Incorporating Biomarkers Into Cancer Epidemiology: a Matrix of Biomarker and Study Design Categories’

Vol. 4. 301-31/. June 1995 Cancer Epidemiology, Biomarkers & Prevention 301

Incorporating Biomarkers into Cancer Epidemiology: A Matrix of Biomarker and Study Design Categories’

Nathaniel Rothman,2 Walter F. Stewart, and study design implications of using hiomarkers in observational Paul A. Schulte studies are generally applicable to other exposures and diseases.

Occupational Studies Section. Environmental Epidemiology Branch. National Cancer Institute. Bethesda. Maryland 20892 IN. R.]: Department of Epidemiology. Johns Flopkins University School of Hygiene and Public Biomarker Categories Health. Baltimore. Maryland 21205 [W. F. S.J; and Screening and Notification Fig. 1 portrays the range of biomarker categories that reflect, Section. National Institute for Occupational Safety and Health. Cincinnati. directly or indirectly, the carcinogenic process due to xenobi- Ohio 45226 [P. A. Sj otic exposures (2). This well described heuristic schema in- eludes external exposure and biomarkers of internal dose, biologically effective dose, early biological effects, altered Abstract structure/function, and disease. Susceptibility, either acquired During the last decade, there has been increasing interest or inherited, is portrayed as potentially modifying the relation- in the use of biomarkers in cancer epidemiology to ship between each step in the progression from exposure to enhance exposure assessment, to gain insight into disease disease. The schema in Fig. J is a powerful, organizing, albeit mechanism, and to understand acquired or inherited one-dimensional, framework for understanding the relationship susceptibility. To facilitate the use of biomarkers in between exposure, biomarkers, and the disease process. It does health research, biomarkers have been divided into not, however, describe the progression of studies required to categories that depict the spectrum of cancer develop and validate a newly developed bioassay in human pathogenesis from exposure to disease. In this paper, we populations, nor does this schema provide insight into why, consider the epidemiological designs most suitable for the when, and how a particular biomarker can be used in a specific study of each type of marker. In particular, we present a type of epidemiological study design. In order to explore this two-dimensional matrix relating the biomarker categories process, we begin with a discussion of each study category. on one axis to four different types of activities (laboratory, transitional, and etiological studies and Study Design Categories public health applications) that develop markers and apply them in human populations. We then use the A series of different types of activities is usually carried out to matrix to review the potential application of biomarkers develop and eventually apply a biomarker in etiological studies in observational studies of cancer etiology, discussing the and in public health practice. These can be categorized as advantages, disadvantages, and logistical considerations in laboratory studies, transitional studies (where the biomarker is using biomarkers to answer research questions. the outcome variable), etiological studies (where disease, or a valid surrogate, is the outcome variable), and public health applications (Table 1). Although biomarker research proceeds Introduction in an iterative fashion and is not strictly linear as conceptualized In 1987, the Committee on Biological Markers of the National by these activities, this schema does serve to highlight the Research Council/National Academy of Sciences proposed a critical steps in the process. classification scheme based in part on previous work by Perera and Weinstein (1) that broadly categorized biomarkers into Laboratory Studies markers of exposure, effect, and susceptibility (2). The emer- Basic research in biochemistry, molecular biology, and toxi- gence of this concept has motivated discussion on the study cology attempts to elucidate causative steps in carcinogenesis design implications of the use of biomarkers in epidemiological and modulators of that process with the use of in vitro and in research (3-18). In this paper, we build on this foundation and 1,ivo models. Many biomarkers that have been applied in human provide an organizational framework for using different types studies were identified initially as components or correlates of of biomarkers in epidemiological studies and public health the tumorigenic process in animals (e.g. , DNA adducts). More applications. Biomarkers associated with xenobiotic chemical recently, molecular changes described in human tissues have carcinogens are the primary examples discussed. However, the become the basis of new assays (e.g., Refs. 19, 20). Increas- ingly, research efforts are focusing on developing biomarker assays for direct use in human studies.

Recetved 9/2/94: revtsed 2/22/95: accepted 2/14/95.

I Presented in part at the American Association for Cancer Research sponsored Transitional Studies conference, “Molecular and Biochemical Methods in Cancer Epidemiology and Transitional studies as first defined by Hulka (10, 1 1) bridge the Prevention: The Path between the Laboratory and the Population,’ Naples, gap between the development of markers in the laboratory and Florida, 1992.

2 To whom requests for reprints should be addressed, at NIH/National Cancer their application in population-based studies. They optimize Institute. EPN 4lh. bl3() Executive Boulevard. MSC 7364. Bethesda, MD 21)592. sample processing, evaluate assay accuracy and precision,

SUSCEPTIBILITY

Exposure DIsease Structurel L1111_*.FunctIon

Fig. 1. A continuum of biomarker categories reflecting the carcinogenic process resulting from xenobiotic exposures (2).

Table / Study design categories determining whether assay reliability and accuracy are accept- able, it is then important to define the optimal conditions for Laboratory studies collecting, processing, and storing biological specimens for Transitional studies eventual assay since variation in sample handling can some- Developmental times introduce large variation in assay results, making the Characterization Applied measures unsuitable for research. Characterization Studies. Biomarker characterization studies Etiological studies Observational are designed generally to address questions about the presence Case-control or levels of a newly developed marker in human populations. In Case-case addition, they serve to identify factors that are confounders or Prospective cohort effect modifiers of a marker (e.g., age, gender, and medica- Experimental tions), which need to be measured and taken into account when Screening using the marker in subsequent studies. Intervention Applied Studies. Applied studies are investigations usually Public health applications performed on healthy subjects3 exposed to particular xenobi- Risk assessment otics (e.g., occupational exposures, smokers, etc.) where the Clinical prevention biomarker is treated as the outcome variable. Most often, cross- sectional or short-term longitudinal designs (which may be

experimental in some instances, e.g. , controlled dietary modu- lation) are used. While this type of study has been used com- collect information about potential confounders and effect monly with many classes of biomarkers in cancer epidemiol- modifiers, and study early biological effects of selected expo- ogy, the following discussion applies primarily to markers of sures in generally healthy human populations. Some of these biologically effective dose and early biological effect. research activities have been referred to by other investigators At this stage of research, the biomarker has not been as validation, methodological, or mechanistic studies. While shown to predict an increase in risk of disease in humans. The these terms apply to several of the above objectives, we prefer marker may have some utility, however, if it has been shown to the more global term “transitional” since it emphasizes that reflect biological changes considered relevant to cancer patho-

these studies, while critically important, have a limited capa- genesis in animals exposed to human carcinogens (e.g. , DNA bility by themselves to provide evidence about disease risk for adducts and cytogenetic damage) or if it has been shown humans. previously to be elevated in human populations exposed to While it is not our intention to introduce new jargon, we known carcinogens (e.g., DNA adducts and chromosomal ab- have found that dividing transitional studies into three broad errations). Under these conditions, biomarkers can be used to

categories (i.e. , developmental, characterization, and applied provide mechanistic insight into well established exposure- studies) clarifies their distinctive research goals. In practice, disease relationships, to supplement suggestive but inconclu- elements of all three types of studies are often incorporated into sive evidence of carcinogenicity of a chemical from epidemi- a single field investigation. ological studies of disease, or to provide a preliminary Developmental Studies. When a promising new biomarker evaluation of compounds introduced recently into the work- emerges from the laboratory, some very basic issues need to be place or the general environment. resolved before considering its application in human studies. Applied studies are not capable, however, in and of them- The first priority in evaluating a marker for use in epidemio- selves, of establishing or refuting a causal relationship between logical studies is to determine its reliability. As long as an assay a given exposure or a given level of exposure and risk for is reliable, the ordering of subjects by the measure is preserved. developing disease. Until a marker is shown to predict disease Since this is all that is required for studying a marker-disease risk, which can be established only by comparing risk of dis- relationship, reliability and not accuracy is of initial impor- ease in individuals with and without the marker, results of tance. Reliability of laboratory assays can be assessed optimally studies using biomarkers as outcome measures are only sug- through the analysis of blind replicate human samples repre- gestive; a biomarker may by overly sensitive (i.e., it may be sentative of the range of values likely to be found in human respond to low levels of chemical exposures that are below the populations. When, however, etiological studies are used to establish regulatory policy or serve as guidelines for clinical

intervention, accuracy also becomes critical (although a “gold 3 An exception is studies that evaluate the association between biomarkers and standard” is often lacking for new molecular markers). After radiation and/or antineoplastic therapy in cancer patients.

threshold for disease risk) or insensitive, or reflect phenomena Several early prospective cohort studies collected blood that are irrelevant to the disease process. samples but banked only the serum. These studies have made substantial contributions to understanding the role of nutrient Etiological Studies intake and, to a lesser extent, infections in cancer causation. Etiological studies are distinct from transitional studies in that Fortunately, large prospective cohort studies initiated in the they involve either clinically ill subjects, asymptomatic subjects 1980s and 1990s are, in general, banking most or all fractions with early disease, or subjects positive for an intermediate of the peripheral blood sample to allow a far wider range of process shown previously to be associated with increased risk biological assays to be performed, particularly those that require DNA. of disease (e.g. , colon adenomas and risk of colon cancer). These include ecological, case-control, case-case, prospective One limitation of many large cohort studies is that re- cohort (with nested variants), family, screening, and inter- sources are often available to collect a biological sample at only vention studies. We focus our discussion on case-control, one point in time. Although this is not a concern for DNA-based case-case, and prospective cohort studies, and then comment assays of inherited susceptibility markers (which remain con- on the use of biomarkers as outcome measures in etiological stant in normal tissue), it poses some limitations for several studies. other categories of markers, particularly for short-term markers of internal dose that reflect exposures which vary from day to day. Case-Control Studies. In case-control studies, the prevalence When transient markers are used to classify individuals, misclas- of an exposure or a biomarker is compared in cases and con- sification is inevitable and is almost always nondifferential. As trols. This study design is used far more frequently than the such, type II errors will be more likely as the degree of misclas- prospective cohort study design because of its relatively greater sification increases (23, 24). efficiency and lower costs, and is logistically ideal for the study of rare diseases. As such, maximizing opportunities to integrate Etiological Studies Using Validated Biomarkers as Out- biomarkers creatively into case-control studies is important. come Measures. The use of intermediate markers (e.g. , bio- Defining which biomarker categories can be used most effec- logically effective dose and early biological effect) as outcome tively in this study design is critical to this process (12, 14) measures in epidemiological research is likely to provide sub- because some markers may be affected by disease itself, which stantial research opportunities in the future since it allows risk raises complications of reverse causality. factors for earlier stages of the disease process to be identified. Case-Case Studies. In a case-case study [Ref. 21 ; also de- From a logistical standpoint, subjects with the predisease mark- scribed as a case series study (15) and a case-only study (16)], ers are more common in the population. A disadvantage of this a series of subjects with the disease of interest are enrolled. strategy, however, is that subjects who are biomarker positive Then, tumor characteristics (at the anatomic, cellular, chromo- (“cases”) cannot be ascertained passively. Population-based somal, or molecular level) of cases with exposure to a known screening or screening of selected exposed populations (e.g., or suspected carcinogen are compared with tumor characteris- specific occupations) is required. tics of cases without a history of such exposures (15, 21). This An intermediate marker must be correlated with disease approach was used in early studies of smoking and lung cancer risk in order to be used as a valid outcome measure. The criteria histology (22) and has come into favor more recently to study for validating intermediate biomarkers have been discussed tumors subclassified at the molecular level. As in the case- extensively by Schatzkin et a!. (8) and include the sensitivity of control study, the odds ratio is used as a measure of association the marker (i.e., the proportion of subjects who develop cancer between the exposure and specific tumor type (15). who are positive for the biomarker), the relative risk of the It should be noted, however, that the odds ratio from a association between the marker and disease, and a judgement case-case design will underestimate the odds ratio derived in a about the extent to which the exposure-disease relationship is case-control design when the exposure of interest is associated mediated through a process reflected directly or indirectly by with more than one tumor type. In this instance, the case-case the marker. Steenland et a!. (14) have discussed the logistical design cannot be used to estimate the relative risk of disease challenges of the validation process and have noted that exter- from a specific exposure. A nondiseased control group is nal exposure data, as well as biological samples, need to be needed for this purpose. collected in epidemiological studies in order to evaluate the Prospective Cohort Studies. In prospective studies, biologi- comparative ability (and interrelationship) of these measures to predict disease in humans. cal samples collected from exposed subjects while they are It may be difficult to interpret a correlation between an healthy either are analyzed at the time of collection or are banked for later analysis. The cohort is then followed forward exposure, a biomarker, and disease risk found in an observa- in time; subjects who develop disease are identified, and pre- tional study since an intermediate marker may be directly on the morbid biomarker levels in these cases are then compared with causal pathway from exposure to disease, may be a surrogate those levels in the controls. Often, a nested case-control ap- for that pathway (reflecting similar biological processes), or proach is used. Samples from cases and only a sample of the may reflect an irrelevant process that is correlated with the controls are analyzed, which reduces the laboratory require- critical pathway (8, 14). Additional data from experimental ments and costs considerably. Although this study design is by studies, including in vitro and in vivo animal studies and, under far the most time-consuming and expensive type of observa- some circumstances, human intervention trials, usually are tional epidemiological study, it is the only method available to needed to clarify these potentially complex relationships. test the biomarker-associated relative risk of cancer when the marker is transient or may be directly or indirectly affected by disease. An additional advantage of the cohort study is that it Public Health Applications can be used to test the association between a marker and risk for Biomarkers can be used at the population level, where research multiple cancer sites, as well as noncancer outcomes, thus results may be incorporated into policy making decisions, and providing a more global understanding of its public health at the individual level, where biomarkers may be used for relevance. screening purposes and ultimately in clinical practice (25).

BIOMARKER CATEGORY I- I

STUDY Externala DESIGN Exposure

Laboratory

.. 1b 5 9 13 17 21 Transitional

Etiological 2 6 10 14 18 22 Case-Control

3 7 11 15 19 23 Case-Case 4 8 12 16 20 24 Cohort Public Health Applications

Fig. 2. A matrix of hiomarker (horizontal axis) and study design (vertical axis) categories. a, evaluated by questionnaire or environmental monitoring; h. cell number keyed to respective notation in text. Susceptibility markers are depicted in only one column for graphical simplicity, placed just prior to disease. Multiple susceptibility marker columns could he used in a variety of locations in the matrix, consistent with Fig. 1.

Although a marker may not be useful for clinical screening, it categories are presented along the horizontal axis (Fig. 2). The may still have substantial utility as a research tool. vertical axis of the matrix can be used to formulate a research Results from transitional and etiological studies can be program that develops a biomarker, explores its use in human used to enhance risk assessment at the population level. For populations, and applies it in cancer etiology studies. The example, the IARC recently concluded that there is only limited horizontal axis of the matrix can be used during the planning evidence in humans for the carcinogenicity of ethylene oxide phases of an epidemiological study to evaluate the potential (26) after considering available epidemiological studies of can- contribution of each biomarker category. cer. Its overall evaluation, however, was that “Ethylene oxide is We now describe in detail some of the main applications carcinogenic to humans (Group 1)” (26), which was based in of biomarkers in human studies of cancer etiology, with a part on the observation that ethylene oxide is associated with a primary focus on transitional and etiological studies. We dis- dose-related increase in hemoglobin adducts, chromosomal ab- cuss the potential uses and limitations of each biomarker cat- errations and sister chromosome exchanges in peripheral egory in these studies and describe the role that each study lymphocytes, and micronuclei in bone marrow cells of design plays in the iterative process of cancer research, where exposed workers (26). laboratory, transitional, and etiological studies serve to extend, A biomarker may have utility to screen populations at high challenge, and ultimately reinforce each other as they reveal risk of disease as part of a primary or secondary prevention patterns of cancer causation. If desired, readers can focus on effort. However, a substantial amount of information is required particular cells of interest using the cell numbers in Fig. 2. The before a biomarker can be used for this purpose. In particular, the application of biomarkers in genetic epidemiology studies of probability that an individual will develop cancer over a defined familial cancers, screening and intervention trials, risk assess- period given a constellation of biological and nonbiological risk ment and clinical prevention, and early cancer diagnosis and factors must be estimated, along with a calculation of its uncer- treatment is outside the scope of this paper. tainty (27, 28). The enormous effort undertaken over the last few decades to evaluate serum lipid markers as screening tools for Markers of Internal Dose atherosclerotic heart disease risk is an instructive example of the Biomarker assays of internal dose are measurements of a parent extensive body of research that is required before a biomarker can compound or its metabolite in an accessible biological matrix, be used effectively for screening and in clinical practice. such as serum or urine. In some instances, they may complement or serve as an alternative to questionnaire and environ- Incorporating Biomarkers into Epidemiological Studies mental exposure data. To facilitate the optimal use of biomarkers into epidemiological Transitional Studies (Fig. 2; Cell 1). Transitional studies studies, we have constructed a two-dimensional matrix; study determine whether markers of internal dose are detectable in designs are presented along the vertical axis, and biomarker human populations, evaluate the exposure-marker relationship,

effect modifiers of that relationship (e.g., nutrition and demo- Cohort Studies (Fig. 2; Cell 4). Markers of internal dose may graphic variables), and measure marker kinetics. Both obser- be useful in prospective cohort studies since the problem of vational (e.g. , smokers and occupationally exposed popula- reverse causality is minimized (and eliminated for subjects tions) and experimental studies (e.g., dietary manipulation) diagnosed several years after providing a biological sample). have been used to evaluate marker characteristics. The utility of an internal dose marker depends in part on its There are several applications of internal dose markers in half-life, the pattern of the exposure it is measuring (e.g., transitional studies. They can determine whether a compound regular and daily exposure versus episodic and infrequent cx- present in a particular environment is absorbed and excreted, posure), and whether secular trends have occurred in that cx- which can help to establish the biological plausibility of asso- posure (e.g., smoking cessation). In addition, the information it ciations described in etiological studies (e.g. , measurement of provides must be compared to the availability and quality of aflatoxin metabolites in urine of exposed subjects; Ref. 29). other sources of data (e.g., questionnaires, environmental mon- They can serve as internal dosimeters to evaluate biomarkers itoring, and medical records). Essentially all exposure measures further downstream in the exposure-disease continuum (e.g., misclassify some subjects; it is the relative ability of different adducts and cytogenetic markers). In addition, they can be used sources of data to correctly place individuals into exposure to validate other sources of exposure data (e.g., questionnaire categories that must be considered. For example, subjects can and environmental monitoring data), even if the markers generally report average smoking habits and smoking duration themselves are not used in subsequent etiological studies (9). in an accurate manner; therefore, cumulative exposure can be Case-Control Studies (Fig. 2; Cell 2). It is reasonable to calculated. Since internal dose markers associated with tobacco assume that in most instances cumulative exposure is the single smoke have relatively short half-lives and thus reflect only best predictor of cancer risk. In general, questionnaires are used recent exposure (e.g., cotinine), they have limited utility by to obtain exposure data. In some instances, however, a measure themselves to directly assess a smoking-cancer relationship. of internal dose may be more suitable or may be a valid In contrast, it is difficult to obtain accurate information surrogate of cumulative exposure, even when blood or biopsy about dietary exposure to aflatoxin by questionnaire because material is obtained from cases and controls at the time of the exposure is sporadic and is present in a spectrum of different study. The ideal biomarker should persist over time [e.g., fat- food types (31). In this instance, even a short-term internal dose marker might be expected to classify the long-term exposure soluble substances such as 1 , 1 , 1 -trichloro-2,2-bis(p-chloro- phenyl)ethane (DDT) metabolites, which have been measured status of subjects more accurately than questionnaire data. For in case-control studies of breast cancer (30)] and not be affected example, a nested case-control study conducted in Shanghai, by disease status. Many xenobiotic compounds, however, are China, demonstrated that aflatoxin exposure, assessed by mea- metabolized to excretable substances that have relatively short suring several aflatoxin metabolites (and an N7-guanine afla- half-lives (e.g., aflatoxin metabolites; Ref. 29). toxin adduct) in banked urine samples, was associated with In general, short-term markers have limited use in case- an increased risk of HCC, while aflatoxin exposure assessed control studies. For example, the measure of aflatoxin metab- by questionnaire was not associated with elevated risk (31). olites in urine would have little value in a case-control study of HCC;4 patients are ill to the extent that recent nutritional Markers of Biologically Effective Dose patterns are unlikely to be representative of usual patterns, and Macromolecular adducts, which are considered markers of bi- the disease itself may potentially alter aflatoxin metabolism. ologically effective dose, integrate both external exposure and Biomarkers of internal dose have some utility in case-control a spectrum of processes that activate or detoxify procarcinogens studies of patients with early, asymptomatic disease since cur- and repair DNA damage (1). They can be used to evaluate the rent exposure is likely to reflect usual exposure patterns, at least ability of genotoxic compounds to form adducts in exposed in the recent past. In addition, absorption, metabolism, storage, humans and may have utility as exposure markers in etiological and excretion of these substances are less likely to be influ- studies. enced by the disease process in its early stages. Theoretically, the association between carcinogen-DNA Case-Case Studies (Fig. 2; Cell 3). In a case-case study, the adduct levels and disease risk should be stronger than the association between exposure patterns and tumor characteris- association between external exposure (or internal dose) and tics is evaluated in a series of cancer patients. Defining expo- disease. Practically, however, obtaining a sample from the sure status is an essential feature of these studies. Exposure target tissue is possible for only a few sites (e.g., hematopoietic status may be defined by exposure patterns in particular geo- and lymphatic systems, bladder, cervix, and oral cavity). While graphic areas, by questionnaire data, or potentially, by mark- more readily accessible surrogates have been developed (e.g., ers of internal dose. The case-case study has a particular peripheral WBC DNA adducts), their correlation with DNA advantage in that pathological material from the target site is adducts measured in relevant tissues should be assessed. generally available from all subjects. If a biomarker can be Transitional Studies (Fig. 2; Cell 5). Transitional studies measured in organ tissue (e.g. , organochlorine pesticide me- determine whether markers of biologically effective dose are tabolites), then marker levels in normal tissue adjacent to a detectable in the general or specific target populations, assess tumor can be compared in tumors with and without particular marker kinetics, evaluate the dose-response relationship be- histological or molecular characteristics. As described in the tween the exposure and the marker [e.g. , PAH exposure and preceding section, however, the use of these markers is adduct formation (32, 33)], and assess potential effect modifiers subject to the constraints of marker kinetics and potential of that relationship. disease influence. Since DNA adduct formation is thought to play a central role in early stages of tumorigenesis for many genotoxic agents (34, 35) and elevated adduct levels have been found in human populations exposed to several known or suspected human 4 The abbreviations used are: HCC, hepatocellular carcinoma; ASF, altered structure/function: EBE. early biological effect; PAH, polycyclic aromatic carcinogens (35), it is reasonable to assume that a population hydrocarbons. with elevated DNA adduct levels may be at higher risk of

developing cancer. To date, however, only one study has ac- exposed at the time samples were collected since adducts in tually demonstrated that DNA adducts measured in any bio- peripheral blood and urine probably reflect only recent logical matrix are associated with excess cancer risk in humans exposure (29, 37, 38). (31, 36). Until there is a greater body of evidence that supports When reliable data are not available about external expo- the ability of measured macromolecular adducts to predict sure at the time the biological sample is collected, the adduct- excess cancer risk and characterizes that risk quantitatively, the associated relative risk for cancer can be calculated, but the interpretation of studies that use adduct biomarkers as outcome relevance of interindividual variation in adduct formation can- measures will be somewhat uncertain. not be tested directly because the population cannot be stratified Case-Control Studies (Fig. 2; Cell 6). There is only a limited on exposure. Further, if this variation reflects processes that are role for the use of adduct biomarkers to assess exposure in unrelated to cancer risk, the strength of the exposure-disease case-control studies since most available markers have rela- relationship (assessed via adduct formation) may be weakened tively short half-lives and are therefore subject to the same unknowingly. If, however, a measure of internal dose is avail- limitations as many internal dose markers. For example, pe- able, then the internal dose-associated relative risk for cancer ripheral WBC PAH-DNA adducts probably reflect exposure can be compared to the adduct-associated relative risk (31). that has occurred over several weeks to months (Refs. 37, 38; although adducts in longer-lived mononuclear cells may reflect Markers of Early Biological Effects and Altered more distant exposure), while hemoglobin adducts reflect cx- Structure/Function posure that has occurred over the last 120 days, the life span of Markers of EBE and ASF represent processes that are interme- RBC. diate between exposure and disease. Early biological effect Macromolecular adducts, however, do pose one advan- markers reflect cytogenetic damage and somatic cell mutation tage over markers of internal dose; they may function as a frequency, while ASF markers reflect premalignant alterations phenotypic measure of cancer susceptibility by evaluating in cell cycle control, measured at the morphological (e.g., the tendency to activate and detoxify carcinogens and repair hyperproliferation) or at the molecular (e.g., altered oncogene DNA for subjects exposed to the compound of concern at the expression) level. While some cytogenetic markers can be time of biological sampling. For example, two studies of studied at the target site (e.g., micronuclei in exfoliated bladder PAH-DNA adducts in peripheral WBC of lung cancer pa- cells and oral/nasal mucosa), most cytogenetic and somatic cell tients and healthy controls have suggested that given similar mutation assays are performed on peripheral blood cells; as current smoking patterns, cases have a greater tendency to with macromolecular adducts, the correlation between markers form adducts than do controls (39, 40). The tendency to form in peripheral blood cells and relevant events occurring at target adducts, taking into account current exposure level, could be sites may vary by exposure category, assay type, marker kinetics, modeled as an effect modifier of the relationship between and target site. lifelong cigarette exposure (assessed by questionnaire) and Transitional Studies (Fig. 2; Cells 9, 13). Transitional studies disease. determine whether newly developed EBE or ASF markers are Case-Case Studies (Fig. 2; Cell 7). Markers of biologically detectable in the general population or in particular subgroups. effective dose may complement exposure assessment in case- These studies, performed on subjects with exposures of inter- case studies. They can be measured in blood or urine samples ests, can potentially determine early biological effects of the taken at the time of diagnosis or, potentially, in normal tissue exposure and provide insight into possible disease mechanisms. adjacent to tumor in pathological materials (e.g., PAH-DNA As in vitro, animal, and human evidence accumulates that a adducts in lung; Refs. 41, 42). As noted previously, the use of given marker reflects carcinogen-mediated damage, there is these markers is subject to the constraints of marker kinetics usually a greater tendency to consider the marker as a surrogate and potential disease influence. for disease. However, until the marker-associated relative risk Cohort Studies (Fig. 2; Cell 8). Markers of biologically ef- for disease is known, the marker cannot be used to establish fective dose may be useful in prospective cohort studies since directly an association between a given exposure and cancer the problem of reverse causality is minimized. When a macro- risk. molecular adduct is correlated highly with external exposure Case-Control Studies (Fig. 2; Cells 10, 14). Markers of EBE [e.g., aflatoxin B1 exposure and N7-guanine aflatoxin adducts in and ASF may have use in case-control studies if the markers are urine (29)1, it can be used to address the primary question of not altered by the presence of disease, are relatively stable, and whether that exposure is associated with excess cancer risk, are evaluated on patients who have not yet been treated (par- directly and/or through interactions with other exposures [e.g., ticularly by radiation therapy or chemotherapy). Certain per- risk of HCC due to both aflatoxin B1 and hepatitis B exposure sistent markers, such as stable chromosomal translocations, (31, 36)]. As noted previously, the utility of an exposure bi- potentially integrate both exposure and biological processes omarker is based on its ability to classify subjects into long- that modify exposure-mediated chromosomal damage. The in- term exposure categories relative to other available measures of terpretation of markers used in this setting is enhanced when exposure. data are available on external exposure. Marker levels and Since adduct formation in humans is often variable, ad- external exposures can be evaluated independently for their ducts may reflect information about individual susceptibility, as association with excess cancer risk and then combined in a well as exposure (1). In this instance, the association between stratified analysis. In this instance, the research goal is to adduct levels and cancer risk then becomes a test of whether determine whether, within a given level of exposure, subjects this observed variation is relevant. If high quality data are with higher marker levels have an increased cancer risk. In available on external exposure at the time of biological sample general, such investigations should be regarded as hypothesis collection (as well as on a long-term basis), then the tendency generating, to be followed up in prospective studies if feasible. to form adducts, given current exposure, can be modeled as an Case-Case Studies (Fig. 2; Cells 11, 15) Markers of EBE and effect modifier of cumulative exposure. This approach, how- ASF that persist over time and are unaffected by disease status ever, can only be used to study cases and controls who were can be compared in subgroups of cases defined by exposures

and/or tumor characteristics under the conditions described ulation and are generally considered polymorphisms (i.e., the above. minor allele frequency is >0.01), are probably associated with Cohort Studies (Fig. 2; Cells 12, 16). Prospective cohort relative risks under 10 and as such do not exhibit familial studies are the optimal observational design to incorporate EBE patterns of inheritance, and may interact with particular expo- and ASF markers (subject to the feasibility of collecting the sures (46, 47). Under these circumstances, which apply to a appropriate tissue) since the problem of reverse causality is wide spectrum of potential susceptibility genes, the case-con- minimized. These markers may provide mechanistic insight trol study (known as an association study in genetic epidemi- into the exposure-disease relationship, can be tested for their ology parlance) is an effective approach to evaluating these potential to function as disease surrogates in subsequent stud- factors (46-49). A discussion of methods used in genetic ies, and may be shown to have potential as screening tools for epidemiology studies of familial cancer, which evaluate gene primary and secondary prevention. mutations generally present at low frequencies and associated In the absence of data on external exposure, the goal of with high cancer risk, is outside the scope of this paper (for these studies is relatively straightforward: to determine whether reviews see Refs. 18 and 48). elevated marker levels are associated with overall and site- Markers of susceptibility include polymorphisms in genes specific excess risk of cancer. As noted previously, if data are responsible for chemical activation or detoxification, DNA available on potentially important external exposures (e.g., repair, and genomic stability, among other processes. Some from questionnaires, medical records, and environmental mon- categories of susceptibility genes may interact with very spe- itoring), these studies can then explore the more refined ques- cific types of chemical exposures (e.g., cytochrome P-450 tion of whether, given the same level of exposure to a particular enzyme subtypes and Phase Il-conjugating enzymes), while agent, subjects with higher biomarker levels have an elevated others may confer more general susceptibility. Markers can be cancer risk (14). measured at the DNA level (if the genetic basis of a polymor- As an example, chromosomal aberrations in peripheral phic phenotype has been identified) or the phenotype level lymphocytes were evaluated recently for their ability to predict [e.g., drug probes of hepatic enzyme activity, DNA repair excess cancer risk in two small cohort studies that analyzed measured in peripheral lymphocytes (50), in vitro susceptibility samples collected from individuals tested at various clinics over of lymphocytes for drug-induced chromosomal damage (51)]. about a 20-year period (43, 44). The study by Hagmar et a!. Transitional Studies (Fig. 2; Cell 17). Transitional studies (43), carried out in four Nordic countries, found an elevated play an important role in the study of genetic susceptibility cancer risk associated with the highest tertile of chromosomal markers in that they survey the prevalence of particular alleles aberration frequency for all sites combined, while the study by in specific racial subgroups. Further, they can evaluate the Bonassi et a!. (44), carried out in Italy, found significantly correlation between genotype and phenotype assays (52), esti- elevated risk for all cancer types and for respiratory tract tumors mate the likelihood and impact of allele misclassification (53), and lymphatic/hematopoietic tumors. These studies provide and evaluate potential induction effects (54). This information support for the hypothesis that chromosomal aberrations mea- can be used in designing subsequent etiological studies by sured in peripheral lymphocytes reflect (or are correlated with) estimating sample size needs and by identifying exposures relevant, pathogenic, exposures and/or processes and suggest (e.g., enzyme inducers such as alcohol and certain drugs) that that they may be useful as an outcome measure in subsequent must be collected via questionnaire. studies. In contrast, Hagmar et a!. (43) found that sister In addition, transitional studies can potentially evaluate the chromatid exchange and micronuclei formation were not biological plausibility of gene-environment interactions ob- associated with excess cancer risk (although there was lim- served in etiological studies. For example, the GSTMI gene ited power to test the latter marker). As additional years of encodes the cytosolic enzyme glutathione S-transferase M 1, follow-up are accumulated by this cohort, it will be inter- which conjugates activated aflatoxin metabolites. A case-con- esting to follow the relative risk of cancer associated with trol study in a geographic area with high exposure to aflatoxin each marker. demonstrated recently that subjects with the GSTMJ null gen- The studies by Hagmar et a!. (43) and Bonassi et a!. (44) otype (who lack a functional enzyme) were at increased risk of are an important first step in the process of exploring the HCC (55). To support the biological plausibility of the case- relevance of EBE biomarkers for disease risk in humans and control results, these investigators showed that the same gen- will undoubtedly be extended by studying candidate markers in otype was associated with higher levels of aflatoxin albumin ongoing large prospective cohort studies. However, the appli- adducts in healthy subjects currently exposed to aflatoxin (55). cability of particular assays will depend on the method that was Similarly, Vineis et a!. (56) have shown that among subjects used to process and store peripheral blood samples. For exam- who smoke cigarettes, the slow (versus rapid) NAT2 acetylation ple, many currently available cytogenetic and somatic cell phenotype is associated with a greater tendency to form 4-ami- mutation assays require cultured lymphocytes. Although most nobiphenyl hemoglobin adducts, particularly at lower levels of if not all recent prospective studies have stored the major blood cigarette use. This supports the observation from several case- fractions, only a few have cryopreserved lymphocytes (which control studies that slow acetylators in the general population maintains their viability) so that they can be cultured in the are at increased risk for bladder cancer (57). The underlying future. Fortunately, recently developed PCR-based somatic cell assumption of this study design is that surrogates (e.g., protein mutation assays (19, 45) can be used in studies that have stored adducts in peripheral blood) reflect similar activation and de- nonviable WBC since these assays require only DNA for toxification processes occurring at the target site (i.e., DNA analysis. adduct formation). Case-Control Studies (Fig. 2; Cell 18). The case-control de- Markers of Susceptibility sign is highly efficient for examining the role of genetic sus- There has been increasing interest in evaluating the impact of ceptibility markers, particularly when high quality exposure genetic susceptibility in cancer epidemiology. In this section, data are available from questionnaires or environmental mon- we focus on susceptibility genes that are common in the pop- itoring. Some studies have suggested that there may be subtle

exposure-gene interactions, where particular alleles are associ- date, it should be tested, if possible, on a sample of healthy ated with greater risk of cancer at lower s’ersus higher levels subjects who are exposed to various levels of the chemical. (58) of exposure to carcinogens, or vice versa (59). These Finally, case-case studies can be an efficient first step to interactions are likely to be missed if adequate attention is not determine whether a particular chemical exposure-genetic in- given to exposure assessment. There is a critical need for more teraction is present in a case-control study, particularly with a valid and reliable methods of obtaining data on dietary, envi- large number of subjects. The prevalence of a given allele can ronmental, and occupational exposure to carcinogens by ques- be evaluated and compared among a subsample of cases with tionnaire (60, 6l), given that there will be many opportunities high, middle, and low level exposure. If the frequency of the to study exposure-gene interactions in future studies. allele in each group of cases is similar, it may not be worthwhile The following examples are drawn from the study of to analyze DNA from remaining cases and all controls. It may, polymorphisms in enzymes that oxidize or conjugate chemical however, still be useful to study a subsample of controls to carcinogens. When a phenotype is fixed and determined by a determine whether the allele is associated with increased risk of known genetic allele, DNA-based assays can be used to eval- disease independent of exposure. uate cancer risk associated with a given polymorphism [e.g., Cohort Studies (Fig. 2; Cell 20). If WBC or blood clots from GSTMI (62, 63)] since disease status does not affect genomic serum tubes have been stored in prospective cohort studies, DNA DNA. The interaction of particular alleles and environmental can be extracted and analyzed for genetic polymorphisms. Clearly, exposures can then be assessed. Techniques developed recently since genotype assays can be performed in case-control studies, that use buccal swabs to obtain oral epithelial cells for PCR- they are not a rationale for performing costly and time-consuming based genomic DNA analysis are a promising cost-efficient and prospective studies. If, however, DNA samples are already avail- feasible strategy for case-control studies. They are ideally able, it may be worthwhile to evaluate relevant genetic poly- suited, in particular, to collecting samples on population-based morphisms, since most analyses require only nanogram quantities subjects. of DNA and the interaction between particular genetic polymor- When an enzyme is inducible and there is a genetically phisms and exposures ascertained prospectively can be explored. conferred tendency toward inducibility that has been identified In addition, a cohort study can test the association between a given [e.g., CYP/Al (64)], DNA-based assays can be used to measure polymorphism and risk for multiple diseases. susceptibility of a subject for induction. If the inducers of the Genotype assays potentially can be used in retrospective enzyme are known (e.g. , cigarettes), then exposure to these cohort studies if tumor samples from cases have been stored inducers can be estimated by questionnaire and modeled with and can be retrieved. In this instance, the distribution of genetic the presence or absence of the inducibility polymorphism to alleles among cases with and without the exposure of interest develop an integrated measure of enzyme activity. can be compared. In addition, a control group of living subjects When the genetic basis of a measured phenotype is un- matched to cases on ethnicity, age, and sex could be assembled known, a phenotype assay must be used [e.g., CYPJA2 (65)]. and used for comparison, assuming that the genetic make-up of Phenotype assays can be applied usually to patients with early, the target population had not changed over recent years. asymptomatic disease. The application of phenotype assays to subjects with advanced disease may, however, have more limitations; disease status may alter assay results directly, and, Markers of Disease if the phenotype is inducible, the current pattern of exposure to Markers of disease include tumor characteristics such as tumor enzyme inducers may not be representative of the usual expo- histology and the presence and nature of oncogene and tumor sure pattern. suppressor gene mutations. Although these mutations are Case-Case Studies (Fig. 2; Cell 19). The use of case-case causally associated with the cancer process itself, their char- studies to evaluate genetic markers has been discussed recently acteristics may have etiological relevance (66). by Piegorsch et a!. (16). These studies can be used to compare Transitional Studies (Fig. 2; Cell 21). Transitional studies the prevalence of a given allele among cases with different can be used to evaluate the reliability of tumor marker analysis patterns of exposure since a polymorphism that modifies a and compare various analytical techniques to each other. specific chemical-tumor (or tumor subtype) relationship may be Case-Control Studies (Fig. 2; Cell 22). Case-control studies present in a higher proportion of cases exposed to that agent can use tumor markers to separate cancer cases into more compared to cases without that exposure. Any source of homogenous categories and then determine whether particular genomic DNA can be used for analysis. Even if the study is categories have a stronger association with external exposure. limited to the analysis of archived, formalin-fixed, paraffin- For example, when Taylor et a!. (67) compared the occupa- embedded tumor samples, the presence of some alleles may be tional histories of leukemia patients with mutations in any ras measurable in normal tissue adjacent to the tumor. oncogene and population-based controls, they found an excess Case-case studies assume that the probability of a healthy risk for having worked in an a priori high-risk occupation, person being exposed to a particular compound or level of a which was not detected when all leukemia patients (both with compound is independent of the probability that they have the and without ras mutations) were analyzed together as one polymorphism under study (16). More explicitly, this assumes group. that genetic predisposition to the disease is not related to greater risk for acute or subchronic toxic reactions to the exposure. Case-Case Studies (Fig. 2; Cell 23). In case-case studies, While this is a reasonable assumption for the majority of tumor characteristics of cases with exposure to a known or chemical exposures and genetic polymorphisms described to suspected carcinogen are compared with tumor characteristics of cases without a history of such exposures (Ref. 21). For example, in the case-control study of leukemia and occupation described above (67), results similar to the case-control analysis were obtained when the occupational histories of cases with and R. Sinha, N. Rothman, E. Brown, M. Knize, S. Rossi, D. Rhodes, 0. Levander, and J. Felton. Heterocyclic amine content in red meat cooked by various tech- without ras mutations were compared directly to each other. A niques to different degrees of doneness, manuscript in preparation. similar approach has been used to study the association of

BIOMARKER CATEGORY I. I

#{231}ob0 #{231}O’0 \04 . 0

STUDY External ., DESIGN Exposure , $ v#{149}0

QuestIonnaIre MetaboIltes Macromolecular Cytogenetic Altered Oncogene GOIOfYPO Tumor Adducts Aberrations ExpressIon Markers

Case-Control High Lowe LowModerateb Low.Moderatec LowModerated High Moderate-HIgh#{176} High

Prospective High High High High High Cohort HIgh NA High

Fig. 3. Application of the matrix to evaluate the use of biomarkers in a case-control and a prospective cohort study of cigarette smoking and lung cancer. Low, Moderate, High, NA (not applicable). indicate potential use. a, current cigarette metabolite levels generally do not reflect usual or historical exposure in hospitalized, incident cases; I,, adduct levels in peripheral blood cells generally do not persist more than a few months (although adducts in longer lived mononuclear WBC may reflect more distant exposure). They may. however, serve as susceptibility markers; c, stable, persistent chromosomal aberrations in peripheral lymphocytes may provide insight into cumulative exposure and its interaction with activation and repair processes (14); d, it is difficult to determine if altered or abnormal levels of oncogene proteins found in serum reflect early or late stages in tumorigenesis: e, phenotype assays may be altered by disease state; f, it is generally not feasible to apply a phenotype assay that requires the active participation of study subjects (e.g., taking a drug probe to measure a metabolic polymorphism) to an entire cohort at enrollment.

specific chemical exposures and the presence and pattern of Table 2 Application of aflatoxin (N7-guanine) DNA adduct assay, a marker

mutations in the p53 gene [e.g. , aflatoxin exposure and codon of biologically effective dose 249 mutations in HCC (68), and cigarette use and G:C to T:A Study design Study objective” transversions, consistent with PAH exposure, in non-small cell lung cancer (69)]. Laboratory studies Develop assay for N7-guanine adduct in urine Cohort Studies (Fig. 2; Cell 24). The association of external Transitional studies exposure and risk among cases with similar tumor markers can Developmental Optimize processing urine samples for long-term storage be evaluated in prospective and retrospective cohort studies, if Characterization Detect adducts in Chinese and East African tumor samples can be retrieved. populations Applied Correlate adducts with dietary exposure

Discussion Etiological studies We believe that presenting biomarker categories and study Observational, Evaluate aflatoxin exposure and risk of hepato- prospective cellular cancer designs as a two-dimensional matrix provides an organizational Experimental, Determine if oltipraz therapy alters adduct levels framework for reviewing the literature about a particular expo- intervention sure-disease relationship and evaluating the potential applica- Public health applications tion of biomarkers in new studies. Specifically, the matrix can Risk assessment Potential to screen exposed populations and estimate be used in two ways. By using the vertical study design axis, disease risk at the group level one can summarize the development and use of a biomarker via Clinical prevention Potential to evaluate and modify risk among laboratory, transitional, and etiological studies and public individuals health applications. For example, Table 2 depicts this process ‘, Reviewed in Groopman et al. (70). for a bioassay that measures the N7-guanine aflatoxin DNA adduct in urine [reviewed by Groopman et a!. (70)]. Alternatively, when designing an epidemiological study, the horizontal biomarker axis can be used to systematically and tumor markers have the greatest application in this study, consider whether biomarkers from each category can (or while other markers have a more limited role. In contrast, the should) be incorporated into a study and to compare the utility cohort study (using a nested design for efficiency) can evaluate of a given marker in different types of studies. For example, the association of each marker category with disease, as well as Fig. 3 evaluates the use of biomarkers from each category in a the relationship between sequential markers. Markers reflecting case-control study and a prospective cohort study evaluating the later stages in lung tumorigenesis can be evaluated in samples association between cigarette smoking and lung cancer. In the from subjects who develop cancer within a relatively short case-control study, exposure assessment relies almost exclu- period of time after enrollment (i.e., several years), markers sively on questionnaire data. Inherited susceptibility markers reflecting earlier stages in carcinogenesis can be studied in

samples from subjects who develop cancer in subsequent years, 20. Bell, D. B., Liu, Y., and Cortopassi, G. A. Occurrence of bcl-2 oncogene and DNA-based markers of inherited genetic susceptibility can translocation with increased frequency in the peripheral blood of heavy smokers. J. NatI. Cancer Inst., 87: 223-224, 1995. be evaluated in all cases diagnosed since they do not vary over 21. Taylor, J. A. Oncogenes and their applications in epidemiologic studies. Am. time in healthy tissue. The efficiency of these various goals J. Epidemiol., 130: 6-13, 1989. will, however, depend on the size and age structure of the 22. Kreyberg, L. Lung cancer and tobacco smoking in Norway. Br. J. Cancer, 9: cohort. 495-510, 1955. 23. Hall, A. J., and Wild, C. P. Aflatoxin biomarkers. Lancet, 339: 1413-1414, 1992. Conclusions 24. Ross, R. K., Yu, M. C., Henderson, B. E., Yuan, J-M., Qian, G-S., Tu, J-T., We have found this matrix valuable for reviewing the literature Gao, Y-T., Wogan, G. N., and Groopman, J. D. Aflatoxin biomarkers. Lancet, and for designing our own studies. In addition, after several 340: 119, 1992. years of use, we have found it to be an effective instructional 25. Shields, P. G., and Harris, C. C. Molecular epidemiology and the genetics of environmental cancer. JAMA, 266: 681-687, 1991. tool in the classroom. We hope that it will help others consider why, when, and how to use biomarkers in their own research. 26. IARC. Ethylene oxide. In: IARC Monographs on the Evaluation of Card- nogenic Risks to Humans. Some Industrial Chemicals. Vol. 60, pp. 73-159. Lyon, France: IARC, 1994. Acknowledgments 27. Gail, M. H., Brinton, L. A., Byar, D. P., Corle, D. K., Green, S. B., Schairer, C., and Mulvihill, J. J. Projecting individualized probabilities of developing breast We thank Drs. Douglas Bell, Neil Caporaso, John Groopman, Richard Hayes, cancer for white females who are being examined annually. J. NatI. Cancer Inst., Robert Hoover, Miriam Poirier, John Potter, Mark Schiffman, Martyn Smith, Paul 81: 1879-1886, 1989. Strickland, Jack Taylor, and Sholom Wacholder for helpful discussions. 28. Benichou, J., and Gail, M. H. Methods of inference for estimates of absolute risk derived from population-based case-control studies. Biometrics, in press, 1995. References 29. Groopman, J. D., Zhu, J., Donahue, P. R., Pikul, A., Zhang, L-S., Chen, J-S., I. Perera, F. P., and Weinstein, I. B. Molecular epidemiology and carcinogen- and Wogan, G. N. Molecular dosimetry of urinary aflatoxin DNA adducts in DNA adduct detection: new approaches to studies of human cancer causation. people living in Guangxi Autonomous Region, People’s Republic of China. J. Chron. Dis., 35: 581-601), 1982. Cancer Res., 52: 45-52, 1992. 2. Committee on Biological Markers of the National Research Council. Biolog- 30. Falck, F. C., Jr., Ricci, A., Jr., Wolff, M. S., Godbold, J., and Deckers, P. ical markers in environmental health research. Environ. Health Perspect., 74: 3-9, Pesticides and polychlorinated biphenyl residues in human breast lipids and their 1987. relation to breast cancer. Arch. Environ. Health, 47: 143-146, 1992. 3. Schulte, P. A. Methodologic issues in the use of biologic markers in epide- 31. Qian, G-S., Ross, R. K., Yu, M. C., Yuan, J-M., Gao, Y-T., Henderson, B. miologic research. Am. J. Epidemiol., 126: 1006-1016, 1987. E., Wogan, G. N., and Groopman, J. D. A follow-up study of urinary markers of 4. Perera, F. P. Molecular cancer epidemiology: a new tool in cancer prevention. aflatoxin exposure and liver cancer risk in Shanghai, People’s Republic of China. J. NatI. Cancer Inst., 78: 887-898, 1987. Cancer Epidemiol., Biomarkers & Prey., 3: 3-10, 1994. 5. Hulka, B. S., and Wilcosky. 1. Biological markers in epidemiologic research. 32. Perera, F. P., Hemminki, K., Young, T. L., Brenner, D., Kelly, G., and Arch. Environ. Health, 43: 83-89, 1988. Santella, R. M. Detection of polycyclic aromatic hydrocarbon-DNA adducts in white blood cells of foundry workers. Cancer Res., 48: 2288-2291, 1988. 6. Schulte, P. A. A conceptual framework for the validation and use of biologic markers. Environ. Res., 48: 129-144, 1989. 33. Rothman, N., Correa-Villasenor, A., Ford, D. P., Poirier, M. C., Haas, R., Hansen, J. A., O’Toole, T., and Strickland, P. T. Contribution of occupation and 7. Hulka, B. S., Wilcosky, T. C., and Griffith, J. D. (eds.). Biological Markers in diet to white blood cell polycyclic aromatic hydrocarbon-DNA adducts in wild- Epidemiology. New York: Oxford University Press, 1990. land firefighters. Cancer Epidemiol., Biomarkers & Prey., 2: 341-347, 1993. 8. Schatzkin, A., Freedman, L. S., Schiffman, M. H., and Dawsey, S. M. VaIl- 34. Singer, B., and Grunberger, D. Molecular Biology of Mutagens and Carcin- dation of intermediate end points in cancer research. J. NatI. Cancer Inst., 82: ogens. New York: Plenum Publishing Corp., 1983. 1746-1752, 1990. 35. Hemminki, K. DNA adducts, mutations and cancer. Carcinogenesis (Land.), 9. Willett, W. Future research directions. In: W. Willett (ed), Nutritional Epi- 14: 2007-2012, 1993. demiology, pp. 380-385. New York: Oxford University Press, 1990. 36. Ross, R. K., Yuan, J-M., Yu, M. C., Wogan, G. N., Qian, G-S., Tu, J-T., 10. Hulka, B. S. Epidemiological studies using biological markers: issues for Groopman, J. D., Gao, Y-T., and Henderson, B. E. Urinary aflatoxin biomarkers epidemiologists. Cancer Epidemiol., Biomarkers & Prey., 1: 13-19, 1991. and risk of hepatocellular carcinoma. Lancet, 339: 943-946, 1992. 1 1 . Hulka, B. S., and Margolin, B. H. Methodological issues in epidemiologic 37. Haugen, A., Becher, G., Benestad, C., Vahakangas, K., Trivers, G. E., studies using biologic markers. Am. J. Epidemiol., 135: 200-209, 1992. Newman, M. J., and Harris, C. C. Determination of polycyclic aromatic hydro- 12. Schulte, P. A., Rothman, N., and Schottenfeld, D. Design considerations in carbons in the urine, benzo(a)pyrene diol epoxide-DNA adducts in lymphocyte molecular epidemiology. In: P. A. Schulte and F. P. Perera (eds.), Molecular DNA, and antibodies to the adducts in sera from coke oven workers exposed to Epidemiology: Principles and Practices, pp. 159-198. New York: Academic measured amounts of polycyclic aromatic hydrocarbons in the work atmosphere. Press, 1993. Cancer Res., 46: 4178-4183, 1986. 13. Rothman N. Epilogue. In: P. A. Schulte and F. P. Perera (eds.), Molecular 38. Mooney, L. A., Ottman, R., Tsai, W. Y., Santella, R. M., Jeffrey, A., Wazneh, Epidemiology: Principles and Practices, pp. 565-569. New York: Academic L., Covey, L., Glassman, A. H., Cooper, T., Young, T. L., and Perera, F. P. Press, 1993. Decline in PAH-DNA and 4-aminobiphenyl-hemoglobin adducts (4-ABP-Hb) in peripheral blood after smoking cessation. Proc. Am. Assoc. Cancer Res., 35: 101, 14. Steenland, K., Tucker, J., and Salvan, A. Problems in assessing the relative 1994. predictive value of internal markers versus extemal exposure in chronic disease epidemiology. Cancer Epidemiol., Biomarkers & Prey., 2: 487-491, 1993. 39. Perera, F., Mayer, J., Jaretzki, A., Hearne, S., Brenner, D., Young, T. L., Fischman, H. K., Grimes, M., Grantham, S., Tang, M. X., Tsai, W-Y., and 15. Begg, C. B., and Zhang. Z-F. Statistical analysis of molecular epidemiology Santella, R. M. Comparison of DNA adducts and sister chromatid exchange in studies employing case-series. Cancer Epidemiol., Biomarkers & Prey., 3: lung cancer cases and controls. Cancer Rca., 49: 4446-4451, 1989. 173-175, 1994. 40. Tang, D. L., Santella, R. M., Blackwood, M. A., Warburton, D., Luo, J., 16. Piegorsch, W. W., Weinberg, C. R., and Taylor, J. A. Non-hierarchical Young, T. L., Mayer, J., Tsai, W-Y., and Perera, F. P. A case-control molecular logistic models and case-only designs for assessing susceptibility in population- epidemiology study of lung cancer. Proc. Am. Assoc. Cancer Res., 34: 263, 1993. based case-control studies. Stat. Med., /3: 153-162, 1994. 41. Perera, F. P., Poirier, M. C., Yuspa, S. H., Nakayama, J., Jaretzki, A., Cumen, 17. McMichael, A. J. “Molecular epidemiology”: new pathway or new travelling M. M., Knowles, D. M., and Weinstein, I. B. A pilot project in molecular cancer companion. Am. J. Epidemiol., 140: 1-1 1, 1994. epidemiology: determination of benzo[a]pyrene-DNA adducts in animal and 18. Lander, E. S., and Schork, N. J. Genetic dissection of complex trails. Science human tissues by immunoassays. Carcinogenesis (Land.), 3: 1405-1410, 1982. (Washington DC), 265: 2037-2048, 1994. 42. van Schooten, F. J., Hillebrand, M. J. X., van Leeuwen, F. E., Lutgerink, J. T., 19. Liu, Y., Hernandez, A. M., Sihbata, D., and Cortopassi, G. A. BCL2 trans- van Zandwijk, N., Jansen, H. M., and Kriek, E. Polycyclic aromatic hydrocarbon- location frequency rises with age in humans. Proc Natl. Acad. Sd. USA, 9/: DNA adducts in lung tissue from lung cancer patients. Carcinogenesis (Land.), 8910-8914, 1994. II: 1677-1681, 1990.

43. Hagmar, L., Brcsgger, A., Hansteen, l-L., Heim, S., H#{246}gstedt, B., Knudsen, phism and low-level environmental exposure to carcinogens. Nature (Land.), 369: L., Lambert, B., Linnainmaa, K., Mitelman, F., Nordenson, I., Reuterwall, C., 154-156, 1994. Salomaa, S., Skerfving, S., and Sorsa, M. Cancer risk in humans predicted by 57. Hem, D. W. Acetylator genotype and arylamine-induced carcinogenesis. increased levels of chromosomal aberrations in lymphocytes: Nordic study group Biochim. Biophys. Acta, 948: 37-66, 1988. on the health risk of chromosome damage. Cancer Res., 54: 2919-2922, 1994. 58. Nakachi, K., Imai, K., Hayashi, S-I., Watanabe, J., and Kawajiri, K. Genetic 44. Bonassi, S., Abbondandolo, A., Camurri, L, Pm, L D., Ferrari, M. D., susceptibility to squamous cell carcinoma of the lung in relation to cigarette Degrassi, F., Fomi, A., Lamberti, L., Lando, C., Padovani, P., Sbrana, I., Vecchio, smoking dose. Cancer Res., 51: 5177-5180, 1991. D., and Puntoni, R. Are chromosome aberrations in circulating lymphocytes 59. Kihara, M., Kihara, M., and Noda, K. Lung cancer risk of GSTM1 null predictive of future cancer onset in humans? Preliminary results of an Italian genotype is dependent on the extent of tobacco smoke exposure. Carcinogenesis cohort study. Cancer Genet. Cytogenet., in press, 1995. (Land.), 15: 415-418, 1994. 45. Kirsch, I. R., and Lipkowitz, S. A measure of genomic instability and its 60. Stewart, W. F., and Stewart, P. A. Occupational case-control studies: I. relevance to lymphomagenesis. Cancer Res., 52 (Suppl.): 5545s-5546s, 1992. Collecting information on work histories and work-related exposures. Am. J. Ind. 46. Greenberg, D. A. Linkage analysis of necessary disease loci versus suscep- Med., 26: 297-312, 1994. tibility loci. Am. J. Hum. Genet., 52: 135-143, 1993. 61. Stewart, P. A., and Stewart, W. F. Occupational case-control studies. II. 47. Caporaso, N., and Goldstein, A. Cancer genes: single and susceptibility: Recommendations for exposure assessment. Am. J. Ind. Med., 26: 313-326, exposing the difference. Pharmacogenetics, in press, 1995. 1994. 48. Khoury, M. J., Beaty, T. H., and Cohen, B. H. Fundamentals of Genetic 62. Comstock, K. E., Sanderson, B. J., Claflin, G., and Henner, W. D. GSTI gene Epidemiology. New York: Oxford University Press, 1993. deletion determined by polymerase chain reaction. Nucleic Acids Res., 18: 3670, 49. Khoury, M. J., and Beaty, T. H. Applications of the case-control method in 1990. genetic epidemiology. Epidemiol. Rev., 16: 134-150, 1994. 63. Bell, D. A., Thompson, C. L., Taylor, J., Miller, C. R., Perera, F., Hsieh, L. 50. Wei, 0., Matanoski, G. M., Farmer, E. R., Hedayati, M. A., and Grossman, L., and Lucier, 0. W. Genetic monitoring of human polymorphic cancer suscep- L. DNA repair and aging in basal cell carcinoma: a molecular epidemiology tibility genes by polymerase chain reaction: application to glutathione transferase study. Proc. NatI. Acad. Sci. USA, 90: 1614-1618, 1993. ,.1_ Environ. Health Perspect., 98: 1 13-i 17, 1992. 51. Wu, X., Hsu, T. C., Annegers, J. F., Amos, C. I., Fueger, J. J., and Spitz, M. 64. Petersen, D. D., McKinney, C. E., Ikeya, K., Smith, H. H., Bale, A. E., R. A case-control study of nonrandom distribution of bleomycin-induced chro- McBride, 0. W., and Nebert, D. W. Human CYPIAJ gene: cosegregation of the matid breaks in lymphocytes of lung cancer cases. Cancer Res., 55: 557-561, enzyme inducibility phenotype and an RFLP. Am. J. Hum. Genet., 48: 720-725, 1995. 1991. 52. Rothman, N., Hayes, R. B., Bi, W., Caporaso, N., Broly, F., Woosley, R. L, 65. Butler, M. A., Lang, N. P., Young, J. F., Caporaso, N. E., Vineis, P., Hayes, Yin, S., Feng, P., You, X., and Meyer, U. A. Correlation between N-acetyltrans- R. B., Teitel, C. H., Massengill, J. P., Lawsen, M. F., and Kadlubar, F. F. ferase activity and NAT2 genotype in Chinese males. Pharmacogenetics, 3: Determination of CYPIA2 and NAT2 phenotypes in human populations by anal- 250-255, 1993. ysis of caffeine urinary metabolites. Pharmacogenetics, 2: 1 16-127, 1992. 53. Rothman, N., Stewart, W. F., Caporaso, N. E., and Hayes, R. B.: Misclas- 66. Hollstein, M., Sidransky, D., Volgelstein, B., and Harris, C. C. p53 mutations sification of genetic susceptibility biomarkers: implications for case-control in human cancers. Science (Washington DC), 253: 49-53, 1991. studies and cross-population comparisons. Cancer Epidemiol., Biomarkers & 67. Taylor, J. A., Sandier, D. P., Bloomfield, C. D., Shore, D. L., Ball, E. D., Prey.. 2: 299-303, 1993. Neubauer, A., McIntyre, 0. R., and Liu, E. Ras oncogene activation and occu- 54. Sinha, R., Rothman, N., Brown, E. D., Mark, S. D., Hoover, R. N., Caporaso, pational exposures in acute myeloid leukemia. J. NatI. Cancer. Inst., 84: N. E., Levander, 0., Knize, M. G., Lang, N. P., and Kadlubar, F. F. Pan-fried meat 1626-1632, 1992. containing high levels of heterocyclic aromatic amines but low levels of poly- 68. Hsu, I. C., Metcalf, R. A., Sun, T., Welsh, J. A., Wang, N. J., and Harris, C. cyclic aromatic hydrocarbons induce cytochrome P4501A2 activity in humans. C. Mutational hotspot in p53 gene in human hepatocellular carcinomas. Nature Cancer Res., 54: 6154-6159, 1994. (Land.), 350: 427-431, 1991. 55. McGlynn. K. A., Hu, Y., Shen, F. M., Chen, G. C., Xia, X. L., Baffoe-Bonnie, 69. Suzuki, H., Takahashi, T., Kuroishi, T., Suyama, M., Ariyoshi, Y., Taka- A., Rosvold, E. A., Zhou, T., and Beutow, K. H. Genetic susceptibility to primary hashi, T., and Ueda, R. p53 mutations in non-small cell lung cancer in Japan: liver cancer. Proc. Am. Assoc. Cancer Res., 35: 293, 1994. association between mutations and smoking. Cancer Res., 52: 734-736, 1992. 56. Vineis, P., Bartsch, H., Caporaso, N., Barrington, A. M., Kadlubar, F. F., 70. Groopman, J. D., Wogan, G. N., Roebuck, B. D., and Kensler, T. W. Landi, M. T., Malaveille, C., Shields, P. 0., Skipper, P., Talaska, 0., and Molecular biomarkers for aflatoxins and their application to human cancer Tannenbaum, S. R. Genetically based N-acetyltransferase metabolic polymor- prevention. Cancer Res., 54 (Suppl.): 1907s-191 is, 1994.

Downloaded from cebp.aacrjournals.org on September 26, 2021. © 1995 American Association for Cancer Research. Incorporating biomarkers into cancer epidemiology: a matrix of biomarker and study design categories.

N Rothman, W F Stewart and P A Schulte

Cancer Epidemiol Biomarkers Prev 1995;4:301-311.

Updated version Access the most recent version of this article at: http://cebp.aacrjournals.org/content/4/4/301

E-mail alerts Sign up to receive free email-alerts related to this article or journal.

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Subscriptions Department at [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://cebp.aacrjournals.org/content/4/4/301. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.