberts et al., 1989; van Popple 1993; Vogelstein and Kinzler, 1993), so that the expression or function of genes or their products, or both, is altered (Scully and Field, 1997). Tumour initiators include chemical, viral or physical carcinogens. Chemical carcinogens typically cause simple local changes in the nucleotide sequence, and viruses introduce foreign DNA. Ionizing radiation acts through the production of reactive hydroxyl radicals (in aqueous media) which attack the ring structure of guanine in DNA, resulting in chromosome breaks and translocations (Alberts et al., 114 1989; Lawley 1994), (see also Section VII.E.2.a). Specific chemicals are designated as genotoxic when they or their metabolites react with DNA to induce mutations, or to generate reactive oxygen or hydroxyradicals which interfere with DNA repair. Usually, the altered or damaged DNA is repaired; however, if the rate of cell division is rapid, such as in growing organisms, the chances for successful repair are decreased (Alberts et al., 1989; Lawley 1994). In addition, there may be an impaired ability to repair DNA damage, which in some cases occurs as an inherited trait (Scully and Field, 1997). The altered DNA undergoes further changes during replication so that proto-oncogenes and suppressor genes may be affected. A proto-oncogene may be converted to an oncogene by several mechanisms but typically, only at a limited number of sites within the gene. The gene may be altered by a simple point mutation (change in a single nucleotide pair such as AT to CG), insertion or deletion of nucleotide pairs, chromosome rearrangement or translocation, or by insertion of additional, foreign genetic material such as from a virus (Alberts et al., 1989; Weinburger and Williams, 1995). If these gene changes occur in an area coding for a protein, the activity of the protein may be affected (eg. p53 protein); if these changes occur in control regions, the gene may be simply overexpressed (increased transcription), (Alberts et al., 1989). The gene may also amplified due to anomalies during chromosome replication which result in multiple copies of the same gene (Alberts et al., 1989). Particular genes are affected by specific abnormalities in response to particular carcinogens. For example, members of the myc gene family are typically over-expressed or . amplified (Weinburger and Williams, 1995; Scully and Field, 1997). The ras gene can be altered in only 3 of its 188 codons (the triplet of nucleotides that specifies one amino acid), and in specific types of skin tumours, an 'A' to 'T' point-mutation occurs (Greenblatt et al., 1994; Weinburger and Williams, 1995). For tumour suppressor gene p53, mutations (predominantly base substitutions) in 95 of its 353 codons can cause inactivation; tumour suppressor gene Rb is usually affected by deletions or mutations and its inactivation requires involvement of both alleles (Lawley 1994; Weinburger and Williams, 1995). 115 The altered genotype is usually not fully expressed unless a tumour promoter is present. Tumour promotion is considered reversible and involves an epigenetic change which is a change in the pattern of gene expression without a change in the DNA sequence. In order to express their cancer-associated genes, cells with abnormal DNA must divide and develop, which will determine the rate of cancer expression and the cancer risk. Promoters alter the balance between the growth of normal cells and initiated cells by providing a growth stimulus to cells with the altered genotype, favouring their clonal expansion and development (Lawley 1994; Weisburger and Williams, 1995). As the numbers of initiated cells increases, the probability of additional mutagenic events which may confer malignancy also increases because rates of mutation appear to be higher for already-initiated cells. For example, cells with inactivated p53 genes are genetically less stable and accumulate mutations and chromosomal rearrangements faster, leading to rapid selection of malignant clones (Lawley 1994). Promoters exert their effects through a variety of mechanisms that includes disturbing gap junctions and intercellular communication, and enhancing intracellular signalling pathways for growth stimuli. To be effective, promoters must be present at high levels for a long time but fortunately, their effects are usually tissue specific, reversible, and not permanent (Alberts et al., 1989; reviewed by van Popple 1993; Weinburger and Williams, 1995). A single somatic mutation is insufficient to cause cancer. An estimated three to seven independent genetic and epigenetic changes are required before slightly abnormal cells evolve into the invasive end-stage cancer cell (Vogelstein and Kinzler, 1993). For those cancers that have a discernible etiology, there is almost always a long delay between the causal event (initiation) and the onset of clinical disease. During this period of tumour progression, the prospective cancer cell(s) that are already favoured for clonal expansion, acquire additional mutations that may further enhance their ability to proliferate. Eventually a cell may accumulate sufficient genetic mutations to render it malignant (Alberts et al., 1989). Some genetic changes may be primary or essential to the 116 malignant process, whereas others may appear during tumour progression and be secondary (Scully and Field, 1997). Some sites on chromosomes are particularly liable to mutations and are termed "fragile" sites; these fragile sites are increased in frequency in individuals who smoke or chew tobacco (Scully and Field, 1997). C. Molecular Basis and Biomarkers for Squamous Cell Carcinoma of the Head and Neck Neoplastic cell populations are characterized by relatively uncontrolled proliferation that is the result of specific gene mutations, amplifications or losses of one or more of the cell cycle genes. For example, persistent mitogenic signalling results from the activation of several proto-oncogenes and overproduction of growth factors and their receptors (eg. Ras, c-myc, EGF/c-erbB; see Tables 1.5, 1.6 in Section IV.); a cell may also become less responsive to negative growth factors (eg. TGFp) with the loss of tumour suppressor genes (eg. p53, Rb). Not all combinations of oncogenes will transform cells; if both oncogenes act through the same pathway, the cell may maintain adequate control of the cell cycle. There are also synergistic oncogene combinations such as c-myc and bcl-2 which commonly occur in lymphoma. Overexpression of c-myc is associated with neoplastic growth but under conditions of poor vascularity which reduce available nutrients and growth factors, cells overexpressing c-myc undergo apoptosis. If these cells also overexpress bcl-2, they are rescued from apoptosis due to its effects on p53 (Vogelstein and Kinzler, 1993), (Table 1.5 Section IV). A mutation that creates an oncogene may have an effect on cell behaviour even in the presence of a normal copy of the corresponding proto-oncogene but typically, an oncogene exerts its cancer-provoking effect only if function of tumour suppressor genes is already perturbed (Alberts et al., 1989). In most epithelial cells, neoplastic pathways are guarded by suppressor genes rather than oncogenes (Vogelstein and Kinzler, 1993) so that cancer may be explained in terms of the loss of a tumour suppressor gene rather than acquisition of an oncogene. Normal DNA has two copies or 117 alleles of every sequence, but specific regions in a chromosome may be lost or deleted, resulting in allelic loss (Scully and Field, 1997). Tumour suppressor genes usually lose their function upon loss of both wild-type alleles; usually one allele is lost, followed by a missense mutation in the other (substitution of a single nucleotide pair), (Vogelstein and Kinzler, 1993). For example, in the hereditary form of retinoblastoma, multiple independent tumours affect both eyes. These children have a predisposition for the disease because one copy of their Rb gene is abnormal; a single somatic mutation in the remaining normal gene of even one cell is sufficient to initiate a cancer. In contrast, the nonhereditary form of retinoblastoma involves a single tumour in only one eye. The condition is very rare because two somatic mutations must occur in a single retinal cell in order to destroy both copies of the Rb gene (Alberts et al., 1989). Mutations in tumour suppressor gene p53 may operate by a different mechanism which is discussed below (Section V.C.I.). DNA is a receptor for carcinogens and specific covalent chemical bonds or products called adducts identify the interaction between the carcinogen and DNA. Adducts can be identified and quantified by special physical/chemical analyses, and different carcinogens produce a "fingerprint" that is linked to the mutational activity of that agent (Lawley 1994; Greenblatt et al., 1994). Adducts may be the first stage in cancer initiation, causing DNA polymerase to misread the base pairing due to altered hydrogen bonding properties of a base that contains an adduct. Adducts may also render the bases unreadable and "stall" the replication process (Greenblatt et al., 1994). The study of adducts, specific gene mutations and other biomarkers (see below) in the population forms the basis of molecular epidemiology (Perera 1993) in which standard epidemiological principles of investigational design and statistical analysis are applied to detection of adducts and biomarkers (Section II), (Greenblatt et al., 1994). The expectation is that adducts and biomarkers can be incorporated into cohort and case-control studies to identify causal relationships between exposure to a specific agent and the increased risk of malignancy. Ultimately, certain adducts and biomarkers may replace clinical disease as an endpoint, and thereby provide an earlier outcome so that those individuals at highest potential risk can be identified for intervention and follow-up 118 (Perera 1993). Abnormal karyotypes (chromosomes) are seen in 50-96% of oral/pharyngeal SCCs (Burkhardt 1996) and although all chromosomes may be involved, the most frequently altered chromosomes encode for some of the well-known oncogenes and tumour suppressor genes (Table 1.7), (reviewed by Burkhardt 1996; Scholes and Field, 1996; Scully and Field, 1997). Chromosomal alterations include the loss or gain of whole chromosomes, rearrangements of chromosomes such as inversions or translocations, deletions or amplifications (Scully and Field, 1997). Whereas a consistent chromosome abnormality common to all SCCs of the head and neck has not been identified, genetic aberrations involve, in order of frequency, chromosomes 3, 9, 11, 13 and 17 (Table 1.7; Scully and Field 1997). Allelic losses on chromosomes 3, 9 and 17 are seen as early events in SCC of the head and neck; but losses may be seen on additional chromosomes, consistent with a generalised increased rate of DNA replication errors and defective repair during tumour progression. Significantly, it is the accumulation of genetic changes and not the sequence of genetic changes that determines progression to malignancy (Scully and Field, 1997). If alterations in specific genes and in the protein products they control reflect a different stage or mechanism in carcinogenesis, then changes in gene/protein expression and function may serve as biological markers of malignancy that ultimately replace clinical malignancy as an endpoint in diagnosis (Perera 1993; Slootweg 1996). However, the mere association of a biomarker with malignancy cannot establish whether the altered biomarker is the cause or result of the malignant transformation (see also Sections HE and II.H). To serve as a reliable diagnostic tool, a biomarker should (1) be present in cells that do not demonstrate conventional histological signs of malignancy but are already at risk for transformation as shown by the subsequent development of dysplasia or carcinoma and (2) be useful in identifying dysplasia at risk for progressing to invasive SCC (Slootweg 1996). 119 A brief overview of potential biomarkers in association with oral/pharyngeal premalignancy and/or oral/pharyngeal SCC is presented below and in Table 1.8 (see also Tables 1.3-1.6 in Section IV). Unfortunately, much of the literature is contradictory and confusing due to the lack of uniformity in the techniques used to assess the changes in either the gene or protein biomarker (Greenblatt et al., 1994; Slootweg 1996; Dowell and Ogden 1996; Raybaud-Diogene et al., 1996). For example, the diversity of immunohistochemical (IHC) techniques, in epitope specificities of the antibodies employed, in antigen retrieval, and counting methods has failed to establish whether or not the overexpression of p53 protein is associated with increased cell proliferation, aneuploidy, tumour grade or stage, metastasis, response to chemo or radiotherapy, recurrence, or survival time (reviewed by Greenblatt et al., 1994; Slootweg 1996; Raybaud-Diogene et al., 1996). Until a standardized quantitative analysis of gene/protein expression and function can be linked to histomorphology, the use of biomarkers for diagnostic and prognostic assessment of dysplasia or SCC is debatable (Slootweg 1996; Dowell and Ogden 1996). 1. Tumour Suppressor Gene p53 and Protein p53 a. p53 Protein Normally, p53 protein functions as a checkpoint control in Gl , preventing duplication of damaged DNA either by facilitating its repair by blocking entry to S phase, or initiating apoptosis. The normal function of p53 protein can be abolished with loss of one p53 allele and a mutation in the second allele. However, the mutant protein from one mutated allele, despite retention of the second wild-type allele, can block function of the wild-type protein and may be actively involved in cell transformation (Raybaud et al., 1996). Moreover, wild type protein has a short half-life of only 20-30 minutes whereas the mutant conformationally-altered protein has a prolonged half-life up to 24 hours (Hollywood and Lemoine, 1992), allowing accumulation to levels detectable by IHC. Since wild-type protein levels are not usually detectable, the detection of p53 protein is assumed to indicate overexpression and is used as a surrogate marker for the presence of a p53 gene mutation or deletion (eg. Sauter et al., 1992; Greenblatt et al., 1994; Dowell and Ogden, 120 1996). However, the possibility that p53 expression is a result rather than the direct cause of cell proliferation has not been established (Warnakulasuriya and Johnson, 1994). Cells expressing mutant p53 protein are not blocked at Gl/S phase of the cell cycle and cannot undergo DNA repair or enter apoptosis. Consequently, the mutant p53 protein can function as an oncogene promoting cell proliferation and conferring a growth advantage to p53-positive neoplasms (Gopalakrishnan et al., 1997), but tumour growth could also be the result of lengthened survival by the inhibition of apoptosis of transformed cells, rather than a stimulation of cellular proliferation (Riva et al., 1995). p53 protein has been reported in 11% to 80% of oral SCCs; it has also been identified in adjacent areas of dysplastic and normal epithelium (eg. Regezi et al, 1995; Yan et al., 1996), in lichen planus and in chronic inflammation (reviewed by Slootweg 1996; Raybaud et al., 1996; and Warnakulasuriya and Johnson, 1996). In specific regard to SCC of the tongue, overexpression of protein p53 has been associated witinmproved survival, including stage IV tumours (Sauter et al., 1992) but has failed in predicting angiogenesis or lymph node metastasis (Leedy et al., 1994). b. p53 Gene Mutations in p53 gene have been described as an early event in SCCs of the head and neck (Greenblatt et al., 1994; Scully and Field, 1997). p53 gene mutations have been found in histologically normal epithelium at significant distances from the primary tumour, and they may be detected early during transition from normal to dysplastic epithelium (Greenblatt et al., 1994; Raybaud et al., 1996; Slootweg 1996). By sequencing p53 gene, 27% to 60% of oral dysplastic lesions demonstrate p53 mutations which precede histologic changes and are proportional to the change of dysplasia (reviewed by Scully and Field, 1997). As compared to the frequency of p53 mutations observed in noninvasive lesions (19% reported by Boyle et al., 1993), there is also an increased frequency of mutations in invasive carcinomas (43%, Boyle et al., 1993). Moreover, a 121 majority (64%) of the mutations were changes associated with carcinogens from cigarette smoke (guanine adducts), (Boyle et al., 1993). Several studies demonstrated an increased frequency of gene p53 mutations in smokers as compared to nonsmokers with head and neck SCC and premalignant lesions (reviewed by Slootweg 1996), and in one study p53 mutations were observed only in premalignant lesions of tobacco smokers (Lazarus et al., 1995). In smokers the p53 mutations are widespread throughout the gene whereas in non-smokers with similar clinical diagnoses, the spectrum of mutations are limited to sites in the gene that were characteristically seen with spontaneous mutations (Slootweg 1996). For users of betel quid, the data regarding gene p53 mutations are contradictory and may reflect different chewing habits (reviewed by Slootweg 1996). The prevalence of p53 mutations also varies geographically: only 7% of oral carcinomas from the Indian subcontinent demonstrate p53 mutations compared to 47% to 81% of oral carcinomas from Europe and USA. Geographic differences are also seen in ras mutations (see No. 3 below) and may also be attributed to oral carcinomas with different etiologies (reviewed by Paterson et al., 1996). The association of human papillomavirus (HPV) with alterations in p53 gene is also controversial and although HPV -16 and HPV-18 appear to be risk factors (Section V.F.6) associated with oral SCC, their mode of operation is unclear. Possible mechanisms include the binding and degradation of p53 protein by HPV proteins and the co-expression of HPV-DNA and mutant p53 (reviewed by Slootweg 1996). p53 mutations have been equivocally associated with angiogenesis (Leedy et al., 1994; Slootweg 1996) and although mutations in p53 gene can occur at different sites within the gene, these sites are maintained during tumour progression and metastasis (Burns et al., 1994; Sakata 1996; reviewed by Slootweg 1996; Raybaud et al., 1996). Consequently, the use of p53-mutant specific probes may be a useful tool in detecting tumour cells that are not visualised by conventional histopathology, and may offer a method for distinguishing multiple primary tumours from recurrent or metastatic tumours (Greenblatt et al., 1994; Slootweg 1996). 122 The subdivision of head and neck tumours by anatomic region of origin demonstrates differing patterns of p53 mutations among the various primary mucosal sites (Greenblatt et al., 1994). In laryngeal and pharyngeal primary SCCS, the mutation prevalence is low (34%) and the spectrum of mutations resembles that of lung cancer (predominantly G:C to T:A transitions). In contrast, primary tumours of the oral cavity have a high prevalence of mutations (81%) and a different spectrum of mutations (G:C to A:T, and A:T to G:C transitions as well as deletions and insertions), (Greenblatt et al., 1994). Greenblatt et al. (1994) concluded that the grouping together of anatomic sites (eg. as oral/pharyngeal or head/neck cancers) when addressing carcinogenesis and therapy, ignored mutational differences which could reflect differences in carcinogen exposure, variation in carcinogen activation, or variation in DNA repair between distinct mucosal sites. In specific regard to SCC of the tongue, genetic changes in p53 gene have been correlated to tumour size and histological differentiation (Atula et al., 1996), and metastasis to lymph nodes (Burns et al., 1994). 2. c-erbB Gene, Epidermal Growth Factor Receptor and its Growth Factors Both TGFa and EGF are growth factors promoting epithelial proliferation (Table 1.5, Section IV) and they compete for binding to EGFR (eg. Scully and Field, 1997). Levels of EGF are often increased in premalignant oral lesions and malignant oral SCC (Kannan et al., 1996). TGFa is usually expressed in low levels in the normal adult but may be increased in SCC (Sauter et al., 1992). In malignant tissues, TGFa appeared increased relative to levels of EGF (Kannan et al., 1996), but TGFa levels were either not significantly related to (Sauter et al., 1992) or correlated inversely (Scully and Field, 1997) with EGFR expression. TGFa has been implicated in abnormal cell growth and induction of anchorage independence (Kannan et al., 1996). TGFa produced by an epithelial cell may act in autocrine fashion by acting on autologous receptors as well as in paracrine fashion by acting on receptors of adjacent cells. Thus, these growth loops may increase proliferation and simultaneously provide a stimulus for progression to malignancy (Kannan et al., 123 1996). The receptor for EGF (EGFR) is often overexpressed in basal cells as well as other cell layers with a tendency for greater EGFR expression in poorly differentiated carcinomas, and strong EGFR expression in oral SCC has also been associated with a short survival time (reviewed by Burkhardt 1996; Warnakulasuriya and Johnson, 1996). Increased EGFR levels do not necessarily correlate to amplification of the c-erbB gene although a progressive increase in c-erbB expression is noted in normal, hyperplastic, dysplastic and malignant human oral mucosa, particularly at the time of early invasion, with increased tumour burden, and with more aggressive tumour behaviour (reviewed by Burkhardt 1996; Warnakulasuriya and Johnson, 1996). However, expression of EGFR and c-erbB in precancerous oral lesions and oral SCC are too inconsistent (Burkhardt 1996; Scully and Field, 1997) to be utilized as standard diagnostic procedures, and correlations with prognosis are not established (Burkhardt 1996). In specific regard to SCC of the tongue, Sauter et al. (1992), reported that EGFR was overexpressed in 60% of SCCs of the BOT (20 patients) but no statistically significant trends were observed in relation to survival. In addition, 35% of the BOT SCCs demonstrated overexpression of TGFa; and overexpression of TGFa in patients with stage IV tumours may be related to a poorer prognosis (p=0.1, Sauter et al., 1992). 3. ras Oncogene Family Three closely-related genes of the ras family encode proteins which are involved in coupling growth factor receptors to effector proteins in the cell. Mutant ras protein is involved in prolonging the signal for cell proliferation and its occurrence is often paralleled by EGFR expression (Alberts et al., 1989; Burkhardt 1996). Several studies have demonstrated geographical differences in ras gene changes in oral SCC which may reflect differences in etiology (Burkhardt 1996; Scully and Field, 1997). For example, ras amplification was demonstrated in 100% of SCCs associated with 124 tobacco chewing in India, 18% of SCCs from betel quid chewers in Taiwan and in 66% of buccal mucosa SCC in elderly Japanese. In the United States, ras changes were observed in 32-68% of primary oral carcinomas and were associated with increased tumour size and later stages of disease (reviewed by Burkhardt 1996). Oral carcinomas from India and South East Asia are generally characterized by the absence of p53 mutations and the involvement of ras oncogenes including ras mutations (35% frequency), loss of ras heterozygosity (30%), ras amplification (28%) as well as myc amplification (29%); in contrast, ras and myc changes are uncommon in the West (reviewed Paterson et al., 1996). Overall, overexpression of ras proteins may be important in the early stages of malignancy but the relationship of ras mutation to etiology and stage of disease is not established (reviewed by Burkhardt 1996). 4. myc Oncogene Family The myc family of genes encodes for transcription factors that activate other growth-promoting genes, and in particular, high levels of c-myc protein maintain cell proliferation, c-myc protein has been identified in early oral dysplasias with increased prevalence as the degree of atypia increased (Eversole 1993), and most studies support the overexpression of the c-myc gene in oral carcinomas (Burkhardt 1996; Scully and Field, 1997). Amplification of c-myc may indicate an increased metastatic potential and poor prognosis but further studies are required for establishing a prognostic link (reviewed by Burkhardt 1996; Warnakulasuriya and Johnson, 1996). 5. Bcl-1 and -2 Oncogenes Bcl-1 codes for cyclin Dl , a stimulatory component of the cell cycle. The amplification of Bcl-1 was demonstrated in 35% -48% of head and neck SCCs, and was seen more often in poorly differentiated, aggressive tumours associated with a poor prognosis (reviewed by Burkhardt 1996; Warnakulasuriya and Johnson, 1996). 125 Bcl-2 inhibits apoptosis; its protein product is not evident in basal cells of normal oral mucosa but it may be expressed in hyperplastic lesions like leukoplakia where it delays terminal differentiation in epithelial cells with subsequent hyperkeratosis (reviewed by Burkhardt 1996). In regards to SCC of the tongue, bcl-2 expression was detected in only 16% of samples. No expression was found in tissue of non-smokers but there was a correlation between smoking and bcl-2 expression which seemed to predict more aggressive tumour behaviour and a poorer prognosis (Atula et al., 1996). 6. TGF(3 Growth Factor and Attachment Receptors TGFp is produced by both normal and transformed keratinocytes. TGFp is involved in the regulation of ECM proteins and enzymes that modify the matrix by promoting synthesis of collagens and TIMP but depressing synthesis of metalloproteinases (Table 1.6, Section IV). TGFp also has an inhibitory effect on keratinocyte proliferation by regulating the expression of receptors for EGF. Malignant epithelial cells may become refractory to inhibition by TGFp (Burkhardt 1996); loss of this negative regulation may be linked to the invasion of stroma by the epithelium due to elaboration of metalloproteinases which degrade collagen in the basement membranes, and by increased keratinocyte proliferation (Gudas et al., 1994). Invasion does not always connote metastatic potential but alterations in integrin expression by transformed cells have been linked to a subset of head and neck carcinomas with a high risk for recurrence (Eversole 1993). The loss or reduction in cadherin expression has also been correlated to detachment and separation of cells from the primary tumour and to cell invasion and metastasis in SCC of the head and neck (reviewed by Kinsella et al., 1994). 7. Retinoic Acid Receptors and Keratin Markers Retinoids may also be involved in the invasion process because during the progression of cells from premalignant to malignant, the expression of different RARs may change (reviewed by Love and Gudas, 1994), (Tables 1.4 and 1.6 in Section IV). For example, relative to normal control 126 mucosa and adjacent normal and hyperplastic tissues, only minor changes were detected in the expression RAR-y mRNA and RAR-oc mRNA in head and neck carcinomas (Lotan 1994; Xu et al., 1994). In contrast, the expression of RAR-p mRNA was reduced or lost in carcinomas and dysplastic tissues (Lotan 1994; Xu et al., 1994). Moreover, the decreased expression of RAR-p in premalignant lesions (eg. leukoplakia) in patients without SCC suggested that decreased RAR-p expression may be an early event in head and neck carcinogenesis and may indicate a high risk for tumour development (Xu et al., 1994). Significantly, treatment with retinoic acid can increase expression of RAR-p which has been correlated with clinical response (Lotan 1994). In normal keratinocytes, RAR-p is linked to the expression of K19 and several investigations (eg. Lindberg and Rheinwald, 1989; Hu et al., 1991; Heyden et al., 1992) have suggested that K19 as well other keratins (Section IV., Table 3) would be useful indicators of oral premalignancy and SCC. However, reports (reviewed by Morgan and Su, 1994; Su et al., 1996) are equivocal and findings are summarized in Table 1.9A, B. In general, the potential value of keratins for diagnosing early tumours is limited by the observation that there is no keratin marker that is present in all malignant lesions, but is not present in normal oral mucosa (Ogden 1997). Nevertheless, Ogden et al. (1994) used keratin profiles in combination with DNA profiles (see 8. DNA Content below) of exfoliative cytology as a potential screening test for the early detection of oral SCC in high-risk communities. In combination with an abnormal DNA content, K8 and K19 were strongly associated with malignancy (Ogden et al., 1994), (see also Section II.H.2). To date, the most widely used application of keratin markers in the diagnostic field has been with anaplastic tumours where they may aid in clarifying the tissue of origin. As noted previously, keratins represent only two of six major classes of intermediate filaments. Fortunately, intermediate filaments are a constant feature of all carcinomas, irrespective of their degree of differentiation, and they are usually sufficiently conserved in anaplasia. Consequently, intermediate filaments serve as tissue-specific markers in distinguishing epithelial 127 tumours from nonepithelial tumours, and in distinguishing different types of epithelial tumours (Moll 1987, Morgan and Su, 1994). 8. DNA Content Normal germ cells are haploid and contain 23 chromosomes; somatic cells are diploid and contain 23 pairs of chromosomes. In contrast, malignant cells are often aneuploid because they have an abnormal number of chromosomes that is not a multiple of 23 (see also Burkhardt 1996; Scully and Field, 1997). DNA content can be quantitated by flow cytometry in which a fluorochrome, bound stoichiometrically to DNA, emits light directly proportional to the DNA content. Flow cytometry is often used in conjunction with other markers such as c-myc expression, p53 protein, EGFR, keratins (Ogden et al., 1994), etc. but the use of flow cytometry as a prognostic indicator is not established (reviewed by Warnakulasuriya and Johnson, 1996). Some studies of SCC of the tongue (Saito et al., 1994; Baretton et al., 1995) reported no relationship between ploidy, prognosis, survival or recurrence of tongue SCC. Other studies of tongue and larynx SCC reported moderate correlation between ploidy and tumour grade, stage and survival where diploid tumours were well-differentiated, were a lower grade and had a better prognosis than aneuploid, poorly differentiated higher-grade tumours (Gandour-Edwards et al., 1994) . Flow cytometry also measures the fraction of cells in the S-phase; in SCC of the tongue, no correlation between percent S-phase cells and histological grade was found but a high percent S-phase fraction was associated with increased clinical stage and nodal involvement (Monasebian et al., 1994). Aneuploidy rates of 42% (Saito et al.,1994) and 50% (Baretton et al., 1995) were reported for SCC of the tongue, with tongue carcinomas being diploid more often than other oral SCCs (Baretton et al., 1995). There was also a higher incidence of cervical lymph node metastasis in nondiploid cases than in the diploid cases (Saito et al., 1994; Baretton et al., 1995; King et al., 1995) . 128 During carcinogenic damage to DNA, chromosome and DNA-containing chromatin fragments called micronuclei may be created in proliferating cells. Many abnormally-prohferating cells contain extranuclear micronuclei that can be quantitated to provide information on tissue- or organ-specific abnormalities (Lippman et al., 1990; Pillai et al., 1992). For example, in the aerodigestive tract, micronuclei are formed in the basal cells that migrate to the surface where they are exfoliated. Hence, micronuclei can be analyzed in samples obtained noninvasively such as from mucosal brushings. The presence and frequency of micronuclei can quantitatively reflect ongoing DNA damage. A high frequency of micronuclei has correlated well with cancer risk in high-risk individuals such as smokers and patients with premalignant lesions, but correlated inconsistently with clinical response to chemopreventive agents (Chapter 5). Consequently, the use of micronuclei as an intermediate endpoint marker of carcinogenesis was considered premature (Lippman et al., 1990; Pillai et al., 1992). D. Tumour Kinetics Malignancy is associated with the loss or abnormal expression of normal differentiation pathways or the expression of new differentiation pathways. In addition, the genetic material of malignant cells is more unstable and demonstrates increased mutation rates as compared to normal cells. As a tumour progresses, mutations accumulate and the blocks in differentiation may become progressively more complete resulting in both genotypic and phenotypic heterogeneities (eg. Alberts et al., 1989). Characteristically, the tumour becomes less differentiated or more anaplastic and more difficult to treat. In addition, there may be important differences in cell proliferation throughout a tumour and between primary tumours and metastatic growths (Ansari and Hall, 1992). Malignant tumours are not masses of rapidly dividing cells; instead, only some cells in the tumour will be cycling in the cell cycle (Section IV.D.2; Figure 1.5)', and the majority will be in GO, incapable of dividing (sterile), or dead. In most instances, there are no consistent, significant 129 differences in the durations of the S, G2 and M phases of the cell cycle between normal and malignant cells that could be exploited therapeutically (Fleming et al., 1995). The most variable and longest phase is Gl which can range from 2-3 hours to several days, and the time between mitoses (intermitotic time) for most normal human cells is 1-2 days compared to the intermitotic time of 2-3 days for most malignant cells. The proportion of cells in the cell cycle and synthesizing DNA at a particular time can be determined by exposing the cells to an isotope (autoradiography) and calculating the labelling index (LI); most solid tumours have a LI of 1% - 8% whereas epithelium of the normal gut has a LI of 16%. If the duration of the S phase and intermitotic time of the tumour cells are also known, then the fraction of proliferating cells in the tumour or its growth fraction can be determined (Fleming et al., 1995). During the early preclinical phases of a tumour, the rate at which cells are produced and the rate at which cells are lost from the tumour are proportional to the number of cells in the tumour at a given time. Cell production exceeds the rate of cell loss and for a large part of its life, a tumour grows exponentially so that by the time it is detected clinically, the tumour has already doubled in size about 30 times and has a mass of 109 cells or about 1 gram (Gregory 1992; Fleming et al., 1995). In general, the more anaplastic the tumour, the higher the growth fraction but at the clinical stage, many tumours have a low growth fraction because the growth rate has begun to decline exponentially, eventually resulting in a maximum volume of tumour. The exact mechanisms underlying this growth pattern are unclear but as the tumour increases in size, a high percentage of the daughter cells die, probably because of inherent genetic instabilities of the transformed cells, ischemic necrosis due to tumour growth outstripping the vascular supply, decreased availability of nutrients and hormones, and accumulation of toxic metabolites (Gregory 1992; Fleming et al., 1995). The growth pattern of a tumour throughout its life span can be described by a mathematical model and plotted as an asymmetric sigmoidal curve call the Gompertzian growth curve (Gregory 1992; 130 Fleming et al., 1995). By the time tumours are detected clinically, they are 'high' on the Gompertzian curve where the growth fractions are low. The rationale for treatment of malignancies by radio or chemotherapy relies on Gompertzian growth curves because these therapies target primarily cycling cells, and the responsiveness of a tumour to treatment is dependent on the point in its growth curve at which therapy is initiated. Thus, the aim of initial or inductive therapy is to increase tumour susceptibility to therapy by increasing its growth fraction, i.e. moving the tumour to a lower point on the Gompertzian curve. Surgical debulking or initial radio or chemo therapy will reduce the number of cells in the tumour; consequently, cells previously in GO will re-enter the cell cycle so the growth fraction and growth rate of the tumour are increased which increases the tumour's susceptibility to therapy (Gregory 1992; Fleming et al., 1995), (see also Section VILE). 1. Implications for Radio and Chemo Therapy The guiding principles of cancer therapy are based on the observation that a given dose or course of chemo or radio therapy will kill a constant fraction or proportion of the cell population, rather than a constant number of cells. Hence, multiple courses of therapy are needed to eradicate the tumour because tumour cell regrowth occurs between each treatment dose, and small changes in treatment dose translate into large changes in cell survival. To increase the cell kill at a given dose, the duration of therapy exposure must be prolonged to allow more cells to enter the susceptible phase of the cell cycle (characteristically the S or M phase; see Section VILE). Ideally, the intensity, frequency and duration of radio or chemo therapy should be matched to the tumour's growth rate and to the point reached in the tumour's growth curve (Gregory 1992; Fleming et al., 1995). As early tumours are generally rapidly-growing and demonstrate an exponential growth pattern, cancer therapy produces a greater fractional cell-kill to fast-growing tumours than to slow-growing tumours (Gregory 1992). Thus, the more rapid the growth rate, the more intensive, frequent and short-lived should be the therapy. In addition, for rapidly-growing tumours, therapy should be initiated as soon as possible because even a short delay could permit a large increase in 131 tumour size with a severe reduction in chances of a cure. In general, large tumours are slowly growing and therefore less responsive to therapy. For such tumours, initial therapy may be aimed at eliminating the dividing cells which may be only a small fraction of the tumour. If the tumour is reduced to a smaller size, the tumour is moved to a lower point on the Gompertzian growth curve which may result in a smaller, faster-growing tumour that offers new possibilities for treatment and cure (Gregory 1992). It was initially assumed that radiation therapy and many chemotherapies killed malignant cells directly by disrupting their DNA but in fact, DNA is damaged to a relatively minor extent such that the DNA could be easily repaired (Weinberg 1996). It was also assumed that since p53 was involved in the cellular response to DNA damage, switching off DNA replication and allowing extra time to repair, that loss of p53 function in malignant cells was responsible for the relative success of conventional radio and chemotherapies. That is, tumour cells with altered or absent p53 were expected to be more susceptible to the killing effects of DNA-damaging agents because their genome was unstable, they had a reduced Gl delay, and they failed to repair DNA damage before attempting to replicate through the damage (Murnane and Schwartz, 1993; Lawley 1994). Recently, it has been understood that radiation exposure causes wild-type p53 protein to be stabilised, leading to elevated p53 protein levels and associated increases in transcription of p53-responsive genes (including apoptosis-inducing genes) which results in the induction of growth arrest and apoptosis (Lane 1993; Wilson et al., 1995). In addition, cells with DNA damage as a result of irradiation mistakenly perceive that the inflicted damage to their DNA cannot be repaired easily, and therefore cells with normal p53 undergo apoptosis. These discoveries imply that cancer cells with p53 mutations may be able to evade therapy-induced apoptosis, and may be far less responsive to radiation treatment than cells homozygous for normal p53 (Greenblatt et al., 1994; Weinberg 1996). In fact, "gene dosage" (homo or heterozygosity for mutated or absent p53) was clearly related to resistance to radiation-induced apoptosis in in vitro (thymocyte, Clarke et al., 132 1993; Lowe et al., 1993) and in vivo (mouse) models (reviewed by Lane 1993). However, two separate studies used IHC to determine p53 protein levels of human head and neck cancers but neither demonstrated a correlation with tumour response to chemotherapy (Riva et al., 1995), radiotherapy (Riva et al., 1995; Wilson et al., 1995), or to long term survival (Riva et al., 1995). Although the Gl checkpoint control by p53 may play a minor role in determining radiation sensitivity of head and neck cancers (Murnane and Schwartz, 1993; Wilson et al., 1995), the effectiveness of existing radiation and chemotherapy treatments may nevertheless be improved if therapies could restore a cell's capacity for apoptosis (Weinberg 1996). Cancer therapies are directed against cycling cells but radio or chemotherapy can not selectively target tumour cells with the exclusion of normal cells that are also proliferating. Moreover, there may be no intrinsic differences in the radio or chemosensitivity of normal and malignant cells (Gregory 1992; Fleming et al., 1995); consequently, the initial direct cytotoxic effects of therapy are manifested by rapidly proliferating cells such as normal oral lining epithelium. Proliferation rates in some oral lining tissues may be 1.5 to 5 times greater than in masticatory mucosa (Section IV.B) and these differences are reflected clinically in the rapid appearance of therapy-induced mucositis of noncornified mucosa. Mucositis represents ulceration or breakdown of the epithelium which occurs because cell division in the damaged proliferative basal compartment can no longer match or replace desquamation at the surface. The basement membrane region may also break down and blisters may form so that the complete thickness of epithelium is lost. There is also an acute vascular response and edema due to damage sustained by the endothelium, and damage to salivary glands reduces salivary volumes and its protective functions (Baker 1982; Squier 1990). At a minimum, mucositis is painful and diminishes the quality of life, but it may compromise the patient's nutritional status, may lead to interruptions in therapy, and may allow pathogenic organisms to gain entry and thereby facilitate secondary systemic infections, particularly in individuals treated with systemic chemotherapy which also causes myelosuppression (Baker 1982; Squier 1990; reviewed by Epstein 1992 and Epstein 1994). Radiotherapy also causes late tissues 133 changes which are discussed in Section VII.E.2.C. E . Invasion and Metastasis In SCC, attention is conventionally focused on malignant transformation of the keratinocyte which results in its uncontrolled clonal expansion, penetration of the basement membrane, invasion of the underlying connective tissue and finally, metastasis. Although overt histological and biomarker changes are manifest in epithelial cells, changes may also occur in the subepithelial connective tissue. Smith (1980) argued that changes in the epithelium developed due to altered influences of the subepithelial connective tissue upon the epithelium. For example, if carcinogens exerted their effect primarily upon the connective tissue, then connective tissue influences on the epithelium may be disturbed in such a way as to support malignant change. Conversely, even if malignant characteristics developed in the epithelium, invasion may be resisted by the connective tissue influences (Smith 1980). Some support for stromal interaction in development of oral carcinoma is seen in submucous fibrosis (Section VI.B.2.b.) in which an atrophic, dysplastic epithelium accompanies fibroelastic changes in the connective tissue that include increased collagen production, decreased collagen breakdown, a reduced number of fibroblasts with an altered phenotype and a variable chronic inflammatory cell infiltrate. In addition, the possible role of subepithelial fibroblasts in the modulation of epithelial differentiation and proliferation has already been discussed in Section IV. Tumour growth, invasion and metastasis involves changes in both the epithelium and connective tissue. During carcinogenesis, malignant cells produce proteolytic enzymes that degrade components of the ECM as they penetrate across the basement membrane, through connective tissue and between healthy cells. Oral SCCs have demonstrated increased collagenase production as collagen in basement membranes is degraded ahead of infiltrating tumour cells (Johnson et al., 1980). Ultrastructurally, gaps are seen in the basal lamina with basal cell pseudopodia extending into the connective tissue through the gaps. The collagen content of the underlying connective 134 tissue is reduced and appears related to an increased susceptibility to tumour development (Smith 1980). Whereas invasion of the connective tissue by transformed epithelial cells defines the progression to invasive SCC; invasion does not necessarily connote metastatic potential as evidenced by the behaviour of most ameloblastomas of the jaws and basal cell carcinoma of the skin (Eversole 1993). Metastasis requires the invasion of lymphatics and/or vascular channels and depends upon events that are separate from invasion of subepithelial connective tissue. Integrins and cadherins are involved in normal celhcell adhesion, and integrins are also involved in cell:matrix adhesion and cell motility. Many transformed cells lose some or all of their inter-cellular cadherins, display an altered array of integrins and secrete ECM molecules which facilitate motility. For example, in a subset of head and neck carcinomas with high risk for recurrence, expression of integrin 0^4 is increased and associated with cells that are highly invasive, and expression of integrin 013 is associated with increased motility (reviewed by Eversole 1993). Moreover, for tumour cells to cross a basal lamina and invade or metastasise, they must express integrins for laminin in order to adhere to the lamina, and they must also secrete collagenase type IV to digest the lamina (Alberts et al., 1989). Transformed cells have become anchorage independent so that their cyclin/kinase complex remains continually active, permitting continual passage through the restriction point. These cells have also escaped the normal "social" controls exerted by the adjacent cells; therefore, once they breach the basement membrane, they are motile within the connective tissue. Malignant cells can readily penetrate the thins walls of lymphatic vessels which carry the cells downstream to lodge in one or more local lymph nodes. Lymphatic capillaries are more permeable than blood capillaries because unlike the typical blood capillaries, lymph capillaries lack a basal lamina and are essentially tubes of endothelium (Ross and Reith, 1985). The lack of a basal lamina can be correlated to the extreme permeability of lymph capillaries although as lymph vessels become larger, their walls thicken due 135 to connective tissue and smooth muscle bundles (Ross and Reith, 1985). Lymphatic capillaries are very numerous under the epithelium of the skin and mucous membranes, including the oral cavity and pharynx (Ross and Reith, 1985). Thus, the rich lymphatics of the oral/pharyngeal region (Atkinson and White, 1992; see also Section ni.C.6), and the increased permeability of lymph capillaries relative to blood capillaries (Ross and Reith, 1985) may account for the high proportion of metastases to the cervical lymph nodes (Section VII.B). However, the malignant cells also encounter the basement membranes of small blood vessels and if they can penetrate this barrier as well as the endothelial cells lining the blood vessel, they are free in the blood circulation. Expansion of the tumour mass also relies on the tumour's ability to stimulate angiogenesis which it requires for nutritional support, dissemination of waste products and as a pathway for metastasis. In turn, the new capillary endothelium stimulates tumour cells to produce growth factors which further promote angiogenesis. The newly proliferating capillaries may also be more permeable and hence more likely than mature vessels to be penetrated by tumour cells (reviewed by Hirshberg and Buchner, 1995; Ruoslahti 1996). Once within the circulation, tumour cells must escape the immune system but often, the cells are trapped in the first capillary bed they encounter. Because all organs, other than the intestines, send their blood first to the lungs, the lungs are the most common site of metastasis, followed by the liver. However, some tumours demonstrate a striking preference for unexpected sites that cannot be explained by the pattern of circulation alone, and their metastasis may be mediated by integrins that are expressed by the tumour for specific ECM components (reviewed by Hirshberg and Buchner, 1995; Ruoslahti 1996; Horwitz 1997). For example, metastasis to the oral region is uncommon, but about 30% of oral metastases are the first sign of a metastatic process (Hirshberg and Buchner, 1995). The breast is the most common primary source of tumour metastasising to the jawbones and in particular, the posterior mandible. In contrast, the lung is the most common primary source for metastases to the oral soft tissues, particularly the attached gingiva, followed by the mobile tongue and then the base and/or the posterior border (reviewed by Hirshberg and Buchner, 1995). 136 After reaching the target organ, the tumour cells extravasate from the capillaries and depend upon local growth factors for further proliferation (reviewed by Hirshberg and Buchner, 1995; Ruoslahti 1996). After a period of growth, a tumour mass may remain dormant and remain clinically undetectable for months or years. During dormancy, cell proliferation is balanced by cell death. Inhibitors of angiogenesis may control metastatic growth but the dormant tumours nevertheless, pose a continuous risk of recurrence and metastasis (reviewed by Hirshberg and Buchner, 1995). F. Risk Factors Associated with Oral/Pharyngeal Squamous Cell Carcinoma. At most, only 5% of all cancers can be traced directly to environmental or occupational exposure; throughout the world, lifestyle and related behaviours are much more likely to cause or promote cancer development (Weinburger and Williams, 1995). These conclusions have been reached on the basis of research into organ-specific cancers in relationship to environmental conditions associated with their development. In most cancers including oral/pharyngeal SCC, the associated etiologic factors are not single chemicals, but rather, complex lifestyle-related mixtures. Lifestyle refers to self-imposed habits such as tobacco smoking or other tobacco uses, alcohol use, areca nut habits and dietary patterns (Weinburger and Williams, 1995) Chemical carcinogens may be genotoxic because they react with DNA and cause heritable mutations, and most human carcinogens are genotoxic. Carcinogens may also be epigenetic because their mechanisms of actions do not involve DNA; instead, they may be cytotoxic, cause chronic tissue injury, hormonal imbalances or immunological effects, or they may act as promoters (Weinburger and Williams, 1995). Table 1.10 lists the various risk factors and the relative risks (RR) or odds ratios (OR) associated with oral SCC ; a brief discussion is presented below and in Chapter 1, Section VI. It should be recognized that in calculating risks, many studies have grouped together all cancers of the oral cavity and/or pharynx (eg. Brugere et al., 1986; Blot et al., 1988; Winn et al., 1991; see 137 also Section n.D). An obvious advantage of this approach is the inclusion of a greater number of cases. However, this approach precludes the estimation of relative risks by site and obscures the effect that well-defined carcinogens like tobacco and alcohol may have in different sites (eg. Oreggia et al., 1991; Greenblatt et al., 1994). The apparent differences in risk calculations probably reflect inherent differences in the populations studied, but the selection of cases and controls is subject to a multitude of biases, as is defining and ascertaining exposure to the risk factors of interest (see Sections II. D and F). 1. Tobacco For oral premalignant lesions as well as oral/pharyngeal SSC, tobacco use is regarded as the predominant etiologic factor (Eversole 1993). Tobacco has been used in different forms throughout the world for centuries. Chewing tobacco and areca nut (Section VI.B.2.b.) are used in India and Southeast Asia by both males and females; smokeless tobacco (ST) is also popular with males in Scandinavia and with young males in the United States and Europe. Smoking of tobacco increased significantly with the introduction of commercially-manufactured cigarettes at the turn of the century; men became heavy users during World War I and women during World War II (Eversole 1993; Weinburger and Williams, 1995). Not surprisingly, the prevalence of oral SCC has increased among males born after 1920; the rate in women has tripled since the 1930's and for young American males, the rate has increased four-fold since 1960 (Eversole 1993), (see also Section VII.D). Use of black tobacco carries a greater risk for oropharyngeal, laryngeal, lung and bladder cancer than blond flue-cured tobacco. Black tobacco not only contains a higher concentration of carcinogens than blond tobacco but because of its higher alkalinity, black tobacco is less easily inhaled and therefore it remains in contact with the oral mucosa for a longer time (Oreggia et al., 1991). Tobacco and tobacco smoke contain a number of powerful genotoxic carcinogens such as nitrosamines, polycyclic aromatic hydrocarbons and heterocyclic amines that interact with DNA 138 (Weinburger and Williams, 1995). Nitrosamines are produced from nicotine during the curing process of tobacco, during the combustion of tobacco, and in vivo from the reaction of nitric oxides with nicotine and related alkaloids (Weinburger and Williams, 1995). There are wide differences between individuals in their ability to metabolize nitrosamines to damaging intermediates and individuals who most effectively activate these carcinogens, may be at higher risk (Hechtetal., 1994). Tobacco and its smoke also contain phenols and terpens which are not genotoxic, but they may act as cocarcinogens and promoters, activities that are highly dose dependent and reversible to some extent. Thus, their effects account for the long latency associated with smoking, the steep increase in disease risk with consumption above 20 cigarettes/day, and the decreased risk of disease upon cessation of tobacco use (Spitz 1994; Weinburger and Williams, 1995). Tobacco smoke also contains chemicals that induce enzymes; as a result, the metabolism of hormones and substances such as vitamins are increased so that smokers may have increased requirements for vitamins and other essential nutrients (Weinburger and Williams, 1995). Smoking also appears to have immunosuppressive effects in that smokers have lower levels of NK (natural killer) cell activity in their peripheral circulation than nonsmokers, and experimental exposure of animals to cigarette smoke reduces cell-mediated immunity which is associated with accelerated tumour progression (Browman et al., 1993). Furthermore, cigarette smoking increases the carboxyhemoglobin content of blood, causing a leftwards shift of the hemoglobin-oxygen dissociation curve and a relative hypoxia of the tissues. Tissue hypoxia may have a general compromising effect on tissue resistance and healing, and during radiotherapy of the tumour, smoking-induced hypoxia may interfere with the oxygen-dependent effects of radiation (Browman et al., 1993), (see also Section VII.E.2.a). a. Smoking tobacco Tobacco smoking is the major cause of cancers of the tongue, salivary gland, mouth and pharynx, 139 and tobacco use is directly attributable for about 92% of the oral cavity tumours in men and 61% of these tumours in women (Newcomb and Carbone, 1992). As shown in Table 1.10, the risks of oral/pharyngeal cancer increase with the number of cigarettes smoked per day and the duration of smoking but show little relationship to age started smoking; risks decline following cessation of smoking (Blot et al., 1988; Bundgaard et al., 1994). After adjusting for age, duration, amount smoked and alcohol use, males who smoked only filter cigarettes experienced 50% of the risk of smokers of only nonfilter cigarettes; mixed filter and nonfilter smokers experienced 80% of the risk compared to pure nonfilter smokers (Blot et al., 1988). Unfortunately, a similar trend was not observed for females; filter and mixed smokers experienced 120% and 90%, respectively, the risk of nonfilter smokers (Blot et al., 1988). Smokers of cigars and pipes tend to inhale less than to cigarette smokers but their overall risk estimates approximate those of cigarette smokers for buccal SCCs but not for pharyngeal malignancies (Spitz 1994). Risk assessments by anatomic site and tobacco smoking vary. In a large study of 1114 drinking and smoking patients with SCC of the oral cavity and/or pharynx, and 1268 population-based controls, Blot et al. (1988) found that among males, risk trends with smoking were weaker for tongue cancer (OR range 0.8-3.2) than other oral sites (OR range 1.2-5.2) or the pharynx (OR range 1.5-5.8). Pipe and cigar smoking were more closely associated with cancers of the FOM and buccal mucosa than either tongue or pharyngeal cancer. Among females, the effects of smoking were stronger for pharyngeal cancer (OR range 1.6-36.7) than oral cancer (OR range 0.8-9.7) or tongue cancer (OR range 1.3-8.1), (Blot et al., 1988). In a study of 57 males with lingual SCC and 353 controls, Oreggia et al. (1991) examined the possible risk determinants for SCC of the tongue alone, excluding other sites of the oral cavity. Current smokers had a RR of 29.4 and former smokers had a RR of 11.8 compared with nonsmokers (RR 1.0). The type of tobacco smoked was a strong determinant of risk; exclusive users of black tobacco had a RR of 4.0 140 compared with a RR of 1.8 (not significant, NS) for smokers of mixed black and blond tobacco, and a RR of 1.0 for exclusive smokers of blond tobacco (Oreggia et al., 1991). In a case-control study of northern Italian males, 102 with tongue cancer, 104 with cancer of the mouth (FOM, gingiva, retromolar trigone), and 726 healthy controls, differences between patients with cancer of the tongue and those with cancer of other oral sites, were found in the type of tobacco smoked (Franceschi et al., 1992). The adjusted odds ratio for cancer of the tongue were lower for pipe/cigar smokers (OR 3.4; NS) than for current cigarette smokers, but for cancer of the mouth, the odds ratios were higher for cigar/pipe smokers (OR 21.9) than for current cigarette smokers (OR 11.8). However, for both tongue and mouth cancers, smoking-related risks increased markedly with increased amount and duration of smoking, but smokers of high tar cigarettes had a 10-fold increased risk of tongue cancer and a 14-fold increased risk for cancer of the mouth, as compared to nonsmokers (Franceschi et al., 1992). In a case series of 150 American patients with oral SCC, 97% patients with SCC of the FOM were smokers compared to 64% of tongue-cancer patients and 50% of gingival-cancer patients (Barasch et al., 1994). The odds of smoking among persons with FOM cancer were 38 times the odds of smoking among persons with gingival cancer and 21 times the odds of smoking among persons with lingual cancer (Barasch et al., 1994). In addition to malignancy, tobacco smoking is also a risk factor for oral premalignancy (Section VI). The risk (OR) of a having a dysplastic lesion for smokers compared with nonsmokers or exsmokers for over 10 years, was calculated at 7.0 with a dose-response relationship for tobacco dependent upon the level of cigarette consumption. Moderate smoking (<20 cig/day) produced an OR of 3.7 and heavy smoking (>20 cig/day) an OR of 13.8 (Kulasegaram et al., 1995). Theoretically, marijuana is carcinogenic because its smoke is quantitatively similar to cigarette smoke but marijuana smoking results in a greater tar burden to the respiratory tract, especially 141 because of the rapid, deep inhalation used in smoking marijuana (reviewed by Spitz 1994). The potential carcinogenic effects of marijuana smoking are difficult to separate from tobacco smoking and alcohol because most abusers of marijuana also consume tobacco and alcohol (confounding, Section II. F.2), (Spitz 1994). b. Smokeless Tobacco With the exception of ST use among teenage males and professional baseball players, ST is not as popular as other tobacco products. The prevalence of ST use is only 6.1% of American males and of these, only 3.6% limit their tobacco use to ST and do not use other tobacco products (Spitz 1994). ST or snuff is carcinogenic but some forms of ST have low correlations with oral malignancy. In a recent review of the literature, Eversole (1993) reported that leukoplakia was found in 13% - 78% of ST users and most lesions correlated to the site of tobacco placement, usually the mucobuccal fold. The prevalence of leukoplakia was directly related to the duration and frequency of use and type of ST, snuff being more common than chewing tobacco (reviewed by Eversole 1993). Snuff-related lesions appeared to show significant dysplastic changes less commonly than other forms of leukoplakia and erythroplakia; carcinomas arising from snuff lesions tended to be well-differentiated, slow growing and with late metastasis (Mincer et al., 1972). Overall, the incidence of carcinoma among ST users is not high, particulary among those using snuff; only 1.4% of SCCs were related to ST use and only 7.7% of verrucous carcinoma patients were ST users (reviewed by Eversole 1993). The use of tobacco in conjunction with areca nut is reviewed in Section VI.B.2.b. 142 2. Alcohol a. Alcoholic Beverages Alcohol is considered a promoter and a possible cocarcinogen of tobacco because it increases the amounts of reactive carcinogenic metabolites from procarcinogens in tobacco by acting as a solvent for tobacco and other carcinogens (Winn et al., 1991; Newcomb and Carbone, 1992; Weinburger and Williams, 1995), and enhancing their penetration by increasing the permeability of the oral mucosa to tobacco-associated nitrosamines and polycyclic hydrocarbons (reviewed by Eversole 1993). Chronic alcohol use also induces microsomal enzymes that enhance the metabolic activation of tobacco and other carcinogens, and through this mechanism, may affect tobacco carcinogenesis at distant organs (Newcomb and Carbone, 1992). Alcohol may suppress the efficiency of DNA repair after exposure to nitrosamine compounds (Shaha and Strong, 1995), and alcoholic beverages themselves, especially hard liquors, also contain congeners and carcinogenic contaminants (Weinburger and Williams, 1995). Moreover, chronic alcohol use is often associated with nutritional deficiencies which may contribute independently to oral carcinogenesis (see below), (Blot et al., 1988). Risks associated with alcohol use increase with total alcohol consumption and the smoking-adjusted excess risk associated with high alcohol consumption among men and women is approximately 9-fold (Blot et al., 1988). The effects of alcohol are interrelated with smoking (see Section 3 below) but there is also a deleterious effect of heavy alcohol consumption among nonsmokers, indicating that tobacco is not a requisite cofactor for alcohol-related cancer (Blot et al., 1988). For example, in nonsmokers, a metabolite of alcohol, acetaldehyde, can induce cancer of the oesophagus and this risk is higher in heavy drinkers of hard liquors (Weinburger and Williams, 1995). Overall, the risks of alcohol consumption are greatest for hard liquor and beer drinkers, and least for wine drinkers, suggesting that ingredients other than alcohol are involved as beer and hard liquor contain nitrosamines and polycyclic hydrocarbons (Blot et al., 1988; Weinburger and Williams, 1995). 143 In a large study in the United States (Blot et al., 1988), the majority of patients consumed beer and hard liquor as compared to wine; there was a small decrease in risk among light and moderate wine drinkers but risks rose sharply beyond 2 drinks/day (Blot et al., 1988). In contrast, in a large study in France (Brugere et al., 1986) 92% of patients consumed wine although only 30% drank wine only; overall, average alcohol consumption exceeded amounts observed in other countries and was reflected by the large relative risk (Table 1.10; RR range 2.7-70.3). In a case-control study in Northern Italy (Franceschi et al., 1992), consumption of wine was considerably greater than consumption of beer or hard liquor and this trend was reflected in the risks for alcohol consumption and oral cancer: wine had a higher risk (OR range 1.0-6.8) than either beer (OR 1.1-NS) or hard liquor (OR 0.5-NS). Risk assessments by anatomic site and alcohol consumption also vary. After correction for tobacco smoking, risk trends with drinking for females were greatest for pharynx cancer (OR range 1.0-15.0), followed by tongue (OR range 1.0-11.0) and other oral sites (OR range 1.0 -6.6), (Blot et al., 1988). In contrast, risks for males were greatest for cancer of other oral sites (OR range 1.0 - 12.3), followed by pharynx (OR range 1.0 - 10.4) and tongue (OR range 1.0-6.0), (Blot et al., 1988). Similar risks for tongue cancer and alcohol consumption up to 200 ml/day were observed in Uruguay with RR ranging 1.5-6.8; consumption above 200 ml/day resulted in RR of 11.6 (Oreggia et al., 1991). In northern Italy, the overall risks for tongue and oral cancer were similar (OR 3.4 and 3.3 respectively) for total alcohol consumption which was predominantly wine consumption (Franceschi et al., 1992). The relationship between alcohol consumption and oral dysplasia is not as strong as between alcohol consumption and oral cancer, as Kulasegaram et al.(1995) were not able to demonstrate an increased risk of oral dysplasia among self-reported consumers of alcohol. Nevertheless, alcohol consumption was greater among cases than controls and the proportion of hard liquor drinkers amounted to 33% of cases compared to 13% of controls (Kulasegaram et al., 1995). 144 b. Alcohol-containing Mouthwashes Common brands of mouthwash contain from 1% - 30% alcohol along with flavouring and sweetening. Oral swishing with a mouthwash containing 25% ethanol may provide a local alcohol exposure to the oral mucosa that is similar to drinking a 50% ethanol beverage (Winn et al., 1991). However, Shapiro et al. (1996) argued that the ingestion of alcohol had many effects, other than topical ones, that appeared to play a role in alcohol-associated carcinogenesis. Moreover, the quantity of alcohol ingested from mouthwash use would be exceedingly low and in addition, there were major differences in other ingredients between alcoholic beverages and mouthwashes (Shapiro et al., 1996). Several studies have evaluated the association between mouthwash use and oral/pharyngeal cancer and results were inconsistent (reviewed by Shapiro et al., 1996). Winn et al. (1991) reported that risks for oropharyngeal cancer and mouthwash use were elevated only for mouthwashes with an alcohol content of 25% or higher. Risks tended to increase with early age at start of use and with increased duration of use. Adjusted risks for site and mouthwash use were similar for tongue (OR range 1.2-1.4), other oral sites (OR range 1.2-2.0) and pharynx (OR range 1.4-1.5), (Winn et al., 1991). Significantly, mouthwash use was reported more often among smokers than nonsmokers; among denture wearers versus those without dentures, and among individuals with periodontal disease than those without (Winn et al., 1991). Mouthwash may be used to mask the use of tobacco and alcohol and typically, these habits are under-reported by study participants and under-ascertained by study investigators (Section HF). Consequently, confounding by tobacco and alcohol use results in spuriously elevated odds ratios for mouthwash use and oral/pharyngeal cancer (Shapiro et al., 1996). 3. Tobacco Smoking and Alcohol It is diffic